MATB41 - Full Course Note

Download as pdf or txt
Download as pdf or txt
You are on page 1of 49

MATB41 1 CURVED LINES AND SURFACES

1 Curved Lines and Surfaces


Curved Lines
An equation involving x and y, gives a curved line in the plane.
By replacing x with x a, the graph is shifted to the right by a.
By replacing y with y b, the graph is shifted upward by b.

The parabola (with vertex) at the origin is defined by


y = ax2 x = by 2
y = ax2 opens up for a > 0, and opens down for a < 0.
x = by 2 opens to the right for b > 0, and opens to the left for b < 0.

The ellipse (with center) at the origin is defined by


⇣ x ⌘2 ⇣ y ⌘2
+ =1
a b
with (a, 0) being the right-most point on the x-axis,
and (0, b) as the top-most point on the y-axis.
When a = b, an ellipse becomes a circle, with a = b = R as the radius of the circle.

The hyperbola (with center) at the origin is defined by


⇣ x ⌘2 ⇣ y ⌘2 ⇣ y ⌘2 ⇣ x ⌘2
=1 =1
a b a b
A hyperbola has 2 curves, and a center line that is not crossed. The 2 curves
goes in the opposite direction of the center.
Since the right side is positive, and a square is positive, the position of the
negative sign determines the direction of the curves.
To determine the direction, we can either set x = 0 or y = 0.
For example, in the case of x2 y 2 = 1, by setting x = 0, we attain y 2 = 1
is impossible, so the verticle line x = 0 is the center and is never crossed.
Thus the 2 curves opens toward right and left.

To find the slant asymptotes when x and y are both large, change the
number 1 on the right side of the equation to 0.

(The above 3 curves together are sometimes called: Conic Sections)

Author: Xiao Page 1


MATB41 1 CURVED LINES AND SURFACES

Curved Surfaces
An equation involving x, y, z, gives a curved surface in 3D space.
In general, this surface is difficult to imagine and sketch in 3D.

The 3D Sphere (with center) at the origin with radius R is given by

x2 + y 2 + z 2 = R 2

The 3D ellipsoid (with center) at the origin is given by


⇣ x ⌘2 ⇣ y ⌘2 ⇣ z ⌘2
+ + =1
a b c
(a, 0, 0), (0, b, 0), (0, 0, c) are the 3 points on the ellipsoid, similar to the ellipse.
When a = b = c, an ellipsoid becomes a sphere, with a = b = c = R as the radius.

Surfaces Attained Through Rotation: r2 = x2 + y 2


When the variables x and y appear together as x2 + y 2 , this signals that the
surface is attained through rotation. Set a new variable r, with r2 = x2 + y 2 .
We graph the equation involving z and r in the r z plane, with r > 0 only.
We then revolve the curve around the z-axis to attain the surface in 3D:
The r-axis stands for both x-axis and y-axis.

Question 1
Sketch the following curves.

1. x = 2y 2 4. x2 + 3y 2 + 2x 12y + 10 = 0
x2 y 2 5. y 2 x2 = 1
2. + =1
2 9
(x 2)2 (y + 2)2
2 2
3. x + 2y = 4 6. =1
4 9
Question 2
Sketch the following surfaces.

1. x2 + y 2 = 1 z2
4. x2 + y 2 + =1
4
2. z = x2 + y 2 p
5. z = ( x2 + y 2 1)2
3. z 2 = x2 + y 2 6. x2 + y 2 + z 2 = 2z

Author: Xiao Page 2


MATB41 2 VECTOR AND MATRIX

2 Vector and Matrix


Vector
A point in 2D needs 2 coordinates. It is typically written as, for example:

1
~a = (1, 2) or ~a =
2
This is the point x = 1 and y = 2 on the 2D xy plane.

A point in 3D needs 3 coordinates. It is typically written as, for example:


2 3
0
~b = (0, 2, 1) or ~b = 425
1
This is the point x = 0, y = 2, z = 1 in 3D space.

The point can also be interpreted as a vector, written in the same way.
We can also think of a vector as an arrow from the origin to the point.

Typically, it doesn’t matter if we write the vector horizontally or vertically.

Matrix
A matrix is a block of numbers written in a rectangle (or square) in a specific
order. For example, A is a 2 ⇥ 3 (2 by 3)matrix, with 2 rows and 3 columns:

1 2 0
A=
0 5 2
We may think of vectors as 1 ⇥ n or n ⇥ 1 matrix. (for n = 2 or 3)
We can add matrices, and multiply matrices by a scalar (number).
   
1 2 5 6 1+5 2+6 6 8
+ = =
3 4 7 8 3+7 4+8 10 12
  
1 2 c·1 c·2 c 2c
c = =
3 4 c·3 c·4 3c 4c
Standard arithmetic rules apply. For matrices A, B, and scalar c:
A+B =B+A cA = Ac c(A + B) = cA + cB

Author: Xiao Page 3


MATB41 3 DETERMINANT

3 Determinant
The determinant of a square matrix is a number.

Start at the first row, first position, write down this value, and remove this
row and this column from the original matrix. Multiply this value to the
determinant of the remaining matrix.
2 3
1 2 3 
4 5 5 6
det 4 5 6 = 1 · det ...
8 9
7 8 9
Now move right, write down this value, and remove this row and this column
from the original matrix. Multiply this value to the determinant of the
remaining matrix with a minus sign.
2 3
1 2 3  
4 5 5 6 4 6
det 4 5 6 = 1 · det 2 · det ...
8 9 7 9
7 8 9
Now move right again and repeat until we reached the last column. The
positive and negative sign need to alternate.
2 3
1 2 3   
5 6 4 6 4 5
det 44 5 6 5 = 1 · det 2 · det + 3 · det
8 9 7 9 7 8
7 8 9
Each time we apply the algorithm, we end up with several new determinants
to calculate, but with matrices of smaller sizes.

a b
det = ad bc
c d
Determinant does not behave well with addition and scalar multiplication:

det(A + B) 6= det(A) + det(B) det(cA) 6= c · det(A)

Question 3
Find the determinant. 2 3 2 3
 1 3 3 3 1 0
3 1 4
1. 2. 3 5 25 3. 41 3 45
2 5
4 4 6 0 0 4

Author: Xiao Page 4


MATB41 4 DOT PRODUCT

4 Dot product
2 vectors can form a dot product, and the result is a scalar (number).
2 3 2 3
x1 y1
6 x2 7 6 y2 7
~x · ~y = 4 5 · 4 5 = x1 y1 + x2 y2 + ...
.. ..
. .
Note that this dot product is di↵erent from scalar multiplication, as we are
multiplying 2 vectors together, not a scalar with a vector.

The length (absolute value) of a vector using the Pythagoras Theorem:


p p
|~x| = ~x · ~x = (x1 )2 + (x2 )2 + ...
Similarly, define the distance between two points using pythagoras Theorem:
p
d(~x, ~y ) = |~x ~y | = (y1 x1 )2 + (y2 x2 )2 + ...
Going from a point a to point c, is always shorter than going to some other
point b first and then back to c. This is called the Triangle Inequality:
d(~a, ~c)  d(~a, ~b) + d(~b, ~c)
|~x + ~y |  |~x| + |~y |
Let ✓ be the angle between ~x and ~y , the dot product is also given by,
~x · ~y = |~x| |~y | cos✓
Thus, 2 vectors are orthogonal (perpendicular) if the dot product is zero.

The dot product have the usual properties of multiplication:


~a · ~b = ~b · ~a
~a · (~b + ~c) = ~a · ~b + ~a · ~c
c(~a · ~b) = (c~a) · ~b = ~a · (c~b)
Note that since the dot product needs two vectors and produce a number, a
quantity such as ~a · ~b · ~c is not well defined.

Question 4
Find the length of the vectors. Find ~a · ~b and the angle between them.
1. ~a = (4, 3), ~b = (2, 1).
2. ~a = (4, 0, 2), ~b = (2, 1, 0).

Author: Xiao Page 5


MATB41 5 CROSS PRODUCT

5 Cross product
2 vectors (in 3D) can form a cross product, and the result is a vector (in 3D).

Let ~x = (x1 , x2 , x3 ), ~y = (y1 , y2 , y3 ) in 3D.


We typically write the cross product using determinant:
2 3
i j k
~x ⇥ ~y = det 4x1 x2 x3 5
y1 y2 y3
~x ⇥ ~y = i(x2 y3 x3 y2 ) j(x1 y3 x3 y1 ) + k(x1 y2 x2 y1 )
~x ⇥ ~y = (x2 y3 x 3 y 2 , x 1 y 3 + x 3 y 1 , x1 y 2 x2 y1 )
The notations i = (1, 0, 0), j = (0, 1, 0), k = (0, 0, 1) are very easy to use,
where the quantity attached to i is the first component of the vector, and
the quantity attached to j would be the second, and k would be the third.

The vector produced by the cross product, ~x ⇥ ~y , would be orthogonal


to both vectors ~x and ~y . Notice that in most cases, there would be 2 such
vectors with this property. They are exactly
~x ⇥ ~y and ~x ⇥ ~y
In fact, we have
~x ⇥ ~y = ~y ⇥ ~x
(Notice that the order of a dot product does not matter, but the order of
cross product matters up to a negative sign.)
The other usual properties of multiplication hold:
(c~a ⇥ ~b) = c(~a ⇥ ~b) = a ⇥ (c~b)
~a ⇥ (~b + ~c) = ~a ⇥ ~b + ~a ⇥ ~c
Similar to the dot product, we can also relate the angle between the 2 vectors,
~x ⇥ ~y = ~x ~y sin ✓
Note that since ~x ⇥ ~y is a vector, we need to take its absolute value.
So 2 vectors are parallel if the cross product is zero.

Question 5
Let ~x = (3, 2, 1), ~y = (1, 1, 1). Find ~x ⇥ ~y .

Author: Xiao Page 6


MATB41 6 LINES AND PLANES

6 Lines and Planes


Line
For a line to be defined, we need a direction ~v and a point on the line r~0 .
2 3
x
4y 5 = ~r = t~v + r~0
z
The vector ~r and the value t do not need to be determined.
~v and r~0 are not unique, any correct pair would give the right equation.)

Plane
For a plane to be defined, we need a normal vector ~n = (a, b, c) and a point
on the plane r~0 .
~n · ~r = ~n · r~0
ax + by + cz = ~n · r~0
where ~r is the same as above and does not need to be determined.
The normal vector ~n is orthogonal to the plane.
~n and r~0 are not unique, any correct pair would give the right equation.)

Plane Defined by 3 Points


Consider 3 points ~u, ~v , w.
~ We can connect any 2 pairs of points together to
create 2 direction vectors which are inside the plane. For example, we may
take ~x = ~u ~v , and ~y = w~ ~v . From there we can form the cross product
to attain a vector that is orthogonal to both vectors inside the plane, so the
cross product would be the normal vector, which is orthogonal to the plane.

Question 6
Find the equation of the line or plane.
1. The line through ( 8, 0, 4) and (3, 2, 4).
2. The plane through the origin and perpendicular to the vector (1, 5, 2).
3. The plane through (2, 4, 6) parallel to the plane x + y z = 5.
4. The plane through (0, 1, 1), (1, 0, 1), (1, 1, 0).
5. The plane equidistant from point (3, 1, 5) and ( 2, 0, 0).

Author: Xiao Page 7


MATB41 7 MULTIVARIABLE FUNCTION

7 Multivariable Function
In first year, we study 1D functions: functions of one variable, as f (x).
This can be plotted in 2D plane, with y = f (x).
Now, we can study functions of multiple variables.

2D Function
A 2D function, f (x, y), has 2 inputs being x, y, and 1 output.

Sometimes, we can label the output as z = f (x, y).


We can plot this 2D function in 3D space, where the height z above the point
(x, y) on the xy-plane is equal to f (x, y).

This would give a graph, which is a surface.


Alternatively, z = f (x, y) can be thought of as an equation which involves
x, y, z, which would give a surface in 3D space.

However, similar to 1D functions, given an equation involving x, y, z, it is


not always possible to isolate z into z = f (x, y).

Examples: f (x, y) = x2 +2y f (x, y) = xexy f (x, y) = sin(xy 2 )

3D Function
A 3D function, f (x, y, z), has 3 inputs being x, y, z, and 1 output.

It is possible to label the output as w = f (x, y, z), but this is typically not
useful as this introduces a 4th variable. The ”plot” of this function would be
in 4D space, which does not exist.

Examples: f (x, y, z) = x2 + 2z 1 f (x, y, z) = yexz

Level Set
The level set of f at k is given by setting f to be equal to the constant k.
This will reduce the dimension of f by 1.

Question 7
1. Draw the level set of f (x, y) = 4x2 + y 2 + 1 for k = 2, k = 5.
2. Draw the level set of f (x, y, z) = x2 + y 2 z for k = 0 and k = 2.

Author: Xiao Page 8


MATB41 8 LIMITS AND CONTINUITY

8 Limits and Continuity


f (x, y) is continuous at the 2D point ~a if

lim f (x, y) = f (~a)


(x,y)!~a

Every standard function we know are continuous in their domains.


Any function transformations (such as +, , ⇥, /, composition) of
continuous functions, are also continuous. (except division by 0)

As most functions are continuous, most limits can be obtained by putting


(x, y) = ~a in the formula of f(x,y).

Di↵erent from 1D functions, computing limits in 2D is much more difficult.


Most limit evaluation techniques from 1D functions does not work for 2D
functions. In particular, there is no L’hopital’s Rule for 2D functions.

Fortunately, since most functions are continuous except when dividing by


0, we typically only need to focus on the point where the function is dividing
by 0 (and state the function is continuous everywhere else).

It turns out that it is much easier to show a limit does not exist.

Test for limit that does not exist in R2


Given f (x, y), compute the limit as (x, y) approaches ~a = (0, 0).
1. Replace y with a easy curve y = g(x), passing through ~a = (0, 0),
(such as y = 0, y = x, y = x2 ), then let x approach 0 to attain a (1D) limit.

2. Replace x with a easy curve x = h(y), passing through ~a = (0, 0),


(such as x = 0, x = y 2 ), then let y appraoch 0 to attain a (1D) limit.

3. Try with several di↵erent g(x) and h(y) to attain many (1D) limits.

4. If you find 2 di↵erent (1D) limits generated by 2 di↵erent curves, then


the (2D) limit of (x, y) approaches ~a = (0, 0) of f (x, y) does not exist.
Note: If all the 1D limits are the same, that is NOT enough for you to
conclude the 2D limit exists. To show the limit exists, we typically must use
Squeeze Theorem.

Author: Xiao Page 9


MATB41 8 LIMITS AND CONTINUITY

Since most limit evaluation techniques does not work for 2D functions, it is
a very fortunate fact that Squeeze Theorem does work for 2D functions. The
formulation is almost the same as the 1D version.

Squeeze Theorem
To attain lim f (x, y), can try to find g(x, y) and h(x, y) so that:
(x,y)!~a

1. g(x, y)  f (x, y)  h(x, y) near the point ~a.


2. lim g(x, y) = L = lim h(x, y)
(x,y)!~a (x,y)!~a

Then we conclude lim f (x.y) = L.


(x,y)!~a

Use Squeeze Theorem to show limit exist


In practice, we typically want to show lim f (x, y) = 0.
(x,y)!~a

Start at |f (x, y)|, create a chain of inequality, and simplify |f (x, y)| to
attain |g(x, y)|, which has limit 0 as (x, y) approaches ~a :
|f (x, y)|  ...  |g(x, y)| ! 0 i.e. lim |g(x, y)| = 0
(x,y)!~a

Then we conclude lim f (x, y) = 0.


(x,y)!~a

The typical strategy in constructing the above inequality is to remove


positive quantities from the denominator of f (x, y).

Question 8
x2 y
f (x, y) = 4 and f (0, 0) = 0
x + y2
Show the limit at (x, y) = (0, 0) does not exist.

Question 9
xy
f (x, y) = p and f (0, 0) = 0
x4 + y 2
Show the function is continuous at (0, 0).

Question 10
x2 sin2 y
Find lim
(x,y)!(0,0) 2x4 + y 2

Author: Xiao Page 10


MATB41 9 DERIVATIVE

9 Derivative
Consider multivariable function f (x, y) or f (x, y, z).
Define the derivative to be:
@f @f
For f (x, y) rf = ( , )
@x @y
@f @f @f
For f (x, y, z) rf = ( , , )
@x @y @z
where rf is called ’del’ f, or gradient of f.
We can interpret rf as a vector, even though f (x, y) or f (x, y, z) is a scalar.
@f
The quantities in rf such as , are called partial derivatives.
@x
When taking partial derivatives with respect to a variable,
we regard all other variables as constants and take derivative as usual.
@f @f @f
Sometimes, we use the notation: = fx = fy = fz
@x @y @z
Example
Let f (x, y) = x2 y 3 + x.
@f
To take the partial derivative with respect to x, , we view y as constant.
@x
@f
To take the partial derivative with respect to y, , we view x as constant.
@y
@f @f
= 2xy 3 + 1 = 3x2 y 2
@x @y
Let f (x, y, z) = x2 z + yz 2 .
@f
To take the partial derivative with respect to x, , we view both y, z as constants.
@x
Similarly, for other partial derivatives:
@f @f @f
= 2xz = z2 = x2 + 2yz
@x @y @z
Question 11
Compute rf .
2
1. f (x, y) = (x2 1)(y + 1) 2. f (x, y) = (xy 2)2 3. f (x, y) =
x + 3y
4. f (x, y) = ex+4y 5. f (x, y, z) = x2 + 2z 1 6. f (x, y, z) = yexz

Author: Xiao Page 11


MATB41 10 DIRECTIONAL DERIVATIVE

10 Directional Derivative
Remember one dimensional derivatives in first year?
f (a + h) f (a)
f 0 (a) = lim
h!0 h
where f is defined on the real line.
We start at the point x = a, and we ask, if we move just a tiny bit, away
from x = a, how much does the function change?

We generalize this idea of moving a tiny bit away in higher dimensions.


In the real line, we can move only left or right.
In 2D, we can move in any direction we like (on the plane).

We can describe the direction of movement by a unit vector, similar to


the direction vector of the equation of a line.
The directional derivative toward the direction ~u at the point ~a:
f (~a + h~u) f (~a)
D~u f (~a) = lim
h!0 h
where ~u is a unit vector.
This is also the rate of change of the function in the direction ~u at point ~a.
~v
Given a vector ~v , we can turn it into a unit vector: ~u =
|~v |
Note: This is a 1D limit, as h is a number (the length of the movement).

Facts:
1. Partial derivatives are directional derivatives with ~u being in a
coordinate direction. Example, for f (x, y):
@f @f
= D~u f = D~u f
@x @y
with ~u = (1, 0) with ~u = (0, 1)
2. If f is di↵erentiable, then all directional derivative exist and
D~u f = rf · ~u
where ~u is a unit vector.
This formula gives an easy way to calculate directional derivative.

Author: Xiao Page 12


MATB41 10 DIRECTIONAL DERIVATIVE

3. We can interpret rf (~a) as a vector.


If we evaluate rf at a point ~a, and if we interpret the vector rf (~a)
as a direction, then the directional derivative in the direction of rf (~a)
at the point ~a is largest, and the value of this maximum directional
derivative is equal to |rf (~a)|.
4. Recall equation of plane: ~n · ~r = ~n · r~0
For f (x, y, z) = k for some constant k, gives a level set surface.
Define the tangent plane at some fixed point r0 = (x0 , y0 , z0 ) by the
normal vector ~n = rf (x0 , y0 , z0 ).
Question 12
f (x, y) = e2x y + y 2 + 4. At the point (0, 0), find the directional derivative
along the direction given by ~v = (1, 3). (First turn ~v into a unit vector)

Question 13
f (x, y, z) = x2 y + z. At the point (2, 2, 1), find the maximum rate of change,
and the direction with the maximum.
(The direction with maximum rate of change is rf , and the value of this
maximum rate of change is |rf |.)

Question 14
f (x, y, z) = xy 2 ez . Let f (x, y, z) = e.
At the point (1, 1, 1), find the equation of the tangent plane.

Question 15
Find all directional derivatives at (0, 0) (including the partials), if they exist.
Is the function continuous at (0, 0)?
8
< xy (x, y) 6= (0, 0)
a) f (x, y) = x2 + y 2
:0 (x, y) = (0, 0)
p
b) f (x, y) = |xy|
Strategy:
When the function has division by 0, or some other problems at the origin
(such as absolute value or square root), we must use the definition of di-
rectional derivative, because f may not be di↵erentiable at (0, 0) so the
formula rf ·~u does not work. Set ~a = (0, 0) and ~u = (a, b), where a2 +b2 = 1.

Author: Xiao Page 13


MATB41 11 DIFFERENTIATION

11 Di↵erentiation
f is di↵erentiable if the gradient vector rf (~a) satisfies
f (~a + ~h) f (~a) rf (~a) · ~h
lim =0
~h!0 |~h|
Note: This is a 2D limit. This limit is typically used when ~a = (0, 0), so
we may set ~h = (x, y), and show the limit is zero (using Squeeze Theorem).

Facts:
1. If f is C 1 at ~a, then f is di↵erentiable at ~a (rf exists).
C 1 : All partial derivatives exist near ~a and including at ~a,
and the partial derivatives are continuous (as functions) at ~a.
2. Being di↵erentiable is a stronger condition than having directional derivative.
Recall, if rf exist, the directional derivative is given by D~u f = rf · ~u
3. If a function is di↵erentiable at ~a, then f is continuous at ~a.
Question 16 8 2
< xy (x, y) 6= (0, 0)
f (x, y) = x4 + y 2
:
0 (x, y) = (0, 0)
Show the directional derivatives exist at (0,0), but f is not di↵erentiable at (0,0).
Strategy:
There are 2 ways to approach the problem.
1. If the function is not continuous at (0, 0), then it is not di↵erentiable at (0, 0).
So we may try to show it is not continuous at (0, 0). However, some functions
are continuous at (0, 0), so we can attain no conclusion in such cases.
2. Compute all the directional derivatives. If f is di↵erentiable, then
every directional derivative should satisfy D~u f = rf · ~u.
1 1
Check this equation for ~u = (1, 0), ~u = (0, 1), ~u = ( p , p ) (easy unit vectors)
2 2
Question 17
Show whether the function is di↵erentiable at (0, 0).
8 8 4
< p x|y|
> 4
(x, y) 6= (0, 0) <x + y (x, y) 6= (0, 0)
a) f (x, y) = x2 + y 2 b) f (x, y) = x2 + y 2
>
:0 :
(x, y) = (0, 0) 0 (x, y) = (0, 0)

Author: Xiao Page 14


MATB41 12 HIGHER ORDER DERIVATIVE

12 Higher Order Derivative


We may take (partial) derivatives on top of partial derivatives.

In 2 dimensions, f (x, y), there are 4 ways of taking second derivatives.


We put them into the Hessian Matrix:
2 2 3 2 ⇣ ⌘ 3 2
@ f @ 2f @ @f @ ⇣ @f ⌘ 3
6 @x2 @xy 7 6 @x @x fxx fxy
H =6 7 @x ⇣ @y ⌘7 = 4 5
4 @ 2 f @ 2 f 5 = 4 @ ⇣ @f ⌘ @ @f 5
fyx fyy
@yx @y 2 @y @x @y @y
Facts:
If f is C 2 , then mixed partial derivatives commute:
The order of taking derivatives does not change the final answer: fxy = fyx
! Taking derivative in x then y, is the same as taking in y then x.

In other words, if f is C 2 , the Hessian Matrix is symmetric.

C 2 : All 2nd order partial derivatives exist near ~a and including at ~a,
and the 2nd order partial derivatives are continuous (as functions) at ~a.
(In practice, a function can fail to be continuous when there is division by zero.)

We can also construct the Hessian for f (x, y, z), which will be a 3 ⇥ 3 matrix.
We can also construct higher order derivatives, such as 3rd order fxyz or fxyy .
In general, if f is C k , then mixed partial derivatives commute (up to order k).

Question 18
Find the Hessian Matrix.
1. f (x, y) = 3x2 + 4xy + 5y 2 2. f (x, y) = cos(x + 2y) 3.f (x, y) = e2x+y
x2 +y
4. f (x, y) = e 5. f (x, y) = e sin(y) 6. f (x, y, z) = x2 y + xz + z 2
x

Question819
3 3
< x y xy (x, y) 6= (0, 0)
f (x, y) = x2 + y 2
:
0 (x, y) = (0, 0)
Find fx , fy at (x, y) = (0, 0) and (x, y) 6= (0, 0) . Then show fxy 6= fyx at (0, 0).
i.e. Taking 2nd derivatives in di↵erent order give di↵erent answers.
Is f C 2 at (0, 0)? (and have we reached a contradiction?)

Author: Xiao Page 15


MATB41 13 CRITICAL POINTS

13 Critical Points
In first year, we found critical points of functions by setting the derivative
to zero. Then we check their second derivative to conclude whether critical
points are max or min. In higher dimensions, it is very similar.

Let f (x, y) be a 2D function (which is C 2 ).


We set rf (~x) = ~0. Solve ~x, these are the critical points.
Then we check second derivatives: the Hessian Matrix.
2 2 3
@ f @ 2f
6 @x2 @xy 7
H =6 4 @ 2f @ 2f 5
7

@yx @y 2
Note that H can contain x and y, so H would be di↵erent for di↵erent points.
For each critical point ~a,
if Det(H (~a)) > 0, look at the top left entry:
@ 2f
if (~a) > 0, then it is a local minimum.
@x2
2
@ f
if (~a) < 0, then it is a local maximum.
@x2
if Det(H (~a)) < 0, then it is a saddle point.
if Det(H (~a)) = 0, then it is inconclusive.

To find the (absolute) maximum and minimum of f in a region D:


1. Find all local extrema inside D using critical points
2. Find all extrema on the boundary of D
3. Compare all the function values to attain the absolute extrema

Question 20
Find and classify all critical points.
1. f (x, y) = 3x2 + 4xy + 5y 2
2. f (x, y) = x2 + 3y 4 + 4y 3 12y 2
3. f (x, y) = (x 1)(x2 y 2 )
2 2
4. f (x, y) = (x2 + 2y 2 )e x y

Question 21
Let f (x, y) = x2 + y 2 4x.
Find the max and min inside the triangle D defined by (0, -3), (0, 3), (3, 0).

Author: Xiao Page 16


MATB41 14 LAGRANGE MULTIPLIER

14 Lagrange Multiplier
Sometimes, we want to maximim/minimize a function on some specific curve,
or there are some constraints on the variables that the function takes in.

Take the function we want to maximize/minimize to be f .


Take the equation that describes the constraint to be g = 0.
(Perhaps we have more than 1 constraint, so the second constraint is h = 0)
The Lagrange Multiplier Formula is:
rf = rg(+µrh + ...)
Each constraint will create an extra r term on the right side with some
constant in front. (We need every r term on the right to be non-zero.)

and µ are constants that you need to solve for, along with n variables of f .
The equation is an equation for vectors, so each of the components are equal.
So, there are n equations in the formula since rf has n components.
For each constant like or µ, there is a equation of constraint for each.
So if there are 2 constraints, there are n + 2 variables that you need to solve,
and you have n + 2 equations to do it.

You are not allowed to divide by zero when you cancel terms.
When attempting to divide a quantity x, you need to split into 2 cases:
Case 1: when x 6= 0 and you can divide
Case 2: when x = 0

Question 22
Find the max and min of the function f (x, y) subject to constraints given.
f (x, y) = 3x 6y with constraint 4x2 + 2y 2 = 25

Question 23
Consider the curve in R2 , C : y = x2
Find the point p on C so that the distance between p and (0, -1) is smallest.
2
Strategy: We minimize f (x, y) = distance2 = (x 0)2 + y ( 1)

Question 24
Consider 2 curves in R2 , C1 : y = x2 and C2 : y = x 1
Find points p on C1 and q on C2 so that the distance between them is smallest.

Author: Xiao Page 17


MATB41 15 REGULAR INTEGRAL IN 2D

15 Regular Integral in 2D
1 Dimensional Integration
In 1 dimensional integration, we have a function that depends only on x.
We can plot the function in 2D, by letting the y = f (x).
This is called the graph of the function.

When we integrate a function, the domain of integration needs to be an


interval [a, b]. The interval can be interpreted as a 1 dimensional object.
The dimension of the domain must match the dimension of the function.

The result of the integral represents the (2 dimensional) area under the curve
y = f (x) above the interval [a, b]. (if f (x) > 0)

2 Dimensional Integration
In 2 dimensional integration, we have a function that depends on x and y.
For example, f (x, y) = xy + x2 , f (x, y) = ex sin(xy + 1).

Similar to the case that f (x) = 10 can be considered as a 1D function,


f (x, y) = 2x, f (x, y) = ey , f (x, y) = 5, can all be considered as a 2D
function, even though they may only depend on one or none of the variables.

If we were to calculate the2D integral of these functions by considering them


as 2D functions, the answer would be di↵erent than their 1D integral (where
we consider them as 1D functions).

To plot a 2D function, we can let z = f (x, y).


The value of the function will be the height of the point plotted in 3D space.
This is the graph of a 2D function. It will be a 2D surface in 3D.

z = sin(y) + 2x

Author: Xiao Page 18


MATB41 15 REGULAR INTEGRAL IN 2D

When we integrate a 2D function f (x, y), the domain of integration needs


to be a patch of area on the xy-plane. Perhaps the domain is a square, or a
circle, or some area that is irregularly shaped.
Regardless, the domain must be a 2 dimensional object on the xy-plane.
The dimension of the domain must match the dimension of the function.

The value of the integral over the domain corresponds to the 3D volume
under the curved surface f (x, y) above the domain of integration.

In the above case, the domain is the rectangle on the xy-plane,


with x between [a, b] and y between [c, d].
For the volume, the base is always flat, but the top will usually be curved,
since the function z = f (x, y) tends to create a curved surface in 3D.
If the domain had been a triangle, then the value of the integral would be
the volume of a triangular prism, except the top will be curved.

Thus, when we try to integrate a 2D function f (x, y) over any domain that
is a 1D curve in the xy-plane, we would be trying to find the volume above
the 1D curve. By agreeing that a curve is infinitely thin, we conclude that
the volume above the curve must be zero.
This is why our domain of integration for f (x, y) must be a 2D area.

Author: Xiao Page 19


MATB41 15 REGULAR INTEGRAL IN 2D

Compute 2D Integrals
A 2D integral typically will look like this:
Z
f (x, y)
D

where D is some 2D domain on the xy-plane.

To compute a 2D integral, we don’t plot the function in 3D space.


We split the 2D integral into 2 components, with dx and dy.

1. Draw the domain in the xy-plane.

2. For all points inside the domain, determine the maximum and minimum
values of x. These two values will be constant numbers xmin , xmax .

3. Pick a random value of x between the minimum and maximum. Con-


sider the verticle line that passes through the point x. Consider the
maximum and minimum values of y on the verticle line.

4. Repeat step 3 several times. For di↵erent values of x chosen, the max
and min values of y would be di↵erent. Typically, all the minimum
values of y would lie on a smooth curve, ymin = m(x). Similarly, the
maximum values would be the same, ymax = M (x). In other words,
the max and min of y is a function of x.

5. Thus the integral can be evaluated as


Z Z Z !
x=xmax y=M (x)
f (x, y) = f (x, y) dy dx
D x=xmin y=m(x)

where the 1D integral inside the bracket is evaluated first. When


carrying out integration with dy, we pretend that the variable x is
constant in the integration. We use the results of the inner integral, to
carry out the integral outside the brackets with dx afterwards.

In the above steps, we first fixed x, and let y vary between the lower bound
curve and the upper bound curve while integrating. Then we add (integrate)
all possible values of x.

Author: Xiao Page 20


MATB41 15 REGULAR INTEGRAL IN 2D

We can also do the reverse by first attaining the min and max values of y in
the domain.
1. Draw the domain in the xy-plane.
2. For all points inside the domain, determine the maximum and minimum
values of y. These two values will be constant numbers ymin , ymax .
3. Pick a random value of y between the minimum and maximum. Con-
sider the horizontal line that passes through the point y. Consider
the maximum and minimum values of x on the horizontal line.
4. Repeat step 3 several times. For di↵erent values of y chosen, the max
and min values of x would be di↵erent. Typically, all the minimum
values of x would lie on a smooth curve, xmin = m(y). Similarly, the
maximum values would be the same, xmax = M (y). In other words,
the max and min of x is a function of y.
5. Thus the integral can be evaluated as
Z Z y=ymax Z x=M (y) !
f (x, y) = f (x, y) dx dy
D y=ymin x=m(y)

where the 1D integral inside the bracket is evaluated first.


Notice in the inner integral, x is integrated first, and y is regarded as constant.
Here, y is integrated after x, whereas in the previous way, x is integrated
after y. This is sometimes called changing the order of integration.

This is not a good name since, when the order is changed, the limits of
integrationg are almost never the same. In fact, the outer integral would
always have limits that are constant numbers.

Fubini’s Theorem tells us that the 2 results will always be the same regard-
less of the order we choose, so we are free to choose the order of integration
to our convenience when calculating 2D integrals.
(Fubini’s Theorem requires all relevant integrals to be integrable.
Typically no division by zero will ensure the function is integrable.)

Sometimes this may be used to our advantage, as some functions do not


have 1D antiderivatives.

Author: Xiao Page 21


MATB41 15 REGULAR INTEGRAL IN 2D

Properties of Integrals
For integrals over some domain D in Rn with integrable functions:
Z Z Z Z Z
(f + g) = f+ g cf = c f
D D D D D
If f  g on all of D, then Z Z
f g
D D

|f | is integrable and Z Z
f  |f |
D D
If domain C is contained in D and f 0, then
Z Z
f f
C D

For domain C and D (that we usually see)


Z Z Z Z
f= f+ f f
C[D C D C\D

Notice that all properties of integral are shared with the integral in 1D.

Sharp points on the boundary of domain


In the case where the domain has a sharp point in the boundary, we can
split up domain at the sharp point into 2 pieces, and integrate the 2 pieces
separately, then add up the answers.

n-dimensional Volume
For any set D in Rn , define the n-dimensional volume of D:
Z
V = 1
D

(Note: here f = 1, the constant function)

If n=1, then V is the length of D. (1D interval with 1-dimensional volume)


If n=2, then V is the area of D. (2D area with 2-dimensional volume)
If n=3, then V is the volume of D. (3D volume with 3-dimensional volume)

Author: Xiao Page 22


MATB41 15 REGULAR INTEGRAL IN 2D

Example 1
Let f (x, y) = 10. Let D be the unit square between 0 and 1.

By noticing that the function is constant, we can see that the plot in 3D
space will be a flat surface.
The integral will represent the volume of the rectangular prism, which
can be calculated as base area ⇥ height.
Thus, the answer will be 10 · 1 = 10.

We can also evaluate the integral using the above steps.


Drawing the domain, we see that the min and max values of x is 0 and 1.
For any fixed value of x, we see that the min and max values of y is 0 and 1.
Z Z Z Z x=1 Z y=1
f (x, y) = 10 = 10( 1) = 10 ( 1 dy) dx
D D D x=0 y=0

Rules of 1D integration apply, we can move the constant 10 outside.


Z x=1
= 10 1 dx = 10
x=0

Example 2
Let f (x, y) = x. Let D be the triangle with vertices (0, 0), (1, 0), (1, 2).

Drawing the domain, The min and max values of x is 0 and 1.


For any fixed value of x, the min value of y is 0.
However, the max value of y changes.
The max value of y all lie on the curve y = 2x = M (x).
Z Z Z x=1 ⇣ Z y=2x ⌘ Z x=1 ⇣ Z y=2x ⌘
f (x, y) = x= x dy dx = x 1 dy dx
D D x=0 y=0 x=0 y=0

Notice that the bound inside the inner integral is a function of x.


Since we are integrating with dy, we regard x as constant, and move it out
of the integral.
Z x=1 ⇣ y = 2x ⌘ Z x=1
2x3 x=1 2
= x y dx = x · 2x dx = =
x=0 y=0 x=0 3 x=0 3

Author: Xiao Page 23


MATB41 15 REGULAR INTEGRAL IN 2D

Domain Defined Using Inequalities


When domains are defined using an inequality, we take the inequality and
make it equality. This will create a curve in the xy-plane which we can draw.
We start on the curve, and move a little bit to the right of the curve. This
increased the value of x, so one side of the equality has become larger.
Using this information, we can determine the area defined by the equality.

Example 3
Let x > 0, y > 0. Let D be the domain defined by y > x3 .

We start by taking the equality y = x3 .


By moving right, we increased x, so we would have y < x3 .
Since this is not what we want, we conclude D is the domain to the left of
the curve y = x3 .

Domain Bounded Between Curves


Having curves in the xy-plane, would sometimes bound an area.
We draw the curves in xy-plane and calculate the intersections.

Example 4
Let f (x, y) be a function. Let D be bounded by y = x and y = x2 .

After drawing the curve, we see that the intersection is at (0, 0) and (1, 1).
We see that the min value of y lie on the curve y = x2 = m(x).
The max value of y is y = x = M (x).
Z Z x=1 ⇣ Z y=x ⌘
f (x, y) = f (x, y) dy dx
D x=0 y=x2

Example 5
Let D be bounded by y = x and y = x + 2, y = 0.

The max value of y is not a smooth curve due to the sharp point at x = 1.
We can split the integral in 2 pieces, with x 2 [0, 1], and x 2 [1, 2].
Z Z x=1 ⇣ Z y=x ⌘ Z x=2 ⇣ Z y= x+2 ⌘
f (x, y) = f (x, y) dy dx + f (x, y) dy dx
D x=0 y=0 x=1 y=0
We can also interchange the order of integration.

Author: Xiao Page 24


MATB41 15 REGULAR INTEGRAL IN 2D

Question 25
Let D be bounded by y = x and y = x + 2, y = 0.
Compute the integral by changing the order of integration.
Compare with example 5, notice that we no longer need to split up the
domain, because the sharp point is now the endpoint of the integral.

QuestionR 26
Integrate D f (x, y).
1. Let D be bounded by x = 0, y = 0, y = 2x + 2.
2. Let D be bounded by y = x + 1, y = x2 + 1.
3. Let D be the triangle defined by (0, 2), (0, -2), (2, 0).
4. Let D = {(x, y) 2 R2 | x 2 [0, 1], y 2 [0, 1], y > x2 }
Question 27
Evaluate Z y=1 ⇣Z x=y 4
sin x ⌘
y p dx dy
y=0 x=0 (1 x)
Strategy:
The function looks difficult. What is the region of integration?
Try changing the order of integration.

Question 28
Evaluate.
1. Z x=8 ⇣Z y=2
x ⌘
dy dx
x=0
p
y= 3 x y7 + 1
2. Z x=4 ⇣Z y=2 ⌘
p
y cos(y 4 + 3) dy dx
x=0 y= x
3. Z y=1 ⇣Z p
x= y
sin x ⌘
dx dy
y=0 x=y x
4. Z y=1 ⇣Z x=arccos y ⌘
e2 sin x dx dy
y=0 x=0

Author: Xiao Page 25


MATB41 16 REGULAR INTEGRAL IN 3D

16 Regular Integral in 3D
To compute integral of f (x, y, z) over a 3 dimensional domain D, it is similar.
1. Draw the domain in 3D. (This is difficult)
2. Take the projection of the 3D domain onto one of the 3 planes created
by the axis. This creates a 2D domain.
Suppose in this case, we take the projection onto the xy-plane.
3. Compute the limits of integration for the 2D domain in xy-plane
(as a 2D integral). These will be the limits of the outer most integrals.
Match dx and dy according to the order used.
4. Pick a random point (x, y) in the projection. Consider the vertical
line in the z direction. Consider the maximum and minimum values of z.
5. Repeat step 4 several times. The min and max values of z usually
will be on a smooth surface that depend on the point (x, y). Thus
zmin = m(x, y) and zmax = M (x, y).
These are the limits of the inner most integral, with dz.
Z Z Z z=M (x,y) !
f (x, y, z) = f (x, y, z) dz dx dy
D 2D z=m(x,y)

where the order of dxdy depends upon the order chosen for the 2D integral.
In general, inner limits can depend on variables from outer integrals, but
limits of outer integrals never depend on variables from inner integrals.

When an equation involves only 2 variables (in 3D), it gives a curve in 2D


which extends infinitely toward the 3rd dimension, giving a cylinder-like
surface in 3D. With 2 such curves, we will get at least 2 intersection points.
Connecting the intersection points with a ”smooth” curve typically gives the
picture of the domain in 3D.
Question 29
Let D be bounded by Rthe given equations, in the first octant.
Compute the integral D f (x, y, z) in 6 ways:
(3 ways to project , 2 ways to compute the 2D integral of the projection.)
1. z = 1 x, y = 1 2. z = 1 y, x = y 2
3. z = 1 x2 , y = 1 x 4. z = 1 y 2 , y = x
5. y 2 + z 2 = 9, y = 3x 6. 2x + y + 2z = 1

Author: Xiao Page 26


MATB41 17 VECTOR FUNCTION

17 Vector Function
So far, we have considered multi-variable functions.
The function takes in 2 (or 3) inputs, and gives out 1 output:

f (x, y) : R2 ! R

f (x, y, z) : R3 ! R
We can also have vector functions, which gives multiple outputs.
In general, we can have
f : Rm ! Rn
2 3
x1 2 3
6 x2 7 y1
6 7 6 y2 7
6 7 6 7
f 6 x3 7 = 6 .. 7
6 .. 7 4 . 5
4 . 5
yn
xm
It is called a vector function, since f have many outputs, so the outputs as
a whole can be regarded as a vector.

f (~x) = ~y

Each output, may depend on all of the inputs, and so each output is a
coordinate function that depends on all of the input variables.
In other words, a vector function is made up of many multi-variable functions.
2 3
x1 2 3 2 3
6 x2 7 f 1 (x 1 , x 2 , ..., x m ) y1
6 7 6 f2 (x1 , x2 , ..., xm ) 7 6 y2 7
6 7 6 7 6 7
f 6 x3 7 = 6 .. 7 = 6 .. 7
6 .. 7 4 . 5 4.5
4 . 5
fn (x1 , x2 , ..., xm ) yn
xm

This makes extending the definitions from multi-variable functions to


vector functions very easy:

A vector function f is integrable/continuous/di↵erentiable,


when all of its coordinate functions are integrable/continuous/di↵erentiable.

If we focus only within 3 dimensions, there are 2 new types of functions.

Author: Xiao Page 27


MATB41 17 VECTOR FUNCTION

Parametrization
When f goes from a lower dimensional space, to a higher dimensional space,
we sometimes call f a parametrization.

1D Parametrization
f (t) : R ! R2 OR f (t) : R ! R3
where we have used the variable t as the input of f .
Thus for each t, f (t) = (x, y) is a point in 2D, or f (t) = (x, y, z) in 3D.

If we view t as time, then we can view f (t) as the position of an object.


Thus f (t) describes the position of an object, as time goes on, which will
traverse out a curve.
Since a curve is a 1D object, this is called a 1D parametrization.

In fact, this is one of the most important ways to describe a curve in higher
dimensions, and finding a parametrization for a given curve is an important
skill in Calculus.

Example
Given a circle of radius R at the origin, we can choose the parametrization
to be:
f (t) = (R cos t, R sin t) = (x, y) t 2 [0, 2⇡]
where x = R cos t, y = R sin t in polar coordinates.
The parametrization starts at t = 0, which is the point (R, 0). Then as t
increases, f (t) traverses the circle counter-clockwise, as t = ✓ is the angle,
and back to (R, 0) when t = 2⇡.

However, parametrization is not unique. The same circle can be parametrized by:
f (t) = (R sin t, R cos t) = (x, y) t 2 [0, 2⇡]
where x = R cos t, y = R sin t is not the standard polar coordinates.
The parametrization starts at t = 0, which is the point (0, R). Then as t
increases, f (t) traverses the circle clockwise, since t is now the angle
between the point (x, y) and the positive y-axis.
We still traverse the whole circle, but this is the less natural parametrization.

When we need a parametrization of a curce, we typically choose the easiest one.

Author: Xiao Page 28


MATB41 17 VECTOR FUNCTION

2D Parametrization
f (u, v) : R2 ! R3
where we have used the variables u, v as the inputs of f .
Thus for each (u, v) in 2D, f (u, v) = (x, y, z) is a point in 3D.

Given a 2D region in the domain R2 (such as a square), f (u, v) will


create a surface in 3D.
Since a surface is a 2D object, this is called a 2D parametrization.

Going from left to right:


We may think that f lifts up a 2D region in R2 into 3D, and maybe stretches
the region somewhat, depending on the definition of f .

Going from right to left:


Given a surface S in 3D, we can try to find a parametrization for this surface
f (u, v) = (x, y, z). f pulls back this (curved) surface S in 3D with (x, y, z)
as variables, into a flat region in 2D with (u, v) as variables.

We know that an equality involving x, y, z gives a surface in 3D: g(x, y, z) = 0


This gives an alternative way of describing a surface in 3D. Finding a parametriza-
tion for a given surface is also an important skill in Calculus.

Example
If we consider the plane z = x + y + 10 that is above the unit square on the
xy-plane, we may choose the parametrization to be:
f (u, v) = (u, v, u + v + 10) = (x, y, z) u 2 [0, 1] v 2 [0, 1]
This is called the natural parametrization, as we have set u = x and v = y.
For any point (x, y) = (u, v) on the xy-plane, f takes in this point, and gives
out the same point with a new 3rd coordinate, or height, with the 3rd
component defined by f : z = u + v + 10.
This lifts up the unit square from the xy-plane, onto the plane z = x+y+10.
However, notice that the square is now slanted, and also stretched.
In this case, the plane z = x + y + 10 is flat, but in general, f lifts up a flat
region in 2D, into a curved surface in 3D, with some stretching.

In reverse, we can view f pulls back a (possibly complicted) surface in 3D,


(in this case z = x + y + 10), into a flat region in 2D (the unit square).

Author: Xiao Page 29


MATB41 17 VECTOR FUNCTION

Coordinate Transformation
When f goes between spaces of the same dimension, we sometimes call f a
coordinate transformation.
f (u, v) : R2 ! R2 OR f (u, v, w) : R3 ! R3
where f (u, v) = (x, y), and f (u, v, w) = (x, y, z).

Going from left to right:


Given a 2D region in the domain R2 (such as a square), f (u, v) = (x, y) will
produce another 2D region in R2 , with some stretching as defined by f .

Going from right to left:


Given a (possibly complicated) region in 2D with variables (x, y), we can try
to find a coordinate transformation f for this region, where f (u, v)
transforms a simple region in the uv-plane, onto the original region in
the xy-plane.
We have found a new set of coordinates (u,v), to describe the more
complicated region in the xy-plane.

Similarly, we can also do this for regions in 3D.

Example
Given a solid circle of radius R at the origin, x2 + y 2  R2 , it is a region in
2D. We know that representingp this region is complicated in (x, y), where we
get expressions such as y = R2 x2 .

However, we may consider polar coordinates:


f (r, ✓) = (r cos ✓, r sin ✓) = (x, y) r 2 [0, R] ✓ 2 [0, 2⇡]
where for each point (r, ✓) in the ”r ✓” plane, f (r, ✓) = (x, y) is a point in
the solid circle in the xy-plane.
In other words, f transforms a solid rectangle in the ”r ✓” plane, into
the solid circle in the xy-plane. Instead of working with a circle, we now
can work with a rectangle instead, as we have found better coordinates to
describe the original region in the xy-plane.

For f to turn a rectangle into a circle, it must stretch the rectangle at multiple
places, at di↵erent rates, which is based on the definition of f .

Author: Xiao Page 30


MATB41 18 DERIVATIVE MATRIX

18 Derivative Matrix
Given a vector function, for example f (x, y, z),
2 3 2 3
x f1 (x, y, z)
f 4y 5 = 4f2 (x, y, z)5
z f3 (x, y, z)
The derivative of a vector function is Df , called the derivative matrix.

For the first row of the matrix, we take the first coordinate function, and
take the partial derivative with respect to all possible variables (x, y, z) in
order, horizontally. Then for each coordinate function, we do the same thing
to get a new row of the matrix.
2 3
@ @ @
f f f
6 @x 1 @y 1 @z 1 7
6@ @ @ 7
6 7
Df = 6 f 2 f2 f2 7
6 @x @y @z 7
4@ @ @ 5
f3 f3 f3
@x @y @z
Relation to Gradient rf
For a multi-variable function f (x, y, z) : R3 ! R, it returns one value.
(For example, f (x, y, z) = xyz 2 )
@ @ @
Df = rf = ( f, f, f )
@x @y @z
where the derivative matrix has only 1 row.

On the other hand, for a vector function: f (x, y, z) = (f1 , f2 , f3 ), each row
of Df is the gradient of all the coordinate functions:
2 3
—rf1 —
Df = 4—rf2 —5
—rf3 —
where rfi are viewed as horizontal vectors in the matrix.

Question 30
f (x, y, z) = (3x + y, ey z, xyz)
First, state the dimension of the vector f takes in, and the dimension of the
vector f gives out. Then, compute Df .

Author: Xiao Page 31


MATB41 19 CHANGE OF VARIABLES FORMULA

19 Change of Variables Formula


Suppose you need to integrate f over a domain B.
But B is very bad. You don’t want to integrate over a bad domain.
Suppose you can find a function g such that g takes another region A onto B:
g:A!B g(A) = B
and somehow A is a good domain.
So, instead of integrating f over the bad domain B, you can integrate over
the good domain A instead.:
Z Z
f= f g · |Det(Dg )|
B A
f g means replacing the old variables of f in B with the new ones in A.
(We require that g to be a one-to-one transformation, and Det(Dg ) 6= 0.)
The natural question is of course: How do I find such g?

1. Polar Coordinates
Need to integrate f (x, y)

g(r, ✓) = (rcos✓, rsin✓) = (x, y)

f g = f (g(r, ✓)) = f (rcos✓, rsin✓) = f (x, y)


r 2 = x2 + y 2
Det(Dg ) = r

Useful when domain B is (partly) of circular shape, or function given is


already in polar coordinates.

Question 31
Integrate f (x, y) = x over D = {1  x2 + y 2  4} in the 2nd quadrant.
Question Z 32
+1 p
2
Show that e x dx = ⇡
1
Strategy: Z 1
x2 2
The normal distribution has no anti-derivative. Start with ( e ) and
1
write it as a product of 2 same things. Then change the variable in one
integral from x to y, and use the polar coordinate transformation.

Author: Xiao Page 32


MATB41 19 CHANGE OF VARIABLES FORMULA

Polar Curve
r represent the distance from the point (x, y) to the origin,
✓ represent the angle rotated counterclockwise from x-axis to the point.
Negative angles represent angles rotated clockwise.
Thus ✓ = ⇡ is the same as ✓ = ⇡.

If r is negative, it represent the point on the line with angle ✓ through


the origin, except on the opposite side of the line, with radius being |r|.
Example p p
The point (r, ✓) = (1, 3⇡/4) in polar coordinates represent (x, y) = ( 1/ 2, 1/ 2).
The point (r, ✓) = ( 1, 3⇡/4) in polar coordinates has r as negative. We
go on the line of 3⇡/4, which is toward the top left. Since r is negative,
we go on the opposite side of the line, and take the point with
p radius p being |r| = 1.
Thus ( 1, 3⇡/4) in polar coordinates represent (x, y) = (1/ 2, 1/ 2).

Using the above convention, we can define polar curve: r = h(✓).


To draw the polar curve, we pick some values of ✓, and plot the value of r
specified by r = h(✓) on the line represented by ✓.

r = a gives the circle with radius a centered at the origin.


r = acos✓ gives the circle to the right (if a > 0).
The circle is between the points (a, 0) and the origin. The radius is a/2.
r = bsin✓ gives the circle to the top (if b > 0).
The circle is between the points (0, b) and the origin. The radius is b/2.

Question 33
a) Find the area enclosed by polar curve h(✓), but ouside of polar curve k(✓):
h(✓) = r = 1 + cos(✓), k(✓) = r = 3cos(✓)
Strategy:
Plot the curve. Find the intersection between the polar curves. First choose
the min and max value of ✓ as the outer integral, then r would start from
the smaller value (inner polar curve) to the larger value (outer polar curve).

b) Sketch the curve r = 1 + 2cos✓ and find the area it encloses in the outer
loop but outside the inner loop.
Strategy:
This shape is complicated and it crosses itself. We can not apply the formula
directly , we need to use symmetry and subtract the inner area.

Author: Xiao Page 33


MATB41 19 CHANGE OF VARIABLES FORMULA

2. Cylindrical Coordinates
Need to integrate f (x, y, z)

g(r, ✓, z) = (rcos✓, rsin✓, z) = (x, y, z)

For any specific point (x, y, z)


it can be described by a radius, a height, and an angle:

r is the distance between the point and the z-axis.


✓ is the angle between the x-axis and the point (x, y, 0).
(x, y, 0) is the projection of the point onto the xy-plane.
z is the height.

f g = f (g(r, ✓, z)) = f (rcos✓, rsin✓, z) = f (x, y, z)


r 2 = x2 + y 2
Det(Dg ) = r

Useful when domain B is (partly) of cylindrical shape, or has circular projec-


tion, or function given is already in cylindrical coordinates, or the domain can
be obtained through a rotation with the defining equations containing x2 +y 2 .

Graphing surfaces attained through rotation


When the variables x and y in the defining equations can be written
completely with r2 = x2 + y 2 , this signals that the surface is attained
through rotation. We graph the 2D curves involving z and r in the r z
plane. We then revolve the curves around the z-axis to attain surfaces in 3D.
To integrate the volume bounded by the surfaces, we need to use cylindrical
coordinates and perform a 2D integration in the r z plane, and then we
add the ✓ integral on the outside.
Don’t forget to include the term det(Dg ) = r.
We only integrate the portion with r > 0.

Question 34 p
z = h(x, y) = a · ( (x2 + y 2 ) 1)2 a>0
Consider the volume:
inside the cylinder x2 + y 2 = 1, above z = 0, and below z = h(x, y)
Find a so that the volume is 1.

Author: Xiao Page 34


MATB41 19 CHANGE OF VARIABLES FORMULA

3. Spherical Coordinates
Need to integrate f (x, y, z)

g(⇢, , ✓) = (⇢sin cos✓, ⇢sin sin ✓, ⇢cos ) = (x, y, z)

For any specific point (x, y, z), it can be described by the radius and 2 angles:

⇢ is the distance from the point to the origin, as the radius.


is the angle between the z axis and the point (x, y, z).
This is called inclination angle.
✓ is the angle between the x-axis and the point (x, y, 0).
(x, y, 0) is the projection of the point onto the xy-plane.
This is called azimuthal angle.

Note: r and ⇢ both represent ’radius’ in Spherical and Cylindrical


Coordinate, but they stand for di↵erent radius.

f g = f (g(r, ✓, z)) = f (⇢sin cos✓, ⇢sin sin ✓, ⇢cos )

⇢2 = x 2 + y 2 + z 2
Det(Dg ) = ⇢2 sin

Useful when domain B is (partly) of spherical shape, or function given is


already in spherical coordinates.

The maximum value of is 180 degrees, at most to the negative z-axis.


For a full sphere of radius R, the range of the variables are:
⇢:0!R
:0!⇡
✓ : 0 ! 2⇡
If the sphere is not centered at the origin, spherical coordinates usually is
not the choice.

Question 35
Find the 3D-volume of a solid hemisphere in R3 with radius R and x > 0.

Strategy:
Recall integrating the function f = 1 gives the volume of the domain.

Author: Xiao Page 35


MATB41 19 CHANGE OF VARIABLES FORMULA

Question 36 p p p
Consider the surface A characterized by z = 3 (x2 + y 2 ) = 3 · r, where
r is the radius in cylindrical coordinates.
Consider the surface B as the sphere of radius 2, x2 + y 2 + z 2 = 4.
Find the volume bounded by surface A and B in 2 ways:

a) Use Spherical Coordinates.


Strategy:
Plot the curve in the r z plane, and rotate to a 3D volumne.
The volume looks like an ice-cream cone. What is the maximum inclination
angle ? Remember that r in both coordinates are not the same.

b) Use Cylindrical Coordinates. Which method do you think is more suited?


Strategy:
Solve the intersection of the surfaces by pluging a quantity of one equation
into the other. Perform a 2D integral in r z plane, and add the ✓ integral
on the outside.
p
c) Now consider surface C as the plane z = 3. What is the volume be-
tween surface A and C? Which method should you use?

Question 37 p
Compute the volume above the cone z = x2 + y 2 and below the sphere
x2 + y 2 + z 2 = 2z.

a) Use Cylindrical Coordinates.


Strategy:
Why is it a sphere?
Plot the curve on the r z plane. For the sphere, complete the square in z.
Find the intersection and perform a 2D integral in r z plane.

b) Use Spherical Coordinates. Which way do you prefer?

Strategy:
Since the sphere is not centered at the origin, the range of ⇢ is NOT 0 to 1.
Use the defining equation x2 + y 2 + z 2 = 2z and switch variables to spherical
coordinates to attain ⇢ = 2cos( ). Notice ⇢ varies in terms of the value in
3D, similar to polar curves in 2D, so we must put it as the innermost integral.

Author: Xiao Page 36


MATB41 19 CHANGE OF VARIABLES FORMULA

4. Ellipse Transformation
To integrate over an ellipse, we write the equation of ellipse as
⇣ x ⌘2 ⇣ y ⌘2
+ =1
a b
Similar to polar coordinates, we use

g(r, ✓) = a · rcos(✓), b · rsin(✓)

Det(Dg ) = abr
Since the stretching factors of the ellipse are included in function g, the range
of radius is always r : 0 ! 1.

✓ does not exactly represent the angle from polar coordinates anymore.
⇡ 3⇡
It still works as normal for these angles: ✓ = 0, , ⇡, , 2⇡...
2 2
A full ellipse is still ✓ : 0 ! 2⇡.

Useful when a 2D domain is an ellipse, or if the 3D projection of a


volume is of ellipse shape.

Question 38
Verify Det(Dg ) = abr.
Then find the area of the ellipse given by the standard equation above.

Question 39
Integrate f (x, y) = sin(2x2 + 4y 2 ) over the region bounded by 2x2 + 4y 2 = 4.

Question 40 p
Compute the volume above z = 2x2 + y 2 and below z = 2.

Strategy:
This surface is not given by a circular rotation, with r2 = x2 + y 2 .
However, it is similar. We may take the ”radius” to be r2 = 2x2 + y 2 , so
we can still plot on the r z plane and do an ellipse rotation. Taking
the projection of the volume onto the xy-plane gives an ellipse. Instead of
integrating the 2D projection integral directly, we first perform the inner z
integral, and use ellipse transformation onto the 2D integral.

Author: Xiao Page 37


MATB41 19 CHANGE OF VARIABLES FORMULA

5. Parallelogram Transformation
Given a parallelogram in R2 with one vertex being origin.
It is generated by 2 points/vectors, ~a and ~b.
We can turn the unit square into parallelogram by requiring that
 
1 0
g = ~a g = ~b
0 1
Since the parallelogram is linear, this is a linear transformation.
It is characterized by a matrix.
  
u u x
g =A· =
v v y
Due to the contraint on the unit vectors above, A must be
2 3
| |
4
A = ~a ~b5
| |
If the paralellogram does not have one vertex at origin, we can perform a
shift of the parallelogram, by subtracting the coordinates of one point from
all points of the parallelogram. We take note of ~a and ~b that generate the
shifted parallelogram. The transformation g would take 2 steps. First the
transformation of the square to the shifted parallelogram, then from the
shifted parallelogram to the original one.

Example
Given the parallelogram (1, 1), (2, 1), (2, 2), (3, 2) which is a
standard 45 degrees parallelogram that is shifted.  
1 1
The standard 45 degrees parallelogram is defined by the vectors and .
0 1
We can use the above transformation, to turn the unit square into the
standard parallelogram:
  
u 1 1 u
h = ·
v 0 1 v
Then we need to shift the parallelogram upward and right by 1 unit. Thus
   
u 1 1 u 1
g = · +
v 0 1 v 1
In both the non-shifted and the shifted case, Dg = A.

Author: Xiao Page 38


MATB41 19 CHANGE OF VARIABLES FORMULA

6. Tetrahedron Transformation
Generalizing the idea of parallelogram, the tetrahedron transformation is
much more useful in practice.

Given a tetrahedron in R3 with one vectex being origin.


It is characterized by 3 points/vectors, ~a, ~b, ~c.
We can turn the unit tetrahedron into the tetrahedron given by requiring
that 2 3 2 3 2 3
1 0 0
g 405 = ~a g 415 = ~b g 405 = ~c
0 0 1
If we define the names of the variables of g as g(u, v, w), then our domain of
integration is the unit tetrahedron, given by
Z w=1 Z v=1 w Z u=1 v w
f g · |Det(Dg )| du dv dw
w=0 v=0 u=0

Of course, g is again a linear transformation.


2 3 2 3 2 3
u u x
g v = A · v = y5
4 5 4 5 4
w w z
where the matrix A, by the constraints above, is
2 3
| | |
A = 4~a ~b ~c5
| | |
If the tetrahedron does not have one vertex at origin, we perform a shift
similar to parallelogram. Same as the parallelogram transformation, in both
the non-shifted and the shifted case, Dg = A.

Question 41
Find the volume bounded by x + 2y + z = 2, x = 2y, x = 0, z = 0.
Strategy:
(Take any 3 planes, they intersect at 1 point. 4 points form a tetrahedron.)

Question 42 R
Compute the integral D z x y where D is the tetrahedron given by
(0, 0, 0), (1, 2, 3), (0, 2, 2), ( 1, 1, 1).

Author: Xiao Page 39


MATB41 19 CHANGE OF VARIABLES FORMULA

7. Domain Bounded by 2 Expressions


Sometimes, a 2D domain B is defined by the region bounded by some curves
given by: f1 (x, y) = a, f1 (x, y) = b, f2 (x, y) = c, f2 (x, y) = d.

If the curves only intersect at 4 points, we can transform the square onto
the domain by using   
x f1 (x, y) u
h = =
y f2 (x, y) v

u
and since u = f1 (x, y) and v = f2 (x, y), the bounds of is exactly the
v
square A = {u 2 (a, b), v 2 (c, d)}.

However, the direction of h is opposite to the direction neeeded in the


change of variable formula. To use the change of variable formula, we
technically should compute the inverse of h, giving
 
u x
g =
v y
However, computing inverse is sometimes difficult.
Since we need det(Dg ) in the formula, and g = h 1 , we can instead use
1
det(Dg ) = det(Dh 1 ) = det((Dh ) 1 ) = det(Dh ) 1
=
det(Dh )
Thus we can simply compute Dh which is given, and use it in the change of
variable formula without needing to compute the inverse.
(We still need to compute the inverse sometimes.)
Example
Let B given by xy = 1, xy = 2, xy 2 = 1, xy 2 = 2. Then set u = xy, v = xy 2 .

y x
Dh = 2 ) det(Dh ) = 2xy 2 xy 2 = xy 2 = v
y 2xy
Suppose we want to integrate f (x, y) = y, then this correspond to y = v/u.
Z Z Z 2 Z 2
v 1 1
y= ·| |= dv du
B A u v u=1 v=1 u
Question
R 43
1. RD (x + y)(x y) over D bounded by x y = 0, x y = 2, x + y = 1, x + y = 2.
2. D sin 2x y
2x+y
over D bounded by x = 0, y = 0, 2x + y = 1.

Author: Xiao Page 40


MATB41 20 CHAIN RULE IN 1D

20 Chain Rule in 1D
We typically think of chain rule as a rule for computation:
0
f g(x) = f 0 g(x) · g 0 (x)

where we are given a function inside another function such as y = sin(ex ),


and we take the derivative of the outside (not touching the inside), then
multiply the derivative of the inside.

Alternatively, we can define a variable dependence.


For example, define the variables as y = y(t), and t = t(x).
(In the above example, we would set y = sin t and t = ex .)
Then we may express the chain rule (in 1D) as:
dy dy dt
=
dx dt dx
dy
Intuitively, asks how much would y change, if there is a small change in x.
dx
Given the variable dependence above, a small change in x would cause a small
change in t, which in turn would cause a small change in y.
Hence, we have the 2 ”fractions” on the right.

2nd Derivative
We may use the above form of the chain rule to calculate the 2nd derivative.
d dy
We apply another di↵erential operator on top of :
dx dx
d2 y d ⇣ dy ⌘
=
dx2 dx dx
Plug in the above form of the chain rule,
d ⇣ dy dt ⌘
=
dx dt dx
We want to take the derivative of this product, so we need product rule,
d ⇣ dy ⌘ dt dy d ⇣ dt ⌘
= · + ·
dx dt dx dt dx dx
Notice that there are 2 types of derivatives here.

Author: Xiao Page 41


MATB41 20 CHAIN RULE IN 1D

d ⇣ dt ⌘
The 2nd term involves the expression .
dx dx x
Recall that we assumed t = t(x). (such as t = e )
dt
So is also a function of x.
dx
We want to take the derivative of such function against x.
This results in the 2nd derivative of t = t(x) against x:
d ⇣ dt ⌘ d2 t
= 2
dx dx dx
d dy ⌘

The 1st term involves the expression .
dx dt
This term is more difficult. (and requires an additional expansion)
Recall that we assumed y = y(t). (such as y = sin t)
dy dy
So is also a function of t. (such as = cos t)
dt dt
We want to take the derivative of such function against x, on an expression
involving t.
dy
Notice that is an expression involving t, similar to y = y(t).
dt
So to take the derivative against x, on an expression involving t, we know
that for y = y(t):
d⇣ ⌘ dy dy dt d⇣ ⌘ dt
y(t) = = = y(t)
dx dx dt dx dt dx
dy
Since is also an expression involving t, it would behave in the same
dt
way as y(t) when taking derivative against x.
dy
We can substitue y(t) with in the above expression:
dt
d ⇣ dy ⌘ d ⇣ dy ⌘ dt
=
dx dt dt dt dx
dy
Since y = y(t) is a function of t, on the right is also a function of t.
dt
We want to take the derivative of such function against t.
This results in the 2nd derivative of y = y(t) against t:
d ⇣ dy ⌘ d2 y
= 2
dt dt dt

Author: Xiao Page 42


MATB41 20 CHAIN RULE IN 1D

In conclusion, to take the 2nd derivative involving chain rule:


d2 y d ⇣ dy ⌘ d ⇣ dy dt ⌘ d ⇣ dy ⌘ dt dy d ⇣ dt ⌘
= = = · + ·
dx2 dx dx dx dt dx dx dt dx dt dx dx
Taking an additional expansion in the first term:

d ⇣ dy ⌘ dt dt dy d2 t
= · + ·
dt dt dx dx dt dx2
d2 y dt dt dy d2 t
= 2 · + ·
dt dx dx dt dx2
dt
Notice that we get a square of in the first term.
dx
Thus the chain rule formula for 2nd derivative:
d2 y d2 y ⇣ dt ⌘2 dy d2 t
= 2 + ·
dx2 dt dx dt dx2
⇣ dt ⌘2 d2 t
Note: 6= 2 .
dx dx
Of course, given a function y = y(x), one would not go through this long
and complicated process just to take 2nd derivative. It is much easier to just
take the derivative (twice) using the computation rules.

However, this formula shines when at least one of the function is


unknown. This is exactly the case for ODE, where y = y(x) is unknown.
We may use this formula to carry out a change of variables and turn a
complicated ODE into a simple one.

ODE - Euler Equation


Consider y = y(x). For numbers a, b, c,
ax2 y 00 + bxy 0 + cy = 0
This is a 2nd order linear equation, but with non-constant coefficients.
In general, this type of equation has no solution. But here, notice that we
have exactly x2 in front of y 00 , and x in front of y 0 .
Due to this, we can guess the solution to be y(x) = xn .
We plug in y(x) = xn into the ODE, and solve for n.

Typically, there will be 2 solutions for n. The general solution will be a


linear combination of the 2 solutions (with constants c1 , c2 in front).

Author: Xiao Page 43


MATB41 20 CHAIN RULE IN 1D

Example

x2 y 00 xy 0 3y = 0
Take y = xn , then y 0 = nxn 1
and y 00 = n(n 1)xn 2 .
Plugging into the ODE,

x2 n(n 1)xn 2
xnxn 1
3xn = 0

Notice that we can divide both sides by xn .

n(n 1) n 3=0

n2 2n 3=0
We will always get a quadratic of this form. In this case, we get

(n 3)(n + 1) = 0

Thus the general solution is given by

y(x) = c1 x3 + c2 x 1

However, notice that given an equation with di↵erent numbers, the quadratic
we get may have only one (repeated) solution. In this case, we would be
”missing” the second solution. We can instead find the solution by using the
chain rule formula for 2nd derivative.

Example
x2 y 00 3xy 0 + 4y = 0
If we proceed using the previous approach, we would get (n 2)2 = 0, giving
one (repeated) solution.

Instead, we can perform a change of variable, t = ln x.


We get
dy dy dt dy 1
y0 = = =
dx dt dx dt x
2 ⇣ ⌘
d y dt 2 dy d t 2
d y ⇣ 1 ⌘2 dy
2
1
y 00 = 2 + · = + ·
dt dx dt dx2 dt2 x dt x2

Author: Xiao Page 44


MATB41 20 CHAIN RULE IN 1D

Plugging the above into the ODE,

x2 y 003xy 0 + 4y = 0
⇣ d2 y ⇣ 1 ⌘2 dy 1⌘ ⇣ dy 1 ⌘
x2 + · 3x + 4y = 0
dt2 x dt x2 dt x
Notice that all the x cancels out.
d2 y dy dy
3 + 4y = 0
dt2 dt dt
d2 y dy
4 + 4y = 0
dt2 dt
This is a 2nd order linear equation with constant coefficients.
The solution is given by y(t) = ert .
Plugging into the equation,

r2 4r + 4 = 0

(r 2)2 = 0
This is a repeated root solution.
Thus the general solution for y(t) is given by,

y(t) = c1 e2t + c2 te2t

Since t = ln x, we get the general solution for y(x) is given by,


⇣ ⌘2 ⇣ ⌘2
y(x) = c1 e2 ln x + c2 ln xe2 ln x = c1 eln x + c2 ln x eln x

y(x) = c1 x2 + c2 (ln x)x2


In other words, the ”missing” second solution from the repeated root of
n = 2, is attained by multiplying an extra factor of ln x in front of xn .

Question 44
Find the general solution to the ODE, where y = y(x).
1. x2 y 00 3xy 0 + 4y = 0
2. 2x2 y 00 5xy 0 + 5y = 0
3. x(1 x2 )2 y 00 (1 x2 )(1 + 4x2 )y 0 + 2x3 y = 0
1
with the change of variable t = ln(1 x2 ).
2

Author: Xiao Page 45


MATB41 21 CHAIN RULE IN 2D AND 3D

21 Chain Rule in 2D and 3D


Consider an example: z = z(x, y), and x = x(s, t), y = y(s, t).
We can construct the tree diagram of variable dependence.

We start with the variable z on the top, and for each variable that z depends
on, we get an extra branch below. In this case, z = z(x, y) and we get 2
branches for x and y. The 2 branches correspond to the partial derivatives
@z @z
and as we will see.
@x @y
Similarly, we create branches for x and y, and get 4 branches for s and t. The
@x @x @y @y
branches correspond to the partial derivatives , , , .
@s @t @s @t
@z
To compute :
@s
Using the diagram, we start at z, and consider all possible ways to reach s.
The first way is, start from z, go to x, then from x to s.
This gives (using multiply)
@z @x
·
@x @s
The second way is, start from z, go to y, then from y to s.
This gives (using multiply)
@z @y
·
@y @s
The answer is the 2 ways added together
@z @z @x @z @y
= · + ·
@s @x @s @y @s
Similarly,
@z @z @x @z @y
= · + ·
@t @x @t @y @t

Author: Xiao Page 46


MATB41 21 CHAIN RULE IN 2D AND 3D

@z
Intuitively, asks how much would z change, if there is a small change in s.
@s
Given the variable dependence above, a small change in s would cause both
a small change in x and a small change in y. The small change in x would
in turn cause a small change in z, and the the small change in y would also
cause a small change in z. Thus the ”total” small change in z would be the
sum of the 2 changes caused by x and y.
Hence, we get the sum of 2 pairs of ”fractions” multiplied on the right.

Of course, given the functions z = z(x, y) and x = x(s, t), y = y(s, t), one
does not need go through this long and complicated process just to take the
derivative. It is much easier to just plug in the expression of x and y into z,
and take the derivative using the computation rules.

Question 45
@z
1. z = x y, x = sin(t), y = 3t. Find .
@t
@z @z
2. z = xy 2 , x = s cos(t), y = s sin(t). Find , .
@s @t
@u @u
3. u = xy + yz, x = s + t, y = s, z = 2t. Find , .
@s @t
Similar to the Chain Rule in 1D, the chain rule formula shines when at least
one of the function is unknown.

Question 46
Suppose that z = f (x, y), x = 2s + t and y = s 3t. Assume:
f (1, 1) = 2, f (3, 2) = 1, zx (1, 1) = 2, zx (3, 2) = 4, zy (1, 1) = 6, zy (3, 2) = 1.
Find zs and zt at (s, t) = (1, 1).

PDE - Partial Di↵erential Equation


A Partial Di↵erential Equation (PDE) is an equation, which implicitly
defines a function u(x, y) by mixing u, x, y, and (partial) derivatives of u.

Similar to ODE, the function u(x, y) defined by an PDE often can not be
solved explicitly. Di↵erent from an ODE, the general solution of a PDE
would involves some unknown function, instead of unknown constants.

Author: Xiao Page 47


MATB41 21 CHAIN RULE IN 2D AND 3D

First Order Linear Equation with Constant Coefficients


Consider u = u(x, y). Consider the example:

2ux + 3uy = 0

We perform the change of variables: s = 2x + 3y and t = 3x 2y.


(This change of variable works in general: a = 2, and b = 3 in this case.)
Using chain rule, where we think of u as u = u(s, t), s = s(x, y), t = t(x, y)
@u @u @s @u @t @u @u
= · + · = 2+ 3
@x @s @x @t @x @s @t
@u @u @s @u @t @u @u
= · + · = 3+ ( 2)
@y @s @y @t @y @s @t
Plugging into the PDE, we get

2ux + 3uy = 0
⇣ @u @u ⌘ ⇣ @u @u ⌘
2 2+ 3 +3 3+ ( 2) = 0
@s @t @s @t
@u
Notice that the terms involving cancels.
@t
@u
(22 + 32 ) =0
@s
Thus the partial derivative of u(s, t) against s is 0.
However, since we are in 2D, this does not mean u is constant.
@u
For example, u(s, t) = t2 , would have = 0.
@s
@u
Thus, = 0 implies that u must not depend on s, but can still depend on t.
@s
To signal this dependence, we get

u(s, t) = f (t)

where f is a 1D function only depending on t.

Since t = 3x 2y, the general solution of u is given by:

u(x, y) = f (3x 2y)

Notice that the general solution involves an unknown function f .

Author: Xiao Page 48


MATB41 21 CHAIN RULE IN 2D AND 3D

Initial Condition
Indeed, one may check that u(x, y) = f (3x 2y) satisfy the original PDE,
without needing to know what the unknown function f is.

However, one may impose an initial condition on u, which would allow


one to solve for f . This is analogous to imposing an initial condition on an
ODE, which would solve for the constants.

Suppose in the above example, we impose the initial condition: u(x, 0) = x2 .


Then since the general solution is u(x, y) = f (3x 2y):
u(x, 0) = f (3x) = x2
w
We may solve for f by taking a substitution w = 3x, so x = .
3
w2
f (w) =
9
Thus, the particular solution is
(3x 2y)2
u(x, y) = f (3x 2y) =
9
General Case
For the equation aux + buy = 0: Set s = ax + by and t = bx ay
The general solution is u(x, y) = f (bx ay).

Question 47
Consider u = u(x, y). Find the solution to the PDE (general or particular).
1. ux + uy = 0 with u(x, 0) = x.
2. aux + buy = 0
3. ux + uy = 1
4. ux uy + u = 0.
5. ux + uy + u = ex+2y with u(x, 0) = 0
6. ux + 2uy + (2x y)u = 2x2 + 3xy 2y 2

Question 48
Consider 2ux + yuy = 0.
(First order linear equation with non-constant coefficients: a = 2, b = y)
Find the solution with the change of variables: s = x, t = e x/2 y.
dy y b
(C = e x/2 y is the solution to the ODE: = = )
dx 2 a

Author: Xiao Page 49

You might also like