You are on page 1of 181

Multivariable Calculus Notes

David A. Santos

Philadelphia, PA October 22, 2003

ii

Contents

1 Brief Review of Linear Algebra 1.1 Rn . . . . . . . . . . . . . . . . . 1.2 Mmn(R) . . . . . . . . . . . . . 1.3 Determinants . . . . . . . . . . . 1.4 Linear Transformations . . . . . . 1.5 Inner Products . . . . . . . . . . 1.6 Cross product in R3 . . . . . . . 1.7 Lines and Planes in R3 . . . . . . 1.8 Topology of Rn . . . . . . . . . . 1.9 Quadratic Forms . . . . . . . . . 1.10 Quadratic Surfaces . . . . . . . . 1.11 Canonical Surfaces in R3 . . . . 1.12 Parametric Curves and Surfaces 1.13 Frenet-Serret Formul . . . . . . 1.14 Limits . . . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

1 1 7 13 16 22 27 31 37 39 41 44 49 54 56 59 59 67 73 81 85 93 97

2 Differentiation 2.1 Local Study of Functions . . . . . . . . . . . . 2.2 Denition of the Derivative . . . . . . . . . . . 2.3 The Jacobi Matrix . . . . . . . . . . . . . . . 2.4 Gradients and Directional Derivatives . . . . 2.5 Extrema . . . . . . . . . . . . . . . . . . . . . 2.6 Lagrange Multipliers . . . . . . . . . . . . . . 2.7 Arithmetic Mean-Geometric Mean Inequality . iii

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

iv 3 Integration 3.1 Differential Forms . . . . . . . . . . . . . 3.2 Integrating in 0(Rn) . . . . . . . . . . 3.3 Integrating in 1(Rn) . . . . . . . . . . 3.4 Closed and Exact Forms . . . . . . . . . 3.5 Integrating in 2(R2) . . . . . . . . . . . 3.6 Change of Variables in 2(R2) . . . . . 3.7 Change to Polar Co-ordinates . . . . . . 3.8 Integrating in 3(R3) . . . . . . . . . . . 3.9 Change of Variables in 3(R3) . . . . . 3.10 Integration in 2(R3) . . . . . . . . . . . 3.11 Greens, Stokes, and Gau Theorems

CONTENTS 105 105 111 112 122 127 141 149 157 160 165 171

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

Chapter

Brief Review of Linear Algebra


1.1 Rn

Figure 1.1: A point in R2 .


a= a1 a2
Figure 1.2: A bi-point in R2 .

These n-tuples are called points.

1 Denition Rn is the set of real n-tuples x 1 x 2 n , xk R . R = x:x= . . . xn 1

Chapter 1

Thus R2 is the collection of points on the plane (see gure 1.1) and R3 is the collection of points in three-dimensional space. Points on their own are very boring entities. They are devoid of arithmetical properties (you cannot add two points), they are simply a list of real numbers. Suppose now that we are given two points x and y in Rn. Starting from x we move on a straight line towards y, and we denote this ordered movement by the notation [x, y] (read the bi-point x, y). This movement involves a displacement of each of the n co-ordinates, the k-th co-ordinate being displaced yk xk units. If we let ak = yk xk record the displacement of the k-th co-ordinate, then we say that [x, y] is a representative of the vector a1 a2 a = . . . . an Notice that there are innitely many bi-points representing the same vector. Thus in gure 1.2 we see a representative bi-point of a vector in R2. The same vector would represent any parallel displacement of this bi-point. 2 Example Consider the points x1 = 1 , y1 = 2 3 , x2 = 4 3 , y2 = 5 5 . 1

Though the bi-points [x1 , y1 ] and [x2 , y2 ] are in different locations on the plane, they represent the same vector a = 31 53 = . 4 2 1 5

These two bi-points are parallel and have the same length, moreover, they are pointing in the same direction. 3 Denition A vector a1 a2 a = . Rn . . an

Rn

is an equivalence class denoting all those bi-points whose displacement in the k-th co-ordinate is ak.

! We write points of R

4 Denition If a and b are two vectors in Rn their vector sum a + b is dened by the co-ordinatewise addition a1 + b1 a2 + b2 (1.1) a + b = . . . . an + bn

We write vectors in Rn using bold-face letters with arrows on top and square brackets, as in x1 x2 x = . . . . xn

using bold-face letters and parentheses, as in x1 x2 x = . . . . xn

5 Denition A real number R will be called a scalar. If R and a Rn we dene scalar multiplication of a vector and a scalar by the co-ordinatewise multiplication a1 a2 a = . (1.2) . . . an 6 Theorem The operations of vector addition 1.1 and scalar multiplication 1.2 make Rn a vector space. That is, these operations satisfy ( a , b , c ) (Rn)3, (, ) R2,

4 Closure under vector addition: a + b Rn a Rn

Chapter 1

(1.3)

Closure under scalar multiplication:

(1.4)

Commutativity of addition:

Associativity:

a + b = b + a (a + b)+ c = a +(b + c)

(1.5)

(1.6)

Existence of additive identity:

Existence of additive inverses:

0 Rn : a + 0 = 0 + a = a

(1.7)

Distributive Law:

a Rn : a + ( a ) = a + a = 0 ( a + b ) = a + b ( + ) a = a + a 1a = a

(1.8)

(1.9)

Distributive Law:

(1.10)

(1.11)

() a = ( a )

(1.12)

Rn

7 Denition A set of vectors a k Rn, 1 k l is said to be linearly independent if


l

k=1

8 Denition An ordered set

k a k = 0 = 1 = 2 = = l = 0. A = { v 1, v 2, . . . , v n}

of n linearly independent vectors in Rn is called an ordered basis for Rn. This means that every vector in Rn can be uniquely written as a linear combination of the v k, that is, if u Rn then there exist unique scalars k (called the co-ordinates of v under A ) such that 1 n 2 u = k v k := . . . . k=1 n A 9 Example The family A = { i , j , k } with 1 0 0 i = 0 , j = 1 , k = 0 0 0 1

forms an ordered basis for R3 (these is called the natural basis for R3). Any vector u can be written uniquely as a linear combination of these vectors, for example 2 3 = 2 i 3 j + 4 k . 4 A The family B = { i , k , j } forms a different ordered basis for R3. In this case 2 3 = 2 i + 4 j 3 k . 4 B

Chapter 1

! In most cases we will be using the standard ordered basis


with A = { e 1, e 2, . . . , e n, } 0 . . . ek= 1 . . . 0

(a 1 in the k slot and 0s everywhere else). In such cases we will write


n k=1

1 2 k e k = . . . n

(without the A subscript) rather than


n k=1

1 2 k e k = . . . . n
A

10 Example Prove that the family C = { b 1, b 2, b 3} with 1 1 1 b 1 = 0 , b 2 = 1 , b 3 = 1 0 0 1

forms an ordered basis for R3 and nd the co-ordinates of 2 3 4 under C .

Mmn(R) Solution: We have 1 + 2 + 3 = 0 0 2 + 3 = 0 1 b 1 + 2 b 2 + 3 b 3 = 0 3 = 0. 0

Solving this triangular system gives 1 = 2 = 3 = 0, whence C is a linearly independent family of 3 vectors and hence an ordered basis for R3. It is easy to verify that 2 1 1 1 5 3 = 5 0 7 1 + 4 1 = 5 b 1 7 b 2 + 4 b 3 = 7 . 4 0 0 1 4 C

1.2

Mmn(R)

11 Denition An m n (m by n) matrix A with m rows and n columns is a rectangular array of the form a11 a12 a1n a21 a22 a2n A= . , . . . . . . . . am1 am2 amn where (i, j) {1, 2, . . . , m} {1, 2, . . . , n}, aij R.

! As a shortcut, we often use the notation A = [a ] to denote the matrix


ij

A with entries aij. Notice that when we refer to the matrix we put square brackets (as in [aij]), and when we refer to a specic entry we do not use the surrounding parentheses (as in aij).

12 Example A= is a 2 3 matrix and 0 1 1 1 2 3

is a 3 2 matrix

2 1 B = 1 2 0 3

Chapter 1

13 Denition We denote by Mmn(R) the set of all m n matrices with real number entries. If m = n we use the abbreviated notation Mn(R) = Mnn(R). Mn(R) is thus the set of all square matrices of size n with real entries. 14 Denition The n n zero matrix 0n everywhere, 0 0 0 0 0 0 0n = 0 0 0 . . . . . . . . . 0 0 0 Mn(R) is the matrix with 0s 0 0 0 . . . . 0

15 Denition The n n identity matrix In Mn(R) is the matrix with 1s on the main diagonal and 0s everywhere else, 1 0 0 0 0 1 0 0 In = 0 0 1 0 . . . . . . . . . . . . . 0 0 0 1 16 Denition The main diagonal of a matrix matrix A = [aij] Mmn(R) is the set {aii|i min(m, n)}. A matrix is diagonal if every entry off its main diagonal is 0. 17 Denition A Mmn(R) is said to be upper triangular if ((i, j) {1, 2, , n}2), (i > j, aij = 0), that is, every element below the main diagonal is 0. Similarly, A is lower triangular if ((i, j) {1, 2, , n}2), (i < j, aij = 0), that is, every element above the main diagonal is 0.

Mmn(R) 18 Example The matrix A M34(R) shewn M4(R) is lower triangular. 1 1 a b c 1 A = 0 2 3 0 B = 0 0 0 0 1 1

9 is upper triangular and B 0 a 2 1 0 0 3 t 0 0 0 1

19 Denition Let A = [aij] Mmn(R), B = [bij] Mmn(R) and R. The matrix A + B is the matrix C = [cij] Mmn(R) with entries cij = aij + bij. 20 Example Let Then

a 2a c 1 2a c a b , N = a b a b . M= 0 a+b 0 1 ab 0 1 2a 4a 2c 2a 2b . 2M = 0 2a + 2b 0 2

a+1 0 2c b 2a 0 , M+N= a 2a 0 2

21 Denition Let A = [aij] Mmn(R) and B = [bij] Mnp(F). Then the matrix product AB is dened as the matrix C = [cij] with entries cij =
n

ailblj.
l=1

In order to obtain the ij-th entry of the matrix AB we multiply elementwise the i-th row of A by the j-th column of B. Observe that AB is a m p matrix.

! Observe that we use juxtaposition rather than a special symbol to denote matrix multiplication. This will simplify notation.

10 22 Example Let 1 2a a 2a c M= , N = a b 0 0 a b 1 2
2

Chapter 1

be matrices over R.Then

a 2a 2a2 c + 2ab a 2a(a b) + c 2a + 2c , NM = a(a b) 2a(a b) (a b)c . MN = a(a b) + b 2b a 4a c + 2b

! Matrix multiplication is not necessarily commutative.


1 3 1 3 1 3

23 Example Solution: We have 1 3 1 3 1 3 over R.

1 1 1 2 3 3 3 3 0 0 0 1 1 2 1 = 0 0 0 , 3 3 3 3 0 0 0 1 1 1 2 3 3 3 3

! Observe then that the product of two non-zero matrices may be the
zero matrix.

24 Theorem If (A, B, C) Mmn(F) Mnr(F) Mrs(F) we have (AB)C = A(BC), i.e., matrix multiplication is associative. Proof To shew this we only need to consider the ij-th entry of each side. Both are equal to
n r

aikbkk ck j.
k=1 k =1

Mmn(R)

11

Even though matrix multiplication is not necessarily commutative, it is associative.

that is, if A Mn(R), and (r, s) N2, then (Ar)(As) = (As)(Ar) = Ar+s.

! By virtue of associativity, a square matrix commutes with its powers,

25 Denition Let A = [aij] Mn(R). Then the trace of A, denoted by tr (A) is the sum of the diagonal elements of A, that is
n

tr (A) =
k=1

akk.

26 Theorem Let A = [aij] Mn(R), B = [bij] Mn(R), R. Then tr (A) = tr (A) , tr (A + B) = tr (A) + tr (B) , tr (AB) = tr (BA) . (1.13) (1.14) (1.15)

Proof The rst assertion is trivial. To prove the second, observe that AB =
n n

[
k=1

aikbkj] and BA = [
k=1 n

bikakj]. Then
n n n

tr (AB) =
i=1 k=1

aikbki =
k=1 i=1

bkiaik = tr (BA) , u

whence the theorem follows. 27 Example Write 1 2 3 A = 2 3 1 M3(R) 3 1 2

as the sum of two 3 3 matrices E1, E2, with tr (E2) = 10.

12 Solution: There are innitely many solutions. Here is one: 9 2 3 10 0 0 1 2 3 A = 2 3 1 = 2 3 1 + 0 0 0 . 3 1 2 3 1 2 0 0 0

Chapter 1

28 Example Given a square matrix A M4(R) such that tr A2 = 4, and (A I4)2 = 3I4, nd tr (A). Solution: tr (A I4)2 = = = = tr A2 2A + I4 tr A2 2tr (A) + tr (I4) 4 2tr (A) + 4 2tr (A) ,

and tr (3I4) = 12. Hence 2tr (A) = 12 or tr (A) = 6. 29 Denition The transpose of a matrix of a matrix A = [aij] Mmn(R) is the matrix AT = B = [bij] Mnm(F), where bij = aji. 30 Example We have a 2a c a 0 a+b a b , MT = 2a a 0 . M= 0 a+b 0 1 c b 1 31 Theorem Let A = [aij] Mmn(R), B = [bij] Mmn(R), C = [cij] Mnr(R), R. Then ATT = A, (A + B)T = AT + BT , (AC)T = CT AT . (1.16) (1.17) (1.18)

Determinants

13

Proof The rst two assertions are obvious. To prove the third put AT = (ij), ij = aji, CT = (ij), ij = cji, AC = (uij) and CT AT = (vij). Then
n n n

uij =
k=1

aikckj =
k=1

kijk =
k=1

jkki = vji, u

whence the theorem follows.

32 Denition A matrix A Mn(R) is symmetric if AT = A. A matrix B Mn(R) is skew-symmetric if BT = B. 33 Example If 1 2 3 0 2 3 0 5 A = 2 4 5 , B = 2 3 5 6 3 5 0

then A is symmetric and B is skew-symmetric.

34 Theorem Any matrix A Mn(R) can be written as the sum of a symmetric and a skew-symmetric matrix. Proof Observe that (A + AT )T = AT + ATT = AT + A, and so A + AT is symmetric. Also, (A AT )T = AT ATT = (A AT ), and so A AT is skew-symmetric. We only need to write A as 1 1 A = ( )(A + AT ) + ( )(A AT ) 2 2 to prove the assertion.

1.3

Determinants

We now dene the notion of determinant of a matrix A Mn(R). We will use an inductive denition, so that we can effect calculations of determinants quickly.

14

Chapter 1

35 Denition The determinant of a square matrix A Mn(R), denoted by det A is dened inductively as follows. 1. If n = 1, A = [a], then det A = a. 2. If n = 2, A = a c , then det A = ad bc. b d

3. If n 3, A = [aij], let Aij Mn1(R) denote the matrix obtained by deleting the i-th row and the j-th column from A. Then
n

det A =
j=1

aij(1)i+j det Aij,

(the development along the i-th row) or, alternatively,


n

det A =
i=1

aij(1)i+j det Aij.

(the development along the j-th column).

we always obtain the same determinant. This seems like an arbitrary assumption, but in linear algebra courses we do see that this is indeed the case. The result is independent of our choice. It is therefore advantageous to choose that row, or column, with a maximal number of 0s. It also follows that det A = det AT , and that the determinant of a triangular matrix is the product of its diagonal elements.

! We have assumed that no matter which row or column we choose,

36 Example Find 1 2 3 det 4 5 6 7 8 9

by expanding along the rst row. Solution: We have det A = 1(1)1+1 det

5 6 4 6 4 5 + 2(1)1+2 det + 3(1)1+3 det 8 9 7 9 7 8 = 1(45 48) 2(36 42) + 3(32 35) = 0.

Determinants

15

37 Example Find 1 0 1 1 2 0 0 1 det 666 3 1 1000000 1 0 0 1 Solution: Since the second column has three 0s, it is advantageous to expand along it, and thus we are reduced to calculate 1 1 1 3(1)3+2 det 2 0 1 1 0 1 2 1 = 3(1)(1)(1)(1) = 3. 1 1

Expanding this last determinant along the second column, the original determinant is thus 3(1)3+2(1)(1)1+2 det

To nd the eigenvalues of a matrix A it is necessary and sufcient to solve the equation (called the characteristic equation) det(In A) = 0. 39 Example Since 1 1 1 det 1 3 1 = 3 32 4 + 12 = ( 2)( + 2)( 3), 3 1 + 1 the matrix 1 1 1 1 3 1 3 1 1

38 Denition Let A Mn(R). A vector v = 0 is an eigenvector and a scalar is an eigenvalue if Av =v.

has eigenvalues 2, 2, 3.

16

Chapter 1

40 Theorem The eigenvalues of a real symmetric matrix are all real. 41 Denition Let v 1, v 2, . . . , v k be k vectors in Rn. The k-parallelotope spanned by the v i is the set
k i=1

ti v i : ti [0; 1] .

If

is the n k matrix having the k vectors as columns, then det AT A

A = [ v 1, v 2, . . . , v k]

is the k-dimensional volume of the k-parallelotope spanned by the v 1, v 2, . . . , v k.

! We will se later that det A A 0 so the square root of it is dened.


T

42 Theorem The signed area of a triangle in R2 spanned by the vectors v 1, v 2 is 1 det[ v 1, v 2]. 2

1.4

Linear Transformations

43 Denition Let V, W be two vector spaces over the eld of real numbers. A function V W L: , a L( a ) is called a linear transformation if it is Homogeneous: L( a ) = L( a ), for R. Linear: L( a + b ) = L( a ) + L( b ),

Linear Transformations

17

These two properties can be condensed by saying that a linear transformation is a function L : V W such that

44 Example The trace map

L( a + b ) = L( a ) + L( b ).

tr () :

is linear, since in view of Theorem 26,

Mn(R) R A tr (A)

tr (A + B) = tr (A) + tr (B) , tr (A) = tr (A) . 45 Example The transpose map L: Mn(R) Mn(R) A AT

is linear, since in view of Theorem 31,

L(A + B) = (A + B)T = AT + (B)T = AT + BT = L(A) + L(B). 46 Example Let X Mn(R) be a xed square matrix. Prove that the map L: is linear. Mn(R) Mn(R) A XAX

Solution: Let A, B be matrices in Mn(R) and let be a scalar. We have L(A + B) = = = = whence the claim follows. X(A + B)X XAX + X(B)X XAX + XBX L(A) + L(B),

18 47 Example Prove that the map R2 f: x y R3 x + 2y 2x y

Chapter 1

is linear. Solution: Put

and let R. Then

x x v1= 1 , v2= 2 y1 y2 x1 x + 2 y1 y2 x1 + x2 L y1 + y2 (x1 + x2) + 2(y1 + y2) 2(x1 + x2) ( y1 + y 2) x2 + 2y2 x1 + 2y1 2x1 + 2x2 y2 y1 L( v 1) + L( v 2),
n

L( v 1 + v 2) = L = =

= = whence L is linear.

then

If { v i}i[1;n] is an ordered basis for Rn, a = L( a ) = L


n i=1

i=1

meaning that the action of L on an arbitrary vector a Rn is completely determined by the action L has on the given ordered basis of Rn. This in turn gives the following.

i v i

i v i, and L is linear,

=
i=1

iL( v i),

Linear Transformations 48 Theorem Let

19

49 Example Find the matrix representation of the linear map in example 47 if both R2 and R3 have as ordered bases their standard ordered bases. R2 has the ordered basis 1 1 , 0 1 and R3 has the standard basis as ordered basis. Solution: We have L 1 0 1 = 2 , L 0 2 = 0 , 1

be a linear transformation. Then there is a unique matrix AL Mmn(R) such that for all a Rn, L( a ) = AL a .

Rn Rm L: a L( a )

0 1

and hence the desired matrix is 1 2 2 0 . 0 1 We have L 1 0 1 = 2 , L 0 1 1

and hence the desired matrix is 1 3 2 2 . 0 1

3 = 2 , 1

20 50 Example Consider L : R3 R3, with x xyz L y = x + y + z . z z Prove that L is a linear transformation.

Chapter 1

Find the matrix corresponding to L under the standard basis. Find the matrix corresponding to L under the ordered basis 1 1 1 0 , 1 , 0 , B= 0 0 1 for both the domain and the image of L. Solution: x a Let R. Put u 1 = y , u 2 = b . Then z c L( u + u 2) = =

= =

proving that L is a linear transformation.

x + a L y + b z + c (x + a) (y + b) (z + c) (x + a) + (y + b) + (z + c) c z + xyz abc x + y + z + a + b + c c z a x b y + L L c z L( u ) + L( u 2)

Linear Transformations

21

We have

1 0 1 0 1 1 We have L 0 = 1, L 1 = 1 , and L 0 = 1 , whence 1 1 0 0 0 0 the desired matrix is 1 1 1 1 1 1 . 0 0 1 0 1 1 1 1 1 L 0 = 1 = 0 0 + 1 1 + 0 0 = 1 , 0 B 1 0 0 0 0 1 0 1 1 1 2 L 1 = 2 = 2 0 + 2 1 + 0 0 = 2 , 0 0 0 0 1 0 B 1 0 1 1 1 3 L 0 = 2 = 3 0 + 2 1 + 1 0 = 2 , 1 1 0 0 1 1 B

and

whence the desired matrix is 0 2 3 1 2 2 . 0 0 1 B We will also use the following result. 51 Theorem Let

be linear transformations with matrix representations AL1 Mmn(R) and BL2 Mlm(R) respectively. Then the composition map Rn Rl L2 L1 : a (L2 L1)( a )

Rn Rm Rm Rl L1 : , L2 : a L1( a ) a L2( a )

has as matrix representation the product of matrices BL2 AL1 Mln(R).

22

Chapter 1

1.5

Inner Products

52 Denition Let V be a vector space over R. An inner product function V V R : (x, y) x y satisfying ( x + y ) z = x z + y z ( x ) y = x ( y ) = ( x y ), R.

is a

xy = yx xx 0

54 Denition Given a vector space V over R with inner product and norm ||||, the distance d a , b between a and b is d a, b = a b .

53 Denition Given a vector space V over R with inner product , the norm a of a vector a is a a. a =

xx =0 x = 0

55 Example Given a and b in Rn, the usual inner product is the dot product dened by n a b = akbk. (1.19)
k=1

56 Example Let 1 1 a = 2 , b = 4 3 3

Inner Products be vectors in R3. Their dot product is a b = (1)(1) + (2)(4) + (3)(3) = 0, 14, b = (1)2 + (4)2 + (3)2 =

23

their norms are a =

(1)2 + (2)2 + (3)2 =

26,

and their distance is

0 d a , b = a b = 2 = 6

(0)2 + (2)2 + (6)2 = 2 10.

! We took for granted the fact that the dot product in R


inner product. This does require (a very easy) proof.

does dene an

57 Example Let A, B in Mn(R). Prove that the operation AB = tr BT A is an inner product. This is the standard inner product for matrices with real number entries. Solution: Observe that (A1 + A2)B = = = = tr BT (A1 + A2) tr BT A1 + BT (A2) tr BT A1 + tr BT A2 A1B + A2B,

and so (1) and (2) in denition 52 are veried. Also (3) follows from Theorem 26. To check (4) and (5) we only need to observe that
n n

AA =
i=1 j=1

a2 ij.

24 58 Example Given A= 1 1 1 0 ,B = 0 1 0 1

Chapter 1

their standard inner product is A B = tr

1 0 0 1

1 1 0 1

= tr

1 1 0 1

= 2,

their standard norms are ||A|| = and ||B|| = 1 0 0 1 = (1)2 + (0)2 + (0)2 + (1)2 = 1 1 0 1 = (1)2 + (1)2 + (0)2 + (1)2 = 3,

2.

59 Theorem (Cauchy-Bunyakovsky-Schwarz Inequality) Let V be a vector space over R having inner product and corresponding norm ||||. Then for any two vectors x and y we have |xy| x y .

Proof Since the norm of any vector is non-negative, we have x +ty 0 ( x + t y )( x + t y ) 0 x x + 2t x y + t2 y y 0 2 2 x + 2t x y + t2 y 0.


2

This last expression is a quadratic polynomial in t which is always nonnegative. As such its discriminant must be non-positive, that is, (2 x y )2 4( x
2

giving the theorem.

)( y

) 0 | x y | x

y ,

Inner Products

25

60 Corollary (Triangle Inequality) Let V be a vector space over R having inner product and corresponding norm ||||. Then for any two vectors a and b we have a + b a + b . Proof || a + b ||2 = = = ( a + b )( a + b ) a a +2ab + bb || a ||2 + 2|| a |||| b || + || b ||2 (|| a || + || b ||)2,

from where the desired result follows.

61 Denition Let x and y be two non-zero vectors in a vector space over the real numbers. Then the angle ( x , y ) between them is given by the relation x y cos ( x , y ) = . y x

This expression agrees with the geometry in the case of the dot product for R2 and R3. 62 Example Let u , v be vectors in a vector space V over R with inner product . Prove the polarisation identity: 1 u v = || u + v ||2 || u v ||2 . 4

giving the result.

Solution: We have || u + v ||2 || u v ||2 = ( u + v )( u + v ) ( u v )( u v ) = uu +2uv + vv (uu 2uv + vv) = 4u v,

63 Example Let a , b be xed vectors in R2. Prove that if v R2, v a = v b , then a = b .

26

Chapter 1

Solution: We have v R2, v ( a b ) = 0. In particular, choosing v = a b , we gather ( a b )( a b ) = || a b ||2 = 0. But the norm of a vector is 0 if and only if the vector is the 0 vector. Therefore a b = 0 , i.e., a = b .

! The Cauchy-Bunyakovsky-Schwarz (CBS) Inequality applied to the


dot product in Rn gives
n k=1 n 1/2 n 1/2

xkyk

x2 k
k=1 k=1

y2 k

(1.20)

64 Example Assume that ak, bk, ck, k = 1, . . . , n, are positive real numbers. Shew that
n 4 n n n 2

akbkck
k=1 n

a4 k
k=1 k=1

b4 k
k=1

c2 k

Solution: Using CBS on


k=1 n

(akbk)ck once we obtain


n 1/2 2 a2 kbk k=1 1/2 2 a2 kbk k=1 n 1/2

k=1

akbkck
n

c2 k

Using CBS again on


k=1 n

we obtain
1/2 2 a2 kbk n 1/2

k=1

akbkck

c2 k
k=1 n 1/4 n 1/2

k=1 n

1/4

a4 k
k=1 k=1

b4 k
k=1

c2 k

which gives the required inequality.

Cross product in R3

27

1.6

Cross product in R3

We now dene the standard cross product in R3 as a product satisfying the following properties. 65 Denition Let ( x , y , z , ) R3 R3 R3 R. The cross product : R3 R3 R3 is a closed binary operation satisfying Bilinearity: Anti-commutativity: x y = ( y x )

Right-hand Rule: i j = k, j k = i , k i = j . k j

x x = 0

Scalar homogeneity: ( x ) y = x ( y ) = ( x y )

( x + z ) y = x y + z y and x ( z + y ) = x z + x y

Figure 1.3: Right-handed system.

Figure 1.4: Left-handed system.

To study points in space we must rst agree on the orientation that we will give our co-ordinate system. We will use, unless otherwise noted, a right-handed orientation, as in gure 1.3.

28 66 Example Find 1 0 0 1 . 3 2

Chapter 1

Hence

Solution: We have ( i 3k)( j +2k) = i j +2 i k 3k j 6k k = k 2 j +3 i +60 = 3 i 2 j + k. 1 0 3 0 1 = 2 . 3 2 1

Operating as in example 66 we obtain

68 Theorem x ( x y ) and y ( x y ).

y1 x1 67 Theorem Let x = x2 and y = y2 be vectors in R3. Then y3 x3 x y = (x2y3 x3y2) i + (x3y1 x1y3) j + (x1y2 x2y1) k .

Proof We will only check the rst assertion, the second verication is analogous. x ( x y ) = (x1 i + x2 j + x3 k )((x2y3 x3y2) i +(x3y1 x1y3) j + (x1y2 x2y1) k ) = x1x2y3 x1x3y2 + x2x3y1 x2x1y3 + x3x1y2 x3x2y1 = 0, u

completing the proof.

69 Example Let R. Find a vector a of unit length simultaneously perpen0 1 dicular to v = a and w = a . a 0

Cross product in R3 v w v w Solution: Either of or will do. Now || v w || || v w ||

29

v w = (a j + a k ) ( i + a j ) = a( j i ) a2( j j ) + a( k i ) + a2( k j ) 2 = ak + a j a i 2 a = a , a a4 + a2 + a2 = 2a2 + a4. Hence we may take either

and || v w || =

2 a 1 a 2a2 + a4 a 2 a 1 a . 2a2 + a4 a

or

! The cross product of vectors in R


but

is not associative, since

i ( i j )= i k = j ( i i ) j = 0 j = 0.

We have, however, the following theorem. 70 Theorem

a (b c) = (ac)b (ab)c.

30 Proof a (b c) = =

Chapter 1

= = completing the proof.

(a1 i + a2 j + a3 k ) ((b2c3 b3c2) i + +(b3c1 b1c3) j + (b1c2 b2c1) k ) a1(b3c1 b1c3) k a1(b1c2 b2c1) j a2(b2c3 b3c2) k +a2(b1c2 b2c1) i + a3(b2c3 b3c2) j a3(b3c1 b1c3) i (a1c1 + a2c2 + a3c3)(b1 i + b2 j + b3 i )+ (a1b1 a2b2 a3b3)(c1 i + c2 j + c3 i ) = (ac)b (ab)c, u

! Permuting the vectors in the above theorem and adding, we obtain


71 Theorem Let ( x , y ) [0; ] be the convex angle between two vectors x and y . Then || x y || = || x |||| y || sin ( x , y ).
Jacobis Identity: a (b c)+ b (c a)+ c (a b) = 0.

Proof We have || x y ||2 = = = = = =

whence the theorem follows.

(x2y3 x3y2)2 + (x3y1 x1y3)2 + (x1y2 x2y1)2 2 2 2 2 2 x2 2y3 2x2y3x3y2 + x3y2 + x3y1 2x3y1x1y3+ 2 2 2 2 2 + x2 1y3 + x1y2 2x1y2x2y1 + x2y1 2 2 2 2 2 2 (x1 + x2 + x3)(y1 + y2 + y3) (x1y1 + x2y2 + x3y3)2 || x ||2|| y ||2 ( x y )2 || x ||2|| y ||2 || x ||2|| y ||2 cos2 ( x , y ) || x ||2|| y ||2 sin2 ( x , y ),

The following corollaries are now obvious.

Lines and Planes in R3

31

72 Corollary Two non-zero vectors x , y satisfy x y = 0 if and only if they are parallel. 73 Corollary (Lagranges Identity) || x y ||2 = ||x||2||y||2 ( x y )2. 74 Example Let x R3, ||x|| = 1. Find || x i ||2 + || x j ||2 + || x k ||2.
2

Solution: By Lagranges Identity, || x i ||2 = x

and since ( x i )2 + ( x j )2 + ( x k )2 = equals 3 1 = 2.

|| x j ||2 = x

|| x k ||2 = x

( x k )2 = 1 ( x k )2, x
2

( x j )2 = 1 ( x j )2,

( x i )2 = 1 ( x i )2,

= 1, the desired sum

1.7

Lines and Planes in R3

1 76 Example Find the equation of the line passing through 2 in the di3 2 rection of 1. 0

x 3 3 . The line y 75 Denition Let a R and v R \ { 0 }. Put r = z passing through a in the direction of v is the set { r : r = a + t v , t R}.

32 Solution: The desired equation is x 1 2 y = 2 + t 1 . z 3 0

Chapter 1

1 2 77 Example Find the equation of the line passing through 2 and 1. 3 0 Solution: The line follows the direction 3 1 (2) 2 (1) = 3 . 3 30 The desired equation is 3 1 x y = 2 + t 3 . 3 3 z

? Why does
represent the same line as

x 2 3 y = 1 + t 3 z 0 3 x 1 3 y = 2 + t 3 ? z 3 3

! Given two lines in space, one of the following three situations might

arise: (i) the lines intersect at a point, (ii) the lines are parallel, (iii) the lines are skew (one over the other, without intersecting).

Lines and Planes in R3

33

or equivalently,

In three-dimensions a hyperplane is simply a plane. obtain the equation To n1 of a plane in R3 simply notice that if the vector n2 is normal (perpen3 n a1 dicular) to the plane passing through the point a2 then form any other a3 x x a1 point y on the plane, the vector y a2 will be perpendicular to the z z a3 n1 vector n2 and so n3 x a1 n1 n2 y a2 = 0 z a3 n3 n1(x a1) + n2(y a2) + n3(z a3) = 0, (1.21)

78 Denition The set of points x Rn satisfying an equation of the form a x =c ( a Rn and c R xed) is called a hyperplane in Rn.

gives the equation of a plane in R3. 79 Example The equation of the plane passing through the point (1, 1, 2) 3 and normal to the vector 2 is 4 3(x 1) + 2(y + 1) + 4(z 2) = 0.

! From n (x a ) + n (y a ) + n (z a ) = 0 we obtain that the


1 1 2 2 3 3

equation of a plane is of the form

n1x + n2y + n3z = d,

34
n1 where n2 is perpendicular to the plane, and d R. n3

Chapter 1

80 Example Find the equation of plane containing the point (1, 1, 1) and perpendicular to the line x = 1 + t, y = 2t, z = 1 t. Solution: The vectorial form of the equation of the line is 1 1 r = 0 + t 2 . 1 1 1 1 Since the line follows the direction of 2 , this means that 2 is nor1 1 mal to the plane, and thus the equation of the desired plane is (x 1) 2(y 1) (z 1) = 0.

81 Example Find the equation of plane containing the point (1, 1, 1) and containing the line x = 2y = 3z. Solution: Observe that (0, 0, 0) (as 0 = 2(0) = 3(0)) is on the line, and hence on the plane. Thus the vector 10 1 1 0 = 1 1 0 1 lies on the plane. Now, if x = 2y = 3z = t, then x = t, y = t/2, z = t/3. Hence, the vectorial form of the equation of the line is 0 1 1 r = 0 + t 1/2 = t 1/2 . 0 1/3 1/3

Lines and Planes in R3 1 This means that 1/2 also lies on the plane, and thus 1/3 1/6 1 1 1 1/2 = 4/3 3/2 1/3 1 1 4 3 x y + z = 0. 6 3 2

35

is normal to the plane. The desired equation is thus

Given three planes in space, they may (i) be parallel (which allows for some of them to coincide), (ii) two may be parallel and the third intersect each of the other two at a line, (iii) intersect at a line, (iv) intersect at a point.

82 Example Find the equation of the plane passing through the points (a, 0, a), (a, 1, 0), and (0, 1, 2a) in R3. The vectors

and

a (a) 2a 0 1 = 1 a0 a 0 (a) a 11 = 0 2a 0 2a

The equation of the plane is thus given by 2a xa 3a2 y 0 = 0, za a

lie on the plane. A vector normal to the plane is 2a 2a a 1 0 = 3a2 . a 2a a

36 that is, 2ax + 3a2y az = a2.

Chapter 1

83 Example Find the equation of the line perpendicular to the plane ax + a2y + a3z = 0, a = 0 and passing through the point (0, 0, 1). a Solution: A vector normal to the plane is a2. The line sought has the a2 same direction as this vector, thus the equation of the line is a 0 x y = 0 + t a2 , t R. 1 z a2 84 Example Find the equation of the plane perpendicular to the line ax = by = cz, abc = 0 and passing through the point (1, 1, 1) in R3. Solution: Put ax = by = cz = t, so x = t/a; y = t/b; z = t/c. The parametric equation of the line is x 1/a y = t 1/b , t R. z 1/c 1/a Thus the vector 1/b is perpendicular to the plane. Therefore, the equa1/c tion of the plane is 1/a x1 0 1/b y 1 = 0 , 1/c z1 0 or 1 1 1 x y z + + = + + . a b c a b c We may also write this as bcx + cay + abz = ab + bc + ca.

Topology of Rn

37

85 Example (Putnam Exam 1980) Let S be the solid in three-dimensional space consisting of all points (x, y, z) satisfying the following system of six conditions: x 0, y 0, z 0, 2x + 4y + 3z 36, 2x + 3z 24. Determine the number of vertices and the number of edges of S. Solution: There are 7 vertices (V0 = (0, 0, 0), V1 = (11, 0, 0), V2 = (0, 9, 0), V3 = (0, 0, 8), V4 = (0, 3, 8), V5 = (9, 0, 2), V6 = (4, 7, 0)) and 11 edges (V0V1, V0V2, V0V3, V1V5, V1V6, V2V4, V3V4, V3V5, V4V5, and V4V6). x + y + z 11,

1.8

Topology of Rn

86 Denition Let a Rn and let > 0. An open ball centred at a of radius is the set B( a ) = {x Rn : d x , a < }.

87 Example An open ball in R is an open interval, an open ball in R2 is an open disk and an open ball in R3 is an open sphere. 88 Denition A set S Rn is said to be open if for every point belonging to it we can surround the point by a sufciently small open ball so that this balls lies completely within the set. That is, a S > 0 such that B( a ) S. On the real line, an open ball is an interval of length 2r centred at a point p, on the plane, an open ball is an open disk centred about p , in 3-dimensional space, an open ball is a sphere excluding its boundary and centred at p .

89 Denition A set O Rn is said to be open in Rn if x O r > 0 such that Br( x ) O .

38

Chapter 1

That is, a set is open if for all its elements, there exist open balls centred at the elements and totally contained in the set. 90 Example The open interval ] 1; 1[ is open in R. The interval ] 1; 1] is not open, however, as no interval centred at 1 is totally contained in ] 1; 1]. 91 Example The region ] 1; 1[]0; +[ is open in R2.

92 Example The ellipsoidal region {(x, y) R2 : x2 + 4y2 < 4} is open in R2. The reader will recognise that open boxes, open ellipsoids and their unions and nite intersections are open sets in Rn. 93 Denition A set F Rn is said to be closed in Rn if its complement Rn \ F is open. 94 Example The closed interval [1; 1] is closed in R, as its complement, R \ [1; 1] =] ; 1[]1; +[ is open in R. The interval ] 1; 1] is neither open nor closed, however. 95 Example The region [1; 1] [0; +[[0; 2] is closed in R3.

96 Example (Putnam Exam 1969) Let p(x, y) be a polynomial with real coefcients in the real variables x and y, dened over the entire plane R2. What are the possibilities for the image (range) of p(x, y)? Solution: Since polynomials are continuous functions and the image of a connected set is connected for a continuous function, the image must be an interval of some sort. If the image were a nite interval, then f(x, kx) would be bounded for every constant k, and so the image would just be the point f(0, 0). The possibilities are thus (i) a single point (take for example, p(x, y) = 0), (ii) a semi-innite interval with an endpoint (take for example p(x, y) = x2 whose image is [0; +[), (iii) a semi-innite interval with no endpoint (take for example p(x, y) = (xy 1)2+ x2 whose image is ]0; +[), (iv) all real numbers (take for example p(x, y) = x).

Quadratic Forms

39

97 Example (Putnam Exam 1984) Let A be a solid a b c rectangular brick in three dimensions, where a > 0, b > 0, c > 0. Let B be the set of all points which are at distance at most 1 from some point of A (in particular, A B). Express the volume of B as a polynomial in a, b, c. Solution: The set B can be decomposed into the following subsets: The set A itself, of volume abc. Two a b 1 bricks, two b c 1 bricks, and two c a 1 bricks, Four quarter-cylinders of length a and radius 1, four quarter-cylinders of length b and radius 1, and four quarter-cylinders of length c and radius 1, Eight eighth-of-spheres of radius 1. Thus the required formula for the volume is abc + 2(ab + bc + ca) + (a + b + c) + 4 . 3

1.9

Quadratic Forms

98 Denition A matrix A Mn(R) is called positive denite if x Rn \ { 0 }, T x A x > 0. It is positive semi-denite if x Rn, x T A x 0. It is negative denite if x Rn \ { 0 }, T x A x < 0. It is negative semi-denite if x Rn, x T A x 0. A matrix is indenite if it is neither positive nor negative denite (semi-denite). 99 Example The matrix 1 0 0 A = 0 2 0 0 0 3

40 is positive denite, since x 1 0 0 x y z 0 2 0 y = x2 + 2y2 + 3z2 > 0, 0 0 3 z

Chapter 1

for (x, y, z) = (0, 0, 0), being a sum of squares. 100 Example The matrix 1 0 0 A = 0 2 0 0 0 3 0 1 0 0 = 1 > 0, 3 0

is indenite, since 1 0 1 0 0 0 2 0 0

1 0 0 0 0 1 0 0 2 0 1 = 2 < 0. 0 0 3 0

101 Example The matrix 1 0 0 A = 0 2 0 0 0 0 1 0 0 x x y z 0 2 0 y = (x2+2y2) < 0. 0 0 0 z

is negative semi-denite, since 1 0 0 0 0 0 1 0 2 0 0 = 0, 0 0 0 1

102 Denition The n principal minors of a matrix A = [aij] are 1 = a11, 2 = det a11 a12 , a21 a22

a11 a12 a13 3 = det a21 a22 a23 , a31 a32 a33

Quadratic Surfaces . . . a11 a12 a13 a1k a21 a22 a23 a2k k = det . , . . . . . . . . . . . . . . ak1 ak2 ak3 akk . . . n = det A.

41

103 Theorem (Sylvesters Criterion) A matrix A Mn(R) is positive denite if and only if all its principal minors are positive. It is positive semi-denite if all its principal minors are non-negative. It is negative denite if all of its odd order minors are negative and all of its even order minors are positive. It is negative semi-denite if all of its odd order minors are non-positive and all of its even order minors are non-negative. It is indenite if none of the above cases occur.

1.10

Quadratic Surfaces

104 Denition A quadric (or quadratic) surface in R3 is a surface whose cartesian equation is a polynomial of degree 2. Thus the general form of a quadric surface is Ax2 + 2Bxy + 2Cxz + Dy2 + 2Eyz + Fz2 + 2Gx + 2Hy + 2Iz + J = 0 where all the coefcients are real. A B x y z 1 C G This can be written in the matrix form B C G x D E H y = 0, (1.22) E F I z H I J 1

where we have identied a 1 1 matrix with a real number.

! The square matrix of coefcients is symmetric, and as such all its


eigenvalues are real by virtue of Theorem 40.

42 105 Example Write the quadric surface 2x2 3y2 + 10xy + 5 + (z 2)2 = 0 in matrix form. Solution: First observe that

Chapter 1

2x2 3y2 + 10xy + (z 2)2 = 2x2 3y2 + 10xy + 5 + z2 4z + 9. The desired form is x 2 5 0 0 5 3 0 0 y = 0. x y z 1 0 0 1 2 z 1 0 0 2 9

In the case when there are no cross terms, things greatly simplify, and, up to permutations of the letters, we have the following result. 106 Theorem If y2 z2 x2 + q + r = d, a2 b2 c2 then this equation represents p an ellipsoid: p = q = r = d = 1, a, b, c are the lengths of the semi axes. a single-sheet hyperboloid: p = q = d = 1, r = 1. a double-sheet hyperboloid: r = d = 1, p = q = 1. Cone: p = q = 1, r = 1, d = 0. x2 y2 z + q + r 2 = d, 2 2 a b c then this equation represents p an elliptic paraboloid: p = q = 1, r = 1, d = 0. a hyperbolic paraboloid: p = r = 1, q = 1, d = 0. If

Quadratic Surfaces an elliptic cylinder: p = q = 1, r = d = 0. a hyperbolic cylinder: p = d = 1, q = 1, r = 0. a pair of planes: p = 1, q = 1, d = 0. If py2 + qx = d then this equation represents a parabolic cylinder: p, q > 0. a pair of parallel planes: d > 0, q = 0, p = 0. two coinciding planes: p = 0, q = d = 0.

43

x Figure 1.5: Example 107.

107 Example Demonstrate that the surface in R3 S : 4x2 + y2 4 = 0 is a cylinder and draw its graph.

44

Chapter 1

Solution: The variable z is missing. Notice that on the plane, 4x2+y24 = 0 denes an ellipse. Thus the surface is an elliptic cylinder, with the z-axis as directrix. Its graph appears in gure 1.5. 108 Example The surface in R3 given by S : x2 z2 = 1 is a cylinder, as the y-variable is missing. It is a cylindrical hyperboloid. 109 Example The surface in R3 given by S : y .05z2 = 1 is a cylinder, as the x-variable is missing. It is a cylindrical paraboloid. 110 Example (Putnam Exam 1970) Determine, with proof, the radius of the largest circle which can lie on the ellipsoid x2 y2 z2 + + = 1, a > b > c > 0. a2 b2 c2 Solution: The largest circle has radius b. Parallel cross sections of the ellipsoid are similar ellipses, hence we may increase the size of these by moving towards the centre of the ellipse. Every plane through (0, 0, 0) which makes a circular cross section must intersect the yz-plane, and the diameter of any such cross section must be a diameter of the ellipse x = z2 y2 0, 2 + 2 = 1. Therefore, the radius of the circle is at most b. Arguing b c similarly on the xy-plane shews that the radius of the circle is at least b. To shew that circular cross section of radius b actually exist, one may verify that the two planes given by a2(b2 c2)z2 = c2(a2 b2)x2 give circular cross sections of radius b.

1.11

Canonical Surfaces in R3

111 Denition A surface S consisting of all lines parallel to a given line and passing through a given curve is called a cylinder. The line is called the directrix of the cylinder.

Canonical Surfaces in R3

45

To recognise whether a given surface is a cylinder we look at its Cartesian equation. If it is of the form f(A, B) = 0, where A, B are secant planes, then the curve is a cylinder. Under these conditions, the lines generating S will be parallel to the line of equation A = 0, B = 0. In practice, if one of the variables x, y, or z is missing, then the surface is a cylinder, whose directrix will be the axis of the missing coordinate.

112 Example Demonstrate that the surface in R3 S : ex


2 +y2 +z2

(x + z)e2xz = 0,

implicitly dened, is a cylinder. Solution: The planes A : x + z = 0 and B : y = 0 are secant. The surface 2 2 has equation of the form f(A, B) = eA +B A = 0, and it is thus a cylinder. The directrix has direction i k . 1 1 1 + + =1 xy yz zx is a cylinder and nd the direction of its directrix. Solution: Considering the planes A : x y = 0, B : y z = 0, the equation takes the form 1 1 1 1 = 0, f(A, B) = + A B A+B thus the equation represents a cylinder. To nd its directrix, we nd the x 1 intersection of the planes x = y and y = z. This gives y = t 1. The z 1 direction vector is thus i + j + k .

113 Example Shew that the surface S in R3 given implicitly by the equation

114 Denition Given a point R3 (called the apex) and a curve (called the generating curve), the surface S obtained by drawing rays from and passing through is called a cone.

46

Chapter 1

form f( , ) = 0, where A, B, C, are planes secant at exactly one point, C C then the surface is a cone, and its apex is given by A = 0, B = 0, C = 0.

! InA practice, if the Cartesian equation of a surface can be put into the B

115 Example Demonstrate that the surface in R3 given implicitly by z2 xy = 2z 1 is a cone Solution: After rearranging, we obtain (z 1)2 xy = 0, or Considering the planes A : x = 0, B : y = 0, C : z = 1, 0 we see that our surface is a cone, with apex at 0. 1 116 Example The surface in R3 implicitly by z2 = x2 + y2 y x is a cone, as its equation can be put in the form ( )2 + ( )2 1 = 0. z z 0 Considering the planes x = 0, y = 0, z = 0, the apex is located at 0. 0 117 Denition A surface S obtained by making a curve turn around a line is called a surface of revolution. We then say that is the axis of revolution. The intersection of S with a half-plane bounded by is called a meridian. x y + 1 = 0. z1z1

Canonical Surfaces in R3

47

! If the Cartesian equation of S can be put in the form f(A, ) = 0, where


A is a plane and is a sphere, then the surface is of revolution. The axis of S is the line passing through the centre of and perpendicular to the plane A.

118 Example Shew that the surface S in R3 implicitly dened as xy + yz + zx + x + y + z + 1 = 0 is of revolution and nd its axis. Solution: Rearranging, (x + y + z)2 (x2 + y2 + z2) + 2(x + y + z) + 2 = 0, so we may take A : x + y + z = 0, : x2 + y2 + z2 = 0 as our plane and sphere. The axis of revolution is then in the direction of i + j + k . x4 + y4 + z4 4xyz(x + y + z) = 1 is a surface of revolution, and nd its axis of revolution. Solution: Rearranging, 1 (x2 + y2 + z2)2 ((x + y + z)2 (x2 + y2 + z2)) 1 = 0, 2 and so we may take A : x + y + z = 0, : x2 + y2 + z2 = 0, shewing that the surface is of revolution. Its axis is the line in the direction i + j + k . 120 Example Find the equation of the surface of revolution generated by revolving the hyperbola x2 4z2 = 1 about the z-axis.

119 Example Shew that the surface in R3 implicitly dened by

Let (x, y, z) be a point on S. If this point were on the xz plane, it would be on the hyperbola, and its distance to the axis of rotation would be |x| = 1 + 4z2. Anywhere else, the distance of (x, y, z) to the axis of rotation is

48 the same as the distance of (x, y, z) to (0, 0, z), that is have x2 + y2 = 1 + 4z2, which is to say x2 + y2 4z2 = 1.

Chapter 1 x2 + y2. We must

121 Example Find the equation of the surface of revolution generated by revolving the line 3x + 4y = 1 about the y-axis. Solution: Let (x, y, z) be a point on S. If this point were on the xy plane, it would be on the line, and its distance to the axis of rotation would be 1 |x| = |1 4y|. Anywhere else, the distance of (x, y, z) to the axis of 3 rotation is the same as the distance of (x, y, z) to (0, y, 0), that is x2 + z2. We must have 1 x2 + z2 = |1 4y|, 3 which is to say 9x2 + 9z2 16y2 + 8y 1 = 0.

122 Example Find the equation of the surface of revolution S generated by revolving the ellipse 4x2 + z2 = 1 about the z-axis. Solution: Let (x, y, z) be a point on S. If this point were on the xz plane, it would be on the ellipse, and its distance to the axis of rotation would be 1 1 z2. Anywhere else, the distance from (x, y, z) to the z-axis |x | = 2 is the distance of this point to the point (0, 0, z) : x2 + y2. This distance is the same as the length of the segment on the xz-plane going from the z-axis. We thus have 1 x2 + y2 = 1 z2, 2 or 4x2 + 4y2 + z2 = 1.

Parametric Curves and Surfaces

49

123 Example The circle (y a)2 + z2 = r2, (a, r) (R )2 on the yz plane is revolved around the z-axis, forming a torus T . Find the equation of this torus. Solution: Let (x, y, z) be a point on T . If this point were on the yz plane, it would be on the circle, and the of the distance to the axis of rotation would be y = a + sgn(y a) r2 z2, where sgn(t) (with sgn(t) = 1 if t < 0, sgn(t) = 1 if t > 0, and sgn(0) = 0) is the sign of t. Anywhere else, the distance from (x, y, z) to the z-axis is the distance of this point to the point (0, 0, z) : x2 + y2. We must have x2 + y2 = (a + sgn(y a) r2 z2)2 = a2 + 2asgn(y a) r2 z2 + r2 z2. Rearranging x2 + y2 + z2 a2 r2 = 2asgn(y a) r2 z2, or (x2 + y2 + z2 (a2 + r2))2 = 4a2r2 4a2z2 since (sgn(y a))2 = 1, (it could not be 0, why?). Rearranging again, (x2 + y2 + z2)2 2(a2 + r2)(x2 + y2) + 2(a2 r2)z2 + (a2 r2)2 = 0. The equation of the torus thus, is of fourth degree.

1.12

Parametric Curves and Surfaces

124 Denition Let [a; b] R. A parametric curve representation r of a curve is a function r : [a; b] Rn, with x1(t) x2(t) r (t) = . , . . xn(t)

and such that r ([a; b]) = . r(a) is the initial point of the curve and r(b) its terminal point. A curve is closed if its initial point and its nal point coincide.

50

Chapter 1

The trace of the curve r is the set of all images of r , that is, . If there exist t1 = t2 such that r (t1) = r (t2) = p, then p is a multiple point of the curve. The curve is simple if its has no multiple points. A closed curve whose only multiple points are its endpoints is called a Jordan curve. It is important to realise that a curve might have different parametric representations. For example r : s : [0; 2] t [0; 2] t R2 cos t sin t R2 sin 2t cos 2t

and

are two parametrisations for for the unit circle : {(x, y) R2 : x2 + y2 = 1}. Notice that r travels the unit circle once starting at (1, 0) and going counterclockwise, and so r is a simple curve. One the other hand, s starts at (0, 1) and travels the unit circle twice going clockwise. We dene lim x1(t) a t lim x2(t) t a , lim r (t) = . t a . . lim xn(t) provided each of the limits on the right hand side exists and x1(t) x (t) 2 r (t) = . , . . xn(t)
t a

provided each of the entries on the right hand side be differentiable at t. In particular, we dene the differential of r as d x1 d x2 dr = . . . . d xn

Parametric Curves and Surfaces We will keep our practice of writing x(t) r (t) = y(t) x(t) r (t) = y(t) z(t)

51

in R2 and

in R3. 125 Example The trace of

is known as a cylindrical helix.

r (t) = i cos t + j sin t + k t

126 Example Find a parametric representation for the curve resulting by the intersection of the plane 3x + y + z = 1 and the cylinder x2 + 2y2 = 1 in R3. Solution: The projection of the intersection of the plane 3x + y + z and the cylinder is the ellipse x2 + 2y2 = 1, on the xy-plane. This ellipse can be parametrised as 2 x = cos t, y = sin t, 0 t 2. 2 From the equation of the plane, z = 1 3x y = 1 3 cos t 2 sin t. 2

Thus we may take the parametrisation cos t x(t) 2 sin t r (t) = y(t) = . 2 z(t) 2 1 3 cos t sin t 2

52

Chapter 1

127 Example Let P be a point at a distance d from the centre of a circle of radius r. The curve traced out by P as the circle rolls along a straight line is called a trochoid. Find a parametrisation of the trochoid. Solution: Let be the angle (in radians) of rotation of the circle, and let C be the centre of the circle. At = 0 the centre of the circle is at (0, r), and P = (0, r d). Suppose the circle is displaced towards the right, making the point P to rotate an angle of = 0 radians. Then the centre of the circle has displaced r0 units horizontally, and so is now located at (r, r). The polar co-ordinates of the point P are (d sin 0, d cos 0), in relation to the centre of the circle (notice that the circle moves clockwise). The point P has moved X = r d sin 0 horizontal units and Y = r d cos 0 units. This is the desired parametrisation. 128 Example A hypocycloid is a curve traced out by a xed point P on a circle C of radius r as C rolls on the inside of a circle with centre at O and radius R. If the initial position of P is (R, 0), and is the angle, measured counterclockwise, that a ray starting at O and passing through the centre of C makes with the x-axis, shew that a parametrisation of the hypocycloid is (R r) x = (R r) cos + r cos , r y = (R r) sin r sin (R r) r .

Solution: Suppose that starting from = 0, the centre O of the small circle moves counterclockwise inside the larger circle by an angle , and the point P = (x, y) moves clockwise an angle . The arc length travelled by the centre of the small circle is (R r) radians. At the same time the point P has rotated r radians, and so (R r) = r. See gure 1, where O B is parallel to the x-axis.

Let A be the projection of P on the x-axis. From gure 2, OAP = OPO = , OO P = , POA = , and OP = (R 2 2

Parametric CurvesO and Surfaces B P O


O 53 P O A
Figure 1.7: Hypocycloid

Figure 1.6: Hypocycloid

r) sin( ). Hence x = (OP) cos POA = (R r) sin( ) cos( y = (R r) sin( ) sin( Now x = (R r) sin( ) cos( ). 2 ), 2

= (R r) sin( + ) sin (R r) = (cos cos(2 + )) 2 (R r) (cos + cos(2 + )) = (R r) cos 2 = (R r) cos (R r)(cos( + ) cos ). Now, cos( + ) = cos( ) = cos (R r) r and so (R r) r r r = and cos = OO Rr

) 2

x = (Rr) cos (Rr)(cos(+) cos = (Rr) cos +r cos as required. The identity for y is proved similarly.

129 Denition A parametric representation r of a surface S in R3 is a function of the form R2 R3 , r : (u, v) r(u, v)

54 where

Chapter 1

and whose image is S.

x(u, v) r (u, v) = y(u, v) , z(u, v)

130 Example The surface of a sphere S : {(x, y, z) R3 : x2 + y2 + z2 = R2} centred at the origin and of xed radius R > 0 can be parametrised by using spherical co-ordinates, with 0 2 and 0 and x(, ) R cos sin r (, ) = y(, ) = R sin sin . z(, ) R cos For x2 + y2 + z2 = R2(cos2 sin2 + sin2 sin2 + cos2 ) = R2. 131 Example A parametrisation for the torus generated by revolving the circle (y a)2 + z2 = r2 around the z-axis is a cos + r cos cos r (, ) = a sin + r sin cos , r sin

with (, ) [; ]2.

1.13

Frenet-Serret Formul

132 Denition A curve r : [a; b] R3 is said to be smooth if it is differen tiable on [a; b] and r is continuous there. The unit tangent vector T of a smooth curve at t is r (t) . T (t) = ||r (t)|| The unit normal vector N at t is T (t) N (t) = , || T (t)||

Frenet-Serret Formul and the unit binormal vector at t is B (t) = T (t) N (t).

55

The three vectors { T , N , B } are called the Frenet-Serret frame of the curve r. 133 Example Compute the Frenet-Serret frame at t for the cylindrical helix r(t) = i cos t + j sin t + kt. Solution: We have r (t) = i sin t + j cos t + k, and ||r (t)|| = 2. Thus sin t 1 T (t) = cos t 2 1 Also, cos t 1 T (t) = sin t , 2 0

1 and || T (t)|| = . Hence 2

cos t N = sin t . 0 Finally, sin t cos t 1 1 B (t) = cos t sin t = (i(sin t) j(cos t) + k). 2 2 1 0

56

Chapter 1

1.14

Limits

134 Denition A function f : Rn Rm is said to have a limit L Rm at a Rn if > 0 > 0 such that In such a case we write, 0 < ||x a|| < = ||f(x) L|| < .
x a

lim f(x) = L.

The notions of innite limits, limits at innity, and continuity at a point, are analogously dened. Limits in more than one dimension are perhaps trickier to nd, as one must approach the test point from innitely many directions. x2y 135 Example Find lim . (x,y) (0,0) x2 + y2

Solution: We use the sandwich theorem. Observe that 0 x2 x2 + y2, x2 and so 0 2 1. Thus x + y2
(x,y) (0,0)

lim

and hence

x2y = 0. (x,y) (0,0) x2 + y2 lim

(x,y) (0,0)

lim

x2y lim |y|, (x,y) (0,0) x2 + y2

136 Example Find

Solution: Observe that if |x| |y| 0, then y8 x5y3 4 = y4. 6 4 x +y y If |y| |x| 0, then x8 x5y3 = x2. x6 + y4 x6

x5y3 . (x,y) (0,0) x6 + y4 lim

Limits Thus

57

as (x, y) (0, 0). Aliter: Let X = x3, Y = y2.

x5y3 max(y4, x2) y4 + x2 0, x6 + y4 X5/3Y 3/2 x5y3 = . x6 + y4 X2 + Y 2

Passing to polar co-ordinates X = cos , Y = sin , we obtain X5/3Y 3/2 x5y3 = = 5/3+3/22| cos |5/3| sin |3/2 7/6 0, x6 + y4 X2 + Y 2 1+x+y . (x,y) (0,0) x2 y2 lim 1+x +, x2

as (x, y) (0, 0).

137 Example Find

Solution: When y = 0,

as x 0. When x = 0,

as y 0. The limit does not exist. 138 Example Find lim

1+y , y2

Solution: Putting x = t4, y = t3, we nd xy6 1 = 2 +, 6 8 x +y 2t

xy6 . (x,y) (0,0) x6 + y8

as t 0. But when y = 0, the function is 0. Thus the limit does not exist. 139 Example Find ((x 1)2 + y2) loge((x 1)2 + y2) . (x,y) (0,0) |x| + |y| lim

58 Solution: When y = 0 we have 2(x 1)2 ln(|1 x|) 2x , |x | |x | and so the function does not have a limit at (0, 0). 140 Example Find lim sin(x4) + sin(y4) x4 + y4 .

Chapter 1

Solution: sin(x4) + sin(y4) x4 + y4 and so sin(x4) + sin(y4) x4 + y4 as (x, y) (0, 0). x4 + y4 0,

(x,y) (0,0)

141 Example Find

Solution: When y = 0 we obtain

sin x y . (x,y) (0,0) x sin y lim sin x 1, x

as x 0. When y = x the function is identically 1. Thus the limit does not exist.

Chapter

Differentiation
2.1
x a

Local Study of Functions

142 Denition A function : R R is said to be innitesimal as x a if lim (x) = 0. 143 Example sin : R [1; 1] is innitesimal as x 0, since lim sin x = 0.
x 0

145 Denition We say that a function : R R is asymptotic to a function : R R as x a, and we write , if (x) = 1. x a (x) lim sin x = 1. x 0 x

1 is innitesimal as x +, since 144 Example f : R R, x x 1 = 0. lim x + x

146 Example We have sin x x as x 0, since lim


2

x2 + x = 1. 147 Example We have x + x x as x 0, since lim x 0 x 59

60 148 Example We have x2 + x x2 as x +, since lim

Chapter 2 x2 + x = 1. x2

149 Denition (Du-Bois Raymond-Landau small oh notation) We say that : R R is negligible in relation to : R R as x a or that is preponderant in relation to as x a, if We express the condition above with the notation (x) = (read of x is small oh of of x as x tends to a) (x) = 0. x a (x) lim o ((x)) xa

x +

! When the limit to which the independent variable tends is understood,


we abbreviate

In particular, if is innitesimal as x a, then (x) = o (1) as x a. 150 Example We have x2 = o (x) as x 0 since x2 = lim x = 0. x 0 x 0 x lim

o ((x)) as o ((x)) xa

151 Example We have x = o x2 as x + since We write (x) = (x) + o ((x)) as x a if (x) (x) = o ((x)) as x a. 152 Example We have sin x = x + o (x) as x 0 since lim sin x sin x x = lim lim 1 = 1 1 = 0. x 0 x x 0 x 0 x
x +

lim

x 1 = lim = 0. 2 x 0 x x

153 Theorem Let and be innitesimal functions as x a. Then the following hold.

Local Study of Functions The sum of two innitesimals is an innitesimal: o ((x)) + o ((x)) = o ((x)) . The difference of two innitesimals is an innitesimal: o ((x)) o ((x)) = o ((x)) . c R \ {0}, o (c(x)) = o ((x)) . n N, n 2, 1 k n 1, o (((x))n) = o ((x))k . o (o ((x))) = o ((x)). n N, n 1, ((x))no ((x)) = o ((x))n+1 . o (((x))n) n N, n 2, = o ((x))n1 . (x) o ((x)) = o (1). (x)
n

61

If ck are real numbers, then o


k=1

ck((x))k

= o ((x)).

()(x) = o ((x)) and ()(x) = o ((x)). If , then ( )(x) = o ((x)) and ( )(x) = o ((x)). 154 Example In view of theorem 153, we have o 2x3 + 8x2 = o (x) , as x 0.

155 Example In view of theorem 153, we have 2x3 + 8x2 + x x,

as x 0.

62 156 Example In view of theorem 153, we have o 2x3 + 8x2 = o x4 , as x +.

Chapter 2

157 Example In view of theorem 153, we have 2x3 + 8x2 + x 2x3,

as x +.

The following result follows easily from the Maclaurin expansion of the given function. 158 Theorem Let x 0. Then sin x = x

x3 x5 x2n+1 + + (1)n + o x2n+2 . 3! 5! (2n + 1)! In particular, sin x = x + o x2 . x2 x4 x2n + + (1)n + o x2n+1 . 2! 4! (2n)! In particular, cos x = 1 + o (x). x3 2x5 + + o x5 . 3 15 xn x2 x3 + + + + o (xn) 2! 3! n! x2 x3 xn + + (1)n+1 + o (xn). 2 3 n

cos x = 1

tan x = x +

ex = 1 + x +

loge(1 + x) = x (1+x) = 1+x+ o (xn).

( 1) 2 ( 1)( 2)( 3) ( n + 1) n x + + x + 2 n!

Local Study of Functions 159 Example Since tan x = x + o (x), we have tan as x 0. Also, x2 x2 = +o 2 2 x2 2 = x2 + o x2 , 2

63

(tan x)3 = (x + o (x))3 = x3 + 3x2o (x) + 3xo x2 + (o (x))3 = x3 + o x3 . x2 + o x2 , we have 2! 9x4 + o x4 . 2

160 Example Since cos x = 1

cos 3x2 = 1

161 Example Find an asymptotic expansion of cot2 x of type o x2 as x 0. Solution: Since tan x x we have cot2 x We can write this as cot2 x = 162 Example Calculate x2 2 . lim x 0 loge cos 3x sin sin tan x2 + o x2 2 +o x
2

1 . x2

1 +o x2

1 . x2

Solution: We use theorems 158 and 153. sin sin tan x2 = sin sin 2 x2 = sin 2 x2 = sin 2 x2 = +o 2

+o

x2 + o x2 2

+ o x2 x2 ,

64 and 9x2 + o x2 loge cos 3x = loge 1 2 9x2 9x2 + o x2 + o + o x2 = 2 2! 9x2 = + o x2 2 The limit is thus equal to x2 1 + o x2 + o (1) 1 lim 2 2 = . = lim 2 9 x 0 x 0 9 9x + o (1) + o (x2) 2 2 163 Example Find lim (cos x)(cot
x 0
2

Chapter 2

x)

. 1 +o x2 = 1 . Also, x2

Solution: By example 161, we have cot2 x = x2 + o x2 2

loge cos x = loge 1 Hence (cos x)cot


2

x2 + o x2 . 2

= = exp =

exp (cot2 x) loge cos x x2 1 1 + o + o x2 x2 x2 2 1 exp( + o (1)) 2 1 e 2,

164 Example Find an asymptotic development of loge(2 cos x+sin x) around x = 0 of order o x4 .

as x 0.

Local Study of Functions Solution: By theorem 158, 2 cos x + sin x = 2 1 = = and so, loge(2 cos x + sin x) = = = + = + = loge 2 1 + x2 x4 x3 + + o x5 + x + o x4 2 24 6 3 4 x x 2 + x x2 + + o x4 6 12 x x2 x3 x4 + + o x4 , 2 1+ 2 2 12 24

65

x x2 x3 x4 + + o x4 2 2 12 24 x x2 x3 x4 + + o x4 loge 2 + loge 1 + 2 2 12 24 x x2 x3 x4 + + o x4 loge 2 + 2 2 12 24 2 1 x x2 x3 x4 + + o x4 2 2 2 12 24 3 2 1 x x x3 x4 + + o x4 3 2 2 12 24 4 2 1 x x x3 x4 4 + +o x + o x4 4 2 2 12 24 1 x2 x3 x4 x x2 x3 x4 + + loge 2 + 2 2 12 24 2 4 2 6 1 x4 1 x3 3x4 + o x4 3 8 8 4 16 x 5x2 5x3 35x4 loge 2 + + + o x4 2 8 24 192

In the direction of local behaviour of vector functions, we start with the following theorem.

as x 0. Our main concern, as far as asymptotic developments is concerned, will be expansions when the vector tends to 0 . Let f, g : Rn Rm. By v , as v 0 , we mean that f( v ) = g ( v ) + o ||f( v ) g( v )|| lim = 0. v v0

66

Chapter 2

165 Theorem Let L : Rn Rm be a linear transformation. Then L( 0 ) = 0 . Moreover, there exists a constant M > 0 such that ||L( h )|| M h . Proof By linearity, L( 0 ) = L( 0 + 0 ) = L( 0 ) + L( 0 ), h =
n

which gives the rst part. Now, let

k=1

where the e k are the standard basis for Rn. Then and hence by the triangle inequality, and by the Cauchy-BunyakovskySchwarz inequality, ||L( h )||
n

hk e k,

L( h ) =

k=1

hkL( e k),

k=1

( =

|hk| )
k=1

2 1/2 n

|hk|||L( e k)||
n

k=1

h (

k=1

||L( e k)||2)1/2,
n k=1

||L( e k)||2)1/2

and the theorem follows upon putting M = (

||L( e k)||2)1/2. h

166 Corollary With M as in theorem 165, || h L( h )|| M h


2

=o

as h 0 .

Denition of the Derivative

67

2.2

Denition of the Derivative

exists. The limit condition above is equivalent to saying that g(x) g(a) g (a)(x a) = 0, x a xa lim g(a + h) g(a) g (a)(h) = 0. h 0 h lim g(a + h) g(a) = g (a)(h) + o (h) ,

We now dene the derivative in a multidimensional vector space. Recall that in one variable, a function g : R R is said to be differentiable at x = a if the limit g(x) g(a) = g (a) lim x a xa

or equivalently,

This last condition we may write in small oh notation as

as h 0. The above analysis provides an analogue denition for the higher-dimensional case. Observe that since we may not divide by vectors, the corresponding denition in higher dimensions involves quotients of norms. 167 Denition Let A Rn. A function f : A Rm is said to be differen tiable at a A if there is a linear transformation, called the derivative of f n m at a , D a (f) : R R such that (f) Equivalently, f is differentiable at a if there is a linear transformation D a such that (f)( h ) + o , h f( a + h ) f( a ) = D a as h 0 . (f)( x a )|| ||f( x ) f( a ) D a = 0. lim xa || x a ||

68
! The condition for differentiability at a is equivalent to as x a .

Chapter 2

(f)( x a ) + o || x a || , f( x ) f( a ) = D a

(f) is uniquely deter168 Theorem If A is an open set in denition 167, D a mined.

Proof Let L : Rn Rm be another linear transformation satisfying deni (f)( v ). Since A is tion 167. We must prove that v Rn, L( v ) = D a open, a + h A for sufciently small h . By denition, as h 0 , we have (f)( h ) + o . h f( a + h ) f( a ) = D a and Now, observe that (f)( v ) L( v = D (f)( h ) f( a + h )+ f( a )+ f( a + h ) f( a ) L( h ). D a a By the triangle inequality, (f)( v ) L( v || ||D a = = (f)( h ) f( a + h ) + f(a)|| ||D a +||f( a + h ) f( a ) L(h)|| h +o h o , h o f( a + h ) f( a ) = L( h ) + o (||vectorh||) .

as h 0 This means that (f)( v )|| 0, ||L( v ) D a (f)( v ), completing the proof. i.e., L( v ) = D a
a

! If A = { a }, a singleton, then D (f) is not uniquely determined. For || x a || < holds only for x = a , and so f( x ) = f( a ). Any linear transformation T will satisfy the denition, as T ( x a ) = T ( 0 ) = 0 , and ||f( x ) f( a ) T ( x a )|| = || 0 || = 0, identically.

Denition of the Derivative

69

(L) = L, 169 Example If L : Rn Rm is a linear transformation, then D a n for any a R . (L) uniquely deterSolution: Since Rn is an open set, we know that D a mined. Thus if L satises denition 167, then the claim is established. But by linearity

whence the claim follows.

||L( x ) L( a ) L( x a )|| = ||L( x ) L( a ) L( x ) + L( a )|| = ||0|| = 0,

170 Example Let f : Rn R, n 1, f( x ) = ||x|| be the usual norm in Rn, 2 with x = x x . Prove that x v (f)( v ) = , D x ||x||

t Solution: Assume that x = 0 . We use the fact that (1 + t)1/2 = 1 + + o (t) 2 as t 0, from theorem 153. We have f( x + h ) f( x ) = = = = || x + h || ||x|| ( x + h )( x + h ) ||x||
2

for x = 0 , but that f is not differentiable at 0 .

x x

As h 0 ,

+2xh + h

+2xh + h 2 2xh + h

+ x

+2xh + h

x 2 x .

70 Since h
2

Chapter 2 =o h as h 0 , we have
2 2

proving the rst assertion. To prove the second assertion, assume that there is a linear transfor (f) = L, L : Rn R such that mation D 0 as h 0. Recall that by theorem 165, L( 0 ) = 0 , and so by example L( h ) (L)( 0 ) = 0 , (L)( 0 ) = L( 0 ) = 0 . This implies that 169, D D 0 0 h as h 0. Since f( 0 ) = ||0|| = 0, f( h ) = h this would imply that h L( h ) = o L( h ) 1 h h , ||f( 0 + h ) f( 0 ) L( h )|| = o h ,

+2xh + h

2xh + h

+ x

x h +o h

or

= o (1) .

But the sinistral side 1 as h 0 , and the dextral side 0 as h 0 . This is a contradiction, and so, such linear transformation L does not exist at the point 0 . 171 Example Let L : R3 R3 be a linear transformation and Shew that F is differentiable and that (F)( h ) = x L( h ) + h L( x ). D x R3 R3 F: . x x L( x )

Denition of the Derivative Solution: We have F( x + h ) F( x ) = ( x + h ) L( x + h ) x L( x ) = ( x + h ) (L( x ) + L( h )) x L( x ) = x L( h ) + h L( x ) + h L( h )

71

as h 0 by virtue of corollary 166. This h Now, || h L( h )|| = o yields the desired result. The following examples consider derivatives of functions with domains other than Rn. 172 Example Let f: R3 R3 R (x, y) x y

be the usual dot product in R3. Shew that f is differentiable and that D( x , y )f( h , k ) = x k + h y .

As ( h , k ) ( 0 , 0 ), we have by the Cauchy-Buniakovskii-Schwarz in h , which proves the assertion. equality, | h k | h ||k|| = o 173 Example Shew that f: is differentiable and that Mn(R) Mn(R) X X2

Solution: We have f( x + h , y + k ) f( x , y ) = ( x + h )( y + k ) x y = xy + xk + hy + hk xy = x k + hy + hk.

DXf(H) = XH + HX.

72 Solution: We have f(X + H) f(X) = (X + H)2 X2 = X2 + XH + HX + H2 X2 = XH + HX + H2 = XH + HX + o (H) , proving the assertion. 174 Example Let f: M2(R) R . A det A

Chapter 2

Given H M2(R), nd an expression for DA(f)(H) Solution: Put A= Then f(A + H) f(A) = = = = = = a1 + h1 a2 + h2 a a 1 2 a3 + h3 a4 + h4 a3 a4 (a1 + h1)(a4 + h4) (a2 + h2)(a3 + h3) a1a4 + a2a3 a1a4 a2a3 + a1h4 + a4h1 a2h3 a3h2 +h1h4 h2h3 a1a4 + a2a3 a1h4 + a4h1 a2h3 a3h2 + h1h4 h2h3 h1 h2 a4 a2 tr + h1h4 h2h3 h3 h4 a3 a1 tr (Hadj (A)) + h1h4 h2h3, det h h a1 a2 , H= 1 2 . h3 h4 a3 a4

where adj (A) is the (classical) adjoint matrix of A. Now, the ArithmeticMean-Geometric-Mean inequality for real numbers a, b says that |a||b| a2 + b2 and so 2 2 2 h2 + h2 1 2 + h3 + h4 |h1h4 h2h3| |h1||h4| + |h2||h3| 1 = ||H||2 = o (H) , 2 2 as H 02. This implies that DA(f)(H) = tr (Hadj (A)) .

Just like in the one variable case, differentiability at a point, implies continuity at that point.

The Jacobi Matrix

73

175 Theorem Suppose A Rn is open and f : A Rn is differentiable on A. Then f is continuous on A. Proof Given a A, we must shew that Since f is differentiable at a we have and so
xa

lim f( x ) = f( a ).

(f)( x a ) + o || x a || , f( x ) f( a ) = D a f( x ) f( a ) 0 , u

as x a , proving the theorem.

2.3

The Jacobi Matrix

We now establish a way which simplies the process of nding the derivative of a function at a given point. 176 Denition Let A Rn, f : A Rm, and put f1(x1, x2, . . . , xn) f2(x1, x2, . . . , xn) f( x ) = . . . . fm(x1, x2, . . . , xn) Here fi : Rn R. The partial derivative whenever this limit exists.

fi (x) is dened as xj

fi fi(x1, x2, . . . , xj + h, . . . , xn) fi(x1, x2, . . . , xj, . . . , xn) (x) = lim , h 0 xj h

To nd partial derivatives with respect to the j-th variable, we simply keep the other variables xed and differentiate with respect to the j-th variable.

74

Chapter 2

x 177 Example If f : R3 R, and f( r ) = f y = x + y2 + z3 + 3xy2z3 then z D2(f)( r ) = 2y + 6xyz3, D1(f)( r ) = 1 + 3y2z3,

and

! Sometimes we will use the equivalent notations


f D2(f)( r ) = (x, y, z), y f (x, y, z), D1(f)( r ) = x

D3(f)( r ) = 3z2 + 9xy2z2.

etc..

178 Example Suppose g : R R is continuous and a R is a constant. Find the partial derivatives with respect to x and y of x f : R R, f = y
2 x2 y

g(t) dt.
a

Solution: We have

f (x, y, z) = 2xyg(x2y), x and f (x, y, z) = x2g(x2y). y Since the derivative of a function f : Rn Rm is a linear transformation, it can be represented by aid of matrices. The following theorem will allow (f) under the standard us to determine the matrix representation for D a n m bases of R and R .

The Jacobi Matrix 179 Theorem Let

75

Proof Let e j, 1 j n, be the standard basis for Rn. To obtain the Jacobi (f)( e j), which will give us the j-th column of matrix, we must compute D x f = (Jij), and observe that the Jacobi matrix. Let J x J1j J2j (f)( e j) = . . D x . . Jnj and put y = x + e j, R. Notice that (f)( y x )|| ||f( y ) f( x ) D x || y x || (f)( e j)|| ||f(x1, x2, . . . , xj + h, . . . , xn) f(x1, x2, . . . , xj, . . . , xn) D x . = | |

Suppose A Rn is an open set and f : A Rm is differentiable. Then fi (f) each partial derivative (x) exists, and the matrix representation of D x xj n m with respect to the standard bases of R and R is the Jacobi matrix f1 f1 f1 x1 (x) x2 (x) xn (x) f2 f2 f2 (x ) (x) (x) . x2 xn f = J x x1. . . . . . . . . . . . fn fn fn (x) (x) (x) x1 x2 xn

f1(x1, x2, . . . , xn) f2(x1, x2, . . . , xn) f( x ) = . . . . fm(x1, x2, . . . , xn)

Since the sinistral side 0 as 0, the so does the i-th component of the numerator, and so, |fi(x1, x2, . . . , xj + h, . . . , xn) fi(x1, x2, . . . , xj, . . . , xn) Jij| 0. | |

76 This entails that Jij = lim

Chapter 2

This nishes the proof.

fi(x1, x2, . . . , xj + , . . . , xn) fi(x1, x2, . . . , xj, . . . , xn) fi = (x) . 0 xj

180 Example Let f : R3 R2 be given by

Compute the Jacobi matrix of f under the usual basis for R3 and R2. Also, (f) under the the usual basis for compute the matrix representation for D r 4 1 1 3 2 R and the basis , in R . Finally, nd D(1,2,3)f 5. 0 1 6 Solution: The Jacobi matrix is the 2 3 matrix f1 f1 f1 y x+z y (r) (r) (r) 1 x y z . f = 1 = J r f2 f f 2 2 0 (r) (r) (r) x y x y z

xy + yz f( r ) = . loge xy

f is expressed in the standard basis of Each of the column vectors of J r 1 1 2 R . To express them in the basis , of R2 we simply write 0 1

y 1 1 = (y ) x x x+z 1 = (x + z y

1 1 1 , + 0 x 1 1 1 1 1 ) , + 0 y y 1

y 1 1 =y +0 . 0 0 1

The Jacobi Matrix The desired matrix representation is 1 1 y x+z y y = 1 x . 1 0 x y

77

A r

4 To compute D(1,2,3)f 5, we will use the Jacobi matrix, which will give us 6 the result in the standard basis of R2. 4 2 4 2 4 40 5 = 7 . J(1,2,3) 5 = 1 1 0 6 6 2 2 181 Example Let f(, , z) = ( cos , sin , z) be the function which changes from cylindrical co-ordinates to Cartesian co-ordinates. We have cos sin 0 f = sin cos 0 . J (,,z ) 0 0 1 182 Example Let f(, , ) = ( cos sin , sin sin , cos ) be the function which changes from spherical co-ordinates to Cartesian co-ordinates. We have cos sin cos cos sin sin f = sin sin sin cos cos sin . J (,,) cos sin 0 The Jacobi matrix provides a convenient computational tool to compute the derivative of a function at a point. Thus differentiability at a point implies that the partial derivatives of the function exist at the point. The converse, however, is not true.

78 183 Example Let f : R2 R be given by y if x = 0, x if y = 0, f(x, y) = 1 if xy = 0.

Chapter 2

Observe that f is not continuous at (0, 0) (f(0, 0) = 0 but f(x, y) = 1 for values arbitrarily close to (0, 0)), and hence, it is not differentiable there. We f f have however, (0) = (0) = 1. Thus even if both partial derivatives x1 x2 exist at (0, 0) is no guarantee that the function will be differentiable at (0, 0). You should also notice that both partial derivatives are not continuous at (0, 0). We have, however, the following. 184 Theorem Let A Rn be an open set, and let f : Rn Rm. Put f1 f2 f= . . .. If each of the partial derivatives Djfi exists and is continuous fm on A, then f is differentiable on A. Just like in the one-variable case, we have the following rules of differentiation. Let A Rn, B Rm be open sets f, g : A Rm, R, be differentiable on A, h : B Rl be differentiable on B, and f(A) B. Then we have
((h f)) = D (h) (D (f)). Chain Rule: D x x f (x) (f) + D (g). ((f + g)) = D Addition Rule: D x x x

Since composition of linear mappings expressed as matrices is matrix multiplication, the Chain Rule takes the alternative form when applied to the Jacobi matrix.
(h f) = (J h)(J f). J x x f (x)

(2.1)

The Jacobi Matrix 185 Example Let uev u = u + v , f v uv

79

Solution: We have

(f h). Find J r

x2 + y h( r ) = . y+z

and

v e uev f = 1 1 , J (u ,v ) v u
h = J r

2x 1 0 . 0 1 1

Observe also that

ey+z (x2 + y)ey+z . f = 1 1 J h(r) 2 y+z x +y Hence


(f h) = (J f)(J h) J r r h(r) y+z 2 y+z (x + y)e e 2x 1 0 = 1 1 0 1 1 y+z x2 + y y+z 2 y+z 2 y+z (1 + x + y)e (x + y)e 2xe . = 2x 2 1 2xy + 2xz x2 + 2y + z x2 + y

80 186 Example Let u = u2 + ev, v u, v : R3 R u( r ) = xz, v( r ) = y + z. f : R2 R, f


3 2

Chapter 2

h u(x, y, z) Put h( r ) = f . Find (r) for i = 1, 2, 3. v(x, y, z) xi h = f g. Now, Solution: Put g : R R , g( r ) =


g = J r

z 0 x , 0 1 1

u( r ) v( r )

xz . Observe that y+z

Thus

f = 2xz ey+z . J g (r)

f = 2u ev , J (u ,v )

h h h (r) (r) (r) x1 x2 x3

f)(J g) = (J r g (r) z 0 x = 2xz ey+z 0 1 1

h = J r

= Equating components, we obtain

2xz2 ey+z 2x2z + ey+z

h (r) = 2xz2, x1 h (r) = ey+z, x2 h (r) = 2x2z + ey+z. x3

Gradients and Directional Derivatives

81

2.4

Gradients and Directional Derivatives


Rn Rm f: x f( x )

A function

is called a vector eld. If m = 1, it is called a scalar eld. 187 Denition Let Rn R f: x f( x )

The gradient operator is the operator

be a scalar eld. The gradient of f is the vector dened and denoted by f x1 (x) f ( x ) . f)T = x2 f( x ) = (J x . . . f (x) xn D1 D2 = . . . . Dn

188 Theorem Let A Rn be open and let f : A R be a scalar eld, and assume that f is differentiable in A. Let K R be a constant. Then f( x ) is orthogonal to the surface implicitly dened by f( x ) = K. Proof Let
n R R c : t c (t)

be a curve lying on this surface. Choose t0 so that c (t0) = x . Then (f c )(t0) = f( c (t)) = K,

82 and using the chain rule f)(J c ) = 0, (J t0 c(t )


0

Chapter 2

which translates to

Since c (t0) is tangent to the surface and its dot product with f( x ) is 0, we conclude that f( x ) is normal to the surface. u
! Let be the angle between f( x ) and c (t ). Since
0

(f( x ))( c (t0)) = 0.

189 Example Find a unit vector normal to the surface x3 + y3 + z = 4 at the point (1, 1, 2). Solution: Here f(x, y, z) = x3 + y3 + z 4 has gradient 2 3x f( r ) = 3y2 1 3 which at (1, 1, 2) is 3. Normalising this vector we obtain 1 3 19 3 . 19 1 19

f( x ) is the direction in which f is changing the fastest.

|(f( x ))( c (t0))| = ||f( x )|||| c (t0)|| cos ,

190 Example Find the direction of the greatest rate of increase of f(x, y, z) = xyez at the point (2, 1, 2).

Gradients and Directional Derivatives Solution: The direction is that of the gradient vector. Here z ye f( r ) = xez xyez 2 e which at (2, 1, 2) becomes 2e2 . Normalising this vector we obtain 2e2 1 1 2 . 5 2 191 Example Let f : R3 R be given by x = x + y2 z2. y f z

83

Find the equation of the tangent plane to f at (1, 2, 3). Solution: A vector normal to the plane is f(1, 2, 3). Now 1 f( r ) = 2y 2z 1 4 6

which is

at (1, 2, 3). The equation of the tangent plane is thus 1(x 1) + 4(y 2) 6(z 3) = 0, or x + 4y 6z = 9.

84 192 Denition Let

Chapter 2

be a vector eld with

193 Example If f(x, y, z) = (x2, y2, yez ) then

f1( x ) f ( 2 x ) f( x ) = . . . . fn( x ) The divergence of f is dened and denoted by f1 f2 fn divf(x) = f( x ) = (x) + (x) + + (x) . x1 x2 xn
2

Rn Rn f: x f( x )

divf(x) = 2x + 2y + 2yzez . 194 Denition Let gk : Rn Rn, 1 k n 2 be vector elds with gi = (gi1, gi2, . . . , gin). Then the curl of (g1, g2, . . . , gn2) e1 e2 en D1 D2 Dn g11( x ) g ( x ) g ( x ) 12 1n . curl(g1, g2, . . . , gn2)(x) = det g ( g22( x ) ... g2n( x ) 21 x ) . . . . . . . . . . . . g(n2)1( x ) g(n2)2( x ) . . . g(n2)n( x )
2

195 Example If f(x, y, z) = (x2, y2, yez ) then i j k 2 curlf(r) = det D1 D2 D3 = (ez ) i . 2 x2 y2 yez

196 Example If f(x, y, z, w) = (exyz, 0, 0, w2), g(x, y, z, w) = (0, 0, z, 0) then x e1 e2 e3 e4 y 1 D2 D3 D4 2 xyz = det D curl(f, g) ) e 4. 2 = (xz e z exyz 0 0 w w 0 0 z 0

Extrema

85

197 Denition Let A Rn be open and let f : A R be a scalar eld, and assume that f is differentiable in A. Let v Rn \ { 0 } be such that x + t v A for sufciently small t R. Then the directional derivative of f in the direction of v at the point x is dened and denoted by f( x + t v ) f( x ) f . (x) = lim t 0 x t v

! Some authors require that the vector v in denition 197 be a unit vector.

199 Example Find the derivative of f(x, y, z) = x3 + y3 z2 in directional 1 the direction of v = 2. 3 Solution: We have 3x2 f( r ) = 3y2 2z

198 Theorem Let A Rn be open and let f : A R be a scalar eld, and assume that f is differentiable in A. Let v Rn \ { 0 } be such that x + t v A for sufciently small t R. Then the directional derivative of f in the direction of v at the point x is given by f( x ) v .

and so

(f( r )) v = 3x2 + 6y2 6z.

2.5

Extrema

We now turn to the problem of nding maxima and minima for vector functions. As in the one-variable case, the derivative will provide us with information about the extrema, and the second derivative will provide us with information about the nature of these extreme points.

86

Chapter 2

To dene an analogue for the second derivative, let us consider the following. Let A Rn and f : A Rm be differentiable on A. We know (f) is a linear transformation from Rn to Rm. This that for xed x 0 A, D x0 means that we have a function A L (Rn, Rm) T: , (f) x D x

where L (Rn, Rm) denotes the space of linear transformations from Rn to Rm. Hence, if we differentiate T at x 0 again, we obtain a linear transfor2 n n m (f)) = D (D (T ) = D (f) from R to L (R , R ). Hence, mation D x0 x0 x0 x0 n m 2 (f)(x1 ) L (R , R ). Again, this means that given x1 Rn, we have D x0 m 2 (f)(x1 ))( x 2) R . Thus the function given x 2 Rn, D x0 is well dened, and linear in each variable x1 and x2 , that is, it is a bilinear function. Just as the Jacobi matrix was a handy tool for nding a matrix repre (f) in the natural bases, when f maps into R, we have the sentation of D x following analogue representation of the second derivative. 200 Theorem Let A Rn be an open set, and f : A R be twice differenn n 2 (f) : R R R with respect to the tiable on A. Then the matrix of D x standard basis is given by the Hessian matrix: 2f 2f 2f x1x1 ( x ) x1x2 ( x ) x1xn ( x ) 2f 2f 2f (x) ( x ) ( x ) f = x2x1 x x x x H 2 2 2 n x . . . . . . . . . . . . 2f 2 2 f f (x) ( x ) (x) xnx1 xnx2 xnxn
: B x0

Rn Rn L (Rn, Rm) 2 (f)(x1 , x2 ) ( x1 , x2 ) D x0

201 Example Let f : R3 R be given by x f( r ) = f y = xy2z3. z

Extrema Then 0 2yz3 3y2z2 f = 2yz3 H 2xz3 6xyz2 r 2 2 3y z 6xyz2 6xy2z

87

From the preceding example, we notice that the mixed partial deriva2f 2f 2f 2f 2f 2f = , = , and = , are tives x1x2 x2x1 x1x3 x3x1 x2x3 x3x2 equal. This is no coincidence, as guaranteed by the following theorem. 202 Theorem Let A Rn be an open set, and f : A R be twice differ2 2 (f) is continuous, then D (f) is symmetric, that is, entiable on A. If D x0 x0 (x1 , x2 ) Rn Rn we have We are now ready to study extrema in several variables. The basic theorems resemble those of one-variable calculus. First, we make some analogous denitions. 203 Denition Let A Rn be an open set, and f : A R. If there is some open ball Bx0 (r) on which x Bx0 (r), f( x 0) f( x ), we say that f( x 0) is a local maximum of f. Similarly, if there is some open ball Bx1 (r) on which x Bx0 (r ), f( x 1) f( x ), we say that f( x 1) is a local maximum of f. A point is called an extreme point if it is either a local minimum or local maximum. A point t is called a critical point if f is differentiable at t and (f) = 0. A critical point which is neither a maxima nor a minima is called D t a saddle point. 204 Theorem Let A Rn be an open set, and f : A R be differen (f) = 0, that is, x 0 is a tiable on A. If x 0 is an extreme point, then D x0 critical point. Moreover, if f is twice-differentiable with continuous second f is negative denite, then derivative and x0 is a critical point such that H x0 f is positive denite, then f has a local f has a local maximum at x 0. If H x0 f is f is indenite, then f has a saddle point. If H minimum at x 0. If H x0 x0 semi-denite (positive or negative), the test is inconclusive. 2 2 (f)(x2 , x1 ). (f)(x1 , x2 ) = D D x0 x0

88 205 Example Find the critical points of R2 R f: (x, y) x2 + xy + y2 + 2x + 3y

Chapter 2

and investigate their nature. Solution: We have

and so to nd the critical points we solve

f = 2x + y + 2 x + 2y + 3 , J r

2x + y + 2 = 0, x + 2y + 3 = 0, 1 4 which yields x = , y = . Now, 3 3


f = H r

2 1 , 1 2 2 1 = 3 > 0. 1 2

which is positive denite, since 1 = 2 > 0 and 2 = det 1 4 Thus x0 = ( , ) is a relative minimum and we have 3 3

1 4 7 = f( , ) f(x, y) = x2 + xy + y2 + 2x + 3y. 3 3 3

206 Example Find the extrema of f: R3 R . 2 2 2 (x, y, z) x + y + 3z xy + 2xz + yz

Solution: We have

f = 2x y + 2z 2y x + z 6z + 2x + y , J r

Extrema which vanishes when x = y = z = 0. Now, 2 1 2 f = 1 2 1 , H r 2 1 6

89

2 1 = 3 > 0, which is positive denite, since 1 = 2 > 0, 2 = det 1 2 2 1 2 and 3 = det 1 2 1 = 4 > 0. Thus f has a relative minimum at 2 1 6 (0, 0, 0) and 0 = f(0, 0, 0) f(x, y, z) = x2 + y2 + 3z2 xy + 2xz + yz.

207 Example Find the extrema of f : Solution: We have

R3 R . 2 2 (x, y, z) x + y + z2 + xyz

and

f = 2x + yz 2y + xz 2z + xy , J r

2 z y f = z 2 x . H r y x 2

We see that 1(x, y, z) = 2, 2(x, y, z) = det


f = 8 2x2 2y2 2z2 + 2xyz. det H r f = (0 0 0) then we must have If J r

2 z = 4z2 and 3(x, y, z) = z 2

2x = yz, 2y = xz, 2z = xy,

and upon multiplication of the three equations, 8xyz = x2y2z2,

90 that is, xyz(xyz + 8) = 0.

Chapter 2

Clearly, if xyz = 0, then we must have at least one of the variables equalling 0, in which case, by virtue of the original three equations, all equal 0. Thus (0, 0, 0) is a critical point. If xyz = 8, then none of the variables is 0, 8 and solving for x, say, we must have x = , and substituting this into yz 2 2x + yz = 0 we gather (yz) = 16, meaning that either yz = 4, in which case x = 2, or zy = 4, in which case x = 2. It is easy to see then that either exactly one of the variables is negative, or all three are negative. The other critical points are therefore (2, 2, 2), (2, 2, 2), (2, 2, 2), and (2, 2, 2). At (0, 0, 0), 1(0, 0, 0) = 2 > 0, 2(0, 0, 0) = 4 > 0, 1(0, 0, 0) = 8 > 0, and thus (0, 0, 0) is a minimum point. If x2 = y2 = z2 = 4, xyz = 8, then 2(x, y, z) = 0, 3 = 32, so these points are saddle points. 208 Example Find the extrema of f : Solution: We have R3 R . 2 2 (x, y, z) x y + y z + 2x z

and

f = 2xy + 2 x2 + 2yz y2 1 , J r

2y 2x 0 f = 2x 2z 2y . H r 0 2y 0 We see that 1(x, y, z) = 2y, 2(x, y, z) = det


f = 8y3. 3(x, y, z) = det H r f = (0 0 0) then we must have If J r

2y 2x = 4yz 4x2 and 2x 2z

xy = 1,

x2 = 2yz, y = 1,

Extrema

91

1 1 1 and hence (1, 1, ), and (1, 1, ) are the critical points. Now, 1(1, 1, ) = 2 2 2 1 1 1 2, 2(1, 1, ) = 6, and 3(1, 1, ) = 8. Thus (1, 1, ) is a saddle 2 2 2 1 1 1 point. Similarly, 1(1, 1, ) = 2, 2(1, 1, ) = 6, and 3(1, 1, ) = 2 2 2 1 8, shewing that (1, 1, ) is also a saddle point. 2 209 Example Determine the nature of the critical points of f(x, y, z) = 4xyz x4 y4 z4. Solution: We nd 4yz 4x3 f(x, y, z) = 4xz 4y3 . 4xy 4z3

Assume f(x, y, z) = 0. Then

Thus xyz 0, and if one of the variables is 0 so are the other two. Thus (0, 0, 0) is the only critical point with at least one of the variables 0. Assume now that xyz = 0. Then 1 (xyz)3 = x4y4z4 = (xyz)4 = xyz = 1 = yz = = x4 = 1 = x = 1. x Similarly, y = 1, z = 1. Since xyz = 1, exactly two of the variables can be negative. Thus we nd the following critical points: (0, 0, 0), (1, 1, 1), (1, 1, 1), (1, 1, 1), (1, 1, 1). The Hessian is 12x2 4z 4y f = 4z H 12y2 4x . x 4y 4x 12z2

4yz = 4x3, 4xz = 4y3, 4xy = 4z3 = xyz = x4 = y4 = z4.

If 1 = xyz = x2 = y2 = z2, we have 1 = 12x2 = 12 < 0, 2 = 144x2y2 16z2 = 144 16 = 128 > 0, and 3 = = = < 1728x2y2z2 + 192x4 + 192z4 + 128zyx + 192y4 1728 + 192 + 192 + 128 + 192 1024 0.

92

Chapter 2

This means that for xyz = 0 the Hessian is negative denite and the function has a local maximum at each of the four points (1, 1, 1), (1, 1, 1), (1, 1, 1), (1, 1, 1). Observe that at these critical points f = 1. Now f(0, 0, 0) = 0 and f(1, 1, 1) = 7. 210 Example Determine the nature of the critical points of g(x, y, z) = xyzex
2 y2 z2

.
2 2 2

with t(x, y, z) = (ex )(ey )(ez ). Since at the critical points we have 1 2x2 = 1 2y2 = 1 2z2 = 0, the Hessian reduces to (4x3 6x)(yz) 0 0 . g = (e3/2) H 0 (4y3 6y)(xz) 0 x 3 0 0 (4z 6z)(xy) 1 = (4x3 6x)(yz) 2 = (4x3 6x)(4y3 6y)(xyz2) 3 = (4x3 6x)(4y3 6y)(4z3 6z)(x2y2z2). Also, 4 1 2
3

The function is 0 if any of the variables is 0. Since the function clearly assumes positive and negative values, we can discard any point with a 0. 1 1 1 If (x, y, z) = 0, then x = ; y = z = . We nd 2 2 2 (4x3 6x)(yz) (1 2x2)(1 2y2)z (1 2x2)(1 2z2)y g = t(x, y, z) (1 2y2)(1 2x2)z H (4y3 6y)(xz) (1 2y2)(1 2z2)x , x (1 2z2)(1 2x2)y (1 2z2)(1 2y2)x (4z3 6z)(xy)
2 2 2

Solution: To facilitate differentiation observe that g(x, y, z) = (xex )(yey )(zez ). Now 2 2 2 (1 2x2)(yz)(ex )(ey )(ez ) 2 2 2 g(x, y, z) = (1 2y2)(xz)(ex )(ey )(ez ) . 2 2 2 (1 2z2)(xy)(ex )(ey )(ez )

We have

1 2

= 2 2 < 0,

1 4 2

1 6 2

= 2 2 > 0.

Lagrange Multipliers

93

This means that if an even number of the variables is negative (0 or 2), then we the Hessian is negative denite, and if an odd numbers of the variables is positive (1 or 3), the Hessian is positive denite. We conclude that we have local maxima at 1 1 1 1 1 1 1 1 1 1 1 1 ( , , ), ( , , ), ( , , ), ( , , ) 2 2 2 2 2 2 2 2 2 2 2 2 and local minima at 1 1 1 1 1 1 1 1 1 1 1 1 ( , , ), ( , , ), ( , , ), ( , , ). 2 2 2 2 2 2 2 2 2 2 2 2

2.6

Lagrange Multipliers

In some situations we wish to optimise a function given a set of constraints. For such cases, we have the following. 211 Theorem Let A Rn and let f : A R, g : A R be functions whose respective derivatives are continuous. Let g(x0) = c0 and let S = g1(c0) be the level set for g with value c0, and assume g(x0) = 0. If the restriction of f to S has an extreme point at x0, then R such that f(x0) = g(x0).

! The above theorem only locates extrema, it does not say anything
concerning the nature of the critical points found.

212 Example Optimise f : R2 R, f(x, y) = x2 y2 given that x2 + y2 = 1. Solution: Let g(x, y) = x2 + y2 1. We solve f for x, y, . This requires 2x 2x = . 2y 2y x x = g y y

94

Chapter 2

From 2x = 2x we get either x = 0 or = 1. If x = 0 then y = 1 and = 1. If = 1, then y = 0, x = 1. Thus the potential critical points are (1, 0) and (0, 1). If x2 + y2 = 1 then f(x, y) = x2 (1 x2) = 2x2 1 1, and f(x, y) = 1 y2 y2 = 1 2y2 1. Thus (1, 0) are maximum points and (0, 1) are minimum points. 213 Example Find the maximum and the minimum points of f(x, y) = 4x + 3y, subject to the constraint x2 + 4y2 = 4, using Lagrange multipliers. Solution: Putting g(x, y) = x2 + 4y2 4 we have f(x, y) = g(x, y) = 4 2x = . 3 8y x 16 = . y 3

Thus 4 = 2x, 3 = 8y. Clearly then = 0. Upon division we nd Hence x2 + 4y2 = 4 =

The maximum is clearly then 4

3 256 2 16 y + 4y2 = 4 = y = , x = 9 73 73. 16 73 +3 3 73 = 73,

and the minimum is 73. 214 Example Let a > 0, b > 0, c > 0. Determine the maximum and minix2 y2 z2 x y z mum values of f(x, y, z) = + + on the ellipsoid 2 + 2 + 2 = 1. a b c a b c x2 y2 z2 Solution: We use Lagrange multipliers. Put g(x, y, z) = 2 + 2 + 2 1. a b c Then

Lagrange Multipliers

95

a b c x2 y2 z2 It follows that = 0. Hence x = ,y = ,z = . Since 2 + 2 + 2 = 1, 2 2 2 a b c 3 3 we deduce 2 = 1 or = . Since a, b, c are positive, f will have a 4 2 maximum when all x, y, z are positive and a minimum when all x, y, z are negative. Thus the maximum is when a b c x = ,y = ,z = , 3 3 3 and 3 f(x, y, z) 3 and the minimum is when a b c x = , y = , z = , 3 3 3 and 3 f(x, y, z) . 3 215 Example Let a > 0, b > 0, p > 1. Maximise f(x, y) = ax + by subject to the constraint xp + yp = 1. Solution: Put g(x, y) = xp + yp 1. We need a = pxp1 and b = pyp1. Clearly then , = 0. We then have x= Thus 1 = xp + yp = a p
p/(p1)

2x/a2 1/a f(x, y, z) = g(x, y, z) 1/b = 2y/b2 . 1/c 2z/c2

a p

1/(p1)

, y=

b p

1/(p1)

b p

p/(p1)

96 which gives = This gives x= a p


p/(p1)

Chapter 2

b p

p/(p1)

(p1)/p

a1/(p1) b1/(p1) , y = . (a1/(p1) + b1/(p1))1/p (a1/(p1) + b1/(p1))1/p

Since f is non-negative, these points dene a maximum for f and so ax + by bp/(p1) ap/(p1) + . (a1/(p1) + b1/(p1))1/p (a1/(p1) + b1/(p1))1/p

216 Example Find the extrema of f(x, y, z) = x2 + y2 + z2 subject to the constraint (x 1)2 + (y 2)2 + (z 3)2 = 4. Solution: Let g(x, y, z) = (x 1)2 + (y 2)2 + (z 3)2 4. We solve x x f y = g y z z 2x 2(x 1) 2y = 2(y 2) . 2z 2(z 3)

for x, y, . This requires

Clearly, = 1. This gives x =

2 3 ,y= , and z = . Substitut1 1 1 ing into (x 1)2 + (y 2)2 + (z 3)2 = 4, we gather that 1 1
2

2 2 1

3 3 1

= 4,

from where =1

14 . 2

Arithmetic Mean-Geometric Mean Inequality This gives the two points (x, y, z) = and 2 4 6 1 + ,2 + ,3 + 14 14 14

97

2 4 6 1 ,2 ,3 . 14 14 14 12 14 The rst point gives an absolute maximum of 18 + and the second 7 12 14 an absolute minimum of 18 . 7 (x, y, z) = 217 Example Optimise f(x, y, z) = x + y + z subject to x2 + y2 = 2, and x + z = 1. Solution: Put g(x, y, z) = x2 + y2 2, h(x, y, z) = x + z 1. We must nd , such that f(x, y, z) = g(x, y, z) + h(x, y, z), which translates into 1 = 2x + , 1 = 2y, 1 = , and x2 + y2 = 1, x + z = 1. We deduce that x = 0, y = 2, z = 1. We may shew that (0, 2, 1) yields a maximum and that (0, 2, 1) yields a minimum.

2.7

Arithmetic Mean-Geometric Mean Inequality

218 Denition Let a1, a2, an be n non-negative real numbers. Their arithmetic mean or average is a1 + a2 + + a n . n

98 Their geometric mean is (a1a2 an)1/n.

Chapter 2

The Arithmetic-Mean-Geometric-Mean Inequality (AMGM) asserts that (a1a2 an)1/n with equality if and only if a1 = a2 = = an. We will see that this simple inequality allows us to solve many extremum problems without using calculus. We will give two proofs of AMGM, one using Lagrange Multipliers, and one using one-variable calculus due to George Polya. 219 Theorem Let a1, a2, an be n non-negative real numbers. Then (a1a2 an)1/n with equality if and only if a1 = a2 = = an. a1 + a2 + + a n , n a1 + a2 + + a n , n

Proof (Using Lagrange Multipliers) Let f(x) = f(x1, x2, . . . , xn) = x1x2 xn, g(x) = g(x1, x2, . . . , xn) = x1 + x2 + + xn, with x1 0, x2 0, . . . , xn 0 and x1 + x2 + + xn = S. Observe that if any of the xi = 0, then f vanishes. So we may assume that no xi vanishes. Observe that i 1 i n, f(x) = g(x) = f(x) = x1x2 xi1xi+1 xn = . xi

Arithmetic Mean-Geometric Mean Inequality Since f(x) = 0, the equality means that xi =

99

S . Since f is non-negative on the stipulated conditions, n S the extrema at xi = is a maximum, and so n x1x2 xn S n
n

f(x) = imposes x1 = x2 = = xn. This xi

x1 + x2 + + xn n

, u

which gives Arithmetic-Mean-Geometric-Mean Inequality. For Polyas proof we need the following lemma. 220 Lemma For all real numbers x it is veried that x ex1, with equality only when x = 1. Proof Put f : R R, f(x) = ex1 x. Clearly f(1) = e0 1 = 0. Now, f (x) = ex1 1, f (x) = ex1.

If f (x) = 0 then ex1 = 1 implying that x = 1. Thus f has a single minimum point at x = 1. Thus for all real numbers x 0 = f(1) f(x) = ex1 x, which gives the desired result. Proof (George P olya) Put Ak = nak , a1 + a2 + + a n nnGn , (a1 + a2 + + an)n u

and Gn = a1a2 an. Observe that A1A2 An =

100 and that A1 + A2 + + An = n. By the preceding lemma, we have A1 exp(A1 1), A2 exp(A2 1), . . . An exp(An 1).

Chapter 2

Since all the quantities involved are non-negative, we may multiply all these inequalities together, to obtain, A1A2 An exp(A1 + A2 + + An n). In view of the observations above, the preceding inequality is equivalent to nnGn exp(n n) = e0 = 1. n (a1 + a2 + + an) We deduce that Gn which is equivalent to (a1a2 an)1/n a1 + a2 + + a n . n a1 + a2 + + a n n
n

Now, for equality to occur, we need each of the inequalities Ak exp(Ak 1) to hold. This occurs, in view of the preceding lemma, if and only if Ak = 1, k, which translates into a1 = a2 = . . . = an. This completes the proof. u 221 Example Let f(x) = (a + x)5(a x)3, x [a; a]. Find the maximum value of f by means of AMGM.

Arithmetic Mean-Geometric Mean Inequality

101

Solution: If x [a; a], then a + x 0 and a x 0 thus we may use a+x ax AMGM with n = 8, a1 = a2 = = a5 = and a6 = a7 = a8 = . 5 3 We then deduce
5 3

a+x 5 whence

ax 3

a+x 5

+3 8

8 ax 3 = a 4

5533a8 , 48 ax a+x = . and equality holds if and only if 5 3 f(x) 222 Example For any positive integer n > 1 we have 1 3 5 (2n 1) < nn. For, by AMGM, 1 3 5 (2n 1) < 1 + 3 + 5 + + (2n 1) n
n

n2 n

= nn.

Notice that since the factors are unequal we have strict inequality. 223 Example The sequence xn = For the set of n + 1 numbers 1, 1 + has arithmetic mean 1+ and geometric mean 1+ 1 n
n/(n+1)

1+

1 n

, n = 1, 2, . . . is increasing.

1 1 1 ,1 + ,...,1 + , n n n 1 n+1

102 Therefore 1 1+ > n+1 that is to say 1+ or xn+1 > xn, giving the result. 1 n+1
n+1

Chapter 2

1 1+ n

n/(n+1)

>

1+

1 n

224 Example Find the volume of the largest rectangular box with sides parallel to the coordinate planes which can be inscribed in the ellipsoid x a Here a > 0, b > 0, c > 0. Solution: Let 2x, 2y, 2z be the dimensions of this box. We seek to maximise x2 y2 z2 8xyz. Putting n = 3, x1 = 2 , x2 = 2 , x3 = 2 , we have a b c x2 y2 z2 = x1x2x3 a2 b2 c2 Thus 8xyz x1 + x2 + x3 3 8abc 3 3
3 2

y b

z c

= 1.

1 . 27

225 Denition Let a1 > 0, a2 > 0, . . . , an > 0. Their harmonic mean is given by n . 1 1 1 + + + a1 a2 an As a corollary to AMGM we obtain

Arithmetic Mean-Geometric Mean Inequality

103

226 Corollary (Harmonic Mean-Geometric Mean Inequality) Let b1 > 0, b2 > 0, . . . , bn > 0. Then n 1 1 1 + + + b1 b2 bn (b1b2 bn)1/n.

Proof This follows by putting ak =

1 in Theorem 219. For then bk 1 1 1 + + + b b2 bn 1 . n

1 1 1 b1 b2 bn

1/n

Combining Theorem 219 and Corollary 226, we deduce 227 Corollary (Harmonic Mean-Arithmetic Mean Inequality) Let b1 > 0, b2 > 0, . . . , bn > 0. Then n b1 + b2 + + bn . 1 1 1 n + + + b1 b2 bn 228 Example Let ak > 0, and s = a1 + a2 + + an. Prove that n2 s s ak n1 k=1 and ak n . s a n 1 k k=1 Solution: Put bk = s . Then s ak 1 = b k k=1
n n n

s ak =n1 s k=1

104 and from Corollary 227, n n1


n k=1

Chapter 2

s s ak , n ak s 1 = ,we s ak s ak

from where the rst inequality is proved. Since ak = s ak k=1 have =


k=1 2 n n

k=1 n

s 1 s ak s s ak n

n n n1 n . = n1

Chapter

Integration
3.1 Differential Forms

We will now consider integration in several variables. In order to smooth our discussion, we need to consider the concept of differential forms. 229 Denition Consider n variables x1, x2, . . . , xn in n-dimensional space (used as the names of the axes), and let a1j a2j aj = . Rn, 1 j k, . . anj

be k n vectors in Rn. Moreover, let {j1, j2, . . . , jk} {1, 2, . . . , n} be a collection of k sub-indices. An elementary k-differential form (k > 1) acting on the vectors aj, 1 j k is dened and denoted by aj1 1 aj1 2 aj1 k aj 1 aj 2 aj k 2 2 2 dxj1 dxj2 dxjk (a1, a2 , . . . , ak ) = det . . . . . . . . . . ajk 1 ajk 2 ajk k 105

106

Chapter 3

In other words, dxj1 dxj2 dxjk (a1, a2 , . . . , ak ) is the xj1 xj2 . . . xjk component of the signed k-volume of a k-parallelotope in Rn spanned by a1, a2 , . . . , ak .

! By virtue of being a determinant, the wedge product of differential


forms has the following properties anti-commutativity: da db = db da. linearity: d(a + b) = da + db. scalar homogeneity: if R, then da = da. associativity: (da db) dc = da (db dc).1

! Anti-commutativity yields
da da = 0.

230 Example Consider 1 a = 0 R3. 1 dx(a) = det(1) = 1, dy(a) = det(0) = 0, dz(a) = det(1) = 1, are the (signed) 1-volumes (that is, the length) of the projections of a onto the co-ordinate axes. 231 Example Consider the vectors in R4 1 1 0 2 a= 3 , b = 1 . 2 4
1

Then

Notice that associativity does not hold for the wedge product of vectors.

Differential Forms Then, for example, 1 1 2 0 1 1 dx1 dx2(a, b) = d 3 1 = det 2 0 = 2, 4 2 1 1 2 0 2 0 dx2 dx4(a, b) = d 3 1 = det 4 2 = 4, 4 2 1 1 2 0 3 1 dx3 dx1(a, b) = d 3 1 = det 1 1 = 4. 4 2 232 Example Consider the vectors in R4 1 1 1 2 0 0 a= 3 , b = 1 , c = 0 . 4 2 0 Then, for example, dx1 dx2(a, b, c)

107

is undened, since for a 2-form we need two vectors. We have, however, 1 1 1 1 1 1 2 0 0 dx1 dx2 dx4(a, b, c) = d 3 1 0 = det 2 0 0 = 4. 4 2 0 4 2 0 233 Example In R3 we have dx dy dx = 0, since we have a repeated variable. 234 Example In R3 we have dx dz + 5dz dx + 4dx dy dy dx + 12dx dx = 4dx dz + 5dx dy.

108

Chapter 3

! In order to avoid redundancy we will make the convention that if a sum

of two or more terms have the same differential form up to permutation of the variables, we will simplify the summands and express the other differential forms in terms of the one differential form whose indices appear in increasing order.

235 Example In R5 we have dx5 dx4 dx2 = dx4 dx5 dx2 = dx4 dx2 dx5 = dx2 dx4 dx5, d x4 d x2 d x5 = d x2 d x4 d x5 dx2 dx5 dx4 = dx2 dx4 dx5. Hence we write dx2 dx5 dx4 + 5dx4 dx2 dx5 + 2dx5 dx4 dx2 as 8dx2 dx4 dx5. 236 Denition A 0-differential form in Rn is simply a differentiable function in Rn. 237 Denition A k-differential form eld in Rn is an expression of the form =
1j1 j2 jk n

aj1 j2 ...jk dxj1 dxj2 dxjk ,

where the aj1 j2 ...jk are differentiable functions in Rn. 238 Example g(x, y, z, w) = x + y2 + z3 + w4 is a 0-form in R4. 239 Example An example of a 1-form eld in R3 is = xdx + y2dy + xyz3dz.

Differential Forms 240 Example An example of a 2-form eld in R3 is = x2dx dy + y2dy dz + dz dx. 241 Example An example of a 3-form eld in R3 is = (x + y + z)dx dy dz.

109

242 Theorem If is an l-form and is a k-form then is a l + k-form and = (1)lk . 243 Example The product of the 1-form elds in R3 1 = ydx + xdy, 2 = 2xdx + 2ydy, is 1 2 = (2x2 + 2y2)dx dy. 244 Theorem Let 0 k n, and let k(Rn) denote the space of kdifferential forms in Rn. Then k(Rn) is a vector space of dimension n k = n! . k!(n k)! n k elementary forms

A basis for this space is given by the

dxj1 dxj2 dxjn , 1 j1 < j2 < < jk n. 245 Example In R3, 1(R3) has dimension is {dx, dy, dz}. 246 Example In R3, 2(R3) has dimension is {dx dy, dx dz, dy dz}. 3 2 = 3, and a basis for 2(R3) 3 1 = 3, and a basis for 1(R3)

110 247 Example In R3, 3(R3) has dimension is {dx dy dz}. 248 Example In R4, 2(R4) has dimension is 4 2 3 3

Chapter 3 = 1, and a basis for 3(R3)

= 6, and a basis for 2(R4)

{dx1 dx2, dx1 dx3, dx1 dx4, dx2 dx3, dx2 dx4, dx3 dx4}. 249 Denition Let f(x1, x2, . . . , xn) be a 0-form in Rn. The exterior derivative df of f is
n

df =
i=1

f dxi. xi

Furthermore, if = f(x1, x2, . . . , xn)dxj1 dxj2 dxjk is a k-form in Rn, the exterior derivative d of is the (k + 1)-form d = df(x1, x2, . . . , xn) dxj1 dxj2 dxjk . 250 Example If in R2, = x3y4, then d(x3y4) = 3x2y4dx + 4x3y3dy. 251 Example If in R2, = x2ydx + x3y4dy then d = = = = d(x2ydx + x3y4dy) (2xydx + x2dy) dx + (3x2y4dx + 4x3y3dy) dy x2dy dx + 3x2y4dx dy (3x2y4 x2)dx dy

252 Example Consider the change of variables x = u + v, y = uv. Then dx = du + dv,

Integrating in 0(Rn) dy = vdu + udv, whence dx dy = (u v)du dv.

111

253 Example Consider the transformation of co-ordinates xyz into uvw co ordinates given by u = x + y + z, v = Then du = dx + dy + dz, z y dv = dy + dz, 2 (y + z) (y + z)2 dw = x x y+z dx + dy + dz. 2 2 (x + y + z) (x + y + z) (x + y + z)2 y+z z , w= . y+z x+y+z

Multiplication gives du dv dw = y(y + z) zx (y + z)2(x + y + z)2 (y + z)2(x + y + z)2 z(y + z) xy + dx dy dz 2 2 2 (y + z) (x + y + z) (y + z) (x + y + z)2 z2 y2 zx xy dx dy dz. = (y + z)2(x + y + z)2

3.2

Integrating in 0(Rn)

254 Denition A 0-dimensional oriented manifold of Rn is simply a point x Rn, with a choice of the + or sign. A general oriented 0-manifold is a union of oriented points. 255 Denition Let M = +{b} {a} be an oriented 0-manifold, and let be a 0-form. Then = (b) (a).
M

112

Chapter 3

! x has opposite orientation to +x and


=
x +x

256 Example Let M = {(1, 0, 0)} +{(1, 2, 3)} {(0, 2, 0)}2 be an oriented 0-manifold, and let = x + 2y + z2. Then = ((1, 0, 0)) + (1, 2, 3) (0, 0, 3) = (1) + (14) (4) = 17.
M

3.3

Integrating in 1(Rn)

257 Denition A 1-dimensional oriented manifold of Rn is simply an oriented smooth curve Rn, with a choice of a + orientation if the curve traverses in the direction of increasing t, or with a choice of a sign if the curve traverses in the direction of decreasing t. A general oriented 1-manifold is a union of oriented curves.

! The curve has opposite orientation to and


=

We now turn to the problem of integrating 1-forms. 258 Example Calculate xydx + (x + y)dy

where is the parabola y = x2, x [1; 2] oriented in the positive direction.


Do not confuse, say, {(1, 0, 0)} with (1, 0, 0) = (1, 0, 0). The rst one means that the point (1, 0, 0) is given negative orientation, the second means that (1, 0, 0) is the additive inverse of (1, 0, 0).
2

Integrating in 1(Rn) Solution: We parametrise the curve as x = t, y = t2. Then xydx + (x + y)dy = t3dt + (t + t2)dt2 = (3t3 + 2t2)dt, whence =
2

113

(3t3 + 2t2)dt
1 2

2 3 3 4 = t + t 3 4 1 69 = . 4 What would happen if we had given the curve above a different parametrisation? First observe that the curve travels from (1, 1) to (2, 4) on the parabola y = x2 . These conditions are met with the parametrisation x = t 1, y = ( t 1)2, t [0; 9]. Then 3 2 1 ) t 1 ) + (( t 1 ) + ( t 1 ) t 1)2 d ( ) d ( xydx + (x + y)dy = ( t = (3( t 1)3 + 2( t 1)2)d( t 1) 1 = (3( t 1)3 + 2( t 1)2)dt, 2 t whence

1 (3( t 1)3 + 2( t 1)2)dt 0 2 t 9 3t2 7t3/2 5t = + t 4 3 2 0 69 = , 4


9

as before.

! It turns out that if two different parametrisations of the same curve


have the same orientation, then their integrals are equal. Hence, we only need to worry about nding a suitable parametrisation.

259 Example Calculate the line integral y sin xdx + x cos ydy,

where is the line segment from (0, 0) to (1, 1) in the positive direction.

114

Chapter 3

Solution: This line has equation y = x, so we choose the parametrisation x = y = t. The integral is thus
1

y sin xdx + x cos ydy =


0

(t sin t + t cos t)dt


1

= [t(sin x cos t)]1 0 = 2 sin 1 1, upon integrating by parts. 260 Example Calculate the path integral x+y xy dy + 2 dx 2 2 x +y x + y2
0

(sin t cos t)dt

around the closed square = ABCD with A = (1, 1), B = (1, 1), C = (1, 1), and D = (1, 1) in the direction ABCDA. Solution: On AB, y = 1, dy = 0, on BC, x = 1, dx = 0, on CD, y = 1, dy = 0, and on DA, x = 1, dx = 0. The integral is thus =
AB 1

+
BC

+
CD 1 1

+
DA

x+1 dx + 2 1 x + 1
1

x1 dx + 2 1 x +1 1 1 dx = 4 2 1 x + 1 = = 4 arctan x|1 1 = 2.

y1 dy + y2 + 1

y+1 dy 2 1 y + 1

! When the integral is along a closed path, like in the preceding example, it is customary to use the symbol

rather than

. The positive

direction of integration is that sense that when traversing the path, the area enclosed by the curve is to the left of the curve.

Integrating in 1(Rn) 261 Example Calculate the path integral x2dy + y2dx,

115

where is the ellipse 9x2 + 4y2 = 36 traversed once in the positive sense. Solution: Parametrise the ellipse as x = 2 cos t, y = 3 sin t, t [0; 2]. Observe that when traversing this closed curve, the area of the ellipse is on the left hand side of the path, so this parametrisation traverses the curve in the positive sense. We have
2

((4 cos2 t)(3 cos t) + (9 sin t)(2 sin t))dt


0 2

=
0

(12 cos3 t 18 sin3 t)dt

= 0. 262 Example Find

zdx + xdy + ydz where is the intersection of the

sphere x2 + y2 + z2 = 1 and the plane x + y = 1, traversed in the positive direction. Solution: The curve lies on the sphere, and to parametrise this curve, we dispose of one of the variables, y say, from where y = 1 x and x2 + y2 + z2 = 1 give x2 + (1 x)2 + z2 = 1 = 2x2 2x + z2 = 0 1 = 2 x 2 1 = 4 x 2
2 2

+ z2 =

1 2

+ 2z2 = 1.

So we now put x= 1 cos t + , 2 2

sin t z= , 2

y=1x=

1 cos t . 2 2

116

Chapter 3

We must integrate on the side of the plane that can be viewed from the 1 point (1, 1, 0) (observe that the vector 1 is normal to the plane). On 0
2

1 the zx-plane, 4 x + 2z2 = 1 is an ellipse. To obtain a positive 2 parametrisation we must integrate from t = 2 to t = 0 (this is because when you look at the ellipse from the point (1, 1, 0) the positive x-axis is to your left, and not your right). Thus zdx + xdy + ydz =

sin t 1 cos t d + 2 2 2 2 0 1 cos t 1 cos t + d + 2 2 2 2 2 0 sin t 1 cos t d + 2 2 2 2 0 sin t cos t cos t sin t 1 = + + 4 4 2 2 2 2 2 = . 2

dt

Given a curve how can we nd its length? The idea, as seen in gure 3.1 is to consider the projections dxi, 1 i n at each point. The length of the vector d x1 dx 2 dx = . . . d xn

is ||dx|| =

(dx1)2 + (dx2)2 + + (dxn)2.

Integrating in 1(Rn) Hence the length of is given by ||dx|| =


117

(dx1)2 + (dx2)2 + + (dxn)2.

(3.1)

Similarly, suppose that is a simple closed curve in R2. How do we nd the (oriented) area of the region it encloses? The idea, borrowed from nding areas of polygons, is to split the region into triangles, each of area x dx 1 1 = (xdy ydx), det 2 2 y dy

and to sum over the closed curve, obtaining a total oriented area of 1 2 x dx 1 = det 2 y dy (xdy ydx).

(3.2)

Here

denotes integration around the closed curve.

dx

Figure 3.1: Length of a curve.

Figure 3.2: Area enclosed by a simple closed curve

118

Chapter 3

263 Example Let (A, B) R2, A > 0, B > 0. Find a parametrisation of the ellipse y2 x2 : {(x, y) R2 : 2 + 2 = 1}. A B Furthermore, nd an integral expression for the perimeter of this ellipse and nd the area it encloses. Solution: Consider the parametrisation : [0; 2] R2, with x A cos t = . y B sin t

This is a parametrisation of the ellipse, for

y2 A2 cos2 t B2 sin2 t x2 + = + = cos2 t + sin2 t = 1. 2 2 2 2 A B A B Notice that this parametrisation goes around once the ellipse counterclockwise. The perimeter of the ellipse is given by
2

||dx|| =
0

A2 sin2 t + B2 cos2 t dt.

The above integral is an elliptic integral and quite hard to evaluate. We will have better luck with the area of the ellipse, which is given by 1 2 (xdy ydx) =

1 (A cos t d(B sin t) B sin t d(A cos t)) 2 1 2 (AB cos2 t + AB sin2 t)) dt = 2 0 1 2 = AB dt 2 0 = AB.

Integrating in 1(Rn) 264 Example Find a parametric representation for the astroid : {(x, y) R2 : x2/3 + y2/3 = 1},

119

in gure 3.3. Find the perimeter of the astroid and the area it encloses. Solution: Take
3 x cos t = 3 y sin t

with t [0; 2] R2. with Then The perimeter of the astroid is


2

x2/3 + y2/3 = cos2 t + sin2 t = 1.

||dx|| =
0

9 cos4 t sin2 t + 9 sin4 t cos2 t dt


2

=
0

3| sin t cos t| dt 3 2 6
0 2

= = =

| sin 2t| dt
0 /2

sin 2t dt 6.

Figure 3.3: Example 264.

120 The area of the astroid is given by 1 2 (xdy ydx) =

Chapter 3

= = = = =

1 (cos3 t d(sin3 t) sin3 t d(cos3 t)) 2 1 2 (3 cos4 t sin2 t + 3 sin4 t cos2 t)) dt 2 0 3 2 (sin t cos t)2 dt 2 0 3 2 (sin 2t)2 dt 8 0 3 2 (1 cos 4t) dt 16 0 3 . 8

265 Example Find the area enclosed by the curve x = sin3 t, y = (cos t)(1+ sin2 t). Solution: Observe that the parametrisation traverses the curve once clockwise if t [0; 2]. The area is given by 1 2 x dx = 1 xdy ydx det 2 y dy 4 = 2
0 /2

(sin3 t( sin t(1 + sin2 t) + 2 sin t cos2 t)

cos t(1 + sin2 t)(3 sin2 t cos t))dt


0

= 2 = 2

(3 sin2 t + sin4 t)dt


/2 0 /2

9 1 + cos 2t + cos 4t dt 8 8

9 . 8

Integrating in 1(Rn)

266 Denition Let be a smooth curve. The integral f(x)||dx||

is called the path integral of f along .


3

-3

-2

-1

Figure 3.4: Example 267.


2 1 1 2 3 -1 -2 -3

121

267 Example Find

x||dx|| where is the triangle starting at A : (1, 1)

to B : (2, 2), and ending in C : (1, 2). Solution: The lines passing through the given points have equations LAB : x 4 y= , and LBC : y = 4x + 6. On LAB 3 x||dx|| = x (dx)2 + (dy)2 = x and on LBC x||dx|| = x (dx)2 + (dy)2 = x( 1 + (4)2)dx = x 17dx. 1 1+ 3
2

x 10dx dx = , 3

122 Hence x||dx|| =


LAB 2

Chapter 3

x||dx|| +
LBC

x||dx||
1 2

x 10dx + 3 1 10 3 17 = . 2 2

x 17dx

3.4

Closed and Exact Forms

268 Lemma (Poincar e Lemma) If is a p-differential form of continuously differentiable functions in Rn then dd = 0. Proof We will prove this by induction on p. For p = 0 if = f(x1, x2, . . . , xn) then d = and dd =
k=1 n n

f d xk xk k=1

f xk
n

d xk d xk

=
k=1 n j=1

2f d xj xjxk

=
1jkn

2f 2f d xj d xk xjxk xkxj

= 0, since is continuously differentiable and so the mixed partial derivatives are equal. Consider now an arbitrary p-form, p > 0. Since such a form

Closed and Exact Forms can be written as =


1j1 j2 jp n

123

aj1 j2 ...jp dxj1 dxj2 dxjp ,

where the aj1 j2 ...jp are continuous differentiable functions in Rn, we have d =
1j1 j2 jp n

daj1 j2 ...jp dxj1 dxj2 dxjp


n

=
1j1 j2 jp n i=1

aj1 j2 ...jp d xi xi

dxj1 dxj2 dxjp ,

it is enough to prove that for each summand d da dxj1 dxj2 dxjp = 0. But d da dxj1 dxj2 dxjp = dda dxj1 dxj2 dxjp +da d dxj1 dxj2 dxjp = da d dxj1 dxj2 dxjp , since dda = 0 from the case p = 0. But an independent induction argument proves that d dxj1 dxj2 dxjp = 0, u

completing the proof.

269 Denition A differential form is said to be exact if there is a continuously differentiable function F such that dF = . 270 Example The differential form xdx + ydy is exact, since xdx + ydy = d 1 2 (x + y2) . 2

124 271 Example The differential form ydx + xdy is exact, since ydx + xdy = d (xy) . 272 Example The differential form x2 is exact, since x y d x + dy = d x2 + y2 x2 + y2 1 loge(x2 + y2) . 2 y x dx + 2 dy 2 +y x + y2

Chapter 3

says that for certain domains (called d = ddF = 0. A result of Poincare star-shaped domains) the converse is also true, that is, if d = 0 on a star-shaped domain then is exact.

! Let = dF be an exact form. By the Poincare Lemma Theorem 268,

273 Example Determine whether the differential form = is exact. Solution: Assume there is a function F such that dF = . By the Chain Rule dF = This demands that F F dx + dy. x y 2x(1 ey) ey d x + dy (1 + x2)2 1 + x2

2x(1 ey) F , = x (1 + x2)2

Closed and Exact Forms

125

F ey . = y 1 + x2 We have a choice here of integrating either the rst, or the second expression. Since integrating the second expression (with respect to y) is easier, we nd ey F(x, y) = + (x), 1 + x2 where (x) is a function depending only on x. To nd it, we differentiate the obtained expression for F with respect to x and nd 2xey F = + (x). x (1 + x2)2 Comparing this with our rst expression for (x) = that is (x) = where c is a constant. We then take F(x, y) = ey 1 + c. 1 + x2 F , we nd x

2x , (1 + x2)2 1 + c, 1 + x2

274 Example Is there a continuously differentiable function such that dF = = y2z3dx + 2xyz3dy + 3xy2z2dz ? Solution: We have d = (2yz3dy + 3y2z2dz) dx +(2yz3dx + 2xz3dy + 6xyz2dz) dy +(3y2z2dx + 6xyz2dy + 6xy2zdz) dz = 0,

126 so this form is exact in a star-shaped domain. So put dF = Then

Chapter 3

F F F dx + dy + dz = y2z3dx + 2xyz3dy + 3xy2z2dz. x y z

F = y2z3 = F = xy2z3 + a(y, z), x F = 2xyz3 = F = xy2z3 + b(x, z), y F = 3xy2z2 = F = xy2z3 + c(x, y), z Comparing these three expressions for F, we obtain F(x, y, z) = xy2z3. We have the following equivalent of the Fundamental Theorem of Calculus. 275 Theorem Let U Rn be an open set. Assume = dF is an exact form, and a path in U with starting point A and endpoint B. Then
B

=
A

dF = F(B) F(A).

In particular, if is a simple closed path, then = 0.

276 Example Evaluate the integral x2

2x 2y dx + 2 dy 2 +y x + y2

where is the closed polygon with vertices at A = (0, 0), B = (5, 0), C = (7, 2), D = (3, 2), E = (1, 1), traversed in the order ABCDEA. Solution: Observe that d x2 2y 2x dx + 2 dy 2 +y x + y2 = (x2 4xy 4xy dy dx 2 dxdy = 0, 2 2 +y ) (x + y2)2

and so the form is exact in a start-shaped domain. By virtue of Theorem 275, the integral is 0.

Integrating in 2(R2) 277 Example Calculate the path integral (x2 y)dx + (y2 x)dy,

127

where is a loop of x3 + y3 2xy = 0 traversed once in the positive sense. Solution: Since 2 2 (x y) = 1 = (y x), y x

the form is exact, and since this is a closed simple path, the integral is 0. 278 Theorem If is a p-form and is a q-form on n-dimensional space, then d( ) = d + (1)p d.

3.5

Integrating in 2(R2)

279 Denition A 2-dimensional oriented manifold of R2 is simply an open set (region) D R2, where the + orientation is counter-clockwise and the orientation is clockwise. A general oriented 2-manifold is a union of open sets.

! The region D has opposite orientation to D and


=
D D

In this section, unless otherwise noticed, we will choose the positive orientation for the regions considered. This corresponds to using the area form dx dy.

Let D R2. Given a function f : D R, the integral f(x, y)dx dy


D

128 is the sum of all the values of f restricted to D. In particular,

Chapter 3

dx dy
D

is the area of D.

In order to evaluate double integrals, we need the following. 280 Theorem (Fubinis Theorem) Let D = [a; b] [c; d], and let f : A R be continuous. Then
b d d b

f(x, y)dx dy =
D a c

f(x, y)dy dx =
c a

f(x, y)dx dy

Fubinis Theorem allows us to convert the double integral into iterated (single) integrals. 281 Example
1 3

xydx dy =
[0;1][2;3] 0 1 2

xydy dx xy2 2
3

=
0 1

dx
2

=
0

9x 2x dx 2
1 0

5x2 4

5 = . 4 Notice that if we had integrated rst with respect to x we would have ob-

Integrating in 2(R2) tained the same result:


3 2 1 3

129

xydx dy =
0 2 3

x2y 2 y dx 2
3 2

dy
0

=
2

y2 4

5 = . 4 Also, this integral is factorable into x and y pieces meaning that


1 3

xydx dy =
[0;1][2;3] 0

xd x
2

ydy

= = 282 Example We have


4 3 1 4

1 2 5 4

5 2

(x + 2y)(2x + y) dx dy =
0 3 0 4

(2x2 + 5xy + 2y2) dx dy 2 5 + y + 2y2 3 2 dy

=
3

= 283 Example Find

409 . 12

(x + y)(sin x)(sin y)dx dy


D

where D = [0; ]2.

130

Solution: The integral equals x sin x sin ydx dy +


D

-3

-2

-1

Figure 3.5: Example 284. Integration order dydx.


Chapter 3

y sin x sin ydx dy = 2

y sin y dy

sin x dx

= 4.

3 2 1

3 2 1

-3

-2

-1

-1 -2 -3

-1 -2 -3

Figure 3.6: Example 284. Integration order dxdy.

In the cases when the domain of integration is not a rectangle, we decompose so that, one variable is kept constant. 284 Example Find
D

xy dxdy in the triangle with vertices A : (1, 1),

B : (2, 2), C : (1, 2). The lines passing through the given points have equations LAB : y = 3x + 1 x 4 , LBC : y = 4x + 6, LCA : y = . Now, we draw the re3 2 gion carefully. If we integrate rst with respect to y, we must divide the region as in gure 3.5, because there are two upper lines which the upper value of y might be. The lower point of the dashed line is (1, 5/3). The

Integrating in 2(R2) integral is thus


1 (3x+1)/2 2 4x+6

131

x
1 (x4)/3

y dy dx +
1

x
(x4)/3

y dy dx =

If we integrate rst with respect to x, we must divide the region as in gure 3.6, because there are two left-most lines which the left value of x might be. The right point of the dashed line is (7/4, 1). The integral is thus
1 (6y)/4 2 (6y)/4

y
2 43y

x dx dy +
1

285 Example Consider the region inside the parallelogram P with vertices at A : (6, 3), B : (8, 4), C : (9, 6), D : (7, 5), as in gure 3.7. Find 1 3
P
7 6 5 4 3 2 1

xy dx dy.


x dx dy =
(2y1)/3

11 . 8

11 . 8

10

Figure 3.7: Example 285.

Solution: The lines joining the points have equations LAB : y = x , 2

LBC : y = 2x 12, x 3 LCD : y = + , 2 2

132 LDA : y = 2x 9. The integral is thus 1 3


4 3 2y

Chapter 3

xy dxdy+
(y+9)/2

1 3

5 4

(y+12)/2

xy dxdy+
(y+9)/2

1 3

5 4

(y+12)/2

xy dxdy =
2y3

409 . 12

286 Example Find


D

(2x+3y+1) dxdy, where D is the triangle with vertices

at A(1, 1), B(2, 4), and C(1, 3). Solution: The line joining A, and B has equation y = x 2, line joining B, and C has equation y = 7x + 10, and line joining A, and C has equation y = 2x + 1. We split the triangle along the vertical line x = 1, and integrate rst with respect to y. The desired integral is then
1 2x+1

(2x + 3y + 1) dxdy =
D 1

(2x + 3y + 1)dy dx
x2 7x+10 2

+
1 1 x2

(2x + 3y + 1)dy dx 21 2 3 x + 9x 2 2
2 1

=
1

dx

+ = 41 = 3. 287 Example Find

60x2 198x + 156 dx

(xy(x + y))dx dy
D

where D = {(x, y) R2|x 0, y 0, x + y 1}.

Integrating in 2(R2) Solution: The domain of integration is a triangle. The integral equals
1 1x

133

xy(x + y)dx dy =
D 0 1 0

xy(x + y) dy
1x

dx

y2 y3 = x x + dx 2 3 0 0 1 x(1 x)2 (1 x)3 + = x 2 3 0 1 = . 30 288 Example Find 1 dx dy (x + y)4


D

dx

where D = {(x, y) R2|x 1, y 1, x + y 4}. Solution: The integral equals 1 dx dy = (x + y)4


D 3 1 3 4x 1

dy dy (x + y)4
4x

dx

1 (x + y)3 dx 3 1 1 1 1 3 1 dx = 3 3 1 (1 + x) 64 1 = . 48 = 289 Example Find xd x d y


D

where D = {(x, y) R2|y 0, x y + 1 0, x + 2y 4 0}.

134 Solution: The integral equals


2/3 x+1 4 2 x 2

Chapter 3

xd x d y =
D 1 2/3 0

dy x dx +
2/3 4

dy x dx
0

=
1

x(x + 1) dx +
2/3

x 2

x 2

dx

275 . 54

290 Example Find xydx dy


D

where D = {(x, y) R2|y x2, x y2}. Solution: The integral equals


1 x

xydx dy =
D 0 1

x
x2

y dy

dx

=
0

1 x(x x4) dx 2

1 = . 12 291 Example Find loge(1 + x + y)dx dy


D

where D = {(x, y) R2|x 0, y 0, x + y 1}.

Integrating in 2(R2) Solution: Integrating by parts,


1 1x

135

loge(1 + x + y)dx dy =
D 0 1 0

loge(1 + x + y) dy

dx

=
0 1

x [(1 + x + y) loge(1 + x + y) (1 + x + y)]1 dx 0

=
0

(2 loge(2) 1 loge(1 + x) x loge(1 + x) + x) dx 1 . 4

= 292 Example Find

xydx dy
D

where D = {(x, y) R2|x 0, y 0, xy + y + x 1}. Solution: The integral equals 1x 1 1+x xy dy dx


0

xydx dy =
D

= =

(t 1)(t 2)2 dt t2 1 11 = 4 loge 2 . 4

0 2

1 x 2

1x 1+x

dx

293 Example Find loge(1 + x2 + y)dx dy


D

where D = {(x, y) R2|x 0, y 0, x2 + y 1}.

136 Solution: Using integration by parts,


1 1x2

Chapter 3

loge(1 + x + y)dx dy =
D 0 1 0

loge(1 + x2 + y) dy

dx

=
0

(2 loge(2) 1 loge(1 + x2))dx


1

+
0

(x2 loge(1 + x2) + x2) dx

2 8 = loge 2 + . 3 9 3 294 Example Find x2


D

y dx dy +1

where D = {(x, y) R2|x 0, x2 + y2 1}. Solution: The integral is 0. Observe that if (x, y) D then (x, y) D. Also, f(x, y) = f(x, y). 295 Example Find 2x(x2 + y2)dx dy
D

where D = {(x, y) R2|x4 + y4 + x2 y2 1}. Solution: The integral is 0. Observe that if (x, y) D then (x, y) D. Also, f(x, y) = f(x, y). 296 Example Find |x y|dx dy
D

where D = {(x, y) R2||x| 1, |y| 1}.

Integrating in 2(R2) Solution: Let D1 = {(x, y) R2| 1 x 1, x y}, D2 = {(x, y) R2| 1 x 1, x > y}. Then D = D1 D2, D1 D2 = and so f(x, y)dx dy =
D D1

137

f(x, y)dx dy +
D2

f(x, y)dx dy.

By symmetry, f(x, y)dx dy =


D1 D2

f(x, y)dx dy,

and so f(x, y)dx dy = 2


D D1 1

f(x, y)dx dy
1

= 2
1 1 x

(y x) dy (1 2x + x2) dx

dx

= 8 . = 3

297 Example Find


4 0 y

ey/x dx dy.

y/2

Solution: We have 0 y 4, y x y = 0 x 2, 2 x2 y 2x.

138 We then have


4 0 y 2 2x

Chapter 3

ey/xdx dy =
0 2 x2

ey/xdy dx xey/x
0 2 2x x2

y/2

= =
0

dx

(xe2 xex) dx

= 2e2 (2e2 e2 + 1) = e2 1

298 Example Find


2 1 x

sin
x

x dy 2y

dx +
2

sin
x

x dy 2y

dx.

Solution: Upon splitting the domain of integration, we nd that the integral equals
2 1 y2 y

x sin dx dy = 2y

2 1

2y x cos 2y
2

y2

dy
y

2y y cos dy 2 1 4( + 2) = , 3

upon integrating by parts. 299 Example Find the area of the region R = {(x, y) R2 : x+ y 1, 1x+ 1 y 1}.

Integrating in 2(R2) Solution: The area is given by


1 1(1 1x)2 (1 x)2 1

139

dx dy =
D 0

dy dx

= 2
0

( 1 x + x 1)dx

2 . = 3 300 Example Find


D

xy dx dy, where

D = {(x, y) R2 : y 0, (x + y)2 2x}. 1 Solution: Observe that x (x + y)2 0. Hence we may take the positive 2 square root giving y 2x x. Since y 0, we must have 2x x 0 which means that x 2. The integral equals
2 0 0 2xx

xydy dx = =

2 3

2 0

x( 2x x)3/2dx

4 2 2 u (u 2 u2)3/2du 3 0 1 1 (1 v2)3/2(1 + v)2dv = 6 1 1 /2 = cos4 (1 + sin2 )d 6 /2 7 = . 96

301 Example A rectangle R on the plane is the disjoint union R = N k=1Rk of rectangles Rk. It is known that at least one side of each of the rectangles Rk is an integer. Shew that at least one side of R is an integer.

140 Solution: Observe that


a

Chapter 3

sin 2x dx =
0

if a is an integer

Thus

a 0

1 (1 cos 2a) if a is not an integer 2

Now

sin 2x dx = 0 a is an integer.
N

sin 2x sin 2y dx dy = 0
k=1 Rk

since at least one of the sides of each Rk is an integer. Since


N

sin 2x sin 2y dx dy =
R k=1 Rk

sin 2x sin 2y dx dy,

we deduce that at least one of the sides of R is an integer, nishing the proof. 302 Example Evaluate
1 0 1 0 1

(x1x2 xn)dx1 dx2 . . . dxn.

Solution: We have
1 0 1 0 1 n 1

(x1x2 xn)dx1 dx2 . . . dxn =

xkdxk =
k=1 0

1 1 = n. 2 2 k=1

303 Example Evaluate


1 0 1 0 1

(x1 + x2 + + xn) dx1 dx2 . . . dxn.

Change of Variables in 2(R2) Solution: This is


1 0 1 0 1 n n 1 1 0 1

141

xk d x1 d x2 . . . d xn =
0 k=1

= =

1 2 k=1 n . 2

k=1 0 n

xkdx1 dx2 . . . dxn


0

304 Example (Putnam Exam 1965) Evaluate


1 n + 0 1 1 0 1

lim

cos2
0

(x1 + x2 + + xn) dx1 dx2 . . . dxn. 2n

Solution: Make the change of variables xk = 1 yk. Then


1 0 1

I=
0

cos2
0

(x1 + x2 + + xn) dx1 dx2 . . . dxn 2n

equals
1 0 1 0

sin2
0

(y1 + y2 + + yn) dy1 dy2 . . . dyn. 2n

1 Since sin2 t + cos2 t = 1, we have 2I = 1, and so I = . 2

3.6

Change of Variables in 2(R2)

We now perform a multidimensional analogue of the change of variables theorem in one variable. 305 Theorem Let (D, ) (Rn)2 be open, bounded sets in Rn with volume and let g : D be a continuously differentiable bijective mapping such 1 g = 0, and both | det J g |, that det J are bounded on . For u u g| | det J u g| is integrable on and f : D R bounded and integrable, f g| det J u
D

f =

g |, f g| det J u

142

that is

7 6 5 4 3 2 1


Chapter 3 f(x1, x2, . . . , xn)dx1 dx2 . . . dxn
D

g|du1 du2 . . . dun. f(g(u1, u2, . . . , un))| det J u

7 6 5 4 3 2 1

10

10

Figure 3.8: Example 306. xy-plane.

Figure 3.9: Example 306. uv-plane.

One normally chooses changes of variables that map into rectangular regions, or that simplify the integrand. Let us start with a rather trivial example. 306 Example Evaluate the integral
4 3 1

(x + 2y)(2x + y)dx dy.


0

Solution: Observe that we have already computed this integral in example 282. Put u = x + 2y = du = dx + 2dy, v = 2x + y = dv = 2dx + dy,

Change of Variables in 2(R2) giving du dv = 3dx dy. Now, u 1 2 x = v 2 1 y

143

is a linear transformation, and hence it maps quadrilaterals into quadrilaterals. The corners of the rectangle in the area of integration in the xy-plane are (0, 3), (1, 3), (1, 4), and (0, 4), (traversed counter-clockwise) and they map into (6, 3), (7, 5), (9, 6), and (8, 4), respectively, in the uv-plane (see gure 3.9). The form dx dy has opposite orientation to du dv so we use dv du = 3dx dy instead. The integral sought is 1 3
P

uv dv du =

409 , 12

from example 285. 307 Example The integral


1

(x y )dx dy =
[0;1]2 0

1 y4 dy = 0. 5

Evaluate it using the change of variables u = x2 y2, v = 2xy. Solution: First we nd du = 2xdx 2ydy, dv = 2ydx + 2xdy, and so du dv = (4x2 + 4y2)dx dy. We now determine the region into which the square D = [0; 1]2 is mapped. We use the fact that boundaries will be mapped into boundaries. Put AB = {(x, 0) : 0 x 1},

144 BC = {(1, y) : 0 y 1}, CD = {(1 x, 1) : 0 x 1}, DA = {(0, 1 y) : 0 y 1}.

Chapter 3

On AB we have u = x, v = 0. Since 0 x 1, AB is thus mapped into the line segment 0 u 1, v = 0. v2 On BC we have u = 1 y2, v = 2y. Thus u = 1 . Hence BC is 4 v2 mapped to the portion of the parabola u = 1 , 0 v 2. 4 On CD we have u = (1 x)2 1, v = 2(1 x). This means that u = v2 1, 0 v 2. 4 Finally, on DA, we have u = (1 y)2, v = 0. Since 0 y 1, DA is mapped into the line segment 1 u 0, v = 0. The region is thus the v2 v2 area in the uv plane enclosed by the parabolas u 1, u 1 with 4 4 1 u 1, 0 v 2. We deduce that (x4 y4)dx dy =
[0;1]2

(x4 y4)

4(x2

1 du dv + y2)

1 4 1 = 4 1 = 4 = = 0, as before. 308 Example Find e(x


D
3 +y3 )/xy

(x2 y2)du dv

udu dv
2 0 1v2 /4

udu dv
v2 /41

dx dy

Change of Variables in 2(R2) where D = {(x, y) R2|y2 2px 0, x2 2py 0, p ]0; +[ xed}, dx = 2uvdu + u2dv, dy = v2du + 2uvdv, dx dy = 3u2v2du dv. The region transforms into = {(u, v) R2|0 u (2p)1/3, 0 v (2p)1/3}. The integral becomes f(x, y)dx dy =
D

145

using the change of variables x = u2v, y = uv2. Solution: We have

exp
3

u6v3 + u3v6 (3u2v2) du dv u3v3


3

= 3

eu ev u2v2 du dv
(2p)1/3 0 2 2 u3

1 = 3 =

3u e

du

1 2p (e 1)2. 3

As an exercise, you may try the (more natural) substitution x3 = u2v, y3 = v2u and verify that the same result is obtained. 309 Example Find
D

f(x, y)dx dy where

D = {(x, y) R2|a xy b, y x 0, y2 x2 1, (a, b) R2, 0 < a < b} and f(x, y) = y4 x4 by using the change of variables u = xy, v = y2 x2.

146 Solution: Here we argue that du = ydx + xdy, dv = 2xdx + 2ydy. Taking the wedge product of differential forms, du dv = 2(y2 + x2)dx dy. Hence f(x, y)dx dy = (y4 x4) = 2(y2 1 du dv + x2)

Chapter 3

1 2 (y x2)du dv 2 v = du dv 2

The region transforms into = [a; b] [0; 1]. The integral becomes f(x, y)dx dy =
D

v du dv
1

b 1 du 2 a ba = . 4

v dv
0

310 Example Let D = {(u, v) R2 : u 1, u v u}. Consider R2 : R2 u+v uv , 2 2 .

Find the image of on D , that is, nd D = (D ).

(u, v)

Change of Variables in 2(R2) Find (x + y)2ex


D
2 y2

147

dx dy.

Solution: Put x = u+v uv and y = . Then x + y = u and x y. Observe 2 2 that D is the triangle in the uv plane bounded by the lines u = 0, u = 1, v = u, v = u. Its image under is the triangle bounded by the equations x = 0, y = 0, x + y = 1. Clearly also 1 dx dy = du dv. 2 From the above (x + y)2ex
D
2 y2

dx dy =

1 u2euvdu dv 2 D 1 1 u 2 uv u e dudv = 2 0 u 1 1 2 2 = u(eu eu )du 2 0 1 = (e + e1 2). 4

311 Example Use the following steps (due to Tom Apostol) in order to prove that 2 1 = . n2 6 n=1 Use the series expansion 1 = 1 + t + t2 + t3 + 1t in order to prove (formally) that
1 0 1 0

|t| < 1,

d xd y = 1 xy

1 . n2 n=1

148

Chapter 3

Use the change of variables u = x + y, v = x y to shew that


1 0 1 0

d xd y =2 1 xy

1 0

dv du+2 2 2 u 4 u + v

2 1

2u u2

dv du. 4 u2 + v2

Shew that the above integral reduces to


1

2
0

2 u arctan du + 2 4 u2 4 u2

2 1

2 2u arctan du. 4 u2 4 u2

2 by using the substitution Finally, prove that the above integral is 6 u = arcsin . 2 Solution: Formally,
1 0 1 0

d xd y = 1 xy

xy2 x2y3 x3y4 = + + + y+ 2 3 4 0 1 x x2 x3 + + )dx = (1 + + 2 3 4 0 1 1 1 = 1 + 2 + 2 + 2 + 2 3 4

0 0 1

(1 + xy + x2y2 + x3y3 + )dxdy


1 0

dx

This change of variables transforms the square [0; 1] [0; 1] in the xy plane into the square with vertices at (0, 0), (1, 1), (2, 0), and (1, 1) in the uv plane. We will split this region of integration into two disjoint triangles: T1 with vertices at (0, 0), (1, 1), (1, 1), and T2 with vertices at (1, 1), (1, 1), (2, 0). Observe that 1 dx dy = du dv, 2

Change to Polar Co-ordinates

149

and that u + v = 2x, u v = 2y and so 4xy = u2 v2. The integral becomes


1 0 1 0

d xd y 1 = 1 xy 2
T1 T2 1

du dv 2 v2 1 u 4 dv du + 2 2 2 u 4 u + v
u 2 1 2u u2

= 2
0

dv du, 4 u2 + v2

as desired. This follows by using the identity


t 0

d = arctan t. 1 + 2

This is straightforward but tedious!

3.7

Change to Polar Co- ordinates

One of the most common changes of variable is the passage to polar co ordinates where x = cos = dx = cos d sin d, whence y = sin = dy = sin d + cos d, dx dy = ( cos2 + sin2 )d d = d d.

312 Example Find xy x2 + y2dx dy


D

where D = {(x, y) R2|x 0, y 0, y x, x2 + y2 1}.

150

Chapter 3

We use polar co-ordinates. The region D transforms into the region = [0; 1] 0; Therefore the integral becomes
/4 1

. 4

cos sin d d =
0

cos sin d
0

4 d

= 313 Example Find

1 . 20

x2 + y2dx dy
D

where D = {(x, y) R2|x 0, y 0, x2 + y2 1, x2 + y2 2y 0}. Using polar co-ordinates,


/6 1

f(x, y)dx dy =
D 0 2 sin

2 d d

1 /6 (1 8 sin3 ) d 3 0 16 = + 3. 18 9

314 Example Find (x2 y2)dx dy


D

where D = {(x, y) R2|(x 1)2 + y2 1}.

Change to Polar Co-ordinates Solution: Using polar co-ordinates,


/2 2 cos

151

x2 y2dx dy =
D /2 /2 0

3 d (cos2 sin2 )d cos4 (cos2 sin2 ) d


0

= 8 = . 315 Example Find


D

xydx dy

where D = {(x, y) R2|(x2 + y2)2 2xy}. Solution: Using polar co-ordinates,


D /4 0 sin 2

xydx dy = 4
0

2 cos sin d d
/4 0

4 3

( sin 2)3 cos sin d

4 /4 2 sin 2 d = 3 2 0 2 = . 12 316 Example Find


D

f(x, y)dx dy where

and f(x, y) = x3 + y3.

D = {(x, y) R2 : b2x2 + a2y2 = a2b2, (a, b) ]0; +[ xed}


2 1

Solution: Using x = a cos , y = b sin , the integral becomes (ab)


0

a3 cos3 + b3 sin3 d
0

4d

2 (ab)(a3 + b3). 15

152 317 Example Find


D

Chapter 3 f(x, y)dx dy where

D = {(x, y) R2|y 0, x2 + y2 2x 0} and f(x, y) = x2y. Solution: Using polar co-ordinates the integral becomes
/2 0 2 cos 0

4 4d cos2 sin d = . 5

318 Example Find


D

f(x, y)dx dy where

D = {(x, y) R2|x 1, x2 + y2 2x 0} and f(x, y) = (x2 1 . + y2)2

Solution: Using polar co-ordinates the integral becomes


/4 /4

1 d d = 3 1/ cos
2 xyy2

2 cos

/4

cos2
0

sec2 4

d =

. 8

319 Example Find


D

ex

dxdy, where

D = {(x, y) R2 : x2 + xy + y2 1}. Solution: Completing squares 2 3y y + x2 + xy + y2 = x + 2 2 3y y . The integral becomes Put U = x + , V = 2 2 ex


{x2 +xy+y2 1}
2 xyy2

2 d xd y = 3

e(U
{U2 +V2 1}

2 +V2 )

dUdV.

Change to Polar Co-ordinates Passing to polar co-ordinates, the above equals 2 3


2 0

153

2 2 e dd = (1 e1). 3 0

320 Example Let D = {(x, y) R2 : x2 + y2 y 0, x2 + y2 x 0}. Find the integral (x + y)2dx dy.
D

Solution: Put D = {(x, y) R2 : y x, x2 + y2 y 0, x2 + y2 x 0}. Then the integral equals 2


D

(x + y)2dx dy.

Using polar co-ordinates the integral equals


/2 cos 2 0

2
/4

(cos + sin )

1 /2 d d = cos4 (1 + 2 sin cos )d 2 /4 3 5 = . 64 48


3

321 Example Let D = {(x, y) R2|y x2 + y2 1}. Compute dx dy . 2 2 2 D (1 + x + y ) Solution: Observe that D = D2 \ D1 where D2 is the disk limited by the equation x2 + y2 = 1 and D1 is the disk limited by the equation x2 + y2 = y. Hence dx dy = 2 2 2 D (1 + x + y )
D2

dx dy (1 + x2 + y2)2

D1

dx dy . (1 + x2 + y2)2

154 Using polar co-ordinates we have dx dy = (1 + x2 + y2)2


2 0 1 0

Chapter 3

D2

d d = 2 2 (1 + ) 2

and dx dy = 2 (1 + x2 + y2)2 =
/2 0 + /2 sin2 d d d = 2 (1 + 2)2 0 1 + sin 0 dt 2 dt = . t2 + 1 2t2 + 1 2 4 sin

D1

(We evaluated this last integral using t = tan ) Finally, the integral equals 2 2 . = 2 2 4 4

322 Example Evaluate x3y3 1 x4 y4 dx dy


{(x,y)R2 :x0,y0,x4 +y4 1}

using x2 = cos , y2 = sin . Solution: We have 2xdx = cos d sin d, 2ydy = sin d + cos d, whence 4xydx dy = d d. It follows that x3y3 1 x4 y4 dx dy = 1 2 2 (x y )( 1 x4 y4)(4xy dx dy) 4 1 3 = ( cos sin 1 2)d d 4

Observe that x4 + y4 1 = 2 cos2 + 2 sin2 1 = 1.

Change to Polar Co-ordinates

155

Since the integration takes place on the rst quadrant, we have 0 /2. Hence the integral becomes
/2 0 1 0

1 3 ( cos sin 4

1 2)d d = =

1 4

/2

cos sin d
0 0

3 1 2d

1 1 2 4 2 15 1 . = 60

323 Example William Thompson (Lord Kelvin) is credited to have said: A mathematician is someone to whom + x2 e dx = 2 0 is as obvious as twice two is four to you. Liouville was a mathematician. Prove that + x2 e dx = 2 0 by following these steps. Let a > 0 be a real number and put Da = {(x, y) R2|x2 + y2 a2}. Find Ia =
Da

e(x

2 +y2 )

dx dy.

Let a > 0 be a real number and put a = {(x, y) R2||x| a, |y| a}. Let Ja =
a

e(x

2 +y2 )

dx dy.

Prove that Ia Ja Ia2. Deduce that


+

x2

dx =

. 2

156 Solution: Using polar co-ordinates


2 a 0

Chapter 3

Ia =
0

e d d = (1 ea ).

The domain of integration of Ja is a square of side 2a centred at the origin. The respective domains of integration of Ia and Ia2 are the inscribed and the exscribed circles to the square. First observe that Ja =
a a 2

x2

dx

Recall from formula 3.2 that the area enclosed by a simple closed curve is given by 1 xdy ydx. 2 Using polar co-ordinates

Since both Ia and Ia2 tend to as a +, we deduce that Ja . This gives the result.

xdy ydx = ( cos )(sin d + cos d) ( sin )(cos d sin d) = 2d. This will be used in the next problem. 324 Example Prove that every closed convex region in the plane of area has two points which are two units apart. Solution: Parametrise the curve enclosing the region by polar co-ordinates so that the region is tangent to the polar axis at the origin. Let the equation of the curve be = f(). The area of the region is then given by 1 2
0

1 d = 2
2

1 (f()) d = 2 0
2

/2

((f())2 + (f( + /2))2)d.


0

Integrating in 3(R3)

157

By the Pythagorean Theorem, the integral above is the integral of the square of the chord in question. If no two points are farther than 2 units, their squares are no farther than 4 units, and so the area < a contradiction. 325 Example (Putnam Exam 1976) In the xy-plane, if R is the set of points inside and on a convex polygon, let D(x, y) be the distance from (x, y) to the nearest point R. Show that
+ +

1 2

/2

4d = ,
0

eD(x,y) dx dy = 2 + L + A,

where L is the perimeter of R and A is the area of R. Solution: Let I(S) denote the integral sought over a region S. Since D(x, y) = 0 inside R, I(R) = A. Let L be a side of R with length l and let S(L ) be the half strip consisting of the points of the plane having a point on L as nearest point of R. Set up co-ordinates uv so that u is measured parallel to L and v is measured perpendicular to L. Then
l

I(S(L )) =

0 0

ev du dv = l.

The sum of these integrals over all the sides of R is L. If V is a vertex of R, the points that have V as nearest from R lie inside an angle S(V ) bounded by the rays from V perpendicular to the edges meeting at V . If is the measure of that angle, then using polar co-ordinates

I(S(V )) =
0

e d d = .

The sum of these integrals over all the vertices of R is 2. Assembling all these integrals we deduce the result.

3.8

Integrating in 3(R3)

326 Denition A 3-dimensional oriented manifold of R3 is simply an open set (body) V R3, where the + orientation is in the direction of the outward

158

Chapter 3

pointing normal to the body, and the orientation is in the direction of the inward pointing normal to the body. A general oriented 3-manifold is a union of open sets.

! The region V has opposite orientation to D and


=
V V

! In this section, unless otherwise noticed, we will choose the positive


orientation for the regions considered. This corresponds to using the volume form dx dy dz.

Let V R3. Given a function f : V R, the integral f(x, y, z)dx dy dz


V

is the sum of all the values of f restricted to V . In particular, dx dy dz


V

is the oriented volume of V . 327 Example Find x2yexyz dx dy dz.


[0;1]3

Solution: The integral is


1 0 1 0 1 1 1

x ye
0

xyz

dz dy dx =
0 1 0

x(exy 1) dy dx (ex x 1)dx


0

5 = e . 2

Integrating in 3(R3) 328 Example Find


R

159 z dx dy dz if

R = {(x, y, z) R3|x 0, y 0, z 0, x + y + z 1}. Solution: The integral is


1 (1 z)2 (1 z x)2

zdx dy dz =
R 0 1

z
0 (1 z)2 0

dy dx (1
0

dz

=
0

x)2dx

dz

1 1 z(1 z)4 dz 6 0 1 . = 840 =

329 Example Prove that xdxdydz =


V

a2bc , 24

where V is the tetrahedron V = (x, y, z) R3 : x 0, y 0, z 0, Solution: We have


c bbz/c 0 aay/baz/c

x y z + + 1 . a b c

xdxdydz =
V 0 0

xdxdydz
2

1 c bbz/c ay az a 2 0 0 b c 3 c 2 1 a (z + c) b dx = 6 0 c3 a2bc = 24 =

dydz

160

Chapter 3

3.9

Change of Variables in 3(R3)

330 Example Find (x + y + z)(x + y z)(x y z)dx dy dz,


R

where R is the tetrahedron bounded by the planes x+y+z = 0, x+yz = 0, x y z = 0, and 2x z = 1. Solution: We make the change of variables u = x + y + z = du = dx + dy + dz, This gives w = x y z = dw = dx dy dz. du dv dw = 4dx dy dz. v = x + y z = dv = dx + dy dz,

These forms have opposite orientations, so we choose, say, du dw dv = 4dx dy dz which have the same orientation. Also, 2x z = 1 = u + v + 2w = 2.

The tetrahedron in the xyz-co-ordinate frame is mapped into a tetrahedron bounded by u = 0, v = 0, u + v + 2w = 1 in the uvw-co-ordinate frame. The integral becomes 1 4
2 0 1v/2 0 2v2w

uvw du dw dv =
0

1 . 180

Consider a transformation to cylindrical co-ordinates (x, y, z) = ( cos , sin , z). From what we know about polar co-ordinates dx dy = d d.

Change of Variables in 3(R3) Since the wedge product of forms is associative, dx dy dz = d d dz.

161

331 Example Find


R

z2dx dy dz if

R = {(x, y, z) R3|x2 + y2 1, 0 z 1}. Solution: The region of integration is mapped into = [0; 2] [0; 1] [0; 1] through a cylindrical co-ordinate change. The integral is therefore
2 1 1

f(x, y, z)dx dy dz =
R 0

d
0

d
0

z2 dz

= . 3 (x2 + y2)dx dy dz over the rst octant region


D

332 Example Evaluate

bounded by the cylinders x2 + y2 = 1 and x2 + y2 = 4 and the planes z = 0, z = 1, x = 0, x = y. Solution: The integral is
1 0 /2 /4 2

3d d dz =
1

15 . 16

333 Example Three long cylinders of radius R intersect at right angles. Find the volume of their intersection. Solution: Let V be the desired volume. By symmetry, V = 24V , where V =
D

dx dy dz,

162

Chapter 3

D = {(x, y, z) R3 : 0 y x, 0 z, x2+y2 R2, y2+z2 R2, z2+x2 R2}. In this case it is easier to integrate with respect to z rst. Using cylindrical co-ordinates = (, , z) 0; [0; R] [0; +[, 0 z R2 2 cos2 . 4
/4 R 0 R

Now,

R2 2 cos2

=
0 /4 0

d z d d
0 /4 0

= =

R2 2 cos2 d d
R 0

1 (R2 2 cos2 )3/2 3 cos2 0 R3 /4 1 sin3 = d 3 0 cos2 2 = 2 1 u2 R3 /4 [tan ]0 + du u = cos 3 u2 1 = = Finally V = 16V = 8(2 R3 1 u1 + u 3 21 3 R . 2
2 2

2)R3.

Consider now a change to spherical co-ordinates x = cos sin , y = sin sin , z = cos . We have

dx = cos sin d sin sin d + cos cos d, dy = sin sin d + cos sin d + sin cos d, dz = cos d sin d.

Change of Variables in 3(R3) This gives dx dy dz = 2 sin d d d.

163

From this derivation, the form d d d is negatively oriented, and so we choose dx dy dz = 2 sin d d d instead. 334 Example Let (a, b, c) ]0; +[3 be xed. Find R= (x, y, z) R3 : xyzdx dy dz if
R

x2 y2 z2 + + 1, x 0, y 0, z 0 . a2 b2 c2

We use spherical co-ordinates, where (x, y, z) = (a cos sin , b sin sin , c cos ). We have dx dy dz = abc2 sin d d d. The integration region is mapped into = [0; 1] [0; ] [0; ]. 2 2 The integral becomes
/2 1 /2

(abc)2
0

cos sin d
0

5 d
0

cos3 sin d

(abc)2 . 48

335 Example Let V = {(x, y, z) R3 : x2 + y2 + z2 9, 1 z 2}. Then


2 /2arcsin 1/3 /2arcsin 2/3 2/ cos

dx dy dz =
V 0 1/ cos

2 sin d d d 63 . 4

164 336 Lemma Let m, n be non-negative integers. Then


1

Chapter 3

xm(1 x)n dx =
0

m!n! . (m + n + 1)!

337 Example (Putnam Exam 1984) Find x1y9z8(1 x y z)4 dx dy dz,


R

where R = {(x, y, z) R3 : x 0, y 0, z 0, x + y + z 1}. Solution: We make the change of variables u = x + y + z = du = dx + dy + dz, This gives uvw = z = uvdw + uwdv + vwdu = dz. x = u(1 v), y = uv(1 w), z = uvw, u2v du dv dw = dx dy dz. To nd the limits of integration we observe that the limits of integration using dx dy dz are 0 z 1, 0 y 1 z, 0 x 1 y z. This translates into 0 uvw 1, 0 uv uvw 1 uvw, 0 u uv 1 uv + uvw uvw. Thus 0 uvw 1, 0 uv 1, 0 u 1, uv = y + z = udv + vdu = dy + dz,

Integration in 2(R3) which nally give 0 u 1, 0 v 1, 0 w 1. The integral sought is then, using Lemma 336,
1 0 1 0 1

165

u20v18w8(1 u)4(1 v)(1 w)9 du dv dw,


0

which in turn is
1 1 1

u20(1 u)4du
0 0

v18(1 v)dv
0

w8(1 w)9dw

1 1 1 265650 380 437580

which is =

1 . 44172388260000

3.10

Integration in 2(R3)

338 Denition A 2-dimensional oriented manifold of R3 is simply a smooth surface D R3, where the + orientation is in the direction of the outward normal pointing away from the origin and the orientation is in the direction of the inward normal pointing towards the origin. A general oriented 2-manifold in R3 is a union of surfaces.

! The surface has opposite orientation to and


=

In this section, unless otherwise noticed, we will choose the positive orientation for the regions considered. This corresponds to using the ordered basis {dy dz, dz dx, dx dy}.

166

Chapter 3

339 Denition Let f : R3 R. The integral of f over the smooth surface (oriented in the positive sense) is given by the expression f d2x .

Here d2x = (dx dy)2 + (dz dx)2 + (dy dz)2 is the surface area element. 340 Example Evaluate

z d2x where is the outer surface of the sec-

tion of the paraboloid z = x2 + y2, 0 z 1. Solution: We parametrise the paraboloid as follows. Let x = u, y = v, z = u2 + v2. Observe that the domain D of is the unit disk u2 + v2 1. We see that dx dy = du dv, dy dz = 2udu dv, dz dx = 2vdu dv, and so d2x = Now, z d2x =
D

1 + 4u2 + 4v2du dv.

(u2 + v2) 1 + 4u2 + 4v2du dv.

To evaluate this last integral we use polar co-ordinates, and so


2 1

(u2 + v2) 1 + 4u2 + 4v2du dv =


D 0 0

3 1 + 42dd 1 (5 5 + ). 12 5

Integration in 2(R3) 341 Example Evaluate x 1, 0 y 2.

167 y d2x where is the surface z = x + y2, 0

Solution: We parametrise the surface by letting x = u, y = v, z = u + v2. Observe that the domain D of is the square [0; 1] [0; 2]. Observe that dx dy = du dv, dy dz = du dv, dz dx = 2vdu dv, and so d2x = The integral becomes
2 1

2 + 4v2du dv.

y d x

=
0 0 1

v 2 + 4v2du dv
2

= =

du 13 2 . 3
0 0

2 + 4v2dv

342 Example Evaluate

x2 d2x where is the surface of the unit sphere

x2 + y2 + z2 = 1. Solution: We use spherical co-ordinates, (x, y, z) = (cos sin , sin sin , cos ). Here [0; 2] is the latitude and [0; ] is the longitude. Observe that dx dy = sin cos d d, dy dz = cos sin2 d d, dz dx = sin sin2 d d, and so d2x = sin d d.

168 The integral becomes


2

Chapter 3

x2 d2x

=
0 0

cos2 sin3 d d 4 . 3

343 Example Evaluate between z = 0 and z = 1.


S

z d2x

over the conical surface z =

x2 + y2

Solution: Put x = u, y = v, z2 = u2 + v2. Then dx = du, dy = dv, zdz = udu + vdv, whence u v dx dy = du dv, dy dz = du dv, dz dx = du dv, z z and so d2x = = = Hence z d x =
u2 +v2 1 2

(dx dy)2 + (dz dx)2 + (dy dz)2 1+ u2 + v2 du dv z2

2 du dv. 2 2 d d = . 3 0
1 2

u2 + v2

2 dudv =

2
0

344 Example Find the area of that part of the cylinder x2 + y2 = 2y lying inside the sphere x2 + y2 + z2 = 4. Solution: We have x2 + y2 = 2y x2 + (y 1)2 = 1.

Integration in 2(R3)

169

We parametrise the cylinder by putting x = cos u, y 1 = sin u, and z = v. Hence dx = sin udu, dy = cos udu, dz = dv, whence dx dy = 0, dy dz = cos udu dv, dz dx = sin udu dv, and so d2x = = (dx dy)2 + (dz dx)2 + (dy dz)2 cos2 u + sin2 u du dv

= du dv. The cylinder and the sphere intersect when x2 + y2 = 2y and x2 + y2 + z2 = 4, that is, when z2 = 4 2y, i.e. v2 = 4 2(1 + sin u) = 2 2 sin u. Also 0 u 2. The integral is thus
2 22 sin u 2

d x

=
0

22 sin u 2 0

dvdu =
0

2 2 2 sin udu

= 2 2 = 16.

1 sin u du

345 Example Evaluate xdy dz + (z2 zx)dz dx xydx dy,

where is the top side of the triangle with vertices at (2, 0, 0), (0, 2, 0), (0, 0, 4). Solution: Observe that the plane passing through the three given points has equation 2x + 2y + z = 4. We project this plane onto the co-ordinate

170 axes obtaining


4 2z/2 0

Chapter 3

xd y d z =
0

8 (2 y z/2)dy dz = , 3
42x

(z2 zx)dz dx =
0 0 2

(z2 zx)dz dx = 8,
2y 0

xydx dy =
0

2 xydx dy = , 3

and hence xdy dz + (z2 zx)dz dx xydx dy = 10.

346 Example Evaluate xydy dz x2dz dx + (x + z)dx dy,

where is the top of the triangular region of the plane 2x + 2y + z = 6 bounded by the rst octant. Solution: We project this plane onto the co-ordinate axes obtaining
6 3z/2

xydy dz =
0 0

(3 y z/2)ydy dz =
3 62x

27 , 4

x dz dx =
0 3 3y 0

x2dz dx =

27 , 2 27 , 2

(x + z)dx dy =
0 0

(6 x 2y)dx dy =

and hence xydy dz x2dz dx + (x + z)dx dy =

27 . 4

Greens, Stokes, and Gau Theorems

171

3.11

Greens, Stokes, and Gau Theorems

We are now in position to state the general Stokes Theorem. 347 Theorem (General Stokes Theorem) Let M be a smooth oriented manifold, having boundary M. If is a differential form, then =
M M

d.

In R2, if is a 2-form, this takes the name of Greens Theorem. 348 Example Evaluate (x y3)dx + x3dy where C is the circle x2 + y2 = 1.
C

Solution: We will rst use Greens Theorem and then evaluate the integral directly. We have d = d(x y3) dx + d(x3) dy = (dx 3y2dy) dx + (3x2dx) dy = (3y2 + 3x2)dx dy. The region M is the area enclosed by the circle x2 + y2 = 1. Thus by Greens Theorem, and using polar co-ordinates, (x y3)dx + x3dy =
C

(3y2 + 3x2)dx dy
M 2 1

=
0 0

32d d

3 . = 2 Aliter: We can evaluate this integral directly, again resorting to polar co ordinates.
2

(x y3)dx + x3dy =
C

(cos sin3 )( sin )d + (cos3 )(cos )d


0 2

=
0

(sin4 + cos4 sin cos )d.

172

Chapter 3

To evaluate the last integral, observe that 1 = (sin2 + cos2 )2 = sin4 + 2 sin2 cos2 + cos4 , whence the integral equals
2 2

(sin4 + cos4 sin cos )d =


0 0

(1 2 sin2 cos2 sin cos )d

3 = . 2

349 Example Evaluate


C

x3ydx + xydy where C is the square with vertices

at (0, 0), (2, 0), (2, 2) and (0, 2). Evaluating this directly would result in evaluating four path integrals, one for each side of the square. We will use Greens Theorem. We have d = d(x3y) dx + d(xy) dy = (3x2ydx + x3dy) dx + (ydx + xdy) dy = (y x3)dx dy. The region M is the area enclosed by the square. The integral equals
2 2

x3ydx + xydy =
C 0 0

(y x3)dx dy

= 4.

350 Example Consider the triangle C : (2, 2).

with vertices A : (0, 0), B : (1, 1),

If LPQ denotes the equation of the line joining P and Q nd LAB, LAC, and LBC. Evaluate y2dx + xdy.

Greens, Stokes, and Gau Theorems Find (1 2y)dx dy


D

173

where D is the interior of Solution:

1 4 LAB is y = x; LAC is y = x, and LBC is clearly y = x + . 3 3 We have


1

y2dx + xdy =
AB

(x2 + x)dx
0 2

=
2

5 6

y2dx + xdy =
BC 1 0

4 1 x+ 3 3

y2dx + xdy =
CA

(x2 x)dx
2

1 15 x dx = 3 2 14 = 3

Adding these integrals we nd y2dx + xdy = 2. We have


0 x/3+4/3

(1 2y)dx dy =
D 2 x 1

(1 2y)dy dx
x/3+4/3

+
0 x

(1 2y)dy dx

44 10 27 27 = 2.

351 Example Use Greens Theorem to prove that (x2 + 2y3)dy = 16,

where is the circle (x 2)2 + y2 = 4. Also, prove this directly by using a path integral.

174 Solution: Observe that d(x2 + 2y3) dy = 2xdx dy. Hence by the generalised Stokes Theorem the integral equals
/2 4 cos

Chapter 3

2xdx dy =
{(x2)2 +y2 4} /2 0

22 cos d d = 16.

To do it directly, put x 2 = 2 cos t, y = 2 sin t, 0 t 2. Then the integral becomes


2 2

((2 + 2 cos t)2 + 16 sin3 t)d2 sin t =


0 0

(8 cos t + 16 cos2 t +8 cos3 t + 32 cos t sin3 t)dt

= 16. In R3, if is a 2-form, the above theorem takes the name of Gau or the Divergence Theorem. 352 Example Evaluate (x y)dy dz + zdz dx ydx dy where S is
S

the surface of the sphere x2 + y2 + z2 = 9 and the positive direction is the outward normal. Solution: The region M is the interior of the sphere x2 + y2 + z2 = 9. Now, d = (dx dy) dy dz + dz dz dx dy dx dy = dx dy dz. The integral becomes dx dy dz =
M

4 (27) 3

= 36.

Greens, Stokes, and Gau Theorems Aliter: We could evaluate this integral directly. We have (x y)dy dz =

175

xdy dz,

since (x, y, z) y is an odd function of y and the domain of integration is symmetric with respect to y. Now,
3 2

xd y d z =
3 0

||

9 2dd

= 36. Also zdz dx = 0,

since (x, y, z) z is an odd function of z and the domain of integration is symmetric with respect to z. Similarly ydx dy = 0,

since (x, y, z) y is an odd function of y and the domain of integration is symmetric with respect to y. The classical Stokes Theorem occurs when is a 1-form in R3. 353 Example Evaluate
C

ydx + (2x z)dy + (z x)dz where C is the in-

tersection of the sphere x2 + y2 + z2 = 4 and the plane z = 1. Solution: We have d = (dy) dx + (2dx dz) dy + (dz dx) dz = dx dy + 2dx dy + dy dz + dz dx = dx dy + dy dz + dz dx. Since on C, z = 1, the surface on which we are integrating is the inside of the circle x2 + y2 + 1 = 4, i.e., x2 + y2 = 3. Also, z = 1 implies dz = 0

176 and so d =

Chapter 3

dx dy.

Since this is just the area of the circular region x2 + y2 3, the integral evaluates to dx dy = 3.

354 Example Let denote the curve of intersection of the plane x + y = 2 and the sphere x2 2x + y2 2y + z2 = 0, oriented clockwise when viewed from the origin. Use Stokes Theorem to prove that ydx + zdy + xdz = 2 2.

Prove this directly by parametrising the boundary of the surface and evaluating the path integral. Solution: At the intersection path 0 = x2+y2+z22(x+y) = (2y)2+y2+z24 = 2y24y+z2 = 2(y1)2+z22, which describes an ellipse on the yz-plane. Similarly we get 2(x 1)2+ z2 = 2 on the xz-plane. We have d (ydx + zdy + xdz) = dydx+dzdy+dxdz = dxdydydzdzdx. Since dx dy = 0, by Stokes Theorem the integral sought is
2(y1)2 +z2 2

dydz
2(x1)2 +z2 2

dzdx = 2( 2).

(To evaluate the integrals you may resort to the fact that the area of the (x x0)2 (y y0)2 elliptical region + 1 is ab). a2 b2 If we were to evaluate this integral directly, we would set y = 1 + cos , z = 2 sin , x = 2 y = 1 cos .

Greens, Stokes, and Gau Theorems The integral becomes


2

177

(1 + cos )d(1 cos ) +


0

2 sin d(1 + cos ) + (1 cos )d( 2 sin )

which in turn
2

=
0

sin + sin cos

2+

2 cos d = 2 2.