CONTENTS

A,

Linear spaces Linear independence Gauss-Jordan elimination Parametric equations of lines and planes in Vn Parametric equations for k-planes in Vn

A1
A7
A17
A25
A32

> Matrices Systems of linear equations Cartesian equations of k-planes in V n The inverse of a matrix
Determinants

Construction of the determinant when n s 3

Cramer I rule and a formula for A'
s Construction of the determinant function
The cross product in V3

Supplementary exercises Derivatives of vector functions Differentiating inverse functions Implicit differentiation The second-derivative test for extrema The extreme value and small-span theorems Exercises on line integrals

CO' C1 C10 C14

C22

C28
c34

Notes on double integrals

Double integrals over general regions
Area

Green's theorem in the plane Conditions under which Pi + Qj is a gradient The change of variables theorem Isometries

El E6 E21
E27

Stokes' theorem
Exercises on the divergence theorem
Grad, curl, div and all that

Linear Spaces

we have seen (12.1-12.3 of Apostol) that n-tuple space

V ,

has the following properties : Addition:

1. (Commutativity) A + B = B + A.
2. (Associativity) A

+ (B+c) = (A+B) +
for all A.

C.
0
-

3. (Existence of zero)

There is an element

such 'that A + - = A 0
B

4. (Existence of negatives) Given A, such that A + B = 0.

there is a

Scalar multiplication:

5. (Associativity) c (dA) = (cd)A.

6. (Distributivity) (c+d)A = cA

+ dA,

c(A+B) = cA + cB. 7. (Multiplication by unity) More generally, let V 1A = A. be any set of objects

Definition.

(which we call vectors).

V,

And suppose there are two operations on

+)

as follows:

The first is an operation (denoted A, B

that A + B.

assigns to each pair

of vectors, a vector denoted

The second is an operation that assigns to each real number c and each vector
A,

a vector denoted

cA.

Suppose also that the with these two operaThe seven

seven preceding properties hold.

Then V,

tions, is called a linear space (or a vector space).

properties are called the axioms - a linear space.

for

n ' The study of linear spaces and their properties is dealt with in a subject called Linear Algebra.

There are many examples of linear spaces besides n--tuple space

W ! shall treat only those aspects of linear algebra needed E

for calculus. Therefore we will be concerned only with n-tuple space Vn

and with certain of its subsets called "linear subspaces" : Definition. Let W be a non-empty subset of Vn ; suppose W is closed under vector addition and scalar multiplication. Then W is called a linear subspace of V n (or sometimes simply a subspace of Vn .) To say W is closed under vector addition and scalar multiplication

means that for every pair A, B of vectors of W, and every scalar c, the vectors A + B a ~ dcA belong to W. Note that it is automatic that
I

the zero vector Q belongs to W, since for any A

W, we have Q = OA.

Furthermore, for each A in W, the vector - A is also in W. This means (as you can readily check) that W is a linear space in its own right (i.e.,
f

it satisfies all the axioms for a linear,space).

S~bspaces of Vn m y be specified in many different ways, as we shall
see.

Example 1.

The s u b s e t o f

Vn

c o n s i s t i n g o f the

9-tuple

alone i s a subspace of
space.

Pad of course

Vn; it i s ths " s m a l l e s t p o s s i b l e " sub V , i s by d e f i n i t i o n a subspace of Vn;

it i s the " l a r g e s t p o s s i b l e " subspace.

W;ample 2 .

Let A be a fixed non-zero vector, The subset of Vn

is a subspace of

consisting of all vectors X of the form X = cA

n' It is called the subspace spanned by A In the case n = 2 or 3 , it can .

be pictured as consisting of all vectors lying on a line through the origin.

Example 3 Let A and B be given non-zero vectors that are not

.
1

parallel. The subset of Vn consisting of all vectors of the form

is a subspace of V n o It is called the subspace spanned by A and B. In the case n = 3, it can be pictured as consisting of all vectors lying in the plane through the origin that contains A and B.

---A/
We generalize the construction given in the preceding

examples as follows:

Definition. Let S = $A?,

...

be a set of vectors in Vn

A vector X of Vn of the form

X = c1 1 +
A

...

+*
.
The set W of

is called a linear combination of the vectors Alr...,A,c

all such vectors X is a subspace of Vn, as we will see; it is said to be the subspace spanned by the vectors Al,...,% linear span of A l ,

It is also called the

...,+
then

and denoted by
W

L(S).

Let us show that belong to

is a subspace of

V. n

If

and

W,

X = clAl
I

-*
+

=kAk

and

Y = d1A1

+ * * *

+ dkAkf

for soma scalars ci

and

di.
f

We compute

* * *

X . + Y = (cl+dl)A1

+ (ck+dk)Akr

ax = (ac ) A + 1 1

* * *

+ (ack)Akf
W
by definition. Thus

so both

+ Y

and

ax

belong to

is a subspace of

Vn.

Giving a spanning set for W is one standard way of specifying W . Different spanning sets can of course give the same subspace. Fcr example, it is intuitively clear that, for the plane through the origin in Example 3,

any_ two non-zero vectors C and D that are not parallel and lie in this
plane will span it. We shall give a proof of this fact shortly.

Example 4. The

n-tuple space Vn

has a natural spanning set,

namely the vectors

En = (0,0,0,.

.. ,1).

These are often called the unit coordinate vectors in -

is e a s y to see that they span Vn,

an element of
V , ,

for i f . X = (xl,

...,

Vn. It
x ) is n

then

In the case where coordinate vectors respectively.

E2,

n = 2, E2

we often denote the unit in V2

by

El

and

I and

3,

In the case where by

t
1,

n = 3, we often denote respectively.

El,

and- E3

and

They are pic-

tured as in the accompanying figure.

Example 5. The subset W the form are cX.

I

of

V3

consisting of all vectors of V3. For if 0, X and

X + Y

(a,b,O) is a subspace of

y and

3-tuples whose third component is

It is easy to see that

so are

is the linear span of

(1,0,0)

and

(O,l,O).
V3 consisting of all vectors of the V3.
It consists

Example 6. The subset of form

X = (3a+2b,a-b,a+7b) is a subspace of

of all vectors of the form

X = a(3,1,1)

+ b(2,-1,7),
3 1 , and (2,-1,7).

so it is the linear span of

Example 1 The set W of all tuples (x1,x2,x3.x4) svch that

,

i a subspace of Vq s
I

as you can check. Solving this equation for x4 ,

WE

see

that a 4-tuple belongs to W if and only if it has the form

(xl,X2, x3, -3x1 + X2 - 5 ~ ~ ) . .

where xl acd xL and x3 are arbitrary. This element can be written in the form
,

It follaris that (1,O.O.-3) and (OtlrOtl) and (O,Ofl,-5) sy:an W.

Exercises

1 .

Show that the subset of V3 specified in Example 5 is a subspace

s~ecifiedin Example 7. What can
such that alxl +

of V-. 3

Do the same for the subset of V4

you say about the set 'of all x ,

...x n

... + anxn = 0

...,an) in general? (Here we assume A = (al.

give a geometric interpretation?

is not the zero vector.) Csn you

2. In each of the following, let W denote the set of all vectors (x,y,z) in Vj satisfying the condition given.

(Here we use element of

(x,y,z)

instead of

(xl,x2,x3)

for the general

Vj.

V3.)

Determine whether

is a subspace of

If it is, draw a picture of it or describe it geometrically,

and find a spanning set for (a) x = 0.
(b) x + y = O .
(c)
X

W.

(e) x = y

or

2x=z.

(f)
(4)

x2
X2

y 2 = 0.

+ y = 1.

+ *2 = 0

(dl

x = y

and

2x =

2.

3.

Consider the set

of all real-valued functions

defined on the interval

[arb].

(a) Show that

is a linear space if cf

denotes the usual sum of functions and product of a function by a real number. vector?

denotes the usual

What is the zero

(b) Which of the following are subspaces of

(i) All continuous functions.
(ii) All integrable functions.
(iii) All piecewise-monotonic functions.
(iv) All differentiable functions.

(v)

F?

All functions

such that

f (a) = 0.

(vi) All polynomial functions.

Ljnear independence

Dc?f - inition.

We say that the set S = I A ~ ,

...,q;\

of vectors of

vn

spans the vector X if X belongs to L(S),

X = cA + 1 1

that is, if

... + ck+

for some scalars ci. If S spans the vector X, we say that S spans X
uniquely if the equations

X=
i=l

ciAi

and

imply that ci = di

for all i.

It is easy to check the following:

, Theorem 1 Let S = < A ~ ,

X be a vector in L(S).
i

...

be a set of vectors of Vn;

let

Then S spans X uniquely if and only if S spans

the zero vector

2 uniquely.

Proof. Mte that - = 0

i

2OAi .

This means that S spans the zero

vector uniquely if and only if the equation

implies that ci = 0 for all i . Stippose S syms

k

uniquely. To show S spans X uniquely, suppose and

X=
1= 1

c.A. 1 1

X =

i=l

dilli.

Subtracting, we see that

whence ci - d. = 0, or ci = di , for all i.

1 Conversely, suppose S spans X uniquely. Then

for some (unique) scalars xi ' Now if

it follows that

Since S spans X uniquely, we must have xi = xi + ci , or ci = 0, for all 1 0 . This theorem implies that if S spans one vector of L(S) uniquely,
then it spans the zero vector uniquely, whence it spans every vector of L(S)
uniquely. This condition is important enough to be given a special name:
of vectors of Vn is said to be linearly independent (or simply, independent) if it spans the zero vector Definition. The set S =
I

ZA] ,...,%J

uniquely. The vectors themselves are also said to be independent in this

situation.

If a set is not independent, it is said to be dependent.

Banple

8 If a subset .

T of a set S is dependent, then S itself

(Just add on the

is dependent. For if T spns Q ncn-trivially, so does S . additional vectors with zero coefficients.)

This statement is equivalent to the statement that if S is independent, then so is any subset of S. Example 9 Any set containing the zero vector Q is dependent. For
. example, if S = A ~ , . . . , a i A1 = 0, then rd

Example

The unit coordinate vectors E1,...,En

in Vn span Q

uniquely, so they are independent.

i

Pample & Let S = A , .

..

If the vectors Ai are non-zero

and mutually orthogonal, then S is independent. For suppose

Taking the dot product of both sides of this equation with A1 gives the equation

0 =
C1

AlOA1

(since A. *A1 = 0 for i # 1) 1

whence cl = 0.
j

NGW A1 # Q by hypothesis, whence A1 *A1 # 0,

2

Similarly, taking the dot product with Ai for the fixed index

shows that c = 0 . j
Scmetimes it is convenient to replace the vectors

Bi = A ~ / ~ I A ~ ~ \

Then the vectors B1,

...,Bk

are of

&

Ai by the vectors
length and are mutually

orthogonal. Such a set of vectors is called an orthonormal set. The coordinate vectors E l
form such a set. n A set ccnsisting of a single vector A is independent

B.mple

if A # Q.
I

A set consisting of two non-zero vectors ARB is independent if and

only if the vectors are not parallel. More generally, one has the following result: Theorem 2- The set S = { A ~, of the vectors A j

...,\I

is independent if and only if none

can be written as a linear combination of the others.

Proof. Suppose first that one of the vectors equals a linear

combination 06 the others. For Al = c2A2 +
* * *

instance, suppose that

+ ckAk:

then the following non-trivial linear combination equals zero:

.. . ...

Conversely, if

where not all the that cm # 0,

ci

are equal to zero, we can choose

so

and obtain the equation

where the sum on the right extends over all indices different
from

m.

Given a subspace W of Vnr there is a very important relation that

holds between spanning sets for W and independent sets in W Theorem 21 Let W be a subspace of Vn vectors A1,
k vectors.
i

that is spanned by the k

...,\ .

Then any independent set of vectors in W contains at most

Proof. Let B

,...,B

be a set of vectors of W;

let m 2 k. We

wish to show that these vectors are dependent. That is, we wish to find scalars xl,...,xm '

-- n ; all zero, ct

such that

Since each vector B belongs to W, we can write it as a linear combination of

j We do so, using a "double-indexing" notation for the coefficents, the vectors Ai

as follows:

Multiplying the equation by x and summing over j, and collecting terms, we

j
have the equation

In order for <x .B to equal 2 , it will suffice if we can choose the x j j

so that coefficient of each vector Ai in this equation equals 0. Ncw the
are given, so that finding the x. is just a matter of solving a
numbers aij 3 (homogeneous) system consisting of k equations in m unknowns. Since m > k, there are more unknowns than equations. In this case the system always has a non-trivial solution X (i.e., one different from the zero vector). This is a standard fact

about linear equations, which we now prove. First, we need a definition.

a
...,xn)

Definition. Given a homogeneous system of linear equations, as in ( * ) following, a solution of the system is a vector (xl, that satisfies

each equation of the system. The set of all solutions is a linear subspace of
/

Vn
(as you can check).

It is called the solution space of the system.

It is easy to see that the solution set is a subspace.

If we let

be the n-tuple whose components are the coefficerts appearing in the

jth equation of the system, then the solution set consists of those X
~uch that A . * X = 0 for all J

j. If X and

are two solutions, then

and

Thus X

+Y

and

cX are also solutions, as claimed.

a bornogeneous system of k linear equations

in n utlknow~ls. If k is less than n, then the solution space con-
tains some vector other t2ian 0.

I'ronf.. We are concer~tcd here only v i t h proving the existence of some solutioli otlicr tJta110, not with nctt~nlly fitidirtg such a solution it1 practice,

Theorem 4. Given

nor wit11 finditig all possildd solutiot~s. (We 11-ill study the practical problem in nnuch greater rlctail in a later scctioti.) in We start with a system of lc ecluatio~is ?t uriknowns:

Our procedure 11411 1)c to reduce tlic size of this system step-by-step by elimitit~tingfirst XI, tlleri x2, and so on. After k - 1 steps, we mil1 be reduced to solvitig just one c q t t a t i o ~ ~ this will be easy. But a certain and nmount, of care is ticeded it1 the dcscriptiorl-for instance, if all = . . = akl = 0, i t is nonset~se to spcak of "elirninnting" XI, since all its coefi cierits are zero. \Ve I ~ a v eo a l l o ~ for this possibility. t i~ 'L'o begirt then, if all the cocflicic~tts st are zero, you may verify t h a t of the vector (fro,. . . ,0) is n solution of the system which is different from 0 , and you nre done. Othcr~risc, t Icast one of the coefiicielits of s is a t nonzero, attd 1t.e rncly s~tj~posc cortvetlier~cethat the equations have for beerr arranged so that this happetls ill the first ec~uation, with the result that 0 1 1 0. We rnultiply the first crlt~ation the scalar azl/afl and then 1)y suhtract i t from the second, eli~nitiat~itlg XI-term from the second tghe et~uatiori.Si~rtilarly,we elirninatc the xl-term in each of the remaining equations. 'I'he result is a ttcw system of liriear equatioris of the form

Now any soltttiolr of this 11c1vssystoc~n cqttntiol~s also a solution of the of is old system (*), because we cat1 recover the old system from the new one: we merely multiply the first erlttatiorl of tlre systcm (**) by the same scalars we used before, aild.then tve add i t to the corresponding later equations of this system. The crucial ttlirig about what n c hnve done is contained in the follorvirlg statement: If the smaller system etlclosed in the box above has a solution other than the zero vector, tlictr thc Ia>rgcrsystem (**) also has a solution other than the zcro ~ c c t o r[so tJ1lnt the origitinl system (*) tve started o r wit21 llns a solutio~l t l ~ c t han the zcro vcctor). $Ire prove this as follows: stna1.ller system, different from Sr~ppose(d2, . . . , d.,) is a solutiolr of t l ~ o , . . . , . We su1)stitutc itito tllc first equation and solve for XI, thereby obtailiirlg the follo~vingvector,

i n w1ricI1 yo11 may verify is a s o l ~ ~ t ~ofo t,hc Iargcr systcm (**). In this v a y we havc rc!clnc:e~l tlio sixo of our problcrn; we now tlccd only to prove otir ttlcorcrn for a sysf,ern of Ic - 1 ecluntions in n - 1 unknowns. n If ~ v c apply this reductio~l scct)~~ci tilnc, tve reduce the prol~lem provto ing the theorem for a systern of k - 2 ccl~tatiol~s n - 2 unkno~v~rs. in Con tinuing in this way, after lc - 1 eIimiriation steps it1 all, we will be down 1 unlrno~vns.Now to a system cot~sistitig orlly one ecll~nt,ion, n - k of it1 n - Ic 1 2 2, because IVC ass~trned our hypothesis that n > Ic; thus ns our problem retluccs to proving the follo~trillg ststemcrrt: a "system" consistitig of otze li~icnr ornogcneous ccluntion it1 two or ntore unkno~vrls lways h a has a solutioll other than 0.

WE! leave it to you to show that this statement holds.(Be sure you

ccnsider the case where one or more or all of the coefficents are zero.)
E -2ample

span all

13.

We have already noted that the vectors

El,

...,En

of

Vn.

It. follows, for example, that any three vectors in V2

are dependent,

that is, one of them equals a linear combination of the others. for any four vectors in Vj.
I

The same holds

The accompanying picture ~ k e s these facts plausible.

Similarly, since the vectors Elt..-,E are independent, any spanning

n
set of Vn must contain at least n vectors. Thus no two vectors can span
and no set of three vectors can span v4

V3 '

Theorem 5. Let W be a subspace of Vn that does not consist of

0 - alone. Then:
(a)

The space W has a linearly indepehdent spanning set.

>

(b) Any two linearly independent spanning sets for W have the same

number k of elements; Proof. --

k<n

unless W is all of Vn.

in W. Then the set { 1 is independent. A> If

(a) Choose A1 # 0

In general, suppose A l .A

is an independent set of vectors of W.

this set spans W, we are finished. Otherwise, we can choose a vector Ai+l of W that is not in L(Alr.. ,A.) Then the set 1A1, ,A A ~ + ~ ] is 1 i' independent: For suppose that

...

for some scalars ci not all zero. If c ~ =+0 , ~ this equation contradicts independ ecce of A for A

..A

while if c ~ #+0 , we can solve this equation ~

contradicting the fact that Ai+l does not belong to L(Alt

....A,). 1

Cc~ntinuingthe process just described, we can find larger and larger

independent sets of vectors in W The process stops only when the set we obtain
. spans W. Does it ever stop? Yes, for W is contained in Vn, ar;d Vn contains

no more than n independent vectors. indefinitely!

(b) Suppose S =

Sc,

the process cannot be repeated

IA~, and ... \\

T= B

... B

are two

ljnearly independent spanning sets for W. spans W, we must have k <_ j

Because S is independent and T

, by the preceding theorem. Because S sy:ans

W and T is independent, we must have k z j . Thus k = j.

N l Vn contains no more than n independent vectors; therefore we cw must have k 5 n. Suppose that W is not all of Vn. a vector 2$+1 set A Then we can choose

of Vn that is not in W. By the argument just given, the

+

...

is independent. It follows that W1

< n,

so that k i 0 n.

Definition. Given a subspace W of Vn that does not consist of Q alone, it has a linearly independent spanning set. Any such set is called a
)

basis for W, and the number of elements in this set is called the dimension of W . We make the convention that if W consists of - alone, then the dimension of 0

W :

is zero.

4 Example 1 . The space Vn has a "naturaln basis consisting of the

. vectors E1,...,E n
It follows that Vn has dimension n are many other bases for Vn., For instance, the vectors

(Surprise!) There

form a basis for

I

V ,

as you can check.

~xercises

1.

Consider the subspaces of V3

listed in Exercise 2, p A6. Find bases for .

each of these subspaces, and f r d spanning sets for them that are - bases. ii not
2. Check the details of Example 1 . 4

3 Suppose W has dimension k. (a) Show that any independent set in

.

consisting of k vectors spans W.

(b) Show that any spanning set for W

consisting of k vectors is independent.

Let S = I A ~ , ,A be a spanning set for W. Show that S m
contains a basis for W. [Hint: Use the argument of Theorem 5.1
4.
5.

...

>

Let

IA;,
be ...,G

an independent set in Vn

Show that this

set can be extended to a basis for Vn

6.

[Hint: Use the argument of Theorem 5 . 1

If V acd W are suSspaces of Vn and Vk, respectively, a

:

function T
i

V +W

is called a linear transformation if it satisfes the usual T X + Y) = T(X) + T Y , ( ()

linearity properties:

If T is one-to-one and carries V onto W, it is called dl: linear.

isomorphism of vector spaces.
Stippose All.. vectors of W.

.,A,,

is a basis for V; let B1p.afBk be arbitrary

:

(a) Slim? there exists a linear transformation T

V +W

such that T(Ai) = Bi fc>rall i.

(b) Show this linear transformation is unique.

7. L e t W be a subspace of V : n

let Al,...,% be a basis for W .

and Y =

Let X, Y be vectors of W .

Then X = xjAi

2 yiAi

for unique

scalars xi m d y.. These scalars are called the camponents of X and Y, 1 respectively, relative to the basis Aif...f\.

(a)

Note that
--)

and

Conclude

that the function T : Vk 1inear isomorphism

W defia~dby T(xl,. ,%) =

..

'f xiAi

is a

A1 7

(b) Suppose that the basis A ,

is an orthonormal basis. Show

that

X*Y = 2 x i Y i

Conclude that the isomorphism T of (a) preserves the

= X*y

dot product, that is, T(x).T(Y)

8 Prove the following:

.

Theorem. If W is a subspace of Vn, then W has an orthonormal basis.

Froof. Step 1 Let . in Vn
;

.be7 L
' B1,..., B be mutually orthogonal non-zero vectors m
Given scalars

( let Am+l be a vector not in L B l,...,B m ). cl,. ,c '

let m

..

Btl
n+

Show that Bmcl is different from L(B~,

and that L(B1,.

...,Bm ,Am+l ) -

Then show that the ci m y be so chosen that Bm+l is m

orthogonal to each of B

..,B, .

Steep2 Show that if W is a subspace of Vn of positive dimension. . then W has a basis consisting of vectors that are mutually orthogonal.
i

[Hint: Proceed by induction on the dimension of W ] . Step 3 Prove the theorem. . Gauss-Jordan elimination

If W is a subspace of Vn, specified by giving a spanning set for

W,
we have at present no constructive process for determining the dimension

of W nor of finding a basis for W, although we b o w these exist. There

is a simple procedure for carrying out this process; we describe it n w .

~efinitfon. The rectangular a r r G of numbers

A ~ + C1B1 + + ~

"

CmBm
=

..,Bm ,Bm+l)

is called a matrix of size called the entry of

A

by

n.

The number row and

j-th

a ij

is

in the

th i -

column.

Suppose we let Ai. be the vector

for i = 1,. .lc.

Theh Ai

is just the ith row of the matrix A.

The

subspace of Vn spanned by the vectors A l l...,% of the matrix A.

is called the row space

WE now describe a procedure for determining the dimension of this space.

It involves applying operations to the matrix A,

(1) Interchange two rows of A
.

of the following types:

(2) Replace row i of A by itself plus a scalar multiple of another row,

say rcm m.

(3) Multiply

row i of A by a non-zero scalar.

These operations are called the elcmentary from the following fact: Theorem

& operations.

Their usefulness cclmes

6.

Suppose B

is the matrix obtained by applying a sequence

of elementary row operations to A,successively. Then the row spaces of

A

=d . B

are the same. It suffices to consider the case where B Let All...,% is obtained by

Proof. --

applying a single row operation to A. and let BI,

be the rows of A,

...,Bk

be the rows of 8.

If the operation is of t y p ( I ) , these two sets of vectors are the

same (only their order is changed), so the spaces they span are the same. If the operation is of type
Bi=cAi

(3), then
B = A
j

and

for j # i.

Clearly, any linear combination of B1,...,Bk can be written as a linear combination of Al.

...,+

Because c # 0 ,

the converse is also true. Then

Finally, suppose the operation is of type Bi = Ai + dAm md

(2).

Bi=A for j f i . j

Again, any linear combination of Bl,...,Bk can be written as a linear combination of Al.

...,+

Because

and
A. = B 3 j the converse is also true. for j # i

The Gauss-Jordan procedure consists of applying elementary row operations to the matrix A until it is brought into a form where the dimension of its It is the following:

row space is obvious.

IGaussJordan

Examine the first column of your matrix.

elimination.

I
%

(I) If this column consists entirely of zeros, nothing needs to ba

done. Restrict your attention now to the matrix obtained by deleting the first column, and begin again.

(11) If this column has a non-zero entry, exchange rows if necessary

to bring it to the top row. Then add multiplesof the top row to the lower rows so as to make all remaining entries in the first column into zeros. Restrict your attention now to the matrix obtained by deleting the first column and first row, and begin again.
The procedure stops when the matrix remaining has only one row.

k t us illustrate the procedure with an example.

Pr-oblem. Find the dimension of the row space of the matrix

Solution. First step. Alternative (a) applies. Exchange rows (1) and ( 2 ) , obtaining

,!

Replace row (3) by row (3) + row ( 1 ) ; then replace (4)by (4)+ 2 times ( I )

Second step. Restrict attention to the matrix in the box. (11) applies. Replace row (4) by row (4) - row (2)

obtaining

Third step. Restrict attention to the matrix in the box.

L _ .

(I) applies,

so nothing needs be done.

One obtains the matrix

Fourth step. Restrict attention to the matrix in the box.

(11) applies. 7 Replace row (4) by row (4)- 7 row (3) , obtaining

--

The procedure is now finished. The matrix

__3

s we end up with is in what i called

echelon or "stair-stepUfom. The entries beneath the steps are zero. And

the entries -1, 1, and 3 that appear at the "inside cornerst1 the stairsteps
of
i

are non-zero. These entries that appear at the "inside cornersw of the stairsteps are often called the pivots in the echelon form.
Yclu can check readily that the non-zero rows of the matrix B

are

independent. (We shall prove this fact later.) It follows that the non-zero rows of the matrix B form a basis for the row space of B, and hence a basis for the row space of the original matrix A. Thus this row space has dimension 3.

The same result holds in general. If by elementary operations you

reduce the matrix A to the echelon form B, then the non-zero rows are B
are independent, so they form a basis for the row space of B, and hence a
b~.sisor the row space of A. f

Now we discuss how one can continue to apply elementary operations to

1

reduce the matrix B to an even nicer form. The procedure is this:

Begin by considering the last non-zero row. By adding multiples of this row
to each row above itr one can bring the matrix to the form where each entry lying
above the pivot in this row is zero. Then continue the process, working now
with the next-to-last non-zero row. Because all the entries above the last
pivot are already zero, they remain zero as you add multiples of the next-to-
last non-zero row to the rows above it. Similarly one continues. Eventually
the matrix reaches the form where all the entries that are directly above the
pivots are zero. (Note that the stairsteps do not change during this process,
nor do the pivots themselves.)
Applying this procedure in the example considered earlier, one brings
the matrix B into the form

Note that up to this point in the reduction process

, we have used only

elementary row operations of types (1) and (2). It has not been necessary to multiply a row by a non-zero scalar. This fact will be important later on.
WE? are not yet finished. The final step is to multiply each non-zero

row by an appropriate non-zero scalar, chosen so as to make the pivot entry

. into 1 This we can do, because the pivots are non-zero. At the end of

this process, the matrix is in what is called reduced The reduced

echelon form.

echelon form of the matrix C above is the matrix

As we have indicated, the importancs of this process comes from the

' !

following theorem:
Theorem 7 Let A be a matrix; let W be its row space. Suppose
. we transform A by elementary row operations into the or into the reduced echelon matrix B,

echelon matrix D. Then the non-zero rows of B

are a basis for W, m d so are the non-zero rows of D.

Pr-oof. The rows of B span W, as we noted before; and so d the o -rows of D. It is easy to see that no non-trivial linear combination of the rmn-zero rows of D equals the zero vector has an entry of

, because

each of these rows

1 in a position where the others all have entries of 0.

Thus the dimension of W equals the number r of non-zero rows of D. This is the same as the number of non-zero rows of B

If the rows of B

\ifre not independent, thon one would equal a linear combination of the others. Piis would imply that the row space of B could be spanned by fewer than
\

r rcws, which would imply that its dimension is less than r.

Exercises
I. Find bases for the row spaces of the following matrices:

2.

Reduce the matrices in Exercise 1 to reduced

echelon form.

*3. Prove the following:

P~eorern. The reduced

echelon form of a matrix is unique.

echelon matrices, w5ose

Proof. Let D and D 1 be two reduced

rows span the same subspace W of Vn. We show that D = D'. Let R
be the non-zero rows of D ; and suppose that the
k t

pivots (first non-zero entries) in these rows occur in columns jl,...,j respectively.

(a)

=ow that the pivots of D 1 wrur in the colwols jl,...,jk.

[Hint: Lst R be a row of Dl; suppose its pivot occurs in column p. We have R = c R + + c & for some scalars ci (Why?) Show that 11 c = 0 if ji < p. Derive a contradiction if p is not equal to any of i

...

(b) If R

is a row of D 1 whose pivot occurs in columr.~.jm

, show

that R = Rm.

[Hint: We have R = c1 1 + . R Show that ci = 0 for i Z m, and cm = 1 1 .

... + c k s

for some scalars ci g

parametric equations - lines - planes - Vn

of and in Given through points
X
P

n-tuples

and

A,

with

A # Q,

the line

determined - A by such that

is defined to be the set of all

for some scalar

L(P;A).

t. A

It is denoted by
is called a direction vector for the
L is simply the 1-dimensional subspace
'

The vector

line. Note that if P = of Vn spanned by

A.

2, then

, ,

The equation for the line, and As t t

(*)

is often called a parametric equation

is called the parameter in this equation.

X

ranges over all real numbers, the corresponding point

ranges over all points of the line L. When t = 0, then X = P; when

t = 1, then X = P + A; when t = $,

then X = P +

%A;

and so on. All

these are points of L. Occasionally, one writesthe vector equation out in scalar
form as follows:

A26

n
the scalar parametric equations for the line.

where

P = p , , p n

and

A = (al,...,a

).

These are called

Of course, there is no uniqueness here: a given line can

be represented by many different parametric equations.

The

following theorem makes this result precise:

Theorem 8. The lines -L(P;A) and L(Q:B) are if - equal B.

and if have - only - they - Proof. --

a point - common and A in

is to - parallel -

If L(P;A) = L(Q;B),

then the lines obviously have a point

in comn. Since P and P + A lie on the first line they also lie on the second line, so that

for distinct scalars ti and t2. A is parallel to B .

Subtracting, we have A = (t2-tl)B, so

Conversely, suppose the lines intersect in a point R, and suppose

A and B are parallel. We are given that

for some scalars tl ard t2, and that A = cE for some c # solve these equations for P in terms of Q and B : P = Q + t 2 B - tlA = Q + (t2-tl&JB. Now, given any point X = P + tA of the line L(P;A),
X = P + tA = Q + (t2-tlc)B+ tcB.
WE

0.

We can

can write

Thus X

belongs to the line L (Q;B) Thus every point of

L (P; A)

.
L (Q:B)

belongs to

The
O

symmetry of the argument shows that the reverse holds as well. Definition.
It follows from the preceding theorem that

given a line, its direction vector is uniquely determined up to

a non-zero scalar multiple.

We define two lines to be parallel

if t h e i r d i r e c t i o n v e c t o r s a r e p a r a l l e l .

Corollary 9 .

D i s t i n c t p a r a l l e l l i n e s cannot i n t e r s e c t ,

Corollary 1 0

exactly

. a l and - Given - - i n e L - - a p o i n t Q , one l i n th t to --e c o n t a i n i n g Q -a-i s p a r a l l e l Suppose

Q

there is

L.

Proof. L(Q;A)

is t h e l i n e

L(P;A).

Then t h e l i n e

contains

and i s p a r a l l e l t o
Q

L.
L

By Theorem 8 , any

other l i n e containing one.

0

and p a r a l l e l t o

i s equal t o t h i s

there

two - Given - d i s t i n c t p o i n t s is one l i n t hem. - e x a c t l y --e c o n t a i n i n g 0. Proof. L e t A = Q - P; then A # Theorem 11.

P

and -

Q,

The l i n e (since

L(P;A)

contains both
Q = P

(since

P = P + OA)

and

+
Q.

LA). L(R;B)

Now suppose and Then

i s some o t h e r l i n e c o n t a i n i n g

f o r d i s t i n c t scalars

tl

and

t2. It follows t h a t

s o t h a t the v e c t o r

A = Q

-P

is parallel t o

B.

I t follows

from Theorem ' 8 t h a t

Now we. study planes i n . , V

Definition.
B
P

If

i s a p o i n t of
Vnr

Vn

and i f . A

and

a r e independent v e c t o r s of determined

w e def i n e t h e p l a n e through
X

and -

t o be t h e s e t of a l l p o i n t s

of

t h e form

where

and

run through a l l r e a l numbers.

W denote t h i s e

p l a n e by

M(P;A,B).
(*)

The e q u a t i o n p l a n e , and tion. s and

i s c a l l e d a parametric equation f o r the

a r e c a l l e d t h e parameters i n t h i s equan s c a l a r equations, i f desired.

I t may be w r i t t e n o u t a s

W e s = t = 0, then X = P; hn
s = 0

when s = 1 and t = 0, then X = P + A; when

SO

and t = 1, then X = P + B; and

on.

Ncte that

if

P =

2,

then t h i s plane is j u s t the 2-dimensional subspace

of Vn

span&: by A and B. J u s t as f o r l i n e s , a p l a n e has many d i f f e r e n t p a r a m e t r i c

representations. Theorem 12. e q u a l i f and o n l y span

More p r e c i s e l y , one h a s t h e following theorem: The -planes if have - they - M(P;A,B)

and -

M(Q:C,D)

are D.

-A

a p o i n t i n common and t h e l i n e a r

of

P roof.

and -

e q u a l s t h e l i n e a r span of

and -

I f t h e p l a n e s a r e e q u a l , t h e y obviously have a

A29

point in c o m n . Fcrthermore, since P and P +

ard

P + B all lie

on the first plane, they lie on the second plane as well. Then

P + B = Q + s3c + t p ,
Subtracting, we see that are some scalars s ar-d t i' i

A = (sZ-sl)C + (t2-tl)Df
B (s3-s1)C + (t3-tl)D.

Thus A and B lie in the linear span of C and D. Symmetry shows that
C and D lie in the linear span of A and B as well. Thus these linear

spans are the same.

Conversely, suppose t h a t t h e p l a n e s i n t e r s e c t i n a p o i n t
R

and t h a t

L(A,B) = L(C,D).
=

Then
Q+s2C+t2D

P+slA+tB 1

for some scalars si ard ti. We can solve this equation for P as follows:

P
=

Q + (linear combination of A,B,C,D).

Then if X is any point of the first plane M(P;A,B), we have

X P+sA+tB =

for some scalars s and t,

= Q + (linear combination of A,B,C,D) + sA + tB = Q +

(linear combination of CID), L(c,D),

since A
Thus

and B

belong t o

belongs to

M (Q;C 1D)

.
as w e l l .
0

Symmetry o f t h e argument shows t h a t every p o i n t of

M(Q;C,D)

belongs t o

M(P;A,B)

Definition.
A

Given a p l a n e

M = M(P;AtB),

the v e c t o r s

and

a r e n o t uniquely determined by
W e say t h e planes

M,

but t h e i r l i n e a r

span is.

M(P;A,B)

and

M(Q;C,D)

are .

parallel i f

L(A,B) = L(C,D).

Corollary

1 . TKO distinct parallel planes cannot 3

Given - plane a M and --

intersect. Q, there is
M.

14. corollary -

a point

exactly one plane containinq Q Proof.

that to --is parallel -

Suppose M = M ( P ; A , B ) and

Then

M(Q;A,B) M. M

is a

plane that contains Q

is parallel to
Q

By Theorem 12
is equal to

any other plane containing this one.

parallel to

Definition. WE'say three points P,Q,R are collinear if they lie on a line.
Lemma 1 . 5

The points P,Q,R are collinear if and only if the vectdrs

Q-P =d

R-P are dependent (i.e., parallel). Proof. The line L(P; Q-P) is the one containing P and Q, and

theliae;
i

L(P;R-P) istheonecontaining P and R. If Q-P m d R-P are the same, by Theorem .8,so P, Q,and R

are parallel, these lines

are collinear. Conversely, if 'P,Q, and R are collinear, these lines must
be the same, so that Q--P and ' R-P must be parallel.

a,
P, Q, R,

Theorem 16-

. '

Given three non-collinear points A = Q

there - exactly - plane containing is one them. Proof.

A and

Let

and

B = R

P;

then
A = Q

B are independent. The plane M(P; A,B)

cGntains P and P +

and P + B = R *
Now suppose

M(S;C,D)

is 'another plane containing P,

Q,

and

R.

Then

'I

for some scalars si and ti Subtracting, we see that the vectors Q - P = A and R - P = B belong to the linear span ~f f and D. By 2 symmetry, C and D belong to the linear span of A ad
12 implies that these two planes are equal.

B. Then Theorem

Exercises
1. We say the line L is parallel to the plane L belongs to L(A,B). then L

M = M(P;A,B)

if the direction vector of

Show that if

is parallel to M.

and intersects M,

is contained in 2.

Show that two vectors

Al

and

A2

in Vn

are

linearly dependent if and only if they lie on a line through

the origin.
3. Show that three vectors All
A2,

A3

in Vn

are

linearly dependent if and only if they lie on some plane through the origin.

Let

A = 1 - 1 0
(a)

B = (2,0,1).

Find parametric equations for the line through P and Q, and

for the line through R with direction vector A. Do these lines intersect?
(b) Find

parametric equations for the plane through PI Q, and

R.

and for the plane through P determined by A and B.

5.

Let L be the line in Vj

through the points P = (1.0.2) and

Q = (1,13). Let L1 be the line through

A =
I

parallel to the vector

( 3 - 1 ) Find parametric equations for the line that intersects both L

and Lf and is orthogonal to both of them.

432

Parametric equations

for

k-planes

Vn.

Following t h e p a t t e r n f o r l i n e s and p l a n e s , one c a n d e f i n e , more generally, a k-plane i n

Vn

a s follows:
P

Definition.
A ~ , . . . , A ~ of

Given a p o i n t

of

Vn V,,

and a s e t

k
P

independent v e c t o r s i n determined

we define the
t o be t h e set of

k-plane through a l l vectors

&

Al,

....Ak

of t h e form

f o r some s c a l a r s M(P;A1,..

ti.

W e d e n o t e t h i s s e t o f p o i n t s by

., A k ) .
X

Said d i f f e r e n t l y i f and o n l y i f

is i n the

k-plane

M (P;Al,

i s i n t h e l i n e a r span of

A1,

...,Ak.

...,Ak'

Note that if P =
1

2,

then t h i s k-plane is just the k- dimensional

linear subspace of

Vn

s p ~ e by A1, d

...,%.
( 1 - p l a n e s ) and p l a n e s

J u s t as w i t h t h e c a s e o f l i n e s

(2-planes), one has t h e following r e s u l t s : Theorem 1 , 2

be two k-planes i n

Let MI = M(P;A
Vn.

and

3 = M(Q;B~'

1 * * * 1 ~

k1

Then M1 = M2
AII...I%

i f and only if they have a point i n equals the linesr span of

M1 B1,

commn and the linear span of Definition.

...,Bk.

W say that the k-planes e

and M2

of t h i s theorem

are p a r a l l e l i f the linear span of A I I . . - I %

equals the linear span of

B1?

...,Bk.
Theorem 1 8 . Given k-plane k-plane i n
M
V, Vn

5 point
Q

Q,

t h e r e i s e x a c t l y o ne

containing

parallel to

- M.. Lemma 19. --

.
they are contained i n

Given points POt..,Pk in Vn,

a plane of dimension l e s s than k i f and only i f the vectors

P1 - Po,

..., Pk- Po

are dependent.

Theorem 2Q. Given k+l distinct points Po,

...,Pk

in Vn.

If these points.do not lie in any plane of dimension less than k, tten
s there i exactly onek-plane containing them; it is the k-plane

More generally, we make the following definition:

Definition. If M1 M2 = M(Q;B~
MI
= M(P:Al, ...,$)

is a k-plane, and

,...,Bm )

is an m-plane, in Vn

, and if k s m, we say
is contained

is parallel to M2 if the linear span of Al,...,%

in the linear span of B1,

...,, . B

Brercises 1. Prove Theorems 1 and 18.

7
9 2. Prove Theorems 1 and 20.

3 Given the line L = L(Q;A) .

in Vj

where A = (1.-1,2).

Find parametric equations for a 24lane containing the point P = (1,1,1) that is parallel to L. Is it unique? C m you find such a plane containing both the point P and the point Q = (-1,0,2)?

4 Given the 2-plane MI. in V4 containing the points P = (1,-I, 2,-1) .

and Q = (O.l,l,O)wd

R = (lrl,0,3).Find parametric equations for a 3-plane

in Vq that containsthe point S = (l,l,l.) and is parallel to MI.

Is it unique? Can you find such a 3-plane thatcontains both S and the
point T = (0,1,0,2)?

Matrices
We have already defined what we mean by a matrix. In this section,

we introduce algebraic operations into the set of matrices.

Definition. If A
k by

and B are two matrices of the same size, say n matrix obtained by adding

n,

we define A

B to be the k by

the corresponding entries of A

and B, m d we define cA to be the matrix

obtained from A by multiplying each entry of A by c. That is, if aij are the entries of A and B, respectively, in row i and column
and bij j , then the entries of A + B and of cA in row i and column j are aij respectively. P o ? that for fixed Jt
1:

bij

and

caij '

ar:d n

the set of all k by n matrices

satisfies all the properties of a linear space. This fact is hardly surprising, for a k by n matrix is very much like a Ic-n tuple;

that only difference is that the components are written in a rectangular array instead of a linear array. Unlike tuples, however, matrices have a further operation, a product
operation. It is defined as follows:
Cefinition. If

is a k by n mtrix, and B

is an n by

p matrix, we define the product D = A * B of A of size k by the formula

and B to be the matrix is given by

p whose entry dij in row i and column j

Here

i = k

and

j = I,...,p.

dij is computed, roughly speaking, by tak ing the "dot product" of the 1 'th row of A with the j-th column of
B.

The entry

Schematically,

This definition seems rather strange, but it is in fact e,xtremely useful. Motivation will come later! One important justification for this definition is the fact that this product operation satisfies some of the familar "laws of algebra" : Theorem '1. Matrix multiplication has the following properties: k t -A, B, C ,-D be matrices.
(1) (Distributivihy)

If A-(B + C) - defined, then is -

Similarly; if
. -

(B + C). D is defined,
. then

(El + C)* D = B * D

C.D.

(2) (Homogeneity) If A * B - defined, then is

(3) (Associativity) If A * B - B 6 C are defined,then and .

(4) (Existence of identities) Fcr each m, there - -

i an s

m by m

matrix -

s . that for m ~ch -- l' ii

matrices A
and B, we have - and B *Im = B

I .A = A

whenever these products are defined.

--. Proof. We verify the first distributivity formula. In order for

B + C to be defined, B and C must have the same size, say n by p.

m e n in order for A - ( B + C) to be defined, A must have n columns. Suppose A has size 'x by n. .Then A - B and A - C aredefinedand have size k

by p; thus their sum is also defined. The distributivity formula now follows from the equation

a (b.+c s=l is SJ sj

)=I

a b s=l is sj

+I

n
a c s=l is sj'

The other distributivity formula and the homogeneity formula are proved similarly. We leave them as exercises.
Now let us verify associativity.

- .
and
B

If A
B

is

by

is
C

by

p,

then

is

k
C

by

p.

The product
by q.

(AeB)

is thus defined

( B * C ) is

provided

has size p
I (

The product A

defined in precisely the same circumstances. is an exercise in summation symbols: column

j

Proof of equality
i and

The entry in row

of

(A*B)

is

and the corresponding entry of

(B=C)

is

These t w o expressions are equal. that act as identity elements.

I* Given m, let Im be the m by r matrix whose general entry is n Finally, we define matrices

4j,

where

&

1j

= 1

if

i = j and

6: = 0 lj

if i # j. The matrix Im is a square

matrix that has I4

1's down the "main diagonal" and 0's elsewhere. For instance,

is the matrix

Now the product Im*A

is defined in the case where A has m

rows. In

this case, the general entry of the product C = A Im -

is given by the equation

Let i and j be fixed. Then as s ranges from 1 to m, of the terms of this sunanation vanish.

a1.l but one

The only one that does not vanish

is the one for which s = i, and in that case

cij =

0+. . . + 0 +

Ti
A

&. s
= 1

1.

We conclude that

Qlj

O+ . . a +

0=

Q5.
if B has rn columns.
B
A

Ar, entirely similar proof shows that

B * Im = B

Remark. defined. be equal.

If

is defined, then

need n o t b e

And e v e n i f it i s d e f i n e d , t h e two p r c d u c t s need n o t

For example,

Remark. A natural qestion to ask at this point concerns the existence

of multiplicative inverses in the set of matrices. WE'shall study the answer
to this question in a later section.
Exercises

1.
2.

Verify the other half of distributivity.

Verify homogeneity of matrix multiplication. [Hint: If I and I ;

3. Show the identity element is unique.

are two possible choices for the identity element of size m by m, compute
1 ;

I ;

,I
4 Find a non-zero 2 by .
2 matrix A

such that A * A is the zero

matrix. Conclude that there is no matrix B such that B O A = 12.

5. Consider the set of m by m matrices; it is closed under addition

and multiplication. Which of the field axioms (the algebraic axioms that the real numbers satisfy) hold for this set? in modem algebra a "ring with identity. I ! ) (Such an algebraic object is called

Systems - linear equations of

Given numbers and given numbers cc , , l

aij

for

i = 1

and

j = l,--=,n,

we wish to study the following, which is called k , a system of k linear equations 9 n unknowns:

A solution of this system is a vector X = (xl,

...,xn)

that satisfies each

equation. The solution - of the system consists of all such vectors; it is set a subset of Vn

We wish to determine whether this systemhas a solution, and if so, what the nature of the general solution is. Note that we are not assuming anything about the relative size of k and n; they may be equal, or one may be larger than the other. Matrix notation is convenient for dealing with this systemof equations. rd Let A denote the k by n matrix whose entry in rcw i a ? column j is aij
Let X and C denote the matrices

B7

These are matrices with only one column; accordingly, they are called column matrices

The system of equations ( * ) can now be written in matrix form as

A solution of this matrix equation is now, strictly speaking, a column matrix

rather than an n-tuple. However, one has a natural correspondence

between n-tuples and column matrices of size n by

1. It is a one-to-one

correspondence, and even the vector space operations correspond. What this means is
that we can identify Vn with the space of all n by all this amounts to is a change of notation.
Representing elements of Vn as column matrices is so convenient that
1 matrices if we wish;

we will adopt it as a convention throughout this section, whenever we wish.

Example 1.

Consider the system

[Here we use
X3r

x, y, z

for the unknowns instead of

xl, x2,

for convenience.]

This system has no solution, since

the sum of the first two equations contradicts the third equation.

D-ample2. Consider the system 2 x + y + z = l

This system has a solution; in fact, it has mora than one solution. In
solving this sytem, we can ignore the third equation, since it is the sum of the
first two. Then we can assign a value to y arbitrarily, say y = t, and solve the first two equations for x and z We obtain the result .
*

The solution set consists of all matrices of the form

Shifting back to tuple notation, we can say that the solution set consists of
all vectors X such that

This expression shows that the solution set is a line in

V3, and in "solvingt1

the system, we have written the equation of this line in parametric form.

Now we tackle the general problem.

We shall prove the following

result: Sbppose one is given a system of k linear equations in n unknowns. Then the solution set is either (1) empty, or (2) it consists of a single point, or (3) it consists of the points of an m-plane in Vn, for some m7O. In case (11, we say the system is inconsistent, meaning that it has no solution.

In case ( 2 ) , the solution is unique.

I

in case ( 3 ) , the system has infinitely

many solutions.
F;C shall apply Gauss-Jordan elimination

to prove these facts. The

crucial result we shall need is stated in the following theorem: Tlieorem 2, Consider the system of equations
1 matrix.

A'X = C, where A

is

a k by' n matrix and C is a k by

Let B be the matrix obtained by

rd applying an elementary row operation to A, a r let C' be the matrix obtained by applying the same elementary row operation to C. Then the solution set of the system B - X = C' is the same as the solution.setof the system A'X = C.

Proof. Exchanging rows i and j of both matrices has the effect of simply exchanging equations i and j plus c plus c
d

of the system. Replacing row i by itself

times row j has the effect of replacing the ith equation by itself times the jth equation. And multiplying rcw i by

a non-zero scalar

has the effect of multiplying both sides of the ith equation by d. Thus each

solution of the first system is also a solution of the second system. Nc~wwe recall that the elementary operations are invertible. Thus the system A'X = C can be obtained by applying an elementary operation to both sides of the equation B w X = C1. It follows that every solution of the second system is a solution of the first system. Thus the two solution sets are identical.

WE*consider first the case of a homogeneous system of equations, that is,

a system whose matrix equation has the form

AGX =

In this case, the system obviously has at least one solution, namely the trivial solution X = -. 0 Furthermore, we know that the set of solutions is a

linear subspace of Vn

, that is, an m-plane through the origin for some m.

We wish to determine the dimension of this solution space, and to find a basis

for it.

Dflfinition. Let A be a matrix of size k by

n.

Let W be the row

space of A; let r be the dimension of W. Then r equals the number of non-zero rows in the echelon form of A. true that r It follows at once that r
k.

It is also is called

5 n, because W is a subspace of Vn
(or sometimes the -- of A). row rank

The number r

the rank of A

Theorem - Let A be a matrix of size k by 3. of A.

n. Let r be the rank

Then the solution space of the system of equations A 0 X = 2 is of dimension n - r.

a subspace of Vn

Proof. The preceding theorem tells us that we can apply elementary operations to both the matrices A and -
without changing the solution set. 0

Applying elementary operations tr - leaves it unchanged, of course.

c 0 So let us apply elementary operations to A so as to bring A into reduced echelon form D, and consider the system D - X = Q

The number of

non-zero rows of D equals the dimension of the row space of A, which is r. Now for a zero row of D, the corresponding equation is automatically satisfied, no matter what X we choose. Orily the first r equations are relevant. Suppose that the pivots of D appear in columns jl,...,jr ' Let J
denote the set of indices l j from the set {I,.

,...,jr\ and let K consist of the remaining indices

for which j is in J appears with a Each unknown x j non-zero coefficient in only - of the equations of the system D - X = one 0. Therefore, we can Itsolveufor each of these unknowns in terms of the remaining
unknowns xk

... ) n ,

, for k in K. Substituting these expressions for x ,

into the n-tuple X = (xl, ...,xn ) ,

jl we see that the general solution of the

..., x

jr

system can be written as a vector of which each component is a linear combination of the

xk

for k in K.

(Of course, if k is in K, then the linear

cathination that appears in the kth component consists merely of the single term

9! )

k!t

us pause to consider an example.

EJ-ample 3 -

Let A

be the 4 by

5 matrix given on p.A20.

The

equation A'X = - represents a system of 4 equations in 5 unknowns. Nc~w A 0 reduces by row operations to the reduced echelon matrix

Here the pivots appear in columns 1,2 and 4; thus J K is the set f3.53

The unknowns xl x2, and x4

is the set 11,2.4]

and

each appear in only

one equation of the system. We solve for theese unknowns in terms of the others
as follows:

X1
X*

= 8x3 3x5 = -4x3 2x5

=

x ,

0.

The general solution can thus be written (using tuple notation for convenience)

The -solution space is thus spanned by two vectors

The same procedure we followed in this example can be followed in general. Once we write X as a vector of which each component is alinear combination of the xk

then we can write it as a sum of vectors each of which involves

only one of the unknowns coefficients

5,

and then finally as a linear combination, with

%I of vectors in Vn

(8,-4.1,O.O)

and (3,-2,0.0,1).

There are of course n - r of the

unknowns

and hence

n - r of these vectok.

It follows that the solution space of the system has a spanning set consisting of n - r vectors. We now show that these v~ctorsare independent; then the theorem is proved.. To verify independence, it suffices to shox? that if xe take the vector X, which equals a linear combination with coefficents of these vectors, then X = - if and only if each 0 equals 0.

% (for k in K)

This is easy. Consider the first expression for X tkat we wrote down,

where each component of X is a linear combination of the unknowns The kth component of X is simply

5.

rc-

It follows that the equation X = 2

implies in particular that for each k in K, we have

= 0 .

For example, in the example we just considered, we see that the equation

X = - implies that x3 = 0 and x5 = 0 , because x3 is the third component 0 . . of X and x5 is the fifth component of X.
This proof is especially interesting because it not only gives us the
dimension of the solution space of the system, but it also gives us a method
for finding a basis for this solution space, in practice. All that is involved is
Gauss-Jordan elimination.
Corollary& Let A be a k by n matrix. If the rows of A are

0 independent, then the solution space of the system A-X = - has dimension n - k.

Now we consider the case of a general system of linear equations, of the

form A'X = C

For the moment, we assume that the system has at least one

solution, and we determine what the general solution looks like in this case. Theorem - Let A be a k by n matrix. 5. Let r equal the rank of A.

If the system A*X = C has a solution, then the solution set is a plane in Vn of dimension m = n - r.

Proof. Let X If X is a

P be a solution of the system. Then A-P = C

column matrix such that A'X = C, then A.(X - ? ) = 2 , and

ccnversely. The solution space of the system A'X = - is a subspace of n' 0 of dimension m = n - r; let All ,A be a basis for it. Then X is a solution ... m of the system A'X = C if and only if X - P is a linear combination of the vectors Air that is, if and only if X = P + t1A1
t...
+

tmAm

for some scalars ti. Thus the solution set is an m-plane in Vn

Nciw let us try to determine when the system A'X = C has a solution.

One has the

follow in^ general

result:
k by

Theorem of A '

6. k t A be a

n matrix. Let r equal the rank

(a) If r 4 k , then there exist vectors C in Vlc

system A X = C has no solution. '

such that the

(b) If r = k, tl~enthe system A-X = C al~~ays a solution.

has

Proof.

We consider the system A X = C and apply elementary row ' into echelon form

operations to both A . and C until we have brought A

.
C'

(For the moment, we need not yo all the hay to reduced echelon form.)

Let

be the column matrix obtained by applying these same row operations to C.

Consider the system B'X = C r . Consider first the case r < k.

3

In this case, the last row at least of

is zero. The equation corresponding to this row has the form

where c' is the entry of C 1 in row k. If c' is not zero, there are no k k values of xl, xn
satisfying this equation, so the system has no solution.

...,

Let us choose C * to be a k by

1 mstrix whose last entry is non-zero.

Then apply the same elementary operations as before, in reverse order, to both B and C*. These operations transform B back to A; when we apply them to C * , the result is a matrix C such that the system A X = C has no ' solution. Now in the case r = k, the echelon matrix B has no zero rows, so

the difficulty that occurred in the preceding paragraph does not arise. We shall show that in this case the system has a solution. More generally, we shall consider the following two cases at the same time: Either ( 1) B has no zero rows, or (2) whenever the ithrow of B is zero, then the corresponding entry ci of C' is zero. We show that in either of * these cases, the system has a solution. Let
US

consider the system B - X = C 1 ard apply further operations to

both B and C', so as to reduce B to reduced echelon form D. Let C" be the matrix obtained by applying these same operations to C'. Note that the zero rows of B, and the corresponding entries of C', are not affected by these operations, since reducing B to reduced echelon form requires us to work only with the non-zero rows. Consider the resulting system of equations D'X = C t t . We now proceed as in the proof of Theorem 3. Let J be the set of column indices in which the pivots of D appear, and let K be the remaining indices. Since each xi
2

for j in J, appears in only one equation of the system, we can solve for each

in terms of the numbers c; a d the unknowns We can now assign j values arbitrarily to the \ and thus obtain a particular solution of the system. The theorem follows.

The procedure just described actually does much more than was necessary

to prove the theorem.

It tells us how to determine, in a particular case, whether

or not there is a solution; and it tells us,when there is one,how to express the solution set in parametric form as an m-plane in n'

Ccnsider the following example: E2:ample 4 Consider once again the reduced echelon matrix of Example 3:
.

The system

has no solution because the last equation of the system is

On the other hand, the system

does have,a solution. Fc~llowingthe procedure described in the preceding proof,

we solve for the unknowns xl, x2, and x4 as follows:

The general solution is thus the 2-plane in V5

specified by the parametric equation

Remark. Solving the system A-X = C in practice involves applying elementary operations to A

, and applying these same operations to

C.

A convenient way to perform these calculations is to form a new matrix from A by adjoining C as an additional column. This matrix is often called the
auqmented matrix of the system. Then one applies the elementary operations to this matrix, thus dealing with both A and C at the same time. is described in 16.18 of vol. I of Apostol. This procedure

Exercises 1.

Let A be a k by n matrix.

(a) If k < n, show that the system

A X = - has a solution different from G.(Is this result familiar?) What ' 0
can you say about the dimension of the solution space? (b) If k > n, show that
/--

therearevalues of C such that the system A X = C has no solution. '

2, Ccnsider the matrix A of p. A23.

of the system AbX = 2. arbitrary C?
3.

(a) Find the general solution

(b) Does the system A - X = C have a solution for

Repeat Exercise 2 for the matrices C, D l and E of p. A23. B be the matrix of p. A23.

4. &t the system

.-. 113.

(a) Find the general solution of

(b) Find conditions on a,b, and c

that are necessary and sufficient for the

C =

system B - X = C to have a solution, where C when you reduce B to echelon form?]

[~int:What happens to

5. Let A be the matrix of p. A20. Find conditions on a,b,c, and d

that are necessary and sufficient for the system A'X = C to have a :solution, where

@ k t A be a by n mti: ar:; let r be the rank of : A Let R be the set of all those vectors C of Vk fc,rwhich the system
has a solution. (That is, R A-X is the set of all vectors of the form

X ranges over Vn . ) ( a ) Show that R is a subspace of vk

(b) Show that R has dimension r. [Hint: Let W be the solution Then W has dimension m = n - r. C oose space of the system A-X = 9 . a basis A f A for W By adjoining vectors one at a time, extend m

, as

Show the vectors this to a basis Al,...,Am ,B1,...,Br for all of Vn A.B1 , A.3r span R; this follows from the fact that A.Ai = - for 0 all i. Show these vectors are independent.]
(c) Cclnclude that if r k, there are vectors C in Vk such that the system A - X = C has no solution; while if r = kt this system has a solution for all C. (This provides an.alternate proof of Theorem 6.)

...,

Let A be a k by n tnztrix. The columns of A, when looked at as elements of Vk , span a subspace of Vk that is called the column The row space and column space of A are very different, space of A have the same dimension ! Prove but it is a totally unexpected fact that they --this fact as follows: Let R be the subspace of Vk defined in Exercise 6. Show that R is spanned by the vectors A-E i,...,A-En ; conclude that R equals the column space of A.

Csrtesian equations

of !:-planes in

n '

'

There are two standard ways of specifying a k-plane One is by an equation in parametric form:

in Vn.

X = 2 + t l A L, + . . . +

vI.ere A , . . . ,

t?<q.l;,

ar2 in3ependent vectors in n '

(If tkess vectors xere

not independent, this equation would still specify an n-plane for some m, 5ut some work would be required to determine m. We normally require the vectors to b2 independent 51 the parametric form of the equation of a ?<-plane.) bother way to specify a plane in system of linear equations
Vn
is as the solution set of a

where the rows of A are independent. I f

has size k by

n, then

the plane in question has dimension n - k. The equation is called a caretesian form for the equation of a plane.

(If the rows of A were not

independent, then the solutionsetwould be either empty

, or

an m-plane

for some m, but some work would be required to determine m.) The- process of "solving" the system of equations A S X = C that

we described in the preceding section is an algorithm for passing from a cartesian equation for M to a parametric equation for M. 0 . can ask re

whether there is a process for the reverse, for passing from a parametric equation for M to a cartesian equation. Tk~eanswer is "yes," as we ' f yshould one shall see shortly. The other question one might ask is, ' I n care?" The answer is that sometimes one form is convenient, and other times the other form is more useful. Particularly is this true in the case of 3-dimensional space V3

, as we shall see.

Dflfinition. Let A be a matrix of size k by n. be the rows of A;

Let

let W be the subspace of Vn they span. Now the vector

X is a solution of the system A e X = g if and only if X is orthogonal

to each of the vectors that X
Ai

This statement is equivalent to the statement

is orthogonal to every vector belonging to W. The solution s2ace

of this' system is for this reason sometimes called the orthoqonal complement of W. It is often denoted ' W (read "W perp".)

We have the following result:

Theorem 1 If W is a subspace of Vn of dimension k, then , its orthogonal complement has dimension n - k. Fwthermore, W is the

orthogonal complement of W I ; that is, (WL ) I = W. Proof. Ttiat Theorem 3 j

Ai

wL

has dimension n

-k

is an imnediate consequence of

for W is the row space of a k by n matrix A with independent rows is the solution space of the system A - X = has dimension n - (n - k)

, whence w L

0.

W) The space (L'

, by what we just proved.

And it contdns each vector Ai

(since Ai *X = - for each X in W I. 0 )

,

Therefore it equals the space spanned by Theorem equation

X
=

Suppose a k-plane M

in Vn

is specified by the parametric

P + tA

1 1

...

- - tkq; I

where the vectors Ai and let B , B1,".,B m '

are independent. Let W be the space they span;

Elrn be a basis for W L . If B

is the matrix with rows

Or

then the equation

B*(x-P) = 0 ,

is a cartesian equation for M.

BiO

Proof. The vector X lies in M

W. This occurs if and only if X

if and only if X - P belongs to

- P is orthogonal to each of the vectors

a~d this occurs if and only if B*(X - P) = 0 Bit

.a

The preceding proof actually tells us how to find a cartesian equation for M. 0 . takes the matrix A whose rows are the vectors Ai; one finds re

a basis B1,...,B for the solution space of the system A - X = 2, using m the Gass-Jordan algorithm; and then one writes down the equation B.X = B O P

*
k

-.

We now turn to the special case of V3, whose model is the familiar 3-dimensional space in which we live. In this space, we have only lines (1-planes) and planes (2-planes) to deal with. A P ~ e can use either the w parametric or cartesian form for lines and planes, as we prefer. in this situation we tend to prefer: parametric form for a line, and
cartesian form for a plane.
Let us explain why.
If L is a line given in parametric form X = P + tAI then A is uniquely determined up to a scalar factor. (The point P is of course le not determined.) T i equation itself then exhibits some geometric information about the line; one can for instance tell by inspection whether or not two lines are parallel.
Or1

However,

the other hand, if M is a plane given in parametric form by the

equation X = P + sA + tB

one does not have as much geometric information to find a cartesian equation is

immediately at hand. However, 1st us seek

for this plane. We note. that the orthogonal complement of L(A,B) one-dimensional, and is thus spanned by a single non-zero vector

N = (a

a2, a3)

We call N a normal vector to the plane M

it is

uniquely determined up to a scalar factor. (1n practice, one finds N by solving the system of equations A-N = 0
-1

B = N = - .) 0
Then a cartesian equation for M is the equation

If P is the point has the form

(*)

(pl, p2, p3) of the plane MI

'

thisequation

al(xl - pl)

a2(x2 - p2)

a3(x3 - P3 ) = 0 .

We call this the equation -- plane throuqh

p = (pll P2' p3) with of the normal vector N = (a a2, a3 1.

WE have thus proved the first half of the following theorem: Theorem% If M is a 2-plane in V3, then M has a cartesian

equation of the form a1x1 + a2x2 + a3x3 - b

where N = (al, a2, a3) is non-zero. Conversely, any such equation is the cartesian equation of a plane in V3; the vector N to the plane. Proof. To prove the converse, we note that this equation is a system consisting of 1 equation in 3 unknowns, and the matrix A = [a1 a2 a3] has rank 1. Therefore the solution space of the system A.X = [b] is a plane
of dimension 3

is a normal vector

- 1

= 2.

a
useful; it

Nctw we see why the cartesian equation of a plane is

contins some geometric information about the plane. For instance, one can tell by inspection whether two planes given by cartesian equations are parallel.

For they are parallel if and only if their normal vectors are parallel,
and that can be determined by inspection of the two equations.
Similarly, one can tell readily whether the line X = P + tA is parallel to a plane 1 ; one just checks whether or not A is orthogonal

to the normal vector of M.

Mzny theorems of 3-dimensional geometry are now easy to prove.
us consider some examples.
Theorem 10. Three planes in V3 intersect in a single point

if and only if their normal vectors are independent. Proof. Take a cartesian equation for each plane; collectively, they form a system A - X The rows of
=

of three equations in three unknowns. The solution space of the system

are the normal vectors.

(which consists of the points common to all three planes) consists of a a single point if and only if the rows of Theorem --line.
Proof. -the two planes. Let Nl - X = bl and 11.

are independent.

Two non-parallel planes in V3

intersect in a straight

N2 -X = b 2 be ~artesianequations for

Their intersection consists of those pints X that satisfy

both equations. Since N matrix having rows N1

and N2

are not zero and are not parallellthe

aad N2 has rank 2. Hence the solution of this

system of equations is a 1-plane in V, Theorem

. 17
V,.

& Let L be a line, and M a plane, in

If L

is

paralleLto M, then their intersection is either empty or all of L. If

i not parallel to M, then their intersection is a single pint.

s
Proof. -LFgt L have parametric equation X = P +.tA; let M have We wish to determine fcr what values of

~artesianequation N - X = b. the point X

=

P + tA lies on the plane M; that is, to determine the

solutions of the equation

Now if L is parallel to MI

then the vector A

is perpendicular to

the normal vector N to M; that is, N . A = 0 .

In this case, the equation

holds for all t if it happens that N.P = b, and it holds for no t if N-P # b. it is empty. Thus the intersection of L and M is either all of L, or

On the other hand, if L is not parallel to M, then N - A # 0 .

In this case the equation can be solved uniquely for t. Thus the intersection

of L and M consists of a single point. Ex-ample 5 .

a
in V3

Ccnsider the plane M = M(P;A,B)

, where

P = (1, -7, 0) and A = (1, 1 1) and B = (-1, 2, 0). TO find a normal

vector N = (al, a2, a3) to M, we solve the system

One can use the Gauss-Jordan algorithm, or in this simple case almost by inspection. One can for instance set a2 = 1.

, proceed

Then the second

: equation implies that al = 2 and then the first equation tells us that a3 = -a1

- a2

-3.

The plane thus has cartesian equation

Exercises

1.

The solution set of the equation

,

is a plane in
2.

V3; write the equation of this plane in parametric form,

Write parametric equations for the line

through

(1,0,0)that

is perpendicular to the plane

3.

x1
- x3 = 5.
md

Write a parametric equation for the line through (0,5,-2) that

5x1+ x
2

is parallel to the planes 2x2 = x3

-7x = 4 . 3
MI

4 Show that if P and Q are two points of the plane .

the line through P and Q
5.

then

is coniained in M.
\

Write a parametric equation for the line of intersection of the

planes of Exercise 3.
6.

Write a cartesian equation for the plane through P = (-1,0,2)

and Q = (3,1,5) that is parallel to the line through R = (1,1,1)with direction vector A = (1,3,4). 7 Write cartesian equations for the plane M(P;A,B) in V4, . where P = (11 -11 0, 2) and A = (11 0, 1, 0) and B = (2, 1, 0, 1).
8.

Show that every n - 1 plane in Vn

6

is the solution set of

(all.. .an) #

an equation of the form alxl + and conversely.

...

anXn = b, where

2;

7.

Let M1 a ~ dM2 be 2-planes in V4;

assume they are not

parallel. What can you say about the intersection of MI and M2 ? Give examples to illustrate the possibilities.

The --

inverse of a matrix

We now consider the pro5lern of the existence of multiplicatiave

inverses for matrices. A t this point, we must take the non-commutativity of matrix.multiplicationinto account.Fc;ritis perfectly possible, given a matrix A, that there exists a matrix B such that A-B equals an identity matrix, without it following that B - A equals an identity matrix. Consider the following example:

r
-Then
I

Example 6. Let A and B be the matrices

A - B = I2

but B-A # I3

, as

1-1]
0

you can check.

Definition. Let A be a k b n matrix. A matrix B of size n y by k is called an inverse for A if both of the following equations hold:

A-B = Ik

and

B-A = In

W e shall prove that if k # n,

then it i impossible for both these s

equations to hold. Thus only square matrices can have inverses.

We also show that if the matrices aro, square and one of these equations

holds, then the other equation holds as v e l l ! Theorem 13. Let A be a matrix of size k by n. Then A has an inverse if and only if k = n = rank A. inverse is unique.

If A has an inverse, that

,Proof. Step 1. right inverse for A

B-A =

If B is an n by k matrix, we say B is a

if A-B = Ik

We say B

is a leEt inverse for A

if

In

QA&a.k
be the rank
that if A h a s a right inverse,
I

then r = k; a . if A has a left inverse, then r = n. rd of the theorem follows. Then A - B = I It k * follows that the system of equations A.X = C has a solution for arbitrary Fjrst, suppose B is a right inverse for A C, for the vector X = B.C is one such solution, as you can check.

Theorem 6 then implies that r must equal k. Second, suppose B is a left inverse for A. Then B - A = In

It

follows that the system of equations A - X = - has only the trivial solution, 0 for the equation A - X = g implies that B-(A*X) =

2,

whence X =

N ; the dimension of the solution space of the system A.X = Q is cw it follows that n
*

d. n- r ;

- r = 0.

Step 2. Now let A be an n by n matrix of rank n. We show there

is a matrix B such that A.B = In

.
In particular, it has a solution Let us choose

Because the rows of A are independent, the system of equations

A.X = C has a solution for arbitrary C.

when C is one of the unit coordinate vectors Ei in Vn. Bi

SG

that

for i = l...,n.

Then if B

is the n by n matrix whose successive equals the matrix whose successive

columns are B1,. ,Bn columns are

.. ' El,...,En i

the product A B . that is, A.B

=

I ,

.
.

Step 3. W show that if A and B are n by n matrices and e

A-B = I ,

then BmA = In.

The " i E n part of the theorem follows.

Let us note that if we apply Step 1 to the case of a square matrix

of size n by n

, it says that

if such a matrix has either a right

. inverse or a left inverse, then its rank must be n

Now the equation A.B = In says that A has a right inverse and that

B has a left inverse. Hence both

Step 2 to the matrix B,
B.C = In

and B must have rank n. Applying

we see that there is a matrix C such that

Now we compute

The equation B * C = In now becomes B - A = In

, as desired.

Step 4.

The computation we just made shows that if a matrix has

an inverse, that inverse is unique.

B has an left inverse A

Indeed, we just showed that if then A = C .

and a ri~htinverse C,

k:tus state the result proved in Step 3 as a separate theorem: Theorem 1 . If A and B are n by n matrices such that 4
A-B = I

':

n '

then B . A = I n .

El

We

now have a t h e o r e t i c a l criterion for t h e e x i s t e n c e

A-I p r a c t i c e ? in
B = d l 'if A

of

But how can one f i n d I. i n s t a n c e , how does one' compute

A .

For

is a given
it will suffice

nonsingular

by

matrix?

By Theorem 14,

to find a matrix

such t h a t

B = 13.

But t h i s problem is j u s t t h e sroblem of

s o l v i n g three systems of l i n e a r equations

Thus the Gauss-Jordan algorithm applies. An efficient way

to apply this aqlgorithm to the computation af
A -I

is out

lined on p . 612 of Apostol, which you should read now.

There is a l s o a f o n u l a f o r
determinants.
R~.mark

A-'

t h a t involves

It is given in the next s e c t i o n .

It remains to consider the question whether the existence

of the inverse of a matrix has any practical significance, or whether it is of theoretical interest only. In fact, the problem of finding the inverse of a matrix in an efficient and accurate lay is of great importance in engineering. One way to explain this is t o note that often in a real-life situation, one has a fixed matrix A, and one wishes to solve the system
A.X =
C

repeatedly, for many different values of C. Rather than solving

each one of these systems separately, it is much more efficient to find the inverse of A, for then the solution X = A".C can be computed by

sirple matrix multiplication.

Exercises

1.

Give conditions on a,b,c,d,e,E such that the matrix

is a right inverse to the matrix A of Example 6. Find two right inverses for A .
2.

Let A be a k by n matrix with k < n Show that A has .

no left inverse. *o .w

that if A has a right inverse, then that right inverse

is not unique.
3. Let B be an n by k matrix with k 4 n Show that B has
.
no right inverse. Show that if B has a left inverse, then that left
inverse is not unique.

Determinants -

The determinant is a function that assigns, to each square matrix

A, a real number. It has certain properties that are expressed in the

following theorem:
Theorem 15. There exists a function that assigns, to each n by n rlti tarx
A,

a real number that we denote by det A.

It has the following

properties:
(1)

If B is the matrix obtained from A by exchanging rows

i and j of A, then det B =

(2)

det A.
A

If B is the matrix obtained form

by replacing

row i of

hy itself plus a scalar multiple of row j

(3) If B
i

(where i # j), then det B = det A

i the matrix obtained from A by multiplying row i s

c,

of A by the scalar
4

then det B = c-detA

. .

If In

is the identity matrix, then det In = 1

We are going to assume this theorem for the time being, and explore
some of its consequences. We will show, among other things, that these
four properties characterize the determinant function completely. kter

we shall construct a function satisfying these properties.

First we shall explore some consequences of the first three of these properties. We shall call properties (1)-(3) elementary listed in Theorem 15 the

row properties of
f

the

detsrminant function.

Theorem 1 . t 6 mti: ar;

5e a function that assigns, to each

n by n

A, a real number. Scppose f satisfies the elementary row

properties of the determinant function. Then for every n by n matrix A,

( *)

f(A)

f(In).det A

This theorem says that any function f that satisfies properties (I), ( 2 ) , and (3) of Theorem 15 i a scalar multiple of the determinant s function. It also says that if f satisfies property (4)as well, then

E must equal the determinant function. Said differently, there is at

most one --

function that satisfies all four conditions. 1 First we show that if the rows of A are dependent, . Equation ( * ) then holds trivially in this case.

Proof. St= - -

then f ( A ) = 0 and det A = 0 .

Let us apply elementary row operations to A to brin~it to echelon form B. We need only the first two elementary row operations to do this,

and they change the ~ l u e s of f and of the determinant function by at

. most a sign. Therefore it suffices to prove that f ( B ) = 0 and det B = 0

The last row of

is the zero row, since A has rank less than n.

If

we multiply this row b the scalar c, we leave the matrix unchanged, and y hence we leave the values of f and det urlchanged. On the other hand, this operation multiplies these values by c . conclude that f ( B ) = 0 acd d~t = 0 B . Since c is arbitrary, we

Step 2 .
independent.

Now let us consider the case where the rows of A are Again, we apply elementary row operations to A. Hcwever, and det do not

we will do it very carefully, so that the values of f change.

As

usual, we begin w i t h t h e first column.

If a l l

e n t r i e s are zero, nothing remains t o be done with t h i s column.

We move on to consider columns 2,...,n and begin the process again.

Otherwise, w e f i n d a non-zero e n t r y i n t h e f i r s t column.

I f necessary, we exchange rows t o bring t h i s entry up t o the

upper left-hand corner: this changes the sign of both the func

tions

and

det,

s o we then m u l t i p l y this r o w by

-1

to

change the s i g n s back.

Then we add m u l t i p l e s of t h e f i r s t row

t o each of t h e remaining rows s o as t o make a l l t h e remaining e n t r i e s i n t h e f i r s t column i n t o zeros.

By t h e preceding theorem
f

and i t s c o r o l l a r y , this does n o t change t h e values of e i t h e r

or

det.
Then w e repeat the process, working w i t h t h e second

column and w i t h rows

2 . n .

The o p e r a t i o n s we a p p l y w i l l

n o t a f f e c t t h e z e r o s w e already have i n column 1.

Ssnce the rows of the original matrix were independent, then we do

not have a zero row at the bottom when we finish, and the "stairsteps"

of the echelon form go wer just one step at a time.

In this case, w e have brought t h e matrix t o a form where a l l of

the e n t r i e s below the main diagonal a r e zero. c a l l e d upper t r i a n g u l a r f orm.) e n t r i e s are non-zero.

(This i s what i s

Furthermore, all the diagonal

f

Since the values of

and
B,

det

remain

the same i f w r e p l a c e e

by this new matrix

it now s u f

fices t o prove our formula f o r a m a t r i x of the form

where t h e diagonal e n t r i e s a r e non-zero.

St.ep 3 We show that our formula holds for the matrix B. .

To do

this we continue the Gauss-Jordan elimination process. By adding a multiple

of the last row to the rows above it, then adding multiples of the next-

to-last row to the rows lying above it, and so on, we can bring the matrix to the form where

all the

non-diagonal entries vanish. This form is called

diaqonal form. The values of both f and det remain the same if we replace

by t h i s new matrix

C .

So now it suffices to prove our

formula for a m a t r i x of the form

0

..
B

'
.

bnn
.

(Note that the diagonal entries of

remain unchaaged when

we apply the Gauss-Jordan process to eliminate a11 t h e

non-zero entries above the diagonal. entries of

Thus the diagonal

are the same a s t h o s e o f

l/bl

B.)
This action m l t iplies the
,

WE?multiply.the first row of C by

values of both f

and

det by a Eactor of

l/bll. Then we multiply the

second row by

l/b22,

the third by

l/b33,

and so on. By this process, We conclude that

and

we transform the matrix C into the identity matrix In.

det In

(l/bll) ...( l/bNI) det C.

Since det In = 1 by hypothesis, it follows from the second equation that

det C = bll b22

... bnn

'

Then it follows from the first equation that

E(C)
=

f(In)- det C,

as desired.

Besides proving the determinant function unique, this theorem also tells us one way to compute determinants. O r d applies this version of the Gauss-Jordan algorithm to reduce the matrix to echelon form. If the matrix that results has a zero row, then the determinant i zero. Otherwise, the matrix that results is in upper triangular s form with non-zero diagonal entries, and the determinant is the product
of the diagonal entries.
.....
< ,

, : -

-lz .

. :

.-

.,; '

:,-::=. ,-

.). . : *.. ... <,,, .

.. .....! =. : :c ,

'-'2.

2 cr:. , , l

The proof of this theorem tells us something else:

If the rows of

A are not independent, then det A = 0, while if they are independent, then det A # 0. We state this result as a theorem: Theorem 16. Let A be an n by n matrix. if and only i f det A # 0 Then A h . rank n as

.
.
is non-singular.

An n by n matrix A fur which det A # 0 is said to be non-sinqular

This theorem tells us that A has rank n if and only if A Now we prove a totally unexpected result: Theorem 17.
Let

A m d B k n by n matrices. det (A-B)

=

Then

(det A . (det B) )

Proof. This theorem is almost impossible to prove by direct computation. Try the case n = 2 if you doubt me ! Instead, we proceed in another direction:

Let B be a fixed n by n matrix. Let us define a function f of

n by n matrices by the formula

f(A) = det(A9~).

We shall prove that f satisfies the elementary row properties of the

determinant function. From this it follows that

f(A) which means that det(A*B) = det(In.B)* det A
= =

f(In)- det A

det B

. det A

and the theorem is proved.

are the rows of A, considered

as row matrices, then the rows of A-B are (by the definition of matrix
Now exchanging rows multiplication) the row matrices A . , . , ; 1 B..AB
i and j of A, namely Aj ai A rd has the effect of exchanging rows j' i and j of A.B. Thus this operation changes the value of f by a factor of -1. Similarly, replacing the ithrow Ai of A by Ai + FA

First, let us note that if A1,

...,An

has the effect on A-B of replacing its ith row Ai.B by

( A ~ CA.).B t

A ~ . Bt c A , - B
3

(row i of A . B )

+ c(row j of

A-B).

Hence it leaves the value of f unchanged. Finally, replacing the ith row Ai of A by cAi h s the effect on A.B
by

of replacing the ith row Ai.B

(cAi)- B = c (Ai-B) = c (row i of A-B). Hence it multiplies the value of f by c.

The determinant function has many further properties, which we shall

not explore here. (One reference book on determinants runs to four volumes!)

We shall derive just one additional result, concerning the inverse matrix.

Exercises 1.

Suppose that

satisfies Lhe elementary row properties of

the determinant function.

suppose also t h a t

x, y, z are numberssuch t h a t

mrnpute the value o f

for each o f the following matrices:

2.

Let

f b the Function of Esercise 1. e

Calculate

f(In).

Express

i n terms of the determinant function.

7 .

Compute the determinant of the following matrix, using GaussJordan elimination.

4.

Determine whether the following sets o f vectors are l i n e a r l y

independent, using determinants .,

(a) Al = ( l , - l , O ) r
(b)

= (Oflf-llr

A3. = ( 2 t 3 f - 1 ) 9
f

= (1,-1.2,1),

A.2

= (-lt2,-1,O)

A3 = (3t-ltltQ)

A4 = . ( 1 , 0 , 0 , 1 )

(c) P ,

= ( ~ . O , O ~ , O 42 = , ( ,~)
A5

I ~ ~ ~ o ~ A3
~= o ( )I o

A4 = ( l . l f O , l t l ) )

= (1,010,010)
A3

~ O ~ ' ~ ~ O , ~ ) ,

(dl

= (1,-11,

A2 =

( O t l ) ,

= (lfl).

A - formula for A
'j

We know that an n by n matrix A has an inverse if and only if

it has rank n, and we b o w that A has rank n if and only if det A # 0. Now we derive a formula for the inverse that involves determinants directly. We begin with a lema about the evaluation of determinants.

Lemma --

18.

Given the row matrix [ a (n-1) by

... an] , let us define a

function f of

(n-1) mtrtrices B by- the formula

f(B) = det

, I

where B1 consists of the first j-1 columns of B, arid B cc~nsists 2 of the remainder of B. Then

Proof. You can readily check that f satisfies properties ( 1)-( 3) of the determinant function. Hence- f (B) = f(I 1) -det B. nW

compute

f(In) = det n-j where the large zeros stand for zero matrices of the appropriate size.
A sequence of .j-1 interchanges of adjacent rows gives us the equation

One can apply elementary operations to this matrix, without changing the
value of the determinant, to replace all of the entries al,...,aj-l,aj+l,...,a
n
by zeros. Then the resulting matrix is in diagonal form. We conclude that

Corollary

Consider a n by n

n mtrix of the form

where Bl,

...,B4

are matrices of appropriate size. Then det A = (-1)

Proof. A sequence of i-1 interchanges of adjacent rows wilt king the matrix A to the form given in the preceding l m a .

Definition. In general, if A matrix of size (n-1) by the j th column of A

I

is an n by n matrix, then the

(n-1) obtained by deleting the ith row and

is called the (i ,j )-minor of A, and is denoted Ai j.

me preceding corollary can then be restated as follows:

Corollary 20. If all the entries in the jth column of A are zero except for the entry a i j in row i, then det A = (-1) i+ j aij-det Aij.
~ t n w r ~ e

( - 1) i + j

det Aij

that appears in this corollary is also

Note that

given a special name. It is called the (i,j)-cofactor of A. the signs (-l)itj follows the pattern

N ; we derive our formula for A - ~ . cw

Tl~eorern21. k t A be an n by n matrix with det A # 0.

1

If A * B = I n'

then

= (-l)jci det A . ./det A . bi j 31 (That is, the entry of B in row i and column j equals the ( j ,i)

cofactor of A,

divided by det A.
A

This theorem says that you can compute

B by computing det
(n-1) matrices. low dimensions!)

and the determinants of n2 different (n-1) by

This is certainly not a practical procedure except in

Proof.

k t

X denote the jth col-

Because A-B = In '

of B. Then xi - 'j i. the column matrix X satisfies the equation

e ere
of 1
I

is the .column matrix consisting of zeros except for an entry

Furthermore, if Ri denote the ith mlumn of A, then
,

in row j . )

because A - In =

, we1 have

the equation In) = A.Ei = A 1 .

,

~ - ( column ~ i ~ of

Now we introduce a couple of weird matrices for reasons that will become
clear. Using the two preceding equations, we put them together to get
the following matrix equation:

It turns out that when we take determinants of both sides of this equation,
we get exactly the equation of our theorem! First, we show that
det [El

... Ei-l

X Ei+l

... En]

= xi

Written out in full, this equation states that

det

If x. = 0, 1

this equation holds because the matrix has a zero row. If

xi # 0, we can by elementary operations replace all the ectries above

and beneath xi

in ts column by zeros. The resulting matrix will k

i

in diagonal form, and its determinant will be xi.

TFus the determinant of the left side of sqxation ( * ) equals (det A) .xi, which equals (det A)*bij. We now compute the determinant of the right side of equation ( * ) . Corollary 20

applies, because the th column of this matrix consistsof zeros except for
i

an entry of

in row j

Thus the right side of ( * ) equals (-l)jti times

the determinant of the matrix obtained by deleting raw j and column i.

This is exactly the same matrix as we would obtain by deleting rcw j and
column i of A. Hence the right side of ( * ) equals (-l)jti det R . . , J1

and our theorem i proved. s

If A is a matrix with general entry a i (denoted

in

-1.
Rc;mark

row i and colrmn j

then the transpose of A is a , ,

J1

i the matrix s

whose entry In row i and column j

Thus i f
A

.
has sire n by k; it

has size

K by n, then A'=

can be pictured as the mtrix obtained by flipping A around the line

y

-x.

For example,

Of murse,i A as A.

is square, then the transpose of A has the same dimensions

Using this terminology, the theorem just proved says that the inverse of

A can be computed by the following four-step process:

(1) Fcrm the matrix whose entry in row i and column j is the

(This is called the matrix zf minor determinants.)

det *ij.
( 2 ) Prefix the sign (-1) to the entry in row i and column j , for

n* rr

each entry of the matrix. (This is called the matrix

-of

cofactors.)

(5) Transpose the resulting matrix.

( 4 ) Divide each entry of the matrix by

det

A.

In short, this theorem says that

A-1 =

-
A ) ~ ~ .
(cof det A

This formula for A ' ~ is used for rornputational purposes only for 2 by 2
or 3 b 3 matrices; the work simply gets too great otherwise. But it is
y important for theoretical purposes. For instance, if the entries of A

are contin~ousfunctions of a parameter t, this theorem tells us that

the e ~ t r i e s of A-'

are also continuous functions of t, provided det

is never zero.

R ~ m r k2.

This formula does have one practical consequence of great

deb A

importance. It tells us that i f of A,

is small as cunpared with the entries is likely to result in a

then a small change i the.entries of A n

large change in the ccmputed entries of A-I. This means, in an engineering problem , that a small error in calculating A (even round-off error) may result i a n

gross error in the calculated value of A-l.A matrix for which d e t A relatively small is saidtobeill-conditioned.

i s

If such a'mtrix arises in practice,

one usually tries to refomlate the problem to avoid dealing with such a matrix.

aL3

Exercises

use t h e formula f o r

to f i n d the inverses of the follow

LOW

ing matrices , assuming the usual definition of the deter-,inant in

dirnens ions.

(b)

c :n
0 c d
A

,assmFng

ace

f O .

2 . L e t

ba a square matrix all of whose entries are integers.

show that i f
integers.

dat A =

21,

then all the entries of

A-'

are

3 . Consider the matrices A,B,C,D,E

of p. A.23.

Which of these

matrices have inverses?

4 Consider the following matrix function!

.

For what values of

t does A-'

exist? Give a formula for A-Iin t e r m

5.
Or

Show that the conclusion of Theorem20 holds if

A has an entry

ail

in row i and c l a m j, =d all the other entries in

row

i equal 0.

'. b
m
by
I(,

Theorem L e t A,
and
m bl
mr

BI

C be matricas of .size lc by

k,

and

respectively. Then

rB z]

= (det

A)

(det (2).

(Here 0 is the zero matrix of appropriate size.)

Prwf.
k t

B and C k fixed. e

Fcr each

k by

matrix

A,

define

( a ) Show

satisfies the elementary row properties of the determinant

function.
(b) U s e Exercise 5 to show that
( c ) Cctrrplete the proof.

f ( I ~ = det C. )

Ccnstruction of the determinant when n 5 3. T'i~eactual definition of the determinant function is t r least interesting 'e

part of this entire discussion. The situation is similar to the situation with respect to the functions sin x, cos x, and ex. You trill recall that
0

their actual definitions (as limits of

Pbrseries) were not nearly as interesting

as the properties we derived from simple basic assumptions about them.

We first consider the case where n < 3, which isdoubtless familiar

to you. This case is in fact all we shall need for our applications to calculus.

We begin with a lema:

L m a 21. Let e m
f(A)

be a real-valued function of n by

n matrices.

Suppose that:
( i ) Exchanging any two rows of A changes the mlue of f by a factor

of -1.
( i i ) For each i ,

is linear as a function of t h e

ith row.

Then f satisfies the elementary row properties of the determinant function. Proof. By hypothesis, f satisfies the first elementary row property.

We check the other tu'0.

Let
A,,

...,A,

be t h e rows

of A.

To say that

is linear

as a function of row f

alone i to s

say t h a t (when

f is written as a,function

of the rows of
(*)

A):

f A ,,..., cx + dY, (

... ,An)

= cf(A1,

...,X,. ..,An)

+ dE(A1,...,Y,...,A,,).

where cX + dY and X and Y appear in the ith component. The special case d = 0 tells us that multiplying the ith row of A b y c has the effect of multiplying the mlue of f by c .

We now consider the third type of elementary operation.

Suppose that B is the matrix obtained by replacing row i

of A 5 y

itself plus c times row j. We then compute (assuming j > i for convenience in notation),

The second term vanishes, since two rows are the same. (Exchanging them does
not change the matrix, but by Step 1 it changes the value of f by a factor
of - 1 . ) ( J Definition. We define

det [ a ]
=

a.

det

bl

b2

a l b ~ - a2bl.

q,eorem 22.

The preceding definitions satisfy the four conditions

of the determinant functi on.

Proof. - The fact that the determinant of the identity matrix is follows by direct computation. of the preceding theorem hold
.

It then suffices to check that (i) and (ii)

.
,

Irl the 2 by 2 case, exchanging rows leads to the determinant bla2- b2al

which is the negative of what is given.

In the 3 by 3 case, the fact that exchanging the last two rows changes the sign of the determinant follows from the.2 by 2 case. The fact t h a t exchanging

the first two rows also changes the sign follows similarly if we rewrite the formula defining.t h e determinant in the form

Finally, exchanging rows 1 and 3 can be accomplished by three exchanges of

adjacent rows [ namely, (A,B,C)
the sign of the determinant.

->

(A,C,B)-9 (C,A,B) -3 (C,B,A) 1, so i changes t

s y To check (ii) i easy. Consider the 3 b 3 case, for example. We

h o w that any function

of

the form
= axl + b w + a 2 3

E(X) = [ a b c ] - X
is linear, where X is a vector in V
f(X)

The function

det

has this form, where the coefficients a, b, and c involve the constants
bi and c
j *

Hence f is linear as a function of the first row.

is linear as a function

The "row-exchange propertyw then implies that E

of each of the other rows. 0

Exercise

*l.

Let us define

det

(a) Show that det Iq = 1 .

(b) Show that excha 'ng any two of the last three rows changes the sign of the

determinant.

(c) Shwthat exchanging the first two rows changes the sign.
expression as a sum of terms involving det

[Hint:

Write the

pi

bjJ
(d) Show that exchanging any two rows changes the sign.

(e) Show that det

P i

'7 j.

is linear as a function of the first row.

(f) Conclude that det is linear as a function of the

ith row.

( g ) Conclude that this formula satisfies all the properties of the determinant

function.

tive integers 1, 2, . . . , k and write them down in some arbitrary order, say jl, j z , . . . , j h . This new ordering is called a permutation of these integers. For each integer ji in this ordering, let us count how many integers follow it in this ordering, but precede it in the natural ordering 1, 2, . . . , k. This number is called the number of inaersions caused by the integer j;. If we determine this ilumber for each integer ji in the ordering and add the results together, the number we get is called the total number of inversions which occur in this ordering. If the number is odd, we say the permutation is an odd permutation; if the number is even, we say it is an even permutalion. For example, consider the following reordering of the integers between 1 and 6:
2, 5 , 1, 3, 6, 4.

Construction of fhe Determinant

unction^^ Suppose we take the posi-

If me count up the inversions, we see that the integer 2 causes one inversion, 5 causes three inversions, 1 and 3 cause no inversions, 6 causes one inversion, and 4 causes none. The sum is five, so the permutation is odd. Xf a permutation is odd, me say the sign of that permutation is - ; if A useful fact about the sign of a permuta it is even, we say its sign is tion is the following:

+.

Theorem 22.If we interchange two adjacent elements of a per mutation, we %hange the sign of the permutation.

jl, tation

ProoJ. Let us suppose the elements ji and ji+1 of the permutation . .' , ji, ji+l, . . . ,j k are the two we interchange, obtaining the permu-

j ~ ,.

. .

jlj~, +,

. . . ,jk.

The number of inversions caused by the integen j l , . . . , ji-1 clearly is the same in the new permutation as in the old one, and so is the number of inversions caused by ji+t, . . . ,js. I t remains to compate the number of inversions caused by ji+land by ji in the two permutations. Case I: j r precedes j,-+ln the natural ordering 1, . , k. I n this case, i the number of inversions caused by ji is the same in both permutations, but the number of inversions caused by ji+lia one larger in the second in permutation than in the first, for ji follows j4+* the second permutation, but not in the first. Hence the total number of inversions is increased by one.
Case 11: j i follows js+l in the natural ordering 1, . , k. I n this case, is the number of inversion caused by jicl the same in both permutations,

. .

. .

but the number of inversions caused by ji is one less in the second permutation than in the first.
I n either case the total number of inversions changes by one, 80 t h a t the sign of the permutation changes. U

EXAMPLE.f we interchange the second and third elements of the I permutation considered in the previous example, we obtain 2, 1, 5, 3, 6, 4, in which the total number of inversions is four, so the permutation is even.
Definition.

Consider a k by k matrix

Pick out one entry from each row of A ; do this in such a way that these entries all lie in different columns of A . Take the product of these entries, and prefix a sign according as the permutation jl, . . , j k is even or odd. (Note that we arrange the entries in the order of the rows they come from, and then"we compute the sign of the resulting permutation of the column indices.) If we write down all possible such expressions and add them together, the number we get is defined to be the determinant of A .

REMARK.We apply this definition to the general 2 by 2 matrix, and obtain the formula

If we apply i t to a 3 by 3 matrix, we find that

The formula for the determinant of a 4 by 4 matrix involves 24 terms, and for a 5 by 5 matrix it involves 120 terms; we will not write down these formulas. The reader will readily believe that the definition we have given is not very useful for computational purposes!

The definition is, however, very convenient for theoretical purposes.

Theorem 24. The determinant of the identity matrix is 1.

. ,:!, -" Proof. Every term in the expansion of det In has a factor
.

, of zero in it except for the term alla22...alck and this term equals 1.
/

Theorem $5.

If A '

is obtained from A by interchanging rows

i and i+l, then det A ' =

det A.

Proof. Note that each term

in the expansion of det A' also appears in the expansion of det A, because we make all possible choices of one eritry from each row and column when we write down this expansion. The only thing we have to do is to compare what signs this term has when i t appears in the two expansions. Let a~r, - . n i j r a i f l , j , + ,. . . nkjk be a term in the expansion of det A . If we look a t the correspondi~igterm in the expansion of det A', we see that we have the same factors, but they are arranged diflerenlly. For to compute the sign of this term, we agreed to arrange the entries in the order of the row8 they came from, and then to take the sign of the correspondiilg permutation of the column indices. Thus in the expansion of det A', this term mill appear as

l'lie permutation of the columr~indices here is the same as above except ji that eleme~lts and j i + l have been interchanged. By Theorem 8.4, this means that this term appears in the expansion of det A' with the sign opposite to its sign ill the expansion of det A . Since this result holds for each term in the expansion of det A', we have det A' = - det A . - . 0

The function det is linear as a function of the ith row.

th
Proof. -- Suppose we take the constant matrix A, and replace its i
Theorem 2 ' 6. row by the row vector

[xl

...

]\A

When we take the determinant of this

new matrix, each term in the expression equals a constant times x , for j some j. (This happens because in forming this term, we picked out exactly one
entry from each row of A . )

Thus this function is a linear combination

of th2 components xi; that is, it has the form

LC, ....

ck . x l

for some constants ci

'

Exercises

1,

Use Theorem 2s.' to show that exchanging

two rows of A

changes the sign of the determinant.

2. Consider the term

the determinant.

all; a2j2 "' %jk

in the definition of

(The integers j l , j2, . . . , j k are distinct.) Suppose

we arrange the factors in this term in the order of their column indices, obtaining an expression of the form

Show that the sign of the perm-tion permutation jl,j2,...,jk Ccnclude that det
3.'
= det A

il,i2,

...,ik

equals the sign of the

in general.

k t A

be an n by n matrix, with general entry aij in

Let m be a fixed index. Show that

row i

and column j.

Here A denotes , as usual, the (m,j )-minor of A. This formula i s mj

called theMformula for expanding det A according to the cofactors o f the nth row. [Hint: Write the mth row as the sum of n vectors, each

of which has a single non-zero component. Then use the fact that the determinant function is linear as a function of the mth row. 1

Tl~ecross-product 2 -

If A = (alIa2' a 3 ) and B = (blIbZI b )

3

are vectors in

v3'

we define their cross product AXB

=

(det

r2

be the vector

a] . 2 3

,-

det

2)

dtb al a2 ) e[l

bd

We shall describe the geometric significance of this product shortly.

But first, we prove some properties of the cross product:
Theorem 27. Fc'rall vectors A, B in V3,
(a)

we have

B%A

-AxB.

(11) Ar(B + C) = A X B + A X C
(Ei+C)XA

= BkA
C(AXB)

C%A.

(c)

(CA)X

B =

= AX(CB)

(d)

AXB

i orthogonal to both A and B

s .

Proof.

(a) follows becauseexhanging two rows of a determinant

ck-angesthe sign; ar:d (b) and (c) follows because the determinant is linear

as a function of each row separately. To prove ( d ) , Lie note that if

C = (cIIc2, c3)

then

fcl

'31

by definition of the determinant. It; follows that A-(AxB) = B - ( A X B ) = 0 because the determinant vanishes if two rows are equal. The only proof that requires some work i (e). For this, we recall that s

(a + bl2 = a2 + b2 + 2ab, and (a + b + c12 =

Equatiori ( e ) can be written in the form

a2 +

b2 + c2 c 2ab + 2ac + 2bc

We first take the squared terms on the left side and show they equal
the right side. Then we take the "mixed" terms on the left side and show
they equal zero. The squared terms on the left side are

which equals the right side,

x3

2 (a,b.)
1 3

i,j = 1

The mixed terms on the left side are

In the process of proving the previous theorem, we proved also

the following:

.: Theorem 28. Given A, B, C

Proof. --

, we have

A- (Bx C) = ( A x B) .C.

This fbllows from the fact that

&finition.
of vectors of V3

The ordered 3-tuple of independent vectors (A,B,C) is called a positive triple if

A-(B% C)

> 0.

Otherwise, it i called a neqative triple. A positive s

triple is sometimes said to be a riqht-handed triple, and a negative one is said to be left-handed.

The reason for this.terminology is the following: (1) the triple

(i, &) i,

is a positive triple, since i - ( d x k ) = det I3 = 1

and

(2) if we draw the vectors

i, i, and &

in V3

in the usual way,

and if one curls the fingers of one's right hand in the direction from the

first to the second, then one's thumb points in the direction of the third.

&" j- >
1

Furthermore, if one now moves the vectors around in V

3'

perhaps changing their

lengths and the angles between them, but never lettinq them become dependent,
=d if one moves one's right hand around correspondingly, then the
fingers still correspond. to the new triple (A,B,C) in the same way, and
this new triple is still a positive triple, since the determinant cannot
have changed sign while the vectors moved around.(Since they did not become
dependent, the determinant did not vanish.)

Theorem .29. Let A and B be vectors in V are dependent, then A X B = 0 Otherwise, AXB .

3'

If A and: B :

is the unique vector

3

orthogonal to both A and B having length llA ll IlBll sin

(where 8

is the angle between A and B), such that the triple (A,B,A%B) forms a positive (i.e.,right-handed) triple.

ProoF. also have

W know that e

AXB

is orthogonal to both

and

B. \v'e

II~xaU2 =
=
Finally, i f

~ I A C - I I B~ ?A.B) 2 (
\(A[\~~ . (A,B,C)

( 1 1 - ros 1

1 I

1 ~ 1 1~ (~ B A s i, n 2 e ~

C = A X B , then

is a positive t r i p l e , since

Polar coordinates

Let A = (a,b) be a point of V2 different from Q. We wish to define what we mean by a "polar angle" for A. The idea is that it should be the angle between the vector A
and the unit vector i = (1,O). But we also wish to choose it so its value reflects whether

A lies in the upper or lower hdf-plane. So we make the following definition: Definition. Given A = (a,b) # Q. We define the number

(*I
and to be -if b

8 = * arcos (Aei/llAII)

t o be a polar annle for A, where the sign in this equation is specified to be

+ if b > 0,

< 0. Any number of the form 2mn + 0 is also defined to be a polar angle

for A.

If b = 0, the sign in this equation is not determined, but that does not matter. For if A = (a,O) where a

> 0, then arccos (A-i/llAII) = arccos 1 = 0,

SO

the sign does not

matter. And if A = (-a,O) where a > 0, then arccos (A.L/IIAII) = arccos (-1) = T . Since

the two numbers

+ T and - T differ by a multiple of 27, the sign does not matter, for
T.

since one is a polar angle for A, so is the other.

Note: The polar angle B for A is uniauely determined if we require -n < f? <
But that is a rather artificial restriction. Theorem. 2 2 'I2 a ~ o i n of V2. Lgt r = (a +b ) t = IIAll; l a 8

A = (a,b) # Q

a - polar annle for A. Then A = (r cos 8, r sin 8).

Proof. If A = (a,O) with a > 0, then r = a and 0 = 0 + 2ms; hence r cos 8 = a and r sin 0 = 0. If A = (-a,O) with a

> 0, then r = a and 0 = a +2m7r, so that

r cos 0 = -a and r sin B = 0.

Finally, suppose A = (a,b) with b f 0. Then A.C/I(AII = a/r, so that

0 = 2ma e arccos(a/r).
Then a/r = cos(*(&2rn~))= cos 0, or a = r cos 0. Furthermore,

b2 = r2 - a2 = r2 (l-cos2 8) = r2 sin2 8,
so
b = *r sin 8.

We show that in fact b = r sin 8. For if b > 0, then 0 = 2ma + arccos(a/r), so that

and sin 19 is positive. Because b, r, and sin 0 are all positive, we must have b = r sin B rather than b = -r sin 8. On the other hand, if b < 0, then 0 = 2m7r - arccos(a/r), so that
2mn-a< B<2ma

and sin 0 is negative. Since r is positive, and b and sin 8 are negative, we must have
b = r sin 0 rather than b = -r sin 8. o

(~lanetarv Motion (

In the text, Apostol shows how Kepler's three (empirical) laws of planetary motion
can be deduced from the following two laws:
(1) Newton's second law of motion:

F = ma.

(2) Newton's law of universal gravitation:

Here m, M are the masses of the two objects, r is the distance between them, and G is a universal constant. Here we show (essentially) the reverse-how Kepler 's . More precisely, suppose a planet Newton's laws can be deduced from

".,.kmLM
xy plane with the

origin. Newton's laws tell us that the acceleration of P is given by the equation

That is, Newton's laws tell us that there is a number A such that
r and that X is the same for all pIanets in the solar system. (One needs to consider other

X
!!=-7"r,

systems to see that A involves the mass of the sun.) This is what we shall prove. We use the formula for acceleration in polar coordinates (Apostol, p. 542):

We also use some facts about area that we shall not actually prove until Units VI and
VII of this course.

Theorem. S u ~ ~ oa ela net P moves in the xy plane with the sun at the orign. s (a) Ke~ler's second law im~lies that the acceleration is radial. (b) Ke~ler'sfirst and second laws imply that

a,=- A~ 2 4 , I

\I

where Xp

~ a number that mav depend on the particular planet P.

vector as the planet moves from polar angIe Q1 to polar angle 02:

(c) Keder's three laws i m ~ l v that Xp is the same for d la nets. Proof. (a) We use the following formula for the area swept out by the radial

Here it is assumed the curve is specified by giving r as a function of 0.

Now in our present case both 0 and r are functions of time t. Hence the area swept out as time goes from to to t is (by the substitution rule) given by

Differentiating, we have dA = 1I2 dB , which is constant by Kepler's second law. That is,

(*I
for some K.
Differentiating, we have

The left side of this equation is just the transverse component (the po component) of a!

Hence a is radial.
(b) To apply Kepler's first law, we need the equation of an ellipse with focus at

the origin. We put the other focus at (a,O), and use

,-,

the fact that an ellipse is the locus of all points (x,y) the sum of whose distances from (0,O) and (a,O) is a constant b

> a.

The algebra is routine:

or

+ Jr2 - 2a(r

cos 0)

+
2

a'=
2

b,

r2 - 2a(r cos 8) + a2 = (b-r)2 = b2 - 2br 2br - 2ar cos 0 = b - a

+ r 2,

r =1

-e

b2 - a2 cos 0 , where c = -26-. and e = a/b.

i

e
(The numberhis called the eccentricity of the ellipse, by the way.) Now we compute the

radial component of acceleration, which is

Differentiating (**), we compute

-1
(1-e cos 0)
Simplifying,

dr = a (-1)r 2(e sin e)=B 1 d Then using (*) from p. B60, we have 1 dr = $e sin 6')K. Differentiating again, we have

d2r = - -(e cos O)zK, 1 d0 C dt

d2r 1 -Z = - &e cos 8) dt

Similarly,

K , using (*) to get rid of de/dt.

-r

[q

= - r[K

] 2 using (*) again to get rid of dO/dt. 7

Hence the radial component of acceleration is (adding these equations)

K~ K~ K~ e cos [ 1 - 3 e cos B ) = - - = - C e + TI ~ 3 r r K e ' = -5[ cos 8 C cOs C r
+

81

Thus, as desired,

(c) To apply Keplerys third law, we need a formula for the area of an ellipse, which will be proved later, in Unit VII. It is Area = T(ma'or axis) (minor axis) a
2

The minor axis is easily determined to be

given by:

minor axis =

= m
d .

It is also easy to see that major axis = b. Now we can apply Kepler's third law. Since area is being swept out at the constant rate ;K, we know that (since the period is the time it takes to sweep out the entire area),

1
Area = (2K)(Period).
Kepler's third law states that the following number is the same for all planets: (major axis)"(major axi s) 3

- -r (major axia)l(minor a w i ~ ) ~ / ~ ~ (major axi's )

4
2
A

16

=b

r2 lT i n* i

major axis

;;Z

Thus the constant

Xp is the same for all planets.

(1) L e t L b e a l i n e i n V, with d i r e c t i o n vector A ; l e t P b e a p o i n t n o t on L. S h o w t h a t t h e p o i n t X o n t h e l i n e L closest to P s a t i s f i e s the c o n d i t i o n t h a t X-P i s p e r p e n d i c u l a r to A.

(2) F i n d p a r a m e t r i c e q u a t i o n s f o r t h e c u r v e C c o n s i s t i n g of a l l p o i n t s of V % e q u i d i s t a n t f r o m t h e p o i n t P = ( 0 , l ) a n d t h e l i n e y =-I. If X i s a n y p o i n t of C, s h o w t h a t t h e t a n g e n t v e c t o r to C' a t X m a k e s e q u a l a n g l e s w i t h t h e v e c t o r X d P a n d t h e v e c t o r j. ( T h i s i s t h e r e f l e c t i o n p r o p e r t y of t h e p a r a b o l a . ) (3) C o n s i d e r t h e c u r v e f ( t ) =(t,t cos ( d t ) ) f o r 0 c t I 1, = (0,O) f o r t = 0.

3

T h e n f is c o n t i n u o u s . L e t P be t h e p a r t i t i o n

P = { O , l / n , l / ( n - 1 ) ,...,1/3,1/2,1].
D r a w a p i c t u r e of t h e i n s c r i b e d polygon x ( P ) in t h e c a s e n = 5. Show t h a t i n g e n e r a l , x(P) has length

I x(P) I
C o n c l u d e t h a t f is n o t rectifiable.

2 1 + 2(1/2

+ 113 +

... + l / h ).

($)

Let q be a fixed unit vector. A particle moves in Vn in such a way that its position ) vector ~ ( tsatisfies the equation r(t)-u = 5t 3 for all t, and its velocity vector makes a

constant angle 0 with

u, where 0 < 9 < 7r/2.

(a) Show that ((I[(15t2

/cos 8. =
(b) Compute the dot product . a ( t ) - ~ ( tin terms o f t and 0. )

(9

A particle moves in i-space so as to trace out a curve of constant curvature

K = 3.

Its speed at time t is e2t. Find Ila(t)ll, and find the angle between q and 3 at time t.

(6) Consiaer r = e-

the curve given in polar coordinates by the equaticn

for 0 5 Q 52UM

where N

is a positive integer.

Find the length of this curve. h3at happens as M becomes

a r b i t r a r i l y large?

( a ) Derive the following formula, vhich can be used t o compute the

cctrvature of a curve i n R":

(t) Find the c u n a t u r e of the curve ~ ( t= ) ( l c t , 3 t , Z t t 2 , 2 t 2 ) .

Derivatives - vector functions.

of Recall that if x is a point of a scalar function of x, R "

and if f

f ( 5 ) is (if it

then the derivative of

exists) is the vector

For some purposes, it will be convenient to denote the derivative

of

by a row matrix rather than by a vector.

Df

When we do this,
?f. Thus

we usually denote the derivative by

rather than

If we use this notation,.the definition of the derivative

takes the following form:

where

0 as

h - ->

0 -.

Here the dot denotes matrix

multiplication, so we must write order for the formula to work;

h -

as a column matrix in

his is the formula that will generalize to vector functions

Definition. Let S be a sybset of R ~ . If

k f f(5) is called a vector function - -a of - : S -> R , then vector variable. form

In scalar form, we can write

f(5) -

out in the

Said differently, n variables."

f -

consists of

"k real-valued functions of

f f is defined in an open ball is differentiable at
a -

Suppose now that a. We say that

about the point

if a -

each of the functions fl(x), ...,fk(~) is differentiable at

!

(in the sense already defined) derivative of

f -

Furthermore, we define the

at

a -

to be the matrix

That is, Df(=) Dfi (=)

row is the derivative

is the matrix whose 1 of the ith coordinate function of f.
Oth

a Said differently,- the'derivative 'Df(5)- o ' . - at - f. f is.the k by n matrix whose entry in row i and column
j

is

i t i s o f t e n c a l l e d t h e J a c o b i a n m a t r i x of

f -( 5 ) .

Another

notation f o r t h i s matrix is the notation

With t h i s n o t a t i o n , many o f t h e formulas w e proved f o r a s c a l a r function f -( 5 ) . i f and -f ( 5 ) h o l d w i t h o u t change f o r a v e c t o r f u n c t i o n

W e c o n s i d e r some of them h e r e :

Theorem 1. only i f

The - function

f is -(5) - d i f f e r e n t i a b l e

at -

a -

where

- g ( & ) ->
(Here

as
_ I

h 0 - -> .

and E a r e w r i t t e n a s column m a t r i c e s . ) Proof: Both s i d e s o f t h i s e q u a t i o n r e p r e s e n t column I f we consider the

I-

-, 5, f

matrices.

th

e n t r i e s of t h e s e matrices,

we have t h e f o l l o w i n g e q u a t i o n :

NOW

f
is.

is d i f f e r e n t i a b l e a t
And

i f and o n l y i f each f u n c t i o n

fi

fi as

is d i f f e r e n t i a b l e a t
h 0 - -> -.
But
Ei ( h ) ->

a
0

i f and o n l y i f as h -> 0 -. 0 -, for

Ei (&) ->
each

i,

i f and o n l y if

E h - (- )

->

0 -

as

h ->

Theorem 2.
is at - continuous -

- -( - ) - d i f f e r e n t i a b l e - -, If f x is at a
a. f is differentiable a t

then -

f -

Proof. function whence

fi.

If

a -,

then s o i s each a -,

Then i n p a r t i c u l a r ,

fi

i s continuous a t

f a - i s continuous a t -. The rule. - general chain Before c o n s i d e r i n g t h e g e n e r a l c h a i n r u l e , l e t us t a k e t h e

c h a i n r u l e w e have a l r e a d y proved and r e f o r m u l a t e it i n t e r m s o f matrices.

Assume t h a t

f (5) = f ( x l , . . a -,

., xn

is a s c a l a r f u n c t i o n

d e f i n e d i n an open b a l l about xn(t))

and t h a t

i s a parametrized curve p a s s i n g

x a - ( t O ) = -.
)

If

f (- ) x

is differentiable

x -( t ) = through -. a a t -, and a

(xl ( t ) ,
Let

...,
is

if

x -( t )

differentiable a t

t o , and we have shown t h a t t h e composite

f ( ~ ( t ) )i s d i f f e r e n t i a b l e a t by t h e e q u a t i o n

t o , and i t s d e r i v a t i v e i s g i v e n

when

t=tO.
W e can r e w r i t e t h i s formula i n s c a l a r form a s follows:

o r w e can r e w r i t e it i n t h e following m a t r i x form:

Recalling t h e d e f i n i t i o n of t h e Jacobian matrix

Df, -

we s e e

t h a t t h e l a t t e r formula c a n be w r i t t e n i n t h e form

(Note t h a t t h e m a t r i x
DX
d

Df

i s a row m a t r i x , w h i l e t h e m a t r i x

is by i t s d e f i n i t i o n a column m a t r i x . )

This i s t h e form of t h e c h a i n r u l e t h a t w e f i n d e s p e c i a l l y u s e f u l , f o r it i s t h e formula t h a t g e n e r a l i z e s t o h i g h e r dimensions. L e t us now c o n s i d e r a composite of v e c t o r f u n c t i o n s of vector variables. t h e following: Suppose taking values f is on an ba l - - d e f i n e d - - open -l-i n
Rk,

For t h e remainder of t h i s s e c t i o n , we assume

R"

about

a -,

&

with

- - ( 5 ) = -. f b

Suppose

p - defined is
Let -

i n an ball - - open - a b o u t h ,

t a k i n g v a l u e s - RP. in

F -( 5 ) =

9 (f

(x) )

d e n o t e t h e composite f u n c t i o n .

W s h a l l write these functions as e

f -(5) =
If f -

f ( x l ~ g e - r x n ) and

g -( x )

=~(ylr...,yk).

and

g -

are differentiable a t

a -

and

b -

r e s p e c t i v e l y , it i s e a s y t o s e e t h a t t h e p a r t i a l d e r i v a t i v e s of

~ -

( 5 ) xist a t e

a -,

and t o c a l c u l a t e them.

After a l l , the

t i-h

coordinate function of

-(- ) F x

i s g i v e n by t h e e q u a t i o n

~f we s e t each o f t h e v a r i a b l e s

xQ

except f o r t h e s i n g l e aQ, then both s i d e s a r e

variable

j*

equal t o the constant x

alone. j
g i v e s us t h e formula

f u n c t i o n s of

The c h a i n r u l e a l r e a d y proved t h e n

Thus

= [i-h t

row of

Dz]

-d i f f e r e n t i a b l e - -s on i t

domains, t h e n t h e composite

F -(5) =

q (f( 5 ))

is - continuously

domain, and

T h i s theorem i s a d e q u a t e f o r a l l t h e c h a i n - r u l e a p p l i c a t i o n s w e s h a l l make. Note: The m a t r i x form of t h e c h a i n r u l e i s n i c e and n e a t , and In practical situations,

it i s u s e f u l f o r t h e o r e t i c a l p u r p o s e s .

one u s u a l l y u s e s t h e s c a l a r formula ( * ) when one c a l c u l a t e s part i a l d e r i v a t i v e s o f a composite f u n c t i o n , however. The f o l l o w i n g proof i s i n c l u d e d s o l e l y f o r c o m p l e t e n e s s shall not need to use it:
Theorem 4. - e t f - g be a s above. - f -
L and If is
;

we

-1

- --

d i f f e r e n t i a b l e -t a

-a - and - g

i s d i f f e r e n t i a b l e - -, at b

then -

is d i f f e r e n t i a b l e a t

a a nd - -, -

DF(5) = Dg(b)

Df ( a ) .

--

Proof.

W e know t h a t

where

E l ( k ) ->

as

- -> k

0.

L e t us s e t

- = - (-+ h ) - -( g ) k f a f

i n t h i s formula,

Then

Thus t h e J a c o b i a n m a t r i x of

F -

s a t i s f i e s the matrix equation

his i s o u r g e n e r a l i z e d v e r s i o n of t h e c h a i n r u l e .
There i s , however, a problem h e r e .
if
W e have j u s t shown t h a t

f -

and

g -

a r e d i f f e r e n t i a b l e , then t h e p a r t i a l d e r i v a t i v e s F exist. But we know t h a t t h e mere Fi

of t h e composite f u n c t i o n

e x i s t e n c e of t h e p a r t i a l d e r i v a t i v e s o f t h e f u n c t i o n enough t o g u a r a n t e e t h a t
Fi

is not

is differentiable.
f and
C J

One needs t o

g i v e a s e p a r a t e proof t h a t i f b o t h t h e n s o i s t h e composite following. )

are differentiable,

g(x) = f(p(x)).

(See Theorem 4

, I
both

One c a n avoid g i v i n g a s e p a r a t e proof t h a t

F -

is

d i f f e r e n t i a b l e by assuming a s t r o n g e r h y p o t h e s i s , namely t h a t f and

g -

a r e continuously d i f f e r e n t i a b l e .
f -

In t h i s case,

t h e p a r t i a l s of

and

g -

a r e c o n t i n u o u s on t h e i r r e s p e c t i v e

domains; t h e n t h e formula

which w e have proved, shows t h a t f u n c t i o n of f o r each

D.F

3 i

i s a l s o a continuous Fi

x -.

Then by o u r b a s i c theorem, F -

is differentiable

if so t h a t

i s d i f f e r e n t i a b l e , by d e f i n i t i o n .

W e summarize t h e s e f a c t s a s f o l l o w s :

Theorem 3 . about a,

L -e t- f- be

d e f i n e d on an open b a l l i n 'R
Rk;

taking values i n

----let a - f (- ) = b. - - g- be - Let
taking values i n

d e f i n e d i n an open b a l l a b o u t a nd -

--

b,

RP.

If f - -

are c o n t i n u o u s l y

d i f f e r e n t i a b l e on t h e i r r e s p e c t i v e

- - (- )I1 . a
N w w e know t h a t o

f -- (a+h)
where

f (a) = Df (5) h

.- + E2(- )ll -ll , h h

Plugging t h i s i n t o ( * * ) , we

E2 (h)

->

as

- -> h

-. 0

get t h e e q u a t i o n

+ E l ( -( a +- ) f - h
Thus

f a - - (-) ) l l- (- +-h ) - - (a)ll . f a

F F -(g+h) - -( 5 )
where

Dz(b)

Df

(5) h + E3(h)llhll, -

W e must show t h a t

E3

->

as

h ->

easy, since

Dg

(b)

i s c o n s t a n t and

-. 0 <gZ- ) (h

The f i r s t term i s ->

as

h ->

0.

Furthermore, as approaches

h ->
(since

-, 0 f

t h e expression

-(a+h) - (a) f -f

i s c o n t i n u o u s ) , so t h a t

E1 (g+h) f ) - (f - - - (5) --> expression

0.

We need finally to show .that the

is bounded as

h -> -

0. -

Then we will be finished. h ,I &,I

Now

--f -l () Il.f(a+h) - - a l -11 2 1

- I1 Df (a) -C

E2 (2)1 1

I1 Df (a) -Now

u l l - + it g2 (&) I1 , 0 as

where

u -

is a unit vector.

E2(h) -> C

->

and it is easy to see that I1 Df (a) u l - l (Exercise!) Hence the expression bounded, and we are finished.

nk max 1 D.f . (a)1

l - (a+h) f --

1 3 f(a)
nl - - l/h U

0, -

is

Differentiating inverse functions.

Recall that if f(x) is a differentiable real-valued
f'(x) > 0

function of a single real variable x, and if for a g x g b, then f

is strictly increasing, so it has an is differentiable and its deriva-

inverse g. '~urthermore, g tive satisfies the formula

Part, but not all, of this theorem generalizes to vector functions. We shall show that if
f - -

has an inverse - and g,

if g - -

is differentiable, then there is a formula for Dg Specifically, we prove the following:

analogcus to this one. Theorem 5.

f -

Let

be -

subset

of

R. "

Suppose that

:A->R

. Suppose and that - (a) = b f - -- -

also - - - that f has. an

inverse g. ~f at a, and if g is - -f -is differentiable - - - - - - differentiable - b, -

at then

Proof.
g (5) - (f )
=

Because

g -

is inverse to f ,

the equation f - and q -(f ( 5 ) ) . g are Thus

5 holds for all - in S. x

Now both

differentiable and so is the composite function, we can use the chain rule to compute

Dq(b)

~f (a) = D (identity) = In.

Since the matrices involved are that

n by

n, this equation implies

Remark 1. This theorem shows that in order for the differentiable function

to have a differentiable inverse,

Df(a)

it is necessary that the Jacobian matrix

have rank

n. to

Roughly speaking, this condition is also sufficient for have an inverse.

More precisely, one has the following result, which

is the famous "Inverse Function Theorem" of Analysis:
Suppose f is defined and continuously differentiable
Rn about a -, f taking values in
R. "

in an open ball of Df (a) has rank ball

B

If

n,

then there is some (probably smaller) open such that in Rn. carries B in a 1-1 fashion

about a, B

onto an open set C g :

Furthermore, the inverse function

C ->

is continuously differentiable, and

Remark 2.

For a function of a single variable, y = f(x),

the rule for the derivative of the inverse function x = g(y) is often written in the form

This formula is easy to remember; the Leibnitz notation for

derivatives "does the work for you". It is tempting to think

that a similar result should hold for a function of several

variables. It does ---not.

For example, suppose

f -(x,y)

= (u,v)

is a differentiable transformation from the the u

plane to

plane.

And suppose it has an inverse; given by

Our theorem tells us that if

f -(=)

b,

then

~q(2) [D~(~)I-'*
=
If we write out these matrices in Leibnitz notation, we obtain
the equation

Now the formula for the inverse of a matrix gives (in the case

of a 2 by 2 matrix) the formula

. npplying this formula, we obtain the equation

This means, for example, that

Thus the simplistic idea that reciprocal of

av/ax

ax/av

"should be" the

The Leibnitz

is very far from the truth.

notation simply doesn't "do the work for you" in dimensions

greater than 1. Matrix notation does..

Implicit differentiation. Suppose g is a function from R~~~ to

R: "

let us

write it in the form

Let - be a point of c

R",

and consider the equation

This equation represents a system of unknowns.

equations in n

In general, we would expect to be able to solve this For

system for - of the unknowns in terms of the others. n solve this system for - in terms of y. x

instance, in the present case we would expect to be able to

We would also expect

the resulting function - = ~ ( y ) be differentiable. x to Assuming this expectation to be correct, one can then calculate the derivative of the resulting function 2 by using the chain rule, One understands best how this is done by workApostol works several in

ing through a number of examples. sections 9.6 and 9.7. a d 6 of 9.7. n

At. this point, you should read 9.6 a d Ekamples 1,2,3, n

A natural question to ask now is the following: to what extent

our assumptions

are correct, that the given equation determines x

as a function of y

WE'discuss that questeon now.

Fjrst let us consider the problem discussed on p 294 of the text. . It involves an equation of the form

where F is continuously differentiable. Assuming that one can in theory s solve this equation for z a ; a function of x and y, say z = f(x,y). Apostol derives equations for the partials of this unlmown function:

af - (x1y) ax
at the point (x,y,f(x,y))
j

- -- a x

3F and

aF az

-J-f - - 3 . L
3~

3~ -

aF az

Here the functions on the right side of these equations are evaluated

.
,
in order

Note that it was necessary to assume that )F/dz # 0

to carry out these calculations. It is a remarkable fact that the condition aF/?z # 0 is also sufficient to justify the assumptions we made in carrying them out. This is a consequence of a fsrnous theorem of Analysis called the Implicit Function Theorem. O i consequence of this theorem is the following: re If one has a point (xO,yO,zO) that satisfies the equation F(x,y,z) = 0

and if a F b z # 0 at this point, then there exists a unique differentiable function f(x.y), defined in an open set B about (xO,yO)

such that

f(xo,yo) = zO and such that F!x,ytf (xty) = 0 for all (x,y) in B. Of course, once one hows that f exists and is differentiable, one can find its partials by implicit differentiation,

as explained in the text.

<

E5(.am~le . Let F(x, y,z) = x2 + y2 + z2 + 1 1

The equation

F(x, y,z) = 0 cannot be solved for z in terms of, x and y; for in

fact there is no point that satisfies the equation.
Exam~Qe2. Let F(xfyfz)= x2 + y2 + z2 - 4 The equation . F(x,y,z) = 0 is satisfied by the point a = (0,2,0). But 3F/hz = 0 at the point a, so the implicit function theorem does not apply. This fact is hardly surprising, since it is clear from the picture that z is not determined as a function of (x.y)
(X~IY~'

in an open set about the point.

= (OI2).

Hc;wever, the point b = (l,l,6) satisfies the equation also, and )F/~Z # 0
Q : ( o , ~ , o ) at

this pint.

The implicit function theorem

3
Y

implies that there is a function f(x,y) defined in a neighborhood of (xO,yO) = (1,l) such that f(l,l) ~= and f satisfies the

equation F(x,y, z) = 0 identically. h-ote that f is not uniquely determined unless we specify its value at (xO.yO). There are two functions f defined in a neighborhood of (1,l)

that satisfy the equation f(x,y,z) = 0, namely,

z

[ 4 - x2 - y 2 ] 4

and

z =

[4 - x2 - y2]5

However, only one of them satisfies the condition f( 1,1) =

fi .

Note that at the point a = (0,2,0) we do have the condition

. JF/Jy # 0 Then the implicit function theorem implies that y is

determined as a function of (x,z) near this point. The picture makes this
,

fact clear.

',
I

Nc~wlet us consider the more general situation discussed on p. 296 of the text. We have two equations

where F acd G are continuously differentiable. (We have inserted an extra variable to make things more interesting.) Assuming there are rd functions x = X(z,w) ar y = Y(z,w) that satisfy these equations for

all points in an open set in the (zfw)plane, we have the identities F(X,Y,z,w) = 0 and G(X,Y,Z,W) =

o ,

whence (differentiating with respect to z , )

These are linear equations for JX/Jz and a~/az we can solve them if the ; coefficient matrix

is non-singular. One can use Cramerlsrule, as in the text, or one

can write the solution in the form

The functions on the right side of this equation are evaluated at the point
, I

(~(z,w),~(z,w),z,w), so that both sides of the equation are functions of z and w alone.

You can check that one obtains an equation for the other partials

of X and Y if one replaces z by w throughout.

All this is discussed in the text. But now let us note that in order to carry out these calculations, it was necessary to assume that the matrix

& F ,G/$ x,y

was non-singular

Again

, it is a

remarkable fact

that this condition is also sufficient to justify the assumptions we have made. Specifically, the Implicit Function Theorem tells us that if ( X ~ , ~ ~ , Z ~isW ~point satisfying the equations ( * ) , and if the matrix , a )

a F,G/ d x,y

is non-singular at this point, then there do exist unique and Y(z,w) defined in an open set about

differentiable functions X(Z,W) (zOIwO), such that

X(zo.wo) = xo and such that

I

and

YizoIwo) = yo

F and G vanish identically when X and Y are substituted

for x and y. Thus under this assumption all our calculations are justified.

r
I

E3mple 3,

Consider the equations

The points

( x ~ , ~ ~ = (1.2t-lrO)~ )a ; (xl Y ~ , ~ ~ I ~ ~ ) ~ z ~ ~ w rd I = (1,5121-2)

satisfy these equations, as you can check. WE calculate

At the point

( x ~ , ~ ~ , z ~ , this matrix equals w~),

[ y],

whichis

non-singular. Therefore, there exist unique functions x = X(z,w) and y = Y(Z ,w)

defined in a neighborhood of (zo,w0) = (-1.0) that

satisfy these equations identically, such that X(-1,O) = 1 and Y(-1,O) = 2 . SSnce we know the values of X and Y at the point (-l,O), we
Indeed,

can find the values of their partial derivatives at this point also.

Om

the other hand, at the point equals

= (xl,yl,zl,wl) (1,f,2,-2) the

matrix hF,G/bx,y

which is singular. Therefore we do not expect to be able to solve for x and y in terms of z d we have
w nc.ar this point. However, at this point,

<

I1

Therefore, the implicit function theorem implies that we and w in terms of y and z near this point.

can

solve for x

Exercises

1. Given the continuously differentiable scalar field f(x, ) , y 2 3 9 let + t = f(t ,t + 1. Find ( 1 , given that Vf(1,2) = 5i - J () )

*.

2 Find the point on the surface z = xy nearest the point (2,2,0). .

3. A rectangular box, open at the top, is to hold

256 cubic inches.

Find the dimensions that minimize surface area.

4. Find parametric equations for the tangent line to the curve of

intersection of the surfaces

at the point
5.

(2,1,2).

Let

be a s c a l a r function of

variables.

Define

Express
f

F8(1)

i n terms o f t h e f i r s t o r d e r p a r t i a l s o f

(3,3,2).
i n terms o f t h e f i r s t a n d s e c o n d o r d e r
a t the point (3,3,2).

a t the point F"(1)

Express

p a r t i a l s of
6. Let
f : R

f
---r

R'

and let

g : R~

Suppose t h a t

f(0,O)

= (1,2)

-( 1 , 2 ) f

= (0'0).

g(O,O) = ( 1 , 3 , 2 )

d l J ) = (-1,Opl)-

Suppose t h a t

a)

If

h ( 4 ) =g(f(x)), f i n d

b)

If

has an i n v e r s e

Dh(0,O). 2 2 : R R ,

find

Dk(0,O).

7 Ccnsider the functions of Example 3. .

Find the partials

JX/&

md

3 ~ / & at

the point

( Z ~ , W ~ ) (-1 =

, 0).
G = 0,

8. Fcr the functions F

and G of Example 3, compute ~ ( F , G ) / . ~ ( x , ~ )

fcr which

at the point

1 , 2 - 2 )

Given the equations F = 0 ,

pairs of variables is it possible to solve in terms of the other two near this point?

The s e c o n d - d e r i v a t i v e

t e s t f o r extrema of a f u n c t i o n of

two v a r i a b l e s . Theorem. Suppose t h a t

--a.

f(xl,x2)
B

has - continuous about a.

second-

order p a r t i a l derivatives i n a b a l l Dlf

Suppose t h a t

a nd

D2f

vanish a t

Let -

A =

D1llf(a)l

B =

DlI2f(a),
f h s -a0,

C = D212f(=).

(a)

If
I f
If -

B2

- AC >
AC

0,

(b)

B~

< 0

then and - A

a saddle point a t then f

a.

>

h s -a- a r e l a t i v e

(c)

B2

AC

<

a nd -

<

0,

then -

h s -a-

a relative

maximum - a . at

(d)

I f

BL

- AC

= OI

the t e s t ---

i s inconclusive.

Proof.

S t e p 1.

W e f i r s t prove a v e r s i o n of T a y l o r ' s theorem

w i t h remainder f o r f u n c t i o n s of two v a r i a b l e s : Suppose ball v B

f (xl, x2) a -.

has c o n t i n u o u s second-order p a r t i a l s i n a Let v be a f i x e d v e c t o r ; say

centered a t Then

= (hlk).

where

* a -

i s some p o i n t on t h e l i n e segment from

g ( t ) = f (=+tv) ,

a -

to

a - + tv. -

W e d e r i v e t h i s formula from t h e s i n g l e - v a r i a b l e form of T a y l o r ' s theorem. Let

i.e .

Let

F ( t -) v

denote t h e l e f t s i d e o f t h i s e q u a t i o n .
F(tv)

We

w i l l be concerned about t h e s i g n of

when f

is small,

because t h a t s i g n w i l l depend on whether

mum o r a l o c a l minimum a t

h a s a l o c a l maxi-

a -,

o r neither. v = (h,k)

Step 3 ,

From now on, l e t

be a u n i t v e c t o r .

Consider t h e q u a d r a t i c f u n c t i o n

W e s h a l l determine what v a l u e s

takes a s

v -

v a r i e s over t h e

unit circle. Case 1. t h e same s i g n a s Proof. If

A,

B~

AC

<

0,

t h e n w e show t h a t

(v)

has

f o r a l l u n i t vectors

v.
thus

When

-= v
A

(1,0),

then

Q(v) = A ;

Q(v)

h a s t h e same s i g n a s function
[ O ,21~1 ,

i n t h i s case.
t

~ o n i i d e r h e continuous t

Q(cos t, s i n t ) . A s the vector

ranges o v e r t h e i n t e r v a l ranges o v e r a l l u n i t v e c t o r s

(cos t , s i n t )

in

V2.

I f t h i s f u n c t i o n t a k e s on a v a l u e whose s i g n i s d i f f e r
A,

e n t from t h a t o f t h e r e must be a

t h e n by t h e i n t e r m e d i a t e - v a l u e theorem, Q ( c o s to, s i n to) = 0. That i s ,

to such t h a t

f o r some u n i t v e c t o r t h e number ko/ho

(ho,ko).

Now i f

ho # 0,

t h i s means t h a t

i s a real r o o t o f t h e e q u a t i o n

g ( t ) = f(al + th, a2 + t k ) .

W know t h a t e between
0

g ( t ) = g(0) and
t.

2 g l ( 0 ) - t + g W ( c ) */2! t

where
g

is

C a l c u l a t i n g the d e r i v a t i v e s o f

gives

from which formula between

0

(*)

follows.

Here

a -

= a

- + cv, -

where

is

and

t.

S t e p 2. vanish a t

I n t h e p r e s e n t c a s e , +&e f i r s t p a r t i a l s o f so t h a t

The o n l y r e a s o n this

approximation r a t h e r t h a n

equality

i s t h a t t h e second p a r t i a l s a r e e v a l u a t e d a t the unknown p o i n t

a -

i n s t e a d of a t

a -,

T h i s matter w i l l be d i s p o s e d o f by u s i n g Formally, w e have t h e e q u a t i o n

elementary e p s i l o n i c s .

+ E 1l D .

f ( a * )-A] h2

2 [Dl. Zf (=*) -B] hk

+ [

D ~ f ( a * ) -c] k 2 . 2

Note t h a t t h e l a s t three terms a r e s m a l l i f

i

a -

is close to

-, a

because t h e second p a r t i a l s are continuous.

~ u thi,sequation has a real root only if Similarly, if equation

ko

B~

AC

>

0.

0,

the number ho/ko

is a real root of the

again we conclude that B~ led to a contradiction. Case 2. If B~

AC

>

0.

Thus in either case we are

AC

> 0, then we show that Q (v) takes

has

on both positive and negative values. Proof. If A # 0, the equation Ax2 + Bx + C = 0 Thus the equation. y = Ax
2

two distinct real roots.

2Bx

represents a parabola that crosses the

x-axis at two distinct

points.
B2

On the other hand, if A = 0, then B # 0 (since 2 in this case the equation y = Ax + 2Bx + C AC > 0 ) ; It follows that in

represents a line with non-zero slope. either case, there is a number xo

for which

and a number xl

for which

Let

(ho,ko) be a unit vector with

ho/ko

xo

and let

(hl,kl)

be a unit vector with hl/kl = xl.

Then Q(ho,k,)

< 0 and

Step 4.
i

W e prove p a r t

( a ) of t h e theorem.

Assume
Q(&)

B~

AC

> 0.

Let

% be a u n i t v e c t o r f o r which
(**),

>

0.

Examining formula

w e see t h a t t h e expression
2 [ f(a+tv)

- -

f (a)I /tL approaches

Q(vo) a s
L

t ->

0.

Let

x a v - = - + t - and l e t t -> 0. Then - approaches - a l o n g x a

the

s t r a i g h t l i n e from
f

to

- + G, a x

and t h e e x p r e s s i o n O the n then t h e

(5)

f (2) approaches zero through p o s i t i v e values.

o t h e r hand, i f

vl
to

i s a p o i n t a t which

Q(vl) <

0,

same argument shows t h a t a s

approaches

along t h e s t r a i g h t

l i n e from
0 -

5 +

vl,
f

the e x p r e s s i o n

f (5)

f (5) approaches

through n e g a t i v e v a l u e s .
W e conclude t h a t

has a saddle point a t and (c) o f

W e know t h a t

-. a

S t e p 5.

W e prove p a r t s (b)

theorem.

(Q(x)

Examining e q u a t i o n ( * * ) once again. f o r a l l unit vectors

m,

1 >

v.

Then

IQ

(1) h a s a p o s i t i v e minimum I
(Apply t h e extreme[Q(COS

as

v
G

ranges o v e r a l l u n i t v e c t o r s .

v a l u e theorem t o t h e continuous f u n c t i o n
0

t , s i n t ) 1,

for

<t

2r.)

N w choose o

s m a l l enough t h a t each of the t h r e e

(**)

square-bracketed e x p r e s s i o n s on t h e r i g h t s i d e of than m/3 whenever

i s less

i s within

6
If

of

-. a

Here w e use continu

i t y of the second-order p a r t i a l s , t h e l i n e from

0 < t

< 6,
A

then

-* a
0

i s on

to
(*)

- + 6v; a -

since

is a u n i t vector, then
whenever whenever

the r i g h t s i d e . 0 5
)

h a s t h e same s i g n a s

< t < 6.

If

> 0,

t h i s means t h a t

f (x)

- - f (a) > 0 -

0
I

< Ix-a 1 < -f (51

Ear

then

6,

so

f
0

h a s a r e l a t i v e minimum a t

a.
so

If

<

0,

f (a)

<

whenever

< Ix-a1 <

--

6,

f h a s a rela

t i v e maximum a t

examples illustrating ( d ) , see the exercises. a

a -.

Exercises
1.

Show t h a t t h e f u n c t i o n

a t the o r i g i n , w h i l e t h e f u n c t i o n

x4 x4

y4
y4

has a r e l a t i v e minimum has a saddle p o i n t

there.

Conclude t h a t t h e s e c o n d - d e r i v a t i v e t e s t i s i n c o n c l u s i v e

2.

U s e T a y l o r ' s theorem t o d e r i v e the second d e r i v a t i v e

t e s t f o r maxima and minima o f a f u n c t i o n

f (x)

of a s i n g l e variwhat c a n you s a y f
at

able.

If

f 1( a ) = f" ( a ) = 0

and

f"' ( a ) # 0,

a b o u t t h e e x i s t e n c e o f a r e l a t i v e maximum o r minimum a t
3.
!

a?

Suppose near

f (x) h a s c o n t i n u o u s d e r i v a t i v e s o f o r d e r s Suppose

,...,n+l

x = a.

and

f ("+l) ) # 0 . (a

What c a n you s a y a b o u t t h e e x i s t e n c e o f a r e l a f a t a? Prove your answer c o r r e c t .

t i v e maximum o r minimum. o f
4.

(a)

Suppose

f (xl, x2)

h a s continuous third-order
(*)

p a r t i a l s near

-. a

Derive a t h i r d - o r d e r v e r s i o n o f formula

of

the p r e c e d i n g theorem.

(b)

Derive t h e g e n e r a l v e r s i o n o f T a y l o r ' s theorem f o r

f u n c t i o n s o f two v a r i a b l e s . [The f o l l o w i n g " o p e r a t o r n o t a t i o n " i s c o n v e n i e n t (hDl+kD2) f x=a = h

f
.

I- ~ (-~ + fk ~ (a), f a) ~ 2 (hDl+kD2) 1 - - = h 2DIDlf (a) + 2hkD1D2f x=a

1

(5) + h 2 DZD2f (5),

and s i m i l a r l y f ~ r(hDl+kD2)
I

The extreme-value theorem and the small-span theorem.

The proofs of the extreme-value theorem and small-span
theorem for rectangles given in Apostol at~sufficientlycon-
densed to cause some students difficulty. We shall prove the theorems only for ,R ' through without difficulty in A rectangle closed intervals
Q

Here are the details.

but the proofs go

R ~ .

in

R~
and

is the Cartesian product of two

[c,dl;

[a,b]

Q
!

= [a,b]

[c,d] = {(x,y)

and

c 5 y

d. )

The intervals intervals of

[a,b]

and

[c,d]

are called the comuonent

Q.

If P1 =
{X~'X~'*

.'xn]

is a partition of

[a,b],

and if

P2 = { Y ~ J Y ~ S * . . , Y ~ ~ is a partition of [c,dJS then the cartesian product

Q.

P1

P2

is said to be a partition of into n subintervals and P2

Since

P1

partitions [a,b] [c,d] into Q m into

partitions

subintervals, the partition mn

P = P 1 x P2 subrectangles, namely the rectangles

partitions

Theorem ( s m a l l - s ~ a n theorem). is - continuous on the rectangle in - R2 . t hat - f

Let

be 2 scalar function that

Q = [a,bl
Then, given
6

[c,dl

>
f

0,

there

is

q partition

of

such

is

bounded span

on
of

every subrectanale in every

such that the --is less --- than Proof.

of the partition and subrectangle of the partition

6 0 For purposes of this proof, let us use the If

Qo

following terminology:
Q,

is any rectangle contained in

Po

let us say that a partition o f

is

tt6-nice" if

is

bounded on every subrectangle

.

o f the partition and if the

span of

in every subrectangle o f the partition is less than

r.

We recall that the span of

in the set

is defined b y

the equation

Recall also that i f

S1

is a s u b s e t o f
f spanS f .

S,

then

span

1
To b e g i n , w e n o t e t h e f o l l o w i n g e l e m e n t a r y f a c t :
Qo

Suppose

[ao,boI x [co,d0I
Q.

is any r e c t a n g l e contained i n
ponent i n t e r v a l
I1 = [ a o , p J

L e t u s b i s e c t t h e f i r s t com

[ao,bo]

of

Po

i n t o two s u b i n t e r v a l s where p is t h e midpoint o f i n t o two s u b i n t e r -

and

I2 = [ p , b o ] ,

[ao,bo]. vals
I

Similarly, let us bisect and

J2.

[co,d0]

J1

Then

Qo

is w r i t t e n a s t h e union o f t h e

four rectangles

I1

J1

and

I2

J1

and

I1

and

I2 x J2.
eo-

Now i f e a c h o f t h e s e r e c t a n g l e s h a s a p a r t i t i o n t h a t i s nice,

then w e can put t h e s e p a r t i t i o n s together t o g e t a p a r t i

Qo

t i o n of

t h a t is

ro-nice.

The f i g u r e i n d i c a t e s t h e p r o o f ;

e a c h of t h e s u b r e c t a n g l e s o f t h e n e w p a r t i t i o n i s c o n t a i n e d i n a subrectangle o f one o f t h e o l d p a r t i t i o n s .

Now we prove the theorem. false and derive a contradiction. some nice. eo

We suppose the theorem is That is, we assume that for has no partition that is ao

> 0, the rectangle

Let us bisect each of the component intervals of writing Q as the union of four rectangles.

Q,

Not all of these

so-nice, for if
Let Q1 be one

smaller rectangles have partitions that are they did, then Q

would have such a partition. Q1

of these smaller rectangles, chosen so that

i

does not have a

partition that is

nice.

Bisect each component inter-

At least one of these
eo-nice; let Q2

Now we repeat the process. val of Q1

into four smaller rectangles.

smaller rectangles has no partition that is denote one such.

Continuing similarly, we obtain a sequence of rectangles

none of which have partitions that are

ro-nice.' Consider the

left-hand end points of the first component interval of each of

these rectangles. Let s be their least upper bound. Similar-

ly, consider the left-hand end points of the second component

interval of each of these rectangles, and let upper bound. Then the point (s,t) t be their least

belongs to all of the rec-

tangles

Q,Q1,Q2,.

.. .

Now we use the fact that (s,t).

is continuous at the point r centered at

ro.

We choose a ball of radius

f

(s,t)

such

that the span of rectangles

in this ball is less than

Because the

Qm become arbitrarily small as m increases, and because they all contain the point ( s , t ) , we can choose m large enough that Qm lies within this ball. Since (s,t), Qm is contained in
f in

Now we have a contradiction. the ball of radius

Q m
is less than tion of Qm that is
1

centered at

the span of

But this implies that there

is a parti-

iO-nice, namely the trivial partition!

O

Thus we have reached a contradiction. Corollary. ous on the ---

Let
i

f be a scalar function that

Then . f

continu

rectangle Set

9.

&

bounded

9.

Proof. is a-nice.

= 1,

and choose a partition of

that

This partition divides Rl,

Q
f

into a certain number of is bounded on each of

subrectangles, say

...,R mn'
>Now
for

these subrectangles, by hypothesis; say

I ( ) S Mi fil
1~ e Ri.

Then if

M = max{M1,

...,Mmn }

we have

If(x)l
for all
y E Q.
I3

I M

Theorem (extreme-value theorem).

)

Let

f
Q.

be a scalar
Then there

function that points xo

continuous

J O

the

rectangle

are

and xl - Q such that of

f(xo) I
for all - -
x
E

f(x) -

f(x -1

Q.

Proof.

We know

is bounded on

Q;

let

M = sup{f(x)
We wish to show there is a point f(xl) = M. Suppose there is no

I x e Q). x1 of Q

such that

such a point.

Then the function

- f(x)

is continuous and positive on

Q,

so that the func-

tion

is also continuous and positive on lary that

g

Q.

By the preceding corol-

is bounded on

Q;
e

let 1

be a positive constant such

g(y)

<

for

Q.

Then

for all in
Q.

f(x)

f(x) I Then M f(5) for

- <C, M - (l/C)
(1/C) in
Q,

or

is an upper bound for the contradicting the fact

set of values of that

M

is the least upper bound for this set.

A similar argument proves the existence of a point

of
Q

X o

such that f(so) = inf {f (s)I x e Q).

Exercises on line inteqrals

1.

Find the centroid of a homogeneous wire in shape of the

parabolic arc

y = x

for

-1

x 5

p.

'

[ U s e a table of integrals if you wish.]

2.

Let

on the set
(a)

consisting of all D2f1 = DlfZ

(x,y)

+
S.

0.

Show that

on

(b)
(-s,~)

Compute the line integral C

d from a
x2

(a,O)

to

when

is the upper half of the circle

+ y2 = a2

Compute it when

is the lower half of the same circle.

3.

Let

be as in problem 2.
x

Let

be the set of all

f

(x,y)

with

>

0.
i--- :

Find a potential function for

that is

defined in

U.

fi
+(x,g~=
J - F ~ --. i * ~ n r ~ i t t e A
-T

CLccrYc

4.

Let f be a continuous vector field defined in the open, connected

subset that
I

s of

R".
*2

Suppose that f =

md

=.a% in

s.

%W O

is a constant function. [~int: Apply Thoerem 10.3. ]

Notes - double integrals.

on (Read 11.1-11.5 of Apostol.)
Just as for the case of a single integral, we have the
following condition for the existence of a double integral:
Theorem 1 (Riemann condition). Q = [arb] x [c,d] given any s
E

Then there

>

0,

- defined is on f - integrable - Q - - only is on if and if -- functions s - t with are step and

Suppose f

< f < t - Q, such that on

Let If - A be-a number. - these the - further condition that

step functions - s - t and

satisfy

then -

A =

11

f .

The proof is almost identical with the corresponding proof

for the single integral.
Using this condition, one can readily prove the three
basic properties--linearityr additivity, and comparison--for the
integral

f.

We state them as follows:

(a) Suppose f

Theorem 2. Then - -so is

and

- integrable are on

Q.

cf (x) + dg (x); furthermore,

(b)
Q2.

Then

L e t Q be s u b d i v i d e d -- r e c t a n g l e s i n t o two - i n t e g r a b l e o v e r Q - - -n l y - is i f and o - i f i t
and Q2;

Q1

and -

is

i n t e g r a b l e -- Q1 over both

furthermore,

(c)
over

If -

on -

Q,

and i - -f

and -

are -integrable

Q,

then

TO prove t h i s theorem, one f i r s t v e r i f i e s t h e s e r e s u l t s

f o r s t e p f u n c t i o n s (see 1 1 . 3 ) , and t h e n u s e s t h e Riemann condit i o n t o prove them f o r g e n e r a l i n t e g r a b l e f u n c t i o n s . The p r o o f s

a r e very s i m i l a r t o those given f o r t h e s i n g l e i n t e g r a l .

W e g i v e one of t h e p r o o f s a s a n i l l u s t r a t i o n .

For

example, c o n s i d e r t h e formula

where

and

are i n t e g r a b l e .

s2, tl, t2 such t h a t

s1 C f C tl

on

Q,

and such t h a t

lJQ If

JJ,

I\
J

Q g,
sl,

W e choose s t e p f u n c t i o n s

and

S2 C

< t2

We then find a single partition of sl, s2, tl, t2

relative to whi.ch all of sl + s2 and tl + t2

are step functions; then

are also step functions relative to this partition. one adds the earlier inequalities to obtain

Furthermore,

Finally, we compute

this computation uses the fact that linearity has already been proved for step functions. ~hus

JJQ (f + g)

exists.

TO

calculate this integral, we note that

by definition.

'

Then

here again we use the linearity of the double integral for step
functions. It follows from the second half of the Riemann

conditi.on that

(f + g) must equal ; e number h

up to this point, the development of the double integral

has been remarkably similar to the development of the single
integral. Now things begin to change. We have the following

basic questions to answer:

(1) Under what conditions does
(2) r f
(3)

11

"Q

exist?

f exists, how can one evaluate.it?

Is there a version of the substitution rule for double

integrals?
(4) What are the applications of the double integral? We shall deal with questions (l), ( Z ) , and (4) now, postponing question (3) until the next unit.

Let us tackle question (2) first.

How can one evaluate

the integral if one knows it exists? The answer is that such

integrals can almost always be evaluated by repeated one-dimen-
sional integration. More precisely, one has the following theorem:

Theorem 3 (Fubini theorem). on - -a rectangle Q

integrable on Q.

- f - defined - bounded Let be and = [ a r b ] x [c,d] , - assume - f and that is For each fixed y - [c,d], assume -in that the --

one-dimensional integral -

exists.
1

Then the integral --

A ( y ) dy

exists, ,and furthermore,

proof.

We need to show that

the double integral

Choose step functions s (x,y) and

Q,

,b

X(y)dy

exists and equals

f *

t(x,y) , defined on and

such that

s(x,y) C

(x,y)

t(x,y),

his w e can do because

exists.

For convenience, choose (This

and

so they are constant on the ?artition lines.

does not affect their double integrals.) sional integral

Then the one-dimen-

exists. [For, given fixed y

in

[c,d],

the function s(x,y)

is either constant (if y

function of function
c - y - d. < <

is a partition point) or a step

hence it is integrable.] Now I claim that the b S (y) = s(x,y)dx is a step function on the interval

x;

For there are partitions xo,

...,xm
and

yo,.-.,yn is

of

[arb] and

[c,d], respectively, such that

s(x,y)

constant on each open rectangle and

( x ~ - ~ , x ~ ( ~ ~ - ~ , y Let . y ~) x )

be any two points of the interval

holds for all

Then
is

s(x,y) = s(x.7) in

--

x.

(This is immediate if x

x i , x i ; if x

is a partition point, it follows fram

the fact that

s is constant on the partition lines.)

here

fore

Hence tion.

S (y)

is constant on

so it is a step func-

A similar argument shows that the function

is a step function for

< < c - y - d.

- . .

Now since

s 5 f

for all

(xfy), we have

.
by the comparison theorem. hypothesis.) (T3emiddle integral exists by y That is, for all

in

[c,d] ,

Thus

and

are step functions lying beneath and above A,

respectively,

Furthermore

(see p. 3561, so that

It f o l l o w s t h a t

A(y)dy

exists, by t h e Riemann c o n d i t i o n .
i s i n t e g r a b l e , w e can conclude

Now t h a t w e know

A(y)

from a n e a r l i e r i n e q u a l i t y t h a t

that is,

But i t is a l s o t r u e that

by d e f i n i t i o n .
E

Since t h e i n t e g r a l s of
A ( Y )dy

s
and

and

t
f

a r e less t h a n are within

a p a r t , we conclude L l a t Because
E

of e a c h o t \ e r .

i s a r b i t r a r y , t h e y must be e q u a l .

w i t h t h i s theorem a t hand, one can proceed t o c a l c u l a t e some specific double i n t e g r a l s . i n 1 1 . 7 and 11.8 o f Apostol.
NOW

S e v e r a l examples a r e worked o u t

l e t us t u r n t o t h e f i r s t of o u r b a s i c q u e s t i o n s , t h e
W e readily

one concerning t h e e x i s t e n c e of t h e double i n t e g r a l . prove t h e f o l l o w i n g :

i

Theorem 4 .

The - integral
Q.

exists i f

is -

continuous on t h e r e c t a n g l e

-E',

~ r o o f . A l l one needs i s t h e small-span theorem of Given of

If

p C. 27
E'.

choose a p a r t i t i o n of

such t h a t t h e span

on each s u b r e c t a n g l e of t h e p a r t i t i o n i s less than

is a subrectangle, l e t

Qij

'i j = min f ( x )

on

Qij:

t i j = max f ( x )

on

Qij.

Then

tij

si j

<
s

E'.

U s e t h e numbers

sij

and on
Q.

tij

t o obtain

s t e p functions

and

with

<

f G t

One t h e n has

JJ*( t - s ) < ~ ' ( d c ) ( b - a ) .

I

This number e q u a l s
E

i f w e begin t h e proof by s e t t i n g

E/

(d-c) (b-a)

We

I n p r a c t i c e , t h i s e x i s t e n c e theorem i s n o t n e a r l y s t r o n g enough f o r o u r purposes, e i t h e r t h e o r e t i c a l o r p r a c t i c a l .

s h a l l d e r i v e a theorem t h a t i s much s t r o n g e r and more u s e f u l .

F i r s t , we need some d e f i n i t i o n s :

Definition. d e f i n e t h e area o f

If
Q

Q =

[a,b] x [c,d]

is a rectangle, we

by t h e e q u a t i o n

a r e a Q.=

I\

1;

1 i s a s t e p f u n c t i o n , we can c a l c u l a t e
(d-c) (b-a)

Of c o u r s e , s i n c e
i

t h i s i n t e g r a l d i r e c t l y as t h e p r o d u c t

A d d i t i v i t y of two r e c t a n g l e s
Q1

i m p l i e s t h a t i f we subdivide and
Q2,

into

then

a r e a Q = a r e a Q1 + a r e a Q*.

Applying t h i s formula r e p e a t e d l y , w e s e e t h a t i f one h a s a p a r t i t i o n of

Q,

then

area Q =

li,j area

Qij,

where t h e summation e x t e n d s o v e r a l l s u b r e c t a n g l e s o f t h e p a r t i t i o n .
1t now f o l l o w s t h a t i f
A

and

a r e r e c t a n g l e s and

Q,

then

area A Let

a r e a Q.
D

Definition.

be a s u b s e t of t h e plane.
E

Then

is

s a i d t o have c o n t e n t - i f f o r every zero

s e t of r e c t a n g l e s whose union c o n t a i n s

>

0,

there i s a f i n i t e

and t h e sum of whose

a r e a s does n o t exceed Examples.

E.

(I)
(2)
(3)

f i n i t e s e t h a s c o n t e n t zero.

A h o r i z o n t a l l i n e segment h a s c o n t e n t zero.
A v e r t i c a l l i n e segment h a s c o n t e n t zero.

(4)

A s u b s e t o f a set o f c o n t e n t z e r o h a s c o n t e n t zero.

(5) A f i n i t e union o f sets o f c o n t e n t z e r o h a s c o n t e n t zero.

(6) The graph o f a c o n t i n u o u s f u n c t i o n

y = $(x);
i

< x <.b

h a s c o n t e n t zero.

(7)

The graph of a continuous function

x=$(y);
has c o n t e n t zero.

c G y 9 d

Most of t h e s e statements a r e t r i v i a l t o prove: only t h e

l a s t t w o r e q u i r e some c a r e .
L e t us prove ( 6 ) .

Let

E'

> 0.

Given t h e continuous function

l e t us use t h e small-

span theorem f o r functions of a s i n g l e v a r i a b l e t o choose a

partition span of
$ J

a = xo < x l <

... < xn. = b

of

[a ,b]
E'.

such t h a t t h e Consider t h e

on each s u b i n t e r v a l i s less than

rectangles

for

i = l,...,n.

They cover t h e graph of

Q,

because

( 1

Qx

1 <

whenever

x
Ai

i s i n the i n t e r v a l

,xi]

The t o t a l a r e a of t h e r e c t a n g l e s

equals

i=l

( x ~ - i-1) 2 ' = 2 c 1 ( b x

- a).

T h i s umber e q u a l s n
E

i f w e begin t h e proof by s e t t i n g

= e / 2 (b-a)

W now prove an elementary f a c t about sets of c o n t e n t zero: e Let be Let be of - Q - -a r e c t a n g l e . - D - -a s u b s e t Q -- c o n t e n t - Given E > 0 , t h a t has zero. t h e r e -s-a p a r t i t i o n i of such t h a t of t h that Q -- t h o s e s u b r e c t a n g l e s - -e p a r t i t i o n - c o n t a i n p o i n t s - D have t o t a l --- E. of area l e s s than Lemma 5. Note t h a t t h i s lemma does n o t s t a t e merely t h a t
D
is

c o n t a i n e d - t h e union o f f i n i t e l y many s u b r e c t a n g l e s of t h e parin t i t i o n having t o t a l a r e a l e s s t h a n

E,

b u t t h a t t h e sum of t h e
D

a r e a s o f - t h e s u b r e c t a n g l e s t h a t c o n t a i n p o i n t s of all
i

is l e s s
D

than

E.

The f o l l o w i n g f i g u r e i l l u s t r a t e s t h e d i s t i n c t i o n ;

i s c o n t a i n e d i n t h e union of two s u b r e c t a n g l e s , b u t t h e r e a r e

seven s u b r e c t a n g l e s t h a t c o n t a i n p o i n t s of

D.

Proof.
A1,...,An

F i r s t , choose f i n i t e l y many r e c t a n g l e s
c/2

of t o t a l a r e a l e s s t h a n

whose union c o n t a i n s
i,

D.

"Expandm each one s l i g h t l y .

,J

That i s , f o r each
Ai,

choose a

rectangle
A;

Ai

whoseinterior contains
Ai.

such t h a t t h e a r e a o f

i s no more t h a n t w i c e t h a t o f

Then the union o f t h e

sets

Int

A i

contains
Of

D,

and t h e r e c t a n g l e s

A ;

have t o t a l
may extend

area l e s s than outside

Q,

course, the rectangle

so l e t
A:

A:

denote t h e r e c t a n g l e t h a t i s t h e
Q.
E.

i n t e r s e c t i o n of

and

Then t h e r e c t a n g l e s

A:

also

have t o t a l a r e a l e s s t h a n

N w use t h e end p o i n t s of t h e component i n t e r v a l s of t h e o rectangles

AT

t o define a partition

of t h e r e c t a n g l e

Q.

See t h e f i g u r e .

W e show t h a t t h i s i s o u r d e s i r e d p a r t i t i o n .

Note t h a t by c o n s t r u c t i o n , t h e r e c t a n g l e
by
P,

Aj;

is parti,tioned

s o t h a t it i s a union of s u b r e c t a n g l e s Nw i f a subrectangle o
Qij

Qij

of

P. D,

c o n t a i n s a p o i n t of

then

it c o n t a i n s a p o i n t o f

Int

A1;

f o r some

kt

s o that it a c t u a l l y
B

lies in

and hence i n

Ai;.

Suppose w e l e t

d e n o t e t h e union
D; B C A.

of a l l t h e subrectangles
A

Qij

t h a t c o n t a i n p o i n t s of

and l e t

be the union of t h e r e c t a n g l e s

Then

It follows that

1
NOW

area Qij

Q~j~~

Q~j~~ area Qij*

on the other hand, by additivity of area for rectangles,

It follows that

This last inequality is in general strict, because some subso rectangles i j belong to more than one rectangle , ' their areas are counted more than once in the sum on the right
side of the inequality.
It follows that

as desired. 0
Now

we prove our basic theorem on existence of the double

on -

If is on - f - bounded - Q, and is continuous f exists.

except on a set of content - zero, then ---Theorem 6.

I/',

Proof. S t e ~ We prove a preliminary result:

1. Suppose that given e > 0, there exist functions g and h that are integrable over Q, such
that g(x) I f(x) l h(x) and for x in Q

Then f is integrable over Q. We prove this result as follows: Because h and g are integrable, we can find step functions sl, s2, tl, t2 such that s1 1g 5 tl and s2 5 h j t2, and such that

Consider the step functions sl and t2. We know that s1Iglf1h1t2 so sl is beneath f, and t2 is above f. Furthermore, because the integral of g is between the integrals of sl and of tl, we know that

Similarly, If we add these inequalities and the inequality

we have

Since e is arbitrary, the 'kiemann condition is satisfied, so f is integrable over Q.

S t e ~. Now we prove the theorem. Let D be a set of zero content containing the 2 discontinuities of f. Choose M so that If(x) 1 5 M for x in Q; then given r > 0, set r t =
r/2M. Choose a partition P of Q such that those subrectangles that contain points of D

have total area less than

6).

(Here we use the preceding lemma.)

Now we define functions g and h such that g subrectangles that does not contain a point of D, set

< f < h on Q. If Q.. is one of the 13

g(x) = f(x) = h(x) for x E Q. .. Do this for each such subrectangle. Then for any other x in Q, set
1J

g(x) = -M ThengSfShonQ.

and h(x) = M.

Now g is integrable over each subrectangle Q.. that does not contain a point of D,
1J

since it equals the continuous function f there. And g is integrable over each subrectangle Q.. that does contain a point of D, because it is a step function on such a
1J

subrectangle. (It is constant on the interior of Q. ..) The additivity property of the
U

integral now implies that g is integrable over Q. Similarly, h is integrable over Q. Using additivity, we compute the integral

JLQ

(h-g) =

1JJ
Qij

(h-g) = 2M

1(area Q~ that contain points of D)

Thus the conditions of Step 1hold, and f is integrable over Q. f Theorem 7. Sup~ose & bounded on Q, and f eauals 0 except on a set D of content zero. Then --

JJQf exists and

~ Q U zero ~ S

Proof. We apply Step 2 of the preceding proof to the function f. Choose M so that If(x)l 5 M for x in Q; given
E

> 0, set

E'

= E / ~ M Choose a .

partition P such that those subrectangles that contain points of D have total area less than E'. Define functions g and h as follows: If Q.. is one of the subrectangles that does not
1J

contain a point of D, set g(x) = f(x) = 0 and h(x) = f(x) = 0 on Q. .. Do this for each
1J

such subrectangle. For any other x in Q, set

g(x) = -M ThengSfshonQ.
I

and h(x) = M.

Now g and h are step functions on Q, because they are constant on the interior of each subrectangle Q. .. We compute
1J

JJQ
Similarly,

h =M

(1(area Q.. that contain points of D))

13

E, SO

that f is integrable over Q. Furthermore,

Since E is arbitrary, Corollary 8. set of -- content

a IJQf exists, and if g

g bounded function that eauds f except on a

Proof.

W write e
g

g = f
f

+ (g-f).

Nw o

is integrable

by h y p o t h e s i s , and

i s i n t e g r a b l e by t h e preceding

corollary.

Then

i s i n t e g r a b l e and

Double i n t e g r a l s extended o v e r more g e n e r a l r e g i o n s . (Read s e c t i o n 1 1 . 1 2 of Apostol.) ~ p o s t o l efines d set S, In t h i s section, f d e f i n e d on a bounded

[IS f

f o r a function

but then he quickly r e s t r i c t s himself t o t h e s p e c i a l S

c a s e where

i s a r e g i o n of Types I o r 11.

W discuss here e

the general case. F i r s t , w e prove t h e f o l l o w i n g b a s i c e x i s t e n c e theorem: Theorem 9. B S d of has - content

Let - S be-a zero, and i - - -f
$JS f

bounded s e t i n t h e p l a n e .

I n t S,

P roof. - Let

then -

is a t each -1 c o n t i n u o u s -

S. As

JmZnlU -,-.Ll MEbsYl

point

If

exists.

be a rectangle c o n t d h i n g

usual.
S.
S

let
N

equal

on

S,

and l e t
xo

equal

outside

Then
(because

i s continuous a t each p o i n t
f

of t h e i n t e r i o r of
xo,

it equals

i n an open b a l l about

and

is continuous xl

at

x0).

The f u n c t i o n
S,

i s a l s o continuous a t each p o i n t

of t h e e x t e r i o r of about xl.

because it equals z e r o on an open b a l l

The only p o i n t s where

S,

can' fail t o b e continuous

and this set, by assumption, exists.

are p o i n t s of ths boundary of

has c o n t e n t zero.
1

Hence

! f

E l

Note: value of

Adjoining or deleting boundary points of f

changes the value of

only on a set of content zerol so that

/Js f remains unchanged. Thus

)'IS

f =

/Irnt
1

for instance.
Let us remark on a more general existence theorem than
that stated in Theorem 9. if ~d
S

If

is a bounded set, and -

has content zero, and if f is continuous on Int S

* except on a set D of content zero, then f exists. For
in this case the discontinuities of the extended function f
lie in the union of the sets content zero because both Bd Bd S
S

and D

D,

and this set has

and

do.

There are more general existence theorems even than this,

but we shall not consider them.

NOW we note that the basic properties of the double

integral hold also for this extended integral: Theorem 10. Let S

be-a bounded set in the plane. -

---

One

has the - - following properties: (a) Linearity,

the left side e x i s t s if the r i g h t s i d e does.

(b) Comparison.

If

f < g

on t h e set

St

then

provided both integrals e x i s t .

(c) Additivity.

Let

S = SL

S2.

If

S1 n S2

has content

zero, then

provided the right side exists.

Proof.
ft

(a) Given

f, g

defined on
0 0

S,

let

2, g equal
Then Let
Q

gr

respectively, on
dg

S and equal S

otherwise. otherwise.

cl

equals cf

dg on

and

be a rectangle containing S.

We know that

from this linearity follows.

(b) Similarly, if
conclude that

f -
< g, then

from which we

(c) Let

be a rectangle containing
and equal
0

S.

Let fl
f2 equal f on f

equal on
S2,

on S1,

elsewhere. Let
f3 equal

and equal 0 elsewhere. Let

Sf

and equal 0

elsewhere. Consider the function

it equals

on the set S1 n S2, has content zero,

and equals zero elsewhere.

Because zero.

Sl n S2

l/

f4

exists and equals

Now

linearity implies that

How

can one evaluate

iIS

f when
S

is a general

region? The computation is easy when

I or I1 and

is a region of type
S:

is continuous on the interior of

one

evaluates

proved on p ,

1 367 of Apostol. integration. 4 f by iterated

This result is

using additivity, one can also evaluate

llS f

for

many other regions as well.

continuous function f

For example, to integrate a

over the region S pictured, one can

break it up as indicated into

i

two

regions Sl and Since S1

S2 that

intersect in a set of content zero. and and S2

is of type I

is of type 11, we can compute the integrals

f

I[

{I

s,A.

s,

by iterated integration. We add the results to

obtain

Area.
We can now construct a rigorous theory of area. We
x

already have defined the area of the rectangle Q = [a,b] by the equation

[c,d]

area Q =

11

I. .

, I

We use this same equation for the general definition.

Definition.

L~-t be a bounded s e t i n the plane. S

We say t h a t

is Jordan-measurable i f

Sfs

1 e x i s t s ; i n t h i s case, we define

Note that i f
t:y Theorem 9.

Bd S has content zero, then

is Jordan-measurable, by

The converse a l s o holds; the proof is l e f t a s an exercise.

The area function has the following properties:

Theorem 11. plane.

Let

and If

be measurable - -in the sets -A

Jordan-

(1) (Monotonicity).
(2)

S C T, - area S then

< area T.

(Positivity).
S

Area

S 3 0, and equality holds if

and - only if

has content zero.

(3)

(Additivity) - S n T If
'

is - - -a set of

content zero,

then -

Su T

Jordan-measurable and

area(SuT) = area S + area T.

(4) Area S = Area (Int S) = Area (Su Bd S) Proof. Let Let

Q

.
S and T.

be a rectangle containing

is(x) = 1

for

= 0

for

x ft S.

Define

FT

similarly. is contained in T, then

(1) If S

$ (x)

C lT(x)

Then by the comparison theorem,

area

Ifs 11, /I,% j'b

1 =

L <

I .

= area T.

(2) Since

< 1, we have by the comparison theorem,

0 .

0 =

11, 11,
S

1 = area

s,

for all S.

~f

has content zero, then

$1
S

1=

$1 is =
Q

0,

by Corollary 7
.

Conversely, suppose Given on

Q
E

I/

1 = 0.

Then

/I

2=

is = 0.

>
t

0,

there must be a step function t

is defined

such that

f/

t <

E.

Let

P be a partition relative

to which

is a step function.

Now if a subrectangle Qij

S in its interior, 1. S

of this partition contains a point of then the value of t

on this subrectangle must be at least

E.

~hus these subrectangles have total area less than

Now

is contained in the union of these subrectangles (of total area less than zero.
(3) Because
e)

and the partition lines.

Thus

has content

I, !

1 and

$1 T

exist and

S n T has 1 exists
SUT

content zero, it follows from additivity that andequals 1 + $ $ 1. S T

(4) Since the part of

$1

$1

not in

Int S

lies in

Bd S,

it has content zero.

Then additivity implies that

area S = area(1nt S)

area (S

Int S)

= area (Int S)

A similar remark shows that

area ( S u Bd SJ = area (Int S )

+ area (Bd S)

= area (Int S )

Remark.

Let

be a bounded s e t i n t h e plane.
S,

d i r e c t way of d e f i n i n g t h e a r e a of i n t e g r a t i o n t h e o r y , i s a s follows: taining

S.

without developing
Q

Let

be a r e c t a n g l e con

Given a p a r t i t i o n

of
P

Q,

let

a(P)

denote t h e t o t a l .

a r e a of a l l s u b r e c t a n g l e s of let
A(P)

t h a t a r e contained

&

S,

and
P

denote t h e t o t a l a r e a of a l l s u b r e c t a n g l e s o f S. Define t h e i n n e r a r e a of

that

contain points of

be t h e supremum

of t h e numbers

a (P) , a s

ranges o v e r a l l p a r t i t i o n s o f
S

Q;

and d e f i n e t h e o u t e r a r e a o f
A(P)

t o b e t h e infemum of t h e numbers
S

If t h e i n n e r a r e a and o u t e r a r e a of

a r e equal, t h e i r

common v a l u e i s c a l l e d t h e a r e a of

S.

W e leave it a s a ( n o t t o o d i f f i c u l t ) e x e r c i s e t o show t h a t

t h i s d e f i n i t i o n o f a r e a i s t h e same as t h e one we have given. Remark. There i s j u s t one f a c t t h a t remains t o be proved W e w o u l d c e r t a i n l y w i s h it t o be t r u e

about o u r n o t i o n of a r e a .

t h a t i f two s e t s S and T i n t h e p l a n e a r e "congruent" i n t h e s e n s e

1

of elementary geometry, t h e n t h e i r a r e a s a r e t h e same.

This f a c t i s

n o t immediate from t h e d e f i n i t i o n of a r e a , f o r we used r e c t a n g l e s w i t h s i d e s p a r a l l e l t o t h e c o o r d i n a t e axes t o form t h e p a r t i t i o n s on which we based o u r n o t i o n of " i n t e g r a l " , and hence o f " a r e a " . ~t i s n o t immediate, for i n s t a n c e , t h a t t h e r e c t a n g l e s p i c t u r e d below have t h e same a r e a , f o r t h e a r e a of
T
S

and

i s defined

by approximating sides. ing T

by r e c t a n g l e s w i t h v e r t i c a l and h o r i z o n t a l

[Of c o u r s e , w e can w r i t e e q u a t i o n s f o r t h e c u r v e s boundand compute i t s a r e a by i n t e g r a t i o n , i f w e wish.] Proof o f t h e i n v a r i a n c e o f a r e a under "congruence" w i l l

have t o w a i t u n t i l we s t u d y t h e problem o f change of v a r i a b l e s i n a double i n t e g r a l .

Eh erci ses
1.

Show t h a t i f
Q

ISS 1

e x i s t s , then Since

Bd S

hz-s content zero.

[Hint:

Chwse

so that

S CQ.

S&
< E
S

IS

exists

there a r e
P

functions such

s and
that
s

t h a t a r e s t e p functions r e l a t i v e t o a p a r t i t i o n

of

Q,

<, Is jt o
P

Q and

[$ ( t - s )

Show t h a t t h e subrectangles

determined by
2. ( a )

tk.at contain points of

S

hzve t o t a l volume l e s s than

.] .

Let

and

be bounded subsets of

R~

Show t h a t

B (S U T ) C (Bd SVBd T ) . d

Give
S

an example where e q u a l i t y does not hold

(b) Show t h a t i f
S V T and SnT,

and

a r e Jordan-measurable,

then s o a r e

and furthermore

ar.ea(SbT)

areas

+ areaT

area ( S n T ) .

3.

Express in terms o f iterated integrals the double integral

2 2 x y ,

where

is t h e b o u n d e d p o r t i o n o f t h e f i r s t
xy

quadrant lying between the curves lines

4.
y

= 1

and

xy

= 2
2

and the

= x

and

y = 4x.

(Do

not e v a l u a t e t h e i n t e g r a l s . )
z = x

A s o l i d is b o u n d e d a b o v e b y t h e s u r f a c e
xy-plane, and by the plane

below

by the

x = 2.

Make a sketch;

e x p r e s s i t s v o l u m e a s a n i n t e g r a l ; a n d f i n d t h e volume.

5 .

Express i n t e r n of i t e r a t e d i n t e g r a l s t h e volume of t h e region

i n t h e f i r s t a t a n t of z = 0

R~ bcunded by:

(a) The s u r f a c e r z = xy and

z = xy

and

wd

x + 2y + z = 1.

( b ) The s u r f a c e s

z = 0 and

Let Q denote the rectangle [0,1] x [0,1] in the following exercises. @(a) Let f(x,y) = l/(y-x) f(x,y) = 0 Does if x # y,

if x = y.

JJQf

exist? if x # y,

(b) Let g(x,y) = sin (l/(y-x))

Does

JJQg exist?

@ ~ e f(x,y) = 1if x = 112 and y is rational, t

,

Show that

JJQ

f(x,y) = 0 otherwise 1 f exists but f(x,y)dy fails to exist when x = 112. 0

@ Let f(x,y) = 1if (x,y) has the form (a/p,b/p),

where a and b are integers and p is prime, f(x,y) = 0 otherwise. Show that

J J f(x,y)dy dx exists but

f does not.

JJQ

GREEN'S THEOREN AND I T S APPLICATIONS

The d i s c u s s i o n i n 1 1 . 1 9

11.27 o f , Apostol i s n o t complete

We

n o r e n t i r e l y r i g o r o u s , a s t h e a u t h o r himself p o i n t s o u t . give here a rigorous treatment.

nr e e n ' s G

Theorem - -e P l a n e i n th

W a l r e a d y know what i s meant by s a y i n g t h a t a r e g i o n i n e t h e p l a n e i s o f Type I o r o f Type 11 o r t h a t i t i s o f b o t h types simultaneous l y

Aposiso 1 p r o v e s Green ' s Theorem f o r Such a r e g i o n R can be

a region t h a t i s of both types.

d e s c r i b e d i n two d i f f e r e n t ways, a s f o l l o w s :

The a u t h o r ' s proof i s complete and r i g o r o u s e x c e p t f o r one gap, which a r i s e s from h i s u s e of t h e i n t u i t i v e n o t i o n of "counter
1

clockwisen.

S p e c i f i c a l l y , what h e does i s t h e f o l l o w i n g :

For t h e f i r s t :

p a r t o f t h e proof h e o r i e n t s t h e boundary C o f R as f o l l o w s :
(*)

By i n c r e a s i n g x , on t h e c u r v e y = @(Ix ) ; By i n c r e a s i n g y , on t h e l i n e segment x = b; By d e c r e a s i n g x , on t h e c u r v e y = $ 2 ( x ) : and By d e c r e a s i n g y , on t h e l i n e segment x = a .

Then i n t h e second p a r t o f t h e p r o o f , h e o r i e n t s C a s f o l l o w s :
(**)

By d e c r e a s i n g y , on t h e c u r v e x = q L ( y ) ;
By i n c r e a s i n g x , on t h e l i n e segment y = c;
By i n c r e a s i n g y , on t h e c u r v e x = I J ~ ( Y ) a nd
; By d e c r e a s i n g x, on t h e l i n e segment y = d .

(The l a t t e r l i n e segment c o l l a p s e s t o a s i n g l e p o i n t i n t h e prec e d i ng f i g u r e . ) The c r u c i a l q u e s t i o n i s : orientations How d o e s one know ---t h e s e two

a re t h e s ante? ---

One c a n i n f a c t see t h a t t h e s e two o r i e n t a t i o n s a r e t h e same, by simply a n a l y z i n g a b i t more c a r e f u l l y what one means by a r e g i o n o f ~

~ I and 11. ~ e

S p e c i f i c a l l y , s u c h a r e g i o n can b e d e s c r i b e d by f o u r monotonic f u n c t i o n s :

A..

wt..ere

dl

and d4 are strictly

decreasing and

X2

and d3 are
,

U w *

stzictly increasing,

[some o r a l l o f t h e ai can be missing, of course.

Here a r e p i c t u r e s of t y p i c a l such regions:]

T h e curves a1 and a2, a l o n g w i t h t h e l i n e segment y = c , a r e

used t o d e f i n e t h e c u r v e y = + 1 ( ~ ) t h a t bounds t h e r e g i o n on t h e bottom. S i m i l a r l y , a 3 and a4 and y = d d e f i n e t h e curve

y = C $ ~ ( X t h a t bounds t h e r e g i o n on t h e top. )

S i m i l a r l y , t h e i n v e r s e f u n c t i o n s t o a1 and a 3 , along with

x = a , combine t o d e f i n e t h e curve x = Q l ( y ) t h a t bounds t h e

r e g i o n on t h e l e f t ; and t h e i n v e r s e f u n c t i o n s t o ag and a d , a l o n g w i t h x = b, d e f i n e t h e curve x =
NOW

P2 ( y ) .

one can choose a d i r e c t i o n on t h e bounding curve C by

simply d i r e c t i n g each of t h e s e e i g h t curves a s i n d i c a t e d i n t h e f i g u r e , and check t h a t t h i s i s t h e same a s t h e d i r e c t i o n s s p e c i f i e d i n ( * ) and (**I - a s follows: i n c r e a s i n g x = d e c r e a s i n g y on y = al ( x )

increasing x o n y = c

[ ~ o r m a l l y , one d i r e c t s t h e s e curves

i n c r e a s i n g x = i n c r e a s i n g y on y = a 2 ( x )
increasing y o n x = b

d e c r e a s i n g x = i n c r e a s i n g y on y = cr4 (x)
decreasing x o n y = d

d e c r e a s i n g x = d e c r e a s i n g y on y = a g ( x )
decreasing y tJe make t h e f o l l o w i n g d e f i n i t i o n : Definition.
L e t R be a n open s e t i n t h e p l a n e bounded by
W e say t h a t

a s i m p l e c l o s e d p i e c e w i s e - d i f f e r e n t i a b l e curve C.

R i s a Green's r e g i o n i f i t i s p o s s i b l e t o choose a d i r e c t i o n
on C s o t h a t t h e e q u a t i o n Pdx C
, i

Qdy =

[I

(2- 5 dxdy 1
R

holds f o r every continuously d i f f e r e n t i a b l e vector f i e l d P(x,y)r

-t

~ ( x , ~ )t 'hja t i s d e f i n e d i n a n open s e t c o n t a i n i n g

R and C.

The d i r e c t i o n on C t h a t makes t h i s e q u a t i o n c o r r e c t i s

c a l l e d t h e counterclockwise d i r e c t i o n , o r t h e counterclockwise o r i e n t a t i o n , o f C.
I n t h e s e terms, Theorem 11.10 o f Apostul can be r e s t a t e d

a s follows : Theorem 1.
Let be - R - bounded

b~ a s i m p l e c l o s e d o i e c e -

wise-differentiable csrve.
i -s-a

If -R

i s of and - - Types I - 11,

then R

Green's r e g i o n .
As t h e f o l l o w i n g f i g u r e i l l u s t r a t e s , a l m o s t any r e g i o n

you a r e l i k e l y t o draw can be shown t o be a Green's r e g i o n by r e p e a t e d a p p l i c a t i o n o f t h i s theorem. I n such a c a s e , t h e

c o u n t e r c l o c k w i s e d i r e c t i o n on C i s by d e f i n i t i o n t h e one for which Green's theorem holds.

I

For example, t h e r e g i o n R i s a

G r e e n ' s r e g i o n , and t h e counterclockwise o r i e n t a t i o n o f i t s boundary C i s a s i n d i c a t e d . The f i g u r e on t h e r i g h t i n d i c a t e s and

R2

t h e proof t h a t i t i s a Green's r e g i o n ; each o f R1

is o f Types I and 11.

Definition.

L e t R b e a bounded r e g i o n i n t h e p l a n e

whose boundary i s t h e union o f t h e d i s j o i n t p i e c e w i s e - d i f f e r e n t i a b l e s i m p l e c l o s e d c u r v e s C1,

..., Cn.

We c a l l R a

g e n e r a l i z e d Green's r e g i o n i f i t i s p o s s i b l e t o d i r e c t the

curves C1,

...,

Cn so t h a t t h e e q u a t i o n

holds f o r every continuously d i f f e r e n t i a b l e vector f i e l d

~f + (23 d e f i n e d

i n a n open s e t about R and C.

Once a g a i n , e v e r y such r e g i o n you a r e l i k e l y t o draw can

b e shown t o be a g e n e r a l i z e d Greents r e g i o n by s e v e r a l a p p l i -

c a t i o n s o f Theorem 1.
generalized

For example, t h e r e g i o n R p i c t u r e d

is aAGreenls r e g i o n i f its boundary i s d i r e c t e d a s i n d i c a t e d .

The proof i s i n d i c a t e d i n t h e figure on the r i g h t .
One a p p l i e s

Theorem 1 t o each of the 8 r e g i o n s p i c t u r e d and adds the r e s u l t s together.

IXza
Definition. Let C be a piecewise-differentiable curve in the plane parametrized by

- the function d t ) = (x(t),y(t)). The vector T = (x' (t),y8(t))/llgt (t) 11 is the unit tangent
vector to C. The vector

s = (Y' (t),-x'(t))/llsr'(t)ll

is called the unit negative normal to C.

If C is a simple closed curve oriented counterclockwise, then q is the "outward normal"

-8

Iff = Pi

+ Q j is a continuously differentiable vector field defined in an open

set containing C, then the integral line integral

J, ( 1 - ~ ) d iss well-defined; show that it equals the

ss, [E

Show that if C bounds a region R that is a Green7sregion, thenqc (f-p)dS =

+ %]dxdy.

Remark. If f is the velocity of a fluid, then outward through C in unit time. Thus aP/&

5c

(f-p)dS is the area of fluid flowing measures the rate of expansion

of the fluid, per unit area. It is called the divergence off.] Definition. Let $ be a scalar field (continuously differentiable) defined on C. If g is a point of C, then $'(zc;g) is the directional derivative of $ in the direction of p. It is
4

equal to V$(g) -9, course. Physicists and engineers use the (lousy) notation of denote this directional derivative.

2to

@ Let R be a Green's region bounded by C. Let f and g be scalar fields (with

continuous first and second partials) in an open set about R and C. (a) show

CQC 2

ds =

J'J

v2g dx dy
R

where v2g = a 2 g / h 2
(b) Show

+ a2g/lly2.

(c) If V 2f = 0 = V 2g, show

These equations are important in applied math and classical physics. A function f with V 2f = 0 is said to be harmonic. Such functions arise in physics: In a region free of charge, electrostatic potential is harmonic; for a body in temperature equilibrium, the temperature function is harmonic.

C ~ n d i t i o n sUnder which
Let f = P ;

P?

Q j i s a Gradient.

--

p3 be a continuously d i f f e r e n t i a b l e vector

f i e l d defined on an open s e t S i n the plane, such that

aP/ay = 3 4 / a # on S,

In g e n e r t l , we knzw t h a t f need not be

( . -

a gradient on S,

W e do know t h a t f w i l l be a g r a d i e n t i f S

is convex (or even i f S i s star-convex).

W e seek t o e x t e n d

t h i s r e s u l t t o a more general c l a s s o f plane sets.

T h i s more g e n e r a l c l a s s may be informally described as

c o n s i s t i n g o f t h o s e regions i n the p l a n e t h a t have no "holesw.

For example, t h e region St i n s i d e a simple closed curve C1 has

no h o l e s , nor does t h e r e g i o n S2 o b t a i n e d from t h e plane by d e l e t i n g t h e non-negative x-axis, O t h e o t h e r hand, t h e n

r e g i o n S3 c o n s i s t i n g of t h e p o i n t s i n s i d e C2 and o u t s i d e
C3 h a s a h o l e , and s o does t h e r e g i o n S 4 o b t a i n e d from t h e

p l a n e by d e l e t i n g t h e o r i g i n . Needless t o s a y , w e must make t h i s c o n d i t i o n more p r e c i s e i f w e a r e t o prove a theorem about it.

1

This t a s k t u r n s o u t

t o be s u r p r i s i n g l y d i f f i c u l t ,
W e b e g i n by p r o v i n g some f a c t s about t h e geometry of t h e

plane.

Definition.

A s t a i r s t e p curve C i n t h e p l a n e i s a curve

t h a t i s made up of f i n i t e l y many h o r i z o n t a l and v e r t i c a l l i n e

segments

.
Then by u s i n g t h e c o o r d i n a t e s o f t h e end

For such a c u n r e C, w e can choose a r e c t a n g l e Q whose i n t e r i o r c o n t a i n s C.

p o i n t s of t h e l i n e segments o f t h e curve C a s p a r t i t i o n p o i n t s ,

w e can c o n s t r u c t a p a r t i t i o n o f Q such t h a t C i s made up

e n t i r e l y o f edges o f s u b r e c t a n g l e s o f t h i s p a r t i t i o n . process i s i l l u s t r a t e d i n t h e following figure:
s his

Let in t Then t h - C be-a simple c l o s e d s t a i r s t e p curve - h e plane. --e complement - C - - w r i t t e n - -e union - - d i s j o i n t of can be as th of two and h e he open - - o f t h e s e sets i s bounded - t - o t- r i s sets. One Each has a s its unbounded. - o f them- C -- boundary.
Theorem 2 . Proof. Choose a r e c t a n g l e Q whose i n t e r i o r c o n t a i n s C ,

(The Jordan curve theorem f o r s t a i r s t e p c u r v e s ) .

and a p a r t i t i o n o f Q, s a y xo < x 1 < c x < ym, n and yo c yl c s u c h t h a t C i s made up of edges o f s u b r e c t a n g l e s of t h i s p a r t i t i o n . S t e p 1. the partition W b e g i n by marking e each o f t h e r e c t a n g l e s i n

...

...

t h e following rule: t Consider t h e r e c t a n g l e s i n t h e i-h "column" beginning w i t h Mark t h e bottom one w i t h

or

- by

t h e bottom one.

+.

In general, i f a

g i v e n r e c t a n g l e i s marked w i t h

+ o r -, mark t h e one j u s t above

it with t h e same s i g n if t h e i r common edge does n o t l i e i n C ,

and w i t h t h e o p p o s i t e s i g n i f t h i s edge d oes l i e i n C. t h i s p r o c e s s f o r each column of r e c t a n g l e s .

Repeat

In t h e following

f i g u r e , w e have marked t h e r e c t a n g l e s i n columns 1 , 3 , and 6 , t o i l l u s t r a t e t h e process.

Note t h a t t h e r e c t a n g l e s i n t h e bottom row a r e always

1

marked +, and s o a r e t h o s e i n t h e f i r s t and l a s t columns, (since C does n o t touch t h e boundary o f Q

).

SteE 2. of t h - -e

W e prove t h e following:

I f two --

subrectangles

p a r t i t i o n have an edge - common, then they have o p p o s i t e in --

s i no i f t is i n and have t h e same i f th 4 - h a t edge - - C, - they --- s i g n a t edge i s n o t i n C.

T h i s r e s u l t h o l d s by c o n s t r u c t i o n f o r t h e h o r i z o n t a l edges.

---

we

rove

i t holds f o r t h e v e r t i c a l edges, by i n d u c t i o n .

~t i s t r u e f o r e a c h of t h e lowest v e r t i c a l edges, t h o s e

o f t h e form xix[yo,yl].

(For no such edge i s i n C, and t h e

+.)

bottom r e c t a n g l e s are a l l marked f o r t h e r e c t a n g l e s i n row j

. i n row j

Supposing now it i s t r u e

- 1, w e prove i t t r u e f o r r e c t a n g l e s
(! )

There a r e 16 c a s e s t o c o n s i d e r

, of

which we

i l l u s t r a t e 8:

'-wj j-I

(1)

(The other eight are obtained from these by changing all the signs.)

We know in each case, by construction, whether the two horizontal edges are in C, and we h o w from the induction hypothesis whether the lower vectical edge is in C. Those edges that we h o w are in C are marked heavily in the figure. We seek to determine whether the upper vertical edge (marked
"?I1)

is in C or not.

hTe use

the fact that C is a

simple closed curve, which implies in particular that

e a c h v e r t e x i n C l i e s on e x a c t l y two edges i n C.

In case ( I ) ,

t h i s means t h a t t h e upper v e r t i c a l edge i s n o t i n C, f o r o t h e r w i s e t h e middle v e r t e x would be on o n l y one edge o f C. S i m i l a r l y , i n c a s e s (21,

( 3 1 , and ( 4 1 , t h e upper v e r t i c a l edge

i s n o t i n C, f o r o t h e r w i s e t h e middle v e r t e x would l i e on
t h r e e edges o f C. S i m i l a r r e a s o n i n g shows t h a t i n c a s e s ( S ) ,
i

(6), and ( 7 )

t h e upper v e r t i c a l edge must l i e i n C, and i t shows t h a t c a s e ( 8 ) c a n n o t 'occur. Thus S t e p 2 i s proved i n t h e s e 8 c a s e s . The o t h e r 8

a r e symmetric t o t h e s e , s o t h e same proof a p p l i e s . S t e p 3.

I t f o l l o w s from S t e p 2 t h a t t h e t o p row o f

r e c t a n g l e s i s marked rectangles a r e S t e p 4.

+, s i n c e t h e upper marked +, and C does n o t

l e f t and upper r i g h t touch t h e boundary of Q.

W e d i v i d e a l l o f t h e complement of C i n t o two I n t o U w e p u t t h e i n t e r i o r s of a l l i n t o V w e p u t t h e i n t e r i o r s of a l l a l s o p u t i n t o V a l l p o i n t s of t h e
W e s t i l l have

sets U and V a s follows.

r e c t a n g l e s marked r e c t a n g l e s marked

-, and +. W e

p l a n e l y i n g o u t s i d e and on t h e boundary o f Q.
1
I

t o d e c i d e where t o p u t t h e edges and v e r t i c e s of t h e p a r t i t i o n t h a t d o n o t l i e i n C.

C o n s i d e r f i r s t a n edge l y i n g i n t e r i o r t o Q.

I f it does

n o t l i e i n t h e curve C, then both i t s adjacent rectangles l i e i n U o r b o t h l i e i n V (by S t e p 2 ) ; p u t t h i s (open) edge i n

U o r i n V accardingly.
i n t e r i o r t o 9.

Finally, consider a vertex v t h a t l i e s

I f i t i s n o t on t h e c u r v e C, t h e n c a s e (1) o f

t h e p r e c e d i n g e i g h t cases ( o r t h e c a s e w i t h o p p o s i t e s i g n s ) holds. Then a l l f o u r o f t h e . a d j a c e n t r e c t a n g l e s a r e i n U o r

a l l f o u r a r e i n V; p u t v i n t o U o r V a c c o r d i n g l y .
It i s i m m e d i a t e l y c l e a r from t h e c o n s t r u c t i o n t h a t U and V

a r e open s e t s ; any p o i n t o f U o r V ( w h e t h e r i t i s i n t e r i o r t o a s u b r e c t a n g l e , o r o n a n e d g e , o r i s a v e r t e x ) l i e s i n an open b a l l contained e n t i r e l y i n U o r V

U i s bounded and V i s unbounded,
)

.
+

I t i s a l s o immediate t h a t

F u r t h e r m o r e , C i s t h e common

boundary o f U and V, b e c a u s e f o r e a c h edge l y i n g i n C , one o f t h e a d j a c e n t - ' r e c t a n g l e s i s narlred by S t e p 2 . ~ e f i n i t i o n . L e t C be a simple c l o s e d s t a i r s t e p curve i n t h e plane. The bounded open s e t U c o n s t r u c t e d i n t h e p r e c e d i n g
C,

and t h e o t h e r i s marked

-,

proof i s c a l l e d t h e i n n e r region of
I t is true that
U

o r t h e region inside

C.

and V are connected, but the proof is d i f f i c u l t .

W s h a l l not need t h i s f a c t . e
Definition, Then
S

Let

b e a n open c o n n e c t e d s e t i n t h e p l a n e .

is c a l l e d simply connected, i f , f o r every simple c l o s e d

C

s t a i r s t e p curve

which l i e s i n S.

S,

t h e inner region of

is

also a s u b s e t o f

Theorem 3 . s t a i r s t e p curve

If i s th r a - U - -e e g i o n i n s i a e - s i m p l e C, - U -s-a Green's region. then i

closed

~ r o o f . Choose a p a r t i t i o n o f a r e c t a n g l e Q e n c l o s i n g U such t h a t C c o n s i s t s e n t i r e l y o f edges o f s u b r e c t a n g l e s o f the partition. F o r each s u b r e c t a n g l e Q i j of t h i s p a r t i t i o n

l y i n g i n U, i t is t r u e t h a t

i f Ci

i s t h e boundary of Q i j ,
(For Qij

traversed i n a c o u n t e r c l o c k ~ i s e Now each edge of

direction.

i s a t y p e 1-11 r e g i o n ) .

t h e p a r t i t i o n l y i n g i n C a p p e a r s i n o n l y one of t h e s e c u r v e s
Cij,
)

and e a c h edge of t h e p a r t i t i o n n o t l y i n g i n C appears i n with

e i t h e r none o f t h e Ci j ' o r i t appears i n two o f t h e C i j opposikely d i r e c t e d arrows, a s indicated:

I f w e sum over a l l s u b r e c t a n g l e s Qij i n U, w e t h u s o b t a i n t h e equation

I U i n e segments i n

CI

MA'+ Qdy =

- 2) d x d ~ e

T h e o n l y q u e s t i o n i s w h e t h e r t h e d i r e c t i o n s w e have t h u s g i v e n

t o t h e l i n e s e g m e n t s l y i n g i n C combine t o g i v e an o r i e n t a t i o n o f C.

That t h e y d o i s p r o v e d by examining t h e p o s s i b l e c a s e s .
t h e o t h e r seven a r e opposite

Seven o f them a r e a s f o l l o w s ;

t o them.
-

T h e s e d i a g r a m s show t h a t f o r e a c h v e r t e x v o f t h e p a r t i t i o n s u c h t h a t v i s on t h e c u r v e C , v i s t h e i n i t i a l p o i n t o f one o f t h e t w o l i n e segments o f C t o u c h i n g i t , and t h e f i n a l p o i n t of the other. Theorem 4. every p a i r in Let - S. -

0
pla.ne s u c h t h a t

Let be an se i t - S - - o p e n -t-n-h e

of

p o i n t s o f S can be joined

- --

~ - s t a i r s t e p curve a

--

be a v e c t o r f i e l d t h a t i s c o n t i n u o u s l y d i f f e r e n t i a b l e i n S ,

s u c h that

--

--

on a l l of - -S.
1

S.

(a) If S

is - simply

connected

then

is a qradient
b a qradient in S. e

(b) f S

L not s

simply

connectad, then

--

or - may not

Proof.

h e proof of (b) is l e f t a s an exercise.

W prove ( a ) here. e

Assume t h a t
St.= -

is simply c o n n ~ c t e d .
WE show t h a t

1.

Pdx + Qdy =

f o r e v e r y s i m p l e c l o s e d s t a i r s t e p c u r v e C Lying i n S.

We

know t h a t t h e r e g i o n U i n s i d e C i s a Green's r e g i o n . U

W e a l s o know t h a t t h e r e g i o n

l i e s entirely within t h a t i s not i n so t h a t

S
S S,

S. then

(For
C

i f there w e r e a point encircles a point

p

of

U
S,

not i n

has a hole a t

p.

his c o n t r a d i c t s t h e Eact that

is simply connected.) h o l d s on a l l o f
;

There

f o r e t h e equation f o r e conclude t h a t

aQ/ax = aP/ay

we t h e r e

f o r some o r i e n t a t i o n o f

(and hence f o r b o t h o r i e n t a t i o n s o f

S t e p 2.

W e show t h a t i f

---

f o r e v e r y simple c l o s e d s t a i r s t e p c u r v e i n equation holds f o r s t a i r s t e p curve i n

S, S.

t h e n t h e same

Assume

c o n s i s t s of t h e edges of s u b r e c t a n g l e s
C,

i n a p a r t i t i o n o f some r e c t a n g l e t h a t c o n t a i n s

a s usual.

W e proceed by i n d u c t i o n o n t h e number of v e r t i c e s on t h e

c u r v e C.

Consider t h e v e r t i c e s o f C i n o r d e r :

Now C c a n n o t have o n l y one v e r t e x . C i s a p a t h going from vo t o v i n t e g r a l vanishes

1

I f i t h a s o n l y two, t h e n

and t h e n back t o vo.

The l i n e

i n t h i s case. Now suppose t h e theorem t r u e f o r c u r v e s w i t h fewer t h a n n vertices.

L e t C have n v e r t i c e s .
I f C i s a s i m p l e c u r v e , we
,

a r e through.

Otherwise, l e t vk b e t h e first v e r t e x i n t h i s
W e

sequence t h a t e q u a l s some e a r l i e r v e r t e x vi f o r i < k. c a n n o t have vk segment.

It Vk =

vkWl,

f o r then vkmlvk would n o t be a l i n e

Vk-2vk-1' Then t h e i n t e g r a l from v

k-2' t h e n t h e curve c o n t a i n s t h e l i n e segment f o l l o w e d by t h e s a m e l i n e segment i n r e v e r s e o r d e r .

V

~ to vkWl and t h e i n t e g r a l from - ~

k-1 t o vk a r e n e g a t i v e s o f each o t h e r . W e can d e l e t e vk-l from t h e sequence o f v e r t i c e s w i t h o u t changing t h e v a l u e of

the integral.

W have a c l o s e d c u r v e remaining w i t h fewer e

l i n e segments t h a n b e f o r e , and t h e i n d u c t i o n h y p o t h e s i s applies. I f i C k-2, t h e n w e can c o n s i d e r t h e c l o s e d curve w i t h v e r t i c e s vi,

This i s a s i m p l e c l o s e d c u r v e , s i n c e k
' a l l i t s v e r t i c e s a r e d i s t i n c t , s o t h e i n t e g r a l around i t i s
V~+~,...,V

z e r o , by S t e p 1.
/C Pdx

Therefore t h e value of t h e i n t e g r a l

Qdy i s n o t changed i f we d e l e t e t h i s p a r t of C , i.e.,

i f w e d e l e t e t h e v e r t i c e s vi,.

..

,v

k- 1 from t h e sequence.

Then

t h e induction hypothesis applies. Example.

In t h e f o l l o w i n g case,

t h e f i r s t v e r t e x a t which t h e curve touches a v e r t e x a l r e a d y

I

touched is t h e p o i n t

q.

One c o n s i d e r s t h e s i m p l e c l o s e d c r o s s ~ e l e t i n gthis

h a t c h e d curve, t h e i n t e g r a l around which i s zero.

c u r v e , one h a s a c u r v e remaining c o n s i s t i n g o f fewer l i n e segments. You can c o n t i n u e t h e p r o c e s s u n t i l you have a s i m p l e

c l o s e d c u r v e remaining. Step 3.
W e show t h a t i f C1 and C2 a r e any two s t a i r s t e p

c u r v e s i n S from p t o q, t h e n

IC1Pdx

Qdy = JC2 Pdx + Qdy. I f -C2 d e n o t e s C2 w i t h

his f o l l o w s by t h e u s u a l argument.

t h e r e v e r s e d d i r e c t i o n , t h e n C = C1 curve.
W e have

(-C2)

i n a closed s t a i r s t e p

T h i s last i n t e g r a l v a n i s h e s , by S t e p 2. S t e p 4. Now w e prove t h e theorem.

L e t a be a f i x e d

p o i n t o f S , and d e f i n e

$ (x) =

J'c (g Pdx +

Qdy.

where C ( x ) is any s t a i r s t e p c u r v e i n S from a t o x.

There

always e x i s t s s u c h a s t a i r s t e p c u r v e (by h y p o t h e s i s ) , and t h e v a l u e o f t h e l i n e i n t e g r a l i s independent of t h e c h o i c e of t h e c u r v e (by S t e p 3).

I t remains t o show t h a t

a$/ax = P and

= Q.

W proved t h i s once b e f o r e under t h e assumption t h a t e

I

C ( 5 ) was

an a r b i t r a r y p i e c e w i s e smooth curve.

B u t t h e proof works j u s t a s

w e l l i f w require e

C(5)

t o be a s t a i r s t e p curve.

To compute
W e computed

w e f i r s t computed
choosing a c u r v e g r a t e d along curve
Cl.
W e computed

from $(x+h,y)

and i n t e by choosing t h i s same

plus the s t r a i g h t l i n e

from t o be a s t a i r s t e p c u r v e .
C1

t h e p r e s e n t c a s e , we have r e q u i r e d Then we n o t e t h a t i f C1

&2

s t a i r s t e p ' u v e c r

+ C2

is - a l s o

a - stairstep
o u t change.

curve.

T h e r e f o r e t h e e a r l i e r proof goes through with-

Remark.

I t i s a f a c t t h a t i f two p a i r o f p o i n t s o f
S,

can

b e j o i n e d by some p a t h i n
s t e p path.
i

t h e n t h e y can b e joined by a s t a i r I t follows

S

( W e s h a l l n o t b o t h e r t o prove t h i s f a c t . )

t h a t t h e h y p o t h e s i s o f t h e preceding theorem i s merely t h a t connected and s i m p l y connected, Exercises

I.

be

Let S 5e the punctured plane, i.?., the plane with the Show that ths vector fields

orizin deleted.

satisfy the condition (a) Show that

2 2

b?/Jy
f -

a a/ax.
[Hint: F i r s t find ~(3 so that [Hint:

is a gradient in S.

~ a / ) x x/(x =
Compute

c y )

.]

(b) Show that q

is not a gradient in S.

9- dd

where C i s the unit circle centered a t the origin.]

Let
C2

2.

Prove the following:

'Rreorem 5. Let C1 be a simple closed stairstep curve in the plane.

b~ a simple closed stairstep curve that is contained in the inner

region of C1. Show that the region consisting of those points that are in the inner region of C1 and are not on
C2

nor in the inner region of C

C2

is a generalized Green's region, bounded by C1 ard

[Hint: Follow the pattgrn of the proof of Theorem 3.1

3 Let q be the vector field of Exercise 1. . closed stairstep w
J C f*dl # 0 .

Let C be any simple

. e whose inner region contains Q Show that

[Hint: Show this inequality 'holds if C is the boundary

oE a rectangle. Then apply Theorem 5. ]

*4.

Even i f t h e r e g i o n

i s not s i m p l y c o n n e c t e d , o n e c a n

u s u a l l y determine whether a given v e c t o r f i e l d e q u a l s a g r a d i e n t

i

field i n

S.

Here i s o n e example, where t h e r e g i o n

is the

punctured plane, Theorem 6 . d i f f e r e n t i a b l e and Suppose

-. - that f

= Pi

- + Qj - c o n t i n u o u s l y - is

enclosing

i n th Let b - -e p u n c t u r e d p l a n e . - R -e-a f i x e d r e c t a n g l e the c let - o r i g i n ; o r i e n t Bd R -o u n t e r c l o c k w i s e ; A = \

(a)

If

Pdx+Qdy. Bd R C i s any s i m p l e c l o s e d s t a i r s t e p c u r v e n o t touch-

--

the then -o r i g i n , -

e i t h e r equals
O]r
I

5 A

( i f t h e o r i g i n is i n t h e inner region of

C)

(otherwise).

(b)

If -

A = 0,

then -

equals - gradient f i e l d i n the a

--

punctured plane.
(c)

[Hint:

I m i t a t e t h e p r o o f o f Theorem 4 . I differs - - a gradient f i e l d from

If

# 0,

then f --

a -

c o n s t a n t m u l t i p l e of the v e c t o r f i e l d

--

That i s , t h e r e i s a c o n s t a n t

such t h a t

f -+

cq

equals a

g r a d i e n t f i e l d i n t h e punctured plane.

(Indeed,

c = - ~ / 2 n .)

Subtracting, w e obtain
aQ - - -(ax au 3x au a v

ax a - -) Y av au

J(u,v) ax
Since

,
= f

is evaluated a t

2 Q/&

we have our desired result :

One c a n weaken t h e h y p o t h e s i s o f t h i s theorem a b i t i f o n e wishes.

f(XIY)

S p e c i f i c a l l y , it is not necessary t h a t t h e function which i s b e i n g i n t e g r a t e d b e c o n t i n u o u s i n a n e n t i r e r e c S.

tangle containing t h e region of i n t e g r a t i o n if and f(x,y)

It w i l l suffice
S

i s m e r e l y c o n t i n u o u s o n some open s e t c o n t a i n i n g

C.

F o r it i s a s t a n d a r d theorem ( n o t too- d i f f i c u l t t o p r o v e ) g and t h a t is c o n t i n u o u s

t h a t i n t h i s c a s e one can f i n d a function i n t h e e n t i r e p l a n e and e q u a l s t h e theorem t o t h e f u n c t i o n g. f on

S

C.

One t h e n a p p l i e s

I,

1i

aY

aY + -61 av

2 (t)) dt

where the partials are emluated at P_(t). We can write this last integral
as a fine integral over the curve D. Indeed, if we define

then this last integral can be written as

NOW

we apply Green's theorem to express this line integral as a

Since

double integral.

is by hypothesis a Green's region,

this line integral equals

It remains to compute these partials, using the chain rule.

have

We

proof.
I

Let

R = [c,dl

[c',dtl.

Define Then aQ/ax = f(x,y)

Q(x,y) = on all of

f(t,y)dt because

for
f

(x,y) in

Re

R,

is continuous.

We prove our theorem

by applying Green's theorem.

Let

(u,v) = @(t)

be a choose the
T.

parametrization of the curve D, for a < t - b; <

counterclockwise direction, so Green's theorem holds for Then

-(t) a

~(fi(t))

is a parametrization of the t.

curve C .

It may be constant on some Subintervals of the

axis, but that doesn't matter when we compute line integrals.

dr
Also, it may be countsrclock~~ise clockwise.

We apply Green's theorem to S :

I\
S

f (x,y)dx d y =

\\
S

aa/ax ax ay = t

+ + (01 + Qj) ads.

This sign is

if

a -(t)

parametrizes

in the counterclock-

wise direction, and integral,

otherwise.

Now let us compute this line

Ez3
R e - chanqe of variables theorem
7 (The . Dieorem - - chanqe of variables theorem)

- Let S
in the -C

be an open set in the (x.y) plane and let T

--

be - an o w n set

(u, planeI houndfd v)

D

the piecewise-diffe&tiable
=

simile closed curves

(co*ti*uercrly d i & f e r e n f ia bLs\

,respectively. - F(u,v) Let

(X(u.v). Y(u.v))

a transformation

from --an open set of (u,v) plane into the (x,y) plane that carries
T into and onto . As a Sf - carries D = a T - C = 2 S - - transformation of D onto C , b_e F may k constant on some segments o D, but otherwise f Assume

if one-to-one.

and S - T are Green's reqions. Assume that f(x,y) & continuous

---4. 10

in some -- rectanqle

R containing S. Then

if

Here - J(u,v)

= det JX,Y/~U~V

-- siqn The

F carries

clockwise orientation of ctherwise.

D to the clockwise orientation of C, and is

--

1 1 I
I

r I

b&le

1.

Consider the polar coordinate transformation F(rIQ)

=

(r cos 8 , r sin 8)

It carries the rectangle T in the (r,Q) plane indicated in the figure into

the wedge S in the (xIy)plane. It is constant on the left edge of Tt

but . is one-to-one on the rest of T Note that it canies the counterclockwise . orientation of D = JT to the counterclockwise orientation of C =

3 S.

A a l t e r n a t e v e r s i o n o f t h e change of v a r i a b l e s theorem i s n t h e following: Theorem 8 . theorem. region

T.

Assume a l l t h e hypotheses o f t h e p r e c e d i n g does n o -- J ( u , v ) --t change s i g n o n t h e

--

Assume a l s o t h a t

--

If s i g n is

J(u,v)

>

on a l of - -l-

v a r i a b l e s formula i s

- -

+;

while

the i n th of - s i g n - -e change if the - J ( u , v ) C 0 on -l-f T, -al o

T,
-f

Therefore i n e i t h e r c a s e

Proof. f ( x , y ) z 1.
I

W a p p l y t h e p r e c e d i n g theorem t o t h e f u n c t i o n e W o b t a i n t h e formula e

11
J(u,v)
0

dx dy = t

I\

J ( u , v ) du dv.

The l e f t s i d e of t h i s e q u a t i o n is p o s i t . i v e . on a l l o f T,

heref fore i f

t h e s i g n on t h e r i g h t s i d e of t h e J(u,v) C 0 on a l l o f
T,

formula must be s i g n must be

+;

while i f

the

Now we r e c a l l t h a t t h e s i g n does n o t depend on

t h e p a r t i c u l a r f u n c t i o n b e i n g i n t e g r a t e d , o n l y on t h e transforma-

t i o n involved.

Then t h e theorem i s proved.

Remark.
!

The formula we have j u s t proved g i v e s a g e o m e t r i c

i n t e r p r e t a t i o n of t h e J a c o b i a n d e t e r m i n a n t of a t r a n s f o r m a t i o n . ~f J(u,v) # 0 a t a particular point

T

( u O , v O ) , l e t us choose a
S

small rectangle

a b o u t t h i s p o i n t , and c o n s i d e r i t s image If

under t h e t r a n s f o r m a t i o n .

is s m a l l enough,

J(u,v)

will

be v e r y c l o s e t o Assuming S

J(uo,vo)

on

T,

and s o w i l l n o t change s i g n .

i s a G r e e n ' s r e g i o n , w e have

area

//

dx d y =

\I

IJ(u,v)~

du dvt

so

area S

IJ(U~,V~) I

(area T ) .

Thus, r o u g h l y s p e a k i n g , t h e magnitude o f

J(u,v)

measures how

much t h e t r a n s f o r m a t i o n s t r e t c h e s o r s h r i n k s a r e a s a s it c a r r i e s a p i e c e of t h e the sign of

u, v

plane t o a piece of t h e

x, y

plane.

And

J(u,v)

t e l l s whether t h e t r a n s f o r m a t i o n p r e s e r v e s

o r i e n t a t i o n o r not; i f t h e s i g n i s negative, then t h e transformation " f l i p s overn t h e region t o f i t onto S.

before shrinking o r s t r e t c h i n g it

-~cT",o;~~~)s L
AS a n a p p l i c a t i o n o f

=-7

t h e change o f v a r i a b l e s theorem, w e

s h a l l v e r i f y t h e f i n a l p r o p e r t y o f o u r n o t i o n o f a r e a , namely, t h e f a c t t h a t c o n g r u e n t r e g i o n s i n t h e p l a n e h a v e t h e same a r e a . F i r s t , w e m u s t make p r e c i s e what w e mean b y a " c o n g r u e n c e . " ~ e f i n i t i o n . A t r a n s f o r m a t i o n h : R 2 -> or an isomet t o i t s e l f i s c a l l e d a congruence &preserves A points. That is, h is a congruence i f

R~

of t h e plane

d i s t a n c e s between

for every p a i r

a -, b If h - -

of p o i n t s i n t h e plane. has -h -

The f o l l o w i n g i s a p u r e l y g e o m e t r i c r e s u l t : Lemma 9. t h e form -: R~ ->

R2

i -s-a

congruence, then

o r, -

writing vectors

as column matrices,

where ad

- ( a , c ) and (b,d) are u n i t orthogonal vectors. - follows that -It

- M, -Jacobian t he
: R

determinant

of

h,

ec;uals +I.
&(O,O)

k -

Proof. ->
p2

Let

(p,q)

d e n o t e the p o i n t

Define

by t h e e q u a t i o n

I t i s e a s y t o c h e c k that

k -

is a congruence, s i n c e

f o r every p a i r of points
i

2,

which h a s t h e p r o p e r t y t h a t W f i r s t show t h a t e hypothesis,

k -

b. L e t u s &(g)= 0.

s t u d y t h e congruence

k,

p r e s e r v e s norms o f v e c t o r s :

By

Ilall = Ilk(a) -

- -OII

llk(a)ll.
By h y p o t h e s i s ,

Second, we show t h a t

k -

preserves d o t products:

Ilk(=)II2 -2k(a)-&(b) +
Because
k -

llk(b)l12 = 1la1l2

--

- - 2a.b- + llbl12.

p r e s e r v e s norms, w e must have

W e now show t h a t

k -

is a l i n e a r t r a n s f o r m a t i o n .
2 R ; then

Let

gl

and

g2 be the u s u a l u n i t b a s i s v e c t o r s f o r
~ e t

(x,y) =

xg1 + y e Z

e -3 = k (-e ) 1

and

+ e

= k (-2 ) . e k -

Then
1

e3

and

+ e

a r e a l s o u n i t orthogonal vectors, s i n c e Given

p r e s e r v e s d o t p r o d u c t s and norms.

x -=

(x,y), c o n s i d e r

the vector we have

k -(?I)

because

e3

and

g4

form a b a s i s f o r

R2 ,

f o r some s c a l a r s
L e t us compute

a a

and and

6,

which a r e o f c o u r s e f u n c t i o n s of W have e

x. -

B.

because

i s orthogonal

by d e f i n i t i o n o f

e3,

because

k -

preserves d o t products,

because

i s orthogonal t o

e2.

Similarly,

P(?i)

= k(2)

= -( -) *k(g2) 5 k x =

e2 = y.
R2,

W e conclude that f o r a l l p o i n t s

5 = ( x , ~ ) of

Letting

ej =

(a,c)

and

= (b,d) ,

w e can w r i t e

o u t in,

components i n the form

k -( 5 )

= x(a8c) + y(b,d) = (ax

by, c x

dy).

Thus

is a l i n e a r t r a n s f o r m a t i o n .
w e recall

h R e t u r n i n g now t o o u r o r i g i n a l t r a n s f o r m a t i o n , -,

that

k -( 5 )

= h(x)

-- -

(ptq).

T h e r e f o r e w e can w r i t e o u t

h -( 5 )

i n components a s

h -(

~ = (ax )

by

p, c x

dy

q).

TO

compute t h e J a c o b i a n d e t e r m i n a n t o f
= (a,c)

h -,

w e note t h a t because

e -3

and

= (b,d)

are u n i t o r t h o g o n a l v e c t o r s , w e have

the equation

d e t

cj= [,jtOr
a
b

det

1 0

Theorem 1 0 .

Let

self -, c a r r y i n g r e g i o n Green 's r e g i o n s , t h e n

i

--

b e a congruence of the p l a n e t o it region

S..

to -

7.

If bo -t h

--

-T

a nd -

a re -

area S = area T.

Proof.
a one-to-one

The t r a n s f o r m a t i o n c a r r i e s t h e boundary o f f a s h i o n o n t o t h e boundary o f a r e c a r r i e d by

S

in

(since d i s t i n c t
R2).

p o i n t s of

R'

t o d i s t i n c t points of

~ h u st h e h y p o t h e s e s o f t h e p r e c e d i n g theorem a r e s a t i s f i e d . Furthermore,

I J (u,v) I

= 1.

From t h e e q u a t i o n

w e conclude t h a t

area S = a r e a T.

EXERCISES.
I

1.

Let

h x -(x) = A -

be an a r b i t r a r y l i n e a r transformation
M,

of

R~

t o itself.

If S

i s a rectangle of a r e a

what i s t h e

a r e a o f t h e image o f
2.

under t h e t r a n s f o r m a t i o n

h?

Given t h e t r a n s f o r m a t i o n

\ .

(a)

Show t h a t i f
h ad

(a,c)

and

(bed)

are u n i t o r t h o g o n a l

v e c t o r s , then

i s a congruence.

(b) If

- bc

= 21,

show

h -

p r e s e r v e s areas.

Is

n e c e s s a r i l y a congruence?
3.
A translation of

R~

i s a t r a n s f o r m a t i o n o f t h e form

where
;I

is fixed.

r o t a t i o n of

R~

i s a t r a n s f o r m a t i o n of

t h e form

h -( 2 )

= ( x cos

y s i n 0,

x sin 4

y cos

$1,

where (a)

is fixed.

Check t h a t t h e t r a n s f o r m a t i o n
(r,9)

c a r r i e s the point

with polar coordinates

t o the point with polar coordinates

( r ,9+$)
(b) congruences. Show t h a t t r a n s l a t i o n s and r o t a t i o n s a r e Conversely, show t h a t every congruence w i t h J a c o b i a n

+I c a n be w r i t t e n a s t h e composite o f a t r a n s l a t i o n and a r o t a

tion.
i

(c)

Show t h a t e v e r y congruence w i t h Jacobian

-1

c a n be

w r i t t e n a s t h e composite o f a t r a n s l a t i o n , a r o t a t i o n , and t h e r e f l e c t i o n map

4.
A

Let

be a s q u a r e m a t r i x .

Show t h a t i f t h e rows of
A

a r e orthonormal v e c t o r s , t h e n t h e columns of

are also

orthonormal v e c t o r s .

i
r

3.

Let Given

b e the act o f all

y)

( x , ~ ) with

b2x2 f

a2 Y 2

I.

f(x,

expresa t h e integral u2 + v 2 1.

JJ
S

as an integral over

the u nit disc

Bvaluate when

f(x,y) = x 2, -

. .

*.

6.

I.

Let C be a circular cylinder o f radiua a whome central axis is the x-axis. Let D be a circular cylinder o f radius b 5 a w h o s e central axir is the z-axia. Bxpresa the volume conmoa to the two cylinders am an integral in cylindrical coordinates. L ~ v a l u a t e when b = Transform t h e integral in problem 3, p . 0.26 by using the substi tution x = u / v , y = uv w i t h u, v > 0 . Evaluate the integral.

a - w . 1

StokesI Theorem

Our text states and proves Stokes' Theorem in 12.11, but ituses the scalar form for writing both the line integral and the surface integral
involved. In the applications, it is the vector form of the theorem that is
most likely to be quoted, since the notations dxAdy =d the like are not

n in common use (yet) i physics and engineering.

Therefore we state and prove the vector form of the theorem here.
The proof is the same as in our text, but not as condensed.
Definition. Let F =
-. -I

PT + QT + $

be a continuously differentiable

vector field defined in an open set U of R3 field in U, by the equation curla = (Ra )/y

We define another vector

-adz)+ + i

(JP/>Z aR/Jx) -

We discuss later the physical meaning of this vector field.

An easy way to remember this definition is to introduce the symbolic

operator "delM, defined by the equation

m to note that d

curl F can be evaluated by computing the symbolic determinant 7' 7

curl ?

+ V ~ F=
-9

det

Theorem. (Stokes1 theorem).

Let S be a simple smooth parametrized

S

surface in R , parametrized by a function g : T

region in the (u,v) plane. Assume that T is a Green's region, bounded by a simple closed piecewise-smooth curve D, and that has continuous

+ (>Q/& -ap/Jy) 7 .

where T is a

second-order partial deri~tivesin an open set containing T ard D. Let C be the curve L(D).
If F is a continuously differentiable vector field defined in

an open set of R~

containing S and C, then

J

--t

( F a

Jc

T) ds

JJs ((curl T)-$] d~ .

fi
to the surface S points in the same

Here the orientation of C is that derived from the counterclockwise orientation of D; and the normal direction as 3 J ru
Y

1gJv

Remark 1. The relation between T and

is often described

informally as follows: "If youralkaround C in the direction specified by

with your head in the direction specified by ? i

then the surface S

is on your left." The figure indicates the correctness of this informal

description.
/

Remark 2 We note that the equation is consistent with a change

. of parametrization. Suppose that we reparametrize S by taking a function

9 : W -s T carrying a region in the (s,t) plane onto T, and use

the

new parametrization R(s,t) = g(q(s,t))

What happens t the integrals o

/

in the statement of the th?orem? If det Dg

, then the left side of

the equation is unchanged, for we know that g carries the countercloclcwise

orientation of because

2~
-

to the counterclockwise orientation of

2 T. Furthermore,

) R /as

x&/J t

( J dx 3rJk) det Dg , the unit normal

determined by the parametrization g is the same as that determined by g so the right side of the equation is also unchanged.

Orr the other hand, if det D g i 0,

then the counterclockwise orientation so that

of J W goes to the opposite direction on C,

changes

sign. But in that case, the unit normal determined by g is opposite to that determined by

r.

Thus both sides of the equation change sign.

Proof of --- the theorem.

The proof consists of verifying the
following three equations:

m e theorem follo&

by adding these equations together.

We shall in fact verify only the first equation. The others are

f proved similarly. Alternatively, i one makes the substitutions

> 4 -+ 3 -4 1 j and j + k m.d k -+i m d x--sy and y + z and z - e x , then each equation is transformed into the next one. This corresponds to 3 ar? orientation-presedng chmge of ~riables in R , so it leaves the orientations of C and S unchanged. Sa let F henceforth denote the vector field Pi ; we prove Stokes' theorem in that case.
-3

The idea of the proof is to express the line and surface integrals
of the theorem as integrals over D and TI respectively, and then to
apply Green's theorem to show they are equal.
Let r(uIv) = (X(U.V)~ Y(u,v), Z(u,v)), as usual.

l(t)
4

Choose a counterclockwise parametrization of D for a 5 t

call it

< b.

Then the function

is the parametrization of C that we need to compute the line integral

of our theorem.
k

We compute as follows:

Xv where 2 U u and a &

are evaluated at d_(t), of course. We can write

this as a line integral over D. Indeed, if we let p and q be the functions p(uIv) s(u,v)
= P(g(uIv) )*ru(ufv)
= P(g(u,v) 1 ) 8v r u $ -

ax

ax
d

then our integral is just the line integral

New by Green's theorem, this line integral equals

I*)

(J~/JU

J , / ~ v ) du dv

We use the chain rule to compute the integrand. We have

where P and its partials are evaluated at ~(u,v), of course. Subtracting, we see that the first and last terms cancel each other.
The double integral ( * ) then takes the form

Nclw we compute the surfac~ integral of our theorem.

Since

curl F = > P / ~ Z

T->P/>~ 2 , formula

(12.20) on p. 435 of our text

tells us we have

Here

2P/> z

and 2 P/3y

are e~luated t r(u,v) , of course. a

,

Our theorem is thus proved.

Exercises

on tfie diverqence

theorem

1.

Let

be the portion of the surface

z = 9

x2

y2

lying

above t h e xy plane. Let 8 be the unit upward normal to S.

Apply t h e divergence theorem'to the solid bounded by S and the xy-plane to evaluate (a)
(b)

J'I P
S

il
2

d~

if:

1 = ~in(~+z)? +

s X Z J + ( x +y

)d.

y 2 z 3 + yf + zR. Anewers: (a) 8 1 ~ / 2 *

D =

(b) 8ln.

2.
I

Let S2

S1

denote t h e surface
x2

z = 1
y2

Y 2 ; z 2 0.

Let

denote t h e unit d i s c let

z = 0

Let
S ,

= xt
and let

3,
be Evaluate

(2wcy)f + zk;

3,
S2,

b e t h e unit normal t o both with positive

t h e unit normal to y o d l d S and

S1

component.

Jj

* a 2 d s .

s2

Grad ' Curl Div and all that. -' We study two questions about these operations:

I .

- - -

DO they have natural (i.e., coordinate-free) physical

or geometric interpretations?

11.

What is the relation between them?

I.

We already have a natural interpretation of the

gradient.
For divergence. the question is answered in 12.20 of Apostol.
1

The theorem of that section gives a coordinate-free

-+

and the subsequent discussion

+ gives a physical interpretation, in the case where F is the
flux density vector of a moving fluid.

definition of divergence F.

Apostol treats curl rather more briefly.

+

Formula

(12.62) on p 461 gives a coordinate-free expression for

.
r i -

curl F (a), as follows: -L

where

C(r)

is the circle of radius r

centered at

a -

lying in the plane perpendicular to the point

and passing through

a,

and

C(r)

is directed in a counterclockwise

direction when viewed from the tip of

!

n.

This number is

called the circulation of

at a - -b

around the vector

n;

it is clearly independent of coordinates. coordinate-free definition of

+

Then one has a

curl F

as follows:

curl F

at

a -

points in the direction of the vector

around which the circulation of

is

a maximum, and its magnitude equals

this maximum circulation.

YOU

will note a strong analogy here with the relation between

-+

the gradient and the directional derivative.

For a physical interpretation of imagine F
I

curl F,

let us

to be the velocity vector field of a moving fluid.

in the fluid,

-+

Let us place a small paddle wheel of radius P with its axis along n.

Eventually, the paddle wheel settles

down to rotating steadily with angular speed

(considered as

positive if counterclockwise as biewed from the tip of if).

+ + The tangential component F * T of velocity will tend to in-
crease the speed

if it is positive and to decrease

if

i t is n e g a t i v e .
1

O p h y s i c a l grounds, i t i s r e a s o n a b l e t o n

suppose t h a t

average v a l u e of

( F - T ) = -

-b

speed of a p o i n t

on one of t h e p a d d l e s

That is,

&I
I t follows t h a t

C~
+

T ds = r w .

s o t h a t by formula

w e have ( i f
+

i s very s m a l l )

c u r l F ( a ) = 2w.

+ I n p h y s i c a l terms t h e n , t h e v e c t o r k u r l F(=)]

points i n the

d i r e c t i o n of t h e a x i s around which o u r p a d d l e wheel s p i n s

most r a p i d l y (in a c o u n t e r c l o c k w i s e d i r e c t i o n ) , and i t s magni

t u d e e q u a l s t w i c e this maximum a n g u l a r speed.

11.

What a r e t h e r e l a t i o n s between t h e o p e r a t i o n s Here i s one way of e x p l a i n i n g them.

grad, c u r l , and d i v ? Grad - goes goes from v e c t o r

from s c a l a r f i e l d s t o v e c t o r f i e l d s , C url

f i e l d s t o v e c t o r f i e l d s , and Div goes from

his i s s u m a r i z e d i n t h e

vector f i e l d s t o s c a l a r f i e l d s . diagram:

Scalar f i e l d s

0 (5)

Vector f i e l d s curl Vector f i e l d s

a
Jl

I ,(

Scalar f i e l d s

(XI grad
3

L e t us c o n s i d e r f i r s t t h e t o p two o p e r a t i o n s ,

and

curl.

W e r e s t r i c t o u r s e l v e s t o s c a l a r and v e c t o r f i e l d s t h a t U

a r e c o n t i n u o u s l y d i f f e r e n t i a b l e on a r e g i o n

of

Here i s a theorem w e have a l r e a d y proved: Theorem 1.

+
F

i -s-a

gradient i n

i f and --

only i f

-L

F * du

-=

for - e v e r y+c l o s e d

piecewise-smooth p a t h

in - U.

Theorem 2.

If

F = grad

--

f o r some

4,

then

+ + c u r l F = 0.

~ r o o f . W e compute

curl

by t h e formula

curl

w e know t h a t i f

$
1

i s a g r a d i e n t , and t h e p a r t i a l s o f
= D.F j J i

are

continuous, t h e n curlF=O.
-!

D.F

f o r all

i , j.

Hence

If i - c u r l = 8 -n-a- star-convex r e g i o n U , then f o r some in - $ = grad $ -- + d e f i n e d - U . The f unc- f u n c t i o n $(XI = + ( x ) + c i s t h e most g e n e r a l Theorem 3 .
_I-

t i o n such t h a t --Proof.
j.

F = grad $.

If

D.F = D.F I j I i i s star-convex, t h i s f a c t i m p l i e s t h a t

by t h e Poincar6 l e m m a .

If

+ +
c u r l F = 0,

then

for a l l
F

i,

is a

g r a d i e n t i n U, Theorem 4 .

17

The - condition

c u r l $ =

in -

u
3
i -s-a

does o - -n-t

i n g e n e r a l imply t h a t

g r a d i e n t - U. in

Proof.

Consider t h e v e c t d r f i e l d

It is defined i n t h e region

c o n s i s t i n g of a l l of curl

ex

cept f o r the
TO show

z-axis.

I t i s e a s y t o check t h a t
U,

8.

is not a gradient i n

we let

be t h e u n i t

circle

~ ( t=) (cos t , s i n t , 0 ) ;

i n the

xy-plane,

and compute

rt

f o l l o w s from Theorem Remark.

A region

] I

that
U

$
R~

cannot be a g r a d i e n t i n
i s c a l l e d "simply
U

U.

in

connected" i f , roughly speaking, e v e r y c l o s e d curve i n bounds an o r i e n t a b l e s u r f a c e l y i n g i n

3
U.

The r e g i o n

R3

( o r i g i n ) i s simply connected, f o r example, b u t t h e r e g i o n

R

- (z-axis)
8

i s not.
U

rt t u r n s o u t t h a t i f
,

i s simply connected and i f

U.

curl $ =

in

U,

then

is a gradient i n

The proof

goes roughly a s follows: Given a c l o s e d c u r v e able surface i n t o t h a t surface.

U

in

U,

let

be an o r i e n t -

which

bounds.

Apply S t o k e s t theorem

One o b t a i n s t h e e q u a t i o n

Then Theorem 1 shows t h a t

NOW

i s a gradient i n

U.

l e t us consider t h e next t w o operations,

curl

and

Again, w e c o n s i d e r o n l y f i e l d s t h a t a r e c o n t i n u o u s l y 3 d i f f e r e n t i a b l e i n a region U of R There a r e analogues of div.

a l l t h e e a r l i e r theorems:
I

Theorem 5 .

11 5 t
S

dS = 0

If i then - F -s- a- c u r l i n U, -
for closed - e v e r y o r i e n t a b l e -s u r f a c e
S

in -

U.

~roof. Let

be a c lased s u r f a c e t h a t l i e s i n (While we assume t h a t in

U
U,
S

U.
lies

w e do n o t assume t h a t

includes t h e
S

)-dimensional bounds.) Break and

region t h a t

s up i n t o two s u r f a c e s S1
S2 that intersect in their

common boundary, which i s a simple smooth c l o s e d curve C. Now + by h y p o t h e s i s , G = c u r l ? f o r some d e f i n e d i n U. W e corn pute :

n ds =

11

S2 Adding, w e see t h a t
,

curl s2 .

F m i d~ i

1 Hm
C

da.

ema ark.

The convers'e of Theorem 5 h o l d s a l s o , b u t w e then - d i v 2 = 0.

s h a l l n o t a t t e m p t t o prove it. Theorem 6 . P roof. If -

= curl

f o r some --

?,

By assumption,

Then

div

-+

G =

(D1D2F3-D1D 3F 21 - (D2D1F3-DZD3Fl)+(D3DlF2-D3D2F1)

If in - div = 0 - -a- star-convex region U, + then for some in - 2 = curl 3 -- F defined - U. + + The is the most func- function H = F + grad $ --- general + tion such that --- = curl H. Theorem 7. We shall not prove this theorem in full generality. proof is by direct computation, as in the ~oincar6lemma. A proof that holds when U is a 3-dimensional box, or U is all of R3, is given in section 12.16 of Apostol. The

when

This proof also shows how to construct a specific such function in the given cases.
Note that if curl(H-F) = for some
+ +

= curl ?$ and

-+

G = curl H,

-+

then in U,

8.

Hence by Theorem 3, The - condition

= grad 4

4.

Theorem 8 .
-t

divG=O

& U I
is - - if - -a curl in
U.

---

does not in general imply that

Proof.

Let

be t h e v e c t o r f i e l d

which i s d e f i n e d i n t h e r e g i o n except f o r t h e o r i g i n . that d i v G = 0.

c o n s i s t i n g of a l l of

R3

One r e a d i l y shows by d i r e c t computation

If that

i s t h e u n i t s p h e r e c e n t e r e d a t t h e o r i g i n , t h e n w e show

his w i l l imply (by Theorem 5 ) t h a t

If

is n o t a c u r l .

I (x,y,z)B = 1,

so

(x,y,z) i s a p o i n t o f S t t h e n + + + + * G(x,y,z) = xi + yj + zk = n o T h e r e f o r e

11 b*;

dA =

1 dA = ( a r e a of s p h e r e ) # 0.

erna ark.

Suppose w e s a y t h a t a r e g i o n

U U

in

R3

i s "two-

simply c o n n e c t e d w i f e v e r y c l o s e d s u r f a c e i n s o l i d region l y i n g i n
U.*

bounds a

The r e g i o n

U = R -(origin)

i s nott'two
axis)

simply connected", f o r example, b u t t h e r e g i o n

U = R3-(z

is -.

rt
div G = 0
-b

turns out that i f in

U,

U is"two-simply connected and i f

I1

then

is a curl in

U.

The proof goes

roughly a s follows:
Given a c l o s e d s u r f a c e
it bounds.
S

in

U,

let

be t h e r e g i o n
V,

Since

i s by h y p o t h e s i s d e f i n e d on a l l of

w e can apply Gauss' theorem t o compute

Then t h e converse of Theorem 5 implies. t h a t

+
G

is a c u r l i n

U.

There i s much more one can s a y a b o u t t h e s e matters, b u t one needs t o i n t r o d u c e a b i t of a l g e b r a i c topology i n o r d e r t o do so.
It i s a b i t l a t e i n t h e semester f o r t h a t !

*The proper mathematical term for t h i s is ffharalogicallyt r i v i a l in dimension two."

MIT OpenCourseWare http://ocw.mit.edu

18.024 Multivariable Calculus with Theory

Spring 2011

For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.