Mathematical Programming with MATLAB

Monday 20.09.2010 Click to edit Master subtitle style

5/13/12

HA13: Differential Equations

Differential Equation Solvers Function Handles and Anonymous Functions First Order Differential Equation Second Order Differential Equations Initial Value Problems Boundary Value Problems
5/13/12

Solving Differential Equations Numerically

MATLAB includes a number of functions that solve ordinary differential equations of the form

Higher order differential equations and systems of differential equations must be reformulated into a system of first order expressions.
5/13/12

Differential Equation Solvers

Not every differential equation can be solved using the same technique. All of the differential equation solvers in MATLAB have the same format. This makes it easy to try different techniques, by just changing the function name. MATLAB includes a large family of differential equation solvers. A good first try is usually the ode23 or ode45 solver function, which uses a RungeKutta technique.

5/13/12

Family of Solver Functions

A minimum of three inputs is required - function handle to a function that describes the first order differential equation or system of differential equations in terms of t and y

- the time span of interest - an initial condition for each equation in If you do not specify the result array, [t,y] , the system. the functions create a plot of the results
[t,y] = odesolver(function_handle, [initial_time, final_time], [initial_cond_array])

5/13/12

ODE solver function ode45

Type of problems likely to be solved with this technique non-stiff differential equations

Numerical Comments Solution Method Runge-Kutta Best choice for a first guess technique, if you dont know much about the function. Uses an explicit Runge-Kutta (4,5) formula, called the Dormand-Prince pair

ode23

non-stiff differential equations

Runge-Kutta

This techniques uses an explicit Runge-Kutta (2,3) pair of Bogacki and Shampine. If the function is mildly stiff this may be a better approach Unlike ode45 and ode23 which are single step solvers, this technique is a multistep solver. Uses numerical differentiation formulas or backward differentiation formulas. It is difficult to predict which technique will work best for a stiff differential equation modified second order Rosenbock formulation Useful if you need a solution without numerical damping

ode113 ode15s

non-stiff differential equations

Adams

stiff differential equation and differential NDFs (BDFs) algebraic equations stiff differential equations Rosenbrock

ode23s ode23t

moderately stiff differential equations and Trapazoid rule differential algebraic equations stiff differential equations TR-BDF2

ode23tb

This solver uses an implicit Runge-Kutta formula with the trapazoid rule and a second order backward differentiation formula.

ode15i

fully implicit differential equations BDFs

5/13/12

Function Handles

A function handle is a nickname (alias name) for a function. The function handle can refer to

a standard MATLAB function, stored as an M-file

an anonymous MATLAB function. Anonymous functions can be defined in an M-file program, or in the command window.

5/13/12

This function corresponds to:

5/13/12

You can use an anonymous function as input to an ode solver function Because we did not assign a name to the ode function output, a graph is created. See next slide.
5/13/12

Solving Differential Equations Solution to dy/dt = 2 * t Numerically

1 0.8 0.6 y 0.4 0.2 0 -1 -0.5 0 time 0.5 1

This figure was generated automatically by the ode45 function. The title and labels were added in the usual way.
5/13/12

Systems of differential equations

If you want to specify a system of equations, it is probably easier to use a function M-file. The output of the function must be a column vector of first derivative values. For example:

5/13/12

Systems of differential equations: Example

This function M-file represents a system of equations

5/13/12

Systems of differential equations: Example continue

We can now use an M-file function as input to an ODE solver The handle for an existing M-file is @m_file_name ode45(@another_fun, [-1,1],
[1,1])

5/13/12

Systems of differential equations: Example and dx/dt= -y Solution to dy/dt=x continue

2 1 x and y 0 -1 -2 -1 y x

-0.5

0 time

0.5

This system of equations was solved using ode45. The 5/13/12 title, labels and legend were added in the

1st Order Differential Equations

The initial value problem of the 1st order differential equation is of the form y'(t) = f(t), y(0) = a It can be solved in MATLAB using functions ode23 or ode45. Example: RL- circuit (a resistor and an inductor) L I + R I = E,
5/13/12

where

E=12, R=2, L=0.5.

RL Circuit

The function file is diff_i.m

function di = diff_i(t,i) % The differential equation of the RL circuit E=12; R=2; L=0.5; % voltage % resistance % inductanse

di = E/L - (R/L)*i;
5/13/12

RL Circuit: Current as Function of Time

% RL circuit: % calculate the current as a function of time t0 = 0; tl = 5; i0 = 0; % initial condition at time t0 % solve current for values 0...5 s

[t, i] = ode23('diff_i', t0, tl, x0); plot(t,i,'-b') axis([0 2 0 7])

5/13/12

title('RL Circuit')

2nd Order Differential Equation

The initial value problem of the 2nd order differential equation of the form y''(t) = f(t), y(0) = a, y'(0) = b have to be transformed to a system of 1st order differential equations. Let us call the original variable y1= y Introduce a new variable y2 = y1' After substituting to the original equation we get the system of following equations y1 = y2 5/13/12

Simple Harmonic Motion

A typical example of a system that undergoes simple harmonic motion is an idealized spring mass system, in which is a mass is attached to a spring. The differential equation is m x + b x + k x = mg, where k=10, b=2, g=9.81, m=2 x1 = x We x2 = x1 have to solve the system of two equations: x1 = x2 x2 = - (k/m) x1 (b/m) x2 + g
5/13/12

Solution in MATLAB

The function string.m contains the definition of the differential equations

function xe=string(t,x) k=10; b=2; g=9.81; m=2; A=[0 1; -k/m -b/m]; b=[0,g]; xe=A*x+b';
5/13/12

Solution in MATLAB continue

On the command window we call the function ode23

>> t0=0; tfin=10; % the start and end times of calculation >> x0=[0;0]; % initial values >> [t,x]=ode23(string',t0,tfin,x0); >> plot(t,x(:,1)) % plot the position vector x = x_1 >> plot(t,x(:,2))
5/13/12%

plot the velocity vector

x = x_2

Component Form of Equations

Instead of the matrix form the system of differential equations can be written in the function string.m alternatively also in the component form:
function xe=string(t,x) k=10; b=2; g=9.81; m=2; xe(1) = x(2); xe(2) = (-k/m)*x(1) + (-b/m)*x(2) + g;
5/13/12

MATLAB dsolve (from Symbolic Toolbox)

dy =y dt

To pin down the constant C1 we need a boundary condition. To find C1, we need a boundary condition that tells us some value of y(t): e.g. y(1)=2.

Symbolic Toolbox is an optional toolbox that allows symbolic computations. 5/13/12 It must be installed separately on your computer.

Another differential equation example

What function y(t) has as its derivative 2y/t ?

Makes sense since derivative of t2 is 2t and 2t = 2 t2/t

5/13/12

HA14: Other Numerical Techniques

Interpolate between data points, using either linear or cubic spline models. Perform numerical differentiations Perform numerical integrations Solve differential equations numerically Root Finding and Minimization

5/13/12

Interpolation: Summary

MATLAB includes an interpolation technique called interp1. Its is the most commonly used function for linear interpolations. A cubic spline technique is also available.

5/13/12

Linear Interpolation
Assume the function between two points is a straight line.
12

A Data Plot

Measured Data
16 14

10

Interpolated Point
y-axis

12 10 8 6 4

Interpolated Points

y-axis

2 0

-1

x-axis

x-axis

Linear Interpolation Connect the points with a straight line to find y

5/13/12

Linear Interpolation: MATLAB Code

interp1 is the MATLAB function for linear

interpolation.

First define an array of x and the y values. Now define a new x array, that includes the x values for which you want to find y values
new_y=interp1(x,y, x_new)

5/13/12

Interpolation: Function interp1

Interpolation is the technique of creating new data points within existing points MATLAB has built-in interpolation functions for both 1-D and 2-D interpolation. 1-D Syntax
ynew = interp1(xorig,yorig,xnew,method) Where ynew - new y value(s) xorig, - original x values

Where method can be:

'nearest - nearest neighbor interpolation 'linear (default) 'spline (cubic spline interpolation 'cubic (piecewise cubic)

5/13/12 - original y values yorig

Measured Data
16 14 12 10

y-axis

8 6 4 2 0 -1 0 1 2 3 4 5 6

x-axis

5/13/12

Other Interpolation Techniques

Cubic Spline - A cubic spline creates a smooth curve, using a third degree polynomial MATLAB includes other interpolation techniques including

Nearest Neighbor Cubic Two dimensional interpolation: interp2

Three dimensional interpolation: interp3 5/13/12

We can get an improved estimate by using the spline interpolation technique

5/13/12

5/13/12

16 14 12 10

Cubic Spline Interpolation

y-axis

8 6 4 2 0 -1 0 1 2 3 4 5 6

x-axis
Cubic Spline Interpolation. The data points on the smooth curve were calculated. The data points on the straight line segments were measured. Note that every measured point 5/13/12 also falls on the curved line.

Interpolation: Demo
Perform the following: >> x = 0:12;

Create a vector from 0 to 12 Evaluate cosine at every x value. Create a more fine spacing Create new y values at the finer x spacing. Repeat the interpolation but with a cubic spline technique. Plot the three datasets the original, the default interpolation and the cubic spline interpolation

>> y = cos(x);

>> xi = linspace(0,12,50);

>> yi = interp1(x,y,xi);

1 . 5

>> yi2 = interp1(x,y,xi,spline);

1 0 . 5 0

Note the different results from the different interpolation methods

1 2

0 . 5

1 0

1 0

Data Interpolation: Example

x = 0:.5:4*pi; y = sin(x) + sin(2*x)/2; plot(x,y,'bo-') linear % shows

% Now assume have only the points, but not the function xx = 0:.1:4pi sampling yy = interp1(x,y,xx); hold on 5/13/12 % finer

Numerical Derivatives

The diff function finds the difference between adjacent elements in an array It can be used to approximate a derivative.

5/13/12

Function diff

Numerical differentiation uses the diff function. The diff function is easy to understand, even if you have not taken calculus. It just calculates the difference between the points in an array. If we know how y changes with x, we could create a set of ordered pairs for any number of x values. The more values of x and y, the better the 5/13/12 approximation of the slope.

Function diff: Example

S mleDta ap a 1 6 1 4 1 2 1 0 y-axis 8 6 4 2 0 1

2 xa is -x

The derivative of a data set can be approximated by finding the slope of a straight line connecting 5/13/12 data point each

Numerical Derivative

Finite differences can be used to compute a numerical derivative. MATLAB provides a built-in function for calculating numerical derivative using the forward difference method.

It is easy to create your own difference formula by taking advantage of MATLABs vector processing. 5/13/12

Numerical Derivative: Function diff

Syntax
ynew = diff(y,order) Where: y ynew data to differentiate forward differenced data order order of the derivative
To compute a derivative you need to use diff on both the y and x data y x

If you have a function that you want to differentiate, first evaluate the function using a small Dx, then put the resulting data through diff

The slope is an approximation of the derivative in this case based on data measurements
5/13/12

Numerical Derivative: Demo

Perform the following: >> xi = 0:0.25:12; >> yi= xi.*2 + xi; >> yi_diff1 = diff(yi,1); >> yi_diff2 = diff(yi,2); >> xi_diff1 = diff(xi,1); >> xi_diff2 = diff(xi,2); >> plot(yi,b-);
20

Create a vector from 0 to 12 Evaluate your function Calculate the first y difference Calculate the second y difference Repeat for the xs

Plot your three different lines, making sure to divide by the xs.

>> hold, plot(yi_diff1./xi_diff1,r--); >> plot(yi_diff2./xi_diff2,g:);

10 5

15

f(x ) f'( x ) f''(x )

10

15

20

25

30

35

40

45

50

Numerical Integration

MATLAB handles numerical integration with two different quadrature functions

o

quad uses adaptive Simpson quadrature quadl uses adaptive Lobatto quadrature

Both functions require the user to enter a function in the first field.

called out explicitly as a character string defined in an M-file anonymous function.

The last two fields in the function define the limits of integration
5/13/12

5/13/12

Numerical Integration: Geometrical Derivation An integral is often thought of as the area

under a curve
15 An integral can be approximated by the area under a curve 15 A Trapazoid rule approximation

10 y-axis y-axis 5

10

2 x-axis

2 x-axis

The area under a curve can be approximated using the trapezoid rule.
5/13/12

Numerical Integration

You integrate to either determine the area under the curve, or as part of a differential equation. There are times when you cannot integrate a function analytically. The trapezoidal method is a common technique for computing area under the curve. If noise in your data is a problem, fit the data to a function or smooth your data with averaging.

Numerical Integration continue

MATLAB provides several built-in functions for computing area under the curve trapz function calculates area for data sets quad function calculates area for functions Syntax
area = trapz(x,y) Where x vector of x data y - vector of y data area = quad(fn,lo,hi) Where fn the function to evaluate lo lower bound hi upper bound

5/13/12

Calculating an Integral: Demo

Integrate the following function: trapz function:
>> x = 0.001:0.1,15; >> func = x.^3 ./ (exp(x) - 1); >> area = trapz(x,func) area = 6.4926

Normally you would use the trapz function if you did not have a function and only data

quad function:
>> area = quad('x.^3 ./ (exp(x) - 1)',001,15) area = 6.4927 5/13/12
Rounded answer: 6.4939

Data Problems

Real data is typically ugly

Problem: Noisy Data

Do not really want to exactly fit to the points Real question: what is the "best fit" n is "degree" of the polynomial returns coefficients polyval(coeffs,X) evals at vector X using those coeffs go through sparse data (knot points);

'polyfit(x,y,n)' minimizes "squared error"

Splines 5/13/12

Interpolation vs Extrapolation

Interpolation is MUCH better behaved has at least two points to help. Extrapolation is often really ugly

Check different order of polyfit using values of 6 and -1

5/13/12

Numerical Derivative: Smoothing

Noise in the data will impact the calculation and requires work-arounds. Fit the data to a function, then differentiate the function analytically or with numerical methods. Filter (smooth) the data, then differentiate the smoothed data. Filtering can change calculated derivative values. Watch out for a shift in your data when using moving averages. Use the Less is More approach.

0 -2 -4 Pressure -6 100 -8 dP/dT -10 -12 -14 0 10 20 Time -100 -150 0 10 20 Time 30 40 50 0 -50 30 40 150

0 -2 -4 Pressure -6 -8 -10 -12 -14 0 10 20 Time 30 40

Original Data

30 point moving average

5/13/12

Data Averaging (Smoothing)

Data averaging can be accomplished through convn and filter.

These functions use a mathematical technique called convolution. In convolution, you have a data set, and what is termed a kernel. The weighting of the average depends on the values used in the kernel. The kernel is slid along the data set, elements are multiplied together, the results are summed and a new data set is built from the results.

Kernel values should sum to 1 or else you will change the mean level of your data! Unequal kernel values will give a weighted average 5/13/12

Data Averaging (Smoothing)

Syntax (for our case well just deal with vectors)
ysmooth = convn( yorig , kernel , 'shape ) Where: yorig kernel shape original row or column data to smooth row or column vector of filter weights same returns the same length as the data set.

ysmooth = filter( kernel, 1, yorig) Where: convn with the same option does not cause a data shift, whereas filter will row or column by the number of points yorig shift the data data to smooth used in the smoothing. column vector of filter weights kernel row or
5/13/12

Data Averaging (Smoothing): Demo

Perform the following:
Load the decaying sine data file from above (save to your desktop, then double-click it in the Current Directory window, or use the load command)

>> kernel = ones(5,1)./5; >> C >> D = convn(ysine,kernel,'same'); = filter(kernel,1,ysine);

Create a [5x1] vector where each value = 0.2 and the sum =1 Moving avg with convn Moving avg with filter Plot data and build a legend

>> plot(xsine,ysine,'.',xsine,C,xsine,D) >> legend('data','5 pt avg - convn','5 pt avg - filter')

1 .5 d ta a 5p ag- c nn t v ov 5p ag- filte t v r

0 .5

Notice how filter shifted the data

1 5

- .5 0

5/13/12

1 0

Root and Minima Locating

MATLAB provides built-in functionality finding the zeros and minima of a function. Syntax
ymin = fminsearch(fcn,xguess) yzero = fzero(fcn,xguess) Where fcn function to search xguess guess at the minimum

Where fcn function to search xguess guess at the zero

5/13/12

Find the root & minimum of this function: fzero function:

>> x = linspace(-5,5,100); >> y = x.^2-4.*x-8; >> plot(x,y) >> zero = fzero('x.^2-4.*x-8',0) zero = -1.4641
40 30 20 10

Root and Minima Locating: Demo

(-1.4641,0) 0 -10 -20 -5 0 (2,-12)

fminsearch function:

>> minx = fminsearch('x.^2-4.*x-8',0) minx = 2 5/13/12

Roots of Function: Example

Example: find roots of the function

f(x)=sin(x)
>> x0 = 3.1 ; >> tol = 0.01; % starting point % desired accuracy % show

>> trace = 1; intermediate results

>> z1 = fzero('sin(x)', x0) >> z1 = fzero('sin(x)', x0, tol) >> z1 = fzero('sin(x)', x0, tol, trace)
5/13/12

System of Non-linear Equations

First write the definition for the system of the equations in the file eqns.m
function eq=eqns(p); %EQNS defines the left hand side of equations

% The system of the nonlinear equations to be solved is % x^2 + y^2 - 3xy = 0 % x^2 + y = 0 x = p(1); y = p(2); eq = zeros(2,1); eq(1) = x^3 + y^3 - 3*x*y; eq(2) = x^2 + y;
5/13/12

Symbolic Derivatives

Symbolic Toolbox is an optional toolbox that allows symbolic computations in MATLAB. It must be installed separately on your computer. 5/13/12

Example: Partial Derivative of Gaussian

5/13/12

Symbolic Integration

5/13/12

Definite Integral: Example

5/13/12

Symbolic Bounds of Integration

5/13/12

Improper Integrals (of one type)

5/13/12

5/13/12