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 11.

Tue, Oct 2, 2007

Dierentials.
Recall in single variable calculus: y =f (x) dy = f (x) dx. Example: y = sin1 (x) x = sin y,
dx = cos y dy, so dy/dx = 1/ cos y = 1/ 1 x2 .
Total dierential: f = f (x, y, z) df = fx dx + fy dy + fz dz.
This is a new type of object, with its own rules for manipulating it (df is not the same as f !
The textbook has it wrong.) It encodes how variations of f are related to variations of x, y, z. We
can use it in two ways:
1. as a placeholder for approximation formulas: f fx x + fy y + fz z.
2. divide by dt to get the chain rule: if x = x(t), y = y(t), z = z(t), then f becomes a function
df
dx
dy
dz
of t and
= fx
+ fy
+ fz
dt
dt
dt
dt
Example: w = x2 y + z, dw = 2xy dx + x2 dy + dz. If x = t, y = et , z = sin t then the chain rule
gives dw/dt = (2tet ) 1 + (t2 ) et + cos t, same as what we obtain by substitution into formula for w
and one-variable dierentiation.
Can justify the chain rule in 2 ways:
1. dx = x (t) dt, dy = y (t) dt, dz = z (t) dt, so substituting we get dw = fx dx + fy dy + fz dz =
fx x (t) dt + fy y (t) dt + fz z (t) dt, hence dw/dt.
2. (more rigorous): w fx x + fy y + fz z, divide both sides by t and take limit as
t 0.
Applications of chain rule:
Product and quotient formulas for derivatives: f = uv, u = u(t), v = v(t), then d(uv)/dt =
fu u + fv v = vu + uv . Similarly with g = u/v, d(u/v)/dt = gu u + gv v = (1/v) u + (u/v 2 ) v =
(u v uv )/v 2 .
Chain rule with more variables: for example w = f (x, y), x = x(u, v), y = y(u, v). Then
dw = fx dx + fy dy = fx (xu du + xv dv) + fy (yu du + yv dv) = (fx xu + fy yu ) du + (fx xv + fy yv ) dv.
Identifying coecients of du and dv we get f /u = fx xu + fy yu and similarly for f /v.
It's not legal to simplify by x.
Example: polar coordinates: x = r cos , y = r sin . Then fr = fx xr + fy yr = cos fx + sin fy ,
and similarly f .
18.02 Lecture 12.

Thu, Oct 4, 2007

Handouts: PS4 solutions, PS5.

Gradient.
dw
dx
dy
dz
dw
dr
Recall chain rule:
= wx
+ wy
+ wz . In vector notation:
= w .
dt
dt
dt
dt
dt
dt
Denition: w = wx , wy , wz GRADIENT VECTOR.

Theorem: w is perpendicular to the level surfaces w = c.

Example 1: w = ax + by + cz, then w = d is a plane with normal vector w = a, b, c.

Example 2: w = x2 + y 2 , then w = c are circles, w = 2x, 2y points radially out so circles.

Example 3: w = x2 y 2 , shown on applet (Lagrange multipliers applet with g disabled).

w is a vector whose value depends on the point (x, y) where we evaluate w.

Proof: take a curve r = r(t) contained inside level surface w = c. Then velocity v = dr/dt is in
the tangent plane, and by chain rule, dw/dt = w dr/dt = 0, so v w. This is true for every v
in the tangent plane.
Application: tangent plane to a surface. Example: tangent plane to x2 + y 2 z 2 = 4 at (2, 1, 1):
gradientis 2x, 2y, 2z = 4, 2, 2; tangent plane is 4x + 2y 2z = 8. (Here we could also solve
for z = x2 + y 2 4 and use linear approximation formula, but in general we cant.)
(Another way to get the tangent plane: dw = 2x dx + 2y dy 2z dz = 4dx + 2dy 2dz. So
w 4x + 2y 2z. The level surface is w = 0, its tangent plane approximation is
4x + 2y 2z = 0, i.e. 4(x 2) + 2(y 1) 2(z 1) = 0, same as above).
Directional derivative. Rate of change of w as we move (x, y) in an arbitrary direction.
Take a unit vector u
= a, b, and look at straight line trajectory r(s) with velocity u
, given by
x(s) = x0 + as, y(s) = y0 + bs. (unit speed, so s is arclength!)
dw
Notation:
.
ds |u
Geometrically: slice of graph by a vertical plane (not parallel to x or y axes anymore). Directional
derivative is the slope. Shown on applet (Functions of two variables), with w = x2 + y 2 + 1, and
rotating slices through a point of the graph.
dw
dr
Know how to calculate dw/ds by chain rule:
= w
= w u
.
ds |u
ds
Geometric interpretation: dw/ds = w u
= |w| cos . Maximal for cos = 1, when u
is in
direction of w. Hence: direction of w is that of fastest increase of w, and |w| is the directional
derivative in that direction. We have dw/ds = 0 when u
w, i.e. when u
is tangent to direction
of level surface.

18.02 Lecture 13.

Fri, Oct 5, 2007 (estimated written before lecture)

Practice exams 2A and 2B are on course web page.

Lagrange multipliers.
Problem: min/max when variables are constrained by an equation g(x, y, z) = c.

Example: nd point of xy = 3 closest to origin ? I.e. minimize x2 + y 2 , or better f (x, y) =

x2 + y 2 , subject to g(x, y) = xy = 3. Illustrated using Lagrange multipliers applet.
Observe on picture: at the minimum, the level curves are tangent to each other, so the normal
vectors f and g are parallel.
So: there exists (multiplier) such that f = g. We replace the constrained min/max
problem in 2 variables with equations involving 3 variables x, y, :

fx = gx
2x = y

i.e.
here
fy = gy
2y = x

g = c

xy = 3.

2x y = 0
In general solving may be hard and require a computer. Here, linear algebra:
x + 2y = 0
2
requires either x = y = 0 (impossible,
det = 4 = 0. So = 2. No solutions
since xy =3), or
for = 2, while = 2 gives ( 3, 3) and ( 3, 3). (Checked on applet that f = 2g at
minimum).
Why the method works: at constrained min/max, moving in any direction along the constraint
df
= f u
= 0, i.e.
surface g = c should give df /ds = 0. So, for any u
tangent to {g = c}, ds
|u

u
f . Therefore f is normal to tangent plane to g = c, and so is g, hence the gradient
vectors are parallel.
Warning: method doesnt say whether we have a min or a max, and second derivative test doesnt
apply with constrained variables. Need to answer using geometric argument or by comparing values
of f .
Advanced example: surface-minimizing pyramid.
Triangular-based pyramid with given triangle as base and given volume V , using as little surface
area as possible.
Note: V = 13 Abase h, so height h is xed, top vertex moves in a plane z = h.
We can set up problem in coordinates: base vertices P1 = (x1 , y1 , 0), P2 , P3 , and top vertex
P = (x, y, h). Then areas of faces = 12 |PP1 PP2 |, etc. Calculations to nd critical point of
function of (x, y) are very hard.
Key idea: use variables adapted to the geometry, instead of (x, y): let a1 , a2 , a3 = lengths of
sides of the base triangle; u1 , u2 , u3 = distances in the xy-plane from the projection of
P to the
sides of the base triangle. Then each face is a triangle with base length ai and height u2i + h2
(using Pythagorean theorem).

So we must minimize f (u1 , u2 , u3 ) = 12 a1 u21 + h2 + 12 a2 u22 + h2 + 12 a3 u23 + h2 .

Constraint? (asked using ashcards; this was a bad choice, very few students responded at
all.) Decomposing base into 3 smaller triangles with heights ui , we must have g(u1 , u2 , u3 ) =
1
1
1
2 a1 u1 + 2 a2 u2 + 2 a3 u3 = Abase .
Lagrange multiplier method: f = g gives
a1
u
a1
1
=

, similarly for u2 and u3 .

2 u21 + h2
2
u1
u2
u3
We conclude = 2
= 2
= 2
, hence u1 = u2 = u3 , so P lies above the
2
2
u1 + h
u2 + h
u 3 + h2
incenter.