Physics With Excel and Python

Dieter Mergel
Physics with Excel

and Python
Using the Same Data Structure
Volume I: Basics, Exercises and Tasks
Physics with Excel and Python
Dieter Mergel
Physics with Excel

and Python
Using the Same Data Structure
Volume I: Basics, Exercises and Tasks
Dieter Mergel
Fakultät für Physik
Universität Duisburg-Essen
Duisburg, Nordrhein-Westfalen, Germany
ISBN 978-3-030-82324-5 ISBN 978-3-030-82325-2 (eBook)

https://doi.org/10.1007/978-3-030-82325-2
Based on the German language edition: Physik mit Excel und Visual Basic by Dieter Mergel,
© 2017 2017. Published by Springer-Spectrum. All Rights Reserved, and extended with Python solutions.
© Springer Nature Switzerland AG 2022
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilms or in any other physical way, and transmission or information
storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology
now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or
the editors give a warranty, expressed or implied, with respect to the material contained herein or for any
errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.
This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Preface
Ψ E xcel—as powerful as necessary, Python—as simple as possible.

This book treats a series of physics exercises that were developed for
courses at the University of Duisburg-Essen for students training to become
physics/mathematics teachers or physics engineers. The exercises were intended to
introduce computational physics based on spreadsheet calculation combined with
simple VBA macros and also to broaden the beginner’s knowledge of physics.
This approach garnered positive reactions from practitioners and resulted in a text-
book in German.1 Furthermore, the methods developed turned out to be powerful
enough to treat a broad range of topics in undergraduate physics, resulting in a
second volume with exercises on particles, waves, fields, and random processes.2
Referees found the exercises interesting and ambitious enough to serve for
undergraduate education. However, concerns arose that spreadsheet techniques,
although useful in the business world, might be a dead end for students who would
be required to use scientific computing in their future research work. Therefore,
the concept has been changed for the present English version. Programming in
Python is now included from the beginning, while the same topics are addressed
as in the German textbook.
The key to all of the exercises is data structures, developed in introductory sec-
tions that explain the physical problems. They serve as an interface to both Excel
and Python, and potentially also to other applications for scientific computation.
To enable this approach, the Excel solutions in this edition use vectorized code
and matrix formulas to mimic broadcasting, an essential Python technique for
creating new arrays.
We feel that this approach is suitable as a low-threshold introduction to scien-
tific computing as early as high school all the way up to undergraduate physics
1 Dieter Mergel, Physik mit Excel und Visual Basic Grundlagen, Beispiele und Aufgaben, Springer
Spektrum (2017), https://doi.org/10.1007/978-3-642-37857-7.
2 Dieter Mergel, Physik lernen mit Excel und Visual Basic, Anwendungen auf Teilchen, Wellen,
Felder und Zufallsprozesse, Springer Spektrum (2018), https://doi.org/10.1007/978-3-662-575

13-0.
v
vi Preface
classes at the university and may also be a good start for students who later choose
to specialize in computational physics.
Our approach is intended to make the student fit for a computer-oriented world,
be it for spreadsheet calculations in business, scientific computing in research, or
mathematics and physics teaching in high school. We take into account that not all
students have the same attitude towards programming; some have to be encouraged
to venture into a new world, whereas others have to be cautioned not to rush into
blind programming.
Duisburg, Germany Dieter Mergel

Contents
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 A Two-Track Didactical Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 What Can You Expect? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 What Do You Need? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.4 Tim, Alac, and Mag . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.5 Didactic Concept . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.6 Subject Matter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.7 Getting Started with Excel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.7.1 Start Menu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.7.2 Spreadsheet Presentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.8 Getting Started with Python . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.9 Skills to Be Trained . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2 Data Structures, Excel and Python Basics . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.1 Introduction: Named Ranges in Excel, Arrays in Numpy . . . . . . . 15
2.2 Characteristics of a Parabola . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2.1 Different Definitions of a Parabola . . . . . . . . . . . . . . . . . . . . 17
2.2.2 Data Structure and Nomenclature . . . . . . . . . . . . . . . . . . . . . 19
2.3 Basic Exercise in Spreadsheet Calculation . . . . . . . . . . . . . . . . . . . . . 19
2.3.1 Cell Addressing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.3.2 Graphical Representation of a Function . . . . . . . . . . . . . . . 22
2.3.3 Smart Legends in Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.3.4 Scroll Bars . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.3.5 Summary: Cell References and Name Manager . . . . . . . 26
2.3.6 What Have We Learned so Far, and How
to Proceed Further? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.3.7 Python Program . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.4 Python and NumPy Basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.4.1 Basic Exercise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.4.2 Data Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.4.3 Python Libraries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.4.4 Numpy Constructions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
vii
viii Contents
2.4.5 Standard Plot Program . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

2.4.6 Formatted Output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
2.5 Matrix Calculations in Excel and Python . . . . . . . . . . . . . . . . . . . . . . 41
2.5.2 Operations on Arrays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.5.3 Matrices in Spreadsheets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.5.4 Matrices in Python . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.6 Four Parabolas and Their Upper Envelope . . . . . . . . . . . . . . . . . . . . . 49
2.6.1 Graphical Representation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
2.6.3 Spreadsheet Calculation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
2.6.4 Python Program . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
2.6.5 Extrema Along Different Axes . . . . . . . . . . . . . . . . . . . . . . . . 56
2.7 Sum of Four Cosine Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
2.7.1 Sound and a Cosine Identity . . . . . . . . . . . . . . . . . . . . . . . . . . 57
2.7.3 Spreadsheet Layout . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
2.7.4 Python Program . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
2.7.5 Producing Labels (as Strings) in Excel and Python . . . . 68
2.8 Questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
3 Formula Networks and Linked Diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . 75
3.1 Introduction: Well-Structured Sheets and Programs . . . . . . . . . . . . 75
3.2 Image Construction for Focusing and Diverging Lenses . . . . . . . . 78
3.2.1 Straight Line Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
3.2.2 Geometrical Image Construction for a Thin
Focusing Lens . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
3.2.3 Imaging Equation with Correct Signs . . . . . . . . . . . . . . . . . 82
3.2.4 Beam Through a Converging Lens that Really
Contributes to the Image . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
3.2.7 Python Program . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
3.3 Doppler Effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
3.3.1 A Formula for All Cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
3.3.2 A Sound Source Passes a Remote Receiver . . . . . . . . . . . 95
3.3.4 Spreadsheet Calculation “Remote Receiver” . . . . . . . . . . . 97
3.3.5 Python Program “Remote Receiver” . . . . . . . . . . . . . . . . . . 97
3.4 Exponentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
3.4.1 Explosive Character of Exponentials . . . . . . . . . . . . . . . . . . 99
3.4.2 General Exponential Function . . . . . . . . . . . . . . . . . . . . . . . . . 101
3.4.3 Representation in a Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . 102
3.4.4 Diode Characteristics I(U) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
Contents ix

3.4.7 Python Program . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
3.5 Questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
4 Macros with Visual Basic and Their Correspondences in Python . . . 115
4.1 Introduction: For, If, Sub/Def . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
4.2 Basic Exercise: For- Loops . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
4.2.1 Visual-Basic-Editor 1: Editing . . . . . . . . . . . . . . . . . . . . . . . . 118
4.2.2 Programming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
4.3 Macro-Controlled Drawings with For, Sub, If . . . . . . . . . . . . . . . . 123
4.3.1 Macro Recorder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
4.3.2 Visual-Basic Editor 2: Macro Recording,
Debugging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
4.3.3 Programming Elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
4.4 A Checkerboard Pattern (Excel) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
4.4.1 Checkerboard, Same-Colored and Multi-colored . . . . . . . 132
4.4.2 Global Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
4.5 A Checkerboard Pattern (Python) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
4.5.1 Turtle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
4.5.2 Differences to Visual Basic . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
4.5.3 Checkerboard with Squares, Triangles, and Circles . . . . 139
4.6 Drawing Densely-Packed Atomic Layers; Crystal Physics . . . . . . 143
4.6.1 Program Structure and Geometry . . . . . . . . . . . . . . . . . . . . . 143
4.6.3 Excel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
4.6.4 Python . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
4.7 Text Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
4.7.1 Cutting and Joining Strings . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
4.7.2 Data Structure and Program Flow . . . . . . . . . . . . . . . . . . . . . 153
4.7.3 Excel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
4.7.4 Programming Step by Step . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
4.7.5 VBA Constructs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
4.7.6 Python . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
4.8 Processing the Protocol of a Measuring Device . . . . . . . . . . . . . . . . 160
4.8.1 Protocol of a Measuring Device . . . . . . . . . . . . . . . . . . . . . . . 161
4.8.2 Detection of Code Words . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
4.8.4 Excel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
4.8.5 Python . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
4.9 User-Defined Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
4.9.1 User-Defined Functions as Add-In . . . . . . . . . . . . . . . . . . . . 168
4.9.2 Scalar Product and Vector Product . . . . . . . . . . . . . . . . . . . . 170
4.9.3 Python . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
4.10 Questions and Tasks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
x Contents
5 Basic Mathematical Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181

5.1 Introduction: Calculus, Vectors, and Linear Algebra . . . . . . . . . . . 181
5.2 Straight-Line Segment Under a Magnifying Glass . . . . . . . . . . . . . 182
5.2.1 Under a Magnifying Glass . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
5.2.4 Plotting Vectors with Python Matplotlib . . . . . . . . . . . . . . . 185
5.3 Differentiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
5.3.1 First and Second Derivative . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
5.3.4 Python Program . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
5.4 Integration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
5.4.1 Area Under a Curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
5.4.2 Length of a Curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
5.4.3 Data Structure and Nomenclature for the Arrays
in the Integration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
5.4.4 Python Program . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
5.4.5 Spreadsheet Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
5.5 Vectors in the Plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
5.5.1 Vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
5.5.4 Python Program . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
5.6 Tangents to and Perpendiculars on a Curve . . . . . . . . . . . . . . . . . . . . 204
5.6.1 At/On a Polynomial and an Ellipse . . . . . . . . . . . . . . . . . . . 204
5.6.3 Python Program . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
5.7 Banked Curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
5.7.1 Cross-Section of the Road . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
5.7.3 Python Program . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
5.8 Weighted Average . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
5.8.1 A Mobile with Two Arms . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
5.8.3 Python Program . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
5.9 Systems of Linear Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
5.9.1 Polynomial and Electrical Network . . . . . . . . . . . . . . . . . . . 215
5.9.3 Spreadsheet Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
5.9.4 Python Programs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
Contents xi
5.10 Some Mathematical Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222

6 Superposition of Movements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
6.1 Introduction: Translations and Rotations . . . . . . . . . . . . . . . . . . . . . . . 227
6.2 Projectile Trajectory with Velocity Vectors (T-T) . . . . . . . . . . . . . . . 229
6.2.1 Projectile Trajectory and Velocity Vectors . . . . . . . . . . . . . 229
6.2.3 Spreadsheet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
6.2.4 Python . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
6.2.5 Animation of Figures with FuncAnimation . . . . . . . . . . . . 234
6.3 Cycloid, Rolling Curve (R-T) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
6.3.1 Trace of a Writing Point Fixed at a Rolling Wheel . . . . 238
6.3.3 Excel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
6.3.4 Python . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244
6.4 Foucault’s Pendulum (T-R) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246
6.4.1 A Lecture Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246
6.4.3 Excel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 248
6.4.4 Python . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250
6.5 Anchor, Deflected Out of Its Rest Position (R-R) . . . . . . . . . . . . . . 251
6.5.1 Deflected Anchor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
6.5.3 Excel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255
6.5.4 Python . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 256
6.6 Wavefronts, Sound Barriers, and Mach Cone (T-T) . . . . . . . . . . . . 258
6.6.1 Emitting Sound Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
6.6.4 Python . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 264
7 Integration of Newton’s Equation of Motion . . . . . . . . . . . . . . . . . . . . . . . . 269
7.1 Introduction: Approximated Mean Value Instead of Exact
Integration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
7.1.1 Newton’s Equation of Motion . . . . . . . . . . . . . . . . . . . . . . . . . 269
7.1.2 Four Methods for Estimating the Average
Acceleration in a Time Segment . . . . . . . . . . . . . . . . . . . . . . 271
7.1.3 Tactical Approaches in Python and Excel . . . . . . . . . . . . . 272
7.2 Harmonic Oscillation with “Progress with Look-Ahead”
and “Runge–Kutta” . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272
7.2.1 Equation of Motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
7.2.4 Python . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 276
xii Contents
7.3 Falling from a (Not Too) Great Height . . . . . . . . . . . . . . . . . . . . . . . . 278

7.3.1 Limiting Cases, Analytically Solved . . . . . . . . . . . . . . . . . . 279
7.3.3 Spreadsheet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 280
7.3.4 Python . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282
7.4 Stratospheric Jump . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283
7.4.3 Python . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 286
7.5 A Car Drives with Variable Power . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 286
7.5.1 Various Types of Power . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287
7.5.3 Excel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291
7.5.4 Python . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295
7.6 Bungee Jump . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 298
7.6.1 Simulation of the Motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 298
7.6.2 Analytical Calculations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 300
7.6.4 Excel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302
7.6.5 Python . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 304
8 Random Numbers and Statistical Reasoning . . . . . . . . . . . . . . . . . . . . . . . 307
8.1 Introduction: Statistical Experiments Instead of Theoretical
Derivations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307
8.2 Equi-Distributed Random Numbers, Frequencies
of Occurrence, Chi2 Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 310
8.2.1 A Spreadsheet Experiment with Random Numbers . . . . 310
8.2.3 Python . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315
8.3 Points Randomly Distributed in a Unit Square . . . . . . . . . . . . . . . . . 318
8.3.1 Creation and Distribution of the Points . . . . . . . . . . . . . . . . 318
8.3.3 Excel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321
8.3.4 Python . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324
8.3.5 Why Calculate Twice? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 326
8.4 Set Operations in Numpy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327
8.4.1 Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327
8.4.3 Python . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 330
8.5 Normally Distributed Random Numbers . . . . . . . . . . . . . . . . . . . . . . . 332
8.5.1 Normal Distribution, Probability Density
and Distribution Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333
8.5.2 Random-Number Generator and Frequencies
of Occurrence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 336
Contents xiii
8.5.3 Where Do Observed and Theoretical Frequencies

Fit Better Together? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 337
8.5.5 Python . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341
8.5.6 Excel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 342
8.6 Random-Number Generator, General Principle . . . . . . . . . . . . . . . . . 345
8.7 Diffraction of Photons at a Double-Slit . . . . . . . . . . . . . . . . . . . . . . . . 348
8.7.1 Physical Background: Wave-Particle Dualism . . . . . . . . . 348
8.7.2 Cos2 Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 350
8.7.4 Python . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 351
8.7.5 Excel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 353
8.7.6 Simulation in a Spreadsheet . . . . . . . . . . . . . . . . . . . . . . . . . . . 355
8.8 Chi2 Distribution and Degrees of Freedom . . . . . . . . . . . . . . . . . . . . 358
8.8.1 Data Structure, Nomenclature . . . . . . . . . . . . . . . . . . . . . . . . . 359
8.8.2 Python . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 360
9 Evaluation of Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 365
9.1 Introduction: We Know Everything and Play Stupid . . . . . . . . . . . 365
9.2 Weighing a Glass Substrate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 369
9.2.1 Discussion on the Accuracy of a Balance . . . . . . . . . . . . . 369
9.2.3 Excel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 370
9.2.4 Python . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 371
9.3 A Procedure for Rounding to Relevant Digits . . . . . . . . . . . . . . . . . . 372
9.3.1 Numerical Evaluations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 372
9.3.3 Python Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373
9.3.4 VBA Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373
9.4 Increasing the Measuring Accuracy Through Repetition . . . . . . . . 374
9.4.1 Standard Deviation and Standard Error
of the Mean Value of a Measurement Series . . . . . . . . . . . 375
9.4.3 Python Program . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 379
9.4.4 Spreadsheet Layout for This Task . . . . . . . . . . . . . . . . . . . . . 380
9.4.5 How to Report a Measurement Result . . . . . . . . . . . . . . . . . 381
9.5 The t Statistics Connects Confidence Interval
with Confidence Level . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 382
9.5.1 Student’s t Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 382
9.5.4 Python Program . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 388
9.6 Combining Results from Several Measurement Series . . . . . . . . . . 389
9.6.1 Combining Two Measurement Results . . . . . . . . . . . . . . . . 390
xiv Contents
9.6.2
Data Structure and Nomenclature . . . . . . . . . . . . . . . . . . . . . 393
9.6.3
Spreadsheet Calculation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 394
9.6.4
Python, Internally and Externally Consistent Error
of the Combined Result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 395
9.7 Propagation of Standard Deviations . . . . . . . . . . . . . . . . . . . . . . . . . . . 397
9.7.1 Rules for Propagation of Standard Deviations . . . . . . . . . 397
9.7.4 Python Program . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 404
9.8 Propagation of Confidence Intervals . . . . . . . . . . . . . . . . . . . . . . . . . . . 407
9.8.1 From Variance to Confidence . . . . . . . . . . . . . . . . . . . . . . . . . 407
9.8.2 Sum and Product of Two Measurands . . . . . . . . . . . . . . . . . 408
9.9 Mass of a Thin Film on a Glass Substrate . . . . . . . . . . . . . . . . . . . . . 409
9.9.1 Instructions for Use for Accurate Measurements
and Their Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 410
9.9.4 Python Program . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 415
10 Fitting Trend Curves to Data Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 419
10.1 Introduction: Linear and Nonlinear Regression . . . . . . . . . . . . . . . . . 419
10.1.1 Straight Line Through Data Points by Sight . . . . . . . . . . . 419
10.1.2 Multilinear Regression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 420
10.1.3 Nonlinear Regression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 421
10.1.4 Coefficient of Determination R2 . . . . . . . . . . . . . . . . . . . . . . . 422
10.1.5 C-spec Error with Iterative t Adaptation . . . . . . . . . . . . . . . 423
10.2 Linear Trend Line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 424
10.2.1 Creating Data Points and Evaluating Them . . . . . . . . . . . . 424
10.2.3 Spreadsheet Calculation with Linest . . . . . . . . . . . . . . . . . . . 426
10.2.4 Python Program . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 429
10.3 Fitting a Polynomial Trend Line to Data Points
with Multilinear Regression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 431
10.3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 431
10.3.4 Python Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 437
10.4 Exponential Trend Line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 440
10.4.1 Exponential and Logarithm . . . . . . . . . . . . . . . . . . . . . . . . . . . 440
10.4.2 Exponential or Polynomial? . . . . . . . . . . . . . . . . . . . . . . . . . . . 444
10.4.4 Python Program . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 445
Contents xv
10.5 Solving Nonlinear Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 448

10.5.1 Intersection of Straight Lines with a Parabola . . . . . . . . . 448
10.5.4 Python Program . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 455
10.6 Temperature Dependence of the Saturation Magnetization
of a Ferromagnet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 456
10.6.1 Langevin Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 456
10.6.4 Python . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 459
10.7 Fitting Gaussians to Spectral Lines with Nonlinear
Regression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 460
10.7.1 Fitting the Sum of Two Gaussians to Data Points . . . . . . 460
10.7.2 C-spec Errors of the Coefficients by a Statistical
Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 463
10.7.4 Python . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 465
10.7.5 Spreadsheet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 468
10.7.6 C-spec Error of the Optimized Coefficients
by Simulation-Based t Adaptation . . . . . . . . . . . . . . . . . . . . 470
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 475
About the Author
Dieter Mergel studied physics in Göttingen, obtained his doctorate at the Techni-
cal University of Clausthal in the field of solid-state physics, and worked 11 years
in the Philips Research Laboratories Hamburg/Aachen on automatic speech recog-
nition and optical data storage. Since 1993, he is Professor of Technical Physics
at the University of Duisburg-Essen. His professional activities include research
in the field of solid-state layers and lectures for students in teaching and medical
professions.
xvii
Introduction
1
Possible errata and corrections in the internet at: go.sn.pub/Ob4vCR.

For every chapter, solutions for each two exercises in Excel and Python can be
found at internet adresses.
1.1 A Two-Track Didactical Approach
History
The exercises in this book arise from a German textbook that emerged from courses
for prospective teachers and students of Technical Physics at the University of
Duisburg-Essen with the intention to prepare the students for a computerized world.
The participants in the courses had already been studying physics for at least one
year. However, the explanations of the exercises are so explicit that they should also
be suitable for beginners.
Said courses are based on excel and Visual Basic (VBA). The current English
version includes Python from the very beginning so as to make it more generally
useful for students who later choose to dive deeper into Scientific Computation.
Exercises
The subject matter is presented in nine chapters as a series of exercises. Every exercise
consists of three steps:
1. The physical concept is introduced with mathematical equations and diagrams.
2. An adequate data structure is set up independent of the implementation in a
particular programming platform, but taking care that the same nomenclature
can be used in both mathematical equations and programming. This serves as an
interface to any programming application.
3. Solutions in excel and Python are designed so that a solution in one application
can directly be translated into the other one.
© Springer Nature Switzerland AG 2022 1

D. Mergel, Physics with Excel and Python,
https://doi.org/10.1007/978-3-030-82325-2_1
2 1 Introduction
To enable this approach, training in excel emphasizes vectorized code, matrix for-
mulas, and constructs that allow for broadcasting in the same way as Python.
Furthermore, programming VBA macros interacting with spreadsheets introduces
looping, logical queries, and functions.
Didactical advantages of the two-track system

We strive to combine the didactical advantages of both programming applications:
– Spreadsheets are interactive; charts react immediately to changes, and data

structures (but not formulas) are immediately visible in spreadsheets.
– In Python, formulas (but not data structures) are immediately visible.
– The VBA interpreter allows us to run the program step by step and to watch
intermediate results in a spreadsheet, or read the content of variables by mouse-
over in the code.
Our guideline: spreadsheet calculations as powerful as necessary, Python program-

ming as simple as possible.
Emphasis is not on mere computational techniques, but on exercises that may
be regarded as small projects so that project-related difficulties manifest and can be
addressed, e.g., by checking the consistency of equations and numerical solutions
with limiting cases that can be solved analytically.
We maintain that this approach is not only suitable as a low-threshold introduction
to scientific computing as early as High School and up to undergraduate Physics
classes at University. It should also be a good start for students who later choose to
specialize in Computational Physics or, more generally, for professional use.
1.2 What Can You Expect?
What can you, dear reader, expect from this book?

You can expect to be introduced to the world of Python and excel by:
– training to work with numpy arrays, list slicing and broadcasting in Python,
– working with similar constructs, vector structures, and matrix operations also
in excel,
– learning how to write programs with looping, logical queries, and functions in
Python and VBA for excel,
– training how to lay out spreadsheets clearly so that they are apt for simple
scientific computing,
– developing VBA macros that exchange data with spreadsheets,
– applying standard mathematical methods numerically.
and, with that,
– getting a better understanding of certain mathematical and physical concepts.

1.2 What Can You Expect? 3
After having successfully completed the exercises, you should have gained so
much self-confidence that you can answer the question “Programming practice?”
with an enthusiastic “Yes!”.
1.3 What Do You Need?
You will need a Physics textbook (anyone will do, e.g., the one you have at hand
during your studies anyway) and two more books on programming as indicated
below.
EXCEL
To work with excel, you only need a computer in which excel has been imple-
mented (any version; the exercises in this book have been checked in excel 2010
and excel 2019) and an introduction to excel (do not buy one before having done
the basic exercise in Sect. 2.3). In particular, you do not need a special development
environment for visual basic, because it is included in all versions of excel.
Python
You will need to install Anaconda, a free and open-source distribution of the Python
and R programming languages that also comprises the Jupyter Notebook by
default. The examples in this book were obtained with Python 3.7 in Jupyter.
You are advised to use both a book and internet courses to broaden your training
systematically. Make your choice after having gone through Exercises 2.4 and 2.5.
1.4 Tim, Alac, and Mag
You will soon meet two types of students and a tutor who will accompany
us throughout this book. The character named Tim (which stands for “timidus”
or “timida”, meaning shy)) represents those students who are somewhat hesi-
tant, fearing that they may fall short of the requirements, although they study
hard. The character named Alac (which stands for “alacer”, meaning alacritous,
high-flying) is typical of those vehemently self-confident students (men are gen-
erally over-represented) who believe that they already have a superior overview
and do not have to deal with what they consider mere bits and pieces. Mag
(for Magister/Magistra, i.e., the tutor who runs the course) tries to engage with
both characters, encouraging Tim and cautioning Alac, and clarifying that both
approaches are valuable and that every Physics student should venture into the
Computer world.
4 1 Introduction
Tim: Computers are not my thing
Tim I see how well some fellow students are juggling programming tools, but
I’d rather stand back. I prefer to learn the stuff from textbooks.
Mag This course is not intended to turn you into a computer nerd. You will not
learn any cool tricks. We restrict ourselves to some basic techniques that are prac-
ticed again and again. The computational techniques do not stand by themselves,
but are always taught in connection with physical problems.
Tim But I have often heard that programming is a black art for which you have
to be specially talented.
Mag Here, you will learn the most basic computer techniques that every
scientist, engineer, and science teacher must master to succeed in their profession.
Alac Why excel and Python?
Mag All algorithm-oriented computer languages have a similar structure.

Knowledge of specific commands is not the most important thing. You have to
learn how to translate physical and technical problems into a computational struc-
ture. Furthermore, the mistakes that beginners make are always the same in all
computer languages. The most important thing is to track, correct, and, finally,
avoid them. Anyhow, as we shall have spreadsheet solutions and Python pro-
grams in parallel, you will be sensitized more towards common structures than the
peculiarities of specific software.
Alac: How do I become a master programmer?
Mag A master can be recognized by how he/she deals with errors. Any unno-
ticed error in spreadsheet formulas and programs can lead to disaster. It is essential
that you gain experience with data structures and programming constructs.
Alac And that is what this course will accomplish?
Mag Yes! By using data structures in spreadsheets and Python programs and
setting up graphical representations that are comprehensible, even when you look
at them after some time. And by developing simple procedures that control the
program flow.
Tim Data structures, procedures, controlling; that sounds pretty challenging.

How can I learn all of that?
Mag Let’s compare this course with learning a foreign language. How do you
learn foreign languages?
1.4 Tim, Alac, and Mag 5
Alac Learning? For foreign languages, academic learning is useless in the long
run. You simply have to go abroad; the rest follows by itself.
Tim Oh, I couldn’t learn like that. I couldn’t form a proper sentence in a for-
eign language without profound foreknowledge. I would have to learn the correct
grammar and vocabulary first before I would dare to speak.
A good balance
Mag We are trying to find a good balance. You will learn the most straightfor-
ward “sentence” structures, but will also be “sent abroad” right off, and you will
have to make your way there. If you pass this test, you can be confident of being
able to learn the more complex “grammar” if necessary.
Tim Is that thorough enough?
Alac Will I learn the more tricky constructions too late?
Mag Don’t worry! Working through this book will make you fit for a computer-
oriented world, be it for spreadsheet calculations in business or scientific comput-
ing in research. This can be tedious, but it will be worthwhile, whether it be as
early as learning at school or working for a Bachelor’s or Master’s, or even as late
as working on a Ph.D. thesis.
Tim Can I manage this in addition to my studies in Physics?
Mag I think so. Anyway, this course is about physics and will help you to pass
your exams.
1.5 Didactic Concept
Workshop atmosphere
Having cleared up the doubts harbored by Tim and Alac, we now explain the didactic
concept of this book.
In the courses at the University of Duisburg-Essen on Physics with Excel and
visual basic, learning was mostly done in a workshop, such as in physics labs for
beginners. The students dealt with the tasks alone or in pairs while in a computer
lab, ideally also helping each other out across groups and consulting the supervisor
when needed. Students could continue to work on their tasks outside of attendance
time so that everyone could work according to their learning progress.
Experience shows that the students enjoy the tasks, and the learning progress is
fastest when all three aspects—programming, physics, and mathematics—are com-
bined. The systematic practice of various isolated spreadsheet and programming
techniques is often perceived as too dull. The combination of calculations and graphs,
realized in nearly all exercises in this book, proved to be particularly instructive.
6 1 Introduction
Courses with 30 attendance hours

At the University of Duisburg-Essen, two excel-based courses were offered, each
with 30 attendance hours:
– a basic course for beginners, in which two tasks from each of the six Chaps. 2, 3,
4, 8, 9 and 10 were worked on and had to be presented to the supervisor;
– an advanced course with two tasks each from Chaps. 5, 6 and 7, and one task that
had not yet been worked on from the chapters of the beginner’s course. Sometimes,
two short exercises were combined into one task.
Subjects from a one-year physics course

The exercises rely on the subject matter from the first year of undergraduate physics.
We do not intend to repeat physics that can be found in standard textbooks. Therefore,
the introductions to the tasks are concise, but the solutions are presented in great
detail. Experience shows that this creates the risk that the students might work through
the exercises mechanically without caring about the physics context. To counteract
this tendency, simple questions regarding physics and programming are asked in the
middle of the text and answered in footnotes.
In Chaps. 8, 9 and 10, statistical concepts are illustrated in greater detail through
simulations, because many students lack basic knowledge in this area. Although
no theorems are logically derived, their structure should become clear, because the
simulations follow the mathematical ideas.
Simple solutions preferred

The material is presented in nine chapters, each featuring about five detailed exer-
cises. The aim is to pursue clear and simple solutions in which the physical
justification for each step is traceable. To achieve this, suboptimal solutions, sub-
optimal with respect to computational efficiency and numerical precision, are often
presented instead of solutions that are perfect from the outset. It has proven to be
didactically more efficient to point out the shortcomings of this first approach and
give the reader tools for improvements.
Broom rules
To many beginners, spreadsheet calculations and, especially, computer programming
seem like witchcraft. We like to address this idea by setting up “broom rules” that
the students hopefully will not forget so easily. Some examples: “ Half, half, full;
the halves count twice” (Runge–Kutta of the 4th order) or “ Mostly, not always.
(“fundamental rule” of statistical reasoning, no statement is 100% sure).
In addition, Mag puts stumbling blocks along the learning path, in talks with the
two student characters, Tim, who learns the material from the beginner’s course dili-
gently, and Alac, who does not hesitate to implement premature ad-hoc solutions. It
is important to emphasize that both attitudes have their advantages and shortcom-
ings, and neither student should feel denigrated. It is just that some students have
to be encouraged to venture into the programming world, whereas others have to be
cautioned against rushing too quickly into coding.
1.5 Didactic Concept 7
Exam questions
At the end of every chapter, a collection of rehearsals and tasks is presented, typically
requested in written and oral examinations.
1.6 Subject Matter
Block A, Fundamentals (Chaps. 2, 3 and 4)

Ψ The dollar makes it absolute.
The student will learn how to organize spreadsheets, design charts reasonably,
and implement simple programming procedures. The necessary computational tech-
niques are embedded in Physics tasks and trained with clearly arranged formula
calculations, presentations, interpretations of curves, and simple mathematics. The
reader should consult, in parallel, systematic introductions to Excel and Python
for help.
Block B, Physics and Mathematics

Ψ Half, half, whole, the halves count twice.
In Chap. 5, the reader will find exercises for analysis and vector and matrix
calculations in the form of a spreadsheet-specific introduction to mathematics with
parallel Python programs.
In Chap. 6, the knowledge gained in Chaps. 2 to 5 is applied to the kinematic
superposition of movements, including animated charts.
In Chap. 7, we deal with various methods for solving Newton’s equation of motion
and apply them to one-dimensional motions, such as a jump from the stratosphere,
Exercise 7.4, or a bungee jump, Exercise 7.6.
Block C, Simulation and analysis of experiments

If in doubt, count!
Evaluation of measurements is regarded as a critical skill to be exercised at the
beginning of studies in Physics. Therefore, this block is particularly detailed and
illustrated with simulations based on chance, because, as experience shows, many
students’ most significant knowledge gaps are in the field of probability and statistics.
Furthermore, statistics is the branch of mathematics that is most important outside
of technical professions.
We intend to develop a good understanding of concepts through statistical experi-
ments with random numbers without going deep into formal mathematics. Statistical
rules are intended to be illustrated and checked through multiple repetitions of sim-
ulations designed to test the hit rate (“Does the error range capture the true value?”).
For this purpose, random number generators are introduced in Chap. 8, e.g., for
normal distributions.
The student will learn how to analyze and graphically represent measurements
(Chap. 9), emphasizing the precise meaning of error ranges (“C-spec errors” related
to confidence levels). Before this can be done, the measurement process must be
8 1 Introduction
simulated realistically to obtain data that can be evaluated. Our tools for simulation
are random numbers generated according to the desired distribution.
With linear regression, mathematical functions are fitted to sets of measured values
to get trend lines through data points (Chap. 10). Furthermore, an introduction to
the important technique of non-linear regression with solver functions will be given,
again, in both excel and Python.
Follow-up book A follow-up book, “Physics with Excel and Python, Using the
Same Data Structure. Applications”, is being prepared, dealing, in the same style,
with advanced topics, structured according to physical and mathematical aspects,
such as:
– properties of oscillations,
– motions in the plane,
– the steady-state Schrödinger equation,
– partial differential equations,
– Monte Carlo methods,
– wave optics,
– statistical physics, and
– variational calculus.
1.7 Getting Started with Excel
1.7.1 Start Menu
In Fig. 1.1, you see the start menu of excel 2019, where the main tab for-
mulas has been activated, and the cursor has been positioned over the group
Ribbon
Formula bar
Command
Column header
Active cell
Handle
Row number
Fig. 1.1 The start menu of excel, with the main tab formulas activated
1.7 Getting Started with Excel 9
function library to show the command math&trig. Arrows indicate the dif-
ferent elements of the start menu, namely ribbon, tab, group, formula bar, and
command, as well as elements of the working area, column header, row num-
ber, active cell, and handle of a cell. To indicate a “click path” in the text,
we write a sequence tab/group/command/function, e.g., formulas/function
library/math&trig/cos to call the cosine function.
Throughout this book, we will take screenshots from excel 2019. Experience
has shown that students can work with these instructions in every version of excel
without major difficulties.
1.7.2 Spreadsheet Presentation
structure of a spreadsheet
In Fig. 1.2 (S), a spreadsheet organization in the structure, often employed in our
exercises, is shown. With (“gamma”), we refer to the straight lines above C14:G14
and to the left of C14: C174.
Above :
– the parameters of the task are defined in the range C2:C6,

– these cells get the names in B2:B6, with which they can be called in formulas,
– the most informative parameters of the exercise are integrated into a text, here, in
cell E4 (with the formula in E5) that can be taken as a legend in a figure.
Left of :
– the values for the independent variable t are in B14: B174.
A B C D E F G H
1 Prespecifications
2 Amplitude of pendulum Ap 1.50
3 Period of pendulum Tp 1.20
4 Period of rotation Tr 9.00 Tp=1.2; Tr=9
5 Time interval dt 0.0173 ="Tp="&Tp&"; Tr="&Tr
6 Suspension point vs. rot. axis xSh 0.00
7 Calculated therefrom
8 Angular frequency pendulum wP 5.24 =2*PI()/Tp
9 Angular frequency rotating disc wR -0.70 =-2*PI()/Td
10
c disc
ot. dis n rot.
. on r tylo o
ndu lum ace pend c e s
11 Pe Tr Tra
t) +xSh *t) t)
(wP * (wR wR*t) (wR* wR*t)
+dt COS xP*COS SIN( Ap*COS Ap*SIN(
12 =B14 =Ap* = =xP* = =
13 t xP xT yT xSt ySt
14 0.000 1.50 1.50 0.00 1.50 0.00
15 0.017 1.49 1.49 -0.02 1.50 -0.02
174 2.768 -0.52 0.18 0.49 -0.53 -1.40
Fig. 1.2 (S) Typical structure of a spreadsheet, here, for the calculation of the trace of a Foucault
pendulum; rows 16–173 are hidden
10 1 Introduction
1 Sub Protoc() Range("C3") = Tp 8

2 r2 = 13 Cells(r2, 10) = Tp 9
3 Cells(r2, 10) = "Tp" Cells(r2, 11) = Range("D174") 10
4 Cells(r2, 11) = "xT" Cells(r2, 12) = Range("E174") 11
5 Cells(r2, 12) = "yT" r2 = r2 + 1 12
6 r2 = r2 + 1 Next Tp 13
7 For Tp = 1 To 9 End Sub 14
Fig. 1.3 (P) Log procedure, changes the period of the rotational motion in Fig. 1.2 (S) and logs
the values of x T and yT at the last instant of the calculation period
Below :
– values are calculated from the parameters and independent variables,

– the range C14:G174 contains five columnar vectors of length 171 with the names
in row 13,
– row 12 includes, in oblique orientation and in italics, the text of the formulas in the
bold-printed cells of the column below. If no cell is printed in bold, the formula
applies to the entire column.
Nomenclature
Python-typical terms are printed in the Courier font.
When excel-typical terms are referred to in the text, e.g., function names, they are
set in small caps; examples: if(condition; then; else) . Spreadsheet formulas
are given in the form B15 = [=B14 + dt]. The expression in rectangular brackets
corresponds exactly to the entry in the cell, including the equal sign. The equal sign
specifies that it is a formula that is in the cell.
Three types of figure are distinguished, two of which are denoted by suffixes, (S)
for spreadsheets, e.g., Fig. 1.2 (S), and (P) for the code of Visual Basic programs,
e.g., Fig. 1.3 (P). Figures without a suffix are line drawings or screenshots, e.g.,
Fig. 1.1.
Names given by the programmer are printed in the text in italics, e.g., f, d. Some-
times in excel, names are used that contain a dot, e.g., “T.1” or “x.2”. This is because
T1 and X2 are reserved for cell addresses. The associated variables are referenced
in the text without a dot, but with subscripts, i.e., as T 1 and x 2 .
Physical units
Sometimes, no physical units are specified in the axis labels of the figures. They can
then be deduced from the physical units of the parameters.
1.8 Getting Started with Python
You first have to install Anaconda with Python. There are many instructions on
the Internet as to how to achieve that, e.g., https://docs.anaconda.com/anaconda/ins
tall/windows/ or https://www.jcchouinard.com/install-python-with-anaconda-on-
windows/ (2020-09-02).
1.8 Getting Started with Python 11
Fig. 1.4 Window opened to create a new program file
When the Jupyter Notebook is opened, an overview of the filers and single
files on the localhost is shown. The programs used for this book are in a sub-filer
“Py PhExI” of the main filer “Python”. When we click Python/Py PhEx I, the
window in Fig. 1.4 pops up. To edit an already existing file, we have to click on
that file.
To create a new file, we open the list “New” and click on “Python 3”. A new
window pops up, opening a new file “Untitled22” with an empty program
cell “In [ ]”. The version in Fig. 1.5 is displayed after a small program has
been written into that cell. “In [5]” indicates that the 5th version of the code is
shown. This program is executed by clicking the button “Run”. The result of the
instruction print[x] is displayed in the output cell created automatically below
the program cell.
1.9 Skills to Be Trained
The different programming techniques are distributed over various exercises. For
the purpose of learning about them and how to revise them, the following lists
of keywords and broom rules have been compiled. They are meant to assist
the readers with the revision of subjects and, of course, their preparation for
examinations.
Spreadsheet operations (Exercise 2.3)
– Using cell addressing, absolute, relative, indirect

– The dollar makes it absolute.
– Naming cell ranges and using the names in formulas
12 1 Introduction
Fig. 1.5 A program creating an output just below the program cell
– Using sliders to change cell contents

– Scaling and formatting XY scatter diagrams
– Creating smart legends by linking text and variables, “Text” & Variables.
– Gamma structure of tables (Sect. 1.7.2)
– Empty lines separate curves.
– Ctrl + Shift + Enter. Magic “chord” to complete the entry of matrix functions
in excel (Exercise 2.6).
Python constructs (explained mainly in Exercises 2.4 and 2.5)
– Use of numpy arrays

– Ab-initio constructors np.arange, np.linspace
– Creating one and two-dimensional arrays with np.array (row vectors, column
vectors, matrices)
– Broadcasting row vectors, column vectors, and matrices together in algebraic
operations
– Slicing of lists
– List comprehension
– Creating smart legends by linking text and variables
– Applying a standard function to create scatter diagrams
– Creating animated figures (Exercise 6.2).
VBA-macros and Python instructions (Chap. 4)

The terms ‘routines’, ‘programs’, and ‘procedures’ are all used synonymously here.
The term ‘macro’ is for VBA only.
1.9 Skills to Be Trained 13
– For, if , Sub/def , basic structures of programming: loops, logical queries, subrou-

tines (Exercises 4.4 and 4.5)
– Loop2i, loops with a loop index and a running index incremented within the loop
(Exercises 4.2 and 4.8)
– Systematically modifying parameters and recording the results of the spreadsheet
calculations with rep-log procedures
– Processing and decoding texts for evaluation of the protocols of measuring
instruments (Exercises 4.8 and 4.9)
– Writing formulas into cells with procedures (Exercises 4.1 and 4.2)
– Creating user-defined functions and using them in spreadsheet calculations
(Exercise 4.9)
– Linking macros to control elements (command buttons, sliders) in spreadsheets
(Sect. 4.3.3).
Mathematical techniques
Ψ Imaging equation for lenses with plus and minus! (Exercise 3.2).
– Using the line equation constructively (Exercise 3.2)

– Calculating with vectors in the plane and displaying them in diagrams (Exercises
5.5, 5.6, 5.7)
– Calculating with matrices (Exercise 5.9)
– Converting polar coordinates and Cartesian coordinates into one another (Chap. 6)
– Ψ Doppler effect with plus and minus (Exercise 3.3)
– Differentiating (Exercise 5.3) and integrating (Exercise 5.4) numerically
– Weighted sum (Exercises 5.8 and 6.5) and weighted mean (Exercise 6.5).
Functions
– Properties of the exponential function (Exercise 3.4)

– First, the tangent at x = 0! (Exercise 3.4)
– Plus 1 becomes times e. (Exercise 3.4)
– Use of the logarithm function for different computing tasks (Exercise 9.3)
– Addition of sines and cosines: overtones and beats (Exercise 2.7)
– Cos plus Cos equals mean value times half the difference. (Sum rule of cosines,
Exercise 2.7).
Integration of Newton’s equation of motion (Chap. 7)
– Approximated average value instead of exact integral

– Four numerical methods (Sect. 7.1.2, Exercise 7.2):
– Euler
– Progress with look-ahead (our standard procedure in a spreadsheet calculation)
– Half-step
– Runge-Kutta of fourth order, Ψ half, half, whole, the halves count twice. (our
standard procedure as a Python function).
14 1 Introduction
Statistics (mainly Chap. 8)
– Mostly, not always. Fundamental rule of statistical reasoning

– If in doubt, count!
– Ψ Come to a decision! You may be wrong. (Exercise 9.6)
– Generating random numbers with specified distribution (Exercises 8.5, 8.6, 8.7)
– Ψ Chance is blind and checkered. (Exercise 8.3)
– Empirical frequency distribution (Exercise 8.2), in excel: Always one more!
Yes, but of what and than what? In Python: Always one less!
– Chi2 test for comparing theoretical and empirical frequency distributions (Exer-
cises 8.2, 8.8)
– Multiple repetitions of random experiments to test for uniform distribution of the
results of Chi2 tests (Exercise 8.2).
Evaluation of measurements (mainly Chaps. 9, 10)
– We know everything and play stupid.

– Simulating measuring processes and evaluating the generated data sets statistically
(Exercises 9.2, 9.9).
– Mean, standard deviation, Two within, one out of (the standard error range)
(Exercises 9.4 to 9.8)
– Specifying measurement uncertainty (Exercises 9.2, 9.7 and 9.8)
– Multiple repetition of random experiments to test for error rate (Exercises 9.4,
9.5 and 10.3)
– For only a few repetitions of measurements, taking the t-value into account
(Exercise 9.5).
– Twice as good with four times the effort (Exercise 9.4).
– Error propagation (Exercises 9.7, 9.8), Ψ Calculate with variances, report the
C-spec error!
– Ψ From variance to confidence with Student’s t value (Exercise 9.8).
– Reducing measurement uncertainty by combining measurement series (Exercise
9.6).
– Worse makes good even better.
– Linear regression, trend lines, coefficients with uncertainty (Sect. 10.2, 10.3 and
10.4).
– Applying non-linear regression using solver in excel and curve_fit in
Python (Exercises 10.5, 10.6 and 10.7).
– Decoding textual logs of measuring instruments and writing the relevant data into
tables using VBA macros or Python programs (Exercise 4.8).
Data Structures, Excel and Python
Basics 2
This chapter aims to develop computational solutions for physics problems,

parallel in excel and Python based on the same data structure and the
same type of list processing. In this way, excel may serve as a low-threshold
entry into scientific computation with a smooth transition to professional
platforms. We proceed in three steps: (1) Basic spreadsheet techniques are
introduced: absolute and relative addressing and naming of cells, creating
charts, and using sliders (scroll bars), with the didactic goal of addressing
variables in functions by their names, just as in mathematical formulas. (2)
Python basics are explained, with emphasis on the manipulation of Numpy
arrays essential for scientific computation. (3) Matrix operations are intro-
duced in excel, equivalent to those in Python. Finally, in one exercise,
a set of four parabolas and, in another exercise, a group of four cosines
(to simulate acoustic phenomena) are calculated and displayed in parallel in
both applications.
2.1 Introduction: Named Ranges in Excel, Arrays in Numpy
Solutions of Exercises 2.3 (Excel), 2.4 (Python), 2.5 (Excel), and 2.6 (Python) can
be found at the internet address: go.sn.pub/9Rtzxi.
Spreadsheet technology
This chapter is about how cells are addressed, figures are created and formatted,
and sliders are used to change cell contents. This will be easy for you if you are
already familiar with excel and know how to write formulas into cells. If you are
less experienced, you will first have to go through the basic exercise step by step. If
necessary, consult the excel help guide, and, finally, after having gone through the

16 2 Data Structures, Excel and Python Basics
basic exercise, find a textbook about excel techniques that is best suited for your
learning style.
Required and practiced excel techniques are:
– relative and absolute cell addressing,

– direct and indirect cell addressing,
– naming of cells and cell ranges,
– creation of diagrams,
– application of sliders.
We first exercise different types of cell addressing. Our goal, however, is to write
formulas as mathematically as possible, i.e., with letters representing variables.
Then, they will be identical to formulas in Python. For this purpose, individual
cells, ranges of cells within rows (row vectors) or columns (column vectors), and
two-dimensional cell ranges (matrices) are to be designated by names. All of these
techniques will be introduced step by step in the individual exercises and summa-
rized again in Sect. 2.3.5. The systematic use of vectors and matrices is the reason
why spreadsheet calculations can be translated nearly literally into Python.
Figures representing spreadsheets are characterized by the supplement (S), e.g.,
Fig. 2.2 (S).
Formulas in spreadsheets
Formulas in spreadsheets are reported in italic and often in oblique orientation, valid
for a cell in the neighborhood in bold font. They will be written in the text in brackets;
e.g., for the content of cell A11, we write A11 = [=A10 + dx]. We have to distinguish
whether or not an equality sign is written in the cell. For the expressions A9 = [x]
and A10 = [3], no equal sign is written in the cell; [x] is thus interpreted as text and
[3] as a number.
Python constructs
We will learn Python programming by working with the program cell structure
of the Jupyter notebook, first dealing with list processing and then focusing on
operations on arrays in Numpy. The explanations are less technically detailed than
for excel, because list processing is the core business of Python and has been
well described in numerous textbooks and online courses. However, our examples
are designed so that the essence of the definitions and procedures should become
obvious.
The Python constructs for list generation (np.linspace and np.arange
in the numpy library) will be used to define vectors that are later transformed with
standard functions into other vectors using list comprehension and broadcasting. To
mimic the column vectors in spreadsheets, a two-dimensional list with only one row
has to be introduced and transposed.
2.1 Introduction: Named Ranges in Excel, Arrays in Numpy 17
Matrix operations
Having provided excel with the necessary matrix formulas, we can finally demon-
strate broadcasting for algebraic operations and some operations of linear algebra
parallel in excel and Python (Exercise 2.5).
Applications
We practice our newly acquired knowledge in an exercise on four parabolas and
their upper envelope. The chapter concludes with a physically meaningful exercise
treating the sum of four cosines so as to demonstrate overtones, beats, and the addition
theorem of cosines.
2.2 Characteristics of a Parabola
Starting from its vertex form, we set up a data structure to tabulate and plot
a parabola, together with its characteristic features focus and directrix. The
data structure set up here is to be used in Exercise 2.3 for a single parabola
and in Exercise 2.6 for a set of 4 parabolas.
2.2.1 Different Definitions of a Parabola
Parabola from vertex

A parabola is to be presented in a diagram. Its standard form is defined as
y = a + b · x + c · x2 (2.1)
It is, however, more intuitive to start from its vertex form
y = yV + c · (x − x V )2 (2.2)
because its shape is immediately clear: its vertex is at (x V , yV ), and its curvature is
proportional to c (positive or negative). Transforming into the standard form yields
a = yV + c · x V2 and b = −2 · c ∗ x V (2.3)
In Fig. 2.1a, a parabola is shown with its vertex marked with a diamond.
Question
What are the coordinates (x V , yV ) of the vertex and the value of c in Fig. 2.1a?1
10 10
y y
8 8
6 6
4 4
directrix
vertex
6.2+-0.4x+-0.1x²
2 2 4.572+-1.56x+-0.3x²
focus
rays
0 0
-8 -4 0 x 4 8 -8 -4 0 x 4 8
Fig. 2.1 a (left) A parabola with its maximum at (−2.6, 6.6). b (right) Connecting the parabola
with its focus and its directrix; compare Fig. 2.10
Focus and directrix

A parabola may also be defined as the locus with equal distance to the focus (x F , yF )
and to the directrix (y = yD ), both of which are depicted in Fig. 2.1b. The coordinates
of the focus and the directrix are:
c c
x F = xV y F = yV + y D = yV − (2.4)
4 4
Ignoring reflection, all rays incident parallel to the symmetry axis of the parabola
have the same length up to the directrix, e.g., when calculated from the line y = 0.
By definition, they all have the same length up to the focus when reflected at the
parabola. This is why light incident parallel to the optical axis is focused in the focus
of a parabolic mirror.
Calculation points of a curve

Equations 2.1 and 2.2 represent continuous functions. In numerical calculations,
however, the function values y are calculated on only a finite number of discrete
x-values, x i (see the inset picture in Fig. 2.1a). The points (x i , yi ) are called the
calculation points of the function. They are the vertices of the polyline representing
the graph. In most of our exercises, the adjacent calculation points are equidistant on
the x axis , i.e., they all have the same distance to their respective neighbors, usually
specified in a parameter dx.
1 x V ≈ −2, yV ≈ 6.5 from visual inspection in the coordinate system; c = −0.1 from the legend
in the figure.
2.2 Characteristics of a Parabola 19
2.2.2 Data Structure and Nomenclature

(x V , yV ) coordinates of the vertex of the parabola
a, b, c coefficients of the parabola in standard form
yD ordinate value of the directrix of the parabola
(x F , yF ) focus of the parabola
dx distance between adjacent x values of calculation points
x sequence of x values
yP values of the polynomial for x
yA , yB, yC, yD values of the same polynomial but calculated in Excel with
column names
yMax upper envelope of yA , yB , yC , yD .
2.3 Basic Exercise in Spreadsheet Calculation
With the example of tabulating a parabola and displaying it in a diagram, we

train the basic spreadsheet techniques: absolute and relative cell addressing,
providing cells with names, connecting text and numeric variables to get
informative legends for diagrams, applying scroll bars to change cell contents
without typing numbers. After performing these exercises, the reader should
be able to select a more detailed textbook on excel techniques that is best
suited to her/his taste and needs. At the end of the exercise, an analogous
Python program with the same data structure is presented.
2.3.1 Cell Addressing
Spreadsheet layout
A spreadsheet layout for generating a parabola yP = f (x) is displayed in Fig. 2.2 (S).
We are going to review three regions successively: A1:B6 (to explain cell addressing),
D1:H3 (to name cells and apply sliders), and A8:E169 (to name cell ranges). All
relevant cells and the ranges x, yP , and yA get names from the beginning. The formulas
in cells are printed in neighboring cells in italic to keep track of the calculations.
Relative and absolute cell addressing

The coefficients of the vertex form of the parabola, x V , yV , and c, are defined in B1:B3
and recalculated in B4:B5 into b and a of the standard form, with the formulas in
C4:C5. The value in B6 specifies the horizontal distance dx between the calculation
points.
Fig. 2.2 (S) Tabulation of a parabola with the parameters c, b, a in B3:B5; F1:H3, sliders to deter-
mine the parameters x V , yV , c; E8:E169, alternative tabulation of the parabola using the named
cells a, b, c_ and the column range x
The independent variables x (left of ) are in A9:A169. They are obtained by

entering the initial value, here −8.0, into the first cell A9 of the range and the
formula A10 = A9 + $B$6 or A10 = [=A9 + dx] into the next cell and copying this
formula into the whole range. Copying is done by seizing the handle of A9 (see the
small black square at the lower right edge in the subpicture in column C) with the
pointer and dragging it down to A169.
Task Change the contents of cell B6, named “dx”! All x values in cells
A10:A169 should adjust themselves immediately.
The values for yP are obtained by entering the formula reported in B7 into B10
and dragging it up to B9 and down to B169. The formula is most conveniently
obtained by first entering [=], and then by clicking on the corresponding cells and
continuing with the operators * for multiplication and ˆ for potentiation, resulting
in B10 = [=B5 + B4*A10 + B3*A10ˆ2]. In the last term, the variable taken from
A10 must be squared. This is done with the power operator ˆ. You have to press
the button with the ˆ-sign and then the desired power, “2” in our case. Only after
the second step does the operator [ˆ2] appear in the cell.
When this formula is copied into another cell, the cell addresses change accord-
ingly. Copying into C11 would yield C11 = [=C6 + C5*B11-C4*B11ˆ2), realizing
relative addressing but not giving the desired result, because we would like to keep
the cells with the coefficients constant. This is achieved by making the references
to B3:B5 absolute, with dollars as prefixes: $B$5, $B$4, $B$3, either by introduc-
ing the $ sign explicitly before the column letter or the row number or by pressing
the function key F4, resulting in the formula reported in B7.
2.3 Basic Exercise in Spreadsheet Calculation 21
Having now copied this into C11, we would get C11 = [=$B$5 + $B$4*B10-
$B$3*B10ˆ2], with the values for x copied incorrectly (A10 becomes B10),
because they are still relatively addressed. Making the address of column A abso-
lute is achieved with $A10, B10 = [=$B$5 + $B$4*$A10-$B$3*$A10ˆ2]. If we
now copy this formula into another cell, only the row number 10 changes, e.g.,
into 12 when copied into any column in row 12.
Making a cell reference absolute can also be achieved by pressing the function
key F4 several times. Key words for the excel help: Absolute, relative, and mixed
cell references.
There is a more elegant way to copy a formula down a column: clicking onto
the “fill handle” (the bottom right corner) of the cell that contains the formula.
Then, the cell contents are immediately continued down to the 169th row, i.e., for
all cells for which there is an entry in the neighboring column, here column A,
until the first empty neighbor cell is encountered.
Questions
Questions concerning Fig. 2.2 (S):

From A9 to A10, x increases by dx = 0.1. Why is the increase 15.7 for the
next jump from −7.9 to + 8.0?2
Where is dx defined?3
Why is the name c_ in E7 provided with an underscore?4
Having gotten this far, we have programmed our first function. Changing the values
of x V , yV , and c in B1, B2, B3, the function values in column B adapt imme-
diately. We may now proceed to the section “Graphical representation” to see the
resulting curve, but should come back to learn about naming cells.
Naming cells
In the range B1:B6 of Fig. 2.2 (S), parameters are defined that are accessed in various
parts of the worksheet. To call them like variables in mathematical equations, we
provide the cells with the names written to the left of them. This is done by activating
the range A1:B6 comprising names and values and clicking through (excel 2019):
formulas/defined names/create from selection.
A prompt appears, “create names from values in the left column?”.
The answer is yes, that the agent has correctly detected, and we confirm this by
clicking ok. For more about the name manager, see Sect. 2.3.5. We can now refer
to these cells by their names, anywhere in the spreadsheet and, indeed, throughout
the whole book.
When writing, e.g., [=a] into a cell somewhere in the spreadsheet and pressing
enter, this cell immediately gets the numerical value corresponding to a, in our
2 Rows 11 to 168 are hidden. The jump is over 159 advances of dx; -7.9 + 15.9 = 8.0.
3 The value for dx is set in cell B6, which is given the name in A6.
4 The name c is protected for excel-internal use. A name c in a cell intended to become an
identifier is automatically extended by an underscore.

example, 6.2. When the content of cell B5 is changed, the value of all cells with [=a]
changes as well.
We can even name cell ranges, e.g., A9:A169 with the name x and B9:B169 with
yP. When activating A8:B169 and proceeding as above, the Name Manager prompts
us: "create names from values in the top row?”, and we confirm this
by clicking OK.
We may now write = a + b*x + c_*xˆ2 into E9, more elegant and clearer than
the formula in B7, with absolute and relative cell addressing, and copy down to D
169 to get the same values as for yP . For x, the formula in a cell takes the value in
the same row in column A. The name manager has changed c into c because the
letter c conflicts with a protected name in excel.
2.3.2 Graphical Representation of a Function
After having created the function table of a parabola so beautifully, we would like
to visualize the curve. To do so, we set the cursor into an empty cell, away from
the filled cells, and click in the insert tab on insert/ charts/, and on scatter
within the charts section. A blank chart is inserted.
Upon our activation of design/select data (see Fig. 2.5b), a select data
source window opens. We click add, and a window as shown in Fig. 2.3 opens.
For series name, we click on cell B8 of Fig. 2.2, and for series x values, we
activate A9:A169 and hit return. For series y values, we activate B9:B169 and
hit return. The empty chart changes to that shown in Fig. 2.4.
As the legend “yP” has been taken from a cell in the spreadsheet, it will adapt
immediately when the cell entry is changed. If a legend in a chart is identical to
that in the spreadsheet indicating the data, it helps in keeping the overview.
Spreadsheet
symbol
Fig. 2.3 Insertion of a data series into a chart; the series name is best taken from the worksheet
by activating the relevant range, not by entering it as text
yP
08
06
04
02 yP
00
-10 -05 00 05 10
-02
-04
Fig. 2.4 Scatter plot after completing Fig. 2.3
Fig. 2.5 Tabs, which are important for diagrams, after insert/charts/scatter/with only
markers in the Start menu (excel 2019, excel 2010 similar) or after activating an existing chart,
a (left) format/current selection, to the left of the start bar, to format an element of the dia-
gram, b (center left) design tab / data group /select data to select data to be entered into
the diagram, c (center right) design tab /chart styles, d (right) format/size, appears after
activating a diagram, to the right of the start bar, to specify the size of the diagram
Formatting the chart

We reshape the diagram according to our taste, e.g., as in Fig. 2.1a. After clicking
on the diagram and home/format, the following components change:
– Size (7 cm high, 8 cm wide) (in the start bar to the far right, see Fig. 2.5d).
Also, after clicking on the relevant element of the diagram or selecting it from
the leftmost register in the format tab (Fig. 2.5a, at first, only chart area
appears, but after opening the list by clicking on ▼, all items of the diagram
appear), we choose before clicking format selection:
– Format/Chart Area/Border /No line

– Format/Plot Area/Border Color/Solid line/Color/Black and Width/1pt
– Format/Data Series/Marker Options / Built in/Diamond, Size 4 and Marker

Fill/Solid Fill/ Black
– Format/Horizontal (Value) Axis/Axis Options/Minimum /-8, /Maximum /8,
/Major Unit/4.
Task Change the value of the parameters x V , yV , c in Fig. 2.2 (S) and observe
how the spreadsheet entries and the chart change!
Alac That’s cool! The diagram is alive!
Tim Once created, it’s always up-to-date!
2.3.3 Smart Legends in Figures
Smart legends are created by concatenating variables and text. In Fig. 2.4, we
have specified the parabola with its name in the spreadsheet, in simple text. In
Fig. 2.1a, however, the legend contains the actual parabola equation generated in
D6 of Fig. 2.2 (S) with “Text” &variable.
= a&” + ”&b&”x + ”&c_&“x” in D6 yielding 6.2 + −0.4x + −0.1x
If text is to be inserted, it must be enclosed in quotation marks, e.g., as above, “x

+ “. The concatenation operator is “&”. When the parameters a, b, or c_ change,
the text in D6 and the legend in the chart will immediately follow.
excel Ψ “Text“& Variable concatenation with &
Python Ψ „Text “ + Str(Variable) concatenation with +
Often, float values have to be rounded. When, e.g., B3 = [= 1/3], then only 0.33
is displayed in the cell. However, if B3 is inserted into a legend, 0.333333333
appears. Setting round(B3, 2), 0.33 is returned: [= “c_ = “&round(B3; 2)]
results in the cell content [c_ = 0.33].
2.3.4 Scroll Bars
Cell contents are changed with a slider (a scroll bar)

We can play even more impressively with the curves when we change their parameters
with sliders. To introduce sliders, we select developer/controls/insert/activex
controls to get the tabs in Fig. 2.6a.
If the tab does not appear in your toolbar, you have to supplement the toolbar by
making a tick in file\options\customize ribbon in the box before developer.
We need the slider to be an activex control element, and we click on the
icon for the slider in the top line on the far right, with the design mode turned on,
Fig. 2.6 (excel 2019, excel 2010 similar) a (left) Tab after going through devel-
oper/controls/insert/activex controls. A slider is listed in the upper row on the right,
at the foot of the arrow. b (center) In J1:L1, a slider has been installed by clicking on the control
element and pulling it up in said spreadsheet area. c (right) Menu for defining the properties
of the slider (scrollbar). It appears after we click on properties in a; important parameters:
linkedcell, minimum and maximum values
and then pull up a rectangle with the mouse at the desired place in the spreadsheet.
In Fig. 2.6b, this is done in cells J1:L1. Now, with the design mode still on, we
can configure the slider. In Fig. 2.6c, the properties list of the activated scrollbar
is shown. We specify that it is cell I1 (linkedcell) into which the number is to be
written, and that the number should be between 0 (min) and 100 (max). We then
turn the design mode off by clicking on this icon again (see Fig. 2.6a) and move
the slider’s thumb with the mouse. Immediately, a number appears in I1.
design mode is activated and deactivated by clicking on the icon. When
activated, existing control elements can be modified, or new ones added. When
deactivated, the control elements can be operated.
When we grab the thumb (the rectangular bar in the middle of the slider)
with the cursor and move it, the output in the linked cell changes. SmallChange
specifies the jumps (to the left or the right) of the numbers when we click on the
(left or right ) edge of the slider. LargeChange sets the jumps’ size when
⍓
we click within the slider bar left or right of the thumb. Try it out!
Conversely, if we change the contents of a linkedcell, the new value is entered
into the slider’s memory, and its thumb will move.
We can use a slider (scroll bar) to enter integers between 0 and 32,767 (=
215 − 1) into a cell (LinkedCell). Settings are specified in the properties group:
For linkedcell, specify the address of the cell to be written into. min and max
can limit the value range.
Sliders in Fig. 2.2 (S)

In Fig. 2.2 (S), three sliders are introduced with linked cells E1:E3, wherefrom the
parameters a, b, and dx in B1:B3 are calculated. When one of the sliders is operated,
the cells’ value and the curve in Fig. 2.1a adapt immediately.
A slider can produce only positive integers. If other numbers are needed, plug the
value of the linkedcell into a formula in another cell. In Fig. 2.2 (S), the value of
the linkedcell (E3), set between 0 and 100, is changed with B3 = [=(E3−50)/10]
to the range −5 to 5, with intervals of 0.1.
Questions
Which numbers can appear in I1 (range and minimum distance to each other),
according to the information in Fig. 2.6c?5
What is the range of the numbers for x V in Fig. 2.2 (S), provided that Min
= 0 and Max = 80 for the slider in F1:H1. What is the minimum step size?6
Connecting a VBA macro to a control element

The following will be important in later chapters: We can connect a VBA routine to the
slider (an example is given in Exercise 10.5). A routine sub scrollbar1_change(
) … end sub is executed each time we move the scrollbar1 slider. To enter a
code, click on view code (Fig. 2.6b) and select the considered slider’s name in the
left-hand drop-down window that has popped up, then click change (or some other
action to trigger the routine) in the right-hand drop-down window.
2.3.5 Summary: Cell References and Name Manager
The dollar makes it absolute.
Cell references, relative, absolute, indirect

The following formulas in a cell, C3 = [=A4], C3 = [=$A4], C3 = [=A$4], C3 =
[=$A$4], yield the same result: the value of cell A4 is written into the active cell C3.
However, if these formulas are copied into another cell, they change. For example, in
D4, there will be: [=B5], [=$A5], [=B$4], [=$A$4]. A $ sign before a column label
or a row number has the effect that the label or number are held fixed when copied.
A dollar makes a cell reference absolute. If the $ is missing, the label is incremented
5 MIN = 0; MAX = 100; integers between 0 and 100 can appear in I1, distance = 1.
6 B1 = [=(E1-40)/5] (see D1), range -8 (for E1 = 0) to + 8 (for E1 = 80) in steps of 1/5 = 0.2.
by the column or line spacing between the old and new cells; this is a relative cell
reference.
The spreadsheet function indirect(cell) expects a cell address as an argument.
It writes the contents of the cell with this address into the current cell. For example,
with A4 = [=indirect(A5)] ; A5 = [X7]; X7 = [3.4] the value in A4 will be 3.4.
Assigning names to cell ranges

[=a*x + b] instead of [=B$2*$A6 + B$3]
Individual cells, ranges in a column (“column vectors”) or a row (“row vec-
tors”), and rectangular ranges (“matrices”) can be named and then inserted with
their names into formulas and worksheet functions as arguments. Naming is done
with the name manager by going through formulas/name manager. For details
on the name manager, see below. Make extensive use of this feature! Doing so, you
can write formulas and functions in mathematical language, e.g., [= A*sin(k*x)]
instead of [=A$1*sin($A5*B$2)]! Such a formula is valid for a set of sine functions
whose amplitudes A and wave number vectors k are stored in two row vectors with
names “A” and “k” and where the independent variable x is stored in a column vector
with the name “x”.
Name Manager of Excel

As an example to demonstrate the properties of the name nanager, we calculate
the electric field E x in x-direction of a point charge at (x 1 , 0) in the xy-plane:
x − x1
Ex = 3
(2.5)
(x − x1 )2 + y 2
The definition of variables and constants and the calculation are distributed over
two sheets, “Dist” and “E.x”. The values of r = (x − x1 )2 + y 2 , representing the
planar distance to point (x 1 , 0) for x and y from -2 to 2, are calculated in Fig. 2.7a
(S) (Sheet “Dist”).
The x values are in B3 to F3, the y values in A4 to A8. The value for x 1 is specified
in B1. The calculation can be performed with mixed cell references, as in cell B5:

B5 = = SQRT (B$3 − Dist!$B$1)∧ 2 + A$5∧ 2 (2.6)
The formula is, however, more intuitive when cell names are used, as in D6,
displayed in D2:

D6 = = SQRT (x − x.1)∧ 2 + y∧ 2 (2.7)
To achieve this, we have to designate, e.g., range B3:F3 with the name x,
already present in cell G3. We activate B3:G3 and follow the menu formu-
las/defined names/create from selection. To “activate” the range means
Fig. 2.7 (S) a (left) Sheet “Dist”; the values in the matrix B4:F8 are calculated with the values in
the horizontal vector x = B3:F3 and the vertical vector y = A4:A8, together with the value of x 1 in
cell $B$1, using mixed cell references in B5 and variable names in D6. b (right) After activating
B3:G3 in a, the dialogue box of formulas/ defined names suggests providing B3:F3 with the
name in G3 ( right column)
that the cells are marked with the left mouse button pressed. A window like the
one shown in Fig. 2.7b (S) pops up. The assistant has already recognized a poten-
tial name in the immediate neighborhood of the activated range, namely, in the
right-most column of the activated range. This name corresponds to our intention,
and we click OK.
The matrix range B4:F8 is named by activating it and selecting formu-
las/defined names/define name. A window marked new name pops up with
the refers to field already filled in, because a cell range was activated before the
selection. We have to fill in the name field, in our case, with “r.0”.
The formulas/defined names/name manager window, displayed in
Fig. 2.8 (S), gives us an overview of all of the named ranges.
Fig. 2.8 (S) The name manager lists all names and ranges (“refers to”) and the scopes for which
they are valid
A B C D E F G I J K L M N O
1 Sheet E.x 14
0^3 0^3} rmula
.1)/r. .1)/r. trix fo
2 =(x-x -x a
15 {=(x as m
3 ### ### ### ### ### #VALUE!
16 -0.1 -0.1 0.0 0.1 0.1 #N/A #N/A
4 -0.1 -0.1 0.0 0.1 0.1 #VALUE!
17 -0.2 -0.3 -0.1 0.4 0.2 #N/A #N/A
5 -0.2 -0.3 -0.1 0.4 0.2 #VALUE!
18 -0.2 -0.8 -100.0 1.2 0.3 #N/A #N/A
6 -0.2 -0.8 -100 1.2 0.3 #VALUE!
19 -0.2 -0.3 -0.1 0.4 0.2 #N/A #N/A
7 -0.2 -0.3 -0.1 0.4 0.2 #VALUE!
20 -0.1 -0.1 0.0 0.1 0.1 #N/A #N/A
8 -0.1 -0.1 0.0 0.1 0.1 #VALUE!
21 #N/A #N/A #N/A #N/A #N/A #N/A #N/A
9
22
Fig. 2.9 a (left, S) Sheet “E.x”. The x component of the electric field of a point charge at (x 1 ,
0) is calculated. In row 3 and column G, the index is outside of the permitted range. b (right, S)
Alternative calculation in an arbitrary position with a matrix formula
The names are valid in the entire workbook when they are first defined. The
names “x” and “y” appear twice. When they were first defined in the “Dist” spread-
sheet, they were valid throughout the workbook, as shown in the scope column.
When they are defined a second time in another sheet, here, “MMult”, their scope
is limited to this spreadsheet, and the previous definitions of x and y do not apply
here, although they still do in the rest of the workbook.
Calculating an electric field

In sheet “E.x” [Fig. 2.9 (S)], the x-component E x of the electric field is calculated:
x − x1
E x (x, y) =
r03
Tim Which definitions for x, x 1 , and r 0 are valid in sheet “E.x”?
Mag You can find this out by trial and error or by using filter in the name
manager.7
Referring to names in a cell

The ranges (vectors x and y, matrix r 0 , constant x 1 ) defined in the two sheets can be
called in each sheet cell-wise, however, they cannot be so within the complete range
of the sheet, but only within a range that matches the range in the sheet in which the
name is defined, for example, the matrix r 0 only within the range B4:F8. Outside of
this range, errors are reported (see Fig. 2.9a). The horizontal vector x may, for the
current definition of its coordinates, only be called in columns B to F, the vertical
vector y only in rows 4 to 8. The constant x 1 designates only a single cell and may
be called in the whole file without restriction.
7The named ranges in sheet “Dist” are valid throughout the whole workbook, except for “MMult”,
so they are also so in “E.x”.
Referring to names in a matrix function

The matrix r 0 can be called within any range of a spreadsheet when entering the
formulas as a matrix function, e.g., in I16:M20 in Fig. 2.9b (S). To do so, activate
the desired range, write the desired formula into the formula bar, in our case [=(x-
x.1)/r.0ˆ3], and finish with the “magic chord"! The matrix formula is enclosed in
curly brackets: {=(x − x.1)/r.0ˆ3}.
Magic chord for completing matrix functions: ctrl + shift +

enter.
2.3.6 What Have We Learned so Far, and How to Proceed

Further?
Alac That’s all really super easy. Worksheet calculation seems to be a children’s
game. That wasn’t clear to me until now.
Tim Well. Might we have acquired knowledge in only a narrow section for
particular tasks?
Mag We have traveled quickly across a wide area on a narrow path. This is
actually a fast track to success, at least for the tasks we intend to tackle.
Tim Is it not better to learn thoroughly so that one does not become lost when
the tasks are set a little differently?
Alac You can always tackle modified tasks through trial and error.
Mag Yes, trial and error is a possibility. You should do this anyway with all of
the programming constructs with which you are not already familiar. Nevertheless,
you should also go to a bookstore or into the internet and find books that instruct
you in excel. Browse along the learning path you have just gotten to know. It will
not take long for you to figure out which of the books explains the computational
procedures in a way that you can understand. You should buy that book!
The same advice holds for Python, which you will get to know in the next
section.
2.3.7 Python Program
Table 2.1 presents a Python program corresponding to Fig. 2.2 (S) for complete-
ness. The reader may study it after having gone through Exercise 2.4.
The program consists of a list of simple assignments of type x V = -2, with the
name x V being called the identifier, or simple formulas assigned to an identifier,
Table 2.1 Specifications corresponding to Fig. 2.2 (S)

1 xV=-2 #Coordinates of vertex
2 yV=6.6
3 c=-0.1 #Curvature
4 b=-2*c*xV #Coefficients of standard form
5 a=yV+c*xV**2
6 dx=0.1 #Horizontal distance between x
7 x=np.arange(-8,8+dx,dx) #Array of x values
8 yP=a+b*x+c*x**2 #Array of y values
9 print('a ={:5.2f}'.format(a),'; b ={:5.2f}'.format(b))
a = 6.20 ; b =-0.40
Table 2.2 Calculating the coordinates (for Fig. 2.1b) of the focus and the directrix
10 xF=xV #Coordinates of the focus
11 yF=yV+1/4/c
12 yD=yV-1/4/c #y Coordinate of directrix
e.g., b = −2*c*x V recurring to variables specified earlier. The calculated values

of a and b (printed into the second cell) are the same as in Fig. 2.2 (S).
The term ‘a = {:5.2f}’ indicates that the printout starts with the text string ‘a
= ’ and that the value of the variable a in format(a) is to be printed right aligned
(:) in float format (f) with length 5 and 2 decimal places (5.2).
The coordinates of the focus and the directrix are calculated in Table 2.2 with
the same formulas as in Fig. 2.2 (S). A plot corresponding to Fig. 2.1b is produced
with the standard program FigStd explained in Sect. 2.4.5.
2.4 Python and NumPy Basics
We will learn how to work with program cells in the Jupyter notebook,
and get acquainted with arrays in numpy. For more detailed information
about Python specifics, the reader is referred to Stewart, J. (2014). Python
for Scientists. Cambridge: Cambridge University Press. https://doi.org/10.
1017/CBO9781107447875, Chap. 3 (A Short Python Tutorial), Chap. 4
(NumPy), and Chap. 5 (Two-Dimensional Graphics) or other textbooks
treating the same subjects.
2.4.1 Basic Exercise
Table 2.3 displays a simple Python program developed in the Jupyter

notebook.
Table 2.3 Cell structure of Python in the Jupyter notebook
We can start running the program with any cell. However, if referring to vari-
ables or functions, we have to define them ahead of time. Starting Cell 1 before
Cell 2, or Cell 3 before Cell 4, leads to a #NameError.
We are now running the program cell by cell in sequence.
Cell 1: The print statement tries to get yA from the memory. As yA is not yet
defined, a #NameError message is returned.
2.4 Python and NumPy Basics 33
Cell 2: Three lists, characterized by square brackets, are defined, and their con-
tents are printed out. When we now run Cell 1 again, no error message appears,
but the contents of yA are printed out.
Cell 3: The statement tries to assign a value to variable z by calling a function
add2 that is not yet defined, so that a name error occurs.
Cell 4: The function add2 is defined. Now, running Cell 3 does not result in an
error message.
Cell 5: Function add2 is applied to variables x and y, and the result is printed
out. When applied to lists, the formula 2*a + b is not an algebraic function; list x
is concatenated twice and then list y once to yield the list [x, x, y].
Cell 6: The numpy library is imported under the abbreviation np. Two numpy
arrays are created from the lists x and y. The function add2 is applied to x np
and ynp , yielding an error message. When applied to arrays, it is the algebraic
operations addition and multiplication that are performed element-wise. As the
two arrays have different size, this does not work, and a #ValueError message
results: “Operands could not be broadcast together with shapes (4,) and (3,)”.
Cell 7: List y is extended with a new element to have the same length as xnp.
The function add2 can now be applied, with the result now being an array of the
same length and a linear combination of the two initial arrays. The formula 2*a
+ b is now interpreted as an algebraic operation and performed element-wise on
the arrays.
Copy
Parts of the memory may get various names, e.g., the statement AddTwo = add2
assigns the additional identifier AddTwo to the function add2 defined in Cell 4 of
Table 2.3, so that this function may also be called by AddTwo. Such assignments
are different from making a copy, as is demonstrated in Table 2.4. The statement YA
= Y[3:5] produces only a name (an identifier) for part of list Y, whereas with YC
= np.copy(Y[3:5]), a new object with its own memory space is created. Y C
is an object of its own and is not affected by subsequent changes of Y.
Table 2.4 Identifier of a subarray vs. copying a subarray into a new object; only one print state-
ment is reported in the first cell; the other print statements are similar. The results of all print
statements are reported in the second and the fourth cells
1 Y=np.array([0,1,2,3,4,5,6,7,8,9])
2 YA=Y[3:5]
3 YC=np.copy(Y[3:5])
4 print("Y ", Y)
5 Y [0 1 2 3 4 5 6 7 8 9]
6 YA [3 4]
7 YC [3 4]
8
9 Y[3]=8 #Another program line
Y [0 1 2 8 4 5 6 7 8 9]
YA [8 4]
YC [3 4] #YC is independent of Y
2.4.2 Data Structures
Data types
The following data types are available in Python:
– Int (integer, unlimited size),

– float (8 Bytes),
– bool (boolean),
– string (text),
– complex (only in numpy).
Lists
Lists contain one or more items. They:
– may contain items of different type: z = [1, 3.14, ‘abc’. 2 + 4],
– are mutable, e.g., can be changed by append or delete;

x.append(‘new’) → [x, ‘new’],
– can be concatenated by +: [x] + [y] → [x, y],

– can be concatenated by *: 3*[x] → [x, x, x],
– can be sliced, i.e., subarrays can be addressed with new identifiers: x2 =
[3:-1], from the fourth (indices 0,1,2,3,...) to the
penultimate element of x.
Append, add, and multiply result in longer lists. The new identifier x2 points to a
sublist of x from the 3rd to the penultimate (−1) element. The method append, as
well as the operators + and *, are demonstrated in Table 2.3 (def add2). Lists can
be multidimensional, but in our exercises, only one- and two-dimensional lists are
used.
Sets {}
Sets contain unordered collections of unique elements to which standard mathe-
matical set operations can be applied. These are intersection, union, difference, and
symmetric difference. Examples are given in Exercise 8.4. Sets do not record ele-
ment position and, as a consequence, do not support indexing, slicing, or other
sequence-like behavior.
Dictionaries {}
Dictionaries comprise pairs of an identifier (data type string) and an object (arbitrary
data type). The objects are addressed by their identifiers:
We do not use dictionaries in this book.

2.4.3 Python Libraries
Python libraries are collections of pre-compiled functions. They are open-source,

supported by a community of programmers, who are always there to answer
questions, e.g., on stackoverflow.com. We shall sometimes directly refer to such
advice.
Numpy
Numpy is a fast and efficient array-processing package designed for numerical com-
puting that provides functionalities comparable to MATLAB. We are using it as our
standard. It is usually imported under the name np: import numpy as np, see
Cell 6 in Table 2.3.
Numpy.random
Random number routines produce pseudo-random numbers. We import
Numpy.random as npr and make use of two functions:
– npr.rand(N) generates an array of N random numbers between 0 and 1,

(Exercises 8.2 and 8.3).
– npr.randn(N) generates an array of N normally distributed random num-
bers (Exercise 8.5).
Matplotlib
Matplotlib is a package for designing a variety of charts or even arrays of charts. We
import matplotlib.pyplot as plt. We restrict ourselves to producing
simple scatterplots with a user-defined function (Sect. 2.4.5) that is able to display
nearly all results of our exercises, similar to excel charts. In order to plot arrows,
we introduce a second user-defined function (Sect. 3.2.7).
Scipy
Scipy is designed for scientific computing and is especially suited for machine
learning. We need only certain functions for optimization [minimize (Exer-
cise 10.6), fsolve (Exercise 10.5), curve_fit (Exercise 10.7)], linear algebra
[solve (Exercise 5.9)], and statistics (Chisquare from Scipy.stats).
Pandas
Pandas mimics spreadsheet calculation within Python, enabling input to and
output from excel files and text files. We use it only occasionally, e.g., in Exercise
4.8, to make the reader aware that such things exist and are useful.
Table 2.5 Data types in numpy demonstrated with arrays built with lists from Cell 2 of Table 2.3
1 import numpy as np
2 xInt = np.array(x,dtype=int)
3 yFloat = np.array(y,dtype=float)
4 yAfloat = np.array(yA,dtype=float)
5 xBool = np.array(x,dtype=bool)
6 xComplex = np.array(x,dtype=complex)
7 xStr = np.array(x,dtype=str)
xInt [ 0 1 2 3] shape (4,)
yFloat [ 0.00 4.00 5.00 -1.00] shape (4,)
yAfloat [ 0.00 4.00 5.00] shape (3,)
xBool [ 0 1 1 1]
xComplex [0.+0.j 1.+0.j 2.+0.j 3.+0.j]
xStr ['0' '1' '2' '3']
xInt*yFloat [ 0. 4. 10. -3.]
xInt*yAfloat ValueError: operands could not be broadcast
together with shapes (4,) (3,)
2.4.4 Numpy Constructions
Numpy is usually imported under the name np: import numpy as np.
Ndarrays
Ndarrays are, in general, n-dimensional arrays. One- and two-dimensional ones
have analogies in spreadsheets. For scientific computing, they have an advantage
over spreadsheets when the data becomes large.
They:
– are immutable, with size and datatype specified when the object is introduced,
– are operated element-wise in algebraic operations: x + y → [x + y],
– can be arguments in mathematical functions.
In Table 2.5, one-dimensional arrays of various data types are built with the func-
tion np.array(.) expecting a list as an argument, here taken from Table 2.3.
Two numerical arrays can be multiplied element-wise when they have the same
shape, e.g., x Int and yfloat , but not x Int and yAFloat .
Question
How do the entries of xBool in Table 2.5 arise?8
8 Numerical 0 becomes False, = 0 becomes True. False is output as 0, True as 1.

Ab-initio constructors
The following five functions construct an array from the specifications in the
argument list.
– np.ones(N) 1D-array of N ones

– np.zeros(N) 1D-array of N zeros
– np.linspace (start, stop, number of steps, endpoint =
True)
np.linspace(1.5, 3.5, 3) → [1.5 2.5 3.5], True is default.
np.linspace (1.5,3.5,2, endpoint = False) → [1.5 2.5].
– np.logspace
np.logspace(1, 3, 3, base = 10) → [ 10. 100. 1000.]
– np.arange (start, stop, step)

np.arange (0.0, 4.5, 1.5) → [0.0 1.5 3.0].
In np.arange, the stop value is not included. This is favorable when stacking
np.aranges together:
np.hstack([np.arange(1,4,1),np.arange(4,10,2)])
–
→ [1 2 3 4 6 8]
We can start the second np.arange with the stop of the first np.arange and
have the stop/start value only once.
The following two functions construct arrays of the same shape as Array in the
argument:
– np.ones_like(Array) 1D-array of ones

– np.zeros_like(Array) 1D-array of zeros.
From the numpy.random library, imported as npr, we shall use:
– npr.rand(N) generating an array of N random numbers between 0 and 1.

– npr.randn(N) generating an array of N standard normally distributed random
numbers.
Identifiers to elements of 2-dimensional arrays

In Table 2.6, we construct 1-dimensional arrays with np.linspace and stack
them together to form a 2-dimensional array consisting of two rows. Furthermore,
identifiers to single rows, elements, columns, and 2D subarrays are specified.
Table 2.6 Constructing a 2D array with two 1D arrays obtained with linspace; identifiers to
single rows, elements, columns, and 2D subarrays; only one print statement reported in Cell 1
1 X=np.linspace(0,3,4)
2 Y=np.linspace(10,13,4)
3 Lis=np.array([X,Y])
4 print(“np.shape(Lis) “, np.shape(Lis))
np.shape(Lis) (2, 4) #2 Rows with 4 entries each

Lis
[[ 0.00 1.00 2.00 3.00]
[ 10.00 11.00 12.00 13.00]]
Lis[0] [ 0.00 1.00 2.00 3.00]

Lis[0][2] 2.0
Lis[:,1] [ 1.00 11.00]

Lis[0][2:3] [ 2.00]
Lis[1][1:3] [ 11.00 12.00]
Lis[0:1][1:3]
[[ 1.00 2.00]
[ 11.00 12.00]]
Recommendation: Use this type of presentation, printing out the shape

and content of matrices, to get an overview of your data structure!
Lis[0][2] means: from the first row (index 0), take the third element (index 2).
Alternative interpretation: take the matrix element from row = 0, column = 2.
Lis[1] [1:3] means: from the second row (index 1), take the second (index 1)
to the third (index 2, index 3 exclusive) elements.
Lis[0][1:-1] means: from the first row (index 0), take the second (index 1) to
the penultimate (index -1) elements.
Lis[:,1] means: take the second column (index 1).
Information on arrays
– np.size(array) returns the total number of elements.

– np.shape(array)returns the shape, for 2-dimensional arrays in the form (r,c),
number r, c of rows and columns, respectively.
Stacking
– np.hstack ([list of arrays]) concatenates the 1D-arrays horizon-

tally into a long 1D-array, is defined more generally for multidimensional
arrays
– np.stack([list of 1D-arrays]) stacks the arrays (all of the same size
by necessity) in the list as rows into a 2D-array, is defined more generally for
multidimensional arrays.
Functions
– np.sin, np.cos, np.tan, np.atan2

– np.power
– np.dot.
The function np.dot(A, B) performs a matrix multiplication of the two

matrices A and B. The same can be achieved with the operator@: A@B. Matrix
multiplication is dealt with in Exercise 2.5 in detail.
Atan2 is arcus tangens calculating the angle from the x-axis given the point
(x, y) . The order of arguments is atan2(y;x), different from excel:
np.arctan2(y, x) (Python) = atan2(x; y) (excel)
np.arctan2(3, 2) = 0.980.98 = atan2(2; 3).
np.arctan2(2, 3) = 0.590.59 = atan2(3; 2).
np.arctan2(4, 1) = 1.331.33 = atan2(1; 4).
2.4.5 Standard Plot Program
We create charts by using the library matplotlib.pyplot imported under the

name plt. As we deal almost exclusively with scatter diagrams for functions of
the form y = f(x), we do not need the full versatility of matplotlib. To profit
from this restriction, we have devised a function to plot a standard figure, FigStd,
in Table 2.7, where all necessary style information is coded within the function
Table 2.7 User-defined function for creating a scatter plot; this cell is run in all programs at the
beginning
2 import matplotlib.pyplot as plt
3 np.set_printoptions(precision=2, threshold=10,edgeitems=3,
4 formatter={'float': '{: 7.2f}'.format})
5
6 def FigStd(xlabel, xmin, xmax, dx,
7 ylabel, ymin, ymax,dy,
8 xlength=4,ylength=4):
9 plt.figure(figsize=(xlength, ylength))
10 plt.axis([xmin, xmax, ymin, ymax])
11 plt.rcParams.update({'font.size': 10,
12 'font.style':'italic'})
13 plt.xlabel(xlabel)
14 plt.xticks(np.arange(xmin, xmax+dx, dx))
15 plt.ylabel(ylabel)
16 plt.yticks(np.arange(ymin, ymax+dy, dy))
17 plt.plot([xmin,xmax],[0,0],'k-',lw=1) #x Axis through 0
18 plt.plot([0,0],[ymin,ymax],'k-',lw=1)
Fig. 2.10 Parabola (the same

as in Fig. 2.1b) specified in
Tables 2.1 and 2.2 and plotted
with Table 2.8
Table 2.8 Applying our standard function for scatter plots

1 FigStd('x',-8,8,4,'y',0,10,2)
2 lblP="y=%4.2f+%4.2fx+%4.2fx²"%(a,b,c) #Label for parabola
3 plt.plot(x,yP,'k-',label=lblP) #’k-‘ Black full line
4 plt.plot(xV,yV,'kd') #’kd’ Black diamond
5 plt.plot([-8,8],[yD,yD],'k-',lw=0.5)
6 #’lw’ Line width
7 plt.plot(xF,yF,'kx')
8 for i in range(-7,-3,1):
9 plt.plot([i,i,xF],[yD,a+b*i+c*i**2,yF],
10 'k--',lw=0.5) #’k—-‘ Black dashed line
11 plt.legend()#Plots the labels for the curves within the fig
body, and the information concerning the x-axis and the y-axis has to be speci-
fied in positional arguments. The axes’ lengths in the figure have default values
(xlength = 4, ylength = 4) that can optionally be specified otherwise.
The plot in Fig. 2.10, corresponding to Fig. 2.1b, is produced with the program
in Table 2.8 (continuation of Table 2.1 and Table 2.2).
In the instruction [yD, a + b*i + c*i**2,yF], the values are cre-
ated implicitly within the list.
Extensions
In Sect. 3.4.7, we introduce two extensions of our standard figure: a secondary y-
axis, a logarithmic scaling of the y-axis with the statement plt.yscale(value
= ”log”).
Table 2.9 Formatted output with %

1 a1,a2,a3=1.2345,3.4567,-4.5678
2 label="y=%4.2f*x**%4.2f +%6.1f"%(a1,a2,a3)
3 print(label)
y=1.23*x**3.46 + -4.6
2.4.6 Formatted Output
The example in Table 2.9 shows how a formatted output can be achieved. There is
a format string with three text variables “y = $*x**$ + $”. The $ here is a proxy;
it starts with a “%” and is followed by a format, e.g., 4.2f for a floating number
4 characters long and to be displayed with 2 decimal places. Following the string
is a % sign and the name of the variables as a tuple, here %(a1, a2, a3) with the
three entries replacing the three % in the format string.
2.5 Matrix Calculations in Excel and Python
We define row and column vectors and 2-dimensional matrices, and explain
operations on them, parallel in excel and Python. We apply broadcasting
in excel and Python to adapt the shape of operands so as to fit with each
other in the intended operation. Finally, we get to know and apply the matrix
operations transposition and inversion of linear algebra.
Data structure
The data structure in scientific computing is based on arrays. Python, especially its
numpy library, is designed for vectorized code, so that the programmer is forced to
work with it from the first line of a code. This is its core business, described in all
introductory textbooks, so we do not need to explain it here in detail.
Spreadsheet software, in contrast, is primarily designed for business calculations.
In order to make it suitable for scientific computing, we have to create said vectorized
data structure. This is achieved by identifying cell ranges with names so that most of
the operations on vectors and matrices known from Python programs can also be
applied in spreadsheet calculations. This is the essence of our approach to making
spreadsheet calculations suitable for elementary scientific computing.
The treatment in this book is restricted to 1-dimensional arrays called vectors and
2-dimensional arrays called matrices. Vectors are of two types, row and column. We
designate the shape of vectors and matrices with a tuple (r, c), where r and c are the
number of rows and columns, respectively. In Python, a row vector is regarded as
a 1-dimensional array, e.g., of shape (2,), whereas a column vector is represented as

a 2-dimensional array with only one column, e.g., of shape (3, 1).
Nomenclature
We not only strive to use the same notation of variables in Python and excel, but
also to use a similar one in mathematical equations in the text. For this exercise, we
have chosen the following nomenclature:
U, V column vectors, shape (3, 1) 3 rows, 1 column

W, X row vectors, shape (2, ) 2 entries in the array
L, M matrices, shape (3, 2) 3 rows, 2 columns
N, O square matrices, shape (2, 2).
In the text, vectors are characterized by an underline, matrices by a double underline.

Additionally, we apply subscripts and underlines to make the variable names more
similar to the usual mathematical notation.
2.5.2 Operations on Arrays
Element-wise operation
Functions with scalar arguments, e.g., cos(x), are applied element-wise on vectors
and matrices, resulting in an output of the same shape as the argument. The same
holds for the algebraic operations addition and multiplication on arrays of the same
shape. Function names and operations are usually the same in excel and Python,
with one important exception: arcus tangens, as already mentioned in Sect. 2.4.4.
Broadcasting
Broadcasting is a technical term for an essential tool in Python, but the underlying
operations are also available in excel. If two operands of an operation are of different
but compatible shapes, they can be broadcast together. For a row vector of size n
and a column vector of size m, this is done by repeating the row vector vertically
and the column vector horizontally so as to get the same shape as an (n, m)-matrix.
When such vectors are, e.g., multiplied, the result is a matrix of shape (n, m), with
the elements being the element-wise product of the broadcast matrices.
Also, other operations, such as the multiplication of two vectors or the exponen-
tiation of a vector with another vector, are performed as if each of the operands were
repeated, so that matrices with a common shape are obtained that are then processed
element-wise.
Linear Algebra
Functions applied in linear algebra comprise determinant, inverse, and matrix
multiplication. Their names are listed in Table 2.10.
2.5 Matrix Calculations in Excel and Python 43
Table 2.10 Functions of linear algebra

EXCEL Python
TRANSPOSE(M) np.transpose(M)
INDEX(M; r; c) M[r, c] or M[r][c]
Indexing starts in EXCEL with 1, and in Python with 0 in square brackets:
EXCEL: INDEX(W;1) Python W[0]
MDETERM(N) np.linalg.det(N)
MINVERSE(N) npl.linalg.inv(N)
MMULT(M1, M2) M1 @ M2 or np.dot(M1,M2)
In Python, a matrix is constructed as a list of row vectors. So, indexing M[2][0]

points to the first element [0] in the third row M[2]. The abbreviation stands for
numpy.linalg that is imported under that name.
Spreadsheet first
We start our overview of matrix calculations with spreadsheet constructs because the
data structure there is immediately visible on the screen.
2.5.3 Matrices in Spreadsheets
Row and column vectors, matrices

In the upper half of Fig. 2.11 (S), two column vectors U, V of shape (3,1), two row
vectors W, X of shape (2,), and two matrices L, M of shape (3,2) are defined, using
the notation for shapes in Python. In the lower half of the figure, elements of the
defined entities are singled out, with spreadsheet functions index(vector; index)
and index(matrix; row index; column index). In the last two rows, the scalar
product of the vectors W and X is calculated in Cartesian and polar coordinates.

The length l of a vector is determined with the equation l = xi2 , obtained
by nesting two functions: sqrt(sumsq(W). A 2-dimensional vector can be inter-
preted as a straight line in the xy-plane directed from (0,0) to (x,y). Its angle to the
horizontal is determined with the arcus tangens function, = atan2(x,y), which,
for vector W, reads as = atan2(index(W;1);index(W;2).
U V W 1 2 L M
1 4 1 2 6 5
2 5 X 3 4 3 4 4 3
3 6 5 6 2 1
)
W;2)
);IN DEX(
) W;1
;1;2) Q(W) DEX(
EX(U
;3)
EX(M RT(SUMS N2(IN
=IND =IND =SQ =ATA
3 5 lW 2.24 aW 1.1
lX 5.00 aX 0.93
11.0 {=SUM(W*X)}
11.0 =lW*lX*COS(aW-aX)
Fig. 2.11 (S) Column vectors U, V, row vectors W, X, and matrices L, M are defined as named
ranges, and some operations are applied to them in the lower half of the figure
Broadcasting
Addition and multiplication of a row vector and a column vector, and of two matrices,
is demonstrated in Fig. 2.12 (S).
In the upper half of Fig. 2.12 (S), two vectors are added with the operator + or
multiplied with the operator *. When the operations are performed on two vectors
of the same type, row or column, they must be of the same size. The operations
are then performed element-wise, and the result is a vector of the same shape.
Activating a larger range leads to a repetition of the 1-dimensional result [see U*V
in Fig. 2.12 (S)]. Algebraic operations on matrices of the same shape result in a
matrix of the same shape (see L ∗ M).
All operations in Fig. 2.12 (S) are of the spreadsheet matrix type, as indicated
by the fact that they are enclosed in curly brackets. To recall: You have to activate
a range of size suitable for the result, enter the formula, and finish with the magic
chord: Ψ Ctrl, Shift, Enter.
Multiplying or adding a column vector (r,) (e.g., V) and a row vector (,c) (e.g.,
W) results in a matrix of shape (r,c) (see the results for W*V and V + W). The
operations are performed as if each of the operands is repeated, so that matrices
with a common shape are obtained that are then multiplied or added element-wise.
{=U*V} {=V*W} {=W*V} {=V+W}

4 4 4 8 4 8 5 6
10 10 5 10 5 10 6 7
18 18 6 12 6 12 7 8
scal 2
{=L+M} {=L*M} {=L*scal} {=L^2} {=SQRT(L)}
7 7 6 10 2 4 1 4 1.00 1.41
7 7 12 12 6 8 9 16 1.73 2.00
7 7 10 6 10 12 25 36 2.24 2.45
Fig. 2.12 (S) Arithmetic operations on vectors and matrices defined in Fig. 2.11 (S)
{=L*V} {=V*L} {=L*W} {=TRANSPOSE(L)}

4 8 4 8 1 4 1 3 5
15 20 15 20 3 8 2 4 6
30 36 30 36 5 12
Fig. 2.13 (S) Operations on vectors and matrices with different shapes, defined in Fig. 2.11 (S)
after broadcasting
} } )}
N;O) W;N) SE(O
MMULT( MMULT( MINVER
{= {= {=
N O I
1 2 4 3 8 5 7 10 -0.5 1.5
3 4 2 1 20 13 7 10 1 -2
#N/A #N/A 7 10
-2 =MDETERM(N)
1 0 =MMULT(O;I)
0 1
Fig. 2.14 (S) Mathematical matrix operations on the square matrices N and O
The same holds for the multiplication of a vector or a matrix with a scalar (see
L*scal). Functions applied to vectors or matrices are also performed element-wise
and yield a range of the same shape as the argument (see {=Lˆ2} or {=sqrt(L)}).
In Fig. 2.13 (S), operations combining a vector and a matrix are performed.
Before an element-wise operation, the vectors are repeated so as to obtain the
same shape as the matrix that also becomes the result’s shape.
A matrix is transposed with the function {=transpose(Matrix)}. In Fig. 2.13
(S), this is applied to L.
Operations of Linear Algebra

Operations and functions on mathematical matrices are shown in Fig. 2.14 (S). On
square matrices, the determinant can be obtained with mdeterm and its inverse
matrix with minverse.
Two matrices of suitable shapes, (n1 ,n2 ) and (n2 ,n3 ), can be multiplied with
MMult (Matrix1, Matrix2) ; the result is of shape (n1 , n3 ). Multiplying the original
matrix O with its inverse I yields the unit matrix.
2.5.4 Matrices in Python
Before operating with numpy, we have to import this library (see line 1 in Table
2.11, where we also set the printoptions for arrays that are to be printed,
here, in the float format 0.2f with 2 decimal places and one blank between the
printed numbers. The “0” in 0.2f indicates that the number of characters is not
fixed but adapted to the current number. Additional separating blanks are obtained
by inserting them between ‘ and {, or after {: as in line 3. In the following, print
Table 2.11 Importing numpy and setting a print option for arrays
2 np.set_printoptions(formatter={'float':'{: 0.2f}'.format})
3 x=np.linspace(0,10,4)
4 print(‘x \n’,x) #\n Makes a line feed
x
[ 0.00 3.33 6.67 10.00]
statements similar to that in line 5 of Table 2.11 are not explicitly reported, but
only the results, as in the bottom cell of Table 2.11. The key word “\n” induces a
line feed.
Row vectors
In Table 2.12, row vectors W and X are created as an array with the same
elements as in Fig. 2.11.The length of W, calculated correspondingly with
np.sqrt(np.sum(W*W)), is the same. The scalar product of two vectors
can be calculated with the function np.dot or with the operator @ for matrix
multiplication: W @ W.
Column vectors
Row vectors can be transformed into column vectors by transposition, most
simply with the function np.transpose(), but also with the extension
.transpose(1,0) fixed to the identifier of a vector (or, more generally, of a
matrix) (see Table 2.13). The original vector UR is not defined as a row vector, but as
a matrix with only one row, characterized by enclosing the list within double square
brackets, [[ …]]. This (1, 3)-matrix can be transposed with np.transpose(UR)
or UR.transpose(1, 0) to become a (3,1)-matrix equivalent to a column
vector. The transpose operation can also be applied to more-dimensional matrices.
In said table, we also demonstrate the element-wise multiplication of two column
vectors U, V and the broadcasting that occurs when a row vector (W from Table
2.12) and a column vector V are multiplied. An instruction [1,2,3]*[1,2] results in
an error message: “Operands could not be broadcast together”. The operation * is
commutative: W*V = V*W.
Table 2.12 Creating row vectors in Python and determining their scalar product; print state-
ments are not reported in Cell 1, and only their results are reported in Cell 2
1 W = np.array([1, 2]) W, np.shape(W) [1 2]
2 X = np.array([3, 4]) (2,)
3 lW = np.sqrt(np.sum(W*W)) lW 2.24
4 ld = np.dot(W, W) ld 5
1 W @ W 5
Table 2.13 Creating column vectors in Python, multiplication of two column vectors
and multiplying column vector V with row vector W (from Table 2.12). The entries in the
lower cells are obtained by print statements in the top cell that are not reported there, e.g.,
print("np.shape(UR)", np.shape(UR))
1 UR = np.array([[1, 2, 3]])
2 U1 = np.transpose(UR)
3 U = UR.transpose(1, 0)
4 V = np.array([[4, 5, 6]]).transpose(1,0)
np.shape(UR) (1, 3) V
UR [[4]
[[1 2 3]] [5]
[6]]
np.shape(U1) (3, 1)
U1 U*V
[[1] [[ 4]
[2] [10]
[3]] [18]]
np.shape(U) (3, 1) V*W

U [[ 4 8]
[[1] [ 5 10]
[2] [ 6 12]]
[3]]
np.sum(U*V) 32
U[0][0] [1]
Matrices
In Table 2.14, we see how 2-dimensional matrices are created with the function
np.array( …) . The argument of np.array is a list of 1-dimensional lists all
of the same size. Algebraic operations such as L*M on matrices with equal shape
are again performed element-wise.
In Table 2.15, multiplication of L [shape (3,2)] with V [shape (3,1)] and W

[shape (2,)] is demonstrated. These are some further examples of broadcasting.
The effect of matrix transposition, np.transpose(L) , is also reported. Shape
(3,2) indicates 3 rows with 2 elements each, and may also be interpreted as a
matrix with 3 rows and 2 columns.
Linear Algebra
In Table 2.16, we learn about linear algebra operations on square matrices (shape
(n, n)) using the functions inv and det of the linalg sublibrary of numpy
that is often imported as np. Np.linalg.det returns the determinant and
np.linalg.inv() the inverse of a square matrix.
Table 2.14 Creating 2-dimensional matrices in Python and performing algebraic operations on
them; the entries in the bottom cells are obtained by print statements in the top cell that are not
reported there, e.g., print("L\n",L)
1 L = np.array([[1, 2], [3, 4], [5, 6]])
2 M = np.array([[6, 5], [4, 3], [2, 1]])
np.shape(L) (3, 2) L*M
[[ 6 10]
L [12 12]
[[1 2] [10 6]]
[3 4]
[5 6]] L**2
[[ 1 4]
M [ 9 16]
[[6 5] [25 36]]
[4 3]
[2 1]] np.sqrt(L)
[[ 1.00 1.41]
[ 1.73 2.00]
[ 2.24 2.45]]
Table 2.15 Product of a 2-dimensional matrix (L defined in Table 2.14) with 1-dimensional
vectors V, W defined in Tables 2.13 and 2.12, respectively; the entries are obtained with, e.g.,
print("L*V\n", L*V)
L L*V
[[1 2] [[ 4 8]
[3 4] [15 20]
[5 6]] [30 36]]
V L*W
[[4] [[ 1 4]
[5] [ 3 8]
[6]] [ 5 12]]
W np.transpose(L)
[1 2] [[1 3 5]
[2 4 6]]
The determinant of matrix N is 1·4 – 2·3 = −2. Python deviates from that value
in the 16th decimal. This is because the calculations are performed in the binary
system.
The operator @ stands for matrix multiplication. The matrix product of a matrix
with its inverse yields the unit matrix (see lines 18 to 20 in Table 2.16). The differ-
ence to the algebraic multiplication with the operator *, operating element-wise, is
demonstrated in Table 2.17, L*W vs. L@W.
Table 2.16 Operations of linear algebra on square matrices

1 N=np.array([[1, 2], [3, 4]])
2 O=np.array([[4, 3], [2, 1]])
3 I=np.linalg.inv(N)
N np.linalg.det(N)
[[1 2] -2.0000000000000004
[3 4]]
O
[[4 3] N @ I #Mathematical matrix multiplication
[2 1]] [[ 1.00 0.00]
[ 0.00 1.00]]
I
[[-2.00 1.00]
[ 1.50 -0.50]]
Table 2.17 Array multiplication with * versus mathematical matrix multiplication with @
L L*W
[[1 2] [[ 1 4]
[3 4] [ 3 8]
[5 6]] [ 5 12]]
W L @ W
[1 2] [ 5 11 17]
Questions concerning Tables 2.15, 2.16, and 2.17
What are the shapes of L, W, L*W, np.transpose(L)? 9
What is the shape and the type of N@I?10

What are the shapes of L, W, L@W ?11
2.6 Four Parabolas and Their Upper Envelope
Four parabolas are represented in a figure, together with their upper

envelope, first produced by a spreadsheet calculation and then by an anal-
ogous Python program intended to serve as a basic exercise in Python
programming and developed step by step in great detail.
9 Shape(L) = (3, 2), shape(W ) = (2,), shape(L*W ) = (3, 2), shape (np.transpose(L)) = (2, 3).
10 N@I is a diagonal matrix, shape = (2, 2).
11 Shape(L) = (3, 2), shape(W ) = (2,), shape(L@W ) = (3,).
2.6.1 Graphical Representation
Figure 2.15 shows four parabolas with their vertices marked with diamonds and
their upper envelope always running on the, at the respective position, top parabola.
The spreadsheet for producing this figure is given in Fig. 2.16 (S). It is briefly
described in the following section on the basis of Exercise 2.2. The corresponding
10 J K L M
(xV, yV)
y yA 17 -10.0 2.4 4.8 -96.4
8 yB 18 -8.8 2.4 4.7 -94.8
yC 19 -7.7 2.3 4.7 -93.3
yD 177 -330.0 5.0 1.0 -0.4
6 yMax
0
-8 -4 0 x 4 8
Fig. 2.15 a (left) Four parabolas for the parameters specified in Fig. 2.16 (S). b (S, right) Coordi-
nates for the four parabolas obtained with a matrix formula, J17:M177 = [{=a + b*x + c_*xˆ2}]
A B C D E F G H
1 xV -5 -2 4 5
2 yV 8 1 0.5 5 (xV, yV)
3 c_ -2 0.04 0.03 -0.6
4 b -20 0.16 -0.24 6
5 a -42 1.16 0.98 -10
6 dx 0.1
^2 10^2 3*$A11^2 *$A10^2

*$A9 3*$A $ 3
9+B$3 $A10+C$ *$A11+D $A10+E$ E10) 2
*$A * * : _*x^
7 =A9+d
x +B$4 C$5+C$4 D$5+D$4 E$5+E$4 MAX(B10 *x+c
=B$ 5 = = = = =a+b
8 x yA yB yC yD yMax
9 -8.0 -10.0 2.4 4.8 -96.4 4.8
10 -7.9 -8.8 2.4 4.7 -94.8 4.7
11 -7.8 -7.7 2.3 4.7 -93.3 4.7
169 8.0 -330.0 5.0 1.0 -0.4 5.0
170
Fig. 2.16 (S) Four parabolas defined by the parameters in row vectors in rows 1 to 5; the formula
reported in G7 is a better alternative for the formulas in B7:E7
2.6 Four Parabolas and Their Upper Envelope 51
Python program is developed in detail in Sect. 2.6.4, serving as the basic exercise
in Python programming.
x V , yV arrays of the coordinates of the vertices of 4 parabolas

a, b, c arrays of the coefficients of 4 parabolas y = a + bx + cx 2
x array of x values (here, 161 from -8 to 8), separated by dx
yA , yB , yC , yC four parabolas defined over x
G matrix comprising yA , yB , yC , and yD as columns
yMax maximum of yA , yB , yC , and yD at the x values.
2.6.3 Spreadsheet Calculation
Figure 2.16 (S) shows a spreadsheet tabulating four parabolas for the values of x in
A9:A169 and the parameters in rows 1 to 5, valid for the y-values of the parabolas
in the respective column.
Question
What are the formulas for b in B4:E4 and a in B5:E4 of Fig. 2.16 (S)?12
The task is to write a formula into cell B9 with relative and absolute addresses
so that it can be copied into the whole range B9:E169 to produce the values of
the four parabolas yA , yB , yC , yD with the parameters c_, b, and a in range B3:E5.
Parameter vector c_ is specified directly by entering the desired values into B3:E3.
Parameter vectors b and a are obtained from the coordinates of x V and yV of the
vertex with the same formula as in Fig. 2.2 (S), realizing Eq. 2.3. In column F,
the maximum yMax of the four parabolas is built, giving their upper envelope. The
result is shown in Fig. 2.15a.
The column vector x is defined in A9:A169 in 161 steps of dx = 0.1. The
values of the four parabolas are then generated by typing the equation reported in
B7 into B9 and copying right and down to E169, the range spanned by x and the
row vectors a, b, c_. The resulting curves are shown in Fig. 2.15a, together with
the vertices (x V , yV ) and the curve yMax calculated in column F.
Naming cell ranges

Row and column ranges can be provided with names (see Sect. 2.3.5). Activating
A1:E5 and going through formulas/create from selection/create names
12 b = -c·x V ; a = yV + c·x V 2 , see Fig. 2.2 (S) or Table 2.1.

Table 2.18 Specification of the coefficients of 4 parabolas and composition of their labels
2 #Specify position (xV,yv) of vertices!
3 xV=np.array([-5.0,-2.0,4.0,5.0])
4 yV=np.array([ 8.0, 1.0,0.5,5.0])
5 #Define coefficients of y=a+bx+cx³
6 c=np.array([-2.0,0.04,0.03,-0.6])
7 b=-2*c*xV
8 a=yV+c*xV**2
9 #Compose labels for the figure
10 lbl_1=str(a[0])+"+"+str(b[0])+"⋅x+"+str(c[0])+'*x²'
14 print(lbl_1)
15 -42.0+-20.0⋅x+-2.0*x²
from values in the/left column does the job. Proceeding correspondingly

with A8:A169 gives us a column vector named x. The formula can now be B9 = [=a
+ b*x + c_*xˆ2] instead of B9 = [B$5 + …], and remains the same when copied
into the whole range B9:E169.
Application as a matrix formula

The formula [=a + b*x + c_*xˆ2] can be applied as a matrix formula within any
range of suitable size. In Fig. 2.15b, the range J17:M177 has been activated, the
formula entered and the process finished with the magic chord: Ψ Str + Alt + Enter.
The same numbers appear as in B9:E169 of Fig. 2.16 (S).
Now, how to do all this in Python?

In the first cell of Table 2.18, we import the numpy library. We run the cell so
that the features and functions of numpy are available in the following cells.
Questions
What are the formulas to get a and b from the coordinates of the vertex?13
What are the shapes of c, b, and a in Table 2.18?14
13a = yv + c · xv2 and b = −2 · c · xv .

14Shape(c) = (4,), c with 4 explicit entries; b and a have the same shape, because they are
constructed with c.
Table 2.19 Creation of a column vector x

1 xS=np.linspace(-8,8,161) #Is a vector = 1D array
2 xR=np.array([xS]) #Is a matrix = 2D array with one row
3 x=xR.transpose(1,0) #Creates a column vector
4
5 np.set_printoptions(edgeitems=2,
6 formatter={'float': '{: 6.1f}'.format})
7 print(‘xS ‘,xS)
8 print(‘xR ‘,xR,'\n')
9 print(" x\n",x,'\n')
xS [ -8.0 -7.9 ... 7.9 8.0] x
xR [[ -8.0 -7.9 ... 7.9 8.0]] [[ -8.0]
[ -7.9]
...
[ 7.9]
[ 8.0]]
The next cell contains that part of the main program that reproduces the def-
initions and the data structure of the spreadsheet solution. The parameters x V ,
yV and a are specified exactly as before in numerical row vectors with four ele-
ments. The parameters b and c are obtained with the formulas mentioned above
(in Sect. 2.2.1), also as row vectors.
In the third cell, the labels of the four curves in Fig. 2.15 are composed.
Remember: the concatenation operator in Python is + (contrary to & in
excel), and numerical values have to be converted explicitly into a string, e.g.,
str(a[0]). They are used as the legend in the chart.
In Table 2.19, the column vector x, representing the independent variable, is
constructed. We could achieve that in one statement:
x = np.array ([np.linspace(0.0,1.0,11)]).transpose(1,0)
but use the three lines in the first cell instead, so as to make the construction
clearer. First, a simple list x S with 161 items equally spaced between −8 and 8,
including 8 as the last element, is generated. In the output (bottom) cell, it shows
up as a 1-dimensional row vector. To get a column vector, we first have to make
it a matrix (2-dimensional array) x R by defining an np.array with just one row
showing up in the output cell as a list within double square brackets [[…]]. This
matrix is transposed, with x = xR.transpose(1,0), indicating that the axes
0 and 1 are interchanged, now yielding a matrix x with just one column. This is
the desired column vector.
In Table 2.20, a two-dimensional array named G is constructed with G = a
+ b*x + c*x**2. It has the same shape and the same content as the range
B9:E169 in Fig. 2.16 (S). The four columns of array G get additional names yA to
yD , as in the spreadsheet. Remember: Python is zero-addressing; first index is 0.
In column F of Fig. 2.16 (S), we have calculated the maximum of the four
parabolas for each value of x. The same is achieved in Python with the statement
Table 2.20 Creating the y values of all 4 parabolas; G[:,0] indicates first column
1 G=a+b*x+c*x**2 5 print(G,'\n')
2 yA=G[:,0] #1st col. 6 print(yA)
3 yB=G[:,1] #2nd col. 7 yMax=G.max(axis=1)
4 yC,yD=G[:,2],G[:,3] #Max. across the columns for every x
8 print(yMax)
G
[[ -10.0 2.4 4.8 -96.4]
[ -8.8 2.4 4.7 -94.8]
...
[-324.8 4.9 1.0 -0.0]
[-330.0 5.0 1.0 -0.4]]
yA [ -10.0 -8.8 ... -324.8 -330.0] #First column
yMax [4.82 4.74 ... 4.92 5.0] #For every x
yMax = G.max(axis = 1). G is a two-dimensional array, and we have to

specify the axis along which the maximum value has to be found. Axis = 1
indicates that it is across columns for every entry in a row. The resulting curve is
shown in Fig. 2.17 as a bold gray line.
Figure 2.17 is similar to Fig. 2.15 obtained with excel, but now created by the
Python program FigStd described in Sect. 2.4.5.
Fig. 2.17 Similar to

Fig. 2.15, but produced with
the Python program in
Table 2.21
Table 2.21 Program for plotting Fig. 2.17

1 import matplotlib.pyplot as plt
2 FigStd('x',-8,8,4,'y',0,10,2)
3 plt.plot(xV,yV,'kd')
4 #’kd’ means points as black diamonds
5 plt.plot(x,yMax,color='gray',lw=4,label='yMax')
6 plt.plot(x,yA,'k-', lw=1, label=lbl_1) #’k-‘ Black Line
7 plt.plot(x,yB,'k--',lw=1.5,label=lbl_2) #-- Dashed line
8 plt.plot(x,yC,'k-.',lw=2, label=lbl_3) #-. Dash-dotted
9 plt.plot(x,yD,'k:', lw=3, label=lbl_4) #: Dotted
10 plt.legend()
11 plt.savefig('PhEx2-2 parabolas.png',dpi=1200)
Plotting four parabolas

The program for plotting Fig. 2.17 is given in Table 2.21, with the function FigStd
called in the second line. It produces a figure of size 4 cm × 4 cm with scaling and
labeling of the axes as specified in the arguments. The figure is made with the entries
label, minimum, maximum, and distance between the ticks, each for the x-axis and
the y-axis.
We add five curves with the statement plt.plot(). The plot statements all
get the pre-syllable plt, the short form under which the matplotlib library
has been imported. The x-values and the y-values of the curve to be plotted are
positional arguments and have to be the two first arguments of plt.plot. The
third argument specifies the color and style of the curve. The keyword ‘ks’ in line
2 specifies that the points (x Ext , yExt ) are to be marked with black (‘k’) squares
(‘s’). The next plot statements contain more keyword arguments, lw for linewidth,
and label for the curve’s label to be displayed in the legend. Additionally, in the
last plot statement, the curve’s color is explicitly specified with color=’gray’.
For more information on plot styles and options, consult Python Help!
The legend with all labels is plotted by calling the function plt.legend().
Generally, you can specify its position in the figure. The default is ‘loc’ =
0, indicating that the program should choose an optimum position. Specifying
‘loc’ = 2 would place the legend at the top-left part of the figure. Various style
specifications can also be incorporated with keyword arguments, e.g., fontsize
= 10. With the last statement, the figure is saved under the indicated name and
with the indicated resolution.
Question
The statement plt.legend will not plot the legend. Why?15
15The statement plt.legend() calls a function, and function identifiers have to be supplemented with
parentheses.
Table 2.22 Extrema of matrix G along different axes

1 yMax=G.max(axis=1) #Across columns, for every x
2
3 yMaxCol=G.max(axis=0) #Down the rows, for every column
4 iMaxCol=G.argmax(axis=0) #Index of Max in the row
5 xMaxCol=x[iMaxCol] #Corresponding x value
6
7 yMinCol=G.min(axis=0)
8 xMinCol=x[G.argmin(axis=0)] #Two statements in one
9 np.set_printoptions(precision=2)
xMaxCol yMax [4.82 4.75 ... 4.92 5. ]
[[-5.] #1 Max. for every column:
[ 8.] yMaxCol [ 8. 5. 4.82 5.]
[-8.] xMaxCol[:,0] [-5. 8. -8. 5.]
[ 5.]] #1 Min. for every column:
yMinCol [-330. 1. 0.5 -96.4]
xMinCol[:,0] [ 8. -2. 4. -8. ]
2.6.5 Extrema Along Different Axes
In Table 2.22, we determine the extrema of the 2-dimensional matrix G, defined

in Table 2.20 and containing the y values of the parabolas. The variable yMax
contains the maxima for axis 1, i.e., across the columns (index 1), for every entry
in the rows, i.e., for the same value of x. This is the upper envelope of the four
parabolas.
Next, we will find the individual extrema (minima and maxima) of the four
parabolas. To do so, we have to build the maxima yMaxCol and minima yMinCol for
every single curve. This is done down the rows (index 0), for axis = 0. The
arrays yMaxCol and yMinCol contain 4 entries each. The indices iMaxCol at which the
maxima occur in the columns of G are found with iMax = G.argmax(axis
= 0); x MaxCol = x[iMaxCol ] gives the corresponding x values. For the coordinates
yminCol and x minCol of the minima, we proceed accordingly with argmin.
In the printout cell, we realize that x Max is a column vector, i.e., a matrix with
only one column. The slice xMaxCol [:,0] extracts the column within the matrix
and displays it as a row.
Questions
Are the values for (x Min , yMin ) and (x Max , yMax ) consistent with the prespecified
values of the vertices (x V , yV )? Compare with Fig. 2.17!16
16 The four vertices show up as the 1st and 4th point in (x Max , yMax ) and the 2nd and 3rd point in
(x Min , yMin ). The other points are at the boundaries of the x range.
Fig. 2.18 Illustration to

demonstrate indexing of G[0,:]
axis = 0
matrices
G[:,0]
axis = 1
The indexing of rows, columns, and axes seems not to be very intuitive.
Therefore, we illustrate it in Fig. 2.18. G[0,:] addresses the first row: axis = 0
means across the rows yielding a number of values corresponding to the num-
ber of columns. G[:,0] addresses the first column, axis = 1 means across the
columns yielding a number of values corresponding to the number of rows , in
our example for every value of x.
2.7 Sum of Four Cosine Functions
We sum up four cosine functions with different angular frequencies ωi . When

the frequencies are multiples (overtones) of a fundamental frequency, such
sums mimic the time signals of sound. Beats are generated when the fre-
quencies are equally spaced within a small frequency range. The formula
for the addition of cosines is illustrated by setting each two frequencies as
equal, with the result being described by the broom rule “Cos plus cos
yields mean times half the difference”.
2.7.1 Sound and a Cosine Identity
Vibrations of a string
Mag Take a look at the microphone signal in Fig. 2.19a! How would you
describe the signal?
Alac Well, a peak is repeated periodically, and in between, there is a lot of

fidgeting.
Mag Yes and no. Yes, there is a fundamental frequency of repetition, and no,
sound is not fidgeting; it is composed of harmonics.
Figure 2.19a is the record of the sound of a guitar string. What is the
fundamental frequency in this case?
Fig. 2.19 a (left) Oscilloscope image of a microphone signal of a vibrating guitar string as a func-
tion of time (courtesy of Norbert Renner, University of Duisburg-Essen), time unit of the grid =
5 ms., b (right) Fundamental) and first harmonic oscillations (first overtones) of a reed of length l
fixed on one end (top), l = (λn /4)·(2n + 1) and a string of length l fixed on both ends (bottom), l
= (λn /2)·(n + 1); n indicates the number of internal nodes
Tim The period is one-and-a-half grid distance, 7.5 ms, corresponding to a

frequency of 133 Hz. But what are harmonics?
Mag Let’s consider the vibrations of a guitar string. How does a string vibrate?
Tim It is sinusoidally excited, as in the bottom part of Fig. 2.19b.
Mag Yes.
Figure 2.19b suggests that there are only discrete values of the wavelength with
which a string can vibrate. You can determine them by considering the boundary
conditions.
Tim The sine must go through zero, where the string is clamped.
Alac In between, it may also go through zero.
Mag Exactly. The zero-crossings are called nodes. The boundary conditions
require that the string length l be a multiple of half the wavelength, l = (n + 1) ·
λn /2, n = 0, 1, 2, … The possible frequencies are multiples of the fundamental fre-
quency c/(2l)), where c is the velocity of sound on the string. The corresponding
vibrations are called harmonics. The mode with n = 0 is called the fundamental
mode. Now, let me repeat my question: How do you now describe the microphone
signal?
2.7 Sum of Four Cosine Functions 59
3 3
ω0=1, all ω0=0.5, odd
2 2
y y
1 1
0 t 0 t
0 10 20 30 0 10 20 30
-1 -1
-2 -2
Fig. 2.20 a (left) Sum of a fundamental tone with ω0 = 1 and three overtones at 2 x, 3 x, 4 x ω0 .
The time unit is 1 s when the values for ω are given in 1/s. b (right): Sum of a fundamental tone
with ω0 = 0.5 and three overtones with 3 x, 5 x, 7 x ω0 . Time unit = 1 s
Alac I guess it is the sum of the harmonics.
Mag Right, it is the sum of the allowed vibrations with individual amplitudes.
In the case of a string clamped on both ends, the frequencies are multiples (1, 2,
3, …) of the fundamental frequency, f n = f 0 · (n + 1), or correspondingly for the
circular frequencies ωn . In cases in which one end of the vibrating medium is free,
such as, for example, in a saxophone reed clamped on one side, the frequencies
are odd multiples (1, 3, 5, …) of the fundamental frequency, f n = f 0 · (2n +
1). Examples are given in Fig. 2.20. There, you see the sum of four cosines with
a fundamental frequency ω0 = 1 and multiples 2ω0 , 3ω0 , 4ω0 (Fig. 2.20a) or
multiples 3ω0 , 5ω0 , 7ω0 (Fig. 2.20b).
Task in this exercise
Tim Now, are we going to emulate the microphone signal?
Mag Yes, but in the more general context of summing up four cosine functions
whose frequencies satisfy certain conditions.
In this exercise, we shall use the form y = A·cos(ω · t + φ) with the parameters
amplitude A, circular frequency ω, and zero phase φ. We set up a calculation
model with four cosines, with their circular frequencies specified by a fundamental
frequency and either three multiples thereof (to get harmonics) or three more
frequencies at a distance of dω (to get beats or demonstrate the addition theorem
of cosines).
Beats and the uncertainty relation

Beats are shown in Fig. 2.21a. The four frequencies are again equidistant, but packed
together in a small frequency range ω. As control parameters for the calculation,
to be systematically varied later, we take the lowest frequency ω1 and the width
of the frequency range ω, i.e., the difference between the highest and the lowest
w.0=4; Delta.w=3 w.0=2; Delta.w=1

4 4
y y
2 2
0 0
0 10 20 30 0 10 20 30
t t
-2 -2
-4 -4
Fig. 2.21 a (left) Beats arising from four cosine functions; the initial angular frequency ω0 = 4,
the width in the frequency range ω = 3, b (right) another beat with initial angular frequency ω0
= 2, and width in the frequency range ω = 1
frequencies. As a result, the signal clusters together into wave packets. In Fig. 2.21,
they are separated by black markings. We shall determine the width of such wave
packets based on an uncertainty rule.
Mag The amplitudes of the cosine components of a beat are not freely chosen,
but rather derived from binomial coefficients. How do you get such coefficients?
Alac Well, with Pascal’s triangle. Why so precise rules for the amplitudes?
Can’t we choose them as we like?
Mag With said choice of amplitudes, we get a clear picture of the beat in the
time domain, with the envelope being similar to a bell curve. Initially, you are to
change only the lowest frequency ω0 and the spectral width ω and observe the
function’s behavior in the time domain. Later, you may select the amplitudes at
will and see whether the observed regularity is preserved.
Task You are to insert points into the diagrams at the position of the nodes
of the oscillations. To do so, you may create formulas that specify the marker
points’ coordinates when the width of a packet t and an initial time offset t 0 are
specified. In a spreadsheet, it would be best for you to use sliders that change the
coordinates so that the points in the diagram lie precisely on the nodes of the beat.
Mag Have you figured out the rule for the width of a wave packet?
Tim The time interval t for a packet is proportional to the inverse of the
spectral width t = 3π/ω = 1.5/ f . The width is certainly smaller, because
the signal goes to zero well before the marker points.
Mag With that statement, you have found an uncertainty relation: f · t =

3/2 ≥ 1.
Tim Heisenberg’s uncertainty relation?
Mag Yes, it is related to the commutation relation of time t and energy E =

h f , E · t = 23 h ≥ h.
Sum of cosines
Mag How do you write cos(x) + cos(y) as a product of two trigonometric

functions?
Alac Is it necessary to know such things? You can look it up in handbooks.
Tim I’ve memorized it: “the mean times half the difference.”
Cos plus Cos yields the mean, times half the difference.
Mag This broom rule is a useful mnemonic if you can reconstruct the full form
from it:

x+y x−y
cos(x) + cos(y) = 2 · cos cos (2.8)
2 2
Alac I’m certainly never going to forget such a crazy saying. The first cosine
in the product has the average of the two primary cosines, the second half the
difference.
Mag Exactly, and if x = 0 and y = 0, then 1 + 1 = 2 must be the result.

This explains the pre-factor on the right-hand side of Eq. 2.8. We can test the sum
formula for two cosines with our calculation model by making each two of our
four frequencies equal so that only two different frequencies remain and setting all
amplitudes to 0.5. The result of such a calculation can be seen as the thick gray
curve in Fig. 2.22.
The black dotted curve corresponds to the function.

dw
2 cos w0 + t (the mean frequency)
2
and the dashed curve to.

dw
2 · cos ·t (half the difference of the frequencies).
2
We see that the cosine with half the difference frequency (dashed line)
envelopes the sum. The curve oscillates with the mean frequency (dotted line).
The product of the two curves experiences a phase shift π after each zero crossing
of the envelope.
2
y
1
0
0 10 20 30
-1
t
-2
w.0=2; Delta.w=1 w=w.0+dw/2 w=dw/2 black diamonds
-3
Fig. 2.22 Thick gray curve: the sum of two cosine functions with the frequencies ω1 and (ω1 +
ω); black dotted curve: cosine function with the mean frequency; black solid curve: envelope
with half the difference of the frequencies
dt distance between adjacent points of time

t vector of 801 equidistant points of time
c1 , c2 , c3 , c4 cosine functions at t
c matrix [c1 , c2 , c3 , c4 ]
A array of 4 amplitudes
ω array of 4 circular frequencies
ω0 lowest frequency
Δω width of the frequency range of a beat
dω distance between frequencies
φ array of 4 phase shifts.
sumC sum c1 + c2 + c3 + c4
2.7.3 Spreadsheet Layout
From the very beginning: a clear layout!

We set up a worksheet calculation to sum up four cosine functions, with their ampli-
tudes A, angular frequencies ω, and zero phases φ to be chosen freely. The functions
are to be calculated from t = 0 to t = 32 s for 801 sampling points. The basic -
structure of a suitable spreadsheet set-up can be seen in Fig. 2.23 (S). The names c1 ,
c2 , c3 , c4 of the cosine functions are written in the spreadsheet with a dot separating
the letter and the number, c.1, etc., because C1, etc., are cell addresses.
Overtones
The problem of generating overtones (higher harmonics) is a special case of the
general task. The functions are to be calculated within a matrix range of, in our
A B C D E F G H I J
1 =w.0 =2*w.0 =3*w.0 =4*w.0 , all
2 A 0.23 0.75 0.63 0.72
3 w 1.00 2.00 3.00 4.00 w.0 1.00 ω0
4 phi 0 0 0 0 ω0=1, all =J3&"="&w.0&G1
5 0.04
i) i) i) i)
t+ph (w*t+ph (w*t+ph (w*t+ph 9:E9)
$A$5 O S(w*
O S O S O S (B
6 =A8+ =A*C =A*C =A*C =A*C =SUM
7 t c.1 c.2 c.3 c.4 sumC
8 0.00 0.23 0.75 0.63 0.72 2.33 Write formula into B8!
9 0.04 0.23 0.75 0.63 0.71 2.31 Dragg into B8:E808!
808 32.00 0.19 0.29 -0.11 -0.50 -0.13
Fig. 2.23 (S) Four cosine functions c1 to c4 are calculated in the matrix range B8:E808 (below ),
the cells of which always contain the same formula. The four functions are summed up in column F.
The formulas in the columns are displayed in row 6 in oblique orientation. The time t (the indepen-
dent variable) is defined as column vector A8:A808 (left of ) with the name t. The amplitudes A,
the angular frequencies ω, and the zero phases ϕ are defined as row vectors B2:E2, B3:E3, B4:E4,
respectively (above ). Here, the angular frequencies are multiples of the fundamental frequency
ω0
particular case, width 4 (number of functions) and height 801 (number of data points)
. We write the desired values for amplitudes A, circular frequencies ω, and zero phase
φ in rows above the matrix range and the independent variable t in a column to the
left of the matrix range.
You are to organize the worksheet calculation such that you need to write a formula
into only one cell that is then copied to the entire calculation area (here B8:E808)
for the four functions by dragging down and to the right.
Beats
The parameters for beats are specified in H2:I4 of Fig. 2.24 (S), namely, the first
frequency ω1 and the frequency range ω, from which dω, the distance between
neighboring frequencies, is obtained. The amplitudes A are binomial coefficients (up
to a factor of 2) and can be determined using Pascal’s triangle.
Cosine identity
Figure 2.25 (S) presents a spreadsheet layout for obtaining the cosine identity shown
in Fig. 2.22, the same as for beats. Here, however, c1 = c3 ; c2 = c4 , and all amplitudes
A B C D E F G H I J
1 =w.0 =B3+dw =C3+dw =D3+dw =Delta.w/3
2 A 0.5 1.5 1.5 0.5 Delta.w 1.00
3 w 2.00 2.33 2.67 3.00 dw 0.33
4 phi 0 0 0 0 w.0 2
5 0.04 w.0=2; Delta.w=1
Fig. 2.24 (S) Parameters for a beat; ω (“omega”) is coded as w

Fig. 2.25 (S) Parameter set for a sum of two cosine functions; c1 = c3 ; c2 = c4 ; H5 displays the
legend for one of the curves in Fig. 2.22, while there are formulas in K6:L6 for the angular fre-
quencies of the fast oscillation and the envelope according to the sum formula, Eq. 2.8, applied in
K8:K808 and L8:L808; the amplitudes are taken from cell $F$8. K4 and L5 contain the legend for
the other two curves in Fig. 2.22
are equal (= 0.5). The legends for the fast oscillation (with medium frequency) and
the envelope (with half the difference frequency) are assembled in K1:L5.
Program flow in Jupyter

The program flow in the cell structure of Jupyter is shown in Fig. 2.26. The
parameters for five subtasks are specified in five program cells. The identifiers of the
parameters are the same in all five cells shown in Table 2.26 and Table 2.27. The
main program gets the parameter values from the cell that has been run immediately
before the main program is started. To keep track of the program flow, labels that
contain all relevant parameters are created in all subtasks and displayed in the main
program as legends in figures.
Fig. 2.26 Program flow: the

Overtones, all
parameters for five subtasks
are specified in five cells and
alternatively fed into the main Overtones, odd Main program
program
Columnar vector t
Beats 1
Calculate c, sumC
Beats 2
Plot sumC
Add Cosines
Table 2.23 Main program for calculating 4 cosine functions and plotting their sum sumC ;yMin ,
yMax , and FigName are specified in Tables 2.25–2.27
1 t0=np.linspace(0,32,801)
2 t1=np.array([t0])
3 t=t1.transpose(1,0)
4 c=A*np.cos(w*t+phi)
5 sumC=np.sum(c,axis=1)
6
7 FigStd('t',0,30,10,'y',ymin,ymax,dy)
8 plt.plot(t,sumC,'k-',label=lbl_0)
9 plt.legend(loc=0,fontsize=11) #loc = 0, best free place
10 plt.savefig(FigName) #Store figure as a file!
Main program
The main program for calculating the four cosine functions and plotting their sum is
given in Table 2.23. We assume that we have imported the two libraries numpy and
matplotlib.pyplot under the shortcuts np and plt, as well as the function
FigStd described in Sect. 2.4.5. This shall be done in all future programs and shall
not be reported explicitly. The coefficients A, ω, φ, as well as the label lbl 0 for the
curve and the name FigName of the file name under which the figure is saved, have
to be specified ahead of time. Figures 2.20, 2.21, and 2.22 all have different scaling
of the y axis, so that the corresponding axis parameters ymin , ymax , dy must also be
specified in the subtask cells before calling FigStd.
This is done in the program cells for the different situations: harmonics (Table
2.25), beats (Table 2.26), and the addition of cosines (Table 2.27).
In order to get the same data structure as in the spreadsheet solution, we have
to construct t as a column vector. We first define our discrete time points accord-
ing to column A in Fig. 2.23 (S). This is done with t0 = np.linspace(0,
32, 801) specifying that the range from 0 to 32 is scanned with 801 equidis-
tant points. The endpoint 32 is included by default. If we do not want that, we
have to include “endpoint = False” as an entry behind the three positional
arguments: np.linspace (0, 32, 801, endpoint = False), but we
won’t do that here.
The variable t 0 is now a row vector. To make it a column vector, we first
transform it into a two-dimensional array by including the array t 0 within square
brackets as the argument for t1 = np.array(), with one row only (shape =
(1, 801); axis 0 has one element, a list with 801 elements. We then transpose the
array between the two axes 1 and 0 by t1.transpose(1,0). The shape and the
first and last elements of the three vectors are reported in the second cell of Table
2.24. The column vector t has the same shape and contains the same numbers
as the variable t in A8:A808 of Fig. 2.23 (S). The same holds for the matrix c
Table 2.24 Second cell: data structure of the variables of the main program, printed with instruc-
tions similar to those in the first cell. Third cell: matrix of the y values of four cosines
print("t0,
1 shape:", np.shape(t0))
print(t0)
2
t0, shape: (801,) c, shape: (801, 4)
[ 0.00 0.04 ... 31.96 32.00] [[ 0.23 0.75 0.63 0.72]
[ 0.23 0.75 0.63 0.71]
t1, shape: (1, 801) ...
[[ 0.00 0.04 ... 31.96 32.00]] [ 0.20 0.35 -0.04 -0.41]
[ 0.19 0.29 -0.11 -0.50]]
t, shape: (801, 1)
[[ 0.00] sumC,shape: (801,)
[ 0.04] [ 2.33 2.31 ... 0.10 -
... 0.13]
[ 31.96]
[ 32.00]]
calculated as c = A*np.cos(w*t + phi) and for sumC reported in the

third cell of Table 2.24.
Question
Which parameters relevant in the main program in Table 2.23 have to be spec-
ified in the sub-programs executed immediately before the main program?17
To calculate the four cosines, we apply the same formula, an operation on

three row vectors and one column vector as in range B8:E808 in the spreadsheet
(Fig. 2.23 (S)).
excel: [range] = [=a*cos(w*t + phi)]
Python:c = A*np.cos(w*t + phi)
Both operations yield a 2D array, excel by dragging into a 2D range, Python
by automatic broadcasting.
You should have noticed that our spreadsheet layout has been translated line-
by-line into Python. The striking similarity between the two platforms when using
vector notation is the reason why mathematical-physical calculations can be sensi-
bly performed in excel. training such calculations is a good introductory exercise
for computational physics.
There is a difference between the spreadsheet and the Python solution
presented here:
17(1) Labels that contain information on the characteristic frequency and frequency range. (2)
Scalings of the y-axis. (3) Name under which the resulting figure is stored. (4) Parameters A, w,
phi.
– In excel, the four column ranges get the names c.1, c.2, c.3, c.4, with which
they can be called.
– In Python, all four curves are stored in the columns of the two-dimensional
list c and can be called by c[:,i] with i = 0, 1, 2, 3.
Summing up the four cosines is done in excel with the formula in F9 =

[=Sum(B9:E9)] of Fig. 2.23 (S) and in Python with np.sum(c,axis = 1).
The resulting variable sumC is a 1D vector. It is the sum across the columns (axis
= 1) for a specific value of t, and has the rows’ length. Attention: The choice of
axes does not seem intuitive, but becomes apparent when sumC is plotted versus
t; both have to be of the same length. For an overview, see Fig. 2.18.
Same data structure in Excel and Python

In spreadsheets, the formulas behind the values in the cells are usually hidden. To
report them in our figures, we have copied their text in italic into neighboring cells.
Contrastingly, in Python, the formulas are evident, but the data are hidden. To
make them visible, we print them out in a well-structured manner (see Table 2.24).
A comparison with Fig. 2.23 (S) shows that we have reached our goal to implement
the same data structure in excel and Python.
Harmonics
In Table 2.25, we specify the harmonics’ parameters in the same way as in the
spreadsheet, in the first cell for all multiples (of ω0 = 1) and in the second cell
for odd multiples of the lowest frequency ω0 = 0.5. The values for amplitude A,
circular frequency ω, and zero phase φ are specified as vectors, in Python realized
as lists (characterized by square [] brackets). From these parameters, we compose a
label lbl 0, later to be reported in the figure showing the result of our calculation.
Furthermore, we specify in FigName the name under which the corresponding chart
is to be stored. In lines 5 and 11, we specify the scale of the y-axis according to
Fig. 2.20.
Table 2.25 Specifications of the parameters for harmonics

1 A=[0.23,0.75,0.63,0.72]
2 w=[1,2,3,4] #Overtones, all
3 phi=[0,0,0,0]
4 lbl_0="ω0="+str(w[0])+", all"
5 ymin,ymax,dy=-2,3,1 #Scaling the y-axis
6 FigName="PhExI 7-4 Overtones all" #File name for later use
7 A=[0.23,0.75,0.63,0.72]
8 w=[0.5,1.5,2.5,3.5] #Overtones, odd
9 phi=[0,0,0,0]
10 lbl_0="ω0="+str(w[0])+" (odd)"
11 ymin,ymax,dy=-3,3,1
12 FigName="PhExI 7-4 Overtones odd" #In plt.savefig(FigName)
Table 2.26 Specifications of the parameters for beats

1 #Beats 1
2 A=[0.5,1.5,1.5,0.5]
3 Delta_w=1
4 dw=Delta_w/3
5 w1=2
6 w=[w1,w1+dw,w1+2*dw,w1+3*dw]
7 phi=[0,0,0,0]
8 lbl_0=("ω1="+str(w1)+"; Δω="+str(Delta_w)) #Label for curve
10 FigName="PhExI 2-7-4 Beats1" #File name in savefig()
11 #Beats 2
12 A=[0.5,1.5,1.5,0.5]
13 Delta_w=3
14 dw=Delta_w/3
15 w1=4
16 w=[w1,w1+dw,w1+2*dw,w1+3*dw]
17 phi=[0,0,0,0]
18 lbl_0=("ω1="+str(w1)+ "; Δω="+str(Delta_w))
20 FigName="PhExI 2-7-4 Beats2" #File name
Table 2.27 Specification of the parameters for the addition of cosines

1 #Addition of cosines
2 A=[0.5,0.5,0.5,0.5]
3 Delta_w=1
4 dw=Delta_w/3
5 w1=2
6 w=[w1,w1+dw,w1,w1+dw]
7 phi=[0,0,0,0]
8 lbl_0=("ω1="+str(w1)+"; dω="+str(np.round(dw,2)))
9 ymin,ymax,dy=-3,3,1 #Scaling of y axis
10 FigName="PhExI 2-7-4 Sum Cos" #File name
Beats and the addition of cosines

The parameters for the other situations are set in Table 2.26 (beats) and Table 2.27
(addition theorem of cosines).
2.7.5 Producing Labels (as Strings) in Excel and Python
When producing labels, the following differences between excel and Python
have to be taken into account:
excel (G4 in Fig. 2.23 (S)) A1 = [=”w0 & w.0 & “, all”]
The concatenation operator is & (ampersand), and numeric values are automat-
ically converted into a string.
Python (Table 2.25):lbl_0 = “ω0” + str(w[0]) + “, all”
The concatenation operator is + ; numeric values have to be converted explicitly
into a string. In both applications, text is enclosed in quotation marks.
When numbers x have to be rounded to n decimal places, we can use
round(x;n) in excel and np.round(x,n) in numpy.
2.8 Questions
Cell references
1. What does the broom rule Ψ The dollar makes it absolute tell us?
2. What formula must be written in cell B5 of Fig. 2.27 (S), with absolute and relative
references, so that copying this formula into the range B5:E205 creates four sine
functions?
3. (Python) Specify arrays A, ω, t, so that an instruction C = A*np.cos(ω*t)
generates the four cosines C a , C b , C c , C d of Fig. 2.27(S) in one matrix C. How
do you replace the # in F sum = np.#(C, axis = #) to get the sum of the four
cosines?
A B C D E F G H
1 1 2 3 4 Amplitude
2 4 3 2 1 Angular frequency
3
4 Ca Cb Cc Cd
5 0 =A*COS(w*t)
6 1
7 2
205 200
Fig. 2.27 (S) structure of a spreadsheet for displaying four cosine functions
Fig. 2.28 The sum of the

four cosines specified in
Fig. 2.27 (S)
A B C D E F
1
2
a
m
2.00
1.00 10.00 y=1x+2
3 y=1x+2 8.00
4 x y
6.00
5 1.00 3.00
6 2.00 4.00 4.00
7 3.00 5.00 2.00 y=1x+2
8 4.00 6.00
9 5.00 7.00 0.00
10 6.00 8.00 0.00 2.00 4.00 6.00 8.00
Fig. 2.29 a (left, S) A straight line is defined in the spreadsheet and displayed in the diagram. b
(middle) Standard chart of the data in a. c (right) The design/select data/edit dialog box,
which is used to insert the data series x, y from a into the diagram
4. (Python) Write a program that realizes the spreadsheet calculation in

Fig. 2.27 (S), using an instruction C = A*np.cos(w*t)! Calculate the sum
of the four cosines and plot them as shown in Fig. 2.28, with a statement
plt.plot(t,Ctot,’k-’)!
Spreadsheet function Indirect
5. Look at the entries in eight cells of a table: A1 = 5; B2 = K; E1 = 5; K5 = 7; W1

= “E"&1; W2 = indirect(W1); W3 = B2&A1; W4 = indirect(W3). Which
numbers appear in cells W2 and W4?
Arrays in Python
6. Array x is specified as x = np.arange (-8,8 + dx,dx) with dx = 1.

How many elements does x comprise and what are its first and last elements?
7. What are the elements of np.linspace (0,3,4)?
8. What is the shape of arrays U = np.array([1,2,3]), UR =
np.array([[1,2,3]]) and UT = UR.transpose(1,0)?
9. What is the shape of np.array([[1,2],[2,3],[3,4]])?
10. Let V = np.array([[1,2,3]] and W = np.array([3,2,1]).
What is the shape of U = V * W and V.transpose(1,0)?
Diagrams
In Fig. 2.29a (S), you see data series x and y, in Fig. 2.29b, the corresponding chart,
and in Fig. 2.29c, the dialog box with which the data series was inserted into this
diagram.
11. Which spreadsheet ranges contain series name, series x values, and series
y values?
12. How do you create the expression y = 1x + 2 in the spreadsheet, and how do
you insert it into the chart in Fig. 2.29b as a legend?
13. (Python) Below, you find a program for plotting a chart similar to the one
in Fig. 2.29a, additionally with labels ‘x’ for the horizontal and ‘y’ for the
vertical axis. Fill in the missing entries! The graph should be a straight line with
diamonds, all black.
2.8 Questions 71
A B C D E F
27 1000 5.00
A B C D E F
27 0 -5.00
Fig. 2.30 Two settings of a slider
FigStd( … )
x=[ … ]
y=[ … ]
plt.plot(x,y, … )
14. (Python) How do you produce the string y = 1x + 2 when a = 2.0023 and m
= 1.001 are specified?
15. (Python) How do you produce a string 3.0*exp(t/-30.0) when A = 3.001 and
t A = -30.0 are specified?
Sliders (scroll bars)

In Fig. 2.30, you can see two settings of a slider.
16. Which is the linked cell?

17. What are the minimum and maximum values of the slider?
18. The formula in F27 accesses cell D27. How does it look like?
19. The formula = (A5-500)/100 is used to generate decimal numbers between -5
and 5 recurring to a slider. What is the linked cell, and what are min and max
of this slider? What is the distance between two decimal numbers.
Polar coordinates
20. The coordinates of a circle are best given in polar coordinates with the angle φ
and the radius r. How do you get the cartesian coordinates x and y needed for
an xy diagram?
The figure in Fig. 2.31a is generated by the spreadsheet organization in Fig. 2.31b
(S). Cells C11 and E11 have the names shown to their left. The column area B14:B26
gets the name phi.
21. How big are the numbers dPhi and r K ?
Apply names for cell ranges, if defined, in the answers to the next three questions!
22. What formulas are in cells B15 and B26?

23. Which formula is in column C below x?
15
y
B C D E
10 11 dPhi 0.5236 rK 10
12
5 13 phi x y
14 0.00 10.00 0.00
0 15 0.52 8.66 5.00
-15 -10 -5 0 5 10
x
15 16 1.05 5.00 8.66
-5 17 1.57 0.00 10.00
24 5.24 5.00 -8.66
-10
25 5.76 8.66 -5.00
26 6.28 10.00 0.00
-15
Fig. 2.31 a (left) Representation of a circle with 12 line segments; b (right, S) Coordinates for the
circle in a
24. Which formula is in column D below y?

25. (Python) Complete the following program by replacing # to get a figure similar
to Fig. 2.31a:
rK=10
phi = np.arange(0,#,dPhi)
x=rK*#
y=rK*#
FigStd(#,#,#,#,#,#,#,#)
Cosine functions
Figures 2.32a and b display two cosine functions.
15 15
CosA CosB
10 10
y y
5 5
0 0
-10 0 10 20 t 30 -10 0 10 20 t 30
-5 -5
-10 -10
-15 -15
Fig. 2.32 a (left) Cosine function CosA . b (right) Cosine function CosB
2.8 Questions 73
26. What are the amplitudes and cycle times of the functions CosA and CosB shown
in Fig. 2.32?
27. What are the angular frequencies of the two functions shown in Fig. 2.32?
28. What are the overtones to the fundamental with the frequency f = 100 Hz?
29. How do you interpret the broom rule: Ψ Cos plus Cos = mean value times half
the difference?
30. A second cosine function is added to a cosine function with f = 100 Hz. What
frequency must the second cosine function have to produce a beat of 1 Hz?
Formula Networks and Linked
Diagrams 3
In this chapter, we practice clearly-structured calculations with formulas. Our

aim for spreadsheets:—formulas in cells should be similar to mathematical
formulas, a feat that is achieved by naming variables, calling them by their
name, and clearly separating independent and dependent variables. As in
every chapter, parallel solutions in Python are presented. For both excel
and Python, we aim to document intermediate results step by step from
top to bottom and accompany them with charts. This way, the results are
checked during the implementation of the formulas, and the calculation is
easy to understand, even weeks later. With these rules in mind, we treat
image construction for lenses, the Doppler effect, and exponential growth.
3.1 Introduction: Well-Structured Sheets and Programs
Solutions of Exercises 3.2 (Excel), 3.3.1 (Python), 3.3.5 (Python), and 3.4 (Excel)
can be found at the internet address: go.sn.pub/McoItP.
Physical tasks with networks of formulas

Many tasks in secondary education, or in physics courses at colleges, or even in
regard to minor subjects at universities, are based on simple formulas used to solve
practical tasks. In this chapter, we solve such tasks through spreadsheet calculations
and Python programs:
– image constructions for optical lenses,

– the Doppler effect with a general formula and for an observer off a race track, and
– exponential diode characteristics.

76 3 Formula Networks and Linked Diagrams
3
A B C D E F G
y [m]
1 Designation Name Value Unit x y
2
2 Width w 2 m -1 -1.5
3 Height h 3 m -1 1.5
1 4 1 1.5
5 Area A 6 m² =h*w 1 -1.5
0 6 -1 -1.5
-3 -2 -1 0 1 2 3
-1 x [m]
-2
-3
Fig. 3.1 a (left) A rectangle around the origin of the coordinate system b (right, S) Calculation of
the area of a rectangle; Independent variables in B2:C3; coordinates x, y for the representation in
a are calculated in columns F and G. b (right) Spreadsheet solution; coordinates for the rectangle
shown in a, symmetrical to the origin of the coordinate system
For this purpose, a system of formulas has to be built using the results of other
formulas. The spreadsheet set-ups and the Python programs should reflect the line
of thought and be easily traceable, even weeks later.
Solve step by step and write it down in mathematical language!

The student shall learn to keep an overview by developing the solution step by step
and displaying the results graphically, e.g., as optical ray construction. The diagrams
should adapt automatically to any change in the parameters of the task.
Illustration: Draw a rectangle!

We consider a simple task: Specifying the width w and height h of a rectangle,
calculating its area A, and drawing the rectangle as a chain of straight lines with
vertices (x, y), with its center at the origin of the coordinate system (see Fig. 3.1a).
Python program
In Python, every object has to have an identifier (a name) so that, from the outset, the
implementation is similar to mathematical formulas. For Fig. 3.1a, we apply Table
3.1.
Formula network in spreadsheets

In spreadsheet calculations, we have to pay special attention in order to achieve our
goal. It is necessary to call all variables by names such that the cell formulas can
be written like mathematical formulas. It is often useful to present the calculation
in four columns: designation of the quantity, variable name as an identifier for cells,
numerical value, and physical unit. An example is given in Fig. 3.1b. All independent
variables are in a block at the top left of the sheet, with names in B2:B5 and units in
D2:D5.
If one of the independent variables is changed, the entire calculation and all
diagrams should follow, without corrections having to be made somewhere in the
3.1 Introduction: Well-Structured Sheets and Programs 77
Table 3.1 Specifying the coordinates of a rectangle around (0, 0)

1 w=2 #[m] Width
2 h=3 #[m] Height
3 A=h*w #[m²] Area
4 x=[-w/2, -w/2, +w/2, +w/2, +w/2]
5 y=[-h/2, +h/2, +h/2, -h/2, -h/2]
sheet. We call a spreadsheet structure corresponding to these requirements a ‘formula

network’. It is often useful to change independent variables with sliders that can
quickly be used to get an impression of the trend of the solution (not realized in
Fig. 3.1b).
The area is calculated as F = h * w = height by width. If you write “= h * b”
into cell C5, C5 = [= h * w], then immediately “6” appears after you hit Enter.
In columns F and G of Fig. 3.1b, the coordinates for the graphical representation in
Fig. 3.1b are calculated. The formulas are the same as for x and y in Table 3.1. Cells
and cell areas must be named, e.g., F2:F6 with “x”. A good formula network must
not only be correct, but also clear!
Provide cells in spreadsheets with names

There are several ways to name cell areas. We already became acquainted with this in
the last chapter in Exercise 2.1 and Sect. 2.3.5. You can also find out more about this
in the excel help under the keyword “Create a name”. We prefer to use the variant
formulas/defined names/create from selection, which has the advantage
that the names given in the name manager are visible in the spreadsheet, and thus
contribute to the clarity of the calculation.
Check the solution with diagrams

For the representation of the rectangle in a diagram, here, Fig. 3.1a, the coordinates
x, y are calculated from the values for width w and height h. The drawings should
adapt automatically to any change in the parameters of the task. Therefore, in the
spreadsheet of Fig. 3.1b, columns F and G contain formulas and not just numbers.
The entries for w and h control all computations in the sheet and the figure, giving
the impression of a “living spreadsheet”.
Experience has shown that it is especially appealing to many students to work on
the tasks until the diagram “obeys on command”. In the case of more complex tasks,
it is also easier to see whether the formula is correct.
Question
What are the formulas for x and y in Fig. 3.1b (S)?1
1 x = +w/ 2 or x = –w/ 2; y = +h/ 2 or y = –h/ 2, always correctly combined.

Mathematical functions
In this chapter, we apply four mathematical functions:
– straight lines for image constructions of geometrical optics,

– exponentials for diode characteristics,
– polar coordinates with sine and cosine.
3.2 Image Construction for Focusing and Diverging Lenses
We construct the image point that a lens generates from an object point with
three characteristic rays, applying the general imaging equation valid for both
a focusing lens and a diverging lens. Lens equation with plus and minus.
We draw the bundle of rays through the lens, which actually contributes to
the image point. After this exercise, the reader should be able to master the
straight-line equation blindfolded.
3.2.1 Straight Line Equation
In this exercise, a straight line is defined by two points (x 1 , y1 ) and (x 2 , y2 ). For a

given third coordinate x 3 , a coordinate y3 is to be calculated so that the point (x 3 ,
y3 ) lies on the straight line.
Straight line equation

Given two points (x 1, y1 ) and (x 2 , y2 ) of a straight line, the straight-line equation is
y(x) = y1 + m · (x − x1 ) (3.1)
or
y(x) = y2 + m · (x − x2 ) (3.2)
both times with the slope
y2 − y1
m= (3.3)
x2 − x1
We call the respective points (x 1 , y1 ) in the first straight-line equation and (x 2 , y2 )

in the second the reference points of the straight line.
3.2 Image Construction for Focusing and Diverging Lenses 79
A B C D E F 6
1 x.1 -1.91 209
2 y.1 3.00 y.1; y.2 y.1; y.2
3 4 y.3
4 x.2 2.00 Straight line
5 y.2 -2 200 2
6
7 m -1.28 =(y.2-y.1)/(x.2-x.1)
8 x.3 1.63 563 0
9 y.3 -1.53 =y.2+m*(x.3-x.2) -4 -2 0 2 4
10
11 Straight line -2
12 -10 13.35 =y.2+m*(B12-x.2)
13 10 -12.23 =y.3+m*(B13-x.3)
-4
Fig. 3.2 a (left, S) Spreadsheet layout for the diagram in b; the values of x 1 and y1 are obtained
by means of sliders. b (right) The point “y.3” is to lie on a straight line given by the two points
“y1; y2”
We demonstrate Eq. 3.1 through the spreadsheet in Fig. 3.2a. The y-value y1 of
the first and the x-value x 2 of the second defining point are directly written into
cells B2 and B4, respectively. The associated values x 1 and y2 in B1 and B5 are
determined using the two sliders in D1:F1 and D5:F5. From these coordinates, the
slope m is calculated in B7 with Eq. 3.3. The x value x 3 of the third point in B8
is selected with the slider in D8:F8, and the corresponding y-value y3 in B9 is
obtained with the straight-line equation Eq. 3.2. These three points are represented
in Fig. 3.2b with diamonds.
In range B12:C13 of Fig. 3.2a (S), the straight-line coordinates are calculated
for x values −10 to 10 extending beyond the range of the x-axis in Fig. 3.2b so
that the straight line goes through the whole picture. The straight line is entered
into the figure with series x- values: (B12:B13), series y- values: (C12:C13).
Questions
Questions concerning Fig. 3.2a (S):

What are the linked cells for the three sliders?2
Which number range (the same for all) is presumably covered by the
sliders?3
If the coordinates, to be set by sliders with min = 0 and max = 800, take
on values between −4 and 4, what are the formulas in cells B1, B5, and B8,
with which the coordinates are calculated from the cells linked to the sliders?4
2 The linked cells are C1, C5, and C8.

3 The number range of all sliders runs from 0 to 800.
4 B1 = [=(C1 − 400)/100]; B5 = [=(C5 − 400)/100]; B8 = [=(C8 − 400)/100].
10
0
-20 -16 -12 -8 -4 0 4 8 12 16 20
Focal points
Rays -5
Object arrow
Image arrow
-10
Fig. 3.3 Image construction for a focusing lens with parallel, central and focus rays
In C12 and C13, the y coordinates of the straight line are calculated with
two different formulas. Why do both formulas describe the same straight line?5
If we have done everything correctly, the graphic presentation in Fig. 3.2a

should adapt to every change of the values for y1 , x 2 , x3 by the three sliders,
and the three points should lie on a straight line every time.
3.2.2 Geometrical Image Construction for a Thin Focusing Lens
Image construction by ray drawing

Figure 3.3 illustrates how the image of an object point is constructed with three
characteristic rays in the xy-plane.
The x-axis represents the optical axis and also the axis of the circularly shaped
lens. A thin lens is represented by its principal plane and its focal length. The shape
of the lens does not play a role anymore. The principal plane is the plane x = 0 in
which the y-axis is situated.
Question
How does the parallel ray in an image construction run?6
5 The slopes are the same for both straight lines. As reference points, (x 2 , y2 ) has been chosen for
C12 and (x 3 , y3 ) for C13. As both lie on the straight line, the third point also lies on the same line.
6 The parallel ray runs from the object point parallel to the x-axis (that is, the optical axis) up to x
= 0 and then through the image-side focus.

10 10
5 5
0 0
-10 -5 0 5 10 -10 -5 0 5 10
Focal Focal
points -5 points -5
Rays Rays
-10 -10
Fig. 3.4 a Converging lens, a (left) real inverted image for an object distance outside double
the focal length. b (right) Virtual, upright image for an object distance within the focal length;
corresponds to a look through a magnifying glass from the right
Mag Do you remember how to determine the image point of an object point
geometrically?
Alac Yes, as in Fig. 3.4a. We draw two rays starting from the object point, one
through the center of the lens and another parallel to the optical axis up to the
principal plane and then through the focal point on the right side of the lens. The
image point is where the two rays intersect.
Mag So you can do it. Your construction is valid for a converging lens. How-
ever, a lens has two focal points, one on the image-side and another one on the
object-side. In the image construction just described, you have exploited the fact
that all rays incident parallel to the optical axis go through the image-side focus
after having passed through the lens.
Tim We have often drawn a third ray from the object point through the focal
point on the left of the lens, which imagine is called the object-side focal point.
After passing through the lens, this ray is parallel to the optical axis and then
passes through the image point.
Mag Yes, all three construction rays intersect at the image point, as in Fig. 3.4.
That’s what we want to reproduce with our exercise.
In Fig. 3.4, the image construction for a converging lens using the principal rays
(parallel, central, and focal) is represented in a Cartesian coordinate system. By
convention, the optical axis is the x-axis. The optical center and the lens’s principal
plane are respectively located in the origin of the coordinate system in the plane
x = 0. We take it as given that the object is always to the left of the lens, i.e., the
object distance is always negative.
We know (from physics courses) that an inverted real image is formed to the
right of the lens when the object distance is bigger in magnitude than the focal
length (Fig. 3.4a). If the object distance is smaller in magnitude than the focal
length, the result is an erect virtual image to the left of the lens (Fig. 3.4b). The
general imaging equation for optical lenses considers these relationships by sign
conventions for the variables that enter the imaging equation, Eq. 3.4.
Lens equation with plus and minus!
3.2.3 Imaging Equation with Correct Signs
Mag Do you know the imaging equation for lenses?
Alac Sure, I’ve already learned it at school:
1 1 1
+ = (3.4)
xO xI f
where x O and x I are the object and the image distance, respectively, and f is the
focal length.
Mag This equation is useful only for handmade geometric constructions. For
an analytical calculation, we must use a more accurate one, namely, Eq. 3.5, in
which two modifications with respect to Eq. 3.4 have been introduced. Now, f I
is the image-side focal length, x I the image distance, and x O the object distance.
The object distance is, in principle, negative, because the object is, by convention,
placed to the left of the lens.
Alac With the old equation, we always got the correct values for image distance
and image size.
Mag Yes, the absolute values are calculated correctly. However, no signs, plus
or minus, are considered. Let’s adopt the more general notation. The object dis-
tance x o is negative if the object is to the left of the lens. The image is often upside
down. This is automatically considered in Eq. 3.6, which calculates the image size
yI from the object size yO .
Furthermore, Eqs. 3.5 and 3.6 are also valid for a diverging lens if a negative
image-side focal length is introduced, f I < 0.
General imaging equation for lenses

For the analytical computation of images of lenses, the imaging equation has
to be written with signs:
1 1 1
− + = (3.5)
xO xI fI
The x-axis is the optical axis. The principal plane of the lens is in the
plane x = 0; the object distance x O is negative. The image-side focal length
f I is positive for converging lenses and negative for diverging lenses. The
image distance x I may result positive or negative. The imaging scale is
yI xI
= (3.6)
yO xO
with yO being the object size and yI the image size that can be positive
or negative.
Converging lens
In Fig. 3.4, you see the usual image construction for a focusing lens (f I > 0) employing
parallel, center, and focus rays.
Diverging lens
For the image construction of a diverging lens, you can use the same spreadsheet
calculation or Python program as for a converging lens. You only have to enter
a negative image-side focal length f I . Figure 3.5 shows two examples (obtained
from Table 3.3).
3.2.4 Beam Through a Converging Lens that Really Contributes

to the Image
Mag In your geometric ray constructions, you did not draw the cross-section of
the lens. What can you say about the lens?
Alac In all cases, the center is thicker than the edge; otherwise, it would not be
a converging lens.
Tim The principal plane of the lens is located in the plane x = 0. The center of
the lens is at the origin of the coordinate system.
Mag How big should the diameter of the lens be, e.g., in Fig. 3.4?
Fig. 3.5 Imaging with a diverging lens, the geometric construction being the same as for a con-
verging lens, but with negative focal length (f I < 0) (drawings obtained with the Python program
in Table 3.4). a (left) upright image for an object distance larger than the focal length. b (right) As
with a, but for an object distance smaller than the focal length
Table 3.2 Specifications for an image construction with a converging lens, resulting in a figure
similar to Fig. 3.3
6 #Converging lens
7 fI=6.0
8 xO=-15
9 yO=6
10 xMin, xMax, Dx = -20, 20, 5 #Scaling of figure axes
11 yMin, yMax, Dy = -10, 10, 2.5
12 FigName='Converging lens' #File name in plt.savefig()
Table 3.3 Specifications for image constructions with a diverging lens, resulting in Fig. 3.5
1 #Diverging lens, Object beyond focal length
2 fI=-6
3 xO=-10
4 yO=12
5 xMin, xMax, Dx = -12, 12, 4
6 yMin, yMax, Dy = -8, 16, 4
7 FigName='Diverging lens, outside' #In plt.savefig()
8 #Diverging lens, Object within focal length
9 fI=-6
10 xO=-5
11 yO=12
12 xMin, xMax, Dx = -12, 12, 4
13 yMin, yMax, Dy = -8, 16, 4
14 FigName='Diverging lens, inside' #In plt.savefig()
Alac I would draw the lens from y = −7 to y = 7 so that the three constructing
rays pass through the lens. The diameter would be roughly the same as those of
the lenses used for lecture experiments.
Mag Be cautious; think of cameras! In that case, the lens diameter is much
smaller than, for example, the elephant you are photographing.
Tim That’s right. But does it mean that the construction rays do not go through
the lens?
Mag They don’t, indeed. They exist only in thought and on paper. Which rays
actually do contribute to the image point for a camera lens?
Alac Only the central ray, or perhaps other rays that really do pass through the
lens.
Mag Yes, the image point is formed by a bundle of rays through the lens, as
we will draw now. The lens itself has not shown up in the figures presented so
far. The size of the lens is irrelevant to the image construction; only the principal
plane and the focal length are needed to construct the image point.
Figure 3.6 shows the ray construction of the image point, together with the
cross-section of the lens and eleven rays going from the object point through posi-
tions in the lens’s full span and finally focusing in the image point. The parallel
ray and the focus ray run outside the lens. They do not exist in physical reality.
Fig. 3.6 (By program in Tables 3.4 and 3.5) Light beam contributing to the image formation in
Fig. 3.3
Cross-section of the lens

We are going to add the cross-section of the lens to the drawings. To do so, we need
to construct segments of a circle. We make use of the circle equation:
x 2 + y2 = r 2 (3.7)
The parameters are the coordinates x 0 of the center point on the x-axis, the radius
the lens’s curvature, and the lens diameter DL = 2r L (in
r K of the circle determining
front view). With x0 = r Lens
2 − r L2 , we calculate the distance of the center of the
sphere, limiting the surface of the lens to the origin of the drawing. With

x= 2
r Lens − y 2 − x0 , (3.8)
we get the x-coordinate for a given y-coordinate on the surface of the lens.
fI image-side focal length

(x O , yO ) coordinates of the object point
(x I , yI ) coordinates of the image point, to be calculated with the imaging
equation.
Three rays for the geometrical construction of the image, without prior knowledge
of the coordinates of the image point, are defined by f I and the object point from
which characteristic slopes have to be calculated.
x Par , yPar 3 characteristic points of the parallel ray

x Cen , yCen 3 characteristic points of the central ray
x Foc , yFoc 3 characteristic points of the object-side focal ray
mPar , mCen , mFoc slopes of the non-horizontal parts of the three rays
x Lens , yLens cross-section of a converging lens, calculated with Eq. 3.8.
Imaging equations
We calculate the ray path coordinates for imaging with a converging lens, and there-
with set up an image construction that should adapt automatically whenever the
parameters are changed. The coordinates of the three constructing rays’ defining
points are shown in Fig. 3.7 (S).
A B C D E F G H I J K L M
1 Specifications Rays
2 Image-side focal length fI 6.0 Central ray
3 Object distance xO -15.0 slope -15 =xO 6 =yO
4 Object height yO 6.0 0 =0 0 =0
5 Imaging equation mCen -0.4 20 -8 =mCen*J5
6 Image distance xI 10.0 =(1/fI+1/xO)^-1 =yO/xO Parallel ray
7 Image height yI -4.00 =yO*xI/xO -15 =xO 6 =yO
8 0 =0 6 =yO
9 Focal points 6 =fI 0 mPar -1 20 -14 =yO+mPar*J9
10 -6 =-fI 0 =-yO/fI Focal ray
11 Object arrow -15 =xO 0 -15 =xO 6 =yO
12 -15 =xO 6 =yO -6 =-fI 0 =0
13 Image arrow 10 =xI 0 =0 mFoc -0.67 0 -4 =yO+mFoc*-xO
14 10 =xI -4 =yI =yO/(xO+fI) 20 -4 =L13
Fig. 3.7 (S) Imaging equation for a converging lens; the quantities related to the object and the
image are designated with the indices O and I. x = 20 in column J indicates the right border of
Fig. 3.3
The five parameters (focal length, object distance, and height, as well as the
image distance and height that are dependent on them) are provided with the names
in column B, with which they are entered into the image equations (B6:C7) and the
coordinates of the construction rays (J:M).
The coordinates of the focal, central, and parallel rays are introduced as data series
into the diagrams of Fig. 3.4a, b. We may enter the column range J3:J14 as series
x values and L3:L14 as series y values to get three separate straight lines,
because empty rows separate their coordinates. The designations Central ray, etc.,
are in column K, not in column J or column L.
Ψ Empty rows separate curves.
Note that, in most cells, there are formulas. So, you cannot simply copy the
numbers from the spreadsheets displayed in this text. If you have done the
implementation correctly, images such as those in Fig. 3.4a, b should result, auto-
matically adapting whenever you change the parameters of focal length, object
distance, and object height.
Alac A fascinating experience!

The parameters in Fig. 3.7 (S) are for a converging lens (f I = 6.0) and an object
distance (x O = −15) beyond the focal length (Fig. 3.3). With f I = 3.0, x O = −9,
and yO = 7, Fig. 3.4a results. With f I = 3.0, x O = −2 (within the focal length),
and yO = 3, Fig. 3.4b results.
For a diverging lens (Fig. 3.5), we choose f I = −6 and x O = −10 (a) or x O =
−5 (b).
Fig. 3.8 (P) Drawing rays from the object point through the lens to the image point; the coordi-
nates of the object point and the image point are read from column C of a spreadsheet, e.g., Fig. 3.7
(S). Dy is the distance between rays at x = 0
Ray bundle through the lens

We use a VBA subroutine7 as in Fig. 3.8 (P) to draw the ray bundle that physically
contributes to the image. The rays run from the object point (x O , yO ) to a point in the
lens, and finally to the image point (x I , yI ). The coordinates of the points are in C3:C7
of a spreadsheet; they are read in the first lines of the sub-routine. The parameters
of the lens are specified within the subroutine. The defining points of the rays are
written into columns 6 (c2 , F) and 7 (c2 + 1, G) of the spreadsheet.
Specifications for three different types of images

The Python program is organized into four cells. The first three cells contain the
specifications and a filename to store the resulting figure, each for
– a converging lens (Table 3.2, resulting in a figure similar to Fig. 3.3),

– a diverging lens with the object outside the focal length (Table 3.3 top, resulting
in Fig. 3.5a),
– a diverging lens with the object inside the focal length (Table 3.3 bottom, resulting
in Fig. 3.5b).
The fourth cell, represented in Tables 3.4 and 3.5, draws arrows representing
the image and the object, together with the image construction, with rays using
the specifications of one of the three initial cells that were run earlier.
The specifications in the three cells comprise not only focal length and the
coordinates of the object point, but also parameters for axis scaling of the image
and the name of the file wherein the image is to be stored. When the program in
one of these cells is run, the resulting parameters are valid for the following image
construction in Fig. 3.4. So, each of the three situations can be the basis of an
image.
7 VBA macros are introduced in Chap. 4.

Table 3.4 Drawing object and image arrows, function ArrowP presented at the end of this section
1 FigStd('x',xMin,xMax,Dx,'y',yMin,yMax,Dy,xlength=8)
2 plt.plot((-fI,fI),(0,0),'ko',
markersize=4,label="$f_I=$"+str(fI))
3 ArrowP((xO,0),(xO,yO),lw=1.5) #Object
4 lbl_1=r'$x_O$='+str(xO)+r', $y_O$='+str(yO)
#’$x_O$=’ becomes xO=
5 plt.text(xO,yO+0.5,r"($x_O$,$y_O$)",fontsize=10)
6 #Calculated image
7 xI=1.0/(1.0/fI+1.0/xO)
8 yI=yO*xI/xO
9 Arrow((xI,0),(xI,yI),lw=1.5,ls='--') #Image
10 lbl_2=(r'$x_I$='+str(round(xI,2))
11 +r', $y_I$='+str(round(yI,2))) #$x_I$ as xI in legend
12 plt.text(xI+0.5,yI+0.5,"($x_I$,$y_I$)",fontsize=10)
Table 3.5 Continuation of Table 3.4; setting up the image construction with the parameters spec-
ified in other cells without explicitly referring to the image point
13 #Parallel ray
14 xPar=[xO,0,1.5*xI]
15 mPar=-yO/fI #Slope in image space
16 yPar=[yO,yO,yO+mPar*xPar[2]]
17 plt.plot(xPar,yPar,ls='-',color='k',
18 lw=1,label=lbl_1)
19 #Central ray
20 xCen=[xO,0,1.5*xI]
21 mCen=yO/xO #Slope in whole space
22 yCen=[yO,0,mCen*xCen[2]]
23 plt.plot(xCen,yCen,ls='-',
24 color='k',lw=1.,label=lbl_2)
25 #Ray through object-side focus
26 xFoc=[xO,-fI,0,1.5*xI]
27 mFoc=-yO/(xO+fI) #Slope in object space
28 yg0=yO+mFoc*xO
29 yFoc=[yO,0,yg0,yg0]
30 plt.plot(xFoc,yFoc,ls='-', color='k',lw=1.)
31 plt.legend(loc=4,fontsize=10) #loc= 4 ,”Lower right”
32 plt.axis('scaled')
33 plt.savefig(FigName)
General program for drawing the image construction

The main program is in Table 3.4; it performs the image construction according to
the specifications in the cell executed earlier. Arrows representing the object and the
image are drawn. The coordinates of the object follow directly from the specifications
Table 3.6 User-defined function for drawing an arrow from point P0 to point P1 in the xy-plane
(construct explained in Chap. 4)
1 def Arrow(P0,P1,c="k",ls='-',lw=1,hw=0.4):
2 (x0,y0)=P0
3 (x1,y1)=P1
4 #c has to be given as c="k", not c='k'
5 plt.arrow(x0,y0,x1-x0,y1-y0,
6 length_includes_head=True,
7 head_width=hw,fill=False,
8 linestyle=ls, color=c,linewidth=lw)
in Table 3.2 or Table 3.3, whereas those of the image have to be calculated with the
image equation (lines 7 and 8).
The program reproduced in Tables 3.4 and 3.5 calculates with the values
obtained in one of the three cells in Tables 3.2 and 3.3. The ray constructions
in Fig. 3.1 are obtained with the specifications in Table 3.5. The arrows are drawn
with a function reproduced in Table 3.6 at the end of this section.
The three characteristic rays are drawn with the Python program in Table 3.5.
Plotting an arrow in Python

A user-defined function for drawing an arrow from point P0 to point P1 is shown in
Table 3.6.
The arrow is plotted from P0 to P1 , with color c, linewidth lw, and headwidth
hw entered as keyword arguments. If the function call does not specify the values
of the keywords, the default values specified in the header are taken. The coor-
dinates of the two points have to be translated into the expected entries of the
function plt.arrow of the MatPlotLib library.
Cross-section of the lens

The program for drawing the cross-section of a lens is given in Table 3.7.
The arrays y and x specify only one-quarter of the cross-section (see the values
in the bottom cell). They run from (2.0, 0.0) up to (0.0, 6.0). For a complete
cross-section, we need three more curves. They are obtained with the help of the
functions
np.flipud (“flip up down”) reversing the order of the elements in an array,
and
np.hstack concatenating arrays to one long array, x Lens resp. yLens .
The concatenated arrays x Lens and yLens are used for the drawing performed in
Table 3.8, line 21. The curve begins at (0.0, 6.0), runs to (2.0,0.0), continues to
(0.0, −6.0), then to (−2.0, 0.0), and closes the cross-section by running to (0.0,
6.0).
Table 3.7 Coordinates of the cross-section of a lens with radius r L of the disk and radius r O of
curvature of the surface of the lens
1 rL=6
2 rO = 10
3 xO=np.sqrt(rO**2-rL**2)
4 y=np.linspace(0,rL,3)
5 x=np.sqrt(rO**2-y**2)-xO
6 xf=np.flipud(x) #First becomes last
7 yf=np.flipud(y)
8 FigStd('x',-20,20,5,'y',-10,10,2.5,xlength=8)
9 xLens=np.hstack([ xf, x,-xf,-x]) #One long array
10 yLens=np.hstack([ yf,-y,-yf, y])
x [ 2.00 1.54 0.00]
y [ 0.00 3.00 6.00]
Bundle of rays through the lens

The coordinates of the nR = 11 rays going through the lens are calculated in Table
3.8 with pre-specified coordinates of the object and the image point, e.g., in Table
3.4. The rays are drawn in a for-loop, starting with a y-coordinate at the bottom of
the lens and increasing it by y = 2·r L /(nR − 1). The cross-section of the lens is
drawn with the plot-statement in line 21, using the coordinates x Lens , yLens calculated
in Table 3.7. The complete drawing, i.e., image construction, lens, and ray bundle,
is shown in Fig. 3.6.
Table 3.8 Drawing a bundle of rays from the object point through the lens to the image point
11 #Bundle of rays through the lens
12 #Object point and image point are known.
13 nR=11 #Number of rays
14 rL=3 #Diameter of lens
15 x=np.zeros(3)
16 y=np.zeros(3)
17 (x[0], y[0])=(xO, yO) #Object point
18 (x[2], y[2])=(xI, yI) #Image point
19 x[1]=0
20 FigStd('x',-20,20,5,'y',-10,10,2.5,xlength=8)
21 plt.plot(xLens,yLens,'k')
22 Arrow((xO,0),(xO,yO),lw=1.5) #Object
23 Arrow((xI,0),(xI,yI),lw=1.5,ls='--') #Image
24 Dy=2*rL/(nR-1)
25 for i in range(nR): #Bundle of rays
26 y[1]=-rL+i*Dy #Position in lens
27 plt.plot(x,y,'k-',lw=0.5)
3.3 Doppler Effect
When a sound source (a “sender”) and a receiver move relative to air, the
receiver perceives a frequency that is different from the transmitted one. We
set up a formula for all cases of movements of the two agents on a straight
line. We determine the frequency trajectory recorded at a receiver off the
sender’s track.
3.3.1 A Formula for All Cases
When a sound source (in the following, designated as sender S) and a receiver R
are approaching or moving away from one another on a straight line, the receiver
perceives a frequency different from that emitted. In the following, we develop a
formula for the cases when sender and receiver move on the same straight line.
Formula for intuitive use (Doppler)

The relationship between frequencies f (frequency ratio) and speed v is given by
the following formula:
fR c ± vR
= (3.9)
fS c ∓ vS
The letters f, c, and v denote the frequency, the speed of sound, and the speed
(≥0) of the agents relative to air. The upper sign in the formula is valid when the
agents are approaching each other and the lower sign when they are moving apart.
Note, as a mnemonic, that, above the fraction bars, there are quantities with index R
and, below the fraction bars, quantities with index S, for both sides of the equation.
Doppler effect with plus and minus
It is best to consider which signs are to be used for every individual case. An
example: S → R →; the sender moves towards the receiver, and the frequency
increases (/(c − vS )); the receiver moves away from the sender (c − vR ), and the
frequency decreases; thus f R /f S = (c − vR )/(c − vS ).
Questions
How may Eq. 3.9 be simplified when the receiver is stationary and the sender
is approaching him?8
8 f R / f S = c/(c − v S ), the received frequency becomes higher.

3.3 Doppler Effect 93
What frequency does the receiver hear when he travels at the same speed as
the sender, (a) in front of and (b) behind the sender?9
Sender overtakes receiver

Let us apply Eq. 3.9 to the situation in which the sender and receiver both move in
the same direction, and the sender overtakes the receiver.
Before overtaking, the sender is approaching the receiver, thereby increasing the
received frequency (minus sign in the denominator). The receiver is moving away
from the sender, also reducing the received frequency (minus sign in the numerator):
fR c − |v R |
= (3.10)
fS c − |v S |
After overtaking, it is the other way around: The sender is moving away, the
receiver is approaching, and a plus sign must be inserted in both the numerator and
denominator:
fR c + |v R |
= (3.11)
fS c + |v S |
Remember: |vR | und |vS | are speeds (amount of the velocities).
Motion on a straight line, analytical formula (Doppler)

We formulate a general formula in which the signs are automatically correct:
fR c − v R · sgn(x R − x S )
= (3.12)
fS c − v S · sgn(x R − x S )
with x S , x R being the positions of the sender and the receiver, respectively, on the x-
axis. The mathematical function sgn (“signum”) is available as a spreadsheet function
sign and as np.sign in numpy. To be able to better compare the formula with the
previous calculations, we rewrite it with the speeds (amounts):
fR c − |v R | · sgn(v R ) · sgn(x R − x S )
= (3.13)
fS c − |v S | · sgn(v S ) · sgn(x R − x S )
Tim I could never develop a formula like that. I would always set the wrong
sign or change the correct order.
Mag Nor could I. I’ve been toying around with this, checking whether the
outcome corresponds to the intuitive formula Eq. 3.9 in ten different situations.
9 (a) f R / f S = (c−v R )/(c − v R ) = 1; (b) f R / f S = (c+v R )/(c + v R ) = 1; the received frequency

is, in both cases, equal to the sent frequency.
1 =(c.s-vR*s.vR*xRel)/(c.s-v.S*s.vS*xRel)
2 =SIGN(x.R-x.S)
3 fE/fS s.vR s.vS x.S x.R xRel
4 cs 340 m/s S--> R* 1.05 =(cs)/(cs-vS) 0 1 -1 1 1 1.05
5 vS 17 m/s R* S--> 0.95 =(cs)/(cs+vS) 0 1 1 -1 -1 0.95
6 vR 10 m/s S* R--> 0.97 =(cs-vR)/(cs) 1 0 -1 1 1 0.97
7 R--> S* 1.03 =(cs+vR)/(cs) 1 0 1 -1 -1 1.03
8 S--> R--> 1.02 =(cs-vR)/(cs-vS) 1 1 -1 1 1 1.02
9 R--> S--> 0.98 =(cs+vR)/(cs+vS) 1 1 1 -1 -1 0.98
10 S--> <--R 1.08 =(cs+vR)/(cs-vS) -1 1 -1 1 1 1.08
11 <--R S--> 0.92 =(cs-vR)/(cs+vS) -1 1 1 -1 -1 0.92
12 <--S R--> 0.92 =(cs-vR)/(cs+vS) 1 -1 -1 1 1 0.92
13 R--> <--S 1.08 =(cs+vR)/(cs-vS) 1 -1 1 -1 -1 1.08
Fig. 3.9 (S) Frequency ratio for ten cases, in column F, calculated with Eq. 3.9 and with individual
considerations for each case, in column M, calculated with the general formula Eq. 3.13
Alac Proof by trial and error? You can’t do that in math!
Mag With trial and error, proof is not possible, but serious mistakes can be
uncovered. You can later rigorously prove the formula.
Check the analytical formula in a spreadsheet

In Fig. 3.9 (S), ten situations are listed (D:E) in which sender and receiver move to
the left or the right or one of them is at rest, and it is also distinguished as to whether
the sender is to the left or the right of the receiver. In column F, the frequency ratio
is calculated according to Eq. 3.9, “for intuitive use”, considering which signs are to
be used for each formula. In column M, the analytical formula Eq. 3.13 is applied
relating to the values 1, 0, or −1 indicating the direction of the movements (s.vR,
s.vS) and the relative position (x.S, x.R), with respect to the zero of the straight line,
of receiver and sender. The quantity x Rel is defined as sign(x R − x S ).
Checking in Python
In the Python program of Table 3.9, a function Doppler is defined realizing Eq. 3.12,
taking the velocities vS and vR (with correct sign + or −), respectively, of the sender
and the receiver, together with a keyword argument pos as input and returning the
frequency ratio f R /f S . The string argument pos specifies the relative position of
sender and receiver, ‘SR’ indicating that the receiver is to the right and ‘RS’ to the
left of the sender. The same parameters as in Fig. 3.9 (S) are specified successively
in a list var1 that is passed to Doppler expecting three positional arguments; so, the
list var1 has to be unwrapped (*var1 in line 11). The results of Doppler are the same
as in Fig. 3.9 (S).
Unwrapping a list with *

The function Doppler defined in Table 3.9 expects three positional arguments. The
parameters of our lists are, however, specified in one object, the list var1. Calling
Doppler(var1) results in an error message: TypeError: Doppler() missing
1 required positional argument: ‘vR’. In the call of Doppler in line
11, list var1 has, therefore, been unwrapped (*var1) so that the elements of var1
are transferred and not the list as one object.
Table 3.9 A general formula for calculating the Doppler shift for sender and receiver moving on
the same straight line
1 def Doppler(vS,vR,pos = 'SR'):
2 if pos == 'SR': sgnX=1
3 if pos == 'RS': sgnX=-1
4 fR=c-vR*sgnX
5 fS=c-vS*sgnX
6 return fR/fS
7 c=340.0
8 vS=17.0
9 vR=10.0
10 var1=[vS,0.0,'SR']
11 print(var1,'{:5.2f}'.format(Doppler(*var1)))
[17.0, 0.0, 'SR'] 1.05 [17.0, 10.0, 'RS'] 0.98
[17.0, 0.0, 'RS'] 0.95 [17.0, -10.0, 'SR'] 1.08
[0.0, 10.0, 'SR'] 0.97 [17.0, -10.0, 'RS'] 0.92
[0.0, 10.0, 'RS'] 1.03 [-17.0, 10.0, 'SR'] 0.92
[17.0, 10.0, 'SR'] 1.02 [-17.0, 10.0, 'RS'] 1.08
3.3.2 A Sound Source Passes a Remote Receiver
Figure 3.10a illustrates the ride of a car on a straight road, on the line y = 0 from x
= −100 m to x = 100 m. At a distance of yR = 30 m off the road, a receiver is at
position (0, 30). The car constantly sends out a tone of 200 Hz. Which frequency
does the receiver perceive when the car passes by?
1.2
Track fR/fS
Connection to receiver vS=55.56m/s
1.1 ; yR=20m
y [m] yR=50m
30 yR=100m
1.0
0.9
0 0.8
-100 -50 0 50 100
-100 -50 0 50
x [m] x [m] 100
Fig. 3.10 a (left) An observer (receiver) at point (0; 30) hears a car (sender) passing on the x-axis
(in the figure compressed). The velocity along the current connection line (dashed) determines the
perceived frequency. b (right) The perceived frequency, relative to the frequency of the source,
when a sound source passes the receiver at different distances yR at speed vS = 55.56 m/s
Velocity on line connecting sender and receiver

The velocity along the current connection line (dashed) determines the perceived
frequency. The quantities vR (receiver) and vS (sender) in Eq. 3.9 are the velocity
components on the current connecting line between sender and receiver, i.e., on the
dashed lines in Fig. 3.10a.
The velocity must have the correct sign. The upper signs in Eq. 3.9 apply when
the car is approaching, and the lower signs when it is driving away. For the situation
described above, the sender’s velocity component in the direction of the stationary
receiver is to be determined by differentiating the distance with respect to time. This
yields a negative velocity when the car is left of x = 0 and a positive velocity when
is it right of x = 0, so that the frequency ratio
fR 1
= (3.14)
fS 1 + vcS
is valid for the whole track.

Frequency curves for the distances yR = 20, 50, and 100 m are displayed in
Fig. 3.10b. The closer the receiver is to the track, the stronger the frequency varies
with the position of the car. When the sender is at the receiver’s height, the received
frequency is identical to the sent frequency.
The interesting thing about the computation model in this exercise is that the time-
dependent distance of the sender to the receiver is to be calculated (with Pythagoras)
at each interval boundary. Then, the velocity component on the connecting line is
obtained through numerical differentiation. In this way, more complicated geometries
can also be treated, e.g., when the transmitter is moving at varying speed on a circular
path, or if transmitter and receiver are moving on different paths.
c speed of sound in air

vS velocity of sender in x direction
vR velocity of receiver in x direction
fS frequency of sender
fR frequency at receiver
xS position of sender
xR position of receiver
x array of 201 equidistant positions of a sender on the x-axis
t(x) points of time corresponding to x
yR distance of the receiver to the x-axis
dist current distance sender-receiver, dist = x 2 + y 2R .
v relative speed along the line connecting sender and receiver
A B C D E F G H I
1 200.00 km/h
2 Velocity of sender vS 55.56 m/s
3 Frequency of sender fS 200.00 Hz =B2&"="&ROUND(vS;2)&"m/s; "
4 Speed of sound c_ 340.00 m/s &B5&"="&yR&"m"
5 Distance of receiver yR 20.00 m vS=55.56m/s; yR=20m
2) 9)
/vS +yR^ /(t-C )
1 +(x-B9) QRT(x^2 dist-D9) S/(1+v/cs /fS B9)/2
7 =B9+ =C9 =S =( =f =f =(x+
8 x t dist v f fNorm xC
9 -100 0 101.98
10 -99 0.02 101.00 -54.47 238.15 1.19 -99.50
209 100 3.60 101.98 54.47 172.38 0.86 99.50
Fig. 3.11 (S) A sound source moves on the x-axis past an observer at a distance d R . The x position
in column B is the independent variable. From that, the time (column C), the distance source-
observer (column D), the velocity of the sender in the direction of the connecting line (E), and the
observed frequency (F, G) are calculated with Eq. 3.14. Attention: The time in column C depends
on the speed of the source! In column H, x C is the center of the intervals
3.3.4 Spreadsheet Calculation “Remote Receiver”
A possible calculation model is shown in Fig. 3.11 (S), where we have chosen the
x-coordinate of the car on the track as the independent variable, and time, distance,
and velocity along the connecting line as dependent variables. You could just as
easily choose time as the independent variable. The x in the formula C10 = [=
C9 + (x - B9)/vS] refers to B10, the entry of the column vector x in the same
row.
Mag The frequency ratio is calculated numerically in columns F and G. Over

which local coordinates do you plot the calculated frequencies? Perhaps over x in
column B?
Tim Well, since you asked it in that way, it probably isn’t. I remember this
much: We take the centers of the considered distance intervals because the speeds
were calculated with the (t, x) coordinates of the interval boundaries.
Mag Right! The centers of the intervals are calculated as x C in column H.
3.3.5 Python Program “Remote Receiver”
A Python program that solves the task is given in Table 3.10, the main program in
the upper cell, the function FreqLine for realizing Eq. 3.14 for the complete fre-
quency curve in the lower cell. List slicing is used to calculate the interval centers
x C in the main program, and the velocities v when differentiating the distance dist
in FreqLine.
Table 3.10 a (top) Specifications of the situation in which a receiver is at rest at a distance yR off
the sender’s track. b (bottom) Function for calculating the frequency curve when yR is given
1 vS=55.56 #Speed of sender
2 fS=200 #Frequency of sender
3 c=340 #Speed of sound
4
5 x=np.linspace(-100.0,100.0,201)
6 #Center xC of path segments
7 xC=(x[1:]+x[:-1])/2
8
9 #Time segments Dt
10 Dt=(x[1:]-x[:-1])/vS
11 #Time t
12 t=np.cumsum(Dt) #Integrates over dt
13 def FreqLine(yR):
14 dist=np.sqrt(x**2+yR**2)
15 v=((dist[1:]-dist[:-1])/Dt)
16 fRS=1/(1+v/c)
17 return fRS #Frequency ratio
Questions
concerning Table 3.10:

What are the arguments and the global parameters accessed in FreqLine?10
The local variable x C over which the received frequency is to be plotted
is calculated in the main program. Would it be more consistent to calculate
x C in FreqLine and return it together with the normalized frequency (def
FreqLine(x, yR) …. return xC,fRes)?11
Question
concerning Table 3.11

What does the keyword argument lw=1.5 in the plot function specify?12
What does the label in line 8 look like?13
How do we specify that the figure is saved with a resolution of 1200 dpi?14
The plot program in Table 3.11 calculates and displays the frequency curves
for the three distances yR = 20, 50, and 100 m.
10 FreqLine, argument: yR, global parameters x, dist, c, Dt.

11 Discuss!
12 Line width lw = 1.5 point.
13 The label is “yR = 20”.
14 plt.savefig (’Doppler off, multiple.png’, dpi=1200), dpi is a key
word argument.
3.4 Exponentials 99
Table 3.11 Plotting several frequency trajectories with the parameters specified in Table 3.10
1 FigStd('x',-100,100,25,'f/fQ',0.8,1.2,0.1)
2 plt.plot([-100, 100],[1,1],ls='-',color='k',lw=1)
3 #Horizontal through y=1
4 yR=20 #Distance to track
5 fRS=FreqLine(yR)
6 plt.plot(xC,fRS,'k-', lw=1.5,label='yR='+str(yR))
7 yR=50; fRS=FreqLine(yR)
8 plt.plot(xC,fRS,'k--',lw=1.5,label='yR='+str(yR))
9 yR=100; fRS=FreqLine(yR)
10 plt.plot(xC,fRS,'k-.',lw=1.5,label='yR='+str(yR))
11 plt.legend(loc=0,fontsize=10)
12 #plt.savefig('Doppler off, multiple.png')
3.4 Exponentials
For exponential functions, apply Plus 1 yields times e; plus 1 in the argu-
ment yields times e in the value. An exponential function A · exp(−t/t0 )
is best drawn by hand (Really? Yes, also by hand!), beginning with its tan-
gent at t = 0. Diode characteristics seemingly exhibit a “kink voltage” that
depends on the scaling of the y-axis; more generally, exponential functions
seem to explode.
3.4.1 Explosive Character of Exponentials
Rice grains on a chessboard

In a classic bet from far eastern literature, a clever chess player agreed with his
king that his winnings would be paid by placing a rice grain on the first square of
a chessboard and then doubling the number of grains repetitively on each of the
following 63 squares.
In Fig. 3.12a (S), the doubling of the number of rice grains is simulated with a
spreadsheet calculation. In A3:A66, the 64 squares of the checkerboard are numbered
from n = 0 to 63. In B3, a rice grain is placed on the first field (n = 0, B3 = [1]). In
the following cells, the preceding number is doubled until the huge number 263 =
9.22 × 1018 is reached in B66 (n = 63), on the 64th square.
Ψ Plus (in the argument) yields times (in the result) for exponentials.
For y = 2x , plus 1 (in the argument) yield times 2 (in the result).
A B C D E 1,E+19
1 =A3+1 =B3*2 3
=2^n =2^D
2 n y y =2^n 2^62.4 8,E+18 y =2^n
3 0 1 1 62.4 6.1E+18 2^62,4
y
4 1 2 2
5 2 4 4 6,E+18
6 3 8 8
7 4 16 16
4,E+18
8 5 32 32
9 6 64 64
66 63 9.22E+18 9.22E+18 2,E+18
67 64 1.84E+19 1.84E+19
1025 1022 4.5E+307 4.5E+307
1026 1023 9.0E+307 9.0E+307 0,E+00
1027 1024 #NUM! #NUM! 0 16 32 48 n 64 80
Fig. 3.12 a (left, S) Powers of 2, y = 2n , obtained by repetitive multiplication by 2 (column B)

and by potentiation (C: E). The formula in E1 refers to E3 b (right) Graphical representation of the
powers of 2 of a
The values are graphically displayed in Fig. 3.12b. You can see that the explosion
by a factor 1019 takes place on the last few squares. On squares 0–62, there are
263 –1 grains, on all squares together, 264 −1 grains.
Profit
Question
What would be the share of the internationally traded rice in 2016 if the loser
of the game (a wealthy medieval Sultan) had been able to deliver?15
What is the value of 10E3 in excel and 10e3 in Python? Be careful!16
Mag Can the winner satisfy his hunger with his win?
Alac Perhaps once.
Tim I’ve heard that exponential growth means explosion. So, maybe the winner
can live well on his heap of rice for a week.
Mag World rice production in 2015/16 was 470 million metric tons, but only
about 5% were traded on the world market. Unlike wheat, rice is consumed by
more than 95% of the population in the cultivating countries.
15 For 64 fields, the winner would have received 264 –1 = 18 × 1018 grains, corresponding to about
1018 g = 1012 tons of rice. This is the 2000-fold amount of the rice harvest 2015/16 of 470 million
tons.
16 10E3 = 10 × 103 = 104 = 10,000, the same with 10e3 in Python.
3.4 Exponentials 101
Tim I’ve counted. One kilogram of rice contains about 40,000 = 4 × 104 grains.
According to the rule of the game, more than 10,000 times the volume of one
year’s world trade of rice should pile up on the chessboard. Incredible!
Alac Crazy! A disaster! This cannot be true! What’s the catch?
Mag There is no catch. The catastrophe results from the rule plus 1 yields
times 2 governing the exponential 2n .
The power function y = 2n is calculated in column C of Fig. 3.12a (S). It

is entered as the worksheet formula [= 2ˆn] and yields the same results as the
repetitive multiplication by 2 in column B. The argument of the power function
does not have to be an integer. In E3, the value for the power in D3 (= 62.4) is
calculated and inserted as a filled diamond in Fig. 3.12b.
Maximum float number

We continue doubling the preceding number in the worksheet beyond n = 63 until
the application can no longer store the resulting number. As of row 66 in Fig. 3.12a
(S), the numbers are represented in exponential form, 9.22E+18+18 = 9.22 × 1018 .
The number 21024 can no longer be calculated in excel2019; see the error message
#num! in row 1027.
In Python 3, the int format can store arbitrarily large numbers. However, if
the number is to be calculated as float, the same limit holds as in Excel: 2**1023 gives
8.988e+307 and 2**1024 returns the OverflowError: int too large to
convert to float.
3.4.2 General Exponential Function
The power function can be generalized to y = ax , where a and x are real num-
bers. If a is Euler’s number e = 2.718, then the power function becomes the
known exponential function with the formula [= exp(…)] (excel) or np.exp()
(Python).
Exponential function with characteristic length

The normalized exponential function is usually written in mathematical textbooks
as follows:
f (x) = |a| · exp(ax) = |a| · eax (3.15)

1
The letter e denotes Euler’s number e = ∞ 0 n! = 2.718. Normalized means that
the integral from 0 to +∞ (for a < 0) or from 0 to −∞ (for a > 0) is 1. The integral
is dimensionless because the product of the units of dx and a is 1.
It is, however, often physically more sensible to write the exponential with a
characteristic x value x 0 :

1 x 1 x

f (x) = · exp = · e x0 (3.16)
x0 x0 x0
Thus, the unit of x 0 is equal to the unit of x, e.g., a length or a time, and has an
intuitive meaning: the tangent at x = 0 intersects the y-axis at the amplitude |1/x0 |,
and the x-axis at the characteristic length x 0 .
When the function is specified with an amplitude |xA0 | , its integral from 0 to,
respectively, +∞ (for a < 0) or −∞ (for a > 0), is A.
3.4.3 Representation in a Diagram
First, the tangent at t = 0!

Two exponential functions, together with their tangents at t = 0, are shown in
Fig. 3.13a. Here, the independent variable is time t.
Mag Do you now know how to draw an exponential function Ae ·exp(t/t 0 ) by

hand on a piece of paper?
Alac Sure! First, mark the amplitude Ae on the vertical axis and the charac-
teristic time t 0 on the horizontal and pass a straight line through the two points.
The exponential curve approaches the tangent at t = 0 and the horizontal axis for
t → ∞ or t → –∞, depending on the characteristic parameter’s sign.
3 3·exp(t/-30)
y 1·exp(t/15)
2
tangents
0
-20 0 20 40 60 80
t
Fig. 3.13 a (left) Two exponential functions Ae ·exp(t/t 0 ) with their tangents at the intersections
with the y-axis. b (right) How to draw an exponential function by hand: First, the tangent as a
straight line with its intersections with the x- and y-axes!
Mag Correct, just as in Fig. 3.13b! Keep in mind:
Plus one in x, times e in y

Expo with kink and straight line
Tim “Plus one” and “Straight line” are clear, but why “kink”?
Mag You have seen this in Fig. 3.12b; it will be explained in Sect. 3.4.4. An
essential feature of an exponential is an explosion on a suitably scaled y-axis.
3.4.4 Diode Characteristics I(U)
The current I through a semiconductor diode depends exponentially on the applied

voltage U. The I (U) characteristics of a semiconductor diode are described by an
exponential function passing through zero:

U
I = Is · exp −1 (3.17)
UT
This function has two parameters: the strength of the reverse current I s also
called the saturation current, and the thermal voltage U T , which is given by k B T/e,
with k B being the Boltzmann constant, e the elementary charge, and T the absolute
temperature. At room temperature, U T = 25 mV. The current through a diode is
zero when the applied voltage is zero.
We will represent such a function for I s = 1 × 10–14 A and U T = 0.025 V in
various plots (see Fig. 3.14).
Fig. 3.14 a (left) Diode characteristics and associated exponential function, representation for
small currents. b (right) Twice the same exponential as in a, but with different scaling of the I axes;
left y-axis for the left curve, right y-axis for the right curve; the vertical grid lines have a distance
U T = 25 mV
Questions
Which of the curves in Fig. 3.14a, b are exponential functions? Which are
shifted on the I-axis? Which are shifted on the U-axis?17
At which U-values in Fig. 3.14b are you most likely to find the “kink points”,
in electronic engineers’ jargon?18
Also shown in Fig. 3.14a is an exponential function without the term −1 in

the parentheses of Eq. 3.17. This function intersects the I-axis at I s . It grows by
a factor of exp(1) ≈ 2.7 whenever U progresses by dU. The curves in Fig. 3.14a
appear to be exponential, the way they are usually represented.
For exponential functions to the base e, the broom rule is Plus 1 yields times
e. In the concrete case, this means that, if U advances by U T (the distance between
the vertical gridlines in Fig. 3.14), exp increases by a factor of about 2.7.
In Fig. 3.14b, the same data as in Fig. 3.14a are shown twice, only with different
scaling of the two I-axes: I Max is now at 2.5 × 10–5 A (left vertical axis) or 2.5 ×
10–3 A (right vertical axis); the curves seem to exhibit a kink at about 0.5 and
0.6 V. In electrical engineering, this voltage is called the “kink voltage”, “knee
voltage” or “cut voltage”.
The increase of the U-value by a factor of 2.7 when progressing from U =
0.475 to U = 0.500 V looks like a steep rise. It increases, however, by the same
factor when it progresses from U = 0 to U = 0.025 V but then appears like the
familiar soft curvature of the exponential. The position of the seeming kink on the
U-axis is a function of the scaling of the I-axis.
In Fig. 3.15a, the I-axis is logarithmic. In this representation, the diode charac-
teristics appear as a straight line, except for the points below U = 0.05, because,
there, the term −1 is quantitatively significant. A kink is nowhere to be seen.
Exponential function
A, B amplitudes of functions expA and expB

tA, tB time constants of expA and expB
dt length of time interval
t array of instants of time, separated by dt
expA , expB values at times t.
17 All curves represent exponential functions. The diode curve in Fig. 3.14a is shifted downwards
on the I-axis by the saturation current I s , so that it passes through zero. The curves in Fig. 3.14b
have not been shifted on the U-axis.
18 At about 0.5 and 0.6 V.
Fig. 3.15 a (left) Diode characteristics, logarithmic scale of the I axis (semi-log plot). The curve
can only be represented for I > 0, because the logarithm is defined only for positive values. b (right)
Spreadsheet layout for calculating diode characteristics, I s = saturation current, U T = temperature
voltage, dU = U T
Diode characteristics
Is saturation current
UT thermal voltage, 25 mV at room temperature, characteristic parameter of the
exponential
dU interval width, here, dU = U T
U sequence of voltages, dU apart
exp exponential function with I s and U T as amplitude and characteristic voltage
I diode current, I(U).
Initial slope of the exponential function

In Fig. 3.16 (S), two exponential functions are calculated for 51 points equidistant
in t,

t t
ex p A (t) = A · ex p = A · e tA (3.18)
tA
and expB correspondingly. The characteristic time, t A or t B , may be positive or

negative. The distance dt between the points is set in B3.
For the exponential function y = ex , applies; Plus 1 (in the

argument) yields times e (in the value).
For the power of 2, y = 2n holds: Plus 1 yields times 2.
A B C D E F G H I
1 A 3.0 tA -30.0 3·exp(t/-30) =A&"·exp(t/"&tA&")"
2 B 1.0 tB 15.0 1·exp(t/15)
3 dt 2.0
) tB)
t P(t/tA XP(t/
4 =A6+d =A*EX =B*E
5 t expA expB tangents
6 -20.0 5.8 0.3 30 =-tA 0.0
7 -18.0 5.5 0.30 0.0 3 =A
8 -16.0 5.1 0.34 -30 =tA 6 =2*A
56 80.0 0.2 207.13
Fig. 3.16 (S) Two exponential functions whose independent variable t is specified as column vec-
tor A6:A56 and their parameters amplitude A, B and time constant t A , t B are given in A1:D2. The
legends for the functions are compiled in F1 and F2 with the formula in G1 of type “Text” &
Variables
The table in Fig. 3.16 (S) has a typical layout. The 51 values of the independent
variable time t are located to the left of in a column vector named “t”. The
parameters A, B, and t A , t B of the curves are specified above , as well as the
time interval dt, the distance on the horizontal axis between the calculation points.
The initial value of t, here, −20, has to be entered into cell A6. The values for the
remaining 50 t values are determined successively from the respective predecessor.
The t-predecessor for cell A7 is cell A6, A7 = [=A6 + dt]. In the formula, A6 is
not provided with a dollar sign. It is a relative reference, so that the address of the
addressed cell adapts during copying. Therefore, A56 = [=A55 + dt].
Questions
Suppose that A6 contains an initial time t = 5 and the length of a time segment
is stored in a cell with the name dt and has the value 2. What values for time t
are in cells A7 and A8?19
How can the coordinates of the tangent to the exponential Ae ·exp(t/t 0 ) at t
= 0 be derived from the parameters of the exponential function?20
What is the distance on the horizontal axis between the functions’ calculation
points in Fig. 3.13a created 21
∞ from Fig.
t3.16 (S)?
What is the value of 0 3 · exp − 30 dt? 22
Task Change the parameters A, t A , and dt and check whether the diagram reacts
accordingly! This is the case if each cell contains the correct formula. Remember:
19 A6 = [5], A7 = [= A6 + dt] = 7; A8 = [= A7 + dt] = 9.

20 Straight line through the points (0, A) und (−t e , 0).
21 The distance of adjacent points is dt = 2 (see Fig. 3.16 (S), B3).
22 The definite integral is 90, based on Eq. 3.16, because 3 = 1 · 90.
30
you cannot merely transfer the numbers from the figures to your worksheet. Most
cells contain a formula; only sometimes are numerical values entered directly.
A possible calculation model is presented in Fig. 3.15b (S).
The parameters I s and U T of the diode characteristics are set in the named
cells C1:C2. In C3, the horizontal distance between neighboring sampling points
is defined. Here, we have chosen dU = U T = 0.025 V that is valid for room temper-
ature. The I-U characteristics are illustrated in Fig. 3.14. They intersect the I-axis at
0, showing that no current flows without applied voltage.
Formatting the axis of a diagram

The axis of a diagram is formatted by activating it with the left mouse button and
then clicking format. In the current selection group to the left of the ribbon,
vertical (value) axis shows up in the bar. Upon clicking format selection,
a window opens that allows you to set the minimum, maximum and other param-
eters of the axis. If the axis is to be scaled logarithmically, the appropriate box,
logarithmic scale, must be activated with a checkmark.
Initial slope
A Python program for drawing the exponentials of Fig. 3.13a and their slopes at t
= 0 is given in Table 3.12.
The list t of time instants is created by np.arange (−20, 80 + dt, dt), mimicking
the construction of the time vector in the spreadsheet of Fig. 3.16 (S) with, e.g., A8
= [=A7 + dt]. The lower limit −20 corresponds to the value in A6. To come to
80 (in A56), we have to specify 80 + dt as the upper limit, because np.arange
creates an interval that is open at its right end with the endpoint excluded. ExpA and
expB are constructed from t with the corresponding amplitudes and time constants
Table 3.12 a (top) Specifications of two exponential functions, equivalent to rows 1 through 3 in
Fig. 3.16 (S); b (bottom) labels for the two functions created in lines 7 and 8
1 dt=2.0
2 A,tA= 3.0,-30.0 #Amplitude and time constant
3 B,tB=1.0,15.0
4 t=np.arange(-20,80+dt,dt)
5 expA=A*np.exp(t/tA)
6 expB=B*np.exp(t/tB)
7 lblA=(str(A)+'*'+'exp(t/'+str(tA)+')')
8 lblB=(str(B)+'*'+'exp(t/'+str(tB)+')')
lblA 3.0*exp(t/-30.0)
lblB 1.0*exp(t/15.0)
Table 3.13 Specifications for diode characteristics, the same as in Fig. 3.15b
1 Is=1e-14
2 UT=2.5e-2
3 dU=2.5e-2
4 U=np.arange(-0.2,1.05+dU,dU)
5 exp=Is*np.exp(U/UT)
6 I=Is*(np.exp(U/UT)-1) #Diode characteristics
(A, t A ) and (B, t B ). The curves can be plotted with our standard function StdFig.
In Fig. 3.13a, they are represented with open symbols. In Python, this is specified
by plt.plot(…., fillstyle=’none’, …).
Questions
What are the instructions for plotting the tangents in Fig. 3.13a?23
What are the size and last element of:
– L1 = np.arange(−20, 80+dt, dt) with dt = 1,24

– L2 = np.linspace (−20, 80, 100),25
– L3 = np.linspace (−20, 80, 101)26
– L2 = np.linspace (−20, 80, 100, endpoint=False).27
A Python program corresponding to the spreadsheet layout in Fig. 3.15b is shown
in Table 3.13.
Subplots
A plot like Fig. 3.14b with two vertical axes cannot be achieved with our standard
figure, the function FigStd defined in Sect. 2.4.5. We have to refer to the function
subplots of the pyplot library (see Table 3.14). I(U) is plotted twice, with
ax1.axis([0.3, 0.7, 0, 2e-5]) for the primary (left) y-axis from 0 to
2e−5 and with ax2.axis([0.3,0.7,0,2e-3]) for a second y-axis from 0 to
2e−5, declared with ax2 = ax1.twinx() as the secondary (right) y-axis.
Logarithmic scaling
Logarithmic scaling of an axis can be achieved with plt.yscale
(value=”log”) and plt.xscale(value=”log”).
23 plt.plot([0,tA],[A,0],’k-‘), compare F5:H8 in Fig. 3.16 (S)!

24 L1, size 101, last element L1[−1] is 80.
25 L2, size 100, last element 80.
26 L3, size 101, last element 80.
27 L4, size 100, last element 80 − 100/100 = 79.
Table 3.14 Setting up a diagram similar to Fig. 3.14b with primary and secondary axes, variables
specified in Table 3.13
1 fig, ax1 = plt.subplots()
2 ax1.plot(U, I, 'k-')
3 ax1.axis([0.3, 0.7, 0, 2e-5])
4 for x in np.arange(0.3,0.7,dU):
5 ax1.plot([x,x],[0,2e-5],'k-',lw=0.6)
6
7 ax2 = ax1.twinx() #Secondary axis
8 ax2.axis([0.3,0.7,0,2e-3])
9 ax2.plot(U, I, 'r--')
10 plt.show()
3.5 Questions
General advice
1. To practice programming, translate the spreadsheet solutions of this chapter into

Python and compare with the suggested programs!
Python-specific
2. Let x=np.linspace(−100,100,101). What is the distance between

neighboring elements? What does x[1:]−x[:−1] look like? What are the first
and last elements of (x[1:] + x[:−1])/2 ?
3. Let A = 3.001 and t A = −30.0. How can we produce a label “3.0*exp(t/−30)”?
Concerning Table 3.15:
4. What is the size of the list col?

5. What are the first two elements of col?
6. What are the last two elements of col?
7. How do you compile the four lines into one line with list comprehension?
Check your answers by programming!
Table 3.15 Producing a list x*y

1 col=[]
2 for x in range(1,11):
3 for y in range(1,6):
4 col.append(x*y)
2 2
y.1 y.2 y y.1 y.2
y.1 y.2 y.3 y Einheitsvektoren
unit vectors
y.3 1 1
0 0
-2 -1 0 1 x 2 -2 -1 0 1 x 2
-1 -1
-2 -2
Fig. 3.17 a (left) The segment defined by two points (x 1 , y1 ) and (x 2 , y2 ) is extended to a third
point whose x-coordinate x 3 can be selected arbitrarily. b (right) The unit vectors in the direction
of the line and perpendicular to the line are attached to the line, defined in Fig. a, at the left point
Broom rules
8. Explain the broom rules:

Ψ Lens equation with plus and minus.
Ψ Blank lines separate curves.
Ψ Doppler effect with plus and minus.
Straight-line equation
A straight line is defined by two points (x 1 , y1 ) and (x 2 , y2 ) (see Fig. 3.17a).
9. What is the equation for determining the distance between the two points?
10. Which equation must be used to find the value y3 for a given horizontal position
x 3 so that (x 3 , y3 ) lies on the straight line?
11. Deduce from the coordinates of the two points the direction vector (Dx , Dy ) of
the straight line normalized to length 1!
12. How do the coordinates of the two points result in the vector (Px , Py), the
perpendicular to the straight line and normalized to length 1?
A spreadsheet layout for three points on a straight line is shown in Fig. 3.18
(S).
13. What is the linked cell and min and max of the slider in F4:H4?
14. What formulas are in B3 (input from E3) and D4 (input from E4)?
15. Write a Python program that performs the calculations of Fig. 3.18 (S)! Replace
the function of the sliders with simple assignments with random functions: E1=
… ; E4= … !
3.5 Questions 111
A B C D E F G H I J K M N P
1 Three points on a straight line Unit vectors
2 x.1 -0.6 y.1 -0.5 along the straight line
3 x.2 0.92 y.2 0.57 92 x.1 -0.6 y.1 -0.5
4 x.3 1.5 y.3 0.98 57 x.p -0.03 y.p -0.1
5 perpend. to the straight line
6 Length of segment 1-2 Slope of segment 1-->2 x.1 -0.6 y.1 -0.5
7 l.12 2.66 m.12 0.70 x.v -0.2 y.v -1.07
Fig. 3.18 (S) Spreadsheet layout used to create the coordinates for Fig. 3.17a and b, B3 and D4
depend on the outputs on the sliders. The slider in F4:H4 goes from 0 to 80
0
-8 -4 0 4 8
Object and image
Lens
-4
Fig. 3.19 Incomplete image construction for imaging with a converging lens
Image construction with converging and diverging lenses
16. In school, one usually learns the equation 1/f = 1/o + 1/i for imaging with
converging lenses (o is object distance, i is image distance). How is this imaging
equation modified according to DIN 133528 and made suitable for numerical
calculation in spreadsheets and Python both for converging and diverging
lenses?
17. How is the magnification factor defined in DIN 1335?
18. What characterizes a converging lens in the imaging equation?
19. What characterizes a diverging lens in the imaging equation?
20. Draw the rays for the image construction in Fig. 3.19!
21. What is the image-side focal length?
22. Draw the bundle of rays that contributes to the image formation!
28 Equations 3.5 and 3.6.

Fig. 3.20 Gravitational F g

and centrifugal F c force 8 Bank line
when driving through a curve
Fg, Fc
0
0 2 4 6 8 10
Forces when driving through a curve
23. What are the formulas for the gravitational force F g and the centrifugal force
F c on a car of mass m, traveling with speed v through a curve with radius r?
24. What do you have to do to get a true-angle display when the x-axis is scaled
from −2 to 8 km and the y-axis from 10 to 15 km, in excel and in Python?
25. How is the static friction force defined? What does a static friction coefficient
μ = 0.5 mean?
26. Draw in Fig. 3.20, with a triangle ruler, the resulting force, the force in the
plane, and the force perpendicular to the plane! Which force determines the
static friction?
27. Draw a vector in the bank line and another one perpendicular to it!
28. What are the vector equations for determining the quantities of Question 26?
Doppler effect
fE c ± vE
= (3.19)
fQ c ∓ vQ
29. What do the letters in Eq. 3.19 stand for? Which signs are to be used when?
Adjust the formula for the three cases in Fig. 3.21!
A car (sender S) drives along the x-axis. Its position at time t is specified in an array
named x S . A pedestrian (receiver R) moves along the y-axis towards the x-axis. Its
position at time t is indicated in an array named yR .
S R R S R S
Fig. 3.21 A sound source S and a sound receiver R move on a straight line
3.5 Questions 113
30. What is the distance d SR between car and pedestrian as a function of t?

31. What is the formula for calculating the relative velocity vSR in the direction of
the link line? To answer this question, you have to define vectors x R , yR and d SR .
Exponential function
32. To draw an exponential function freehand on paper, it is useful to start with a

straight line as a guide. How is this straight line determined by the exponential
function parameters, amplitude A and time constant τ ?
33. An exponential function increases from 1 to 2 when the argument is increased
from t = 0.0 to 0.1 s. How much does it increase if the argument is increased
from 0.0 to 0.2 s? Think binary! How much does it increase if the argument is
increased from t = 0.8 to 1.0 s? Sketch the same exponential function twice,
each time with the t-axis from 0 to 1.2, but with a y-axis scaling from 0 to 4 for
the first sketch and from 0 to 16 × 4 for the second sketch! What do you see in
the second sketch that is not obvious in the first sketch?
Macros with Visual Basic and Their
Correspondences in Python 4
We practice the basic programming structures: loops, branches, sub-routines

(FOR, IF, SUB/def), with particular emphasis on data exchange between
spreadsheets and procedures. We will learn to obtain the sequences for
EXCEL-typical spreadsheet operations by recording the associated commands
with a macro recorder. With this knowledge, we will:
– draw dense-packed crystal planes,

– decode protocols of measuring instruments and compile clear summaries
of the results,
– systematically modify the parameters of calculation models with log and
control procedures and continuously enter the results of the calculations
into another range of the spreadsheet, and
– outsource complicated formulas into user-defined spreadsheet functions
in order to arrange the tables more clearly.
We will present parallel solutions in Python creating the drawings with

the library turtle.
4.1 Introduction: For, If, Sub/Def
be found at the internet address: go.sn.pub/gTtbiH.
Tim worries, Alac brags
Mag This chapter will teach us to program, to let macros interact with
spreadsheet calculations, and to realize parallel solutions in Python.

116 4 Macros with Visual Basic and Their Correspondences in Python
Tim That sounds pretty demanding. Is it at all manageable for beginners like
me?
Mag Quite clearly: Yes. Many people have already achieved that, even students
who had never before written a line of code. Visual Basic is a good-natured pro-
gramming language that does not require much knowledge, at least not for the
tasks that we want to tackle. This chapter will not only take away your fear of
programming, but you will also find it fun to write programs that do amusing
things.
Tim Well, people who had just finished the course told me that they had to deal
with routines, macros, programs, and procedures.
Mag Don’t worry, we don’t make any distinction among those terms, we
use them all synonymously. Our programs include both EXCEL-typical command
sequences and classical algorithmic structures.
Alac I’m not afraid of VISUAL BASIC or Python; after all, I already took a
course about another programming language and used it to write funny programs.
Mag That is certainly a good prerequisite for faster success. You will have an
easier time than Tim. Nevertheless, don’t take the tasks for granted. In our course,
programming tasks are combined with physics exercises (so much for our hopes
for just “funny”:-)). I’ve often experienced instances in which mere programmers
have been discouraged by their limited progress in this kind of programming and
have given up.
Alac So, I’m essentially learning more physics?
Mag Not only that. You should certainly understand more about physics after
the course than before. Nevertheless, the exercises will also familiarize you with
good programming skills: to develop systematically, to document carefully, to
detect and correct mistakes.
Tim One more question. Many workplaces require programming skills in spe-
cial programming languages. Wouldn’t it be better for me to learn such languages
from the onset?
Mag Don’t worry.’ In our tasks with VISUAL BASIC and Python, you will have
room to make a sufficient number of mistakes to learn from so that you can be con-
fident in becoming a computer expert. The algorithmic constructions are the same
in all programming languages. More important than acquiring special knowledge
at an early stage is that you gain the ability to cope with “hard” programming
tasks and master the rules for good programming.
4.1 Introduction: For, If, Sub/Def 117
How do we proceed?
This course cannot give a general introduction to programming, because that would
require an excess of repetition of things that are already well described in specific
textbooks. As in Chap. 2, we will do a basic exercise that you should follow step by
step and convert into your own program. It contains all of the commands and con-
structions that we will need later, but not much more. However, this basic knowledge
will enable you to find your way around in EXCEL help or Python internet aid and
choose suitable textbooks for programming with VISUAL BASIC and Python.
Only a few algorithmic constructions

In our programs, we use constructions that are the same in all algorithmic languages:
Loops, logical branches, sub-routines (FOR, IF, SUB/def) . Special EXCEL instructions,
e.g., for handling files or creating drawings, will be provided to us by the macro
recorder, which records commands that the user makes when setting up spreadsheet
calculations.
Three types of program will be used repeatedly in the following chapters: log
procedures, formula procedures, and user-defined functions.
Rep-log and scan-log procedures

Two tools can be used sensibly for every task in EXCEL: Sliders and log procedures.
With a slider, variables in a spreadsheet can easily be changed manually, and the
user can see how the results change, as we have learned in the basic Exercise 2.2. A
rep-log procedure systematically changes independent variables in the spreadsheet
and continuously writes the calculation results into another spreadsheet range. The
spreadsheet calculation works like a function. The equivalent in Python is functions
that are repetitively called in loops. Scan-log procedures scan a range of a spread-
sheet and write the data into another range in a structured manner. The program in
Fig. 4.5 (P) is a simple example. In Chaps. 8 to 10, log procedures repeat stochastic
experiments to illustrate statistical rules and laws.
User-defined functions in visual basic

We generally recommend performing calculations step by step in several rows or
columns. If the calculation runs without errors, spreadsheet space and calculation
time can be saved if you execute this calculation in a user-defined spreadsheet func-
tion, with values transferred from the spreadsheet into the program, and with values
written into the spreadsheet by the function, exactly as we learned with built-in
spreadsheet functions, e.g., COS(x) or SIN(x). As examples, we shall implement func-
tions for the scalar and the vector products of three-dimensional vectors, as usual, in
both Visual Basic and Python.
Pandas
Pandas (“Python Data Analysis”) is a library for Python, based on NumPy.
It is designed for data management and analysis and works with structured data
(DataFrame (2-dimensional)) and time series (Series (1-dimensional)), thus
mimicking spreadsheet calculations. We shall use it in this book only in Exercise 4.8
“Processing the protocol of a measuring device”.
4.2 Basic Exercise: FOR-Loops
We get to know the VISUAL-BASIC-Editor and practice reading cell con-

tents and filling in cells. We are using For-loops to execute tasks one after
the other with systematically changed parameters. This often involves incre-
menting a running index in the loop that indicates a cell’s position in
the spreadsheet. Such loops are called loop2i, obeying the broom rule
Continue counting in the loop!
4.2.1 Visual-Basic-Editor 1: Editing
In the menu ribbon (see Sect. 1.7), click on the main group DEVELOPER and then
on the tab VISUAL BASIC; the “SHEET1 (CODE)” window appears (Fig. 4.1).
If the right lower rectangle (below “(General)”) is gray, double-click the SHEET1
line in the MICROSOFT EXCEL OBJECTS so that it turns white. Instructions that you
write into this sheet in the VISUAL BASIC editor are executed in Sheet1.
Fig. 4.1 Spreadsheet and associated VISUAL BASIC sheet after executing DEVELOPER/VISUAL BASIC
(EXCEL 2019)
4.2 Basic Exercise: FOR-Loops 119
4.2.2 Programming
Enter values and formulas into cells

Write SUB Name into the first line of the white area! For “Name”, enter your own
name. The editor adds one line END SUB. You can now write statements between SUB
Name and END SUB. In Fig. 4.2b, SUB Annegret has been created with four instructions.
Cell B1 in a spreadsheet can be addressed in two ways, with CELLS(1,2) or with
RANGE(“B1”). CELLS(r,c) addresses the cell in the rth row and cth column. Cell A1
(= CELLS(1,1)) is filled with a text in quotation marks, here, with “Annegret”. Of
course, you should write your own name.
Individual cells can also be addressed with instructions of the type
[RANGE(“A1”) = …], as in lines 3 and 5 of the macro in Fig. 4.2b. In loops,
the addressing with CELLS(r,c) can be used more effectively, if, e.g., the row index
r or the column index c is to be scanned systematically.
Task First, write only this one line [CELLS(1,1) = “Annegret”] into the VBA sheet
and start the program by pressing the start button (high-lighted in Fig. 4.2b). For
the procedure to be executed, the pointer | must be somewhere in the procedure.
Then, insert the other lines of SUB ANNEGRET one after the other, execute the macro
after each line and observe what happens in the spreadsheet:
– a number is written into Cell B1, here, 12.25.

– CELLS(2,1) = CELLS(1,2) means that cell A2 (CELLS(2,1)) is being filled in with
the contents of cell B1 (CELLS(1,2)). The content of B1 is transferred once to
A2 by the program, and A2 remains unchanged, even if the content of B1 is
changed later.
– The text “ = A1” is written into cell B2, interpreted in the spreadsheet as a
formula. The corresponding cell is filled in with the contents of cell A1, B2 =
Fig. 4.2 SUB Annegret in b (right) writes into range A1:B3 in a (left). The text after the apostro-
phe’ is interpreted by the VBA interpreter as a comment and not as program code
[=A1]. If the contents of A1 in the spreadsheet are now changed, the new value
also appears in cell B2.
Task Delete all entries in the spreadsheet again and then run through the pro-
gram step by step by placing the cursor | in the program and repeatedly pressing
the function key F8 (also obtained and explained in the VBA developer, Fig. 4.9, by
DEBUG/ STEP INTO). Step by step, the previously deleted entries will appear again in
the spreadsheet.
FOR loops The macros in Fig. 4.4 (P), Fig. 4.5 (P), and Fig. 4.6 (P), wherein
FOR loops are used, fill in the spreadsheet in Fig. 4.3 (S).
SUB Protoc1 in Fig. 4.4 (P) fills in column A. Line 2: the text “x” is written
into A1. In the FOR loop, the variable x is incremented from 3 to 9.5 in steps of
0.25 and written into cells in A. The variable x takes on 3 as the first value and
is then incremented by 0.25 each time the loop is traversed until the value 9.5 is
reached. The variable of the FOR loop, here, x, is called the loop index. In line 9,
a formula is entered into C6, normal text enclosed in quotation marks but starting
with an equal sign.
A B C D E F G H I J K
1 x 9.50 x Cos(x) Sin(x) Tan(x)
2 3.00 -1.00 =COS(C1) 3.00 -0.99 0.14 -0.14 3 9.50
3 3.25 -0.08 =SIN(C1) 3.25 -0.99 -0.11 0.11 4
4 3.50 0.08 =TAN(C1) 3.50 -0.94 -0.35 0.37 5 x
5 3.75 3.75 -0.82 -0.57 0.70 6 Cos(x)
6 4.00 0.99 =COS(C1)^2 4.00 -0.65 -0.76 1.16 7 Sin(x)
7 4.25 4.25 -0.45 -0.89 2.01 8 Tan(x)
8 4.50 4.50 -0.21 -0.98 4.64
23 8.25 8.25 -0.39 0.92 -2.39 3 0.08
24 8.50 8.50 -0.60 0.80 -1.33 4 =TAN(C1)
25 8.75 8.75 -0.78 0.62 -0.80 5 3.50
26 9.00 9.00 -0.91 0.41 -0.45 6 -0.94
27 9.25 9.25 -0.98 0.17 -0.18 7 -0.35
28 9.50 9.50 -1.00 -0.08 0.08 8 0.37
Fig. 4.3 (S) Column A is filled in by SUB Protoc1 in Fig. 4.4 (P). The formula in C6 has also been
entered by SUB Protoc1. The list in columns E:H is obtained from the spreadsheet calculation in
C1:C4 by executing SUB Protoc2 in Fig. 4.5 (P). The range C1:H4 is transferred by SUB ScanCopy
in Fig. 4.6 (P) into the two columns J and K
1 Sub Protoc1() r2 = r2 + 1 6
2 Cells(1, 1) = "x" Next x 7
3 r2 = 2 8
4 For x = 3 To 9.5 Step 0.25 Range("C6") = "=Cos(C1)^2" 9
5 Cells(r2, 1) = x End Sub 10
Fig. 4.4 (P) SUB Protoc1 fills in A in Fig. 4.3 (S) and a formula into C6. Syntax for calling a cell:
CELLS (ROW, COLUMN)
1 Sub Protoc2() Cells(r2, 6) = Cells(2, 3) '6 = column F 7

2 r2 = 2 'Running index for writing to cells Cells(r2, 7) = Cells(3, 3) '7 = column G 8
3 Cells(1, 6) = "Cos(x)" Cells(r2, 8) = Cells(4, 3) '8 = column H 9
4 For x = 3 To 9.5 Step 0.25 r2 = r2 + 1 10
5 Cells(1, 3) = x 'Column C Next x 11
6 Cells(r2, 5) = x 'Column E End Sub 12
Fig. 4.5 (P) SUB Protoc2 changes the value in cell C1 (line 5) and writes the function values from
C2:C4 consecutively into the columns F (6th) to H (8th) of the spreadsheet in Fig. 4.3 (S)
Loop2i structure
In Fig. 4.4 (P) , we have introduced a running index r 2 , which specifies the row
of the cell to be filled in. It is set in line 2 to 2 before the start of the loop and is
incremented by 1 in line 6 at the end of each loop cycle, so that the values of x are
sequentially written into lines 2 to 28. We will often use such structures, call them
Loop2i, because they comprise two indices, and we memorize them with a broom
rule .
Loop2i: Continue counting (the running index) in the loop!
Spreadsheet calculation used as a function

In cells C2:C4 of Fig. 4.3 (S), formulas with the trigonometric functions COS, SIN,
and TAN are written by hand with the argument in C1, e.g., C1 = [9.50] and C2 =
[=COS(C1)]. These spreadsheet calculations are used by SUB Protoc2 in Fig. 4.5 (P)
as a function. It is a typical rep-log procedure, changing a parameter in a spreadsheet
calculation and writing the calculation results to another range of the spreadsheet.
SUB Protoc2 changes the value in cell C1 in the (x = )-loop, which is used as an
argument in the functions in C2:C4, and transfers the results of the spreadsheet
calculation from C2:C4 (one below the other) to F to H (side by side).
The statement in line 3 is: CELLS(1,6) = ”Cos(x)”. The quotation marks indi-
cate a text in between, which is to be written as text into the cell. An instruction
CELLS(1,6) = COS(x) would cause the program first to calculate the cosine of the
variable x, to which one would have to have assigned a value somewhere earlier
in the procedure, and then write the result, a number, into the cell. An instruction
CELLS(1,6) = ‘ = COS(x)’ would write a formula into the cell, similar to line 9 in
Fig. 4.4 (P).
All cell references with fixed row and column indices can also be expressed
with RANGE, e.g., RANGE(“C2”) instead of CELLS(2,3).
Questions
concerning SUB Protoc2, Fig. 4.5 (P)

1 Sub ScanCopy() Cells(r2, c2 + 1) = Cells(r, c) 7

2 r2 = 2 'Row 2 r2 = r2 + 1 8
3 c2 = 10 'Column J Next c 9
4 For r = 1 To 4 'Rows 1 to 4 r2 = r2 + 1 10
5 For c = 3 To 8 'columns C to H Next r 11
6 Cells(r2, c2) = c End Sub 12
Fig. 4.6 (P) SUB ScanCopy writes the contents of the range A1:F4 of Fig. 4.3 (S) consecutively
into columns J and K of the same spreadsheet. CELLS (2,3) corresponds to C2 in the spreadsheet
What do the instructions in lines 3 and 5 have to be if you want to address

with RANGE?1
Which instructions must be added to the code to write the headings in G1
and H1 of Fig. 4.3 (S)?2
How would you have to change the for-loop in lines 4 to 11 if you want to
read the x values from column A?3
Nested loops
SUB ScanCopy in Fig. 4.6 (P) transfers the range (r = 1 to 4: c = 3 to 8), i.e., C1:H4
of the table, to columns J and K of Fig. 4.3 (S).
Range C1:H4 is read horizontally, row by row, and written consecutively verti-
cally into column K (c2 + 1 = 11) with the (c = ) loop; line 8: The running index
r 2 is incremented, indicating the next free row in the spreadsheet.
To be read, the range with the two coordinates row number r (from 1 to 4) and
column number c (from 3 to 8) must be scanned with the cells being addressed
with CELLS(R,C). This is done with two nested loops, an outer loop (FOR r = ), and
an inner loop (FOR c = ) that is called within the outer loop and ends in line 9
with NEXT C. SUB ScanCopy also writes the index c into column J (line 6, c2 = 10
from line 3).
The line index r 2 is incremented by one at the end of each of the two loops FOR
c = and FOR r = . The increment in the inner loop (FOR r = ) causes the adjacent
entries in a row of the table, e.g., C1:H1, to be written consecutively into rows 2
to 7 of J, J2:J7. The increment in the outer loop (FOR c = ) causes a row, e.g., row
8 in the table in Fig. 4.3 (S), to be skipped.
1 RANGE(„F1“) = “Cos(x)”: RANGE(„C1“) = x.

2 RANGE(„G1“) = “Sin(x)”: RANGE(„H1“) = „Tan(x)“; do not forget the quotation marks!
3 FOR r = 2 TO 28: x = CELLS(r, 1): … To put multiple statements on one line in VBA, separate the
statements by a colon “:”!
Questions
Using the variables x and r 2 in the loop2i in SUB Protoc2 in Fig. 4.5 (P), explain
the broom rule: Loop2i: Continue counting (the running index) in the loops!4
It would have been easier to specify the columns in SUB ScanCopy in Fig. 4.6
(P) as numbers, i.e. CELLS(r 2 , 11) instead of CELLS(r 2 , c2 + 1). Does the variant
CELLS(r 2 , c2 + 1) offer any advantage?5
Why is J8:K8 in Fig. 4.3 (S) not filled in?6
What is the value of r 2 at the end of SUB ScanCopy in Fig. 4.6 (P)?7
How do we proceed further?

In this basic exercise, we have learned how to read content from cells into proce-
dures and fill in cells. We have also become familiar with FOR loops and the special
construction of loop2i with a loop index and a running index. In this chapter’s next
exercises, we will get to know sub-routine calls (SUB) and logical queries (IF) . FOR,
SUB, IF are already the essential basic structures of programming, which we will
repeat in Python (def instead of SUB) and apply in the following chapters over and
over again.
4.3 Macro-Controlled Drawings with FOR, SUB, IF
We construct a macro for drawing filled circles of variable diameter at dif-

ferent coordinates in the spreadsheet. The required instructions are obtained
by recording macros generated when a circle is inserted and formatted as
a shape by hand. They are combined in a sub-routine to be called from
the main program, specifying the circles’ position. Similarly, we get the
instructions for drawing rectangles and triangles. We are practicing the basic
structures of programming: FOR, IF, SUB.
4 The rows from r = 1 to 4 and the columns from c = 3 to 8 are scanned in the nested for-loops. The
24 scanned values are stored in successive rows. The index of these rows, r 2 , must be incremented
in the inner FOR loop after every entry.
5 If the data is to be output to another range of the spreadsheet, only one parameter for the columns,
namely, c2 , must be adjusted in addition to r 2 .

6 Because, in ScanCopy in Fig. 4.6 (P), at the end of the loop FOR r = 1 to 4, the index r is
2
incremented without data having previously been written into cells in that row.
7 At the end of ScanCopy, the following applies: r = 2 (initial value) + 4 × 6 = 24 (c-loop) + 4
2
(r-loop) = 28 (Row 28 in Fig. 4.3 (S)).
4.3.1 Macro Recorder
We are going to record the commands that are executed when we insert an ellipse
into a spreadsheet and format it. The recorded macro is converted into a sub-
routine that is called several times by the main program with modified coordinates.
In Fig. 4.7a, you see a decorative spiral drawn with the tools acquired in this
exercise. The starting point is a macro (Fig. 4.7b), recorded when an ellipse was
inserted and formatted.
What do We Learn in This Exercise?
Mag Once you have completed this task, you can create images like the one in
Fig. 4.7a.
Alac Great, that will amaze my friends!
Mag More importantly, you will master Visual Basic statements such as those
in Fig. 4.7b (P).
Tim Terribly complicated! I will never be able to keep all of that in my head at
the same time.
Mag You’re not supposed to. Figure 4.7b (P) contains a series of instructions
that the Macro Recorder has recorded when an ellipse has been inserted by hand.
Fig. 4.7 a (left) Decorative spiral, drawn by a macro. b (right, P) Macro recorded by the macro
recorder while an ellipse is inserted into the spreadsheet. Superfluous instructions have been
deleted. If possible, do not write such code by hand! Get it using DEVELOPER/ RECORD MACRO and
modify it as needed!
4.3 Macro-Controlled Drawings with FOR, SUB, IF 125
Alac So, everything is done by the macro recorder?
Mag No, you still have to modify the recorded code, introduce variables and
learn the basic program constructions: loops (FOR i = … TO …), logical branches
(IF THEN … ELSE …), and sub-routines. (CALL SUB(a, b, c, …)).
The tab DEVELOPER/RECORD MACRO

We want to apply a VBA macro to draw a series of filled circles. First, we have to
get the elementary commands for drawing a circle. These can be found, in principle,
in manuals for VISUAL BASIC FOR APPLICATIONS. Nevertheless, we make life easier for
us and use the macro recording function. You can find it in the main register tab
DEVELOPER, (see Fig. 4.8). Further explanations can be found in EXCEL help under
the keyword CREATE A MACRO.
If the main tab DEVELOPER does not appear in your ribbon, you must activate it in
the EXCEL options, with FILE/OPTIONS/CUSTOMIZE RIBBON/MAIN TABS/ DEVELOPER.
Circles, Squares, Triangles, by Hand and by Macro

After turning on the macro recording function, we draw a circle by hand
(INSERT/ILLUSTRATIONS/SHAPES) and format it. For example, we select the color of the
filling and the thickness and color of the border. When we have finished the drawing,
we end the macro recording by clicking the STOP RECORDING button, which appears in
the toolbar in place of RECORD MACRO. The recorded macro is in a project MODULE (see
Fig. 4.9 under “Modules”), not in a VBA sheet connected with a spreadsheet.
Fig. 4.8 The DEVELOPER/RECORD MACRO tab records all program code associated with
the spreadsheet operations performed by the user, e.g., introducing a rectangle as in
F1:H2 (INSERT/ILLUSTRATIONS/SHAPES). The VISUAL BASIC button (far left) activates the
VISUAL BASIC EDITOR (see Fig. 4.1)
Fig. 4.9 Visual basic editor. You have to activate PROJECT EXPLORER (with VIEW/PROJECT EXPLORER)
to see all open files. The recorded macro is located in Module 1 (hidden in the group MODULES) of
the VBA project (4–2 Basic Exercise Annegret)
VBA is the abbreviation for “Visual Basic for Applications”. The addition “for
applications” indicates that the application’s instructions, here, EXCEL, are available
as internal instructions, e.g., ACTIVESHEET. ADDSHAPE, with which a geometric form
is inserted into the spreadsheet.
Select Objects and Edit them Together

You can select objects with the arrow cursor. To do this, click
HOME/FIND AND SELECT/SELECT OBJECTS at the ribbon’s far-right. With the
new mouse pointer held down, you can now span a rectangle within which
all drawing objects are selected and edit this set of objects as a whole, for
example, color them, group them, or delete them.
Within a VBA macro, use the command ACTIVESHEET.DRAWINGOBJECTS.SELECT
to select all drawing objects in the active worksheet. With SELECTION.DELETE, you
can delete all objects.
4.3.2 Visual-Basic Editor 2: Macro Recording, Debugging
We can review the result of our macro recording in the Visual Basic editor. This
editor is activated when you click on DEVELOPER/VISUAL BASIC (far left in Fig. 4.1)
or press ALT F11. A window like that in Fig. 4.9 appears when the program page of
a sheet or a module is additionally double-clicked. SHEET1 has been clicked here,
which already contains SUB Annegret from Sect. 4.3.2.
Upon clicking on the “View” tab, a menu opens up that has been placed over
SUB Annegret in Fig. 4.9.
We click on the PROJECT EXPLORER button, and the PROJECT—VBAPROJECT sub-
window appears. In this window, each worksheet (SHEET1, SHEET2, SHEET3, SHEET4)
is assigned a VBA sheet in which Visual Basic code can be generated and edited.
In Fig. 4.9, SHEET1 (Tabelle1(Annegret)) has been clicked, and in the editor, the
macro SUB ANNEGRET from Sect. 4.2.2 has popped up.
Since we have already recorded a macro, another object MODULE1
appears under Modules . It contains the program code SUB MACRO1, which we
have transferred to Fig. 4.7b (P), with four instructions:
– Lines 3 and 4: An ellipse (MSOSHAPEOVAL) has been created. The first two num-
bers in the argument list are the x and y coordinates; both are measured from
the upper left corner of the spreadsheet. The next two numbers in the list are
the two diameters of the ellipse.
– Line 7: The area within the ellipse is colored.
– Line 13: The border of the ellipse is colored.
– Line 18: The thickness of the border of the ellipse is specified.
You can edit the macro commands in the editor like normal text. The syntax must,
of course, comply with the rules of the VBA interpreter.
Debug/Step Into
Let the macro run again; best if you do it step by step! If you place the cursor in a
program in the Visual Basic Editor and press the function key F8, each step of the
program is executed individually (DEBUG/STEP INTO). You can then see exactly what
is happening and check whether the drawing is changing as you expect. You can
also change the instructions before they are executed. Going through a macro step
by step is a good way to detect programming errors.
If you place the cursor on a variable name, the value of that variable will pop up.
Task Change the coordinates and the size of the diameters by modifying the
instructions!
Mag Now, the real programming starts, with loops and sub-routines!
4.3.3 Programming Elements
Variables Instead of Numbers

In SUB MACRO2 in Fig. 4.10 (P), we have replaced the current numbers in ADDSHAPE
with variables x, y, dx and dy, to which we have assigned values in lines 2 to 5. If
we run this macro, one of the shapes in Fig. 4.11a (S) is created, or a similar one if
other values have been chosen.
The colors in lines 9, 12, 32, and 35 are composed of red, green, and blue com-
ponents (intensity between 0 and 255) via the specification RGB (red, green,
blue).
1 Sub Macro2() Sub Circles() 20

2 x = 400 For i = 1 To 3 21
3 y = 20 Call Disc(i * 50 + 60, i * 25) 22
4 dx = 100 Next i 23
5 dy = 50 End Sub 24
6 ActiveSheet.Shapes.AddShape(msoShapeOval, _ 25
7 x, y, dx, dy).Select Sub Disc(x, y) 26
8 With Selection.ShapeRange.Fill 'x and y are the coordinates of the center 27
9 .ForeColor.RGB = RGB(220, 220, 220) d = 50 'diameter of the circle 28
10 End With ActiveSheet.Shapes.AddShape(msoShapeOval, _ 29
11 With Selection.ShapeRange.Line x - d / 2, y - d / 2, d, d).Select 30
12 .ForeColor.RGB = RGB(180, 0, 0) With Selection.ShapeRange.Fill 31
13 .Weight = 1 .ForeColor.RGB = RGB(220, 220, 220) 32
14 End With End With 33
15 End Sub With Selection.ShapeRange.Line 34
16 .ForeColor.RGB = RGB(180, 0, 0) 35
17 Private Sub CommandButton1_Click() .Weight = 1 36
18 Call Circles End With 37
19 End Sub End Sub 38
Fig. 4.10 a (left, P) Variable names are introduced, MACRO1() from Fig. 4.7b (P) becomes
Macro2(). The macro SUB CommandButton1 is triggered by the command button in Fig. 4.11b.
b (right, P) Macro2() is converted into a sub-routine Disc(x,y), which is called repeatedly by the
main program Circles with various values for x and y, with the result in Fig. 4.11b
A B C D A B C D
1 1
2 2
3 3
4 4
5 5
CommandBuon1
6 6
7 drawn by Macro1 7
Fig. 4.11 a (left, S) Circle and ellipse after executing SUB Macro2 in Fig. 4.10a (P). b (right, S)
Result of the procedure Circles in Fig. 4.10b (P)
Questions
Which color is created with RGB(180, 0, 0)?8

Which color is created with RGB(220,220,220)?9
How can you tell that a circular disc, and not an elongated ellipse, is
produced with SUB Disc?10
Sub-routines
We want to summarize the relevant instructions in a sub-routine “Disc”, which con-
tains, in the procedure header, the coordinates (x, y) of the center of the circle in
the parameter list, Disc(x, y). This sub-routine is called from a main program with
different values for (x, y). Figure 4.10b (P) suggests a solution for this task, with the
main program SUB Circles and the sub-routine SUB Disc. Superfluous specifications
in the recorded macro have been deleted.
The diameter of the circular disc is set to d = 50 in SUB Disc (line 28). Grey
is now selected as the fill color, lines 31, 32. The line width remains as before
(… LINE.WEIGHT = 1). These parameters cannot be changed by the main program,
because they are not in the procedure header.
When placing the circular disc in the spreadsheet, note that the center of the
circle is passed via the procedure header (SUB Disc(x,y)), but that it is the upper
left corner of the shape that must be specified in the drawing command.
The ratio of the scaling in Visual Basic to the grid scale in the spreadsheet can
be seen from the following data:
– A circle with diameter 100 points has a diameter of 3.53 cm.

– 28.4 point correspond to 1 cm.
– 28.5; 28.6; 28.7 point all correspond to 1.01 cm.
– 28,8 point correspond to 1.02 cm.
Transferring parameters to sub-routines
When parameters are passed to sub-routines, the order in the argument

list in the procedure header is decisive; the names in the main program
are not significant.
A procedure header in Fig. 4.10b (P) reads SUB Disc(x,y). This sub-routine is called
in SUB Circles() with CALL Disc(I*50 + 60, I*25). The first entry in the header in SUB
Disc is taken over as x, and the second entry as y. We often name the variables
8 RGB(180, 0, 0) is a strong red, intensity 180 of 255.

9 RGB(220,220,220) is a light grey; red, green and blue
are equally present.
10 In lines 29, 30, ACTIVESHEET.SHAPES.ADDSHAPE(…,., d, d), the same variable d is used for both
diagonals of the ellipse.
in the main program the same as in the sub-routine. So, we could write x = i*50
+ 60 and y = i*25 within the loop in SUB Circles() and then call CALL Disc(x, y),
with the same result as above.
If we executed CALL Disc(y, x), the first entry, here, y from the main program,
would be interpreted as x in the subprogram and the second entry as y. The row of
the three circular discs would start at A8 and go down more steeply. In the main
program, we could also choose completely different variable names, e.g., a and b,
and then proceed with CALL Disc(a, b) or CALL Disc(b, a).
Name the variables such that you are best able to keep an overview!
Questions
concerning Fig. 4.10b (P)

Which three circle centers are passed to SUB Disc(x,y) in SUB Circles?11
Which argument in CELLS(a,b) stands for the row index in the spreadsheet?12
Main Program
A main program is characterized by the fact that it contains no parameters in the
procedure header. Only main programs are executable programs. Sub-routines gen-
erally contain parameters in the header that must be assigned values by a higher-level
program. Examples:
– SUB circles() in Fig. 4.10b (P) is a main program that the user can start.
– SUB disc(x,y) in Fig. 4.10b (P) is a sub-routine with x and y in the procedure
header. It cannot run on its own, but can only be called by another procedure with
specified values for the parameters x and y.
Task Change the procedure so that, in addition to the coordinates of the cen-
ter point, the diameter d of the circle and the thickness w of the boundary
(SHAPE OUTLINE) are selected in the main program and are transferred to the
sub-routine as parameters in the procedure header!13
FOR loop
The main program Circles calls the sub-routine DISC in the loop (FOR i = ) three
times. The centres of the circular disks are set to (110, 25), (160, 50) and (210,75)
for i = 1, 2, 3. The drawing resulting from these specifications is shown in Fig. 4.11b.
11 (x, y) = (110, 25), (160, 50) and (210, 75).

12 The first argument, a, stands for the row: CELLS(row, columns).
13 SUB DISC(X,Y,D,W), LINE 36, WEIGHT = W.
A FOR loop is used in the macro Circles. The general syntax for a FOR loop is:
FOR x = xmin TO xmax STEP delta_x
{LIST OF COMMANDS}
NEXT x
An example with integers:
r2 = 10
FOR N = –211 TO 453 STEP 12
CELLS(R2, 2)=N
R2 = R2 +1
NEXT N
When this loop is executed, the loop index n assumes the values −211, −199,
…, 437, 449. CELLS(10,2) to CELLS(65,2) are filled in. In the argument of CELLS,
the row number comes first and the column number second. Cells B10 to B65 are
therefore filled in with −211, −199, …, 449.
A further example is the loop in SUB Circles(), in which the sub-routine Disc is
called three times:
FOR I = 1 TO 3
CALL DISC(I*50+60, I*25)
NEXT I
Task Develop a macro for drawing a row of rectangles! “Develop” means that
you get the instructions with RECORD MACRO and redesign the recorded macro using
variables, sub-routines, and loops.
Task Write a macro that draws a (4 × 4) array of filled circles, the colors thereof
being composed of fractions of red and green, with the green fraction systemati-
cally increasing in each row and the red fraction systematically increasing in each
column!
Command Button, Design Mode On/Off

In Fig. 4.11b, a command button has been inserted in "Design mode" in A5:B6 with
DEVELOPER/INSERT/ACTIVEX CONTROLS/COMMAND BUTTON (visible when the mouse is
over the ▭ icon).
In the PROPERTIES card, revealedby right-hand clicking on

COMMAND BUTTON/PROPERTIES, COMMANDBUTTON1 (as text) is assigned both as
a NAME and a CAPTION for the command button. As is usually the case with
controls, with the DESIGN MODE turned on (click the DESIGN MODE button on the
DEVELOPER tab, see Fig. 4.1 and Fig. 1.1 of Sect. 1.7), the PROPERTIES can be
changed, e.g., name and caption. When the design mode is switched off (by
clicking the DESIGN MODE button again), the control can be operated.
The procedure SUB COMMANDBUTTON1_CLICK in Fig. 4.10a (P) is assigned to

the command button. In detail, proceed as follows: In the VISUAL BASIC Edi-
tor (Fig. 4.9), click on the arrow ▿ at (GENERAL). A list opens up in which
SUB COMMANDBUTTON1 appears. This entry is activated by clicking on it. Next,
click on the arrow ▼ next to the cell with the inscription “Annegret” (as shown in
Fig. 4.9, or the name you have chosen). A list opens up in which CLICK appears,
together with other commands. Click on this entry, and SUB COMMANDBUTTON1
is immediately completed to SUB COMMANDBUTTON1_CLICK. The upper line in the
VBA editor will now read COMMANDBUTTON1 ▼; CLICK().
In our case, in Fig. 4.10a (P), only one procedure, SUB Circles, is called. We
could omit SUB COMMANDBUTTON1_CLICK by naming the command button Circles
and completing SUB Circles() to SUB Circles_CLICK(). Please note that we would
have to change the name of the button, which is independent of its caption.
4.4 A Checkerboard Pattern (Excel)
We obtain the VBA commands for drawing elementary geometric shapes

by recording macros and incorporate them into a sub-routine that executes
the drawing in the desired layout. The shapes’ position in the spreadsheet is
passed to the procedure via its header or via global variables. The respec-
tive shapes, as well as the color of their borders and interiors, are selected
randomly.
4.4.1 Checkerboard, Same-Colored and Multi-colored
Checkerboard with same-colored shapes

In this exercise, a checkerboard of rectangles, triangles, and circles, as shown in
Fig. 4.12a, is to be drawn. The procedure for this is shown in Fig. 4.13 (P).
SUB DRAWI1 in Fig. 4.13 (P) is the main program that randomly calls one of the sub-
routines Rect, Ova, or Tria, ten times in each of eight rows, drawing a rectangle,
ellipse, or triangle at the current position of x and y. Its core is a nested loop with
two loop indices, k for the row and i for the column address within a row.
The variable ROT in SUB drawi1 determines whether a rectangle, an ellipse
(oval) or a triangle shall be drawn. In line 5, the variable ROT is randomly assigned
a value 0, 1, or 2. Chance is brought in by the function RND() generating a random
number between 0 and 1, which is then multiplied by 3. This real number is turned
into an integer by INT (into the variable ROT). To give some examples: INT(0.75*3)
= INT(2.25) = 2; INT(0.22*3) = INT(0.66) = 0; INT(0.54*3) = INT(1.53) = 1.
4.4 A Checkerboard Pattern (Excel) 133
Fig. 4.12 a (left) A checkerboard pattern of rectangles, circles and triangles, all equally formatted,
drawn with SUB drawi1 in Fig. 4.13 (P). b (right) Like a, but with forms differently formatted, filled
with different colors, and surrounded with borders of different thickness and different color, drawn
with SUB drawi in Fig. 4.14 (P)
1 Sub drawi1() If ROT = 2 Then Call Tria(x, y) 8

2 x = 100 x = x + 15 9
3 For k = 1 To 8 'next row Next i 10
4 For i = 1 To 10'within row x = 100 'reset left position 11
5 ROT = Int(Rnd() * 3) '0, 1 or 2 y = y + 15 'advance top position 12
6 If ROT = 0 Then Call Rect(x, y) Next k 13
7 If ROT = 1 Then Call Ova(x, y) End Sub 14
Fig. 4.13 (P) Procedure SUB drawi1 with which Fig. 4.12a is drawn
In lines 6 to 8, logical IF queries determine which shape is drawn. After the

shape has been drawn, the x value is increased by 15 (line 9).
Task First, draw only one row by omitting the loop (FOR k = …)! The sub-
routines Rect(x,y), Ova(x,y) and Tri(x,y) should be written according to the model
of sub Disc(x,y) in Fig. 4.10b (P). Apart from MSOSHAPEOVAL, MSOSHAPERECTANGLE
and MSOSHAPETRIANGLE have to be used.
Task Draw the complete checkerboard pattern!
A randomly more colored checkerboard pattern

We draw 8 rows with 10 shapes each, such as shown in Fig. 4.12b. The format of the
shapes, namely, the color to be filled in and the color and thickness of the border, is
now determined using a random number. In addition, the positions (left, top) of the
shapes are not passed to the sub-routines through the procedure header, but via global
variables. The main procedure can be found in Fig. 4.14 (P), a typical sub-routine in
Fig. 4.15 (P).
1 Private x, y As Single If ROT = 0 Then Call Rect 10

2 'Position of the shape to be currently drawn If ROT = 1 Then Call Ova 11
3 If ROT = 2 Then Call Tria 12
4 Sub drawi() Next i 13
5 x = 100 'left position x = 100 'reset left position 14
6 y = 100 'top position y = y + 15 'advance top position 15
7 For k = 1 To 8 'next row Next k 16
8 For i = 1 To 10 'within row End Sub 17
9 ROT = Int(Rnd() * 3) '0, 1 or 2 18
Fig. 4.14 (P) SUB drawi is the main program calling, 10 times in each of 8 rows, one of the sub-
routines Rect, Ova, or Tria, which draw a rectangle, ellipse, or triangle at the current position of x
and y. Similar to Fig. 4.13 (P), but with parameters stored in global variables x, y defined in line 1
1 Sub Rect() ' Draws a rectangle of width 10 and height 10 1

2 ActiveSheet.Shapes.AddShape(msoShapeRectangle, x, y, 10, 10).Select 2
3 Call Lin(255 * Rnd(), 63 * Rnd(), 63 * Rnd(), 10) 'Rim of shape 3
4 'Formats rim: Red fully varied, green and blus, half intensity, strength 4
5 Call Interi(63 * Rnd(), 128 + 127 * Rnd(), 63 * Rnd()) 'Interior of shape 5
6 'Formats interior: Green always more intense than 50% 6
7 x = x + 15 'Advances position in the row 7
8 End Sub 8
Fig. 4.15 (P) SUB Rect draws a rectangle with a fixed size, but with randomly selected colors for
the border (line 3) and interior (line 5). The instructions for coloring are executed in the Lin and
Interi sub-routines in Fig. 4.16 (P)
4.4.2 Global Variables
Global variables are also valid in sub-routines. They must be declared as PRIVATE
or PUBLIC before the routines (see line 1 in Fig. 4.14 (P)). Variables of type PRIVATE
are only available in the module in which they are declared, those of type PUBLIC
in the whole workbook. The positions (Left, Top) of the shapes are now stored in
global variables x and y that can be read and modified by each sub-routine.
The data type SINGLE in line 1 of Fig. 4.14 (P) denotes a single-precision
floating-point number stored in 4 bytes. Decimal numbers of the data type DOUBLE
are stored in 8 bytes. You can find out more about other data types with EXCEL
help in the VBA-Editor.
Don’t just copy the macros if you already have some programming
practice! Rehearse the sequence of instructions in your mind and get
the commands for drawing shapes through macro recording!
In SUB Rect in Fig. 4.15 (P), a square is drawn, and its interior and border are for-
matted with the sub-routines Lin and Interi in Fig. 4.16 (P). The position is taken
from the global variables (x, y) and passed to MSOSHAPERECTANGLE through the
procedure header of AddShape. Both side lengths are fixed to 10. The arguments
4.4 A Checkerboard Pattern (Excel) 135
1 Sub Lin(r, g, b, w) Sub Interi(r, g, b) 8

2 red, green, blue and weight w 'red, green, blue 9
3 With Selection.ShapeRange.Line With Selection.ShapeRange.Fill 10
4 .ForeColor.RGB = RGB(r, g, b) .ForeColor.RGB = RGB(r, g, b) 11
5 .Weight = w End With 12
6 End With End Sub 13
7 End Sub 14
Fig. 4.16 (P) SUB Lin and SUB Interi color the border and the interior of the shape (line 2), respec-
tively, according to the variables r (red), g (green), and b (blue). The thickness (weight) of the
border is specified in w, set to 10 in Fig. 4.15 (P)
in the headers of Lin and Interi are generated in SUB Rect(...) with RND() which
returns a random number between 0 and 1.
Questions
concerning Fig. 4.15 (P):

The procedure header of SUB Rect is empty. How does the sub-routine know
the position in the spreadsheet where the rectangle is to be drawn?14
Which color dominates the border of the shapes?15
4.5 A Checkerboard Pattern (Python)
We draw a multi-colored checkerboard pattern using the Python library

turtle, setting up a similar program structure as for EXCEL in Sect. 4.4, but
considering the differences in code from Visual Basic.
4.5.1 Turtle
In order to draw a set of shapes with Python, we use the library turtle. This
simple plot program’s illustrative idea is that of turtles running across the screen,
thereby creating colorful traces.
In Table 4.1, the libraries turtle and numpy.random are imported and a
turtle named t is created. This will be the first cell in every program; it must
be run before the functions are compiled, because they resort to these libraries.
Generally, several turtles can be active at the same time. We use, however, only
14 SUB Rect accesses the global variables x, y.

15 Line 5 in Fig. 4.15 (P): Green = 128 + 127*RND() is represented at least with strength 128.
Table 4.1 Importing relevant libraries; creating a screen with a title; the internet address in line
1 points to an introduction to Turtle graphics
1 #https://docs.python.org/3.3/library/turtle.html
2 import turtle
3 import numpy.random as npr
4 t=turtle.Turtle() #Creates a turtle with name t
5 turtle.title("Checkerboard")
Fig. 4.17 a (left) Turtle screen created by the Python program in this section, in a frame spanned
by the points (−400, −400) and (400, 400). b (right) Triangles of different size and at different
positions in a square
one turtle instance called t. In the second cell in Table 4.1, a screen is created
with a program-specific name, here, Checkerboard.
The standard size of the screen is 1000 pt × 800 pt spanned between (−500,
−400) to (500, 400). Such a screen with a checkerboard pattern is shown in
Fig. 4.17a. Turtle and EXCEL apply coordinate systems with different origins.
In EXCEL, it is at the upper left corner of the spreadsheet, and all coordinates are
positive. In Turtle, the origin is at the center of the screen, and the coordinates
are positive or negative.
Attention: The turtle window may be below the Python window.
Different types of triangle

The centroid (x C , yC ) of a triangle, corresponding to its center of gravity, is calculated
as
x1 + x2 + x3 y1 + y2 + y3
xC = , yC =
3 3
It is marked in Fig. 4.17b with dots.
The function drawTria, used to draw Fig. 4.17a, draws an equilateral triangle
(bottom left in Fig. 4.17b) whose centroid does not coincide with the square’s center,
4.5 A Checkerboard Pattern (Python) 137
in contrast to the equilateral triangle at top left. The bottom right triangle has its top
point at the midpoint of the upper side of the square. It is shifted in the upper right
figure so that its centroid coincides with the square’s center.
Basic functions and measurements

The basic functions (always completed with parentheses ()), attributed to the instance
of a turtle, e.g., t.penup() or t.pos(), are.
Settings:
.pen( … ) specifying speed, pensize, pencolor, fillcolor

.pu() pen up
.pd() pen down
.setpos(x,y) moving to the specified position
.setheading(φ) setting the orientation as angle φ in ° with respect to the
x-axis
.rt(φ) right turn by angle φ in °
.lt(φ) left turn by angle φ in °
.fd(r) step forward by r pixels
.bk(r) step backward by r pixels
.dot(s,c) plots a dot with diameter s and color c at current position
.begin_fill() beginning to fill in the contour
.end_fill() ending to fill in the contour
Measurements:
.pos() returns Cartesian coordinates of turtle

.heading() returns angle of direction
.distance(x,y) returns distance to point (x, y)
Question
How do you get the position of a turtle named doro in polar coordinates? There
are two possibilities.16
4.5.2 Differences to Visual Basic
We will demonstrate some syntactic differences between Python and

VISUAL BASIC by means of the function drawSquare, which creates a square shape.
16r = doro.distance(0,0), phi = doro.heading()/180*np.pi() or x

= doro.pos(0), y = doro.pos(1), r = np.sqrt(x**2 + y**2), phi =
np.arctan2(y,x)
Table 4.2 Function drawSquare in Python, drawing a square with center (x, y) and size sz that
is filled if the keyword variable fill is set to True
1 def drawSquare(x,y,sz,fill=False):
2 r=sz/2
3 t.pu() #Pen up!
4 t.setpos(x-r,y-r) #Run to position!
5 t.pd() #Pen down!
6 if fill==True: t.begin_fill()
7 for i in range(4):
8 t.fd(sz) #Forward!
9 t.lt(90) #Turn left!
10 if fill==True: t.end_fill() #Fill square with color!
11 t.pu()
12 t.setpos(x,y)
In Table 4.2, it is implemented in Python with four segments of equal length

and a 90° turn after each segment; the features of the pen have been specified
in a superordinate program. In Fig. 4.18 (P), this is realized in a Visual Basic
sub-routine that is called from the main program SUB DSQ().
Positional and keyword arguments

The procedure headers in both cases, Python and VBA, contain positional arguments
x, y, sz, and a keyword argument fill that decides whether or not the contour is
filled (with the default set to False).
The first three positions in the header must contain appropriate values when the
function is called. It is the position in the header that determines the variable in the
procedure to which the value is assigned. In Fig. 4.18 (P), a variable y with value
40 is passed through the third position to the procedure, where it is assigned to a
variable named sz.
The keyword argument fill is defined in the procedure header with a default value
taken in the procedure, if not specified otherwise.
1 Dim col(2) As Integer Sub drawSquare(x, y, sz, _ 13

2 Optional fill As Boolean = False) 14
3 Sub dSq() ActiveSheet.Shapes.AddShape( _ 15
4 col(0) = 100 msoShapeRectangle, x, y, sz, sz).Select 16
5 col(1) = 200 Selection.ShapeRange.fill.Visible = msoFalse 17
6 col(2) = 0 If fill = True Then 18
7 y = 40 With Selection.ShapeRange.fill 19
8 For a = 100 To 200 Step 50 .Visible = msoTrue 20
9 Call drawSquare(a, 100, y, fill:=True) .ForeColor.rgb = rgb(col(0), col(1), col(2)) 21
10 Next a End With 22
11 Call drawSquare(100, 100, y, fill:=True) End If 23
Fig. 4.18 (P) DrawSquare realized in visual basic for EXCEL

Global parameters
The first line in the VISUAL BASIC program in Fig. 4.18, before the procedures, defines
an array col of integers with three elements accessible in the whole module. It is
written in the main program on lines 4 to 6 and read in drawSquare on line 21.
In Python, arrays can be declared anywhere in the program, e.g., by col = [100,
200, 0]. All variables, as well as arrays, are valid in subordinate functions unless the
variable name is again declared in a function with an equal sign, e.g., col = [10, 100,
10], creating a new object with own memory space.
Grouping blocks of code

In Python, indentation has a syntactic function. A block of code is necessar-
ily grouped by the same amount of indentation. In Table 4.2, there are two examples:
the statements of the function are all indented by 4 spaces; the two statements to be
executed in the for-loop are further indented by another 4 spaces.
EXCEL uses code words to state the end of a block. In Fig. 4.18 (P), the procedure
(sub-routine) ends with END SUB, the IF block with END IF, and the WITH block with
END WITH. The (FOR A = ) block in the main program ends with NEXT A. Although
indentation does not have a syntactic function in Visual Basic, we use it to maintain
a better overview of the program structure.
Case-sensitivity
Python is case-sensitive: True and False have to be written with capital T and F;
x and X are two different variables. Visual Basic is case-insensitive: an input “true”
is automatically changed to “TRUE”; x and X are regarded as the same variable. When
we change the case of the first letter anywhere in the program, names in other places
will automatically adapt.
4.5.3 Checkerboard with Squares, Triangles, and Circles
User-defined functions for square, triangle, circle, dash

To draw a checkerboard pattern as in Fig. 4.17a, we need functions that draw, besides
the rectangle already realized in Table 4.2, a triangle and a circle. They should have
the same procedure header as drawSquare, with the center point (x,y) and the size
sz as positional arguments and fill as a keyword argument.
Drawing a triangle in Table 4.3 is similar to how it is done in drawSquare. Drawing
a circle with radius r is achieved with the built-in function circle(r) (see Table
4.4). There, the turtle runs along a circle of radius r, starting at its current position
and with its current direction.
In the three functions mentioned above, the turtle runs along a shape, starting
at its current position and with its current direction. It was, however, intended by
the programmer that the turtle start running straight to the right. This is indeed
Table 4.3 Drawing a triangle with center (x, y)

13 def drawTria(x,y,sz,fill=False):
14 r=sz/2
15 t.pu()
16 t.setpos(x-r,y-r) #Left lower edge
17 t.pd()
20 t.fd(sz) #Forward!
21 t.lt(120) #Turn left!
22 if fill==True: t.end_fill()
23 t.pu()
24 t.setpos(x,y)
Table 4.4 Drawing a circle with center (x, y)

25 def drawCircle(x,y,sz,fill=False):
26 r=sz/2
27 t.penup()
28 t.setpos(x,y-r)
29 t.pendown()
31 t.circle(r) #Is a function within turtle
32 if fill==True: t.end_fill()
33 t.penup()
34 t.setpos(x,y)
assured in our current main program, but nevertheless, not making this intention
explicit is considered a big mistake in Software Engineering.
Questions
The turtle named t in Tables 4.2, 4.3, and 4.4 is a global instance accessed
within the functions. This is possible because you have only one turtle running.
What do you do if several turtles are on the field?17
In Table 4.4, it is implicitly assumed by the programmer that the turtle is
heading straight to the right at start, a big programming mistake. How do you
avoid this bug?18
17 The turtle name has to be an argument, e.g. t, def drawSquare(t, x, y, sz, fill
= false), so that it is no longer regarded as a global instance in the functions.
18 Introduce t.setheading(0) before the turtle starts running. If necessary, store the orig-
inal direction at the beginning, e.g., phi0 = t.heading() and reset it at the end with
t.setheading(phi0).
Table 4.5 Drawing a dash

35 def dash(ds):
36 t.rt(90) #Right turn
37 t.fd(ds)
38 t.rt(180) #180° to the right
39 t.fd(2*ds)
40 t.rt(180)
41 t.fd(ds) #Back to zero
42 t.lt(90) #90° to the left, original direction
Table 4.6 Drawing the checkerboard pattern with a nested loop; when lines 3 and 20 are activated,
the turtle runs faster
1 turtle.clearscreen()
2 t=turtle.Turtle()
3 #turtle.tracer(0, 0)
4 #Draw checkerboard pattern
5 for rn in range(-280,285,80):
6 for c in range(-360,365,80):
7 r=npr.rand()
8 g=npr.rand()
9 b=npr.rand()
10 tup=(r,g,b) #Red, green, blue
11 tup2=(g,b,r) #g=Red, b=green, r=blue
12 t.pen(pencolor=tup2, fillcolor=tup,
13 pensize=4, speed=0)
14 if r<0.33:
15 drawTria(c,rn,60,fill=True)
16 elif r<0.67:
17 drawSquare(c,rn,60,fill=True)
18 else: drawCircle(c,rn,60,fill=True)
19 t.dot()
20 #turtle.update()
The function dash in Table 4.5 draws a dash at the current position perpendic-
ular to the current turtle heading, of extension ds to both sides.
Main program
The main program has three parts. The first part is shown in Table 4.1, importing the
necessary libraries and creating a screen with the title Checkerboard. Then, in Table
4.6, the checkerboard pattern is drawn with a nested loop over 10 x positions and 8
y positions, with randomly choosing one of our three shapes.
Before we call the draw* functions, the pen specifications have to be set in the main
program, pencolor and fillcolor in our program, by rgb (red, green, blue) in
standard mode with values between 0 and 1, randomly chosen with npr.rand().
Table 4.7 Drawing a frame with dashes around the checkerboard

21 #Draw frame with dashes
22 t.pen(pencolor="black", pensize=1, speed=10)
23 t.penup()
24 t.setpos(-400,-400) #Left bottom corner
25 t.pendown()
26 for k in range(4): #4 straight lines
27 for i in range(10): #10 segments with dashes
28 dash(6)
29 t.fd(80) #Forward!
30 dash(6)
31 t.lt(90) #Left turn 90°!
32 #turtle.update()
Table 4.8 Drawing the axes of the coordinate system

33 #Axes of the coordinate system
34 t.pu(), t.setpos(-400,0), t.pd()
35 t.fd(800) #Forward!
36 t.pu(), t.setpos(0,-400), t.pd()
37 t.setheading(90) #Direction 90° to x axis
38 t.fd(800)
39 #turtle.update()
The keyword variable pencolor expects a tuple with 3 elements to specify the
color. The individual variables r (red), g (green), b (blue) are set in lines 7 to 9 and
assembled into two different tuples, tup for fillcolor and tup2 for pencolor.
Frame around the figure in Turtle

The main program is continued in Tables 4.7 and 4.8, through drawing of a frame
around the checkerboard and the axes of the coordinate system, respectively.
Questions
What type of triangle is specified in drawTria, Table 4.3: equilateral, or acute-

angled?19
How do you draw a triangle touching a square at two neighboring corners
and the center of the opposite side (see Fig. 4.17b, bottom right)?20
19 The function drawTriangle draws a triangle with all angles equal to 60° so that it becomes
equilateral.
20 Let the turtle run with setpos(..) along (x-sz/2,y-sz/2), (x + xz/2,
y-sz/2), (0, y + sz/2), back to (x-sz/2,y-sz/2)!

What is the direction of the turtle after having been guided by the code
snippet in Table 4.7?21
How do we speed up Python’s turtle function?

The answer is found with an internet search (2020):
https://stackoverflow.com/questions/16119991/how-to-speed-up-pythons-turtle-
function-and-stop-it-freezing-at-the-end
with the answers:
– (1) Set turtle.speed() to fastest.

– (2) Use the turtle.mainloop() functionality to do work without screen
refreshes.
– (3) Disable screen refreshing with turtle.tracer(0, 0), then, at the end,
do turtle.update()
We trigger variant (3) when we activate lines 8 and 25 in Table 4.6. The instruction
turtle.tracer(0,0) eliminates the millisecond delays that occur when the
screen is updated after every turtle change. The screen is refreshed with the complete
picture by turtle.update() (Table 4.8).
4.6 Drawing Densely-Packed Atomic Layers; Crystal Physics
We draw two different stackings of three planes with densely packed spheres
that correspond to the cubic face-centered (fcc) or hexagonal dense-packed
(hdp) crystal structure.
4.6.1 Program Structure and Geometry
In this task, a top view of a close-packed plane of atoms is to be drawn with a

program based on a nested For-loop. The subordinate loop draws a horizontal row
of discs, representing the atoms, touching each other. This is achieved by shifting
a new disk to the right by the diameter d relative to the previous disc. In a higher-
level loop, the rows are to be shifted one after the other in the plane so that the
circular disks touch each other in a hexagonal arrangement, as shown in Fig. 4.19.
In a second development step, the main program resulting from the first step is
to be converted into a procedure that draws a plane and to which the coordinates
21 The turtle turns 90 + 180 + 180 – 90 = 360° = 0° toward its original direction.
A B C D E F G H
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
Fig. 4.19 (S) Hexagonally packed plane, drawn with SUB DIEB from Fig. 4.22 (P), here, however,
in half-size
Fig. 4.20 Two densely packed planes and only two “atoms” in the third plane; dark grey “atom”
with dotted border top left: position as in the hexagonal close packing (hcp); middle grey “atom”
with black border: position as in the face-centered cubic (fcc) shape; the drawing is obtained with
Tables 4.11 and 4.12
of the first disc are passed. An extended main program puts a second plane onto
the gaps in the first plane, and ultimately places two discs, one at a position typ-
ical of the face-centered cubic (fcc) structure and the other one characteristic of
hexagonal densest packing (hdp), resulting in Fig. 4.20.
Questions
How many neighbors does a sphere in a close-packed plane have?22
22 Six nearest neighbours.

4.6 Drawing Densely-Packed Atomic Layers; Crystal Physics 145
Fig. 4.21 a (left) Geometry of a hexagonal packing in a plane, displacement of the second row of
atoms with respect to the first row of atoms. b (right) Position of an atom in the second plane
How many neighbors does a sphere in a stack of close-packed planes have?23
The drawings in Fig. 4.21 indicate the coordinates of the centers of the circular
disc in the planes. Figure 4.21a gives the position of a disc in a row relative to
the previous row, and Fig. 4.21b that of a disc in a plane relative to the previous
plane.
Three stacked planes

The first row of circular disks starts at (x 0 , y0 ). The second row is offset from the first
row in the x- and y-directions by distances indicated in the geometric construction in
Fig. 4.21a. A second plane is to be placed over the first one, with the circular disks
lying over the gaps in the first plane (see Fig. 4.21b), resulting in the assembly of
light grey discs in Fig. 4.20.
For the third plane, there are two possibilities:
– It lies exactly above the first plane, as in the crystal structure of hexagonal close
packing (hcp). For the drawing in Fig. 4.20, only one circular disk is placed in
the correct position (dark grey, top left in the picture).
– It lies above the still visible gaps in the first plane, as in the cubic face-centered
(fcc) crystal structure. In Fig. 4.20, this is done only for one circular disk,
drawn in middle grey with a black border. The displacement of the third plane
in the x- and y-directions is, for fcc, twice as large as the displacement of the
second plane, both with respect to the first plane.
23 12 nearest neighbours, 6 of them in its own plane, 3 below and 3 above.

The positions of the discs in the following list are deduced from Fig. 4.21:
d disc diameter
(x 0 , y0 ) position of the first disc, top left
(delX, delY ) shift of a plane with respect to (x0 , y0 )
delX = 0, delY = 0 for the 1st plane
delX = d/2 for the 2nd plane (see Fig. 4.21b)
delY = −d/2·tan(30/180·π) for the 2nd plane (see Fig. 4.21b)
x = x 0 + delX + dx + i·d position of disc i.
dx = delX + 0 1st and 3rd rows (see Fig. 4.21a)

= delX + d/2 2nd and 4th rows (see Fig. 4.21a)
y = y0 + dy y position of disc i (see Fig. 4.21a)
dy = delY + d 34 ·n nth row (see Fig. 4.21a and Eq. 4.1)
with:

d34 = d · 3/4 (4.1)
read from Fig. 4.21a.
4.6.3 Excel
A row of discs
We draw four rows of circular disks so that the disks touch each other, using four
consecutive FOR-loops (see SUB DiEb Fig. 4.22 (P)).
The sub-routine SUB Disc(x,y), already reported in Fig. 4.10b, is called with
individual positions x, y passed via the procedure header. Contrary to the figure,
the disc’s diameter is set in the main procedure to d = 100. In each row, the x-
position of the following circle is shifted to the right by a circle’s diameter. The
y-position is always the same for a row.
Stacking planes
To draw several planes, one on top of the others, we convert the main program DiEb
described above into the sub-routine SUB Plane(delX, delY) (repeated in Fig. 4.23
(P)) to which the coordinates of the top left disc are transferred. The displacements
delX and delY can be determined with the help of Fig. 4.21b.
1 Sub DiEb() 15
2 x0 = 100 '3rd row of discs 16
3 y0 = 100 dx = 0 'shift with respect to 1st row 17
4 d = 100 'diameter of the disc dy = d * Sqr(3 / 4) * 2 18
5 '1st row of discs For i = 0 To 7 19
6 For i = 0 To 7 Call Disc(x0 + dx + i * d, y0 + dy) 20
7 Call Disc(x0 + i * d, y0) Next i 21
8 Next i '4th row of discs 22
9 '2nd row of discs dx = d / 2 'shift with respect to 1st row 23
10 dx = d / 2 'shift with respect to 1st row dy = d * Sqr(3 / 4) * 3 24
11 dy = d * Sqr(3 / 4) For i = 0 To 6 25
12 For i = 0 To 6 Call Disc(x0 + dx + i * d, y0 + dy) 26
13 Call Disc(x0 + dx + i * d, y0 + dy) Next i 27
14 Next i End Sub 28
Fig. 4.22 (P) SUB Dieb for drawing a plane; the four rows of atoms are drawn using four loops,
result shown in Fig. 4.19 (S)
1 Sub Plane(delX, delY) 15

2 x0 = 100 'Offset to left upper corner dx = 0 + delx 16
3 y0 = 100 'of the worksheet dy = d * Sqr(3 / 4) * 2 + dely 17
4 d = 100 'Diameter of the disc For i = 0 To 7 18
5 dx = delx Call Disc(x0 + dx + i * d, y0 + dy) 19
6 dy = dely Next i 20
7 For i = 0 To 7 dx = d / 2 + delx 21
8 Call Disc(x0 + dx + i * d, y0 + dy) dy = d * Sqr(3 / 4) * 3 + dely 22
9 Next i For i = 0 To 6 23
10 dx = d / 2 + delx Call Disc(x0 + dx + i * d, y0 + dy) 24
11 dy = d * Sqr(3 / 4) + dely Next i 25
12 For i = 0 To 6 End Sub 26
13 Call Disc(x0 + dx + i * d, y0 + dy) 27
14 Next i 28
Fig. 4.23 (P) SUB DiEb is converted into a sub-routine Plane to which the initial coordinates are
transferred by a higher-level program
In the main program SUB hcp fcc in Fig. 4.24 (P), SUB Plane is called twice,
for the initial layer with delX = 0 and delY = 0 and for the second layer with its
1 Sub hcp_fcc() dy = 0 10
2 x0 = 100 dx = 100 * 1 11
3 y0 = 100 Call Disc(x0 + dx, y0 + dy) 12
4 Call Plane(0, 0) 'fcc 13
5 delx = 100 / 2 dy = 100 * Sqr(3 / 4) - 100 / 2 * Tan(30 _ 14
6 dely = 100 / 2 * Tan(30 _ / 180 * 3.14159265) 15
7 / 180 * 3.14159265) dx = 100 * 5 16
8 Call Plane(delx, dely) Call Disc(x0 + dx, y0 + dy) 17
9 hdp End Sub 18
Fig. 4.24 (P) Main program, which calls SUB Plane twice, places two atoms on top (with SUB Disc
from Fig. 4.10b (P)), and thus draws a picture similar to Fig. 4.20
discs on the gaps of the first layer. There are two possibilities for the third plane,
hcp or fcc, represented with one disc each in lines 12 and 17.
Questions
In Fig. 4.23 (P), the variables in the header are called delX and delY. In the
body of the procedure, the formulas refer to different names, delx and dely.
Will this discrepancy lead to error messages?24
How do you have to change SUB Disc in Fig. 4.10b (P) so that d becomes a
global variable?25
Grouping and copying shapes in various picture formats

To group the shapes into an integrated picture, activate the white arrow
SELECT OBJECTS in the FIND&SELECT tab (far right in the HOME tab of the EXCEL rib-
bon, Fig. 1.1 in Sect. 1.7), drag the selection rectangle around the shapes, then click
FORMAT/GROUP. You can now copy the group as one graphic into other applications,
e.g., into a Word file or a PowerPoint file.
Task Group your drawing into an image of type png, tif , or some other image
format, and copy this image to another area of the spreadsheet or to another appli-
cation, e.g., to a PowerPoint file! To do so, select the object, click COPY, move
the cursor to another location in the table, click PASTE/PASTE SPECIAL, and select the
desired format.
4.6.4 Python
In the first cell of Table 4.9, the relevant libraries are imported. In the next cell,
a screen with the name “Crystal planes” is created, and global parameters are
specified, with the disc diameter d being set to 50 and the position of the first disc
in the upper left corner at (x 0 , y0 ) = (−175, 175). In the third cell, the first row of
discs is drawn, comprising eight discs drawn from left to right, using the function
drawCircle in Table 4.4. When developing the program, you should check this
snippet of code and see whether a row of gray discs is really plotted from left to
right.
Comments concerning Table 4.9:
– Line 1, link for an introduction into the basic features of turtle,

– Line 5, the screen is cleared. This is important when developing a program,
and you have to improve the code and repeat it again and again until it runs
error-free.
24No, Visual Basic is case-insensitive, contrary to Python.

25Skip line 28 “d = 50”, insert dim d AS INTEGER as the first program line, and specify d = …
somewhere in the program before SUB Disc is called for the first time.
Table 4.9 Importing relevant libraries; creating a screen and setting global parameters; drawing
1st row of discs, function disc from Table 4.4
1 #https://docs.python.org/3.3/library/turtle.html
2 import turtle
5 turtle.clearscreen()
6 t=turtle.Turtle() #Create turtle with name t!
7 turtle.title("Crystal planes")
8 tup=(0.9,0.9,0.9) #Light grey
9 t.pen(pencolor="black", fillcolor=tup, pensize=1, speed=10)
10
11 x0=-175
12 y0=175
13 d=50 #Disc diameter
14 #1st row of discs
15 dx=0
16 dy=0
18 drawCircle(x0+dx+i*d,y0+dy,d,fill=True)
19 print(t.pos()) #Current position
Table 4.10 Function for drawing a close-packed plane; delX, delY position of top left disc with
respect to (x 0 , y0 ), drawCircle from Table 4.4
1 def Plane(delX,delY):
2 #1st row of discs
3 dx=0+delX
4 dy=0+delY
7 #2nd row of discs
8 dx=d/2+delX
9 dy=-d*np.sqrt(3/4)+delY
12 #3rd row of discs
13 dx=0+delX
14 dy=-d*np.sqrt(3/4)*2+delY
17 #4th row of discs
18 dx=d/2+delX
19 dy=-d*np.sqrt(3/4)*3+delY
Table 4.11 Plotting the first and second planes

22 Plane(0,0)
23 delX=50/2
24 delY=-50/2*np.tan(30/180*np.pi)
25 Plane(delX,delY)
Questions
concerning Table 4.9

What is the position of the turtle after having drawn the first row?26
What are the pen color and fill color of the turtle?27
What is the global parameter accessed in drawCircle?28
To produce a figure like that in Fig. 4.20, two planes and, additionally, two
discs have to be placed at appropriate positions. The function for drawing a close-
packed plane is shown in Table 4.10. Its arguments are the shifts delX and delY of
the first disc’s position with respect to (x 0 , y0 ). The four rows of discs are drawn
with loops calling drawCircle (from Table 4.4), always with the same variables,
but with dx and dy set for each row individually, according to Eq. (4.1).
The first two planes are drawn with the program in Table 4.11. Two atoms in the
third plane representative of the hexagonal-dense packed and face-centered cubic
structures, respectively, are drawn in Table 4.12. For hdp, the position is just on
top of a disc in the first plane, whereas for fcc, x and y get a double shift from a
disc in the first plane, whereby the x position becomes identical to the neighboring
disc in the first plane and the y position is in the center of a gap in the second
plane.
Question
In line 29 of Table 4.12, dx = 50*1 is specified; the number 50 is precisely

the disc diameter. Nevertheless, what is bad about this instruction, and likewise
about the instructions in lines 36 and 37, and lines 23 and 24 in Table 4.11?29
26 The position of the turtle after the first row is x = -175 + 7·50 = +175; y = 175, as can be
deduced from lines 18 and 19 in Tab. 4.9.
27 pencolor = "black", fillcolor = tup, tup = (0.9, 0.9, 0.9), a light gray.
28 The disc diameter d is a global parameter accessed in drawCircle. It is also used in line 18
to specify the shift of the position of the current disc with respect to the preceding one.
29 When the value of the global parameter d is changed, these instructions do not follow. Set dx =
d*1 in lines 47, 48, 53, 60, 61 instead!

4.7 Text Processing 151
Table 4.12 Drawing an atom of the third plane, either hdp or fcc
26 #hdp
27 tup=(0.3,0.3,0.3) #Dark grey
28 t.pen(pencolor="black", fillcolor=tup,pensize=1, speed=0)
29 dx=50*1
30 dy=0
31 drawCircle(x0+dx,y0+dy,d,fill=True)
32
33 #fcc
34 tup=(0.6,0.6,0.6) #Middle grey
35 t.pen(pencolor="black", fillcolor=tup,pensize=1, speed=0)
36 dx=50*5
37 dy=-50/2*np.tan(30/180*np.pi)*2
38 drawCircle(x0+dx,y0+dy,d,fill=True)
4.7 Text Processing
Operations on strings (texts) are practiced by the example of swapping letters

within words. We will use the functions Len, Split, Join in both Visual Basic
and Python. The functions LEFT, RIGHT, and MID are used in VISUAL BASIC to
cut out pieces of strings, achieved in Python through list slicing.
4.7.1 Cutting and Joining Strings
Swirling characters
Text It is siad taht a txet can be undesrtood eevn if you lvaee olny the begiinnng
and the end lteter in ecah wrod in plcae but exahcnge middle lettsre. Do you
beileve taht or is it nonnesse?
Task Write a pogrram that fsirt rades a text from a sersadehept. This text is then
to be bokren down into wsodr. The idnuvidial words are tnorsfarmed so that the
frsit and last letrets raiemn in pcela, but the iennr ltrtees are ramlondy swapped.
We are going to write a program that exchanges letters in an originally correctly

written text. Such a program has already swirled the first two paragraphs of this
description. The number of letter swappings in the above text: 1 (Text), 2 (Task).
Plain text of the first two paragraphs
Text It is said that a text can be understood even if you leave only the beginning
and the end letter in each word in place but exchange middle letters. Do you
believe that, or is it nonsense?
Task Write a program that first reads a text from a spreadsheet. This text is then
to be broken down into words. The individual words are transformed so that the
first and last letters remain in place, but the inner letters are randomly swapped.
VBA instructions for text processing

We need the following VBA instructions that affect character strings:
LEN(…), SPLIT(…), JOIN(…) , LEFT(…), RIGHT(.) and MID(…).
An overview is given in the box. For more information, refer to VBA help! There,
you can find out, for example, about the MID function:
Returns a Variant (String) containing a specified number of characters from a string.
Syntax MID (STRING, START [, LENGTH]).
STRING Required. String expression from which characters are returned.

START Required; Long. Character position in STRING at which the part to be
taken begins.
LENGTH Optional; Variant (Long). Number of characters to return.
Text processing in VBA and Python

VISUAL BASIC Python
SPLIT(STRING, [SEPARATOR…) String.split(“separator”)
Splits a string expression into words and stores them in an array. Unless
otherwise specified, a space is interpreted as a separator between words.
LEN(STRING) len(String)
Determines the length of a given string.
LEFT(STRING, LENGTH) String[:Length]
Cuts out a piece of length Length from a character string starting from the
left.
RIGHT(STRING, LENGTH) String[-Length:]

Cuts out a piece of length Length from a character string starting from the
right.
MID(STRING, START, LENGTH) String[Start: Stop]
Cuts out a piece of length LENGTH or (Stop – Start) from a charac-

ter string starting from position START and to the right of it. Attention:
VISUAL BASIC starts indexing from 1, Python from 0.
JOIN(ARRAY[, DELIMITER) ‘Delimiter’.join(Array)
Returns a string created by joining several substrings contained in an array.

If ‘Delimiter’ is omitted, the space character (“ ”) is used.
Questions
Let the variable Tx contain the text “Cutting out”. With which instructions do
you get the first, last, and second characters of Tx? How do you copy the string
“ing” from Tx to a new variable Wd? In Visual Basic,30 in Python?31
Consider the string Sente = “This is. Our goal.” How do you separate the
string into the two sentence fragments terminated by full stops? Use the Split
command! How do you get a new string JS = “This is.” including a full stop?
EXCEL?32 Python?33 Compare!
4.7.2 Data Structure and Program Flow
Text text to be processed, words separated by spaces

Words list of the words in Text
For every Word in Words:
Exchange two letters with function ExchLett(Word)

Split Word → Letters list of letters in Word
Join Letters → NewWord
Join sequence of NewWord → NewSentence
30 LEFT(Tx, 1), RIGHT(Tx, 1), MID(Tx,2, 1), Wd = MID(Tx, 5, 3); the first character has index 1.
31 Tx[0], Tx[-1], Tx[1], Wd = Tx[4:7]; the first character has index 0.
32 SINGSENT = SPLIT(SENTE, "."): DIM NEWSENT(1) AS STRING: NEWSENT(0) = SINGSENT(0): NEWSENT(1) = ".":
JS = (NEWSENT, ""), 5 statements; the colon is the separator between statements in a line.
33 SingSent = Sente.split(“.”); JS = ‘’.join ([SingSent[0], “.”]),
2 statements; the semicolon is the separator between statements in a line.
In the VISUAL BASIC program in Sect. 4.7.3, we use one array of words containing
the original words at the beginning and the new words at the end. In Python
in Sect. 4.7.6, we are using two arrays: Words, with original data remaining
unchanged, and WordsNew obtained consecutively by appending one scrambled
word after the other.
4.7.3 Excel
In the following, first, a complete program (main program Scramble and sub-
routine XWord) that solves the task is introduced. Do not copy it, but rather
continue reading! Then, the program is developed step by step in test macros to
follow the effect of the individual instructions.
Please be aware that, in the following continuous text, according to our spelling
convention, the words in SMALL CAPS are VBA internal terms, while the words in
italics are invented by the programmer.
Split (Sentence)
The main program Scramble() (Fig. 4.25 (P)) reads a text from cell A1 of the spread-
sheet (Text = CELLS(1,1)), splits it into words (Words = SPLIT(Text), line 3) and passes
the words one by one to the sub-routine XWord(Word) (line 7) which exchanges
two letters. If two letters are to be exchanged for a second time, line 8, now com-
mented out, has to be activated. The instruction SPLIT specifies the variable Words
automatically as an array. For the data type array, see Sect. 4.7.5.
The sub-routine XWord (Word) (Fig. 4.26 (P)) splits the word transferred via
the header into individual letters (lines 18–22), randomly exchanges two inner
letters (lines 23–28), puts together the new word in lines 30 to 34, and returns
the modified word to the higher-level procedure from which the sub-routine was
called.
1 Sub Scramble() Call XWord(Words(lS)) 'Letters are exchanged 7

2 Text = Cells(1, 1) 'Call XWord(Words(lS)) '... a second time 8
3 Words = Split(Text) Next lS 9
4 For lS = 0 To UBound(Words) newText = Join(Words, " ") 10
5 'Cells(lS + 1, 2) = Words(lS) Cells(7, 1) = newText 11
6 'activate for Scramble_test End Sub 12
Fig. 4.25 (P) SUB Scramble reads a text from cell A1, splits it into words stored in the array Words,
and passes the words one by one to the sub-routine XWord. The words returned by XWord (in the
variable Word in the header) are assembled into a new sentence in line 10 and output to cell A7
(CELLS(7,1)) of the spreadsheet. Line 8, now a comment, has to be activated when two letters are
to be exchanged for a second time
13 Sub XWord(Word) 'Random positions 2 to lWord-1 25

14 Dim Letter(20) As String L0 = Letter(n1) 26
15 Debug.Print (Word) Letter(n1) = Letter(n2) 27
16 lWord = Len(Word) '#Letters in the word Letter(n2) = L0 28
17 If lWord >= 4 Then 29
18 For n = 1 To lWord For n = 1 To lWord 30
19 'Letters are singled out. Word = Join(Letter, "") 31
20 Letter(n) = Mid(Word, n, 1) 'Cells(n, 4) = Letter(n) 32
21 'Cells(n, 3) = Letter(n) 'Output to spreadsheet in Xword_test 33
22 Next n Next n 34
23 n1 = Int(Rnd() * (lWord - 2)) + 2 End If 35
24 n2 = Int(Rnd() * (lWord - 2)) + 2 End Sub 36
Fig. 4.26 (P) SUB XWord detects the length of the transferred word (line 16), splits it into letters
(lines 18–22), swaps two inner letters (lines 23–28), and reassembles the letters into the modi-
fied word (line 31). Commented lines 21 and 32 have to be activated to obtain SUB XWord Test,
mentioned in Sect. 4.7.4
Join (Sentence)
The main program SUB Scramble in Fig. 4.25 (P) reassembles the modified words
into a text (newText = JOIN(Words, “ ”), line 10) and writes it into cell A7 (Cells(7,1)
= newText). The second entry “ ” in JOIN causes a space to be inserted after each
element of the array Words.
VBA terms and user-defined variable names

In the procedures Scramble and XWord, there are terms that VBA assigns a precisely
defined meaning to:
– CELLS(r, c): cell in the row r and column c of the current spreadsheet,
– FOR … TO …; DO WHILE … LOOP; ON ERROR GOTO;
– The functions SPLIT(…); JOIN(…); INT(…); RND().
Such terms are printed in the text in SMALL CAPS.

There are also eleven variable names, which the programmer has invented
himself/herself:
– Text, Words, lSent, newText, lWord, Letter, n1 , n2 , L 0 , x, n.
He might as well have taken eleven letters:
– a, b, c, d, e, f, g, h, i, j, k,
or eleven combinations of letters and numbers:
– a1, a2, a3, a4, b5, b6, b7, b8, × 1, × 2, × 3.

All combinations of letters and numbers are allowed as variable names, but the
first character must be a letter. Such names are italicized in the text. To keep the
program clear, you should choose variable names that easily convey their meaning
in the program.
Attention: l (small el) and I (capital i) can easily be confused! Variable names
in different places of the program then look the same but designate two different
variables. So, it is better to use a capital L: LWord instead of lWord.
Questions
What should lines 2 and 11 of Fig. 4.25 (P) be if the statement is formulated
with RANGE instead of CELLS?34
Which variable names in Fig. 4.25 (P) did the programmer come up with
himself/herself?35
Which variable names in Fig. 4.26 (P) did the programmer come up with
himself/herself?36
Do lines 23 and 24 of Fig. 4.26 (P) guarantee that two letters are always
exchanged?37
4.7.4 Programming Step by Step
Step by step, we develop a program that performs the text swirling described in
Sect. 4.7.1. It interacts with the spreadsheet, i.e., reads from cells and fills in the
spreadsheet cells as shown in Fig. 4.27 (S).
A B C D
1 A sentence is to be decomposed. A s s
2 sentence e e
3 is n n
4 to t t
5 be e n
6 decomposed. n e
7 A sentnece is to be decomposed. c c
8 e e
Fig. 4.27 (S) A1 contains the sentence to be processed. B, C, D, and cell A7 are filled in by the
program. The individual words are in B, the individual letters of the second word are in C, those
of the swirled word in D
34 Text = RANGE(“A1”); RANGE(“A7”) = newText.

35 Text, Words, lS, newText.
36 Word, Letter, lWord, n, n , n , L
1 2 0.
37 No, n and n can be identical, so that there is no effective exchange.
1 2
We modify the procedures SUB Scramble in Fig. 4.25 (P) by activating the still
out-commented line 5 (filling in column B of Fig. 4.27 (S)) and SUB XWord by
activating lines 21 (filling in column C) and 32 (filling in column D).
The sub-routine XWord Test(Word) is the same as XWord(Word) in Fig. 4.26
(P), however, with the lines 21, 22, and 33 to 35 activated that now output
intermediate results into the spreadsheet.
Question

Which program lines guarantee that the first and last letter of a word are not
displaced? From which range of the spreadsheet is the text to be processed read?
Into which ranges of the spreadsheet are the individual words of the text, the
individual letters of a word, the swirled letters, and the modified text written?38
After checking these macros to see if they do what we want them to do, we
transform them into a procedure that reads a sentence from cell A1 and outputs the
changed sentence in A2, like SUB Scramble in Fig. 4.25 (P). We now only swap
letters from the word’s interior, i.e., leave the first and last letters as they are.
Task Do this exercise with other texts as well, and surprise your friends with
playful letters!
4.7.5 VBA Constructs
The data type Array in VBA

Arrays are declared in VBA as follows:
DIM Variable name(shape) AS data type
For example, [DIM Fel(2) AS DOUBLE] defines an array with three cells (to be
addressed with 0, 1, 2), where each cell can contain a real number of type DOUBLE.
DIM AR(2,3) AS INTEGER defines a two-dimensional array of integers of shape 3 rows ×
4 columns.
4.7.6 Python
The basic functions and methods for text processing, namely, splitting a text into
words, a word into letters, and, the other way around, joining letters to form a new
38 Text read from A1 (CELLS(1,1)), words written into B, letters of the selected word into C, swirled
letters into D (CELLS(R,4)), modified text into A7. Compare with Fig. 4.27 (S)!
Table 4.13 Basic functions for text processing

1 Text="""A sentence is decomposed."""
2 Words=Text.split()
3 Letters=list(Words[1])
4 L=Letters[4]
5 NewWord=':'.join(Letters[1:-1]) #Concatenate with “:”
6 print(Words) ['A', 'sentence', 'is', 'decomposed.']

7 print(Letters) ['s', 'e', 'n', 't', 'e', 'n', 'c', 'e']
8 print(L) e
9 print(NewWord) e:n:t:e:n:c
word, are presented in Table 4.13. The join() method creates a new string from
“precursor strings”, e.g., a list of letters. The letters in the new word are separated
by a colon as specified in a prefix to join (see lines 5).
These basic functions are again applied in Table 4.14 to split a longer text
into words, pass each individual word to the function ExchLett (reported in Table
4.15), and join the scrambled words into a new text, with blanks as separators. The
content of the variable Text starts with three quotation marks """, indicating that
the following text covering several lines up to the next three quotation marks """
is a string. To enter multi-line strings, use triple codes to start and end them!39
The function ExchLett in the third cell of Table 4.15 uses the function
random.sample from the random library to choose two different internal
letters and then exchange them. This function was found with a search in stack-
overflow.com. Its syntax and mode of action can be deduced from the second cell,
which presents the result of line 23.
In lines 30 to 32, the temporary variable L 0 is introduced in order to swap two
variables. The code in line 34 does the same, but without the use of any temporary
variable. In Python, a backslash (\) indicates that the instruction line is continued.
Statements can also be split up after a comma.
39 https://stackoverflow.com/questions/10660435/pythonic-way-to-create-a-long-multi-line-
string.
Table 4.14 Rewriting scrambled words in a text a (left) program; b (right) result
10 Text="""It is said that you can read a text also if you
leave only the beginning and the end letter in each word as
they are and swirl two middle letters. Do you believe that
or is it nonsense?"""
11
12 Words=Text.split()
13 WordsNew=[]
14 for i in range(len(Words)):
15 WN= ExchLett(Words[i])
16 WordsNew.append(WN)
17 sentNew=' '.join(WordsNew) #Concatenate with blank
18 print(sentNew, "\n")
It is siad taht you can raed a txet aslo if you lvaee olny the
beginning and the end lteter in ecah wrod as tehy are and
swril two mildde ltteers. Do you belveie taht or is it
noesnnse?
Table 4.15 Swapping internal letters

19 #https://stackoverflow.com/questions/9755538/
20 #how-do-i-create-a-list-of-random-numbers-without-
duplicates
21
22 import random
23 random.sample(range(0,10), 10)
[7, 2, 5, 9, 0, 6, 3, 4, 8, 1]
24 def ExchLett(Word):
25 #Split a word into letters
26 Letters=list(Word)
27 #Exchange two internal letters of a word
28 if len(Letters)>=4:
29 n=random.sample(range(1,len(Letters)-1), 2)
30 L0=Letters[n[0]]
31 Letters[n[0]]=Letters[n[1]]
32 Letters[n[1]]=L0 #Join letters into a word!
33 return ''.join(Letters)
34 Letters[n[0]],Letters[n[1]=\
Letters[n[1]],Letters[n[0]]
Table 4.16 Scrambling a text passage

1
Task="""Write a program that first reads a text from a
spreadsheet. This text is then to be broken down into
words. The individual words are transformed so that the
first and last letters remain in their place, but the inner
letters are randomly swapped."""
2 Words=Task.split()
3 WordsNew=[]
4 for i in range(len(Words)):
5 WN= ExchLett(Words[i])
6 WordsNew.append(WN)
7 sentNew=' '.join(WordsNew)
8 print(sentNew)
Wrtie a porgram taht frist rdaes a txet form a sprdaesheet.
Tihs txet is tehn to be broekn dwon itno worsd. The indaviduil
wodrs are transofrmed so taht the fisrt and lsat lettres
remian in thier palce, but the inner lettres are rdnaomly
seappwd.
Questions
Write lines 30 to 32 of Table 4.16, swapping two variables using the temporary
variable L0, as one statement!40
How can lines 5 and 6 of Table 4.15 be merged into one statement?41
How can you introduce an additional line of code into Table 4.16 to achieve
two letter swappings in a word?42
Table 4.16 processes the second text passage, now performing two letter
exchanges.
4.8 Processing the Protocol of a Measuring Device
Continuous text is decomposed and rearranged in tables. Knowledge

acquired by processing texts in the previous exercise is applied to separate
text and numbers in reports created by measuring instruments, with the aim
to recognize code words and rearrange the essential results in tables.—The
Python program is based on the library Pandas.—A piece of advice: With
40 Letters[n[0]], Letters[n[1]] = Letters[n[1]], Letters[n[0]].

41 WN = EXCHLETT(EXCHLETT(WORDS[I]).
42 Introduce WN = ExchLett(WN) after line 5!
4.8 Processing the Protocol of a Measuring Device 161
the knowledge gained in this exercise, you may earn a small bit of extra
income from a side job in scientific projects!
4.8.1 Protocol of a Measuring Device
Many measuring devices output a plain-text file as a protocol containing both text
and numbers. As an example, we will use the output of chemical analysis with RBS
(Rutherford Back Scattering) concerning the composition of four Nb-doped TiO2
layers on a silicon substrate (spreadsheet in Fig. 4.28 (S)). Every layer contains
different fractions of Ti (titanium), O (oxygen), Nb (niobium), and Ar (argon). We
will convert this information into a table, as in Fig. 4.29 (S).
A B C D
1 T1.lay T2.lay T3.lay T4.lay
2 !----------------- !----------------- !----------------- !------------------
3 d=0.20E18 d=0.25E18 d=0.30E18 d=0.35E18
4
5 Ti#,1 Ti#,1 Ti#,1 Ti#,1
6 O#,2.5 O#,2.4 O#,2.3 O#,2.2
7 Nb#,0.03 Nb#,0.05 Nb#,0.07 !------------------
8 Ar#,0.01 !----------------- Ar#,0.008 s=
9 !----------------- s= !----------------- Si#,1
10 s= Si#,1 s=
11 Si#,1 Si#,1
Fig. 4.28 (S) Protocol of RBS measurements, transferred into an EXCEL spreadsheet; row 1 =
names of four different samples; parameters: d = number of atoms per cm2 ; Ti, O, Nb, Ar =
elements found in the layer with their indices in the chemical formula
A B C D E F
1
2 SampNam NAtoms Ti O Nb Ar
3 T1 2,00E+17 1 2,5 0,03 0,010
4 T2 2,50E+17 1 2,4 0,05
5 T3 3,00E+17 1 2,3 0,07 0,008
6 T4 3,50E+17 1 2,2
7
Fig. 4.29 (S) The data from Fig. 4.28 (S) have been written into this table in a spreadsheet with
the name “TabLay”. Each sample has its own row. The first row is left blank in order to insert an
index for the next free row later
36 For r = 3 To r2 cll = Left(cl, 2) 'take first two letters 39

37 cl = Cells(r, sample) 'e.g.: d=0.20E18 If cll = "d=" Then NAtoms = Right(cl, Le - 2) 40
38 Le = Len(cl) 'length of the string Next r 47
Fig. 4.30 (P) Cuts off the first two letters of a data string (line 39) and checks whether this is the
code word “d =”
1 Sub DecodeRBS() For sample = 1 To 4 15

2 'Original data in sheet "RBS-data" Sheets("RBS-data").Select 'original data 16
3 'Table with sample characteristics in "TabLay" 'Get sample name! 17
4 Dim cl, cll As String SampNam = Cells(1, sample) 'sample name 18
5 'Write headers! Le = Len(SampNam) 'length of the name 19
6 r3 = 2 SampNam = Left(SampNam, Le - 4) 20
7 Sheets("TabLay").Select ' ".lay" is removed 21
8 Cells(r3, 1) = "SampNam" 'Identify range with information on sample! 22
9 Cells(r3, 2) = "NAtoms" For r1 = 3 To 30 'scans rows 3 to 30 23
10 Cells(r3, 3) = "Ti" cl = Cells(r1, sample) 'Content of cell 24
11 Cells(r3, 4) = "O" cll = Left(cl, 5) 25
12 Cells(r3, 5) = "Nb" If cll = "!----" Then r2 = r1 - 1 26
13 Cells(r3, 6) = "Ar" 'r2 = last row of information on the sample 27
14 r3 = r3 + 1 ' next free row for output Next r1 28
Fig. 4.31 (P) Complete program for rearranging the raw data from Fig. 4.28 (S) into a table as in
Fig. 4.29 (S); continued in Fig. 4.32 (P)
4.8.2 Detection of Code Words
The information in a column of Fig. 4.28 (S) is to be decoded and stored in a row
of Fig. 4.29 (S). The main task is to identify certain code words that indicate the
physical or technical quantity to which the following numbers refer. An extract
from the complete Visual Basic decoding program (Fig. 4.32 (P)) can be found in
Fig. 4.30 (P).
The data set for a layer is read line by line. The data strings “d = 0.20E18” and
“Ti#,1” contain information about the number of atoms per cm2 and the titanium
content in the sample, respectively. When the program processes the file, it is not
clear from the outset what type of data string is currently involved. For decoding,
therefore, the first parts of the data line are separated (line 39 in Fig. 4.30 (P)),
and it is queried as to whether this part is “d =” (line 40). If this is the case,
the following string is interpreted as a number and written into the corresponding
variable; for “d =”, this is NAtoms (line 40 in Fig. 4.30 (P)).
In Python, the task is tackled with the library Pandas, which mimics
spreadsheet calculation.
wb name in Pandas of the workbook containing the data

sh name in Pandas of the worksheet within the workbook
T1, T2, T3, T4 identifiers of samples
„!----“ code word to indicate the end of useful information

“d=” two-character code word for atomic coverage, atoms per
cm2
“O#,” three-character code word for oxygen
“Ti#,”, “Nb#,”, “Ar#,” four-character code words for titanium, niobium, and
argon
r2 last row of useful information
r3 current row in the spreadsheet for output
NAtoms number of atoms per cm2
O, Ti, Nb, Ar fraction of the respective element
4.8.4 Excel
The complete decoding procedure can be found in Fig. 4.31 (P) and Fig. 4.32 (P).
SUB DecodeRBS writes headings into the spreadsheet “TabLay” (lines 7 to 13),
successively reads rows 3 through 30 from the spreadsheet “RBSData” (line 23,
index r1), determines the last row of useful data (line 26), decodes the useful data
(FOR loop in rows 36 through 47), and finally writes the decoded data row by row
(index r 3 ) into the spreadsheet “TabLay” (rows 49 through 55).
The loops are of type loop2i, with the loop index r and the running index r 3 ,
indicating the next free row in the table and set to 2 at the beginning (line 6)
and incremented in line 56 after the extracted values for the parameters have been
entered into the output table within the For-loop with the index r scanning the
input table.
Since the code words have different lengths (“d =” has two letters, “Ti#,” has
four letters), in lines 39, 41, and 43, two, three, and four letters are, one after the
other, cut off from the beginning of the data string and it is checked as to whether
they correspond to one of the code words. The part of the data string following
29 'Remove old information! If cll = "Ti#," Then Ti = Right(cl, Le - 4) 44

30 NAtoms = Empty If cll = "Nb#," Then Nb = Right(cl, Le - 4) 45
31 O = Empty If cll = "Ar#," Then Ar = Right(cl, Le - 4) 46
32 Ti = Empty Next r 47
33 Nb = Empty 'Write decoded data into a different sheet! 48
34 Ar = Empty Sheets("TabLay").Select 49
35 'Decode information in rows 3 to r2! Cells(r3, 1) = SampNam 50
36 For r = 3 To r2 Cells(r3, 2) = NAtoms 51
37 cl = Cells(r, sample) 'e.g.: d=0.20E18 Cells(r3, 3) = Ti 52
38 Le = Len(cl) 'length of the string Cells(r3, 4) = O 53
39 cll = Left(cl, 2) 'take first two letters Cells(r3, 5) = Nb 54
40 If cll = "d=" Then NAtoms = Right(cl, Le - 2) Cells(r3, 6) = Ar 55
41 cll = Left(cl, 3) 'take first three letters r3 = r3 + 1 'next free row 56
42 If cll = "O#," Then O = Right(cl, Le - 3) Next sample 57
43 cll = Left(cl, 4) 'take first four letters End Sub 58
Fig. 4.32 (P) Continuation of Fig. 4.31 (P)

the cut-off contains a number separated with RIGHT(..) and then assigned to the
corresponding variable (e.g., NAtoms, Ti, …). This separation is easy because all
numbers have the same format.
Questions
Which code word is queried in Fig. 4.30 (P)?43

Which code words in Fig. 4.32 (P) have length 4?44
Module in VBA
Since the program SUB DecodeRBS refers, with SHEETS(“RBS-data”).SELECT and
SHEETS(“TabLay”).SELECT, to two different spreadsheets, it must be operated in a
module. If it stands in the VBA sheet associated with a spreadsheet, it operates only
in that spreadsheet with instructions like CELLS(r,c) = .
VBA keyword empty

In lines 36 to 47 of Fig. 4.32 (P), the variables O, Ti, Nb and Ar are only filled in if the
assigned code words occur in the original protocol. If a code word does not occur in
the current data line, the content of the associated variable is not overwritten, and the
old value persists. To prevent this from happening, we have entered the assignments
in lines 30 to 34, e.g., Nb = EMPTY. EMPTY is a VBA keyword, to ensure that the
variables do not contain any value when a new sample is processed.
Sample T4 does not contain any Nb; cell E6 in Fig. 4.29 (S) thus remains
empty. If we had not set Nb = EMPTY before decoding, then Nb would still contain
the value 0.07 of the previous sample and would have been incorrectly entered into
the table of results. This would be a grave error in content!
The table is continued the next time the macro is called
Tim The presented program sorts the data of exactly four samples into another
table. What if a new set of samples comes in and the table is to be continued? Can
the program remember the value of the index r 3 for the next free row and use it
for the next call?
Mag No, the program forgets the values of the variables when it finishes its
execution.
Alac Then, I will simply adapt the code before each new call. In line 14, the
next free row in the table of Fig. 4.29 (S) is entered, r 3 = 7, and for the last FOR
loop index in line 15, I will enter the current number of samples.
43 „d = “ is queried.
44 The code words „Ti#,“, „Nb#,“ and „Ar#,“ have the length 4.
Mag That’s a practical idea, and it works. But it can also be done more elegantly
with the following two pieces of program.
You can use the first row in Fig. 4.29 (S), still empty, e.g., cell A1, to store the
number of the first free row after the previous entries, and cell D1 to specify the
number of new samples. This information is then read with r 3 = RANGE(“A1”) in
line 14 and … TO RANGE(“D1”) in line 15. Cell A1 is now overwritten with the
value of r 3 by a new instruction at the end of SUB DecodeRBS. The number of new
samples must be entered manually in D1 when a new data series is to be decoded,
or the programmer can devise a query to automatically determine the number of
samples in the raw data of Fig. 4.28 (S).
DO … LOOP UNTIL
The next free row can also be determined by querying the current cell content in a
loop to see if the current cell is empty:
DO
r3 = r3 + 1
LOOP UNTIL CELLS(r 3 , 1) = EMPTY
Similarly, the number of NOT EMPTYs in a new raw data file can be obtained.
For more information on the instructions DO … LOOP, DO WHILE … LOOP, and

DO … LOOP UNTIl, see EXCEL help!
4.8.5 Python
Pandas (“Python Data Analysis”) is a library for Python, based on NumPy.

It is designed for data management and analysis and works with structured data
(DataFrame (2-dimensional)) and time series (Series (1-dimensional)), thus
mimicking spreadsheet calculations. We use it only in this exercise to read data
from an EXCEL book, decode the measurement protocol, and write the results into
an EXCEL sheet, in the same form as in Fig. 4.29 (S).
R1C0 in Pandas
In Table 4.17, the EXCEL workbook ‘RBS_data.xlsx’ (Fig. 4.28 (S)) is opened, so that
its data are available in Pandas. Without further specification, the workbook must
be in the same directory as the Python program. The data in Sheet1 are entered into
a two-dimensional matrix sh. This matrix can be addressed in A1 or R1C0 reference
style (see second cell). In A1 style, a column is addressed by a letter and a row by a
number starting at 1; in R1C0 style, both are addressed by numbers, with columns
being numbered starting at 0.
With the statement in line 5, the contents in the cells will be copied directly. If
a cell contains a formula, this formula will be transferred. If it is desired that all
Table 4.17 a (top cell) opening an EXCEL workbook; b (bottom cell) addressing cells and ranges
and reading their content
2 import pandas as pd
3 import openpyxl #A Python library to read/write Excel 2010
4
5 wb=openpyxl.load_workbook(‘RBS_data.xlsx')
6 sh=wb['Sheet1']
Sh: <Worksheet "Sheet1">
sh['A1'].value: T1.lay
sh[1][0].value: T1.lay
7 for r in range(1,4):
8 print(sh[r][0].value)
range(1,6):
T1.lay
!--------------------------
d=0.20E18
None
Ti#,1
formulas be evaluated and only the results transferred, the opening has to include
a keyword argument data_only = True:
wb = openpyxl.load_workbook( RBS_data.xlsx , data_only = True).
In Table 4.18, the sample name is extracted from the first entry in c = 0 (column
A), reported in the second cell. The content of the bottom cell is produced by the
print statements shown explicitly in lines 5 to 9. In contrast to this, print statements
Table 4.18 Extracting the sample name from the first entry in column c; print statements resulting
in the output cell (bottom cell) are explicitly reported in lines 5 to 9
1 c=0
2 SN=list(sh[1][c].value)[:-4] #Delete “.lay”
3 SampNam=''.join(list(SN))
4
5 print("sh[1][c] ",sh[1][c])
6 print("sh[1][c].value ",sh[1][c].value)
7 print("list(sh[1][c].value)",list(sh[1][c].value))
8 print("SN ",SN)
9 print("SampNam ",SampNam)
sh[1][c] <Cell 'Sheet1'.A1>
sh[1][c].value T1.lay
list(sh[1][c].value) ['T', '1', '.', 'l', 'a', 'y']
SN ['T', '1']
SampNam T1
Table 4.19 Getting the last row of useful information; line 5 contains an error
1 def LastRow(c):
2 for r1 in range(2,15):
3 cl=sh[r1][c] #Cell address
4 if cl.value!=None:
5 cll=''.join(list(cl.value)[:5])
6 #Indentation 4 spaces!
7 if cll=='!----': r2=r1-1
8 return r2
are usually omitted in our tables reporting Python programs; only the results are
usually reported in an output cell.
Questions
What are the values of sh[1][0].value and

list(sh[1][0].value)[0]?45
Line 5 of Table 4.19 contains a bug. Which one?46
In Table 4.19, column c is scanned for the occurrence of “!––”, the code word
signaling the end of the information on the first layer.
Table 4.20 reports the function for decoding the string in the cell in column c,
row r 2 . The Python code mimics the VBA procedure in Fig. 4.32 (P).
In Table 4.21, a data frame out is created reproducing the EXCEL sheet in
Fig. 4.29 (S). A For loop runs over sample with sample data, determining the
last row r 2 with useful information and decodes the range rows 3 to r 2 of the
column. The print statement in line 11 produces the output in the lower cell.
Output from Pandas to an Excel file

Writing our results into an EXCEL file requires some care. The simple statement
in Table 4.22a creates a new file RBS data3.xlsx and writes our frame out into a
sheet with the name Sheet3. If a file with said name already exists, it is overwritten.
The keyword arguments header = False, index = False cause row 1 and
column 1 of Table 4.21 (bottom cell) not to be output.
To write the data into an already existing file, we have to open that file and
specify a writer (lines 8 and 9 in Table 4.22).
45 sh[1][0].value -> T1.lay, list(sh[1][0].value)[0] -> T.

46 Indentation with respect to the if line is only 3 spaces; it must be 4 spaces.
Table 4.20 Function for decoding a string

1 def decode(c,r2): #Column c, row r2
2 nAtoms=None
3 O=None
4 Ti=None
5 Nb=None
6 Ar=None
7 #Decode information in rows 3 to r2!
8 #First index in array starts with 0
9 for r in range(2,r2+1):
10 cl=sh[r][c] #Cell address
11 if cl.value!=None: # != means “not equal”
13 if cll=='d=':
14 nAtoms=''.join(list(cl.value)[2:])
16 if cll=='O#,': O =''.join(list(cl.value)[3:])
18 if cll=='Ti#,': Ti=''.join(list(cl.value)[4:])
19 if cll=='Nb#,': Nb=''.join(list(cl.value)[4:])
20 if cll=='Ar#,': Ar=''.join(list(cl.value)[4:])
21 nAtoms=float(nAtoms)
22 return(nAtoms,Ti,O,Nb,Ar)
4.9 User-Defined Functions
We code functions in VISUAL BASIC that can be applied in spreadsheets just

like built-in functions. As examples, we realize the vector operations scalar
product and cross-product with three-component vectors as arguments and
with, respectively, a scalar and a vector as the return variable.
4.9.1 User-Defined Functions as Add-In
Functions in modules
We often call built-in functions in cells, e.g., trigonometric functions with formulas
like B5 = [=B$1*cos(B$2*$A5]). We can also create functions ourselves and apply
them in the same way. As an example, we implement a function CosSq(x), calcu-
lating the square of a cosine. This has to be done in a module that we create with
INSERT/MODULE in the project explorer (see Fig. 4.33a).
The function is implemented in the corresponding Visual Basic sheet MODULE1. The
qualifier is FUNCTION, not SUB, as for procedures. For a function, a value must be
4.9 User-Defined Functions 169
Table 4.21 Data frame reproducing the structure of the EXCEL sheet in Fig. 4.29 (S); NaN stands
for “Not a number”
1 out = pd.DataFrame(index=range(1,8), s=list('ABCDEF'))
2 title=['SampNam','NAtoms','Ti','O','Nb', 'Ar']
3 out.iloc[1]=title
#iloc is integer position from 0 to length-1 of the axis
4 for sample in range(4):
5 SampNam=sh[1][sample].value
6 SampNam=SampNam[:-4] #Remove .lay
7 r2=LastRow(sample)
8 nAtoms,Ti,O,Nb,Ar=decode(sample,r2)
9 result=[SampNam,nAtoms,Ti,O,Nb,Ar]
10 out.iloc[sample+2]=result
11 print(out)
12 out.to_excel("output.xlsx",sheet_name="Sheet2",
13 header=False,index=False) #A…F and 1…7 not transferred
#NaN means “Not a number”
A B C D E F
1 NaN NaN NaN NaN NaN NaN
2 SampNam NAtoms Ti O Nb Ar
3 T1 2e+17 1 2.5 0.03 0.01
4 T2 2.5e+17 1 2.4 0.05 None
5 T3 3e+17 1 2.3 0.07 0.008
6 T4 3.5e+17 1 2.2 None None
7 NaN NaN NaN NaN NaN NaN
Table 4.22 a (top cell) Creates a new file and writes the data into the specified sheet; b (bottom
cell) Opens an existing file and adds a new sheet
14 out.to_excel('RBS_data3.xlsx',sheet_name="Sheet3",
15 header=False,index=False)
16 #Creates a new workbook with one sheet “Sheet3”.
17 #Overwrites 'RBS_data3.xlsx' if it exists already.
18 """https://stackoverflow.com/questions/20219254/
19 how-to-write-to-an-existing-excel-file-
20 without-overwriting-data-using-pandas"""
21 wb2 = openpyxl.load_workbook('RBS_data2.xlsx')
22 writer = pd.ExcelWriter('RBS_data2.xlsx',
engine='openpyxl')
23 writer.book = wb2 #Necessary for not deleting other sheets
24 out.to_excel(writer, "data3",header=False,index=False)
25 writer.save()
Fig. 4.33 a (left) PROJECT-EXPLORER window; the user-defined spreadsheet function CosSq is in
MODULE1. b (right) CosSq pops up after typing “=Cos” among the other functions starting with cos
assigned to the function identifier within the function’s body, in our case cosSq =
COS(x)ˆ2. When now “=Cos” is written into a cell, a list pops up with all functions
starting with Cos, including our CosSq (see Fig. 4.33b).
If you want to use your functions in every EXCEL file, they must be saved as an
ADD-IN. To do so, create an EXCEL file, enter your function codes into VBA mod-
ules, and finish with: SAVE AS/EXCEL ADD-IN. This add-in must be activated in the
EXCEL options. Upon selecting FILE/OPTIONS/ADD-INS/ a list appears with an entry “
Dieter’s Functions” that must be included by ticking the box . In the VBA edi-
tor, VBA PROJECT Dieters Funktionen.xlam now also appears in the project explorer
under PROJECT–VBA PROJECT (see Fig. 4.33a, bottom line).
4.9.2 Scalar Product and Vector Product
We are developing functions for the scalar and vector products of two three-
dimensional vectors stored in cell ranges that are entered as arguments in the
functions. Let’s consider two three-dimensional vectors:
r 1 = (x1 , y1 , z 1 ) and r 2 = (x2 , y2 , z 2 )
Their scalar product is defined as.
r 1 · r 2 = x1 · x2 + y1 · y2 + z 1 · z 2 (4.2)
Their vector product (or cross product) is defined as.
r 1 × r 2 = (y1 · z 2 − y2 · z 1 , z 1 · x2 − z 2 · x1 , x1 · y2 − x2 · y1 ) (4.3)
1 Function Scl(r1 As Range, r2 As Range) Function Crsm(r1 As Range, r2 As Range) 12

2 Scl = r1(1) * r2(1) + r1(2) * r2(2) + r1(3) * r2(3) Dim cs(2, 2) 13
3 End Function cs(0, 0) = r1(2) * r2(3) - r1(3) * r2(2) 14
4 cs(1, 0) = r1(3) * r2(1) - r1(1) * r2(3) 15
5 Function Crs(r1 As Range, r2 As Range) cs(2, 0) = r1(1) * r2(2) - r1(2) * r2(1) 16
6 Dim cs(2) cs(0, 1) = cs(1, 0) 17
7 cs(0) = r1(2) * r2(3) - r1(3) * r2(2) cs(0, 2) = cs(2, 0) 18
8 cs(1) = r1(3) * r2(1) - r1(1) * r2(3) Crsm = cs 19
9 cs(2) = r1(1) * r2(2) - r1(2) * r2(1) End Function 20
10 Crs = cs 21
11 End Function 22
Fig. 4.34 (P) User-defined functions for the scalar product Scl and the vector product Crs of two
three-dimensional vectors r 1 and r 2; ; the function Crsm can process and output both column and
row vectors
The output of the scalar product is one number returned into the cell with the
corresponding formula; that of the vector product is a set of three components to
be entered into a row range or a column range.
These two products are calculated with the two user-defined spreadsheet
functions Scl and Crs in Fig. 4.34 (P).
Questions
How many components does the array cs(2) in Fig. 4.34 (P) have?47
What are the differences between the arrays named cs in the functions Crs
and Crsm in Fig. 4.34 (P)?48
Scalar product
The scalar product is easy to program. It can be calculated in one code line (line 2
in Fig. 4.34 (P)). Two three-dimensional cell ranges must be entered as arguments.
These can both be column ranges or both row ranges or one column range and one
row range (see Fig. 4.35 (S)). Consequently, the variables in the function header are
declared as RANGE.
Vector product
The result of a vector product is again a vector. In Fig. 4.36 (S), two column vectors
a and b are defined in range A2:B4. The row vectors c and d in range B6:D7 contain
the same coefficients as a and b. In column D, the cross-product a x b is calculated
with spreadsheet formulas.
47The array DIM cs(2) has the three components cs(0), cs(1), cs(2).
48In Crs, a one-dimensional array (type cs(2)) is written, in Crsm, a two-dimensional array (type
cs(2,2)).
A B C D E F G H I J K L
1 a b 12 =Scl(A2:A4;B2:B4)
2 1 2 c_ 1 2 3 12 =Scl(A2:A4;E3:G3)
3 2 2 d 2 2 2 12 =Scl(a;b)
4 3 2 12 =Scl(c_;d)
5 12 =Scl(a;d)
Fig. 4.35 (S) Contains the results of the user-defined spreadsheet function Scl, which calculates
the scalar product of two three-dimensional vectors
A B C D E F G H I J K L M N O
) )
;b) _;d) _;b) (a;b) (c_;d (c_;b
1 a b axb =crs(a =crs(c =crs(c =crsm =crsm =crsm
2 3,0 -2,0 -10,0 =A3*B4-B3*A4 -10,0 -10,0 -10,0 -10,0 -10,0 -10,0
3 2,0 8,0 -20,0 =A4*B2-B4*A2 -10,0 -10,0 -10,0 -20,0 -20,0 -20,0
4 2,5 5,0 28,0 =A2*B3-B2*A3 -10,0 -10,0 -10,0 28,0 28,0 28,0
5
6 c_ 3,0 2,0 2,5 -10,0 -20,0 28,0 =crs(a;b) -10,0 -20,0 28,0 =crsm(a;b)
7 d -2,0 8,0 5,0 -10,0 -20,0 28,0 =crs(c_;d) -10,0 -20,0 28,0 =crsm(c_;d)
8 -10,0 -20,0 28,0 =crs(a;d) -10,0 -20,0 28,0 =crsm(a;d)
Fig. 4.36 (S) The vector product a x b is calculated in column D using spreadsheet formulas; the
range G2:I8 contains results of the user-defined function crs, which can only output row vectors
(wrong results in G2:I4); the range L2:N8 contains results of the user-defined spreadsheet function
crsm, which can accept and output row and column vectors
In columns G to I, the function Crs is used. As you can see from the results, this
function accepts row and column vectors as input, but only returns correct values
if they are output as row vectors (see G6:J8); the results in G2:I4 are wrong.
In columns L to N, the function Crsm from Fig. 4.34 (P) is applied, which can
output the result either as a row vector (e.g., L6:N6) or as a column vector (e.g.,
L2:L4). This is because, in this function, a 3 × 3 matrix is written into the range
declared with DIM cs(2,2), of which only one row or one column is output if only
one row range or one column range is activated.
The functions Crs and Crsm must be called as matrix functions. In Fig. 4.36
(S), for example, the area G6:I6 was activated, the formula entered according to
J6 and closed with the magic chord (Ctl + Shift) + Enter. In L2:L4, a column
area was activated, and in L6:N6, a row area, so that in each case, vectors with
three components are returned by crsm.
Questions
Why is cs(2,2) in FUNCTION Crsm sufficient as an array for a 3 x 3 matrix?49
49 DIM cs(2,2) is a (0, 1, 2) x (0, 1, 2)-Matrix. The indices begin at 0.

Table 4.23 Two column vectors a, b and two row vectors c, d are specified
1 a=np.array([[1,2,3]]).transpose(1,0)
2 b=np.array([[2,2,2]]).transpose(1,0)
3 c=np.array([1,2,3])
4 d=np.array([2,2,2])
a b
[[1] [[2] c [1 2 3]
[2] [2]
[3]] [2]] d [2 2 2]
Table 4.24 Scalar product, line 8 deactivated

5 def Scl(r1,r2):
6 o=r1[0]*r2[0]+r1[1]*r2[1]+r1[2]*r2[2]
7 out=o
8 #if type(o) == np.ndarray: out = o[0]
9 return out
Scl(a,b) [12]
Scl(c,d) 12
Scl(a,d) [12]
How can this function be used to output row and column vectors to a
spreadsheet?50
4.9.3 Python
Scalar product
The specifications in Table 4.23 for column vectors a, b, and row vectors c, d are the
same as in Fig. 4.35 (S).
The function Scl (for “scalar product”) as reported in Table 4.24 corresponds
literally to the Visual Basic function of the same name (Fig. 4.34 (P)). Its output
is, however, only a scalar if two row vectors are multiplied.
Column vectors are two-dimensional arrays; a scalar product with one of them
is broadcast into a one-dimensional array (see lines 10 and 12 of Table 4.24). If a
scalar is always desired, line 8 has to be activated by removing the # character.
Vector product or cross product

A function Crsm for calculating the cross-product of two three-dimensional vectors
is reported in Table 4.25.
50 Because internally a 3 × 3-matrix is created, see explanations for Fig. 4.36 (S)!
Table 4.25 Cross product of two three-dimensional vectors, output optionally transposed in line
16 to become a column vector
1 def Crsm(r1, r2, C):
2 cs=np.empty(3)
3 cs[0]=r1[1]*r2[2]-r1[2]*r2[1]
4 cs[1]=r1[2]*r2[0]-r1[0]*r2[2]
5 cs[2]=r1[0]*r2[1]-r1[1]*r2[0]
6 if C==True:cs=np.array([cs]).transpose(1,0)
7 return cs
Table 4.26 Cross-product of row and column vectors

1 a=np.array([[3,2,2.5]]).transpose(1,0)
2 b=np.array([[-2,8,5]]).transpose(1,0)
3 c=np.array([3,2,2.5])
4 d=np.array([-2,8,5])
Crsm(c,b,True)
Crsm(a,b) [-2.00 4.00 -2.00] [[-2.00]
[ 4.00]
Crsm(c,d) [-2.00 4.00 -2.00] [-2.00]]
In Table 4.26, again, two column vectors a, b and two row vectors c, d are
specified. Their pairwise cross-product, obtained with Crsm, is reported in the
lower cells of the table. If only the vectors are transferred to the function, row
vectors are returned (bottom left cell), whereas a column vector is returned if the
optional parameter C is assigned “True” (bottom right cell).
4.10 Questions and Tasks
Densely packed planes
1. How many neighbors does a sphere have in a closely-packed plane?

2. How many neighbors does a sphere have in a stack of closely-packed planes?
Program-controlled drawings
3. Write a macro that writes the numbers 1 to 20 in a diagonal of a table, e.g., in

cells A1, B2, etc.
4. What does the broom rule Empty lines separate curves mean?
Record macro
The diagram in Fig. 4.37 (S) has been created with the macro recorder switched on.
The program code can be found in Fig. 4.38 (P). The diagram has been formatted
4.10 Questions and Tasks 175
A B C D E F G H I
1 Curve
2 x y 120
3 1 1 100
4 2 4
80
5 3 9 y
6 4 16 60
7 5 25
40
8 6 36 Curve
9 7 49 20
10 8 64 0
11 9 81 0 5 10 x 15
12 10 100
Fig. 4.37 (S) Diagram of the data in columns A and B
1 Sub Macro1() 1
2 ActiveSheet.Shapes.AddChart2(240, xlXYScatterLines).Select 2
3 ActiveChart.SeriesCollection.NewSeries 3
4 ActiveChart.FullSeriesCollection(1).Name = "=Diagram!$B$1" 4
5 ActiveChart.FullSeriesCollection(1).XValues = "=Diagram!$A$3:$A$12" 5
6 ActiveChart.FullSeriesCollection(1).Values = "=Diagram!$B$3:$B$12" 6
7 End Sub 7
Fig. 4.38 (P) Instructions recorded by the macro recorder when the diagram in Fig. 4.37 (S) was
created
with the programs in Fig. 4.38 (P) and Fig. 4.40 (P). Your task is to analyze the
VISUAL BASIC programs and redraw the diagram with our standard FigStd and plt
(matplotlib.pyplot) commands in Python.
Plotting a diagram
5. Of what type is the diagram in Fig. 4.37 (S) (LINE, BAR, or SCATTER)?
6. What is the equation for y?
7. How do you create the arrays x and y in Python?
8. With SUB MACRO1 in Fig. 4.38 (P), retrace how the chart was created and interpret
the program lines 3 to 6!
9. (Python) What does a header in FigStd (numpy and matplotlib) look
like when leading to a diagram like that in Fig. 4.37 (S)?
10. SUB MACRO2 in Fig. 4.39 (P) has recorded the instructions executed to format the
data series in the diagram of Fig. 4.37 (S). Interpret the formatting instructions!
How do you implement them in the plot command of plt.plot(?) of the
plotlib library?
11. SUB MACRO3 in Fig. 4.40 (P) has recorded the commands executed to format the
x-axis in the diagram of Fig. 4.37 (S). Interpret the instructions that follow the
two WITH SELECTION commands!
8 Sub Macro2() .ForeColor.ObjectThemeColor _ 13

9 ActiveChart.FullSeriesCollection(1).Select = msoThemeColorText1 14
10 Selection.MarkerStyle = 2 .Weight = 1.25 15
11 Selection.MarkerSize = 7 End With 16
12 With Selection.Format.Line End Sub 17
Fig. 4.39 (P) SUB MAKRO2 contains the commands that were recorded by the macro recorder when
the data series for the diagram in Fig. 4.37 (S) was formatted
18 Sub Macro3() ActiveChart.Axes(xlValue).Select 25

19 ActiveSheet.ChartObjects("Chart 1").Activate With Selection.Format.Line 26
20 ActiveChart.Axes(xlCategory).Select .ForeColor.ObjectThemeColor \ 27
21 With Selection.Format.Line '= msoThemeColorBackground1 28
22 .ForeColor.ObjectThemeColor \ .ForeColor.Brightness = -0.5 29
23 '= msoThemeColorText1 End With 30
24 .Weight = 1# End Sub 31
Fig. 4.40 (P) SUB MACRO3 contains commands recorded by the macro recorder when the diagram
in Fig. 4.37 (S) was formatted
12. Write a Python program that produces a similar diagram with FigStd
(numpy and matplotlib) ! It should include formatting the data series in
the function header and changing the thickness of the x-axis within FigStd.
Text processing in VISUAL BASIC and Python

The following three questions refer to both Visual Basic for EXCEL and Python.
The variable Tx contains the text “We are cutting.”
13. With which instructions do you get the first, the last, and the 4th letters of Tx?
14. How do you copy the fragment “re cu” from Tx to a new variable Wd?
15. Of what type are the variables A, B, C, and D in the commands A = Split(B) and
C = Join(D)?
Loop2i; continue counting in the loop!
16. Write a macro SUB XY1() that writes all products x·y from x = 1 to 10 and from
y = 1 to 5 successively into a spreadsheet, with x and y being integers!
17. Do the same in another macro SUB XY2() for x and y being half-integers (1, 3/2,
2, 5/2, …)!
18. Do the same as in SUB XY2() in a new macro SUB XY3(), but insert a blank line
after every third entry into the spreadsheet!
19. Create a similar Python program using the.append method and nested loops
running over arrays x and y created by np.linspace! For the equivalent of
inserting an empty row into a spreadsheet, append ‘None’ to the list!
20. Make a hand-drawn sketch of straight-line segments in the xy plane, with the x
values in line 1 and the y values in line 2 of Fig. 4.41 (S)!
Fig. 4.41 (S) Coordinates for straight-line segments in the xy plane
Rep-log procedure
In Fig. 4.42a, a circle is represented, calculated in the spreadsheet in Fig. 4.43 (S)
in columns G:I. The coordinates of the centers of these circles are taken from the
table in Fig. 4.41 (S). Figure 4.42b shows four circles whose coordinates have been
obtained from the table in Fig. 4.43 (S) with a rep-log procedure that systematically
changes the center point and the radius.
20 20
y Circle y
10 10
0 0
-10 0 10 20 30 -10 0 10 20 x 30
x
-10 -10
Circles Centers
-20 -20
Fig. 4.42 a (left) A circle with r 0 = 4 and x 0 = 3. b (right) The circle from a was enlarged three
times and shifted along the x-axis; in the diagram, all four circles are represented as one data series
1 r.0 4.00 c.s 4.00
2 x.0 3.00 v.F 3.00 Circle Circles Centers
+x.0
hi) hi)
dphi COS(p SIN(p
3 =G5+ =r.0* =r.0*
4 30 ° phi x y x y xC yC
5 dphi 0.524 0.00 7.00 0.00 7.00 0.00 3 0
6 0.52 6.46 2.00 6.46 2.00 6 0
17 6.28 7.00 0.00 7.00 0.00
Fig. 4.43 (S) Spreadsheet calculation for Fig. 4.42. The variable cs contains the factors with
which the radius of the circle is increased. The variable vF is the velocity with which the circle
center is shifted on the x-axis
21. What are the radii and the coordinates of the circles’ center points in
Fig. 4.42b?
22. The coordinates of the circle are to be calculated four times by changing the
parameters in the table and then to be stored successively in columns Q and
R (column indices 17 and 18, respectively) of the spreadsheet and graphically
displayed in a diagram as one data series (to yield Fig. 4.42b). The centers’
coordinates are in columns N and O (column indices 14 and 15, respectively).
The circle radius should increase with r 0 = vs ·t, and the center point should
be shifted on the x-axis with x 0 = vF ·t. Write such a log procedure in Visual
Basic!
23. For the answer of Question 23 in Python, create empty lists Q and R and
extend them with the.append method in a nested loop! Instead of an empty
line, insert “None”! Finally, plot (Q, R) with a correct label and observe
whether four separated circles show up!
Formula-generating routine
In Fig. 4.44b, you see a parabola connected to the horizontal axis by vertical lines.
The spreadsheet calculation for the coordinates is shown in Fig. 4.44a (S).
24. The y-value of the parabola is calculated in the usual way with named cell
ranges (see the formula in B5). Every tenth point of the parabola is connected
to the horizontal axis with a vertical line (“Dashes”). Write a routine that
writes the formulas for the vertical lines’ coordinates into columns D and E!
25. Create a corresponding Python program with two variants. (a) The vertical
lines are plotted one after the other in a loop. (b) The coordinates of all vertical
lines are stored in lists named D and E separated by empty cells, and plotted
as one data series.
A B C D E 80
1 a.1 2.00 c.1 4.00
y=2(x-4)²+4
2 b.1 4.00 dx 0.10 y
3
4 y=2(x-4)²+4 Dashes
60 Dashes
+c.1
.1)^2
*(x-b
5 =a.1 =A17 =B17 40
6 x y x.s y.s
7 0.00 36.00 0.00 36.00
8 0.10 34.42 0.00 0.00
9 0.20 32.88 20
10 0.30 31.38 1.00 22.00
11 0.40 29.92 1.00 0.00
12 0.50 28.50
13 0.60 27.12 2.00 12.00 0
107 10.00 76.00 0 2 4 6 8 x 10
Fig. 4.44 a (left, S) Polynomial y = a1 (x − b1 )2 + c1 in columns A and B; coordinates x s , ys for

the vertical lines of Fig. b in columns D and E. b (right) Display of the data from a
4
A B C D E F G H I J
y 1 G, 2P
2 x y
3 1 -1.8 2.3
2 4 2 3.20 -1 4
5
6 G, 3.P S S, mirrored
0)
0 $4;B1
x 4 $4;$C
-4 -2 0 2 $C$3;$B4 }
) 3;
ND( ht($B$ D()*$F$ D()*$F$ 4 ;F10)
G, 2P 8*RA ig r(E10
7 =-4+ =Stra =RAN =RAN ={Mir
G, 3.P
-2 8 x y xS yS xSm ySm
S 9 1.89 -0.14 3.72 2.58 -3.72 -2.58
S, mirrored 10 3.27 -1.05 3.60 3.85 -3.60 -3.85
11 1.63 0.03 1.81 1.11 -1.81 -1.11
12 0.76 0.61 2.97 0.62 -2.97 -0.62
-4
Fig. 4.45 a (left) Two points “G, 2P” define a straight line (18a); the points in the third quadrant
are mirrored at the zero point into the first quadrant (18b). b (right, S) Spreadsheet calculation for
a; the coordinates for the points on the straight line are in columns B and C; the coordinates in
columns H and I are the mirror images of the coordinates in columns E and F
User-defined spreadsheet functions

In Fig. 4.45a, the results of two user-defined spreadsheet functions are displayed.
One function adds additional points, “G, 3.P”, onto a straight line “G, 2P” defined
by two points. The other function mirrors points in the first quadrant at (0, 0) into
the third quadrant. The coordinates of the points in Fig. 4.45a are calculated in the
spreadsheet of Fig. 4.45b (S).
26. What are the names of the two user-defined functions reported in line 7 in
Fig. 4.45b (S)?
27. Write a function (VISUAL BASIC or Python) of the type y3 = f(x 1 , y1 , …)
that calculates the y-value y3 of a third point from the coordinates of the two
defining points of a straight line and the x-value x 3 of the third point!
28. Write a function (VISUAL BASIC or Python) of the type (x sp , ysp ) = f(x,y)
that mirrors the coordinates x and y of a point at the origin of the coordinate
system!
Macros
29. You want to trigger a macro whenever a slider named SCROLLBAR1 is changed.
What is the name of the associated macro?51
30. At a mail-order company, some data from all outgoing packets are entered in the
spreadsheet of Fig. 4.46 (S). Write a protocol routine (Visual Basic or Python)
that reads some data from Fig. 4.46 (S) and enters it into a table as in Fig. 4.47
(S)! The packets have to be numbered consecutively.
51 SUB SCROLLBAR1_CHANGE().
A B C D E F G H I
1 Name Mary B.
2 Running number 46
Protoc
3
4 Width (cm) w 5 Volume (l) V 1.80 =w*l*h/1000
5 Length (cm) l 18 Surface (m²) S 0.11 =(w*l+w*h+l*h)*2/10^4
6 Hight(cm) h 20 Time 07.09.2020 17:21 =NOW()
Fig. 4.46 (S) Table section in which the width, length, and height of packages, as well as the
sender’s name, are to be entered
K L M N O P Q R S
6 12 next free row
m m m²
7 Numb
er e h/cm ength/c eight/c olume/l urface /
Nam Time Widt L H V S
8 43 Otto L. 7.9.20 17:03 17 17 14 4.05 0.15
9 44 James L. 7.9.20 17:16 20 10 10 2.00 0.10
10 45 Henry M. 7.9.20 17:16 18 28 8 4.03 0.17
11 46 Mary B. 7.9.20 17:21 5 18 20 1.80 0.11
Fig. 4.47 (S) The data from Fig. 4.46 (S) are to be reorganized in this way
Basic Mathematical Techniques
5
With the methods learned in Chaps. 2–4 (list processing, programming

constructs for, if, sub/def), we practice differentiation, integration, and cal-
culating with vectors. We get to know a new technique: solving systems of
linear equations with matrix calculation.
5.1 Introduction: Calculus, Vectors, and Linear Algebra
Solutions of Exercises 5.3 (Excel), 5.5 (Python), 5.6 (Python), and 5.7 (Excel) can
be found at the internet address: go.sn.pub/VYYbJL.
Straight-line segment
Straight-line segments are central for vector calculation, and also for calculus,
because we approximate all curves by sequences of such elements. In the intro-
ductory Exercise 5.2, we calculate vector entities related to a straight-line segment,
e.g., line vector and mid-perpendicular, as well as length, slope, and area enclosed
with the x-axis.
Differentiation and integration

We get to know simple techniques with which the first and second derivatives of a
function can be obtained (Exercise 5.3) and with which a function can be integrated.
With integration, the area between the curve of a function and the x-axis (Sect. 5.4.1)
and the length of a curve (Sect. 5.4.2) can be calculated. We pay special attention to
the x-value over which the results must be plotted: at the beginning, in the middle,
or at the end of the interval for which the elementary calculation was done.

182 5 Basic Mathematical Techniques
Vectors in the plane

We discuss the addition and the scalar product of two vectors and convert planar
Cartesian and planar polar coordinates one into the other (Exercise 5.5). In Exercise
5.6, the tangents to and the perpendiculars on a polynomial are drawn. As a physically
relevant example, we calculate the forces acting on a car running through a banked
curve (Exercise 5.7). In Exercise 5.8, we calculate a mobile’s equilibrium by using
the mathematical construction of a weighted average.
Systems of linear equations

Systems of linear equations (with small rank) are solved by bringing them into matrix
form and forming the inverse matrix of the coefficient matrix (Exercise 5.9). With
this method, we get the coefficients of a polynomial from a given set of points on
the curve and determine currents in electrical networks with Kirchhoff’s rules. In
Python, we also use functions from the numpy.linalg library for these tasks.
Mathematical functions
In Sect. 5.10, some useful mathematical functions are listed, both in excel and
numpy notation, together with short descriptions.
5.2 Straight-Line Segment Under a Magnifying Glass
A straight-line segment is specified by its two endpoints. Its center, line

vector, and perpendicular vector are obtained with vector operations. The
basic pieces of calculus, dx, dy, dA (the area under the curve), are calculated.
5.2.1 Under a Magnifying Glass
The functions we are investigating in this textbook are all continuous. They are
approximated as polylines, i.e., as sequences of straight-line segments. Therefore,
all operations of calculus, such as differentiation, integration, and integration along
a line, are based on such properties of segments as length, slope, line vector, and
mid-perpendicular.
In this exercise, we specify a straight segment in a plane by its two endpoints A,
B formulated as positional vectors. The unit line vector AB of the segment and its
perpendicular AB_p to be erected at the midpoint AB_C are calculated with matrix
operations. An example is shown in Fig. 5.1.
The equation of the line passing along the segment is
y = y A + m · (x − x A ) (5.1)
5.2 Straight-Line Segment Under a Magnifying Glass 183
4 4
y y
2 2
0 0
-4 -2 0 2 x 4 -4 -2 0 2 x 4
segment AB segment AB
AB_ AB_
-2 -2
AB_p AB_p
{=AB_p+AB_C} {=AB_p+AB_C}
on line on line
-4 -4
Fig. 5.1 Line vector and mid-perpendicular of a straight-line segment. The cross is drawn with a
function based on the coordinates of the two endpoints, a (left) with the settings from Fig. 5.2 (S),
b (right) other settings
with m being the slope of the segment and (x A , yA ) the coordinates of point A.
In Fig. 5.1a, the point (ys , x s ) on this line for x s = −1.76 is represented by a
cross. We use vector calculation, e.g., P·AB to obtain the unit vector perpendicular
to the segment with P, the 90° rotation matrix, and AB, the unit line vector.
A, B the two endpoints of a straight segment in the plane

x A , yA coordinates of point A
x B , yB coordinates of point B
dx = (x B – x A ) their distance in the x-direction
dy= (y B – yA ) their distance in the y-direction
ds = dx 2 + dy 2 length of the segment
m slope dy/dx
y = yA + m · (x − xA ) function describing the line through the segment
A = (yA + yB )/2 · dx area between the segment and the x-axis
A = (x A , yA ) vector representation of point A
B = (x B , yB ) vector representation of point B
AB C = (A + B)/2 center of the segment
AB =(B - A)/ds
unit line vector
0 −1
P= 90°- rotation matrix.
1 0
AB_ p_ = P · AB unit vector perpendicular to the segment.
In the spreadsheet calculation of Fig. 5.2 (S), the coordinates of the starting point
A are determined with sliders in A2:B3, ranging from 0 to 100. The values in the
linked cells are transformed into coordinates ranging from −4 to 4 (G2:G3). The
coordinates of the endpoint B are typed directly into cells J2:J3.
The cells G2:G3 get multiple identifiers: x A and yA refer to single cells, whereas
A refers to the whole range and can be processed as a column vector. The same
applies to B.
The length ds of the segment is calculated in A6 from the coefficients of A and
B. It can also be calculated with the matrix formula {=sqrt(sumxmy2(B;A))}.
sumxmy2 stands for “Sum of all individual (x – y)2 ”. The innermost operation A
– B involves two matrices, but the output is only a scalar. So, we have to enclose
the formula in curly brackets and finish with the magic chord Ψ Ctl + Shift +
Return.
The result of this calculation is shown in Fig. 5.1a. The perpendicular unit
vector AB p is also drawn from the center AB C of the line to AB C + AB p
(J6:J7).
In Fig. 5.3 (S), we calculate the primary segments of calculus dx, dy, and the
area dA between the segment and the x-axis. Furthermore, we set up the equation
1 A_ B_
2 55.00 xA 0.40 =(D2-50)/12.5 xB -3.00
3 92.00 yA 3.36 =(D3-50)/12.5 yB -3.50
2)
-xA)^
)^2+(xB gth_A
B} _)
P_;AB
yB }
((yA- _)/len )/2} B_C
RT A_-B ULT( _+B_ =AB_p+A
4 =S Q {=( =MM {=(A {
5 ds AB_ AB_p AB_C
6 7.66 0.00 0.44 0.00 0.90 =D7 -1.30 -0.40
7 0.00 0.90 0.00 -0.44 =-D6 -0.07 -0.51
8
Fig. 5.2 (S) Coordinates of points A and B, line vector AB and center position vector AB C; the
perpendicular vector AB p is obtained from AB through matrix multiplication with P, the 90°
rotational matrix P presented in Fig. 5.3 (S)
L M N O P Q R S T
5 area dA 0.24 =(yA+yB)/2*dx P_
6 dx -3.40 =xB-xA 0 -1
7 dy -6.86 =yB-yA 1 0
8 slope m 2.02 =dy/dx
9
10 on line x -1.76 28
11 y -1.00 =yA+m*(x-xA)
Fig. 5.3 (S) Continuation of Fig. 5.2 (S). Characteristics of a straight-line segment important for
calculus, and the equation for the line running along the segment
5.2 Straight-Line Segment Under a Magnifying Glass 185
y = yA + m·(x − x A ) of the line along the segment. The value of x is determined

with a slider so that the cross in Fig. 5.1 runs along this line.
5.2.4 Plotting Vectors with Python Matplotlib
To plot vectors with Python, we have to use the function arrow of the library
pyplot,to be imported with import matplotlib.pyplot as plt. We inte-
grate this function into a user-defined function ArrowP with standard formatting
parameters (see Table 5.1).
In excel, arrowheads are a design feature of a line: … line/end arrow type.
The constructor arguments of plt.arrow comprise, among others, the
keyword arguments:
width float (default: 0.001) width of full arrow tail

fill bool
linestyle or ls {’-’, ’–’, ’-.’, ’:’, ”, (offset, on–off-seq), …}
linewidth or lw float or None
head_length float or None (default: 1.5 * head_width) length of
arrow head
overhang float (default: 0, triangular) fraction that the arrow is swept
back
Some arrows are plotted in Fig. 5.4 to demonstrate the effect of the construc-
tors. The arrow pointing upwards is drawn with our standard function ArrowP
(overhang = 1) in Table 5.1.
5.3 Differentiation
We learn how to approximate the first and second derivatives of a function

f(x) numerically with difference quotients between neighboring calculation
Table 5.1 User-defined function ArrowP for drawing an arrow from point P0 to point P1 in a
plane; the argument list of plt.arrow does not contain all possible keyword arguments (similar
to Tables 5.3, 5.4, 5.5 and 5.6)
1 def ArrowP(P0,P1,c="k",ls='-',lw=1,hw=0.2):
2 (x0,y0)=P0 #Decomposes the foot position vector
3 (x1,y1)=P1 #Decomposes the tip position vector
4 print(lw,hw)
5 #c has to be given as c="k", not c='k'
8 head_width=hw,overhang=1,fill=False,
9 linestyle=ls, color=c, linewidth=lw)
Fig. 5.4 Arrows plotted with

different constructor
arguments in the procedure
head of plt.arrow
points. The first derivative must be plotted over the middle between two grid
points, the second derivative over the central of three grid points.
5.3.1 First and Second Derivative
The first derivative of a curve is the slope of its tangent to the curve at the specified
x-value. The derivative of a function f(x) with respect to x is defined as
d f (x) f (x + x) − f (x)

= lim (5.2)
dx x→0 x
More specifically, Eq. 5.2 is called the right derivative.

Such an approach to the limit cannot be carried out for discrete functions; they
are specified as a list of x-values (x = […x i …]) and y-values (y = […yi …]).
Instead, we calculate the difference quotient of neighboring grid points i and i +
1:

d f xi +x2 i+1 f (xi+1 ) − f (xi ) yi+1 − yi
≈ = (5.3)
dx xi+1 − xi xi+1 − xi
For our discrete functions, the first derivative in the center of a segment is
approximated by y/x, the slope of the segment between adjacent points. The
values are to be plotted over the center of the interval. It is evident that the smaller
x is, the better the accuracy of the approximation.
5.3 Differentiation 187
The second derivative of a function is the derivative of the first derivative, thus,

d 2 f (x) d d
= f (x) (5.4)
dx2 dx dx
which is equivalent to applying Eq. 5.2 twice. The second derivative describes
the change of the slope, and is thus a measure of the curvature of the curve. The
difference equation for grid points x i at equal intervals x is

d 2 f (x) 1 f(xi+1 ) − f(xi )) f(xi ) − f(xi−1 )
≈ −
dx2 x x x
f(xi+1 ) − 2f(xi ) + f(xi−1 )
= (5.5)
x2
The second derivative at point x can be calculated directly without a detour via
the first derivative with function values at the grid points x, x – dx and x + dx.
It must be plotted over point x (x i in Eq. 5.5), the coordinate of the middle grid
point.
Sine function
To give an example, we differentiate the sine function, knowing beforehand that its
first and second derivatives are the cosine and the negative sine, respectively. So, we
can check whether our numerical calculations reproduce this result. This is indeed
confirmed in Fig. 5.5, with 100 calculation points in one period 2π. The shape of a
cosine is clearly visible in the numerically calculated y1d in Fig. 5.5a, and that of a
negative sine in y2d in Fig. 5.5b.
Oscillation of a mass-spring-system
Consider the oscillation z(t) of a mass-spring system. The second derivative z̈ with
respect to time is the acceleration a, which, in turn, is proportional to the restoring
1.5 1.5
y = sin(x) y = sin(x)
y2d
1.0 y1d 1.0 -Sin(x)
0.5 0.5
0.0 0.0
0 2 4 6 0 2 4 6
-0.5 -0.5
x x
-1.0 -1.0
-1.5 -1.5
Fig. 5.5 Derivatives of a sine function a (left) first derivative, b (right) second derivative, numer-
ically (dashed line) and theoretically (x) calculated, dx = 2π/100
force −kz of the spring. The curvature of the displacement is thus proportional to the
force. This leads to the simple equation of a harmonic oscillator based on Newton’s
law:
F = −k · z = m · a = m · z̈
or

k
z̈ = − ·z (5.6)
m
The curvature of the displacement is thus proportional but opposed to the force.
A sine function is a solution to this differential equation, as can be seen in Fig. 5.5b.
Question
Let y = sin(x) in Fig. 5.5 be the displacement of an oscillator. At which posi-

tions x is the speed of the oscillator maximum, and at which positions is it
zero?1
Composite function
We are now considering a composite function, i.e., for which the argument is not
just an independent variable, but also a function. For the numerical derivative, there
is no difference from a simple function. We build the two arrays x and y = sin(f(x))
and proceed with y/x.
In Fig. 5.6a, the function y = A · sin(kx) with A = 0.1 and k = 2π is differen-
tiated first once and then twice. The results are again the cosine and the negative
sine, however, with different amplitudes, 0.1 · 2π and -0.1 · (2π)2 , respectively.
Questions
What is the amplitude A of the curve A · sin(kx) in Fig. 5.6a?2

Let A be 10 times bigger than it is in Fig. 5.6a. How do you have to change
the scale of the left and the right y-axes to get the same appearance in the
figure?3
Why might it be advantageous to display a function f (t) = sin(2π t) instead
of f (x) = sin(x)?4
1 The speed is maximum at zero crossings (x = 0, π, 2π in Fig. 5.5, at the extrema in y1d ) and zero
at turning points (x = π/4, 3π/4 in Fig. 5.5).
2 A = 0.1.
3 Both axes also have to be scaled by a factor of 10: −4 to 4 and −40 to 40.
4 Then, a period duration has the length 1, a quarter period (π/2) the length 0.25, and a half period
(π) the length 0.5, thus always at simple rational numbers that are clearly visible on the x-axis.
0.4 4 0.4 4
<-- y=A∙sin(kx) <-- y=A·sin(b∙x²)
y1d --> y1d -->
0.2 y2d -->
2 0.2 y2d --> 2
y2d_th
0.0 0 0.0 0
0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 x 0.75 1.00
x
-0.2 -2 -0.2 -2
-0.4 -4 -0.4 -4
Fig. 5.6 a (left) y=A · sin(2π x) and its first and second derivative. b (right) y = A · sin(bx 2 ) with
A = 0.1 and b = 4. its first and second numerical derivative, together with the analytical second
derivative Eq. 5.7. For the derivatives, the right y-axis is valid.
Why might it be advantageous to divide the values of f (t), the second

derivative of sin(2π t), by 4π2 ?5
In Eq. 5.7, we have chosen a parabola as an argument for the sine:
y = A · sin bx 2 (5.7)
The function y = A · sin bx 2 and its numerically obtained derivatives are

displayed in Fig. 5.6b with continuous lines.
Analytical derivatives with the chain rule

The derivatives of a composite function y = f (z), z = g(x) are obtained with the
chain rule:
dy d f dg
= · (5.8)
dx dz d x
For y = sin(2π x), we get
y = − cos(2π x) · 2π and y = − sin(2π x) · (2π )2 (5.9)
and for y = sin b · x 2 :
y = − cos bx 2 · 2bx
y = − sin bx 2 · (2bx)2 + cos bx 2 · 2b (5.10)
5 Then, we would expect an amplitude −1, which is easier to verify in the diagram.
A B C D E F A B C D E F
3 2 A 0.10
4 dx 0.0628 =2*PI()/100 3 k 6.28
4 dx 0.01
7)
9)/2 9-x) D7)/(xc-Cx)
dx x)
5 =A7+ =SIN( =(x+A =(B9- =(y1d =-SIN
y)/(A - ( 7)
N(k*
x) 9)/2 9-x) D7)/(xc-C
6 x y xc y1d y2d -sin(x)
dx
5 =A7+ =A*SI =(x+A =(B9- =(y1d
y)/(A -
7 0.00 0.00 0.03 1.00 0.00 6 x y xc y1d y2d
8 0.06 0.06 0.09 1.00 -0.06 -0.06 7 0.00 0.00 0.01 0.63
9 0.13 0.13 0.16 0.99 -0.13 -0.13 8 0.01 0.01 0.02 0.63 -0.25
105 6.16 -0.13 6.19 1.00 0.13 0.13 105 0.98 -0.01 0.99 0.63 0.49
106 6.22 -0.06 6.25 1.00 0.06 106 0.99 -0.01 1.00 0.63
107 6.28 0.00 0.00 107 1.00 0.00
Fig. 5.7 (S) Spreadsheet layout for calculating the first (y1d ) and second (y2d ) derivatives of the
sine function a (left) for sin(x) and b (right) for sin(kx)
Some points of y are also displayed in Fig. 5.6b with crosses. They lie on the curve
obtained with numerical differentiation, indicating that our procedure is correct,
especially our decision to plot the result over the middle of the three points from
which the second derivative is obtained.
x, y arrays specifying the points of the curve, x values dx apart,

xC array with the x-values of the center of the segments,
y1d array containing the values of the first derivative,
y2d array containing the values of the second derivative.
A spreadsheet layout for the differentiation of the sine function is given in Fig. 5.7
(S).
Questions
concerning Fig. 5.7 (S):

Interpret the formula in C5, valid for C8!6
What are the lengths of the arrays x, x C , y1d , and y2d ?7 Also, compare with
Table 5.2!
6 The spreadsheet formula in C5 calculates the x-value of the centre of the interval.
7 len(x) = R107 – R6 = 101, len(xC) = 100, len(y1d) = 100, len(y2d) = 99 (see also Table 5.2).
Table 5.2 a (top) Function deri for determining the derivative of y = f(x); b (bottom left) First
and second derivatives of y = sin(2π x); c (bottom right) Lengths of the arrays x, y1d, and y2d
1 def deri(x,y):
2 dx=x[1:]-x[:-1]
3 dy=y[1:]-y[:-1]
4 xC=(x[1:]+x[:-1])/2 #x Center of segment
5 return xC,dy/dx
6 x=np.linspace(0,1,101, len(x) 101
endpoint=True) len(y1d) 100
7 y=np.sin(2*np.pi*x) len(x1) 100
8 x1,y1d=deri(x,y) len(y2d) 99
9 x2,y2d=deri(x1,y1d) len(x2) 99
The formula in D8 is D8 = [=(B9-y)/(A9-x]. Which value does excel take

for x and y?8
How do you get a selection of –sin(x) in column F of 10 points, as presented
in Fig. 5.5b?9
We plot (x c , y1d ). Show that the formulas in E5 (valid for E8) is correct!10
In the Python program corresponding to Fig. 5.7 (S), the derivatives are built
with a function deri, shown in Table 5.2a, requiring x and y as input and returning
x C , the center of the intervals, and y/x. The Python program in Table 5.2b
treats y = sin(2π x).
The quantities dx, dy, and x C are obtained by slicing the arrays x and y. Their
length is, by construction, one less than that of x and y, as can be verified in Table
5.2c. We no longer have to think about the values of x for which the derivative is a
good approximation, because this is already done within the function and returned
as x C . For the second derivative, we apply deri twice.
Composite function
In Table 5.3, we are considering the composite function y = A · sin bx 2 (Eq. 5.10).
The results for the 10 x-values specified in line 7 are shown in Fig. 5.6b. The
numerically calculated values (marked –) and the theoretical ones (marked with x)
8 For x and y in C8 and D8, the values in the same (8th) row are taken from the column ranges with
names x (A8) and y (B8).

9 With a VBA procedure comprising a loop of the type: for r = 7 to 107, …, r2 = r2 +
1.
10 Row n-1: x-dx, xc = x-dx/2, y1d.
Row n: x, xc = x + dx/2,y1d,y2d.
To get y2d at the position x, we have to calculate the difference quotient for y at x + dx/2 and
x-dx/2.
Table 5.3 First and second derivatives of A · sin bx 2

1 x=np.linspace(0,1,101,endpoint=False)
2 A=0.1
3 b=4
4 yf=A*np.sin(b*x**2)
5 x1,yf1d=deri(x,yf)
6 x2,yf2d=deri(x1,yf1d)
7 xx=np.linspace(0,1,10)
8 yf2d_th=(A*np.cos(b*xx**2)*2*b
9 -A*np.sin(b*xx**2)*(2*b*xx)**2)
10 #Theoretical second derivative
of the second derivative coincide, indicating that our simple numerical recipe yields
sufficiently accurate derivatives.
Question
Over which arguments, type x or x c , must the function yf2d th in Table 5.3 be
plotted so that it correctly represents the second derivative of the function y?11
5.4 Integration
We determine the area under a sine curve and the length of a polynomial.
5.4.1 Area Under a Curve
Integral function, definite integral

The integral of a function f(x) between x 1 and x 2 corresponds, in the simple cases we
are dealing with, to the area limited by the curve, the x-axis, and two vertical bound-
aries. It can be positive or negative. The area under a curve between two adjacent
interpolation points is calculated using the trapezoid rule visualized in Fig. 5.8a. The
area of a trapezoid of width x is
f (x − Δx) + f (x)
FΔx (x) = · Δx (5.11)
2
The integral from x 1 to x 2 is the sum of all trapezoids in that region. It must be
represented in the diagram over x 2 , the end of the integration interval.
11The function yf2d th has to be plotted over xx, the equivalent of x, because it is the analytically
determined second derivative of A · sin(bx 2 ) at position xx.
5.4 Integration 193
1.0
yS
y
2 YsInt
d(YsInt)/dx
y
1
0.9 Trapez; dx
=0.251
0
0 2 4 x 6
0.8
x -1
1.0 1.2 1.4 1.6 1.8
Fig. 5.8 a (left) Numerical calculation of an integral with the trapezoidal rule, dx = 0.251. b
(right) Integral of the sine function and differentiation of the integral displayed over x (column A
in Fig. 5.11 (S))
Figure 5.9 shows the integral function (a) of a sine function and (b) of a poly-
nomial. The integral function of a sine is a cosine. The integral in Fig. 5.9a is
theoretically F(x) = (1 − cos(x)). The integral function of a polynomial of the nth
order is a polynomial of the (n + 1)th order.
Fig. 5.9 Curves obtained with the Python program in Table 5.4. a (left) Integration of the sine
function. b (right) Integration of a 3rd order polynomial (lower curve) (here, negative values of the
integral)
Questions
What is the analytical integral function of a sine?12

What is the area under a sine arc?13
In Fig. 5.8b, the numerically obtained derivative of the integral function is
shifted with respect to the original curve, contrary to the fundamental theorem
of calculus. What, do you suspect, is the reason for this?14
5.4.2 Length of a Curve
The length s of a straight segment is easily calculated as
s = x 2 + y 2 (5.12)
2
dy
ds = 1+ · dx (5.13)
dx
The length of a curve may be approximated by the length of the sequence of
segments used to approximate it, i.e., the sum of the lengths of all segments.
How is the function "length of a curve" correlated with the derivative of that
curve? Looking at Fig. 5.10a, we may state that the slope of the length is always
positive; and the bigger the absolute value of the derivative of the function, the
bigger the derivative of the length. Looking at Eq. (5.13), we see that the slope
ds/dx of the length is equal to the absolute value of the slope of y(x) if (dy/dx)2
1.
Question
Where does the curve "Length of a circle" as a function of the y value of the
circle in Fig. 5.10b cross the x-axis?15
5.4.3 Data Structure and Nomenclature for the Arrays

in the Integration
dx horizontal distance between the vertices
12 sin(x)d x = cos(x).
13 The area under a sine arc is 2, as can be seen form the value of the integral function at x = π in
b
Fig. 5.9a, based on the theorem a f (x)d x = F(b) − F(a).
14 The derivative is wrongly plotted over the end of the intervals, not correctly over their center.
15 The “Length of a circle” crosses the x-axis at x = 0 (for y = 0), x = π (for y = 0 after having
gone through a half circle, and x = 2π (for y = 0, after having gone through a full circle).
5.4 Integration 195
Fig. 5.10 a (left) Length of a polynomial, calculated with Eq. (5.12); with def lenCurve and
a second time as len2 in line 12 in Table 5.5. b (right) Length of a circle (reported on the x axis)
as a function of the y-value of the circle; attention: the horizontal axis represents the dependent
variable
xP x coordinates of a polynomial
yP y coordinates of the polynomial
Y pInt integral function of the polynomial
lenP length of yP
x x coordinates of a sine and a cosine
yS y coordinates of a sine function
yc y coordinates of a cosine function
Y sInt integral function of the sine function
lenCirc length of the circle defined by polar coordinates: x = yC and y = yS .
Area under a curve

A Python function, def inte, for performing the integration of a discretized function
y = f(x) with the trapezoid approximation is shown in Table 5.4. The returned array
integ has the same length as x. Its first element is zero, because that was introduced by
np.zeros(len(x)), and the first element is not overwritten. With the Python
program in the next cells, we perform integrations of a sine function (middle cell)
and a 3rd order polynomial (bottom cell). The results are displayed in Fig. 5.9.
Length of a polynomial
The approximation of Eq. 5.12 is implemented in the Python function def
lenCurve(x,y) in Table 5.5, returning the cumulated sum of the individual elements
ds (np.cumsum(ds)) as a function of x.
In the second cell of Table 5.5, the length of the polynomial yP (x) of Table 5.4 is
calculated, to be displayed in Fig. 5.10a. First, with lenP = lenCurve(x,yP) and then
(line 12) as len2, applying Eq. (5.13).
Table 5.4 First cell: function for performing an integration of y = f(x); second cell: f(x) = sin(x;
third cell: f(x) is a 3rd order polynomial
1 def inte(x,y):
2 dx=x[1:]-x[:-1]
3 yC=(y[1:]+y[:-1])/2 #y Center of segment
4 integ=np.zeros(len(x))
5 integ[1:]=np.cumsum(yC*dx)
6 return integ
7 x=np.linspace(0,2*np.pi,101, endpoint=True)
8 ys=np.sin(x)
9 yc=np.cos(x)
10 YsInt=inte(x,ys)
11 x=np.linspace(0,2,100,endpoint=True)
12 a,b,c,d=-1,2,-9,8
13 yP=a+b*x+c*x**2+d*x**3
14 YpInt=inte(x,yP)
Table 5.5 Function returning the length of the curve y = f(x); x and yP are defined in Table 5.4
1 def lenCurve(x,y):
2 ds=np.zeros(len(x))
3 dx2=(x[:-1]-x[1:])**2
4 dy2=(y[:-1]-y[1:])**2
5 ds[1:]=np.sqrt(dx2+dy2)
6 return(np.cumsum(ds))
7 FigStd('x',0,1,0.2,'y',-1.5,2.5,0.5)
8 plt.plot(x,yP,'k--')
9 lenP=lenCurve(x,yP)
10 plt.plot(x,lenP,'k-')
11 xC,y1d=deri(x,yP) #Derivative
12 len2=inte(xC,np.sqrt(1+y1d**2)) #Length accord. to formula
13 plt.plot(x[:-1],len2,'k+')
14 plt.show
Questions
Express lines 3 and 4 of Table 5.4 in one instruction! An nd.array is to be

returned!16
Design suitable labels for the plots in lines 8, 10, and 13 of Table 5.5!
Length of a circle
In Table 5.6, we calculate the length of a circle. The full length is reported in the
second cell. It is within 0.2 ‰ of the value of 2π. The length lenCirc (value on the
x-axis) as a function of the y coordinate yC of the circle (regarded as the independent
variable) is displayed in Fig. 5.10b.
16 Integ = np.array([0,*np.cumsum(yC*dx)])
5.4 Integration 197
Table 5.6 Calculating the length of a circle and plotting it as a function of the y-values of the
circle
1 phi=np.linspace(0,2*np.pi,101, endpoint=True)
2 yS=np.sin(phi) #x-coordinates of the circle
3 yC=np.cos(phi) #y-coordinates of the circle
4 lenCirc=lenCurve(yC,yS)
5 FigStd('x, lenCirc',-1,7,1,'y',-4,4,1)
6 plt.plot(yC,yS,'k--')
7 plt.plot(lenCirc,yS,'k-')
lenCirc[-1] 6.282
(2*np.pi) 6.283
5.4.5 Spreadsheet Solution
Area under the curve

A spreadsheet solution corresponding to Table 5.4 for the sine function is given in
Fig. 5.11a. The integration is performed in column C by operating on the individual
cells of y and taking the constant dx for all x i+1 − x i . If the horizontal distance
between the vertices is not constant, we have to replace dx in D8 by (x-A7), and so
on.
The integration in Fig. 5.11a (S) starts at x = 1.005 (A7). The corresponding start
value for the integration, zero, was entered into C7. The following cells accumulate
the areas of the trapezoids. The resulting integral Y sInt is shown in Fig. 5.8b, together
with yS , this, however, for dx = 0.251/10.
Question
According to the first fundamental theorem of calculus, the derivative of the

integral over f(x) should again yield the function f(x). In Fig. 5.8b, however,
A B C D E F G Sub Trapez() 1
4 dx 0.251 r2 = 7 2
For r = 7 To 11 Step 1 3
dx
7)/2* t-C7)/dx () Cells(r2, 6) = Cells(r, 1) 'x 4
dx ( x) (y+B
I n rapez
5 =A7+ =SIN =C7+ =(Ys Sub T Cells(r2, 7) = 0 5
6 x yS YsInt d(YsInt)/dx Trapez r2 = r2 + 1 6
7 1.005 0.844 0 1.005 0.000 Cells(r2, 6) = Cells(r, 1) 'x 7

Cells(r2, 7) = Cells(r, 2) 'y 8
8 1.256 0.951 0.23 0.90 1.005 0.844
r2 = r2 + 1 9
9 1.508 0.998 0.47 0.97 1.256 0.951
Cells(r2, 6) = Cells(r + 1, 1) 'x.next 10
10 1.759 0.982 0.72 0.99 1.256 0.000 Cells(r2, 7) = Cells(r + 1, 2) 'y.next 11
11 2.010 0.905 0.96 0.94 1.256 0.951 r2 = r2 + 1 12
12 2.262 0.771 1.17 0.84 1.508 0.998 Next r 13
107 26.138 0.844 0.00 0.76 End Sub 14
Fig. 5.11 a (left, S) Integration of y = sin(x), columns F and G contain the data series for the trape-
zoids shown in Fig. 5.8a. b (right, P) VBA procedure for writing the coordinates of the trapezoids
into columns F and G
d(YsInt)/ dx is shifted relative to the function yS (x) to the right. Inspecting

Fig. 5.11a, find out why that’s the case!17
In column D, the integral is differentiated, and the result is also displayed

in Fig. 5.8b as a function of x listed in column A. In this plot, the derivative
d(Y sInt )/dx is shifted to the right with respect to the original function. The reason
for this flaw is that the derivative is plotted over x (D8 is plotted over A8), the
end of the interval in which the derivative is built, instead of over the center of the
interval.
Coordinates of the trapezoids by a scan-log procedure

Figure 5.11b (P) shows the program code for a procedure that writes the coordinates
for the trapezoids from the table data for x and dY sInt /dx in Fig. 5.11a into the columns
F and G of Fig. 5.11a (S). Rows 7–12 of columns A and C are scanned with the loop
index r, and three points of a trapezoid are transferred with each loop cycle to F and
G, with the running index r 2 being incremented three times in each run. The result
“Trapez” is shown in Fig. 5.8a.
5.5 Vectors in the Plane
Polar and Cartesian coordinates are converted one into the other. Two vec-
tors are added, and their scalar product is built. Perpendicular bisectors
are erected on line segments. Arrows representing forces are attached to
application points in the xy-plane.
5.5.1 Vectors
Vectors in polar and Cartesian coordinates

Vectors have a magnitude l and a direction that, in the plane, can be determined by
the angle α to the positive x-axis. Alternatively, a vector can be defined by Cartesian
coordinates (V x , V y ). The two coordinate systems can be transformed one into the
other by
Vx = l · cos(α)
Vy = l · sin(α) (5.14)
17The derivative d(Y sInt )/ dx is plotted versus x listed in column A. The derivative calculated for a
segment is, however, to be plotted over the horizontal center of this segment.
5.5 Vectors in the Plane 199
l= Vx2 + Vy2
α = ar cus tangens Vx , Vy (5.15)
Attention: The order of the arguments in arcus tangens is different in the excel
and Python functions:
excel Python
atan2(x, y) np.arctan2(y, x)
Vector addition
Two vectors are added by adding their Cartesian coordinates individually:
W = V + U, Wx , Wy = Vx + Ux , Vy + Uy .
Vector addition is illustrated for forces in Fig. 5.12a with the axes F x and F y
scaled in units of N (Newton). The vectors are represented as arrows with their bases
in the origin of the coordinate system and the coordinates of their head points being
the coordinates of the vector. The resulting vector W points to the corner of the
parallelogram spanned by V and U.
In Fig. 5.12b, the plane xy is displayed, scaled in units of m with the arrows
representing the vectors of Fig. 5.12a attached at a point of application, here, (2, 4),
after being scaled with a scalar with the physical dimension m/N (here, 0.8 m/N)
to get the same physical unit as the axes, namely, m (meter). The scaling factor is
chosen so that arrows of convenient length result that fit into the chart.
10 10
Fy [N] y [m]
5 5
0 0
-10 -5 0 5 Fx [N]10 -10 -5 0 5 x [m] 10
U_
W_
-5 -5
V_ scal=0.8;
Attack=(2,4)
-W_
-10 -10
Fig. 5.12 a (left) Vector addition of forces in the (F x , F y ) plane with axes in physical units N;
two vectors U and V are added to produce the resulting vector W. b (right) Vector arrows attached
to a point in the xy-plane with axes in physical units m
Question
What is the physical unit of the scaling factor for forces in the xy-plane?18
Scalar product
The scalar product of two vectors V and U can be calculated in two ways:
– by multiplying their Cartesian coordinates and summing up the products:
U · V = Ux · Vx + U y · Vy (5.16)
– or using polar coordinates with the included angle γ = α U - α V :
U · V = lU · l V · cos(γ ) (5.17)

U, V, W vectors with two components
lU , lV , lW length of U, V, and W
aU , a V , a W angles of U, V, and W to the x-axis
Attack point of application in the plane, vector with two components (x
and)
Uat , Vat , Wat coordinates of heads of arrows attached to the point of application.
Fig. 5.13 (S) shows the spreadsheet layout for the drawings in Fig. 5.12.
The lengths and the angles of the two vectors are specified with sliders (scroll
bars). These polar coordinates are transformed into Cartesian coordinates in rows
7 and 8, where the scalar product of the two vectors is also calculated in two ways
(U·V and Scp), according to Eqs. (5.16) and (5.17), respectively. W is the sum
of U + V, written into C11:C12 with the matrix formula in curly brackets {=U_
+ V_}. The Cartesian coordinates are transformed into the polar coordinates (lW ,
α W ) in I10:I12. The length lW is calculated with W (1)2 + W (2)2 .
Questions
18 The physical unit of the scaling factor is m/N, so that the length of the arrows is in m.
A B C D E F G H I J
1 50 length lU 5.00 =E1/10
2 140 angle aU 0.87 =(E2-90)/180*PI()
3 71 lV 7.10 =E3/10
4 17 aV -1.27 =(E4-90)/180*PI()
5
6 U_ V_
7 .x 0.00 3.21 =lU*COS(aU) 0.00 2.08 =lV*COS(aV) U∙V -19.33 =SUMPRODUCT(U_;V_)
8 .y 0.00 3.83 =lU*SIN(aU) 0.00 -6.79 =lV*SIN(aV) Scp -19.33 =lU*lV*COS(aU-aV)
9
10 W_ -W_ lW 6.0614 =SQRT(INDEX(W_;1)^2
11 0.00 5.29 {=U_+V_} 0.00 -5.29 {=-W_} +INDEX(W_;2)^2)
12 0.00 -2.96 0.00 2.96 aW -0.51 =ATAN2(C11;C12)
Fig. 5.13 (S) Specification of two vectors U and V by their length and their angle to the x-axis with
sliders (rows 1–4); transformation to Cartesian coordinates (C7:F8); their scalar product (twice in
H7:I8); their sum (C11:D12); and the polar coordinates of the sum (I10:I12)
What is the apparent range of the numbers generated by the slider in

A2:D2?19
How do you get positive and negative angles between –π/2 and π/2 from an
always positive output of a slider with a range of 0–180?20
Where are the length and angle of the vector W = U + V calculated?21
With which instruction do you get the first entry in the named range W ?22
In rows 11 and 12 of Fig. 5.13 (S), the vector sum of the two vectors is built
with a matrix formula W_ {= U + V}. Addition is not possible in polar coordi-
nates; if the vectors are specified by length and angle, they must be converted to
Cartesian coordinates before addition.
Figure 5.14 (S) contains the extension of the calculation for obtaining the coor-
dinates of arrows representing vectors applied at a point (Attack) in the plane.
Attack is specified in Z4:Z5. The vector arrows in the xy-plane go from Attack to
Attack_ + U · scal, and so on. All four arrows can be entered into a chart together
as one series by specifying AB4:AL4 as series x- values and AB5:AL5 as
series y- values, because the respective ranges in the spreadsheet are separated
by empty cells.
Empty cells separate curves.
19 The slider in A2:D2 ranges from 0 to 180, angles in degree.

20 With a formula as in I2 = [=(E2-90)/180*Pi()], one can get positive or negative values.
21 The length and angle of the new vector W are calculated in I10 (l ) and I12 (a ).
W W
22 index(w_;1); in excel, the first entry is indexed as 1, contrary to Python where indexing starts
with 0.
Z AA AB AC AD AE AF AG AH AI AJ AK AL AM
2 scal 0.8 {=U_*scal+Attack_} {=V_*scal+Attack_} {=W_*scal+Attack_}{=-W_*scal+Attack_}
3 Attack_ U_ V_ W_ -W_
4 2.00 2.00 4.57 2.00 3.66 2.00 6.23 2.00 -2.23
5 4.00 4.00 7.06 4.00 -1.43 4.00 1.63 4.00 6.37
Fig. 5.14 (S) Calculating the coordinates of arrows representing the vectors U, V, W in the xy-
plane of Fig. 5.12a
The Python program corresponding to Fig. 5.13 (S) is given in the first two cells
of Table 5.7. The results of the calculation are shown in the third cell (bottom right)
of that table. They correspond exactly to the values in Fig. 5.13 (S). Lines 16-19
calculate the coordinates of the vector arrows in the plane, similar to Fig. 5.14
(S). W_.shape = (2,) indicates that W is an array with two elements. The values
for the scalar product and the polar coordinates of W are exactly the same as in
Fig. 5.13 (S).
Notice! The argument order of arcus tangens is different in the excel and
Python functions (see Sect. 5.5.1).
The main program in Table 5.8 calls ArrowP (Table 5.1 in Sect. 5.2.4) four
times. It is a continuation of Table 5.7 and can refer to the data, e.g., U at, specified
therein. The resulting diagram is shown in Fig. 5.15.
Table 5.7 a (top) Python program for defining two vectors with their length and their angle
to the x-axis, being transformed into Cartesian coordinates and their scalar product being built; b
(bottom left) the sum of the two vectors is built and the coordinates of all vectors are calculated
when attached to a point of attack in the plane; c (bottom right) reports values and shapes of some
variables
1 lU=5.0 #Length of vector U
2 aU=0.8727 #Angle to x-axis
3 U_=lU*np.array([np.cos(aU),np.sin(aU)])
4 lV=7.10
5 aV=-1.274
6 V_=lV*np.array([np.cos(aV),np.sin(aV)])
7
8 Scp=lU*lV*np.cos(aU-aV)
9 UV=U_@V_ #Dot product
10 W_=U_+V_ UV -19.33
11 lW=np.sqrt(np.sum(W_**2)) Scp -19.33
12 aW=np.arctan2(W_[1],W_[0])
13
W_ 5.29 -2.96
14 Attack_=np.array([2.00,4.00])
15 scal=0.8 W_.shape (2,)
16 U_at=U_*scal+Attack_ lW 6.06
17 V_at=V_*scal+Attack_ aW -0.51
18 W_at=W_*scal+Attack_
19 W_at_opp=-W_*scal+Attack_
Table 5.8 Continuation of Table 5.7; the program plots four arrows and the parallelogram of
forces, ArrowP from Table 5.1
20 FigStd('x',-10,10,5,'y',-10,10,5)
21 ArrowP(Attack_,U_at,lw=1.5)
22 ArrowP(Attack_,V_at,lw=1.5)
23 ArrowP(Attack_,W_at,lw=1.5)
24 ArrowP(Attack_,W_at_opp,lw=1.5)
25 plt.plot([U_at[0],W_at[0],V_at[0]],
[U_at[1],W_at[1],V_at[1]],'k--')
#Parallelogram, 3 sides
26 plt.axis('scaled') #Axis lengths to scale
Fig. 5.15 Vector diagram

corresponding to Fig. 5.12b,
but drawn with the Python
program in Table 5.8
Questions
How do you produce the labels shown in Fig. 5.12b (excel) and Fig. 5.15
(Python)?23
What is the effect of the statement plt.axis(‘scaled’)?24
23 Excel: “scal = ”&scal&”; Attack = (“&Index(Attack_,1)&”,”&Index(Attack_,2)&”)”.

Python: lbl = "scal = " + str(scal) + "\nAttack = " +
str(Attack_).
24 The lengths of the axes in the figure correspond to the scaling of the axes specified in the program
(see Fig. 5.15).

-1 0 1 2 3 4 5 -2 -1 0 1 2
2 2
x=1.6·cos(φ); y=1·sin(φ)
0.5+1x+-1x²+0.2x³
1
1
0
-1
-1 -2
Fig. 5.16 Tangents and perpendiculars, a (left) for a polynomial as obtained from the spreadsheet
in Fig. 5.18 (S), and b (right) on an ellipse constructed using polar coordinates (see legend)
5.6 Tangents to and Perpendiculars on a Curve
Tangents are fixed to a curve and perpendiculars are erected on it, taking a
third-degree polynomial as an example. All calculations are performed with
vectors. Python uses a loop to draw a multitude of segments. In contrast,
in the excel realization, the curve is calculated in a spreadsheet, and the
segments’ coordinates are generated with a scan-log procedure.
5.6.1 At/On a Polynomial and an Ellipse
We determine and display tangents to and perpendiculars on a polynomial, first

with a Python program and then through the combination of a spreadsheet cal-
culation and a VBA procedure. An example is shown in Fig. 5.16a. In Fig. 5.16b,
the same construction is shown for an ellipse.
The coordinates of the ellipse are obtained with
x = a X · cos(φ); y = aY · sin(φ) (5.18)
Questions
What are the coefficients aX and aY of Eq. 5.18 for the ellipse in Fig. 5.16b25
25 From the legend, we infer aX = 1.6 and aY = 1.

5.6 Tangents to and Perpendiculars on a Curve 205
Table 5.9 Program for determining the endpoints of segments representing the tangents to and
the perpendiculars on a curve
1 #Define polynomial of 3rd order
2 a,b,c,d=0.5,1.0,-1,0.2
3 dx=0.1
4 x=np.arange(0,4+dx,dx)
5 y=a+b*x+c*x**2+d*x**3
6
7 #Segments, their centers and lengths
8 xC=(x[1:]+x[:-1])/2
9 yC=(y[1:]+y[:-1])/2
10 Dx=x[1:]-x[:-1]
11 Dy=y[1:]-y[:-1]
12
13 scal=5 #Scaling the length of arrows
14 #Tangentials 19 #Perpendiculars
15 xL=xC[1:]-Dx[1:]*scal 20 xpL=xC[1:]-Dy[1:]*scal
16 xR=xC[1:]+Dx[1:]*scal 21 xpR=xC[1:]+Dy[1:]*scal
17 yL=yC[1:]-Dy[1:]*scal 22 ypL=yC[1:]+Dx[1:]*scal
18 yR=yC[1:]+Dy[1:]*scal 23 ypR=yC[1:]-Dx[1:]*scal
In Fig. 5.16a, the perpendiculars do not seem to be perpendicular to the

curve, contrary to Fig. 5.16b. What is the reason for this?26
a, b, c, d coefficients of the polynomial y = a + b · x + c · x 2 + d · x 3

x x-coordinates of the vertices, dx apart
y y values of the function
x C , yC center of the segments
dx, dy lengths of the segments in the x and y directions
x L , yL left coordinates of the tangential segments
x R , yR right coordinates of the tangential segments
x P L, yP L left coordinates of the perpendicular segments
x P R, yP R right coordinates of the perpendicular segments.
A Python program for our task is given in Table 5.9. The segments’ centers and
lengths are elegantly obtained with one instruction each through slicing, as are,
again, the arrays for the coordinates of the tangential and perpendicular segments.
26 In Fig. 5.16a, the lengths of the x- and y-axes do not conform to the scaling (-1 to 5) and (-1 to
2). In Fig. 5.16b, the lengths of the axes have been adjusted to the scaling.
Table 5.10 Program for plotting the results of Table 5.9

1 FigStd('x',-1,5,1,'y',-1,2,0.5)
2 plt.plot(x,y)
3 for i in range(0,len(xC)-1):
4 plt.plot([xL[i],xR[i]],
[yL[i],yR[i]],'k-',lw=0.5)
5 for i in range(0,len(xpL)-1):
6 plt.plot([xpL[i],xpR[i]],
7 [ypL[i],ypR[i]],'k-',lw=0.5)
8 plt.axis('scaled') #Only effective downstream (at the end)
9 plt.savefig('PhEx 5.5 polynomial.png',dpi=1200)
Fig. 5.17 Tangents to and

perpendiculars on a
polynomial; the coordinates
are calculated in Table 5.9
and drawn with the program
in Table 5.10
Tangents are drawn through the center (x C , yC ) of a segment ranging from
(xC − d x · scal, yC − dy · scal) to (xc + d x · scal, yc + dy · scal).
Scal is a scalar chosen to give a suitable length to the segments representing

the tangents and perpendiculars in the figure. No explicit for-loop is necessary to
determine the various coordinates, contrary to the VBA code in Sect. 5.6.4. In
Python, this is implicitly done with slicing.
The result of the calculation in Table 5.9 is plotted by the program in Table
5.10 with the resulting chart in Fig. 5.17. Here, looping over all tangents and
all perpendiculars is chosen, contrary to Fig. 5.18b (S), where all coordinates are
written into one column for x and another one for y, with empty cells separat-
ing segments so that they can be entered as one series into the figure. With the
statement plt.axis(’scaled’), the axes’ lengths are adapted to the axes’
scaling. Here, the perpendiculars are visibly orthogonal to the curve, contrary to
Fig. 5.16a. With the last statement in Table 5.10, the diagram in Fig. 5.17 is stored
as a png file.
In the spreadsheet of Fig. 5.18a (S), the coefficients of a polynomial of the 3rd
degree are specified, and 41 points (x, y) on that curve are calculated similar to
5.6 Tangents to and Perpendiculars on a Curve 207
A B C D E 1 Sub TangVert() Cells(r2, 5) = yC - dy 16

1 a 0.50 cc -1.00 2 scal = 5 r2 = r2 + 1 17
2 b 1.00 d 0.20 3 r2 = 8 Cells(r2, 4) = xC + dx 18
3 0.5+1x+-1x²+0.2x³ 4 For r = 8 To 47 Cells(r2, 5) = yC + dy 19
4 dx 0.10 5 xA = Cells(r, 1) r2 = r2 + 2 20
5 6 yA = Cells(r, 2) 'perpendiculars 21
6 =a+b*x+cc*x^2+d*x^3 7 xB = Cells(r + 1, 1) Cells(r2, 4) = xC - dy 22
7 x y by VBA procedure 8 yB = Cells(r + 1, 2) Cells(r2, 5) = yC - dx 23
8 0.0 0.50 -0.15 0.36 9 'center r2 = r2 + 1 24
9 0.1 0.59 0.25 0.73 10 xC = (xA + xB) / 2 Cells(r2, 4) = xC + dy 25
10 0.2 0.66 11 yC = (yA + yB) / 2 Cells(r2, 5) = yC + dx 26
47 3.9 1.05 0.66 0.98 12 dx = (xB - xA) * scal r2 = r2 + 2 27
48 4.0 1.30 0.64 0.58 13 dy = (yB - yA) * scal Next r 28
14 'tangentials End Sub 29
15 Cells(r2, 4) = xC - dx 30
Fig. 5.18 a (S, left) Table for calculating the x and y coordinates of a 3rd order polynomial; the
columns D and E from row 8 contain the outputs of the VBA procedure in (b). b (P, right) VBA pro-
cedure for calculating the coordinates of the segments representing the tangents and perpendiculars
to the curve y = f(x) and storing them all in columns D and E
Table 5.9. The coordinates of the segments, representing the tangents at and the
perpendiculars on the curve, are calculated in a VBA procedure sub TangVert in
Fig. 5.18b (P).
The subroutine is of the type scan-log and applies the construction loop2i The
loop index r runs down A8:A47, scanning the values in columns 1 and 2. The
index r 2 specifies the row of the output in columns D and E and is incremented
in lines 17, 20, 24, and 27. In each cycle, the coordinates of the tangent and
those of the perpendicular are calculated and written one after the other into the
same columns, always separated by blank lines so that they are plotted as isolated
segments when entered as one series into a chart.
5.7 Banked Curve
We calculate the forces acting on a vehicle running through a banked curve.

The gravitational and centrifugal forces are decomposed into vectors on the
road and perpendicular to it. The components are combined to get the forces
pressing the vehicle onto the road and pushing it perpendicular to its track on
the road.
5.7.1 Cross-Section of the Road
Turns on roads are generally banked so that the surface is inclined towards the
inside of the turn. The reason for this is that the car is in less danger of being
pushed out of the turn if the centrifugal force C is not fully working parallel to the
road’s surface. In addition, the gravitational force G also has a component parallel
0 0
0 2 4 x 6 8 10 0 2 4 6 x 8 10
plane
-2 -2
I, O
y y
GC
-4 -4
G+C
Itot, Otot
-6 -6
plane
-8 -8 GC
Components in the plane
orthogonal to the plane
-10 -10
Fig. 5.19 a (left) The total force G + C is decomposed into components parallel and orthogonal to
the line (cross-section of the road). b (right) The two vectors G and C are decomposed individually
into components parallel and orthogonal to the bank line. Inclination angle α = 40°
to the surface towards the inside of the turn. We represent the situation with a
straight line ("bank line") describing the road’s cross-section.
We treat the task with 2-dimensional vectors, namely, unit vectors parallel and
orthogonal to the bank line and vectors representing the forces acting on the car,
gravitational mg and centrifugal –mv2 /r, where m, v, and r are the mass and veloc-
ity of the vehicle and the radius of the curve, respectively. The signs are valid for
the situation of Fig. 5.19. As the mass occurs in both forces, we need not consider
it explicitly when interested in comparisons between the two forces. So, we use
accelerations with sizes G = g and C = v2 /r. The components of the accelerations
in the bank line and perpendicular to it are calculated with scalar products with
the in-plane and out-of-plane unit vectors.
The acceleration parallel to the road drives the car out of its track. The accel-
eration perpendicular to the bank line presses the vehicle onto the road, thus
determining the frictional force.
Questions
What are convenient physical units of G and C?27

Is it sensible to interpret G and C as accelerations?28
27 As we have defined these quantities as accelerations, their unit is N/kg.

28 Open to debate. Pro: The mass does not play any role. But does the frictional force depend on
the mass?
5.7 Banked Curve 209
α inclination of the track towards the horizontal

I = [cos(α), -sin(α)] unit line vector
O = [sin(α), cos(α)] unit vector orthogonal to the bank line
G = [0, 9.81] gravitational acceleration [N/kg] = [m/s2 ]
rr radius of the curve
v speed of the car
C = [−v2 /rr, 0] centrifugal acceleration
Attack point of attack of the forces
GI , CI components in the plane
I tot = GI +CI total force in the plane
GO , CO components orthogonal to the plane
Otot = GO +CO total force orthogonal to the plane.
In the first cell of Table 5.11, the four basic vectors, line vector I, orthogonal vec-
tor O, gravitational acceleration G, and centrifugal acceleration C, are determined
from the parameters of the exercise: inclination angle α, gravitational acceleration
g, speed v of the car and radius rr of the curvature of the road. In the second and
third cell of that table, vector operations are applied to decompose the accelera-
tions into components parallel and orthogonal to the bank line. G@I is a matrix
multiplication, for one-dimensional vectors equivalent to the scalar product. The
resulting values are listed in Table 5.12. They have the same values as those in the
spreadsheets in Sect. 5.7.4.
In Table 5.13, the coordinates of the arrows, representing the vectors in the xy-
plane with suitable length, are calculated. The starting point Attack of all arrows
is given as multiple PoA of the unit line vector, starting at the origin (0, 0) of
Table 5.11 First cell: specifying line vectors I (in-plane) and O (orthogonal to plane), and grav-
itational G and centrifugal C acceleration
1 a=40 #Angle of inclination, degrees
2 lbl_1="α="+str(a)+"°"
3 a*=np.pi/180 #Angle in radian
4 rr=100 #[m], Radius of curve
5 v=40 #[m/s], Speed of car
6
7 I_=np.array([np.cos(a),-np.sin(a)])
8 O_=np.array([-I_[1],I_[0]])
9 G_=np.array([0,-9.81]) #Gravitational acceleration
10 C_=np.array([-v**2/rr,0]) #Centrifugal acceleration
11 GI_=G_@I_*I_ 14 GO_=G_@O_*O_
12 CI_=C_@I_*I_ 15 CO_=C_@O_*O_
13 Itot_=GI_+CI_ 16 Otot_=GO_+CO_
Table 5.12 Numerical values of the vectors specified in Table 5.11, values for GO_ and CO_ are
the same as in Fig. 5.21 (S)
I_ [ 0.77 -0.64] GI_ [ 4.83 -4.05]
O_ [ 0.64 0.77] CI_ [-9.39 7.88]
G_ [ 0.00 -9.81] Itot_ [-4.56 3.83]
C_ [-16.00 0.00] Otot_ [-11.44 -13.64]
Table 5.13 Calculating the coordinates of the arrows in Fig. 5.19; PoA is a scalar determining the
point of attack on the bank line
1 PoA=7 7 CI=CI_*scal+Attack_
2 scal=0.3 8 Itot=Itot_*scal+Attack_
3 Attack_=np.array(I_*PoA) 9 GO=GO_*scal+Attack_
4 G=G_*scal+Attack_ 10 CO=CO_*scal+Attack_
5 C=C_*scal+Attack_ 11 Otot=Otot_*scal+Attack_
6 GI=GI_*scal+Attack_ 12 GpC=(G_+C_)*scal+Attack_
Table 5.14 Plotting the arrows that represent the various accelerations in Fig. 5.19a with ArrowP
1 FigStd('x',0,10,2,'y',-10,0,2)
2 PoA=7
3
4 plt.plot([0,20*I[0]],[0,20*I[1]],c='0.7',lw=3,ls='-')
#Plane
5 At_=I_*3 #Attack point of the unit vectors
6 scal=1
7 Arrow(At_[0],At_[1],
8 At_[0]+I_[0]*scal, At_[1]+I_[1]*scal)
9 Arrow(*At_,*(At_+O_*scal)) #* Decomposes the array
10
11 At_=Attack_
12 Arrow(*At_,*G_)
13 Arrow(*At_,*C_)
14 Arrow(*At_,*GpC_)
15 Arrow(*At_,*Itot_)
16 Arrow(*At_,*Otot_)
17 #Parallelogram of forces:
18 plt.plot([C[0],GpC[0],G[0]],[C[1],GpC[1],G[1]],'k:')
19 plt.plot([Itot[0],GpC[0],Otot[0]],
[Itot[1],GpC[1],Otot[1]],'k:')
the bank line. The endpoints are obtained with instructions like G = G *scal +
Attack. Plotting the arrows is achieved with Table 5.14.
Question
Interpret the statement in Table 5.11: G_@I_*I_!29
29 G_@I_is a matrix multiplication of two vectors, equivalent to G_[0]*I_[0] + G_[1]*I_[1],

yielding a scalar. G_@I_*I_is the multiplication of the vector I_ with this scalar.
5.7 Banked Curve 211
A spreadsheet solution of the banked-curve problem is shown in Fig. 5.20 (S).

We specify the line vectors I and O and the forces G and C as column vectors
with two components. We have to pay attention to operator precedence when cal-
culating the centrifugal force –v2 /rr. The values of the coordinates are the same as
in Table 5.12, left cell.
In excel, the sign operator – has precedence over the power operator
ˆ. We therefore have to enter = −(vˆ2/rr). In Python, −v**2/rr will
accomplish the task.
In Fig. 5.21 (S), G and C are decomposed into components parallel and orthog-
onal to the banked curve’s characteristic line. The scalar product is obtained with
sumproduct. In Fig. 5.22 (S), the components of the corresponding arrows in the
plane are calculated.
Attack is the starting point of all arrows, and G, C, GI, CI, I tot , GO, CO, Otot are
the endpoints of the arrows representing the various accelerations in the xy-plane.
We can create a living figure by introducing sliders to vary the parameters α,
rr, and v.
5.8 Weighted Average
Inclinaon of road angle 40 ° I_ O_ G_ C_

a 0.70 radian 0.77 =COS(a) 0.64 =SIN(a) 0 -16 =-(v^2/rr)
Radius of curve rr 100 m -0.64 =SIN(a) 0.77 =-COS(a) -9.81 0
velocity of car v 40 m/s plane
144 km/h 0 15.32 {=I_*20}
0 -12.86
Fig. 5.20 (S) Specifying line vectors I and O, and gravitational G and centrifugal C acceleration,
same specifications as in Table 5.11, same values as in Table 5.12
Components in the plane orthogonal to the plane
)*I_ _;I_)*I_ _)*O_ _;O_)*O_

(G_;I_ (C (G_;O (C
PRO DUCT PRODUCT CI_ PRO DUCT PRODUCT +CO_
=SUM =SUM =GI_+ =SUM =SUM =GO_
GI_ CI_ Itot_ GO_ CO_ Otot_
4.83 -9.39 -4.56 -4.83 -6.61 -11.4
-4.05 7.88 3.83 -5.76 -7.88 -13.6
Fig. 5.21 (S) Continuation of Fig. 5.20 (S), decomposing the gravitational and centrifugal forces
into components parallel and orthogonal to the bank line
poA 7 scal 0.3

l}
_} _} k_} Aac
k_} )*sca
ack ack Aac _+C_
* poA _* scal+A I_*s cal+A o t_ *scal+ tot_*scal+ ack_+(G
=I_ {=G {=G {=It {=O {=A
Aack_ G C GI CI Itot GO CO Otot G+C
5.36 5.36 0.56 6.81 2.55 3.99 3.91 3.38 1.93 0.56
-4.50 -7.44 -4.50 -5.72 -2.14 -3.35 -6.23 -6.86 -8.59 -7.44
Fig. 5.22 (S) Continuation of Fig. 5.21 (S), calculating the coordinates of the vector arrows to be
drawn in the xy-plane; the values of the coefficients are to be compared with the corresponding
values in Table 5.12, right cell
The equilibrium of a mobile with two arms is calculated with the law of
the lever. The mathematical construct is a weighted average. The calculation
uses vectors for forces and arms.
5.8.1 A Mobile with Two Arms
Figure 5.23a shows a mobile with one horizontal crossbar and weights attached at
its left and right ends, balanced by a counter-force applied at the center of gravity
of the construction. The crossbar is supposed to be weightless. The equilibrium is
calculated with the law of the lever:
g · m L · xL = g · m R · xR
2
Crossbar
Left weight
1 Right weight
Counterweight
y
-1
-0.5 0.0 x 0.5
Fig. 5.23 a (left) Mobile with one horizontal crossbar. b (right) excel menu for editing the
series “Crossbar” in a; the arrowheads are introduced by format data series/series
options/fill&line/end arrow type
5.8 Weighted Average 213
where mL and mR are the masses attached at the ends of the bars and x L and x R
their distance to the fulcrum.
We use the vector formulation for the equilibrium of torques:
WL × XL = WR × XR
The construction is to be done with all positions and forces specified as two-
dimensional vectors. Weights and counter-force are to be represented as arrows in
the xy-plane.
The x-coordinate x C of the equilibrium point, where the counter-force has to
be applied to hold the mobile in equilibrium, is calculated as a weighted average
with the masses as weights:
m L · xL + m R · xR
xC = (5.1)
mL + mR
L left arm of the crossbeam

R right arm of the crossbeam
xL , xR length of the arms of the crossbeam from x = 0
mL , mR masses at the crossbeam
wL weight at the left arm
wR weight at the right arm
CoG center of gravity (x C , 0)
xC horizontal position of center of gravity, from x = 0
wAnti upward counter force applied at CoG.
All underlined entities are two-dimensional vectors represented in the programs

with an underscore at the end, e.g. L_.
In Table 5.15, the horizontal positions (distance to x = 0) of the end of the arms
of the crossbar on the horizontal y = 0 and the forces are specified.
In Table 5.16, the arrows representing the weights at the crossbar’s ends and the
counter-force to be applied at the equilibrium point are plotted with ArrowP from
Sect. 5.2.4. A diagram like the one in Fig. 5.23a results (lines 16–18 ignored).
Table 5.15 Specifying the vectors (arms of the crossbar, weights and counter-force) of the mobile
in Fig. 5.23a
1 #Scalars 7 #Vectors
2 xL = -0.25 8 L_ = np.array([xL,0])
3 xR = 0.274 9 R_ = np.array([xR,0])
4 mL = 0.88 10 CoG_ = np.array([xC,0])
5 mR = 0.33 11 wL_ = np.array([0,-mL])
6 xC =(mL*xL+mR*xR)/(mL+mR) 12 wR_ = np.array([0,-mR])
13 wAnti_= np.array([0,mL+mR])
Table 5.16 Plotting the vectors of Table 5.15 as arrows in the plane of Fig. 5.23a
14 FigStd('x',-0.5,0.5,0.5,'y',-1,2,1)
15 plt.plot([xL,xR],[0,0],'k-')
16 plt.plot(*(L_+wL_),'kd',fillstyle="none")
17 plt.plot(*(R_+wR_),'kd',fillstyle="none")
18 plt.plot(*(CoG_+wAnti_),'ks',fillstyle="none")
19 ArrowP(L_,L_+wL_,hw=0.03)
20 ArrowP(R_,R_+wR_,hw=0.03)
21 ArrowP(CoG_,CoG_+wAnti_,hw=0.03)
Questions

What are the arguments transferred by *(L_+wL_) to plt.plot(…)?30
What do the instructions in lines 16–18 do (result not appearing in
Fig. 5.23a).31
A spreadsheet calculation corresponding to the Python program in Table 5.15 is

laid out in Fig. 5.24 (S). All vectors are specified as column vectors with two
coordinates, one for the x and the other for the y direction. In the lower half from
left to right, we have the arms L and R of the crossbar, the arrows representing the
weights at the left and the right end, and the counter-force applied at the center of
gravity.
30 L_ + wL_= [xL, -mL] (an array), *[xL, -mL] = xL, -mL (the elements of an
array).
31 The instructions plot diamonds at the tips of the weights and a square at the tip of the counter-
force.
5.8 Weighted Average 215
1 xL -0.25 {=-(wL_+wR_)}
2 xR 0.274 L_ R_ CoG_ wL_ wR_ wAn_
3 mL 0.88 -0.3 =xL 0.27 =xR -0.1 =xC 0 0 0
4 mR 0.33 0 0 0 -0.9 =-mL -0.3 =-mR 1.21
5
6 xC -0.107 =(mL*xL+mR*xR)/(mL+mR)
7
8 =L_ =R_ =L_ =L_+wL_ =R_ =R_+wR_ =CoG_ =CoG_+wAn_
9
10 x -0.3 0.27 -0.3 -0.3 0.27 0.27 -0.11 -0.1
11 y 0 0 0 -0.9 0 -0.3 0 1.21
12
Fig. 5.24 (S) Defining the elementary vectors and calculating the anti-force
Question
How do you enter the three arrows into a figure as in Fig. 5.23a?32
5.9 Systems of Linear Equations
We treat two problems resulting in a system of linear equations solved

with matrix inversion: calculating the standard form of a polynomial of the
3rd degree from four points and determining the currents in a network of
resistances, voltage, and current sources with Kirchhoff’s rules.
5.9.1 Polynomial and Electrical Network
Polynomial
A third-degree polynomial is specified by 4 coefficients:
y = f (x) = a + b · x + c · x 2 + d · x 3 (5.19)
They can be calculated if 4 points (xP i , yP i ) are specified in a plane, leading to 4

linear equations:
yi = a · 1 + b · xi + c · xi2 + d · xi3 i = 1, 2, 3, 4 (5.20)
Two examples are shown in Fig. 5.25.
32 G10:N10 as an X-series and G11:N11 as a Y-series.

Fig. 5.25 3rd-order polynomials generated, a (left) with the Python program in Sect. 5.9.4 and b
(right) with the excel spreadsheet in Sect. 5.9.3
Matrix of powers of x
The system of linear equations is put into matrix form, with the matrix P composed
of the powers (0, 1, 2, 3) of x i (“matrix of powers”) and the known yi as a vector on
the right side of the equation:
⎡ ⎤ ⎡ ⎤ ⎡ ⎤
1 x1 x12 x13 a y1
⎢ · ⎥ ⎢ ⎥ ⎢ ⎥
⎢ ⎥ · ⎢ b ⎥ = ⎢ y2 ⎥ (5.21)
⎣ · ⎦ ⎣ c ⎦ ⎣ y3 ⎦
2
1 x4 x4 x4 3 d y4
P · coe f f = y P
The vector of unknowns contains the coefficients: [a, b, c, d]. The coefficients
may be obtained by applying P−1 , the inverse of P, to yP :
coe f f = P −1· y P (5.22)
The matrix of powers has to be non-singular; otherwise, its inverse cannot be

built. A square matrix is nonsingular if and only if its determinant is nonzero. The
determinant is obtained with mdet (excel) or npl.det (Python).
Question
What is the determinant (< 0, = 0, > 0) of the matrix of powers if two x values
of the four defining points are identical?33
33The determinant is zero because it is impossible to draw a polynomial through two points when
one is above the other one.
5.9 Systems of Linear Equations 217
Fig. 5.26 Electrical circuit I2

comprising a current source
I0, a voltage source U0, and
three ohmic resistors R1, R2, I1 R2 I4 I3
I0
and R3
U0 R3
R1
Electrical Networks
In Fig. 5.26, an electrical network with three ohmic resistors R1 , R2 , and R3 , a
constant current source I 0, and a constant voltage source U 0 is shown. We are going
to calculate the currents I 1 , I 2 , I 3 , and I 4 in the different branches. The directions of
these currents (i.e., the directions of the arrows in the figure) are arbitrary, but they
must be chosen at the very beginning. The signs of the currents are with respect to
these directions.
Kirchhoff’s rules as matrix equation

The four linear equations for the four unknown currents I 1 , I 2 , I 3 , and I 4 are obtained
with the help of Kirchhoff’s rules, namely, with two mesh rules (the voltage around
a closed circuit must be zero):
R3 · I3 − U0 = 0 and − R1 · I1 + R2 · I2 + U0 = 0 (5.23)
and two junction rules (the currents flowing into a junction must equal the currents
flowing out of the junction):
−I1 − I2 + I0 = 0 and I2 + I4 − I3 = 0 (5.24)
These four equations are transformed so that the known source voltage U 0 (or 0)
is on the right side of the mesh rules and the known source current I 0 (or 0) is on the
right side of the junction rules so that they can be expressed as a matrix equation:
Res · I = Sr cs (5.25)
Res is a matrix with components 0 or 1 or the resistances of the circuit, I is the

vector of the unknown currents, and Srcs is a vector whose coefficients are zero or
the known source voltages and currents. An example for the network in Fig. 5.26 is
given in the spreadsheet of Fig. 5.27 (S).
Questions

Fig. 5.27 (S) Eq. (5.25) applied to the network of Fig. 5.26. The matrix equation is presented in
rows 1–5 in general form and in rows 8–11 with the concrete values from column B. The ver-
tical lines, and the dots in column H, are intended to indicate matrix calculation. They have no
computational function in the spreadsheet
Verify that the matrix calculation in rows 8–11 is identical to Eqs. 5.23 and
5.24!
What are the formulas in D10:G10?34
Polynomial
x P , yP coordinates of 4 points in the plane, specified as two vectors

A 4 × 4 matrix (“matrix of powers of x"), powers of x P in rows
AInv the inverse of matrix A
coeff solution a, b, c, d of the system of linear equations, coefficients of the
polynomial
Electrical circuit
R1 , R2 , R3 , R4 four ohmic resistances

I0 current of a constant-current source
U0 voltage of a constant-voltage source
Res square matrix containing resistances, zeros, and ones
Res I nv inverse of Res
Srcs vector containing the voltage and current sources or zero
Currents vector containing the currents to be determined with the system of
linear equations.
34 D10 = [=-R.1], E10 = [=R.2], F10 = [0], G10 = [0].

We need functions to build the inverse of a square matrix and to multiply matrices.
excel: minversemmult
Python: np.inv@ operator (or np.matmul)
5.9.3 Spreadsheet Solutions
Polynomial
The solution of Eq. 5.22 is implemented in the spreadsheet of Fig. 5.28 (S). The
coordinates of the defining points are stored in column vectors named xP and yP . The
matrix A is composed of column vectors obtained as powers of xP . To solve for the
coefficients of the polynomial, we have to build the inverse AInv of A to be applied
to yP in a matrix multiplication.
Electrical network
For the network of Fig. 5.26, the spreadsheet layout in Fig. 5.27 (S) applies, in rows
2–4 in general form and in rows 8 to 11 with the values from A7:B11. The indices
of the resistances and the constant current and constant voltage are separated from
the letters by a dot, because a name like R1 would be interpreted as a cell address.
The unknown currents are obtained by forming the inverse matrix to M to be
applied to the vector S of sources:
M −1 ·M · I = M −1 ·S or M −1 ·S = I
The solution for the matrix equation in Fig. 5.27 (S), applying the excel matrix
functions minverse to get the inverse matrix and mmult to perform matrix multipli-
cation, is given in Fig. 5.29 (S) where, in column N, we check whether Kirchhoff’s
rules are satisfied.
A B C D E F G H I J K L M N O P
)
E(A)} inv;yP
VERS ULT(A
2 =xP^3 =xP^2 =xP 1 {=MIN =MM
3 xP yP A Ainv coeff
4 1 1 1 1 1 1 -0.01 0.03 -0.02 0.01 0.14 dd
5 3 5 27 9 3 1 0.23 -0.44 0.29 -0.07 -2.04 cc
6 6 2 216 36 6 1 -1.24 1.92 -0.87 0.19 8.30 bb
7 9 9 729 81 9 1 2.03 -1.50 0.60 -0.13 -5.40 aa
8
Fig. 5.28 (S) Spreadsheet implementation of Eq. 5.22; Ainv is the inverse matrix of A; the result-
ing polynomial is presented in Fig. 5.25b
C D E F G H I J K L M N O
12
13 {=MINVERSE(Res)} {=MMULT(ResInv;Srcs)}
14 ResInv Scrs Cur
15 0.40 0.00 -0.01 0.00 I.0 0.23 I.1 -5.00 =-R.1*I.1+R.2*I.2
16
17
0.60
0.00
0.00
0.01
0.01
0.00
0.00
0.00
∙ U.0
-U.0
= 0.27
0.06
I.2
I.3
-0.50
5.00
=-I.1-I.2
=R.3*I.3
18 -0.60 0.01 -0.01 -1.00 0 -0.21 I.4 0.00 =I.2+I.4-I.3
19
Fig. 5.29 (S) Solving the matrix equation in Fig. 5.27 (S) with the inverse of Res. In column N,
Kirchhoff’s rules are checked
Table 5.17 Calculating the coefficients coeff of a polynomial of the 3rd order when 4 points (in
arrays (x P , yP )) are given
1 import numpy.linalg as npl
2 xP=np.array([1,3,6,9]) #Defining points
3 yP=np.array([[2,6,2,2]])
4 xP0_= np.array([1,1,1,1])
5 xP1= xP
6 xP2= xP**2
7 xP3= xP**3
8
9 A=np.array([xP0,xP1,xP2,xP3]).transpose(1,0)
10 yPt_= yP.transpose(1,0)
11 coeff_= npl.solve(A,yPt)
A coeff yPt
[[ 1 1 1 1] [[-4.00] [[2]
[ 1 3 9 27] [ 7.67] [6]
[ 1 6 36 216] [-1.78] [2]
[ 1 9 81 729]] [ 0.11]] [2]]
Question
The calculations in column N (reported in column O) of Fig. 5.29 (S) are

intended to check whether Kirchhoff’s rules are fulfilled. Which of the four
equations Eq. 5.23 left or right and Eq. 5.24 left or right are checked in
N15–N18?35
5.9.4 Python Programs
Polynomial from defining points

In Table 5.17, the coordinates of the defining points are specified in the arrays xP and
yP . The matrix A is put together in line 9, first as a list of the rows x P0 , x P1 , x P2 , x P3
that is then transposed to get a form as in Eq. (5.21); see also the printout in the lower
cell. Likewise, the vector yP is generated. In line 11, the system of linear equations
35 N15: Eq. 5.23 right; N16: Eq. 5.24 left; N17: Eq. 5.23 left; N18: Eq. 5.24 right.
Table 5.18 Python program for plotting the polynomial specified by the coefficients in Table 5.17
2 y=coeff[0]*x**3+coeff[1]*x**2+coeff[2]*x+coeff[3]
3 FigStd('x',0,10,2,'y',0,10,2)
4 plt.plot(x,y,'k-')
5 plt.plot(xP,yP,'kx') #Defining points of the parabola
is solved with the function solve of the numpy.linalg library. Alternatively,

the coefficients can be obtained by applying the inverse of the "powers of x" matrix
to the vector of pre-specified y values:
coe f f 2 = P Inv @ y P
The results of the two methods are identical. The coefficients are used to generate
the coordinates of the polynomial shown in Fig. 5.25a. The entities in the lower cell
of Table 5.17 are arranged so that they correspond to the equation:
P · coe f f = y Pt
The entity yPt is the transposed row vector yP . At least the first coefficient of yPt
can be checked by mental calculation.
Questions
The coefficients in Fig. 5.28 (S) are the same as in the Python solution in
Table 5.17, however, in reverse order. What is the reason for this?36
What is the more versatile version: (a) y = coeff[0]*x**3 + … or (b)
y = coeff[0] + coeff[1]*x?37
The program for realizing the plot in Fig. 5.25a is shown in Table 5.18.
Electrical network
A Python program for solving Kirchhoff’s equations corresponding to the spread-
sheet layout in Fig. 5.27 (S) and Fig. 5.29 (S) is presented in Table 5.19. The matrix
Res is defined as in the spreadsheet in Fig. 5.27 (S), whereas the vector of the sources
is first defined as a row vector Src and then transposed into a column vector Srcs
equal to the one in Fig. 5.27 (S) and to Srcs in column K of Fig. 5.29 (S). Lines 9
and 10 can be combined into one instruction.
36The matrix of powers P in Fig. 5.28 (S) contains the columns in reverse order to Table 5.17.
37Version (b) is more versatile, because it can easily be extended to higher order, … +
coeff[3]*x**4 + ….
Table 5.19 Solving Kirchhoff’s equation

1 import numpy.linalg as npl 9 Srcs=Src.transpose(1,0)
2 R1,R2,R3=100,68,90 10 ResInv = npl.inv(Res)
3 I0,U0=0.5,5 11 Cur=ResInv@Srcs
4 Res=np.array([[1,1,0,0], 12 CurS=npl.solve(Res,Srcs)
5 [0,0,R3,0],
6 [-R1,R2,0,0],
7 [0,-1,1,-1]])
8 Src=np.array([[I0,U0,-U0,0]])
ResInv:
[[ 0.40 -0.00 -0.01 -0.00]
[ 0.60 0.00 0.01 0.00]
[ 0.00 0.01 0.00 0.00]
[-0.60 0.01 -0.01 -1.00]]
Cur with ResInv (inverse matrix) CurS with linalg (solve)
[[ 0.23] [[ 0.23]
[ 0.27] [ 0.27]
[ 0.06] [ 0.06]
[-0.21]] [-0.21]]
The system of linear equations is solved in two ways, in line 11 with the
inverse matrix ResInv as in the spreadsheet calculation and in line 12 with the
function solve of the numpy.linalg library. Both methods yield the same
result (second cell in Table 5.19) as the spreadsheet procedure.
5.10 Some Mathematical Functions
Numpy, np.* excel

Basic numeric information
abs abs(x) absolute value
sign sign(x) sign (+1, −1 or 0)
Basic mathematical operations
sum sum(x; x 2 ) x + x 2 (no matrix function)
prod product(x;x 2 ) x · x 2 (no matrix function)
quotient(x;x 2 ) integer portion of x/x 2 , no matrix function
mod (x; x 2 ) Modulo, remainder from x/x 2
sqrt sqrt(x) Positive square root
Rounding functions
round round(x; N) Up or down, to N digits
roundup(x; N) Up to bigger absolute value, to N digits
rounddown(x; N) Down to smaller absolute value, to N digits
mround(x; x 2 ) Up or down to a multiple of x 2 , no matrix function
5.10 Some Mathematical Functions 223
Numpy, np.* excel

Ceil ceiling(x; x 2 ) Up versus ∞ to a multiple of x 2
Floor floor(x; x 2 ) Down versus −∞ to a multiple of x 2
Rint int(x) Down to the next integer
fix trunc(x) Towards zero to the next integer
Trigonometric functions
pi pi() Constant value of pi
rad2deg degrees Converts radians to degrees
deg2rad radians Converts degrees to radians
cos cos(α) Cosine of a given angle (rad)
arccos acos Inverse cosine [−1, 1] → [0, π]
sin sin(α) Sine of a given angle (rad)
arcsin asin Inverse sine [−1, 1] → [−π/2, π/2]
tan tan(α) Tangent of a given angle (rad)
arctan atan inverse tangent (arcus tangens)
[−∞, ∞] → [-π/2,π/2]
arctan2(y,x) atan2(x;y) Angle of a given pair of x and y coordinates;
attention: Order of arguments is different in excel
and Python!
Exponents and logarithms
Exp exp(x) e raised to the power of x
Log ln(x) Natural logarithm, inverse of exp
log(x;b) Logarithm of x to base b
log(x; b1 ) = log(x; b2 ) * log(b2 ; b1 )
log10 log10 Base 10 logarithm
log2 log(x; 2) Base 2 logarithm
Sums
Footnote 38 sumif Adds the cells in a supplied range that satisfy a
given criterium
Footnotea sumifs Adds the cells in a supplied range that satisfy
multiple criteria
sum(x * y) sumproduct(x;y) x * y over arrays
sum(x * x) sumsq(x) x 2 over arrays
sum(x2 -y2 ) sumx2my2(x;y) x 2 minus y2 over arrays
sum(x2 + y2 ) sumx2py2(x;y) x 2 plus y2 over arrays
sum((x–y)2 ) sumxmy2(x;y) (x – y)2 r arrays
Cumsum Footnoteb Cumulated sum
Functions of linear algebra
npl.det mdeterm Determinant of a square matrix
npl.inv minverse Inverse of a square matrix
np.matmul(A,B) (a;b) Matrix product of two matrices
or A@B
Numpy, np.* excel

np.dot(x,y) sumproduct(x;y) Dot product of two vectors
a Can be achieved with list comprehension, as in the following example: Rng =
np.linspace(1,20,20) ; x = [×1 for × 1 in Rng if ×1 > 3 if × 1 <
6]; sum(x)
b In a spreadsheet operation, B1 = [=A1]; B2 = [=B1 + A2]; … cumulates in column B the
values in column A
1. How do you get the anchor points of the inverse function of y = y(x)?38
Differentiation and integration
2. You have calculated the difference quotient between the boundaries of an interval.
Over which value do you plot the result?
3. What is the formula for the numerical second derivative of a function, specified
at equidistant positions with distance dx?
4. Over which positions of an interval (beginning, middle, end) are values for the
first derivative, the second derivative, and the integral to be plotted?
5. What does the trapezoidal rule of integration stand for?
Vectors
6. What are the Cartesian coordinates of vectors of length 1 (one) pointing in the
x direction, the negative y direction, and 45° from the x-axis?
7. What are the components of a vector pointing from (x1 ) = (4, 5) to (x2 ) = (6,
3)?
8. What are the lengths of the two vectors pointing from (0, 0) to (1, 1), and from
(0, 0) to (3, 4)?
9. How do you form a scalar product of two vectors A and B in two different ways,
(i) component by component, and (ii) using the angle φ between the two vectors?
10. What is the angle between two vectors if the scalar product is zero?
11. What are the coordinates of a unit vector in the direction of the segment from
[x 1 , y1 ] to [x 2 , y2 ]?
12. What are the coordinates of the perpendicular to the vector (x, y)?
38The inverse function is x = x(y). To get its interpolation points, you only have to swap the two
columns for x and y in the spreadsheet or the two lists in Python.
A B C D E F G H I J K L M N O P Q R S T U V
1 A B D
2 1 0 1 1 0 0 1 2 3 =MMULT(E2:G4;I2:K4)
3 0 1 0 0 1 0 4 5 6
4 0 0 1 7 8 9 =MMULT(A2:C3;I2:K4)
5
Fig. 5.30 (S) Five matrix ranges, named cell ranges: A = [A2:C3], B = [E2:G4], C = [I2:K4], D
= [I2:J4], x = [I2:K2], y = [I3:K3]
System of equations
13. Which arguments does the excel function mmult(?) have, which multiplies two
matrices? What is the equivalent in Python? What are the relations between
the widths and heights of the two matrices?
14. You have set up a system of equations M · I = I , with known matrix M and
known source vector S, but unknown vector I. By which instructions of type
matrix operation do you get the coefficients of I in excel and Python?39
Functional expressions
Calculate the values of the following expressions in excel:
15. arctan2(1;1), log10(0.001),

16. product(2;3;4;5), power(10;3).
17. round(3.74638,2), int(17.453), remainder(127; 2).
The functions in Questions 18 and 19 refer to Fig. 5.30 (S). What values result
for:
18. sumproduct(x;y) , sumxmy2(B;C)? What are the corresponding expressions

in Python?
19. mmult(B;C), mmult(A;D)? What are the corresponding expressions in
Python?
20. You are to determine the derivative dy/dx of a function defined by x =
np.linspace (0,10,11) and y = x**2. How do you determine dy and
dx with list slicing? What are the first two elements of dy/dx, and over which
values do they have to be plotted?
21. Define the arrays A, B, and C, in Python corresponding to the three
ranges A2:C3, E2:G4, and I2:K4, respectively, in Fig. 5.30 (S)! Calculate
B@C and A@C!
22. What are the two arguments in the function to calculate the angle of the vector
(1, 2) to the x axis in excel and Python?
excel: I = mmult (M);S). Python: I = npl.inv(M)@S; npl stands for numpy.linalg,

39
M and S are of type np.array.

Superposition of Movements
6
We learn how to compose complicated movements from simple ones, namely

translations and rotations in a plane. The exercises honor famous scientists:
Bernoulli (cycloid), Foucault (pendulum) , and Steiner (moment of inertia).
In spreadsheets, we systematically use sliders and macros with which we
have familiarized ourselves in previous chapters.
6.1 Introduction: Translations and Rotations
Solutions of Exercises 6.2 (Python), 6.3 (Excel), 6.4 (Excel), and 6.5 (Python) can
be found at the internet adress: go.sn.pub/or1CXF.
Simple movements
In this chapter, we put together movements in a plane from two simple movements:
Translations T, straight-line movements in one direction, generally defined by a
two-dimensional velocity vector (vx , vy );
Rotations R, rotations in the xy-plane, described by an angular velocity ωz and
the radius r of the trajectory.
Polar coordinates
We use polar coordinates (r, ϕ) to describe rotations and convert them, e.g., for
graphical representation in charts, to Cartesian coordinates (x, y):
x = r · cos(φ); y = r · sin(φ) (6.1)
Projectile motion, T-T (Exercise 6.2)

A projectile trajectory is composed of two linear movements. If friction is not taken
into account, these are a vertical one accelerated by gravity and a uniform horizontal

228 6 Superposition of Movements
one. We attach a velocity vector and its vertical and horizontal component to the
trajectory at a freely chosen point.
Cycloid, rolling curve, T-R (Exercise 6.3)

We calculate the trace of a point on a wheel rolling on a plane, resulting from a
translation of the wheel axis on a straight line and a rotation about the wheel axis,
with translation speed and angular speed being dependent on each other.
The resulting curve, called a rolling curve or cycloid, represents the brachis-
tochrone (the fastest path to fall from one point to a lower point) , as shown by
Johann Bernoulli. This will be treated in the follow-up book Physics with Excel and
Python, Using the Same Data Structure. Applications in the chapter “Calculus of
Variations”.
Foucault’s pendulum, T-R (Exercise 6.4)

What is the trace of a swinging pendulum on a rotating surface? It is obtained by
superposing the movement of the linear oscillation in the laboratory system with a
rotation of the base table on which the motion is recorded. Oscillation and rotation
are independent of each other.
This experiment has historical significance. Michel Foucault demonstrated with
a 67-m long pendulum suspended from the dome of the Pantheon in Paris that the
earth rotates against the fixed stars.
Swinging anchor, R-R (Exercise 6.5)

We consider an anchor in the form of a hanging T with three mass points attached,
one at each end of the T and one at the junction of the two lines. The mass points are
supposed to be connected by massless struts. We hang the rigid anchor at the upper
end of the stem or hold it at its center of gravity.
The anchor’s motion results from a superposition of the rotation of a selected
point of the anchor, e.g., the center of gravity, around the suspension point, and a
rotation of the anchor around the selected point. A rotational matrix describes this
motion.
We calculate the center of gravity of the anchor and, with Steiner’s theorem,1 its
moment of inertia when rotating around the upper end of the stem and when rotating
around its center of gravity.
Sound emitted from a moving source, T-T (Exercise 6.6)

In the last exercise of this chapter, we investigate the circular wavefronts emitted
from a moving acoustic source.
1 Also known as the parallel axis theorem or Huygens-Steiner theorem.

6.1 Introduction: Translations and Rotations 229
Animations
The movements treated in this chapter are well suited to be animated. The basic tech-
nique for this with FuncAnimation of the matplotlib.animation library
is explained in Sect. 6.2.5 in connection with the projectile trajectory.
6.2 Projectile Trajectory with Velocity Vectors (T-T)
We calculate and plot the trajectory of a projectile composed of two lin-

ear motions (T), a uniform horizontal one and a vertical one accelerated by
gravity. The parameters are launch height, angle, and speed. Vectors for the
horizontal, vertical, and total velocities are attached at three points to the
trajectory.
6.2.1 Projectile Trajectory and Velocity Vectors
Trajectory and attached arrows

In Fig. 6.1a, a projectile trajectory is shown with arrows representing velocity vectors
attached for three different time points. The parameters are launch height and angle,
and speed. In Fig. 6.1b, two trajectories for different launch angles are displayed.
The trajectories are, in all cases, downwardly open parabolas. Figure 6.1b shows that
maximum height and maximum distance depend on the launch angle.
4 4
aL=65° ; vL=5m/s;
y y aL=65° ; vL=5m/s;
velocity vectors
aL=25°; vL=5m/s
2 2
0 0
0 2 4 x 6 0 2 4 x 6
-2 -2
Fig. 6.1 a (left) Projectile trajectory, coordinates at different equidistant times; velocity vectors
at three different times. b (right) Two trajectories for two different launch angles
Our task is to plot a projectile trajectory with given values for launch height yL ,
angle α L , and velocity vL . In Fig. 6.1a, we have chosen yL = 2 m, α L = 65°, and vL
= 5 m/s. The time distance dt between the calculation points is 0.05 s.
Coordinates of the projectile trajectory

The coordinates of the parabola are calculated from
vx = v L x with v L x = v L · cos(α L );
v y = v L y − g · t with v L y = v L · sin(α L ) (6.2)
as
x(t) = vLx · t (6.3)
1
y(t) = yL + v L y · t − g · t 2 (6.4)
2
In the x-direction, there is a uniform motion with the horizontal initial velocity
vLx . The movement in the y-direction is the sum of a uniform motion with velocity
vLy , determined by the initial velocity and a downward motion accelerated by gravity.
Tasks For a given speed and launch height, change the angle such that (a) the
height and (b) the width reached will be maximum!
Calculate the impact velocity for the two trajectories in Fig. 6.1b!
Determine the maximum height and width analytically and compare them with
the value of the simulation!
Attach the tangential vector of the velocity and the decomposition of this vector
into x and y components at the point of the trajectory corresponding to a specific
time t = tt! The arrow representing the velocity vector (vXtt , vYtt ) is drawn in the
xy-plane
from (xtt , ytt ) to (xtt , ytt ) + (VXtt , VYtt ) · scal (6.5)
The lengths of the arrows in the figure are to be adapted to the diagram with a
scaling factor scal.

y L , aL , v L launch height, angle, speed
vLx , vLy horizontal and vertical components of vL
t equidistant (dt) series of instants of time
x(t), y(t) trajectory as a function of t
tt list of three specific times
x tt , ytt position at tt
6.2 Projectile Trajectory with Velocity Vectors (T-T) 231
A B C D E F B C D E F
1 Time interval dt 0.05 s 11 ="aL="&aL&"° ; vL="&vL&"m/s;"
2 Prespecifications 12 aL=65° ; vL=5m/s;
3 Launch height yL 2 m
^2
Launch angle aL 65 ° g/2*t
4 dt t v Ly*t-
5 Launch speed vL 5.00 m/s 13 =B15+ =vLx* =yL+
6 Gravitational acceleration g 9.81 m/s² 14 t x y
7 Deduced therefrom 15 0 0.00 2.00
8 vLx 2.11 m/s 16 0.05 0.11 2.21
9 vLy 4.53 m/s 44 1.45 3.06 -1.74
Fig. 6.2 a (left, S) The parameters for the task are defined and used to calculate the horizontal, vLx ,
and vertical, vLy , initial velocities. The launch angle α L is set with a slider. b (right, S) Continuation
of a. Coordinates of the parabola (t, x, y) in equidistant time steps dt in columns B, C and D; the
label for the figure is generated in B12 with the formula reported in B11
vXtt , vYtt velocity vectors at tt (3 instances)

scal scaling factor [s] for the velocity vectors in the xy-plane
x Att , yAtt coordinates of the tips of the arrows representing the vectors.
6.2.3 Spreadsheet
Trajectory
A possible spreadsheet calculation is shown in Fig. 6.2 (S).
The freely selectable parameters initial height, angle, and speed of the launch are
in C3:C6. From this, the initial horizontal and vertical velocities, vLx, and vLy , are
calculated in C8:C9.
Questions
Which formulas are in C8 and C9 of Fig. 6.1a (S)?2

How is the legend “aL = ...” in Fig. 6.2b (S) generated?3
Velocity-vector coordinates
Task Calculate the x- and y-components of the velocity for a given time tt! In
Fig. 6.3 (S), this is done for tt = 0.818 s, a time set with a slider.
In Fig. 6.3a, the coordinates of the arrows representing the velocity vector at t =
tt and its horizontal and vertical components are calculated according to Eq. 6.2.
This is repeated in Fig. 6.3b for two other time points. The resulting arrows are
displayed in Fig. 6.1a. As of excel 2007, line segments in charts can be provided
with arrowheads (excel 1019: format data series/fill & line/end arrow
type), and consequently our vector arrows.
2 vLx = C8 = [= vL*cos(aL/180*pi())]; vLy = C9 = [= vL*sin(aL/180*pi())].

3 See cell B11! Concatenation of text and variables.
H I J K L H I J K
14 velocity vectors 27 0.1 0.21 2.40
15 scaling factor scal 0.2 28 0.1 0.63 3.11
16 tt xAtt yAtt 29
17 818 30 0.21 2.40
18 0.818 1.73 =vLx*tt 2.42 =yL+vLy*tt-g/2*tt^2 31 0.63 2.40
19 0.818 2.15 =I18+scal*vLx 1.73 =K18+scal*(vLy-g*tt) 32
20 33 0.21 2.40
21 1.73 =I18 2.42 =K18 34 0.21 3.11
22 2.15 =I19 2.42 =K18 35
23 36 1.1 2.32 1.05
24 1.73 =I18 2.42 =K18 37 1.1 2.75 -0.20
25 1.73 =I18 1.73 =K19 38
26 39 2.32 1.05
Fig. 6.3 a (left, S) Coordinates (in I18:K19) of an arrow representing a velocity vector and its
vertical and horizontal components (I21:K25) at time tt, defined in H18; tt, x Att , and yAtt are the
names for the areas H18:H42, I18:I42, and K18: K42, containing the coordinates of the three vec-
tors in Fig. 6.1a. b (right, S) Continuation of a. Another velocity vector at another time, this time set
directly in H27 without a slider; the formulas are structurally the same as those reported in J18:J25
and L18:L25, but with references to different cells
Questions

The time at which velocity vectors are calculated and attached to the trajec-
tory is set with a slider. What is the linked cell of this slider, and what
are probable min and max? What are the formulas in H18 and H19?4
Change the time tt with the slider so that the height y = 0 is reached for
the discharge height and speed in Fig.6.1a and a launch angle 65°. At what
time does this occur, and at what speed does the projectile hit the ground?5
What is the purpose of the quantity scal in the formulas in row 15? How
big is it, and what physical unit does it have?6
An Excel trick
When you want to specify vectors for several instants of time, you can copy the
range H18:K25 in Fig. 6.3a if the formulas are written with relative and absolute
cell references so that they remain valid when copied. In Fig. 6.3 b (S), the formulas
have been copied into the area H27:K34. Regarding cells H27:H28, corresponding
4 In Fig. 6.3a (S), H17 is the cell linked to the slider (linked cell). min = 0, max = 1500, as
can be estimated from the position of the rider and the number in H17. H18 = H17/ 1000; H19 =
H18.
5 The projectile reaches the ground at t = 1.25 s and hits with v = 8.01 m/s, calculated with.
2
i
v = v0x + v0y − g · ti
2
6 Scal = 0.2 s. This parameter determines the length of the velocity vectors in their representation
in the plane (x [m], y [m]); see also Exercise 5.5. It occurs in an equation of the kind x [m] = x 0
[m] + scal * v [m/s]; scal has the unit [s].
to H18:H19, the desired time is entered directly into H27 and H36. As the range
H18:K25 has been copied twice, vectors are attached at a total of three points.
6.2.4 Python
Projectile trajectory
In the left cell of Table 6.1, the parameters of the exercise are specified. A label lbl1
for the legend in a figure is generated. In the right cell, initial horizontal and vertical
velocities are determined and the projectile trajectory (x(t), y(t)) is calculated.
Drawing arrows that represent vectors

Table 6.2 draws a figure that is similar to Fig. 6.1a. The time instants for the three
velocity vectors are specified in list t 2 . The coordinates of the arrows representing
these vectors are calculated in a for-loop.
In Python, in order to draw an arrow, we have to make use of the function
plt.arrow, which requires, among others, the initial position (x 0 , y0 ) and the
length of the vectors in the x- and y-directions as input. As we prefer to enter begin-
and end-points, we have defined a new function ArrowP, reported in Table 6.3, with
two positional arguments P0 and P1, and some keyword arguments with default
values.
ArrowP has two positional arguments, tail point P0 and head point P1 , and three
keyword arguments, ls = line style, lw = line width, and hw = head width. In the
current situation, the default head width is too large and we have specified hw = 0.1
in the function calls (lines 30–32 in Table 6.2).
Table 6.1 Projectile trajectory, with the same data structure as in the spreadsheet of Fig. 6.2
1 dt=0.05 #Time interval 8 #Deduced
2 #Prespecified 9 aL*=np.pi/180 #Radian
3 yL=2.0 #Launch height 10 vLx=vL*np.cos(aL)
4 aL=65 #Launch angle 11 vLy=vL*np.sin(aL)
5 vL=5.0 #Launch speed #Projectile trajectory:
6 g=9.81 #Gravit. accel. 12 t=np.arange(0,1.45+dt,dt)
7 lbl1="aL="+str(aL)+"; 13 x=vLx*t
vL="+str(vL) 14 y=yL+vLy*t-g/2*t**2
Table 6.2 Python program for drawing arrows representing velocity vectors at the trajectory;
the loop in the right cell corresponds to Fig. 6.3
15 FigStd('x',0,6,1,'y',-2,4,1)
16 plt.plot(x,y,'kD-',ms=2,label=lbl1)
17 plt.legend()
18
19 #Velocity vectors at three time instants t2
20 t2=[0.1,0.8,1.1]
21 scal=0.2
22 for tt in t2:
23 vXtt=vLx #Constant horizontal velocity
24 vYtt=vLy-g*tt #Uniform vertical motion
25 xtt=vXtt*tt
26 ytt=yL+vLy*tt-g/2*tt**2
27 P1=[xtt,ytt] #Foot position of arrow
28 Ax=xtt+vXtt*scal #Tip position of arrow
29 Ay=ytt+vYtt*scal
30 ArrowP(P1,[Ax,Ay], hw=0.1)
31 ArrowP(P1,[Ax,ytt],hw=0.1)
32 ArrowP(P1,[xtt,Ay],hw=0.1)
Table 6.3 User-defined function for drawing arrows

1 def ArrowP(P0, P1, c="k", ls='-' ,lw=1, hw=0.4):
2 #c has to be given as c="k", not c='k' (2020)
3 (x0,y0)=P0
4 (x1,y1)=P1
7 head_width=hw, fill=False,
8 linestyle=ls, color=c, linewidth=lw)
6.2.5 Animation of Figures with FuncAnimation7
We are going to set up an animated version of Fig. 6.1a by extending the program
presented in Table 6.1 that provides all data x(t), y(t), vx (t), and vy (t) that are
accessed in the following program as global arrays.
Creating a figure and a subplot object

In Table 6.4, the sublibrary animation is imported from matplotlib. A figure object
fig is set up and its default font size is set to 7 points (lines 4 and 5). In general,
an array of r x c subplots can be introduced into the frame of a figure. The instruc-
tion is add_subplot(rcn) with r and c specifying the number of rows and
columns. The index n indicates the individual subplot, starting at 1 in the upper left
7 Matplotlib.pyplot.subplots—Matplotlib 3.4.1 documentation

Table 6.4 Setting up a figure and a subplot object

1 %matplotlib notebook
2 import matplotlib.animation as animation
3 cm = 1/2.54 #Centimeter in inches
4 fig=plt.figure(figsize=(9*cm,9*cm))
5 plt.rcParams.update({'font.size': 7})
6 ax=fig.add_subplot(111) #1 row, 1 column, 1=top left
7 ax.set_xlim(0,6)
8 ax.set_ylim(-2,4)
9 ax.set_xlabel('x [m]',size=9)
10 ax.set_ylabel('y [m]',size=9)
Fig. 6.4 Presenting the objects arr, dot, and ln a (left) with toy data, b (right) at the end of the
animation
corner and increases to the right. We need only one subplot, so the instruction is
add_subplot(111). Resulting figures are shown in Fig. 6.4 with different plot
objects.
%matplotlib is a magic function that renders the figure in a notebook instead
of displaying a dump of the figure object.8
Creating plot objects

We first create some toy data. In Table 6.5, three two-dimensional print objects arr,
dot, ln are created. In Table 6.5a, the lists provided for the x and y data are empty
but the styles (line or dot) are already specified. Furthermore, a legend object leg is
created with captions for the three plot objects. In Table 6.5b, the three plot objects
are provided with some toy data. Running this cell, yields Fig. 6.4a.
8 https://stackoverflow.com/questions/43027980/purpose-of-matplotlib-inline.
Table 6.5 Creating plot objects a (top) with empty lists b (bottom) with toy data, result shown in
Fig. 6.4a
11 #Creating plot objects
12 arr,=ax.plot([],[],'k-',lw=0.5)
13 dot,=ax.plot([],[],'ko',ms=3)
14 ln, =ax.plot([],[],'k--',lw=1)
15 leg = ax.legend(['arr','dot','ln'])
16
17 leg.set_title('trajectory',prop={'size':6})
18 arr,=ax.plot([1,2,1,1,1,2],[0,0,0,-0.5,0,-0.5],
'k-',lw=0.5)
19 dot,=ax.plot([3],[3],'ko',ms=3)
20 ln, =ax.plot([2,3,4,5],[2,1,0,-1],'k--',lw=1)
21 plt.savefig("PhEx 6-2 Trajectory initial",dpi=1200)
Making an animation
We get a “living figure” by applying the function FuncAnimation
that creates an animation by repeatedly calling a function.9 Its variables
are FuncAnimation(fig, func, frames = None, interval = 200,
init func = None, …). Fig is a figure object. func is a callable, or, more precise,
a function that refreshes the plot objects for every frame. The first argument in func
will be the next value in frames. Interval specifies the delay between successive
frames in ms.
In Table 6.4, we have specified the figure as fig and in Table 6.7a, the refresh
function as animDotLine. The delay time is set to 100 ms. The frames parameter is
given as an integer that gives the index of the position and velocity arrays up to a
certain time. The final result is shown in Fig. 6.4b (from Table 6.6).
The refresh function animDotLine is given in Table 6.7a. The variable in the
function header is taken as index for the global arrays for position and velocity.
The plot objects are refreshed with the instruction set_data. For the projec-
tile (dot) it is just the current position (one point) . The trajectory itself (ln) is
represented by the curve traversed so far; the data are obtained by appending the
current position to the lists x Lin and yLin . The data x Arr and yArr for the velocity
arrows are calculated in another function coArr, reported in Table 6.7b.
We take the plot objects in this exercise as prototypes for other exercises:
dot single point,
arr new picture,
ln continued curve.
9 https://matplotlib.org/stable/api/_as_gen/matplotlib.animation.FuncAnimation.html
Table 6.6 Creating an animation object recurring to the figure object fig and the function anim-
DotLine; plotting arrows for indices in a list (here only one element)
1 xLin, yLin =[],[]
2 xArr, yArr =[],[]
3 scal=0.2
4 dotAnim=animation.FuncAnimation(fig,animDotLine,
frames=int(len(t)*0.6),interval=100, repeat=False)
5
6 for k in [2,11,20]:
7 xArr,yArr=coArr(k)
8 for i in np.arange(0,5,2):
9 ArrowP([xArr[i],yArr[i]],[xArr[i+1],yArr[i+1]])
10 plt.show()
11 plt.savefig("PhEx 6-2 Trajectory Anim partly",dpi=1200)
Table 6.7 a (left) Refresh function in the animation b (right) Function called in animDotLine for
each frame
1 def animDotLine(i): 10 def coArr(i):#Coeff. arrow
2 xLin.append(x[i]) 11 xi,yi=x[i],y[i]
3 yLin.append(y[i]) 12 vxi,vyi=vx[i],vy[i]
4 dot.set_data(x[i],y[i]) 13 xArr=[xi,xi+vxi*scal]
5 ln.set_data(xLin,yLin) #2 elements
6 xArr,yArr=coArr(i) 14 yArr=[yi,yi+vyi*scal]
7 arr.set_data(xArr,yArr) 15 xArr+=(xi,xi+vxi*scal)
8 #x,y from Table 6.1 #4 elements
9 #xLin,yLin from Table 6.6 16 yArr+=(yi,yi)
17 xArr+=(xi,xi)
#6 elements
18 yArr+=(yi,yi+vyi*scal)
19 return xArr, yArr
Questions
The font size for the axis titles in Fig. 6.4b seems to be too small. What is the
reason and how can you change it?10
How to provide the velocity vectors with arrow heads?11
10 The default font size was set with plt.rcParams.update({font.size’:7}). With

such an instruction you can change the whole font, not only its size.
11 Use unit-line and perpendicular vectors, see Chap. 5.
Does the time development of the animation reflect the true development?12
In this exercise and most easily in all other exercises, all coordinates of the
plot objects are calculated in the main program and stored in arrays that are then
accessed for the animation. The animation is thus an addition to the main program.
The advanced programmer can calculate the coordinates on the run, individually
and temporarily for every frame.
6.3 Cycloid, Rolling Curve (R-T)
We consider the movement of a point fixed on a wheel rolling along a straight

horizontal line. Viewed from the laboratory system, it is composed of a rota-
tion about the wheel axis and a uniformly progressing translation of the axis
parallel to the line. The speed of the point depends on its current altitude
above the line.
6.3.1 Trace of a Writing Point Fixed at a Rolling Wheel
We are going to examine the movement of a point (“writing point”) on a wheel

that rolls along a straight line in a plane as, e.g., shown in Fig. 6.5. The movement
of the rolling wheel is composed of a rotation (R) about its axis and a translation
(T) of the axis parallel to the plane (“the road”). Rotation period T W of the wheel
and velocity vA of its axis are related as follows:
2πr W
vA = with rW = radius of the wheel (6.6)
TW
10
y [m]
trace of point
5
wheel at t=0
wheel at t=tt
0
-10 0 10 20 30 x [m] 40 50
Fig. 6.5 Trace of a writing point on a wheel rolling along a straight line, always with the same
time interval between two adjacent marks
12The time development of the animation reflects the true time development because the frames
variable i is proportional to the time.
6.3 Cycloid, Rolling Curve (R-T) 239
During a circulation period T W , the wheel unrolls once on the road so that the
axis covers a distance 2πr W , the length of the circumference.
For the coordinates (x P , yP ) of the writing point and its speed vT along its
trajectory, the following equations hold:
xP = xArA + xC = rp · cos(ωW t) + xA
(6.7)
yP = yArA = rP · sin(ωW t)

s x2 + y2
vt = = (6.8)
t t
The index “ArA” designates rotation around the center (the axis), and x A the
translation of the axis.
Tasks Represent the wheel and the writing point graphically at any time, to be
set with a slider in the spreadsheet! Set-up a corresponding Python program with
animation!
Show the trajectory y = y(x) of the writing point in the same diagram!
Determine the writing point’s speed along its trajectory by numerical differen-
tiation!
Speed of the writing point

The speed of a writing point on the wheel’s rim along its trajectory is shown in
Fig. 6.6a, together with its average speed.
3.2 2.5
2.0
2.4
vP
1.5
vAv/vA
1.6
1.0
0.8 vt (1, 1.27)

0.5
vAv
0.0 0.0
0 10 20 30 40 0.0 0.5 1.0 1.5 2.0
t [s] rP/rW
Fig. 6.6 a (left) Speed vP of a point on the rim of the wheel in [m/s] as a function of time (r P =
r W = 5 m). b (right) Mean trajectory velocity vAv of the point relative to the velocity vA of the
axis, as a function of the distance r P of the point P to the wheel axis relative to the radius r W of
the wheel
Questions
At which points of the wheel is the writing point’s speed minimum and at which
is it maximum?13
How do you calculate the speed vT (t) of the writing point along its trajec-
tory numerically from x P (t) and yP (t) at time t, when the distance between the
interpolation points is t?14
Over which t value is the numerically calculated path velocity to be
plotted?15
What is the orbital period of the wheel in Fig. 6.6a?16
What is the speed vA of the axis in Fig. 6.6a?17
When the writing point touches the ground, its speed disappears, here, at T =
10 s and 30 s. Its speed is maximum (v = 2vAxis = π m/s) when it is at the rotating
wheel’s highest point.
The average speed of the writing point is indicated in Fig. 6.6a by the horizontal
dashed line. To get the average speed numerically, it is essential to average over
whole cycle times, e.g., over two cycle times, as in Fig. 6.6a.
The average speed depends on the distance of the writing point from the axis
of the wheel. In Fig. 6.6b, the average speed of the point relative to the speed of
the axis is plotted as a function of the distance of the writing point from the axis.
When the writing point is on the axis, its average speed is equal to the speed of
the axis. When it is on the rim, the speed ratio is 1.273 = 4/π.
Task Determine the point’s mean path velocity as a fraction of the axis velocity
for different distances of the point to the axis as in Fig. 6.6b!
Several trajectories with a procedure
Task Vary the distance of the writing point r P from the axis at time t = 0
systematically, e.g., as in Fig. 6.7. The distance can be greater than the radius, and
also negative. In excel, use a rep-log procedure; in Python, define a function
with r P as the argument!
13 Minimum speed: point on the road (v = 0); maximum speed (v = 2vAxis ): at the highest point
of the wheel.

T = vx ; v y = ((x(t) − x(t − t))/t; (y(t) − y(t − t))/t) then |vT | = vx2 + v 2y .
14 v
15 The speed is plotted over the center of the interval for which the velocity is numerically calcu-
lated.
16 T = 20 s, period of the speed profile, speed of the axis = 2·π · 5/20 = 1,57 [m/s].
17 Speed of the axis v = 2π ·r A = 2π ·5 = π m

A Tw 20 2 s
16
rP = 9 to -3
12
y
8
0
0 10 20 30 x 40 50 60
-4
Fig. 6.7 Trajectories of writing points on a wheel with various distances r P to the axis
Brachistochrone
Johann Bernoulli discovered, in 1696, that cycloids solve the brachistochrone prob-
lem. A brachistochrone curve is a trajectory on which a body in the homogeneous
gravitational field of the earth glides fastest from a starting point to a lower endpoint.
To describe such trajectories with our data structure, we would have to set ωW
positive and yR = −r R , so that the wheel, hanging from a ceiling, unrolls to the right.
The resulting curve is essentially part of one of the curves in Fig. 6.7 reflected at the
x-axis. It indeed has the property of a brachistochrone, as we shall see in the chapter
“Variational Calculus” of the follow-up textbook “Physics with Excel and Python.
Applications”.
rW radius of the wheel

rP distance of the writing point to the wheel axis
T W , wW cycle time, and angular velocity of the wheel
vA =(2πr W )/T W velocity of the axis
t series of equidistant (dt) time instants
tC center of the time intervals of t
x ArA , yArA rotation of the point around the axis
xA horizontal position of the axis
x P , yP trace of the writing point in the xy-plane
vP velocity of the writing point along its trajectory
tt one specific time point.
A B C D E F
1 Prespecifications
2 Radius of the wheel rW 5.00 m
3 Distance of the point from the axis rP 5.00 m
4 Cycle time of the wheel TW 20.00 s
5 Time interval dt 1.00 s
7 Angular frequency wW -0.31 1/s =-2*PI()/TW
8 Velocity of the axis vA 1.57 m/s =2*PI()*rW/TW
Fig. 6.8 (S) Parameters for the rolling curve in Fig. 6.5
6.3.3 Excel
Parameters of the motion

In Fig. 6.8 (S), the movement parameters are specified in named cells, in particular,
the radius r W and the cycle time T R of the wheel. From the given parameters, we
derive angular frequency ωW , height yP, and velocity vA of the axis. The distance
of the writing point to the axis of the wheel r P need not be equal to the radius r R
of the wheel. It may be bigger or smaller. The wheel will move on the plane to the
right, rotating clockwise. The angular frequency ωR , with which the polar angle is
calculated, is therefore negative.
Questions
From which axis is the angle ϕ of the plane polar coordinates measured?18
What is the angle ϕ when the writing point is at the highest point of the
wheel?19
What is the angle ϕ when the writing point is at the height of the axis?20
What sign does ωW have in a right-handed coordinate system when the
wheel rotates clockwise?21 Compare with Fig. 6.8 (S)!
Trace of the writing point

In Fig. 6.9 (S) , the coordinates (x P yP ) of the writing point are calculated for 41
instants of time, with intermediate calculations for the rotation ( x ArA , yAA ) of the
point about the axis of rotation (C:D) and for the translational motion (x A ) of the
axis (column E).
18 The angle ϕ is measured from the positive x axis.

19 ϕ = π/2 = 90°, 90° + 360°, 90° + 360° + 360°, …
20 ϕ = 0 = 0°.
21 The angular frequency is negative when the point runs clockwise.
B C D E F G H I J
9 vAv 1.99 m/s =AVERAGE(vP)
10 wheel at t=0.5 trace of point 1.268 =vAv/vA
)/dt
3)^2
I()/2) ()/2)+
rW
+( yP-G1
*t+P wW*t+PI ^2
W -F13)
dt
11 =B15+ =rP*C
(w
OS rP*SIN( vA*t A+xA yArA T((xP (t+B13)/2
= = =xAr = =SQR =
12 t xArA yArA xA xP yP vP tC
13 0.5 0.78 9.94 0.79 1.57 9.94
14 1.5 2.27 9.46 2.36 4.63 9.46 3.10 1
53 40.5 0.78 9.94 63.62 64.40 9.94 3.14 40
Fig. 6.9 (S) Coordinates of a writing point when unrolling the wheel; rotation of the point around
the wheel axis (x ArA ; yArA ); translation of the axis of the wheel (x A ); addition of the two move-
ments yields (x (t); y (t)); speed along the path vt
The corresponding equations are
xar A (t) = r P cos(ωt) + x R ; yar A (t) = r P sin(ωt) + r W (6.9)
Here, it is assumed that, at t = 0, the writing point is vertically above the axis.
An example can be found in Fig. 6.5. You can extend the solution by allowing the
selected point to assume any position at t = 0 by choosing the phase shift within
the circular functions!
In column E, there is the displacement of the x-coordinate of the center of the
wheel over time (translation in the x-direction). The coordinates of the rotating
point in the laboratory system are in columns F and G, calculated from the super-
position (component-wise addition of the Cartesian coordinates) of the rotation
about the wheel axis (x arA , yarA ) and the translation of the axis (x M , 0):
x(t) = xA + xarA y(t) = yarA (6.10)
The resulting trajectory is called a cycloid (rolling curve).
Question
In which cell of Fig. 6.9 (S) is the mean path velocity calculated?22
Task Determine the mean path speed of the point relative to the axis’ speed for
different distances of the point to the axis! A typical evaluation can be seen in
Fig. 6.6a, b. It is best to use a VBA rep-log procedure that varies the distance and
logs the average speed!
22 The mean speed vAv is calculated in H9 of Fig. 6.9 (S).

Fig. 6.10 (S) Position of the K L M N O P

point (x(t 2 ); y(t 2 )) at time tt; 10 point at t=tt axis at tt wheel at t=tt
wheel at t = 0 (x rad ) and tt =
t
t 2 (x t2 ); tt is set with the +xAt +rW
tt +PI()/2) t+PI()/2)
slider. The formulas are the wW * W * t tt
COS( rP*SIN(w vA*tt A+xA
same as in Table 6.10 11 =rP* = = =xAr
12 tt xPtt yPtt xAtt xWtt
13 15 18.56 5.00 23.56 24.34
14 25.83
15 27.10
53 24.34
Wheel at time tt
In Fig. 6.5, the wheel is shown at t = 0 and at a second time tt. In Fig. 6.10 (S), the
coordinates of the wheel are determined for tt specified with a slider. The writing
point has the coordinates (x Ptt , yPtt ). The coordinates of the wheel are (x Wtt , yW ),
with yW being the y-coordinates calculated for t = 0. If you put the pointer into the
right part of the slider and keep it pressed down, the wheel rolls to the right.
6.3.4 Python
The Python program exhibits the same structure as the formula network in the
spreadsheet. In Table 6.8, the parameters of the task are set.
In Table 6.9, the wheel’s coordinates at the start and the trace of the writing
point are calculated. The velocity vP = ds/dt along the trace of the point is calcu-
lated in line 15. The length ds of the trace sections is obtained by slicing x P and
yP .
The coordinates of the wheel and the writing point at a specific time t = tt
are obtained with the function WheelAt in Table 6.10, which has only tt as an
argument and resorts otherwise to global variables.
The plot program for yielding a figure like Fig. 6.5, with the arrays calculated
in Table 6.9, is given in Table 6.11. The coordinates for the wheel and the writing
point at a specific time t = tt are obtained by calling the function WheelAt(...) in
Table 6.10.
Table 6.8 Setting the parameters for the rolling wheel, the same as in Fig. 6.8 (S)
1 #Prespecified:
2 rW=5.0 #Radius of the wheel
3 rP=5.0 #Distance of the point from the axis
4 TW=20 #Period of wheel rotation
5 dt=1.0 #Time increment
#Calculated therefrom:
6 wW=-2*np.pi/TW #Angular frequency
7 vA=2*np.pi*rW/TW #Velocity of the axis
Table 6.9 Calculation of the coordinates of the wheel at its start and of the trace of the writing
point
8 #Coordinates of point vs. time:
9 t=np.arange(0,40+dt,dt)
#Coordinates of the wheel at start:
10 xArA=rW*np.cos(wW*t+np.pi/2)
11 yArA=rW*np.sin(wW*t+np.pi/2)+rW
12 xA=vA*t #From velocity of axis
13 xP=xArA/rW*rP+xA #Writing point
14 yP=rP*np.sin(wW*t+np.pi/2)+rW
#Veloc. along the trajectory of the point:
15 vP=np.sqrt((xP[1:]-xP[:-1])**2+(yP[1:]-yP[:-1])**2)/dt
16 tC=(t[1:]+t[:-1])/2 #Valid for vP
Table 6.10 Function for specifying the coordinates of the wheel and the writing point at t = tt
1 def WheelAt(tt):
#Wheel and point at t = tt:
2 xAtt=vA*tt #Position of axis
3 xPtt=rP*np.cos(wW*tt+np.pi/2)+xAtt
4 yPtt=rP*np.sin(wW*tt+np.pi/2)+rW
5 xWtt=xArA+xAtt #x Coordinates of the rim
6 return xAtt,xPtt,yPtt,xWtt
Table 6.11 Plot program yielding a figure like Fig. 6.5

1 FigStd('x',-10,50,10,'y',0,10.0,2.5,xlength=12,ylength=4)
2 plt.plot(xArA,yArA,'k-') #Wheel at t = 0
3 plt.plot(xP[0],yP[0],'ko') #Point at t = 0
4 plt.plot(xP,yP,'k-') #Trace of point
5
6 xAtt,xPtt,yPtt,xWtt=WheelAt(tt=15)
7 plt.plot([xP[0],xAtt],[rW,rW],'ko', fillstyle='none')
8 #Pos. of axis
9 plt.plot(xWtt,yArA,'k--') #Wheel
10 plt.plot(xPtt,yPtt,'ko') #Selected point
A meaningful animation could comprise the functions dot for the writing point,
ln for the cycloid, arr for the wheel from Sect. 6.2.5.
6.4 Foucault’s Pendulum (T-R)
We calculate the trace of a pendulum swinging in the laboratory system (T)

on a rotating table (R).
6.4.1 A Lecture Experiment
In a lecture experiment about Foucault’s pendulum, a thread pendulum swings over

a rotating plate, writing a trace thereon. In the laboratory system, the pendulum
swings in a plane, with its suspension point being located in the axis of rotation
of the plate.
Figure 6.11a shows the trace of the pendulum for a period of oscillation of T p
= 1.2 s and a rotation time of the table of T r = 9 s. The partial circle “Stylo”
represents the trace of a pen, resting in the laboratory system, along the rotating
plate to indicate the sense of rotation.
2 2
Trace pend. on rot. disc Trace pend. on rot. disc
Trace stylo on rot. disc Trace stylo on rot. disc
y on disc
y on disc
1 1
0 0
-2 -1 0 1 2 -2 -1 0 1 2
-1 -1
-2 -2
x on disc x on disc
Fig. 6.11 a (left) Traces of a pendulum oscillating in the laboratory system (T P = 1.2 s) and of
a stylus at rest in the laboratory system on a rotating plate (T T = 9 s), the suspension point of the
pendulum being in the axis of rotation of the plate; the unit length is 1 cm, as explained in the main
text. b (right) Closed track of a pendulum whose suspension point is not in the axis of rotation (T P
= T T /9)
6.4 Foucault’s Pendulum (T-R) 247
Questions
A thread pendulum is swinging with a period of 12.7 s. How long is the

pendulum?23
In what period of time does the earth rotate by 1°?24
What are the amplitude of the pendulum and the horizontal displacement of
the suspension point against the rotation axis in Fig. 6.11b?25
Under what condition do closed tracks occur in the lecture experiment on
Foucault’s pendulum? What is the concrete condition in Fig. 6.11b?26
For simplification, we assume that the trace of the pendulum on the plate at rest
or, more generally, in the laboratory system is described by
x p = A p · cos(ω p t) (6.11)
Then, we let the plate rotate around its vertical axis. The trace of the pendulum
on the rotating table is composed of the oscillation in the x-direction (T, in the
laboratory system) and an angular displacement on the table according to its
rotation (R). The equations for the conversion of the coordinates (x L , yL ) in the
laboratory system into the coordinates (x T , yT ) of the rotating table are
x T = x P · cos(ωT t) and yT = xp · sin(ωT t) (6.12)
where ωT is the angular frequency of the rotating table. These equations are a
special case of the general form for a counter-clockwise rotation by φ applying a
rotational matrix:

xT cos φ − sin φ x
= · P (6.13)
yT sin φ cos φ yP
Where have all the units gone?
Tim In Fig. 6.11, we have not specified any physical units for the lengths.
Alac That’s no problem: Times in seconds, lengths in meters. That’s standard.
√
23 T = 12,7 s, ω = 2π/T = g/l, → length l of the pendulum = 40 m.
24 The earth rotates by 360° in one day, → in 4 min, by 1°; t = 1°/360° * 24 * 60 * 60 s = 240 s.
25 The deflection of the pendulum is from 0.5 cm to 1.5 cm. The suspension point of the pendulum
is shifted by a distance 1 cm against the center. The amplitude is 0.5 cm.

26 The ratio of the period of the pendulum and the circulation time of the plate must be a natural
number. In Fig. 6.11b, the ratio is 9 to 1. The pendulum makes nine oscillations during one turn of
the plate.
Mag Concerning calculation, everything is clear. But does that make sense
physically? How long is the pendulum?
Tim From the oscillation period, T P = 1.2 s, as stated in the caption, it follows
that l = 36 cm.
Mag How does the maximum swing fit in with that?
Alac I admit: A deflection of 1.8 m does not fit with the pendulum length. So,
let’s decide that the pendulum should be deflected by just 1.8 cm.
Mag So, the unit of length in Fig. 6.11 is 1 cm. Indeed, this is experimentally
difficult to record, but at least our calculation is consistent.
Tasks Create a spreadsheet calculation/a Python program for the experiment

described above and vary the pendulum’s oscillation duration and the table’s cycle
time!
Check if the pendulum track is as expected (a) when the rotation period of the
table is large compared to the period of oscillation and (b) when the oscillation
period and rotation period are identical!
Change the calculation for the case in which the table’s rotational axis is still in
the plane of the swinging pendulum but no longer passes through the suspension
point of the pendulum! An example can be seen in Fig. 6.11b.
Ap amplitude of the pendulum

T p , wP oscillation period of the pendulum and corresponding angular velocity
T r , wr rotation period of the table and corresponding angular velocity
x Sh shift of the suspension point with respect to the rotational axis
t series of equidistant (dt) time instants
xP position of the pendulum at t in the lab system
x T , yT trace of the pendulum on the table
x St , ySt trace of a stylus at rest in the lab system on the table, to check the
direction of rotation of the table.
6.4.3 Excel
Setting the parameters

The parameters for the movement in Fig. 6.11 are specified in Fig. 6.12 (S). The
quantity x Sh (in C6) determines the displacement of the suspension point against the
plate’s axis of rotation.
A B C D E
1 Prespecifications
2 Amplitude of pendulum Ap 1.50 0.50
3 Period of pendulum Tp 1.20 1.00
4 Period of rotation Tr 9.00 9.00
5 Time interval dt 0.0173 0.09
6 Suspension point vs rot. axis xSh 0.00 1.00
8 Angular frequency pendulum wP 5.24 =2*PI()/Tp
9 Angular frequency rotating disc wR -0.70 =-2*PI()/Td
Fig. 6.12 (S) Specifications for the movement presented in Fig. 6.11; values in column C for
partial picture a, those in column E for partial picture b
B C D E F G H
xSh *t)
P*t)+ R wR*t
) (wR*
t)
wR*t
)
dt OS(w P*COS(w SIN( Ap*COS Ap*SIN(
12 =B14+ =A p*C
=x =xP* = =
13 t xP xT yT xSt ySt
14 0.000 1.50 1.50 0.00 1.50 0.00
15 0.017 1.49 1.49 -0.02 1.50 -0.02
174 2.768 -0.52 0.18 0.49 -0.53 -1.40
Fig. 6.13 (S) In column C, the pendulum motion x P (t) is calculated in the laboratory system. In
columns D and E, this movement is transformed (to x T , yT ) into the coordinate system of the rotat-
ing table. In columns F and G, the coordinates on the rotating plate of a point (“Stylo”) fixed in the
laboratory system are calculated
Question
How long is the pendulum in Fig. 6.12 (S) when the oscillation period T P is
given in seconds?27
Trace of pendulum
The movement itself is calculated in Fig. 6.13 (S) for the pendulum swinging in the
x-direction.
Task Complete the diagram with two points representing the pendulum’s posi-
tions and the pen at a selectable time! In Fig. 6.11a, this was done for t = 0.2249
(arrow close to x = 1). A suggestion: Use a slider to select a row from 14 to 174
and copy the coordinates from that line to an area added to the diagram as a point.
You can use the reference type indirect for this purpose.
√
27 T = 1.2 s, ω = 2π/T = g/l, length l of the pendulum = 0.36 m.
Table 6.12 Foucault’s pendulum, specification of the parameters, the same as in Fig. 6.12 (S)
1 #Prespecified:
2 Ap=1.5 #Amplitude of the pendulum
3 Tp=1.2 #Period of the pendulum
4 Tr=9.0 #Period of the rot. disc
5 dt=0.0173 #Time interval
6 xSh=0.0 #Shift of suspension point
7 #Calculated therefrom:
8 wP=2*np.pi/Tp #Circ. freq. of the pendulum
9 wR=-2*np.pi/Tr #Circ. freq. of the disc
Table 6.13 Setting up the arrays describing the motion of the pendulum in the laboratory system
and its trace on the rotating table
1 t=np.arange(0,2.768+dt,dt)
2 #Traces:
3 xP=Ap*np.cos(wP*t)+xSh #Pendulum in lab
4 xT=xP*np.cos(wR*t) #Trace of pendulum on table
5 yT=xP*np.sin(wR*t)
6 xSt=Ap*np.cos(wR*t) #Trace of stylus on table
7 ySt=Ap*np.sin(wR*t)
6.4.4 Python
In Table 6.12, the parameters of the swinging pendulum are specified, with the
same values as in Fig. 6.12 (S).
The arrays describing the motion of the pendulum in the laboratory system and
its trace on the rotating table are set up in Table 6.13, together with the trace of a
stylus fixed in the laboratory system
To get figures such as those in Fig. 6.11, we apply the program in Table 6.14.
The parameters in Table 6.12 are for Fig. 6.11a. To get Fig. 6.11b, the parameters
in column E of Fig. 6.12 (S) have to be inserted into Table 6.12.
Remember: Within brackets or parentheses, line breaks are allowed after punc-
tuation marks, as is applied in Table 6.14, line 5. Explicit line breaks are possible
after a backslash (\) as in line 2.
Questions

How many positional arguments are in the header of ArrowP?28
What is the first argument in the header of ArrowP in line 9?29
28 ArrowP(P0,P1,…) has two positional arguments, foot point P0 and head point P1 .
29 P0 = [xT[−2],yT[−2]].
Table 6.14 Plotting the arrays obtained in Table 6.13 to get a picture like that in Fig. 6.11a,
ArrowP from Table 6.3
1 FigStd('x',-2.0,2.0,0.5,'y',-2.0,2.0,0.5)
2 plt.plot(xT,yT,'k-x',ms=3,label='pend\
3 ulum on disc')
4 plt.plot(xSt,ySt,'k--',label='stylo on disc')
5 ArrowP([xSt[-2],ySt[-2]],[xSt[-1],
6 ySt[-1]],hw=0.1)
7 i=10
8 ArrowP([xT[i],yT[i]],[xT[i+1],yT[i+1]],hw=0.1)
9 ArrowP([xT[-2],yT[-2]],[xT[-1],yT[-1]],hw=0.1)
10 plt.legend()
To which arrows in Fig. 6.11a do the three calls of the function ArrowP
correspond?30
Task Set up an animation in the laboratory system, with the pendulum swinging
horizontally and the table rotating! The frames should be the equidistant time
instants t to mimic the oscillation.
6.5 Anchor, Deflected Out of Its Rest Position (R-R)
The rotation of an anchor about its suspension point is described as a rotation

(R) of the center of gravity about the suspension point and a rotation (R) of
the anchor about the center of gravity (if the anchor is not suspended there).
The moment of inertia is calculated using Steiner’s theorem.
6.5.1 Deflected Anchor
Coordinates of the deflected anchor

We consider the rotation of an anchor about a point located in the origin of the
coordinate system. In Fig. 6.14a, the anchor is held at the end S of the stem, in
Fig. 6.14b, at its center of gravity C g . The construction of the anchor is simplified
with four mass points attached to the ends of a hanging T (see the inset in Fig. 6.14b).
30First: end of trace “stylo”; second: initial phase (i = 10) of trace “pend”; third: end of trace
“pend”.
1 1
α=40°
Center of gravity Cg
0 0
-1 0 1 2 3 -1 0 1 2 3
α α
α S
-1 -1
α L -M-R
-2 -2
α=40°
Center of gravity Cg
-3 -3
Fig. 6.14 An anchor is deflected by an angle α = 40°, held a (left) at the upper end of the stem,
b (right) at its center of gravity. The four characteristic point masses are designated by the letters
S, L-M-R. The center of gravity is marked by an open circle
In the following programs, the anchor’s characteristic points are listed as [L, M,
S, M, R] (Left, middle, top, middle, right; see Fig. 6.14b). We have doubled M
because the anchor can then be drawn in one uninterrupted line. The distances L-M
= M-R and S-M are specified, respectively, as r A and lA , so that the coordinates
of the anchor at rest are defined as
xA = [−rA , 0, 0, 0, rA ]
yA = [−lA , −IA , 0, −lA , lA ] (6.14)
Rotation about the suspension point S

If the anchor is rotated about its top point S (Fig. 6.14a) , the displacement of any point
R of the anchor, with Cartesian coordinates (−r A , −lA ) at rest, may be considered
the sum of two rotations:
Rotation of point R around M : (−rA , 0) → (−rA · cos(α), rA · sin(α))

Rotation of point M around S : (0, −lA ) → (lA · sin(α], −lA · cos(α)
(−rA · cos(α) + lA · sin(α),
With the sum : (−rA , −lA ) →
rA · sin(α) − lA · cos(α))
The total rotation can be presented as a matrix multiplication:
= (−r A , −l A ) · [Rot Mat] (6.15)
with the rotational matrix defined as

cos(α) sin(α)
Rot Mat = (6.16)
− sin(α) cos(α)
6.5 Anchor, Deflected Out of Its Rest Position (R-R) 253
The rotational matrix can be applied to any point in the plane:
(x R , y R ) = (x, y) · [Rot Mat] (6.17)
Here, we have chosen to present the coordinates as row vectors, because this is
more convenient in spreadsheet calculations where coordinate vectors (a pair of two
numbers) can more clearly be stored in successive rows.
Task Calculate the coordinates of the anchor for the freely selectable parame-
ters: length lA of the stem (A-M), half the length r A of the crossbar (M-R or M-L),
and angle α!
Center of gravity as weighted sum

The center of gravity is defined as the sum over the coordinates of the characteristic
points weighted with their mass:

i m i · xi
xG =
mi
i
i m i · yi
yG = (6.18)
i mi
This is achieved in excel with
yG = sumproduct(y, m)/sum(m)
and in Python with
yG = np.dot(y, m)/np.sum(m)
The function np.dot returns the dot product of two arrays. For 1-D arrays, it is
the inner product of the vectors, i.e., the sum of the products of the components, the
equivalent of sumproduct. For 2-D arrays, it is equivalent to matrix multiplication.
The anchor is held at its center of gravity

In Fig. 6.14b, the anchor is held at its center of gravity by a rod (considered to
be massless) that can rotate around S and is currently rotated by an angle α. The
coordinates of the deflected anchor are the coordinates of the anchor at rest shifted
by the coordinates of the center of gravity after rotation by α according to Eq. 6.17
applied to (x G , yG ). Now, only one rotation is effective.
Moment of inertia
The moment of inertia I is defined as
I = ri2 · m i (6.19)
i
where r i is the distance to the axis of rotation, here, ri2 = xi2 + yi2 .
It is calculated
in Python as: np.sum((x**2+y**2)*m)

in excel as: sumproduct(x; x; m) + sumproduct(y; y; m)
We calculate the moment of inertia by applying the formulas:
– (a) to a rotation of the anchor about the point S (corresponding to Fig. 6.14a),
– (b) to a rotation of the anchor about its center of gravity,
– (c) to a rotation of the total mass, concentrated in the center of gravity, about S
(corresponding to Fig. 6.14b).
The sum of (b) and (c) should be equal to (a), according to Steiner’s law.
Animation
We set up an animation with the anchor swinging correctly with sinusoidal time
dependence about its suspension point.
rA distance L-M = M-R

lA distance M-S
x A , yA coordinates of the anchor at rest (5 elements, combinations of r A and
lA )
xyA = [x A , yA ] (2D range)
α angle of rotation
x R , yR coordinates of the deflected anchor, rotated around the suspension point
x G , yG coordinates of the center of gravity for the anchor at rest
x Gr , yGr coordinates of the center of gravity rotated around S, at the origin of
the coordinate system
RotMat rotational matrix
Equivalence:
excel: [x R ,yR ] = mmult(xyA ,RotMat), xyA = [x A , yA ]

Python: [x R ,yR ] = RotMat @ [x A ,yA ] or np.matmul (RotMat,[xA,yA])
6.5.3 Excel
Specs
The parameters of the exercise are specified in columns A:E of Fig. 6.15 (S), with
the rotation angle being adjusted with a slider. The coordinates of the anchor at rest
are specified in columns G:H by inserting—±lA or ±r A , or 0 when appropriate. The
masses attached to these points are given in column I. The coordinates of point M
appear twice in x and y, but the second time with zero mass so that the center of
gravity and moment of inertia may be calculated in formulas in Fig. 6.16 (S), taking
the whole arrays x A , yA , and m as input.
The anchor can be made to oscillate by running the time t with a slider and
calculating the deflection angle α in C4 with α = A · cos(ω · t) with suitable A and
ω.
Rotational matrix
The coordinates (x R , yR ) of the deflected anchor are calculated by applying the
rotational matrix RotMat in Fig. 6.16 (S) to the coordinates (x A , yA ) of the anchor
at rest.
Center of gravity
The center of gravity C g and the moments of inertia are calculated in Fig. 6.17 (S). B5
= [“Deflection angle =”&C3&“°”. The moment of inertia for a rotation about the
suspension point S is calculated twice:
A B C D E F G H I
1 Length of rod lA 2.50 m xA yA m
2 Half-length of cross bar rA 0.50 m L -0.50 -2.50 1.0
3 220 40.00 ° =B3-180 M 0.00 -2.50 1.0
4 a 0.70 rad =C3/180*PI() S 0.00 0.00 1.0
5 Deflection angle α=40° M 0.00 -2.50 0.0
6 R 0.50 -2.50 1.0
Fig. 6.15 (S) Anchor parameters in B1:C4; the deflection angle α is determined with the slider in
A3 (linked cell = B3) . he coordinates x, y of the line connecting the characteristic points are in
columns G and H with the associated masses in column I, set to 1 except for the second reference
to point M. B5 = “Deflection angle =”&C3&“°”
K L M N O P
1 RotMat xR yR
2 0.77 0.64 1.22 -2.24 =MMULT(xyA;RotMat)
3 -0.64 0.77 1.61 -1.92
4 0.00 0.00
5 =COS(a) =SIN(a) 1.61 -1.92
6 =-SIN(a) =COS(a) 1.99 -1.59
Fig. 6.16 (S) Continuation of Fig. 6.15 (S). The rotational matrix RotMat is applied to the coor-
dinates x A and yA , bound together in Fig. 6.15 (S) in one matrix xyA = [G2:H6] in Fig. 6.15
(S)
AA AB AC AD AE
1 Center of gravity Cg Moment of inertia for rotation about Cg
2 xG yG 5.19 =SUMPRODUCT(xA-xG;xA-xG;m)
3 0.00 -1.88 =SUMPRODUCT(yA;m)/SUM(m) +SUMPRODUCT(yA-yG;yA-yG;m)
4 xGr yGr for rotation of Cg
5 1.21 -1.44 =MMULT(xGyG;RotMat) 14.06 =(xG^2+yG^2)*SUM(m)
6
7 Moment of inertia for rotation about S
8 19.25 =SUMPRODUCT(x;x;m) 19.25 =AD2+AD5
9 +SUMPRODUCT(y;y;m)
Fig. 6.17 (S) Continuation of Figs. 6.15 (S) and 6.16 (S). AA:AC: center of gravity, moment of
inertia for rotation about the point S = (0, 0); AD:AE: moment of inertia with Steiner’s theorem
as the sum of two rotations calculated with the spreadsheet functions sumproduct and sum
– in AB8, with the coordinates (x, y) of the anchor at rest,

– in AD2:AD8, as the sum of the moments of inertia for rotation about the center
of gravity C g and a rotation of the total mass in the center of gravity about (0,0).
Both calculations should yield, according to Steiner’s theorem, the same result, and
they do.
6.5.4 Python
Specs and rotational matrix

A Python solution corresponding to Figs. 6.15 (S) and 6.16 (S) is given in Table
6.15.
Table 6.15 Swinging anchor, specs as in Fig. 6.15 (S)

1 #Prespecified:
2 lA=2.5 #Length of stem
3 rA=0.5 #Half-length of crossbar
4 a=40 #Angle of deflection in °
5 a*=np.pi/180 # in rad
6
7 #Coordinates of the anchor at rest:
8 # L-M-S-M-R
9 xA=np.array([-rA,0,0,0,rA])
10 yA=np.array([-lA,-lA,0,-lA,-lA])
11
12 #Rotational matrix:
13 RotMat=np.array([[np.cos(a),-np.sin(a)],
14 [np.sin(a), np.cos(a)]])
15 #Coordinates of the deflected anchor:
16 [xR,yR]=RotMat@[xA,yA]
Animation
In Table 6.16, a function ancRot is defined that is applied in the animation. It cal-
culates the coordinates of the anchor and its center of gravity for an angle α. The
function anim called within FuncAnimation converts the frame number i into the
angle α through a sine function that mimics an oscillation.
Center of gravity
In Table 6.17, the coordinates of the center of gravity C g are calculated, at rest and
rotated. The dot product is used to calculate the nominator in the fraction for the
center of gravity (Eq. 6.18).
A program for drawing the rotated anchor as shown in Fig. 6.14a is presented in
Table 6.18.
In Table 6.19, the moments of inertia are calculated when the anchor is rotated:
– (a) about S, the upper end of the stem,

– (b) about Cg, the center of gravity; and
– (c) when the center of gravity is rotated about (0, 0).
Table 6.16 The anchor swings with sinusoidal time dependence about S
1 def ancRot(a): #Angle a=alpha
2 #Rotational matrix:
3 RotMat=np.array([[np.cos(a),-np.sin(a)],
4 [np.sin(a), np.cos(a)]])
5 #Coordinates of the deflected anchor:
6 [xR,yR]=RotMat@[x,y]
7 [xGr,yGr]=RotMat@[xG,yG]
8 return xR, yR, yGr, yGr
9 anc,=plt.plot(xR,yR,'k-o')
10
11 def anim(i):
12 a=np.pi/4*np.sin(i*0.1)
13 xR, yR, xGr, yGr = ancRot(a)
14 anc.set_data(xR,yR)
15
16 AnchorAnim=animation.FuncAnimation(figA,anim,
17 frames=range(180),interval=100,repeat=False)
Table 6.17 Coordinates of the center of gravity, at rest and rotated

1 #Center of gravity Cg
2 m=np.array([1,1,1,0,1]) #Masses of L-M-S-M-R
3 xG=np.dot(xA,m)/np.sum(m) #Coordinates of c of g
4 yG=np.dot(yA,m)/np.sum(m)
5 [xGr,yGr]=RotMat@[xG,yG] #Coordinates of rotated anchor
Table 6.18 Program for drawing a figure like Fig. 6.14a

1 FigStd('x',-1.0,3.0,1,'y',-3.0,1.0,1.0)
2 plt.plot(xA,yA,'k-o',label="α=0") #At rest
3 plt.plot(xR,yR,'k:o',fillstyle='none',
4 lw=2,label="α="+str(round(a,2))) #Rotated by α
5 plt.plot(xG,yG,'ko',ms=4,
6 label='c of grav.') #Center of gravity
7 plt.plot(xGr,yGr,'ko',ms=4)
8 plt.legend()
Table 6.19 Moment of inertia

1 #Momentum of inertia, around S
2 IS=np.sum((x**2+y**2)*m) IS 19.25
3
4 #Rotated coord. of center of gravity: xG 0.0
5 [xGr,yGr]=RotMat@[xG,yG] yG -1.875
6
7 #Moment. of inertia, around c of grav
8 IarCg=np.dot((x-xG)**2,m) \ IarCg 5.19
9 +np.dot((y-yG)**2,m)
10
11 #Rotation of center of gravity
12 ICg=(xG**2+yG**2)*np.sum(m) ICg 14.06
13
14 Itot=IarCg+ICg Itot 19.25
We confirm again that (b) is smaller than (a) and that (b) + (c) equals (a).
Table 6.20 displays a program for plotting the anchor held at its center of gravity,
as in Fig. 6.14b.
6.6 Wavefronts, Sound Barriers, and Mach Cone (T-T)
We draw the wave crests of sound waves emitted by a source moving at a

certain speed and direction in the xy-plane and demonstrate the breaking of
the sound barrier for supersonic speed. Polar coordinates are used for the
calculation, and Cartesian coordinates for the scatter diagrams.
6.6 Wavefronts, Sound Barriers, and Mach Cone (T-T) 259
Table 6.20 Program for plotting the anchor gripped at its center of gravity (Fig. 6.14b)
1 x_Cg=xA+(xGr-xG)
2 y_Cg=yA+(yGr-yG)
3
4 FigStd('x',-1.0,3.0,1,'y',-3.0,1.0,1.0)
5 plt.plot(xA,yA,'k-o') #Anchor at rest
6 plt.plot(xG,yG,'kx',ms=8,label='c of g') #Cent. of gravity
7 plt.plot(x_Cg,y_Cg,'k:o',fillstyle="none")#Rotated anchor
8 plt.plot([0,xGr],[0,yGr],'k--',label="α="+str(a))
#Rotated rod
9 plt.plot(xGr,yGr,'kx',ms=8)
10 plt.plot()
11 plt.legend()
6.6.1 Emitting Sound Waves
In Fig. 6.18, crests of sound waves in a plane emitted from a moving source are
shown, in b, for supersonic speed. One wave crest is emitted in every period of
the sound signal.
Sound barrier and Mach cone
Mag Are the motions linear or rotational?
3000 3000
vS=80; alpha=30° vS=500; alpha=10°
Centers Centers
y [m]
y [m]
2000 2000
1000 1000
0 0
-3000 -1000 1000 3000 -6000 -4000 -2000 0
-1000 x [m] -1000x [m]
-2000 -2000
-3000 -3000
Fig. 6.18 a (left) Wavefronts of acoustic waves emitted by a moving sound source that has
reached the position (0, 0) at t = 0; airspeed vS (here, below the velocity of sound) and angle α of
flight direction against the horizontal axis. b (right) As in a, but with α = 10° and at supersonic
speed
Alac The motion of the airplane is along a straight line.
Mag Ok, so it’s translational. But what about the sound?
Tim Sound is a longitudinal wave, so it’s a linear motion.
Alac Ok. But how do the circles arise?
Mag As wave propagation is isotropic, the wave crests are circles.
Tim And the circles are described with polar coordinates.
Alac I understand, polar coordinates but no circular motion.
Tim Ok, that’s clear. The airplane, modeled as a point in the plane, moves
linearly in two dimensions. Sound is a wave and propagates isotropically in air.
Mag Another point: What happens if the aircraft flies at exactly the speed of
sound?
Alac Then a sound barrier builds up, and there is a loud bang.
Mag Simulate this situation! The best way to do so in a spreadsheet is to install

a slider and increase the speed vS , of the sound source slowly, starting from zero
up to the speed of sound!
Tim What does “breaking the sound barrier ” mean?
Mag When an airplane speeds up to the speed of sound, all sound waves arrive
at a particular place at the same time and enforce each other to become the “sound
wall”.
Coordinates of the circular wave crests

The flying object (the “source”) is traveling at speed vS relative to the air and at
an angle α to the horizontal axis (the x-axis), emitting sound waves in every period
that propagate in the air at the speed of sound c. In our representation, Fig. 6.18, the
source t is located at t = 0 at the site (0, 0), and circular wave crests are calculated
for every second, going back in time (negative time).
The center of the circle is given by the position of the flying object at the (negative)
time of emission:
x S (t) = (v S · cos(α)) · t
y S (t) = (v S · sin(α)) · t (6.20)
The expressions in parentheses decompose the distance vQ ·t traveled into hori-

zontal (x) and vertical (y) components. x S (t) and yS (t) are smaller than 0 because t <
0 for our settings.
Starting from the trajectory of the object, waves are spreading with velocity c and
have covered a distance r up to time t:
r = −c · t for t ≤ 0 (6.21)
so that (in our drawing plane) a circular wavefront arises. In three-dimensional

reality, the wavefronts are, of course, spherical surfaces. For our two-dimensional
representation, the quantities r become the radii of the circles in Fig. 6.18.
c speed of sound
vS speed of sound source
α angular deviation of the linear track of the source from the x-axis
t array of instants of time, negative; the current time is 0
r array of radii of wave crests emitted at times t
x S , yS arrays of coordinates of the source at times t; the position at t = 0 is (0,
0)
phi list of the polar angles for drawing the wave crests
x, y 2D matrices containing the coordinates of a set of wave crests, shape
size(phi) · size(t).
Calculating circles
In the spreadsheet of Fig. 6.19 (S), we produce eight circles to represent the crests
of waves that have been sent out at instants t = −1, −2, …, −8 s. Two examples are
shown in Fig. 6.18.
The definition of the coordinates of a circle is best done in polar coordinates. In
Fig. 6.19 (S), the angle is defined in [A13:A43] from ϕ = 0 to 2π in 31 steps of dϕ
= 0.209 = 2π/30, set in A10. The three parameters of the task (sound speed c, speed
vS of source, flight angle α) are defined in B1:B4 and get named in A1, A2, A4.
The worksheet in Fig. 6.19 (S) has the typical structure indicated by the bold
- shaped line. The calculation range below spans B13:R43. The column-specific
parameter set, coded as row vectors r, x S , yS , for the eight functions is in B7:I9 above
, controlled by the time t (see the formulas reported in J7:J9). The independent
variable ϕ, the polar angle of the circles, is in A13:A43, to the left of .
A B C D E F G H I J K L R S
1 c_ 340 m/s Sound speed
2 vS 500 m/s Speed of source ="vS="&vS&"; alpha="&B3&"°"
3 10 ° Direction angle vS=500; alpha=10°
4 alpha 0.175 rad =$B$3/180*PI()
5
6 t -1 -2 -3 -4 -5 -6 -7 -8
7 r_ 340 680 1020 1360 1700 2040 2380 2720 =-c_*t
8 xS -492.4 -985 -1477 -1970 -2462 -2954 -3447 -3939 =vS*COS(alpha)*t
9 yS -86.82 -174 -260.5 -347.3 -434.1 -520.9 -607.8 -694.6 =vS*SIN(alpha)*t
10 0.209 =2*PI()/30
hi)+x
S +yS}
10
+$A$ r_*COS(p (phi)
*SIN
11 =A13 = {=r_
12 phi x y
13 0.000 -152 -305 -457 -610 -762 -914 -1067 -1219 -87 -174 -695
14 0.209 -160 -320 -480 -639 -799 -959 -1119 -1279 -16 -32 -129
43 6.283 -152 -305 -457 -610 -762 -914 -1067 -1219 -87 -174 -695
Fig. 6.19 (S) Coordinates (x, y) of circles with their centers shifted in the xy-plane (below );
starts at B13 and is extended to the right and downwards; the x-coordinates in B13:I43 are gen-
erated with the formula reproduced in B11. The corresponding y values are in columns K:R. The
angular coordinates ϕ for all curves are in A13:A43 (independent variable left of ) . The time in
row 6 controls the radius (row 7) and the coordinates of the center (in rows 8 and 9)
This table can be enlarged row by row beyond row 43, for example, to allow for
a finer angular scale, because it is downwards open. It cannot be broadened column
by column, because, starting with column K, another calculation range follows, in
which the y-coordinates of the circles are calculated.
Remember: If you have written the formula in a cell correctly with

relative and absolute references or with variable names, you can drag
it into a larger cell range without changing it.
Questions

What is the meaning of the formula in cell B4, reported in D4?31
What is the formula in cell D7?32
What is the formula in cell D9?33
Wouldn’t it be nicer to have the wave crests cover the whole cone in
Fig. 6.18b? How would you do that?34
31 Transformation of the angle from degrees to radians: 360° = 2π, 180° = π.

32 [D7] = [= −c_*t], see J7!
33 [D9] = [=vS*sin(alpha)*t], see J9!
34 Specify the number of time instants to go from −1 to −12 in 12 steps.
Below , eight circles around the origin with the radii r from line 7 are calculated
from the column vector polar angle φ and the row vectors r, x S and yS .
The x values in B13:I43 are calculated with the simple formula [=r_*cos(phi)
+ xS]; the current values are taken from the same row or the same column as the
current cell.
Matrix formula
The formulas in range B13:I43 of Fig. 6.19 (S) refer, for each cell, to the same
column (when named row vectors are addressed) and the same row (when the named
column vectors are addressed). The same formula network can be generated in any
range with a matrix formula. An example is given in K13:I43 (= [{=r_*cos(phi) +
yS}]). To do so, activate the range, enter the formula, and complete with Ctrl +
Shift + Enter!
Task sliders Install two sliders to adjust the source’s direction and speed and
observe how the diagram reacts to changes in both parameters and a change in
sound velocity.
Discussion EXCEL
Tim The wave crests in my diagram do not change when I change the speed of
the aircraft.
Mag You have copied the numbers from Fig. 6.19 (S) into range B7:I9. How-
ever, these cells have to contain formulas, not just numbers. Then, the values in
these cells change as the parameters of the problem are changed. The formulas for
column I can be found in column J. If you drag cells I7:I9 (with the formulas in
J7:J9) to the left all the way to column B, the formula network for the parameters
is complete.
Tim What exactly is our task?
Mag The polar angle is in column A, independent of the time. Your task is to
enter a formula into cell B13 that creates the x coordinates of all eight circles at
time t by dragging to the right and down to cell I43. To get the y coordinates, you
have to apply a matrix formula, such as in K11 in Fig. 6.19.
At time t = 0, the aircraft shall be at the origin (0, 0) of the coordinate system.
The time in row 6 of Fig. 6.19 (S) is counted backward. So, the coordinates (x S ;
yS ) in rows 8 and 9 indicate where the flying object was 1, 2, etc., seconds ago.
They are calculated, as discussed above, from the speed vS of the source, the angle
α, and the time t. The respective location of the flying object at this past instant is
also the center of the circles representing the crests of the emitted sound waves.
Table 6.21 Coordinates of wave crests

1 c=340 #[m/s] speed of sound
2 vS=500 #[m/s] speed of source
3 alpha=10 #[Degree]
4 alpha*=np.pi/180 #[rad]
5
6 t=np.linspace(-1,-8,8,endpoint=True)
7 r=-c*t #Radius
8 xS=vS*np.cos(alpha)*t #Center
9 yS=vS*np.sin(alpha)*t
10
11 ph=np.array([np.linspace(0,2*np.pi,31)])
12 phi=ph.transpose(1,0)
13
14 x=r*np.cos(phi)+xS #Coord. of circle
15 y=r*np.sin(phi)+yS
6.6.4 Python
The Python program for calculating the coordinates of the wave crests is shown
in Table 6.21. After specifying the parameters speed of sound c, speed of source
vS, and deviation of the track of the source from the horizontal by an angle α,
we specify the instants t of time at which the source is supposed to emit a signal.
From t, we get the radii r of the circular sound crests and the Cartesian coordinates
of their centers x S and yS , i.e., the current positions of the sender, all as arrays
broadcast from t.
In ph, we define, as an array, the 31 angles with which circles are to be drawn
as regular polygons. This array is transposed into a column vector phi. With these
constructs and the two instructions
x = r*np.cos(phi) + xS
y = r*np.sin(phi) + yS
we reproduce the spreadsheet data structure of Fig. 6.19 (S) with the four row
vectors t, r, x S , yS, and one column vector phi. The values of the different arrays
are reported in Table 6.22. They coincide with those of the spreadsheet calculation
in Fig. 6.19 (S).
Tangent to a circle
With Fig. 6.20, we complete Fig. 6.18b with tangents to the circles that represent the
Mach cone. The construction of the tangent to a circle is illustrated in Fig. 6.20a and
realized in Table 6.23.
Task Verify the instructions in Table 6.23 with the help of Fig. 6.20a!
Table 6.22 Structure of the arrays in Table 6.21

alpha 0.175
t [-1 -2 -3 -4 -5 -6 -7 -8]
r [ 340 680 1020 … 2040 2380 2720]
xS [-492 -985 -1477 … -2954 -3447 -3939]
yS [ -87 -174 -260 … -521 -608 -695]
x[0] [-152 -305 -457 … -914 -1067 -1219]
Fig. 6.20 a (left) Construction of a tangent to a circle. b (right) Mach cone with the sound barrier
drawn as a tangent to the circles
Table 6.23 Tangent to a circle, drawn as straight lines from (0, 0) to (x T , yT ) and from (0, 0) to
(x Tu , yTu )
1 #Tangent, angle beta 6 xT=-np.cos(-alpha+beta)*b
2 xM0=vS*t[-1] 7 yT=np.sin(-alpha+beta)*b
3 r0=r[-1] 8 xTu=-np.cos(-alpha-beta)*b
4 b=np.sqrt(xM0**2-r0**2) 9 yTu=np.sin(-alpha-beta)*b
5 beta=np.arcsin(r0/-xM0)
1. What are the polar coordinates for the Cartesian coordinates (0, 5) and (1, 1)?
2. What are the Cartesian coordinates for the polar coordinates r = 2, and φ =
45° or φ = 135°?
3. The spreadsheet formula = cos(90) returns −0.44807362. Is excel thus
wrong? However, np.cos(90) similarly returns −0.4480736161291701.
Why?
4. Given the vector (3, 4), you are to attach this vector and its x- and y-
components as arrows to point (1, 1) . What does the data series in a
spreadsheet look like? What are the three lists in Python for the coordinates?
Fig. 6.21 Vector (2, 1) attached to (1, 0.5)
Table 6.24 Program snippet for attaching arrows at a point in the plane
1 FigStd('x',0,4,1,'y',0,2,0.5)
2 plt.plot(xO[0],xO[1],'kx')
3 plt.plot([xO[0],Arrow[0]],[xO[1],Arrow[1]],'k-')
4 plt.plot( … )
5 plt.plot([…,'k-.')
6 plt.axis('…')
5. In Fig. 6.21, the vector vA = (2, 1) and its components are attached to the
point x O = (1, 0.5). Specify the variables x O , vA , and Arrow and complete the
program snippet in Table 6.24!
6. You are to draw the path (x, y) of a point in the plane, which moves with speed
v in the xy-plane at an angle α = 30° to the x axis. What are the formulas in
the parameter representation (x, y) = f (v,α)?
7. Describe the rotation of a point on a circle around the origin with rotation
time T, both in polar and in Cartesian coordinates!
8. Which formulas apply to the Cartesian coordinates of a circle with diameter
d, moving with velocity v along the x-axis?
9. A point moves with constant velocity v along the y axis of a laboratory system.
What are its polar and Cartesian coordinates, in the laboratory system, and
in a system moving relative to the laboratory system, with constant angular
velocity ωD , around an axis through the origin of the laboratory system?
10. Calculate the moments of inertia of the two dumbbells in Fig. 6.22! The two
points represent masses of equal size. The connections between the masses
and to the suspension points are supposed to be massless.
Fig. 6.22 A dumbbell, (left)

suspended on the (massless)
middle stem fixed to it,
(right) suspended on two
threads at its ends
4 4
4 4
11. Explain Steiner’s theorem on the basis of Fig. 6.22a, b, together with Fig. 6.14!
12. How do you calculate the center of gravity of an arrangement of point masses
mi at (x i , yi )?
Integration of Newton’s Equation
of Motion 7
The Newtonian equation for one-dimensional motions of point masses is

numerically integrated by estimating the mean acceleration <a> in the time
intervals between the supporting points of the time trajectory. Four methods
are used: Euler, half step, “Progress with look-ahead” and Runge-Kutta of
the 4th order. Three examples from an adventurous life are simulated real-
istically: Stratosphere jumping, bungee jumping, and car in a racing start.
Different friction models (dry, viscous, Newtonian) are applied.
7.1 Introduction: Approximated Mean Value Instead

of Exact Integration
Solutions of Exercises 7.2 (Python), 7.3 (Excel), 7.4 (Python), and 7.6 (Excel) can
be found at the internet adress: go.sn.pub/THdoLU.
7.1.1 Newton’s Equation of Motion
We investigate the motion of a point mass on a straight line. According to Newton’s

laws, the acceleration at a certain point in time is determined by the forces acting
on the body at that point in time. These forces F are generally dependent on the
location x, e.g., due to a spring, or on the velocity ẋ, e.g., due to friction, or on
both, so that, generally, we have F = F(x, ẋ).
Newton’s equation of motion for one dimension is
F(x, ẋ)
ẍ(t) = (7.1)
m

270 7 Integration of Newton’s Equation of Motion
where ẍ = a is the acceleration, ẋ = v the velocity, x the location, and m the

mass of the body. If all forces within a period of time are known, as well as
the location and velocity of the point mass at the beginning of the period, the
differential equation of motion, Eq. 7.1, can be solved. Then, the values of x and
ẋ at the end of the time period can be calculated.
The forces in some of the exercises in this chapter depend only on velocity or
only on location, making the calculations simpler and easier to follow.
We learn how Newton’s equation of motion can be solved numerically in a
sufficiently precise way over a period of time. To introduce the different methods,
we work on tasks that can be analytically solved so that we can check the results
of our calculations: Vibrations of a mass-spring system without friction (Exercise
7.2) and falling through air with friction (v2 -proportional, Exercise 7.3).
Mean value instead of integral

The numerical solution we are striving for is based on difference equations in which
time progresses by finite amounts t. If the location and the velocity of the body at
the beginning t 0 of a time interval are known, the values at the end t 0 + t of the
interval follow as
t0 +t
v(t0 + t) = v(t0 ) + a(t)dt = v(t0 ) + a · t (7.2)
t0
t0 +t
x(t0 + t) = x(t0 ) + v(t)dt = x(t0 ) + v · t (7.3)
t0
Equations 7.2 and 7.3 are exact. The expressions with the mean val-
ues <a> and <v> correspond exactly to the integral, because the mean value is simply
defined like that.
Our task is to estimate the mean acceleration <a> and the mean velocity <v> in
the time interval under consideration. The numerical estimation is an approximation
for which we present four different methods in the next section.
Approximated mean instead of exact integral.
Friction
To consider friction, we use the general formula:
a Fric = −sgn(v) · a f · |v|n
The acceleration due to friction is proportional to a power n of the speed (the

absolute value of the velocity). The first term −sgn(v) guarantees that friction always
tries to decrease the speed. The exponent n depends on the type of friction:
– n = 0 for internal friction, applied for the losses in a bungee rope (Exercise 7.6)
– n = 1 for viscous friction, applied for the damping of a harmonic oscillator
(Exercise 7.2)
7.1 Introduction: Approximated Mean Value Instead of Exact Integration 271
– n = 2 for Newtonian friction, active when dragging a body rapidly through a

viscous medium, in our exercises, air (Exercises 7.3, 7.4, and 7.5).
Observe the influence of the different types of friction on the time trajectory!
7.1.2 Four Methods for Estimating the Average Acceleration

in a Time Segment
In the simple Euler method, the mean acceleration in the time segment [t n , t n+1 ]
is approximated by the acceleration at the beginning of the time segment:
v(tn+1 ) = v(tn ) + a(tn ) · t (7.4)
x(tn+1 ) = x(tn ) + v(tn ) · t (7.5)
This procedure can be improved if the values from Eq. 7.4 and Eq. 7.5 are taken
only as a “preview” for the velocity v p (tn + t) and the location x p (tn + t),
from which the acceleration a p (tn + t) at the end of the considered time period
is calculated. For this “Progress with look-ahead”, the quantities at the beginning
of the next interval are estimated as:
a(tn ) + a p (tn + t)
v(tn+1 ) = v(tn ) + · t (7.6)
2
v(tn ) + v p (tn + t)
x(tn+1 ) = x(tn ) + · t (7.7)
2
The half-step procedure has a similar structure; the values at the beginning of
an interval are used to estimate the values in the middle of the interval, which then
represent the entire interval.
A further improvement can be achieved with the fourth-order Runge-Kutta
method, in which three projections are made in an interval, the first two into the
middle and the third to the end of the interval. Then, a weighted average of four
values is taken as representative of the whole interval.
All methods are trained

We shall apply all four methods in Exercise 7.3 (Falling from a great height; the
force depends only on the speed) and compare the results with each other. There are
analytical solutions for this task, so that we can check the precision of our numerical
methods.
All methods provide sufficiently accurate solutions if the distance between the
supporting points, which mark the time segments’ boundaries, is made sufficiently
small. The bigger the effort with which the mean acceleration in a time segment is
calculated, the longer the time segments can be. Consequently, it must be checked,
for each method, whether the selected length of the time segment is short enough.
As standard procedure in excel, we shall use Progress with look-ahead, because
it proves to be sufficiently efficient and can be implemented clearly in a spreadsheet.
However, experienced readers may also use the half-step procedure, which provides
the same accuracy and is also explained in detail in Exercise 7.3.
In Python, we shall use a function that implements the 4th-order Runge–Kutta
method.
Adventurous life
In the course of this chapter, we will deal with three cases from an adventurous life:
stratospheric jumping, bungee jumping, and a car in a racing start. In the follow-up
volume Physics with Excel and Python, Using the Same Data Structure. Applications,
the procedures from this chapter come into full fruition, with the treatment of motions
in the plane, all kinds of oscillation, field lines, and wave functions of the Schrödinger
equation.
7.1.3 Tactical Approaches in Python and Excel
In Python, the steps from t to t + dt are done in a progress loop that calls a
progress function, invariably called progr in all exercises. It does not have a proper
code, but is assigned to one of the existing functions for Euler, look-ahead, or
Runge-Kutta. Within the progress functions, the acceleration is invariably called
acc, again, without proper code, but with an assignment to a function that is spe-
cific for the physical problem under consideration (accSpring, accFall, accJump,
accPwr, accBungee). This approach is typical of Software solutions in which
functions are embedded within a larger body.
The excel approach is less general and must be adapted individually to each
problem. The preview calculations for the current time interval are done in a row
with specific formulas for acceleration. The values at the beginning of a new time
interval in a new row have to be calculated from values in the preceding row.
Animations
The motions calculated in this chapter can easily be animated with the methods
presented in Sect. 6.2.5.
7.2 Harmonic Oscillation with “Progress with Look-Ahead”

and “Runge–Kutta”
We integrate Newton’s equation of motion for a mass-spring system by cal-

culating the acceleration in a time interval, (a) as an average of two values,
7.2 Harmonic Oscillation with “Progress with Look-Ahead” … 273
one at the beginning of the interval and the other calculated for the end of
the interval with the values at the beginning, and (b) as an average over four
values with the Runge-Kutta method.
7.2.1 Equation of Motion
The equation of motion for a mass fixed to a linear-elastic spring is as follows:
D
a = ẍ = −d ·v =−f ·x −d ·v (7.8)
m
Here, x is the deflection out of the rest position, and D and m are the spring
constant and the mass. Damping is set proportional to the velocity v and opposite
to it: –d·v.
Questions
What is the physical unit of the “spring constant” f in Eq. 7.8 in SI units?1
What is the physical unit of the damping constant d in Eq. 7.8 in SI units?2
What are the initial conditions x(0) and v(0) in Fig. 7.1a?3
The solution for vanishing friction (d = 0) is a stationary oscillation with con-

stant amplitude. We use this knowledge to check whether our numerical solutions
are good enough. An example is shown in Fig. 7.1 for an oscillation with f = 0.10
calculated with LookAhead.
The zoom in Fig. 7.1b shows that the amplitude for n = 100 has increased
over the initial value, indicating that 100 points for a time span of 80 are not good
enough, but that n = 400 does the job. The position of the fourth maximum √ is at t
= 79.5, corresponding reasonably well to the theoretical value of 4π/ (0.1) =
79.48. We may therefore rely on LookAhead with this segmentation of time and
can play around with the parameters.
In Fig. 7.2, we have doubled the spring constant and introduced damping, caus-
ing a strong decay of the amplitude. In the zoom, we see furthermore that the
period of the damped oscillation is bigger than for the undamped oscillation.
Task Vary the constant f = D/m and observe √ whether the period duration
behaves as predicted by the formula T = (2π)/ f (for vanishing friction)!
1 [f ] = [a/x] = (m/s2 )/m = 1/s2 .

2 [d] = [a/v] = (m/s2 )/(m/s) = 1/s.
3 x(0) = 1 (maximum deflection), v(0) = 0 (slope of x(t)).
(a) (b)
1.5 1.05
n=100
n=400
1.0
x
x
0.5
1.00
0.0
0 20 40 60 t 80
-0.5
n=100
n=400
-1.0
0.95
78 79 t 80
-1.5
Fig. 7.1 a (left) Oscillation calculated with LookAhead. b (right) Zoom of a
(a) (b)
1.5 0.05
f=0.2, d=0
f=0.2, d=0
d=0.2
1.0 d=0.2
x
x
0.5
0.0 0.00
0 20 40 60 t 80 65 70 75
t
-0.5
-1.0
-1.5 -0.05
Fig. 7.2 Oscillation of a harmonic oscillator. a (left) Undamped and damped. b (right) Zoom of
damped oscillation
In the following sections, we present LookAhead in a spreadsheet and LookA-

head, Euler, and Runge-Kutta in Python. Varying the number n of points as in
Fig. 7.1, reliable results are obtained with:
n = 20,000 for Euler,
n = 400 for Progress with Look-ahead,
n = 60 for Runge-Kutta.
There is a reduction of about 50 going from Euler to LookAhead and about
8 going further to Runge-Kutta. Therefore, we choose LookAhead as standard in
spreadsheet calculations with only four additional columns (see Fig. 7.3), thus
keeping the programming effort to be repeated in every new spreadsheet small. We
choose Runge-Kutta as standard in Python because the 12 necessary statements
(Table 7.2) have to be implemented only once in a function that can be used in all
further exercises.
A B C D E F G H
1 n 400 f 0.10 spring constant
2 dt 0.20 =80/n d 0.00 damping constant
3 n=400 ="n="&n
t t
)/2*d 6+G6)/2*d v
t 6+F6 * dt *dt -d*vD
4 =A6+d =B6+
(C
=C6+
(D
=-f*x
-d
=x+v* =v+aA =-f*x
D
5 t x v aA xD vD aD
6 0.00 1.00 0.00 -0.10 1.00 -0.02 -0.10
7 0.20 1.00 -0.02 -0.10 0.99 -0.04 -0.10
406 80.00 0.98 -0.06 -0.10 0.97 -0.08 -0.10
Fig. 7.3 (S) An oscillation of a mass-spring system is calculated with LookAhead. The time runs
vertically down, the look-ahead to the end of the current time segment is performed horizontally
to the right. The mass is assumed to be m = 1
n number of points for a time span of 80 s

f, d spring and damping constants
dt = 80/n, length of the time interval
t array of time points, (n + 1) elements, dt apart
x deflection of the mass at t (array)
v velocity of the mass at t (array)
a acceleration at t (array)
x A , vA deflection and velocity at a specific time t A
x D , vD deflection and velocity predicted for time t + dt
xB , xC deflection predicted at time t + dt/2 (Runge-Kutta)
vB , vC velocity predicted for time t + dt/2 (Runge-Kutta)
All calculations are performed for t = 0 to t = 80 s.
The basic spreadsheet layout for integrating Newton’s equation of motion with
LookAhead is shown in Fig. 7.3 (S), with the resulting deflection displayed in
Fig. 7.1.
The time t in column A runs from top to bottom, and thus so does the deflection
x(t) of the mass in column B and its velocity v(t) in column C. For a step from t n
to t n+1 :
– the acceleration a at the beginning of the time segment beginning with t n is

calculated in each row from the values of x and v at that time (column D),
– with these values, the deflection x D and the velocity vD at the end t n + dt of
the time segment are predicted (columns E and F),
– with these new values, the acceleration aD at the end of the time interval is
estimated in column G.
All calculations reported for columns D to G use formulas with named variables
referring to the same row. The new values for x and v at the next instant t n+1 of
time are calculated in the following row with information from the previous row;
see the formulas in B4 and C4 valid for B7 and C7, e.g., B7 = [=B6 + (C6 +
F6)/2*dt].
The estimation of the mean value at the time t n+1 = t n + dt can be improved by
predicting further values for speed and acceleration in the current time segment [t n ,
t n + dt) in further columns, especially with the more efficient fourth-order Runge-
Kutta method. However, due to the simpler table layout, we use LookAhead as our
standard in spreadsheet calculations.
7.2.4 Python
Progress functions
In the second cell of Table 7.1, we have implemented the LookAhead method as a
function, similar to the spreadsheet calculation of Fig. 7.2 (S). In the first cell of
Table 7.1, the Euler method is implemented. In both functions, the acceleration is
not calculated explicitly, but is rather outsourced to another function acc that must
be specified in the main program.
The Runge-Kutta method of the 4th order is implemented in a function reported

in Table 7.2, with two jumps into the center of the interval (x B , vB and x C , vC ) and
one jump to the end of the interval (x D , vD ). The values (x R , vR ) for the next time
instants to be returned are calculated as a weighted average over the velocities or
the accelerations.
Ψ Half, half, whole; the halves count twice.
The main program is shown in Table 7.3. In the first 4 lines, the parameters of
the exercise are defined, followed by the definition of the function accSpring for
the acceleration which makes use of the global parameters f and d. The arrays t,
x, v are defined with their length corresponding to the specified number n of time
instants, with the first entries containing the initial conditions set in lines 12–14.
Table 7.1 Functions performing the Euler and the lookAhead methods; the current values of x(t)
and v(t) are passed as x A and vA
1 def Euler(xA,vA): 1 def lookAhead(xA,vA):
2 aA=acc(xA,vA) 2 aA=acc(xA,vA)
3 xR=xA+vA*dt 3 xD=xA+vA*dt
4 vR=vA+aA*dt 4 vD=vA+aA*dt
5 return xR,vR 5 aD=acc(xD,vD)
6 xR=xA+(vA+vD)/2*dt
7 vR=vA+(aA+aD)/2*dt
8 return xR,vR
Table 7.2 Function implementing the Runge-Kutta method

1 def RungeKutta(xA,vA):
2 aA=acc(xA,vA)
3 xB=xA+vA*dt/2 #Half step into center of interval
4 vB=vA+aA*dt/2
5 aB=acc(xB,vB) #Accel. at center of interval
6 xC=xA+vB*dt/2 #Second half step
7 vC=vA+aB*dt/2 #
8 aC=acc(xC,vC) #Accel. at center of interval
9 xD=xA+vC*dt #Full step
10 vD=vA+aC*dt #
11 aD=acc(xD,vD) #Accel. at end of interval
12 xR=xA+(vA+2*vB+2*vC+vD)*dt/6
13 vR=vA+(aA+2*aB+2*aC+aD)*dt/6
14 return xR,vR #Position, velocity at end of interval
Table 7.3 Calculating the motion of an oscillator

1 n=30 #Number of time instants
2 dt=80/n
3 f=0.10 #Spring constant
4 d=0.0 #Damping constant
5
6 def accSpring(x,v):
7 return -f*x-d*v
8
9 t=np.zeros(n)
10 x=np.zeros(n)
11 v=np.zeros(n)
12 t[0]=0 #Initial conditions
13 x[0]=1
14 v[0]=0
15 acc=accSpring #acc = Name in progress function
16
17 for i in range(n-1): #Progress loop
18 t[i+1]=t[i]+dt
19 #progr=Euler; lbl1="Euler, n="+str(n)
20 #progr=lookAhead; lbl1="lookAhead, n="+str(n)
21 progr=RungeKutta; lbl1="RungeKutta, n="+str(n)
22 x[i+1],v[i+1]=progr(x[i],v[i])
Progress loop
The integration is performed in a progress loop (line 17–22) progressing by time
steps dt. It recurs to the progress function progr that has no code of its own but is
linked to one of the progress functions Euler, lookAhead, or RungeKutta. To keep
Fig. 7.4 Oscillation of a mass-spring system calculated with 4th order Runge–Kutta (with Table
7.3)
track of the program flow it is necessary to create the labels for the corresponding
figures in the same program cell.
Within the progress functions, in Table 7.1, the acceleration is invariably taken
from a function acc that, again, has no code of its own but must be linked to a
function that is specific to the physical problem under consideration, here, in line
15, to accSpring. This increases the versatility of programming in that different
approaches can be tried out.
To get stable results, we have to choose n = 20,000 for Euler, 400 for LookAhead,
and 60 for 4th order Runge-Kutta (n = 30 is not good enough). The results of
Runge-Kutta calculations are shown in Fig. 7.4.
7.3 Falling from a (Not Too) Great Height
Two forces act on a body falling from a great height: constant gravity and
friction proportional to the square of the velocity. For these settings, the
acceleration depends only on the velocity of the body, so that the location
does not have to be calculated synchronously but can be determined subse-
quently by integration over the velocity. Initial acceleration and stationary
velocity are compared with analytical solutions.
7.3 Falling from a (Not Too) Great Height 279
7.3.1 Limiting Cases, Analytically Solved
We drop a body from a great height and let two forces act on it: the gravity –m·g,
assumed to be independent of the height, and a friction force k·v2 , proportional
to the square of the speed (Newtonian friction) and opposite to its direction. With
these assumptions, the equation of motion results as
k 2 k
a = −g − v · sgn(v) = −g − v · |v| (7.9)
m m
where sgn (signum) is a function that calculates the sign of the argument. The
force m·a depends only on the speed, because we assume, simplifying, that gravity
and friction forces do not depend on the height. So, we don’t have to consider
the location in the calculation and may calculate it afterward by integrating the
velocity trajectory. For a fall through air, k = ρ A/2 is valid with the density ρ of
the air and the cross-sectional area A of the body.
In Fig. 7.5a, velocities v(t) obtained with Runge-Kutta of the 4th-order (“RK4”) for
two interval lengths dt = 2 and dt = 4 and with LookAhead for dt = 0.5 coincide
at their points of calculation. Halving the interval length does not change the RK4
solution, so dt = 4 is good enough. For LookAhead, an interval length eight times
smaller is necessary to obtain the same result.
In Fig. 7.5b, v(t) trajectories are shown for three different initial velocities: zero,
upwards, and downwards. The final speed is independent of the initial velocity.
(a) (b)
-40
5 10 t [s] 15
v [m/s]
RK4; dt=2
RK4; dt=1
lookAhead; dt=0.5
-45 final speed. for
k/m=0.004
-50
Fig. 7.5 a (left) Velocity of the falling object calculated in a spreadsheet with Runge-Kutta of the
4th order (RK4) and lookAhead. b (right) Velocity of the falling object as a function of time for
different initial velocities, calculated with RK4 in Python
If, at the beginning of the fall, the velocity is 0, v(0) = 0, the initial gradient of
the velocity curve is -g; the initial acceleration is the acceleration due to gravity
close to the surface of the earth.
After a certain time, the body falls at a constant speed, because the frictional
force equals gravity. This final velocity results from the condition a = 0 (no further
acceleration):

m·g
v Fin = − (7.10)
k
g acceleration due to gravity

cV = k/m, (see Eq. 7.9) pre-factor of friction
t array of time instants dt apart
vA velocity at t
Accelerations
aA at the beginning of the calculation interval

aB , aC in the center
aD at the end.
7.3.3 Spreadsheet
Runge-Kutta of the 4th order

The Runge-Kutta method implemented in a function requires 10 additional statements
(1 for A, 3 × 3 for B, C, D, see Table 7.1). A corresponding spreadsheet layout would
require 10 additional columns. As our problem is independent of the location, we
need only 1 + 3 × 2 = 7 columns. This is realized in Fig. 7.6 (S).
The values in any row within the range 6–26 of columns C to I are valid for
the interval [t n , t n+1 ) and are obtained as follows:
– (1) The acceleration aA at time t n is calculated from the velocity at that time
(column C).
– (2) The velocity vB is calculated with the values vA and aA for the middle of
the interval, i.e., for t n + dt/2, and from that, the corresponding acceleration
aB (first half-step, column E) is obtained.
A B C D E F G H I J
1 g 9.81 m/s² cV=k/m:=0.004
2 cV 0.004 1/m
3 dt 2.00 RK4; dt=2 ="RK4; dt="&dt
t/6
I6)*d ) ) )
+2*G6+ IGN(vA N(vB N(vC (vD)
*E6 2* S 2 2*SIG 2 2*SIG ABS
+ d t +( C6+2 g-cV*vA^ A+aA*dt/ g-cV*vB^ A+aB*dt/ g-cV*vC^ A+aC*dt g-cV*vD*
4 =A6 =B 6 =- =v =- =v =- =v =-
5 t vA aA vB aB vC aC vD aD
6 0.0 0.00 -9.81 -9.8 -9.43 -9.4 -9.45 -18.9 -8.38
7 2.0 -18.65 -8.42 -27.1 -6.88 -25.5 -7.20 -33.1 -5.44
26 40.0 -49.52 0.00 -49.5 0.00 -49.5 0.00 -49.5 0.00
Fig. 7.6 (S) Calculation with 4th order Runge-Kutta; vB , aB , vC and aC are predicted values in the
middle of the time interval, vD and aD are predicted values at the end of the time interval
– (3) The velocity vC and, from that, the acceleration aC for t n + dt/2 are calcu-
lated a second time (second half step, columns F and G), now with the values
aB and vB .
– (4) The velocity vD at the end of the interval, i.e., for t n + dt, is calculated with
the values aC and vC and, from that, the acceleration aD (whole step, columns
H and I)
The new value of the velocity at t n+1 is calculated in the next row (e.g., in B7
with the formula in B5) from the velocity at t n and a weighted average of the four
accelerations calculated in the previous row (here row 6) for the interval [t n , t n+1 ):
vn+1 = vn + a · dt with a = (1a A + 2a B + 2aC + 1a D )/6 (7.11)
The accelerations from the half steps count twice. We remember the procedure
with a broom rule:
Half, half, whole, the halves count twice. (4th order Runge-Kutta).
“Halves” means half-steps.
Questions

Which accelerations from steps (1)–(4) are used in the formulas for
velocities reported in F4, H4, and B4?4
Why is the sum in B7 divided by 6 to get the average, although only four
accelerations enter the formula?5
4 aB in the center of the time interval after the first jump; ac in the center after the second; aA , 2aB ,
2aC , aD for t = 2.
5 Half, half, whole; the halves count twice. The sum of the weights is 6.
Fig. 7.5a shows two solutions obtained with Fig. 7.6 (S) for dt = 4 and dt = 2.
Task Determine the height fallen until you reach 95% of the final speed! You
have to integrate the speed over time!
To get the three functions v(t) displayed in Fig. 10.5b, we have to run the
calculation three times with different initial values written into B6 and store the
results vA (t) in extra columns (copy, paste special/contents).
7.3.4 Python
Table 7.4 reports the preparatory instructions for the progress loop. A special fea-
ture of the code is the stacking of x and v into a 2D matrix to store the fall
trajectories for three initial conditions.
In lines 1–4 of Table 7.4, the fall parameters and the function for the accelera-
tion are defined. In lines 8–10, the elementary arrays t, x, v are specified. In order
to calculate the motion for three different initial velocities, position x and velocity
v are replicated in np.stack constructions so that they can be addressed by the
same index as the curve itself. Labels and line styles for the three curves in a
diagram are also organized as lists with three entries in lines 16 to 19 in Table 7.5.
The progress of motion for the three initial velocities 0, 40, and—80 m/s is
calculated in Table 7.5 in a nested loop. In the first loop over k, line 17, labels are
created for the three initial velocities and in line 19, line styles before the progress
is calculated.
The program for plotting the three time-curves is given in Table 7.6. The line
styles and the labels have been specified in Table 7.5 so that curves, styles, and
labels can all be addressed with the same index.
Table 7.4 Parameters for the fall and function for calculating the acceleration during the fall
1 g=9.81 #[m/s²] Gravitational acceleration
2 cV=0.004 #[1/m] Friction coefficient
3 def accFall(x,v):
4 return -g-cV*v**2*np.sign(v)
5 acc=accFall #Link to acc in progress function
6
7 dt=0.05 #[s] Time increment
#t sets the structure also for x and v, and determines the
length of the progress loop.
9 x=np.zeros(len(t))
10 v=np.zeros(len(t))
11 xM=np.stack((x,x,x)) #For 3 initial conditions
12 vM=np.stack((v,v,v))
Table 7.5 Progress loop for the fall; “Runge–Kutta” is the function defined in Table 7.2
13 vM[0,0]=0
14 vM[1,0]=40
15 vM[2,0]=-80
16 lbl1=[0,1,2]
17 for k in range(3): #Labels for three curves
18 lbl1[k]='cV='+str(cV)+' v[0]='+str(vM[k,0])
19 linStyle=['k-','k--','k-.']
20 for k in range(3): #Over all initial conditions
21 for i in range(len(t)-1): #Progress loop over t
22 xM[k,i+1],vM[k,i+1]= RungeKutta(xM[k,i],vM[k,i])
Table 7.6 Plotting three time-curves

23 FigStd('t[s]',0,20,5,'v[m/s]',-80,40,40)
24 for k in range(3): #Over initial conditions
25 plt.plot(t,vM[k],linStyle[k],label=lbl1[k])
26 plt.plot([0,20],[0,-20*g],'k:', label = "slope -g")
#Straight line with slope -g
27 plt.legend()
Questions
What is the shape of x M and vM in Table 7.4 and what are the first and last two
elements of t?6
What does “linStyle[k]” in line 25 of Table 7.6 do?7
How can the two k-loops in Table 7.5 be merged into one?8
7.4 Stratospheric Jump
When considering jumps from great altitudes, it must be taken into account
that the coefficient of friction changes with height according to changes in
air density. The friction force is, therefore, a function of both location and
velocity. We model the air density in a simplified way with the barometric
formula for a constant temperature (15 °C). In this model, the maximum
speed for a jump from 39 km altitude is reached after a 50 s fall.
6 The shape of x M and vM is (401, 3), len(t) = 401, t = [0, 0.05, …, 19.95, 20.00].
7 linStyle[k] selects the k-th element of the list linStyle defined in line 19.
8 Insert line 18 between lines 20 and 21. The indentation with respect to line 20 is already correct.
(a) (b)
v [m/s] 0 40,000
h [m]
-100 30,000
hE=8400m
-200 20,000
f = fF
-300 hE=8400m 10,000
The engineer's
pulse
-400 0
0 100 200 t [s] 300 0 100 200 t [s] 300
Fig. 7.7 a (left) Velocity for a fall from 38,969 m, compared with the data from Baumgartner’s
jump (“The engineer’s pulse”, Fig. 7.8a); the dashed line “f=fF” is valid for constant friction. b
(right) Altitude as a function of time for the velocity calculated in a
A B C D E F G H I J
3 dt 0.4 s fF 0.004 1/m
4 g 9.81 m/s² hE 8400 m hE=8400m
*dt /hE) *dt E)

H7)/2 h I7)/2 hD/h ^2
5 =A7+d
t
B 7 +(D7+ fF*EXP(- D7+(E7+ -g+f*v^2 h+v*dt fF*EXP(- v+a*dt -g+fD*vD
= = = = = = = =
6 t h f v a hD fD vD aD
7 0.0 38969 0.0000 0.00 -9.81 38969 0.0000 -3.92 -9.81
8 0.4 38968 0.0000 -3.92 -9.81 38967 0.0000 -7.85 -9.81
807 320.0 -150 0.0041 -49.45 0.15 -170 0.0041 -49.39 0.15
Fig. 7.8 a (top) Measured data of the jump of October 14, 2012 (The engineer’s pulse, Oct.15,
2012, Mechanical analysis of Baumgartner’s dive (Part B)). b (bottom, S) Spreadsheet layout for
a jump from a high altitude, when the friction coefficient (f and f D in columns C and G) is height-
dependent
On October 14, 2012, the Austrian adventurer Felix Baumgartner ascended into
the stratosphere in a helium balloon. He jumped out at a height of about 39 km
and fell about 34 km in free flight, reaching supersonic speed before opening
his parachute. Some measurement data are reported in Fig. 7.8a and plotted in
Fig. 7.7a as a dashed line. We are going to simulate this jump with a simple
model for height-dependent friction.
Height-dependent friction force

The coefficient of friction depends on the altitude, because the density of air, and, with
it, the coefficient of friction, decreases with increasing altitude. We apply a height-
dependent friction coefficient that directly corresponds to the barometric formula:

h
f (h) = f F · exp − (7.12)
he
7.4 Stratospheric Jump 285
with f F = 0.004 1/m valid for h = 0 (the same as in Exercise 7.3). With he = 8400 m
(valid for an air temperature of 15 °C), we get the best coincidence with the empirical
data for the velocity as a function of altitude (see Fig. 7.7a).
Our simple model can well reproduce the essential characteristics of the time
trajectory of the velocity. In the beginning, the fall is free, i.e., without friction,
showing up as a linear increase in speed. After about 50 s, the maximum fall speed
is reached; in the real experience, it exceeded the speed of sound. After that, the
speed decreases sharply because of increasing friction due to increasing air density
and approaches the stationary speed we calculated for f = f H = c0 in Exercise
7.3.
For a more realistic simulation, we have to consider that the atmosphere consists
of different air layers with different temperatures. But we are content to find the
main characteristics of the jump when applying just the barometric formula with
constant temperature.
t array of time points, dt apart, independent variable

hE characteristic height of the barometric formula
fF friction at h = 0
h height for t
f friction coefficient at h
v velocity at t
a acceleration at t
h D , f D , v D , aD values predicted for the end of the interval.
The spreadsheet layout for the integration of the equation of motion is presented in
Fig. 7.8b. The integration method is LookAhead. The main change from Exercise
7.3.2 is that the height is now calculated simultaneously with the velocity. This is
necessary because the coefficient of friction is height-dependent.
The height is calculated in column B from the height and the speed in the
preceding rows; from that, in column C, the coefficient of friction f , which is then
used to calculate the acceleration a in column E. Analogously, the same applies to
the “looked-ahead” values in columns F to I.
Table 7.7 Parameters for the stratospheric jump, together with a function for the acceleration
1 n=801 #Number of calculation points
2 dt=320/(n-1) #Time span 320 s
3 g=9.81 #[m/s²], Gravitational acceleration
4 hE=8400 #Characteristic barometric height
5 fF=0.004 #[1/m] Damping constant
6
7 def accJump(xi,vi): #x represents height
8 f=fF*np.exp(-xi/hE) #Barometric formula
9 ac=-g+f*vi**2
10 return ac
Questions
Which thermodynamic quantities determine the characteristic height hE ?9

The formulas for a and aD in Fig. 7.8b (S) are valid only when the velocity
is always negative, so that the body is always falling. Sloppy programming!
How do the formulas have to be changed when the body can also rise?10
Interpret the formula in B8:B807 in the form h(t+dt)=!11
The calculated velocity is shown in Fig. 7.7a as a function of time and compared
with the empirical data of the stratospheric jump by Baumgartner.
7.4.3 Python
The parameters for the Python program are specified in Table 7.7, together with
a function for the acceleration, accJump.
The main program is reported in Table 7.8 with the specification of the arrays
t, h, and v and their initial values for t = 0. The for-loop is identical to all our
programs for integrating the equation of motion. We have to assign the name acc,
called in our standard function RungeKutta, to the acceleration function specific to
the current problem.
7.5 A Car Drives with Variable Power
A driver wants to accelerate his vehicle at full throttle, but making sure that
the wheels are not spinning (a “racing start”). We consider two approxima-
tions for this behavior, constant power and speed-proportional power. We
9 The height hE arises from the Boltzmann distribution of air molecules in the atmosphere: air den-
sity ρ(h) = ρ(0) · exp(−mgh/k B T ); h E = k B T /mg, where m is the average mass of the air
molecules.
10 Instead of [=… + v2 ] one sets [=… + abs(v)ˆ2*SIGN(v)] or [=…−v*abs(v)].
11 Formula in B8:B807: h(t + dt) = h(t) + (v(t) + v (t))/2·dt.
D
7.5 A Car Drives with Variable Power 287
Table 7.8 Progress loop for the stratospheric jump

11 t=np.empty(n) #n determines the length of t, h, and v
12 h=np.empty(n)
13 v=np.empty(n)
14 t[0]=0
15 h[0]=38969 #[m] Jump height
16 v[0]=0
17
18 acc = accJump #acc() accessed within RungeKutta
19 for i in range(n-1):
20 t[i+1]=t[i]+dt
21 h[i+1],v[i+1]=RungeKutta(h[i],v[i])
answer the following two questions: How does the final speed depend on
the frictional resistance? How do longer-term fluctuations in power become
noticeable? The programming challenge is to realize case distinctions.
7.5.1 Various Types of Power
Analytical solution for vanishing sliding friction

In this section, we characterize the racing start of a car (“full throttle”) by a constant
power P during the whole process (not by constant acceleration, as is usually assumed
in textbook exercises). Without friction, the work W is converted into kinetic energy:

m 2P √
W = P · t = v2 → v = · t (7.13)
2 m
The resulting velocity v and the acceleration a = dv/dt are shown in Fig. 7.9a
for P = 100 kW and a car with mass m = 1500 kg. The velocity increases with the
square root of time, and the acceleration becomes infinitely large at t = 0.
Non-zero motion friction and limited acceleration

In the real case, two effects have to be considered: (1) motion-inhibiting friction
(rolling friction, driving resistance) and (2) the condition that the driving force must
never be greater than the static friction force between the tire and the road. The
second effect is taken care of by limiting the acceleration to a maximum value aMax .
We generally set the frictional force that inhibits motion proportional to vnF so that
the exponent nF can later be selected as a parameter, e.g., nF = 2 for friction caused
by an air stream.
100 6 100 6
v [m/s]
80 80
v [m/s]
4 4
60 60
11.55√t
40 a --> 40 <-- v; μ=0
v-End
a [m/s²]
2 2
a [m/s²]
<-- v; μ=0.5
20 20 v=aMax·t
a -->
0 0 0 0
0 20 40 t [s] 60 80 0 20 40 t [s] 60 80
Fig. 7.9 a (left) Speed and acceleration for a constant power of P = 100 kW, without friction
losses, calculated according to 7.13. b (right) Speed and power if driving friction is taken into
account and acceleration is limited to amax ; numerically calculated
The power needed to increase the kinetic energy and to overcome friction is
d m 2
P= v + v · FR (7.14)
dt 2
dv
FR = µ · v n F → P = m · v · + µ · vn F +1 (7.15)
dt
The acceleration
dv P µ
a(t) = = − · v(t)n F (7.16)
dt m · v(t) m
does not depend on the location. It can be calculated from the power and the current
velocity alone.
Question
The velocity in this task is always greater than or equal to zero. How do
Eqs. 7.15 and 7.16 have to be changed if negative velocities are allowed?12
For stationary motion (v = const.), we get
dv
n F +1
a(t) = = 0 → vs (t) = P/µ (7.17)
dt
The stationary velocity vs can easily be calculated analytically with Eq. 7.17,
which we use to check the numerical calculation. This check also works the other
12 v(t) → Abs(v(t); v(t)n F → Abs(v(t))n F · sgn(v(t))

way around, because, if numerically and analytically calculated values do not

match, this may also indicate that our mathematical derivations are erroneous.
It is, therefore, doubly good to compare the results of the two methods; we check
whether our logical reasoning and our numerical calculation are consistent.
The calculated curves for the acceleration and the speed under friction can be
seen in Fig. 7.9b. The acceleration at the beginning of the journey is at its pre-set
maximum value 5 m/s2 (for the acceleration, the right ordinate is valid) and goes
practically to zero within 80 s. The speed is limited; the numerically calculated
curve converges towards the stationary value obtained from Eq. 7.17. We can,
therefore, assume that we did not make any gross programming mistakes. Without
friction (μ = 0), the velocity would continue to increase as in Fig. 7.9a.
Fluctuating power
What is the impact of power fluctuations on the speed? The answer is found in
Fig. 7.10a, similar to Fig. 7.9b, yet with the power always fluctuating by 10% after a
time span DelT = 3.5 s. Although the acceleration fluctuates on this scale, the speed
hardly fluctuates. Speed is the integral of acceleration and, as such, averages over
fluctuations. The fluctuations of the acceleration influence the speed more strongly
when they remain constant over a longer duration.
Task Change the time period DelT during which the power remains constant
and observe the time course of a and v!
Speed-proportional power
For internal combustion engines, the power depends on the number of revolutions of
the engine, and thus on the speed. If a function is known for this dependence, it can
be included in the formula for the power. We investigate the simple case in which
100 6
a [m/s²]
vEnd
80 <-- v;μ=0.5
a; DelT=3,5 -->
4
60
v [m/s]
40
2
20
0 0
0 20 40 t [s] 60 80
Fig. 7.10 a (left) Five different starting processes with a fluctuation of the power by about 10%;
the power remains constant within DelT = 3.5 s = 10 x dt. b (right) A starting process for speed-
proportional power for the parameter values Pmin = 1000, Pmax = 100,000 and pV = 2000
the power P is proportional to the speed v:

P = pv · v (7.18)
Tim If the car is at a standstill, the power is zero, i.e., the car does not start at
all.
Alac That’s why you put your foot down when you’re idling and then let the
clutch come.
Tim How are we supposed to include that in our calculation model?
Mag With a case distinction: If the power, according to Eq. 7.18, is too small,
we use a constant value Pmin ; in real life, this would be achieved by letting the
clutch slide.
Alac That’s for low speed. On the other end, at high revs, the engine runs out
of breath.
Mag Then we apply a constant power when exceeding a certain speed.
Tim Is that really the way to do it?
Mag Anyway, this is our model assumption. If this were a research project,
we would have to compare the curves calculated according to the model with
measured curves, and probably also consider that the gear would be changed at a
certain speed.
In our model, we distinguish three cases:
– v · pv ≤ Pmin P = Pmin , clutch slides

– v · pv ≥ Pmax P = Pmax , engine at power limit
– Pmin < v · pv ≤ Pmax P = v · pv , speed - proportional power
The time curves velocity v(t) and acceleration a(t) are shown in Fig. 7.10b. The dif-
ferences from Fig. 7.9b (and Fig. 7.10a) are clearly visible. At the beginning, with
sliding clutch, the acceleration remains approximately constant, and then decreases
linearly with time.
Question
What is the physical unit of pV ?13
13 [pV ] = [P/v] = Ws/m = J/m.

m mass of the car

μ friction coefficient for air resistance
nF power in friction law
P default constant power
Pmin power when the clutch slides
Pmax engine at the power limit
pV coefficient for speed-proportional power
aMax maximum acceleration to avoid sliding
t sequence of times, dt apart, independent variable
Pvar varying power as a function of t
a acceleration at t
v velocity at t
vS stationary velocity
x distance travelled since t = 0
7.5.3 Excel
Constant power
For the numerical calculation, we use our standard method, “Progress with look-
ahead” (see Fig. 7.11 (S)). The condition “driving force < static friction force” is taken
into account by limiting the acceleration using a min function, a = Min(a(t); a Max ),
where, for a(t), Eq. 7.16 is to be used. The corresponding spreadsheet formula, valid
for C6:C235, is reported in C4.
We do not set the velocity at time t = 0 in B6 to zero, but rather to 0.00001,
so that, in C6, there is no division by 0. The value calculated in column C is
A B C D E F G H
1 dt m µ n P aMax
2 0.35 1500 0.5 2 100000 5.00
3 <-- v; μ=0.5 a -->
)
ax) aMax
d t *v ^n;aM m *vD^n;
* m /
E6)/2 /v-μ/ /vD-μ
dt (C6+ MIN(P/m v+a*dt MIN(P/m
4 =A6+ =B6+ = = =
5 t v a vD aD <-- v; μ=0
6 0.00 0.00001 5.00 1.75 5.00 0.00001
7 0.35 1.75 5.00 3.50 5.00 1.75
235 80.15 58.25 0.01 58.26 0.02 102.09
Fig. 7.11 (S) Acceleration and velocity are calculated using the method “Progress with look-
ahead”. The parameters of the problem are defined in row 2 with their names in row 1. The values
in column G were copied from column B when μ was set to 0
thereby not affected, because, from the min condition, the maximum permissible
acceleration aMax results anyway.
Remember: In the calculation of the velocity v in a row (time t), only values
from the previous row (time t−dt) are used. Therefore, the formula in B7 (reported
in B4) gets input only from B6, C6, and E6. Formulas should only be entered into
the cells in bold, here in A7 and B7. The rest of the respective columns is obtained
by copying (“dragging down”). The initial values in A6 and B6 are entered as
numbers. The formulas reported in C4:E4 are valid in their entire columns, e.g.,
C6:C235, because they do not refer to cell addresses, but rather to names for
variables (cells and column ranges).
Questions

When is the acceleration in column C of constant?14
Examine and describe the expressions for a and aD !15
Another technical note for the spreadsheet calculation: 10/5/2 = 1. So, the
formula is divided consecutively: 10/5 = 2, then 2/2 = 1. According to the rules
of fractional calculation, the formula could be interpreted as 10/(5/2) = 4, which
is not so in excel.
Task Determine time and distance as a function of the power needed to accel-
erate to 95% of the final speed! To calculate the distance, you must integrate the
velocity.
Temporally fluctuating power with the Mod function

How do the curves a(t) and v(t) change when the power does not remain constant,
but rather fluctuates over time?
When the power fluctuates over time, we cannot treat it as a global constant P,
but must insert an extra column in the spreadsheet, Pvar , into which the power can
be entered at any time. To calculate the acceleration, we must access this variable,
and not the global parameter P.
In Fig. 7.12 (S), a calculation is presented in which the power fluctuates over a
longer period of time, i.e., remains constant over a time interval t (DelT). The two
variables r T and Pvar have been inserted into columns B and C for this purpose.
The formulas for v, a, vp and ap are exactly the same as in Fig. 7.11 (S), but with
the time-dependent power PVar instead of the constant power P.
14 As long as the acceleration calculated in column C according to Eq. 7.16 is greater than the value
aMax given in F2.
15 Think about it and read on!
1 dt DelT P DelP m µ n aMax
2 0.35 3.4 100000 10000 1500 0.5 2 5.00 <-- v;μ=0.5
3 a; DelT=3,5 -->
6)
elP);C
D();P;D x) n;aM
ax)
R AN ;aMa *vD^
lT)) RM.INV( *dt μ/ m *v^n -μ/ m
O - D
(t;De 0.2;N G6)/2 r/m/v r /m /v
dt (MOD IF(rT <= D6+(E6+ MIN(Pva v+a*dt MIN(Pva
4 =A6+ =ABS = = = = = a; DelT=3,5
5 t rT Pvar v a vD aD t.R v.R a.R
6 0.00 0.00 115530 0.00001 5.00 1.75 5.00 0.00 0.00 5.00
7 0.35 0.35 115530 1.75 5.00 3.50 5.00 1.75 8.75 5.00
15 3.15 3.15 115530 15.72 4.82 17.40 4.32 15.75 40.17 1.23
16 3.50 0.10 125449 17.32 4.73 18.97 4.29 17.50 42.20 0.98
235 80.15 1.95 104604 58.13 0.07 58.16 0.07 71.75 56.94 -0.11
Fig. 7.12 (S) The calculation of Fig. 7.11 (S) is extended with columns B and C, in which a time-
dependent power is calculated, remaining constant over the time period DelT. Columns I, J and K
are written by sub Vari in Fig. 7.13 (P) with a selection of the data from columns A, B and C
In our model, we require that the power remain constant over a period of time
DelT. This condition can be fulfilled with the mod function (performing a modulo
operation) , whose action is visible in column B of Fig. 7.12 (S). The time t is
divided by DelT and the remainder is returned; e.g., in row 16 t/DelT = 3.5/3.4
= 0.1, remainder (modulus) is 0.1, and 0.1 is returned.
The expression in C4 of Fig. 7.12 (S), valid for C7, outputs a new value for
Pvar only at certain times, otherwise, the value from the previous time is carried
over. It is structured as follows:
if (rT ≤ 0.2) then (Pvar new) else (old value from the previous cell C6)
with Pvar new = norm.inv(rand (); P; DelP), setting a new power,
fluctuating around the mean value P with a standard deviation DelP.
In columns I, J, and K, time t R , velocity vR, and acceleration aR are entered suc-
cessively for five start processes using a macro (protocol procedure sub Vari in
Fig. 7.13 (P)): The curves are graphically represented in Fig. 7.10a. In every run,
Fig. 7.13 (P) Rep-log procedure, copies five start operations (for rep = 1 to 5) continuously
into columns 9 to 11 (= I, J, K) of Fig. 7.12 (S). Only the data for every fifth time are transferred
(line 11, … step 5)
Fig. 7.14 (S) Speed-proportional power Pv = v· pV if it is bigger than Pmin and smaller than Pmax
only every fifth data point (“Step 5”) is transmitted.
Questions
What would the curves of v in Fig. 7.10a look like if, in Fig. 7.13 (P), line 17
were missing?16
How does delP appear as a standard deviation in the fluctuating power in
Fig. 7.12 (S)?17
To get a power that is proportional to the speed of the car, Fig. 7.14 (S), we replace
the formula in C4 of Fig. 7.11 (S) , valid for C6:C235, with a nested if-loop with the
outputs v·pV , PMax , PMin :
[= IF(v∗ pv > Pmin;

(IF(v∗ pv < Pmax; v∗ pV;
Pmax));
Pmin)]
The power is v·pV , i.e., proportional to the speed, if it is bigger than Pmin and
smaller than Pmax .
16 There would be no empty line separating the data sets, so that the last point of a curve would be
connected to the first point of the following curve.
17 See formula in C4 where a formula for normal noise is applied.
Table 7.9 Parameter specification for driving with constant power

1 dt=0.35
2 m=1500 #[kg] Mass of the car
3 mu=0.5 #Friction coefficient
4 nF=2 #Power coefficient in friction law
5 P=100000 #[W] Power
6 aMax=5.0 #[m/s²] max. acceleration
7
8 def accPwr(x,v): #v is always positive
9 a=min(P/m/v-mu/m*v**nF,aMax)
10 return a
11
12 t=np.arange(0,80.15+dt,dt)
13 x=np.zeros(len(t))
14 v=np.zeros(len(t))
15 a=np.zeros(len(t))
Table 7.10 Progress loop with RungeKutta for constant power

16 v[0]=0.00001
17 acc=accPwr #acc is name for acceler. within RungeKutta
18 for i in range(len(t)-1):
19 a[i]=accPwr(x[i],v[i])
20 x[i+1],v[i+1]=RungeKutta(x[i],v[i])
7.5.4 Python
Constant power
The setting of the parameters for driving with constant power is shown in Table
7.9. The size of the arrays is specified with np.arange(0,80.15 + dt,dt)
to yield exactly the same range as in the excel spreadsheet of Fig. 7.11 (S). The
acceleration is executed in a function accPwr that is assigned the additional name
acc (line 17 in Table 7.10) expected in RungeKutta.
Progress of motion is achieved in Table 7.10 with the usual progress loop.
Question
Why do we calculate a[i] separately in line 19 of Table 7.10, although it is

already calculated within Runge-Kutta?18
18 We want to plot a(t), e.g., in Figs. 7.9 and 7.10.

Fluctuating power
To calculate the car’s velocity for fluctuating power, the additional parameters DelT,
DelP, and Pc are specified in Table 7.11. To store the results of five different runs,
matrices vM and aM are built by stacking five times the arrays v and a, respec-
tively (lines 6 and 7). Their shape is 5 rows × 231 columns. The usual progress-loop
contains an inner loop (k =) running over the 5 rows and calling RungeKutta succes-
sively with the individual vM[k,i] so that, in the end, five time series of velocity
and acceleration will have been calculated.
The contents of aM and vM are completely calculated in Table 7.11 and plotted
with the program in Table 7.12.
Table 7.11 Calculation of five time series of v and a for motion with fluctuating power
2 DelT=3.5 #Fluctuation period
3 DelP=10000 #Fluctuation amplitude
4 Pc=100000 #Mean power
5 P=Pc
6 vM=np.stack((v,v,v,v,v))
7 aM=np.stack((a,a,a,a,a))
9 for k in range(5):
10 aM[k,i]=accPwr(x[i],vM[k,i])*10
11 PBef=P #Setting the (new?) power
12 rem=np.mod(t[i],DelT) #0 ≤ rem ≤ DelT
13 P=Pc+DelP*npr.randn() if rem<=DelT else PBef
14 x[i+1],vM[k,i+1]=RungeKutta(x[i],vM[k,i])
np.shape(vM) (5, 231)
np.shape(vM[0]) (231,)
Table 7.12 Plotting five time series of velocity and acceleration for motion with fluctuating power
1 FigStd('t',0,80,20,'v',0,60,20)
2 print('np.shape(t) ',np.shape(t))
3 print('np.shape(vM[0]) ',np.shape(vM[0]))
4 for k in range(5):
5 plt.plot(t,vM[k],'k-') #Full black
6 plt.plot(t,aM[k],'k:') #Dotted black
np.shape(t) (231,)
np.shape(vM[0]) (231,)
Table 7.13 Definitions for speed-proportional power

1 Pv=np.zeros(len(t))
2 Pmin=1000 #Min. power
3 Pmax=100000 #Max. power
4 pv=2000 #Pn=v*pv, proportionality factor
5 aMax=1.5 #Max. acceleration
6
7 def accPwrV(x,v): #Pn is specified in progress loop
8 a=min(Pn/m/v-mu/m*v**nF,aMax)
9 return a
Table 7.14 Loop for progress with RungeKutta for speed-proportional power
1 v[0]=0.00001
2 acc=accPwrV #acc is name in RungeKutta
4 if (Pmin<v[i]*pv<Pmax): Pn=v[i]*pv
5 elif (v[i]*pv<Pmin): Pn=Pmin
6 else: Pn=Pmax
7 Pv[i]=Pn
8 a[i]=accPwrV(x[i],v[i])
9 x[i+1],v[i+1]=RungeKutta(x[i],v[i])
Questions
How do you change the arguments in line 1 of Table 7.12 to get the same axis
titles as in Fig. 7.10?19
How does the changing power P enter accPwr(...)?20
The additional parameters for driving with speed-proportional power are given in
Table 7.13, together with a function accPwrV(x, v) for calculating the accelera-
tion with the input variables x and v, as is expected for the function acc applied
in RungeKutta. The function accPwrV makes use of global parameters. To keep
control of the situation, the global parameters and the function should be defined
in the same program cell. This is actually the case in Tab. 7.13, except for Pv , the
speed-dependent power that depends on the speed calculated in the progress loop.
The progress loop for speed-proportional power in Table 7.14 contains a nested
if … elif … else query to assign the correct value of the power to the
variable Pn accessed as a global parameter in accPwrV.
19 FigStd(’t [s]’,0,80,20,’v [m/s]’,0,60,20).

20 The power P is a global variable changed within the progress loop.
7.6 Bungee Jump
A bungee jumper falls in free fall until the rope becomes tight; only gravity
acts. When the rope is stretched, two additional forces come into play: the
back-driving rope force and the friction force due to the inner friction of the
rope. We neglect air friction so that the force depends only on the location,
and not on the speed. The form of motion is, at times, a free fall or a damped
vibration.
7.6.1 Simulation of the Motion
In a bungee jump, a person hangs on an elastic rope and lets her/himself fall from
a great height into the depths. As an example, we consider a bungee jumper (mass
m = 60 kg, size 1.65 m) jumping from a height of 50 m. The bungee rope is l
= 25 m long in relaxed condition and has an elastic constant (corresponding to
a spring constant) of k = 100 N/m. The zero of the z-axis is at the point where
the rope is fixed. The jumpers move in free fall until the elastic rope is fully
expanded but not yet stretched. For l = 25 m, this is at z = −25 m.
No friction
We first consider the situation without friction in which the jumper starts at height
z = −25 m (fully elongated rope, not yet stretched; see Fig. √ 7.15a). The jumper
oscillates like a mass-spring system with a period of T = 2π/ k/m = 4.87s or 4·T
= 19.47 s.
(a) (b)
-20 0
z [m]
z [m]
0 5 10 15 t [s] 20 0 5 10 15 t [s] 20
-20
-30
-40
zero rope
zero rope
l25; k100; fR0; m60
l25; k100; fR0; m60 Min (Formula)
-40 -60
Fig. 7.15 a (left) Oscillation when the jumper starts with a fully elongated rope. b (right)
Sequence of free fall and oscillation when the jumper starts at the fixing point of the rope
7.6 Bungee Jump 299
When the jumpers start at the fixing point of the rope at the height of z = 0 m,
Fig. 7.15b, they:
– fall freely until the rope is fully elongated (z = −25 m),

– complete a half-period oscillation driven back by the elasticity of the rope to
the zero-rope position (z = −25 m),
– and overshoot up to the original height z = 0 in free fall (starting upwards
because of the initial upward velocity), where the height is a parabolic function
of time.
This sequence of motions repeats periodically.
Friction force of the rope

We assume a friction work W F proportional to the elongation or relaxation path of
the rope, because, during elongation and relaxation, rope components are shifted
against each other, thus dissipating energy. Consequently, the frictional force f R =
dW F /dh is constant. For the example of Fig. 7.16, we have chosen f R = 220 N.
The motion with unexpanded rope (above z = -l) is still a free fall, whereas,
with expanded rope, a dampened oscillation is observed.
The horizontal lines in Fig. 7.16 are obtained from analytical calculations of:
– the maximum height fallen,

– the top speed of the jumper,
– the rest position at the stretched rope.
(a) (b)
0 20
z [m]
v [m/s]
0 5 10 15 t [s] 20
-20 0
0 5 10 15 t [s] 20
-40 -20
zero rope
v vMax
rest position
vD
l25; k100; fR220; m60
Min (Formula)
-60 -40
Fig. 7.16 a (left) Height as a function of time during a bungee jump, “zero rope” = length of
unexpanded rope. b (right) Velocity as a function of time; vD (dotted line) = look-ahead to end
of the time segment, calculated with the acceleration at the beginning. Parameters specified in
Fig. 7.18 and Table 7.15.
Question
What do you think: is rope friction effective only for elongation or also for
relaxation of the rope?21
Reverse problem: The properties of the rope are deduced from the observed
motion
In this exercise, we have specified values for the spring constant and the friction force
of the rope and therewith calculated the time curves for the position and velocity of
the jumper.
More interesting from a physical point of view is the reversal of the problem, in
which the properties of the rope are deduced from an actually possible observation.
A possible approach is to observe the oscillation period and measure the time it takes
for the jumper to come to rest, and then adjust the parameters of the simulation model
so that the simulation fits the measured results. We could then possibly see whether
our assumptions about the rope’s elastic and frictional properties and our neglect of
air friction are justified.
7.6.2 Analytical Calculations
We calculate the minimum and the rest position, as well as the maximum speed
by setting up the equation of motion with the starting point at z = 0:
m · z̈ = −m · g for z > −l (7.19)
m · z̈ = −m · g − (z − l) · k − f R for z ≤ − l (7.20)
The parameter k is the spring constant coming into play when the jumper’s
position is lower than the length of the unexpanded rope.
First minimum by energy balance

The lowest point of the jump can be calculated from the energy balance. The kinetic
energy disappears at the lowest point, being a reversal point; the gravitational energy
has passed into the elastic energy of the rope and the friction energy. With h = height
fallen, the following quadratic equation applies:
(h − l)2 k
m·g·h = + f R (h − l) (7.21)
2
21 Elongation: gravitational energy is converted into kinetic, elastic, and friction energy.
Relaxation (rope becomes shorter): elastic and kinetic energy is converted into gravitational and
frictional energy.
7.6 Bungee Jump 301

2 fR 2mg 2l
h 2 + h · −2l + − + l2 − f R =0 (7.22)
k k k
It is solved in Fig. 7.17 (S), Eq. A, for the parameters of Fig. 7.18.
Maximum speed of the Bungee jumper

The maximum speed of the jump is obtained by considering the dependence of the
kinetic energy on the height of the fall:
m 2 k
E kin = v = mgh − (h − l)2 − f R (h − l) (7.23)
2 2
The gravitational energy goes into the kinetic energy of the jumper, the elastic
energy and the friction work of the rope.
The derivative of the kinetic energy with respect to the height gives the position
of the maximum velocity:
d E kin
= mg − k(h − l) + f R = 0 (7.24)
dh
mg + kl + f R
h= (7.25)
k
The maximum speed is calculated from the kinetic energy T max at this altitude:

2Tmax
vmax = (7.26)
m
The value of vmax is calculated in Fig. 7.17 (S), Eq. (B), for the parameters of
Fig. 7.16, specified in Fig. 7.18 (S).
Eq. (A) Terms of the p -57.37 =-2*l+2*fR/k-2*m*g/k

quadratic equation q 515.0 =l^2-fR*2*l/k
h+ -11.14 =p/2+SQRT((p/2)^2-q)
Deepest point h- -46.23 =p/2-SQRT((p/2)^2-q)
Eq. (B) Hight for max. speed hVm 33.09 =(m*g+k*l+fR)/k

Max. kinetic energy Tmax 14426 =m*g*hVm-k/2*(hVm-l)^2-fR*(hVm-l)
Max. speed vMax 21.93 =SQRT(2*Tmax/m)
Eq. (C) Rest position hRest 30.89 =m*g/k+l
Fig. 7.17 (S) Analytical calculation of the lowest point, the maximum speed, and the rest position
of the bungee jump; p is the first parenthesis of Eq. 7.22, q the second. Eq. (A) = Eq. 7.22, Eq. (B)
= Eq. 7.25 and Eq. 7.26, Eq. (C) = Eq. 7.28
Final position of the bungee jumper

When the jumper has come to rest, there is a balance between the weight force and
the elastic rope force:
m · g = k · (h − l) (7.27)
m·g
h= +l (7.28)
k
calculated in Eq. (C) of Fig. 7.17 for the parameters of Fig. 7.18 (S).
Task Determine the time until the jumper is at rest as a function of jump height!
Task Vary k and F R and observe how the number of oscillations and the time
until rest change!
g gravitational constant 9.81 m/s2

l length of the rope
k elastic constant of the rope
fR friction force of the rope, constant, energy = f R ·(h – l)
m mass of the jumper
t sequence of times, dt apart
z position of the jumper at t, rope fixed at z = 0
zMin minimum position of the jumper
zEnd final rest position of the jumper
v velocity of the jumper as a function of t
vMax maximum speed
T Max maximum kinetic energy.
7.6.4 Excel
The parameters m, l, k, f R, and g (= 9.81 m/s2 ) are specified in named cells

in A1:F2 of Fig. 7.18 (S). H3 displays the formula (concatenation of text and
variables) for obtaining the legend in H2.
The equation of motion of the bungee jump is integrated with our standard
method, “Progress with look-ahead”. The result of the calculation is shown in
Fig. 7.16.
7.6 Bungee Jump 303
1 dt g l k m fR ;
2 0.05 9.81 25 100 60 220 l25; k100; fR220; m60
3 s m/s² m N/m kg N =C1&l&H1&D1&k&H1&"fR"&fR&H1&"m"&m
4
5 a=IF(z<-l;-g-SIGN(v)*fR/m-k/m*(z+l);-g)
6 aD=IF(zD<-l;-g-k/m*(zD+l)-SIGN(vD)*fR/m;-g)
*dt *dt t
G9)/2B9+E9)/2 )/2*d 01)
dt (D9+ ( *dt z+(v+vD (v) (z) EX(z;4
7 =A9+ =B9+ =C9+ =v+a = =MIN =MIN =IND
8 t v z a vD zD aD vMax zMin zEnd
9 0.00 0.00 0.00 -9.81 -0.49 -0.01 -9.81 Simu -22.70 -46.28 -30.95
10 0.05 -0.49 -0.01 -9.81 -0.98 -0.05 -9.81 Formula -21.93 -46.23 -30.89
409 20.0 0.15 -30.9 -3.56 -0.03 -30.9 3.77
Fig. 7.18 (S) Calculation model for a bungee jump; l = length of the rope; k = spring constant;
m = mass of the jumper; f R = friction constant of the rope; acceleration is inhibited by friction
only when the rope is stretched; the formulas a = if(…) in row 5 and aD = if(…) in row 6 are
valid for the column vectors a and aD , respectively. Bewegung wird gehemmt
Questions
Analyze the formula “a = … ” in B5 of Fig. 7.18 (S) applied in column D.

What is the logical structure of the formula? When does only gravity act?22
Which forms of the time trajectory do you observe for an unstretched and a
stretched rope?23
The spreadsheet set-up presented here applies the same friction force for relax-
ing and stretching rope. Friction losses are likely to occur only when the rope is
longer than in the unstretched state (discuss!); this is taken into account through
an if query.
The calculated height is shown in Fig. 7.16a as a function of time. For check-
ing purposes, the prominent points of the jump calculated analytically, namely,
maximum depth (= minimum height) and rest position, are entered into the dia-
gram as horizontal lines. The agreement with the simulated curve is quite good,
which indicates that neither our analytical calculations nor our simulation within
the framework of our model contain gross errors;-) or, less likely, both errors
cancel each other out.
In Fig. 7.16b, the velocity is plotted as a function of time, and also, as a hor-
izontal line, the maximum velocity analytically calculated according to Eq. 7.26.
The dashed curve represents the velocity foreseen at the end of the interval calcu-
lated with the acceleration at the beginning of the interval. The full curve shows
the velocity at the beginning of the next interval, calculated from the mean of the
two accelerations, the one at the beginning of the interval and the estimated one
22 if(logical_test, [value_if_true], [value_if_false]). If the amount of the deflection is

smaller than the rope length, then only the acceleration due to gravity acts; neither the elastic rope
force nor the friction force are active. In the if function, this is the [value_if_false] case, i.e.,
when the condition (z < −l) is not fulfilled.
23 Unstretched: free fall, parabola; stretched: damped oscillation around the rest position.
Table 7.15 Parameters and function for calculating the acceleration during a bungee jump
1 g=9.81 #[m/s²]
2 l=25 #[m] Length of rope
3 k=100 #[N/m] Spring constant
4 m=60 #[kg] Mass of jumper
5 Fr=220 #[N] Frictional force
6
7 def accBungee(z,v): #v positive, negative, or 0
8 if z<-l: #Tensioned rope
9 a=-g-k/m*(z+l)-np.sign(v)*Fr/m #np.sign(0)==0
10 else: a=-g
11 return a
12
13 lbl_1=("l"+str(l)+"; k"+str(k)+"; fR"+str(Fr)+"; m"+str(m))
lbl_1 l25; k100; fR220; m60
Table 7.16 Progress of the bungee jump

14 dt=0.251 #[s] Time increment
16 v=np.zeros(len(t)) #Velocity
17 z=np.zeros(len(t)) #Height
18 acc=accBungee #acc(…) is accessed in RungeKutta
19 v[0]=0 #Initial conditions
20 z[0]=0
22 z[i+1],v[i+1]=RungeKutta(z[i],v[i])
at the end of the interval. The difference between the two curves illustrates the
difference between the Euler and the lookAhead methods.
7.6.5 Python
The parameters of the bungee jump and the function accBungee for calculating
the acceleration of the jumper are given in Table 7.15. The progress of the motion
of the jumper is calculated in Table 7.16.
From integration to numerical averaging
1. How is the velocity v(t) related to the acceleration a(t)?

2. How is the mean value of a continuous function f (t) within the range t 1 to t 2
defined?
3. Interpret the formula v(tn + dt) = v(tn )+[a(tn ) + a(tn + dt)]/2·dt with respect
to the previous two questions!
4. What does the broom rule (Ψ half, half, whole, the halves count twice) tell us
about the 4th order Runge-Kutta method?
Half-step procedure in a spreadsheet

In Fig. 7.19 (S), you can see a spreadsheet structure for the numerical integration
of Newton’s equation of motion for a fall with friction using the half-step method.
The spreadsheet calculation is structured in the same way as for our "Progress with
Forecast" procedure. However, the predicted acceleration aP is calculated for the
middle of the interval and is considered representative of the entire time interval.
5. What are the formulas for x P , vP, and aP ?

6. What are the formulas for x and v at the beginning of the next time interval (in
C7 and D7)?
Function for “Progress with look-ahead” or “half step”

In Table 7.17, you can see the code of a Python function for calculating the progress
from t to t + dt. However, in lines 4 to 8, the formulas have been replaced by 1.
7. Insert the appropriate formulas if “Progress with look-ahead” is to be used!
B C D E F G H I J
1 g 9.81 m/s² gravitational acceleration
2 cV 1.5 1/m friction coefficient
3 dt 0.005 s time increment
)
N(v) *dt^
2 N(vP
*d t 6*dt v^ 2*SIGdt/2+a/8 dt/2 vP^2*SIG
dt G6 H V* * * V*
4 =B6+ =C6+ =D6+ =-g-c =x+v =v+a =-g-c
5 t x v a xP vP aP
6 0.000 4.000 1.000 -11.31 4.00 0.97 -11.23 4.000
7 0.005 4.005 0.944 -11.15 4.01 0.89 -10.99 4.005
406 2.000 -0.361 -2.557 0.00 -0.37 -2.56 0.00 -13.620
Fig. 7.19 (S) Falling with friction, half-step procedure: The velocity vP and the acceleration aP
in the middle of the interval are calculated. Formula in C4 is valid for C7
Table 7.17 Function for numerical integration of the Newtonian equation of motion
1 dt=0.1
2 def ForcA(x,v): 6 aA=1
3 a=acc(x,v) 7 xN=1
4 xA=1 8 vN=1
5 vA=1 9 return(xN,vN)
8. Insert the appropriate formulas if the half-step method is to be used!
Power and work
9. How are work W and power P related to the force F, the body’s displacement x,
and the velocity v?
Bungee jumping
10. What are the formulas for: a) the elastic energy of a spring with constant k, b)
the gravitational energy of a mass m near the earth’s surface, and c) the kinetic
energy of a mass m?
Consider the statement: “The frictional energy when stretching or relaxing a rope is
proportional to the change in length (absolute value) of the rope.”
11. How is this approach justified?

12. Which function results for the frictional force?
Mathematical pendulum
The oscillation equation for a mass-spring system is
∂ 2 x(t) k
= − · x(t) (7.29)
∂t 2 m
The oscillation equation for a mathematical pendulum is
∂ 2 φ(t) g
= − · sin(φ) (7.30)
∂t 2 l
13. Why does no mass appear in Eq. 7.30 as does in Eq. 7.29?
14. Why does sin(φ) turn up on the right side of Eq. 7.30, and not simply φ, as one
might expect in analogy to Eq. 7.29?
Random Numbers and Statistical
Reasoning 8
We perform statistical experiments by applying functions that generate

uniformly distributed, normally distributed, and cos2 distributed random
numbers. We set up frequency distributions of a set of random num-
bers and compare them quantitatively with model distributions by means
of the Chi2 test, the interpretation of which will be explored. (Random-
number generators are used in later chapters to simulate measurement
inaccuracies and noise, e.g., in Chap. 9 (Evaluation of measurements)
and Chap. 10 (Trend lines).) Required spreadsheet functions are: RAND(),
FREQUENCY() as a matrix function, and CHISQ.TEST(), as well as the logi-
cal functions AND and OR, together with their counterparts in the Python
libraries numpy.random and scipy.stats.
8.1 Introduction: Statistical Experiments Instead

of Theoretical Derivations
be found at the internet address: go.sn.pub/zHV7Ko.
Russian proverb: Dovep, no ppovep! Trust but verify!
German variant: Vertrauen ist gut, Kontrolle ist besser. Trust is good, control is
even better.
Ψ If in doubt, count!
Ψ Mostly, not always. “Fundamental rule of statistical reasoning”.
Random-number generators
In Python, two libraries have to be imported to make statistical functions available:

308 8 Random Numbers and Statistical Reasoning
numpy.random (npr) for the functions rand (random numbers between 0

and 1) and randn (normally distributed random numbers).
scipy.stats (sct)for the functions norm(0,1) (normal distribution)
and chisquare (Chi2 test).
The spreadsheet function RAND and the Python function npr.rand return ran-
dom numbers x that are equally distributed between 0 and 1 (0 ≤ x < 1). Other
distributions are obtained when these random numbers are redistributed with suitable
functions, namely, with the inverse of the distribution function of the desired proba-
bility density. For the Gaussian, exponential and Cauchy-Lorentz distributions, these
inverse functions are available as NORM.INV, LN and TAN in EXCEL, and sct.norm
(0,1).ppf, np.log, and np.tan in Python.
The distribution function can be approximated with a polyline in a user-defined
function. This is important for cases in which no predefined function for the inverse of
a cumulated distribution function exists. As an example, we calculate the diffraction
image of photons that have passed through a double-slit where they are distributed
along a line perpendicular to the slits in cos2 -shaped maxima.
What is to be learned?
After having worked through this chapter, you should be able to handle the following
safely:
– generating a set of random numbers that obey a given distribution model,

– determining the frequencies of occurrence of a data set, and displaying them
graphically,
– quantitatively comparing observed frequency distributions with model distribu-
tions. To this end, we practice the cautious use of the Chi2 test with the correct
degree of freedom dof .
Statistical experiments instead of theoretical derivations
Alac Numbers, numbers, numbers. They’re quite tiring. Is it worth the effort?
Mag Yes. This is the basis of the evaluation of experiments.
Alac Evaluation of Lab-course experiments? I’m happy when I have the annoy-
ing protocols behind me. I just want to memorize a few answers from the
introduction to the exercises to get through the exam talks scot-free.
Mag We only practice what is absolutely necessary—but understanding that

well is quite important, not only for studying physics, but for all empirical sciences
and political statements.
Alac Understand it well? Do we have to reproduce mathematical proofs?

8.1 Introduction: Statistical Experiments Instead … 309
Mag No. We perform experiments to grasp the essence of certain theorems of

statistics.
Tim “Statistical experiments”, that sounds interesting. Nevertheless, probability

theory is not my thing. Nobody in my study group understands the Chi2 test yet.
Mag I can understand that. Appearances are often deceptive, and no statement
is a hundred percent certain. Nevertheless, you have to be able to move on this
unsteady ground. In this chapter, we will start with gait exercises.
We use two types of statistical test in which a random experiment is repeated many
times:
– multiple tests for equal distribution and

– multiple tests for error probability.
In Exercise 8.2, multiple tests for equal distribution will be used to check whether
the results of Chi2 tests are equally distributed, as it should be when the model
distribution corresponds to the population from which the samples are taken.
Multiple tests for error probability are used in Chap. 9, “Evaluation of measure-
ments”, and Chap. 10, “Fitting of trend curves to measurement points”, to check
whether the experimental confidence intervals of a measurement result match the
assumed confidence levels.
Statistical functions in Python and Excel

import numpy.random as npr
import scipy.stats as sct.
Functions concerning the normal distribution:
sct.norm(xm ,xd ).pdf(x) NORM.DIST (x;xm ;xd ; FALSE)

−∞ < x < ∞, probability density function
sct.norm(xm ,xd ).cdf(x) NORM.DIST(x;xm ;xd ; TRUE)
–∞ < x < ∞, cumulative density function (distribution function)
sct.norm (xm , xd ).ppf(p) NORM.INV(p; xm ; xd )
0 ≤ p ≤ 1, percent point function, inverse of cdf .
Functions generating random numbers:
npr.rand(n) RAND() (matrix function)

random numbers equi-distributed between [0 and 1)
npr.randn(n) NORM.INV(RAND();0;1) (matrix function)
normally distributed
RANDBETWEEN (no matrix function)
between two given integers,
npr.choice(2, 100, p = [0.2, 0.8])
chooses numbers from a list with specified probabilities, here, 100 numbers 0 or
1
npr.choice (10, 10, replace = False)
without replacement; every number occurs only once.
Nomenclature
pdf probability density function

cdf cumulative density function (distribution function)
ppf percent point function, inverse of cdf
dof degrees of freedom, by default number of measurements minus 1
ddof delta degrees of freedom, 0 by default, greater than 0 if parameters of
the model distribution other than the sample mean are estimated from the
empirical distribution of the frequencies of occurrence.
8.2 Equi-Distributed Random Numbers, Frequencies

of Occurrence, Chi2 Test
We generate numbers randomly between 0 and 1, determine their frequency

of occurrence in intervals of width 0.1, and check, with the Chi2 test, whether
they are equally distributed.
8.2.1 A Spreadsheet Experiment with Random Numbers
In this exercise, we shall perform experiments with random numbers. They will
be explained by means of the spreadsheet in Fig. 8.1, where the data structure is
clearly laid out, and repeated in Sect. 8.2.3 in Python.
(1) We generate 1000 random numbers named Rnd in column A, equally

distributed between 0 and 1, with the spreadsheet function RAND().
(2) We determine the empirical frequency distribution FrqObs of these numbers in
intervals with 11 specified boundaries I b = 0.0, 0.1, …., 0.9, 1.0 in column C.
The 10 intervals between 0 and 1 all have the same width of 0.1. Principally,
however, the intervals may be of different widths.
(3) We perform a Chi2 test to check how close the observed frequencies FrqObs
are to FrqXpt, the frequencies expected for an equi-distribution. This is done in
Range H6:I9. In H9, the Chi2 test for the equi-distribution. It yields the same
value as performed with the function ChiSq.Test on FrqObs and FrqXpt. In H8,
this is performed via the value of ChiSqr in I6 and the function CHISQ.DIST.RT.
8.2 Equi-Distributed Random Numbers, Frequencies of Occurrence, Chi2 Test 311
)} rqXpt
nd;Ib )^2/F
() U EN CY(R
bs-FrqXpt
2 =RAND EQ O
{=FR =(Frq
3 Rnd Ib FrqObs FrqXpt CSq FrqStep
4 0.38 0 0
5 0.57 0.1 111 100 1.21 dof ChiSqr 90
6 0.80 0.2 92 100 0.64 9 10.10 =SUM(CSq) 90
7 0.34 0.3 90 100 1 90
8 0.72 0.4 85 100 2.25 0.34 =CHISQ.DIST.RT(ChiSqr;dof) 90
9 0.35 0.5 98 100 0.04 0.34 =CHISQ.TEST(FrqObs;FrqXpt) 90
10 0.95 0.6 117 100 2.89 110
11 0.44 0.7 102 100 0.04 0.30 =CHISQ.TEST(FrqObs;Frqstep) 110
12 0.04 0.8 97 100 0.09 110
13 0.54 0.9 95 100 0.25 110
14 0.05 1 113 100 1.69 110
15 0.04 0
1003 0.69
Fig. 8.1 (S) In A4:A1003, named Rnd, 1000 random numbers are generated. In D4:D15, the
spreadsheet function FREQUENCY is used to count how many of these random numbers fall into
intervals whose limits I b are specified in C4:C14. Column E shows the frequencies expected for
equal distribution. In H8, the Chi2 test with FrqXpt is performed. In H11, the Chi2 test is performed
against a step function FrqStep in column K, dof (in H6) = degrees of freedom
Empirical frequency distribution

In EXCEL, empirical frequencies of occurrence are determined with the matrix func-
tion FREQUENCY(DATA_ARRAY; BINS_ARRAY). It outputs 12 frequencies (in D4:D15 in
Fig. 8.1 (S)) for the 11 boundaries I b (BINS_ARRAY), because it also determines the
frequencies of the data below the lowest and above the highest boundaries.
The first value in BINS_ARRAY gives the number of occurrences in the examined
data set that lie below the first interval boundary, that is, from −∞ on. The last
value specifies the number of occurrences that lie above the last interval limit, i.e.,
up to ∞.
Ψ Always one more in Excel! O.k., but of what and than what?1
The spreadsheet function FREQUENCY is a matrix function. Before it is entered,

a spreadsheet range must be activated that can capture all output data, and the
operation must be concluded with a magic chord:-) , Ψ (Ctl + Shift) + Enter.
Read the box to get an answer to the important question: “What is a matrix
function?”
What is a matrix function in Excel?
1For the output of the matrix function Frequency, a range has to be activated comprising one cell
more than the number of interval boundaries.
FREQUENCY is a matrix function. In terms of spreadsheet calculation, this

means that a cell range must be selected that is large enough to contain
the amount of the returned data. In the spreadsheet of Fig. 8.1 (S), the
area D4:D15 was marked, then the function was entered (see D2), and
the process was completed with the “magic chord” for matrix functions:
(CTL + SHIFT) + ENTER.
Questions
What is the formula in A4:A1003?2

What set of numbers is sorted into the intervals in Fig. 8.1 (S).3
The Chi2 test as a judge

If the random numbers are evenly distributed between 0 and 1, we may expect that,
below 0 and above 1, there will be no values, and that in each of the ten intervals of
width 0.1, 100 values will occur. This is specified in the variables FreqXpt. A Chi2
test can be used to check how well the observed distribution, here, FrqObs, and the
expected one, here, FrqXpt, match.
To this end, we calculate the value of χ 2 (Chi2 , “Chi square”) defined as
N (Oi − E i )2
χ2 = (8.1)
i=1 Ei
with Oi and E i being the observed and expected frequencies, respectively, and N
the number of intervals considered for the comparison. In our implementations, Oi
= FreqObs = [D5:D14], excluding the intervals with zero occurrence, and E i =
FreqExp = [E5:E14] (only non-zero numbers), with the same size as FreqObs. χ 2
is termed ChiSqr and calculated in I6 as the sum over the individual terms CSq in
column F.
The Chi2 test is listed in the literature under various names, e.g., χ 2 -Test,
Chitest. We call it the Chi2 test, because it is based on a distribution of the quantity
χ 2 = chi2 (“chi squared”), defined in Eq. 8.1.
We perform such a Chi2 test in two ways:
– by introducing the calculated value of ChiSqr into the distribution function

CHISQ.DIST.RT (in H8), and
– by applying a special function for that test, CHISQ.TEST (in H9).
2 = RAND().
3 The 1000 numbers in A4:A1003.
The values of such tests are between 0 and 1, giving the probability that χ 2 of
the current sample fits better with the model distribution of the whole population
than any other sample. If the Chi2 test results in values significantly smaller than
0.01, it is generally concluded that observation and expectation do not match, i.e.,
the sample is not from the assumed model population. The values in H8 and H9
are 0.34, so that there is no reason to doubt that the random numbers are equally
distributed.
The result 0.34 means that, for 34% of other samples of the same population,
χ 2 will be greater, i.e., the frequency distribution of these samples fits worse with
the theoretical one. A more precise explanation is given in Exercise 8.8.
In H8, the Chi2 test is performed by explicitly entering the value of ChiSqr and
the degrees of freedom dof into the inverse function of the integral distribution
function of Chi2 . In EXCEL, we use CHISQ.DIST.RT(CHISQR,DOF). The suffix.RT means
that a right-tailed distribution function is to be used. The left-tailed distribution
function is addressed with CHISQ.DIST. The meaning of these terms will become
clearer in Exercise 8.8.
DOF is the degree of freedom. For our test, dof = 9, corresponding to the 10
intervals in which the frequencies are compared minus 1; minus 1 because the
frequency in the tenth interval is no longer independent of the other frequencies,
but is rather determined by the total number of events.
Questions
FrqObs in Fig. 8.1 (S) is defined for 12 intervals. Why is the degree of freedom
for the Chi2 test not 11, but rather 9?4
The spreadsheet function ChiSq.Test

EXCEL provides the spreadsheet function CHISQ.TEST(ACTUAL_RANGE, EXPECTED_RANGE)
for the Chi2 test. The test is employed in H9 of Fig. 8.1 (S) and results in the same
value p = 0.34 as the first test. This function calculates the value of ChiSqr internally
and assumes a degree of freedom one less than the number of intervals. It must not
be applied when parameters of the expected distribution are calculated from the
data set.
To repeat: The spreadsheet function CHISQ.TEST may only be used if the parame-
ters of the model distributions are defined in advance, as is the case in our example,
and not estimated from the sample.
Does a step distribution also fit?

In addition to comparing our observed frequencies with an equi-distribution, we also
compare them with the step distribution FrqStep in column K of Fig. 8.1 (S).
4The comparison is only between numbers in the 10 intervals with nonzero value, D5:D14 and
E5:E14; dof = 10 − 1 =9.
Tim Looking at Fig. 8.1 (S), we see that the Chi2 test provides a larger value
(p = 0.34 in H9) for the correct distribution than for the incorrect step distribution
(p = 0.30 in H11).
Alac Well, that’s to be expected. A good reliability check must deliver a lower
probability value for a wrong model distribution than for the right one!
Mag Let’s wait and see! The correct distribution does not always fit best.
Repetition of the statistical experiment

We are going to generate the random numbers anew. To this end, we place the cursor
into an empty cell and “delete” its content. EXCEL reacts as if the calculation had been
changed, generates all 1000 random numbers anew, and recalculates the distribution
FreqObs. The Chi2 test sometimes gives higher values for the comparison with
FrqStep than for the one with FrqXpt.
Within the VBA rep-log procedure in Fig. 8.2b, we copy 1000 results of Chi2
tests into columns 13 and 14. Furthermore, we count how often the step function
fits better with the observed data than the equi-distribution. This holds true in about
10% of the tests.
The frequency distribution of the results of the two times 1000 Chi2 tests is
shown in Fig. 8.2a. The values for the comparison with the equi-distribution scatter
around 100, i.e., they seem to be equi-distributed, whereas the test against the step
distribution delivers mostly values below 0.1. These results are also reflected in
the average values of the Chi2 tests: 0.5 for the equi-distribution and 0.13 for the
step distribution.
Repetitions of statistical experiments are generally not possible in real life, in
which there is mostly only one sample and one result of a Chi2 test. This is, how-
ever, possible in our exercises, because we invent our population ourselves with the
1,000
1 Sub protoc() 1
2 StepBetter = 0 2
800 3 For r = 4 To 1003 3
cEq 4 ChiSqr_Eq = Range("H9") 4
Frequency
600 cSt 5 Cells(r, 13) = ChiSqr_Eq 5

6 ChiSqr_Step = Range("H11") 6
400 7 Cells(r, 14) = ChiSqr_Step 7

8 If ChiSqr_Step > ChiSqr_Eq _ 8
200
9 Then StepBeer = StepBeer + 1 9
10 Next r 10
11 Range("N1") = StepBeer 11
0
12 End Sub 12
0 0.2 0.4 0.6 0.8 1
chi²-test
Fig. 8.2 a (left) Frequency distribution of the results of Chi2 tests against an equi-distribution
(cEq) and a step distribution (cSt). b (P, right) The VBA procedure for 1000 repetitions of the two
Chi2 tests in Fig. 8.1 (S)
Table 8.1 Interval boundaries and theoretical frequencies of occurrence

1 db=0.1
2 Ib=np.arange(0,1+db,db)
3 FrqXpt=np.ones(10)*100 #Equidistribution, expected freqs.
4 FrqStep=np.ones(10) #Step distribution
5 FrqStep[5:]*=110 #Entries 5 to 9
6 FrqStep[:5]*=90 #Entries 0 to 4
function RAND(), meaning we have a didactical method at hand for demonstrating

the peculiarities of statistics.
Rnd 1000 numbers supposedly randomly distributed between 0 and 1

Ib 11 boundaries for 10 intervals (“bins”) between 0 and 1
FrqObs frequencies of occurrence for Rnd
FrqXpc frequencies expected for an equi-distribution
FrqStep frequencies expected for a step distribution
dof Degrees Of Freedom, number of intervals minus 1
ddof Delta Degrees Of Freedom, to be deduced from dof if parameters of
the distribution are estimated from the sample.
8.2.3 Python
Standard histogram in Python

The interval boundaries I b and the expected frequencies of occurrence for an equi-
distribution (FrqXpt) and a step function (FrqStep) are specified in Table 8.1. We
know that there are 10 intervals with numbers greater than zero, so that we specify
FrqXpt and FreqStep with size 10.
Questions
What is the value of FrqStep[5] in Table 8.1?5
In Table 8.2, 1000 Chi2 tests are performed in a for loop on Rnd,
an array with 1000 random numbers created with the Python func-
tion npr.rand (1000). The frequency distribution is obtained with
np.histogram(Data;boundaries). This function returns an array of shape
(2, 10) with the frequencies reported in row number 0 and the boundaries repeated
5 FrqStep[5] is 110. FrqStep[:5] relates to the first 5 elements, FrqStep[0] to FrqStep[4].

Table 8.2 Performing 1000 Chi2 tests of 1000 random numbers Rnd versus equi-distribution
FrqXpc and versus step function FrqStep
7 import scipy.stats as sct
9 N=1000
10 cEq=np.zeros(N) #Chi² values for equi
11 cSt=np.zeros(N) #Chi2 values for step
12 StepBetter=0
13 for i in range(N): #N Chi² tests
14 Rnd=npr.rand(1000) #Sample
15 FrqObs=np.histogram(Rnd,Ib)[0] #Freq. of occurrence
16 ChiSqr_Eq=sct.chisquare(FrqObs[0],FrqXpt,ddof=0)[1]
17 ChiSqr_Step=sct.chisquare(
FrqObs[0],FrqStep,ddof=0)[1]
18 cEq[i]=ChiSqr_Eq
19 cSt[i]=ChiSqr_Step
20 if ChiSqr_Step > ChiSqr_Eq: StepBetter+=1
21 print(„StepBetter „, StepBetter)
StepBetter 57 #Better Chi² test for step distribution
in row number 1. We therefore specifiy the returned variable with index 0:

FrqObs = np.histogram (Rnd,Ib)[0] to get the observed number of
frequencies.
Contrary to the EXCEL function FREQUENCY, only the 10 frequencies in the
intervals between the lowest and highest boundaries are returned, and not the fre-
quencies of occurrence below the lowest and above the highest interval boundaries.
The size of FreqObs[0] is, therefore, one less than the number of the specified
boundaries (Ib, reproduced in FreqObs[1]).In a diagram, we would plot them
over the 10 centers of the intervals with, e.g., plt.plot(Ic, FrqObs).
The Chi2 test is performed with the function sct.chisquare
(FreqObs,FreqExp,ddof).The parameter ddof specifies a reduction
in dof , the degrees of freedom. If no parameter of the theoretical distribution is
calculated from the observed sample, ddof = 0. This is the case for our model
distributions because we estimate only the mean from the sample. So, dof is set
internally to the number of sorting intervals (“bins”) minus 1.
The tests are performed within a for-loop with N = 1000 iterations and their
results are stored in arrays cEq and cSt (lines 20, 21). Plots of cEq and cSt over the
center of the intervals look like Fig. 8.2a. We get qualitatively the same results:
The chi2 values of the “Test of the 1000” versus the equi-distribution (FrqEq) are
equally distributed between 0 and 1, while those vs. the step distribution (FrqStep)
increase strongly for Chi2 test approaching zero.
Ψ Always one more in Excel! O.k., but from what and than what?6
6For the output of the matrix function FREQUENCY, a range has to be activated comprising one cell
more than the number of interval boundaries. Frequencies of occurrence are returned for values
below the lowest interval boundary and for values above the highest boundary.
Ψ Always one less in Python! O.k., but from what and than what?7
All-including intervals
The Python function np.histogram can cover, similarly to the EXCEL func-
tion FREQUENCY, the whole range of real numbers from minus infinite to infinite by
extending the definition of the interval boundaries with [-np.inf, ….,]. This is
recommended, because it allows for checking whether the sum of all frequencies is
equal to the total number of data points considered.
Questions
Regarding Table 8.2: In what percentage of cases does the Chi2 test in our
Python program give a higher value for Step than for Eq?8
How does the code have to be changed if 10,000 random numbers are to be
considered?9
Degree of freedom
What is the degree of freedom for the Chi2 tests in lines 16 and 17 of Table 8.2?
From Fig. 8.1 (S) and Fig. 8.2a (S), we deduce that the degree of freedom is 9. Is it
correct to set ddof = 0? We may not be sure what the description means and check
this question with the program in Table 8.3, a modified version of parts of Tables 8.1
and 8.2. In lines 8 and 9, we sum up the results of Chi2 tests of 10,000 repetitions
of the statistical experiment, in ChiS0 with ddof = 0 and ChiS1 with ddof =
1. The average values in the lower cell show that, with ddof = 0, we are closer
to the theoretical average 0.5 of Chi2 tests than with ddof = 1. So, ddof = 0
seems to be true, so that we may conclude that the default degree of freedom in
sct.chisquare is 9, number of intervals minus 1, dof = n – 1, as for the EXCEL
function CHISQ.TEST.
Consider also the improvement of the code in Table 8.3 over that in Tables 8.1
and 8.2: The number of repetitions can be changed by changing N alone.
7 Numpy’s np.histogram outputs a number of frequencies that is one less than the number of
boundaries; frequencies of occurrence are returned only for internal intervals.
8 Bottom cell of Table 8.2: StepBetter = 57, of 1000 runs, makes about 6%. StepBetter is updated
in line 20 within the for-loop.

9 Line 9: N = 10,000; line 14: npr.rand(10,000), line 16 FrqXpc*10; line 17: FrqStep*10. It would
be better to use the variable N instead of specific numbers.

Table 8.3 Checking the degree of freedom with N = 10,000

1 N=10000
2 FrqXpt=np.ones(10)/10 #Freq. in each of ten bins = FrqXpt*N
3 ChiS0=0
4 ChiS1=0
5 for i in range(N):
6 Rnd=npr.rand(N)
7 FrqObs=np.histogram(Rnd,Ib)[0]
8 ChiS0+=sct.chisquare(FrqObs,FrqXpt*N,ddof=0)[1]
9 ChiS1+=sct.chisquare(FrqObs,FrqXpt*N,ddof=1)[1]
10
11 print("ChiS0 ", np.round(ChiS0/N,3)) #Average value of
12 print("ChiS1 ", np.round(ChiS1/N,3)) # Chi² tests
ChiS0 0.501 #0.50 expected for correct dof
ChiS1 0.429
8.3 Points Randomly Distributed in a Unit Square
We create coordinates of points randomly distributed in a unit square, use

the logical functions AND, OR, NOT to separate points in sub-regions of the
unit square, and illustrate the broom rule: Chance is blind and checkered.
8.3.1 Creation and Distribution of the Points
Chance is blind and checkered

Figure 8.3a displays a sample of 2000 points (x, y), randomly distributed in the unit
square. The cross + indicates the center of the distribution obtained with the mean
(x m , ym ) of the coordinates. The vertical bars (“devi bars”) are at x = x m ± x Sd , with
x Sd being the standard deviation of the sample. The region within the vertical bars is
called V. Figure 8.3b displays the same points, with the addition of horizontal bars
at y = ym ± ySd (y devi bars), and marks points in the inner rectangle, called VH,
with open diamonds. The region within the horizontal bars is called H.
Questions
concerning Fig. 8.3:

Do the points give the impression of being equally distributed?
Do the white spots and the point clusters disturb the impression of equal
distribution?
What fraction of points in Fig. 8.3a lies within the vertical lines? More or
less than 50%?
8.3 Points Randomly Distributed in a Unit Square 319
1.0 1.0
0.8 0.8
y y
0.6 0.6
0.4 0.4
0.2 0.2
0.0 0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
x x
Fig. 8.3 a (left) 2000 points (x, y) are randomly distributed in the unit square; the empirically
determined center is marked by “+”. The vertical lines mark the standard deviation from the center
in the x-direction. b (right) As in a, but with the addition of horizontal lines marking the standard
deviation from the center in the y-direction and the points in the inner rectangle marked with open
diamonds
How many points lie within the inner rectangle vH in Fig. 8.3b? Is it more
or less than 25%?10
Calculate a√theoretical estimate for the number of points in the rectangle
with x Sd = 1/ 12!11
How do you generate the sets of numbers x and y anew, in Python and in
EXCEL?12
Mag Take a look at Fig. 8.3, where the distribution of points is intended to be
random. You need to place another point, but in such a way that the distribution
looks genuinely random. Where would you put it?
Alac In no case would I place it where there are already many points heaped
up, but rather into one of the white areas, where there aren’t any points yet.
Mag You must not do that under any circumstances. Chance is blind and does
not see where it has already struck before.
10 Solution in Fig. 8.5 (S), cell P3, 672/2000 = 0.336.

11 √2 · √2 = 1 (see also Eq. 8.6).
12 12 3
12 EXCEL: Place the cursor in an empty cell and press “Del”, i.e., “delete” the contents of an already
empty cell. Then, all formulas in the spreadsheet are recalculated, including the spreadsheet func-
tion RAND(). -Python:The numpy function np.rand(..) has to be executed in a loop or in a
user-defined function.
Tim Now I’m confused. So that I don’t make any mistakes, I’ll close my eyes
before I set my point.
Alac But then you’re much more likely to hit a spot outside the predefined
rectangle.
Tim So, I’ll keep putting down points until you tell me that I’ve landed a hit
within the rectangle.
Mag This is, in fact, a way of randomly placing points within the target
rectangle.
Tim But using this blind method, even more clusters can arise.
Mag That’s all right. Ψ : Chance is blind and checkered.
Statistical characteristics of a sample

The average value or mean x m of a sample is defined, with the number n of elements
in the sample, as

i xi
xm = (8.2)
n
The variance σ 2 (or var X in the notation that we will later use) of a sample is
calculated with the mean as

(xi − xm )2
σ = var x = i
2
(8.3)
(n − 1)
and therefrom the standard deviation as the square root:

i (x i − xm )2
x Sd = σ = (8.4)
(n − 1)
The number of elements in the denominator of the variance is reduced by 1, (n

– 1), because one parameter, the mean, has been estimated from the sample and not
been fixed a priori.
In mathematical literature, the standard deviation is usually called σ. We use the
term x Sd (or ySd ) instead, because it has the same physical unit as the sample
members x i (yi ).
In EXCEL, the corresponding functions are AVERAGE for the mean, VAR.S for
the variance of a sample with (n-1) in the denominator, and, correspondingly,
STDEV.S for the standard deviation. In Python, the functions are np.mean and
np.std(x, ddof = 1), ddof = 1 being necessary to get the correct denom-
inator (n − 1). When the standard deviation of an entire population is to be
determined, we have to use STDEV in EXCEL and ddof = 0 in the Python
function.
According to statistical theory, for a uniform distribution between 0 and 1, we
get
1
Mean xm = (8.5)
2
1
Standar d deviation x Sd = √ ≈ 0.289 (8.6)
12
x, y random coordinates of points in the unit square

x m , ym mean of x and y
x Sd , ySd standard deviation of x and y
inX true if x m – x Sd ≤ x < x m + x Sd
inY true if ym –ySd ≤ y < ym + ySd
inXY true if inX == True and inY == True
V inner vertical stripe of points with inX == True
H inner horizontal stripe of points with inY == True
VH inner rectangle, points with inXY == True
inV , inH, inVH number of points in V, H, VH
x I , yI coordinates of points in VH
L, R, B, T left, right, bottom, top stripe defined by x ≤ x m - x Sd , x ≥ x m +
x Sd , …
8.3.3 Excel
With the spreadsheet function RAND(), 2000 random numbers are generated each in
columns A and B of Fig. 8.4 (s), designated x and y, and displayed as 2000 points
(x, y) in the plane (Fig. 8.3a).
The mean values x m and ym of 2000 between 0 and 1 equally distributed ran-
dom numbers are (not surprisingly) about 0.5. The standard deviations are found
to be x Sd = 0.29 and ySd = 0.29, corresponding within 2% to the theoretical
value of 0.289 (Eq. 8.6). So, there is a fraction 2 × 0.29 = 0.58 to be expected
bars or within the horizontal bars, and a fraction 0.58 ≈ 0.33
within the vertical 2

√2 · √2 = 1 within the inner rectangle of Fig. 8.3b.
12 12 3
d) Sd)
m+xS ym+y
<=x;x<x d<=y;y< )
S d yS 4 A()) A()) I)
()
1 =RAND =RAND
() (xm-x AND(ym- AND(G4;HIF(I4;x;NIF(I4;y;N COUNT(y
=AND = = = = =
2 x y xm 0.51 =AVERAGE(x) inX inY inXY xI yI
3 0.24 0.57 xSd 0.29 =STDEV.S(x) TRUE TRUE TRUE 0.24 0.57 672
4 0.38 0.55 ym 0.49 =AVERAGE(y) TRUE TRUE TRUE 0.38 0.55 0
5 0.89 0.22 ySd 0.29 =STDEV.S(y) FALSE TRUE FALSE #N/A #N/A =SUM(inY)
2002 0.12 0.02 FALSE FALSE FALSE #N/A #N/A
Fig. 8.4 (S) In columns A and B, each 2000 random numbers between 0 and 1 are generated,
and in C3:F3, their mean value (x m , x m ) and standard deviations x Sd and ySd are determined. The
columns G to I check whether the points are within certain boundaries. L4 contains the formula
SUM(INY). “#N/A” means “Not available”. The name of this worksheet is “calc” and is addressed as
such in VBA procedures
M N O P Q R S T U
d) ) d)) )
) ) E) m-xS &xm+xSd "&(ym-yS &ym+ySd
TRUE TRUE ;TRU ="&x " = "
TIF(inX;TIF(inY; TIF(inXY /2000 TIF(x;"< TIF(x;"> TIF(y;"< TIF(y;">
N N N H N N N N
1 =COU =COU =COU =inV =COU =COU =COU =COU
2 inV inH inVH fVH nL nR nB nT
3 1157 1179 672 0.336 409 434 408 413
Fig. 8.5 (S) COUNTIF is applied, counting how often the data x and y in Fig. 8.4 (S) are less away
from the center than the standard deviation, in M:O with the logical data in x in , yin, and xyin . In
Q:T, the data x and y themselves are taken for the logical query if the data are further away. We
get inV + nL +nR = 2000
Check with And

In column G, for each individual x-coordinate, we check whether it deviates less than
the standard deviation downwards or upwards from the mean value x m . In column
H, the same is done for the y-component. For these checks, the logical function
AND(LOGICAL1, [LOGICAL2], …) is used. The corresponding formulas are.
inX = AND(x m -x Sd < = x; x < x m + x Sd )

iny = AND (ym -ySd < = y; y < ym + ySd )
In column I, we mark the points within the devi limits of both x m and ym . We cannot
use the statement AND(x in ; yin ), because it returns TRUE if all elements of the column
vectors x in and yin are true. So, we have to write, e.g., I4 = AND(G4;H4), and so on,
to check individually whether the contents of both cells are true.
Logical functions
In columns J and K, we extract the coordinates of the points within the inner rectangle
with a statement like K4 = [=If(I4; x; NA())], returning the value of x if I4 = = True
and #N/A else. The column ranges in K and J can be entered as data series into a
figure with the cells with #N/A being ignored.
A check COUNT(yI) (in L3) yields 672, although the queried range yI comprises
2000 entries, because only the cells with numbers, and not those with #N/A, are
counted. The function SUM(inY ) returns 0 because only numbers are summed up and
logical values are ignored.
Logical Functions
We have used two logical functions so far:
IF(LOGICAL_TEST, [VALUE_IF_TRUE], [VALUE_IF_FALSE]);
in [VALUE_IF_TRUE] and [VALUE_IF_FALSE] may appear numbers, or functions
whose result is then processed.
AND(LOGICAL1, [LOGICAL2], …)
expects logical values as input and outputs TRUE or FALSE.
Please refer to EXCEL help to inform yourself about the other functions in
the category LOGIC: OR; FALSE; TRUE; NOT; IFERROR.
Count the Trues

In Fig. 8.5 (S), the logical calculations of Fig. 8.4 (S) are continued. The spreadsheet
function COUNTIF(RANGE, CRITERIA) counts the non-empty cells of a range whose
contents match the search criteria. In M3:O3, the ranges inX , inY , and inXY are
queried as to how often their cells contain TRUE. In P3, the relative frequency f VH of
points within the inner rectangle in Fig. 8.3b is reported (f VH = 0.336).
In the subsequent columns Q to T, the number of points in the lower nL ,
right nR , bottom nB , and top nT stripe is calculated. Here, logical expressions
formulated as a character string are used as criteria. In the formula S3 =
[=COUNTIF(y&" < = "&ym − yd )], the criterion is [" < = "&ym − yd ]. The expres-
sion is of the form “Text”&Variable, composed of the comparison symbol in
quotation marks “<” and the arithmetic expression ym − yd .
In Fig. 8.6 (S), the results of Fig. 8.5 (S) are evaluated to estimate the relative
amount of points within the inner rectangle with a probabilistic calculation. In
columns V and W, the number of points outside the x and y devi boundaries is
obtained through addition; in X and Y, it is converted into the numbers inside the
devi bars (the same as in Fig. 8.5 (S)) from which the relative amount is obtained
in Z and AA. These relative amounts are interpreted as probabilities.
Fig. 8.6 (S) Continuation of V W X Y Z AA AB

Fig. 8.5 (S)
nR nB+nT 2000-U3 2000-V3 W3/2000 X3/2000 pV*pH
1 =nL+ = = = = = =
2 ==inV ==inH pV pH pVH
3 843 821 1157 1179 0.58 0.59 0.341
Product of probabilities
As the x and y coordinates are independent of each other, the probability of finding a
point in the inner rectangle may be calculated as the product of the two probabilities
mentioned above:
pVH = pv · pH
We get pVH = 0.341 (AB3). This is to be compared with the relative frequency
of occurrence f VH = 0.336 in Fig. 8.5 (S) and with the theoretical value of 1/3 =
0.333 (from Eq. 8.6).
Keep in mind the differences:
– The theoretical probability is logically derived from the theoretical values of

the mean and the standard deviation of an equi-distribution independent of the
current experiment with 2000 points.
– The quantity pVH is estimated from the relative occurrence of points in V and
H in our current experiment.
– The relative frequency of occurrence in VH is empirically determined in our
current experiment.
Due to these methodological differences, the three values are not necessarily the
same; but, as they evaluate the same set of points, their values should be close
together.
8.3.4 Python
In Python, we have to import the numpy.random library as npr (see Table

8.4) for our exercise. The generation of the 2000 points is straightforward with
npr.random (2000).
The logical queries as to whether the points are in the areas V or H are
performed in three different ways, all resulting in a Boolean array with length
2000:
– in line 6, “inX = np.logical_and ((xm-xSd) < x,

x < (xm + xSd))”, identical to the spreadsheet formulas in Fig. 8.4
(S), with the result being shown in the bottom cell of Table 8.4,
– in line 7, “inX2 = ” as list comprehension with identical results to those
in line 6, as checked with the instruction “inX == inX2", resulting in a
Boolean array of Trues,
– in line 8, “inY = ” as a different type of list comprehension.
The query as to whether a point is in the inner rectangle is done by means of the
list comprehension in line 9.
Table 8.4 Python: 2000 points (x, y) randomly distributed in a unit square, the points (x, y) are
depicted in Fig. 8.3a, as are the points (x I , yI ) in Fig. 8.3b
2 x=npr.random(2000)
3 xm,xSd = np.mean(x),np.std(x)
4 y=npr.random(2000)
5 ym,ySd = np.mean(y),np.std(y)
6 inX=np.logical_and(xm-xSd<x,x<xm+xSd) #Within std error?
7 inX2=[(xm-xSd<x) & (x<xm+xSd)]
8 inY=[(ym-ySd < yi < ym+ySd) for yi in y]
9 inXY=[inX & inY]
#Arrays of coordinates of points within error range:
10 xI=np.extract(inXY,x)
11 yI=np.extract(inXY,y)
inX [ True True True ... False False True]
len(inX) 2000
inX==inX2
[[ True True True ... True True True]]
len(yI) 649
The coordinates of the points in the inner rectangle are obtained, in lines10
and 11, with the function np.extract(condition, array). The condition
here is that inXY == True. The length of x I , i.e., the number of points in the
inner rectangle, is 649, to be compared to 672 = COUNT(YI) in the EXCEL sheet
of Fig. 8.4 (S). The theoretical expectation for a binomial question (point in VH
√
or not in VH) is xm = n · p = 667 and x Sd = npq = 21, so that the above
values are within the standard error range. The meaning of these statements will
be discussed in Chap. 9.
Logical operations on lists and arrays in Python

The operators & (and), | (or), and ~ (not) operate item-wise on lists of Booleans
and create new lists of Booleans. In Table 8.4, however, the numpy functions
np.logical_and, np.logical_or, and np.logical_not are applied;
they do the same, except that they create arrays. The difference becomes apparent
when we try to combine two of them, e.g., with &. This is not possible for lists; in
trying to do so, we get the error message "Unsupported operand type(s)
for &: ’list’ and ’list’". But it works well with arrays, e.g.,
inXY=[inX & inY]
yielding a Boolean array with "True" if the point lies within the inner rectangle
in Fig. 8.3b. With the instruction.
xI = np.extract(inXY,x)
Table 8.5 Size of the Boolean arrays denoting the various partial areas in Fig. 8.3b
12 inV=sum(inX) inV 1124
13 inH=sum(inY) inH 1153
14 inVH=sum(inXY) inVH 649
fVH 0.32
15 inL=sum(xi<xm-xSd for xi in x) inL 437
16 inR=sum(xm+xSd<xi for xi in x) inR 439
17 inB=sum(yi<ym-ySd for yi in y) inB 433
18 inT=sum(ym+ySd<yi for yi in y) inT 414
a new array is created containing the x coordinates of the points in the inner
rectangle.
The points in the inner square are plotted with an open diamond (see Fig. 8.3b)
with the instruction
plt.plot(xI,yI,’kd’, fillstyle=’none’).
In Table 8.5, we have calculated the numbers of points within the various partial
areas V, H, VH, L, R, B, T of the unit square, defined in Sect. 8.3.2.
The values TRUE and FALSE in a Boolean array are coded as 1 and 0 in a
binary list. So, the number of TRUEs is equal to the sum over the list. This is not
possible in Excel; = SUM(INY) in L4 of Fig. 8.4 (S) yields 0.
Questions
How are inV and inH related to inL , inR , inB , inT ?13
How do you estimate the probability of finding points in VH from inV and
inH ? Compare the constructions in lines 6 to 9 in Table 8.4!14
8.3.5 Why Calculate Twice?
Alac In Sect. 8.3.3, we calculate the probability of finding a point in the inner
rectangle one time too many, once as f VH in O3 of Fig. 8.5 (S) and a second time
as pV · pH in Fig. 8.6 (S). We could have saved work.
Tim There is a difference: f VH estimates the probability from the relative

frequency in VH, whereas pVH relies on the product rule of probability.
Mag Tim is right. Anyhow, it is always good to follow two calculation paths
that are expected to deliver the same result. In this way, we can check whether
13 inV = 2000 – inL – inR ; inH = 2000 – inT - inB.

14 pVH = pV · pH = inV /2000 · inH /2000 = 0.324.
our thoughts have been logical and whether we have correctly transferred our
thoughts into the calculation. This is especially advisable when using functions
and formulas for the first time.
Practical advice Before you seriously use functions for the first time, check
their operation with examples for which you can do the arithmetic mentally!
Product rule of probability

As can be seen in Fig. 8.5 (S) (and Table 8.5), only about pH = 59% (inH/2000
= 0.59) of the points lie within the two horizontal lines in Fig. 8.3b. Within the
vertical lines, the proportion is also 0.59 (0.56). Within the rectangle (pVH ), it is less.
According to the rules of probability, it should be the product of the probabilities
within the vertical and horizontal lines, pVH = pV ·pH , provided the two events are
independent of each other. This is indeed the case, because the x- and y-coordinates
have been created independent of each other. We estimate the probabilities with
the point frequencies (0.59 × 0.59) = 0.348. If there is a difference between this
probability and that obtained with the number of points in the inner rectangle, it is
due to the fact that the probabilities are not exact, but simply estimated from the
number of points in the considered areas.
8.4 Set Operations in Numpy
Python provides a data format for sets, and the operations that will be
performed on them. With such sets, we illustrate basic rules for the prob-
abilities of unions and intersections, as well as Bayes’ rule on conditional
probabilities and apply logical queries.
8.4.1 Sets
In the program underlying Fig. 8.7, 400 points are created in the unit square, with
their x- and y-coordinates being two arrays of independent, equally distributed
random numbers. In Fig. 8.7a, only the points in three subsets are displayed: the
top stripe T with y > 0.8 (marked with + ), the right stripe R for x > 0.6 (marked
with x, and the points in the right upper rectangle RT (additionally marked with
open squares ).
The points in the ranges R and T are defined in Python as sets with their
coordinates obtained through logical queries x > 0.6 and y > 0.8 (see Table 8.7).
The number of points in the sets is obtained with the function len:
len(R) → 158, len(T) → 85.
Fig. 8.7 Points in T, R, and RT. a (left) Plotted with their coordinates, e.g., plt.plot(xT, yT,
…); b (right) Plotted as two sets, symdiff and sect = RT mirrored at (0.5, 0.5) (see Table 8.8)
The probability that a new point will be in one of the sets may be estimated as:
p(R) = 158/400 = 0.395, p(T) = 85/400 = 0.213
Now, we perform set operations on R and T (Table 8.8):
– Union, defined in Python as R|T; comprising the points in R or T

– Section, defined in Python as R&T, comprising the points in R and T, marked
above in Fig. 8.7a with open squares.
For the situation in Fig. 8.7, we get:
len(R|T) = 211, p(R|T) = 211/400 =0.528

len(R&T) = 32, p(R&T) = 32/400 = 0.080
Applying the addition rule for the union yields:
p(R|T ) = p(R) + p(T )− p(R&T ) (8.7)
which holds exactly true for our special case:
211/400 == 158/400 + 85/400 – 32/400
The multiplication rule for independent sets reads:
p(R&T ) = p(R) · p(T ) (8.8)

8.4 Set Operations in Numpy 329
The precondition for the multiplication rule is that the two events are inde-
pendent of each other. R and T are indeed independent sets, because x and y are
generated independently, and the membership in R and T depends exclusively on
x or y. As an illustrative counter-example, consider RT where x and y are not
independent of each other.
Checking the multiplication rule for our special case gives us
p(R&T) = 32/400 = 0.080

p(R) · p(T) = 158/400 × 85/400 = 0.084
There is no exact coincidence between these probabilities, because they are

based on an estimate using a sample size of 400. Increasing the number of points
lets the two probabilities come closer together.
Other set operations are:
– Difference, defined as R-T, points in R but not in T:

– len(R-T)= len(R) - len(R&T)
– Symmetric difference, defined in Python as RˆT, points either in R or in T,
indicated in Fig. 8.7b with x:
– len(RˆT)=len(R|T)-len(R&T).
For our specific case, we get: len(R-T) = 126, len(RˆT) = 179.
Conditional probabilities, Bayes’ rule

What is the probability p(T\R) of finding a point in set T when we know that it is in
set R? Graphically, this is the area of RT divided by the area of T, formulated with
probabilities as
p(R&T )
p(T \R) = (8.9)
p(R)
For our specific case, we get
p(T\R) = (32/400)/(158/400) = 0.202

p(R\T) = (32/400)/(85/400) = 0.376
The two conditional probabilities are interconnected by Bayes’ rule:
p(R)
p(R\T ) = p(T \R) ∗ (8.10)
p(T )
Questions
What is the theoretical probability of finding a point in the upper-right rectangle

RT of Fig. 8.7?15
Check the rules for difference and symmetric difference for the numbers of
our specific case!16
Check Bayes’ rule for the numbers of our specific case!17
x, y coordinates, random between 0 and 1

R set of points in the right stripe
x Rt left boundary of R
inR Boolean array, TRUE if the point is in the right stripe
x R , yR coordinates of points in R
Rtup array of tuples (x R , yR )
T set of points in the top stripe
yTp bottom boundary of T
RT section of R and T
x T , yT coordinates of points in T.
8.4.3 Python
In Table 8.6, random coordinates (x, y) of N = 400 points are generated from
which those points are selected that lie in a vertical stripe R right of x Rt . To achieve
this, first, a logical array inR is determined containing True when the condition
x Rt < x is fulfilled. Then, the coordinates x R and yR are extracted from the True
positions.
In Table 8.7, two sets R and T are built based on the coordinate lists x R , yR and
x T , yT . We first zip the coordinates into tuples (result in Table 8.10) and then define
R and T as sets with these tuples as elements. In our specific case, an array with
158 (out of 400) entries results. The same is done for a horizontal stripe T above
yTp . In the following, the set operations union, section, difference, and symmetric
difference are performed. The size of the resulting sets is reported in the right cell
of Table 8.7.
15 The probability of finding a point in RT is its area 0.2 × 0.4 = 0.08 in the unit square.
16 len(R-T) = len(R) - len(R&T); 126 = 158 -32.
len(RˆT) = len(R|T)-len(R&T); 179 = 211 - 32.
17 32/85 = 32/158 * 158 / 85.
8.4 Set Operations in Numpy 331
Table 8.6 Coordinates of the three sections R, T, and RT in Fig. 8.7

1 N=400
2 x= npr.random(N)
3 y= npr.random(N)
4
5 xRt = 0.6 #Right border
6 inR=[xRt<x] #Right stripe
7 xR=np.extract(inR,x)
8 yR=np.extract(inR,y)
9
10 yTp = 0.8 #Top border
11 inT=[yTp<y] #Top stripe
12 xT=np.extract(inT,x)
13 yT=np.extract(inT,y)
14
15 inRT=np.logical_and(inR,inT) #Top-right rectangle
16 xRT=np.extract(inRT,x)
17 yRT=np.extract(inRT,y)
Table 8.7 Defining the sets R and T and combined sets

18 Rtup=zip(xR,yR) #Right str.
19 R=set(Rtup) len(R) 158
20
21 Ttup=zip(xT,yT) #Top stripe
22 T=set(Ttup) len(T) 85
23
24 union = R|T len(R|T) 211
25 sect = R&T len(R&T) 32
26 differ = R-T len(R-T) 126
27 symdiff = R^T len(R^T) 179
28 un_m_sc = union-sect len(un_m_sc) 179
Questions
In Table 8.6, two variables with nearly identical names occur, xRt (a scalar) and
xRT (an array). Why is there no name conflict?18
Table 8.8 reports the instructions for plotting Fig. 8.7b.

Bayes’ rule is numerically checked in Table 8.9.
18Python is case-sensitive; xRt and xRT are two different names. Nevertheless, this is bad
naming.
Table 8.8 Plotting the sets in Fig. 8.7b

1 FigStd('x',0,1,0.1,'y',0,1,0.1)
2 x4,y4=zip(*symdiff)
#For effect of zip(* see:
https://stackoverflow.com/questions/29139350/difference-
between-ziplist-and-ziplist
3 plt.plot(x4,y4,'kx',label="symdiff")
4 x5,y5=zip(*sect)
5 x6,y6=np.array(x5),np.array(y5)
6 plt.plot(1-x6,1-y6,'kD',fillstyle="none",
label="sect, mirrored")
Table 8.9 Bayes’ rule

1 T_given_R=len(sect)/len(R) 0.203
2 R_given_T=len(sect)/len(T) 0.376
3 Bayes=T_given_R*len(R)/len(T) 0.376
Table 8.10 Data structure of Rtup and R defined in Table 8.7

xR [ 0.86 0.78 0.86 ... 0.99 0.73 0.97]
yR [ 0.79 0.24 0.14 ... 0.09 0.43 0.48]
*zip(xR,yR) (0.86, 0.79) (0.78, 0.24) …
Rtup <zip object at 0x000001E3C57F7B48>
R {(0.85, 0.17), (0.9, 0.09), … }
Some Python constructs

The function np.extract (Boolean_array, Value_array), applied in
Table 8.6, extracts those values of the Value_array for which the corresponding
position in a Boolean_array contains True.
Table 8.10 elucidates the data structure of a zip object and a set. The zip()
function pairs together the items of iterators passed as arguments. An iterator is an
object that contains a countable number of values; in our case, we pair the two arrays
x R and yR . The set R contains the same tuples as Rtup , but in a different order because
the order does not play any role in sets.
8.5 Normally Distributed Random Numbers
The functions NORM.INV(RAND();0;1) (EXCEL), as well as

sct.norm.ppf(npr.random(…))and npr.randomn(…)
8.5 Normally Distributed Random Numbers 333
(Python), generate numbers that are standard-normally distributed.

We check this statement by comparing a frequency distribution of a set
of such numbers with the theoretical distribution using the Chi2 test. The
expected frequency of occurrence in an interval is calculated in two ways:
Exactly, with the values of the cumulative distribution function (cdf ) at the
interval boundaries, and approximately, with the probability density (pdf )
in the middle of the interval. The inaccuracy of the approximation will only
become visible with more than 10,000 random numbers.
8.5.1 Normal Distribution, Probability Density and Distribution

Function
This section presents the mathematical background of the formulas applied later
in this exercise.
Normal distribution
The following nominations and specifications hold:
x m , x Sd mean value and standard deviation

–∞ < x < ∞ argument range
0<p<1 probability, value range
EXCEL provides two spreadsheet functions, one with a Boolean parameter

CUMULATIVE?for calculations with normal distributions, corresponding to three
Python functions available from the scipy library:
– probability density function pdf , x → p
xM, xSD , CUM), cum = False or 0

NORM.DIST (X,
sct.norm(xm , xSd ).pdf(x)
– cumulated density function cdf (distribution function), x → p
xM , xSD ,CUM),), cum = True or 1,

NORM.DIST (X,
sct.norm(xm , xSd ).cdf(x)
– the inverse of the distribution function, p → x
xM , xSD ) ,
NORM.INV(P,
sct.norm(xm , xSd ). ppf(p)
1.0 1.0
pdf(0,1) pdf(0,1)
cdf(0,1) 0.8 cdf(0,1) 0.8
∫ x≤1
p p(Ic)
ppf
0.6 0.6
p
0.4 0.4
0.2 0.2
0.0 0.0
-4 -2 0 2 4 -4 -2 0 2 4
x x
Fig. 8.8 a (left) Solid line (bell curve): pdf (0, 1)) , probability density of the standard normal
distribution (i.e., x m = 0 and x d = 1); dashed line: cdf (0, 1), distribution function of the normal
distribution; the content of the grey area under the probability density is indicated by the diamond
at the end of the vertical bar ending in a diamond on the distribution function cdf . b (right) The
grey area below the probability density pdf is equal to the difference of the two y-values of the
diamonds on the distribution function cdf . The horizontal bar indicates the value of the probability
density in the middle of the considered interval
The extension ppf stands for “percent point function”. However, most people
in statistics use “quantile function”.
In mathematical textbooks, the normal distribution is often referred to as N(μ,
σ 2 ), where μ is the mean value and σ 2 is the variance. However, since it is the
standard deviation that has to be entered as a parameter into the spreadsheet and
Python functions, we shall use the abbreviations pdf (xm , xSd ) in the text and
in legends for the probability density of the standard normal distribution that is
defined by the equation
2
1 1 x − xm
pd f (xm , x Sd ) = √ · exp − (8.11)
2π · x Sd 2 x Sd
The bell-shaped pdf (0, 1) is shown in Fig. 8.8, together with the corresponding
S-shaped cdf (0, 1) (dashed line).
Questions
What is the physical unit of the argument of exp in Eq. 8.11?19

What is the physical unit of pdf and pdf ·dx?20
19 The argument in exp is dimensionless, because the nominator and denominator have the same
unit.
20 [pdf ] = 1/[x ], a density; [pdf ·dx] = unit-less, a probability.
d
Do the following two spellings yield different results in EXCEL: Y = EXP(-2ˆ2)

and Y = EXP(-(2ˆ2))?
Same question for np.exp (-2**2) and np.exp (-(2**2)) in
Python.21
Distribution function or cdf

In mathematical literature, the integral of a probability density is called the distribu-
tion function; it is monotonously increasing and spans the value range (the range of
the output values) from 0 to 1. In Python, it is called cdf , “cumulated (probability)
density function”. We shall adopt this notation together with pdf and ppf in this text.
The distribution function of the normal distribution cannot be represented in a
closed form, but this is not a disadvantage for us, since EXCEL offers a spread-
sheet function NORM.DIST(x; x m ; x d ; TRUE) and Python offers sct.norm(x M ,
xSd ).cdf(x), which are good approximations.
In Fig. 8.8a, the probability density of the normal distribution is shown up as
a bell curve and the distribution function as a monotonously increasing S-shaped
curve. The open diamond marks the distribution function at x = 1, corresponding
to the value of the integral over the probability density from −∞ to 1, which is
represented as a grey-filled area under the bell curve, representing the probability
of finding a value of the random number below x = 1.
The function ppf(p) is the inverse of the distribution function cdf(x), giving the
value of x for a specified p, as demonstrated by the kinked line in Fig. 8.8a: A value
p0 on the vertical axis between 0 and 1 is assigned a value x 0 on the horizontal
axis (see arrow). It is the probability of finding a value x ≤ x 0 , corresponding to
the grey area under the probability density.
Probability that a random number falls into an interval

What is the probability p(x 1 , x 2 ) that a normally distributed random number lies
in the interval [x 1 , x 2 )? To calculate this, we have to integrate over the probability
density pdf(x) :
x2
p(x1 , x2 ) = ∫ pd f (x)d x = cd f (x2 ) − cd f (x1 ) (8.12)
x1
The integral in Eq. 8.12 is the difference of the distribution function at the two
interval limits. To give an example: in Fig. 8.8b, an interval from 0.2 to 0.7 has been
selected. The probability of finding a normally distributed random number in this
interval corresponds to the grey area under the probability density pdf , which, in
turn, corresponds to the difference of the values of the distribution function cdf at
the two positions marked with open diamonds.
21 Yes, there is a difference, but only in EXCEL. In a spreadsheet, the argument of the Gaussian
function must be spelled as (-(xˆ2)), because negation has operator precedence over potentiation,
quite surprisingly for mathematically-educated readers and an annoying error source, so: –2ˆ2 =
4; –(2ˆ2) = –4.
The integral over an interval can be approximated by the product of the

probability density in the center x C of the interval and the width x of the interval:

x2 − x1
p(x1 , x2 ) ≈ pd f · (x2 − x1 ) = pd f (xC ) · Δx (8.13)
2
In Fig. 8.8b, this value is indicated by the horizontal line between the interval
limits. The more linear the probability density between the selected interval limits
and the narrower the interval, the better this approximation. The frequency of
occurrence in that interval for a sample of size N is pd f (xC ) · Δx · N .
8.5.2 Random-Number Generator and Frequencies of Occurrence
The function npr.randn(N) of the library numpy.random returns an array

of N numbers randomly distributed according to the standard normal distribution
with x m = 0 and x Sd = 1. In EXCEL, there is no simple function for doing this, and
thus a nested function, NORM.INV(RAND();0;1), has to be used. The literal equivalent
in Python is sct.norm.ppf (npr.random(N)).
Figure 8.9 displays the two frequency distributions of 10,000 random numbers
created with the two Python functions randn and ppf (random), together
with the theoretically expected distribution. We see that, indeed, the two functions
act alike, and the numbers provided seem to be normally distributed.
Fig. 8.9 a (left) Frequency distribution of the numbers generated with the two generators of
normally distributed random numbers in Table 8.12 (randn and ppf (random)); the vertical bars
represent the interval boundaries; the polyline connects the frequencies freqXpC expected for a nor-
mal distribution (obtained with cdf (0, 1)). b (right) Data for pdf (x) , cdf (x), and ppf (p) of the
standard normal distribution obtained from Table 8.11.
250 25000
pdf(0;1)*N*dx pdf(0;1)*N*dx
N=1024
frequency
frequency
N=102400
200 20000
150 15000
100 10000
50 5000
0 0
-3.0 -1.5 0.0 1.5 3.0 -3.0 -1.5 0.0 1.5 3.0
x x
Fig. 8.10 a (left) Frequency distribution for 1024 random numbers. b (right) Frequency distribu-
tion for 100 times more random numbers than in a
8.5.3 Where Do Observed and Theoretical Frequencies Fit Better

Together?
Figure 8.10a displays experimentally determined frequencies of occurrence for

1024 random numbers generated in EXCEL with NORM.INV(RAND();0;1) as data points
over the interval centers x c . The numbers of events outside of the minimum and
maximum interval boundaries are shown in the picture on these interval boundaries
(here, −3 and 3). Figure 8.10b shows a similar picture for a hundred times more
random numbers.
The polylines in Figs. 8.10a and Fig. 8.10b represent the frequencies of
occurrence in the interval [x 1 , x 2 ):
f r eq(x1 to x2 ) = pp f (xc ) · N · x (8.14)
where N is the total number of data and x c and x are the center and width,
respectively, of the sorting intervals, also called the “bins”.
Questions
What additional information would significantly increase the information con-

tent of the legends Fig. 8.10a, b? Compare Fig. 8.16b!22
Mag Where do observed and theoretical frequencies fit better together, in the
left or right picture of Fig. 8.10?
Alac Quite clearly, in the right picture. The experimental and theoretical points
are much closer together than in the left picture.
22 Indicating the result of a Chi2 test would considerably increase the information in the figure.
Tim This surely is a trick question.
Mag Indeed. Many unbiased observers believe, like Alac, that, in Fig. 8.10b,
model distribution and experimental frequencies coincide better than in Fig. 8.10a.
Alac Sure.
Mag Well, it’s an optical illusion. For Fig. 8.10a, the Chi2 test yields a value
of 0.333, giving no reason to doubt that Eq. 8.14 correctly describes the ran-
dom experiment with 1024 single values. For 100 times more single values, e.g.,
Fig. 8.10b, the Chi2 test yields a value of 0.0001, i.e., only in 1 of 10,000 cases
does an even greater deviation occur between the data of a sample and the model
for the population.
Alac Then, we can presumably no longer claim that the theoretical distribution
describes the experimental distribution well. Why not?
Mag For 102,400 individual values, the approximation of Eqs. 8.13 and 8.14 is
not good enough. For so many data, statistical science expects a deviation that is
even smaller than we can note with the eye.
Exactly calculated frequencies of occurrence

The probability of finding a random number in an interval is determined exactly by
the difference of the cumulative distribution function cdf at the interval boundaries,
as prescribed in Eq. 8.12. To determine the frequency in the interval, the probability
must be multiplied by the total number N of the analyzed data:
f r eq(x1 to x2 ) = (cd f (x2 ) − cd f (x1 )) · N (8.15)
In this formula, the interval width x does not show up, in contrast to Eq. 8.14.
A comparison of the frequencies expected according to Eq. 8.15 with the exper-
imental frequencies in the intervals in Fig. 8.10b results in CHISQ.TEST = 0.37. So,
everything is fine again.
The more individual data are available for a statistical test, i.e., the
larger N is, the easier it is to detect discrepancies between experimental
frequencies and those predicted by a model distribution. For only a few
measurements, deviations from a wrong model distribution may not be
noticeable, because deviations from the correct model distribution can
also be large.
What is exact, and what is practical?
Mag We have often calculated the theoretical frequency in an interval with

Eq. 8.14 as (pdf in the center of the interval) times (interval width) times (total
number of data). What can you say about this approach?
Tim It’s wrong. You have to take the difference of the distribution function at
the interval limits times the total number, according to Eq. 8.15.
Alac However, this is a little cumbersome to program. The approximate method

gives faster results, and you can graphically display the probability density with a
polygon line.
Mag That’s true. The path via probability density is an approximation. But in
many cases, especially in experiments with smaller amounts of data, it does not
lead to noticeable differences from the exact method with the distribution function
cdf .
Alac That’s what I’m saying. So, it’s more practical.
Tim Agreed! Nevertheless, I only use this as a first approximation and will take
the trouble to apply the method with the cdf at the end of my work.
What can we learn philosophically ;-)?

The philosopher Karl Popper stated that scientific theories cannot be confirmed, but
only falsified. We find an example of this in our exercise. Let us forget the arguments
in the above dialogue and again accept Eq. 8.14 as a valid theory.
In Chi2 tests, this model provides sufficiently large values for small amounts of
data. However, this finding does not confirm the theory; it only gives us no reason
to reject it. It is only for large amounts of data that the Chi2 test yields such small
values that we may reject the theory with a very small probability of error.
In the history of physics, there have always been cases in which only a refined
measuring technique was able to falsify a theory that then turned out to be merely
an approximation. An example: In Newtonian mechanics, the mass of a body is
independent of the velocity of the body, but this is not so in Einstein’s theory of
special relativity.
Chi2 test with ever bigger sample sizes

We are going to perform an experiment in which, four times for each integer Smp100N
(= 1 to 100), we:
– create an array with 1024*Smp100N normally distributed random numbers

– determine their observed frequencies of occurrence f Obs
– compare f Obs with frequencies f Xpt expected from the probability density pdf (x c )
in the center x c of the intervals by means of Chi2 tests.
Fig. 8.11 Results of

Chi2 tests for sets with ever
more samples (1024*N)
The results of the Chi2 test are shown in Fig. 8.11 as a function of N. The number
of random numbers is 1024* Smp100N. For a small size of the set (≤ 10*1024 for
Smp100N ≤ 10), the values of the Chi2 test seem to be equally distributed, so that
there is no reason to doubt that our approximation describes the empirical results
correctly. For increasing size, the values of the Chi2 test tend to be low, indicating that
the theoretical model is not correct. The bigger the size of the set, the less probable
the high values of the Chi2 test are. Nevertheless, values above 0.5 may occur.
Take some time to think about this exercise! It will tell you a lot about the reli-
ability of a Chi2 test, a model for other statistical tests, about statistical reasoning
and the Philosophy of Science.
The normal distribution
x m , x Sd mean and standard deviation of the normal distribution

pdf(x m , x Sd ) x → p, probability density function
cdf(x m , x Sd ) x → p, distribution function
ppf(x m , x Sd ) p → x, inverse of cdf , “percent point function”
x array of arguments, equally spaced
pD pdf of x
pC cdf of x
p array of probabilities, equally spaced between 0 and 1.
Normally distributed random numbers
N size of the set of random numbers

xN array of numbers created with randn(N)
freqN frequency of occurrence of x
xP array of numbers created with ppf(npr.random(N))
xb boundaries of sorting intervals (bins)
freqP frequencies of occurrence of x P
freqXpC frequencies of occurrence expected from cdf
freqXpP frequencies of occurrence expected from ppf .
8.5.5 Python
Normal distribution
Table 8.11 shows the program for generating arrays with values of these func-
tions to be displayed as three data series in Fig. 8.9. The functions concerning the
normal distribution have to be made available from the library scipy.stats, usu-
ally imported under the abbreviated name sct with import scipy.stats as
sct.
Random Gaussian generator

The Python program in Table 8.12 creates a random sample of 10,000 normally
distributed numbers in two ways, first, with the function npr.randn(10,000) , and
second, with the function npr.random (generating random numbers between 0
and 1) inserted into sct.norm.ppf, the inverse of the distribution function. The
second version corresponds to the nested EXCEL function and can generally be applied
to other distributions, e.g., to get the cos2 -distribution needed in Exercise 8.7.
Next, the empirical frequency distribution of the two samples is determined within
23 intervals with the boundaries Ib = −∞, −3, −2.7, …, 2.7, 3, ∞. This is done with
the function np.histogram that expects the sample and the interval boundaries as
input and returns the empirical frequencies on row 0 (freq[0] in Table 8.12) and the
unchanged interval boundaries on row 1. The data obtained on x and x P are displayed
in Fig. 8.9a, together with the theoretically expected frequencies freqXpC .
Table 8.11 Python functions related to the normal distribution

2 xN=np.linspace(-4,4,41)
3 pD=sct.norm(0,1).pdf(xN) #Probability density function
4 pC=sct.norm(0,1).cdf(xN) #Cumulated density function
5 xP=sct.norm(0,1).ppf(pC) #Percent point function
Table 8.12 Normally distributed random numbers x and x P

6 N=10000
7 xN=npr.randn(N)
8 xP= sct.norm.ppf(npr.random(N)) #Corresp. to Excel formula
9
10 b=np.linspace(-3,3,21)
11 dx=b[1]-b[0] #Interval width
12 xb=np.empty(len(b)+2) #Interval boundaries
13 (xb[0], xb[1:-1], xb[-1])=(-np.inf,b, np.inf)
14 freqXpC=(sct.norm(0,1).cdf(xb[1:]) #Expected freqs.
15 -sct.norm(0,1).cdf(xb[:-1]))*N
16 xc=np.ones(len(xb)-1) #Centers of intervals (“bins”)
17 xc[1:-1]=(xb[1:-2]+xb[2:-1])/2
18 xc[0],xc[-1]=xb[1],xb[-2]
19
20 freqN =np.histogram(xN,bins=xb)
21 freqP =np.histogram(xP,bins=xb)
x [ 1.20 0.88 1.67 ... -0.78 -1.82]
xb [ -inf -3.00 -2.70 ... 3.00 inf]
freqN[0] [ 17 22 50 ... 16 18]
freqXpC [ 13.5 21.2 47.3 .. 21.2 13.5]
8.5.6 Excel
Normal distribution
We create maps of the probability density of the normal distribution and its distribu-
tion function. In the following spreadsheet, Fig. 8.12 (S), 41 values of the probability
density pdf and the distribution function cdf are generated and shown in the diagrams
of Fig. 8.8.
A B C D E F G
2 dx 0.2
)
ALSE ;1;TRUE)
0;1;F 0 ;0;1)
.DIST(x; .DIST(x; NV(p
3 A5+dx
N ORM N OR M N ORM.I
= = = =
4 x pdf(0,1) cdf(0,1) p
5 -4.0 0.000 0.000 0.0 #NUM!
6 -3.8 0.000 0.000 0.1 -1.28
14 -2.2 0.035 0.014 0.9 1.28
15 -2.0 0.054 0.023 1.0 #NUM!
45 4.0 0.000 1.000
Fig. 8.12 (S) Function table for the normal distribution (probability density pdf(0,1)) with the
mean value 0 and the standard deviation 1, column B), the associated cumulative (integral) nor-
mal distribution or distribution function (cdf(0,1), column C), and the inverse of the distribution
function in column F
A B C D E F G H
1 =COUNT(xN) dx 0.5
2 1024 N 1024 =SUM(freq)
) *dx ;freqXpC)
);0;1 b)} ;0)*N 7
V(RAND( (D5:D6) N CY(xN;x ST(xc;0;1 ST(E6:E1
3 N ORM.I
N
AVE RAGE D5+dx =FREQUE NORM.DI CHISQ.TE
= = = { = =
4 xN xc xb freq freqXpP
5 -0.38 -3 -3 0
6 -1.72 -2.75 -2.5 4 4.7 0.51
17 1.41 2.75 3 6 4.7
18 -0.64 3 0
1028 -1.14
Fig. 8.13 (S) Table layout used to generate normally distributed random numbers and to deter-
mine their experimental (column E) and theoretical (column F) frequencies; the Chi2 test is
performed in G6
A spreadsheet function for a Gaussian random generator

To create a set of normally distributed numbers in EXCEL, we use
Norm.Inv(Rand(); 0; 1)
corresponding to sct.norm(0,1).ppf (npr.rand()) in Python. It claims

to return random numbers distributed according to the probability density of a Gaus-
sian bell curve with the mean value 0 and the standard deviation 1 (standard normal
distribution).
We check this claim in Fig. 8.13 (S) by entering the above function into 1024
cells named x N , thus generating 1024 random numbers. A frequency distribution
freq is calculated from them for the interval boundaries x b and displayed in a
diagram over the interval centers x c (see Fig. 8.10a). The expected theoretical fre-
quency distribution freqXpP is calculated with the probability density in the center
of the intervals.
For a normal distribution with mean value 0 and standard deviation 1 (standard
normal distribution) , frequencies as those for freqXpP in column F are expected.
The corresponding formula (F3) is composed of three terms:
– NORM.DIST(x C ;
0; 1; 0) , the probability density pdf in the center x c of the
interval, where the second and third positions indicate the mean value (here,
0) and standard deviation (here, 1) of the Gaussian curve. The fourth position
(FALSE or 0, TRUE or 1) determines whether the probability density or, otherwise,
the distribution function is to be returned.
– The width dx of an interval (here, 0.5).
– The total number N of random numbers (here, 1024).
These expected frequencies are shown in Fig. 8.10a as a polyline. The Chi2
test in G6 gives a value of 0.51. So, we have no reason to doubt that
our function NORM.INV(RAND();0;1) produces standard normally distributed (=
A B C D E F 1 Sub SumHist() 1
1 50 15000 =SUM(fSum) 2 For rep = 1 To 100 2
}
x;xb) 3 Application.Calculation = xlCalculationManual 3
NCY( ro
D() EQUE from Mac
2 =RAN {=FR 4 For r = 4 To 13 4
3 x Ib Freq fSum 5 Cells(r, 5) = Cells(r, 5) + Cells(r, 4) 5
4 0.30 0.1 3 1511 6 Next r 6
5 0.31 0.2 6 1530
7 Application.Calculation = xlCalculationAutomatic 7
11 0.73 0.8 9 1523
12 0.17 0.9 7 1468 8 Next rep 8
13 0.71 6 1538 9 End Sub 9
53 0.76 10 10
Fig. 8.14 a (left, S) The frequency of 50 numbers x in column A is determined in column D.

The sum of the frequencies in column E has been obtained with SUB SumHist in b (P). b (right,
P) Procedure adding up the frequencies in column D into f Sum . Two nested loops apply: (FOR r
= …) adds up the frequencies once, the superordinate loop (FOR rep = …) repeats this process.
CELLS(4,5) IN THE LOOP (FOR R = 4 …) in the procedure is cell E4 in the spreadsheet
Gaussian-distributed) random numbers, and that their frequency of occurrence is

theoretically described using the probability density in the center of the intervals.
We shall use this function in later chapters with various standard deviations to
simulate noise during measurements.
Questions
The distribution of 102,400 random numbers is displayed in Fig. 8.10b. How

can you supplement the EXCEL solution in Fig. 8.13 (S) designed for 1024
random numbers so as to also get such a large number?23
Summing Up Frequencies with a VBA Routine

It is often practical to set up a calculation model in which frequencies are first
determined on a small sample. If the spreadsheet calculation runs without errors,
then the statistics can be made more extensive by repeating the random experiment
several times and adding up the frequencies found. An example is given in Fig. 8.14.
The procedure SumHist adds, in the inner loop (r = …), the values in Freq to
the values in f Sum and repeats this 100 times. Before the (r = …) loop, the auto-
matic calculation is switched off (line 3), because, otherwise, all random numbers
would be generated anew and every entry in the spreadsheet and all frequencies
recalculated. This would make the check number in E1 unequal to the total number
of sample points as it should not be. Such a discrepancy could therefore be used to
identify incorrect programming. After finishing the (r = …) loop, the automatic
calculation is switched on again to generate the random numbers x anew.
Questions
Questions concerning Fig. 8.14:
23 With a rep-log procedure adding up the frequencies in freq 100 times. Continue reading!
How often was SumHist of (P) called to yield the results in Fig. 8.14a (S)?24
Why is the automatic calculation in SUB SumHist Fig. 8.14b (P) switched off
before adding the frequencies?25
What would happen if the automatic calculation were not switched on again
after a summation of the frequencies?26
Task Write a log procedure for repeating the random experiment of the previous
task in Fig. 8.13 one hundred times and add up the frequencies! Do not forget to
switch off the automatic calculation while the current frequency distribution is
added to the sum! Next, create a frequency distribution and do a Chi2 test!
8.6 Random-Number Generator, General Principle
A random-number generator for a desired probability distribution pdf can be

created if the inverse function ppf of the associated cumulative distribution
function cdf exists and is “fed” with random numbers equally distributed
between 0 and 1.
Why do Norm.Inv(Rand();0;1) and sct.norm.ppf(npr.random) generate nor-

mally distributed random numbers?
The answer to this question is made plausible with Fig. 8.15. There, the probability
density pdf of the standard normal distribution is represented with a bell curve, and
the kink points of the right angles lie on the distribution function cdf . The interval
limits of the uniformly subdivided x-axis are transferred to the y-axis by the cdf of
the normal distribution.
The width of the intervals on the y-axis is y = dcdf (x)/ dx)·x. So, if the y-
axis is “fed” with random numbers between 0 and 1 and transferred by ppf to
the x-axis, a distribution of the x-values corresponding to dcdf (x)/dx results, i.e.,
according to the derivative of the cdf , and thus a distribution proportional to the
associated probability density pdf (x) = dcdf (x)/dx.
24 SUB SumHist ran three times. Each time, the random experiment was repeated one hundred
times, so that the summed frequency distribution captures 3 × 100 × 50 = 15,000 numbers (see
cell E1 in Fig. 8.14a).
25 Every single summation would otherwise lead to a recalculation of all random numbers, and thus
to changed frequencies of occurrence. This would be noticed thanks to the fact that, in cell E1 of
Fig. 8.14a (S), there would be no number corresponding to 3 x 100 x 50. The sum in cell E1 is
therefore a check as to whether an error has been made in the spreadsheet or in the procedure.
26 The random numbers would not be determined anew, and the same frequencies would always
be added up.
1 0.98
0.96
0.94
0.92
0 0.90
-2 0 2 1.2 1.4 1.6 1.8 2.0
x
Fig. 8.15 a (left) Bell curve: Normal distribution. b (right) Blow-up of a, y = (dy/ dx)·x
A B C D E F G H 3000
1 dx 0.3 N 10000 =SUM(fSum)
b)} fSum fSum
x
)) CY(x;x edure )*N*d
AND( D4+xb)/2 4+dx QUEN -proc (-xc/1
2 =-LN(R =( =D {=FRE by log =EXP fTheo, Chi² test=0.67
2000
3 x xc xb freq fSum fTheo Chi² test
4 4.60 0 0 0
5 1.30 0.15 0.3 301 2579 2582 0.67
6 0.24 0.45 0.6 196 1932 1913 1000
13 0.58 2.55 2.7 16 254 234
14 0.15 2.85 3 14 171 174
15 0.04 47 465
1003 0.34 0
0 0.6 1.2 x 1.8 2.4 3
Fig. 8.16 a (S) Generating exponentially distributed random numbers. b (right) Display of data
from a
Distributions generated with simple mathematical functions

We generalize our findings: The inverse function ppf (p) of a distribution function
cdf (x) can be used as a random generator for the associated probability density ppf (x)
:
y = PI (x) (distribution function, cd f )

dy
= p(x) (probability density, pd f )
dx
x = PI−1 (y) (ppf, inverse of the distribution function)
All distributions for which the inverse function of its distribution function can
be built can be generated using standard functions. We still have to make sure that
the argument range of the inverse function is correctly covered by the range from
0 to 1. Examples can be found in Table 8.13.
8.6 Random-Number Generator, General Principle 347
Table 8.13 Some functions, their integrals, and the inverse functions of the integral, suited to
generating random number distributions (EXCEL)
SGI [ FGI [ SSI S
(TXLGLVWULEXWLRQWR 5$1'
1RUPDOGLVWULEXWLRQ 1250',67 ;;0;'758( 1250,19 5$1'
([SRQHQWLDO ([SRQHQWLDO /1 5$1'
QDWXUDOORJDULWKP
&RVLQXV 6LQXV $UFXVVLQXV 5DQG

1
[ IRU[ WR [t ( ())2
1
[tIRU[ WR [u ( ())3
2)
1/( ⋅ (1 + $UFXVWDQJHQV [ 7DQ ˭ 5DQG
&DXFK\/RUHQW]
Questions
How do you construct a distribution p(x) = c·x 3 for 0 ≤ x < 1? How big is c?27
Exponential distribution
As an example, we construct a random-number generator that produces a decreasing
exponential distribution from x = 0 to ∞. We know that
pdf(x) = exp(−x) → cdf(x) = 1 − exp(−x) → ppf(p) = −ln(x)
So, –IN(RAND()) should do the job in EXCEL; it is implemented in Fig. 8.16a (S)
with the results shown in b.
Questions
How many random numbers are generated in Fig. 8.16a under the name x?28
What is the total size of f Sum displayed in Fig. 8.16b?29
How was f Sum most likely calculated from freq?30
Task Generate random numbers that are distributed according to a cosine arc
and check the results with a Chi2 test!
√ p(x) is P(x) = c/ 4· x ; P(1) must be 1 → c = ¼. The inverse

27 The antiderivative of 4
function of P(x) is 4 Rand(). The random function is thus [=RAND()ˆ0.25] in EXCEL or

np.random(N)**0.25 in Python.
28 The array x comprises 1000 numbers (rows 4 to 1003).
Sum comprises N = 10,000 numbers (cell F1).

29 The array f
30 By a rep-log procedure summing up freq 10 times (similar to Fig. 8.14b).
Task Generate random numbers distributed between 0 and 1 according to p(x)

= 3x 2 . Determine the theoretically expected frequencies, approximated with the
probability density, Eq. 8.14, as well as exactly with the distribution function,
Eq. 8.15. Perform Chi2 tests to see if the formulas given in the table are correct
and for what number of random numbers the approximation of Eq. 8.14 is good
enough! In EXCEL, use a rep-log procedure that repeats the random experiment!
8.7 Diffraction of Photons at a Double-Slit
We simulate the diffraction of photons at a double-slit, intending to demon-

strate the wave-particle duality of light. For this, we need a random-number
generator distributing the impact points of the phonons on a screen according
to a cos2 probability density. The goal concerning programming is to learn
how to implement such a random-number generator using a finite polyline.
8.7.1 Physical Background: Wave-Particle Dualism
In Fig. 8.17a, b, the diffraction of electrons or photons at a double slit is simulated

to illustrate the wave-particle dualism. The diffraction image is created by many
particles that hit a screen behind the double-slit randomly. The image does not
consist of stripes behind each slit, but is rather an interference pattern with a
maximum behind the middle position between the two slits.
Fig. 8.17 a (left) Two diffraction images of ten photons each. b (right) Top: Diffraction image
of 1000 photons, bottom: Distribution of the x-coordinate of the 1000 photons (the y-coordinate is
evenly distributed)
8.7 Diffraction of Photons at a Double-Slit 349
In physical reality, the distance of the screen to the slits is large against the
distance between the slits, and the independent variable of the interference pattern
is the angle with respect to the mid-perpendicular of the double-slit.
We may imagine the screen upon which the diffraction image is captured as
a gelatine film with silver grains or as a modern CCD detector with pixels. The
impact of a photon is shown by the fact that a single silver grain or a single pixel
is activated. Figure 8.17a shows two typical images after the impact of 10 photons.
Such experiments in the real world show that light is “granular”, i.e., consists of
energy packets that are locally confined.
The events in Fig. 8.17a seem to be evenly but randomly distributed over the
entirety of the detector surface. However, after 1000 photons have been detected,
a pattern like that in Fig. 8.17b emerges; the positions x = −1/2 and x = 1/2 have
never been hit by any photon. This is a consequence of the wave character of light
resulting in a diffraction image consisting of a central maximum and secondary
maxima of the same width. The distance between the maxima is determined by
the reciprocal of the distance between the two slits. The ratio of the intensities
depends on the width of a single slit. Here, we have arbitrarily chosen a ratio of 2
to 1 between the central maximum and the side maxima.
If … then … else …
The x-coordinates x Cos generated by a cos2 -generator are to be distributed according
to the diffraction figure:
– 50% in the central maximum of width 1 (position at x = 0),

– 25% in the right side-maximum of width 1 (position at x = 1), and
– 25% in the left side-maximum of width 1 (position at x = –1).
This can be achieved with a random number rnd between 0 and 1 and a logical query:
If 0.5 < rnd, then x = x Cos (in 50% of cases).

else If rnd < 0,75, Then x = x Cos + x 0 (in 75% – 50% = 25% of cases).
else, x = x Cos - x 0 (in the remaining 25% of cases).
Questions
How do the critical numbers in the IF query have to be changed when the
maxima are to occur at x 0 , x 0 + 1, and x 0 + 2 and intensity ratios of 6:3:1 are
to be obtained?31
In a spreadsheet, we can make a “living picture”; with each change of the

spreadsheet (e.g., when an already empty cell is “deleted”), the coordinates are
31 If rnd < 0.6, then x = x Cos + x 0 ; Else, if rnd < 0.9, then x = x Cos + x 0 + 1; Else, x = x Cos +
x 0 + 2.
recalculated and the points in the diffraction image with ten photons, Fig. 8.17a,
jump around erratically. The attentive observer may suspect that the points stay
away from the straight lines x = −1 and x = 1, but this observation is not
convincing.
In Python, we can create such a “living picture” through animation as shown
in Sect. 6.2.5.
8.7.2 Cos2 Distribution
Our task is to build a random generator that distributes the particles according
to the diffraction pattern. We only want to simulate the central maximum of the
diffraction figure at x = 0 and the first two secondary maxima around −x 0 and
+ x 0 (here x 0 = 1, arbitrary units on the screen). The intensity distribution in
each maximum shall be approximated by the same cos2 probability density (see
Fig. 8.18a),
Cos Sq_ pd f (x) = 2 · cos 2 (π x) (8.16)
however, with different amplitudes.

The distribution function cdf (x) is the integral of this probability density func-
tion pdf (x) with an argument range from -0.5 to 0.5 and a value range from 0 to
1:
sin(2π x) 1
Cos Sq_cd f (x) = +x+ (8.17)
2π 2
It is shown in Fig. 8.18a (“CosSq_cdf”).
Fig. 8.18 a (left) Functions related to the cos2 distribution, data from Table 8.15. b (right) The
frequencies of occurrence of 100,000 outputs of the function CosSq ppf(rand)) are actually cos2
distributed
Questions
How big are the values CosSq cdf (−0.5), CosSq cdf (0), CosSq cdf (0.5) in
Eq. 8.1732
How big is the area under the curve CosSq pdf in Fig. 8.18a?33
For our random-number generator, we need the inverse function of

CosSq cdf(x) in Eq. 8.17.
Solution in three steps
– We create a function that distributes cos2 -distributed values in the range −0.5 to
0.5.
– We generate ten points according to the desired diffraction image (as in Fig. 8.17a).
– We repeat the random experiment with the ten points one hundred times and
determine the frequency distribution of the x-values (as in Fig. 8.17b).
The gradual emergence of the diffraction pattern can be demonstrated with animation.
CosSq_pdf (x) p(x) = 2*cos2 (π·x)

CosSq cdf (x) integral of CosSq pdf (x)
CosSq ppf (p) inverse of CosSq cdf (x)
xI array of x, 33 values from −0.5 to 0.5
pI CosSq cdf (x I )
m[0], … slope of pI in interval x I [0] to x I [1]
x Cos x-values, random according to a cos2 distribution.
8.7.4 Python
We define the three functions related to the cos2 distribution:
– CosSq pdf(x) , probability density function, argument range from −

0.5 ≤ x ≤ 0.5, normalized so that the area under the curve is 1.
32p(–0.5) = 0; p(0) = ½; p(0.5) = 1.

33(Area under cos2 ) = (triangle (0.5 – (-0.5))·2/2) = 1. This is the condition for a probability
density function.
Table 8.14 Python functions related to the cos2 distribution

1 def CosSq_pdf(x):
2 return 2*(np.cos(x*np.pi)**2)
3
4 def CosSq_cdf(x):
5 return np.sin(2*np.pi*x)/(2*np.pi)+x+1/2
6
7 xI=np.linspace(-0.5,0.5,33)
8 pI=CosSq_cdf(xI)
9 m=((pI[1:]-pI[:-1])/
10 (xI[1:]-xI[:-1])) #Slope
11 #m[0] is slope of interval pI[0] to pI[1]
12
13 def CosSq_ppf(p):
14 for i in range(1,33):
15 x0=0
16 if pI[i-1]<=p<=pI[i]:
17 x0=xI[i-1]+1/m[i-1]*(p-pI[i-1])
18 return x0 #Why return within if?
– CosSq cdf(x), cumulative density function (distribution function), argument

range from −0.5 ≤ x ≤ 0.5, monotonously increasing from 0 to 1.
– CosSq ppf(p), percent point function, the inverse of cdf , argument range
0 ≤ p ≤ 1.
These are all displayed in Fig. 8.18a.

The three functions are defined in Table 8.14. The probability density function
CosSq pdf essentially returns a value of cos2 ; CosSq cdf is its antiderivative. For
the inverse of the distribution function (cdf ), a closed expression does not exist.
We therefore approximate it with a polyline with 33 vertices. CosSq cdf for the x-
values of the vertices is calculated in array pI and the slope within the 32 intervals
in m. The three arrays thus obtained are used to get the linear approximation for
Cos2 ppf , with p as the independent variable and x as the dependent variable, and
the slope 1/m between the vertices.
Questions
Is there a variant of CosSq ppf that can suffice with 5 if queries for the 32
intervals?34
How do you plot the results of Table 8.15?35
34 See the program code in Table 8.16.

35 Plot with: plt.plot(xC,hist[0],’kx’); plt.plot(xC,theo,’k- ‘).
Table 8.15 N = 100,000 outputs of the cos2 -generator

19 N=100000
20 rn=npr.random(N)
21 xCos=list(CosSq_ppf(r)for r in rn)
22 dx=0.05 #Interval width
23 xb=np.arange(-0.5,0.5+dx,dx) #Endpoint not included
24 xc=(xb[1:]+xb[:-1])/2
25 hist=np.histogram(xCos,xb) #Empirical frequencies
26 theo=N*dx*CosSq_pdf(xc) #Theoretical frequencies
Table 8.16 Interval search with five queries

1 def CosSq_ppf2(p):
2 iLow=0
3 iHigh=32
5 iMid=np.int((iLow+iHigh)/2)
6 if p <= pI[iMid]:iHigh=iMid
7 else:iLow=iMid
8 x0=xI[iLow]+1/m[iLow]*(p-pI[iLow])
9 return x0
CosSq ppf is used in Table 8.15 to create 100,000 random numbers, the
frequency distribution of which is displayed in Fig. 8.18b, together with the
theoretical curve obtained with CosSq ppf .
Interval search with five queries

Table 8.16 realizes the search in 32 intervals with 5 queries.
8.7.5 Excel
User-defined spreadsheet function

We develop a user-defined spreadsheet function that approximates the random-
number generator for a cos2 -distribution piecewise linearly (see Fig. 8.19 (P)), as has
been done in the Python program. Remember: a user-defined spreadsheet function
must be in a module. It must not be assigned to a specific worksheet (see Exercise
4.9).
Global data arrays for x I , pI , and m are defined at the top of the VBA sheet, in
lines 1 to 3 of Fig. 8.19 (P), and can be called by all procedures and functions in
the same VBA sheet or module. They must be initialized with the correct numbers
before the first call of the function.
1 Public xI(32) As Single Sub init() 14

2 Public pI(32) As Single xI(0) = -0.5 15
3 Public m(32) As Single xI(1) = -0.46875 16
4 …….. 17
5 Function CosSq_ppf(p0) xI(32) = 0.5 18
6 For i = 1 To 32 19
7 If (pI(i - 1) <= p0 And p0 <= pI(i)) Then _ pI(0) = 0 20
8 co = xI(i - 1) + m(i) * (p0 - pI(i - 1)) pI(32) = 1 21
9 Next i 22
10 CosSq_ppf = co For i = 1 To 32 23
11 End Function m(i) = (xI(i) - xI(i - 1)) / (pI(i) - pI(i - 1)) 24
12 Next i 25
13 End Sub 26
Fig. 8.19 (P) Function cossq ppf approaches the inverse function of pI (x), CosSq cdf (x) in
Eq. 8.17, piecewise linearly as a polyline. The global fields x I , pI , and m must be initialized with
the correct values, here, with SUB init, before the first call of the function. The text of the statements
in init is generated in a spreadsheet (see Fig. 8.20 (S), columns E and F)
The function COSSQ_PPF corresponds to the Python function of the same name.
The (FOR i = ..) loop queries in lines 7 and 8 for each of the 32 intervals whether
p0 lies in this interval. If this is the case, the value of the function is calculated
as a linear interpolation between the vertices of the segment, the coordinates of
which are stored globally in the data arrays x I (32) and pI (32). The slopes in the
intervals are in the global array m(32).
Questions
Is there a variant of Cos2 ppf that can suffice with five If queries for the 32
intervals?36
VBA code generated in a spreadsheet

Initialization of the global arrays x I and pI is done in a total of 66 lines in the procedure
init(), also shown in Fig. 8.19 (P). The array m(I) contains the slope of the curve
between points i − 1 and i and is calculated from x and pI within init (lines 23 to
25).
We could enter all 66 lines by hand, which is, of course, tedious and unpleasant,
although it would not be much more time-consuming than tracking down errors in
spreadsheets and other programs. But there is a more elegant way: we let EXCEL
work for us. In various tasks, we have had a VBA procedure write formulas into a
spreadsheet. Text in cells preceded by an = sign is interpreted as a formula. Now,
we will simply do it the other way around: A text is assembled in a spreadsheet and
copied into a VBA procedure to be interpreted there as a formula. This is done in
Fig. 8.20 (S).
36 Compare the Python code in Table 8.16.

A B C D E F G H
1 0.031 =2/32
0.5
)+xI+
2/PI( xI pI
1 I()*xI)/ n&")
= "&
n&")
= "& pf(B5
)
+$A$ (2*P ("& ("& Sq_p
2 =A4 =S IN ="x I ="pI =cos
3 xI pI n Array x Aray y
4 -0.500 0.000 0 xI(0) = -0.5 pI(0) = 0 0.000
5 -0.469 0.000 1 xI(1) = -0.46875 pI(1) = 0.00020041090 0.000
6 -0.438 0.002 2 xI(2) = -0.4375 pI(2) = 0.00159404009 0.000
35 0.469 1.000 31 xI(31) = 0.46875 pI(31) = 0.9997995890 0.000
36 0.500 1.000 32 xI(32) = 0.5 pI(32) = 1 0.000
Fig. 8.20 (S) In columns A and B, there are 33 points on the curve pI (x), Eq. 8.17; in columns D,
E, and F, the VBA code for initialization of the arrays x I and pI is generated from columns A and
B. This text is to be copied into the VBA editor, SUB Init in Fig. 8.19 (P)
A program is just a text, composed of code words, that is translated into

computer instructions.
The column vectors x I and pI correspond to the lists with the same names as in
the Python program (Table 8.14). The code for the VBA -function of the inverse
ppf (p) of the distribution function cdf (x) (x I (0) = …, pI 0) = …) in columns E and
F is generated in the spreadsheet with text processing and copied into SUB init()
IN Fig. 8.19 (P), which initializes the global data arrays x I , pI , and m.
In columns E and F of Fig. 8.20 (S), text corresponding to VBA code is gener-
ated. The spreadsheet formulas in the individual cells consist of text elements and
numbers, e.g., in E5: [="xI("&n&") = "&xI], and yields [xI(1) = −0.4687]. The
ranges E4:E36 and F4:F36 written in this way are then transferred to the Visual
Basic editor by text copying.
Alac The values can be calculated in the routine itself using Eq. 8.17.
Mag Yes, that is possible. But by detouring via the spreadsheet, we have
practiced the way in which code is generated as text in a spreadsheet and trans-
formed into formulas. Code is nothing more than structured text interpreted by a
programming language interpreter.
8.7.6 Simulation in a Spreadsheet
We will simulate the evolution of an interference pattern in a spreadsheet with two

figures like the upper two in Fig. 8.17. Ten new photon impacts are shown every
second in a snapshot (left); they are accumulated for the right figure.
Random experiment with 10 to 1000 photons

In Fig. 8.21 (S), we calculate the coordinates of ten photon impacts to be presented
in the snapshot. The preliminary to the x-component is calculated as x Cos with the
user-defined spreadsheet function [= cossq ppf(rand()]. The random numbers x 0
1 =IF(rn<0.5;xCos;IF(rn<0.75;xCos+1
xC 1004 400;bnd) 0.16 1000
0.75;
N D()) s;IF(rn< 4)
pf(RA ;xCo (J5;J
Sq_p RAND() IF(rn<0.5 RAND() RAGE $J$1 QU
2 =cos = = = =AVE =J4+ =FRE
3 xCos rn x y x.1000 y.1000 bndC bnd freq
4 -0.22 0.09 -0.22 0.08 -0.28 0.51 -2.00 -2.00 0
5 -0.09 0.84 -1.09 0.75 -0.22 0.29 -1.92 -1.84 0
13 -0.32 0.90 -1.32 0.91 -0.04 0.56 -0.64 -0.56 20
29 0.98 0.23 1.92 2.00 0
30 1.14 0.81 2.00 0
31 -0.07 0.86
Fig. 8.21 (S) Ten random points within the diffraction image of a double-slit are generated in C:D.
The total 1000 points in F, G have been accumulated (a 100 times) by SUB More10 in Fig. 8.22 (P).
The formula in K2 is = FREQUENCY (x.1000;bnd)
are distributed over the three maxima at 0, 1, and −1 by means of a second random
number rn with (see formula in C2 and C1).
x = [= IF(0, 5 < rn; xCos;

IF(m < 0, 75; xCos + 1;
xCos − 1))]
according to the specified ratio of the intensity of the maxima 0.5:

0.25: 0.25. Remember: the structure of the logical query in EXCEL is
[=IF(LOGICAL_TEST; VALUE_IF_TRUE; VALUE_IF_FALSE)].
The y-component is distributed uniformly between 0 and 1. The points (x, y)

are shown in Fig. 8.17a; they change their position with every change in the
spreadsheet.
Sub More 10
The upper picture in Fig. 8.17b shows the accumulated photon impacts with the
coordinates (x.1000, y.1000) from Fig. 8.21 (S). This range (columns F and G)
is successively filled by SUB More10 in Fig. 8.22 (P). It transfers the ten random
coordinate pairs (x, y) from columns C and D consecutively to columns F and G
using an index r 2 updated in G1 of the spreadsheet.
Sub Run, random experiment with 1000 photons

With a rep-log procedure, SUB Run in Fig. 8.23 (P), we repeat the random experiment
More 10 with 10 photons 100 times and display the increasing number of points in
the cumulating diagram, Fig. 8.17b (top), getting at recognizing the pattern better
and better.
Over the course of time, we recognize the diffraction pattern more and more
clearly. The frequency distribution of the 1000× values in Fig. 8.17b reveals the
cos2 distribution in the main maximum and in the two side maxima.
1 Sub More10() 1
2 r2 = Sheets("Coord").Cells(1, 7) 'G1 2
3 Application.Calculation = xlCalculationManual 3
4 For i = 0 To 9 4
5 Sheets("Coord").Cells(i + r2, 6) = Sheets("Coord").Cells(i + 4, 3) 'C4 -> F4, etc. 5
6 Sheets("Coord").Cells(i + r2, 7) = Sheets("Coord").Cells(i + 4, 4) 'D4 -> G4, etc. 6
7 Next i 7
9 Wait 9
10 Sheets("Coord").Cells(1, 7) = r2 + 10 'index for columns F und G 10
11 End Sub 11
Fig. 8.22 (P) SUB More10 writes the ten random coordinate pairs in columns C and D of Fig. 8.21
(S) successively into columns F, and G. SUB Wait is reported in Fig. 8.23(P). The pointer r 2 for the
next free row is updated in CELLS(1,7) = G1
1 Sub Run() Sub Wait() 9

2 init Dim m As Integer 10
3 Sheets("Coord").Range("F4:G6000").ClearContents h = Hour(Now) 11
4 Sheets("Coord").Cells(1, 7) = 4 m = Minute(Now) 12
5 For n = 1 To 100 s = Second(Now) + 1 13
6 More10 waittime = TimeSerial(h, m, s) 14
7 Next n Application.Wait waittime 15
Fig. 8.23 (P) In the master procedure SUB Run, the coefficients for CosSq ppf are generated by
calling SUB Init of Fig. 8.19 (P); the old coordinates are deleted in line 3. SUB More10 (Fig. 8.22
(P)) is called in a loop a hundred times
SUB Run is a log procedure that first deletes the old data in columns F and G
in Fig. 8.21 (S). These are the coordinates of the points in Fig. 8.17a (top), so
that the chart is now empty, and then calls the procedures More10 and Wait a 100
times so that the chart fills up again (within 100 s) with 1000 points. We may call
it a “master procedure”, because it controls the program flow.
SUB Wait in Fig. 8.23 (P) stops the calculation for 1 s, so that the viewer can
better follow how Fig. 8.21b top is filled up with new points and the frequency
distribution in Fig. 8.21b bottom takes shape.
Questions

The VBA procedure reads the index r 2 of the next free line in the range
into which the new coordinates of the photon impacts are to be written from
the spreadsheet (G1). Alternatively, the index r 2 could also be updated in the
main program SUB RUN() in Fig. 8.23(P). What is the difference between the two
alternatives?37
How would you change Sub More10 if only 5 photons are to pass the double-
slit?38
8.8 Chi2 Distribution and Degrees of Freedom
Distributions of χ 2 values, obtained from statistical experiments on

normally-distributed random numbers, are compared with theoretical distri-
butions with appropriate degrees of freedom. The degree of freedom is the
number of intervals reduced by 3 if the mean and standard deviation are
estimated from the sample and reduced by 1 if they are fixed a-priori.
In this experiment, we generate N = 1000 standard-normally distributed random

numbers, determine their frequencies of occurrence in 10 intervals, and com-
pare them in Chi2 tests with theoretically expected frequencies obtained from two
models.
In the first model, we use the values 0 and 1 for mean x m and standard deviation
x Sd so that no parameters of the distribution are estimated from the sample, and
the degree of freedom is dof = 9 = number of intervals – 1. The reduction by 1
is due to the fact that the frequency in the last interval is not free but determined
by the total number and the frequencies in the other intervals.
In the second model, we estimate x m and x Sd from the sample so that dof =
9 – 2 = 7. We repeat the statistical experiment with 1000 data N c = 10,000 times
and determine the frequencies of occurrence of the values of Chi2 .
The result of the statistical experiments is shown in Fig. 8.24a. The lines
represent frequencies obtained with the theoretical probability density func-
tion of Chi2 , available from the library scipy.stats, usually imported as
sct, here, sct.chi2.pdf(xc,dof = 9) and sct.chi2.pdf(xc,dof
= 7), where x c are the centers of the sorting intervals.39 We see that the results
of our simulation fit very well with the theoretical curves. So, we have got the dof
right.
Fig. 8.24b presents the pdf of the Chi2 distribution for dof = 5 and dof =
15. The shaded areas represent the results of Chi2 tests. They are the accumulated
37 Update of r 2 in Sub Run(): initial index always starts at a fixed value of r 2 ; old data are
overwritten.
Read r 2 from the spreadsheet: old data remain; new data are appended below the already
existing data.
38 Change: line 4 → i = 0 to 4; line 10 … r2 + 5. The program becomes more flexible if an
additional variable N is introduced: i = 0 to N-1; … r2 + N.

39 The sorting intervals are often called bins into which the data set is to be sorted.
8.8 Chi2 Distribution and Degrees of Freedom 359
Fig. 8.24 a (left) Empirical (x, ) and theoretical (…, —) distribution of values of Chi2 for 7 and
9 degrees of freedom. b (right) Pdf of Chi2 ; the P-value returned in a Chi2 test corresponds to the
shaded area (P = 0.32 for dof = 5 and P = 0.7 for dof = 15)
probabilities that a sample drawn from the population yields a higher value of Chi2
when compared to the theoretical distribution of the population.
Keep in mind that, for 10 sorting intervals for normally distributed random
numbers,
– dof = 9 when x m = 0 and x Sd = 1 are a-priori fixed,

– dof = 7 when x m and x Sd are estimated from the sample.
What is the Chi2 test good for?

We check the hypothesis that empirical frequencies of occurrence arise from a the-
oretical distribution. The error probability to reject that hypothesis is given by the
result of a Chi2 test.
8.8.1 Data Structure, Nomenclature
N number of random numbers

XN sample of N random numbers, standard-normally distributed
xm estimated mean of the sample
x Sd estimated standard deviation of the sample
xb interval boundaries, including −∞ and ∞, defining 10 sorting inter-
vals (“bins”)
fO observed frequencies of x N in the sorting intervals
fx 9 frequencies expected for a normal distribution norm(0, 1) in 10 intervals
fx 7 frequencies expected for norm(x m , x Sd ) with x m and x Sd estimated from
the sample.
N rep number of repetitions of the statistical experiment
dof degree of freedom
cS 7 array for storing N rep Chi2 values comparing f O and fx7

cD7 theoretically expected frequencies of cS 7, pdf of Chi2 for dof = 7
cS 9 array for storing N c Chi2 values comparing f O and fx 9 , pdf of Chi2 for dof
=9
cD9 theoretically expected frequencies of cS 9 .
8.8.2 Python
The basic set-up is given in Table 8.17. 10 intervals from −∞ to ∞ are specified
and the frequencies of occurrence for a standard-normal distribution therein for N
= 1000.
The interval boundaries x b specified in lines 6 and 7 range from −∞ to ∞, to
capture the whole value range of normally distributed random numbers. They are
reported in the lower cell of Table 8.17. The frequencies expected in the intervals
are calculated with the cumulated density function at the interval boundaries. For
the first and last interval, we have to consider that cdf (-∞) = 0 (line 9) and
cdf (∞) = 1 (line 12), which is true, by the way, for any cdf .
The program in Table 8.18 organizes the repetition (N rep = 10,000 times) of the
statistical experiment in a for-loop. For every iteration i, a sample x n of N random
numbers is generated, and their frequency distribution f O , mean x m, and standard
deviation x d are determined. The frequencies f O are compared with a Chi2 test to:
– fx9, expected when 0 and 1 are taken as the mean and standard deviation,
calculated before the loop (Table 8.17), lines 8 to 11;
Table 8.17 Expected frequencies fx 9 of a normal distribution in 10 intervals with boundaries x b ,

x m = 0, x Sd = 1 set a-priori
3 N=1000 #Size of the set of random numbers
4 db=0.5 #Width of the sorting intervals
5 xbb=np.arange(-2.0,2.0+db,db)
6 xb=np.zeros(len(xbb)+2) #Interval boundaries
7 (xb[0],xb[1:-1],xb[-1])=(-np.inf,xbb,np.inf)
8 fx9=np.zeros(len(xb)-1)
9 fx9[0]=sct.norm(0,1).cdf(xb[1])*N #Below first bound.
10 fx9[1:-1]=(sct.norm(0,1).cdf(xb[2:-1])
11 -sct.norm(0,1).cdf(xb[1:-2]))*N
12 fx9[-1]=(1-sct.norm(0,1).cdf(xb[-2]))*N #Above last bound.
xb [-inf -2.00 -1.50 -1.00 -0.50 0.00

0.50 1.00 1.50 2.00 inf]
8.8 Chi2 Distribution and Degrees of Freedom 361
Table 8.18 Histograms of cS9 and cS7, together with the theoretical probability densities CD9Th
and CH7Th of Chi2 for dof = 9 and 7 degrees of freedom
13 Nrep=10000 #Number of repetitions
14 cS9=np.zeros(Nrep)
15 cT9=np.zeros(Nrep)
16 cS7=np.zeros(Nrep)
17 for i in range(Nrep):
18 xn=npr.randn(N) #Std-normal distribution
19 fO=np.histogram(xn,xb) #Empirical frequencies
20
21 ChiSq=sct.chisquare(fO[0],fx9,ddof=0)
22 cS9[i]=ChiSq[0]
23
24 fx7=np.zeros(len(fO[0])) #For expected frequencies
25 xm=np.average(xn)
26 xd=np.std(xn,ddof=1)
27 fx7[0]=sct.norm(xm,xd).cdf(Ib[1])*N
28 fx7[1:-1]=(sct.norm(xm,xd).cdf(Ib[2:-1])
29 -sct.norm(xm,xd).cdf(Ib[1:-2]))*N
30 fx7[-1]=(1-sct.norm(xm,xd).cdf(Ib[-2]))*N
31 ChiSq7=sct.chisquare(fO[0],fx7,ddof=2)
32 cS7[i]=ChiSq7[0]
33 xbC=np.linspace(0,20,21) #For distrib. of Chi²
34 xc=(xbC[1:]+xbC[:-1])/2 #Centers of intervals
35 CD9=np.histogram(cS9,xbC) #Empirical freqs.
36 CD9Th=sct.chi2.pdf(xc,df=9)*Nrep #Theoretical freqs.
37 CD7=np.histogram(cS7,xbC)
38 CD7Th=sct.chi2.pdf(xc,df=7)*Nrep
– fx7, expected when x m and xSd are estimated from the sample, calculated
individually for every iteration within the loop.
The results of the Chi2 tests are stored in cS9 and cS7. The frequency distributions
CD9 and CD7 thereof are determined in Table 8.18. The results are shown in
Fig. 8.24a.
Explain the following broom rules:
1. Chance is blind and checkered

2. Always one more! But of what and than what? (Concerning frequency
distribution.)
3. Come to a decision! Sometimes, it will be wrong.
Frequencies of occurrence
Initial situation: In the range A1:A1000 of a spreadsheet, named data, there are
1000 numbers. In range D2:D10, three interval boundaries I b = 0.1; 0.2 and 0.3 are
specified.
4. How many intervals are defined by the interval boundaries?

5. Which numbers are captured in the second and last intervals?
6. The frequencies are to be calculated in a column range starting with E2. Which
range must be activated for the spreadsheet function FREQUENCY if all numbers
are to be sorted into intervals? With which “chord” do you complete the formula
input?
7. Over which x values should the frequencies be displayed in a diagram?
8. We want to determine the same frequency distribution with the numpy function
np.histogram(data; Ibnp ) . How are the interval boundaries Ibnp to be defined?
The 1000 numbers are now supposed to be random numbers equally distributed
between 0 and 0.5.
9. Which equations do you use in EXCEL and numpy to generate such random
numbers?
10. What mean value do you expect for the 1000 numbers?
11. What frequencies do you expect in the first and last intervals for interval limits
of 0.1, 0.2, and 0.3 in the EXCEL function?
Normal distribution
Figure 8.25a shows the distribution function norm.cdf(0;1) and the probability den-
sity norm.pdf(0;1) of the standard normal distribution. Assume that you have
generated 1 million normally distributed random numbers and answer the following
questions within the reading accuracy of the figure:
1.0 1.0
norm.pdf(0;1) Distribution function
0.8 Probability density 0.8
norm.cdf(0;1)
p
0.6 0.6
0.4 0.4
0.2 0.2
0.0 0.0
-4 -2 0 2 x 4 -4 -2 0 2 x 4
Fig. 8.25 a (left) Distribution function norm.cdf(0;1) and probability density norm.pdf(0;1) of
the standard normal distribution. b (right) A distribution function and its probability density; the
distribution function is not displayed above, and the probability density not below x = 2
12. How many numbers may be expected to have precisely the value 0?
13. How many numbers are expected to have a value between −0.1 and 0.1?
14. How many numbers are expected to have a value between −0.5 and 0.5?
Distribution function and random-number generators
15. Within what range do you expect 95% of the output of a function generating
normally distributed random numbers?
16. Consider the EXCEL function NORM.INV(X;5;3)! What is its argument range?
Where is the maximum of the distribution that results if x = RAND()? How do
you generate 1000 such numbers in a Range B1:B1000? What is the equivalent
function for 1000 numbers in the numpy.random library?
17. Figure 8.25b shows an incomplete distribution function and an incomplete
probability density. Complete the two functions!
18. What are the inverse functions of y = cos(x) for x = 0 to π and of y = exp(−x)
for x ≥ 0? Which are their argument ranges and which are their value ranges?
19. Which distribution is generated by the spreadsheet function (arcus cosinus)
ACOS((RAND()-1)*2)?
User-defined spreadsheet function

In Fig. 8.26a, you find a spreadsheet layout for calculating a house-shaped polyline
(see Fig. 8.26b). Implement a function House(x) for calculating the y-values from
the x-values!
The entry in cell A25 of Fig. 8.26a is 6.38E–16, instead of the expected zero.
This deviation is due to the sum of the rounding errors in the binary addition of
dx = 0.10. In the user-defined spreadsheet function House(x), the value of x must
therefore be rounded, with ROUND(X,14).
A B
y
1 dx 0.10
2
2
x e(A6)
s
3 =A5+d =Hou
4 x y
1
5 -2.00E+00 0
6 -1.90E+00 0
24 -1.00E-01 1.9
25 6.38E-16 2
0
45 2.00E+00 0 -2 -1 0 1 2
x
Fig. 8.26 a (left) The values of y are calculated with a user-defined spreadsheet function House(x).
b (right) Graph of the House(x) function with the data from Fig. 8.26a
Fig. 8.27 a (left) Two shapes produced by a function House partly shown in b (right)
20. Why does the 20-fold addition of 0.10 to −2.00 not yield the smooth value
0.00, but rather 6 × 10–16 ?40
21. Which statement can be used to make the function House(x) zero for x ≤ −1
and x ≥ 1?
22. Write a user-defined spreadsheet function that produces a shape like the
“House” in Fig. 8.26b!
23. Figure 8.27a shows two shapes produced with a Python function House
partly displayed in Fig. 8.27b. Complete the code!
40 The numbers are binary coded so that decimal “smooth” numbers are not binary “smooth”.
Evaluation of Measurements
9
We simulate measurement experiments by assuming “true” (known) val-

ues of a quantity and masking them with different noise levels that enter
as standard deviation into normally distributed random numbers. Then, we
“forget” the true values and estimate the measurand’s now unknown value
from the noisy signal. ψ We know everything and play stupid. This way, the
students should get an impression of how unreliable measurements can be,
learn to indicate the reliability, and understand how measurement uncertain-
ties propagate. The importance of t statistics, connecting confidence interval
and confidence level (or error range and error probability), is illustrated in
several exercises.
9.1 Introduction: We Know Everything and Play Stupid
be found at the internet adress: go.sn.pub/E3k8Ps.
Alac Statistics? That’s all Greek to me.
Mag Ψ If in doubt, count!

Ψ Calculate with variances, report the standard error!
A simple measurement?
We shall determine the mass of a thin film on a glass substrate by weighing the
substrate before and after coating. Estimating the mass is straightforward. However,
stating a confidence interval requires this whole chapter, with t statistics and error
propagation.

366 9 Evaluation of Measurements
How should the poor student learn statistics?

Many students find it challenging to understand statistical statements. They are con-
fronted with mathematically accurate but awkward-looking formulations and have
to work according to strict rules.
In this chapter, the rules for statistical evaluation are not derived mathemati-
cally, but illustrated through simulations, particularly by multiple tests for hit rates.
By repeating a noisy measurement series several times, we determine how often
the standard error of the series’s mean value captures the true value to specify the
confidence level.
We know everything and play stupid

A writer of crime novels thinks up a criminal case, blurs the traces, and lets the
detective reconstruct the case. So do we, by specifying the “true” values ourselves,
turning them into measured values by adding noise, and then evaluating the noisy
data. Finally, we report the measurement result with specified uncertainty. The true
value may lie outside the confidence interval (the “error range” in laboratory jargon)
of the measurement series’ mean value. However, since we know the “true” value,
we notice this and can observe regularities for our error.
Sherlock Holmes or Inspector G. Lestrade?

Some crime novel authors who want to be well received by the reader let the sharp-
witted gentleman detective (Sherlock Holmes) or the curious lady detective (Miss
Marple) appear alongside police officers, who stubbornly follow the rules and come
to false results.
Tim To which type does our evaluation belong?
Alac To the smart detective, I hope.
Mag No, unfortunately not! We must proceed strictly according to the rules.
After all, intuition is often misleading when it comes to statistical issues. To get
the best of both worlds, we follow the broom rule: Ψ Trust (your intuition), but
verify!
Standard formulas
We apply the following formulas for a set of n measurements x = {x i }.
The mean x m of the set:
n
i=1 x i
xm = (9.1)
n
The variance var x of a set x = {x i }:

n
1
var x = (xi − xm ) 2
· , n P = 1 in this chapter (9.2)
n − nP
i=1
9.1 Introduction: We Know Everything and Play Stupid 367
where nP is the number of parameters estimated from the data set. In this chapter,
nP = 1, because only one parameter, the mean, is estimated. In Chap. 10, trendlines
are calculated and nP is equal to the number of their parameters.
The standard deviation of the set:
√
x Sd = var x (9.3)
The standard error of the mean x m of the set:

x Sd
x Se = √ (9.4)
n
Standard error range

The standard error has the same physical unit as the elements of the sample and
their mean value. Therefore, the mean x m plus-minus the standard error x Se is often
reported as the result of a series of measurements:
Standar d err or range = xm ± x Se (9.5)
The standard error x Se is calculated straightforward from Eqs. 9.2 (variance) and
9.3 (standard error).
Not to be forgotten: in your report, you also have to state the number of mea-
surements. If at least eight measurements have been made, the following broom rule
applies:
Ψ Two within and one out of (the standard error range).
The standard error range captures the true value of the measured quantity with a
probability of about 2/3.
C-spec error
For statistically more precise reasoning, we have to connect confidence level and
confidence interval by Student’s1 t statistics. We specify:
C − spec err or range = xm ± t · x Se (9.6)
For example, with eight measurements, the error range xm ± 2.4x Se cap-
tures the true value of the measurand with a probability of 95%. The interval
[xm − t x Se , xm + t x Se ] is called the confidence interval (or error range) of the
mean value x m and the associated probability (95% in the example) confidence level
C. The complementary probability (5% in the example) is called the error probability
E:
C + E = 100%
1 Student is the pseudonym of W.S. Gosset.

We call the error for a well-specified confidence level C-spec error (“confidence-
specified error”).
The parameter determining the t distribution is the degree of freedom dof defined
as
do f = n − n P (9.7)
where n is the number of measurements and nP the number of parameters estimated

from the data series. In the exercises in this chapter, nP is 1 because only 1 parameter,
the mean, is estimated from the data series. In the next chapter, coefficients of trend
lines are estimated from the data and nP > 1.
Multiple tests for obtaining hit rates

Statistical laws specify error probabilities for hypotheses. In this chapter’s exercises,
the hypothesis is that the estimated error range does not capture the true value. We
do not derive such laws from axioms, but apply multiple tests for hit rates instead.
They repeat statistical experiments multiple times and check whether the empirical
hit or miss rates are compatible with the corresponding theoretical probabilities.
Tim If this is not the case, then we have falsified the theoretical assumption.
Mag Or have made a programming error. More about that later.
Simulation-based t adaptation
Many statistical laws are based on assumptions that are often not justified in our
exercises or in real-life experiments, e.g., because the number of measurements is
too small or the noise is too large. In such cases, the formally calculated error ranges
may not exhibit the expected confidence level, e.g., in Exercise 9.9 where error
propagation plays a role. In Exercise 9.8, we present a method to get C-spec errors,
by adapting the t-factor so that a hit rate corresponding to a pre-specified confidence
level is achieved. This method is suited also for real-world experiments.
Calculate with variances, report the C-spec error!

Mathematical formulas concerning confidence are formulated for variances. Statis-
tical rules are based on properties of variances. We calculate with them to grasp the
basic mathematical dependencies. For the final result, we strive to report the C-spec
error because it is related to a confidence level and has the same physical unit as
the measurand. In this way, we simulate the error propagation in sums, products and
powers (Exercises 9.7 and 9.8).
Broom rules for measurements

We start with an exercise on a simple experiment, weighing a glass substrate, and
end with a more challenging one, determining the mass of a thin film on such a glass
9.1 Introduction: We Know Everything and Play Stupid 369
substrate. In between, we have to improve our understanding of statistics with the

following broom rules:
Ψ Mostly, but not always. No statistical statement is 100% certain.
Mag This is the fundamental rule of statistical reasoning. More rules:

Ψ Always round to relevant digits; to convey the result concisely! (Exercises
9.2 and 9.3).
Ψ Twice as good with four times the effort. The measurement inaccuracy is
halved if measurements are made four times as often (Exercise 9.4).
Ψ Two within and one out of. In one-third of all tests, the measurand’s true value
is outside of the standard error range if at least 8 measurements have been made
(Exercise 9.5).
Bad makes good even better, usually, but not always. Even measurement results
with a relatively large statistical uncertainty may improve the overall result by
entering a weighted mean (Exercise 9.6).
9.2 Weighing a Glass Substrate
The weighing process is simulated by adding noise to the glass substrate’s

true weight and estimating the weight from the noisy data. We determine the
standard error of the result and round the result to the relevant number of
digits.
9.2.1 Discussion on the Accuracy of a Balance
Mag We are going to determine the weight of a glass substrate with a typical
weight of 1 g. How precise is this procedure if the measuring precision of the
balance is 1 mg?
Alac A simple calculation: the substrate’s mass is about 1 g ± 1 mg, in other

words, it can be determined with a relative accuracy of 1 per mill.
Mag Be careful! The balance’s specification only states that the mass is dis-
played in grams with 3 digits after the decimal point. However, the accuracy of
the measurement may be lower than the precision of the balance’s display, for
example, due to ventilation or building vibrations.
Tim But we cannot take such influences into account, because they are random
and out of our control.
Mag That’s a good point, because when they are simply random, their influence
on the weighing accuracy can be reduced if we repeat the measurement several
times. With these multiple values, we can also specify the uncertainty of the result,
in addition to the value of the mass.
Tim I’ve experienced that before. In the physics lab course for beginners, we
always have to report all results with measurement errors.
Alac It’s quite simple. We already know the true values from our fellow students in
previous semesters, thus we only need to adjust the error range for our measured
values so that the true value lies within the error bars. In this way, we avoid
annoying questions from the supervisors.
Mag That is similar to, but not exactly like, the way that we will do it in
this exercise. As we generate the measurement data ourselves, we know the true
values in advance, but then estimate them again from noisy data. However (!), we
rigorously (!) calculate the standard error according to the rules.
Tim Textbooks on measuring theory emphasize “confidence interval” and

“confidence level”.
Mag We will learn the meaning of these terms later in Exercises 9.4 and 9.5,
in which we will show that if 9 groups present their lab results with standard error
ranges capturing the true value, 3 will probably have cheated.

mS true mass of the substrate
mNs measurement noise (standard deviation of a normal distribution)
dsp display precision of the scales
mX measurement series
n number of measurements in mX
mM mass of the substrate estimated from the measurements
mSe standard error of mM .
9.2.3 Excel
Fig. 9.1 (S) presents an excel solution, with the parameter specification in a and
the process simulation in b. The raw data are generated in E6:E12 with normally
distributed noise obtained with norminv(rand();0;1) (see Exercise 8.5). They
are transformed into the balance display, our measurement data, in column F. In
G6:J6, the data are evaluated with the preliminary result for the estimated mass:
m S (esti.) = (m M ± m Se )
m s (esti.) = (0.995 ± 0.022)g
9.2 Weighing a Glass Substrate 371
A B C D
1 mS 1g "true" mass of the substrate
2 dsp 3 display accuracy of the balance, typical display: 1.001 g
3 mNs 0.05 g noise of the weighing process
E F G H I J K
)
();0;1
NV ( RAND
.I )
ORM 7;dsp (mX) mX) X) T(n)
Ns*N ND(E RAGE EV.S( NT(m mSd/SQR
4 =mS+m =ROU =AVE =STD =COU =
5 mX mM mSd n mSe
6 1.0062099 1.006 0.995 0.057 7 0.022
7 1.0303005 1.030
12 1.0878611 1.088
Fig. 9.1 (S) Weighing a glass substrate. a (top) Specifications. b (bottom) Simulation of the
process
Table 9.1 Weighing the mass of a substrate

1 mS=1 #[g]
2 dsp=3 #Digits displayed
3 mNs=0.05 #[g] Noise
4 n=7 #Number of measurements
5 mX=np.round(mS+mNs*npr.randn(n),3)
mX [1.076 1.050 1.049 0.990 1.010 0.993 1.018]
6 mM=np.mean(mX) mM 1.027
7 mSd=np.std(mX) mSd 0.030
8 mSe=mSd/np.sqrt(n) #Standard error mSe 0.011
still to be rounded to the relevant digits:
m s ( rounded ) = (1.00 ± 0.02)g.
First, the standard error is rounded to one digit, and then, the mean value is
rounded to the same number of decimal places as the standard error.
9.2.4 Python
The Python solution in Table 9.1 is straightforward.
Question
How do you report the results of Table 9.1 sensibly rounded?2
2 1.027 ± 0.011 becomes 1.03 ± 0.01.

9.3 A Procedure for Rounding to Relevant Digits
A formula network and a Python function are presented that round a mea-
surement result to the relevant number of digits determined by the standard
error. The equations rely on logarithms and powers.
9.3.1 Numerical Evaluations
The numerical evaluation of measurement series generally results in numbers with

many digits, e.g., x m = 0.008702 g and x Se = 0.000602 g. As we have learned
in Exercise 9.2, the final result should be (0.87 ± 0.06) × 10–2 g or (8.7 ± 0.0) mg.
We will obtain such results with formulas. To achieve this, we have to calculate
with logarithms and powers of ten and link text and numbers.
The method is shown as a formula network in Fig. 9.2 (S), based on the example of
Exercise 9.2, in which the mass of a glass substrate is determined. The estimated
mean value x M and its standard error x Se are specified with 3 digits, the display
accuracy of the balance.
The value of x Se (0.022 = 2.22 × 10–2 ) is broken down into power of ten (nSe =
−2) and first digit (xSeR = 2) and reproduced as x SeRR in decimal form with only
one non-zero digit. The value for x M is transformed into an integer x Mr . Thus, the
number of relevant digits corresponds to the precision of x SeR , and is reproduced
in decimal form with the reduced number of digits. The final result displayed in
D7 is obtained with the formula in D8, concatenating text and variables.
Another variant, reporting the final result in exponential form, i.e., (9.8 ± 0.2)E-
1 g, is shown in Fig. 9.3 (S). The value of x M is transformed into a number x MrP
greater or equal to 1 and smaller than 10 and then rounded to the first non-zero
digit of x Se .
A B C D E F G
1 Name Weight Power of xSe nSe -2 =INT(LOG10(xSe))
2 xM 0.977 First digit of xSe xSeR 2 =ROUND(xSe*10^(-nSe);0)
3 xSe 0.022 Reduced xSe xSeRR 0.02 =ROUND(xSeR*10^nSe;ABS(nSe))
4 Unit g xMr 98 =ROUND(xM*10^(-nSe);0)
5 Reduced xM xMrr 0.98 =ROUND(xMr*10^nSe;ABS(nSe))
6
7 Weight = (0.98 ± 0.02) g
8 =Name&" = ("&xMrr&" ± "&xSeRR&") "&Unit
Fig. 9.2 (S) The final result is obtained by rounding with formulas using logarithms
9.3 A Procedure for Rounding to Relevant Digits 373
A B C D E F G
10 Name Weight Power of xM nM -1 =INT(LOG10(xM))
11 xM 0.977 Reduced xM xMrP 9.77 =xM/(10^nM)
12 xSe 0.022 Power of xSe nSe -2 =INT(LOG10(xSe))
13 Unit g Reduced xSe xSeRP 2.2 =xSe/(10^nSe)
14 Rounded to relevant xMred 9.8 =ROUND(xMrP;nM-nSe)
15 Rounded to relevant xSeRed 0.2 =ROUND(xSeRP;0)*10^(nSe-nM)
16
17 Weight = (9.8 ± 0.2) E-1 g
18 =Name&" = ("&xMred&" ± "&xSeRed&") E"&nR&" "&Unit
Fig. 9.3 (S) Same as Fig. 9.2 (S); however, separating the power in the result
Table 9.2 Function for returning the final result rounded to the relevant digits
1 def FinRes(Name,xM,xSe,Unit):
2 #Power of the standard error:
3 n=int(np.floor(np.log10(xSe)))
4 #First digit of xSe:
5 xSeR=np.round(xSe*10**-n,0)
6 #Rounding to the certain digit:
7 xSeRR=np.round(xSeR*10**n,np.abs(n))
8 #Rounding xM to the certain digits:
9 xMr=np.round(xM*10**-n,0)
10 xMrr=np.round(xMr*10**n,np.abs(n))
11 return str(Name) + '=' + str(xMrr) + "±" \
12 + str(xSeRR)+ Unit
9.3.3 Python Function
A user-defined function, performing the calculation of Fig. 9.2 (S), is introduced

in Table 9.2 and applied in Table 9.3 in which two series of (x M , x Se ) pairs are
evaluated to check the validity of the formulas implemented in FinRes.
9.3.4 VBA Function
The rounding can also be performed with the user-defined VBA function FinRes
in Fig. 9.4a (P), which reads-in the name, the value, and the uncertainty of the
measurement result, and outputs the rounded result as shown in Fig. 9.4b. If this
function is inserted into a VBA project module for user-defined functions (in our
case, VBA project (Dieters functions.xlam, Sect. 4.9.1), then it can be called in
any excel file.
Table 9.3 a (top) Continuation; applying the program in Table 9.2; b (bottom) Results
1 xM=[100.234, 1.234, -2.334,-0.004, 10000.1233]
2 xSe= [0.5, 0.0002, 0.02, 0.001, 0.0004]
3 for i in range(len(xM)):
4 outpl=FinRes("a"+str(i),xM[i],xSe[i]," mg")
5 print(outpl)
6 print("\n") #Second series
7 xM=[1234.345,-12345.678,2.383,-0.0991297, 0.000930]
8 xSe=[ 35.023, 34.56, 0.01, 0.00223046,0.000017]
a0=100.2±0.5 mg a6=1230.0±40.0 mg
a1=1.234±0.0002 mg a7=-12350.0±30.0 mg
a2=-2.33±0.02 mg a8=2.38±0.01 mg
a3=-0.004±0.001 mg a9=-0.099±0.002 mg
a4=10000.1233±0.0004 m a10=0.00093±2e-05 mg
1 Function FinRes(Name, xM, xSe, Unit) 1 =FinRes(A4;B4;C4;D4)

2 'Input: Name of the variable, mean, uncertainty, physical unit 2 a0=100,2±0,5 mg
3 n = Int(Log(xSe) / Log(10)) 'power of the uncertainty 3 a1=1,234±0,0002 mg
4 xSeR = Round(xSe * 10 ^ -n, 0) 'first digit of the uncertainty 4 a2=-2,33±0,02 mg
5 xSeRR = xSeR * 10 ^ n 'rounding to the certain digit 5 a3=-0,004±0,001 mg
6 xMr = Round(xM * 10 ^ -n, 0) 'rounding the measured value 6 a4=10000,1233±0,0004 mg
7 xMrr = Round(xMr * 10 ^ n, Abs(n)) 7 a5=-234,6±0,3 mg
8 FinRes = Name & "=" & xMrr & "±" & xSeRR & Unit 8 a6=12350±40 mg
9 End Function 9 a7=-12350±30 mg
A B C D E a8=2,38±0,01 mg
2 =FinRes(A4;B4;C4;D4) a9=-0,099±0,002 mg
3 a0 100.234 0.5 mg a0=100,2±0,5 mg a10=0,0093±0,00002 mg
4 a1 1.234 0.0002 mg a1=1,234±0,0002 mg
5 a2 -2.334 0.02 mg a2=-2,33±0,02 mg
Fig. 9.4 a (left, P) The user-defined function FinRes (top) reads in the name and the value of a
measured quantity (columns A and B in the spreadsheet), as well as its measurement error (column
C), and (bottom) outputs the rounded measurement result as text (column E). The spreadsheet has
to be treated with a main program that calls FinRes. b (right) Ten results of the function FinRes
9.4 Increasing the Measuring Accuracy Through Repetition
We illustrate the meaning of the standard error with repetition procedures

to get the hit rate, i.e., how often the error range captures the true value.
The more often a quantity is measured, the more accurately its value can
be determined. If the same measurement setup is always used, the standard
error of the mean value of the measurement series is inversely proportional
to the root of the number of measurements. Twice as good with four times
the effort.
9.4 Increasing the Measuring Accuracy Through Repetition 375
9.4.1 Standard Deviation and Standard Error of the Mean Value

of a Measurement Series
How can we halve the standard error?
Alac If we want to be twice as good, we just have to measure twice as often.
Mag No, keep the following broom rule in mind and study the next section:
Ψ Twice as good with four times the effort.
Tim I remember the reason for this. It is the variance of the mean value of a
measurement series that is inversely proportional to the number of measurements.
The standard error is the square root of the variance.
Mag One more hint: All mathematically justified theorems make statements
about variances. To estimate the measurement error, however, the standard error
with the same physical unit as the mean must be quoted:
Ψ Calculate with variances, report the C-spec error!
In this section, we learn about the quantities: variance var X of a set of values x
= {x i }, standard deviation x Sd of the data in the set, and standard error x Se of the
mean x m of the set.
The standard error x Se of the mean value x m of a set of measured values, used
to indicate the measurement result’s uncertainty, is smaller than the standard devi-
ation x Sd of the set, measuring the scatter in the set. The standard deviation x Sd
is, in principle, independent of the number n of measurements, while the standard
error x Se decreases with increasing square root of n.
Visual estimate
Question
How big do you estimate the uncertainty of the mean value in Fig. 9.5a to be?3
By what factor is the standard error of the mean smaller than the standard
deviation of the set of measurements?4
Within which range do you expect a 10th measured point in Fig. 9.5?5
Calculated error range

Figure 9.6 displays measured data with calculated means (vertical bar), standard
deviations of the set of measured values (horizontal square brackets above), and
standard errors of the mean (horizontal square brackets below).
3 To check your judgement, have a look at Fig. 9.6!

4 √
By the factor 1/ n, where n is the number of measurements, as you will learn in the course of
this exercise.
5 The 10th measured point should be within the standard deviation of about 1 with a probability of
about 2/3.
4 Meas
-2 -1 0 1 2
16 Meas
-2 -1 0 1 2
Fig. 9.5 Estimate! Estimate mean value, standard deviation, and standard error of a (top) a series
of 4 measurements and b (bottom) a series of 16 measurements! The standard deviation for both
series is about 1
2 2
2 2
1 1
1 1
0 0
-1 -1
4 Meas -1
-2
4 Meas -1
-2
-2 -2
-2 -1 0 1 2 -2 -1 0 1 2
2 2
2 2
1 1
1 1
0 0
-1 -1
16 Meas 16 Meas
-1 -1
-2 -2
-2 -2
-2 -1 0 1 2 -2 -1 0 1 2
2 2
2 2
1 1
1 1
0 0
-1 -1
64 Meas 64 Meas
-1 -1
-2 -2
-2 -2
-2 -1 0 1 2 -2 -1 0 1 2
Fig. 9.6 Results for n = 4, 16, and 64 measurements of a physical quantity whose true value is
0; mean value x m indicated by vertical lines, scatter x m ± x Sd of the measurement series by square
brackets above, and confidence interval (error range) x m ± x Se of the mean value as square brackets
below; the true value 0.00 is a (left) within the standard error range of the mean value or b (right)
outside of the standard error range of the mean value
In the analysis and discussion of measurement series, we must clearly distinguish

between
– the standard deviation x Sd of the data set (Eq. 9.3) and
– the standard error x Se of the mean value x m (Eq. 9.4).
The standard deviation of the series characterizes the scatter of the measured
values within the measurement series. A new measurement is expected to lie
approximately within this range, “approximately” meaning:
Ψ Two within and one out of (the standard deviation).6
6 Namely, with a probability of 2/3 within plus or minus the standard deviation from the mean.
The mean value x m of all measured values can be determined by visual judg-
ment more precisely than ± x Sd , as is visible in the diagrams in Figs. 9.5 and 9.6.
According to theory, the standard error of the mean value x Se is equal to the stan-
dard deviation of the series x Sd divided by the square root of the number n of
measurements, Eq. 9.4.
The corresponding range
[xm − x Se , xm + x Se ] (9.8)
called the standard error range, is indicated in the diagrams by bottom square
brackets.
In the next section, we examine the meaning of the error range in more detail
by determining hit rates.
Mag What is the informative value of the standard error?
Tim The standard error has the same physical dimension as the measured value.
Therefore, it is possible to specify the measurement uncertainty with plus-minus
the standard error.
Mag Is the true value always within the standard error?
Alac No, only if we’ve done everything right.
Tim I’ve learned that this is only true in about 68% of cases, to be remembered
with:
Ψ Two within and one out of.
Alac Is that to say that the true value is outside of the standard error limits for
every third experiment? There was never any trouble like this in our introductory
lab course.
Mag If the true value for nine lab groups lies within the standard error of the
measurement ± x se , then three groups have cheated, I suspect.
Tim We have also learned that the broom rule applies only if the same quantity
is measured at least eight times. We never measured that often in the beginner’s
lab course. What is the error specification good for, then?
Mag We will learn about that in this exercise by determining hit rates. Anyway,
if you specify the number n of measurements, together with x m and x Se , you can
make well-defined statements with t statistics, addressed in Exercise 9.5.
4, 16, or 64 measurements of the same quantity

Let the true value of a quantity be 0 and the standard deviation of the measurement
process be 1. We simulate this process with a Gaussian generator with a mean value
of 0.00 and a standard deviation of 1.00 and determine the mean value x m , standard
deviation x Sd , and standard error x Se for measurement series (samples) with different
numbers of measurements. The results of the simulation are presented in Fig. 9.6.
In the first row of Fig. 9.6, the physical quantity was measured four times, in the
second row, 16 times, and in the third row, 64 times. In the left subpictures, the mean
value’s error ranges include the true mean; in the right subpictures, they do not.
Standard deviation of the measurement series

The measured standard deviation x Sd is approximately the same for all measurement
series, independent of the number of measurements. So, if an additional measurement
is made, the expected deviation of this new measurement from the mean value is
independent of how often the measurement has been made before. The dispersion of
the data set does not depend on the number of data.
How precisely can the mean value of a measurement series be determined?

In Fig. 9.6, two measurement series are plotted for each number of measurements.
The mean value fluctuates more strongly for the series with a smaller number (vis-
ible here when comparing the left and right subfigures). The theory states that the
expected measurement error for the mean value of a measurement series is the stan-
dard deviation of this series divided by the root of the number of measurements
(Eq. 9.4). The error range (also called the confidence interval) of the mean value is
also included in the diagrams mentioned. It corresponds approximately to the range
in which we would locate the mean value with the eye.
Tim Are we really twice as good with four times the effort?
Mag The formula only states that the √ standard error of the mean calculated
according to the rules decreases with 1/ n, narrowing the confidence interval
(“error range”) correspondingly. We still have to know whether the confidence
level (“hit rate”) changes as well.
– The standard deviation of a measurement series is in principle

independent of the number of measurements; it fluctuates only by
chance.
– The standard error of the mean of the measurement series decreases
with the inverse of the square root of n, the number of measure-
ments.
– The probablity that the standard error range captures the true
value depends on the number of measurements and increases with
increasing n up to 68%.
Hit rate by repetition of the statistical experiment

A simulation repeats the standard error range calculation and determines how often
x m ± x Se captures the true value. The “hit rates” are:
0.60 (0.61) for 4 measurements,
obtained with Python (excel) in the following sections.

x meas results of 64 measurements, true value = 0
x m4m mean value of 4 measurements
x Sd4m standard deviation of 4 measurements
x Se4m standard error of the mean of 4 measurements
in4m true, if the error range captures the true value
within4m counts the in4M
and correspondingly for 16 and 64 measurements.
inM 3 values, [in4m , in16m , in64m ].
Hit rate by repetition of the statistical experiment

In the diagrams of Fig. 9.6b, the true value of the measured quantity (namely, 0.0) is
outside of the standard error range. With a rep-log procedure, we examine how often
this happens on average. The function WithinSe in Table 9.4 generates an array of 64
normally distributed random numbers, calculates, in lines 4 to 6, the characteristic
parameters for the first 4 numbers of this array, and determines, in line 7, whether
the standard error range captures the true value (in4m is True or False). The
Table 9.4 Does the standard error range capture the true value? Only the statements for 4 mea-
surements are reported; those for 16 and 64 measurements are the same, but with 16 or 64 instead
of 4
1 def WithinSe():
2 x_meas = npr.randn(64)
3
4 xm4m = np.mean(x_meas[0:4]) #Entries 0, 1, 2, 3
5 xSd4m = np.std(x_meas[0:4],ddof=1)
6 xSe4m = xSd4m/np.sqrt(4)
7 in4m = xm4m- xSe4m < 0 < xm4m+xSe4m
#0 is the true value
……
18 return in4m, in16m, in64m
Table 9.5 Rep-log procedure for calculating hit rates, WithinSe from Table 9.4
1 within4m =0
2 within16m=0
3 within64m=0
4 Ntrials=10000
5 for i in range(Ntrials):
6 inx=WithinSe()
7 within4m+=inx[0] within4m 6018 hit rate 0.60
instructions for the first 16 and first 64 measurements are analogous, resulting in
in16m and in64m, which are also returned in line 18.
In the numpy function np.std, a parameter ddof must be specified, meaning

Delta Degrees Of Freedom. The divisor used in calculating the variance in Eq. 9.2
is n - ddof , where n is the number of data points. By default, ddof in np.std
is zero. If we determine only the mean of a data series, as we do in this chapter,
ddof = nP = 1.
In the main program in Table 9.5, we call WithinSe several times (here, N trials
= 10,000) and increment the variables within4m and the like by 1 if the output of
WithinSe is true. This is possible because the logical values are coded as binary
numbers, 1 for True and 0 for False. We see from the bottom cell output that the
hit rate increases with the number of measurements, from 0.60 for 4 measurements
to 0.68 for 64.
Theoretically, the confidence interval for ± x Se is 68.3% when the number of
measurements approaches infinity. In other words: in more than 30% of the cases,
the “true” value is also theoretically outside ± x Se . We simplify to the broom rule:
Two within and one out of (the standard error range). It applies if the measure-
ment is repeated at least 8 times. If measurements are taken less often, the hit rate
is lower. For 4 measurements, it is only 60%, according to the bottom cell in Table
9.5.
9.4.4 Spreadsheet Layout for This Task
The spreadsheet layout corresponding to the Python program in Table 9.4 is

shown in Fig. 9.7 (S). Columns A to H reproduce the calculation of the Python
function WithinSe with the result in column H.
The role of the main program in Python is played in excel by the VBA
procedure in Fig. 9.8 (P), taking the logical values in H3:H5 as input and, if True,
incrementing the counts in I3:I5 individually.
_m )
+xSe
) ; 0< xm_m
();0;1 0
m < rocedure ials
RAND 3:A6) :A18) T(n) -xSe_ p /Ntr
RM.INV( RAGE(A DEV.S(A3 d_m/SQR (xm_m rep-log in_m
1 =NO =AV E =ST =xS =AN D from =with
2 x_meas n xm_m xSd_m xSe_m in_m within_m hitrate Ntrials
3 2.04 4 0.87 1.22 0.61 4 Meas FALSE 6116 0.612 10000
4 0.30 16 0.12 0.96 0.24 16 Meas TRUE 6738 0.674
5 1.71 64 0.02 0.91 0.11 64 Meas TRUE 6840 0.684
66 0.02
Fig. 9.7 (S) 64 random measurements x meas are generated in column A, while subsets of these are
evaluated in columns D to H. The logical values in column H are read by sub Protoc3 in Fig. 9.8
(P) and summed in column I after a 10,000-fold generation of new measurement series
1 Sub Protoc3() 12
2 Ntrials = Range("K3") 13
3 Range("I3:I5").ClearContents 14
4 For i = 1 To Ntrials 15
5 Application.Calculation = xlCalculationManual 16
6 If Range("H3") = True Then Range("I3") = Range("I3") + 1 17
10 Next i 21
11 End Sub 22
Fig. 9.8 (P) Rep-log procedure for incrementing the number of hits in I3:I5 of Fig. 9.7 (S); if a
value in H3:H5 is True, the corresponding cell in column I is incremented by 1
Questions
concerning Fig. 9.7 (S)

How does the “true” value of the measurand enter the simulation of the
measurement process in A3:A66?7
What is the formula in H5?8
9.4.5 How to Report a Measurement Result
Tim People often say that our rule Two within and one out of only applies
if measurements are taken more than seven times. We never measure that often in
our lab course, at most, four times.
7 As mean of a normal distribution in column H generated with norm.inv (random(), Mean,

StDev). Here, Mean = 0 and StDev = 1.
8 See formula printed in H1; the expression is True if the error range does capture the true value.
The formula in H5 processes entries in row 5 of xm m and xSe m.

Alac That doesn’t matter. Even if you measure a quantity only twice, you can
apply the formulas for mean value and standard error. That’s enough.
Mag Not quite. You have to specify the number of measurements in addition
to the mean value and the standard error. An example of a complete measurement
protocol: The acceleration due to gravity in our laboratory was measured twice
with a pendulum, resulting in an average value of 9.8 m/s2 with a standard error
of 0.2 m/s2 .
Tim Should we report the details of the experimental setup as well? For exam-
ple: How exactly we determined the length of the pendulum and the period of
oscillation.
Mag Of course, to allow for conclusions about systematic errors. The three
aforementioned statistical specifications are enough to make statistically relevant
statements. E.g., for the two measurements referred to by Alac, the calculated
standard error must be doubled to achieve a measurement uncertainty according
to Two within and one out of .9
9.5 The t Statistics Connects Confidence Interval

with Confidence Level
The t-value of Student’s t distribution relates an extended confidence interval

t times the standard error to a confidence level (or an error probability) . For
at least eight measurements, our rule for the standard error range Two
within and one out of (the standard error range) applies.
9.5.1 Student’s t Distribution
Practical handling
The following list explains the terms that play a role in t statistics:
– Standard error rangexm ±x Se with the standard error calculated from the standard
deviation of the measurement series x Sd as x Se = x√Sdn .
– Degree of freedom do f = n − n P with n being the number of measurements,
and nP the number of parameters estimated from the measurement series. In this
chapter, nP = 1, because only the mean x m is estimated.
9The degree of freedom for 2 measurements is dof = 1 and the t value for this is t = 1.84 for a
probability of error of 38% (see Exercise 9.5). For an error probability of 5%, the t value is 12.7.
9.5 The t Statistics Connects Confidence Interval with Confidence … 383
– C-spec error range xm ±tC ·x Se , standard error multiplied by a factor t C determined

by a pre-specified confidence level C or error probability E and the degree of
freedom dof .
– Confidence level C(t, do f ), corresponds to the hit rate in our statistical exper-
iments. The error probability has the complementary value: E(t, do f ) = 1 −
C(t, do f ).
Error probability, confidence level, and t-value are obtained with the following
functions, with dof being the degree of freedom:
excel e(t,dof) = t.dist.2t(t;dof)

c(t,dof)= 1- t.dist.2t(t;dof)
t(e,dof) = t.inv.2t(e,dof)
Python E(t,dof) = (1-sct.t.cdf(t,df=dof))*2

C(t,dof) = 2*sct.t.cdf(t,df=dof)-1
T(E,dof) = sct.t.ppf(1-E/2,df=dof)
In Fig. 9.9a, the error probability E is obtained for different degrees of freedom f
and t values of 1 and 1.96. The t value for pre-specified E and dof is given in b
(F:H).
For 4 measurements, the degree of freedom is 3, and a hit rate 1–0.391 ≈ 0.61
for t = 1 is expected. Actually, in Sect. 9.4.1, hit rates of 0.60 for the Python
simulation and 0.61 for the excel one come close to this expectation.
Mathematics of the t distribution

The entity t is defined as
t = (xm − x T r ue )/x Se (9.9)
A B C D E F G H I
T)
t;dof
/3)*3 V.2T(E
1 =(E-1 =T .IN
2 dof E Et dofT t
3 100000 0.317 =T.DIST.2T(1;dof) -0.05 0.317 100000 1.00
4 7 0.351 =T.DIST.2T(1;dof) 0.05 0.317 7 1.08
5 3 0.391 =T.DIST.2T(1;dof) 0.17 0.317 3 1.20
6 100000 0.050 =T.DIST.2T(1.96;dof) 0.05 100000 1.96
7 7 0.091 =T.DIST.2T(1.96;dof) 0.05 7 2.36
8 3 0.145 =T.DIST.2T(1.96;dof) 0.05 3 3.18
Fig. 9.9 a (left, A:D) Error probability E for t values 1 and 1.96, each for various degrees of free-
dom dof . b (right, F:H) t values for E t values of 0.317 and 0.05, as well as various degrees of
freedom dof T
200
160
120
Frequency
80
40
simulated
=T.DIST(Ic;3;0)*N*dt
0
-3.00 -1.00 1.00 3.00
t
Fig. 9.10 a (left) Frequency of occurrence of t values obtained in a series of simulations with 4
measurements together with the theoretically predicted values for dof = 3; b (right) ppf and cdf
(dashed line) of Student’s t distribution for dof = 3 (dof is named f in the legend); the right-angled
straight lines indicate t.cdf(1,dof) and t.ppf (0.975,dof) (dotted line)
describing the distance of the sample mean x m to the true value x True , divided by
the standard error x Se of the mean value.
In order to get an idea of t statistics, we:
– perform a simulation of a series of 4 measurements with a noise level between

0 and 1 from which a single value of t can be obtained,
– repeat the single experiment to get a representative set of t values, with each
time a new noise level,
– and determine their frequencies of occurrence.
The result (diamonds) is shown in Fig. 9.10a together with the values predicted by
the theoretical t distribution for dof = 3 (line).
Figure 9.10b displays the ppf and cdf of Student’s t distribution for dof =
3; the right-angled straight lines indicate t.cdf(1,dof) starting at t = 1 and
t.ppf (0.975,dof) (dotted line, starting at cdf = 0.975). We get.
sct.t.cdf(1, 3) = 0.8045
C(1, 3) = 2 ∗ sct.t.cdf(1, 3)−1 = 0.609
T(0.05, 3) = sct.t.ppf({0.975, 3})= 3.18}
We see that for t = 1, the cdf is about 0.8, i.e., in about 80% of the experiments
a measurement result is below the upper limit of the error range; 20% are above.
As the pdf is symmetric to t = 0, likewise 20% are below the lower limit of the
error range so that the error probability is E = 0.4 and correspondingly C(1, 3) ≈
0.6.
For dof = 1,000,000, we get

sct.t.cdf(1,1,000,000) = 0.841,
p = 0.16 that x T r ue > xm + 1 · x Se .
C(1, 1,000,000) = 2* sct.t.cdf(1,1,000,000) – 1 = 0.683.
E(1, 1,000,000) = (1–0.683) = 0.317,
p = 0.317 that x True is outside of the error range xm ± 1 · x Se .
T(0.05, 1,000,000) = Sct.t.ppf (0.975,1,000,000) =
1.96.
a confidence level of C = 0.95 (error probability 0.05 = 2·0.025) is
obtained for an error range xm ± 1.96 · x Se .
Set-up of the simulations to determine C-spec errors

We are simulating measurement series according to We know everything and
play stupid by generating normally distributed random numbers with mean = 0
and standard deviation = 1, and determining the experimental hit rate, for series
comprising 2 to 16 measurements, within error ranges with t values that correspond
to error probabilities 0.317 and 0.05.
Proven or not disproven? 8 = ∞?
Tim Summarizing: with the mean value and its standard deviation, and the num-
ber of measurements, statistically correct statements with confidence levels can be
made.
Alac Our simulation proves that.
Tim No, our simulation simply gives us no reason to doubt that statement.
Alac Your nitpicking’s a pain in the neck.
Mag But it is indispensable, because we don’t do logical derivations.
Tim Let’s trust in statistical textbooks.
Alac In the case of measurement series, we are content with a rough estimate:
infinity already starts with 8 measurements.
Tim 8 = ∞? Isn’t that twisting the facts?
Mag Discuss!
Questions
concerning Fig. 9.9 (S) By what factor t should the standard deviation of 4
measurements be increased so that the true value is within the C-spec error
range with a probability of 68.3%?10
What is the probability of error if, for eight measurements, the standard error
is specified as the measurement uncertainty?11
What is the error probability for four measurements if the standard error
times 1.96 is used to specify the error range?12
x True True value of a measurand

n Number of measurements in a series
x Series of n measurements
xm Mean value of the measurement series
x Se Standard error of x m
dof Degrees of freedom, dof = n − 1
C Confidence level
E Error probability, E = 1 − C
t t-value (Student’s) for the C-spec error range, determined by C and dof
x SeT x Se multiplied with a t-value appropriate for C and E
nT Number of trials (repetitions of a statistical experiment)
hitRate Counts how often the error range captures the true value
pOut (nT -hitRate/nT , probability that the error range misses the true value.
In column A of Fig. 9.11 (S), 32 normally distributed random numbers (mean =

0, standard deviation = 1) are generated. Subsets of 2 to 32 of them are specified
by cell range addresses in column C. Indirect addressing of these ranges is used
to calculate their mean (in column D) and their C-spec error for E = 0.317 (in
column F).
Questions
concerning Fig. 9.11 (S)
10 Figure 9.9 (S), the degree of freedom for four measurements is dof = 3, t = 1.20 in cell C5.
11 Figure 9.9 (S), dof = 7, B4 = 35%.
12 Figure 9.9 (S), dof = 3, B8 = 14.5%.
)*t_
T(n_
1 ) ng )) ) n g))/SQRSeT)
;0; T (R 1 (R x ur
ND() IREC 7;n_- RECT ;xm> og proced
.INV(RA " &B 4+2 AGE(IND .2T(0.31 V.S(INDI m<-xSeT ep -L
RM :A R V E x R og
1 =NO ="A3 =AVE =T.IN =STD =OR( from Rep-L
2 x n_ Rng xm t_ xSeT outSeT pOut 100,000 t0.05 pOut0.05
3 -0.44 2 A3:A4 0.34 1.84 1.45 FALSE 0.317 12.7 0.049
4 1.13 4 A3:A6 0.73 1.20 0.60 TRUE 0.317 3.2 0.050
5 0.33 8 A3:A10 -0.19 1.08 0.58 FALSE 0.317 2.4 0.049
6 1.88 16 A3:A18 -0.02 1.04 0.30 FALSE 0.314 2.1 0.050
7 -2.97 32 A3:A34 -0.11 1.02 0.20 FALSE 0.318 2.0 0.049
34 0.35
Fig. 9.11 (S) Evaluation of series from column A with 2 to 32 measurements; in column F, C-
spec error ranges are listed for an error probability of E = 0.317. The formula in column G checks
whether the true value (here, 0) lies outside of the error limits. In column H, the experimental error
rate is recorded after 100,000 repetitions of the statistical spreadsheet experiment (Multiple tests
for hit rates) . The simulation is repeated for an error probability E = pOut of 0.05 (t values in
column J), with the results reported in column K
Which numbers addressed in D6 are averaged?13

How is the argument of the spreadsheet function average in column D
constructed?14
Which cells does the statement in F4 refer to?15
In column E, the spreadsheet function t = t.inv.2t(E;dof ) reports the t value

for an error probability of E = 0.317. The extension “.2T” indicates that deviations
of x m to both sides of the error range are taken into account.
In column F, the mean value’s standard error is multiplied by t, so that the
error rate is expected to be pOut = E = 0.317. Column G contains the logical
expression [=or(x m < -x SeT ; xm > x SeT )], true if x mTrue , namely, 0, is outside of
the error range. In the example of Fig. 9.11 (S), this is the case (by chance) only
for one measurement series. With a rep-log procedure, we perform the random
experiment 100,000 times and get the error rates in column H. They are actually
close to the expected 0.317.
In column J, t values that correspond to an expected error probability of E =
0.05 are calculated. Repeating the simulation with the corresponding C-spec error
range leads to experimental error probabilities, column K, close to the theoretical
value.
13 The mean value of the numbers in A3:A18 is calculated.

14 The argument of average is defined using indirect from the column range with name Rng.
From Rng, the value in the same row is taken.
15 The statement in F4 (1) calculates the standard deviation of the subset A3:A6 (addressed with
√
indirect(rng) , Rng in C4), (2) divides the result by n (B4) to get the standard error, (3) mul-
tiplies the standard error with a t (E4) for E = 0.317 and dof = n − 1 = 3, appropriate for n =
4).
Table 9.6 a (top) Specification of arrays for taking up the results of simulations with t statistics;
b (bottom) A single run of the simulation
1 nList=[2,4,8,16,32]
2 lnL=len(nList)
3 E=0.317
4 t=np.empty(lnL) #t values for E
5 for n in range(lnL):
6 t[n]=sct.t.ppf(1-E/2,df=nList[n]-1)
7 out=np.zeros(lnL)
8 x=npr.randn(32)
9 xm=np.array([np.average(x[0:n]) for n in nList])
10 xSeT= np.array([np.std(x[0:n],ddof=1)/np.sqrt(n)
for n in nList])*t
11 outSeT=np.logical_or((xm<-xSeT),(xm>xSeT))
12 out+=outset #Boolean added as 0 or 1
Table 9.7 Results of a run of the program in Table 9.6b

E 0.317
t [1.839 1.198 1.077 1.035 1.017]
xm [-1.026 -0.871 -0.279 -0.079 -0.04 ]
xSeT [1.784 0.547 0.347 0.277 0.196]
outSeT [False True False False False]
out [0. 1. 0. 0. 0.]
Single experiment
The program for checking the consistency of confidence intervals and confidence
levels and its result is distributed over three cells. In the first cell of Table 9.6, three
lists of the same length (nList, t, out) are generated, with their elements specific to the
5 series with different numbers of measurements. In the second cell, the simulation
is run once; all logged lists are reported in Table 9.7 A numerical array out is created
to sum the logical values of outSeT as numbers 0 or 1. It counts the number of
missed hits, i.e., when the true value is outside of the extended error range when the
simulation is repeated.
Questions
Why does t have to be an array? Why is it not sufficient to specify it as a list?16

Replace lines 4 to 6 of Table 9.6a with a list comprehension!17
Which columns of Fig. 9.11 (S) are equivalent to Table 9.6b?18
16 Line 10 of Table 9.6 multiplies the array of standard errors with t, element-wise. This must be
a numerical multiplication. Therefore, t has to be an array.
17 t = np.array ([sct.t.ppf(1-E/2,df = n-1) for n in nList])
18 Columns A, D, F, G; set x in A, x in D, x
m SeT in F, outSeT in G.
Table 9.8 Multiple repetition of the simulation experiment; a (top cell) program code; b (middle
cell) Table 9.6b changed into a function; and c (bottom cell) results for t values theoretically valid
for an error probability of E = 0.317 (b) and E = 0.05 (c)
13 nOut=np.zeros(lnL)
14 nT=10000 #Number of trials
15 for m in range(nT): nOut+=Check_if_out()
16 np.set_printoptions(precision=3)
1 def Check_if_out():
2 x=npr.randn(nList[-1])
3 xm=np.array([np.average(x[0:n]) for n in nList])
4 xSeT= np.array([np.std(x[0:n],ddof=1)/np.sqrt(n) \
5 for n in nList])*t #nList and t from table above
6 outSeT=np.logical_or((xm<-xSeT),(xm>xSeT))
7 return outSeT
E 0.317 nT 100000
nOut/nT [0.318 0.316 0.319 0.319 0.319]
E 0.05 nT 100000
nOut/nT [0.0501 0.05 0.05 0.0493 0.0493]
How do you have to change the logical query in line 11 of Table 9.6b if you
want to count the hits instead of the misses?19
Error rates
The statistical experiment is multiply repeated in Table 9.8, in a for-loop calling the
function Check if out() and summing up its logical output in a numerical array nOut .
The statistical experiment is performed twice, first with E = 0.317 in line 3 of Table
9.6a (results in Table 9.8b) and second with E = 0.05 (results reported in Table 9.8c).
The experimentally found miss rates pOut = nOut /nT are close to the expectations, E
= 0.317 and E = 0.05.
Questions
Which lines of Table 9.6 have to be integrated into the function Check if out?
What variable has to be returned?20
9.6 Combining Results from Several Measurement Series
A combined result of two measurement series is obtained as a weighted aver-

age of the individual results with the squares of the reciprocal C-spec errors
19 inSeT = np.logical_and(-xSeT < xm,xm < xSeT); the true value of x m is 0.

20 Lines 8 to 11 with return outSeT. The excel equivalent is a rep-log procedure that calcu-
lates pout .
as weights. The C-spec errors are derived from internally and externally
consistent variances.
Ψ Worse makes good even better. Mostly, but not always.
9.6.1 Combining Two Measurement Results
Calculating with variances
Mag Two research groups use different methods, A and B, to determine the
value of the same measurand. They report their non-identical results with different
standard errors. For example, the acceleration due to gravity at a particular location
has been determined by a drop test and a pendulum. As the head of the two groups,
which result do you write in the project’s final report?
Alac Clearly the one with the smaller standard error.
Tim I simply report both results, each with its own number of measurements
and standard errors. The readers can then decide for themselves.
Mag Neither answer is correct. As a supervisor, you should know how to com-
bine both measurements and specify a value for the measurand whose standard
error is even smaller than that for the better of the two measurements.
Tim Alright, I have found a formula in the internet. The final result x mAB is
a weighted mean of the two results x mA and x mB , with the reciprocals of the
variances var A and var B of the measurement series as weights w, Eq. 9.10.
wA · xm (A) + wB · xm (B)
xmAB =
wA + wB
1 1
wA = wB = (9.10)
var A var B
Furthermore, the reciprocal value of the combined results’ variance is calculated

as the sum of the reciprocals of the individual variances:
−1
varAB = var A−1 + var B−1 = w A + w B
√
x Sd (AB) = w A + w B (9.11)
Alac Equation 9.3 is another neat formula, plus, it’s trustworthy, because it fits
our rule Ψ Calculate with variances, report the standard error.
Tim Do we have to take Student’s t-value into account? That’s the t-question.
My internet resource does not deal with it.
9.6 Combining Results from Several Measurement Series 391
Alac Let’s check it with Ψ Two within and one out of .
Mag Good idea. By the way, Eq. 9.11 is called the internally consistent
variance. The greater the variance, the lower the measurement’s weight in the
combined result.
Questions
Due to Eq. 9.11, will the combined result’s variance be greater or smaller than
the variance of the best measurement?21
Provided the noise is the same for two measurement series with n = 4 and n
= 16, by what factor should the standard errors of their mean values differ?22
Calculation with C-spec errors

Equation 9.11 makes a statement about the variances of measurement series. We are,
however, interested in confidence intervals and know already that the standard error
is:
xsd
x Se = √ (9.12)
n
and the C-spec error, with the t value obtained from the confidence level C and the
degrees of freedom dof is:
xCe = x Se · t(C, do f ) (9.13)
Furthermore, we recall our broom rule:

Consequently, we use in the formulas for mean and C-spec error the weights:
1
weight w = 2
(9.14)
xCe
and check the results with hit rates.
Simulation
We set up two independent measurement series for the same measurand, one, A, with
16 individual measurements and the other one, B, with 4 individual measurements,
so that they exhibit different variances. They are combined into a third series C of
size 20. Furthermore, the respective results of A and B are combined into AB using
the method described above.
21The reciprocals of the variances are added; the variance of the overall result becomes smaller.
22The standard errors of the mean values should differ by a factor of two (the ratio of the square
roots of the numbers of measurements).
Fig. 9.12 a (left) A result for one measurement series each for A, B, and AB; the length of the
vertical lines indicates the respective weight in the weighted sum; the true value is x m = 6. b (right)
Deviation of the mean values of the measurement series A, B, C, and AB from the true value,
ordered according to the deviation |x mC -x mTrue |
We calculate the mean values x mA , x mB , and x mC together with the squares of the
C-spec errors according to Sect. 9.5.1. The combined result x mAB is calculated as
the weighted mean according to Eq. 9.10, together with its C-spec error according to
Eq. 9.13, both with the weights in Eq. 9.14. The results for one random experiment
are graphically shown in Fig. 9.12a. The “true” value of the measurand is 6. If the
random experiment is repeated, different positions and heights of the vertical lines
result.
The positions of the vertical lines in Fig. 9.12a correspond to the mean value x m
of the measurement series, the height to 1/x mCe* 2 . In this example, the value of the
combined measurements, x mAB , is closest to the true value of 6.0.
We repeat the random experiment 20 times with a rep-log procedure and note
the deviations of the four mean values for A, B, C, and AB from the true value; the
result is shown in Fig. 9.12b. The experiments are ordered according to the outcome
of experiment C, the direct evaluation of 20 measurements. The value for the series
with four measurements (B) is usually farther away from, but sometimes closer to,
the true value than the results of the series with 16 measurements (A).
Ψ Mostly, but not always! Fundamental rule of statistical reasoning:-).
The weighted mean values of (A) are joined with a solid line. We see that they
are sometimes better and sometimes worse than those for the combined series (AB).
Tim So, it would sometimes be better to report the result of experiment (A)
instead of the combined result?
Mag Keep in mind that we perform simulations according to Ψ We know every-

thing and play stupid. In real life, the true value is unknown, and we have to be
content with Ψ Decide! Sometimes it will be wrong. Let’s have a look at the hit
rates from the rep-log procedures in Table 9.9. and Fig. 9.14 (S).
Table 9.9 Internally (left two cells) and externally (right two cells) consistent evaluation of the
measurement series; the first line corresponds to a variant of line 8 in Table 9.11
vABmax=vABint vABmax=vABint vABmax=vABext vABmax=vABext
Theor. 0.683 Theor. 0.950 Theor. 0.683 Theor. 0.950
Hits A 0.682 Hits A 0.950 Hits A 0.682 Hits A 0.952
Hits B 0.684 Hits B 0.950 Hits B 0.681 Hits B 0.951
Hits C 0.683 Hits C 0.950 Hits C 0.681 Hits C 0.951
HitsAB 0.633 HitsAB 0.932 HitsAB 0.448 HitsAB 0.406
Alac The hit rates of A, B, and C are close to the theoretical value of 0.683
and 0.950, but AB’s hit rates are only about 0.63 and 0.93; they do not meet the
expectation.
Tim Therefore, the estimation of the standard error according to Eq. 9.11, called
“internally consistent”, must be wrong.
Mag It is not wrong but does not fully satisfy our requirements. Before drawing
conclusions, let’s evaluate another formula to estimate a combined mean’s variance
called “externally consistent”. Take a look at Eq. 9.15.
Tim But that’s even worse. Look at the right two cells in Table 9.9 with HitsAB
= 0.406.
Mag That’s right. In the end, we have to calculate both variances and choose
the bigger one to avoid too small an error range. In doing so, we yield hit rates
HitsAB = 0.667 and 0.935, slightly better than those reported when taking the
internally consistent variances alone.
Externally consistent variance

The externally consistent variance is calculated with the mean values x mA and x mB
of series A and B and the combined mean value x mAB :
w A · (xm A − xm AB)2 + w B · (xm B − xm AB)2

v ABext = (9.15)
(M − 1) · (w A + w B )
with M being the number of measurement series whose results are to be combined,
here, M = 2, and wA , wB being the weights defined in Eq. 9.14.
A series A, N A values
B series B, N B values
C A and B together as one series

AB results of A and B combined into one result
x True true value of the measurand
x mA , x mB , x mC mean values of series A and B
x CeA , x CeB , x CeC C-spec errors errors of x mA , x mB , and x mC
wA , wB weights of series x mA and x mB , Eq. 9.14
x mAB mean value of x mA and x mB combined
vABint internal variance of the combined result
vABext external variance of the combined result.
In Fig. 9.13 (S), two data sets A and B are generated, comprising, respectively,
16 and 4 normally distributed random numbers and the union C = A|B (see cell
E3) with mean x mtrue and standard deviation x Ns . They are evaluated for mean
values x m , and C-spec errors x Ce , here for an error probability E = 0.05.
A B C D E F G H
1 xmTrue 6 xmA 6.06 =AVERAGE(A) E 0.05
2 xNs 1 xmB 5.69 =AVERAGE(B) tA 2.13 =T.INV.2T(E;15)
3 xmC 5.99 =AVERAGE(A;B) tB 3.18 =T.INV.2T(E;3)
4 tC 2.09 =T.INV.2T(E;19)
5 =NORM.INV(RAND();xAtrue;xNs) xCeA 0.55 =STDEV.S(A)/SQRT(16)*tA
6 A B xCeB 0.75 =STDEV.S(B)/SQRT(4)*tB
7 6.42 5.15 xCeC 0.44 =STDEV.S(A;B)/SQRT(20)*tC
22 4.98
Fig. 9.13 (S) Sets A and B, normally distributed random numbers; evaluation of the data sets and
C = A|B for mean value (C1:D3), as well as the C-spec error of the mean (F5:G7) for E in G1
I J K L M N O P
1 wA 3.55 =1/xCeA^2 xmAB 6.01 =(xmA*wA+xmB*wB)/(wA+wB)
2 wB 0.51 =1/xCeB^2 vAB 0.25 =(wA+wB)^-1
3 wC 4.62 =1/xCeC^2 xsAB 0.50 =SQRT((wA+wB)^-1)
4 100000
5 hitA TRUE =AND(xmA-xCeA<xmTrue;xmTrue<xmA+xCeA) 0.941
6 hitB TRUE =AND(xmB-xCeB<xmTrue;xmTrue<xmB+xCeB) 0.950
7 hitC TRUE =AND(xmC-xCeC<xmTrue;xmTrue<xmC+xCeC) 0.944
8 hitAB TRUE =AND(xmAB-xsAB<xmTrue;xmTrue<xmAB+xsAB) 0.925
9 0.950 =1-E
Fig. 9.14 (S) In O5:O8 are the hit rates of the experiments with the four sets A, B, C, and AB for
100,000 repetitions. The value expected from statistical theory is given in O9
Table 9.10 Defining four measurement series A, B, C, and AB; calculating their means, vari-
ances, and C-spec errors; t A , t B , t C are global variables from the main program in Table 9.12;
continued in Table 9.11
2
3 def measurements():
4 A=xmTrue+xNs*npr.randn(16)
5 B=xmTrue+xNs*npr.randn(4)
6 C=np.concatenate([A,B]) #Not independent of A and B
7 xmA=np.mean(A)
8 xmB=np.mean(B)
9 xmC=np.mean(C)
#C-spec errors by considering t:
10 xCeA=np.std(A,ddof=1)/np.sqrt(len(A))*tA
11 xCeB=np.std(B,ddof=1)/np.sqrt(len(B))*tB
12 xCeC=np.std(C,ddof=1)/np.sqrt(len(C))*tC
13 hitA=(xmA-xCeA)<xmTrue<(xmA+xCeA)
14 hitB=(xmB-xCeB)<xmTrue<(xmB+xCeB)
15 hitC=(xmC-xCeC)<xmTrue<(xmC+xCeC)
Question
Which two of the three sets A, B, and C are pairwise independent?23
In Fig. 9.14 (S), first, the results of A and B are combined into one final result
x mAB with a calculated error x sAB . Then, the statistical experiment is repeated
100,000 times with a rep-log procedure to count how often the error ranges
x m ± x Ce capture the true value for the sets A, B, C, and AB. The theoretical
rate is 0.950, which is closely reached for A, B, C, but with ≈ 0.925 not for AB.
We conclude that the internally consistent variance is not sufficient to estimate
the error range. Therefore, in the following section with Python programs, the
alternative with externally-consistent error is also considered.
9.6.4 Python, Internally and Externally Consistent Error

of the Combined Result
The Python function in Tables 9.10 and 9.11 simulates the measurement process.
In the function in Table 9.10, the data sets A, B, and C are generated and evaluated
for mean value x m , variance v, and C-spec error x Ce of the mean.
Table 9.11 continues the function measurements begun in Table 9.10. The com-
bined result x mAB , x sAB is built, where x sAB is obtained as the maximum of the
23A and B are independent of each other, because their members are generated in two different
ranges. C, A, and C, B are not independent, because C is the union of A and B.
Table 9.11 Continuation of Table 9.11; line 22 is varied according to which variance is to be
calculated
16 wA=(xCeA)**-2 #Weight
17 wB=(xCeB)**-2
18 xmAB=(xmA*wA+xmB*wB)/(wA+wB)
19 vABint=(wA+wB)**-1 #Variances combined
20 vABext=(wA*(xmA-xmAB)**2
21 +wB*(xmB-xmAB)**2)/(wA+wB)
22 vABmax=max(vABint,vABext) #=vABint
23 xsAB=np.sqrt(vABmax)
#Confid. level not yet clear. Hit rate to be determ.:
24 hitAB=(xmAB-xsAB)<xmTrue<(xmAB+xsAB)
25 return hitA, hitB, hitC, hitAB
hits (True, False, True, True)
Table 9.12 Main program calling the function measurements in a for-loop. The values in the right
cell are for the current specification E = 0.05 and for another run with E = 0.317
2 E=0.05 theor. 0.950
3 tA =sct.t.ppf(1-E/2,df=15) Hits A 0.951
4 tB =sct.t.ppf(1-E/2,df=3) Hits B 0.951
5 tC =sct.t.ppf(1-E/2,df=19) Hits C 0.948
6 Hits AB 0.936
7 xmTrue=6 #True mass
8 xNs=1
9 Nrep=10000
10 hitA,hitB,hitC,hitAB=0,0,0,0 theor. 0.683
11 for rep in range(Nrep): Hits A 0.677
12 hits=measurements() Hits B 0.682
13 hitA+=hits[0] Hits C 0.683
14 hitB+=hits[1] Hits AB 0.688
15 hitC+=hits[2]
16 hitAB+=hits[3]
internally and externally consistent variances. Line 24 checks whether the true
value x mTrue is captured by x mAB ± x sAB , and the four Boolean values are returned.
In the main program in Table 9.12, the statistical experiment is repeated N rep =
10,000 times for a specified error probability E, and the hit rates are determined.
The results are shown in the right cell. The hit rate for AB with 0.936 is not much
closer to the theoretical value of 0.950 than the 0.925 reported in Fig. 9.14 (S) for
the internally consistent variance.
9.7 Propagation of Standard Deviations 397
9.7 Propagation of Standard Deviations
You cannot prevent errors from propagating. It is in their nature. Learn to

live with it! We simulate the propagation of variances and standard devia-
tions in sums, products, and powers with statistical experiments on sets with
100 elements. Note: For products, the relative variances add up; for sums,
the absolute variances. The propagation of confidence intervals is treated in
Exercise 9.8.
9.7.1 Rules for Propagation of Standard Deviations
General Rule
The final result z of a measurement series is often a function of one or more measured
physical quantities x: z = f(x 1 , x 2 , …). The standard theory of error propagation
considers the x i as random variables, normally distributed around the “true” value,
with the result:
The variance of the final result is a weighted sum of the variances

var (xi ) of the measured values of the individual variables x i .
Remember: It is the variances that propagate through the formu-
las to the final result. But, ultimately, the C-spec error has to be
reported together with the estimate of the mean.
For two variables, x and y, the equation is

2 2
∂z ∂z
var z = wx var x + w y var y with wx = and wy = (9.16)
∂x ∂y
which can straightforwardly be extended for more variables. The variables v denote
the empirical variance of the data series, Eq. 9.2. The weights w in the sum are the
squares of the derivatives of z with respect to the associated variables; the greater
the slope, the greater the weight.
As an example, we consider the calculation of the volume V K of a sphere from
its diameter d K : VK = π6 d K3
π 2
var(VK ) = d K2 · var(d K ) (9.17)
2
10 100
8 80 y = 0.93x
R² = 0.99
xTy_Sd_calc
xPy_Sd_calc
y = 1.00x
6 R² = 1.00 60
4 40
xPy; xTrue=10; xTy; xTrue=10;

2 20
yTrue=10 yTrue=10
0 0
0 2 4 6 8 10 0 20 40 60 80 100
xPy_Sd xTy_Sd
Fig. 9.15 a (left) Standard deviation of the sum xPy = x + y, calculated versus empirical val-
ues. b (right) Standard deviation of the product xTy = x·y, calculated versus empirical values. The
standard deviation is up to ten times greater than the mean value. X True , yTrue are the true values
Question
What are the physical units on both sides of Eq. 9.17?24
In the following sections, we examine the propagation of the standard deviations

of sets of 100 measurement points into the final result.
Error propagation in sums

For a sum z = x + y, the weights in Eq. 9.16 are wx = 1 and wy = 1 so that var z =
var x + var y . For a difference z = x − y, the coefficients are the same because the
partial derivatives are squared so that we get the rule:
The variance of a sum or difference is the sum of the variances of the

summands.
For the data in Fig. 9.15a, two data sets x and y with various standard devia-
tions for a normally distributed noise are added element-wise to get xPy (x Plus
y). The standard deviations of the set of the sums, as derived from the standard
deviations of x and y, are plotted against the empirically determined standard devi-
ations xPy Sd of the set xPy. The trend line has a slope 1.00, indicating accordance
with the theoretical expectation. In general, that cannot be taken for granted, as
24 The dimensional analysis m6 = m4 ·m2 shows that Eq. 9.17 does not contain gross errors.
can be seen in Fig. 9.15b for xTy (x Times y), where the trend line through a
corresponding plot has a slope of only 0.93.
Questions
Let the two measurands x and y have standard deviations x Sd and ySd , respec-
tively. How large is, according to the propagation rules, the standard deviation
of the sum x + y, and how large is the standard deviation of the difference x -
y?25
The proportionality factor in the trend line in Fig. 9.15b is, with 0.93, clearly
smaller than 1. Does this contradict theory?26
Task Vary the mean values and standard deviations of x and y and log the mean
value and standard deviation of the sum xPy! Which regularity do you assume?
Derive the expected result from Eq. 9.16!
Error propagation in products

In Fig. 9.15b, the experimentally determined standard deviation xTy Sd of the set
xTy is compared with the theoretical standard deviation xTy Sd calc calculated from
the standard deviations of x and y. For the product z = x·y, Eq. 9.16 yields the weights
2 2
∂z ∂z
wx = = y and w y =
2
= x2 (9.18)
∂x ∂y
hence,
var z = y 2 · varx + x 2 · var y

varz varx var y
2
= 2 + 2
z x y

z Sd 2 x Sd 2 y Sd 2
= + (9.19)
zm xm ym
The square of the relative standard variations of a product is the sum

of the relative standard variations of the factors.
Ψ Calculate with variances, report the standard error!

Ψ Even better: report confidence level and confidence interval!
25 The standard deviations of the sum and the difference are equal: (x + y) Sd = (x − y) Sd =
2 + y2 .
x Sd Sd
26 The formulas for error propagation, here, the square root of Eq. 9.19 for products, are based on
a Taylor series development and are, therefore, valid only for small variances of the independent
variables, a condition no longer satisfied here.
Standard deviations that are too large distort the result

In Fig. 9.15b, the standard deviation of the product, calculated according to the
propagation formula, is plotted against the empirical one. The slope of the regression
line is 0.93, significantly smaller than 1. This is because the standard deviations of
the measurands are too large. The formulas for error propagation are based on a
Taylor series development of the function, in Eq. 9.16 only to the first order. So, they
are valid only for a small interval around the independent variables. The standard
deviations in Fig. 9.15b become too large. For small standard deviations (< 60), the
slope is greater than that of the trend line drawn in the figure, actually 0.99, as an
appropriate check shows.
Error propagation in powers

We now investigate the power function
z = xn (9.20)
The variance of the power function is derived from Eq. 9.16 to

2
∂z 2
var z = ax var x with ax = = n xmn−1 (9.21)
∂x
Thus, the expected standard deviation of z is
z Sd = n · xmn−1 · x Sd (9.22)
Figure 9.16 displays the empirical standard deviation for a power n = 2 (in a) and
n = 4 (in b), both for a noise level x Ns = 0.1, as a function of the mean value x m .
2.0 300
y=x^2; xNs=0,1 y=x^4; xNs=0,1
1.5
xPWn_Sd
xPWn_Sd
y = 0.20x1.00 200 y = 0.47x2.90

R² = 0.99 R² = 1.00
1.0
100
0.5
0.0 0
0 2 4 x 6 8 10 0 2 4 x 6 8 10
m m
Fig. 9.16 a (left) Standard deviation of the second power of the random variable x with the noise
level x Ns = 0.1 as a function of the empirical mean x t of the data set. b (right) Same as a, but with
a power of n = 4
Question
Do the formulas for the trend lines in Fig. 9.16a and b agree with the theoretical
expectation?27
Experimental procedure
We set up two data series (x, y) (size 100 each), defined as a true value (x True or yTrue )
plus a normally distributed noise (standard deviations x Ns , yNs ). We then determine
their mean values (x m , ym ) and standard deviations (x Sd , ySd ).
We add and multiply the two sets pairwise to get the new series xPy (“x Plus y”)
and xTp (“x Times y”) and determine their mean values xPym and xTym and standard
deviations xPySd and xTySd , respectively. Furthermore, we create a new series xPWn
by raising the elements of x to the power n and determine its mean value xPWnm
and standard deviation xPWnSd .
We vary the noise levels and log the mean values and standard deviations of the
sets with a rep-log procedure. This is achieved:
– in excel with a VBA routine that repetitively changes the parameters of the
spreadsheet calculation,
– in Python by outsourcing the parts of the program equivalent to the spreadsheet
into a function, repetitively called by the main program.
We plot the theoretically calculated standard deviations of xPy and xTy ver-
sus the empirical ones in a scatter plot and lay a straight trend line through the
data points (see Fig. 9.15). The slope of such trend lines is 1 if the theoretically
calculated values correspond to the empirical ones.
We set up another set x PW n by raising the elements of the series x to the nth
power and present its standard deviation as a function of the mean value x m of the
series x, together with a power trend line through the data (see Fig. 9.16).
Questions
We have called the collections x, y, xPy, xTy data series, not data sets. Why is
the term “set” inappropriate here?28
As a result of our simulations, we plot the standard deviations of sums and
products against the standard deviations of the summands or factors, without
considering either the number n of measurements or the t values. Why does
that not hinder the illustration of the laws of error propagation?29
27 Yes, we expect z Sd ∝ x 3 for f (x) ∝ x 4 and z Sd ∝ x for f (x) ∝ x 2 and find y = 0.47x 2,90 und
y = 0.20x .1.00 .
28 A set is an unordered collection. However, our data are ordered. The nth element of xPy and xTy
is obtained from the nth elements of x and y.

29 The sets for all terms have the same number n of measurements, so that the factors n and t are
common to all of them and to the results.

x True , yTrue true values

x Ns , yNs noise levels (standard deviation of a normal distribution)
x, y two series of true values plus noise
x m , ym mean of x and y
x Sd , ySd standard deviations of x and y
xPy series x Plus y
xTy series x Times y
xPWn series x to the PoWer of n
xPy_m, xPy_Sd mean and standard deviation of xPy, correspondingly for xTy
and yPWn
xPy_Sd_calc standard deviation of xPy calculated from x s and ys , correspond-
ingly for xTy and yPWn.
In A13:B112 of Fig. 9.17 (S), we create two series x and y of normally distributed
numbers, regarded as measurement values specified by their means (“true values”)
and standard deviations (“noise”) in A1:B4. The series x and y are evaluated for
mean value and standard deviation in D1:E4.
We then create three new data series in D13:E112 by adding x and y (xPy), mul-
tiplying x and y (xTy), and raising x to the power n (xPWn) specified in B5. These
series are evaluated in Fig. 9.18 (S). There, the standard deviations xPy_Sd_calc,
xTy_Sd_calc, and xPWn_Sd_calc are also reported, theoretically derived from the
standard deviations of the original series.
Error propagation in sums

In Fig. 9.18 (S), the empirical standard deviation xPy_Sd is calculated from the
series xPy. The theoretical standard deviation xPy Sd calc is calculated as the root
of the sum of the squares of the summands’ standard deviations. The two values are
identical to the second decimal.
A B C D E F G A B C D E F
1 xTrue 10.00 xm 10.00 =AVERAGE(x) Ns)
2 =STDEV.S(x) rue;x
xNs 0.10 xSd 0.11
AND();xT
3 yTrue 10.00 ym 9.54 =AVERAGE(y) .INV(
R
4 yNs 7.50 ySd 7.56 =STDEV.S(y) 11 =NORM =x+y =x*y =x^n
5 n 2.00 12 x y xPy xTy xPWn
13 10.06 17.89 27.95 180.02 101
112 10.00 17.31 27.31 173.07 100
Fig. 9.17 (S) Two normally distributed data series x and y of size 100, defined by their means
x True and yTrue (“true” values) and standard deviations x Ns and yNs (“noise”).in a Gaussian
We vary the standard deviations of x and y independently of each other sys-

tematically with a rep-log procedure, record the empirical and calculated standard
deviations of the sum, and plot both as data points in Fig. 9.15a. The trend through
the data points is a straight line with a slope very close to 1. Thus, we have retrieved
the theoretical statement with our simulation; at least, we cannot object to it.
Error propagation in products

The product of x and y is stored as a new series xTy in column E of Fig. 9.17 (S) and
evaluated in column M of Fig. 9.18 (S).
We vary the standard deviations of x and y independently of each other systemati-
cally with a log procedure and record the product’s empirical and calculated standard
deviations. The standard deviation xTy Sd calc of the product xTy is calculated from
those of the factors as (see Eq. 9.19)
x T y SdCalc = z SdCalc = (ym x Sd )2 + (xm y Sd )2 (9.23)
The results are shown in Fig. 9.15b. The trend line has a slope of 0.93, less than
the expected 1.00. The reason for this is that the standard deviations x Sd and ySd are
too large (see discussion in Sect. 9.7.1).
Error propagation in powers
Task Change n in Fig. 9.17 (S) or Table 9.13 and review your interpretation!
In Fig. 9.17 (S), column F, x is raised to the nth power; the result is stored as
xPWn. The standard deviation xPWn_Sd of this series is determined in Fig. 9.18
(S) in M8 together with the theoretical one xPWn Sd calc in M9.
In a rep-log procedure, the value of x True is varied from 0.5 to 10 for a noise
level of 0.1. The results for n = 4 and n = 2 are shown in Fig. 9.16a and b,
respectively; they are approximated by a power function as a trend line.
I J K L M N
1 x+y x*y
2 xPy_m 20.34 =AVERAGE(xPy) xTy_m 103.51 =AVERAGE(xTy)
3 20.34 =xm+ym 103.39 =xm*ym
4 xPy_Sd 7.79 =STDEV.S(xPy) xTy_Sd 77.78 =STDEV.S(xTy)
5 xPy_Sd_calc 7.77 =SQRT(xSd^2 xTy_Sd_calc 77.65 =SQRT((ym*xSd)^2
6 +ySd^2) +(xm*ySd)^2)
7 x^n
8 xPWn_m 100 =AVERAGE(xn) xPWn_Sd 1.75 =STDEV.S(xPWn)
9 xm^n 100 =xm^n xPWn_Sd_calc 1.75 =n*xm^(n-1)*xSd
Fig. 9.18 (S) Continuation of Fig. 9.17 (S); evaluation of the data sets xPy (“x Plus y”), xTy (“x
T imes y”) and xPWn (“x to the power of n”)
Table 9.13 Python function, part corresponding to Fig. 9.17 (S) continued in Fig. 9.18 (S)
2 I=np.zeros(14)
3 xTrue, yTrue, n = 10, 10, 2 #True values
4 def simu(xNs,yNs):
5 x=xTrue+xNs*npr.randn(100) #Array with 100 items
6 y=yTrue+yNs*npr.randn(100)
7 xm=np.mean(x)
8 xSd=np.std(x)
9 ym=np.mean(y)
10 ySd=np.std(y)
11
12 xPy=x+y
13 xTy=x*y
14 xPWn=x**n
Question/task
Write the formula in N9 of Fig. 9.18 (S) in mathematical form!30
The spreadsheet calculations can be translated directly into a Python program,

Tables 9.13 and 9.14. The program’s core is the function simu calculating with the
parameters x True , yTrue , n, x Ns , yNs . The parameters x Ns and yNs are chosen to be
handed over in the function head, whereas x True , yTrue , and n are treated as global
parameters.
The arrays x and y are created anew within the function with every call, and con-
sequently, the scalars and arrays derived from them are recalculated. The means,
empirical, and calculated standard deviations for every set are returned as an array
I with 14 elements (Table 9.14).
Error propagation for sums and products

The main program for calculating the standard deviations of the sum, displayed in
Fig. 9.15a, is given in Table 9.15. The standard deviations x Ns and yNs are varied,
whereas the true values x True , yTrue are specified before the for-loop and remain
constant during the whole experiment.
Similar main programs work for the product with:

xTy.append(I[7]).
30 Z = x n ; Z Sd = n x n−1 x Sd .
Table 9.14 Continuation of Table 9.13, part of the function simu corresponding to Fig. 9.18 (S)
15 I[1]=np.mean(xPy)
16 I[2]=xm+ym
17 I[3]=np.std(xPy)
18 I[4]=np.sqrt(xSd**2+ySd**2)
19 I[5]=np.mean(xTy)
20 I[6]=xm*ym
21 I[7]=np.std(xTy)
22 I[8]=np.sqrt((ym*xSd)**2+(xm*ySd)**2)
23 I[9]=np.mean(xPWn)
24 I[10]=xm**n
25 I[11]=xSd**n
26 I[12]=np.std(xPWn)
27 I[13]=n*xm**(n-1)*xSd
28 return I #Mean and standard deviation
(empirical and theoretical) of x*y, x*y, and xn
Table 9.15 Main program for calculating the standard deviation of a sum x + y for varying
standard deviations of x and y
29 xPy_Sd=[]
30 xPy_Sd_calc=[]
31 #Variation of the noise levels xNs and yNs
32 #Constant xTrue, yTrue
33 xTrue=10.0
34 yTrue=10.0
35 for noiseX in np.arange(0.5,8,1):
36 for noiseY in np.arange(0.5,8,1):
37 I=simu(noiseX,noiseY)
38 xPy_Sd.append(I[3])
39 xPy_Sd_calc.append(I[4])
xTy_Sd.append(I[8]).
Table 9.16 presents a program for achieving a plot such as that in Fig. 9.15a.
Power-law trend line

A power-law trend line y = a·x n is obtained with the function fitPowLaw in Table
9.17, by fitting a linear trend line logy = a + m·logx to the logarithmized data logx
= log(x) and logy = log(y). The transformation to y = aR xpR is achieved with pR =
m and aR = exp(a).
Error propagation for powers

Table 9.18 shows the main program for calculating the standard deviation of a power
x n as a function of the true value x True . A program similar to Table 9.16 achieves a
fit to the data points and results in a plot corresponding to Fig. 9.16.
Table 9.16 a (top) Program for producing a figure as in Fig. 9.15a, b (bottom) function to draw
a straight line through (x, y)
1 FigStd('xPy_Sd',0,10,2,'xPy_Sd_calc',0,10,2)
2 plt.plot(xPy_Sd,xPy_Sd_calc,'kD',ms=3,
3 label='StDev Sum;\nxTrue='+str(xTrue)+\
4 ';yTrue='+str(yTrue))
5 PlotLin(xPy_Sd,xPy_Sd_calc,a=False)
6 plt.legend()
7 import statsmodels.api as sm
8
9 def PlotLin(x,y,a=False): #Linear regression line
10 if a==False: #No y-axis intercept
11 model=sm.OLS(y,x)
12 results=model.fit()
13 mR=results.params[0]
14 yLin=mR*np.array(x)
15 plt.plot(x,yLin,'k-',label="y=%.2f*x"%mR)
16 if a==True: #y-axis intercept allowed
17 xx=sm.add_constant(x)
18 model=sm.OLS(y,xx)
20 aR=results.params[0]
21 mR=results.params[1]
22 yLin=aR+mR*np.array(x)
23 lbl="y=%.2f+%.2fx"%(aR,mR)
24 plt.plot(x,yLin,'k-',label=lbl)
Table 9.17 Plotting a power-law trend line by a straight line through logarithmized data
1 def fitPowLaw(x,y):
2 logx=np.log(x)
3 logx=sm.add_constant(logx) #y-axis intercept allowed
4 logy=np.log(y)
5 model=sm.OLS(logy,logx) #Linear regression
7 aR=np.exp(results.params[0]) #Amplitude
8 pR=results.params[1] #Power
9 R2=results.rsquared
10 #print(pR) #Only in test phase
11 yPow=aR*x**pR
12 lbl="y=%.2f*x**%.2f,R²=%.2f"%(aR,pR,R2)
13 plt.plot(x,yPow,'k-',label=lbl)
Table 9.18 a (top) Main program for the power series xPWn, simu from Table 9.13; b (bottom)
Plotting the original data and a power trendline through the data
1 xPWn=[]
2 xPWn_Sd=[]
3 xtp=[]
4 n=2
5 #Variation of xTrue for constant noise level
6 xNs=0.1
7 for x_1 in np.arange(0.25,10,0.25):
8 for rep in np.arange(2):
9 xTrue=x_1
10 I=simu(0.1,0.1)
11 xtp.append(I[10]**(1/n))
12 xPWn.append(I[12])
13 xPWn_Sd.append(I[13])
14
15 FigStd('xTrue',0,10,2,'xPWn_Sd',0,2,0.5)
16 plt.plot(xtp,xPWn_Sd,'kD',ms=3)
17 fitPowLaw(xtp,xPWn_Sd) #Function in table above
18 plt.legend()
9.8 Propagation of Confidence Intervals
We learn how to get from variances of two data sets to a confidence interval
for a combined result, sum or product of the two measurands. We use a
statistical simulation for the last step to get confidence-specified errors (“C-
spec errors”).
9.8.1 From Variance to Confidence
In Exercise 9.7, we investigate the propagation of standard variations in sums,

product, and powers by means of series of 100 elements. In real-world experi-
ments, usually much less measurements are performed so that t factors have to be
taken into account. Furthermore, the number of measurements can be different for
the different operands. As a consequence, the C-spec error of the result cannot be
deduced straightforwardly from the standard deviation of the result; we have to
calculate with C-spec errors also for the operands.
Measurement series for single measurands

In Fig. 9.19 (S), two measurement series x Ai and x Bi are simulated, with the true
values x A and x B and the noise levels nsA and nsB , specified in rows 1 and 2,
1 xAi xBi xA 4 xB 2
2 3.81 1.85 nsA 0.2 nsB 0.1
3 3.98 1.89
4 3.83 1.98 nA 4 =COUNT(xAi) nB 10 =COUNT(xBi)
5 3.84 1.92 xAm 3.87 =AVERAGE(xAi) xBm 1.94 =AVERAGE(xBi)
6 1.98 xAse 0.04 =STDEV.S(xAi)/SQRT(nA) xBse 0.03 =STDEV.S(xBi)/SQRT(nB)
7 1.98
8 2.12 Con 0.683
9 1.83 tA 1.20 =T.INV.2T(1-Con;nA-1) tB 1.06 =T.INV.2T(1-Con;nB-1)
10 1.97 xAce 0.05 =xAse*tA xBce 0.03 =xBse*tB
11 1.88
12 FALSE 6852 FALSE 6865
13 =AND(xAm-xAce<xA;xA<xAm+xAce) =AND(xBm-xBce<xB;xB<xBm+xBce)
14 10000 10000
Fig. 9.19 (S) Confidence intervals x Ace and x Bce for a confidence level Con for two measurement
series x Ai and x Bi
entering our standard function x Ai = [=norm.inv(rand();xa;nsa)] and x Bi =

[=norm.inv(rand();xb;nsb)] for the simulation of noisy measurements.
In rows 4 to 6, the number of elements (nA , nB ), the mean values (x Am , x Bm ),
and the standard errors (x ASe , x BSe ) of x Ai and x Bi are determined. Remember: the
standard error of the mean is obtained as square root of the variance of the series
(their standard deviation) divided by the square root of the number of elements in
the series, Eq. 9.12.
To come from the standard error x Se of the mean to a confidence interval x Ce
around the mean, we have to specify a confidence level. This is done in E8 with Con
= 0.683, valid for the standard error when the number of measurements is large.
Student’s t-values for this confidence level are determined in line 9 and multiplied
in line 10 with the standard error to yield the C-spec errors x Ace and x Bce of the
confidence intervals.
In E12 and I12, we check in the usual way whether the confidence intervals capture
the true values. Furthermore, with a rep-procedure, we check how often that is the
case for 10,000 repetitions of the statistical experiment (reported in F12 and J12).
The numbers are close to the expected values.
9.8.2 Sum and Product of Two Measurands
In Fig. 9.20 (S), we determine product AtP (“A times B”) and sum ApB (“A plus
B”) of the means x Am and x Bm of the two measurement series x Ai and x Bi and
calculate C-spec errors AtBce and ApBce for the errors obtained according to the
rules of error propagation explained in Exercise 9.7. They include tentative t fac-
tors tAxBemp and tApBemp , first set to 1 in N5 and S5. With these intervals, we
get hit rates of more than 7000 for 10,000 repetitions of the statistical experiment,
9.8 Propagation of Confidence Intervals 409
M N O P Q R S T U V
1 AtBtrue 8.00 =xA*xB ApBtrue 6.00 =xA+xB
2 AtBemp 8.10 =xAm*xBm ApBemp 6.01 =xAm+xBm
3 varA 0.001 =(xAce/xAm)^2 varAp 0.01 =xAce^2
4 varB 0.000 =(xBce/xBm)^2 varBp 0.00 =xBce^2
5 tAtBemp 0.959 tApBemp 0.986
6 AtBce 0.26 Apbce 0.12
7 =SQRT(varAp+varBp)*tApBemp
8 =AND(AtBemp-AtBce<AtBtrue;AtBtrue<AtBemp+AtBce)
9 TRUE 6805 10000 TRUE 6833 10000
10 1.03 =T.INV.2T(1-O9/P9;16) 1.00 =T.INV.2T(1-T9/U9;1000)
11 1.03 =T.INV.2T(1-Con;16) 1.00 =T.INV.2T(1-Con;1000)
12 1.01 =O11/O10 1.00 =T11/T10
13 0.96 =O12*tAxBemp 0.99 =T12*tApBemp
Fig. 9.20 (S) Product AtB (“times”) and Sum ApB (“plus”) of the results of the two measurement
series of Fig. 9.19 (S)
significantly larger than expected from a confidence level Con = 0.683. The rea-
son for this discrepancy is that the rules for error propagation set up in Exercise
9.7 are mathematically correct only if the number of measurements is big enough.
C-spec error by adapting the t factor

We do not delve deeper into the mathematical derivation but reduce the hit rates by
adapting the tentative t factors. To do so, we calculate Student’s t value:
– for the empirically obtained hit rates (O12 and T12); they should approach the
confidence level after some iterations,
– for the envisaged confidence level Con (O11 and T11); they do not change in the
course of the simulation.
We build the ratio of the two t values and multiply with the current tentative t values
(results in P13, T13) and copy/paste the results into tAtBemp and tApBemp (N5, S5).
We repeat the statistical simulation and get hit rates 6738 (with t = 0.940 for AB )
and 6708 (with t = 0.962 for ApB). After a second iteration, the results are 6805
(with t = 0.959) and 6833 (with t = 0.986), indeed close to the values expected
for the specified confidence level. The quantities AxBce and ApBce are called the
confidence-specified errors, or short C-spec errors.
9.9 Mass of a Thin Film on a Glass Substrate
Our task is to determine the mass of a thin film on a glass plate. To this
end, the glass plate is weighed several times on a microbalance, before and
after being coated. The mass of the film is obtained with recursion to error
propagation and t statistics.
9.9.1 Instructions for Use for Accurate Measurements and Their

Results
A measuring process …
In an ideal coating process, a glass substrate of mass mSub = 1 g is coated with a
thin film of mass mF = 1 mg. To determine the film’s mass, the substrate is weighed
before and after coating. The balance’s nominal accuracy is 1 μg, which means that
the mass is displayed in grams, with six digits after the decimal point. However, the
measurement’s actual accuracy is significantly lower, e.g., due to disturbances by air
currents or building vibrations.
Questions
What is the ratio of the masses of the substrate and the thin film?31
How many digits after the decimal point can the layer’s mass be determined
nominally, if given in mg?32
By what percentage does the coating increase the mass of the sample?33
… and its simulation

The substrate is weighed n times before and n times after being coated. The process
is simulated by series mBef = mSub + noise and mAft = mSub + mF + noise, where
the noise is normally distributed with mNs as the standard deviation.
excel quotes each number with 15 digits, something like 0.569410526368089,
Python even with 16 digits. Therefore, the results of the simulated weighing have
to be rounded to six digits after the decimal point to mimic the display resolution of
the balance: round(…;6) = 0.569411 and np.round(…,6).
The following explanations are valid for the preceding simulation but also for real
measurements.
Immediate evaluation of the measurement results

The standard errors of the masses of the uncoated and coated substrate are calculated
as the standard deviation of the measurement series divided by the square root of
the number n of measurements in the series. In order to get the C-spec errors mBefCe
and mAftCe (half the width of the confidence intervals), the standard error has to be
31 1 g/1 mg = 1000; the substrate is 1000 times heavier than the film.
32 Displaying an accuracy of 1 μg, the films’ mass could naively be specified as, e.g., 1.001 mg.
33 1 mg of 1 g, corresponding to 0.1%.
9.9 Mass of a Thin Film on a Glass Substrate 411
multiplied by Student’s t-factor that is determined by the degree of freedom (dof =

n – 1) and the specified confidence level Con. In this exercise, we choose Con =
0.683 which corresponds to the confidence level of the standard error if the number
n of measurements is big enough.
The mass of the film is estimated as the difference in the means of the two
measurement series,
m fest = m aft − m Befm
Confidence interval of the estimated mass of the film

In a first approach, we estimate an error range ± mFc0 of the film’s mass with the
rule for sums and differences as square root of the sum of the squares of the C-spec
errors of the summands:
m Fc0 = m 2Be f Ce + m 2A f tCe (9.24)
We know from Exercise 9.8, that the confidence level of this estimate does not
necessarily correspond to the confidence level Con of the estimated masses of the
substrate before and after coating. In order to correct that, we have to multiply mFc0
with a t factor that is to be determined through a statistical simulation.
For this simulation, we take as parameters the estimated masses mBefM of the
uncoated substrate and mFest of the film, and a noise level mNs estimated as the
standard deviation of the measurement series of the uncoated and coated substrate.
In this exercise, these values are the result of a simulation but in laboratory praxis
such are the results of real measurements.
The number of trials in the simulation must be big enough so that the fluctuations
of the hit rate for the same t value are sufficiently small to avoid too large jumping
of the hit rate and assure improvement in the iterative adaptation of the t value.
Rounding the numerical results to relevant digits

The layer’s mass and its C-spec error are to be determined from the simulated mea-
surement data and written as a final result with rounded numbers. To give an example:
If the layer’s mass is determined as the difference of the weighings to be 0.0009748 g
with an error of 6.57 × 10–5 , the final result is 9.7(7) × 10–4 g or 9.7 ± 0.7 × 10–4 g:
0.0009748 0.00097
±0.0000657 ±0.00007
This means that the result can only be expressed sensibly with two digits and
that the uncertainty in the last digit is ± 7. An alternative notation is 0.97(7) mg. For
completeness, the confidence level of the reported error must be noted.
Questions
You get the following results for a single measurand: Mean value 0.0010818
g, standard error of the mean value 4 × 10−5 g. How do you report the final
result?34
Is 1.033 ± 0.037 more precise than 1.03 ± 0,04?35
“Official” provisional measurement result

In the report of the measurement results, you should state the result, here the film’s
estimated mass, and its C-spec error, and the level of confidence:
Mass of the film (C = 0.683) = 1.01 ± 0.06mg
If you do not calculate the C-spec error, you must state all values necessary for
the simulation.
Question
Which results of a measurement series for determining the film’s mass do you
have to state if you do not calculate the C-spec error?36
Syntactical differences between Excel and Python

In excel, [= round(norm.inv(rand();msub;mns);dsp)] is written into n = 5
cells with range name mBef , whereas in python, there is only one statement:
mBef = np.round(mSub + mNs ∗ npr.randn(nBef), dsp)
using the function npr.randn to generate normally distributed random numbers

that have to be multiplied by mNs to get the right scale of the measurement noise.
excel uses the function std.s to get the variance of a sample, whereas Python
uses np.std with ddof, the deduction of the degrees of freedom, to be specified
as a key argument; ddof = 1, in our case, e.g., np.std (mBef,ddof = 1).
34 1.08 × 10–3 g; 0.04 × 10–3 g → (1.08 ± 0.04) mg; uncertainty of measurement given as one stan-
dard error. Ψ Two inside and one out of applies if enough repetitions of the measurement have been
made or the error is extended by a t factor.
35 No, the standard error is also only an estimate and is affected by statistical error.
36 (1) Estimated masses of the uncoated substrate and the film, (2) the noise level of the measure-
ment series. These are the parameters necessary to perform a simulation for getting the C-spec
error.

mSub true mass of the substrate
mF true mass of the film
mNs measurement noise (standard deviation of a normal distribution)
dsp display precision of the scales
nBef , nAft number of measurements before and after coating
mBef array of the results of the weighing before coating
mAft array of the results of the weighing after coating
mBefM , mAftM means of mBef and mAft
mBefSe , mAftSe standard errors of mBef and mAft
Con confidence level
mBefCe , mAftCe C-spec errors of mBefM and mAftM for Con
mFest Estimated mass of the film, = mAftM − mBefM
mFSe standard error of mFest
mFce C-spec error of mFest .
Simulation of the weighing process

The parameters of the exercise are specified and the measurement series simulated
in Fig. 9.21 (S). All values are stored in named cells, with the names listed in column
A and in D1:E1. The uncoated substrate’s mass is measured as mBef = mSub +
noise, that of the coated substrate as msub + mF + noise. The noisy measurement is
simulated with the spreadsheet function norm.inv(rand();msub;mns), with mean
value mSub (or mSub + mF for the coated substrate) and standard deviation mNs .
Evaluation of the simulated or real-world weighing process

The calculation with these parameters is shown in Fig. 9.22 (S). The values in
D:E could also be the results of real measurement. The following evaluation and
simulation would be the same.
A B C D E F
1 mSub 1g mBef mA
2 mF 1.0E-03 g 0.999934 1.000833
3 mNs 1.0E-04 g 1.000047 1.001021 =ROUND(NORM.INV(RAND();mSub;mNs);dsp)
4 0.999899 1.001115
5 dsp 6 0.999952 1.000912 =ROUND(NORM.INV(RAND();mSub+mF;mNs);dsp)
6 1.000154 1.000854
Fig. 9.21 (S) A:C, “True” values of the masses of the substrate and the film and an assumed
value for the noise of the measurement process; dsp = number of digits in the scale display; D:E,
simulated results of measurement series for the uncoated and coated substrate.
H I J K L M
1 mBefM 0.999997 =AVERAGE(mBef)
2 mAM 1.000947 =AVERAGE(mA)
3 nBef 5 =COUNT(mBef) Con 0.683
4 nA 5 =COUNT(mA) tBef 1.14 =T.INV.2T(1-Con;nBef-1)
5 tA 1.14 =T.INV.2T(1-Con;nA-1)
6 mBefSe 4.62E-05 =STDEV.S(mBef)/SQRT(nBef) mBefCe 5.28E-05 =mBefSe*tBef
7 mASe 5.32E-05 =STDEV.S(mA)/SQRT(nA) mACe 6.08E-05 =mASe*tA
8 noise 1.11E-04 =(STDEV.S(mBef)+STDEV.S(mA))/2
Fig. 9.22 (S) Evaluation of five weighing processes each of the uncoated and the coated substrate,
here simulated in Fig. 9.21 (S). The results mBefSe and mAftSe are standard errors. The report could
also be the result of a real-world experiment
O P Q R S T
1 varBef 2.79E-09 =mBefCe^2 TRUE =AND(mFest-mFce<mF;mF<mFest+mFce)
2 varA 3.69E-09 =mACe^2 HitRate 673 From procedure
3 mFc.0 8.05E-05 =SQRT(varBef+varA) Trials 1000
4 tAd 0.891 tHR 1.12 =T.INV.2T(1-HitRate/Trials;4)
5 mFce 7.17E-05 =mFc.0*tAd tF 1.14 =T.INV.2T(1-Con;4)
6 mFest 9.50E-04 =mAM-mBefM tAd.New 0.91 =tAd*tF/tHR
Fig. 9.23 (S) Setting an error mFce and determining its hit rate and therefrom its t factor t HR . The
t factor t F for the confidence level Con is calculated in S5
Questions
Why are the measurement results rounded to six decimal places in Fig. 9.21
(S)?37
Which parameter in Fig. 9.22 (S) characterizes the measuring accuracy of
the weighing process?38
Confidence interval by statistical simulation

In Fig. 9.23 (S), we first calculate an error mFc0 for the film mass (P3) with the rule
of error propagation for sums and differences and then multiply it with a tentative
t-value t Ad to calculate a C-spec error mFce (P5).
In the beginning, we had set t Ad = 1 and obtained a hit rate 731 for 1000 trials,
more than the 683 expected from the confidence level Con. For the first iteration, we
replaced t Ad with t AdNew to get a hit rate of 673, close enough to the expected 683 to
state the resulting mFce = 7E-05 as the C-spec error for the confidence level Con =
0.683. The old tentative C-spec error is multiplied by the ratio of the t value t F for
the desired confidence level and the t value for the current hit rate.
37 This corresponds to the display accuracy of the balance.

38 The accuracy of the current measuring process is characterized by the noise during weighing.
In the case of Fig. 9.22 (S) I8, it is 1.11E-4 g, estimated as the average standard deviation of the
measurement series.
Table 9.19 Parameters chosen for the simulation

1 mSub=1.00 #[g] True mass of the substrate
2 mF=1.0e-3 #[g] True mass of the film
3 mNs=1.0e-4 #[g] Measurement noise
4 dsp=6 # Number of displayed digits
5 nBef=5 # Number of measurements
6 nAft=5
7
8 Con=0.683 # Confidence level
The parameters for the simulation of the weighings are specified in Table 9.19.
The main program to determine the hit rates of the confidence intervals is given
in Table 9.20. It is essentially a for-loop calling a function InRange which takes the
t factor t Ad as input and returns a Boolean inRa that states whether the confidence
interval captures the true value, and the C-spec error mFce .
The function InRange is given in Table 9.21. It recurs to two functions Mass-
Before and MassAfter reported in Table 9.22. MassBefore simulates the measuring
process and returns the estimated mass mBefM of the substrate before coating
together with its C-spec error mBefC . MassAfter does the equivalent for the mass
of the substrate after coating.
The result of the for-loop is a hit rate that is used to calculate a new t value
t AdNew expected to be related to the specified confidence level. For the next
iteration, the value of t AdNew is inserted into line 3 by hand.
The final result is rounded with the function FinRes from Sect. 9.3.3 to the
relevant number of digits (see Table 9.23).
Table 9.20 Main program to determine the hit rate and the C-adjusted t value t AdNew . For a new
run to iteratively improve t, insert t Adnew manually in line 3
1 Trials=10000
2 HitRate=0
3 tAd=1 #To be adjusted for chosen C
4 for i in range(Trials):
5 inRa,mFce=InRange(tAd) #Defined in table below
6 if inRa==True: HitRate+=1
7
8 tF=sct.t.ppf(1-(1-Con)/2,nAft-1) #Target value of t
9 tHR=sct.t.ppf(1-(1-HitRate/Trials)/2,nAft-1)
10 tAdNew=tAd*tF/tHR
tAd 1.00 HitRate 7171
tF 1.14
tHR 1.24
tAdNew 0.92 HitRate 6725
Table 9.21 Function to determine the C-spec error mFce of the film mass and the logical value
whether the confidence interval captures the true value, recurs to functions in Table 9.22 that
perform the simulations
11 def InRange(tAd):
12 #Estimate mass of the film!
13 mBefM,mBefCe=MassBefore() #In table below
14 mAftM,mAftCe=MassAfter() #In table below
15 mFest=mAftM-mBefM #Film mass
16 varBef=mBefCe**2
17 varAft=mAftCe**2
18 mFc0=np.sqrt(varBef+varAft)
19 mFce=mFc0*tAd #C-spec error
20 inRa=mFest-mFce<mF<mFest+mFce
21 return inRa,mFce
Table 9.22 Simulation of the weighing of the substrate before and after being coated
2
3 def MassBefore():
4 #Mass of the substrate before coating:
5 mBef=np.round(mSub+mNs*npr.randn(nBef),dsp)
6 mBefM=np.average(mBef) #Mean
7 mBefSe=np.std(mBef,ddof=1)/np.sqrt(nBef)
8 tBef=sct.t.ppf(1-(1-Con)/2,nBef-1)
9 mBefCe=mBefSe*tBef #C-spec error
10 return mBefM,mBefCe
11
12 def MassAfter():
13 #Mass of the substrate after coating:
14 mAft=np.round(mSub+mF+mNs*npr.randn(nAft),dsp)
15 mAftM=np.average(mAft) #Mean
16 mAftSe=np.std(mAft,ddof=1)/np.sqrt(nAft)
17 tAft=sct.t.ppf(1-(1-Con)/2,nAft-1)
18 mAftCe=mAftSe*tAft #C-spec error
19 return mAftM,mAftCe
mAftM 1.000913 mAftCe 8.06e-05
mBefM 0.999904 mBefCe 4.08e-05
mFest 0.001009
Table 9.23 Rounding the result to the relevant number of digits with the function defined in
Exercise 9.3
1 Result=FinRes('Mass of the film (C=0.683) ',
mFest*1000,mFc*1000," mg")
Mass of the film (C=0.683) =1.01±0.06 mg
Explain the broom rules:
1. Ψ Twice as good with four times the effort.

2. Ψ Two within and one out of .
3. Ψ Worse makes good even better.
4. Ψ Mostly, not always.
5. Ψ Report the C-spec errors but calculate with their squares!
Evaluation of a measurement series

A quantity x was measured 9 times, with the resulting mean value x m = 10 and
standard deviation of the nine individual measurements x Sd = 1.8.
6. Which spreadsheet and which Python function do you use to simulate this
measurement series?
7. What is the standard error of the mean value?
8. A measurement series yields, as the mean value, x m = 7.12546 × 104 and, as
the standard error of the mean value, x Se = 6.28743 × 102 . How do you specify
the measurement result correctly rounded?
9. Does the formula x1 =round(0.847; 1) - round(0.155; 1) yield the same value
as x 2 = round(0.847--0.155; 1)?
10. In a departing airplane, the acceleration a is measured with two different meth-
ods, ten times each. The measurement results are a1 = 20 ± 1 m/s2 and a2 =
21 ± 0.5 m/s2 . Which value and which measurement uncertainty should the crew
report to the ground station?
Error propagation
11. A quantity z is the difference of two quantities z = s1 – s2 . The results of the

summands are s1 = 10 ± 2.24; s2 = 20 ± 2. How large are the difference z and
its error zSe calculated from the standard errors of the summands?
12. A quantity is the product of two quantities, p = p1 ·p2 . Measurements with many
repetitions result in p1 = 10 ± 1 and p2 = 20 ± 3.5. How big are the product p
and its error pSe calculated from the standard errors of the factors?
13. What is the difference between the C-spec error of a quantity that is calculated
as a function of several measurands and its error calculated from the standard
error of the measurands?
Fitting Trend Curves to Data Points
10
We create points on user-specified functions, transform them into measure-

ment points by adding noise to their y-values, and then fit appropriate trend
lines to the noisy data. Confidence intervals of the trend lines’ coefficients
are obtained using t statistics. In doing so, we learn how far we can trust the
parameters of the trend lines. We use functions for linear regression, linest
of Excel and OLS of the statsmodel library of Python, and nonlinear
regression, olver and curve_fit.
10.1 Introduction: Linear and Nonlinear Regression
Solutions of Exercises 10.2 (Excel), 10.3 (Python), 10.4 (Excel), and 10.6 (Python)
can be found at the internet adress: go.sn.pub/26leyH.
10.1.1 Straight Line Through Data Points by Sight
From the beginner’s physics lab course, we are familiar with a procedure for
getting physical parameters from measured data by suitably plotting them and
drawing a straight line through the data points with a ruler. This way, the two char-
acteristic parameters of a straight line, slope and y-axis intercept, can be obtained.
Below are two examples.
(a) The Curie–Weiss law for the temperature dependence of the magnetic suscep-
tibility χ m of a ferromagnet above the Curie–Weiss temperature ,

420 10 Fitting Trend Curves to Data Points
C
χm = (10.1)
T −θ
leads to a linear plot of the reciprocal of the measured data, 1/χ m , versus T
1 1
= ·T − (10.2)
χm C C
from which the parameters C (Curie–Weiss constant) and Θ can be obtained.
(b) The reaction rate R of a chemical reaction as a function of the absolute

temperature T, with k being the Boltzmann constant

E0
R = R0 · exp − (10.3)
kT
is logarithmized and plotted over 1/T (Arrhenius plot):

E0 1
ln R = ln R0 − · (10.4)
k T
from which the collision rate R0 of the reactants and the activation energy E 0 of
the reaction can be obtained.
These are examples of linear regression.
10.1.2 Multilinear Regression
One method for finding an optimum linear function through data points is linear
regression. The linear regression models in excel and Python apply to functions
with several independent variables x i . They are based on the general multilinear
form:
y = a + m 1 x1 + m 2 x2 + m 3 x3 + . . . (10.5)
We will apply multilinear regression to points scattered around a straight line,

a parabola y = a + b·x + c x 2 , and an exponential function ln(y) = ln(A) + m·x.
10.1 Introduction: Linear and Nonlinear Regression 421
Trend lines in charts, spreadsheets, Python

In excel, you can insert trend lines of the linear, exponential, potential, logarithmic,
polynomial, or power type into xy-diagrams (“scatter diagrams”). Coefficients of
a multilinear trend line are obtained, together with their standard errors, with the
linest function of excel or the corresponding OLS (“ordinary least squares”) in
Python.
Both methods, linest and ols, output a standard error for each coefficient whose
exact meaning depends on the number of data points and the number of parameters
estimated from the data set, generally speaking, on the degrees of freedom. In our
exercises, we check, with multiple repetitions of a random experiment, whether we
have correctly recognized the degrees of freedom and correctly set the confidence
limits. If this is the case, our broom rule Two within and one out of (the standard
error range) must hold.
This check is possible because our data points are generated with “true” values
on a curve and newly blurred with random noise for every repetition; then, the
trend parameters are again estimated from the noisy data set. Ψ We know everything
and play stupid.
Procedure of a random experiment

In order to check whether the hit rate corresponds to an expected confidence level,
we
– generate “true” data points (x, y) from x-values and a function y = f(x),
– blur them with normally distributed noise,
– transform the noisy values into a linear form,
– calculate the coefficients of a linear trend line through the noisy data, together
with their standard errors extended by a t-value,
– check whether the error ranges capture the true values,
– determine the hit rate, i.e., how often in a series of simulations the error range of
the trend line’s coefficients does capture the true values.
10.1.3 Nonlinear Regression
If the measurements are subject to background noise, a constant must be added to

the trend function. Then, no straight-line equations, such as those in Eq. 10.2 or
Eq. 10.4, can be obtained by coordinate transformations. In these cases, nonlinear
regression must be applied. Likewise, if the function itself is not linearizable by a
coordinate transformation as, e.g., y = cos ω1 t + cos ω2 t.
In such cases, we generate the trend function to be adjusted with preselected
parameters and vary them so that the function passes through the measured data.
This can be done by hand and visual judgment, but also automatically with the
excel functions goal seek and solver, or with the functions curve_fit from
scipy.stats or minimize from the scipy.optimize library of Python.
For fitting trend lines to measurement data, there are special program packages,
e.g., Origin for the general case, SCOUT for fitting model dielectric functions
to optical spectra, and other programs for Rutherford backscatter spectra or
impedance spectroscopy. However, it has proven to be very useful for improving
their judgment when the students can organize simple adjustments in spreadsheets
or computer programs themselves. This way, they are experiencing nonlinear
regression behavior with examples in which they have all of the parameters in
their own hands.
10.1.4 Coefficient of Determination R2
Explained and residual variance

The coefficient of determination for any trend line is defined as
ex plained variance (vT ot − v Res )

R2 = = (10.6)
total variance vT ot
The terms in this formula are explained by means of Fig. 10.1 for a linear trend
line through the data points yNs . In part a, the distances to the horizontal y = ym
(mean y) are marked with vertical lines. The sum of the square of these distances is
vTot = 61.4, entering Eq. 10.6 as total variance vTot of the data set. In part b, the
trend line within the chart is plotted, together with its formula and the value of the
coefficient of determination R2 = 0.71 in the legend. Here, it is the distance of the
data points to the trend line that is marked with vertical lines. The sum of the squares
of these distances is vres = 17.9, entering Eq. 10.6 as residual variance vRes of the
trendline.
10 10
y = 0.94x + 2.05
R² = 0.71
8 8
y
y
6 6
4 y=1x+2
4
yNs
yNs
2 (xm, ym) 2 Vres=17.9
Vtot=61.4
0 0
0 2 4 x 6 8 0 2 4 x 6 8
Fig. 10.1 a (left) Original curve and noisy data points; distances of the data points to the line y =
ym are marked by vertical lines. b (right) The same data as in (a), together with a linear trend line.
Distances of the data points to the trend line are marked with vertical lines
10.1 Introduction: Linear and Nonlinear Regression 423
The variance explained by the trend line is the difference between the residual
variance and the total variance. In our example, Eq. 10.6 results in R2 = (61.4 –
17.9)/61.4 = 0.71, the same value as reported in Fig. 10.1b.
Adjusted R2
For the above calculation, the same degree of freedom has been supposed for vTot
and vRes , namely, N – 1, the number of data points minus 1. This is justified for vTot
because only one parameter, the mean value, is estimated from the data set. Still,
it is questionable for vRes because more parameters describing the trend line are
estimated from the data set, two for a straight line (ddof = 2 in Python). If the
variances are calculated with the actual degree of freedom N − ddof , adjusted R2
values result:

N −1
vT ot − v Res · N −ddo f
j =
2
Rad (10.7)
vT ot
leading to an expression with the easily obtainable entities R2 , N, and ddof .

N −1
j =1− 1− R ·
2 2
Rad (10.8)
N − ddo f
The term ddof stands for “delta degrees of freedom”. For a set of N data points,
we have the following degrees of freedom:
f =N –1 for the variance of the data set,

f =N –2 for the coefficients of a linear trend line (and an exponential trend line;
see Exercise 10.4),
f =N –3 for the coefficients of a parabolic trend line.
10.1.5 C-spec Error with Iterative t Adaptation
Real-world data
We can apply the procedure of a random experiment presented in Sect. 10.1.2 (
We know everything and play stupid) to real-world data when making two steps
beforehand:
– taking the experimental trend line through the data points as the “true” line and
– estimate the noise level as standard deviation of the residuals. The residuals are
the difference between the y values of the data points and the trend line.
For multilinear trend lines, we can check whether our assumptions about the
confidence level are correct.
Experimental t factors
The solver function in excel does not return standard errors of the optimized coef-
ficients, contrary to the function curve_fit of scipy.stats. Therefore, in
Sect. 10.7.6, we determine C-spec errors (“confidence-specified”) by a random exper-
iment with tentative error ranges that are iteratively adapted to yield a hit rate that
corresponds to the desired confidence level.
10.2 Linear Trend Line
We generate a set of points on a straight line (“generating line”), add nor-

mally distributed noise to their y-values, and then lay a straight trend line
through these “data points”. The coefficients of such a line and their standard
errors are determined with the spreadsheet function linest. In Python, this
is achieved with the function OLS (“ordinary least squares”) of the library
Statsmodels. The didactically most important message: Ψ Two within
and one out of , meaning that in about 1/3 of the cases, the “true” coefficients
of the generating line lie outside the standard error range of the estimated
coefficients.
10.2.1 Creating Data Points and Evaluating Them
A linear trend line is to be laid through data points, minimizing the square devia-
tion of the y-values. We generate the data points ourselves by randomly choosing
x-values within a specific range, calculating the corresponding y-values on a
straight-line ("true values"), and blurring them with normally distributed noise.
Then, we estimate the coefficients of an optimum linear trend line through the data
points and check whether the coefficients’ error ranges capture the true values.
Data points are generated

We generate a set of data points scattered around a straight line defined by
y =a+m·x (10.9)
with m being the slope and a the y-intercept. The independent variable x is randomly
equi-distributed between 0 and x Max . The values of y are blurred with a normally-
distributed noise specified by a standard deviation Ns:
y N s = y + noise (10.10)
The formula for noise is:

10.2 Linear Trend Line 425
– Ns * randn(n) in Python, n being the size of the array generated, and

– norm.inv(rand; 0; Ns) in excel, to be written into a range comprising n cells.
In excel, yNs could also be specified as a whole as

yNs (x) = norm.(rand(); y; Ns) .
The set (x, yNs ) is our data cloud.
Linear trend line through the data points

Our task is to determine a linear trend line
yR = m R · x + aR (10.11)
through the data cloud, which minimizes the sum of the quadratic deviations of yNs
to this line. This procedure is called linear regression.
Two examples are shown in Fig. 10.2, together with the linear trend line (by
definition, going through the center (x m , ym ) of the data cloud) and two straight lines
with coefficients with + or – their standard errors. The equations and the coefficients
of determination R2 of the trend line are reported in the legend. The trend line is
called the “linear regression line”. Its slope mR is called the regression coefficient.
We start with the “true” function y = 1x + 2 to finally get the trend line for the
noisy data:
yr = 1.16x + 0.98 with R2 = 0.78 for a noise level Ns = 1.4,
yr = 0.89x + 2.42 with R2 = 0.28 for a noise level Ns = 2.0.
Hit rates and degree of freedom dof

Statistical theory states that the “true” value is captured by the standard error range,
plus-minus standard error from the estimated value, in 68.2% of the cases. This
10 10
y = 0.89x + 2.42
y = 1.16x + 0.98 R² = 0.28
8 R² = 0.78 8
y y
6 6
4 yNs 4
yNs
(xm, ym)
(xm, ym)
2 a+da; m-dm 2 a+da; m-dm
a-da; m+dm a-da; m+dm
0 0
0 2 4 x 6 8 0 2 4 x 6 8
Fig. 10.2 Noisy data (diamonds) around the “true” line (y = 2 + x) with linear trend lines (solid)
and lines for coefficients + or – their standard errors. a (left) Noise level 1.4. b (right) Noise level
2.0; noise level enters as standard deviation into a normal distribution
statement applies exactly only if the number of measurements goes to infinity, but
our broom rule Two within and one out of applies as of 9 measuring points on a
straight line.
Tim Why now suddenly as of 9 measurement points? For measurement series,

we have learned that the rule holds as of 8 measurements.
Mag That’s because of a different degree of freedom. For measurement series,

the degree of freedom is the number of measurements -1, because only one param-
eter, the mean value, is estimated from the measurement series. For straight lines,
however, two parameters are estimated, namely, slope mR and y-axis intercept aR .
The degree of freedom dof = 7 is thus reached with 9 data points.
x set of independent variables starting at 1 and randomly increased by

a value < dx
a, m coefficients of a straight-line equation, "true values"
y y = a + m·x
Ns noise level entering as the standard deviation in a normal distribution
yNs y blurred with noise
x m , ym average of the sets of x and yNs
aR , m R coefficients estimated from a trend line through yNs
aSe, mSe standard error of aR and mR
R2 coefficient of determination
inA , inM true, if the error range captures the true value
hitA, hitM number of hits, counting how often inA , inM are true.
dof degrees of freedom, number of points – 2 for a linear trend line.
10.2.3 Spreadsheet Calculation with Linest
The program sketched in Sect. 10.2.1 is realized in the spreadsheet in Fig. 10.3 (S).
The values of x start at 1.00 and progress with random interspaces ≤ dx, as can be
seen from the formula in A7 valid for A10. The 10 x-values in A9:A18 start at 1
and are randomly increased by dx*rand(). The corresponding y-values in column
B are calculated with Eq. 10.9. A normally-distributed noise is finally added (yNs
in column C, Eq. 10.10) to yield the “data points” presented in Fig. 10.2 with
diamonds.
The coefficients mR , aR of the linear trend line through the data cloud are
calculated with linest in G4:H6, together with their standard errors mSe , aSe, and
the coefficient of determination R2 . With inM and inA (F9:G9), we check whether
the error range captures the true values of m and a. For inM , we take the standard
A B C D E F G H I J
1 y intercept a 2.00 y=1x+2 =LINEST(yNs;x;;1)
2 Slope m 1.00 mR aR
3 0<= x <=xMax dx 1.40 mSe aSe
4 Noise Ns 2.00 Estimated 0.864 3.362
5 Standard error 0.194 0.883
6 R² 0.713 1.304
)
*aSe
0 ;N s) R+ mSe) <aR+t.95
; m a
ND() m;m< e<a;
AND(
) V(RA ) yNs)R-mSe< R-t.95*aS
dx*R x+a NORMV.IENRAGEV(ExRAGEN(D (m D(a
7 =A9+ =m* =y+ =A =A =A =AN 10000 10000
8 x y yNs xm ym in_m in_a hitM hitA
9 1.00 3.00 4.55 4.03 6.85 TRUE TRUE 6502 9474
10 1.67 3.67 5.98 t.95 2.31 =T.INV.2T(0.05;8)
18 8.17 10.17 9.36
Fig. 10.3 (S) A linear function y = m·x + a is blurred with noise. linest in G4:H6 determines the
coefficients of a linear trend line through the data cloud. In F9:G9, we check whether the standard
error range captures the true values of m or the t-extended error range the true values of a. HitM
and hitA are returned from a VBA routine repeating the spreadsheet calculation 10,000 times
error mSe whereas for inA , we take the confidence interval for an error probability
E = 0.05. Hit M and hit A are returned from a VBA rep-log procedure running the
spreadsheet calculation 10,000 times.
Question
Questions concerning Fig. 10.3 (S):

What is the formula for yNs , the y-values of the data points, in C9:C18?1
Which parameter of the normal distribution is set by the parameter Ns in this
formula?2
Trend line in the diagram

In Fig. 10.2, linear trend lines have been drawn through the “data points”,
together with their equations and coefficients of determination R2 . This is achieved
by activating the data points and going through design/add chart ele-
ment/trendline/linear (excel 2019). Proceed further with activating the trend
line, going through format/series 1 trendline 1/ and setting check-marks in
display equation on chart and display r- squared value on chart.
Each time the spreadsheet is modified (for example, by “clearing” the contents of
an already empty cell), the noise is recalculated, and, accordingly, all data points in
the graph change, yielding a new trend line.
1 C9:D18 = [= y + norm.inv (rand(); 0; Ns)].

2 Ns indicates the standard deviation of the normal distribution.
Linest as matrix function

What are the standard errors in the coefficients of the trend line? They are not reported
in the legend for the trend line. To get them, we apply the linest spreadsheet function
implemented in G4:H6 of Fig. 10.3 (S), performing linear regression. Its syntax is
linest (known_ys; [known_xs]; [const]; [stats]) (10.12)
linest is a matrix function taking, as input, the Y-values (here, yNs from Fig. 10.3
(S)) and the X-values (here, x from Fig. 10.3 (S)) and outputting a matrix with 4 rows
and 2 columns, of which only the first three rows are of interest to us here.
As a reminder: To insert a matrix function into a spreadsheet, first, mark the range
into which the results are to be written; in Fig. 10.3 (S), it is the (3R × 2C) matrix
G4:H6 (highlighted in grey). Then, write [= linest (. After the parenthesis “(“,
the function window opens, indicating the expected input, Eq. 10.12. After entering
the ranges for the Y and X-values, you may enter control parameters that determine
whether the trend line should go through the origin (const = false or 0; default
is true) or whether statistical characteristics should be output (stats = true or 1;
default is false).
If const = 0 or false, then the straight line is drawn through the origin. If stats
= true, additional regression characteristics are output, e.g., the standard errors of
the coefficients and the coefficient of determination R2 . In our case, the trend line
should not be forced to go through zero; therefore, the corresponding site remains
empty. Regression characteristics are to be output; therefore, we set a 1 as the last
argument, thus the function is called as linest (yNs ; x;; 1).
Mag How do you confirm the input of a matrix function?
Alac With the magic chord!3
In the first row of the output matrix of linest are the estimates mR and aR for
m and a (note the order!). In the second row are the standard errors mSe and aSe
of these coefficients. In the third row on the left is the coefficient of determination
R2 , and on the right, the prediction error. The coefficients mR and aR , and R2
correspond exactly to those of the trend line in a diagram with the same values for
x and yNs .
The R2 values reported in the figures and by linest are not adjusted for the
reduced degree of freedom. To get R2 adj , use Eq. 10.8!
Hit rates and degree of freedom

How often does the standard error range capture the true value?
With a rep-log procedure, we count how often in m and in a in F9:G9 are true.
They are true if the respective error intervals (mR – mE , mR + mE ) and (aR – aE ,
3 ctl + shift + return.

aR + aE ) capture the true coefficients. We get hit rates of hitM = 6502 for 10,000
trials, closely corresponding to the theoretical value C = 0.653 for 8 degrees of
freedom obtained with 1-t.dist.2t(1; 8) = 0.653 (2 * sct.t.cdf(1, 8)-1
in Python). The 8 degrees of freedom are given by 10, the number of data points,
diminished by 2, because the 2 parameters mR and aR are estimated from the data
set.
The hit rate for a is calculated with the standard error multiplied with 2.31, the t
factor for E = 0.05. It corresponds closely to the expected 9500.
Rounding to a meaningful number of digits

In the diagrams, the coefficients of the trend line are given with 2 digits. By com-
parison with the standard errors from linest, we see that this assumes an accuracy
that is greater than the uncertainty allows. The equation of the straight trend line is
now, according to the results of linest in G4:H4 of Fig. 10.3 (S) and rounded to the
number of significant digits,
y R = (0.9 ± 0.2)x + 3.4 ∓ 0.9
We have rounded by inspection and written the result down by hand. However,
this can also be achieved with mathematical formulas specified in the spreadsheet
calculation in Exercise 9.3 or with the help of the user-defined function FinRes in
Sects. 9.3.3 and 9.3.4. For more information about linest, see excel help for this
feature.
The Python program in Table 10.1 generates 10 noisy data points (x, yNs ) around
a straight line (cell a) and evaluates them with sm.OLS (cell b). If OLS is called
as = sm.OLS(yNs, x), a trend line yR = mR x is fitted to the data. To get a fit
to yR = aR + mR x, we have to create a new set of independent variables xx =
sm.add_constant(x) . The output in cell c is arranged so that it resembles
the output of linest in excel.
A figure corresponding to Fig. 10.2, as obtained with the program presented in
Table 10.2.
Question
Where are the results in lines 13 and 14 in Table 10.1 needed? 4
In Fig. 10.4a, R2 = 0.74 is reported. How big is R2 adj ?5
4 The coordinates x m and ym of the center of the data cloud are not needed in the presented
program. They are necessary for a figure like Fig. 10.2.
5 With Eq. 10.8 and N = 10 and ddof = 2, we get R2
adj = 0.7075 ≈ 0.71.
Table 10.1 a (top) Generation of noisy data points (x, yNs ) around a straight line; b (bottom left)
Linear regression line through the data points; c (bottom right) Results of the linear regression OLS
3 a,m = 2.0,1.0
4 x=np.zeros(10)
5 dx=1.4
6 Ns=1.4 #Noise level
7 x[0]=1
8 for i in range(1,10):
9 x[i]=x[i-1]+dx*npr.rand() #Uneven spacing of x values
10
11 y=m*x+a
12 yNs=y+Ns*npr.randn(len(x))
13 xm=np.average(x) xx [[ 1.00 1.00]
14 ym=np.average(yNs) [ 1.00 1.18]
15 xx=sm.add_constant(x) . . .
16 model=sm.OLS(yNs,xx) [ 1.00 3.86]
17 results=model.fit() [ 1.00 4.54]]
18 aR=results.params[0] #y=a+mx
19 mR=results.params[1] mR aR
20 aE=results.bse[0] #Errors est 0.98 2.05
21 mE=results.bse[1] mE,aE 0.21 0.57
22 r2=results.rsquared r² 0.74
23 r2_ad=results.rsquared_adj r²_adj 0.70
24 in_a=(aR-aE < a) \ in_m,in_a True True
25 and (a < aR+aE)
26 in_m=(mR-mE < m < mR+mE)
Table 10.2 Python program for plotting Fig. 10.4a, variables specified and results reported in
Table 10.1
1 lblT="y="+str(np.round(mR,2))\
2 +"x+"+str(np.round(aR,2))\
3 +"\n"+"R²="+str(np.round(r2,2))
4
5 FigStd('x',0,6,2,'y',0,8,2)
6 plt.plot(x,yNs,'kx')
7 ylin=x*mR+aR
8 plt.plot(x,ylin,'k-', label = lblT)
9 plt.legend()
10 plt.savefig("PhEx 9-2 trend line",dpi=1200)
Fig. 10.4 Chart with trend

line and formula produced
with the Python program in
Table 10.2
10.3 Fitting a Polynomial Trend Line to Data Points

with Multilinear Regression
We generate points on a parabola and add normally distributed noise to their

y-values. A polynomial of the second degree is fitted to these simulated
data, first, as a trend line within a diagram, and then with the spreadsheet
function linest in excel and with OSL of the statsmodel library of
Python. These functions return, in addition to the coefficients of the poly-
nomial, their standard errors. - Confidence intervals and confidence levels are
interrelated through Student’s t statistics. - Experience teaches us that mea-
surement points must extend beyond the parabola vertex for its coefficients
to be estimated reliably.
10.3.1 Introduction
Noisy data points around a parabola

The formula of a parabola is
y = a + bx + cx 2
y = ax 0 + bx 1 + cx 3 (10.13)
The second variant emphasizes the powers of x.

Figure 10.5 shows 9 data points around a parabola generated with Eq. 10.13 and
blurred with normally-distributed noise to yield values yNs , in a for a noise level
(standard deviation of a normal distribution) of Ns = 1 and in b for Ns = 2.
20 20
y=0.6x²+-3.5x+8 y=0.6x²+-3.5x+8
Ns=1 Ns=2
ym ym
15 15
y = 0.50x2 - 2.56x + 6.79 y = 0.79x2 - 4.55x + 8.05
y R² = 0.95 y R² = 0.95
10 10
5 5
0 0
0 2 4 6 x 8 0 2 4 6 x 8
Fig. 10.5 Nine data points around a parabola (grey lines), together with polynomial trend lines
(thin black lines) of degree 2. a (left) Noise level Ns = 2. b (right) Noise level Ns = 4
Trend parabola
Trend parabolas y R = a R ·x 0 +b R ·x 1 +c R ·x 2 are obtained by multilinear regression,
in which the function is regarded as linearly depending on three variables x 0 , x 1 and
x 2 that formally have to be stacked together in a matrix with three columns, e.g.,
in Python, with xxx = np.array([1, x, x2 ]) . The multilinear regression
functions return the values aR , bR , cR , together with their standard errors aSe , bSe ,
cSe , and the coefficient of determination R2 .
In excel diagrams, polynomial trend lines can be fitted to the data, with their equa-
tions and R2 being reported in its legend, as in Fig. 10.5. Attention: R2 is unadjusted
for a reduced degree of freedom.
Questions
Consider two data sets A and B, with the same number of points around the
same parabola, but with the noise level of B being higher than that of A. What
do you expect for the R2 values for a fit with a parabola to A and B? Which
one is higher?6
What is the degree of freedom for regression analysis of our parabola with
9 data points?7
The coefficient of determination R2 is reported in the figures, together with
the equation of the trend parabolas. Describe how R2 is calculated with the data
points (x, yNs ) and the points (x, yR ) on the trend line! Consider the different
variances, total, residual, explained!8
6 The R2 for a fit to the data set with a lower noise level is expected to be higher, contrary to
Fig. 10.5. Average values of 0.97 for a and 0.90 for b are to be expected.
7 Degree of freedom dof = 9 (number of data points) – 3 (parameters estimated from the data set)
= 6.
Tot = var(yNs ) = (yNs -ym )/dof , residual variance vRes =
8 Consult Eq. 10.6! Total variance v
sum(yNs -yTrend )2 /dof , R2 = (vTot -vRes )/vTot = explained variance/ total variance.
10.3 Fitting a Polynomial Trend Line to Data Points … 433
What are the degrees of freedom to be set for adjusted R2 ?9
Hit rates, with and without the t factor

The hit rates for, e.g., (aR – aE < a < aR + aE ) are around 0.64, as will be seen
later in Fig. 10.7, and thus significantly smaller than suggested by our rule Two
within and one out of . This is because the degree of freedom for the parabola with 9
(points) − 3 (coefficients of the parabola) = 6 is too small for this rule to apply. For
a degree of freedom f = 6, a hit rate of 0.644 (= 1- t.dist.2 t(1; 6) in excel or 1
− (1 – 2 * sct.t.cdf(1, 6) in Python) is actually expected according
to Student’s t-distribution.
Tim I’m confused. What should I state as the result?
Alac Don’t bother. Nobody knows about degrees of freedom and the t-value
anyway.
Mag If you specify the coefficients’ standard error and the number of data
points, an expert can derive statistically relevant statements. You can become such
an expert yourself with our exercises.
The t-value for six degrees of freedom is t = 1.091 obtained with

t.inv.2s(0.318; 6)] (excel) or sct.t.ppf (1–0.318/2,6) (Python) for
an error probability of 0.318 (confidence level 0.682) and t = 2.45 for an error
probability of 0.05 (confidence level 0.95). Tests with an uncertainty given by the
standard error multiplied by the t-values (“t-extended standard error”) indeed give
hit rates that correspond to the confidence levels.
Mag What is the degree of freedom when calculating the variance of yNs or
(yNs – yTrend )?
Alac I would again say 6, but I’m not sure.
Tim To calculate the variance of a set, we have to estimate the mean of this set.
So, I conclude that the degree of freedom is 9 – 1 = 8.

Mag Tim is correct. Calculating vT ot = (y N s − ym )2 /8 and v Res =
(y N s − y R ) /8 yields the same value for R = (vT ot − v Res )/vT ot as that
2 2
reported in the figures or obtained with the multilinear regression functions.
Alac OK, that’s clear.
9 Dof = N – 1 for the total variance, dof = N – 3 for the residual variance, because 3 parameters
are estimated from the data set.
Mag But take care! For the adjusted R2 value, you have to adjust the degrees
of freedom, see Sect. 10.1.4!
Reliable fit of a parabola to the data points

In Fig. 10.6, trend parabolas are drawn through 9 data points. In Fig. 10.6a, the trend
parabola differs more from the original parabola than in Fig. 10.6b, although R2 =
0.99 is the same for both. The reason is that the characteristic feature of a parabola,
the region around an extremum, is less well represented in the data points in a.
Questions
Why do the coefficients of the trend parabola fitted to the data points in
Fig. 10.6a differ so much (≈20%) from the true values (see the legend),
although R2 = 0, 99 is achieved?10
Why does the trend curve in Fig. 10.6b capture the "true"parabola better than
the one in Fig. 10.6a?11
Higher-order polynomials
One can insert polynomials of higher degree as a trend line in diagrams. This is
also possible with linest in the spreadsheet or with OLS in Python. You have
to create column vectors x, x 2 , …, and x n , and enter the whole set as an argument
for known_x’s in linest. For OLS, you have to extend the matrix with x 0 , i.e., a
sequence of ones. This procedure can be generalized to other functions of x.
25 25
y=0.3x²+-3x+8 y=0.4x²+-6x+24
20 yNs 20 yNs
y y = 0.25x2 - 2.36x + 6.09 y y = 0.39x2 - 5.85x + 23.43

15 R² = 0.99 15 R² = 0.99
10 10
5 5
0 0
0 5 10 x 15 0 5 10 x 15
-5 -5
Fig. 10.6 a (left) Original parabola, simulated data points and trend line (second-degree poly-
nomial) for unfavorably located data points. b (right) As in a, but for favorably located data
points
10 The differences in the coefficients are so big because the data points do not capture the minimum
well.
11 In Fig. 10.6b, the characteristic property of the parabola, an extremum, is captured.
A B C D E F G H I J
1 a 8.00 =LINEST(yNs;xx;;1)
2 b -3.50 cR bR aR Estimated
3 c_ 0.60 cSe bSe aSe Stand. Error
4 y=0.6x²+-3.5x+8 0.66 -4.33 10.14 Estimated
5 Ns 1.00 0.04 0.34 0.58 Stand. Error
6 Ns=1 R² 0.98 0.71 #N/A
7 dx 1.00 N NinC NinB NinA t.32 1.09 =T.INV.2T(0.317;6)
8 1000 654 949 956 t.05 2.45 =T.INV.2T(0.05;6)
e) ) )
2*cS *bSe *aSe
();0;1) _<cR+t.3<bR+t.05<aR+t.05
RA N D c _ ;c <b ;b <a ;a
p V( <
Se 05*bSe 05*aSe yNs)
_*x2 RMIN .32*c . . (
9 =A11+ =x^2
dx *x+c y+Ns*NO AND(cR-t AND(bR-t AND(aR-t AVERAGE
=a+b = = = = =
10 x x2p y yNs inC inB inA ym
11 0.00 0.0 8.00 10.06 FALSE FALSE FALSE 7.86
12 1.00 1.0 5.10 6.46
19 8.00 64.0 18.40 18.30
Fig. 10.7 (S) In columns A:D, data points around a parabola y = a + b·x + c· × 2 are created. In
E4:G6, the noisy data yNs = f(x) are fitted with a parabola. The results in E8:G8 are reported by a
rep-log procedure that counts how often inC , inB , inA are true for N repetitions of the experiment.
The formulas for inC , inB , and inA are different! N inC is for t = 1, N inB for t = 1.09, N inC for t =
2.45. Data points and the trend line are shown in Fig. 10.5
a, b, c “true” coefficients of a parabola y = a +b · x + c·x 2

x array of independent variables
x 2p x to the second power
xx [x, x 2p ] for excel
xxx [1, x, x 2p ] for Python
y y(x, x 2p ), function of two variables, values of the parabola for
x and x 2p
Ns noise level, entering as standard deviation in a normal distri-
bution
yNs y blurred with normally distributed noise
a R , bR , c R coefficients of a regression (trend) line through the data points
(x, yNs )
mSe , aSe . cSe standard error of mR , aR , and cR
inA , inB , inC TRUE, if the error range captures the true value
N inA , N inB , N inC number of hits, counting how often inA , inB , inC are true
dof degrees of freedom = number of points – 3 for a parabolic
trend line
t 32 , t 05 t values for confidence levels 0.68 and 0.95.
The argument structure is different for excel and Python:

Excel Linest (yNs;xx;;1)

Python sm.OLS(yNs, xxxT) with xxxT being the inverse of xxx.
Generating the data set

In Fig. 10.7 (S), a spreadsheet layout for fitting a parabola trend line to data points is
presented. The formula for y accesses the two variables x 1 = x and x 2 = x 2 (=x 2p ),
as well as the constant a. The y-values are masked with Gaussian noise with standard
deviation Ns and stored in yNs . Now, our data set is (x, yNs ).
Evaluating the data set

The noisy data set is evaluated in E4:G6 with
linest(yNs;xx;;1).
to obtain the estimated coefficients cR , bR , aR , together with their standard errors cSe ,
bSe , aSe . Note the argument for the independent variables in the formula reported in
E1! The independent variables x and x 2p have to be put together in one range, here,
in a range A11:B19 named xx.
We calculate the error range of the coefficients for different confidence levels, i.e.,
we have to multiply the standard error with an appropriate t-value, and count, with
a rep-log procedure, how often the error ranges capture the true values. The counts
are stored in variables named N in C, N in B, N in A.
The t-value for a confidence level of 0.95 is obtained with t.inv(0.05; dof ), where
dof is the degree of freedom; see the formula in J8, valid for I8.
Question

Which are the independent variables in linest?12
Where are the t-values calculated, and what is their value for inC, inB, inA?
What is the confidence level of inC , inB , inA ? Are the numbers of true in
N in C, N in B, N in A in agreement with expectation?13
12 The independent variables are x in A11:A19 and x 2 in B11:B19, put together as argument xx =
A11:B19 for linest.
13 In is calculated with t.32, valid for a confidence level of 1–0.317 = 0.683 close to the empirical
C
N in C/N = 0.654; inA with t.05 for a confidence level of 1–0.05 = 0.95 close to N in A/N = 0.956.
How do you use linest to determine the coefficients of a trend line for the
function y = a · x15 + b · log(x2 )?14
Trend line in the diagram

To lay a parabola through the nine data points, design/add chart ele-
ment/trendline/ format trendline/trendline options/ offers, among oth-
ers, polynomial/ order 2. With appropriate clicks, the obtained regression
equation and the coefficient of determination R2 of the fit are reported in the diagram.
Hit rates
We apply a rep-log procedure to the spreadsheet calculation reading the values TRUE
or FALSE in E11:G11 N times and returning the counts into N in C, N in B, N in A
(E8:G8). The hit rates 6422/10000 = 0.642, 6837/10000 = 0.684, 9501/10000 =
0.95 correspond to the confidence levels expected for the different t-values applied
for inC , inB , inA (see Footnote 13).
10.3.4 Python Solution
Generation of noisy data

In Table 10.3, noisy data yNs are generated around a parabola, the coefficients of
which are specified all together in a list abc, not separately as a, b, c. The variables
are specified as three row vectors x 0 , x, and x 2P , with x 0 being a list of ones and
x 2P the squares of x; they are gathered in a matrix xxx that is transposed to xxxT
to become a list of column vectors (see lines 7–11) required as entry into the OLS
function, together with yNs .
Polynomial through the noisy data

In Table 10.4, the data set (xxxT, yNs ) is evaluated with the OLS (ordinary least
squares) function of scipy.stats, returning the results into est = sm.OLS(…)
consisting of
– a list of the estimated parameters (est.params),

– a list of their standard errors (est.bse),
– some characteristics of the fit (e.g., est.rsquared).
The results for a run are shown in the bottom cell of Table 10.4.
14 A two-column range is calculated for x 1 5 and log(x 2 ) and entered as known_x’s in linest.
Table 10.3 Python program, generating the data set (xxxT, yNs ) from (x, y)
3
4 abc=[8.0,-3.5,0.6] #Coeffs. of parabola
5 Ns=1
6 dx=1
8 x2p=x**2
9 x0=np.ones(len(x))
10 xxx=np.array([x0,x,x2p])
11 xxxT=xxx.transpose(1,0)
12
13 y=abc[0]*x0+abc[1]*x\+abc[2]*x2p #True parabola
14 yNs=y+Ns*npr.randn(len(y)) #Blurred with noise
yNs:
[ 7.44 6.15 1.76 5.14 3.42 5.13 8.74 12.32 16.80]
xxxT:
[[ 1.00 0.00 0.00]
[ 1.00 1.00 1.00]
[ 1.00 2.00 4.00]
...
[ 1.00 6.00 36.00]
[ 1.00 7.00 49.00]
[ 1.00 8.00 64.00]]
Table 10.4 Evaluation of the data set (xxxT, yNs ) with OLS (ordinary least squares) of scipy.stats
17
18 est=sm.OLS(yNs,xxxT).fit()
19 t05 = sct.t.ppf(0.975,6) #t value for E = 0.05
20 inABC=np.zeros(3) #For the three coeffs.
22 inABC[i]=(est.params[i]-t05*est.bse[i] <
23 abc[i] < est.params[i]+t05*est.bse[i])
#In error range?
24 par=list(est.params)
25 err=list(est.bse)
26 rSq=est.rsquared
abc 8.00 -3.50 0.60
par 7.38 -3.39 0.60
err 0.87 0.51 0.06
rSq 0.97
t05 2.447
in 1 1 1
Table 10.5 Program lines of Tables 10.3 and 10.4 are assembled into a function returning the
logical list in ABC
1 t05 = sct.t.ppf(0.975,6) #t value for E=0.05
2 t32 = sct.t.ppf(0.8415,6) #t value for E=0.318
3 t=[1.0,t32,t05]
4 def FitToPol():
5 yNs=y+Ns*npr.randn(len(y))
6 est=sm.OLS(yNs,xx).fit() #Linear regression
7 inABC=np.zeros(3)
8 for i in range(3): #Coeffs. in error range?
9 inABC[i]=(est.params[i]-t[i]*est.bse[i]
10 < abc[i] <
11 est.params[i]+t[i]*est.bse[i])
12 return inABC
Questions
How do you calculate Student’s t value for a parabola’s coefficients for a

confidence level of 2/3 and 12 data points?15
How is the regression line in Table 10.4 (coefficients in bottom cell) sensibly
reported with uncertainties in the coefficients?16
Which lines in Tables 10.3 and 10.4 have to be gathered in a function, to be
called in a loop to get hit rates for E = 0.05? Which variable has to be returned
and then summed up in the main program?17
Hit rates
In Table 10.5, generation of yNs and evaluation of the data set are transferred into
a function that returns the logical values inABC . This function is called in the main
program of Table 10.6 in a loop in order to determine the hit rates. We have chosen
three different t values for the three coefficients, 1, t 32 , and t 05 , corresponding to
confidence levels of 0.644, 0.683 (= 1 − (1 – 0.8415) * 2), and 0.95, respectively.
In a loop in Table 10.6, we check whether the error ranges capture the true
values. To do this efficiently, parameters concerning the coefficients a, b, c are put
together in lists of shape comparable to the list in est:
– a, b, c in abc, see Table 10.3,

– the logical values inA , inB , inC in inABC, see Table 10.4,
15 sct.t.ppf (1-1/6, 12-3) = 1.02.

16 yR = 7.4(9) − 3.4(5) x + 0.60(6)·x 2 .
17 Lines 14, 18–23 have to be gathered in a function; in
ABC has to be returned and summed up in
the main program, see Table 10.5.
Table 10.6 Main program for getting hit rates for a, b, c; NinA = NinABC[0], etc.
13 nRep=10000
14 NinABC=np.zeros(3)
15 for n in range(nRep):
16 inABC=FitToPol() t 1.00 1.09 2.45
17 NinABC[0]+=inABC[0] N NinA NinB NinC
18 NinABC[1]+=inABC[1] 10000 6438 6888 9524
19 NinABC[2]+=inABC[2]
and are addressed in Tables 10.4, 10.5 and 10.6 with their indices in these lists,
e.g., NinABC[0]+ = inABC[0].
The t-value for a confidence level of 0.95 is calculated according to
t05 = sct.t.ppf (0.975, 6).
t.ppf returns the value for a one-sided test. As we are checking whether the
true values are larger than the lower error limit and smaller than the upper error
limit, we have to enter an error probability 0.05/2 = 0.025 instead of 0.05. The
above formula returns t 05 = 2.45.
The hit rates in the bottom cell of Table 10.6 are, with 0.64, 0.69, and 0.95,
close to the theoretical confidence levels 0.642, 0.69, and 0.950 (see also Footnote
13).
10.4 Exponential Trend Line
We generate noisy data points (x, yNs ) around an exponential and fit a trend
line to the data, (1) as an exponential trend line in an excel diagram, (2)
with linest, and OLS of the statsmodel library of Python as a
fit of a straight line through the logarithmized yNs data, and (3) with the
spreadsheet matrix function logest with a corrected standard error for the
amplitude. For the exponential function to be clearly distinguishable from a
parabola, the data points must cover a sufficiently large x-range.
10.4.1 Exponential and Logarithm
An exponential curve is defined by
y = g · exp(h · x) (10.14)
10.4 Exponential Trend Line 441
200 6
lnYns
yNs, Ns=0.2
8∙exp(3∙x) Linear (lnYns)
150
4
y = 8.97e2.92x
R² = 0.96
100
2 y = 2.92x + 2.19
50 R² = 0.96
0 0
0 0.2 0.4 0.6 0.8 x 1 0 0.2 0.4 0.6 0.8 x 1
Fig. 10.8 a (left) Data points obtained from an exponential function blurred with noise on the
y values, together with an exponential trend line. b (right) Natural logarithm of the values in a,
together with a linear trend line
with g being the amplitude and h a characteristic growth parameter.

To simulate measured data, we add normally distributed noise to the function
values y. A fit to the noisy data is most easily obtained by adding an exponential
trend line in an excel diagram, with the formula and coefficient of determination
R2 displayed as a legend. An example is shown in Fig. 10.8a.
Linear trend of the logarithmized data

By logarithmizing y, Eq. 10.14 can be converted into a straight-line equation of the
form z = a + m·x:
ln(y) = ln(g) + h · x (10.15)
with z = ln(y), a = ln(g), and m = h.

The logarithmized noisy yNs values of Fig. 10.8a are shown in Fig. 10.8b, together
with a linear trend line. The output in the legend is essentially the same as in a,
considering that a = exp(2.19) = 8.97. This indicates that the exponential trend line
is indeed obtained by a least-squares fit to the logarithmized data.
Question
When fitting an exponential trend line in a diagram, sometimes, the following

error message appears: trendline can not be calculated for negative
values or for zero. When and why is that the case?18
18 The yNs values are logarithmized within the function that computes the exponential trend line.
An error occurs when a y-value is negative or zero what occurs if the noise value is negative and
its absolute value bigger than the signal.
Two aspects of this approach are problematic:
– The noise added to the exponential function is independent of the y-value, so

that the logarithmized data scatter more around the trend line for smaller y-
values, indicating that the noise is no longer evenly distributed along the straight
line.
– The coefficient of determination is the same for both scales. This is problematic,
because R2 may be obtained as (Explained variance)/(Total variance), and the
explained variance should be different for different curves.
Uncertainty in the coefficients

We get the coefficients a and m of the linear trend, together with their standard errors,
as usual, with linest and OLS and have to transform them into the coefficients g
and h of the exponential function with h = m and g = exp(a).
As the coefficient m is identical to h, so is the standard error hSe = mSe . The
standard error gSe , however, must be calculated according to the error-propagation
law. We have g = exp(b) and dg E /db = exp(b) so that
g Se = exp(b R ) · b Se (10.16)
Due to the problems arising from logarithmizing the noisy data and deriving the
uncertainty in the coefficient of the exponential from those of the linear trend of the
logarithmized data, we have to be careful with the confidence level.
Influence of noise on the confidence level

For reasonable estimates of the standard error, the hit rate should be about 0.657
( Two within and one out of ). With a rep-log procedure repeating the random
experiment, “Eleven points around an exponential” 1000 times, we check how often
this is actually the case for our experiment. The result can be seen in Fig. 10.9, below
the label “const. noise”.
For the fitting of an exponential trend to 11 data points, the degree of freedom
is dof = 11 − 2 = 9. This results in a confidence level for the standard error range
of 0.657, calculated as 1-t.vert.2s(1; 9) (excel) or 2 * sct.t.cdf(1,9)-1
const. noise prop. noise

h=3 h=1.7 h=0.5 h=0.1 h=3 h=1.7
Total 1000 1000 1000 1000 1000 1000
Hits h 460 547 651 652 650 651
Hits g 383 463 601 641 643 659
Fig. 10.9 How often are the true values of h and g captured by the error range, for signals with
constant noise level (const. noise) and noise proportional to the y-values of the exponential (prop.
noise)?
(Python). With 1000 repetitions of the random experiment, one, therefore,

expects that, in 657 cases, the error range of the estimated coefficients will capture
the true value. However, in Fig. 10.9, the greater the coefficient h in the exponent,
the less the theoretical value of 657 is reached, independent of the noise level.
Figure 10.8b indicates the reason for this. The deviations from the “true” curve
are unevenly distributed on the log-scale. For large values of x, they are smaller,
so that the trend line is closer to the “true” curve than for small values of x. This
is a consequence of logarithmization. However, the unequal distribution of the
deviations from the “true” values does not correspond to the mathematical model
on which linear regression is based.
If we apply a noise proportional to the y-value, e.g., with
yNs = y + norm.inv(rand();0;ns) * y/4(excel)
or yNs = y + Ns * randn(len(y)) * y/4 (Python),
then the deviations in the logarithmic values are more evenly distributed over
the range of t (see Fig. 10.10b). Now, the deviations in the linear representa-
tion, Fig. 10.10a, are distributed unevenly. The number of hits in Fig. 10.9 (under
“prop. noise”) deviates only slightly (less than 4%) from 657, also independent of
the noise level.
Tim When fitting an exponential trend, probably nothing fits at all; not even the
noise is reliable.
Alac I don’t see it that way. You can always give reasonable values for the
coefficients.
Mag In principle, that’s true. For low noise, the uncertainties and the differences
among the various types of noise are not so big.
z.Rs
200
z.
4
Expon. (z.Rs)
y = 10,88e2,92x ln.z.Rs
100 R² = 0,95 2 ln.z
Linear (ln.z.Rs)
y = 2,92x + 2,39
R² = 0,95
0 0
0 0,2 0,4 0,6 0,8 t 1 0 0,2 0,4 0,6 0,8 t 1
Fig. 10.10 a (left) Exponential curve with noise proportional to the y-value. b (right) Logarith-
mized data of a, together with a linear trend line
Tim What exactly does “not so big” mean?
Alac In a given case, we just figure it out with a simulation based on our method
We know everything and play stupid to get the hit rate.
C-spec error by simulation-based t adaptation

We take the regression curve as the “true” curve, estimate the noise level from the
data as well, and perform a statistical simulation with these parameters, similar to
Sect. 10.7.6.
10.4.2 Exponential or Polynomial?
We fit both an exponential and a polynomial to the same set of data points
generated around an exponential function. Figures 10.11a, b show two examples.
In Fig. 10.11a, both trend lines describe the experimental data equally well, with
coefficients of determination R2 of 0.9942 and 0.9953. When repeating the ran-
dom experiment (new measurement points are randomly generated), the parabola
sometimes gets an even higher R2 value, even though the “true” original curve
is exponential. There are not enough measurement points to clearly identify the
exponential character. In Fig. 10.11b, a larger argument range was chosen. Even
now, the parabolic trend line does not differ significantly from the measurement
points; but it would predict completely wrong y-values for x greater than 35.
12 12
y=11∙exp(-0,1x) y=11∙exp(-0,1x)
10 y.Rs 10 y.Rs
Poly. (y.Rs) Poly. (y.Rs)
y y
Expon. (y.Rs) Expon. (y.Rs)
8 8
y = 0,0291x2 - 1,0219x + 11,062 y = 0,0138x2 - 0,7432x + 10,494
6 R² = 0,9942 6 R² = 0,9894
y = 11,243e-0,104x
4 4 R² = 0,9803
2 y = 11,223e-0,107x 2
R² = 0,9953
0 0
0 10 20 30 x 40 50 60 0 10 20 30 x 40 50 60
Fig. 10.11 a (left) Nine points near an exponential function; exponential and parabolic trend line.
b (right) As in a, but with larger argument range
Question
Why is a parabola not an appropriate choice for the data in Fig. 10.11?19
Alac In any case, the exponential trend line in Fig. 10.11a yields a good value
of the decay constant.
Mag If you are convinced that the underlying physical process is actually
exponential, you may determine the decay constant from Fig. 10.11a.
Tim What does it mean to be “convinced”?
Mag That there is a theory of the physical process that predicts an exponential.
Alac …and then we remember “not disproved but also not proven” …
Tim …but preliminarily accepted. This time, that makes sense to me.
Mag That’s the best one can get out of the experiment. Discuss!
x series of independent variables

y exponential values of x; y = g·exp(h·x)
Ns noise level entering as a standard deviation into a normal distribution
yNs y blurred with normally distributed noise
lnY Ns yNs logarithmized
mR , aR coefficient of a straight trend line through (x, lnY Ns )
mSe aSe standard error of mR , aR
h R , gR coefficients of the exponential trend line (regression line)
hSe , gSe standard error of hR , gR , to be calculated from the results of the linear
trend.
The noisy data points around an exponential are generated in the top cell of Table
10.7 and logarithmized in the middle cell. The logarithmized data are fitted with a
linear trendline, the coefficients thereof being presented in the bottom cell.
In Table 10.8, the coefficients of the linear trend fitted to the logarithmized
19 The characteristic of a parabola, its vertex, is not captured (see Sect. 10.2.2 on parabolas).
Table 10.7 a (top) Generation of noisy data points around an exponential; b (middle) Linear fit
to the logarithmized data with OLS; c (bottom) Results of OLS
1 g,h= 10.0, 3.0 #y = g⋅exp(h⋅x)
2 Ns=4 #Noise level
3 x=np.linspace(0,1,11,endpoint=True)
4 y=g*np.exp(h*x)
5 yNs=y+Ns*np.random.randn(len(x))
7
8 lnYns=np.log(yNs) #Natural logarithm
9 xx=sm.add_constant(x) #y intercept allowed
10 model=sm.OLS(lnYns,xx) #Linear regression
12 mR=results.params[1] #Optimized coeffs of y=a+mx
14 mE=results.bse[1] #Standard error
15 aE=results.bse[0]
16 r2=results.rsquared
m a
est. 2.78, 2.45
error 0.22, 0.13
r²= 0.95
Table 10.8 Transformation of the coefficients of the linear trend into those of an exponential trend
17 hR=mR
18 hE=mE h g
19 gR=np.exp(aR) true 3.00, 10.00
20 gE=aE*gR est. 2.78, 11.61
21 inG=(gR-gE<g<gR+gR) error 0.22, 1.49
22 inH=(hR-hE<h<hR+hE) inG, inH False True
data are transformed into coefficients of the exponential trend. The Boolean vari-
ables inH and inG check whether the estimated standard error ranges of the
estimated coefficients capture the true values of h and g.
To get the hit rates, we have to integrate lines 5, 8–16 and 17–22 into a function
that returns inG and inH and have a loop in the main program running over that
function and counting the Trues in inG and inH .
LINEST with the Logarithmized Data

In Fig. 10.12 (S) , noisy data (x, yNs ) are created, logarihmized, and fitted with a
linear trend line with linest in E3:F5. The coefficients mR and aR of the linear trend
and their standard errors are transformed in H3:I5 into the values of the exponential
trend. H5:I5 contains the usual check as to whether the error ranges capture the true
values. The numerical values of the coefficients hR , gR correspond to those of an
exponential trend line in a figure.
We get the hit rates by applying a rep-log procedure that repeats the statistical
experiment by writing a number into a cell to refresh the spreadsheet calculation and
counts the trues in H5 and I5.
Exponential trend with the spreadsheet function logest

excel provides a spreadsheet function logest that determines the parameters of an
exponential trend and their uncertainties, much like linest does for a linear trend.
The equation of the curve to be fitted is
y = a · o x = a · exp(h)x = a · exp(hx) (10.17)
Comparing with Eq. 10.14, the following equations are valid:
h = ln(o) and g = a.
A B C D E F G H I J
1 g 8 mR aR hR gR
2 h 3 mSe aSe hSe gSe
3 Ns 0.2 Estimated 3.08 2.10 3.08 8.19 =EXP(aR)
4 Standard error 0.20 0.12 0.20 0.96 =aSe*EXP(aR)
5 R² 0.96 0.21 TRUE TRUE
s) hSe)
);0;N )} =hR+
AND( s;x;;1 <h;h<
*x) INV(R s) ST(lnYn R-hSe
XP(h ORM yN ↑{=LIN E ND(h
6 =g*E =y+N =LN( ↑=A
7 x y yNs lnYns
8 0 8.00 7.25 1.98 0.657 =1-T.DIST.2T(1;9)
9 0.1 10.80 12.78 2.55
18 1 160.68 207.88 5.34
Fig. 10.12 (S) The noisy data yNs are created in columns A:D and logarithmized in D. A linear
trend line through (x, yNs ) is calculated with linest (formula in E6). The coefficients of the linear
trend are transformed into those of the exponential trend; formula for H5 in H6
Fig. 10.13 (S) Use of the E F G H I

spreadsheet function logest 21 {=LOGEST(yNs;x;;1)} =LN(oR)
to determine the coefficients
22 oR gR hR gR
of an exponential trend and
23 oSe gSe hSe gSe
their standard errors; the
matrix formula in E21 applies 24 21.86 8.19 3.08 8.19
to E24:F26 25 0.20 0.12 0.20 0.12
26 0.96 0.21
For multiple independent variables, the function to be fitted is extended to
y = a · o1x1 · o2x2 · . . . (10.18)
In Fig. 10.13 (S), logest is applied to the data of Fig. 10.12 (S). The same value
R2 = 0.96 is obtained for the coefficient of determination and the same values for hR
= ln(oR ) = 3.08 and gR = 8.19 (calculated in I24) are returned. However, logest
outputs a wrong standard error for gR . Comparison with E4 in Fig. 10.12 (S) shows
that the error of the ordinate intercept of the linear trend is reported as untransformed.
The same transformation as in I4:J4 of Fig. 10.12, based on Eq. 10.16, should be
done.
10.5 Solving Nonlinear Equations
This exercise introduces the nonlinear optimization algorithms solver of

excel and minimize of the Python library scipy.optimize, deter-
mining, as an example, the intersections of a polynomial with a straight
line. This technique is applied in nonlinear regression.
10.5.1 Intersection of Straight Lines with a Parabola
solver of excel is an analysis tool that varies up to 200 independent variables

(“adjustable parameters”) so that the value of a target variable (must be a scalar)
as a function of these parameters becomes optimum, maximum, minimum, or
close to a given value, depending on the setting. The function minimize of
the scipy.optimize library solves the same tasks. We apply it to determine
the intersections of a parabola with straight lines. For the task in this exercise we
apply the function fsolve, also of the scipy.optimize library, that finds the
roots of an equation. Minimize of this library is applied in Exercise 10.6.
10.5 Solving Nonlinear Equations 449
Fig. 10.14 a (left) The intersections of the straight line with the parabola are to be found. The
currently selected x-values are not yet the solution. b (right) The intersections of ten straight lines
with the parabola were determined with a solver algorithm
In Fig. 10.14a, a parabola and a straight line are shown whose intersections are
to be found. This problem leads to a quadratic equation that can be solved analyt-
ically (pq solutions of the reduced quadratic equation). Therefore, the numerical
method with a solver algorithm is not necessary here, but is very convenient, and
also works for more complicated problems that cannot be solved analytically, e.g.,
for the intersections of a cosine function with a third-degree polynomial.
For the currently selected x-values, x L (left of x = 0) and x R (right of x = 0),
the y-values on the straight line yL S and yR S are different from the corresponding
yP L and yP R on the parabola. The x-values are to be modified so that the y-values
become equal.
The equations for a straight line, and a parabola symmetric to x = 0 are
yS = cS + m S · x (10.19)
yP = bP + aP · x 2 (10.20)
We can manually adjust the x-values of the two points, e.g., with sliders, so
that the y-values on the straight line and the parabola match. The same can be
done with solver by minimizing the quadratic deviations of the two y-values.
Figure 10.14b shows the solutions for 10 straight lines with different ordinate
intercept cS . The two intersections for a particular straight line are found in one
run by minimizing the sum of the two quadratic deviations of the y values. The 10
solutions are obtained in a rep-log procedure for the spreadsheet and in a for-loop
in the main Python program.
a P , bP coefficients of a parabola yP = aP ·x2 +bP

cS , mS coefficients of a straight line yS = cS + mS ·x

x sequence of x-values (here 51 values between—6 and 6)
yP , yS values of the parabola and the straight line for x
xL , xR left and right horizontal positions of the intersections
yL P, yL S y-values at x L , P for parabola, S for straight line
yR p, yR S y-values at x R .
Layout
A spreadsheet layout for the task is presented in Fig. 10.15 (S). The graphs of the
parabola yP (x) and the straight line yS (x) are calculated in D11:F61 for 51 points. The
intersections are calculated in D7:F9, first, by adjusting x L and x R with sliders, and
then with solver. Later, we will use a rep-log procedure, triggered with the button
“Intersections”, to get the intersections for the ten straight lines in Fig. 10.14b.
Activating Solver in Excel

The solver function must be activated with file/ options/ add- ins/ solver add-
in, then we find go and click (see Fig. 10.16a). When we call data/solver, the
table in Fig. 10.16b pops up. The solver function optimizes the target cell’s value
(set objective) by varying the parameter cells’ values (by changing variable
cells). The target cell is linked to parameter cells through a set of formulas. If the
value of a parameter cell is changed, the value of the target cell also changes.
1 aP 1.20 yP=1.2x²+2 mS 0.90 yS=0.9x+6
2 bP 2.00 cS 6.00
3 dx 0.24 Intersections
4 3.9E+02 3.9E+02 =G7+G9
/100 bP S )^2
-500) aP*x^2+ mS*x+c (yLP-yLS
5 =(C9 = = = Sub Intersections -->
6 xL yLP yLS cS xLR yLR
7 409 -3.52 16.84 2.84 2.0E+02 2 -5E-07 2
8 xR yRP yRS 0.75 2.675
9 863 4.27 23.84 9.84 2.0E+02
10 x yP yS 4 -0.9694 3.12758
11 -6.00 45.20 0.60 1.71936 5.54742
61 6.00 45.20 11.40
Fig. 10.15 (S) The parameters for the parabola are in A1:B2, those for the straight line in E1:F2;
arguments and function values in D11:F61; in the range D7:G9, the two intersection points are to
be determined; in columns I to K, the intersection points for ten different straight lines with vari-
ous y-axis intercepts cS are stored. The button “Intersections” triggers the procedure Intersectio in
Fig. 10.20 (P)
Fig. 10.16 a (left) Possible add-ins that can be activated in the excel options. “Dieters Funktio-
nen” are user-defined functions that have been saved as add-ins (see Sect. 4.9.1). b (right) Window
after calling the solver function with data/solver (in the Analysis Group); set objective: the
target value in the target cell is maximized (max), minimized (min), or adjusted to a given value
(value of) by varying the values in the changeable cells (by changing variable cells)
Often, the option make unconstrained variables non-negative is activated as

default. This check-mark must be removed, because the variable x to be optimized
can also be negative for our task.
Solver determines intersections

The intersections are now determined with the solver function. To do this, open the
data tab (see Fig. 1.1 in Sect. 1.7) and click on the solver button at the far right
of the Analysis group, data/ solver. A window opens, as in Fig. 10.16b. First, we
take on only one intersection point and enter G7 (of Fig. 10.15 (S)), containing the
left point’s square deviation of the y values of the parabola and the straight line, as
the target cell in the solver tab, and the x-value in cell D7 as the changeable cell.
After pressing the solve button, the two points on the parabola and the straight-line
slide together.
We could now determine the right intersection in the same way; but we choose a
different solution, with both intersections determined simultaneously, by summing
up the square deviations of the two points in G4. We enter G4 as the target cell
to be minimized, and the x-values in D7 and D9 as variable cells showing up with
their names x L and x R . When we press solve, the two intersections are determined
together in one run.
1 Private Sub Intersect_Click() Private Sub ScrollBar2_Change() 8

2 Call Intersectio Cells(7, 4) = "=(C7-500)/100" 'xL 9
4 11
5 Private Sub ScrollBar3_Change() 12
6 Cells(9, 4) = "=(C9-500)/100" 'xR 13
7 End Sub 14
Fig. 10.17 (P) These procedures are triggered when, in Fig. 10.15 (S), the button “Intersections”
is clicked (…_click) or when one of the sliders is changed (…_change), respectively
Question
F4 of Fig. 10.15 (S) contains a function that calculates the quadratic deviation
of the two intersections without accessing column G. Which function can that
be?20 See Section 5.10 (Mathematical and Trigonometric Functions) for advice!
If the option make unconstrained variables non- negative is activated,

solver cannot find any negative x-values. This option must be unchecked if this
is inadequate for the problem under consideration, as it is in this exercise.
There are two ways to select the changeable cells:
– The solver algorithm varies the values in cells D7 and D9 and overwrites the
original formula. However, throughout this exercise, we want to change these
values with the slider again. To be able to do so, we reinsert the formula
with a macro, that is always triggered when the slider is operated; see sub
scrollbar2_change und sub scrollbar3_change in Fig. 10.17 (P).
– We can have solver vary the values in C7 and C9. Then, the formulas in
D7:D9 remain unchanged.
Mag Let’s choose the first variant!
Alac Why would we start there? The second variant is easier, because no macro
is needed. Are we following the motto: Why be straightforward when we can
complicate things?
Tim Well, the first one teaches us how to link a macro to a control element.
Mag Exactly, sometimes we learn via detours.
Sometimes, solver declares the same x-value as the solution for the two inter-
sections. In such cases, the initial x-values have been unfavorably chosen, e.g., both
20 =sumxmy2(E7:E9; F7:F9) “Sum x minus y squared”

1 Sub Macro1() 1
2 Sheets("calc").Select 2
3 SolverOk SetCell:="$G$4", MaxMinVal:=2, ValueOf:=0, ByChange:="$D$7,$D$9", _ 3
4 Engine:=1, EngineDesc:="GRG Nonlinear" 4
5 SolverSolve 5
6 End Sub 6
Fig. 10.18 (P) Recorded macro when initializing the solver function as in Fig. 10.16b
to be greater than zero. Such fallacies are possible in all optimization programs,
because, primarily, only local optima are found.
Use the sliders to change the changeable cells’ initial values so that an
approximate solution is reached before you start solver!
Questions
Why does the formula in D7 or D9 in Fig. 10.15 (S) need to be re-entered after
solver has run?21
How do you re-enter the formula in D7 or D9?22
According to Fig. 10.20, what are the initial values of x for a y-axis intercept
of 20?23
Calling Solver from a VBA procedure

The solver function can be called from a macro. To get the corresponding com-
mands, we turn developer/record macro on before calling solver. The result
for our example is presented in Fig. 10.18 (P).
Before starting the program, we must activate the reference to the solver function
in the VBA editor with tools/references/solver (see Fig. 10.19).
We insert the commands recorded in Fig. 10.18 (P) into the intended procedure
as in Fig. 10.20 (P).
In a loop in sub Intersectio, the ordinate intercept of the straight line is incre-
mented from 2 to 20 in steps of 2 (line 11), and each time, solversolve calls
the solver function (line 16). The addition userfinish:=true causes the solu-
tion proposed by solver to be accepted immediately. If this addition is missing,
a window pops up after each proposal of the solver function in which the user
can click OK.
21 Because, in the current variant of the optimization process, the formulas in these cells are
overwritten by the solver function.
22 By introducing macros like sub scrollbar2_change in Fig. 10.17 (P).
23 The initial values are x = −3 and +3. They are the same for all y-axis intercepts, because they
are always reset within the loop “cg=” in lines 13 and 14 in Fig. 10.20 (P). A better solution could
be to take the previously optimized x-values as the start, because they are closer to the expected
optimized value.
Fig. 10.19 Activating the solver function in the VBA editor via tools/references
5 Sub Intersectio() SolverSolve Userfinish:=True 16

6 r2 = 7 'Row index Cells(r2, sp2) = cS 17
7 sp2 = 9 'Column index 'ordinate section of the straight line 18
8 Range("I7:K400").Clear Cells(r2, sp2 + 1) = Cells(7, 4) 'xL 19
9 SolverOk SetCell:="$G$4", MaxMinVal:=2, _ Cells(r2, sp2 + 2) = Cells(7, 6) 'yLS 20
10 ValueOf:="0", ByChange:="$D$7;$D$9" r2 = r2 + 1 21
11 For cS = 2 To 20 Step 2 Cells(r2, sp2 + 1) = Cells(9, 4) 'xR 22
12 Cells(2, 6) = cS Cells(r2, sp2 + 2) = Cells(9, 6) 'yRS 23
13 Cells(7, 4) = -3 'left x to start r2 = r2 + 2 24
14 Cells(9, 4) = 3 'right x to start Next cS 25
15 End Sub 26
Fig. 10.20 (P) sub Intersectio, working on Fig. 10.15 (S) specifies the parameters for the solver
function in lines 9 and 10 and selects the initial value cS of the ordinate intercept of the straight
line in the loop (for cS = ), calls the solver function in line 16, and saves the coordinates of the
intersections in the spreadsheet in the range below I6:K6, starting with r 2 = 7 (line 6). The spec-
ifications in lines 9 and 10 have to be defined only once, after which they apply to all subsequent
calls of solver
The initial values are x = −3 and +3. They are the same for all y-axis inter-
cepts, as they are always reset within the loop (for cS=) in Fig. 10.20 (P). If they
were set before the loop, the previously optimized x-values would be the next y-
axis intercept’s initial values. This could be better, because the values are closer
to the x-values expected for the next y-axis intercept.
Question
What effect does the instruction r2 = r2 + 2 in line 24 of Fig. 10.20 (P) have
on range I7:K11 of Fig. 10.15 (S)?24
24 Empty cells are introduced between the coordinates of the points, e.g., J9:K9, so that the data
are displayed in the diagram as separate points.
Lambda functions
In Table 10.9, function tables of the parabola (x, yPx ) and the straight line (x, ySx )
are set up using the lambda functions Prb and Str. A lambda function is a small
function associated with a variable. It may take any number of arguments, but must
only contain a single executable expression. Another example is make 0 in Table
10.10.
Furthermore, in Table 10.9, two x values, x L, and x R , are specified, together with
their y-values yRP and yRS , serving later as initial values for the intersections of the
parabola with the straight line.
Table 10.9 Function tables of the parabola and the straight line, and two points on each of them
1 x=np.linspace(-6,6,51,endpoint = True)
2 aP,bP=1.2, 2.0 #Parabola
3 Prb=lambda x: aP*x**2+bP
4 yPx=Prb(x)
5 mS,cS=0.9, 20.0 #Straight line
6 Str=lambda x: mS*x+cS
7 ySx=Str(x)
8
9 xL=-5 #Initial x left
10 yLP,yLS=Prb(xL),Str(xL)
11 xR=5 #Initial x right
12 yRP,yRS=Prb(xR),Str(xR)
Table 10.10 Plotting initial points and optimized intersections

1 FigStd('x',-6,6,2,'y',0,50,10)
2 plt.plot(x,yPx,'k--',label="parabola")
3 plt.plot(x,ySx,'k-',label="straight line, cS="+str(cS))
4 plt.plot([xL,xL,xR,xR],[yLS,yLP,yRS,yRP],'ks',
5 ms=5, fillstyle='none') #Initial points
6
7 from scipy.optimize import fsolve
8 make_0 = lambda x : Prb(x)-Str(x)
9
10 xL=fsolve(make_0, xL) #Optimized left x
11 yLS=Str(xL)
12 xR=fsolve(make_0, xR) #Optimized right x
13 yRS=Str(xR)
14 plt.plot([xL,xR],[yLS,yRS],'ks',ms=5) #ms, Marker size
15 plt.legend()
Function fsolve
In Table 10.10, first, the curves and the initial points are plotted, after which the
intersections are determined by fsolve and then plotted in the same figure. The
function scipy.optimize.fsolve (func, x0, …). finds the roots of a
function of the (in general non-linear) equations defined by func(x) = 0 given
a starting estimate x 0 . In our case, the lambda function make 0 is the root-defining
function.
10.6 Temperature Dependence of the Saturation

Magnetization of a Ferromagnet
The nonlinear Langevin equation relates the saturation magnetization of a

ferromagnet with the temperature. We solve it with solver, called from a
macro, and with the function minimize of the scipy.optimize library
of Python.
10.6.1 Langevin Function
Langevin equation
The Langevin equation describes the temperature dependence of the saturation
magnetization M of a ferromagnet:

μλM
M = N µ · tanh (10.21)
kB T
with:
μ magnetic moment of a magnetic element, e.g., an electron,

N density of magnetic elements,
λ μλN = molecular field,
kB Boltzmann’s constant,
T absolute temperature.
The saturation magnetization M thus appears on both sides of the equation. This is
physically justified because a magnetic dipole aligns itself in the entire field and also
contributes to the entire field.
10.6 Temperature Dependence of the Saturation Magnetization … 457
Equation 10.21 can be simplified by introducing reduced variables
M kB T
m := μ and t := (10.22)
N N μ2 λ
to
m
m = tanh (10.23)
t
The reduced magnetization m is proportional to the magnetization M. The
reduced temperature t is proportional to the temperature T. The solution m =
m(t) is called the Langevin function.
Graphical and numerical solution

The Langevin equation, Eq. 10.23, cannot be solved analytically. Still, it can be
solved graphically by plotting y = tanh(m/t) for a given value of t as a function
of m and by determining the point of intersection with the straight line y = m (see
Fig. 10.21a). It is solved numerically by keeping t fixed and varying m with a solver
function so that the square deviation (m − tanh(m/t))2 is minimized.
We determine 18 points on the curve m(t) by implementing two peculiarities:
(a) all points are optimized simultaneously, and (b) all points are initially on a
quarter circle.
As to (a): The Langevin equation can be solved for each reduced temperature t
independently of the other temperatures. However, we solve it simultaneously for
all points on the curve by adding up the quadratic deviations for the 18 cases and
minimizing this sum by varying the 18 values of m.
1
1.0
m
0.8
0.8
y
0.6
0.6
m
t=0.5
0.4
t=0.6 0.4
Langevin function
t=0.7
0.2 t=0.8 Circular arc
t=0.9 0.2
t=1
0
0.0
0.0 0.2 0.4 0.6 m 0.8 1.0 t 1.0
0.0 0.2 0.4 0.6 0.8
Fig. 10.21 a (left) Graphical solution of the Langevin equation; the y-axis is valid for m and
tanh(m/t). b (right) Langevin function as a varied circular arc
As to (b): Since we expect a strongly bent curve, steeply going towards zero
at t = 1, we select starting points (before minimizing) on a circular arc (see
Fig. 10.21b).
Physical interpretation
The Langevin function in Fig. 10.21b is, with respect to the circular arc, flatter for
small t and steeper for t approaching 1. The magnetic moments first stabilize each
other in parallel alignment, but above a specific temperature, the order collapses.
φ array of 18 angles of a quarter circle

t sin(φ)
m reduced magnetization to be optimized
mC cos(φ), initial values of m
th tanhyp(m/t), tangens hyperbolicus.
The entries m(0) = 1 and m(1) = 0 are held fixed; th(0) is not defined.
A spreadsheet solution of the Langevin equation is presented in Fig. 10.22 (S). The
initial values mC are on a circular arc and copied (with paste/paste values) into
m (E5:E22) before optimization. The solver algorithm is then used to vary these
m-values so that they equal tanh according to the Langevin equation. If something
goes wrong with the optimization, we can start again by copying mC into m.
A B C D E F G H
1 0.0924 =PI()/17/2 Langevin function
2 Circular arc
)
m/t) (m;th
$A$1
3 =A5+ =SIN( =COS
phi) (phi) HYP( SUMXMY2
=TAN =
4 phi t mC m th devi
5 0.00 0.000 1.000 1.000
6 0.09 0.092 0.996 1.000 1.00 2.2E-08
22 1.57 1.000 0.000 0.000 0.00
Fig. 10.22 (S) Solving the Langevin equation; reduced magnetization: initial positions mC on a
circular arc, m after optimization; m is defined as [E6:E22]
10.6 Temperature Dependence of the Saturation Magnetization … 459
Table 10.11 Optimizing the Langevin function

1 from scipy.optimize import minimize
2
3 phi=np.linspace(0.01,np.pi/2,18)
4 t=np.sin(phi) #Values on quarter arc (not varied)
5 mC=np.cos(phi) #Initial values on quarter arc
6
7 def objective(m):
8 th=np.tanh(m/t)
9 return sum((m-th)**2)
10
11 m=minimize(objective,mC,method='SLSQP')
12
13 FigStd('m',0,1.0,0.2,'y',0,1,0.2)
14 plt.plot(mC,t,'k+-')
15 plt.plot(m.x,t,'kx-') #m.x, optimized values
Question
What are the target (objective) and the adjustable variables for solver in
Fig. 10.22 (S)?25
10.6.4 Python
In Table 10.11, the Langevin equation is solved with the scipy function minimize.
The syntax is
m = minimize (objective,mC, method = ’SLSQP’)
where
m list of variables to be varied; shape determined by mC

mC list of initial values for m
objective name of a function with m as the argument and a scalar as the output
to be minimized
SLSQP Sequential Least-Squares Programming, type of solver to be applied.
The optimization works although m[0] and m[-1] are not excluded from the
optimization because mC = tanh(mC /t) is already fulfilled.
25The target is devi (G6) and the variable cells are m, except its first value (E6:E22), because the
magnetization m must be 1 at t = 0 and be kept fixed during optimization.
10.7 Fitting Gaussians to Spectral Lines with Nonlinear

Regression
We fit a sum of two Gaussians to two overlapping EDX spectral lines, taking
advantage of the additional knowledge that, for physical reasons, both lines
must have the same width. The excel tool is solver; the Python func-
tion is curve_fit from the scipy.stats library. The C-spec errors
of the solver solution are obtained through simulation-based t adaptation
(Student’s t) using the parameters obtained from the fit.
10.7.1 Fitting the Sum of Two Gaussians to Data Points
In this exercise, we use the solver tool of excel and the function curve_fit
from the scipy.stats library to fit functions with several fit parameters to
measurement data, generated artificially in the by now well-proven manner.
Generation of two spectrally overlapping EDX signals

In Fig. 10.23a, a spectrum of an EDX analysis is shown. EDX is the abbreviation
for “energy dispersive X-ray analysis”. It means the energy-resolved analysis of
characteristic X-ray radiation with silicon detectors after excitation of the sample
with an electron beam. The count rate y of a detector is recorded as a function of the
energy x of the photons arriving at the detector.
Here, the mapped spectra are artificially generated, but are very similar to the
actual data for a SrTiO3 coated Si wafer. The generation is done by specifying two
Gaussian bell curves with the parameters maximum value, centre, and standard
deviation as A = 600, x A = 1.75, x Ad = 0.05 and B = 150, x B = 1.85, x Bd = xAd
= 0.05, respectively, adding them together with a constant background noise level
yC .
General bell curve

The normal distribution for medium x m and standard deviation x Sd is defined as

1 1 x − xm 2
N (x, xm , x Sd ) = √ · exp − (10.24)
x Sd 2π 2 x Sd
The pre-factor is chosen so that the integral over the function is 1, as it must be for
a probability distribution. The functions for normal distributions are used to model
the spectra, and later to provide fits to the noisy data. To this end, it is advantageous
that maximum A and width x d be chosen independently of each other. The fit function
10.7 Fitting Gaussians to Spectral Lines with Nonlinear Regression 461
is now f (x, xm , x Sd ) = A N · N (x, xm , x Sd ). Consequently, AN for N(x, x m , x Sd )

becomes
√
A N = A · x Sd · 2π (10.25)
When A and x Sd are independent of each other, the initial parameters can be set
with the eye more intuitively.
Question
What is the area under the curve f (x, xm , x Sd ) = A N · N (x, xm , x Sd )?26
Two different fits to the spectrum

The sum of two bell curves, Ga and Gb , and a constant background yAr are fitted to
a simulated spectrum by means of a solver algorithm.
Two different fits to the same measurement data can be seen in Fig. 10.24. For
Fig. 10.24a, the standard deviations of the two bell curves have both been varied
to obtain an optimum fit. For Fig. 10.24b, the two standard deviations have been
forced to be equal. Both fits appear to be equally good, with a very high R2 =
0.99. Therefore, the fit is not unique.
Due to physical considerations, a fit with equal spectral width is more likely,
because the width of the lines is determined by the resolving power of the detector
and should be approximately constant within the considered energy range. So, we
decide to interpret the experimental spectrum with the fit in Fig. 10.24b.
Caution when ftting with many parameters!

The described method of fitting curves with a set of parameters to measurement data
is useful, but also dangerous. If you have enough fit parameters, you can fit almost
any data series without the parameters having a meaningful physical interpretation.
Gain experience! Our exercises offer an excellent practice field for this, as we invent
“true” data ourselves.
Alac Sure, according to the motto: Ψ We know everything and play stupid.
Tim And thereby learn statistics.
Uncertainty in the coefficients

The function curve_fit of the scipy.optimize library returns, in addition
to the parameters of the regression curve, the covariance matrix of the optimized
coefficients from which the standard errors can be obtained as the square root of the
diagonal elements.
26 The area under this curve is AN .

The solver function of excel returns only the optimized parameters. The standard
errors in the fitted curves’ parameters can also be determined in a mathematically
exact way in spreadsheets; see, e.g., E. Joseph Billo, excel for Chemists, Wiley–
VCH (1977) ISBN 0-471-18,896-4, Chap. 17 or John Wiley (2011) ISBN 978–0470-
38,123-6, Chap. 15.
In practice, a coefficient can be changed individually by hand until there is no
longer a good fit upon visual inspection. The deviation from the optimum value of
the coefficient may then be reported as the uncertainty. An example can be seen in
Fig. 10.23b, where the center of the left bell curve has been changed from 1.75 to
1.755. Evidently, the total curve Gabc no longer fits the data points. So, 0.005 is a
rough estimate of the error in x Ar .
800
Ns=20; NsR=20 800
y yA Ga, DxAr=0.005
y yStop
600 yB
600 Gb
yABC
Gabc
yC
400 400
200 200
0 0
1.5 1.6 1.7 1.8 1.9 2.0 1.5 1.6 1.7 1.8 1.9 2.0
x x
Fig. 10.23 a (left) Simulated EDX spectrum of a Si wafer coated with SrTiO3; G1 and G2 are
the underlying “true” spectral lines. b (right) A fit to noisy data with a sum of two bell curves and
a constant; the left bell curve has been shifted by changing x Ar by Δx Ar = 0.005 to estimate the
error in x Ar
800 800
Gabc, R²=0.99 Gabc, R²=0.99
y Ga y Ga
600 Gb 600 Gb
yStop yStop
400 400
200 200
0 0
1.5 1.6 1.7 1.8 1.9 2.0 1.5 1.6 1.7 1.8 1.9 2.0
x x
Fig. 10.24 Fit to an EDX spectrum with two Gaussian curves. a (left) The two standard deviations
are varied independent of each other. b (right) The two standard deviations are forced to be equal
Tim Is that allowed, and is it accurate enough?
Alac Better a rough estimate of the uncertainties of the coefficients than none at
all. In my experience, many labs engage in that practice. The researchers feel that
a report of the optimum coefficients without uncertainty is, in any case, useless.
Nevertheless, even this is done in many laboratories, as I have also observed.
Mag Alac is right. However, we will do better with a simulation described in the
next section by determining the hit rates for the assumed errors in the coefficients
and obtaining the hit rates to the C-spec errors for a specified confidence level by
simulation-based t adaptation.
10.7.2 C-spec Errors of the Coefficients by a Statistical Simulation
Figure 10.25 (S) lists standard errors of the parameters of the two bell curves fitted
to noisy data, obtained with various methods. The values for Python from row
10 on are obtained as the square root of the diagonal elements of the covariance
matrix that is returned by the function curve_fit of the scipy.optimize
library.
The C-spec errors (“confidence-specified”) for excel are obtained with a
statistical simulation in six steps:
(1) Starting from a fit to the spectrum and taking the resulting parameters of the
optimized bell curves (row 2) as values of a “true” noise-less spectrum, as well
as the residual noise of the fit as the “true” noise level. The residual noise is
A B C D E F G
1 Ar xAr xAsdr Br xBr
2 625 1.752 0.050 151 1.86
3 DyAr DxAr DAsdr DyBr DxBr
4 20 0.0050 0.0020 20 0.002 by eye
5 10 0.0019 0.0013 16 0.008 D from row 4
6 11 0.0019 0.0015 16 0.006
7 10 0.0021 0.0015 16 0.007
8 11 0.0016 0.0014 16 0.007 new D from row 7
9
10 Python
11 10 0.0014 0.0014 13 0.006
12 11 0.0013 0.0016 15 0.007
13 12 0.0015 0.0015 19 0.005
Fig. 10.25 (S) [A1:E2]: Results of a fit of two bell curves to experimental data; standard errors
of the coefficients—obtained in excel by eye (row 4) or by adjusting the values from row 4 with
a t-value (rows 5–8)—in rows 11 to 13 with Python as the square root of the diagonal elements
of the covariance matrix
calculated as the standard deviation of the difference between the noisy data
and the fitted curve.
(2) Assuming initial errors in the coefficients by changing them by hand until the
deviation from the data becomes visible (Figs. 10.24 and 10.25 (S) row 4).
(3) Simulating noisy data with the parameters from (1) and (2) to obtain a fit to
these data with our model “two bell curves plus constant background noise”.
(4) Repeating the statistical experiment (3) 100 times or more and so determining
the hit rates for the initial error ranges from (2).
(5) Estimating the confidence level from the hit rates in (4), and from that, the t
values for the desired confidence level.
(6) Adjusting the C-spec errors by dividing the errors assumed in (2) by the t
values from (5).
Row 4 of Fig. 10.25 contains the initial values of the errors obtained by visual
inspection. The next three rows report the results of statistical simulations with
these errors. After that, the initial values of row 4 were replaced with those of the
fit in row 7 to obtain the values in row 8. They are within the error range obtained
with Python.
So, our method Ψ We know everything and play stupid is also successful for a
task occurring in real laboratory life.
Generation of a noisy spectrum
A, x A ,x Asd amplitude, center, standard deviation

yA bell curve with the above parameters
B, x B , x Bsd , yB second bell curve
yC background signal level
Ns noise level
yABC ns yA + yB + yC +noise, noisy spectrum.
Best fit to noisy spectrum
Ar , x Ar , x Asdr amplitude, center, standard deviation of the fit curve

yAr bell curve with the above parameters
Br, x Br , x Bsdr , yBr bell curve with the above parameters
yCr background signal level
yABC new = yAr + yBr + yCr
NsR noise level estimated as standard deviation of (yABC ns −
yABC new).
Visually estimated standard errors of the fit parameters
DyAr , Dx Ar , DAdr estimated standard errors of Ar , x Ar , x Adr

DyBr , DxBr , DBdr estimated standard errors of Br , x Br , x Bdr .
C-spec errors of the fit parameters obtained from a statistical experiment
pout error rate for the estimated standard errors

t Student’st value calculated from pout
st E adapted standard error calculated from pout and t, to get a confidence level
0.68.
10.7.4 Python
For the Python program in Table 10.12, three user-defined functions are needed:
– a bell curve (def gauss) with parameters amplitude A, center x m , and

standard deviation x Sd ,
– a function (def gauss2) adding two bell curves with independent parame-
ters,
– a function (def gauss3) adding two bell curves with the same standard
deviation.
In Table 10.13, the function gauss is used to generate two bell curves yA (x) and
yB (x) intended to mimic ideal spectral lines. A noisy spectrum is simulated by
adding these two curves and a constant background level yC to get yABC , which is
then blurred with normally-distributed noise to become yABC ns. To avoid negative
values that do not occur in reality, a maximum function is applied in line 13.
Table 10.12 Defining a Gaussian and the sums of two Gaussians

2
3 def gauss(x,A,xm,xd):
4 f=sct.norm(0,1).pdf(0)
5 return A/f*xd*sct.norm(xm,xd).pdf(x)
6
7 def gauss2(x,A,xA,xAd,B,xB,xBd,yC):
8 G2=gauss(x,A,xA,xAd)+gauss(x,B,xB,xBd)+yC
9 return G2
10
11 def gauss3(x,A,xA,xAd,B,xB,yC): #Same width
12 G3=gauss(x,A,xA,xAd)+gauss(x,B,xB,xAd)+yC
13 return G3
Table 10.13 Generation of a noisy EDX signal with two spectral lines
2
3 A,xA,xAd =600,1.75,0.05 #Amplitude, position, width
4 B,xB,xBd=150,1.86,0.05
5 yC=20 #Background level
6 Ns=20 #Noise level
7 dx=0.01
8
9 x=np.arange(1.5,2+dx,step=dx)
10 yA=gauss(x,A,xA,xAd)
11 yB=gauss(x,B,xB,xBd)
12 yABC=yA+yB+yC
13 yABCns=np.maximum(yABC+Ns*npr.randn(len(x)),0)
In Table 10.14, a function of type gauss2 is fitted to the noisy data by a

nonlinear least-squares fit with curve_fit, returning the optimized coefficients
(here, stored in popt ) and the covariance matrix (here, stored in cov). The standard
errors D of the coefficients are obtained as square roots of the diagonal elements of
the covariance matrix. The current fit looks like Fig. 10.24a with two bell curves
with different widths, x Asd = 0.047 and x Bsd = 0.67, as reported in the bottom
cell of Table 10.14.
Table 10.14 Fitting gauss2 (independent widths of the two bell curves) to noisy data; “error”
means standard error
1 from scipy.optimize import curve_fit
2
3 p0=[400,1.7,0.05,200,1.9,0.05,130] #Initial guess
4 popt, cov = curve_fit(gauss2, x, yABCns,p0)
5 D=np.zeros(7)
7 D[i]=np.sqrt(cov[i][i]) #√Trace of covariance matrix
A xA xAd B xB xBd yC
true 600 1.750 0.050 true 150 1.860 0.050 20
estim. 534 1.745 0.047 estim. 180 1.828 0.067 21
error 189 0.006 0.004 error 100 0.062 0.027 6
Table 10.15 Fitting gauss3 (equal widths of the two bell curves) to noisy data; “error” means
standard error
8 coef=[A,xA,xAd,B,xB,yC]
9 inC=np.ndarray(6,dtype=bool)
11 ini=[400,1.7,0.05,200,1.85,200] #Initial guess
12 popt3, cov = curve_fit(gauss3, x, yABCns,p0=ini)
13 D=np.zeros(6)
15 D[i]=np.sqrt(cov[i][i])
16 inC[i]=popt3[i]-D[i]<coef[i]<popt3[i]+D[i]
A xA xAd B xB xBd yC
true 600 1.750 0.050 true 150 1.860 0.050 20
estim. 597 1.752 0.054 estim. 144 1.869 0.054 12
error 10 0.0015 0.0015 error 14 0.007 0.001 5
in True False False in True False False False
Question
How can lines 5-7 in Table 10.14 be formulated with list comprehension in one
line?27
A fit with gauss3, Table 10.15, supposing equal widths of the bell curves, results
in coefficients close to the “true” ones. In order to determine the confidence levels
of the calculated standard errors in the coefficients, we check whether the error
ranges capture the true values. The variables “in are reported in the bottom cell.
To do that efficiently in a loop, the coefficients are put together in a list coef, to
be addressed with the loop index. The list is of the same size as the estimated
coefficients popt3 and their errors D, and the result of the logical check is stored
in a list inC of the same size.
Hit rates
Based on the logical check in Table 10.15, we can estimate the hit rates of the error
ranges (see Table 10.16). The statements for a fit with gauss3 are assembled into a
function hitRates that returns the Boolean array inC , stating, for every coefficient,
whether the true value is captured by the standard error range. A loop over rep
= 1000 repetitions counts the number of Trues in N in CC, with the result being
reported in the bottom cell of Table 10.16. The hit rates are close to 668, so that we
have no reason to doubt that curve_fit returns the standard errors of the estimated
coefficients.
27 D = [np.sqrt(cov[i][i]) for i in range(len(cov))].

Table 10.16 Determining the hit rates for the standard errors obtained in Table 10.15; “function
name” should be “variable namefit with gauss3, according”
17 ini=[400,1.7,0.05,200,1.85,200]
18
19 def hitRates():
20 inC=np.ndarray(6,dtype=bool)
22 popt3, cov = curve_fit(gauss3, x, yABCns,p0=ini)
23 D=np.zeros(6)
25 D[i]=np.sqrt(cov[i][i])
26 inC[i]=popt3[i]-D[i]<coef[i]<popt3[i]+D[i]
27 return inC
28
29 coef=[A,xA,xAd,B,xB,yC]
30 NinCC=np.zeros(6)
31 rep = 1000
32 for r in range(rep):
33 inCC=hitRates()
34 #Without () only another variable name
35 NinCC+=inCC
inC:
[ 664.00 679.00 683.00 669.00 680.00 691.00]
Questions
What is the difference between the statements inCC = hitRates and inCC
= hitRates()?28
What is the degree of freedom for a fit with gauss3, according to
Sect. 10.7.2?29
10.7.5 Spreadsheet
Generation of a noisy spectrum

The ideal spectrum yABC without noise is generated in Fig. 10.26 (S) with the parame-
ters specified in B1:E5. A standard normal distribution is used for the noise, addressed
as norm.dist(x; xA,. xAsd;0) where “0” stands for pdf. Its amplitude is calculated
28 The statement inCC = hitRates() calls the function hitRates(), whereas inCC =
hitRates simply assigns another name to this variable.
29 From lines 11 of Table 10.12 and 9 of Table 10.13: dof = 51 (points) – 6 (parameters) = 45.
1 Amplitude A 600 B 150 varTot 46158
2 Center xA 1.75 xB 1.85 varRes
3 StD xAsd 0.05 xBsd 0.05 R²
4 Offset yC 20
5 Noise level Ns 20 nD 0.399 =NORM.DIST(0;0;1;0)
6 dx 0.01 Ns=20; NsR=22
sd;0) ;xBsd;0) BC)/5
0)
xA;xA xB ;0) ABCns;yA
.D IST(x; .DIST(x; );0 ;Ns)
Noise
ORM d*NORM C ND( B+yC+ XMY2(y
sd*N s y V(RA y M
dx D*xA B/nD*xB yA+yB+ NORM.IN MAX(yA+ SQRT(SU
7 =B9+ =A/n = = = = =
8 x yA yB yABC Noise yABCns NsR
9 1.5 0.00 0.00 20.00 -7 13 22.01
10 1.51 0.01 0.00 20.01 -22 0
59 2 0.00 1.67 21.67 -4 17
Fig. 10.26 (S) Generation of a noisy spectrum yABC ns; step (1) of Sect. 10.7.2
in E5 as nD, guaranteeing that the integral over the function is 1. As we want to han-
dle the maxima A and B of the curves as parameters independent of the widths, the
pre-factors for yA and yB are calculated corresponding to Eq. 10.25. With A and the
width x Asd (or B and Bsd ), we have two parameters at hand that have an immediate
visual meaning and can independently be adjusted by hand for the bell curves to fit
the experimental curves.
The ideal “true” spectrum yABC is blurred with noise to become yABC ns in
column G. The signal, a count rate, is never negative in reality; this is guaranteed
with max(*; 0). The residual noise NsR is calculated as the standard deviation of
(yABC ns - yABC ). Its value is close to the pre-specified “true” noise level Ns.
Question
To generate the noisy spectrum in Fig. 10.26 (S), two functions are used:
norm.dist (reported in C7) and norm.inv (reported in F7). Which roles do
they play?30
Fitting bell curves to the spectrum

In Fig. 10.27 (S), the sum of two bell curves, Ga plus Gb plus an offset yCr , is fitted to
the noisy spectrum created in Fig. 10.26 (S). This is done in a separate spreadsheet
into which the vectors x and yABC ns are copied from Fig. 10.26 (S). The variables for
solver are the 7 coefficients in C1:F4 with index r. They are changed in each iteration
of solver so that yABC ns is calculated anew. To have the “experimental” spectrum
fixed, the contents of yABC ns have been value-copied (copy, pastevalues) into
yStop . The variable yABC ns is no longer needed in the current fit. The target cell for
solver is NsR in G9, the residual noise level for (yStop – Gabc ). The coefficient of
determination r Sq of the fit is calculated in H1:H3 according to Eq. 10.6.
30norm.dist produces a smooth Gaussian curve as a function of energy. norm.(rand();0;ns)

produces the normally distributed noise of the y-values.
A B C D E F G H I J
1 Amplitude Ar 405.20 Br 268.20 varTot 45495 =VAR.S(yABCns)
2 Center xAr 1.75 xBr 1.79 varRes 316 =SUMXMY2(yABCns;Gabc)/50
3 Std xAsdr 0.046 xBsdr 0.080 rSq 0.99 =(varTot-varRes)/varTot
4 Offset yCr 10.14
5 Noise level NsR 16.3 =G9 Gabc, R²=0.99
6
r;0) Bsdr;0)
xAsd x 0)
(x;xAr; ST(x;xBr; bc)/5
.DIS T .DI p;Ga
*NORM r*NORM r MY2(ySto
sdr sd C MX
C ns D*xA Br/nD*xB Ga+Gb+y SQRT(SU
7 =x =yAB 427 =Ar/n = = =
8 x yABCns yStop Ga Gb Gabc NsR
9 1.5 21 19 0.0 0.4 10.5 16.3
59 2 26 0 0.0 8.7 18.8
Fig. 10.27 (S) New spreadsheet calculation: fit of the sum of two bell curves to the noisy data of
Fig. 10.26 (S); yStop is the “frozen” result of the fit procedure, reported in Fig. 10.24a
Questions
What is the adjusted coefficient of determination r Sq Adj for Fig. 10.27 (S)?31
Is the procedure reported in Fig. 10.27 (S) a least-squares fit?32
The results for two different fits with the sum of two bell curves, (a) with
individual widths and (b) with identical widths, are shown in Fig. 10.28 (S). Both
fits yield a very high coefficient of determination, R2 = 0.99, so there is no good
reason to prefer one particular fit. However, we know that, for physical reasons,
the width of the spectral lines is determined by the resolution of the detector, and
therefore should be the same for both lines; thus, (b) is preferred.
10.7.6 C-spec Error of the Optimized Coefficients

by Simulation-Based t Adaptation
Fig. 10.29 (S), based on Fig. 10.28 (S), presents a spreadsheet calculation for
obtaining the errors of the optimized coefficients.
Row 4 lists the optimized coefficients of a fit of the sum of two bell curves
to noisy experimental data. They are now regarded as the true values for a new
statistical experiment and used as parameters to generate a spectrum in the calcu-
lation model of Fig. 10.26 (S) that is then entered into Fig. 10.27 (S). The noise
level NsR , calculated as the standard deviation of the difference between the noisy
spectrum and the fitted curve in H9 of Fig. 10.26 (S), is value-copied into G9 of
Fig. 10.27 (S).

2 · N −1 (Eq. 10.8), here, N = 51 (Fig. 10.26 (S)), ddof = 7 and R2 =
j = 1− 1− R
31 2
Rad N −ddo f
2 = 0.989. In this example, the difference between R2 and R2
0.99 (Fig. 10.27 (S)), so that Rad j adj
is unimportant.
32 Strictly speaking, not because the target is the square root of the squares of the deviations. How-
ever, sqrt is a strictly monotonously increasing function so that the minimum in sqrt is also the
minimum in the squares.
A B C D E F G H I J
4 Offset yCr 10.14
A B C D E F G H I J
4 Offset yCr 15.51
Fig. 10.28 (S) Copy of instances from Fig. 10.27 (S); results of fitting with bell curves with a
(top) individual widths, b (bottom) identical widths; F3 = [=xAsdr]
K L M N O P Q R S
3 Ar xAr xAsdr Br xBr
4 405 1.75 0.046 268 1.79
5 DyAr DxAr DAsdr DyBr DxBr
6 10 0.0021 0.0015 16 0.0071
dr)
r+DA
Ar) +DxAr) sd<xAsd yBr) DxBr
)
D y xAr x A D xBr+
A;A<Ar+ <xA;xA< dr<xAsd; B;B<Br+ <xB;xB<
yAr< DxAr r-DA yBr< DxBr
(Ar-D AND(xAr- AND(xAsd AND(Br-D AND(xBr-
7 =AND = = = =
8 inA inxA inxAd inB inxB
9 FALSE FALSE FALSE FALSE FALSE
10 100 67 79 71 66 67
11 0.67 0.79 0.71 0.66 0.67 =P10/$K$10
12 pOut 0.33 0.21 0.29 0.34 0.33 =1-P11
13 t 0.99 1.27 1.07 0.96 0.99 =T.INV.2T(pOut;44)
14 stE 11 0.0016 0.0014 16 0.007 =DxBr/t
Fig. 10.29 (S) Continuation of Fig. 10.28 (S); spreadsheet calculation for determining the stan-
dard errors of the optimized coefficients; L4:P4 copied from D1:F3
The deviation values reported in row 6 are obtained by changing the optimized
parameters until the trend curve differs visibly from the noisy spectrum to be
pre-supposed as errors for the coefficients obtained in new statistical experiments
with spreadsheet calculations as in Fig. 10.26 (s) and Fig. 10.27 (S) in order to
determine whether the supposed confidence intervals capture the true values.
With a rep-log procedure, the new statistical experiment is repeated 100 times
to obtain the hit rates reported in row 10 of Fig. 10.29 (S). From that, we get the
confidence levels in row 11 and the error probability pOut into row 12; therefrom,
the corresponding t-value in row 13; and finally, the standard errors stE for a
confidence level of 0.68 estimated as initial errors (row 6) divided by t.
This simulation with 100 trials is repeated with the stE copied manually as
standard errors in row 6 in the following run. The results of this iterative procedure
are reported in Fig. 10.25 (S).
1. How do you obtain a linear mapping of the function

x 2
y = A · exp
x
and what meaning do the y-axis intercept and slope at x = 0 have?
2. In Fig. 10.30a (S), a straight line is adapted to measurement points with linest.
What is the meaning of the numbers in D1:E2 and D3?
3. A parabola is fitted to 7 data points with linest. How many degrees of freedom
does the fitted parabola have? Does our rule: Two within and one out of ? apply
to the coefficients of the parabola?
4. In Fig. 10.30b, a polynomial yC (x) is fitted to data points (x, yS ) with
solver. Which is the target cell, and which are the adjustable cells? Which
mathematical formula is behind the spreadsheet formula in S5?
5. In Fig. 10.31 (S), a parabola is fitted with linest to 7 data points (x, y). A
rep-log procedure counts 10,000 times how often the true values of a, b, and c
are outside of the error range. Which data are in B7:C13? What is the meaning
D E F G N O P Q R S T U
1 2.97 1.35 =LINEST(y;x;;1)
+cS yS)
2 0.19 1.20 bS*x 2(yC;
5 aS*x^2+SUMXMY
3 0.97 1.76 = =
6 x yC yS
7 0 aS 1.11 1.74 2.49 4.06
8 1 bS 1.66 6.34 5.25
9 2 cS 2.49 10.84 10.24
13 6 52.38 52.41
Fig. 10.30 a (left, S) Output of linest. b (right, S) A polynomial yS is fitted to the data points
(x, yC ) with solver
1 a 1 1.11 1.66 2.49
2 b 2 0.11 0.69 0.88 {=LINEST(A7:A13;B7:C13;;1)}
3 c 3 1.00 1.01 #N/A
4 Ns 1 10000
)
2) 2*F2
E1+E <F1+
a;a< T(1;4
) 2<b;b T(2;4
)
(E1-E2< IST.2 (F1-2
*F
IST.2
5 =AND =T.D =AND =T.D
6 y x x² a.1 b.1 c.1 a.2 b.2 c.2
7 1.74 0 0 FALSE TRUE TRUE TRUE TRUE TRUE
8 6.34 1 1 0.378 0.379 0.383 0.374 0.119 0.120 0.119 0.116
9 10.84 2 4
13 52.38 6 36
Fig. 10.31 (S) A parabola is fitted to the 7 data points (x, y) with linest. The values in E8:G8
and I8:K8 are the result of a rep-log procedure
Table 10.17 Python code snippet for performing a linear least-square fit to data points (x, yNs ) in
arrays x and yNs
1 #Data points (x,y) in lists x and yNs
2 model=sm.OLS(yNs,xx)
5 aE=results.bse[0]
6 r2=results.rsquared
7 r2_ad=results.rsquared_adj
Fig. 10.32 (S) For A B C D E

determining the intersection 1 a.1 1 a.2 3
of two straight lines 2 m.1 2 m.2 -2
3
4 x y.1 y.2
5 1 3 1 4
of the values in E2:G2 and the arguments of the formula for t.dist.2t in H8
and L8? How have the values in E8:G8 and in I8:K8 been obtained?
6. Table 10.17 shows a Python code snippet for performing a linear least-square
fit to data points (x, yNs ) in arrays x and yNs . A linear trend line is fitted to
the data points with the OLS function of the statsmodel.api library. What is the
array xx in the argument of OLS? How do you get the coefficients of the linear
trend line and their standard errors? What is the difference between the two
quantities queried in lines 6 and 7?
7. Figure 10.32 (S) shows the spreadsheet layout for determining the intersection
of two straight lines with solver. What are the formulas in B5, C5, and the
target cell D5? What are the objective (target) cell and the variable cells in
solver?
8. Figure 10.33a shows a spreadsheet layout for calculating the intersections of
three straight lines displayed in b. What are the formulas for y1 , y2 , and y3 ?
The target cell to be minimized is E4. What are the formulas in E6, E7, and
E8?
9. Table 10.18 shows a Python program for solving the problem of Fig. 10.33,
namely finding the intersections of three straight lines with a solver algorithm.
It is incomplete. Introduce the 6 missing statements!
5
A B C D E F G y.1
y.2
1 a.1 2 a.2 -1 a.3 8 3 y.3
2 m.1 1.5 m.2 -1 m.3 -4 y1y2
y1y3
3 1
y2y3
4 8.9E-13 =SUM(E6:E8)
-5 -3 -1-1 1 3 5
5 x y.1 y.2 y.3
6 -1.20 0.20 0.20 12.80 3.3E-13 =(y.1-y.2)^2
-3
7 1.09 3.64 -2.09 3.64 5.7E-13 =(y.1-y.3)^2
8 3.00 6.50 -4.00 -4.00 0.0E+00 =(y.2-y.2)^2
-5
Fig. 10.33 a (S, left) spreadsheet layout for calculating the intersections of y1 , y2 , and y3 using
solver, the column vector x contains the optimized x-values of the intersections. b (right) Straight
lines of a, together with their intersections
Table 10.18 Python program for solving the problem in Fig. 10.33; 6 lines are omitted
8 a1, m1 = 2, 1.5
9 y1 = lambda x: a1+m1*x
10 def target(x):
11 y12=(y1(x[0])-y2(x[0]))**2
12 return y12+y13+y23
13 xI=[-1,1,2.5]
14 x=minimize(target,xI,method='SLSQP')
15 print(x.x)
Index
Symbols A
structure, 9, 69, 261 Ab-initio constructors, 12, 37
Always one more! Yes, but of what and Absolute cell addressing, 16, 19
than what?, 14 Absolute temperature, 103, 420, 456
Approximated average value instead of ACTIVESHEET.SHAPES.ADDSHAPE, 126
exact integral, 13 ActiveX control, 24
ACTIVEX CONTROLS, 24, 25, 131
Calculate with variances, report the C-spec
Add-In, 168, 170, 450, 451
error, 14, 368, 375, 391, 398
Addition of cosines, 57, 65, 68
Chance is blind and checkered, 14, 318, 361 ADDSHAPE, 128, 134
Come to a decision! You may be wrong, 14 Adjusted R 22 , 423, 433, 434
Cos plus Cos equals mean value times half AND, 322
the difference, 13 Angle of inclination, degrees, 209
Ctrl + Shift + Enter, 12, 30 Angular frequency, 60, 242, 247
Doppler effect with plus and minus, 13, 92, Animation, 229, 234–239, 246, 251, 254, 257,
110 272, 350, 351
Empty lines separate curves, 12, 174 Animation object, 237
Antiderivative, 347, 352
First, the tangent at x = 0, 13
Append, 34, 176, 178
From variance to confidence with Student’s Application as a matrix formula, 52
t value, 14 Arcus tangens, 39, 42, 43, 199, 202, 223
half, half, whole, the halves count twice, 7, Area under a curve, 192, 195
13, 281, 305 argmax, 56
If in doubt, count!, 7, 14, 307, 365 argmin, 56
Imaging equation for lenses with plus and ARRAY in VBA, 157
minus!, 13 Arrays in Numpy, 15, 16, 31
Mostly, not always, 6, 14, 307, 417 Arrays in Python, 70, 325
Plus 1 becomes times e, 13 Arrhenius plot, 420
Arrow, 9, 25, 35, 88–90, 126, 132, 148, 185,
The dollar makes it absolute, 7, 11, 69
186, 198–203, 205, 209–215, 217,
Twice as good with four times the effort, 14 229–234, 236, 237, 249–251, 266,
Two within, one out of, 14, 367, 369, 376, 335
377, 380–382, 391, 417, 421, 424, Arrow heads, 185, 212, 231, 237
426, 433, 442, 472 as npr, 35, 37, 309, 324
We know everything and play stupid, 14, ATAN2, 39, 43, 199, 223
365, 385, 392, 421, 423, 444, 461, AVERAGE, 320, 387
464 Axis, 9, 10, 18, 24, 39, 40, 54–57, 65–70,
Worse makes good even better, 14, 390, 417 79–81, 83, 86, 88, 89, 93, 95–97,

https://doi.org/10.1007/978-3-030-82325-2
476 Index
99, 102–109, 112, 113, 137, 142, Close-packed plane, 143–145, 149, 150
175–178, 181, 183, 184, 188, 189, Coefficient of determination, 422, 426, 428,
192, 194–196, 198, 200–203, 206, 432, 437, 441, 442, 448, 469, 470
224, 225, 228, 237–249, 254, Color, 55, 90, 125, 126, 128, 129, 131–135,
258–261, 266, 297, 298, 335, 345, 137, 150
406, 419, 426, 450, 449, 450, 453, Column vector, 12, 16, 27, 41–44, 46, 47,
454, 457, 459, 472 51–53, 56, 63, 65, 66, 106,
Axis=1, 54, 67 171–174, 184, 211, 214, 219, 221,
ax1.twinx, 108 263, 264, 303, 322, 355, 434, 437,
474
COMMANDBUTTON, 128, 131, 132
B Command button, 13, 128, 131, 132
Barometric formula, 283–285 Commutative, 46
Bayes’ rule, 329, 331, 332 Composite function, 188, 189, 191
Beats, 13, 17, 57, 59, 60, 63, 65, 68 Concatenation operator, 24, 53, 69
Bell curve, 60, 334, 335, 343, 345, 346, Confidence interval, 309, 365–367, 370, 376,
460–467, 469–471 378, 380, 382, 388, 391, 397, 399,
Bernoulli, 227, 228, 241 407, 408, 410, 411, 414–416, 419,
Boltzmann, 103, 286, 420, 456 427, 431
Brachistochrone problem, 241 Confidence level, 7, 309, 365–368, 370, 378,
Broadcasting, 2, 12, 16, 17, 41, 42, 44–47, 66 382, 383, 385, 388, 391, 399, 408,
409, 411, 414, 415, 421, 423, 424,
431, 433, 436, 437, 439, 440, 442,
C 463–465, 467, 471
Cartesian coordinates, 13, 71, 81, 137, Confidence-specified error, 368, 407, 409
198–202, 224, 227, 243, 252, 258, Coordinate system, 76, 81, 83, 136, 142, 179,
264–266 198, 199, 242, 249, 251, 254, 263
Case-insensitive, 139, 148 Cosine, 9, 13, 15, 17, 57, 59–70, 73, 78, 121,
Case-sensitive, 139, 331 168, 187, 188, 193, 195, 223, 347,
Cdf, 309, 310, 333–335, 338–342, 345, 352, 449
360, 384 COUNT(, 323, 325
Cell addressing, 11, 16, 19, 22 COUNTIF, 322, 323
Cell references, 21, 26–28, 69, 122, 232 Covariance matrix, 461, 463, 466
Cell references, relative, absolute, indirect, 26 Create a name, 77
Center of gravity, 136, 212–214, 228, Creation of diagrams, 16
251–259, 267 C-spec error, 7, 14, 367, 368, 375, 383,
Central ray, 85–87 385–387, 389–392, 394, 395, 397,
Centrifugal, 112, 207–209, 211 398, 407–412, 414–417, 422–424,
Chain rule, 189 444, 460, 463–465, 470
CHART STYLES, 23 Cumulative density function, 309, 310, 352
Checkerboard, 99, 132, 133, 135, 136, 139, Curie–Weiss law, 419
141, 142 Curly brackets, 30, 44, 184, 200
Chemical react, 420 curve_fit, 14, 35, 419, 421, 424, 460, 461,
Chi2 , 337 463, 466, 467
Chi2 distribution, 358 Cycloid, 227, 228, 238, 241, 243, 246
Chi2 test, 14, 307–314, 316, 317, 333,
337–340, 343, 345, 347, 348,
358–361 D
CHISQ.DIST.RT, 310, 312 Damping, 270, 273
CHISQ.TEST, 307, 312, 313, 317, 338 Data frame, 117, 165, 167, 169
Chisquare, 35, 308 DataFrame, 117, 165, 169
Circle, 71, 72, 86, 123, 125, 128–133, 139, Data series into a chart, 22
140, 146, 177, 178, 194–197, 246, Data types, 34, 36, 134
252, 260–266, 457, 458 Ddof, 310, 315–317, 321, 380, 412, 423
Index 477
Debug, 120, 127, 155 F

DEBUG/ STEP INTO, 120, 127 fcc, 143–145, 147, 148, 150, 151
Decay constant, 445 Fill handle, 21
def ArrowP, 90, 185, 234 fillstyle, 108, 214, 258, 259, 326, 331, 455
Definite integral, 106, 192 FinRes, 373, 374, 415, 429
def StdFig, 108 First derivative, 186, 187, 190, 224
Degrees of freedom, 310, 311, 313, 315, 316, Focal ray, 86
358, 359, 361, 380, 383, 391, 412, Focus, 17, 18, 31, 80, 81, 83, 85
421, 423, 426, 429, 433–435, 472 For-loop, 91, 118, 120, 122, 123, 130, 131,
Delete, 34, 120, 126, 314 139, 143, 146, 163, 164, 167, 206,
Delta degrees of freedom, 310, 315, 380, 423 233, 286, 315–317, 360, 389, 396,
DESIGN MODE, 25, 131 404, 415, 449
Determinant, 42, 45, 47, 48, 216, 223 FORMAT/ CURRENT SELECTION, 23
DEVELOPER, 24, 25, 118, 124, 125, 127, FORMAT DATA SERIES, 212, 231
131, 453 FORMAT SELECTION, 23, 107
Dictionaries { }, 34 Formatted output with %, 41
Difference equations, 270 Formatting the chart, 23
Diffraction pattern, 350, 351, 356 Formula network, 76, 77, 244, 263, 372
DIM, 148, 153, 157, 171, 172 Foucault’s pendulum, 228
direct cell addressing, 16 Frames, 136, 142, 167, 234, 236–238, 251,
257
Directrix, 17–19, 31
Frequencies of occurrence, 308, 310, 311,
Distance to track, 99
315–317, 336–339, 341, 345, 350,
Distribution function, 308–310, 312, 313,
358–360, 362, 384
333–335, 338–343, 345, 346, 348,
FREQUENCY, 14, 57–64, 66, 67, 73, 92–99,
350, 352, 355, 362, 363
307, 308, 310–317, 323, 324, 326,
DO … LOOP UNTIL, 165
333, 336–339, 341, 343–345, 351,
Dollar sign, 106
353, 356–358, 360–362, 384
Dot product, 224, 253, 257
Friction, 112, 227, 269–271, 273, 274,
278–280, 282–285, 287–289, 291,
295, 298–303, 305
E Frictional force, 208, 280, 287, 299, 306
EDX, 460, 462, 466 Friction force of the rope, 299, 300, 302
Electrical network, 182, 215, 217, 219, 221 fsolve, 35, 448, 456
Ellipse, 124, 127–129, 132, 134, 204 Fundamental frequency, 57–59, 63
EMPTY, 164, 165 Fundamental rule of statistical reasoning, 14,
Equi-distribution, 310, 313–316, 324 307, 369, 392
Equilibrium of torques, 213 Fundamental theorem of calculus, 194, 197
Error propagation in powers, 400, 403
Error propagation in products, 399, 403
Error propagation in sums, 368, 398, 402 G
Error range, 7, 14, 365–370, 374–380, 382, Global parameters, 98, 139, 148–150, 276,
384–388, 393, 395, 411, 421, 292, 297, 404
424–428, 435, 436, 439, 442, 443, Global variables, 132–135, 244
446, 447, 464, 467, 472 GOAL SEEK, 421
Euler, 13, 101, 269, 271, 272, 274, 276–278, Gosset, 367
304 Gravitational force, 112, 207
Explained variance, 442 Gravity, 227, 229, 230, 253–255, 257,
Exponential, 13, 75, 78, 99, 101–107, 113, 278–280, 298, 303, 382, 390
308, 347, 372, 420, 421, 423, Grid scale in the spreadsheet, 129
440–448
Exponential distribution, 347
Exponential growth, 75, 100 H
Externally consistent, 390, 393, 395, 396 Half-step procedure, 271, 272, 305
478 Index
Harmonics, 57–59, 62, 65, 67 Linear algebra, 17, 35, 41–43, 45, 47, 49, 181,
Height-dependent friction, 284 223
Hit rate, 7, 368, 374, 377–380, 383, 385, Linear regression, 8, 14, 419, 420, 425, 428,
391–394, 396, 408, 409, 411, 414, 430, 443
415, 421, 424, 425, 428, 429, 433, Linear trend line, 405, 421–427, 441, 443,
437, 439, 440, 442, 444, 447, 463, 447, 473
464, 467, 468, 471 Line breaks, 250
LINEST, 419, 421, 424, 426–429, 431, 434,
436, 437, 440, 442, 447, 472
I Line vector, 181–184, 209, 211
Identifier, 21, 30, 33, 34, 37, 38, 46, 55, 64, Linked cell, 25, 26, 71, 79, 110, 184, 232, 255
76, 162, 170, 184 List comprehension, 12, 16, 109, 224, 324,
If, 10, 13, 14, 115, 117, 123, 167, 224, 303 388, 467
If … then … else, 349 List processing, 15, 16, 181
IF(, 10, 294, 303, 322, 323 Logarithm, 13, 105, 223, 372, 440, 441
Imaging equation for lenses, 13, 82, 83 Logarithmic scale, 105, 107
import numpy as np, 35, 36 Logarithmized data, 405, 406, 441–443,
Indentation, 139, 167, 283 445–447
Independent variable, 9, 10, 20, 27, 53, 63, 76, LOGEST, 440, 447, 448
77, 97, 102, 106, 117, 188, 196, Logical branches, 117, 125
261, 262, 285, 291, 349, 352, 399, Logical IF queries, 133
400, 420, 424, 426, 429, 435, 436, Longitudinal wave, 260
445, 448 LookAhead, 273–279, 285, 304
INDIRECT, 11, 16, 26, 27, 70, 249, 386, 387 Loop2i, 13, 118, 121, 123, 163, 176, 207
Indirect cell addressing, 11, 16
Integer, 25, 26, 34, 101, 131, 132, 139, 148,
157, 176, 222, 223, 236, 309, 339, M
372 Mach cone, 258, 259, 264, 265
Integral function, 192–195 Macro recorder, 115, 117, 124, 125, 174–176
Integration along a line, 182 Magic chord, 30, 44, 52, 172, 184, 311, 312,
Internally consistent, 391, 393, 395, 396 428
Interval search with five queries, 353 Mass-spring system, 187, 270, 272, 275, 278,
Inverse matrix, 45, 182, 219, 222 298, 306
Mathematical functions, 8, 36, 78, 93, 182,
222–224, 346
J Mathematical pendulum, 306
Join, 151–153, 155, 158, 176 Matlab, 35
Jupyter, 3, 11, 16, 31, 32, 64 Matplotlib, 35, 39, 55, 65, 90, 175, 176, 185,
229, 234, 235
matplotlib.pyplot, 175
K Matrices, 12, 13, 16, 27, 38, 39, 41–49, 57,
Karl Popper, 339 184, 219, 223, 225, 261, 296
Keyword argument, 55, 90, 94, 98, 138, 139, Matrix equation, 217–220
166, 167, 185, 233 Matrix formula, 2, 17, 29, 30, 50, 52, 184,
Kirchhoff’s rules, 182, 215, 217, 219, 220 200, 201, 263, 448
Matrix function, 12, 30, 172, 219, 222, 307,
309, 311, 312, 316, 428, 440
L Matrix of powers, 216, 218, 221
Laboratory system, 228, 238, 243, 246, 247, Matrix operations, 2, 15, 17, 41, 45, 182, 225
249–251, 266 MDETERM, 45, 223
Langevin, 456–459 Mean, 13, 14, 27, 38, 57, 61, 62, 73, 79, 85,
Length l of a vector, 43 100, 101, 104, 119, 131, 137, 174,
Length of a curve, 181, 194 239, 240, 243, 260, 269–271, 276,
linalg, 47 281, 293, 303, 305, 307, 310, 312,
Index 479
313, 316, 318, 320–322, 324, 333, NORM.INV, 293, 308, 309, 333, 336, 337,
334, 339, 340, 342, 343, 356, 343, 345, 363, 381, 408, 412, 413,
358–360, 362, 366–369, 371, 372, 425, 427, 443, 469
374–378, 380–382, 384–387, np.append, 109, 176, 196, 225, 388, 432
390–395, 397–402, 404, 407, 408, np.arange, 12, 16, 37, 70, 107, 108, 295
410–413, 417, 422, 423, 426, 433, np.arctan2, 39, 137
444, 445, 460, 461 np.array, 12, 36, 47, 53, 65, 70
Microphone signal, 57–59 np.dot, 39, 46, 224, 253
Mid-perpendicular, 181–183, 349 np.exp, 101, 335
Midpoint, 137, 182 np.extract, 325, 332
Minimize, 35, 421, 425, 448, 456, 459 np.flipud, 90
MINVERSE, 45, 219, 223, 225 np.histogram, 315–317, 341, 362
MMULT, 29, 45, 219, 223, 225, 254 np.hstack, 38, 90
Mobile, 182, 212–214 np.inf, 317
np.linspace, 12, 16, 37, 53, 65, 70, 108, 109,
MOD, 222, 292, 293
224, 225
Module in VBA, 164
np.linalg.det, 43, 47, 49
Modulo, 222, 293
np.log, 37, 308
Moment of inertia, 227, 228, 251, 253–256, np.logical_and, 324, 325, 389
258 np.logical_not, 325
Motion on a straight line, 93 np.logical_or, 325
Multilinear regression, 420, 431–433 np.matmul, 219, 223
Multi-line strings, 158 np.mean, 320
Multiple statements on one line in VBA, 122 npr.choice, 307–309
Multiple tests for error probability, 309 npr.rand, 35, 37, 141, 308, 315, 317, 343
Multiple tests for hit rates, 366, 368, 387 npr.randn, 35, 37, 336, 341, 412
npr.random, 324, 332, 336, 341, 345
np.shape, 38, 47
N np.std, 321, 380, 412
N/A, 322, 323 np.tan, 39, 308
NameError, 32 np.transpose, 46, 47, 49
Name Manager of EXCEL, 27 npl.linalg.inv, 43
Naming cell ranges, 11, 51 numpy.linalg, 43, 182, 221, 222, 225
Naming cells, 21 Numpy.random, 35, 37, 135, 307–309, 324,
Ndarrays, 36 336, 363
Nested loop, 122, 132, 141, 176, 178, 282, 344
Network of formulas, 75
O
Newton’s equation of motion, 7, 13, 269, 270,
Object-side, 81, 86
272, 275, 305
ON ERROR GOTO, 155
Newton’s law, 188, 269 openpyxl, 166
Newtonian friction, 271, 279 Ordinary Least Squares (OLS), 421, 424, 437,
Nodes, 58, 60 438
Noise, 307, 344, 365, 366, 368–370, 384, 391, Orthogonal vector, 209
398, 400–403, 407, 410–414, 419, Oscillation of a mass-spring-system, 187
421, 423–427, 431, 432, 435, 436, out.to_excel, 169
441–445, 460, 463–465, 468–470 OverflowError, 101
Nonlinear equations, 448 Overtones, 13, 17, 57–59, 62, 73
Nonlinear regression, 419, 421, 422, 460
Normal distribution, 7, 308, 309, 333–336,
340–343, 345, 346, 359, 360, 362, P
381, 425–427, 431, 435, 445, 460, Pandas, 35, 117, 160, 162, 165, 167
468 Parabola, 15, 17–20, 22, 24, 40, 49–53, 55, 56,
NORM.DIST, 309, 333, 335, 343, 469 178, 189, 229–231, 420, 431–437,
480 Index
439, 440, 444, 445, 448–451, 455, 378, 382–387, 389, 394, 396, 433,
472 440, 460, 471
Parabola from vertex, 17 Probability Density Function (PDF), 310,
Parallelogram of forces, 203 333–336, 339, 340, 342, 343, 345,
Path velocity, 240, 243 347, 350–352, 358–360, 362, 384
Pendulum, 9, 227, 228, 246–251, 382, 390 Procedure header, 129, 130, 133–135, 138,
Percent point function, 309, 310, 334, 340, 139, 146
352 Product rule of probability, 326, 327
Perpendicular, 110, 112, 141, 182–184, Program flow in Jupyter, 64
204–208, 224, 237, 308 Progress loop, 272, 277, 282, 283, 287, 295,
Perpendicular bisectors, 198 297
Photons, 308, 348–350, 355, 356, 358, 460 Progress with look-ahead, 13, 269, 271, 272,
Physical unit, 10, 76, 199, 200, 208, 232, 247, 274, 291, 302, 305
273, 290, 320, 334, 367, 368, 375, PROJECT EXPLORER, 126, 127, 168, 170
398 Projectile motion, 227
Plain-text file, 161 Projectile trajectory, 227, 229, 230, 233
Plot objects, 234–236, 238 Pythagoras, 96
plt.arrow, 185, 186, 233
plt.axis(‘scaled’), 203
Q
plt.legend, 55
Quantile function, 334
plt.rcParams, 237
plt.savefig, 98
plt.xscale(value”log”), 108 R
plt.yscale, 40, 108 R1C0 in Pandas, 165
Points, 6, 8, 18, 19, 27, 29, 39, 43, 55, 60, 62, Radius of curve, 209, 211
63, 65, 78–81, 85, 86, 88–91, 95, Ray constructions, 76, 83, 85, 90
110, 124, 130, 137, 139, 177–179, RBS, 161, 163
183–185, 187, 199–202, 209–211, rcParams, 39, 235, 237
213, 228, 230, 233, 236, 238–249, RECORD MACRO, 124, 125, 131, 174, 453
251–256, 260, 265–267, 269, 270, Rectangle, 25, 76, 77, 118, 123, 125, 126,
294, 298–301, 318–320, 324–330, 131–135, 139, 148, 318–327, 330
369, 375, 410, 451, 457 Referring to names in a cell, 29
Polar coordinates, 13, 71, 78, 137, 182, 195, Referring to names in a matrix function, 30
200–202, 204, 227, 242, 258, 260, Regression coefficient, 425
261, 265 Relative addressing, 15, 20
Polynomial, 19, 178, 182, 192, 193, 195, 196, Rep-log procedure, 13, 117, 121, 177, 240,
204–207, 215, 216, 218–221, 421, 243, 293, 314, 344, 347, 348, 356,
431, 432, 434, 437, 444, 448, 449, 379–381, 387, 389, 392, 395, 401,
472 403, 427, 428, 435–437, 442, 447,
Polynomial from defining points, 220 449, 450, 471, 472
Power fluctuations, 289 Residual variance, 422, 423, 432, 433
Power-law trend line, 405, 406 RGB, 128
Power operator, 20, 211 Rolling curve, 228, 238, 242, 243
Powers of 2, 100 Rope friction, 300
ppf, 308, 310, 333–336, 340, 341, 343, Rotation, 183, 227, 228, 238, 239, 241–243,
345–347, 350–355, 357, 383, 384, 246–248, 251–256, 266
388, 439, 440 Rotational matrix, 184, 228, 247, 252–256
Printoptions, 45 RotMat, 254
Printoptions for arrays, 45, 46 Round, 222
Probability, 7, 308–310, 313, 314, 323, 324, Round a measurement result, 372
326–330, 333–336, 338–340, Row vector, 12, 16, 27, 41–44, 46, 47, 50, 51,
342–346, 348, 350–352, 358, 359, 53, 63, 65, 66, 171–174, 221, 253,
361–363, 365, 367, 368, 375, 376, 261, 263, 264, 437
Index 481
Runge-Kutta, 6, 13, 269, 271–281, 283, 295, 358–360, 365, 367, 375–378, 382,
305 385, 386, 394, 397–405, 407, 408,
410, 411, 413, 417, 423–426, 431,
435, 436, 445, 460–462, 464, 465,
S 469, 470
Scalar product, 46, 168, 170–173, 182, 198, Standard error, 14, 325, 365–367, 369–372,
200–202, 208, 209, 211, 224 374–380, 382–384, 386, 387, 390,
Scan-log procedures, 117 391, 393, 399, 408, 410–412, 417,
Scipy, 35, 333, 459 421, 424–429, 431–433, 435–437,
scipy.optimize, 421, 448, 456, 461, 463 440, 442, 445–448, 461–463,
Scipy.stats, 35, 307–309, 341, 358, 421, 424, 465–468, 471, 473
437, 438, 460 Standard error range, 367, 369, 370, 376–380,
scipy.stats as sct, 309, 341 382, 421, 424, 425, 427, 428, 442,
SCROLLBAR, 15, 19, 24–26, 71, 179, 200 446
sct.chi2, 358 Static friction, 112, 287, 291
sct.chisquare, 316, 317 STDEV.S, 320
sct.norm, 308 Steiner’s, 228, 251, 254, 256, 267
sct.norm.ppf, 332, 336, 341, 345 Straight line, 9, 43, 70, 76, 78–80, 87, 92–95,
sct.t, 383–385, 388, 396, 415, 416, 429, 433, 102–104, 110, 112, 113, 179, 208,
438–440, 442 228, 238, 260, 265, 269, 350, 384,
sct.t.ppf, 385, 433, 440 403, 406, 419, 420, 423–426,
Secondary y-axis, 40 428–430, 440, 442, 448–451,
Second derivative, 181, 185–187, 189–192, 453–455, 457, 472–474
224 Straight-line segment, 176, 177, 181–184
SET OBJECTIVE, 450, 451 Student’s t, 14, 382, 384, 390, 408, 409, 411,
Sets { }, 34 431, 433, 439, 465
Shape, 17, 36–38, 41–45, 47, 49, 53, 65, 70, Subplots, 108, 234
80, 123, 129, 133, 135, 137, 139, SUB SCROLLBAR1_CHANGE(), 26
144, 187, 283, 296, 315, 357, 364, SUMPRODUCT, 211, 223–225, 253, 254, 256
439, 459 SUMSQ, 223
signum, 93, 279 SUMXMY2, 184, 223
Sine function, 27, 57, 69, 72, 187, 188, 190, Support points, 269, 271
193, 195, 197, 257
Swinging anchor, 228, 256
SI units, 273
Symmetric difference, 34, 329, 330
Slicing, 2, 12, 34, 97, 151, 191, 205, 206, 225,
System of equations, 225
244
Systems of linear equations, 181, 182, 215
Slider, 12, 13, 15, 16, 19, 20, 24–26, 60, 71,
77, 79, 80, 110, 111, 117, 179, 184,
185, 200, 201, 211, 227, 231, 232,
239, 244, 249, 255, 260, 263, 449, T
450, 452, 453 t adaptation, 368, 423, 444, 460, 463, 470
SLSQP, 459 Tangent, 13, 99, 102, 106, 108, 182, 186,
sm.add_constant, 429 204–207, 223, 264, 265
sm.OLS, 429 Tangential vector, 230
SOLVER, 14, 419, 421, 448–454, 456, 459, T.DIST.2T, 383, 473
460, 469, 472–474 Text processing, 151, 152, 157, 158, 176, 355
Solver algorithm, 449, 452, 458, 461, 473 T.INV.2T, 383
Sound, 4, 57, 58, 92, 95–97, 112, 116, 228, Trace to the predecessor, 106
258–261, 263–265, 285, 309 Translation, 227, 228, 238, 239, 243
Sound barrier, 258–260, 265 TRANSPOSE, 45
Sound wave, 258–260, 263 Trapezoidal rule of integration, 224
stackoverflow.com, 35, 158 TRENDLINE, 367, 407, 422, 427, 437, 441,
Standard deviation, 14, 293, 294, 318–322, 445
324, 333, 334, 340, 342–344, TRENDLINE OPTIONS, 437
482 Index
Triangle, 60, 63, 112, 123, 125, 132–134, 136, Vectors, 10, 12, 13, 16, 27, 29, 41–48, 50, 51,
137, 139, 140, 142 53, 63, 65–67, 110, 113, 117, 168,
t statistics, 365, 367, 377, 382, 384, 388, 410, 170–174, 181, 182, 185, 198–202,
419, 431 207–211, 232, 237
Tuple, 41, 142, 330, 332 Vectors in polar and Cartesian coordinates,
Turtle, 115, 135–137, 139–143, 148, 150 198
TypeError, 94 Vectors in the plane, 13, 182, 198
Vectors with Python, 185
Vertex form, 17, 19
U
Uncertainty relation, 59, 61
Uniform motion, 230 W
Union, 34, 327, 328, 330, 394
Wave crests, 258–264
Unit matrix, 45, 48
Wavefronts, 228, 258, 259, 261
Unsupported operand type, 325
Wave-particle dualism, 348
Unwrapping a list, 94
Upper envelope, 17, 19, 49–51, 56 Weighted average, 182, 212, 213, 271, 276,
281, 389
Weighted sum, 13, 253, 392, 397
V
ValueError, 33, 36
Variance, 14, 320, 334, 366, 367, 375, 380, X
390, 391, 393, 395–398, 400, 407, xy-plane, 27, 43, 80, 90, 198–202, 209,
408, 412, 422, 423, 433, 442 211–213, 227, 230, 231, 241, 254,
VAR.S, 320 258, 262, 266
VBA macro to a control element, 26
VBA terms, 155
Vector addition, 199 Z
Vector product, 117, 170–173 zip, 330, 332

Physics With Excel and Python

Uploaded by

Physics With Excel and Python

Uploaded by

Dieter Mergel

Physics with Excel

Physics with Excel

ISBN 978-3-030-82324-5 ISBN 978-3-030-82325-2 (eBook)

Ψ E xcel—as powerful as necessary, Python—as simple as possible.

Felder und Zufallsprozesse, Springer Spektrum (2018), https://doi.org/10.1007/978-3-662-575

Duisburg, Germany Dieter Mergel

2.4.5 Standard Plot Program . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

3.4.6 Spreadsheet Calculation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105

5 Basic Mathematical Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181

5.10 Some Mathematical Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222

7.3 Falling from a (Not Too) Great Height . . . . . . . . . . . . . . . . . . . . . . . . 278

8.5.3 Where Do Observed and Theoretical Frequencies

10.5 Solving Nonlinear Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 448

Possible errata and corrections in the internet at: go.sn.pub/Ob4vCR.

1.1 A Two-Track Didactical Approach

© Springer Nature Switzerland AG 2022 1

Didactical advantages of the two-track system

– Spreadsheets are interactive; charts react immediately to changes, and data

Our guideline: spreadsheet calculations as powerful as necessary, Python program-

1.2 What Can You Expect?

What can you, dear reader, expect from this book?

and, with that,

– getting a better understanding of certain mathematical and physical concepts.

1.3 What Do You Need?

1.4 Tim, Alac, and Mag

Tim: Computers are not my thing

Alac Why excel and Python?

Mag All algorithm-oriented computer languages have a similar structure.

Alac: How do I become a master programmer?

Alac And that is what this course will accomplish?

Tim Data structures, procedures, controlling; that sounds pretty challenging.

Tim Is that thorough enough?

Alac Will I learn the more tricky constructions too late?

Tim Can I manage this in addition to my studies in Physics?

1.5 Didactic Concept

Courses with 30 attendance hours

Subjects from a one-year physics course

Simple solutions preferred

1.6 Subject Matter

Block A, Fundamentals (Chaps. 2, 3 and 4)

Block B, Physics and Mathematics

Block C, Simulation and analysis of experiments

1.7 Getting Started with Excel

1.7.1 Start Menu

1.7.2 Spreadsheet Presentation

– the parameters of the task are defined in the range C2:C6,

– the values for the independent variable t are in B14: B174.

1 Sub Protoc() Range("C3") = Tp 8

– values are calculated from the parameters and independent variables,

1.8 Getting Started with Python

Fig. 1.4 Window opened to create a new program file

1.9 Skills to Be Trained

Spreadsheet operations (Exercise 2.3)

– Using cell addressing, absolute, relative, indirect

– Using sliders to change cell contents

Python constructs (explained mainly in Exercises 2.4 and 2.5)

– Use of numpy arrays

VBA-macros and Python instructions (Chap. 4)

– For, if , Sub/def , basic structures of programming: loops, logical queries, subrou-

– Using the line equation constructively (Exercise 3.2)

– Properties of the exponential function (Exercise 3.4)

Integration of Newton’s equation of motion (Chap. 7)

– Approximated average value instead of exact integral

Statistics (mainly Chap. 8)

– Mostly, not always. Fundamental rule of statistical reasoning

Evaluation of measurements (mainly Chaps. 9, 10)

– We know everything and play stupid.

This chapter aims to develop computational solutions for physics problems,

2.1 Introduction: Named Ranges in Excel, Arrays in Numpy

© Springer Nature Switzerland AG 2022 15

– relative and absolute cell addressing,

2.2 Characteristics of a Parabola

2.2.1 Different Definitions of a Parabola

Parabola from vertex

It is, however, more intuitive to start from its vertex form

e.g., b = −2cx V recurring to variables specified earlier. The calculated values

8 Numerical 0 becomes False, = 0 becomes True. False is output as 0, True as 1.