MIT OpenCourseWare

http://ocw.mit.edu

18.02 Multivariable Calculus

Fall 2007

For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.
 18.02 Lecture 8. – Tue, Sept 25, 2007

Functions of several variables.

Recall: for a function of 1 variable, we can plot its graph, and the derivative is the slope of the
tangent line to the graph.
Plotting graphs of functions of 2 variables: examples z = −y, z = 1 − x2 − y 2 , using slices by
the coordinate planes. (derived carefully).
Contour plot: level curves f (x, y) = c. Amounts to slicing the graph by horizontal planes z = c.
Showed 2 examples from “real life”: a topographical map, and a temperature map, then did the
examples z = −y and z = 1 − x2 − y 2 . Showed more examples of computer plots (z = x2 + y 2 ,
z = y 2 − x2 , and another one).
Contour plot gives some qualitative info about how f varies when we change x, y. (shown an
example where increasing x leads f to increase).
Partial derivatives.
∂f f (x0 + Δx, y0 ) − f (x0 , y0 )
fx = = lim ; same for fy .
∂x Δx→0 Δx
Geometric interpretation: fx , fy are slopes of tangent lines of vertical slices of the graph of f
(fixing y = y0 ; fixing x = x0 ).
How to compute: treat x as variable, y as constant.
Example: f (x, y) = x3 y + y 2 , then fx = 3x2 y, fy = x3 + 2y.

18.02 Lecture 9. – Thu, Sept 27, 2007

Handouts: PS3 solutions, PS4.
Linear approximation
Interpretation of fx , fy as slopes of slices of the graph by planes parallel to xz and yz planes.
Linear approximation formula: Δf ≈ fx Δx + fy Δy.
Justification: fx and fy give slopes of two lines tangent to the graph:
y = y0 , z = z0 + fx (x0 , y0 )(x − x0 ) and x = x0 , z = z0 + fy (x0 , y0 )(y − y0 ).
We can use this to get the equation of the tangent plane to the graph:
z = z0 + fx (x0 , y0 )(x − x0 ) + fy (x0 , y0 )(y − y0 ).
Approximation formula = the graph is close to its tangent plane.
Min/max problems.
At a local max or min, fx = 0 and fy = 0 (since (x0 , y0 ) is a local max or min of the slice).
Because 2 lines determine tangent plane, this is enough to ensure that tangent plane is horizontal
(approximation formula: Δf � 0, or rather, |Δf | � |Δx|, |Δy|).
Def of critical point: (x0 , y0 ) where fx = 0 and fy = 0.
A critical point may be a local min, local max, or saddle.
Example: f (x, y) = x2 − 2xy + 3y 2 + 2x − 2y.
Critical point: fx = 2x − 2y + 2 = 0, fy = −2x + 6y − 2 = 0, gives (x0 , y0 ) = (−1, 0) (only one
critical point).
1
2

Is it a max, min or saddle? (pictures shown of each type). Systematic answer: next lecture.
For today: observe f = (x − y)2 + 2y 2 + 2x − 2y = (x − y + 1)2 + 2y 2 − 1 ≥ −1, so minimum.
Least squares.
Set up problem: experimental data (xi , yi ) (i = 1, . . . , n), want to find a best-fit line y = ax + b
(the unknowns here are a, b, not x, y!)
Deviations: yi − (axi + b); want to minimize the total square deviation D = i (yi − (axi + b))2 .
�

∂D ∂D
= 0 and = 0 leads to a 2 × 2 linear system for a and b (done in detail as in Notes LS):
∂a ∂b
�� � �� � �
x2i a + xi b = xi yi
�� � �
xi a + nb = yi

Least-squares setup also works in other cases: e.g. exponential laws

y = ceax (taking logarithms: ln y = ln c + ax, so setting b = ln c we reduce to linear case); or
quadratic laws y = ax2 + bx + c (minimizing total square deviation leads to a 3 × 3 linear system
for a, b, c).
Example: Moore’s Law (number of transistors on a computer chip increases exponentially with
time): showed interpolation line on a log plot.
18.02 Lecture 10. – Fri, Sept 28, 2007
Second derivative test.
Recall critical points can be local min (w = x2 + y 2 ), local max (w = −x2 − y 2 ), saddle (w =
y − x2 ); slides shown of each type.
2

Goal: determine type of a critical point, and find the global min/max.
Note: global min/max may be either at a critical point, or on the boundary of the domain/at
infinity.
We start with the case of w = ax2 + bxy + cy 2 , at (0, 0).
Example from Tuesday: w = x2 −2xy+3y 2 : completing the square, w = (x−y)2 +2y 2 , minimum.
b b b2 1 b
If a �= 0, then w = a(x2 + xy)+cy 2 = a(x+ y)2 +(c− )y 2 = (4a2 (x+ y)2 +(4ac−b2 )y 2 ).
a 2a 4a 4a 2a
3 cases: if 4ac − b2 > 0, same signs, if a > 0 then minimum, if a < 0 then maximum; if
4ac − b2 < 0, opposite signs, saddle; if 4ac − b2 = 0, degenerate case.
� x x
This is related to the quadratic formula: w = y 2 a( )2 + b( ) + c .
�
y y
2 2
If b −4ac < 0 then no roots, so at +bt+c has a constant sign, and w is either always nonnegative
or always nonpositive (min or max). If b2 − 4ac > 0 then at2 + bt + c crosses zero and changes sign,
so w can have both signs, saddle.
General case: second derivative test.
∂2f
We look at second derivatives: fxx = , fxy , fyx , fyy . Fact: fxy = fyx .
∂x2
Given f and a critical point (x0 , y0 ), set A = fxx (x0 , y0 ), B = fxy (x0 , y0 ), C = fyy (x0 , y0 ), then:
– if AC − B 2 > 0 then: if A > 0 (or C), local min; if A < 0, local max.
– if AC − B 2 < 0 then saddle.
 3

– if AC − B 2 = 0 then can’t conclude.

Checked quadratic case (fxx = 2a = A, fxy = b = B, fyy = 2c = C, then AC − B 2 = 4ac − b2 ).

General justification: quadratic approximation formula (Taylor series at order 2):

Δf � fx (x − x0 ) + fy (y − y0 ) + 12 fxx (x − x0 )2 + fxy (x − x0 )(y − y0 ) + 12 fyy (y − y0 )2 .

At a critical point, Δf � A2 (x − x0 )2 + B(x − x0 )(y − y0 ) + C2 (y − y0 )2 . In degenerate case, would
need higher order derivatives to conclude.
NOTE: the global min/max of a function is not necessarily at a critical point! Need to check
boundary / infinity.
1
Example: f (x, y) = x + y + xy , for x > 0, y > 0.
fx = 1 − x12 y = 0, fy = 1 − xy
1 2 2
2 = 0. So x y = 1, xy = 1, only critical point is (1, 1).
fxx = 2/x3 y, fxy = 1/x2 y 2 , fyy = 2/xy 3 . So A = 2, B = 1, C = 2.

Question: type of critical point? Answer: AC − B 2 = 2 · 2 − 1 > 0, A

= 2 > 0, local min.

What about the maximum? Answer: f → ∞ near boundary (x → 0 or y → 0) and at infinity.