This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
, is defined as |P| |Q| cos 9, where 9 is the angle between P and Q. In coordinates, P (xu yu Zj), Q (x2 y2 z2), =
=
,
(P,Q}
Propositions i/
o.
=
,
x1x2 + y^2 + zxz2
26.
The
nonzero vectors
P and
Q
are
perpendicular if and only
=
Proof. P and Q are perpendicular if and only if the angle fl between them is a right angle, fl is a right angle if and only if cos fl 0, and this holds precisely 0. When
=
=
A
plane through
vector
the
perpendicular
origin
to such a
n=<x.N>=o
is the linear span of two vectors. If N is plane n> then 11 is iven by the ecluat'on
a
98
1.
More
Linear Functions
generally, if p is a point on a plane (not necessarily through orthogonal to FJ, F] is given by the equation
the
origin)
and N is
<x
-
N>
p,
0
=
through the origin is the linear span of a single vector, and can be expressed by two linear equations (since a line is the intersection of two planes). A line
Examples 39. Find the
equation of the plane through (1, 2, 0) spanned by the (1, 0, 1), (3, 1, 2). If N (n1, n2, n3) perpendicular to
two vectors
this
plane
=
must have
we
(1
=
of
1,
,
=
=
solution
N
n1+3 3k1 +n2
=
this
0
2n3=0
+
is (1,-1,-1), so equation of the plane is
system
Then the
1).
<x- (1,2,0), (1, -1, -1)>
=
0
or
P
-
Letting
R
N
=
0
=
(m1, n2, n3),
=
take
equation of the plane through P (1, 0, 1), Q (3, 1, 1). If N is perpendicular to the plane, we have
=
R>
may
x-y-z+l=0
40. Find the
(2, 2, 2),
we
-
we
R>
=
-
=
0
obtain this system of
equations:
-2n1-n2-2n3=0 n1
-
+
n2
which has
<x
-
Ix +
N, R>
4y
+
n3 =0
a
solution N
=
+ 3z
0,
=
41. Find the
or
l(x
=
-
(7, 4. 3). 3) + 4(y
Thus the -
1)
+
3(z
-
equation we 1) 0 or
28
equations of the line L through (4, 0, 0) and perpen Example 40. If x is on L we must have
dicular to the plane in
<x
-
(4, 0, 0),
P
seek is
=
-
R>
=
0
<x
-
(4, 0, 0), Q
-
R>
=
0
1.9 may take these
so we
2x + y + 2z + y +
-x
z
as
the
99
Space Geometry
equations :
8
=
-4
=
Vector Product Given two noncollinear vectors vx, v2 in space, the set of vectors perpen a line. We shall now develop a useful formula for select
dicular to v1; v2 is
ing
a
If N is
vector on that
particular on
that line, and
<x, N>
=
x
for all
x e
Vl
uniquely
a
R3. =
vector ,
product
we
vt
have
0
On the other hand, since x, vx, v2
Now there is
line, called the
is in the linear span of vt and v2
This is
are
coplanar we
have
determined vector N such that
easily
(V, vt2, vt3),
seen
y2
=
using coordinates.
(v2\ v22, v23),
x
=
Write
(x1, x2, x3)
Then
det(
Vj
)
=
W =
2
vt3 v^
.,2
.,3
vx2 22
detj V 1
V2/
x3
x2
Ix1
(x \
1
XKV.W V3V22) *W23 lW) + x3(v11v22-vl2v21) (ix1, X2, X3), ((Vl2V23 1-! V), (V.W V^W), -
~
-
-
-
=
(v.W-v.W))}
x
v2
.
100
1.
Linear Functions
Definition 16.
The vector
V x w
Proposition
(i)
<x,
(ii) (iii) (iv)
Let
product =
v
=
v x w
(v2w3
(v1, v2, v3), is defined
v3w2, v3w1
-
w
(w1, w2, w3)
=
be two vectors in R3.
by
-
v1w3, v1w2
-
v2wx)
27.
v x
v x w
w>
det
=
vl
forallxeR3.
w x v.
=
orthogonal to v and w. The equation of the plane through the equation <x, v x w> 0. is
v x w
the origin
spanned by
v
and
w
has
=
The
proof
discussion. for
example,
of this proposition is completely contained in the preceding The basic property of the vector product is the first; it follows, that for any three vectors u, v, w
w>
v x
Notice that if v, ordered basis u
proposition gives of
=
x
v,
w>
=
collinear,
w x
u>
=
x
w,
u>
0. If they are not collinear, the right handed (see Figure 1.19). The following important geometric interpretation of the magnitude
w are
v x w
=
vxwis
v
an
v x w.
Proposition
(i)
The
28.
area
Let u, v,
w
be three noncollinear vectors.
of the parallelogram spanned by
wx v
Figure
1.19
u
and
v
is
||u
x
v||.
1.9
(ii)
The volume of the
|detQ
parallelepiped spanned by
J
v
This follows
u, v, and
w
101
is
'
Proof. Step (a). The first step is to verify (ii) in case perpendicular. In that case we must show that
det/
Space Geometry
||u||
=
the vectors u, v,
w are
mutually
||v||- ||w||
-
easily from the multiplicative property of the determinant.
First
we
note that
/u\
/ (u,
v
v,
w)
=
\w/
\
0
0
0
0
0
0
\
<w,w>/
(i,j)th entry is the inner product of the rth of the second (see Problem 49). Thus,
since the
;th
row
detjv j =detjv )(u, Step (b). In particular, perpendicular, so
v,
if u,
w)
=
v are
llu
so
||u
x
v||
=
||u||
x
of the first matrix with the
||u||2||v||2||w||2
perpendicular,
(u =
row
x
v\
u
J
then u, v,
u x v are
mutually
||u||- ||v||
vll
||v||, when
is
perpendicular part (i) in general. u
to
v.
angle between prove the of by u and area the spanned Then parallelogram 1.20). (see Figure is the product of the base and the height of the base: Step (c).
and
Now
we
v
area
=
||u||a
=
||u||
||v||sinfl
Let fl be the the
u
v
102
/.
Linear Functions
Now the vector
by
u
and
u
=cosg-fl)
sin
Since
(u x v) is orthogonal to u and orthogonal to u. We have
u x
and is
v
and
u x
HTll
l|U
(u X
X
orthogonal, by Step (b)
u x v are
v,
so
lies in the plane spanned
v)>
x
(U
u x
T)||
we
have ||u
x
(u
x
v) ||
=
||u ||
||u
x
v||
Thus
area
=
||u
Nu
||r|| =
(u
u x
X V, U X
v>
l|u|iVrrruxy ^-
v)>
x
(u
x
x
v)||
lluxv||
=
u
The volume of the parallele Step (d). To prove part (ii) we refer to Figure 1 .21 piped spanned by u, v, w is the product of the area of the base and the altitude: .
volume
Since
u X t
is
,
sin
=
||u
x
y\\b
orthogonal
<6=C0S
(tt -
\2
=
||u
x
to the u,
\
4>\
7
||w||sin <j>
v|| v
|<w,
=--!
plane,
u> xv>| _
llwll-Hu xvll
u
X(u
Figure 1.20
X
v)
1.9
Space Geometry
x V
Figure
1.21
Thus
..
volume
A final
=
|lu
equality
x
v|
<w,uxv>
l|w||-n-T7] w uxt
-=
det
/w\ u
/
which will prove useful is this :
||uxv||2=||u||2||v||2-2 This follows
easily
from the above arguments :
||uxv||2=||u||2||v||2sin24> ||u||2||v||2(l-cos20) =
=
since the
angle
llu||2|
between
2 u
and
v
is (b.
EXERCISES 47. Which
vx=(2, 1,2), t5= (1,3,0),
pairs of the following
vectors
are
orthogonal?
t,=(3, -1,4), v3=(7,0,5), v6= (0,0,1), tt=(-15,5,21)
v*
=
(6, -2,5)
104
1.
Linear Functions
48. Find the vector
products
v, x V/
for all
pairs of
vectors
given in
Exercise 47. 49. Find
vector
a
where vi, v2 v3 50. Find the
such that =
-l
given in Exercise 47. equation of the plane spanned by the (c) t5, v6, (d) v2, t4, where the v,
,
(b)
v
v2, v5,
are
(a) Vi, y6 given in Exer
vectors are
,
cise 47. 51. Find the
of the line
equation
spanned by the
vectors
given in
Exercise 47. 52. Find the equation of the plane
through (3, 2, 1) and orthogonal (7, 1, 2). 53. Find the equation of the line through (0, 2, 0) and orthogonal to the plane spanned by (1, -1, 1) and (0, 3, 1). 54. Find the equation of the line through the origin and perpendicular to the plane through the points (a) E,,E2,E3 (b) (1,1, 5), (0,0, 2), (-1,-1,0) (c) (0, 0, 0), (0, 0, 1), (0, 1, 0) 55. Find the equation of the line of intersection of the two planes, (a) determined by (a), (b) of Exercise 54. (b) determined by (a), (c) of Exercise 54. (c) determined by (b), (c) of Exercise 54. 56. Find the plane of vectors perpendicular to each of the lines deter to the vector
mined in Exercise 55. 57. Let A be
(a)
if the
tions of Ax a
a
3
=
3 matrix.
x
rows
of A lie
b forms
a
line,
(b) if the rows of A lie plane, or is empty.
58. Show that ||v x w|| the two vectors v and w. 59. Is the vector product
=
(u
x
v)
x w
=
u x
(v
x
Show that
on a or
on a
||v||
plane (but
not on a line), the set of solu is empty. line, the set of solutions of Ax b forms =
||w|| sin fl, where fl is the angle between
associative; that is, is
w)
always true? 60. If v, v, v x w,
w are
v x
(v
X
two noncollinear vectors show that the three vectors
w)
are
pairwise orthogonal.
PROBLEMS 48. Prove the identities of Proposition 27. 49. Let vj, v2, v3 be three vectors in R3.
Let A be the matrix whose
1.10 rows are
Vi, v2
(a) the (i,/)th entry (b) det
A
50. Let P be
=
of Linearity
and B the matrix with columns Vi, of AB is
v3
,
Abstract Notions
v2
v3
,
.
105
Show that
,
det B.
given by
the coordinates
(x, y, z) relative to a choice Ei, E2 Show that the point of intersection of the line through P parallel to the Ei axis with the 2-3 plane has coordinates (0, y, z). ,
E3 of basis for space.
Abstract Notions of
1.10
There with
a
are
Linearity
many collections of mathematical objects which are endowed algebraic structure which is very reminiscent of R". To be less
natural
vague, there is defined, within these collections, the operations of addition and multiplication by real numbers. Furthermore, the problems that
naturally arise in these other contexts are reminiscent of the problems on R" which we have been studying. The question to ask then, is this: does the same theory hold, and will the same techniques work in this more general We shall see in this section that for a large class of such objects context ? (the finite-dimensional vector spaces) the theory is the same. We shall see later on that in many other cases, the techniques we have developed can be modified to provide solutions to problems in the more general context. First, let us consider some examples. Examples 42.
[0, 1],
If/ and then
g
are
we can
continuous real-valued functions on the interval cf as follows:
define the functions /+ g,
(f+g)(x)=f(x) + g(x) (cf)(x) cf(x) =
Clearly,/ +
g and
c/are
of addition and scalar
C([0, 1])
also continuous.
multiplication
of all continuous functions
Thus
are
on
that
we see
defined
on
the interval
operations
the collection
[0, 1].
example, if/ and g are differentiable, so are/+ g and cf. Thus the space ^([O, 1]) of functions on the interval [0, 1] with continuous derivatives also has the operations of addition and scalar multiplication. Notice that the operation of differentiation takes functions in CHC0, 1]) into C([0, 1]): if /is in CH[0, 1]) it has a continuous derivative, so /' is in C([0, 1]). Furthermore, 43. In the above
106
Linear Functions
/.
differentiation could be described
as a
linear transformation:
if+g)'=f'+3' (cf)' cf =
is, by the
So
way,
integration
a
linear transformation:
\if+g)=lf+U l(cf)
ctf
=
The fundamental theorem of calculus says that differentiation is the inverse operation for integration:
(J/)' =/ merely a curious way of describing the implied point of view has led to a wide The subject of functional analysis which range of mathematical discoveries. 20th in the was developed early century came out of this geometric-algebraic to long standing problems of analysis. approach These remarks may strike you well-known phenomena, but the
as
Examples 44. If S and T
are
linear transformations of R" to Rm, then
so
is
the function S + T defined by:
(S
+
We
T)(x)
also
can
(cS)(x)
=
=
S(x)
+
T(x)
multiply
a
linear transformation
by
a
scalar:
cS(x)
Thus the space L(R", Rm) of linear transformations of R" to Rm has defined on it operations of addition and scalar multiplication.
45. We have
of
n x n
In
fact,
we
used, in
M" this way it
These
already observed (Section 1.6) that the collection M" on it these two important operations.
matrices has defined
was
an
just
essential way, the fact that when the
same as
examples, together with R", lead
we
viewed
R"2.
to the notion of
an
space : a set together with the operations of addition and scalar We include in the definition the algebraic laws governing these
abstract vector
multiplication. operations.
1.10
Definition 17.
Abstract Notions of Linearity
107
An abstract vector space is a set V with a distinguished on which are defined two operations: and vv are elements of V, then v + vv is a well-defined element
element, 0, called the origin, Addition.
If v
of V. Scalar
multiplication. If v is in V and c is a real number, cv is a well-defined These operations must behave in accordance with these laws :
element of V.
(i) (ii) (iii) (iv) (v) (vi) The of the
v
+
(vv
v
+
vv
t>
+ 0
+ =
=
c(v w) ct(c2 vv) +
\w
x)
(v
=
vv)
+
+ x,
+ v,
vv
v, cv
=
=
+ cw,
(Ci c2)w,
vv.
=
preceding examples are all abstract required laws are easily performed.
vector spaces; the verifications
We
now
want to
investigate
the
extent to which the ideas and facts discussed in the case of R" carry over to abstract vector spaces. First of all, all the definitions carry over sensibly to the abstract
case
vector space V. case :
if
Thus
we
just replace
we
the word R"
take these notions
as
by
the words
an
abstract
defined also in the abstract
linear transformation, linear subspace, span, independent, basis, dimension. one bit of amplification necessary in the case of dimension.
Now there is We have until
now
encountered spaces of only finite dimension.
Example 46. Let R an
be the collection of all sequences of real numbers. an ordered oo-tuple,
Thus
element of I?00 is
(x1,*2,. ..,*",...) R00 is
an
abstract vector space with these
operations:
(x1,x2,...,x,...) + (y1,y2,...,yn,...) x" + y", .) (x1 + y1, x2 + y2, c(x\ x2,...,xn,...) (ex1, ex2, ...,cx",...) =
.
.
.
,
.
.
=
be the has an infinite set of independent vectors. Let This are zero but for the nth, which is 1 entries whose all of sequence entire collection {Eu ...,,.. .} is an independent set. For if there is a relation among some finite subset of these, it must be of the Now R
.
form
c1^
+
+
ckEk
=
0
108
Linear Functions
1.
(of course,
many of the c's may be
c1El
+
so
+
ckEk
if this vector is
indeed the set We
=
must have
{Eu ...,,...}
make the
now
(c1, c2,..., ck, 0, 0,
zero we
following shall
vector space; and
we
holds also in this
more
is
an
general
.
.
c1
.) =
c2
infinite
=
=
ck
independent
=
0.
Thus
set on/?.
restriction to the so-called finite-dimensional
that all of the
see
But
zero).
preceding
information about R"
case.
Definition 18. A vector space V is finite dimensional if there is a finite set of vectors vl,...,vk which span V. That Rx is not finite dimensional follows from some of the observations to be made below. It can also be
verified in the terms of the above definition
(see Problem 53). The important they are no different from
result about finite-dimensional vector spaces is that the spaces R".
Proposition
Let V be
29.
a
finite-dimensional
basis for V, every vector in V
Ifvt,...,vdisa combination of vu v
=
x1v1
(x1, ...,xd) The
.
.
.
,
+
xdvd
is called the coordinate of
correspondence
be
vd:
+
v
of dimension d. expressed uniquely as a
vector space
can
(x1,
.
..,
xd)
is
v
a
relative to the basis vlt...,vd. one-to-one linear transformation
of V onto R*.
Proof.
The definition of basis
(Definition 6) makes this proposition quite clear. (Problem 54).
We leave the verifications to the reader
What is not
so clear is that every finite-dimensional vector space has a and that basis, every basis has the same number of elements. However, once these facts are established the above proposition serves to reduce the
general finite-dimensional space 1.3 through 1.6 carry over. Proposition 30.
to one of the
R", and the results of Section
Every finite-dimensional vector space V has a finite basis, same number of elements, the dimension of V.
and every basis has the
Proof. Suppose V is finite dimensional. Then V has a finite spanning set. Let {vi, vd} be a spanning set with the minimal number of vectors; by definition V has dimension d. We shall show that {vi, vd} is a basis. ...,
...,
Abstract Notions of Linearity
1.10 Since {vu
vd}
span, every vector in V can be written as a linear combination We have to show that there is only one way in which this can be
...,
of these vectors. done.
Suppose for
some
x'Vi H
=
109
vector
h x*vd
=
have two different such ways :
v we
yhi
+ +
y"vd
(1 .48)
Then
(x>
y>i +
-
Vj-i, fJ+i,
equation .
.
.
yd)vd
.
.
.
=
these elements
y' for
some
Thus
j.
Thus it must be
d.
That any two bases have the
same
Hence
{vL,
.
.
.
...,
But this contra
to span all of V also.
serve
in two different ways.
vd
,
0
must have xJ #
we
assumption about
dicts the minimal in terms of Vi,
-
says that vj is in the linear span of the a- 1 elements vit
so
vd,
,
(xd
expressions differ
Since these two
Now this
+
impossible to express vd} is a basis.
v
,
number of elements follows
easily
from
Proposition 28 (see also Problem 55). Let T: V^>Rd be the linear transfor mation associating to each vector its coordinate relative to the above basis {vu...,vd}. If {wu...,wd} is another basis, let S: K1R be the same S T~ is a one-to-one coordinate mapping relative to this basis. Then L 0. Thus (rank + nullity linear mapping of Rd onto R3, so p(L) 5, v(L) =
dimension) :
5
=
=
=
=
d.
PROBLEMS
a
{vi, ...,vk] in 51. Show that for any finite set of vectors S vector weR which does not lie in their linear span [S].
v'
represent the first (k + l)-tuple of entries in
R", there is
=
span
there is
Rk+1,
a
.
.
Since v/,
.
vector w' in Rk+l which cannot be written
nation of v/, vt'. Let 52. Are the vectors E,, .
v.
,
.
w .
.
=
,
(w', 0,
E
,
.
.
.
.
.
.
(Hint: .
v/
,
Let
cannot
combin-
as a
.).)
in R described in
Example
43
a
basis
for Ra ? 53. Let Ro" be the collection of those sequences of real numbers x", .) such that x" 0 for all but finitely many n. Then R0^ (x1, x2, are a E is a linear subspace of R". Show that the vectors Ei =
.
.
.
.
,
.
,
.
.
.
,
,
.
.
.
basis for i?0. 54. Prove
of
a
Proposition
29.
that any two bases by following the arguments in Section 1.4, finite-dimensional vector space have the same number of elements.
55. Prove,
110
1.
Linear Functions Show that the collection L(V,
56. Let V, W be two vector spaces. linear transformations from V to W is
a
W) of
vector space under the two opera
tions :
(a) (b)
if
e L( V, W), (cL)(x) cL(x), L(V, W), (L + L')(x) L(x) + L'(x).
R, L
c g
if L, L'
6
=
=
57. What is the dimension of L(R",
58. Show that
a
Rm)l
vector space V is finite dimensional if there is
a
one-to-one
linear transformation of F0 into J?" for some n. 59. Show that a vector space F is finite dimensional if there is
a
linear
transformation T of R" onto V for some n. 60. Verify that the collection P of polynomials is an abstract vector space. For a positive integer n, let P be the collection of polynomials of degree not Show that P is a linear subspace of P. Show that P is not more than n. finite dimensional, whereas P is. What is the dimension of P1 61 Let x0 c another collec x be distinct real numbers and c0 .
,
.
.
.
,
,
Show that there is
tion of real numbers.
one
.
.
.
and only
,
one
polynomial p in
P such that
p(xt)
=
(Hint:
ct
0
Let L:P
that L has rank 62. Let g be
G(p)
n
J?"+1 be defined
(p(x0)
a
polynomial, and define the function
a
linear function.
63. Define Dk:P^-P:
of A? 64. Let x0 and
only
\
p(x0) =d0
(Hint:
e
Dk(p)
R, and let
one
=
p(x)).
G: P
Show
P:
Describe the range and kernel of G. What are the range and kernel
dfy/dx!'.
c0 ck be given numbers. polynomial p in P such that ,
dp
.
.
,
Use the
same
Show that there is
dkp
d~t(-X)=Cl'""dxk ^ idea
as
=
C"
in Exercise
65. Does Dk:P-+P have any 66. Show that C([0, 1]) is not
1.11
=
=pg
Show that G is
one
by L(p)
+ 1 .)
61.) eigenvalues? a
finite-dimensional vector space.
Inner Products
The notion of
length, or distance, is important in the geometric study of spatial configurations. In Section 1.3 we studied these concepts and related them to an algebraic concept, the inner product. From the point of view of analysis also it is true that these concepts are significant: planar
and
1.11
it is in terms of distance that
"convergence."
(vv1,
.
.
.
By analogy
express "closeness" and in particular with R3 we define the inner product in R",
,
vv"),
product of by
We shall say that v is orthogonal to between v and w is defined by
\y\
=
v
than the
v
=
(v1,
...,
v"),
w
=
as
w
if
w>
=
The distance
0.
d(v, w)
v
and 0,
d(y,0)=V(vt)2-l1<2
d(y, w)2
two vectors
is the distance between
Distance in R" behaves much Pythagorean theorem holds:
when
shall, in this section,
we
{X(vi-wi)2-\1'2
=
|v| of a vector
The modulus
here
v'w'
=
d(y,yv)
we are
The inner
denoted
111
we can
and in terms of it, distance. While introduce some topological terms. Definition 19.
Inner Products
w, sum
=
d(y, x)2
+
as
it does in R2 and
d(y, x)
+
points
are no
further apart
(1 .50)
d(x,yv)
These facts will be verified in the
the
particular,
(1 .49)
=
<
in
d(x, w)2
w x> 0. In any event, two of the distances from a third,
d(v, w)
R3;
problems.
Topological Notions The ball in R" of radius R > 0 and center c, denoted B(c, points whose distance from c is less than R:
Definition 20.
is the set of all
R),
B(c,R)= {xeR":d(x,c)
Thus,
a
c.
a
neighborhood
of
a
point
c
if it contains
A set V is said to be open if it contains
a
points. set S is
a
neighborhood
of
c
if there is
some
R
some
ball
neighborhood
(presumably
of
very
112
Linear Functions
1.
small)
such that
d(x, c)
R
<
implies
x e
S
A set U is open if for every c e U, there is an R such that U => B(c, R). For suppose xeB(c,R). Then d(x, c) that any ball is open. R d(x, c) > 0. Now B(c, R) contains the ball of radius R
Notice < -
-
centered at
For if y is
x.
d(y, c) Here is
a
d(y, x)
^
+
point in
a
that ball, then
d(x, c)
collection of formal
properties
+
R,
so
d(x, c)
by (1 .50),
d(y, c)
=
R
of the collection of open sets.
Proposition 31. (i) R" is open. n U. (ii) IfUu..., U are open, so is Ut n (iii) // C is any collection of open sets, then the set of all points belonging to any of the sets in C is open. (This set is denoted [j U). Proof. (i) Clearly, R" contains a ball centered at every one of its points. Then there are Ri (ii) Suppose Ui,..., / are open, and x is in every U, U => B(x, if). Let R R such that Ui => B(x, Ri), min[.Ri, Rn]. Then if In is in each Thus is in n n U is in each so Ut Ui y B(x, Ri) d(y, x)
=
.
.
.
.
,
.
.
,
.
.
Many of the concepts a mathematician studies are so-called local concepts: They happen in a neighborhood of a point, or are determined by what goes on near a point; far behavior being irrelevant. Differentiation is thus local, whereas
integration is
The
importance of open sets is that it is precisely study these local concepts, since their definition at a point depends on behavior in some neighborhood of the point. If a set is open its complement, the set of all points not in the given set, is said to be closed. Thus, S is a closed subset of R" if R" S {x e R" : x S} is open. Corresponding to Proposition 31 we have this proposition about on
such sets that
we
not.
should
-
closed sets.
Proposition 32.
(i) R" is closed. (ii) If Su Sn .
.
.
,
are
closed,
so
is
S^v
u
S.
=
1.11
Inner Products
(iii) IfC is a collection of closed sets, then the set of all points the sets of C is closed. (This set is denoted [)s c S).
113
common
to
all
6
Problem 67.
Proof.
Notice that there
are
sets which are both open and closed.
There
are
R" and 0 are the only ones. There are also sets which many of them. neither open nor closed, and there are many of them. For example,
not are
an
interval is open in R1 if it contains neither end point, closed if it contains both, and neither open nor closed if it contains only one end point. We
are
acquainted
That
with the notion of
"dropping
a
perpendicular"
in the
line and p is a point not on the line, then we can drop from p to / as in Figure 1 .22. The point p0 of intersection
is, if / is
plane. a perpendicular of the perpendicular with / is the point on / which is closest to p. A more sophisticated way of describing this situation is to say that p0 is the orthogonal projection of p on /. The concept of orthogonal projection generalizes to R" and will prove quite useful there. In order to discuss this problem, we shall generalize even further. a
A Euclidean vector space is an abstract vector space V on which is defined a real-valued function of pairs of vectors, called the inner product, and denoted <, >. The inner product must obey these laws: Definition 21.
(i) (ii) (iii) (iv)
(v, v} > 0. If (v, v}
=
0, then
v
=
0.
=
=
<>!
+ v2
,
w>
=
!, w>
+
\ \
*p \ \ \ \
\ \ \
\
\ \ \
^
\ \
Figure 1.22
114
1.
Linear Functions
Euclidean vector space when endowed with its inner C[0, 1 ] of continuous functions on the unit interval is a
It is clear that R" is The space
product.
a
Euclidean vector space with this inner
, 9>
=
(f(t)g(t)
product:
dt
We leave it to the reader to
verify that the laws (i)-(iv) are obeyed. It is (i)-(iv) are all that is essential to the notion of inner interesting that such function behaving in accordance with those laws ; is, any product will have all the properties of an inner product. Despite the inherent interest in this metamathematical point, we shall not pursue it further, but take it for granted that the above definition has indeed abstracted the essence of that the laws
"
"
this notion. In terms of an inner product length and orthogonality:
llll v
1
on a
vector space
we can
define the notions of
[<, t>>V'2
=
if and
vv
only
if
=
0
The
important bases in a Euclidean vector space are those bases whose are mutually orthogonal. More specifically, we shall call a set {Elt ...,} in a Euclidean vector space V an orthonormal set if
vectors
||,|| Et
1
for all
1
=
i
for all i # /
Ej
If the vectors
Eu E span V we shall call them an orthonormal basis. (Any orthonormal set of vectors is independent Problem 68.) The basic geometric fact concerning orthonormal sets is the following:
Proposition
...,
33.
Let V be
orthonormal set in V. is
orthogonal
to
a
Euclidean vector space and v in V, the vector v v0
For any vector
the linear span
=
-
{l5 ...,}
?=
Sof{E1,...,En}.
n
Proof.
Let
w
=
i
=
2 Ci E, =
-
be in S.
Then
i
2
'=i
=
iv, w>
-
J
i=i
an
1.11
Inner Products
115
Now
< w>
<,
=
,
2 CjEj>
,
=
,
=
c,
n
2 ctE,y
<, ,>
j=i
ft
2 c./
=
j=i
2
=
c,
1=1
1=1
Thus
=
1
2
Theorem 1.8.
be
an
c,
-
1
=
1
Let V be
orthonormal set in V.
v0=
2 =
a
c,
=
o
1
Euclidean vector space, and let
{Eu ...,}
For any vector v, let
<<;,,>,
i=l
Then
\\v\\2=\\v-v0\\2+\\vQ\\2;
(i) (ii)
for any
vv
in the linear span of
{Eu ...,},
\\v-v0\\2<\\v-yv\\2 Proof.
(i)
II" II2
=
<(y v0) + v0,(v- v0) + v0> \\v-v0\\2+ \M\2+ <.V0,V-Vo> +
The last two terms span of
{Ei,
-
=
,
=
...,
(ii) ||y w||2
are zero
by the preceding proposition, since
=
=
vv,
vv>
v
wy Vo + v0 w,v foil2 + \\v0-w\\2+
+
v0
Again, the last two terms are zero for both Thus span of {,,...,}.
v0
,
w
and thus also
||t,_vv||2= ||_0||* + ||t;o-w||2> Hf-Uoll2 (ii) is
proven.
is in the linear
E}.
=
so
v0
v0
w
is in the linear
116
1.
Linear Functions
Gram-Schmidt Process v' is orthogonal to the linear span S of {Eu ..., }. v0 v0 is the vector in S which is closest to v; it is called the orthogonal projection of v into S. It seems, by Theorem 1.8 that one needs an orthonormal basis Notice that
v
=
orthogonal projections; the following proposition gives a obtaining orthonormal basis for finite-dimensional vector thus with it, orthogonal projections.
in order to find
for
procedure spaces, and
Proposition 34. Let Fu ...,F be a basis for a Euclidean can find an orthonormal basis Eu ...,Enso that the linear Fjfor allj. Ej is the same as the linear span ofF1, We
.
.
The
Proof. NPiir'pi. Now in
proof is by induction
general, let Fi,
.
.
.
the linear span W of P\, apply the proposition to W .
.
orthonormal basis with the
F be
,
.
a
on
ofEu
...,
.,
n.
If
basis for
F-L is
vector space V.
span
a
n
=
1,
we
need
only take Ei
Euclidean vector space V.
=
Then
Euclidean vector space also, and we can the inductive hypothesis. Let Eu ..., E-i be an
,
a
by required properties.
Now,
we must
find
a
vector
En
such that 1 11^11 (E,E,)=0 =
all/^K Fn is in the linear span of E, If E is
Thus
a
we
.
.
.
,
E
vector that fulfills the last two
need
En
=
only find
F-
a
vector
conditions, then we can take E ||^ \\~1E filling the last two conditions. That is easy ; take =
.
2(Fn,E3)Ej
ji
Then, for
i < n,
( E,)
=
,
(F Ed
-
,
2 (Fn Ej)(Ej E,) ,
,
J
(Fn E,) ,
-
(F E,)(E, ,,)= 0 ,
Furthermore, F=+
so
>Z(F-,Ej)Ej
the last two conditions
The
are
fulfilled and the proposition is proven.
proof of this proposition provides a procedure for finding orthonormal
1.11 bases in
an
Euclidean vector space, known
as
Inner Products
the Gram-Schmidt process.
111 It
goes like this: any basis
First, pick
1
=
Fu
=
.
,
F of
V.
Take
=
Fj+i
.
.,
Ej
be the vector of
J+1
are
by
the
length
to find
found, take
(Fj+i, i)i
~
and divide
(F2, E^)Elf
-
.
and let
.
11^111-^1
Then choose 2 F2 and so forth. If Eu
E+i
.
length
(FJ+l, E2)E2
-
one
-
collinear with
-
(FJ+l, ,);
+i.
Examples 47.
F1 F2 F3
=
=
=
Apply
the Gram-Schmidt process to this basis ofR3:
(1,0,1) (3, -1,2) (0, 0, 1)
Take
Fx
/
J_
_1_\
El"lFj'V2*0,>/2/ E2
a_1)2)_(3.^
=
-(3. -1.2)-
+
(-i).o
+
2.ii)(-L,o,ii)
(|,0.|) (|. -1.^) -
Then
E2
=
E3
{i) [r~1'^r)
=
(0, 0, 1)
-
=
\Wr2'~\v
'(up)
-^ (-^,0, -^) -^3 (^75.
"(7) '(W71)
+
2,
118
1.
Linear Functions
and
finally ~2
I
-
~3
2
\
\(17)1/2'(17)1/2'(17)1/2j
3
48. Find
an
X(x\ x2, x3, x4)
orthonormal basis for the kernel of X: R4 =
of all,
First
(1, 0, 0,
x1
let
+ us
1/2), (0, 1, 0,
-
Schmidt process,
we
x2 -
+
x3
+
-*
R,
2x4.
pick a suitable basis for K(X); that is, 1/2), (0, 0, 1, 1/2). Applying the Gram-
obtain
2
E1
=
=
7^ (i, o,o,^i)
\(3V)m'\6) '^(mr^J
E:
(30)1 E3
=
(1094)1/2
49. Find the T:
orthogonal projection of (3, 1, 2) into the kernel of
R3^R:
T(x,
y,
z)
=
x
+
2y
+
z
Now the kernel of T is
Applying Ei
=
(l)1/2(2,
Thus the
spanned by Ft
the Gram-Schmidt process, -
1, 0)
E2
=
orthogonal projection
(l)1/25(l)1/2(2, -1,0)
+
=
we
(2,
-
=
(0,
-
1, 2).
obtain the orthonormal basis
(fWj, -1, of (3, 1,
1, 0), F2
1)
2) into this plane is
(f)i/2l(f)^(_i
_
|, 1}
=
m,
_az
f)
1.11 50. Find the
L:
x
+
y
3y
+
point
z
=
0
z
=
0
on
which is closest to
(
line
(the
^(7 1 '
'
closest
0)J'
(7, 1,0).
(-*. -L3)\
/
(26)1'2
L is
the
linear span of the vector of (7, 1,0) on this
orthogonal projection
is
point)
(-4,-1,3)^27 l (26)1'2
26
'
'
'
EXERCISES 61. Which of the
(a) (b) (c) (d) (e) (f) (g) (h)
following sets are open; closed; or neither. {xeR:2< |x-5|<13}. {xeR:0<x<4}. {xeR:x>32}. {X6fi":<x,x>=4}. {x6A3:<x,(0,2,l))=0}. {xe#3:2<||x-(3,0,3)||<14}. {xeR":x1>0,...,x">0}. The set of integers (considered as a subset of R).
(i)
{xeR":
2 *''<)
1=1
x
|>'a'^l}.
(j)
{xeR-.
(k)
{xe/?":2(^)3<2^')2}.
62. Find the + 3y + 2z
on
the plane
4
=
closest to the
point
point (1, 0, 1). point on the line
63. Find the
x
+ 7v +
x
z
=
2
z
=
0
closest to the 64. Find
(a) (b) (c) (d)
point (7, 1, 0).
an
orthonormal basis for the linear span of
Vl
=
Vi
=
v,
=
vi
=
(0, (0, (0, (1,
2, 1, 3, 2,
(1, 0, 2), t, (1, 2, 4). (1, 1, 2, 3). (1, 0, 1, 0), v3 0, 1), v2 (0, 0, 2, (0, 6, 0, 3, 0), y3 0, 0, 0), y2 (2, 1, 4, 3). (4, 3, 2, 1), v3 3, 4), y2 2),
119
the line
Thus the
4, -1,3).
Inner Products
y2
=
=
=
=
=
=
=
=
-
1, 1).
120
/.
Linear Functions
65. Find orthonormal bases for the linear span and kernel of these transformations
(a)
R*:
on
0>)
'8 /8
6
1
0 0\
1
2
1
2
1
2
0
2
2
1
2
1
1
-1
1
1
0
1
0
3
3
0
1
7
4
1
-2
0
PROBLEMS 67. Prove
Proposition
69. Give
an
32.
orthogonal set of vectors is independent. example of a sequence {U} of open sets such that f)"=i U
68. Show that
an
is not open. 70. Give an example of is open. 71. Find
an
vector space
sequence
{C} of closed
sets such that
orthonormal basis for the linear span of 1,
C([0, 1])
In the next four
inner
a
with the inner
problems
product, denoted <
,
V
product , gy
represents
a
=
C
x2, x3 in the
\fg.
vector space endowed with
an
>.
72. Let v, w, x be three points in V such that v x. Show that the Pythagorean theorem is valid :
w
x,
(J"=i
x
is orthogonal to
\\v-W\\2=\\v-x\\2+\\x-w\\2
of
73. Let v, w be two vectors in V. w which is closest to v is
Show that the vector in the linear span
(1.51)
w
v0
(You can verify this by minimizing the function f(t) calculus.) 74. Prove Schwarz's inequality:
\\v
tw\\2 by
\
V.
(Hint: [|f
o
II2
=> 0 where v0 is
(1.50).) 75. Prove the
triangle inequality:
ll-x||^||y-w||+||w-x|l for any three vectors in V
(use Schwarz's inequality).
given by
1.11 76. Let V be
is
vector space with
a
an
Inner Products
inner product.
Suppose that
subspace of V. Let J_( W) {v: 0 for all w e IF}. called the orthogonal complement of W. Show that L(W) is a
=
subspace of
V and
F is finite
(if
121
dimensional) that
W and
W
This is
=
linear
a
1_(W) together
span V.
11. Let T: R"
A.
-*
Show that the 78. Show that
/?m be
a
linear transformation represented by the matrix
of A span (K(T)). linear transformation is one-to-one
rows
a
the
on
orthogonal
complement of its kernel. FURTHER READING
R. E. Johnson, Linear Algebra, Prindle, Weber & Schmidt, Boston, 1968. covers the same material and includes a derivation of the Jordan
This book
canonical form. K. Hoffman and R. Kunze, Linear Algebra, Prentice-Hall, Englewood This book is more thorough and abstract, and has a full
Cliffs, N.J., 1961.
discussion of canonical forms. L. Fox, An Introduction to Numerical Linear Algebra, Oxford University Press, 1965. This is a detailed treatment of computational problems in
matrix
theory. Nickerson, D. C. Spencer, and N. Steenrod, Advanced Calculus, Van Nostrand, Princeton, N.J., 1957. This set of notes has a full treatment of all the abstract linear algebra required in modern analysis. H. K.
MISCELLANEOUS PROBLEMS 79. Show that if A' is obtained from A
by
a
sequence of
row
operations
0. 0, A'x 80. Show that every nonempty set of positive integers has a least element. 81. Show that a set with n elements has precisely 2" subsets.
then these
equations have the
same
solutions: Ax
82. Show that the -fold Cartesian
product of
=
a
=
set with k elements has
k" elements. 83. Can you
interpret the
case
k
=
2in Problem 82
so as
to deduce the
assertion of Exercise 3 ? 84. Let A r
the as
=
(a/)
be
an n x n
Show that A"~r
> 0.
=
0.
i>r for
assumption j
matrix such that
Show that the some r
> 0.
a/
same
Will the
=
0 if
i
j>
r
for
some
conclusion follows from
hypothesis \i
j\
>
r
do
well? 85. Let T: R"
there
->
Rm be
a
linear transformation of rank
r.
Show that
linear transformations S,: Rm-^Rm-', S2: R"-'^Rn such that (a) Si has rank m r and b e R(T) if and only if Sib 0.
are
=
(b) S2 has rank 86. Suppose that T:
-
R"
r ->
and
x 6
K(T) if and only if
R" and Tk
=
/.
x e
Show that T is invertible.
Show that the linear span intersection of all linear subspaces of R" containing S. 87. Let S be
a
subset of R".
R(S2). [S] of
S is the
Linear Functions
1.
Show that
88. Let S, T be subsets of R".
dim([S
r])
u
<
dim([5]) + dim([r]),
equality holds if and only if [S] n [T] 89. Let V and W be subspaces of Rn.
=
and
{0}.
Let X be the set of all
sums
subspace of R". The + V, V + W. If relationship between X and V and W is indicated by writing X the form in written be xeJcan in addition V r\ W={0}, then every we say n W V W with X this 0, In V+ one case, v + w in only way. v
w
with
w 6
v e
W.
Show that X is
linear
a
=
=
=
of V and IF and write X=V@W. 90. Suppose X= V W. Then dim X= dim F + dim IF. 91. Show that if A: R" -> .R is a linear function, there exists a
that X is the direct
sum
w e
.R" such
that A(v)
1(5)
=
{v
e
R":
=
0 for all
s e
S}.
{0}. (a) Show that (S) is a subspace of .R" and that S n _L(S) that Show =(.(S)). [S] (b) F J_(F). (c) If V is a linear subspace of .R", R" 93. Suppose that T: V-> Wis a linear transformation and Fis not finite dimensional. Show that either the rank or the nullity of T must be infinite. =
=
94. Let V be
an
A bilinear function p
abstract vector space.
function of two variables in V with these
on
V is
a
properties:
cp(v, w) p(v, cw) p(cv, w) cp(v, w) p(v, p(vi + v2,w) =p(vu w) + p(v2 w) =
=
,
Wi
+
w2) =p(v, Wi) + p(v, w2)
In fact, the space sum of two bilinear functions is bilinear. Bv of all bilinear functions is an abstract vector space. If V is finite dimen sional, what is the dimension of Bv ? (Hint: See the next problem.)
Show that the
95. Let p be
a
bilinear function
on
R".
Let
a,;j=piE,,Ej) Show
that/? is completely determined by
96. Let V be
(a)
the matrix
(at;j).
abstract vector space. Show that the space V* of linear functions an
under addition and scalar
on
Fis
a
vector space
multiplication.
d also. (b) If dim V=d, show that dim V* (c) Show that to every A e R"* there isaweff such that A(v)
=
=
1.11 Show that ann(F) is a linear subspace of IF*. show that ann(F) has dimension n d. 98. Let V be
linear
a
subspace
linear transformation. R"
-=?
J?m defined
on
If dim W
=
123
dim V
n,
=
d,
of .R", and suppose that T: V->Rm is a a linear transformation 7":
Show that there is
all of R" which extends T.
99. The closure of every
Inner Products
a
neighborhood of
set x
S, denoted 5, is the
contains
points of
S.
set of all
points
x
such that
Find the closure of all the
sets in Problem 61.
1 00. Show that the closure of a set S is the smallest closed set containing S.
101. The boundary of
a
set
S, denoted dS, is the
set of all
points x such complement
that every neighborhood of x contains points of both S and the of S. Find the boundary of all the sets in Problem 61. 102. Show that the
103. Show that the
boundary of a boundary of a
set is
a
closed set.
set S is also the
S n (Rn plement fact, show that 8S 104. Let T: V-> W be a linear transformation of R"
inner
S.
-
In
=
-
boundary of its
com
S). vector space with
a
an
The adjoint of Tis the transformation T*: W-+ V defined
product.
in this way
=
for all
<w, Tv>
(a) Show (b) If T:
that T* is
a
v e
V
well-defined linear transformation.
represented by the matrix A (a/), represented by the matrix A* (a*j), where a*j (This matrix is called the adjoint or transpose of A.) (c) Show that R(T*) is complementary to K(T). v(T), v(T*) p(T). (A) In fact, P(T*) R"
--
Rm is
=
T* : Rm^ R" is
=
then =
atJ.
=
=
105. A bilinear form pona vector space V is called symmetric if it obeys the law: p(v, w) =/>(w, v) for all v and w. An inner product is a symmetric
manipulations with inner products symmetric bilinear forms. For example, the GramSchmidt process (Proposition 32) gives rise to this fact (see if you can work the proof of Proposition 32 to give it): Proposition. Let p be a symmetric bilinear form on V. Suppose Fi, We can find another basis, Eu...,EofV such that the F is a basis for V. Fj for all j, and linear span of Eu...,Ej is the same as that of Fu p(E,,Ej) 0ifi^j. We shall call such a basis Ei,..., E p-orthogonal. 106. Let/? be a symmetric bilinear form on a vector space V, and suppose E is a p-orthogonal basis. Ei, (a) Show that p(v, w) can be computed in terms of this basis as bilinear form and much of the formal remains valid for
...,
.
.
.
,
=
...,
follows : if
p(p, w)
=
v
=
2
v'Et
2 v'w'p(E,
1=1
,
,
w
E,)
=
2
W'E> then >
(1.52)
/.
Linear Functions
(b) Show that
p is an inner
product
on
the linear span of the Et
such that p(E,, E,) >0.
(c) Similarly,
p is an
inner
product
on
the linear span of the Et
such that p(E, E,) < 0. 107. Prove this fact: Let p be a symmetric bilinear form on a finitedimensional vector space V. There is a basis Eu ...,E , integers r, s such that r + s <; n and such that if v 2 v'-Ei then ,
=
,
Xm)=2M!fr
2 (f')2
0-53)
rlr+s
(Hint: Modify the basis {,} in Problem 106 so that (1.52) becomes (1.53).) 108. The integers r, s of Problem 107 are determined by p alone, and are independent of the basis. Here is a sketch of how a proof would go. Suppose Fi, ...,F is another p-orthogonal basis and p is the number of r. Let W be the linear Ft's such that p(Ft Ft) > 0. We have to show p E span of these F's. Expressing points of W in terms of the basis Ei we may consider the transformation T: W->R" given by =
,
TQv'Et)=(v\...,tf) T is one-to-one
on
W, for if
0
so we
w e
W, and
w
= 0,
2 (')2
rir+i
must have
2(')2>o
tsr
on
W.
Since T is one-to-one, it follows that r ;> p. The inequality p >; r same argument with the roles of Eu E and FU...,F
follows from the
.
.
.
,
interchanged. 109. Let A
=
(aci) be
Then A determines
PA(y, w)=
a
a symmetric n x n matrix, that is, a,; j symmetric bilinear form on R" as follows:
=
a,;
,
2 aiww'w-'
i.j
If P is the matrix
corresponding to the change of basis from the standard basis to that described in Problem 105, then P*AP is diagonal. Verify that assertion.
1.11
Inner Products
125
110. Find the p-orthogonal basis and the
107 for the symmetric bilinear forms
[4
(a)
3
p(y, y)
=
p(y, v) >0, =0,
(v1)2 + (i;2)2 + (v3)2 transformation
112. A
T:
v,
w e
Show that if T is
V).
self-
v, w are eigenvectors of a self-adjoint transformation T eigenvalues. Show that
Suppose that
V with different
114. If T is
2(V)2
=
a
=
land
an
2 0>')2
eigenvector for
=
1}
T.
115. Use Problems 113 and 114 to prove the
adjoint operators Theorem. T
a
K(T)R(T)
113. on
given by
self-adjoint if it is self-
V-> V is called
=
=
<0 in R* where p is
(v*)2
-
(b)
l\
0
111. Describe the sets
representation (1 .53) of Problem given by these matrices:
can
be
on
There is
computed
r(2*'E,)
=
Spectral
theorem for self-
R": an
orthonormal basis Ei,
in terms
.
.
.
,
E of eigenvectors of T.
of this basis by
2*'c.E,
1 1 6. Find a basis of eigenvectors in R* for the self-adjoint transformations given by the matrices (a), (b) of Problem 110.
117. Orthonormalize these bases of R*:
(a) (b)
(1, 0, 0, 0), (0, 1, 1, 1), (0, 0, 2, 2), (3, 0, 0, 3). (-1, -1, -1, -D,(0, -1, -1, -1), (0,0, -1, -1),
(0,0,0,-1). (c) (0, 1, 0, 1), (1, 0, 1, 0), (1, 0, 0, 1), (0, 1, 1, 0). 118. Find the orthogonal projection of R5 onto these spaces: (a) The span of (0,1,0,0,1). (b) The (c) The (d) The
(1, 1, 0, 0, 0), (1, 0, 1, 0, 0). (1, 0, 0, 0, 1), (0, 1, 0, 0, 1), (0, 0, 1, 0, 1). of the vectors given in (c) and the vector (0, 0, 0, 1, 1).
span of
span of span
127,216 2) &$/&8/86
&KDSWHU
2QH RI WKH PDLQ PHWKRGV RI WKH PRGHUQ
DSSURDFK
WR PDWKHPDWLFV LV WKH
UHFRJQLWLRQ RI IDPLOLDU FRQFHSWV DW ZRUN LQ XQIDPLOLDU VHWWLQJV 7KXV WKH LGHDV RI OLQHDU DOJHEUD RULJLQDOO\ LQWURGXFHG IRU WKH SXUSRVH RI VROYLQJ V\VWHPV RI HTXDWLRQV ZLOO EH VHHQ DOVR WR KDYH UHOHYDQFH LQ WKH VWXG\ RI ,Q WLPH LW ZLOO EH
IXQFWLRQV
RXU
WKDW PDQ\ VLPSOH FRQFHSWV RI JHRPHWU\ 7KXV LW LV LPSRUWDQW WR XV ZKHUH SRVVLEOH
VHHQ
ORW RI PDWKHPDWLFV
SHUPHDWH WR WU\ WR LVRODWH D
FRQFHSWV DQG
LQLWLDOO\ YHU\ DEVWUDFW VLWXDWLRQ 2I ZH FDQ W MXVW GR WKDW FRXUVH DSSOLFDELOLW\ IDPLOLDULW\ ZLWK WKH EHKDYLRU RI WKRVH FRQFHSWV )RU VHW WKHP LQ DQ
LQ RUGHU WR PD[LPL]H WKHLU ZH
PXVW KDYH KDG VRPH
WKDW
ZH
QL]H
PRUH
QHHG
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
WKH ODWH WK FHQWXU\ :H VKDOO QRW WU\ WR HQFRPSDVV RYHU WZR FHQWXULHV RI VWUXJJOH LQ D IHZ SDJHV ZH VKDOO KDYH WR WDNH VRPH VKRUW FXWV DQG ZH VKDOO
WU\ IRU REYLRXV SHGDJRJLFDO UHDVRQV
WR DYRLG WKH
JUHDW FRQIXVLRQ WKDW LV
VXIIHUHG 7KH
GXULQJ VXFK GHYHORSPHQW EDVLF WHFKQLTXH RI FDOFXOXV LV DSSUR[LPDWLRQ
/HW
XV
JLYH
DQ
LOOXVWUD
1RWLRQV
RI &DOFXOXV
WLRQ RI KRZ LW JRHV 7KH SUREOHPV RI FDOFXOXV DUH VXFK WKDW ZH DUH UHTXLUHG WR SURGXFH D IXQFWLRQ WKDW KDV JLYHQ SURSHUWLHV 7KHUH DUH WZR DVSHFWV WR WKLV D
SUREOHP
VROXWLRQ WR
7KHUH LV WKH WKHRUHWLFDO DVSHFW WR EH DVVXUHG WKDW WKHUH H[LVWV SUREOHP DQG WKH SUDFWLFDO RQH WR GHVFULEH D SURFHGXUH IRU
RXU
HIIHFWLYHO\ FRPSXWLQJ WKDW VROXWLRQ
7KHVH WZR DVSHFWV DUH LQVHSDUDEOH IDFW ZH PDNH D VHTXHQFH RI DWWHPSWV WR VROYH WKH SUREOHP ,I WKHVH DWWHPSWV DUH JRRG LW ZLOO SURYLGH XV ZLWK D VHTXHQFH RI IXQFWLRQV VXFFHVVLYHO\ ,Q
SURYLGLQJ EHWWHU
VROXWLRQV WR WKH
7KHQ IXUWKHU
SUREOHP
VWXG\
RI WKH
JHQHUDO
IRUP RI WKHVH WHQWDWLYH VROXWLRQV PD\ SURYLGH D FOXH WR WKH DFFXUDWH VROXWLRQ 6XSSRVLQJ ZH KDYH D VTXDUH RI VLGH OHQJWK RQH XQLW LQ WKH SODQH VHH )LJXUH
FRQVLGHU WKLV SUREOHP
)LQG
D
IXQFWLRQ GHILQHG
RQ
WKH ER[ ZKLFK KDV
SUHVFULEHG YDOXHV DW WKH YHUWLFHV DQG ZKLFK VDWLVILHV WKLV FRQGLWLRQ )RU HYHU\ SRLQW LQ WKH ER[ DQG DQ\ UHFWDQJOH ZLWK FHQWHU DW WKDW SRLQW WKH YDOXH RII DW WKDW 1RZ
DYHUDJH RI WKH YDOXHV RI DW WKH YHUWLFHV RI WKH ZULWH WKHVH FRQGLWLRQV PRUH SUHFLVHO\
SRLQW LV WKH ZH FDQ
ZKHUH D E F G DUH KDYH
\
IO
D
E
O
JLYHQ QXPEHUV
F
)XUWKHU IRU
O8[ V\W I[ I[V\ W @
I??
G
[ \
DQG
DQ\
V\W I[
UHFWDQJOH
V W
V \
ZH
PXVW
W
1RZ KRZ GR ZH ILQG VXFK D IXQFWLRQ" :H FRPSXWH EDVHG RQ WKH JLYHQ LQIRUPDWLRQ LWV YDOXH DW FHUWDLQ SRLQWV DQG WU\ WR VHH LI ZH REWDLQ D SDWWHUQ
)LJXUH
1RWLRQV
RI &DOFXOXV
)LUVW RI DOO WKH YDOXH DW WKH FHQWHU RI WKH ER[ LV HDV\ WR ILQG
IK L %\ V
D
E
F
G
FRPSXWH WKH YDOXH
ZH FDQ
DW WKH FHQWHU
SRLQWV
RI WKH VLGHV
LW
D
6LPLODUO\
/HW
ZH
WR
PRUH
FRPSOLFDWHG SRLQWV
IRXU VTXDUHV LQ
)LJXUH
YHUWLFHV
FRPSXWH E\
1RZ
SRLQWV VKRZQ LQ )LJXUH H[DPSOH WKH FHQWHUV RI WKH
REWDLQ WKH YDOXHV DW DOO WKH RWKHU FHQWHU
XV PRYH
ZH
E
L> @
IL
PD\
6LQFH
NQRZ WKH YDOXHV DW DOO WKH UHOHYDQW
ZH
IK`
A
AE
AF
AG
P
AD
AE
AF
AG
IKL
A
E
AF
AG
IKL
AD
AE
AF
AG
ZH FDQ VHH
WKDW
ZH
FDQ
EUHDN WKH
FRPSXWH WKH YDOXHV DW WKH FHQWHUV
JLYHQ VTXDUH LQWR VTXDUHV DQG SRLQWV RI WKH IRUP S TM DQG VR
F
D
IRU
G
G
D
D
E F G
)LJXUH
E
E
W
&RQYHUJHQFH RI 6HTXHQFHV
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
IRUWK
:H
FDQ
RI WKH IRUP
T?
,S
S
I?7 7`
T
IO
:
S
T
/
ST
T7
S
-
MDU
WKH FDVH Q S T :H FDQ VKRZ E\ VXFFHVVLYHO\ FRPSXWLQJ WKH YDOXHV DW FHQWHUV RI VTXDUHV WKDW LV YDOLG IRU DOO Q 7KXV UHZULWLQJ ZH FDQ DVVHUW WKDW LI [ \ LV RI WKH IRUP S T ZLWK DQG T DV LQWHJHUV WKHQ
I[ \
[ O
\ D
[?
\ E
[\F
[ \G
$VVXPLQJ WKDW LV D ZHOOEHKDYHG IXQFWLRQ WKLV PXVW WKHQ KROG IRU DOO SRLQWV [ \ )LQDOO\ ZH FDQ VKRZ E\ VXEVWLWXWLQJ LQWR WKH UHTXLUHG FRQGLWLRQV WKDW JLYHV WKH VROXWLRQ 2XU SXUSRVH LQ WKH SUHVHQW FKDSWHU LV WR GLVFXVV WKH WKHRUHWLFDO FRQFHSWV ZKLFK UHPRYH WKH IX]]LQHVV LQ WKH DERYH GLVFXVVLRQ :H VKDOO H[SRVH WKH LGHDV OLPLW DQG FRQWLQXLW\ LQ WKH VHWWLQJ RI IXQFWLRQV RI PDQ\ YDULDEOHV :H VKDOO DOVR SUHVHQW D UHYLHZ RI WKH LQIRUPDWLRQ IURP FDOFXOXV ZKLFK LV QHFHVVDU\ WR WKH
VWXG\
&RQYHUJHQFH
%HIRUH
RI
D
RI WKLV WH[W
RI
6HTXHQFHV
SURFHHGLQJ GLUHFWO\
VHTXHQFH
DUH
WR WKH OLPLW
QRWLRQ
D
IHZ ZRUGV
RQ
WKH QRWLRQ
$ VHTXHQFH RI SRLQWV LQ ; RI SRLQWV LQ ; RQH IRU HDFK
/HW ; EH DQ\ VHW
LQ RUGHU
[ ` VD\LQJ WKDW LV WKLV D VHTXHQFH RI SRLQWV $V D D IXQFWLRQ I3 ; ZKHUH ZH GHQRWH E\ [Q WKH [` VKRUWKDQG GHYLFH ZH ZLOO RIWHQ GHQRWH VHTXHQFH ^[X[
LV
DQ
RUGHUHG FROOHFWLRQ
SRVLWLYH LQWHJHU LQ ; LV JLYHQ E\
PHUHO\ E\
LWV
^[X
[
$QRWKHU ZD\ RI
JHQHUDO
WHUP
^[`
1RWLRQV
RI &DOFXOXV
([DPSOHV
L
L
L
^
'Q
Q
?
IQ
`
^RGGQXPEHUHG [ V`
^HYHU\
^[3Q`
^ `
ILIWK WHUP LQ
[Q
GE
LP
^`
D
VHTXHQFH
^\`
H[WUDFWHG
7KXV WKH FROOHFWLRQV
`
^[QBL`
^[Q`
^[``
ZKHUH SQ LV WKH QWK ZKHUH J LV
LV
^[`
VXEVHTXHQFH RI
D JLYHQ VHTXHQFH FROOHFWLRQ WKH RUGHUHG IURP ^[X
$
^
^ LW`
SULPH
VWULFWO\ LQFUHDVLQJ IXQFWLRQ
D
WKH
RQ
SRVLWLYH
LQWHJHUV DUH DOO VXEVHTXHQFHV RI ^[` ZKHUHDV
^
[X
[X
`
LV QRW
D
VXEVHTXHQFH
7KH DERYH GHVFULSWLRQ RI VXEVHTXHQFH LV IURP LV
SLFWXUHVTXH
^[
;
;
EXW QRW WRR
;
;M ;MT
M
D
;J
;Q
VXEVHTXHQFH RI ^[` "
HYHU ZH
RI
D
,W LVQ W FOHDU IURP WKH
VKRXOG GUDZ WKH OLQH DQG H[FOXGH VXFK
VXEVHTXHQFH
ZLOO EH WKDW LW FRQVLVWV RI
H[WUDFWHG
;QL
[Q D
,V WKH VHTXHQFH
PHDQLQJIXO
;J
7KH SKUDVH
ELW YDJXH
;Q
[Q
SUHFHGLQJ SDUDJUDSK 7KH
QHZ
`
!
+RZ HVVHQFH
VHTXHQFHV RI WKH [ V LQILQLWHO\ PDQ\ 1RZ WR EH UHDOO\ H[DFWLQJ RXU
VRPH
RI WKHP DQG FROOHFWHG LQ WKH VDPH RUGHU QRWLRQ RI VHTXHQFH LWVHOI LV LPSUHFLVH ZH VHHP WR KDYH IDLOHG WR VD\ ZKDW LW LV :H KDYH DOUHDG\ HODERUDWHG $Q RUGHUHG FROOHFWLRQ LV QRW YHU\ VDWLVIDFWRU\ LV JLYHQ E\ D IXQFWLRQ I3 ! ;
EXW ZKDW LV LW" ,W WXUQV RXW WKDW WKLV OLQH RI PHWDSK\VLFDO ERJV GRZQ DQG LV LQ IDFW LUUHOHYDQW :H KDYH DOUHDG\ IRXQG
JLYHQ E\ TXHVWLRQLQJ VRPHWKLQJ ZKLFK FRPSOHWHO\ GHVFULEHV WKH VHTXHQFH WKH IXQFWLRQ 3 ; VR ZK\ QRW GHILQH D VHTXHQFH MXVW DV VXFK D IXQFWLRQ" ,QGHHG ZKHQ ZH GR LW WR DOVR EHFRPHV GHILQH D VR YHU\ HDV\ VXEVHTXHQFH
'HILQLWLRQ
/HW ; EH D VHW $ VHTXHQFH LQ ; LV D IXQFWLRQ 3 RI WKLV VHTXHQFH LV DQRWKHU VHTXHQFH K3 ; ZKHUH K VXEVHTXHQFH DQG J LV D VWULFWO\ LQFUHDVLQJ IXQFWLRQ IURP 3 WR 3
$
;
IJ
7KXV LI ^[[
ZULWLQJ LW
DV
WKH
[
`
LV
IXQFWLRQ
^[J: [J
D
[Q
&RQYHUJHQFH RI 6HTXHQFHV
VHTXHQFH WKLV LV LQ IDFW MXVW DQRWKHU ZD\ RI ,II J LV D VXEVHTXHQFH ZH FDQ HQXPHUDWH
[JQ ` DQ LOOXVWUDWLRQ RI D VWDQGDUG PDWKHPDWLFDO SURFHGXUH
7KH DERYH GHILQLWLRQ LV RI
GHILQLQJ WKLQJV $ FRQFHSW LV PDWKHPDWLFDOO\ DQ REMHFW ZLWK VXFK DQG SURSHUWLHV 2QFH ZH KDYH VWDWHG WKH SURSHUWLHV ZKLFK ZH IHHO GHVFULEH WKH FRQFHSW WKHUH LV QR QHHG WR IXUWKHU LQTXLUH ZKDW WKH REMHFW LV ZH VLPSO\ GHILQH LW E\ WKRVH SURSHUWLHV :H QRZ LQWURGXFH WKH QRWLRQ RI FRQYHUJHQFH RI D VHTXHQFH RI QXPEHUV ZKLFK ZH PD\ WDNH DV FRPSOH[ QXPEHUV VXFK
'HILQLWLRQ /HW ^]` EH D VHTXHQFH RI FRPSOH[ QXPEHUV :H VD\ WKDW WKH VHTXHQFH FRQYHUJHV LI WKHUH LV D ] H & VXFK WKDW WR HYHU\ SRVLWLYH QXPEHU V ! WKHUH FRUUHVSRQGV DQ LQWHJHU 1 VXFK WKDW ?] ]? H IRU Q ! 1 ,Q WKLV FDVH ZH VD\ ^]` FRQYHUJHV WR ] ZULWWHQ OLP ]Q ] RU OLP ] ] RU Q RR
]QA]
6DLG DQRWKHU ZD\ WKH VHTXHQFH
GLVN FHQWHUHG DW ] WKH UDQJH GLVN VHH )LJXUH
RII
I3
RQ
& FRQYHUJHV WR
DOO EXW
)LJXUH
ILQLWHO\
PDQ\
] H
&
LI JLYHQ
LQWHJHUV
DQ\
OLHV LQ WKDW
1RWLRQV
RI &DOFXOXV
7KH
IROORZLQJ SURSRVLWLRQ DVVHUWV WKDW D VHTXHQFH FDQQRW FRQYHUJH WR PRUH SRLQW DQG JLYHV QHFHVVDU\ FRQGLWLRQV IRU FRQYHUJHQFH ZLWKRXW UHIHUHQFH WR WKH OLPLW SRLQW WKDQ
RQH
3URSRVLWLRQ
6XSSRVH
OLP ]
]
L LI DOVR OLP ] Z WKHQ Z ] LL WKH VHTXHQFH LV ERXQGHG WKDW LV
WKHUH LV
DQ
0
!
VXFK WKDW
?]Q?
0
DOO Q
IRU LLL &DXFK\ FULWHULRQ IRU ?] ]P? HIRU DOO QP!1 3URRI L %\
WKH
UIA1L ?]
K\SRWKHVHV JLYHQ Z?
H
IRU
Q
HYHU\
H
A 1
H !
WKHUH 7KXV
!
DQ
79
!
VXFK
1L 1 VXFK WKDW ?]
DUH
Z? H KROGV IRU DOO
LQHTXDOLW\ ?]
LV
]_
WKDW
H
IRU
Z? H
?]Z?A?]]1? ?]1 6LQFH WKH
WKHUH
H
!
ZH
PXVW KDYH
?]
Z?
RU
Z
]
H WKHUH LV DQ LQWHJHU 1 VXFK WKDW _] ]? IRU Q A1 /HW PD[^_]_ _]L_ _]_ _]_`O 7KHQ LI QA1 FHUWDLQO\ _]_A0 ,I Q!1 _]_A_]]_ _]_AO ?]?0 LLL /HW H! EH JLYHQ 7KHUH LV DQ LQWHJHU 1 VXFK WKDW _] ]_ H IRU Q1 7KXV LI Q P ! 1 ZH KDYH
LL 7DNLQJ
0
_]
]P_
&RQGLWLRQ
LLL
_]
]_ _]P
LV FDOOHG
]_
H
FULWHULRQ IRU LW
LPSOLHV FRQYHUJHQFH DV ZH VKDOO LL LPSO\ FRQYHUJHQFH 7KH VHTXHQFH ^ ` LV FOHDUO\ ERXQGHG EXW GRHV QRW FRQYHUJH LW GRHVQ W HYHQ VDWLVI\ WKH &DXFK\ FULWHULRQ _ IRU DOO Q _ VHH
EHORZ
D
1RWLFH WKDW
GRHV QRW
([DPSOHV OLP VR
Q
Q
WKDW 1!Ha
/HW
!
7KHQ IRU
QBA-9B
!
EH Q !
JLYHQ
1
DQG FKRRVH WKH
LQWHJHU
79
OLP
LQ
7KH
&RQYHUJHQFH RI 6HTXHQFHV
SURRI LV
WKH
VDPH
VHH DOVR 3UREOHP
OLP
OOQ /HW
EH
!
JLYHQ
1RZ Q
O
OQ
7KXV
ZH
OO QHHG
Q
WDNH 1!H 0R
RQO\
Q
Q
YHULI\ WKH
FRQGLWLRQ IRU
FRQ
YHUJHQFH
OLP
Q
OLP K $7!
Q
ZH
KDYH
PDWKHPDWLFDO LQGXFWLRQ WKDW K FOHDU VLQFH AI
Q
WKHUH LV
LQWHJHU
DQ
6LQFH WKH VHTXHQFH QQ LV DOO Q ! . 1RZ ZH VKDOO VKRZ E\
.Q
IRU DOO
7KH
Q
1RZ XVLQJ WKH QWK LQHTXDOLW\
FDVH Q
ZH
LV
REWDLQ WKH
O WK
Q
Q
.
.
Q
Q
Q
QQ
LI
7KXV K
!
6LQFH K
.. QQ
Q
[
Q !
IRU QO
VXFK WKDW K
LQFUHDVLQJ
LI
H
Q
!
.Q
LV
JLYHQ
OHW
1A.H
)RU
Q!1
_Q_
FRQYHUJHQFH RI FRPSOH[ VHTXHQFHV LV HDVLO\ UHGXFHG WR WKDW RI UHDO VHTXHQFHV E\ WKH IROORZLQJ IDFW $ FRPSOH[ VHTXHQFH FRQYHUJHV LI DQG RQO\ LI WKH UHDO DQG LPDJLQDU\ SDUWV ERWK FRQYHUJH
7KH
VWXG\ RI WKH
3URSRVLWLRQ OLP ]
]
[
/HW
^]`
EH
L\ LI DQG RQO\
D
VHTXHQFH LI OLP [Q
RI FRPSOH[ QXPEHUV [
DQG OLP \Q
\
]Q
[Q
L\Q
1RWLRQV
RI &DOFXOXV
3URRI 6XSSRVH OLP ] ]? H 6LQFH QA1 ?]
DQG
?[[??]]? ZH
H
WKHUH LV
!
DQ
1 VXFK WKDW IRU
?\\??] ]?
DOVR KDYH
OLP
DQG
[OH
_[ VR
7KHQ JLYHQ
]
[
[
DQG OLP \
&RQYHUVHO\ JLYHQ
LPSOLHV _ \
\?
H
!
\?H
?\Q
IRU Q!1
Y
WKHUH DUH 0 1 VXFK WKDW Q
H9L
!
0 LPSOLHV ?[
[?
H9
7KHQ
OH
H?
?]]? >?[[??\\?
^]`
VR
IDU EHHQ
FRQVLGHULQJ TXHVWLRQV RI WKLV IRUP *LYHQ WKH VHTXHQFH ]" $ GHHSHU SUREOHP LV WKLV *LYHQ
DQG WKH QXPEHU ] LV OLP ] Q FR
SRVVLEOH D QXPEHU ] VXFK WKDW OLP ] ] $ UHTXLUHV D PRUH SURIRXQG XQGHUVWDQGLQJ RI WKH UHDO QXPEHU V\VWHP WKDQ ZH KDYH VR IDU QHHGHG $ TXHVWLRQ RI H[LVWHQFH LV QRZ LQYROYHG 7R UHVROYH VXFK TXHVWLRQV ZH KDYH WR KDYH H[SOLFLW NQRZ OHGJH WKDW WKHUH DUH PDQ\ UHDO QXPEHUV ZKHUHDV XQWLO QRZ ZH KDYH PDGH XVH 7KH H[SOLFLW NQRZOHGJH GHVLUHG RQO\ RI WKH H[LVWHQFH RI WKH QXPEHUV DQG OHDVW KHUH LV WKDW SURYLGHG E\ WKH D[LRP RI WKH XSSHU ERXQG ZKLFK URXJKO\
WKH VHTXHQFH ^]` ILQG LI VROXWLRQ RI VXFK SUREOHPV
JDSV LQ WKH OLQH RU UHDO QXPEHUV $ VHW 6 RI UHDO QXPEHUV LV VDLG WR EH ERXQGHG IURP DERYH LI WKHUH LV D QXPEHU 0 VXFK WKDW ,I 6 LV D VHW ZKLFK LV ERXQGHG DERYH LW LV FRQFHLYDEO\ [ 0 IRU HYHU\ [ H 6 VWDWHV WKDW WKHUH
DUH QR
XVHIXO WR NQRZ WKH VPDOOHVW QXPEHU 0 ZKLFK ZLOO VHUYH :H VKDOO UHIHU WR VXFK D OHDVW XSSHU ERXQG RI WKH VHW 6
XSSHU ERXQG VXS 6 DQG LQI 6
DV DQ
E\
ZLOO GHQRWH WKH JUHDWHVW ORZHU ERXQG LI LW H[LVWV 7KH D[LRP RI WKH OHDVW XSSHU ERXQG DVVHUWV WKDW IRU D VHW 6 ZKLFK LV ERXQGHG DERYH VXS 6 H[LVWV :H VKDOO VWDWH WKLV PRUH
DSSURSULDWH
7KHRUHP [Q A
[QIRU
DOO
VDPH
WR
/HW Q
D[LRP LQ WHUPV RI VHTXHQFHV EHFDXVH LQ WKDW IRUP LW LV
RXU
SUHVHQW
^[` ,I WKH
:H KDYH FDOOHG WKLV
D
EH VHW
FRQWH[W
GHFUHDVLQJ VHTXHQFH RI UHDO QXPEHUV WKDW LV ^[` LV ERXQGHG WKH VHTXHQFH FRQYHUJHV
D
WKHRUHP VLQFH LW
FDQ
EH GHGXFHG IURP WKH D[LRP RI
WKH OHDVW XSSHU ERXQG VHH 3UREOHP ZKLFK ZH FDQ WDNH DV D GHILQLQJ SURSHUW\ RI WKH UHDO QXPEHU V\VWHP $ FRQVHTXHQFH RI WKLV IDFW RI H[LVWHQFH
IRU WKH UHDO QXPEHUV LV WKH IDFW WKDW WKH
&DXFK\ FULWHULRQ VHH 3URSRVLWLRQ
FULWHULRQ IRU FRQYHUJHQFH 7KH SURRI JRHV OLNH WKLV ILUVW ZH ILQG VXEVHTXHQFH RI WKH JLYHQ VHTXHQFH ZKLFK LV GHFUHDVLQJ $Q HDV\ FRQ LV
OLLL D
&RQYHUJHQFH RI 6HTXHQFHV
D
VHTXHQFH RI WKH &DXFK\ FULWHULRQ LV WKDW WKH VHTXHQFH LV ERXQGHG 7KXV E\ 7KHRUHP WKLV VXEVHTXHQFH KDV D OLPLW [ ,W QRZ IROORZV IURP WKH &DXFK\
FULWHULRQ WKDW WKH IXOO 7KHRUHP
VHTXHQFH DOVR KDV WKH OLPLW
JLYHQ
/HW
EH
^[`
&DXFK\
D
VHTXHQFH
IRU HYHU\ H! WKHUH LV DQ LQWHJHU 1 VXFK WKDW ?[ [ 7KHQ WKHUH LV DQ [ VXFK WKDW OLP [ )LUVW
3URRI _[
[P
D
&DXFK\ VHTXHQFH LV ERXQGHG
IRU Q P!1
_
7KHQ 0
/HW
PD[^ _MFL _
RI UHDO QXPEHUV 7KDW LV [P? V ZKHQHYHU QP! 1
WKHUH LV
H
[
?[1 _` LV
D
DQ
1 VXFK WKDW
ERXQG IRU ^[` 0 IRU DOO N VR
VXS^[ Q ! N &OHDUO\ WKH XN DUH GHFUHDVLQJ DQG XN ! X :H VKDOO VKRZ WKDW DOVR E\ 7KHRUHP WKH VHTXHQFH XN FRQYHUJHV VD\ OLP XN /HW XN
OLP [ /HW
X
7KHUH
H !
DUH
0 1 VXFK WKDW
H
?[
IRU Q PA1L
[P?
IRU N!1
_W _
VXS^[ Q!1` WKHUH LV DQ QA1 VXFK 7KHQ FRPELQLQJ DOO WKHVH LQHTXDOLWLHV ZH KDYH IRU Q A 1 6LQFH
1L 1
/HW 1
[ X1
H
[_
?[ X? ?[ %HFDXVH RI
3URSRVLWLRQ
LPDJLQDU\ SDUWV GR &RUROODU\
3URRI
X1
$
X1? _
X?
%
VHTXHQFH FRQYHUJHV LI LWV UHDO DQG GHGXFH WKH VDPH WKHRUHP IRU FRPSOH[ VHTXHQFHV
WKDW
ZH FDQ
&DXFK\
_[
D
VHTXHQFH
FRPSOH[
RI FRPSOH[
QXPEHUV FRQYHUJHV
3UREOHP
(;(5&,6(6 :KDW
DUH
WKDW
WKH OLPLWV
ZKHQ WKH\ H[LVW
RI WKHVH VHTXHQFHV
D
^`
F
MAM
E
^ `
G
^O '`
1RWLRQV
RI &DOFXOXV
H
^
J
I
^QVLQ
K
B
,I
^ ` LV
?Q
D
(+ Q O QA ,
FRQYHUJHQW VHTXHQFH WKHQ OLP[L
[
,V WKLV
D
,, &2
FULWHULRQ IRU FRQYHUJHQFH" ] /HW ^\` EH WKH VHTXHQFH ^[NL [N 6XSSRVH OLP [ ` 6KRZ ] DOVR WKDW OLP \ 6KRZ WKDW LI WKH\ DUH /HW ^ ` ^I` EH WZR FRQYHUJHQW VHTXHQFHV
FRQYHUJHQW VHTXHQFHV ZLWK WKH FRQYHUVH
VDPH
OLPLW WKHQ ?LPV
,V WKH
WUXH"
:KDW LV OLP
Q
O
9Q O
LI DQG
6KRZ WKDW OLP ]
RQO\
LI
OLP_]_
352%/(06 /HW ^[Q` EH D GHFUHDVLQJ VHTXHQFH RI UHDO QXPEHUV 3URYH OHDVW XSSHU ERXQG D[LRP WKDW LI ^[` LV ERXQGHG LW LV FRQYHUJHQW DOVR WKDW DQ LQFUHDVLQJ ERXQGHG VHTXHQFH LV FRQYHUJHQW
OLP
XVLQJ
WKH
'HGXFH
6XSSRVH WKDW OLP ] DQG ^F` LV D ERXQGHG VHTXHQFH 3URYH WKDW ,I ^]` LV D FRQYHUJHQW VHTXHQFH DQG ^F` LV D ERXQGHG VHTXHQFH
F]
QRR
LV
^F ]` FRQYHUJHQW "
RU
ERXQGHG "
'HGXFH IURP WKH IDFW WKDW
&DXFK\
VHTXHQFHV RI UHDO QXPEHUV
FRQ
YHUJH WKDW D &DXFK\ VHTXHQFH RI FRPSOH[ QXPEHUV LV FRQYHUJHQW /HW ^]` EH D VHTXHQFH RI FRPSOH[ QXPEHUV DQG ^F` D VHTXHQFH RI SRVLWLYH UHDO QXPEHUV VXFK WKDW _]_F IRU DOO Q!1 3URYH WKDW LI
OLP
DOVR OLP ] 6XSSRVH OLP ]
F
OLP \Q ] DOVR /HW ^ ` ^ ` V
[ W
D E D
IRU DOO
^\` EH
/HW
]
^W` EH WKUHH
D
VXEVHTXHQFH RI ^]`
VHTXHQFHV RI UHDO QXPEHUV
6XSSRVH WKDW
Q
6KRZ WKDW LI OLP
V
OLP W
6KRZ WKDW LI OLP
V
F
/HW ! 6 ! K !
DQG
F WKHQ DOVR OLP [Q
OLPI
V
6KRZ WKDW WKHUH LV
QA.QQO !K
7KHQ
E /HW ! K ! 6KRZ WKDW OLP QK 6XSSRVH OLP ] ] OLP Z Z 6KRZ WKDW D OLP_]_ _]_ E OLP] Z OLP ] OLP Z F OLP ] Z OLP ] OLP YY
DQ
F
F WKHQ DOVR OLP [ LQWHJHU . VXFK WKDW IRU
6HULHV
6HULHV
$ VHTXHQFH PD\ EH IRUPHG WHUPE\WHUP E\ DGGLQJ D OLWWOH ELW WR HDFK WHUP ,Q WKLV FDVH WKH OLPLW LI LW H[LVWV ZLOO EH DQ LQILQLWH VXP 6XFK VHTXHQFHV DUH
SUREDEO\ WKH PRVW LPSRUWDQW NLQG IRU LQ SUDFWLFH ZKDW ZH XVXDOO\ NQRZ DERXW VHTXHQFH LV WKH GLIIHUHQFH EHWZHHQ WZR VXFFHVVLYH WHUPV JLYHQ WR XV DV WKH VHTXHQFH RI VXPV RI WKHVH GLIIHUHQFHV D
/HW
^]`
EH
D
JLYHQ
7KLV VHTXHQFH LV
7KH VHULHV IRUPHG RI WKLV VHTXHQFH LV WKH
VHTXHQFH
VHTXHQFH RI VXPV Q
=8 =; =
=\
]
] L
7KH VHULHV FRQYHUJHV LI WKH VHTXHQFH RI WKH OLPLW LV GHQRWHG E\ <M L ]L
O
L
VXPV
] FRQYHUJHV LQ WKLV
FDVH
([DPSOH 7KH
JHRPHWULF VHULHV
]
/HW
61
--/
]
7KHQ
61O
]
] ]
6
]1
1RWLFH DOVR WKDW
61
]O
]
]
]61
7KHVH WZR
HTXDWLRQV JLYH XV WKH JHQHUDO
O]61
61
WHUP RI WKH VHTXHQFH
H[SOLFLWO\
]1
RU
61
L/B
1RZ LI ?]?
WKHQ OLP 61
1
R
6
]
]
/B B]
B a
I
LB]
]
BL
_=_
_=__=_
IRU
1RWLRQV
DQG OLP
_]_Z
_]_
_
RI &DOFXOXV
([DPSOH 6R JLYHQ ]? H IRU QA1 DQG WKXV
IRU 1
]
!
H !
ZH
ILQG
1
VR
WKDW
1%
1RWLFH WKDW ZH FDQQRW LPPHGLDWHO\ GHWHUPLQH ZKHWKHU RU QRW WKH JHRPHWULF VHULHV FRQYHUJHV IRU ?]? ! RI FRXUVH IRU ] 61 1 VR WKH VHULHV GLYHUJHV ,Q IDFW WKH JHRPHWULF VHULHV GRHV QRW FRQYHUJH IRU ?]? ! E\ DSSOLFDWLRQ RI WKH IROORZLQJ SURSRVLWLRQ
3URSRVLWLRQ
]HUR WKDW LV OLP ]
H
H[LVWV WKHQ WKH JHQHUDO
WHUP PXVW
FRQYHUJH WR
/HW
3URRI /HW
]
EH
!
Qa1 _]_
^V` EH WKH VHTXHQFH RI SDUWLDO VXPV %\ K\SRWKHVLV OLP V H[LVWV JLYHQ 7KHUH LV DQ 1 VXFK WKDW IRU QP!1 ?V VP _ H 7KXV IRU _MZ L_ H
7KXV
IRU _]_ B
IRU _]_ !
GLYHUJHV )RU LI
_]_!O WKH JHQHUDO
^_]_` JHWV DUELWUDULO\ ODUJH VHTXHQFH
^` GRHV
%\ WKH
]
WHUP LV
_]_
%\ ([DPSOH OLPO_]_ VR ,I _]_ _]_ DQG WKH
DQG GRHV QRW FRQYHUJH
QRW FRQYHUJH WR ]HUR
LV QRI VXIILFLHQW IRU WKH ZD\ WKH FRQGLWLRQ OLP ] FRQYHUJHQFH ] DV WKH IROORZLQJ H[DPSOHV VKRZ
RI WKH VHULHV
([DPSOHV
&HUWDLQO\ WKH
UHZULWH WKLV
VHULHV
WIOOO
+HUH WKH
O Q
GLYHUJHV
DV
JHQHUDO
Q
WHUP WHQGV WR ]HUR
L
QQ
%XW
ZH FDQ
2.2
Let sn
E?=i l/ diverges.
16-
1
?=i 1/n.
1
S2N~Sn~N+~1 Thus, {SN} is
=
not
+
a
'"
+
1
Series
139
Then
1
2N~N2N "2
Cauchy
sequence.
positive numbers is particularly easy to work with. sequence of positive numbers, then the series {Yj=i ck) {cn} is an increasing sequence, so by Theorem 2.1 (as rewritten in Problem 1) this sequence converges if and only if it is bounded. The
sum
of
a
is
For if
series of
a
Proposition 4. Let {c} be a following assertions are equivalent.
(i) (ii) (iii)
of nonnegative
sequence
numbers.
The
ck converges. is bounded. i
{Yj= ck}
e>0, there
For each
k
is
an
N such that for all m>N,
Zcfc< N =
proof of the equivalence of (i) and (ii) is essentially given in the preced ing paragraph. Part (iii) is just the Cauchy criterion restated for positive series (see Problem 11). The
Examples 17.
=i l/!
converges.
1
1 <
2"_1
n!
and thus for all N, N
y
1
<
by (2.5).
N-1
1
y <2
For n!
>
2""1 for all
n,
so
140
2.
Notions
of Calculus
18.
i
ta
AVn
)
cos(l/n)/
+
n
For
converges.
11
1
1
<
cos(l/n)
+
n
n
+ 1
n
n
Thus, for all N V
/I
1
/1
\
1_\
_
cos(l/n)j ,Mn ~
M\n
n
+
n
+ 1
)
1
1
1 _
2+2
'"
+
_
3 1
+
There is
complex putation
no
such
simple
criterion
as
Proposition
4 for
1
]v-]v+-T<1 series of
arbitrary
(or real) numbers, and the question of convergence as well of a limit can become extremely subtle. However, if for
as com a
given
series the series formed of the absolute values converges, the situation is considerably clarified. Ordinarily we shall discuss the convergence of a
series only in the happy circumstance that the values converges. Lef
Proposition 5. converges,
Yjck
be
{ck}
a
sequence
of complex numbers.
If
\c\
also converges.
Proof. Let t be the sequence of partial Notice, for m > n
2 ft
corresponding series of absolute
sums
of 2 I ft I and
s
the
partial
sums
of
.
\Sn
I
S
Thus, if {t} is
a
Definition 3.
Cauchy Let
ft
sequence,
{ck}
absolutely convergent, if converges, we say c is There
are
such
2
<,
things
be
lC*l =tm~t
so
a
also is
{s}.
sequence of complex numbers. c is converges. If |c| diverges, but c
|c| conditionally convergent. as
conditionally convergent
sequences.
In
fact,
Z"=i (-!)"/ converges. But as we have seen in Example 16 the series Yj?m ! 1/n of absolute values is divergent. It is easy to see that L t (- l)"/n
2.2
Series
141
converges. Let {s} be the sequence of partial sums. Then the subsequence {s2n} is decreasing, and bounded below by su and the subsequence {s2n+1} is increasing, and bounded above by s2 Thus, both these subsequences .
Since
converge.
l^n + l
s2n\
~
they have the
<
+ 1
n
same
limit.
It is easy to deduce that the full sequence also Here is the proof in a more general case
converges to that common limit. (known as Leibniz's theorem).
6.
Proposition that lim c
=
Let
Then
0.
{cn} be a decreasing sequence of positive (- l)"c converges.
Proof. Let j=2*=i ( l)*ft. We consider the sequences partial sums separately. The sequence {s2} is decreasing, since *2(n + l) S2
Similarly,
C2n+1
are
<S2 + i
partial bounded, for, given
=S2 C2 + i
of
even
and odd
^0
the sequence of odd
these sequences Sl
C2n + 1
numbers such
sums
{s2+i} is increasing.
Furthermore
any n,
<s2<,s2
so {s2} is bounded below by Si and above by s2 The same is true for the sequence s' both exist. Furthermore, fen+i}. Thus, by Theorem 2.1 lim s2 s, lim s2+i .
=
s'
lim s2 + 1 sequences, of odd s
=
limit, the
lim
s2n
partial
=
s2) lim(c2n +0=0, so .$' .$. Since both and even partial sums converge and have the same
\\m(s2n
sums
=
n-.oo
n-eo
+1
=
=
whole sequence also converges to that limit.
Notice that this argument does not give any hint as to the value of ( l)"/n. Outside of the case of Proposition 4, there is no positive asser -
tion that
can
be made about
tend to behave very
badly,
as
conditionally convergent series. following illustrative example
the
Example 19. The sequence
111111
2+2+4+4+4+4+-
In fact, shows.
they
142
2.
of Calculus
Notions
is the
same
and thus
+ 1-1
1 + 1 H
as
general term is decreasing conditionally convergent:
i
1
-,0,-,0,-,0,
...,,
2n
4
4
2
partial
sums
'
"
2n
2n
2n
4
n
The sequence of
V
h
I----H
44
4
2
2
1
11
111111
c
Since the
diverges.
to zero, by Leibniz's theorem this series is
times
is
0,
However, we may now converges to zero. it that no so longer converges ! Consider rearrange terms of the series we first add the positive terms in each the same series where group and thus
obviously
and then the
negative
terms :
(2.7) The
sequence of
corresponding
n-1
1111 >
2
0,
,
4
,
2
,
4
partial
u,
Thus, there is
.
a
.
.
,
,
2n
.
.
.
1_
,
2n
,
n u,
sums
.
.
is
.
subsequence: {$, \, ...}
and another:
{0,0, ...}
so
We leave to the student we cannot have convergence of (2.7). it to show that can be further (Exercise 9) rearranged so that it once
again
converges, but this time to
one
!
absolutely convergent series. We may please. If we arrive at a limit, it is attempt an the sum. In fact, if is absolutely convergent series we may sum first c the positive terms, and then the negative terms; and c is the sum of these two sums. We conclude this section with the proof of these facts. No such foolishness holds for
to sum the series in any way we
2.2
7. Let
Proposition
c be
absolutely convergent
series
143
of real numbers.
Let
(i)
\ck
+
k
\0
Then the
sums
if
ck
if
c,<0
ck
>
0
_ _
k
f-c
if
c
j
if
c,>0
0
ck~ converge and YJck
,
Let g be
(ii) (Rearrangement.) onto
an
Series
a one-to-one
Then
the positive integers.
(iii) (Regrouping.)
=
cg(n)
Let h be any
=
YJck
<
0
Z CH mapping of the positive integers
c
~
.
strictly increasing function from
P into P.
Let Hn)
Z
d= k=
Then
=
c
Proof. (i) Since the
2
>
Iftl
*=i
{2*=i Iftl)
is bounded by absolute convergence, and
2ft+.2ft"
the sequences
*=i
2*=i
say s, t .
.
sequence
*=i
verge to, Let e > 0
ck
h(n-l)
ft+,
2*=i
c*~
respectively, by
Then there
2ft+-*
*=i
are
are
also increasing and bounded. Thus they con s t. We have to show that 2 ft
Theorem 2.1.
M, N2 such that for
=
n
>
M,
<
2'
and for n>N2, n
2ft"Then for
n
>
<-
max(7v*i, N2),
2ft-(j-0 (ii)
Let ^ be
P onto P into
a a
2ft+- 2
ft"
-fr-0
<
Then g-- is defined and also maps one-to-one map of P onto P. For each n, let JV =max(^(l), ...,g(n)). one-to-one fashion.
144
2.
Notions
of Calculus
Then n
2ift(i^ 2
*=i
for all n,
iftl
*=i
the series
so
^2
2 ft
iftl
is
absolutely convergent.
Nn
n
2 ftV> < 2
Similarly,
it
c*+ ^ 2 c*+ and
2 ft-<*> ^ 2 c*-
for all , so we have 2 ft+c*> ^ 2 c*+> 2 c<*> ^ 2 ft"- Applying the same reasoning but reversing the roles of the two series, we obtain the reverse inequalities so that in fact, 2ft+(*)=2ft+ and 2ft-(*)=2ft"- Thus, by part (i) we obtained the
desired equality; that is, 2 ft<*> 2 ft Part (iii) is actually true for any convergent series. increasing function h be given. Notice that h(ri) ;> =
is
an
N such that
for all
n
2ft-
c
> N.
Thus, for
n>N,
<
24.= 2 2
*=1
and
h(n)
>
N,
so
cj=
*=lj=h(*-l)
that ll
2dn-c
<
*=i
2CJ~C
<
J=l
EXERCISES 7. Show that
n+lf
=i\n
converges. 8. What is
f "=i
(-D" a
where
a2
=
2",
2+i =3"?
2
y=i
cj
Let n
2 ft
for all
=
n.
c, and the
For
e
>
strictly
0, there
2.3
Tests for
145
Convergence
9. Rearrange the series
i_i+i-i+i-1+...+i-i+...-i+
so
4
2
2
it has the
4
4
4
2
2
2
sum one.
10. Can the series
( 1)"/ be rearranged
2
so as
to have
sum
10,000?
PROBLEMS
2 z-
9. Suppose 10.
=
z
Suppose (a) 2 z and lim
(b) 2
z
and
w
w
exist.
exist-
2w 2 z converges
and
11. Prove that
2
=
Snow tnat
w-
2 (z +
w)
=
z
+
w.
2 z w exist ? 2 z" w" ex'st ?
Does Does
if and only if for all
s
>0, there exists
an
N > 0 such that
2
*
<
Deduce that
2.3
Tests for
for all
e
n
>N
Proposition 4(iii)
is true.
Convergence
Since the theory of series is so unwieldy, there has developed convergence which have already given
are more or
some
important and the definition of convergence a large collection of tests (or criteria) for less easy to apply in the relevant cases. We
criteria for convergence.
> 0, if for every c converges if and only (1) Cauchy criterion: > > N. n for all m + cm\ < is an integer N such that |c+1 + then to (-D"cn converges. decreases zero, If the sequence {c} (2) if and only converges is c If the nonnegative, sequence {c} (3) is bounded. sequence {Yj=i ck} of partial sums
there
if the
for absolute The last condition, which can be considered as a condition the basic one. is criterion which convergence, gives rise to the following one (if we The idea is to compare a given series with a known convergent we one (if suspect that it suspect that it converges) or to a known divergent
diverges).
146
2.
Notions
of Calculus
Z I/"'
converges,
Examples 20.
noticed that 1/n!
1
/
have
seen
and since
<2~n+1,
in
Example
17.
There,
is convergent,
Z2_B+1
we
so
is
For
Zl/!. JV
as we
N
\
1
a>
1
Z^I ^r
=l\n!/
forallN
n=lZ.
21.
>
sin
-I
For if
diverges.
is small
x
enough,
sin
x >
x/2.
Thus, there is
an
N such that if n>N,
sin(-)
\nj
>:
2n
and thus for m>N,
fi
\n/
But
we can
enough.
2jv+in
\n/
=i
make the last
Thus,
sum as
Z=i sin(5/n)
large
as we
please by taking
is not bounded, and
so
m
it is not
large con
vergent. 22.
z
*b (1 + 0" is
absolutely convergent.
m
V
,tb|i
1 =
+
i|"
Y1
1 r=-
nh(^2y
The idea behind these
For
|1
+
i\
=
J2,
so
for any m,
i-
< oo
examples
since J v 2
>
1
is contained in the
following
theorem.
2.3 Theorem 2.3.
If
there is
a
(Comparison Test) Let {c}
positive number
K and
Tests for
be
a
sequence of complex numbers.
N, and
an
147
Convergence
a
sequence
{p} of positive
numbers such that
forn>N,
(i)\cn\
(ii) Z
Pn < >
n=l
then
Zc
converges
If instead,
we
absolutely.
have
forn>N,
W \c\>KPn, oo
Z Pn 11=1
(ii)' then
Z k|
>
=
diverges.
In the first
Proof.
2
iftl
=
2
case
k
=
2
iftl
=
2
2 iftl
iftl +
^
t=w + i
s=i
which is unbounded
as n
->
+
K=l
W+l
In the second case, the sequence of
kti
partial
2 ifti<2ifti
iftl +
k=l
k=l
the sequence of
partial
2 iftl
t=i
sums
*2/>
sums
+
is bounded.
I
is unbounded.
p-
t=w + i
oo.
Examples 23.
V^=0z7n!
an
integer
so
that
\z\N+" (N
+
Since a
N
so
z. converges absolutely for any complex all + for n)\ n, (N that JV> 2|z|. Then,
Choose >
(2|z|)n,
Jz\N 2"
n)\~
converges, so does result we obtain lim
1/2"
corollary
have been derived
directly).
|z|7" ! by the comparison test. As z7n! 0 for all z (this however could =
148
2.
of Calculus
Notions
24. Z nkz" converges absolutely for all z, and otherwise diverges. If |z| > 1, then
|z|
<
1 and all
limn*z#0,
integers k;
so
the series
hardly converge. Now suppose |z| < 1. We want to prove the convergence by comparison with the geometric series, so we must can
account for the effect of the coefficients as n
->
oo, thus also
greater than 1
(n
1)* <
+
r-^-
-
(n
.
(n
n
+
,.t
if
<
1
->
Then there is
.
or
s
l)/nk
+
an
nk.
Note that
(Exercise 13).
Let
N such that for all
s
(n
+
l)/n
->
1
be any number
n>N,
k
snk
Thus, by induction we can conclude that, for all n > 0, (N + ri)k < s"Nk. Thus, (N + n)k\z\N+n <(s\z\)"Nk\z\N. We should choose J
Z (s\ z I)" < - With the choice then apply the comparison test to obtain the
that
so
we can
series
Z
25.
of
s:
1
< s <
l/\z\,
convergence of
our
nkz".
Zw'z" diverges
for all z#0. We have seen in Example 2 c, lim c7n! 0, or, replacing c by z~l,
complex number
that for any
=
n->oo
lim
l/n!z"
This
0.
=
precludes
the
possibility
that limn!z"
=
0,
B-*00
the
so
26.
given
series cannot converge.
Z 1/n2
converges.
proof of this,
at
.i\n
1/
n
+
present
\n
n
+
1/
converges to 1 1 n
But
.
1 n
1
+ 1
n(n
thus 00
I
1
A n(n + 1)
=
In
1
+
1)
a
rely
N + 1
Thus, the series
^
we
later section on a
tricky
we
shall
observation.
give
another
2.3
Now, 2n2 1
n2
>
+
n
=
n(n
+
Tests for
Convergence
149
1), thus
1
s<
n2
n(n
+
1)
by comparison
so
Z llnil+l)
27.
that ke
large 1
m(l
>
2.
Z l/2
converges for any e > 0. Let k be Then, for any n ; if m ^ nk,
1 +
:
e)
n*(l+<0
Between nk and
there is
an
an
integer
so
1
<
-
m'"
also converges.
< =
(n
nk+2 +
If there
(n
+
if
,(n
+
If
are
-
nk integers.
Since
1
n0 such that for
n >
n0
<,
2nk, or (n
+
if
-
nk
<
nk.
Thus, n*
1
(BV)k
1
m=ii+im<1+-nk+2~n2 Well,
now we can
show that the sequence of
partial
sums
Ui 7^] is
bounded, for
Nk
Z n=l
nok
J
(l+e)
n
^ 11
Z1 =
N"
1
7(1+7) " JV
+
J
h _
=nofc+l
(B+l)k
n=no m=n+ 1
"
(1+e)
1 *
*o' + Z r2^o'I + Z^2<00 n=no
"
"
Now a special kind of a series is a power series: the geometric series, and the series in Examples 24 and 25 are such series. A power series is a series
150
Notions
2.
of Calculus
of the form 00
Zfl-z* B=0 series has the property that if it converges for some z0 then it con it diverges for some zu then it verges for all z such that |z| < |z0|, and if diverges for all z such that \z\ > IzJ. Thus, the geometric series diverges for |z| < 1 ; the series > 1 and z"/n! converges for all z, for Such
a
\z\
and
,
Z
converges This general property of power series is converges for no z. deduced from the comparison test. We make the following somewhat
Z!z"
easily
stronger statement.
Proposition
8.
Let
{c}
be
a
sequence
of complex numbers.
(i) If {\c \t"} is bounded for some positive number t, then Z cz" converges absolutely for all z, \z\ < t. (ii) If{\c \t"} is unbounded, then Z cnz" diverges for all z, \z\ > t. Proof. (i) Suppose
M>
|c|
t
"
for all
n.
Let
|ftz"l<:|ft|f"(Y<m(^\"
z
be such that |z| <
for all
t.
Then
n
|z|/f
and since
lim cnz"
0,
=
so
2
Definition 4.
ft z" cannot converge.
Let
series associated to
{c} be a sequence of complex numbers. {c} is the series Zb*=o anz"- The radius of
The power
convergence of the power series is the least upper bound R of all real numbers f such that the sequence {| c |f"} is bounded.
According
to
Proposition 8 the series Zb=o anz" converges for z inside jR(|z| < R), and diverges for z outside that disk (see
the disk of radius Problem
12).
Examples has radius of convergence one. For if t > 1, then is unbounded, and if t < 1, t"/n -* 0. Notice that we can make clear assertion for z on the unit circle, since Zb=o 0)7W diverges,
28.
Zb=o z7n
{f7n} no
but
Zb=o ( 1)7"
converges.
2.3 29. If
{c}
Tests for
Convergence
151
is bounded, but does not tend to zero, Zb*=o cbz" nas For clearly {cf"} is bounded for f < 1,
radius of convergence one. and unbounded for t > 1 .
There
are
two final tests of
cn
I)1'"
fr
< r
SOme r <
Z cb converges absolutely.
then
(| then
c
|)1/n
>
for
R
If there is
< r <
Z cb
are as
follows :
some
1
If there R
>
are
infinitely
many
n
such that
1
R
>
an r <
1 such that
eventually
1
absolutely.
converges
>
then
These
Z cn diverges.
Ratio test.
then
importance.
If eventually
Root test.
(I
some
for
1
If
infinitely
many
n
Z c diverges.
by comparison with the geometric series. We leave it to the student to derive these tests (Problem 13). Let us here indicate why the convergence assertions are true. Suppose (| c |)1/n < r < 1, for n large enough (say n > N). Then \cH\ < r" eventually, so the partial sums Z kl These
are
are
both derived
bounded
by 1
n
Z0 Kl =
+ 7 t
by comparison with the geometric series.
< r
for
n >
N
As for the ratio test, suppose
152
Then
2.
we
Notions
of Calculus
have
kjy+il
|Cjv+2l
r"\cN\
<
by induction.
Tj=0 Id
Thus,
<
Z"=o k.1
+
\cM\
Z rk <
since
oo
r <
1.
EXERCISES
following series converge?
11. Which of the
ns + 8
v
(a)
2sin(i).
(e)
(b)
2sin(i).
(f)
v
(0
2^(1).
(g)
2-^2"
(d)
2
(h)
27^r,x,,,x>0. (2n)\
(i)
2
(j)
2(-D"sini.
tenf-) sin(^ \n} n) -
.
^4n6 + n*' ^
n3 + n2 + n+l n* + n5 + n6 + 7
'
n*
x"' k
a
positive integer,
0 <
(1)
2(-i)n ^
+ 1
1
.
+ -LJ\. + z(L (n+1)2 (+2)2/ \n2 ^
(n)
2(-\
"
(-l>- n
<
(m)
n
(k)
x
n
^-7 + -Ul + 2/
+ 1
n
-
(+l)2
12.
Verify directly that
13.
Suppose lim
c
=
c.
lim
z"\n\
=
0 for every
Then for any
z.
integer k,
lim c*
=
c*.
2.4 14. Find the disk of convergence of the
2Z-
00
.?
following
(f)
=
Convergence
153
power series.
2>!z". n
-
in R"
=
0
(g)
2
(h)
2d+n)z".
(2n)2
z"
(C)
Jo^-
(d)
.S,^
n\
/
00
z
2 z"
\"
?o(2nj-
(e)
(j>
2(1+*)"
PROBLEMS 12. Let {c} be a sequence of complex numbers, and let R be the radius of convergence of the power series 2 ft*"- Show that 2 ft*" converges absolutely for |z| < R, 2 ftz" diverges for |z| > R. 13. Derive the convergence and
divergence
assertions of the root and
ratio tests.
2.4
Convergence
in R"
The notion of convergence of a sequence of vectors is easy to conceive, a vector in R" is just an n-tuple of real numbers. Thus, a sequence of vectors is an n-tuple of real sequences, and the question of convergence of since
the vector sequence is just that of the simultaneous convergence of those n real sequences. We might also directly paraphrase Definition 2 of conver gence,
using the
possible notions Definition 5.
notion of distance in R" discussed in are
in fact the
Let
{yk}
be
Chapter
1.
These two
same.
a
sequence of vectors in R".
The sequence
converges if there is a vector v e R" such that to every positive number > 0 there corresponds an integer K such that || yk v|| < e for k > K. We write lim yk v if {yk} converges to v. -
=
ft-* oo
154
Thus, lim
general
in Section 2.1
=
\ck
c\
-
yk can
;'
1,
=
we
.
when
=
said that
0,
->
Now, if
2.
we
precisely that lim || yk
term yk and
way it sounds like
put this
n
v means
=
yk
between the
when
of Calculus
Notions
2.
.
vkJ
,
v
-
zero as
||
=
0 ; that is, the distance When
k becomes infinite.
just the notion we have in mind. Recalling that a complex sequence {ck} converges to c precisely
that this coincides with the above definition when write out the sequence yk of vectors in R" as an n-tuple
(2.8)
(vk\...,vk-) We
n.
->
tends to
we see
we
view the
.
v
vJ for
given sequence as the n real sequences {vkJ}, where v precisely now verify the fact mentioned above, that yk in fact the case is that 2 that Notice all j. Proposition ->
ofR2.
Proposition 9. The if and only if lim vkJ
=
k-*
If
Proof.
w
sequence (2.8) converges to the vector v] for allj.
v
=
(v1, ...,v")
oo
=
(w1,
.
.
.
,
vc")
is
a
vector in
R", then by definition
I|W||=(2(H'')2)1'2 Then, in particular
|ftJ-tfJl
j
(2.9)
\,...,n
that v -> v. Then, given e > 0, there is a vJ\ Thus, by Equation (2.9) for each j, \vkJ precisely that lim vkJ v1.
Suppose
now
for k 5: K. means
=
-
K such that ||v* <
e
for k ^ K.
v
||
< e
But this
=
k->oo
v')2 ->0 for ally, Conversely, if vkJ -+vJ for ally, then (vkJ But then, by Definition 5, v*-^v. v ||-^-0 as k -*oo. llv*
so
[2 OV
02]"2
=
precisely the same way we can verify that if the sequence of vectors (2.8) a Cauchy criterion so do each of the real sequences {vkJ}, and thus are convergent. Hence, by Proposition 9 the sequence of vectors {yk} also so we have a Cauchy criterion for vector sequences also. This converges, fact, as well as some basic algebraic properties of convergence of vectors is easily verifiable. Accordingly, we make these assertions, leaving the proofs In
satisfies
to the reader.
Proposition 10. (Cauchy Criterion) Let {yk} be a sequence of vectors in R". whenever Suppose to every e > 0 there corresponds a K such that \\ vr vs || < both r,s>K. Then the sequence {yk} is convergent.
2.4
Proposition 11. Suppose lim yk {wj are sequences of vectors in R",
y, lim wk
=
and
{Ck}
Convergence w, lim ck
=
is
a
sequence
=
in R"
c, where
155
{vj,
of real numbers.
Then
(i) (ii) (iii)
lim(vfc + yyk) lim
=
=
+ w,
v
cv.
Example 30. Let
find
point of a given plane in R3 which is closest to plane is given by the equation <x, a> c for fixed Let m =g.l.b. {||x||; <x, a> a, c. c). Choose a sequence {x} on the plane such that ||x|| ->m. We shall show that {x} actually
the
us
origin.
a
A
=
=
converges.
||x We
xj|2
-
Now, =
||x||2
+
||xj|2
2
-
(2.10)
<x, xm>
estimate the last term by using the fact that the midpoint + xm) between x and xm must also be on the given plane.
can
i(x
2(xb + 0
+
4
^^ 4
+ ^
2
<x xm> ,
Thus,
-2<x,xm>< ||x||2+ ||xj|2-4m2 and
Combining (2.10) l|x
-
xm||2
<
2(||x||2
Now, since ||x|| n,m>n0, then
||x
-
we
+
->m,
have
(2.11),
we
||xj|2
if
>
||x||
(2.11)
find that
2m2)
-
0 is
(2.12)
given, there is an n0 such that for ||xj| <m + e. Inequality (2.12)
< m + s,
gives
xj|2
<
2((m
+
)2
+
(m
+
b)2
-
2m2)
<
4ms + 2e2
=
fi(4w
+
2e)
please by choosing e small. Thus if n,m large enough, ||x-xj| is small, so the sequence {x} is lim x,then ||x|| m, lim||x|| and thus convergent. Ifx Cauchy, so x is the closest point on the plane to the origin. This
can
be made
as
small
as we
are
=
=
=
156
2.
Let
above
Notions
pause for example, for
us
of Calculus a
moment to consider the reasons,
studying
lem of calculus is to find
an
as
illustrated by the
the convergence of vectors. The central object, usually considered as a point in a
prob given
specified properties (i.e., the maximum given function, or a zero of a function). At least, the theoretical aspect of the problem is to prove the existence of a point with such and such properties. Our technique for doing this is to use the desired properties to develop a sequence of approximations; our hope is that the approximations will converge; and that the limit will have the desired properties. It is thus essential to be able to discuss the question of convergence without already knowing the limit. Hence, for example, we have- the Cauchy criterion. Further, we will need techniques, or criteria, to apply to the given properties in order to be able to extract the desired Cauchy sequence of approximation. For example, we will want to know : (a) If we have a convergent sequence of points having a property, does the limit have that property ? (b) If we have a sequence of points having a property, does the sequence converge? These questions lead or, at least does it have a convergent subsequence? collection of points, which has certain of a
us
to the reconsideration of the closed sets introduced in Section 1.11.
Recall that
precisely,
a
closed set in R" is
S is closed if and
a
set whose
complement
only if corresponding
to every v
is open. More $ S, there is an
0 such that any vector within e of v is also not in S. In particular, if S closed set, and v ^ S, then v cannot be the limit of a sequence of vectors in S. To put it positively, a closed set contains the limits of all convergent >
is
a
sequences it contains.
Proposition 12. equivalent:
This is in fact
Let S be
a
a
Suppose
v
e
=
v.
This is
is
a v
contained in S, then lim yk
nonsense
such that ||v
since
Then there is
vector in S which is within
l//j there
assertions
e
are
S.
Let {yk} be a sequence contained in S and suppose $ S, since S is closed, there is an e > 0 such that no vector
Thus, we must have v e S. Suppose now S is not closed. a
following
S is closed.
it converges to v. If in S gets within e of
there is
criterion for closedness:
The
in R".
set
(i) S is closed. (ii) If {yk} is a convergent sequence Proof.
defining
-
e
of
v
is the limit of
a v
v.
v|| <, l/ and
in v
a
sequence in S.
S such that for every
e
>0
particular, for each n, taking e S. Thus, v ->v so (ii) does
not hold for S.
We are now in a position to state our last basic consequence of the funda mental existence axiom for the real number system. This is that every bounded sequence in R" has a convergent subsequence. It is easy to derive
2.4
Convergence
in R"
157
this from the Cauchy criterion, itself an assertion of existence. Let us illustrate the situation in R2 Suppose {ck} is a sequence of complex numbers which is bounded ; that is, it remains in some fixed square S0 of side length K. Cut that square into four equal squares. At least one of these new squares .
has
equal pieces of the {ck} ; way
many of the
{ck} ; let St be one such square. Cut St into four and let S2 be one of these new squares which has infinitely many now do the same with S2 and so on (see Figure 2.4). In this
infinitely
obtain
we
a
sequence of squares
{S}
with the
properties:
Sm=>S+u
(i) (ii) (iii)
length of S is K/2n, S has infinitely many of the {ck}.
side
Now that this is done, we can, for each integer n, select a k(n) such that and {ckW} forms a subsequence of {ck}. (For this we need to
ck(n)eS,
know that than any For let
have
S contains infinitely many {ck}, so that we can choose k(n) greater previously chosen index.) Now, {ct(B)} is a Cauchy sequence.
e >
0, and choose N
cm,cHm)eSN, IC*(b)
C*(Bl)l
<
so
that
>
Kyj2/2N.
Then, if
n,
m >
N,
we
so
(K\2 \2NJ
(K\2 Kj2 < 2N \2N) ~
Since the sequence {ct(B)} is a Cauchy sequence, by Proposition 10 it con idea of the verges, and the argument for R2 is concluded. This is the basic verification of
Figure 2.4
158
2.
of Calculus
Notions
Theorem 2.4.
subsequence
Every
sequence in
which converges to
a
closed and bounded set S in R" has
a
a
point ofS.
Suppose that S is closed and bounded and {yk} is a sequence in S. We Cauchy subsequence. Since the sequence is bounded, it is contained in some ball B(0, R). This ball can be covered by finitely many balls of radius 1. Since the {yk} are infinite, there is one such ball which contains infinitely many. Call it Bi, and let ykil) e Bi. Bi can be covered by finitely many balls of radius i. Let B2 be one such which contains infinitely many of the {vj and let v*(2) e B2 with k(2) > k(l). Continuing in this way we obtain a sequence {B} of balls, a subsequence {vt<)} of {yk} such that (i) B has radius 1/n, (ii) vJ(n) e B, (iii) B => B+i. Then {vMn)} is a Cauchy sequence, for if n, m> N, yk{n) and vt(m) e BN which has radius 1/N, so
Proof.
shall find
a
2
\\ykw
vt(m)||
for all n,
<
m
>N
N
By Proposition 10 there is a v such that vt() - v as n -^ oo. Since S is and {v*()} 6 S, we also have v e S, so the theorem is proven.
closed set,
a
Example unit
31. The
sphere
S
=
{xeRn: \\x\\
=
1} is closed.
For
if
Now suppose x, then certainly ||x|| -> ||x||, so if x e S, so is x. x T is a linear transformation of R" to R". We want to know if there ->
is
an
xeS at which
||rx|| is a maximum. First of all, ||7x|| with x e S is bounded. Let A representing T, and M max \a/\. Then
numbers of the form the matrix
Tx
=
T(x\
the set of =
(a/)
be
=
.
.
.
,
x")
=
(Z a/xJ,
.
.
.
,
Z af-x*)
so
II Tx||
=
<
Thus,
[(Z *y V)2 \nM2 \\x\2
+ +
+ +
(Z flyV)2]1'2 nM2
nM is the desired bound.
||x||2]1/2
<
(2.13) nM
||x||
By the least upper bound axiom then,
sup{||7x|| : xeS} exists, and there is a sequence {x} c S such that ||7x|| ->w. According to the above theorem there is a sub Since ||7x|| ->w, we also sequence {y} which converges, say to y. have ||7y|| ->m, and by (2.13), in fact ||7y|| m. lim||Ty|| m
=
=
=
2.5
Continuity
159
PROBLEMS 14. Prove
Proposition 10. Proposition 11. 16. Let n be a plane in R3, and suppose x0 is the point on n which is closest to the origin. Show that if x e Tl, then x0 is orthogonal to x x0 (Hint : If not, then one of x x0 x + x0 is closer to the origin than x0.) 17. Find the point on the plane given by the equation <x, (1, 1, 1)> 3 which is closest to the origin. 18. Find the point on the plane <x, (1,0, 1)>=2 which is closest to -0, 1, 1). 15. Prove
-
.
,
=
19. Let L be range of L
are
a
linear function from R" to Rm
20. Let L: RP-+ R be also lim
Show that the kernel and
both closed. linear function.
a
Show that if lim x
=
x, then
L(x) =L(x).
21. Let
v0
be
a
vector in
R", and n the
set of
x
such that <x, v0>
=
c.
Show that IT is closed. 22. Show that for any v0
e
R" and
r
>
0,
{vefl": ||v-v0||
max
v'\
\vk
->
v
in R" if and only if
->0
lslsn
2.5
Continuity
We turn now to the consideration of functions from subsets of R" to Rm. The basic notion of analysis being that of convergence, the fundamental class of functions will consist of those which respect convergence ; that is,
those which take convergent sequences into convergent sequences. are continuous functions.
Definition 6. values in Rm.
Let S be
/is
a
set
continuous
on
These
in R", and /a function defined on S, taking S if whenever yk -> v with vk e S, all k,veS,
then/(v,)-/(v). We shall be concerned most a
given point.
usually
For this purpose
we
with the local
study
of
a
function
make this additional definition.
near
2.
160
of Calculus
Notions
said to be continuous at v0 v->v0
will be / from a set in R", taking values in Rm, of v0 and e R" if /is defined in a neighborhood
A function
Definition 7.
implies /(v) ->/(v0).
Examples
|| v
v||
-
-+0
so
that
||v||
For
continuous.
is
f:R"-*R,f(y)= \\y\\
32.
if
v->-v,
then
||v|| since
->
| ||v||-||v|| |<||v-v||
f:C-*C, f(z) 0, so that z| |z
33. =
34. A linear function
also z on
continuous:
is
z
=
->
-
->
z^z
implies
|z-z|
z.
R" is continuous.
Let
i=l
Then, if yk '
Z"= i
;f
V
have
v we
-?
since the limit of
->
a
u1,
sum
.
.
.
,
vk"
is the
->
v",
sum
so
that
Z?=i a'
of the limits.
-"
Thus,
/(VftWWthe idea of continuity of a function /is this: as a moving point to close p0 the value /(p) off at p gets close to/(p0). That is, we can p gets ensure that /(p) is as close as we please to /(p0) by choosing p sufficiently criterion for continuity, s 8 This leads to the so-called close to p0
Roughly,
,
"
"
-
.
which
we now
give. Let S be
Proposition 13. defined on S.
a
subset
ofR", and let f be
an
Rm valued function
Let x0 e X. f is continuous at x0 if and only if, to every e > 0, there corresponds a5>0 such that \\x x0 1| < 8 implies ||/(x) -/(x0) || < . (ii) If S is open, f is continuous on S if and only if f is continuous at every
(i)
-
point of S. Proof, there is
a
8
>
x0
.
Let x
^ N,
||/(x)
x0
.
0 such that whenever
Since x-*x0, there is n
->
/(xo)||
an
N such
< e, as
S criterion is true,
we shall show that / /(x) -*/(x0). Given e > 0, x is within 8 of x0 we have ||/(x) /(x0)|| < e. that n>N implies ||x x0|| <8. Thus, for
(i) Supposing first that the
is continuous at
e
We have to show
desired.
2.5
Conversely, if the 8
>
0 there is
8
1
an
e
xd
8 criterion is false, then there is x II < 8 but ||/(x)
-
for which ||x
!
l
1
2
3
n
Continuity
an e0
-
-
161
such that for every
/(x0) ||
> e0
Selecting
.
obtain the corresponding sequence Xi, Xi/2, ..., Xi,, which converges to x0. But/(xi/n)+>/(x0) since the/(xi/n) are always outside the ball of radius e0 centered
we
at x0.
Part
(ii) is
left
as an
exercise.
Examples /: R2
35.
/(*>')
-+
R defined
by
=
FT?
is continuous at
For
(0, 0).
5x 1 +
<5|x|<5||(x,^)||
y2
Thus, if
is
given
we can
choose <5
=
e/5.
Then
||(x, y) \\
<
8
implies
5x <
v2
1 +
58
=
36.
f(x,
y,
z)
y3z
=
1 + x2 + z2
is continuous at (0, 0, 0).
|/(x,
y,
z)
-
f(0, 0, 0)|
Thus for each
|/(x,
y,
>
z)\ <84
We have
y3z 1 +
0 choose 8
=
e.
<
=
x2
=
+
e4
z2
=
e.
\y3z\
Then
<
\\(x,
||(x,
y,
y,
z)
z)\\<8 implies
162
Notions
2.
of Calculus
37.
/(*>y)
(4^4-
=
/(0,0)
(x,y)*(0,0)
=
0
This function is not continuous, since
4x2
f(x,x) If
we
=
2
=
redefine
2^0
/(0, 0)
=
2, this
new
function is still not continuous,
since
/(0,y)
^=1^2
=
38. We
by
the
can
easily verify
8 criterion.
s
l/(v) -/(w) |
=
|
the
of the linear function
continuity
Z (' ~w')\< ||(a\
by Schwarz's inequality. Thus, 8 IKa1, ...,a") || h. Then || v
if
.
.
.
-
>
application
concerning convergence study of continuity,
to the
v0
1|
<
a") || || v
,
8
might
w||
given, we can take implies |/(v) -/(v0)| < .
discussed in as
-
0 is
"
=
The facts
(2.14)
For
be
previous sections have expected. In particular,
the assertion that every sequence in a closed bounded set has a convergent subsequence has profound significance for the behavior of continuous func
tions.
Here is
an
important illustration.
Proposition 14. (Intermediate Value Theorem) Let f be a function on the interval {x e R: a < x < b}, and suppose that f (a) Then there is
a c
in the interval such
thatf(c)
=
continuous <
y
y.
We seek (as in Figure 2.5) not just a point at which the value of /is y, precisely the first such point c. We must find a way to describe this point which permits us to use the existence theorem. If x < c we must have f(x) < y, otherwise the graph of /crosses the line y y between a and c. Thus, c is a lower bound for the set of x such that f(x) > y. Since c is in that set, it must be the greatest such lower bound. So if there exists a first c at which /(c) y, it is the greatest lower bound of {x e R : a <, x <, b, f(x) ^ y}. We now show that this point (which exists by the least upper bound property) is the desired c.
Proof.
but
more
=
=
2.5 Let
Continuity
163
c g.l.b.{x :a<x y}. Then c is a limit of a sequence {*} in this Since y (*) we must also have y < lim/( jc) =/(c) since /is continuous. Now, if /(c) # y, we must have f(c) > y. Again, by continuity, there is a 8 such =
set.
that if \x
c\
<
8, then
l/W-/(c)|<^^ from which it follows that for all
x
between
c
f(c 8) > y, contradicting the definition of c as a with/(x) > y. Hence /(c) > y is impossible, so we Now, the that
they
most
important
bounded
are
and
c
8, f(x)
> y.
Thus,
lower bound for the set of must have /(c)
=
x
y.
fact about continuous real-valued functions is
closed and bounded sets.
This follows
easily on the / set S, then, for every positive integer n, there is an x e S such that/(x) > n. If S is closed and bounded, {x} has a convergent sequence {x(t)}. Let lim xn{k) Since / is continuous, /(x0) x0 lim/(xw) > lim n(k). But from Theorem 2.4.
=
If,
on
is continuous and not bounded above
say,
=
.
fc-*00
n(k)
->
oo as
k
-*
oo,
so
this is
impossible.
Thus/is
fc->00
bounded
on
S.
What is
it attains its least upper bound. For if m is this least upper bound, but is not a value of/ then g(x) (/(x) m)'1 is an unbounded function more
=
on
S, again
function
on
a a
contradiction.
To conclude:
such that
/(x1) /(x2)
=
=
if/ is
a
continuous real-valued
closed and bounded set S in i?", then there
sup{/(x):xeS} inf{/(x):x6S}
Figure
2.5
are
xux2e S
2.
164 Here
are
Notions
the
of Calculus
proofs in
=
more
general
context.
continuous Rm -valued function
a
Then the set
bounded set S in R".
f(S)
f be
Let
Theorem 2.5.
slightly
a
of values
off
on
the closed and
on
S,
{f(x):xeS}
is closed and bounded.
Suppose ye/(S) and y^yeRm. We must show that y ef(S). But this is easy. Since y ef(S), there is for each n,xeS Since S is closed and bounded there is a subsequence {zk} of such that /(x) y Since / is continuous, f(zk) ->/(z). On the other which zk^-zeS. converges, {*} lim/(z4) y and hand, [f(zk)} is a subsequence of {y}, so f(zk) -> y. Thus /(z) First, f(S) is closed.
Proof.
=
.
=
=
ye/(S). If f(S) is not bounded, there is for each n an x e S such that ||/(x) || ^ n. But {x} has a convergent subsequence {z}. Let lim zk z. Then lim/(zt) =/(z). But {f(zk)} is a subsequence of {/(x)}, so ||/(zt)ll -+ o, which is impossible since {/(z*)} is convergent. Thus, /(S) must be bounded. =
In
particular,
suppose /is
a
real-valued function defined
on
the closed and
Then/(S) is bounded, so M= sup{f: tef(S)} exists, closed, M e/(S). Thus there is an xt e S such that
bounded set S. since /(S) is
/(Xl) Similarly, we
state
=
sup{/(x):xS}
there is
x2 such
an
that/(x2)
=
inf{/(x) :
x e
S}.
This basic fact
as
Theorem 2.6. on a
and
A continuous
function
attains its maximum and minimum
closed bounded set.
PROBLEMS 24. Let x0
e
Rn.
25. Show that 26. Prove part
a
Show
that/(x)
=
<x, x0> is continuous
linear function L : R"
(ii)
of
27. Show that if / is
-*
on
R".
Rm is continuous.
Proposition 13. a
continuous real-valued function
on a
closed and
bounded set S, there is an x2 such that/(x2) g.l.b.{/(x): x e S}. 28. Suppose that /, g are J?m-valued functions continuous at p0 6 R". If c e R, then also Show that /+ g and , gy are also continuous at p0 =
.
cf is continuous
at p0
.
2.6
2.6
Calculus
of One Variable
165
Calculus of One Variable
Theorem 2.6, which asserts that a continuous function attains its maximum on a closed and bounded set, is the fundamental theoretical
and minimum
tool of the calculus. of
We shall
now
give
a
brief review of the fundamentals to the student's memory.
calculus, leaving the recollection of techniques
We shall
give brief justifications of some of the more basic or special facts. First of all, we studied in the calculus a limit concept which was more general than the sequential limit we have been studying. We recall the definition. Definition 8.
Suppose / is
{x:0 <\x
We say lim f(x) x0\ <8 implies -
First of all, the one:
lim/(x)
L
=
a
set
x0\ <80}
-
L if and
only if, for all > 0, there is <5 > 0 such that \f(x) L\ < e. relationship between the two concepts of limit is an easy if and only if for every sequence x converging to x0 we =
\x
real-valued function defined in
a
-
,
x-*xo
have
lim/(x)
=
L.
We
can
thus
rephrase
the notion of
continuity using
B~*00
Definition 8.
/is
continuous at x0 if and
only
if
lim/(x) =f(x0). x-*xo
Proposition 15. (i) Suppose f is defined
in I
{x:
=
0
<
|x
x0
Then
\ <8}.
lim/(x)
=
L
x-*xq
if and only if, for
limf(x)
(ii) /// is
every sequence
=
also
in I such that x
{xn}
-*
x0
we
have
L
defined
at
x0,f is
if and only if
continuous at x0
\imf(x) =f(x0) X-*Xo
Proof. We will prove only (i). The proof of (ii) is the same and is left as a problem. Suppose first that \im f(x)=L. Let {*} be a sequence such that X-*XQ
x-+x0.
Given
e
that
|jc-x0|<8.
\x
x0
-
\< 8.
>
0, there is
a
8
> 0
Now since x-*x0,
Thus if
n
>
such that \f(x) there is
N, \f(x) -L\<e.
an
-
L\
< e
for any
N such that for
Thus, f(xn) ^L.
x
such
n>N,
166
2.
Notions
Now suppose
of Calculus
lim/(x)
=L is false.
Then there is
such that for every 8
an e0
x-*xo
we can
such that |x,-x0| <8but \f(x)-L\ ^e. Consider the sequence Then |c xl < l/, so certainly c ->x0 1, J, , 1/n, is always outside the interval of width e and center L, so it cannot converge
find
an x
of x's for 8
{c} But/(c)
-
=
...
.
....
toL.
Definition 9. e
x0
R.
/is
Let /be
a
real-valued function defined in
an
interval about
differentiable at x0 if the limit
lim/(X +
"
/(Xo)
f
r->0
If it does the limit is called the derivative of /at x0 and is denoted
exists.
en
^
/ C*o)
^
/
^
~ (*o)
01"
If/is differentiable in an interval J and the derivative/' is also differentiable there, then / is said to be twice differentiable on / and (/')' is the second derivative off and is denoted by /" The
or
dx2
higher derivatives/'",
manner.
.
.
.
,/(n),
...
are
defined
successively
A function which has derivatives of all orders
on
in the obvious
the interval will
infinitely differentiable there. If/ g are n-times differentiable I, are/+ g,fg, and cfTor c a real number. If/ is differentiable in an interval I it is continuous there. If/ is differentiable at a point x0 where it 0. This, together attains a local maximum (or minimum), then f'(x0) with Theorem 2.6 gives this basic existence theorem. be said to be on
so
=
Theorem 2.7.
interval
[a, b~].
(Mean Value Theorem) Let f be differentiable a point t; e (a, b) such that
on
the closed
There is
mJ{b)-f(a) b
(215)
a
Proof. This theorem has a nice geometric interpretation (Figure 2.6). There is a point (|, /(f)) on the graph y =f(x) at which the tangent line is parallel to the line through (a, f(a)) and (b, f(b)). Clearly (see Problem 30), we need only verify this when the latter line is horizontal, that is, f(b) =f(a). In this case, let f0 e [a, b],
2.6
Figure
Calculus
of One Variable
167
2.6
[a, b] be the points at which / attains its maximum and minimum respectively (Figure 2.7). If either f 0 or f i is interior, then/has a local maximum so there, /'(f) 0 for the appropriate f. If this is false, then {f0 f 1} are the points {a, b), so /(a) =f(b) is at once the maximum and minimum of/. Thus, /is constant on [a, b], so /' is identically zero and we can choose any point for our f
fi
e
on
the interval
=
,
.
Now suppose that / is a differentiable function defined on the interval [a, b], and g is a function defined on the range of/ and differentiable there. Then the composed function h g f, defined by =
h(x)=g(f(x)) is also differentiable
h(x)-h(x0) X
~
~
Xn
on
a, b.
For if x0
e
\a, b], then
g(f(x))-g(f(x0)) f(x)-f(x0) f(x)-f(x0)
(W(*,.))
:2_ (.))
Figure 2.7
(2.16)
168
2.
Taking
the limit
Notions
Hm
X
at/(x0),
g(f(x))-g(f(x0)) f(x) f(x0)
Hm
f(x)-f(x0)
~
X0
the limit
implies /(x) ->f(x0)),
have (since x->x0
we
Hm
=
f(x)->f(xo)
the right exist
on
so
both sides,
h(x)-h(x0)
x->x0
The limits
on
of Calculus
since/is differentiable at x0
,
X
X0
and g is differentiable
Thus h is differentiable and
the left exists.
on
x-+x0
we
obtain
the chain rule:
h'(x0) that
(Notice
(g of)'(x0)
=
if/(x) =f(x0),
However, that If /is
exists
a
f(x) fg(y)
g
we
say
us
a
case can
g'(f(x0))f'(x0) (2.16) is invalid and the proof breaks down. separately.) interval [a, b] to the interval [a, j8] and there [a, b~] such that
then
be treated
function from the
function g:
a
=
[a, /?]
->
=
x
for all
=
y
for all y
x e e
that/is invertible and g is
condition under which
a
[a, b] [a, 0] its inverse.
The
mean
value theorem
differentiable function is invertible.
/has an inverse, it must be this will be guaranteed if/' is never for the invertibility of/ function
one-to-one. zero.
From
(2.14)
we see
gives If
a
that
This is the sufficient condition
Theorem 2.8.
Suppose thatf is a continuously differentiable function defined [a, b], andfi is never zero. Letf(a) a andf(b) p\ There is a continuously differentiable function g defined on the interval between a and ft such that on
the interval
3(f(x))
=
=
x
and
g'(f(x))
=
-
f(x)
Proof, f is one-to-one. For if a < ai a f between ai and bi such that
<
for all
=
x
bi ^ b, there is, by the
mean
value
theorem
f(bi)-f(ai)=f'()(bi-ai)*0 Thus
by hypothesis. between
a
and
j8
f(bi) ^/(ai). By the intermediate value theorem every v by /. Now we can define g as follows : let g(y) be that
is attained
2.6
x
such that
f(x)=y.
Clearly, g(f(x))
=
Calculus of One Variable and
x
f(g(y))=y.
169
Now g is differ
entiable:
g(y)-g(yo)
..
hm
x-x0
..
=
y-yo
-ro
lim n-yo
-
f(x0)
l
hm
=
-
f(x)
x-.xo
[f(x)
-
A further fundamental fact to be drawn from the
determined,
this: A function is Theorem 2.9.
and that f'(x)
=
Let h
by
=/ g.
h(c)
-
By hypothesis h'(x) [a, b], there is a f a
=
c e
,
0 for all
<
f
This for all
on
0,
h(c)
so
a
x e
[a, b].
By the
mean
h(a)
Now, given any real-valued function / defined
=
value theorem is
such that
so
'(f )
differs from g by
f'(x0)
its derivative.
[a, b],
But
mean
X0
+ C
value theorem, for any
=
constant,
-
Suppose that f, g are differentiable on the interval [a, b] g'(x) for all x e [a, b]. Then there is a constant C such that
f(x)=g(x) Proof.
up to a
1
f(Xo)]lx
=
h(a).
constant,
as
c e
is constant and thus /
desired.
interval J,
we
consider
=/. By Theorem 2.9, any two such functions differ by a constant; thus by specifying the value of such an/at any point it is completely determined. We denote by Jj / F(x) that function (if it exists) such that F(a) 0 and F'(x) =f(x) for all x e fa, b]. j* /is called the indefinite integral of/. Every continuous function has an indefinite integral, which is given by the process of Riemann integration which we now describe. Let /be a bounded function defined on the interval J. A partition P of I
those differentiable functions F defined
on
I such that F'
=
=
< a such that sequence of points a0 < ay < the approxi to / [a0 a]. We now construct two sums, corresponding 2.8 : in mations to the area under the graph off given Figure
consists of
an
increasing
=
,
S(P,/)=
a(P,f)=
J>(fl|-f-i) Zwi(fli-ai-i)
i=l
170
2.
Notions
of Calculus
Figure where
Mt,mf [fl,-!, a,].
/ is
the maximum and minimum values
are
Definition 10.
interval.
Let
=
be
off
on
the interval
bounded real-valued function defined
a
on
the
integrable if
(2.17)
suPff(F,/)
p
p
(i.e., if we we
/
Riemann
infE(P,/)
2.8
can
please).
the interval
find
partitions
for which the two
sums
2 and
o are as
close
as
In this case the common value is called the definite integral of/ over
/, and denoted
j"7/.
If/ g are integrable on the interval J, then so is/ + g and cf, ceR. Further lif+li9> \icf=c\if- If/is integrable on the interval J, then/is integrable on every interval J c I. If/ is integrable on the intervals [a, i] and [b, c] with a
liW+9)
=
f
he c]
/= f
'[a, 6]
/+ f
J[b, c]
/
Furthermore, if f>g and both functions Finally, if/is integrable on [a, b], then
F(x) is
a
=
f
Jin
/ vi
continuous function of x.
are
integrable,
then
J//^J/^.
2.6
Calculus of One Variable
171
The fundamental theorem of calculus says more: if/ is continuous on that the definite and the is, indefinite integrals of j [a,6]/= \baf; The proof of this is actually quite easy to describe. Define
[a, b], then / coincide.
these functions
on
the interval
[a, b], corresponding
to the two sides of
Equation (2.17); F(x) F(x)
=
=
inf {KP, /) : P
a
partition
of
[a, x] }.
sup{(j(P,/) : P a partition of [a, *]}
that/is Riemann integrable on [a, U] is to prove F(b) F(b). We show, using Theorem 2.9, that in factF(x) F(x) for all x e [a, b~\. First of all F is differentiable in [a, b~\. Let x e [a, ft] and h > 0, then
To prove
=
=
F(x
+
n)
<
F(x)
+ Mh
(2.18)
F(x
+
h)
>
F(x)
+ mh
(2.19)
where M, m are the maximum and minimum of/ in the interval [x, x + ]. These inequalities can be routinely verified (see Problem 32); Figure 2.9 is + h) is just F(x) plus the infimum of all 2 (P,f) over parti [x, x + h]. Any such sum lies between Mh and mh. Now Equations (2.18) and (2.19) give
convincing: F(x tions of
J(x +
m <
h)
-
F(x)
; n
^
<
M
r
*
Figure
2.9
x
+ h
172
2.
Notions
of Calculus
Letting h 0, since / F'(x) exists and is/(x). -*
and has the
F(a)
=
is the
value.
F differ
Thus, F and
0 is obvious, we have that F(x) defined for all x, is differentiable and has
F(a)
J [a>x]/is
same
is continuous, M and m both tend to f(x). Thus Similarly, one verifies that F'(x) also exists for all x
by
a
constant.
Since
F(x) for all x. Thus derivative/. This, then,
=
=
proof of
(Fundamental Theorem of Calculus) Suppose f is contin Then the integral J*/ exists for all x e [a, b]. uous on the interval [_a, b}. This is a differentiable function off, and Theorem 2.10.
d
r*
-r\f dx Ja
f(x)
=
PROBLEMS 29. Prove
Proposition 15(ii).
mean value theorem is proven in the case where The way to do the general case is to compare the graph of/ with the line through f(b) and f(a). More precisely, let g be the function
30. In the text the
f(p) =f(a). whose
graph is that line, (a) Show that
h(x) =/(*) -f(a)
and consider h
-f(bl
~f(a)
b
(x
-
=/ g.
(2.20)
a)
a
(b) Show that h(a) h(b) 0. (c) Now from the text there is a f between a and b such that h'(0 Differentiating (2.20), deduce that =
=
=
0.
m=m-m Suppose that /is differentiable on the interval [a, b], and/'(x)>0 x. Show that /is strictly increasing, that is, f(x) ( y) if x
for all
continuous. 34. Find the real-valued function
/,
such that
f f(t ) dt
=
j
f(t) dt
for all
x e
[0, 1 ]
continuous
on
the interval [0,
1]
2.7
Multiple Integration
35. Suppose /is k times differentiable on R, and fw(x) 1. Verify that /is a polynomial of degree at most k
173
=0 for all
x.
Multiple Integration
2.7
The calculus of many variables results from the attempt to study functions of several variable quantities by generalizing to that context the calculus of a Some notions
variable.
single
of linear
algebra to
closer to that of
be
properly
generalize easily, The
understood.
require some ideas integration theory is much others
variable than is differentiation, hence
one
we
shall describe
it first. A closed
rectangle
in R" is
a
set of the form
{(x1,...,xn)eR":ai<xi
some
case
fixed
points
of intervals,
in the
same
we
a
=
=
(a1, ...,an),
denote the
l(a1,...,d>>,(b1,...,b't)-\ b
=
(b1, ...,bn)
corresponding
open and
As in the
way:
(a,V) lsL,b) (*,K]
=
=
=
{xeRn:ai<xi
rectangle will refer to any of these possibilities. is rectangle R determined by the vectors a and b
The term the
in Rn.
half-open rectangles
Vol(K)
=
The volume of
(b1 -a1) (b"-dr)
or halfNotice that the volume of R is the same whether R is closed, open it should be since the faces contribute no volume. open. Of course, this is as The characteristic function of S, denoted by Xs set. Now let S be any on S and identically zero off S. We should want one is the function which is with the integral of Xsso that the volume of S coincides to define
integral have J Xr Vol(R). The notion particular, for a rectangle R we shall turn out that way. of integral will be built up piece by piece so that things of characteristic functions Now suppose that /is a finite linear combination called of rectangles: /= Z/=i *(*) Such a function is and identically of collection rectangles, finite some It is constant on each of =
In
simPle/u"ctl0,n:
zero
off their union.
174
2.
Notions
Definition 11.
[/=
>
of Calculus
Let/be
a
simple
function.
If/=
Z*=i
ailRi->
Z^Vol(R;)
we
define
(2.21)
;=i
We
immediately have a problem. It may be possible to also write the function in another way, / Zy= i c,- Xsj fr some other collection of rectangles. For Definition 1 1 to make sense, we must be assured that the same
sum
=
Zy=i
and the
Vol(Sj) coincides with (2.21). In case the at and c; are all and {Sj} are nonoverlapping (intersect only in faces),
cj
{Rt}
amounts to the assertion that the volume of
a
set is the
sum
one
this
of the volumes
of its
rectangular pieces, no matter how it is so partitioned. The verification that (2.21) is the same for all expressions of the function /as a combination of characteristic functions is a long verification which is omitted. We now make this general definition of the integral. Definition 12. zero
outside
Let/be a bounded real-valued function which is identically rectangle R. The upper integral off is
some
J /= inf{| The lower
simple
o: a a
function
on
R such that
a
>/}
integral off is
j /= sup{| /is integrable
a: a a
simple
function
on
R such that
a
}
if
j /= j /;
the
common
value is the
integral f /
This is the direct
given to
generalization of the definition of the Riemann integral On the plane and in space it bears the same relation volume as does the Riemann integral to length.
in Section 2.6.
area
and
Definition 13.
Let S be
a
set in R".
If Xs is
integrable,
we
define the
volume of S to be
Vol(S)=J**s Now there
are
sets for which /s is not
and shall not
occur in this text. logical overlapping rectangles contained in the
integrable;
these
are
highly patho
Notice that if Ru ...,Rn are nonset S, then the sum of the volumes
2.7
Multiple Integration
175
J (Z Xr) is less than J Xs since /s > Z X*, Thus the volume of S is greater than the sum of the volumes of any collection of nonoverlapping rectangles contained in S. Similarly, if now Ru Rn are nonZ Vol (Rt)
=
,
.
.
.
,
overlapping rectangles containing S, J *s Z Vol(i?;). S is trapped between the volume of any union of rectangles containing S and the volume of any union of rectangles contained in S. If we can make these <
two volumes as close as we
J
Xs is
integrable (for
Theorem 2.11.
then
J
please by =
Xs
Let R be
proper choices of the
then
rectangles,
]Xs), and its integral is the volume of S.
closed
a
Thus, the volume of
rectangle
in R".
Iff is
continuous
on
R
and zero offR, then f is integrable.
J
Proof. Given e > 0, we must find simple functions < J t + e Vol(.R); for then it will follow that
a,
such that
r
a
>/>
t
and
<j
f /< f
<j
for any >0.
("
<
t
+
e
Vol(7?)
Thus, lf<~lf.
<
f/+
e
Vol(R)
In any case, since the
inequality, J/<J/is
obvious, /is integrable. Such functions a, t are easily found using the basic property of uniform con tinuity (discussed in miscellaneous Problem 80). According to that theorem, given >0, there is a 8>0 such that, if |x-y|<8 then |/(x) -/(y)| < e. Now partition R into a finite set S of rectangles each of which has the property that any two points are within 8 of each other. Thus, if for each such rectangle p, mp and m<e. M are respectively the maximum and minimum of/on p, we must have M ,
Let
a
T=2mX<>0 pes
2 MPXe
=
peS
where p0 is the open
L
=
J
2 Mp Vol(p)
peS
<
f
J
since S is
These
rectangle corresponding
a
t
+
e
Then
a
>/> t certainly, and
2 (rne+ e) Vol(/>)
peS
2 Vol(p)
peS
partition
following
<
to p.
<{t+e Vol(R) J
of R into rectangles.
basic
properties
of the
integral
are
easily derived.
176
2.
Notions
of Calculus
Proposition 16. The collection of integrable functions the integral is a linear function. That is:
is
a vector
(i) Iff is integrable and c e R, then cfis integrable and \cf=c (ii) Iff, g are integrable, so isf+g and \(f+g) J/+ J g(iii) Furthermore, iff< g then [f<\g.
space and
J/.
=
(ii) is certainly true for simple where R,, Sj are rectangles, then 2*jXs,> /=2'Xi> + 2ZbJXsj is also simple, and thus integrable. By Definition 1,
We leave the
Proof.
f+g=2a>Xi
|(/+ 9) More
proof
of
For if
functions.
=
generally,
simple functions
(i)
0
to the reader,
=
2 ", Vol(R,) + 2 bj Vol(Sj) now
let
/,
g be any
cti, a2, ru t2,
CTi>/^(72
=
J>+ \g
integrable functions.
If
>
0, there
are
such that
ti;>0>t2
and
J
<^i
<,
I
(T2
J
Ti
<,
> a2
+
T2
+
J
T2
+
Thus
CTl
+
Ti
>/+ g
so
{ (f+g) ji + jri <|((72 + t2) + <
Since
> 0 was
e
arbitrary,
we
obtain
2e ^
J(/+^)
J (/+ g) <, J (/+ g),
Finally,
j(f+g)^jcr2 + jr2 + 2e<jf+jg+2e so
letting e->-0,
+ 2e
so
/+ # is integrable.
2.7
Multiple Integration
111
Similarly,
j(f+g) 2e>jai + jr2>jf+jg +
so
again letting
e
->0,
j(f+g)>jf+jg (iii) Finally iff
zero
Sg-Sf^0,or!g>jf. We shall
give
the basic tool for
computing integrals: Fubini's theorem. integrate by integrating one variable at a time. For the purpose of showing this, write the variable (x1, ...,x") of R" as (x, y) where x e R"'1 and y e R: x (x1, x"'1), y x". Let / be a function defined on a rectangle R in R", and suppose for each y fixed, f(x, y) is an integrable function of x. Define F(y) J/(x, y) dx. If F is an integrable function of y, its integral According
now
to that result
we can
=
=
...,
=
JF(y)dy j\jf(x,y)dx =
dy
off. We shall now show that iff is integrable generally (after applying this principle n times) J/ functions appearing in the following formula are integrable, then the
is called the iterated integral this is the
if all
More
same as
formula is valid.
jfix1
x") dx1 =
dx"
J J"'" jf(x1,...,xm)dxl
dx2
dx"
(2.22)
This follows from Fubini's theorem. Theorem 2.12.
refer
to
(i) (ii)
Let f be
the coordinates
These functions
an
ofR"
ofy, These functions ofx,
integrable function on a rectangle (x, y), where xeRk,ye R"~k
R in R".
as
|/(x, y) dx, |/(x, y) dx are integrable. j/(x, y) dy, |/(x, y) dy are integrable.
We
178
2.
Notions
of Calculus
(iii) If is given by
any iterated
integral off; for example,
J7(x, y) dx dy j \]f(x, y) dx] dy j \jf(x, y) dy =
Proof.
It is
=
dx
easily verified that the collection of functions for which the
asser
tions (i), (ii), and (iii) are true is a vector space. Furthermore, these assertions are obvious for the characteristic function of a rectangle. Thus, Fubini's theorem holds for simple functions.
Now, suppose /is a bounded, real-valued function on the given rectangle R, and suppose that or is a simple function, and /> a. By definition of the lower integral with respect to the
x
coordinate,
jf(x,y)dx>jo(x,y)dx Now this
inequality is maintained after taking
the lower
integrals with respect
to y,
thus
J"
J*/(x,y)rfxpy>j[ja(x,y)rfx
dy
=
j o(x, y)
dx
dy
(2.23)
since Theorem 2.12 is true for simple functions. Equation (2.23) being true for any o <;/, we can take the least upper bound on the right, obtaining
j jf(x,y)dx
dy>
\f(x,y)dxdy
Now, by considering simple functions kind of
reasoning
we
a
such that
o
>:f and applying the
same
obtain this inequality
j J7(x,y)rfx dy^jf(x,y)dxdy As
a
result,
we
obtain this string of
real-valued function
KlM
on
inequalities,
which is valid for any bounded,
R:
V
J'l
W\fU>
(2.24)
(The second and third inequalities follow immediately from the fact that the upper integral always dominates the lower integral.) Now, if /is indeed integrable, the
2.7 first and last terms of
(2.24)
are
the same,
so
all
are
Multiple Integration the
179
That the second and
same.
top third are equal implies that J /(x, y) dx is integrable. That the bottom third and fourth are equal says that J/(x, y) dx is integrable. The equation
jf(x,y)dxdy j jf(x,y)dx
dy
=
now
just
Now we
states the
equality of the end
shall illustrate the
we
should remark that
defined
only
rectangle; more often such given measurable domain D.
Definition 14.
Let D be
function / defined R
on
/ will
of
a
be
domain contained in
a
we
function is defined
We make the
say /is
integrable
a
or
rectangle
if this is
considered
following definition.
so
Given
R.
a
function/
for the
by
=
a
but rather
D,
a
Before doing that, integrate functions
xe
0
D
xeR,x$
D
JD/=J/
We define If D is
on
=/(x)
/(*)
graph
have the occasion to
on a
on a
defined
of Fubini's theorem.
use
rarely
we
terms with the interior terms.
subdomain of
function,
or
has
a
rectangle
some
integrable if / is. tacitly assume our
other
R bounded
by
surface which is the
a
redeeming property,
then the function
We shall not pursue this theoretical
domains
inquiry,
redeemable.
are
Example 39. Define
{D
(x, y): y) x2
0
=
f(x,
=
+
<
y2
y
if
<
x2,
f(x, y) x2 + y2. 0 otherwise. D, and f(x, y)
0<x
(x, y)
e
=
=
Then
\j-\j- uj>-Hjx-n.c"<x2+/)* since, for fixed x, f(x, y) is We thus obtain x2 + y2
It-?
x2y
zero
r1/
+
if
y-^ dx=L\x
a
x <
0
x6\
or
J
y
>
1
dx
x2 and otherwise is
1
26
+yr*=5+2T= 105
180
2.
of Calculus
Notions
y
=
g(x)
Figure 2.10 Let
J7
=
us
do the
example, iterating this time
f_ [f_ /(x, y) dx\ dy j^ [f_(x2 =
+
1 ~
3
2
1 +
~~
or
dy
_ ~~
15
3
Ibl
7
general technique can be described (Figure 2.10).
as
follows :
=
{(*, y) :
a < x <
b, g(x)
<
y
(Figure 2.11) D
dx\
26
2 ~
in either of these forms D
y2)
in the other order.
t
_
The
same
=
{(x, y):a
Then, given the function/defined
on
rbf rg<-x>
r
f=\
f(x,y)dy
dx
in the first case; and in the second rf
C
c*W
j/ j J =
<x<
f(x,y)dx dy
D,
xjj(y)}
we can
write
Try to write the domain
2.7
Multiple Integration
181
Of course, if neither case can be obtained, then D might have to be broken up into pieces in each of which either representation is possible. The computation of integrals in more than two dimensions is done in pretty much way, but with a certain amount of additional care. For example, should try to pick out one of the coordinates, say z, so that the given domain takes the form g(y) < x (y), where y represents all the other
the
same
one
coordinates and ranges through some domain D0 break down D0 in the same way.
Now
.
one
proceeds
to
Examples
D={(x,y, z): x2 z) xyz.
40.
f(x,
y, Now z ranges between 0 and
D
y2
+
z2
<
1,
(x2
+
y2))1'2,
+
x >
0,
=
{(x, y, z): x2
+
y2
<
(1
D
0,
z >
0},
1, 0
0
< x,
<
y, 0
so
< z <
Thus, continuing the analysis of
A>
>
y
=
{(x, y): x2
=
=
{(x, y, z): 0
< z <
[1
0 -
+
y2
<
< x <
(x2
+
1, 0 1, 0
< x,
<
y
0
<
<
(1
y} -
x2)1/2,
y2)]1/2}
p
x
=
Hy)J x
J
\
a
L,y 1
Figure 2.11
=
[1
-
(x2
+
y2)]1/2}
182
2.
Notions
of Calculus
and "
.1
.
J
f= D
J0
2
V
y2))rfy
dx
y(l-(x2
J0 L
(1
X
y,
0
y2
+
z2
+
z >
^^0
dx
J
4
z): x2
1
x2)2]
-
2
{(x,
+
<
~24
1, (x
0}
y,
z):
0
i)2 y,
We may rewrite this domain
{(x,y, z):(x-i)2
{(x,
-
f(x,
+
< x <
+
1, 0
<
y
<
[i
-
(x 0
z)
r
r[i(x-i)2]'/2
=
<
i
1
as
-
< z <
[1
Jl-(x7
+
< z <
[1
y2)11/2
dx L^o
Figure
2.12
-
(x2
+
y2)]1/2}
iff12,
Thus ri
y2
y20,y> 0, 0
=
dx
Jl-X2)'/2
(see Figure 2.12). =
dy
So
firx(l-x-2)2
D=
dz
z
\_J0
x >
D
.[l-(x2+y2)ll/2
y
2J/[Jo 1
41.
x\ .1
1 =
r.(i-xJ)'/2
rfy
az
-
(x2
+
y2)]1/2}
2.7
Integration is clearly of value study of mass. Suppose
in
Multiple Integration
183
computing volumes;
it also plays a role domain in R3 filled with a certain fluid. shall let (>) be the mass of the fluid contained
in the
is
a
If D is any subdomain in E, we in D. What information do we need in order to compute mass (D), and how do we compute it ? The answer is suggested by comparison of the
properties fact, it is clear that the intuitive properties of mass are the same as the properties of volume ; so we should also expect to be able to compute masses by integration. In fact, we introduce the notion of density: for x0 e E, the density o(x0) of the fluid at x0 is the limit of
mass
with those of volume.
In
mass(i?)
r where
Vol(R)
we mean
by
R
->
the sides of R tend to
density
and
,
that x0 is in the rectangle R, and the lengths of (we might call mass (R)/'Vo\ (R) the relative
zero
of the fluid in the
domain is such
x0
rectangle R).
in terms of this
computable {R J is a almost filling D. Then a
domain and
collection
Now, the
mass
of the fluid in any D is
density function a. Suppose of pairwise disjoint rectangles
in D
approximation to mass (D) and as the size of the rectangles gets smaller smaller, the approximation gets better. On the other hand, this sum is the integral of a simple function approximating a, and thus approximates JB a. Taking the limit we obtain mass (>) JD a. is
an
and
=
EXERCISES 15. Compute the volume of these domains: (a) {(x,y)eR2:x2 + y2<\).
(b) {(x,y)eR2:x2
184
2.
Notions
,a\
of Calculus
tt
x
^
(a)
f(x, y)=
i
n \ f(x,y)=
\
(e)
tfx^y .
.c
if
fx + y
x
v <
+
ifx +
j
1
y>l.
f(x,y) (i+x2 + y2Y'2. Integrate the function /on the domain D in R2. (a) 7J {(x, y):0<x,0
(f)
=
18.
=
=
=
=
=
=
-
=
=
=
=
PROBLEMS
Verify that the integral on R" as defined in this section coincides, 1, with the Riemann integral defined in the previous section. 37. Let /be a bounded, nonnegative, real-valued function defined on the interval /, and let D {(x,y) e R2; x el, 0 < v (*)}. Verify this assertion: /is integrable if and only if D is measurable, and U f= Vol(Z>). Let D be a domain in R2 and suppose 38. Use Problem 37 to verify this. 36.
when
n
=
=
that D is of the form
{(x, y)eR2:a<x
third
-
inequalities of Equation (2.24).
40. State and prove Fubini's theorem in three dimensions. 41 Suppose the unit ball is filled with a fluid whose density is .
proportional
to the distance to the
boundary. Find the radius of the ball centered at the origin which has precisely half the mass. 42. Suppose a cone of base radius r and height h is filled with mud (Figure 2.13). Suppose the density of the mud is equal to the distance from the base.
What is the
43. A beach B is B
=
{(x,y):
1 <x2 +
and the human
mass
of the mud?
shaped in
the form of
a
crescent
(see Figure 2.14)
y2;(x- i)2 + v2
density a increases with the distance from the water. More precisely, a(x, y) (x2 + v2)"1. What is the mass of humanity on that =
beach ?
2.8
2.8
2.13
Figure
2.14
Differentiation
185
Partial Differentiation the
Although it is
Figure
Partial
integral
computed by
in R" is defined without reference to the
succession of integrations,
a
one
coordinate at
coordinates, a
time.
The
notion of differentiation is, to begin with, generalized to R" one coordinate at a time. Later we shall see how to build out of this generalization an
invariant notion of derivation. Let x0
e
R", and suppose that / is
neighborhood x' given by J
(X0
,
.
of x0
.
.
,
X
,
.
a
real-valued function defined in
For each i consider the function of the
x0 )
single
a
variable
186
2.
of Calculus
Notions
If this function is differentiable, we denote the derivative by df/dx1, and call More precisely, it the partial derivative of/in the x' direction. Let /be
Definition 15.
of x0 in R".
df OX
The
v
,
i
(x0)
=
real-valued function defined in
partial derivative of/ with respect
xp'
/(V
,. hm
+ f,
Another way of
This restriction is
partial
function
a
as
the
on
are
partial one
computed merely by considering
constant.
42. =
xy
f(x,y)
=
dx
y
dy
(x, y)
=
x
43.
=
(x2y) f dy
2xy
=
x2
1
44.
/(*, y)
=
cos[x(l
f(x,y)= -(l dx
dy
(x, y)
=
45.
/(*, y)
=
x>
-x
+
+
y)]
y)sin[x(l
sin[x(l
Consider the
through
Examples
fix, y)
derivative is this.
x0 and in the E; direction. variable and df/dx' is its derivative.
the line
function of
derivatives
relevant variable
neighborhood
*
describing
as a
a
to x! at x0 is the limit
,x0")-/(xo1, ...,Xo")
(-.0
function / only These
a
+
y)]
+
y)]
all but the
2.8
x~(x, y)
'
yxy
=
dx
dy
(x, y)
=
Partial
x In
Differentiation
187
x
Of course, if the functions
dx1'"" are
dx"
also defined in
neighborhood
a
of x0
,
partial differentiation, and keep going in this
we
may
way
as
subject them as possible.
far
to further
We shall
refer to any such operation as a partial differentiation and call its order the number of individual partial derivatives involved.
Thus, the order of
\dxJJ
dx'
is 2; the order of
dx2 \dy is 6.
\dz3JJ
We introduce
dx2
notational convention which deletes
\dxj
dx
d2f
a
d
/df\
dx
\dyj
_
dx
dy *2
l2f
d
ldf\
dx1
\dxJj
_
1
dx dxj
d3f dx1 dxj dxk
d6f dx2 dy dz and
so
forth.
((K\\ dx' \dxJ d
3
dx
\dxkjf
d5f \dx dy dz3) /
parentheses.
188
2.
Notions
Suppose
df/dx1,
that
now
...,
df/dx"
of Calculus
/is
open set N in R" and that set all the variables constant except
function defined in
a
all exist in N.
If
we
an
just the derivative off along this line. Thus, if df/dx' 0, / is constant along the line on which only x' varies. In such circumstances we say that /is independent of x', since /does not vary as x' alone varies. If, moreover, df/dx' is zero at all points of N for all i, then/ depends on none of the variables, so is constant. As this is an important one, say x'
then
,
df/dx'
is
=
observation,
make it.
we
Proposition 17. Suppose that f is a real-valued function defined in a neigh of x0 in R". f is constant near x0 if and only if all the derivatives df/dx" exist and are zero near x0 df/dx1, borhood .
..
Proof.
.
,
If /is constant, it is obvious that
df/dx'
=
suppose that these conditions are valid in a ball y (y1, ...,y")e B(x0 r). We will show that/(>0 =
,
the
proof.
0 for all i.
On the other hand, Let B(x0 r) centered at x0 =/(x0). Figure 2.15 illustrates .
,
Consider the function of x" :
f(x0
.
,
.
.
,
Xo-1, x")
This function has derivative
/(xo1,
.
.
.
,
zero
by hypothesis,
XV1, xo") =f(xo'
so
is constant.
xl~\ y")
Now, the function of xn_x,
f(x0\...,x"0-2,x-\y)
(y'.y2,/)
/ (y',-to2,jto
(xn\xir, Jfi>:! )
Figure
2.15
Thus,
2.8
Partial
also has derivative zero, and thus must be constant,
f(x0\
...,
Differentiation
189
so
xTl, v") =f(x0\ ...,y-\ y")
This together with the preceding equation gives
f(x0\
xTl, x0") =f(x0\
,
...,
xl'1, y-\ y")
Continuing in this way, we can replace each x0J by the corresponding y] time, ending up with the desired equation f(x0) f(y). As far fact
only on
the
as
should
we on
higher order verify. they
dy dz
basic
performed.
For
example,
d5f
d5f
dz dx dy dx dz
dy dx dz dz dx
equation; it being clear that all others follow applications of the first one. The verification of (2.25) interesting application of Fubini's theorem.
verify only
the first
succession of
amounts to
an
Theorem 2.13.
Let
f
be
a
function defined in a neighborhood that all first- and second-order partial deriva
real-valued
of(x0 y0) in R2 and suppose off exist and are continuous ,
tives
d2f
d2f
dx dy
dy dx
throughout
R
one
(2.25)
d5f
We shall
are
dy dx
dy
dx
N
concerned, there is
a
This is that each
^
*' dx
a
are
at
partial differentiation depends the number of derivatives with respect to each coordinate, and not now
the order in which
from
differentiations
one
on
N.
Then
N.
Proof. We apply Fubini's theorem to d2f/dx 8y in ((xa yo), (s, t)) contained in N (see Figure 2.16)
a
sufficiently small rectangle
=
,
rs IY
e2/
l
r'
[V
e2/
dy
(2.26)
190
2.
Notions
of Calculus
Figure Now,
we can
rs
Jxo Integrating and (2.27)
easily evaluate the integral
dy
once
\
df
df
Jxo 8x\8y
J
dy
dy
8
again (this time with respect
ay J*o L-'vo dx
dy
right-hand side.
the
on
l8f
f
82f dx
2.16
dy
dx
=
c'
j
y) we obtain from Equations (2.26)
to
8
-[f(s,y)-f(x0,y)]dy
=f(s, 0 -f(x0 t)
-
,
Now,
we can
[f(s, y0) -f(x0 y0)] ,
differentiate this
'
d
8s
L-
d2f 8x8y
'
(x, y) dy dx
-i.
82f 8x
Then, from (2.28)
f'
(2.28)
equation with respect to 5 first, and then t. By the calculus, we know how to differentiate the integral on the the upper limit of integration :
fundamental theorem of
left with respect to
For fixed y,
a2/ =
axey(''y)dy ax(s't)-8x<s'y)
8y
(s, y) dy
2.8
Differentiating
this
equation
with respect to
t , we
Differentiation
191
obtain
82f
82f dx 8y
as
now
Partial
8y
8x
desired.
Another allows
important application
Proposition 18. two
variables F
function
of Fubini's theorem is this result, which
to differentiate under the
us
on
F(x)= Then F is
Suppose that f
integral sign.
is
a
continuously differentiable function of
and y, a < x < b, and y the interval [fl, b~\ by x
e
D,
a
domain in R".
Define
the
|7(x,y)dy
differentiable
and
d-f(x) j8/(x,y)dr =
Jd dx
dx
We shall show that F is the indefinite
Proof. f
integral of
the function
J 8f.
\yx(x,y)dy and thus
by
the fundamental theorem of calculus, the
follows.
proposition
By
Fubini's theorem
i[ll^y)dy\dx=l[V^{x'y)dx by the fundamental f(t,y)-f(a,y). Thus But
theorem of
dy
calculus, the inner integral
on
the
right
is
[f(t,y)-f(a,y)]dy=F(t)-F(a) f'Tf ^(x,y)df\dx=\ JD J
J LJD Let are
us
8x
return
obtained
consideration of the first-order derivatives. to lines differentiating after restricting the function
now
by
to the
These
parallel
192
2.
of Calculus
Notions
to the coordinate axes.
That
along any line.
We
is,
Let x0
Definition 16.
we
e
neighborhood of x0 derivative df(x0 v) to be in
a
.
generalize
this notion to allow differentiation
make this definition.
R" and suppose /is a real- valued function defined If v is a vector in R", we define the directional
,
jff(*o + 'v) This is
clearly
the
Um/(x0 + fv)
same as
/(x0)
-
t
i-o
We leave it
as an
g(x0)
=
exercise to
that
verify
(2.29)
d/(x0,Ej)
Now, in certain pathological
cases
the directional derivatives need not
together in any nice way, but typically we need only know the vatives in order to find any directional derivative. Proposition 19.
Proof.
vary
one
...
,
+
linearly
.
in
v.
looking
at the difference
fv)-/(x0)
variable at
argument with
a a
In order to expose the idea without we consider the two-variable
time.
pile of indices,
encumbering case.
difference
f(x0 + t h, y0 + tk)
{f(x0
deri
a neighborhood ofx0 and the partial Then the directional derivatives df/dx" all exist near x0
The argument consists in
/(x0
hang
Suppose f is defined in
derivatives df/dx1,
df(x0, y)
partial
+
th,
y0 +
f(x0 y0) ,
tk) -f(x0 + th, y0)} + {f(x0 + th, y0) -f(x0 y)} ,
the
Write the
2.8 find
We
can
the
mean
x0
and
better
a
expression for the
+ th,
y0)
f(x0 y0)
-
8f =
,
Similarly, by applying the
8f 8y
is, there is
by applying f0 between
a
,
value theorem to the function f(x0 + th, s),
mean
we can
as
(x0 + th, t]a)tk -q0 between y0 and y0 + tk.
some
rectangle [(x0 yo), (x0 ,
f(x0
+
df
ty)-f(x0) =
--0,
we
Thus,
we
have for suitable
(f0 v) ,
in the
+ th, y0 + tk)],
dx
t
(f
.,8f,
o
,
yo)h +
oy
v.
,. (x0 + th, r)0)k ,
obtain by continuity that
d((x0 y0), (h, k)) ,
Thus the
That
(fo yo)th
rewrite the term in the first set of braces
t
y0) of s.
193
+ th such that
x0
Letting
Differentiation
term in the second set of braces
value theorem to the function f(s,
f(x0
for
Partial
proposition
=
+ j (*o y0)h ,
(x0 y0)k ,
(2.30)
is verified, at least in R2.
This linear function, df(x0 v) of the vector v in R" is called the differential We will make a systematic study of this in a later chapter. The of/ at x0 ,
.
vector-valued function
K\ \dxl'""dx") IK.
gradient off and is denoted by V/ generalization of (2.30) to n variables
is called the 19 that the
df(*o ,v) The
in the what
=
et W
more
<. V/(x0)>
Proposition
t2-31)
total derivative." It is not as powerful one variable and it is some kind of tool. For cumbersome, but it does provide a similar
gradient behaves as a analysis of a function
example,
=
It is clear from
is
sort of as
"
the derivative in
194
2.
Notions
Proposition 20. attains
a
of Calculus
The
maximum
gradient of a function
vanishes at any point at which it
minimum value.
or
(xo1, x0") is (for instance) a maximum value of /, then Proof. If x0 f(x0l, x', x0"), as a function of x', attains a maximum at x0'. Thus, 8f/8x' vanishes at x0'. Since this is true for all i, Vf(x0) 0. =
.
.
.
.
,
.
.
.
.
.
,
,
=
Examples 46. Consider /(x, y,
V/=(2x Thus
x
=
-
zero
x
-
is, only
x2
+ xy +
y2.
when
2y
=
2 that
=
2y)
+ y,x +
V/is
z)
at the
This is the
origin.
only critical point,
and
a
minimum at that.
47.
f(x,
y,
V/= (cos y, is
never
48.
zero,
z)
=
y +
x
sin y,
1)
so
/has
no
f(x, y,z)
V/= (cos(yz),
x cos
=
xz
x cos
z
critical values.
(yz)
sin yz, xy sin
yz)
V/is zero only when x Oandyz n(n + ^foranyintegern. Clearly, / has both negative and positive values near any point on the line {x 0}, so no such point is critical. Thus, /has no critical points. =
=
=
EXERCISES 20. Find the first
partial derivatives of these functions. sin(xy) (c) x'' (d) x2y + y2x 21. Differentiate x*". (Hint: This is the same as finding the directional derivative of x"z at a point (x, x, x) in the direction of (1, 1, 1).) (a)
xyz
(b)
2.9
Improper Integrals
195
22. If /is differentiable at x0, then
8f ,-
(x0)
for all
=
df(x0 Ei) ,
i.
23.
Suppose that/, g are differentiable at x0 in R". Show differentiable and V(fg)(x0) =f(x0)Vg(x0) + g(x0)Vf(x0). 24. If /is differentiable at x0, and/(x0) ^0, then
v(i)(x0) ^ =
V/(x0)
25. What is the minimum of x2 + 26. What is the maximum of
x+3v
that fg is also
y2 + (2v + l)2 ?
n
l+x2+y2' 27. Compute the differentials of the functions in Exercise 20. PROBLEMS 44.
Suppose /is
a
differentiable function of two variables and gi,g2 are one variable so that the range of (gi,g2) is in the
differentiable functions of domain
Find the derivative of
of/.
45. Let /be
function of 46.
x
a
y alone if and
Suppose that
47. Let T: R"
h(t ) =f(gi(t), g2(t)).
differentiable function of two variables.
->
L
:
R"
R" be
->
a
only if 8f/8x + 8f/8y
R is
a
linear function.
linear transformation.
=
Show that
/is
a
0.
What is V/_ ?
Define the function
on
Show that / is differentiable, and V/(x, y) f(x, y)
=
(Tx, y>.
=
=
48. If T: R"-+R" is
a
<7x, x> is differentiable,
2.9
g(x)
=
=
Improper Integrals
We return
be
linear transformation, then the function Tx + Tx. and V#(x)
now
considering
focus
on
the
"
introduce the
to the
study of functions of
functions defined
on "
one
variable; in fact,
the whole real line.
infinity of such functions. notion of lim/(x) as x -* oo. behavior at
we
will
Our interest will
For this purpose
we
196
2.
Notions
If/ is
Definition 17.
{x: x > fl} lim/(x)
=
a
say that
we
if, for
L
of Calculus real-valued function defined in
an
infinite interval
converges to L as x becomes infinite, written > 0 there is an M > 0 such that x> M implies
f(x) e
every
x-*oo
|/(x)
L|
Similarly, if/ is defined
< .
in
{x: x
we
say
lim/(x)
=
L
x-*<x>
(the
limit of
there is
an
f(x)
M
is L
as
x
0 such that
>
becomes M
x <
negatively infinite) if, implies \f(x) L\ < e.
for every
>
0
Examples lim
49. M
x >
1/x
=
For
0.
implies |l/x
0|
-
given
>
0,
we can
take M
=
_1.
Then
< .
50.
4x2
,.
+ 3x + 5
lim
-
ox
jc-+qo
For, 4x2
2
long
so
1 -
=
7
2
0,
as x >
4 +
+ 3x + 5
3/x
~
8x2
Now,
8
7
-
we can
-
5/x2 7/x2 +
(2.32)
compute the desired limit by using the standard algebraic
rules (the limit of a sum is the sum of the limits, etc.). (See Exercise 28.) Since 1/x, 1/x2 tend to zero as x->oo, the limit of (2.32) as x-> oo
is
4/8
=
1/2.
51.
x\x\
,.
hm x-*co
1
T
X
I'm
5=1
X-*
oo
x\x\ t + X
2= 1
If
X2
x|x|
1
=
'
Y+x1
=
TTx1
l +
i/x2
if
x <
0,
x|x|
x2
1 + x2
1 + x2
-1 1 +
1/x2
2.9
lim arctan
52.
x
=
Improper Integrals
197
n/2.
Definition 17 is the analog for functions defined on an infinite interval of the notion of convergence of a sequence (a function defined on the integers). lust as we pass from sequences to series we can pass from infinite limits of
functions
to infinite sums; that
is, integrals
Let/be a continuous / is integrable if lim \xaf exists,
Definition 18. We say
absolutely integrable if lim
J?/ /is
over
infinite intervals.
function
on
in which
case we
\xa \f\
the interval
{x:
a}.
x >
write the limit
as
exists.
Examples 53.
x-2 is integrable
dx
1
x
=
Ji
=
the interval
on
[1, oo).
For
--+1 m
i
so
Cx~2dx=
Jl
54.
x~y
(--+1^1)
lim m-oo\
is not
cos x
=
1
m
absolutely integrable
on
the interval
[1, oo).
For -2HB +
OO
dx
>
X
Jl
Between
2;tn
-
JI/3 QQg x
Z I b=1 J2nn-n/ and
jt/3
dx X
3
2rcn +
1
n/3,
x
cos x >
(2nn
+
n/3)
1
\.
Thus,
f Ji
The
2n
1
cos X
dx>
Z ^i
=
o
2
(27tn
+
tt/3)
infinite intervals is entirely analogous to the We have the following facts (whose counterparts
theory of integration
on
theory of infinite series. theory of series are easily recognized).
in the
oo
3
2.
198
Proposition
of Calculus
Notions
Let f be continuous
21.
on
the interval
{x
: x >
a}.
(i) fis absolutely integrable if and only if the set {\xa |/|} is bounded. (ii) Iff is absolutely integrable, then fis integrable. (iii) (Comparison Test). // there exists a b > a and a constant K and an integrable positive function g defined on {x: x> b} such that Kg > |/|, thenf is absolutely integrable. Proof. (i) If |/| is integrable, clearly {Jj |/|} is bounded. On the other hand, if {J; |/|} is bounded, let L supfjs |/|}. Then for e > 0, L e is not an upper bound, so e. Then for all x>.x0, there exists an x0 such that S |/| > L -
=
-
L>:
j l/l>|
!/!>-
so
-flfl\
(ii) Suppose j? |/| Let
e
> 0.
f Then for n,
=
L.
Then there is
l/l-
m
>: x0
|c-cm|
=
Let c
an x0
=
rS/.
{c} is
a
Cauchy
sequence.
,
f fzf l/l^ f l/l- f l/l
Thus
{c} is Cauchy,
Let
>
n
so
0, and find N
in the
;> x0
,
a
f \f\-L
as
x
<-
-
e
We show that
such that for
so
+
converges, say to
that |c
previous computation.
c|
<
a
c.
j
|/|-.
<
We shall show that in fact
e/2 for n^>N.
Then for
x
>
J/=
c.
max(x0 N), ,
2.9
(iii) Under
the
given hypothesis, if x b
x
j l/I^J" e
Thus
>
Improper Integrals
199
1, then
co
1/1+
AT
J
g
b
a
by (i), /is absolutely integrable.
Here is
easily derived relationship between integrals which provides yet another
an
series and
the absolute convergence of test for the convergence of
series.
Proposition 22. (Integral Test) Let f be a positive, decreasing function + Then Jj f exists if and only ifY.n=if(n) < defined on R .
Proof. For x,n<x
-1
x
i
1
1
ex'
e
1
~
as x ->-ao.
Example 55. OO
1
?2 n(log n)2
< 00
For log
au /'e* du f"*-*
dt J
2
No* 'log 2
f(l0g f)2
I
1
,og2
log
_, =
-"
1 2
logx
Thus
r
/
dt
h f(logf)2
i
_
*-\log2
M=_L logx/
log:
< oo
200
2.
of Calculus
Notions
EXERCISES 28.
Verify
these
algebraic properties of lim.
Suppose lim/(x), lim g(x) JC-.00
X-.00
exist.
(a)
lim f(x) + g(x)
=
x-*o
(b)
lim f(x)g(x)
=
x-*co
(c)
lim -.
lim f(x) + lim
lim f(x) lim
-
=
#(x)
g(x).
x-*<x>
x-*ao
f(x)
g(x).
x-*co
x-*oo
Jim-^ *?
if lim
hm^tx)
#(x) ?t 0.
x-.co
x-oo
29.
Compute these limits
(a)
sin
as x
->
oo.
1
x .
x
(d)
tan-.
(C)
XSinx-
x
x2 + 3x + 1
1
** + l
x2-l
(c)
FT!"
30. Which of these series converge:
2
(a)
n=2
nlogn
n=2
1
"
(>)
1
W
2^r-T (10g lOg ri)2 =
n(logn)2
ii
=
2
n__.
1
OO
Z n
2
=
oo
2
x2
(log n)2(log log n)2
n3'2
n=2(logn)2
1
oo
-J7T
f * IV '
2
W
2
2
(d>
Cg)
^r-^i
1
CO
n
2
(f)
-j
n
=
2
(log rt),2 1
( sin n)2
X-.00
2.70
2.10
The
The
Space of Continuous
Functions
201
Space of Continuous Functions
The mathematician attacks his
problems with a certain store of techniques. problem will require the development of a new technique; more often the problem is solved by viewing it in one way, and then another and then again another until a viewpoint is obtained which allows for the application of one of those techniques. Sometimes if the viewpoint is clever enough, or profound enough or naive enough the applicable technique is quite elementary and surprising and leads to further deep discoveries. This is the case with the contraction lemma (a fixed point theorem) which we shall apply several times in this text to obtain some of the basic facts of calculus. First, in this section, we shall develop the particular viewpoint in the relevant context. It is simple enough instead of looking at continuous
Occasionally
functions Let
in
us
a
one
at
a
time,
we
illustrate this with
finding
a
consider them all.
particular problem. Suppose differentiable function with these properties:
f'(x)=f(x)
for all
a
x
and
/(0)
=
we are
interested
(2.33)
1
function means first of all to verify that a solution to our and secondly to establish some technique for computing it. problem exists, We already have enough experience with calculus to know that this second of objective will be hard to fulfill. What we in fact seek is a means effectively This provides a clue : let us look for a sequence our solution To find such
a
approximating of functions {/}
.
which converges to a function with the properties (2.33). Such a sequence would be a sequence of differentiable function {/} such If we that the sequence {fn(x)} converges for all x, and f'n(x) =f-i(x). had such a sequence, we could take the limit and deduce that
lim/'n(x)
=
lim/n_1(x)
lim/(x) will solve our problem. itself provides the tech a good idea, because Equation (2.33) be Let any function, and define /0 nique for generating such a sequence. Will the sequence so forth. and /,=/' Then let/2=A,/3=A, that Notice 7 i -/ o. /2 {/} converge? Well, that is a problem. we must be very Thus, /3=/'2=/"o, and more generally / =/<5n)is careful to choose an infinitely differentiable function for/0. Suppose f0 so all and 0, Then / + 1 =/o"+1 chosen as a polynomial of degree n. converges, the rest of our functions are zero. Thus, the sequence certainly so/(x)
=
Now this is
=
202 but
2.
hardly
Notions
of Calculus
solution, since the condition /(0)
to a
=
1 is not verified. In
fact,
this present approach has obviously petered out fruitlessly and it may be 1 in our because we have not incorporated the initial condition /(0) =
approach. Can we put all of (2.33) in one statement, and then proceed with this technique of generating an approximating sequence ? The funda mental theorem of calculus says yes; in fact, (2.33) can be rewritten as
f(x)=ff(t)dt
+ l
(2.34)
'o
This
is
operation involving integration rather than differentiation, advantage of not having to choose a very wellbehaved function for the first approximant. Let us try again, with (2.34) rather than (2.33). Letting/, 1, we find and
now
an
have the added
so we
=
/i(x)
=
f2(x)
=
f
1 dt + 1
=
+ 1
x
"o 2
f (f + 1) dt +
1
+
=
x
+ 1
2
Jo
/3(x)=/o(y '+1)d'+1=^+5+x +
+ 1
/b(x)=/o7-i(o^+i=5+(-^ ---+^+-+i +
(2.35) Now we're
getting
converges for any
/(x)
=
somewhere.
x.
lim/(x)=
recognized we
sought
after function.
the solution of
need
the limit in
now
our
problem see
=
that the series
(2.35)
(Of course the reader has long since as
being
the
exponential
is the theoretical mathematics that will allow
Xn
J/(f)d-f+l= Z-f Jo B=on!
function.
that it did in fact turn out that
(2.35) and correctly deduce r*
/(x)
seen
0n!
Thus he should be reassured to
What
already
-.
=
this must be the
We have
Thus, letting
us
way.)
to take
The
2.10
Space of Continuous Functions
203
led to the
question of convergence in the space of continuous proceed to that theory. functions. Let X be a closed bounded set in R", and let C(X) denote the space of all continuous complex-valued functions on X. We know that iff and g are two functions in C(X), then so are/+ g ana\fg and cf, for c, a complex number. In particular, C(X) is a vector space on which multiplication is defined. The vector space C(X) is quite different from the vector spaces C", R" : C(X) is usually infinite dimensional (see Problem 49). C(X) does not have any Thus
we are
We
now
in fact, we wouldn't know how to choose one. however, C(X) is not very different. There is in this particulars,
obvious "standard basis" In other
space a reasonable notion of closeness. Two functions are close if their values are everywhere close ; that is, if the maximum of their difference is
small.
This leads to
Definition 19.
a
notion of
Let X be
a
space of continuous functions
length
and distance in
C(X).
closed and bounded set in jR", and X. If /e C(X), the length of/is
C(X)
the
on
H/ll =max{|/(x)|:xeX} If/ g The
are
in
C(X), the
properties
of
distance
length
between/ and g is \\f-g\\.
and distance
are
just
those of the
corresponding
notions in R" :
\\cf\\
=
\c\ H/ll
ll/+<7ll< 11/11
+
11*11
the ll/n =0, then /=0. What is important is what we can consider notion of convergence of a sequence of continuous functions. We say that term of the /n ->/if ll/n -/II -> . that is> if the distance between the general becomes arbitrarily small. This is the same as saying that and
If
sequence
/ of/n at points
of X converge to the values of/in a uniform manner. The value of these notions lies not only in their naturality, but in the now
the values
possibility of finding specific functions satisfying given properties by techniques of approximation. Let us make this precise.
realizable
Definition 20.
C(X).
Let X be
We say that
{/}
that lim n-* 00
||/
-
/||
=
0
a
closed bounded set, and
{/}
is uniformly convergent if there is
an
sequence in /e C(X) such a
204
2.
of Calculus
Notions
We say that the sequence is such that
II/b /mil
uniformly Cauchy if, for
e >
every
0 there is
an
N
whenever n,m> N
<
Examples 56. Let
x
uniformly
converges
/n'(x)
to
(1 x)x". This sequence ||/,||. compute max|/(x)|
[0, 1], /(x) Let
zero.
us
=
-
=
n(l-x)x"-1-x"
=
so/'(x) 11/.
be the interval
=
0 has the solutions
V
which tends to 57. On the
l/U+1/
+
n
x
=
n
0,
x
l\n
+
n/(n
=
+
+
Thus
1).
lj
zero.
same
interval the sequence fn(x)
=
sin
x/n
tends to zero,
for
||/J
sin
=
-?
-
0
asn->oo
58. Consider the convergence of the sequence {nx sin x/n} on the [0,1]. Now we know that sin x/n ->0 as n - oo, but
interval -
nx
oo,
so
cannot
we
We have to refine
our
n, it is very close to
nx
nxsin
about the For
x/n.
large
product. values of
Thus
x/n.
(2.36)
x
n
n
so we
=
-
make any deduction
information about sin
guess that
nx
sin
x/n
-*
x2.
Let
us
prove it
by computing
(2.37)
x
nx sin n
In order to do
.
x
x
n
n
that, let
<
sin
r
n
us
provide
an
in the interval
estimate to
[0, 1]
our
guess
(2.36).
(2.38)
2.10 Then
The Space
of Continuous Functions
205
(2.37) becomes X
nx sin
x
/
2
x\ 1 + nj
x
.
nxlsin
=
\
n
n
I nx
=
.
x
x\
n
nj
sin
\
X
.
lliwll
<
sm
-x2 n
(2.39)
X
-
n
1 < n
X
nx
(2.40)
n
.
-^
=
n
n2
and since n_1 -*0
as n
->
co, we are thrc)ughL.
59. On the interval It is not does not m
=
2n,
[0, 1] the sequence {sin nx} is not convergent. Cauchy sequence. The distance ||sinnx sinmx|| become arbitrarily small as n, m->co. In particular, if
even a
we
||sin(nx)
have
sin(2nx)||
>
sinl
I
n
\
2n/
sinf 2n \
|
=
1
2n/
The basic theorem about convergence of continuous functions is the which plays the same role in C(X) as the least upper bound axiom
following,
does for R. It provides the assertion of existence of functions with prescribed properties. In order to verify that a sequence of functions has a continuous limit, we need only verify that it is a uniformly Cauchy sequence.
Theorem 2.14.
uniformly
X.
uniformly Cauchy
Suppose {/} is
Proof. on
A
This
n,m~^.N.
means :
This
l/n(x) Thus, for
sequence
of
continuous
-
uniformly Cauchy e > 0, there is an precisely a
for every
x
is
means
fm(x) \<e
for all
x e
sequence of continuous functions N > 0 such that 1 1/ /m 1 1 < e for -
(2.41)
X
each x, {/(x)} is a uniformly Cauchy sequence of real numbers, and Denote the limit, lim/(x) by f(x). We must show that this
thus converges.
function
functions
convergent.
->/(x) is continuous, and that/
converges
uniformly to/.
206
2.
Notions
First of all, if
e
>
of Calculus
0, choose N as above, and let
m
->
in
oo
(2.41).
obtain, for
We
n>N, lim
|/(x) -/(x)|
|/(x) -/(x)|
=
<
for all
e
x e
X
m-* oo
Thus, if n^N, ||/-/||>e.
This
implies
that lim
||/-/|| =0,
desired.
as
n- oo
/ is continuous. Fix x0 e X. Let e > 0 and choose iV so large that < e/3. Since / is continuous, there is a S>0 such that ||x x0||<3 /II ll/ implies |/v(x) /(x0)| < e/3. Then if |x x0| < 8, Now
| f(x) -/(xo)l
l/(x) -/(x)| + |/(x) -/(x)| + IMx) -/(x0)|
^
e
e
e
<3+3+3=, desired.
as
seen one vector space of functions, we can easily see them every The collection of bounded real-valued functions on a set X is a
Having where.
vector space
taking
functions same
over
the reals.
values in R" is also
taking
a
The collection of all bounded functions
on
X
the space of continuous All the spaces here are endowed with the
vector space;
values in R".
similarly,
concept of length :
ll/H =sup{||/(x) ||: xeX} Of
even more
interest
analytic operations.
are
For
the spaces of functions on which is defined some example, if I is an interval, the space of all real-
valued functions which
C](/)
are differentiable on / is a vector space. The space of all functions whose derivative is continuous is also a vector space,
is the space C(n,(/) of all functions which have continuous nth derivatives. The space R(I) of functions which are integrable on / is a vector space. These (and other) examples are further elaborated in the exercises. Suffice it to as
say here that the mathematical
theory which follows this point of view (20th-century) development which has only in foundations of mathematics, but in the
(called functional analysis) is
a
recent
had profound impact, not practical application of mathematics Let
us
return to the space
bounded set X in R". in we
a
space,
are
on
which
easily led
Once are
in all branches of science.
C(X) of continuous functions
we
begin thinking
defined such notions
to consider functions
on
on
of these functions
a
closed
as
points
distance and convergence, that space. Naturally, such a as
function is continuous if it takes convergent sequences into convergent sequences.
2.10
The
Space of Continuous Functions
207
Examples 60. Let g e /-/, that is,
C(X)
and define
\\fng-fg\\< II/.-/H ll^n 61. Define
II/2 -/2|| If
/-?/>
thusalso
=
$
||/-/||-*0,then
i/,: C(X)
ll(/ -/)(/ +/)||
the term
continuous, for
if
->o
C(X),
-
js
<
||/ -/||
,/, js also continuous, -
||/ +/||
||/ +/|| remains bounded
for
(2.42)
while
\\f-f\\->o
||/2-/2||->0.
62. If P is any
polynomial, \/ip(f)
=
P(f) is continuous
on
(Problem 55). 63. Define M:
C(X)
-^
R, M(f)
=
C(X)
||/||.
This is continuous, since
\\M(f)-M(g)\\
=
\ ||/||-|l*ll l
64. Let x0 AT and define F0 : C(X) ^ R, F0(f) =f(x0). Certainly is continuous: for if/->/in C(X), then the maximum over X F0 f
\L(x) -/W| Fo(L) F0(f).
tends to zero; in
particular, |/(x0) -/(x0)| ->0,
so
^
65. The definite /
=
[a, b~]
<=
R.
integral is
J/"-//|-|-f//n_/) so
if/n ->/,
also
J//,
than that of
from
a
continuous function
on
C(I), where
For
->
\,f
<\\L-f\\(b-a) A stronger and more important statement the indefinite integral, as a function
Example 65 is that C(I) to C(I) is continuous.
This is contained in the next pro
position.
Proposition
23. Let I {x e R: a < x < b}. Suppose f is a sequence of functions on I converging uniformly to f. Let F(x) J-* / jafi Then F^F uniformly.
continuous
F(x)
=
=
=
,
208
2.
Notions
of Calculus
Proof. x
I
F(x)-F(x)\=
J
Thus, taking the maximum
(/-/)
the
on
SS
11/. -/IK*
-
<
a)
11/. -/IK*
-
a)
left,
l|F-F||<||/-/||(6-a) if /
so
->/ uniformly
so
also F^F.
Problem 56 is intended to demonstrate that tion is
not a
continuous function
on
C(I).
on
the other hand, differentia
(It isn't
even
everywhere defined;
i.e., there are continuous functions that do not have a derivative.) less, Proposition 23 has this consequence for differentiation.
Neverthe
Proposition 24. Let {/} be a sequence of continuously differentiable functions on the interval [a, b~] and suppose that (i) {/'} is uniformly Cauchy, (ii) f(a) 0 for all n. Then {/} is uniformly convergent to a differentiable function f and f lim/'. =
=
The
Proof.
proof
of this
proposition consists in a rereading By that theorem
of
Proposition
23
via the fundamental theorem of calculus. X
fn(x)=j
/'.
a
by Proposition 23, / is also convergent. If we let g lim/' then lim/, Thus, lim/, is indeed differentiable and its derivative is g lim/' so
=
,
=
Jj g.
=
.
Let
to the consideration of our
original problem. In fact, generalize slightly. Let c be a complex number, and let us seek a differentiable complex-valued function / such that
let
us
return
now
it
us
f'(x) This is,
tinuous
=
cf(x)
for all
x
by the fundamental function/such that
f(x)
=
c\Xf(t)dt
+ l
and
/(0)
=
1
theorem of calculus the
(2.43) same as
seeking
a con
(2.44)
2.10 Now that
follow
we
a more
and define the
Tf(x)
The
Space of Continuous Functions
have the necessary
theory sophisticated approach. function T on C(I):
c
=
ff(t) dt
and
point
of view available,
Let I be the interval I
+ 1
=
209
we
may
\_-R, R],
(2.45)
'o
/ such that/= Tfi that is, a fixed point of the transfor technique is that of successive approximation. Let /0 be any continuous function, and define f Tf0, f2 Tf =T2f0, and in general T"f0. We must show that the sequence {/} converges. If / Tfn_i 1 we can compute the sequence explicitly, and we find that we choose /0 We seek
function
a
Our
mation.
=
=
=
=
=
(cx)"~l
(ex)"
Then if
m >
L(x)
n,
f(x)
-
=
(ex)'1
(cx)m k
+ f + -b^-rv, (m-1)!
[-R, R] the maximum by |c|, and x by R. Thus,
On the interval
replacing
c
m
ii f
Um
All-
'
+
(|c|Kr
i +
k=o
^R)m'1
of this
i +
i +
k\
k=o
+
1)!
expression
is dominated
by
(lcl*>"+1 (n
(m_i)i
m,
(cx)"+i ;
(n
ml
+
l)!
K\
Since the series
(f.
(kl R)k
is a Cauchy sequence, so by (2.46), converges, its sequence of partial sums Since T is {/} is a Cauchy sequence and is thus uniformly convergent.
continuous
on
lim/,
=
C(I), lim
we
have
T(/B_,)
=
Tflim/,-,)
=
r(lim/)
210 so
to
2.
Notions
lim/ solves spend a few
the
The
number
/'(*)
given problem. This function is important enough for paragraphs discussing it. exponential function, denoted exp(cx),
is the solution of the differential
c
cf(x)
=
us
more
Definition 21.
complex
of Calculus
/(0)
=
or
ecx, for any
equation
1
First of all, this definition makes sense, because there is
only
one
solution.
If g also solves, then d
[ecx~\
ecxg'
cecxg
cecxg
cecxg =
92
dx
since
g'
ecx
or
the
Thus e^g'1 is constant. Since its value at 0 is cg. From these discussions we have these additional g.
=
=
exponential
=
b
ex+y
we
from
(ii)
1, of
0
n!
eV.
never zero.
=
Part
(ii) follows
and
=(*)
must have
h(x)
=
e',
'(0)=_=1
so
(ii) is verified.
Part
(iii) follows immediately
:
^CXq-CX
(e")-1
properties
=
"'(*) Thus
=
function
Part (i) follows directly from the argument above. uniqueness. Fix v, and define h(x) e'+yle*. Then
Proof.
so
=
ecx is
from the
1, e^g'1
25.
Proposition
(ii) (iii)
0
92
_
gCX-CX
_
Q
_
1
=
PROBLEMS 49. Let / be
dimensional.
a
nonempty interval in R.
Show that
C(I) is infinite
The Fixed Point Theorem
2.11
50. Show that the sequence of functions
on
211
the closed unit disk in C
defined by n
X
ux)=Lk2 converges.
| 2 Z"/M converge
51. Does the sequence 52. Let {a} be these facts:
a
sequence of
on
the closed unit disk?
complex numbers such
that
2 M
< co-
Verify
For every z, |z| <
(a)
i/Wi
2
<
n=
l,/(z)
2*='
=
ifl-i 1
(b) /is continuous
{z
on
e
C:
\z\
<
1}.
2i?=N+il,.l^0as/V^*>. /
Show that
g be continuous functions
\\fg\\
<
54. Show that
x
x
n
n
Is
\\g\l
the interval
for all
<
sin
ll/ll
on
This is true because
polynomials f(z)=^Liaz\
uniform limit of the 53. Let
az" converges, and
\\fg\\
<
on
since
/is
the
||/-/v||<
the closed and bounded set X.
11/11
lltfll possible?
[0, 1],
n
n2
55. Let xi,
.
.
.
,
xk e X
and p be any
polynomial
in k variables.
Define
r-.C(X)^C
,F(/)=K/(xi), ...,/(*)) Show that T is continuous.
funct.ons which sequence {/} of differentiable is not convergent. that {/'(*)} convergent, but such
56. Find
is
a
uniformly
2.11 The Fixed Point Theorem of the
point theorem is a generalization the discussion approximations described above in The fixed
exponential
function
root of Newton as a technique for finding First, is this. method Newton's
technique was first used by polynomial equations. Simply stated,
This
technique of successive
of the
212
2.
Notions
of Calculus
technique is described by means of which one can transform a given approxi a root into a better approximation. One then chooses a reasonable approximation, applies this technique to it to find a better one. Having this, one again applies the technique : if it's a good one, the result is an even better approximation. Continuing in this way, one obtains a sequence of approximations which should converge to the root. Now, having described the procedure, let us turn to Newton's specific technique for bettering approximations. Let / be a given real polynomial. We want to find a point x0 such that f(x0) 0. Choose a px so that/f^) is small. Now, replace the function by its linear approximation at p1: L(x) =f(Pi) +f'(Pi)(x Pi), and let p2 be the root of L(x) 0. In other words, replace the graph off by its tangent line and let p2 be the x intercept of that line (see Figure 2.17). Now apply this procedure to p2 Let p3 be the root of the linear approximation to / at p2 and so forth. We can describe Newton's technique abstractly as follows : For any point p, let T(p) be the zero of the linear approximation of /at p:T(p) solves the equation f(p) +f'(p)(T(p) p) 0. (We must a
mation to
=
=
.
,
-
Figure 2.17
=
2.11
# 0 for T to be
that/' T(p)
assume
have
we
p, and
=
The Fixed Point Theorem
well-defined
a
conversely, thus
213
function.) Clearly, if f(p) 0, are in reality seeking a fixed =
we
point of T\ T has the property of contraction
Suppose
1 such that
on some
interval I.
There is
a
Ty\ < c\x y\, all x, y e I. Then Newton's method 0 (or/'(x) works. There is a root of/(x) 0) on the interval /, and it is the limit of the sequence x0 Tx0 T2x0 where x0 is any point of /. This is the content of the fixed point theorem. We now state and prove it explicitly for subsets of C(X). It will be clear that the theorem is true for subsets of R", by virtue of the same argument. c <
\Tx
-
-
=
,
,
Suppose S is of sequences
Theorem 2.15.
a
,
closed
a
S contains all limits
which is
=
in S.
contraction, that is, there is
II T(f) Then there is
-
a
T(g) || unique
< c
||/
-
g \\
...,
of functions in C(X) : that Suppose T is a mapping of S onto S
a c <
set
1 such that
for allf,
g
e
S
continuous function f0 such that
T(/0) =/0
.
Proof. Certainly the fixed point is unique. For if T(f0) =/0 and T(f) =/, then l!/o-/ill= lir(/o)-r(/)]|
=
=
ll/.+i-/.ll so we can
=
ll7y.-3/.-ill^c||/.-/.-il|
verify by induction that
ll/.+i-/.ll
m
> n we
11/- -/.!!<
have
ZUi+i-fj)
<mf\\fj+l-fj\\ J=n
c"
#)"*
<
Since
c
lim Tfn
< =
II/1-/0IKII/1-/0I
Since T is continuous, Tf0 {/} is Cauchy, so has a limit f0 e C(X). function. fixed lim/+1 =/0 and thus f0 is the desired 1
-
,
,
n-eo
n-> oo
As an
\IcJ)
x0
an
>
illustration 0 such that
on
x02
the real numbers let =
a,
us
prove that if
by Newton's method.
First,
a >
we
0, there
is
describe the
214
2.
p).
2p(x
Tp
=
of Calculus
0, the linear approximation to x2 Thus, the zero of this linear polynomial is
Let p
T.
map +
Notions
contraction
a
p2
a
+ p ^2p -IH
Clearly, if T has a fixed point x0 that Tis
at p is
a
>
,
1
|Tx-Tv|=;
we
x02
must have
=
Thus,
a.
we
must show
closed interval:
on some
a
a
1
X
V
~2
y +
a
,
(y -X)
x-y + xy
-\ x-y\ 1-^ XV
only ensure that \. {x: x2 >a/2}. (a/xy) Then for x, y e I, xy > a/2, so a/xy < 2, which is the desired inequality. Thus, by the fixed point theorem there is an x0 with x02 > a/2 such that Since a, x,y 1
x02
=
now
give
a
Sometimes
theorem.
j
a
(a/xy)
<
1,
so
contraction with
somewhat a
c
need
we
Let/
=
=
function of
x:
value (0, 1), and near of y. The relations =
1
y
more
subtle
application
of the fixed
point
relation between two real variables determines
function of the other.
as a
are
1
positive,
1, for Tto be
a.
We shall
as a
all
axe
>
=
For
(1, 0)
we
sin(x(log y))
example, x2
x;
+
y2
the relation
=
1
gives
should write
=
x
y
=
(1
=
x
+ v
=
one
0 determines
x2)1/2 y2)1/2 as a (1
near
the
function
0
somewhat less transparent, nevertheless as a function of x.
we can
ask whether
or
not
they
do determine y
Suppose
now, in
F(x, y) defined in the
(2.47)
=
general
we
have
equation (see Figure 2.18)
an
0
(2.47)
plane. We ask : does saying y g(x) ?
amounts to
=
there exist More
a
function g of x such that is there a function g
precisely,
such that
F(x, y)
=
0
if and
It is not hard to find
a
only
if
y
=
#(x)
necessary condition.
For there to be such
a
function
2.11
y is
a
function of
The Fixed Point Theorem
y is not
jc
a
215
function of j:
Figure 2.18 it must be the
that each line
constant intersects the set
F(x, y) 0 F(x, y), as a function of y on lines x The root of constant must take the value 0 only once. 0 is then the value g(x). Now we recall from one-variable theory F(x, y) that a function H(y) will take all values once if H'(y) j= 0. Thus the reason in
only
one
case
x
point (see Figure 2.19).
=
=
Thus the function
=
=
able condition to
impose
on
F is that it has
a
continuous
partial
derivative
with respect to y, and dF/dy = 0. This condition turns out to be enough. More precisely, suppose that F is defined and has continuous partial derivatives in the neighborhood of the origin in R2, and dF/dy(0, 0) ^ 0.
Figure 2.19
216
We seek
a
and
F(x, F(x0 y)
function g defined in g(x)) 0. If we fix This
0.
=
function of y
of F(x0
y)
,
a
brings
neighborhood of x
x
=
,
as a
of Calculus
Notions
2.
=
right T(y)
us
is the
Newton did:
as
at y ; that
+
dy
0 such that a
zero
of the linear
#(0)
=
0
root of
Define T
approximation
is,
dF
F(xo,y)
=
0, then we seek x0 back to Newton's method. near
(xo,y)(Ty-y)
=
0
or
dF
Ty Just
as
Thus,
=
y-
Jy
in Newton's need
we
y for x0
that fixed we now
case
the solution of
point.
that T is
a
that it will have
This
(2.48)
,
,
only verify
near x so
F(x0 y)
(x0 y)
application
F(x0 y)
=
,
contraction in fixed
a
0 is the fixed
some
point
of T.
interval of values of
point ; and we define g(x0) to be point theorem really works, as
of the fixed
shall prove.
Suppose that F has continuous partial derivatives in and that F(0, 0) 0, dF/dy(0, 0) # 0. Then there is 0), neighborhood of(0, in x some interval 6, e) such that g definedfor ( function Theorem 2.16.
=
F(x, y) Proof.
=
0
if and only if
Instead of (2.48)
we
y
=
a
a
g(x)
consider something
slightly simpler.
For
x near
0,
define
Tx(y)=y-
d2L-(0,0) Ty
F(x,y)
(2.49)
We want to find the fixed point, if it exists, of (2.49). e
17 < y < 77 in which Tx is
< x < e,
a
Thus
we
seek suitable intervals,
contraction
\8F (0,0) .
T,( yi)
By the
mean
-
Tx(y2)
=yi-y2-
value theorem there is
a
[F(x,yi)-F(x,y2)]
t, between y1 and y2 such that
dF
F(x, y,)
-
F(x, y2)
=
dy
(x, 0(yt
-
y2)
(2.50)
2.11
Equation (2.50) becomes,
Tx(yi)
Tx(y2)
-
=
-
(0, 0)"1 8-f (x, 0 y2)\lI 3-f dy dy
(0, 0).
we
may choose
s
e <
y2
e so
and
< e
(2.51)
-
Now the term in brackets is continuous in
Thus
217
substitution,
upon
(yi
The Fixed Point Theorem
(x,
that that term is less than is between yy and y2
.
With this choice
(2.51) gives
\Tx(yi)-Tx(y2)\
.
=
=
=
=
=
=
EXERCISES 31. Find, by Newton's method, a sequence of numbers converging to the square root of a, for any a > 0. Now, do the cube root. 32. Find a sequence converging to a root of these polynomials : (c) xb-2x2-3x + 2 (a) x3 + x2 + x+l
(b) (a) F(x, y) 33.
(d) x5 -x- I F(x, y)=x sin(xy). For what values of (x, y) such that 0 defines v as a 0 is it true that nearby the equation F(x, v)
x2-x+l Let =
=
function of x?
(b) Same problem for (ii) F(x, y) x" y, (i) F(x,y)=xy2 + 2xy+\, x2 v2+ (iii) F(x, y) 34. Let F(x, y) be differentiable in a domain D, and (x0 ,y0)e D such that F(xo,yo)=0. Suppose g is differentiable and has the property Show that 0. vo F(x, g(xj) <7(*0) =
=
=
=
,
9{-Xo)-
8F\dx(x0,y0) 8F/dy(x0,yo)
-
218
Notions
2.
35. Find
of Calculus
g' where
g is defined
implicitly by l (c) exy (d) e"=y
xsin(xy)=0 cos(x+v)=y
(a) (b)
=
PROBLEMS 57. Prove the fixed Theorem on
IfS
is
then there is
S,
a
a
point theorem in subset
unique
of
R" and T is
e
S such that
y0
F(x0 y0) ,
0,
differentiable
= 0.
dF/8y(x0 jo) (F(x, y(x))=0 ,
Theorem 2.15
as
defined on S and is T(y0) y<>
and
contraction
.
be
Let g
a
=
partial derivatives
58. Let F have continuous =
R":
(x0 y0)
near
the
,
We
g(x0)=y0).
and suppose
function described in can
prove that g is
follows.
(a) First of all, by the mean value theorem, for any (x, y), there is (f 7j) on the line between (x0 y0) and (x, y) such that ,
,
8F
8F
F(x, y)
F(x0 y0)
-
(I t/)(x ~Xo)
=
,
+
>
(, -q)(y
Why is the mean value theorem applicable? (b) Now, if we substitute y=g(x), y0 =g(x0),
(|, rj)(x -Xo) +
=
-
we
y0)
have
dF
8F 0
a
Yy
(f V)(9(x)
0(Xo))
-
Thus g (x)
-
x
g(x0) Xo
-
dF/8x(lj, ri) 8F/8y(, tj)
Conclude that g is differentiable and
g'(x0)
2.12
8F/8x(xo,g(x0)) =
8F/8y(Xo,g(xo))
Summary z,... of
A sequence zx to C.
complex numbers is a function from {z} converges to z if, for every e
The sequence positive integers there is an N such that \z z\ < e for A convergent sequence is
n >
the >
0
N.
bounded, but
not
conversely. A monotonic Cauchy criterion: a
bounded sequence of real numbers is convergent.
2.12
219
Summary
sequence {z} converges if, for every e > 0, there is an N such that \z for both n,m>N. The series formed of a sequence {zn} is the sequence of sums
zm \
-
< e
(?=i zj-
If the sequence of sums converges, we say that the series converges and denote the limit by "= i z If 2 z- converges, then z -> 0, but not conversely. If {ck} is a sequence of nonnegative numbers, c*. converges if and only if the sequence =1 ck is bounded. A series z converges absolutely if .
XlzJ
Absolutely convergent
< oo.
series may be summed in any convenient
way.
Tests for
Convergence
Suppose |z| < \wB\ for all but finitely (0 E |w| converges, z is absolutely convergent, (ii) if so does 2 |h>|. comparison test.
if
If
root test.
is
|c|1/n
< r
for
some r <
1 and all but
many
2
finitely
Then
n.
|z| diverges,
many n,
2
c
absolutely convergent.
If \cn+1/cn\ < r for some r < 1 and all but finitely many n, is 2 c absolutely convergent. The sequence {yk} of vectors in R" is said to converge to v if, for every e > 0, there is an N such that v || < s for k > N. A sequence of vectors || yk ratio test.
-
converges if and only if it does so in each coordinate. A set S is closed if and only if yk e S, lim yk v implies =
v e
S also.
Every
sequence contained in a closed and bounded set has a convergent subsequence. An .Revalued function defined in R" is said to be continuous at v0 iff is
defined in
a
neighborhood
tion is continuous
on a
of v0 and yk
->
v0
implies f(yk) (v0). point of S.
set S if it is continuous at every
A func If S is
a
closed and bounded set, and / is a continuous real-valued function defined on S, then/is bounded and attains its maximum and minimum. Sections 2.6 and 2.7 the definitions here;
are
only
mainly about integration. major results.
fundamental theorem of calculus.
interval [a,
F(x) exists for all
b}.
=
Then the
We shall not recollect
the
Suppose / is continuous
integral
ff
x e
[_a, b~\.
Fis differentiable
on
(a, b) and F' =/.
on
the
220
R
2.
Notions
fubini's
theorem.
=
Jj
Let /be
I in R".
x
x
of Calculus an
J/can
integrable function defined computed by iteration:
\j ([ \l /(*'.
*") dx" ^"-1
=
Let /be
a
v
is
v
is defined
a
real-valued function defined in
R", the directional derivative
vector in
on a
rectangle
be
'
dx1
neighborhood of x0 in R". If df(x0 v) of/atx0 in the direction a
,
by
lim/(X + ty)'
~
/(Xo)
f->0
(if it exists).
The
,(x0)
=
partial derivative
df(x0,Ei) derivatives
If these
partial
df(x0 y)
is linear in
,
If the
partial
of/with respect to x' at x0 is
v.
are
We
derivatives
can
all defined and continuous
df/dx'
all exist in
an
compute the derivatives d(df/dx')/dxJ. These derivatives. If all first and second derivatives in
an
open set
d2f
d2f dx1 dx'
throughout N. Suppose that / an
F(x)
=
has continuous
dx
partial
interval of reals, and D is
f
f(x, y) dy
Then F is differentiable and
d,.., 00
x0
,
then
open set
are
we
may be able to
the second-order
off exist
and
are
partial
continuous
N, then
dx' dx1
where I is
near
write
=
f
W
J g^0y)dy
a
derivatives in the domain I domain in Rn.
Let
x
D,
2.12
Suppose /is verges to L
a
as x
real-valued function defined ->
written
oo
lim/(x)
=
L if
R.
on
|/(x)
-
We say
L|
221
Summary
that/(x)
con
be made arbi-
can
x->ao
trarily small by taking on
x
sufficiently large.
If now /is
a
continuous function
R such that
f/
lim
^0
x- oo
exists,
say that
we
/is integrable
on
R.
If lim
J* |/|
exists, /is absolutely
X~*QO
If/ is a positive, decreasing continuous function if and only if "= i/(n) < oo. /exists jf We denote by C(X) the collection Let X be a closed and bounded set in R". of all complex-valued continuous functions on X. C(X) is a vector space. If/is in C(X), the /en#fn of/is
Integral
integrable. defined
on
test:
R, then
H/ll =max{|/(x)|:xX} For/ # in C(X) the distance between/and gis \\f- g\\. If {/} is a sequence in C(Z), and ||/ -/|| ->0 as n-> oo for some /e C(Z), we say that {/} Cauchy criterion. Suppose {/} is a sequence converges uniformly to / in C(X) satisfying the following condition: for each > 0, there is an N such that ||/ -/J| < 6 whenever n,m>N. Then there is an/e C(X) such that /->/ uniformly. integration.
also
J*/
The c
-"
If X is
an
interval in R,
and/, -^/uniformly
J/ uniformly.
exponential function,
denoted
is the solution of the differential
exp(cx), or equation y'
ecx for any
=
cy,
y(0)
in
C(X)
complex 1.
=
then
number
It has these
properties : (cx)n =
c(x+y) ecx is
_
o
n!
ecxecy
never zero.
fixed point theorem.
onto S
mapping of S
II T(f) Then there is
-
a
Let S be
which is
T(g) ||
< c
||/
-
a
a
closed
set
offunctions
in
contraction; that is, there is
g II
C(X) and
c <
for all figeS
unique continuous function f0 such that T(f0) =f0
T
a
1 such that
222
of Calculus
Notions
2.
implicit
theorem.
function
Suppose (0, 0), and
derivatives in
a
neighborhood
Then there is
a
function g defined for
F(x, y)
=
if and
0
of
only if
y
x
=
in
that F has continuous that
some
F(0, 0)
=
interval
(
partial
0, dF/dy(0, 0) # 0. e,
e) such that
g(x)
FURTHER READING
M.
Spivak, Calculus, Benjamin,
New
This is
York, 1967.
It is
text in the one-variable calculus.
an
eloquent
an
excellent reference for
a
full
treatment of the material in this
chapter. Lichtenberg, Mathematics for Scientists, Benjamin, New York, 1966. This is a review of the theory of calculus from the point of view of the physical scientist. It includes a chapter on numerical analysis. T. A. Bak and J.
C. W. Burrill and J. R. Knudsen, Real Variables, Holt, Rinehart and Winston, New York, 1969. An advanced text, going thoroughly through the material of this
chapter and beyond to
MISCELLANEOUS 59. Let So also is
the
a
of Lebesque
Then {x +
yn} is also
a
linear
subspace
Show that the collection C of convergent sequences is a linear subAlso C0 the collection of all sequences converging to zero is a
.
,
subspace of B.
These spaces are all infinite dimensional. lim on convergent sequences in the obvious R : lim{x} lim x Show that lim is a linear function.
62. Define the function way : lim : C
-*
"
"
=
.
63. What is the dimension of the space of linear functional annihilate C0? 64. Let xi x+i
sequence.
Show that it is not finite dimensional.
vector space.
space of B.
linear
a
r; thus the collection S of all real
60. Show that the collection B of bounded sequences is of the vector space S of all sequences (Problem 59). 61
integration.
PROBLEMS
{x} and {yn} be sequences. {rx} for any real number
sequences is
theory
=
=4, i(xn + 3/x).
x2
=
on
C which
x are defined, let i(4 + ), and once x2 {x} converges. Assuming that, find the ,
.
.
.
,
Prove that
limit. 65.
(a) Show that for
lim
n"/(n + 1)'
=
every
integer k,
1
limnV(n+l)t+1=0 lim
nk+i/(n
+
1)" does
not exist
(b) Let k be an integer, and 1 > h > 0. (c) Show that lim n/h" does not exist.
Show that lim n"h"
=
0.
2.12 66. Let
x+i
xi
=
Summary
223
1, and in general
1+x,
_,
3
3 + x
Find lim x 67. Suppose lim .
Let y Let A: be
(a) (b)
=
z
=
z.
J(z_, + z). Then a positive integer.
lim ,y z. Now let {y} be defined by =
1 y-
=
+ ~k~+\ ^Zn Zn+1
"'
^
z"+)
Then lim y=z also.
(c) This time take 1 y
=
-
(zi +
+ z)
Once
again lim > z. Suppose that / is lim/(c) =/(c). =
68.
continuous
at
c,
and
lim
c
=
c.
Then
69. Let {c} be a sequence of complex numbers, and suppose (|c|)"" R. Show that R'1 is the radius of convergence of 2 cz". 70. Let {.$}, {?} be two sequences of positive numbers such that lim s tn ' =
exists and is
nonzero. Then 2 s converges if and only if 2 t converges. {c} be a sequence of positive numbers. Suppose that for every we have also sequence of positive numbers {/?} such that 2 Pi < 2 C"P < - Prove that {c} is bounded. 72. Verify Schwarz's inequality:
71. Let
iiaAiV^iw2- !>i2 /
1=1
n=l
n=l
by virtue of the same fact for finite sums, which was dis Chapter 1 .) Is the reverse 73. Prove that if 2 kl2 < , then 2 (Un)\a\ < co. It is true
(Hint:
cussed in Problem 74 of
implication
true?
74. Let S be
a
S} is a closed 75. Suppose that / is
for all
on a
76. R".
Suppose Let x0
there is
,
an x2 e
a
/is
{yeR":
a
R" and
positive real-valued function defined log /is also continuous.
continuous
Show that
that
Xi e
=
set.
s 6
set S in R".
Show that (S)
subset of Rn.
continuous real-valued function defined c e
R" such that
R be such that
f(x2)
=
c.
/(x0)
<
c
(x,).
on
all of
Show that
224
2.
Notions
of Calculus
11. Show that if /is
a
continuous function
on
the interval /
taking only
rational values, then /must be constant. 78. A set S in R" is called connected if every continuous real-valued func tion has the intermediate value property. Show that this is equivalent to the
following definition: A set S is not connected if there is
defined 79.
a
continuous real-valued function /
two values.
S which takes
precisely Verify the following assertions : (a) A ball in R" is connected. (b) The set of integers is not connected. 1} is connected. (c) The sphere {xe R3: ||x || (d) The union of two balls in R" is connected if and only if they on
=
intersect.
(e) An open set is not connected if and only if it can be written as the disjoint union of two nonempty open subsets. (f ) A closed set is not connected if and only if it can be written as the disjoint union of two nonempty closed sets. 80. Let / be a continuous function on the closed and bounded set X. Then/is uniformly continuous; that is, given e > 0, there is a S > 0 such that for all x, y e A'such that |x y\ < S we have |/(x) f(y)\ < e. Supposing not,
derive
we can
a
contradiction
as
"
follows.
There is
an e0
such that
"
for every 8, is not true. Taking |x y | < S implies \f(x) f(y)\ < e0 S 1/n, there are x,yn with |x y\
Show that lim x'
{x\}, {y/}.
=
lim
/ but |/(lim x') /(lim y')|
> e0
,
a
contradiction. 81. Let L be
a
linear functional
on
R" and choose v0 such that
||w0||
=
1
and
L(v0)=max{L(v): N|
=
l}
Show that for every v e R", L(v) L(v0)
each rectangle R, and define F(t) JRt /. Show that / is continuous. Is /differentiable? 83. Let Q {pjq : p, q integers with 0 <,p
,
=
=
thusj Xq
=
84. Let
1-
/ be
integrable nonnegative function defined on the domain D {(x, y,z)eR3; 0 <. z <>f(x, y) ; (x, y)eB}. Verify that Vol(Z>) j /. B
c
R2
and
an
consider =
=
2.12 85.
225
Summary
Suppose that / is a continuous decreasing real-valued function of a 0. Then f
real variable and lim
=
X-.CO
this with Leibniz's theorem for 86.
Suppose that /is
/(x)^ + oo if, for -s-
Hxll-*
as
every M there is
that if /is
|jx||
a
a
series).
real-valued function defined
We say that
R".
oo
if such that
a
f(x)
> M
real-valued continuous function
oo, then /attains
on
a
minimum at
some
on
whenever ||x || > K. R" such
that/(x)
->
Show +
oo as
point.
87. Define
/(x)->0 in
a
way
||x||-*oo
as
suggested by the definition in
continuous function and
a
88.
minimum
on
on
problem.
Show that if a
a
maximum
R".
Suppose / is
a
real-valued function which has continuous
derivatives in the ball {x
<7(x)=f
the above
R" has this property, then it attains both
e
R": ||x|| <
1}.
partial
Show that the function
f(tx)dt
same properties, and find Vg. 89. Let /2 be the space of sequences {c} of real numbers such that
has the
2ici2< n
=
l
Because of the result in Problem 72 are
in
(Schwarz's inequality), if {c}
and
{dn}
I2, then
<{*}, ,}>
2
=
n
=
c*d l
Show that I2 is a Euclidean vector space with that inner product. 90. The space of continuous functions on the unit interval can be made into a Euclidean vector space in this way:
converges.
>=[
f(t)g(t)dt
Corresponding by ||
1 12
so as
product is a notion of length which we denote distinguish it from the modulus || || introduced in the
to this inner to
226
2.
Notions
of Calculus
Show that this length is deficient in these respects : We can have ||/||2-*0 without having ||/||-^0.
text.
(a) (b)
Cauchy
We
can
have
a
converge to
a
sense
of continuous functions which is
{/,}
sequence
sequence in the
of the
continuous function.
length ||
||2
On the other
(c) if ||/. ||. ^0, then ||/||2 -*0 also. 91. Suppose L: C[0, l]-*j? is a linear function. tinuous if and only if there is an M > 0 such that
,
a
but which does not
hand, show that Show that L is
con
\L(f)\^M\\fU 92. Show that there is
f'o(x)
=
a
for all
(/o(x))2
differentiable function
unique x
and
Do it
/o(0)
by applying the fixed point theorem ontheset{/<=C[i ]: ||/||
=
f0 such that
i
to the function T defined below
x
Tf(x)=\ 93. We of (n
f2(t)dt+i
talk of open and closed sets, and convergence in the space M " n) matrices, merely by considering them as vectors in R"2. Doing so these statements : can
x
verify
(a) (b) (c) (d)
The set G of invertible (n x n) matrices is open. The set of triangular matrices is closed. The function A->A2 is continuous.
lfp is
any
polynomial in
one
variable the function
T->p(T) is continuous.
(e) lim (x/n!) 2=o (l/n!)TB
exists for all TeL(R",
Rm).
fl-00
94.
Suppose g is a continuous real-valued function [a, a]. Show that the implication
f J
g(t)f(t)dt
=
on
the interval
0
-0
for all
F
implies g 0 holds whenever F is any one of these classes : (a) F=C([-a,a]). (b)F=Ci([-a,a]). (c) F is the collection of all polynomials. (d) Fis the collection {xi'-Ia subinterval of [a, a]}. (e) Fis the collection of all continuously differentiable functions such that/(-a)=/(a) 0. fe
=
=
ORDINARY DIFFERENTIAL
Chapter
J
EQUATIONS
on the study of the different application to geometry and classical (Newtonian) physics. The motivating problem throughout is the central problem of the subject of differential equations : to find a function on the basis of given information on its derivatives. Observed phenomena in the sciences seem always to involve rates of change. For example, it is observed that the rate of acceleration of a falling body is a constant independent of mass, height, or velocity; the progress of a chemical reaction slows down as it proceeds, dependent on the quantities of the chemicals involved. These observations, when made precise, appear as differential equations. In order to predict (the time it takes for the body to fall a given height, the amount of new chemicals produced before the reaction stops), the function described by the differential equation must be found. The first two sections of the present chapter are devoted to the description
In these next three
ial calculus of
one
chapters
we
shall elaborate
variable and its
of the basic concepts involved ; in the first we shall discuss the differentiation of vector-valued functions, and the second is devoted to approximation and Taylor's formula. We also include a brief excursion into the computation of maxima and minima of functions of several variables subject to constraints
the technique of Lagrange multipliers. The main theoretical tool in this study is Picard's theorem which gives conditions under which a differential equation has a solution and only one
by
solution.
essentially tells us what a well-posed problem is, that well-posed problems are always solvable. The question
This theorem
and asserts
227
228
3
Ordinary Differential Equations
actually producing a formula for the solution, or an algorithm for com puting approximate values for the solution is another matter altogether. Several techniques will be exposed in this chapter and Chapter 5 (successive approximations, series expansions); there are many more very efficient computational techniques which we shall not develop here. It will become clear that the subject of ordinary differential equations has a lot to do with the study of curves (paths of motion). Thus in the next chapter we shall investigate the geometry of curves and its relation with the subject of differential equations. of
3.1
Differentiation
The first
important step
in the
study
of differential
equations
is to consider
vector- valued functions of a real variable as well as real- valued functions.
This is the
appropriate setting for many problems involving differential particularly relevant when studying equations involving
and is
equations,
derivatives of order greater than one. In the first sections we shall consider differentiable vector-valued functions of a real variable and introduce a
special technique Definition 1. in
a
for
approximating
Let x0 e of x0
neighborhood
,. lim
f(*o
+
an
Revalued function defined
f is differentiable at x0 if
f(*o)
t
f-0
exists.
~
Taylor's expansion.
R, and suppose f is
.
0
values:
The limit is called the derivative of f at x0 and is denoted by an open set U, we say f is differentiable (written f is
If f is defined in U if as t
[f(x -+
+
t)
f(x)]/f
converges for all
x on
U to
a
f'(x0).
C1)
in
continuous function f
0.
That this definition is not is demonstrated
Proposition 1.
by
the
so
far from the derivative encountered in calculus
following
assertion.
Let f be
an R"-valued function defined in a neighborhood of (fu ...,/) in coordinates, f is differentiable at x0 if and only iffi,...,fH are differentiable at x0. Further, f'(x0) (f[(x0), fn(Xo))-
x0
e
R.
Write f
=
=
.
.
.
,
3.1
Differentiation
229
Proof. f (Xp +
Q
-
f (Xo)
//l(Xo + Q-/1(Xo) =
/
The limit
on
the left
as t
I
-*
t
)
0 exists if and
(Proposition 10 in Chapter 2), Proposition 1 says. Now if f is
/(x0 + Q-/(x0)\ "'
*
and
only if all the limits on the right exist equality holds also in the limit. That is all that
differentiable function
interval taking values in R", its f'(x0) is a vector in R" and points in the direction of motion of the curve (Figure 3.1). That is, the line through f(x0) and parallel to f'(x0) is the limiting position of the line through f(x0) and a nearby point f(x0 + t). For that line is parallel to t~ 1(f(x0 + t) f(x0)), 0. This line through and by definition this vector has f'(*o) as limit as t f(x0) and parallel to f'(x0) is called the tangent line of the curve at f(x0). From Proposition 1 it easily follows that iff, g are differentiable, so is f + g, and (f + g)'(xo) f'(x0) + g'(*o)- The chain rule also follows easily:
image is
a
a curve
in R".
on an
The derivative
-
=
Proposition 2. (Chain Rule I) Let g be a real-valued function defined in a neighborhood of x0 in R, and differentiable at x. Suppose f is an R"-valued function which is differentiable at g(x0) (see Figure 3.2). Then fgis differ entiable at x0 and (f g)'(x0) g'(x0)f'(g(x0)). (We have written g'(x0) before f'(g(x0)) as this is the customary way of writing the product of a scalar and a vector.) =
Figure
3.1
230
3
Ordinary Differential Equations
i(g(x))
x
g(x)
Figure 3.2 This is of
ordinary
course
true,
just
because it is true in each coordinate, by the ,/), then fg (fig,---,fg), (fu
Thus if f
chain rule.
=
=
.
.
.
so
(f )'
=
=
((/i )',. --,(/s0') (j i0', ,/; 0')
=
0'f
Example 1. Let
f(x)
=
(x, x2, x3), g(t)
=
sin t.
Then
(f g)(t)
=
(sin
t,
sin2f,
sin3 0
(f g)' o
=
cos
t(l, 2 sin t, 3 sin2f)
Now, there is also
chain rule for
differentiable function
taking a real-valued function of a (Figure 3.3). Suppose now g is a continuously defined on an interval / taking values in a domain D
in R".
a
a
vector-valued function
partial
Suppose /
is
real-valued function defined
derivatives continuous.
Then/
g is
a
on
D which has all
real-valued function
on
the
interval /. For
clarity of exposition, let us take the as g(x) (g^(x), g2(x)). Then
coordinates
/(g(*o
case n
=
2.
We
can
write g in
=
+
0) -/(g(*o))
=f(ffi(x0
+
t), g2(x0
+
t)) -f(gi(x0), g2(x0))
=f(gi(x0 0. gi(xo 0) -f(gi(x0), g2(x0 +JXffi(x0), g2(x0 + t)) -f(gi,(x0), g2(x0)) +
+
+
0) (3.1)
3.1
231
Differentiation
f(s, g2(x0 + 0) is differentiable (it is the restriction of / g2(xo + 0)- By the mean value theorem, the first difference is
Now the function to the line y
=
Pi f
-ir (i> 02(*o + 0)[ffi(*o + dx for
some
theorem
between
t,x
we see
0i(*o for
some
+
gt(x0
+
t)
and
ff i(*o)]
Now
gt(xQ).
applying
the
mean
O-0i(*o)=0iOi)'
^r(Si,02(*o
+
dx
9i(x0)
<
Thus the first difference in
.
O)0'i("i)<
$i
Similarly, the second
dy
~
that
nx between x0 + t and x0
df
0
<
0i(*o
+
<
<
x0 + t
0
x0
0
x0
nx
difference is
(0l(*o) >2)9l(*l2)t g2(x0)
<
2
<
g2(x0
+
+ t
/(gU))
Figure 3.3
(3.1)
is
value
232
3
Thus,
Ordinary Differential Equations may rewrite
we
/(g(x0
+
Q
(3.1)
as
/(g(x0))
-
t
=
yx i, 92(x0
Taking the limit as tinuous), and g2(x0
t
+
-0,
+
we
t), 2
0)ffi(li)
+
have
the
on
both tend to
(9i(x0), kteaOh)
right ^ -*#i(x0) (since
Notice that,
using
<*(/g) dx
true
(x0)
along only
in not
Y% (g(*o))0i(*o)
+
=
df(g(x0), g'(x0))
of/ along the
=
exists,
con
so
y (g(*o)) 02'(xo)
the directional derivative
Thus the derivative
derivative
=
gt is
since g2 is continuous. Also lie between x0 and x0 + t. Since all the
g2(x0)
!, w2 bth tend to x0 since they derivatives in (3.2) are continuous, the limit
^^ (xo)
(3.2)
(3.3)
notation, (3.3) becomes
curve x
=
g(x)
is the
same as
(3.4)
its directional
the tangent direction to the curve (Figure 3.4). This is R2, but for all R". The derivation is of course the same,
only with the notational complication of many
Figure
3.4
more
variables.
Thus
3.1
Differentiation
233
Proposition 3. (Chain Rule II) Let g be a continuously differentiable function of a real variable, taking values in a domain D in Rn, and suppose f is a continuously differentiable real-valued function defined on D. Thenfo g is a differentiable function and
(f g)'(0
=
df(g(t), g'(t))
Examples 2. Let
g(t)
=
(sin t,
t),f(x, y)
cos
(a,b))=d-fa 8-fb dx dy
df((x, y), g'(t)
=
+
sin
(cos t,
(f g)'(0
=
=
2t
y2a
+
Then
2xyb
f)
df(g(t), g'(0) cos
=
xy2.
=
cos2
=
r cos t
+ 2 cos t sin
t(-sin 0
cos t
We can, of course, sin t cos2 t
verify
this
by
direct substitution, since f
g(t)
=
.
3. Let
g(r)
=
(t, t2, 2t),f(x,
y,
z)
=
xy +
log z.
c
df((x,
g'(t) (f
=
y,
z), (a, b, c))
=
ya + xb +
-
z
(1, 2t, 2)
g)'(0
=
df((t, t2, 2t), (1, 2t, 2))
=
3t2
+
=
t2
+
2t2
+
-
-
t
4. mum
Then V/(g(x0)) is
Proposition 3, and/ g has a maxi orthogonal to g'(t0). For (/ g)'(t0)
df(g(t0), g'(fo))
Suppose/ at r0
.
g
are
given
as
in
=
0,but
(/ g)'Oo)
=
=
234
3
Ordinary Differential Equations
Lagrange Multipliers This last example serves to provide a method for finding maxima (or minima) of functions subject to certain constraints. This is the process of Lagrange multipliers. Suppose/, # are differentiable functions in a certain domain D in jR". We consider/as the function we are studying and g(x) 0 the constraint. Suppose / has a maximum on g(x) 0 at x0 if T Thus, is a curve in the set {g(x) 0} going through x0 then V/(x0) is orthogonal to the tangent line to T at x0 For if T is the image of a function variable, 0, and (b(t0) x0 is the line to T at Now also '(t0)y point xQ / {g(x) 0}, V/(x0) and Vg(x0) are both orthogonal to all curves through x0 subject to the constraint g(x) 0. If there are enough such curves, say, so that the set of tangent vectors fills out a subspace of R" of dimension n 1 then and must be We not here that collinear. will there are V/(x0) V#(x0) worry of these but take it for After we are not here curves, all, enough granted. studying the theory, but only seeking a technique which will provide candi dates for a maximum point. We can state this principle : if x0 is a maximum (or minimum) point for /subject to the constraint g(x) 0, then there is a A =
=
.
=
,
.
=
=
,
.
=
=
=
,
=
such that
V/(x0) Thus
we can
V/(x) g(x)
=
Mx0)
find =
=
possible
x0
by solving the system of equations
Xg(x) 0
(3.5)
for x, X.
Examples
x2
5. We shall find the maximum value of xyz on the unit 1. + y2 + z2 1. Let/(x) xyz, g(x) x2 + y2 + z2 =
V/(x) Thus
x2
(yz,
+
(yz,
=
xz,
we must
y2
xz,
+
xy)
z2 =
=
xy)
solve =
1
2X(x, y,z)
Vg(x)
=
=
(2x, 2y, 2z)
-
sphere
3.1
X from
Eliminating xz
yz
Equations (3.6),
Differentiation
235
obtain
we
xy
(3.7)
xyz
This
z
=
can
0
be written
or
x
=
0
as
or
y
0
=
or
-
=
Thus either
y
of the coordinates is
one
=
-,-
x
zero
(3.8)
-
z
y
x2
or
=
y2
=
z2
Near
point where one of the coordinates is zero, / changes sign, so these points are disqualified. This leaves any one of the points l/>/3(+l +1, +!) The value of /at any one of these points is any
thus 3_3/2 is the maximum.
3_3/2,
+
l)2
6. Find the
point on the curve 2(x totheorigin. Here#(x,y) 2(x l)2 =
-
+
+
j2
3y2 4
-
=
4 which is closest
and/(x, y) =x2
+
y2.
Thus
V/= (2x, 2y)
Vg
The
equations
x
=
2X(x
y
=
2(x
-
1), 2y)
1)
Xy
l)2
-
gives
x
values
+
y2
4
=
off at
/
7. Find the
+
is 6.
curve on
1
y2
+
either y
2z2
=
8
=
0
or
(1 Ji)2, (1 are
-
X
=
1.
The second
case
V^O), (2, + ^2). The + V2)2 '> and at the s_econd
pair is (1 Clearly, the minimum (see Figure 3.5).
the first
the maximum is 6
=
equation,
Thus, the candidates
=2.
the value of
x2
(4(x
become
From the second
xyz
=
distance is
|1
the intersection of the two surfaces
-
v
2| and
236
3
Ordinary Differential Equations
Figure
3.5
origin. In this problem we have two constraints, through the technique. The tangent vector to the curve is orthogonal to the gradient of both constraining functions, and at the maximum point V(x2 + y2 + z2) is orthogonal to the curve. Thus this gradient must be coplanar with the gradients of the 1, constraining functions. Let f(x) x2 + y2 + z2, g(x) xyz 8. Then V/=2(x, y, z), Vg h(x) x2 + y2 + 2z2 (yz, xz, xy), Vn 2(x, y, 2z). We must solve these five equations for x, y, z,X,\i: which is closest to the but
we
can
see
=
=
=
=
-
=
2(x, y, z) xyz
x2
=
+
X(yz,
=
xz,
xy)
+
zfi(x,
2z)
1
y2
+
2z2
8. Let M
V7/x, x
=
We show this
=
=
=
for all i and /
8
(a/)
be
a
symmetric
n x n
If T is the transformation
matrix.
on
=
a?
27x
by computation:
(3.9)
j
The kth component of VTx, with respect to x\ this gives
Ifl*'x' + Zfl7v j
That
is, af R" defined by M,
2 fly'xV ".
*
y,
x
is found
by differentiating (3.9)
3.1 But since M is
2(Tx)k. /OO The
Then
symmetric, this is the
Vrx, x
=
(Tx, x> must attain a maximum on Lagrange multiplier procedure tells us =
V(2xi2-l)|x=Xo
Thus the transformation T has unit
sphere
2; Ok'x'
+
or
y ak}x>
=
'function
Now, the the unit sphere, say at x. that there is a X such that
2Tx
=
2Xx
eigenvector, namely that
an
237
x0
on
the
which maximizes the function (Tx, x>.
continue this idea in order to prove that a transformation given by symmetric transformation has an orthogonal basis of eigenvectors. For, We
a
same as
2Tx is established.
=
Vrx,x|x=Xo
Differentiation
can
let x1 be the eigenvector found as in 1 subject to the constraints <x, x> =
point subject ||x2 1|
to
=
1,
these
constraints,
<x, Xl>
=
0,
we
Example 8. Now maximize (Tx, x> 0. If x2 is the maximum <x, xx > have X2, \i2 such that =
,
2Tx2
=
2X2x2
,
2Tx2
=
/i2Vx, xx
Thus, by the first two equations, x2 is nonzero and orthogonal to xl5 and by the third, x2 is an eigenvector of T. Now proceed to the constraints 0. The same technique works to produce 1, <x, x^ 0, <x, x2> <x, x> =
=
third
eigenvector. eigenvectors. a
=
We
can
go
on
until
we
have found
n
independent
Examples 9. Let
eigenvectors of M. eigenvector of M if and only if there is a nonzero vector x such that (M XI)x 0. We know the necessary and sufficient condition for that: det(M AI) 0. Thus the eigenvalues of M are 0. Now the roots of det(M XT) and find the X is
an
=
-
-
-
=
=
238
3
Ordinary Differential Equations
After
a
det(M
computation AI)
-
(2
=
we
A)3
-
find that
3(2
-
A)
-
+ 2
=
-(A
We find the
eigenvalues are 1,4. by solving the equations (M
Thus the
I)x
-
=
1)2(A
-
4)
-
corresponding eigenvectors
0, (M
-
4I)x
=
0 for
nonzero
vectors.
1
eigenvalue
:
/I
1
1\
1
1
1
\l
1
1/
M-I=
corresponding eigenvectors: (1, -1, 0), (0, -1, 1) (Any two independent vectors such that vx + v2 + 1-2
1
1
-2
M-4I=
\ sum
0 will
do.)
1\ 1
1-2/
1
is zero, so independent, so the
of the three
and second on
=
4:
eigenvalue
The
v3
are
rows
they are dependent. The first corresponding eigenvector lies
the line
-2x+
2y
x
Such
(0,
y +
a
+
z
=
0
z
=
0
vector is
1, 1)
with
Here
det(M
of
eigenvalues
AI)
-
1, and
eigenvalue
10. Find the
eigenvalue
1
eigenvalue
5: M
:
=
(2
M + I
51
A)2
-
=
=
-
9 which has the roots -1,5,
(3
3\
,
_|
I
1-3 I
eigenvectors of M are (1, (1, 1, 1) with eigenvalue 4.
Thus the
(1, 1, 1).
,
kills the vector
3\
(1,
1).
.1 kills the vector (1, 1).
-
1, 0),
3.1
Differentiation
239
EXERCISES 1. Differentiate these functions and
graph the
curve
function
defined by the
(a) f(t) e", c a complex number. (b) f (f ) (cos f, sin f, f ). (c) f (t) (a cos t, b sin t). (d) t(t) (t2,t3). (e) t(t)=(t,t2,t3). (f) f(f)=(sinf,cosf,0). 2. What is the length of t'(t) in each of Exercises l(a)-(f ) ? angle between f '(f) and f "(f)? =
=
=
=
3. At which
pairs of points
are
and (c) of Exercise 1 parallel? 4. At which pairs of points
the tangent lines to the
are
What is the
curves
(a) (c
the tangent lines to Exercises
=
i),
1(b), (f)
parallel? 5. Find the maximum of xy 6. Find the minimum of
on
the
ax2 +
ellipse
by2
=
1.
+ y on the curve xy 1 in the first quadrant. 7. Find the two points on the curves y x2 and xy 1 which are closest. x
=
=
=
8. Minimize x2 +
y2 + z2 on the ellipsoid ax2 + by2 + cz2 1. straight lines L1 and L2 in space how would you try to find the points PieZ,1, p2eL2 which are closest (i.e., minimize I'p q|| for =
9. Given two
peL\qeL2)? 10. Find the eigenvalues and
(a)
(I !)
eigenvectors
of these matrices:
("i ?)
*>
(b) 11
Find the
.
(a)
eigenvalues
and
1
0
0
1
0
\-l
0
3/
eigenvectors of
-1\ .
(b)
these matrices :
/2
1
3\
1
0
3
\3
3
PROBLEMS 1. Let f, g be differentiable /{"-valued functions defined on an interval I.
h'
=
=
=
are
240
3
Ordinary Differential Equations
rectangular box of maximum volume is to be constructed, with sides parallel to the coordinate planes, one vertex at the origin and the diagonally 1 Find the volume of that opposite vertex on the plane ax + by + cz 3. A
=
.
box.
community consumes water at the rate of sin2(27rf/24) gallons per They wish to build a storage tank of capacity Q with a pump of rate The w gallons per hour, so that the community will never run out of water. Minimize this cost for them. cost is Q + kw. 5. Show that if /is any differentiable function on R3, there are at least two points x on the unit sphere at which V/(x) is parallel to x. 4. A
hour.
3.2
Formula
Taylor's
order derivatives appear for vector-valued functions one- variable calculus.
Higher
just
as
they
do in the usual
Let f be
Definition 2.
an
.Revalued function defined
f is A:-times differentiable
UcJJ.
on
on an open set U if there exist differentiable function
defined on U such that gt f, g2 g't , , gk gk gl5 denote gk by fm. f is Ac-times continuously differentiable f e Ck(U)) if f(t) is continuous on U. =
.
.
=
=
.
.
The
following proposition
is
an
.
.
,
obvious extension of
g'k-! on
We will
U
(written
Proposition
1
by
induction.
Proposition 4. Let f= (ft, ...,/,) be an R"-valued function defined on U. f is k-times (continuously) differentiable on U iffu ,/ are each k-times (continuously) differentiable on U. Further, fm (f^k), ,f(k)). =
.
Knowing be
a
that
a
given
function is differentiable at
great aid in computing approximations
These considerations in turn lead to
a
better
.
.
particular point can nearby points. understanding of the notion of a
to its values at
differentiability. Suppose that/is a differentiable i?"-valued function in a neighborhood of 0. By definition the difference quotient,
defined
-[/(0-/(0)] t
converges to
=
In other
f'(0). -
U(t)
-
/(0)]
-
words, the function e(f) defined for
/'(0)
t
# 0
by
3.2 has limit 0
as t
->
0.
e(t)
8(0
+
Thus
0.
=
Formula
241
Rewriting this,
fit) =/(0) +f'(0)t where lim
Taylor's
a
(3.10)
t
good approximation
to
the value f(t) would be
<->o
/(0)
+ /'(0)' ; how
good depends
of
the function
course on
But since
e(t).
the difference between this approximation and f(t) is e(t) t, it suffices to know just the maximum of |e(f)|. We give an illustration of how to go about determining this.
Suppose /is M
a
C2 function defined in
an
interval \_-R, R].
Let
sup{|/"(x)|:|x|
=
Then
1/(0
0/(0) +/'(0)0l
-
This follows
<
from the
easily
for t
MR\t\ mean
e
(3.11)
[-*, K] There is
value theorem.
between
a
t and 0 such that
Further, there is for
a
given
(0
* e
=
=
an n
between , and 0 such that /'OS)
Thus,
-f'(0) =/"(")
[-R, R],
^C(/(0-/(0))]-/'(0) f'(0
-
/'(0)
=
/"()
r,,Zer_-R,K]
from (3.10) and this inequality. |(f)| < MR. Inequality (3.1 1) follows describe the function to difficult adequately be very Now, although it could In /" is monotonic obtain. to practice, easier much e(0, the maximum M is at the end points -R and R to near 0 so we need only look at its values
Thus
obtain this estimate.
We shall
obtain estimates which
are even more
now
generalize
this argument
in
order to
accurate.
illustration above
we can
assert
special Rereading Equation (3.10) of the a at function a point shows us how the values of that differentiability a firstof values the function at nearby points can be well approximated by that the error is sma order polynomial. (Well approximated here means this well and the
relative to the distance between the two points.) approximability is a criterion for differentiability.
Furthermore,
242
3
Ordinary Differential Equations
Proposition 5. Suppose that fis an R" -valuedfunction defined in a neighbor of x0eR. f is differentiable at x0 if and only if there exists a linear s(t) 0 function L: R^R" and a function e defined for small t such that lim (-.0 hood
=
and
f(xQ
+
t)=f(x0)
+
Furthermore, L(t) =f'(x0)
L(t)
+
e(t)t
t.
Proof. We have seen above that differentiability implies this condition. versely, suppose this condition is verified. Then r
hm
/(* + ') -/(*> =
f
I-.0
for since L is linear,
Now,
L0)
L(t )
fL(l).
=
..
.
, e(t )
,. h lim
,
f
t-*o
=
hm t-o
r-.o
L(t) =
f
L(\)
Thus /is differentiable at x0
evaluation of /(f) for t
approximate
an
,lim
near
Con
,
and /'(x0)
0 with
error
=
L(\).
that is
small relative to |r| may not be as good as required. A better approximation would be one whose error is small as compared to t2, or even better \t\k for
sufficiently large
We shall
now
derivation follows
higher order derivatives come in. gives such approximations. The
This is where the
k.
derive
a
theorem which
by induction directly
from the above remarks.
Theorem 3.1. (Taylor's Theorem) Suppose that f is a (k + l)-times con tinuously differentiable Revalued function defined in an interval I about x0 Then there is a polynomial P (with coefficients in R") of degree k, and a function s defined for t in I such that .
(i)
e(t)
is bounded
by max{|/(lt+1)(x) |
: x
between x0
andx0
+
r},
E(i)tk+1
(ii)
/(x0
+
0
F(0
=
+
A^
(3-12)
Furthermore, P is unique and is given by
P(t) If
we
=
write
/(x0) x
=
+
/'(*o)'
+
^
t2
x0 + t, (3.12) becomes of degree k about x0 :
+
+
i^
tk
a more
familiar
x0)(x
xoy+1
expression,
called
Taylor's expansion fix)
=
I
>=o
rr2 (x i-
-
x0y
+
b(x
-
-
(3.13)
3.2
Taylor's
1 by induction on k. The case k 1 and n above. We now assume the proposition for k applying the induction hypothesis to /'. For simplicity is
proof
The
Proof.
=
=
I
=
243
Formula
already discussed n, by
was
prove it for k we
take
x0
=
=
0, and
{x: \x\
Let tel.
Now let
t}.
e0(t) is bounded by M integrate (3.14) from 0 to x:
+ 1>
since/'(0=/(,
us
Here
The
/""+1>(0) r
"-1
<*
rfr
/'(f) integral
we can
=
T1
I
f*
I
*' dt + -,
max{|/(*+n(x)|
: x
between 0 and
/-,
kn
(3-15)
*o(0" *
is, by the fundamental theorem of calculus, /(x)-/(0).
the left
on
1
=
write
Thus, letting
ix)
we
=
n+ 1 C* -^r x Jo
e0(t)t"dt
obtain from (3.15) n+l
m =/(o) which is
just the
+
2 J
771
^T ^TT
eW
We must show that
(3.12).
same as
+
e(x)
is bounded by M.
But,
"W'-SH* ^(Ol^f<^M{;r.f<M since e0 is bounded by M.
Examples
Taylor expansion
11. Find the
1 +
/(I)
3t4.
t +
=
/"(1)
thus the
f(t)
/'(l)=l
5
=
=
of degree 3 about 1
f"(l)
36
+
=
12'3|,=
5 +
13(t
-
D
+
and
72
Taylor expansion 180
=
i
13
/(4)(0
=
72
is
-
D2
+
120
"
D
,
+
(0
"24
'
4
of
f(t)
=
244
Ordinary Differential Equations
3
where
|e(t)|
<
72.
/(5)(0
Notice that, since
=
> the Taylor expansion of degree 4 is
accurate :
f(t)
5 + 13
=
for all
+
18(1
l)2
-
+
12(f
-
I)3
+
30
-
I)4
t.
12. Find the
(1
1)
-
Taylor expansion of degree
4 about 0 of
t2)-1
+
/(0)
1
=
/'(f)=-2f(l
+
f2)-l
/'(0)
=
0
-2 /"(0) f2)"1 + 4f2(l + r2)"1 2 /'"(0) 8f3(l + f2)~ f'"(t) 4*1 + t2)'2 +St(l + t2)-1 12 /W(f) 4(1 + *T2 + 8(1 + t2)-1 + t[_- ] t2 + f + e(t)t5 1 f{t)
f"(t)
-2(1
=
=
+
-
=
=
/(36)
x
f(x)
\x-112
=
=
=
/'"(36) for
(40)1/2
three decimal
to
places.
y/x about 36.
=
f'"(x)
0
-
13. Calculate
f'(x)
=
=
=
f(x)
/(f)
f"(x)
\x-^2 6
=
/W(x)
/'(36)
=
^x-7'2 f"^
=
^
^ l/WWI^
6 +
+
=
1
Thus
between 36 and 40.
=
\x-3>2
=
1 (x
-L (X 8.6
-
-
36)
36)3
+
^ (x
+
\e(x 6
-
-
36)2 36)4(x
-
36)4
We
expand
=
3.2
where
(40)
e(0
<
15/16.67.
Taylor's Formula
245
Thus
i/2-H+8
67
6 16.67
and the desired
approximation
is 6.334.
14. Calculate e4 to three decimal places. We first write down the have Taylor expansion f(x) ex about 0. Since f'(x) =/(x), we Thus the Taylor expansion of ex, degree n is ex for all x. =
/
=
xn+1
x'
"
/-,
(316)
'-S^^oTTiii
e* we now |e(x)| < max{| e'\ : 0
where
00.xn+1^344l)! _(n +
(n+1)!
3
large that this is bounded by 10" by the following succession of inequalities
We must choose
do,
as we see
4*+ 5
1
3*4"+
n so
(7+7)!
22n+1
zr^
~
~
2~3^ri
n >
41 will
1 ~
ir3"1
1 < ^
Thus
we
irj3/10(n-31)
must have
(3/10)(n
Taylor expansion (3.16) dominated by In the
-
31)
<
3,
or n >
41.
of ex observe that the remainder
term
is
x"+1 e"
(n
+
1)!
and therefore tends to
zero as
-
a..
Thus, if
we
let
n
-
oo
in
(3.16),
we
246
3
Ordinary Differential Equations
obtain (once
i
again)
=
I-
0
Now, this kind of
happen
differentiable in
fix) where M"
+
argument
an
be
can
applied
to know has derivatives of all orders.
1(x)
interval /about x0
/(x0)
=
|e(x)|
an
E
+
(x
max{|/("+1)(f) | :
<
+
-
x0)
+
That is, if
write the
e(x)
between x0 and
(w
x}
/ is infinitely Taylor expansion
+
(317)
1),
valid for every
Let
n.
If
be thus bound.
limM"
t
we can
to any function which we
1(x)(*,~Xi"+1=0 in 1)
(3.18)
+
as n -+ oo in (3.17) and represent /as a Taylor Expansion of/aboutx0. In Chapters 5 and 6 we shall return to the consideration of series expansions for functions. In Section 5.8 we shall construct infinitely differentiable functions which are For the present we mean only not represented by these Taylor expansions. to remark on these approximations of the Taylor expansion as a tool for
then
clearly
series.
take the limit
can
we
This series is called the
approximation. Examples 15. Consider
f'(x)
=
/(4)(x)
/"OO
cos x
=sin x,
and the
.
.
+
/<4" 4>(x)
cos x
=
sin
/'"OO
x
cos x
Thus
/(4n 2)(x)
about
=
sin
x
/(4"+3)(x)
zero
is thus found to be
7
^
X
+
=
x
T
f(x)
sin
We have
x.
+
=
X
x
sin
itself.
Taylor expansion
=
=
=
.
cycle repeats
/<4"+1>(x)
The
now/(x)
X -
+ + Remainder term
=
-cos x
3.2
Since all derivatives of sin
bounded by 1, is bounded by
are
x
one
the remainder for the
so
Taylor's
Formula
of +sinx, +cosx, Taylor expansion of
247
they are degree k
1
(k
iy.
+
which tends to
zero as
k
x5
x7
x3 sinx
=
x
__
accurately compute
+
x2
16. Find sin
bound
on
expansion
\kfk+ 1
(-1/2*
sine
+
to
'"+l2l0r an
ensure
+
calculating
this bound is
<
<
4/5
to
We need to compute terms of the Taylor
n
10" 3.
"
n/4
can
we
'"
10"3.
accuracy of
Similarly,
(see Exercise 15),
Now the remainder
by [_(2k + 1 ) !] \nj4)2k +1. verify that k 3 will work:
after k terms is bounded fact that
...
sum.
for the cosine
the remainder after
and then
infinite
(-l)kx2k
+
6!
n/4
an
as
x6
x4
Taylor expansion
..+___ +
+
Taylor expansion
a
COSX=2!-4!
a
___
expresses
Thus the
->oo.
We shall
use
the
=
I^7
an
n
n3
4
6.64
estimate to sin
+
17. The
n/4
to
within
one
thousandth is
120.45 is infinitely differentiable around the point 1. we infinite Taylor expansion there ? By computation,
logarithm
Does it have
an
find
log(')(x) log<">(x) log('")(x) log<4>(x) log()(x)
=
=
=
x"1
log'(D
=
l
log(")(D=-1 2x"3 log('")(l) 2 13.2x"4 log(4)(l) (- 1)3 2 log(n)(l) (-!)"(-!)! (-l)"(n-l)!x-" -x"2
=
=
=
-
=
=
(3-19)
248
3
Ordinary Differential Equations The
Taylor expansion
of
degree
n
about 1 is thus
^'-S^^^fr-tf ^OT +
Notice that from the first
|e(x)|<(n)!x"<"
+
equation
1
/
x-
1\
x
1
zero
3/2. Thus, Taylor series
=
is bounded
by
n
->
long
1
oo, so
that the
in the interval
1/2
as
> x >
1/2.
Similarly,
remainder goes to zero if < x < 3/2, the logarithm has
l)k
(x
00
log(x)
as
(Exercise 18)
< x <
the
(3.20)
J
show
can
1,
\n+1
1
which tends to we
<
1)
and thus the remainder of
n+
(3.19), if x
of
(3.20,
I(-l)^-X K
i=l
EXERCISES
12. Find the
Taylor expansion about the origin of degree 5 of
tan x;
(l+*)-\ 13. Find sin \
14. Find
V3 accurately
15. Derive the
1 6. Find
(1
+
x)-1
accurately
=
an
to 4 decimal
to 4 decimal
=
places.
Taylor expansion (given after Example 15) of cos
interval about the
origin in which the substitution
l-x
is accurate to three decimal places.
(l+x)-1
places.
l-x + x1?
What about the substitution
x.
of
3.2 17. Find
interval about the
an
x2 e*
l +
=
x
+
T
Taylor's Formula
249
in which the substitution
origin
x3
+
-
is accurate to three decimal
places.
18. Show that the series
(x
1)<
-
k
1=1
represents the logarithm in the interval l/2<x<3/2. series converges for all
in the interval
x
(0, 2).
Observe that the
Does it converge there to
logx? PROBLEMS
Suppose/is a A>times differentiable real-valued function defined on the 0 for all k. Show that / is a polynomial of Suppose /
interval /.
=
.
=
=
Prove that
/0)_/""(0) 9m(0)
9(t) 8.
(Taylor's form of
the interval [ R, R]. such that
1
=
9. Let
m
be any nm
X
fix)
I
=
n=
o
mean
value
theorem) Suppose that /is C on [R, R], there is a between 0 and f
t e
K\
11
0
the
Show that for
(mn
integer and define the functions f0
,
.
.
.
,
fm-i by
+ I
+
i) !
( (a) e*=fi(x)+---+fm(x). l,...,/w-l. (b) //=/i-i for ; (c) /o =/m-l(d) The functions /i, ...,/ are all solutions of equation =
y(m)
_
y
the differential
250
3
Ordinary Differential Equations (a) Suppose that /is continuous
10.
0(0=
f
the interval [ R,
R].
Define
te[-R,R]
f(tx)dx
>0
on
and show that g is also continuous. (b) Suppose that h is C1 on [-R, R]. Prove that there is a con tinuous function k such that h(t ) h(0) + tk(t ). (Hint: Consider tx.) '(T) dr and make the substitution t =
jo
3.3
=
Differential
Now,
Equations
ordinary differential equation is (roughly speaking) an equation function/ and some of its deriva
an
involving the variable x, an "unknown" tives/',/", .,fk\ Thus ..
fix)
=
k(x)
f"+f=0 f'(x)
=
x/(x)
C/(4)(x)]2
e*''w
+
=
|/(3)(x)|
+
log|x
+
1|
examples of differential equations. A solution is a function which makes equation true. For example, |5 k, sin x, exp(^x2) solve the first three equations respectively (as for the fourth, we cannot easily exhibit a solution). We prefer to think about differential equations in this vague sense rather than to try to attempt a formal definition of such, so we shall do so. Many equations do not admit solutions and some equations admit many. are
the
Consider these :
|/|
(y'f y"
|y-x|
+
+ i
+ y
=
The first has and
=
=
0
o
0
no
solution y
=f(x),
because
we cannot
have
both/(x)
=
x
/'(x) supposed
0; the second has no solution because the derivative of the solution would be imaginary. The third equation has as solutions
sin x,
x,
cos
must be
if
we
=
as
well
discarded
as
any linear combination of these.
The first
equation
the second admits solutions
being self-contradictory; permit ourselves to consider complex-valued as
functions.
As
we
shall
3.3 see
this turns out to be
the third The
Differential Equations
very fruitful course, for it
a
251
permits understanding
well.
as
of calculus derives from the fact that it is necessary to the problems (mainly derived from the study of physics and the natural sciences). These problems usually are stated mathematically
importance
solution of concrete
as
differential
equations.
Examples 18.
A bank likes to pay its depositors on the deposited and the length of time they have been able to use these deposits. Thus every (say) June 30 your bank would add to your deposit an amount equal to (say) 5 % of that part of your deposit which they have held for the past year (and if they are decent about it a reasonable fraction of that 5 % for parts left in for fractions of that year). Many years ago, that great financial wizard, L. Waverly Oakes, pointed out that that amount that he kept in his bank for the first half-year was working for the bank and he should be paid for it. Furthermore, argued Mr. Oakes, the payment he
Compound
interest.
basis of the amount
should have received so
also
Finally,
was
also sunk back into the bank's investments
income for the bank, and thus for its depositors. Mr. Oakes pointed out that there is nothing special in half a
earning
was
"
Over His very words were any particular fraction thereof. a the of how of no matter small, particular time, earning any period balance relative to that balance should be directly proportional to year,
or
period of time. In order to best approach the interest due its depositors, our banks should be computing interest as often as possible." The banks all responded to this profound utterance by recomputing their interest every month instead of every year. Some body even suggested that, with an army of secretaries, they could so that
compute the accrued interest every 30 seconds. would have rested
were
it not for
an
And there the matter
obscure student of Isaac Newton
who dabbled in the stock market. at time
Suppose
f(t) according t-y ^
t0
f1
^ t
to n
a sum
of s0
pounds
are
deposited
in the bank.
t0 be the balance accruing from this deposit the Oakes system. Then, Oakes' assertion is, for all
for all times
Let
t >
,
fih) /Pi) fih) -
=
k(t2
-
(3-2D
h)
where k is the earning power
(interest rate)
of money.
The first
252
3
Ordinary Differential Equations
thing this brilliant person remarked is that (3.21) cannot possibly always hold. Let us illustrate his discussion. Suppose that 500 pounds are deposited in the bank at a 5% per
/(0) 500 and at the end of one year, pounds, so/(l) 525. Now, if interest is computed we obtain, by (3.21), half-year, Then
interest rate.
annum
the interest is 25 every
=
=
^-^ 0.05ft)
K2>
500
or/(l/2) /(l)
=
512.50.
Then,
over
the second half of the year,
we
obtain
512.50
-
=
512.50
0.05ft)
by this computation /(l) 525.31. As this is closer to the actual earnings of the initial deposit, this is more like the amount the depositor should get. Furthermore, this semiannual computation has neglected the earnings during the last three-quarters of the 6.25 accrued during the first quarter. In fact, when we compute the interest quarterly we find that the value of/(l) should be no less than 525.504. And so it goes: no matter what period we choose for the computation of interest, we will be neglecting the interest accrued by the growing total during that interest. Thus Oakes' formulation cannot be correct. However, our student was moved by the basic justice of Oakes' ideas and after rewriting Oakes' formula as so
that
f(t2) :
h
~
=
fih) :
_
=
.,v fc/Oi)
h
he asserted that he had found the
precise
statement of the Oakes
Oakes should have said "over any infinitesimally small interval of time ..." rather than over any period of time, no matter formula.
"
of change of the balance time; that is,/' kf, proportional where k is the interest rate (0.05 above). Thus, the problem is to ky 0 so that find a solution / for the differential equation y' how small ..." at any
time is
Precisely, then:
the rate
to the balance at that
=
=
/Oo)
=
*o
Population explosion. Population tends to grow also according differential equation. That is, it is assumed that every individual has the same propensity to reproduce and that propensity 19.
to the above
3.3 is
independent
of time.
Differential Equations
253
Thus
over any infinitesimal period of time population to the initial population is proportional to the time elapsed. (You know what mobs are like: the larger they are the faster they seem to grow.) This assertion is supposed to be true for brief periods of time ; thus we should more precisely assert that the rate of change is directly proportional to the total population; thus if/ is the total population,/' kfi where the constant k is called the growth rate. In some societies the growth rate varies with time; among certain mammals it peaks at certain times of the year. In these cases the population as a function / of time satisfies a differential equation : f'(t) k(t)f(t), where k(t) is the variable growth rate. It may even happen that the growth rate depends on the total population; in a well-regulated society (1984) this would be the case. Then the population function is a solution to a more complicated equation, / Ky)y.
the ratio of the increment in
=
=
=
20. Survival
On
of thefetahs.
Pacific there live only two garibs. These animals are
everpresent undergrowth
to
a
remote volcanic atoll in the
South
species of animals, the fetahs and the essentially vegetarian and there is an feed them. However fetahs especially
garibs and garibs find the succulent fetahs hard to resist. Now each fetah tends to reproduce at the rate of one young each per 7 per year. Conversely year, and consume garibs at the rate of eat fetahs at the rate of and the garibs have only one young per year of fetahs and garibs in a 17 per year. Thus the increment A/ kg year should be given by love to eat
A/=/0
-
Hg0
where f0,g0 the Oakes
are
Ag
=
g0-
the initial
reasoning
lf0
(3-22)
populations of these groups. However, applied to this case; because as the
must be
continually affect the increment. The solution is, as in the above case, to rewrite (3.22) as a differential and garibs at equation. If f(t), g(t) are the populations of fetahs and g: time t, then these equations describe the growth off
population changes,
f'=f-iig
it will
g'=g-7f
less remote island there are n different on the growth patterns species of animals, all of which have some effect of all the others (some feed on others; some house, or protect others). 21. The biotic matrix.
On
a
254
3
Ordinary Differential Equations
This kind of
society
can
be
represented by
a
biotic
nx n
matrix
(i,j)th entry is described as follows: The increment (a/). in the rth species in one year which is attributable to each member of the y'th species is a/. (Thus the effect of one member of the /th species on the rth species in an interval of time Af years, is a- At .) If A
The
=
/0)
=
(f1^),
.
.
.
,/"(0) is the population function equation must be satisfied:
on
this island,
then this differential
(3.23)
f'=Af We consider
22. Particle motion.
in R".
Let
f(f)
be the location of that
Revalued function of at a time
a
t0 is the limit
real variable. as t
->
now
the motion of
particle
at time t
The rate of
a
particle
f is thus
.
change
of
an
position
t0 of
-J_(f(0-f(to)) i0 i
f'Oo), called the velocity of the particle at f0. change of velocity, f ", is the acceleration of the particle.
thus is
f
'
The rate of
The velocity vector has both magnitude and direction; we can write (f) v(t)T(t) (at least when f # 0), where v(t) is a positive function of t, '
=
T(f) is a unit vector. T(f) points out the direction in which the particle traveling at time t and v(t) is the speed at which it is moving. Also, v(t) The length of the path that the can be given the following description. particle traces out in a certain period of time is the distance traveled by the particle. \v(t)\ is also the rate of change of that distance at time t. We will have to await a full discussion of arc length (Section 4.2) before justifying this; however some heuristic arguments are possible (see Problem 20). According to this description, the distance s(t) traveled by the particle from time t0 to t is a solution of the differential equation y' ||f '(t) || with s(t0) 0. We would hope that there is only one solution, for there is no further way to determine this function. (Fortunately, by the fundamental theorem of calculus this problem has a unique solution.) Consider, for example, a particle moving on the unit circle in the plane. Let s(t) be the arc length on Let / be the position function of this particle. the circle from the point (1, 0) to /(f) at time f. Then (since arc length on the unit circle is the same as the angle)
and is
=
f(t)
=
(cos s(t),
sin
s(t))
=
3.3
Differential Equations
255
velocity vector is /'(f) s'(t)(- sin s(t), cos s(t)). Notice that l/'(OI, giving further weight to our description of speed above. k'(OI Notice also that /'(f) is tangent to the circle at the point /(f); this reflects the fact that the motion is constrained to the circle. Differentiating further,
The
=
=
we
find that the acceleration is
f"(t)
s"(t)(- sin s(t),
=
=
cos
s(t))
+
s'(f)2(- cos s(t),
-sin
s(t))
s"(t)T(t)-[s'(t)-]2f(t)
Thus the acceleration has
to the circle
(in the direc motion) magnitude change speed, and a to the direction of whose motion, component perpendicular magnitude is to the For if the is equal speed squared. example, particle rotating around the circle with constant speed, it is accelerating toward the center of the tion of the
a
component tangent
whose
is the rate of
of
circle.
According to Newton's laws of motion the situation is as follows. Given particle at time f 0 situated at p0 and having velocity v0 all further motion is determined uniquely by the forces acting on the object. The motion is determined by this law: the acceleration is directly proportional to the force acting on the particle. Thus, in the absence of any forces, if f(f) is the position of the particle at time t, we have a
,
fOo)
=
Po
f'0o)
=
vo
f"0)
=
0
allf
uniquely determined by these conditions. We say that f is a solu tion of differential equation y" 0 with the initial conditions y(0) p0 y'(0) v0. Newton's laws require that the solution exists and is unique. Thus, in the Mathematics bears this out ; the solution is f(r) p0 + fv0 absence of force, a particle will move with constant velocity, that is, in a straight line at a constant speed. Now, in general, the mechanics of motion can be described as follows. There is a function F defined on R" x R taking values in R". The value F(x, f) represents the force that will act on a unit mass acting at point x at time t The function F is called a force field. A particle of mass m situated According to at the point x will experience the force mF(x, t) at time t. Newton's law it will accelerate in the direction of F. The magnitude of this acceleration is determined by or according to this announcement of Newton's and f is
the
=
=
,
=
=
.
law : Force mF
(a
=
=
=
mass
nza
acceleration).
acceleration,
.
256
3
Ordinary Differential Equations
Suppose
we
place
a
particle
of
mass m
at p0
with
velocity
v0 into this
situation at time t0 Let f be the function describing its subsequent motion to Newton's law. Then at time t it is at f(f) and it experiences a according .
force
Thus
F(f(0, 0f"(0
=
have
we
F(f(0,0
Thus f is the solution of the differential
conditions exist
f(f0)
p0 f'Oo) vo In the next section =
=
uniquely.
,
force fields this is the
equation y" F(y, 0 with the initial require that the solution shall show that for smoothly varying =
Newton's laws we
case.
PROBLEMS 11. Find
all
complex-valued solutions
of the
differential
equation
(/)2+l=0. 12. Solve the differential
y(0)
=
13.
equation y'
=
y
with the initial condition
0.
long will it take 100 dollars to double at a compound % per year ? (b) How long will it take 350 dollars to double at the same rate? (c) How long will it take 100 dollars to double at a rate of 10% per year? 14. It is observed that radioactive elements decay into heavy metals. It is assumed that the probability of any given atom decaying is independent of the particular atom. Let k be the probability that a given atom of a particular element will decay within one year. Show that the function / is governed by the differential equation y' ky if /(f) is the mass of the given element after time f.
(a)
How
interest rate of 5
=
15. The time it takes for
the half-life.
If
an
a
radioactive element to halve in
element has
a
mass
is called
half-life of 14 million years, find the
constant k of Problem 14.
Why is Oakes' formulation of the interest problem wrong ? Can you equation (3.21) so that it holds for a specified period; that is, given n, find/so that (3.21) holds for ti k/n, t2=(k+ \)/n, 0^k
solve
=
"
"
=
=
3.3
Differential Equations
257
Figure 3.6 displace the
by an amount ha and let it go. Using Newton's equation governing the subsequent motion.
mass
the differential
law find
18. A certain insect lays its eggs in the flesh of a mammal. Each insect Now every time one of these eggs hatches in a
hatches h eggs per year. horse, it kills the horse.
Assuming the total mammalian population is a equations governing the growth of this insect and horse population if we also know the natural death rate (dt) of the insect and the natural birth and death rates (bH dH) of the horse. Let I(t), H(t) be the population of the insect, horse, respectively. During a period of time Af bH H- Af horses are born, and dH- H At horses die of natural causes. Now each insect hatches hAt eggs during this interval; the probability that its host is a horse is HjT. Thus there are hl(H\T) Af horse deaths attributed to the insect during this time interval. The change AH in the horse population is thus constant T,
derive the differential
we can
,
,
AH
=
bHH At
-
dH At
-
hi
(f)"
3
Ordinary Differential Equations
Find the corresponding
change in the insect population and deduce that these
differential equations govern the growth: hH H'
V
=
I
(bH-da)H-
=hT-dII
by Galileo that the gravitational attraction of the small, we may assume the world is flat, thus we 0 is the surface of the take as a model R3, and assume that the plane z earth. The gravitational attraction then is a force field F(x, v, z) (0, 0, g). Suppose a particle of mass m is at p0 and has a velocity v0 at What is the time f0 Let f (f ) be the position of this particle at time f. differential equation governing the motion of the particle? Can you 19. It
was
observed
earth is constant.
In the
=
=
.
solve for f ? 20. Suppose there is a (c, 0, 0) on our particle, equation of motion. 21.
Suppose that
on
wind
the
coming
out of the east which exerts
matter what the
no
plane there is
a
position is.
centripetal
a
force
Now find the
force field propor
tional to the distance from the origin. At the time f 0, a particle is placed at the point z0 and has a velocity v0. What is the equation of =
motion ? 22. We
ing it by x
=
x(f)
a
can
try to find
line segment.
v
=
a
formula for the
length of a
curve
by approximat
Let
v(f)
be the equations of a curve, and let (x(t 0), y(t0)) be a point on the curve. For a very short period of time. Af, the curve can be replaced by its tangent line
(see Figure 3.7). The length of the curve between (x(t0), y(t0)) and (x(t0 + Af ), y(t0 + Af )) is then approximately equal to ((Ax)2 + (Ay)2)1'2.
V(ax)2
+
(aj) (jr(f
Figure
3.7
+
A/),y(f + if))
3.4 Then the rate of
Some
change of
arc
259
Techniques for Solving Equations length
over
the interval Af is
((Ax)2 + (Ay)2)1'2 Af
Letting Af -> 0 deduce that the rate of change of is the length of the vector (x'(f ), y'(t )).
3.4
for
Techniques
Some
arc
length along the
curve
Solving Equations
The fundamental theorem of calculus is of course the basic existence on solutions for differential equations, and integration is the primary
theorem
tool. Thus
y'
an
h(x)
=
has the solution Let
constant.
of the form
equation
f(x)
us
state the
Let h
Definition 3.
the closed interval
\* h(t) dt + c,
=
=
la, b~]
same
unique
but for
a
result for vector-valued functions.
continuous Revalued function on Define the integral of h over the interval [_a, b~] to be
(hu
.
and this solution is
.
.
.
,
h)
be
a
r>-(jy...j\.) continuous Revalued function defined and leta
Theorem 3.2.
interval (a,
y'
b)
=
h(x)
Let h be
h + p0-
and Proposition 1 By the fundamental theorem of calculus
f(x)=
f
the open
(3-24)
y(c)=Po
has the unique solution
Proof.
on
a
,
h+Po
is another solution, then is differentiable and satisfies the conditions (3.24). If g derivative and thus is constant. g' =/' on (a, b) so each coordinate of* -/has zero Since g(c) =p0 =/(c), this constant is zero, so g =/.
260
3
Ordinary Differential Equations
Separation of Variables There is a class of differential equations which can be solved simply by integration, just by recalling the chain rule. This is the class of first-order equations (only the first derivative of the unknown function y appears) in which the variables separate ; that is, these are equations of the form
(3.25)
h(y)y'=g(x) The left-hand side appears to be the result of we can rewrite (3.25) as
h
we
let H be
[#OyOO)]'
H(yOO) we can as a
=
an
indefinite
integral
of h, H
=
j h,
then
(3.25)
becomes
0OO
integrate :
so we can
y
rule;
000
=
dx
If
of the chain
y(x)
d_ Thus, if
application
fg
=
solve
(3.26) for
(3.26)
y(x),
we
will have the desired
explicit expression
of
function of x.
Examples 23.
yy'
rewritten
=
1.
Let
H(y)
[//(yOO)]'
as
=
=
1,
Jy or
=
y2/2.
H(y)
=
Then the
y2/2
=
x
constant to be determined
by the initial conditions.
solution of
(2(x
24.
y"2y'
y' =
yy'
=
x2y2.
x2
Integrate : -i 1
-y
*3 =
y
=
+
^
1 is y
=
Again,
we
+
write
c))1'2.
equation
+ c,
where
Thus the
can c
is
be a
general
'.4
Some
Techniques for Solving Equations
261
so
-3
y'x^Tc 25.
sin y or
y
y' =
=
y
=
sin
cos x
+
c
cos
sin(c
arc
A
x).
cos
-
After
x.
integrating
this becomes
particular solution is/(x)
=
x
-
n.
26. 1 +
,
After
y +
It is
x
integration
y2 Y
we
have
x2 =
+
x
now a
+
Y
c
bit difficult to write the solution
x, but it is
possible using
explicitly
function of
as a
the formula for roots of
a
quadratic
polynomial :
y
=
-2(4
+ 8x +
4x2
+
c)1/2
(128)
2
The constant
c
is
presumably
determined by the initial conditions, and
with it the function y. Notice however, that each value of c gives two candidates for the solution, but they may not both be solutions. For example, suppose we seek the solution of (3.27) with the initial condition x
=
0,
y
=
y(0) 0,
=
we
0.
We
arrive
at
(3.28)
and
upon
substituting
obtain
Q_-2(4 + c)1'2 2
so we must choose c 0 and the positive sign before the radical. If the initial condition is y(0) -2, again This boils down to y x. c 0, but we must take the negative root, obtaining y= -(x + 2). =
=
=
=
262
3
Ordinary Differential Equations 1 Notice also that upon substituting the initial condition y( 1) this to we find c 8 both roots solutions and problem; (3.28), give =
into
=
that is, both functions y x and y (x + 2) are solutions with this initial value. Thus it is not always true that the initial conditions =
=
uniquely determine the solution of the differential equations. Looking back at the original equation (3.27) we find what might be a clue to this bizarre behavior: the function (1 + x)(l + y)"1 is ill-behaved at y
Uses
=
1.
-
of Exponential
We shall
now
Recall
techniques. equation V has
a
=
study of the exponential function ; because it is simple differential equation it gives rise to several from Chapter 2 (Definition 21) that the differential
turn to the
the solution of such
a
c
unique solution, denoted
(ecx)'
=
cecx, (ecx)"
These remarks suggest A
=1
XO)
cy
a
=
complex
any
...,
(ecx)(s)
method of attack
=
on
csecx
+
ak_iy
++
a,y'
+ aoy
=
(3.30)
another class of
homogeneous constant coefficient equation is
y(k>
(3.29)
Notice that
ecx.
c2ecx,
number
one
equations.
of the form
0
(3.31)
We shall consider this class in greater detail in Section 3.6. Let us compute ecx. By (3.30),
the left-hand side of (3.31) under the substitution y
akecx
+ =
ak-xck~1ecx (ak
We find that ecx is in
+
a
+
a^^'1
+
+ a0 ecx
axcecx
+
=
a0)ecx
+ a,c +
solution of (3.3 1) if c is
a
(3.32)
root of the
polynomial appearing
(3.32). Examples 27. Find solutions for
Substituting
y
=
ecx,
y"
we
y
=
obtain
0.
(c2
-
\)ecx
=
0, thus
We conclude that ex,e~x are solutions. for any a, b, aex + be~x is also a solution. c=
+1.
we
must have
Notice also that
3.4
Some
28. Find solution of
y'"
Techniques for Solving Equations
+ y
=
263
0.
Here substitution of y ecx yields (c3 + l)ecx 0, cube root of 1 Thus we obtain three solutions : =
=
so c must
be
a
.
einlix
e~x
e-'"l3x
Of course, all functions of the form ae~x + be'",3x + ce~iK/3x solutions. 29. Solve the initial value
y'"+v'
=
o
y(0)
Substituting c
0
=
or c
Let
y(0)
=
can
=
obtain
\
(c3
y"(0) +
c)ecx
l
=
=
0,
(3.33) so
we
must have
Thus all functions of the form
if we
can
solve for a, b,
c
by substituting
the
:
0:a + b +
=
c
0
-b-c=\
y"(0)=l:
a
=
=l:ib-ic=l
y'(0)
We
us see
problem
y'(0)
we
i.
=
solutions.
initial conditions
ecx,
=
ce-'x
+
beix
i
y
+
ae0x are
or c
=
0
=
are
solve this system,
b
1
c
-
=
obtaining
-
=
2
2
Thus the function
/(X)
=
i-!V-^Vfa
will solve
our
problem.
30. Solve the initial value
v" + y
=
0
y(0)
=
l
problem
/(0)
=
0
(3-34)
264
3
Ordinary Differential Equations Here
have,
we
initial conditions, a
+ b
1
=
and thus
f(x)
Notice that
we
b
=
i(eix
=
we
ib
ia
a
+
general solution aeix
as
=
=
0 Thus
1/2.
we
obtain
as
solution
e~ix)
already
know from calculus the solution /(x)
shall learn in the next section that the initial value
unique
cos x
\(eix
=
+
=
e~ix)
sinx
e'x
=
=
cis
Now if /is =
ef,
wards
we
x
=
=
a
cos x
+
/
sin
(3.37)
x
Equations
differentiable function,
how to solve
so
is
ef,
(ef)' =f'ef.
and
an
g
z'
=/. =
=
new
function
Since
eHg
we can see are
how
differential
<7(x)
(3.38)
continuous in
are
consider the since H'
Letting
Thus, working back
equation y' =/'y. equation of the form
g(x)y
y'+/00y /
relationships
(3.36)
Namely, exp(J g) is a solution. With a little more ingenuity to explicitly solve any linear first-order equation. These equations of the form
where
a
functions:
obtain the differential
we see
/
exponential
2^(e'je-e-'x)
First-Order Linear
y
and
trigonometric
has
(3.35)
We shall leave to the exercises the verification of these other
between the
We
cos x.
problem (3.34)
interesting equation follows:
Thus this
solution.
Substituting the
ix.
+ be
obtain
z
=
an
interval about
e"y.
Then z'
by (3.38), y' +fy
=
g,
we
=
a.
eHy'
Let +
have this
H(x)
H'eHy
=
=
\* /
and
eH(y' +fy),
equation in
z:
3.4
which is solvable
Some
Techniques for Solving Equations
265
by integration :
J.XeHg
+
c
a
Finally,
y
=
e~Hz
=
y
e~Hz,
=
thus the general solution of
f eHg
e~H
where H is the indefinite
(3.38)
is found:
ce~H
+
(3.39) and
integral of/
is to be found
c
by substituting for
the initial condition.
Examples 31.
y'
Here the
z'
+ xy
we
=
y(0)
x,
take H
=
j" x
0
=
x2/2
=
corresponding equation y' exp(x2/2)
=
+ yx
in
z
and consider
z
=
y
exp(x2/2).
Thus
is
exp(x2/2)(y'
xy)
+
=
exp(x2/2) x
Thus
z
| exp(x2/2) xdx +
=
c
exp(x2/2)
=
+
c
so
y
z
=
exp(-x2/2)
Substituting so c
=
1.
-
32.
y'
Here
x"2y.
=
we
c
exp(-x2/2)
0 the initial condition y 1 is thus The solution y =
=
x,
take H=
Thenz'
=
y(l)
J 2/x
-2x"3y
+
=
In
x
+
c
and
y
=
x2 In
exp(-02/2) exp(-x2/2).
=
c
+
=
c
+
1,
0.
=
=
-
2 In
x-2y'
x
+
x
=
We obtain
z
-
=
2x"Jy
-
1 +
ex2
and consider
x"2(y'-2x-1y)
nx
z =
=
ye
=
^"2^=^ '
266
3
Ordinary Differential Equations
EXERCISES 19. Solve these differential
equations: l (a) x2exp(x2)y' x3,v(0) 0 (b) y' x sin x + cos x, y(0) (0, 1 0) (x(0), y(0), z(0)) =(t,t2,t 3), (c) (xm /(f)) 1 (d) z(t) e" + ((1 + i)t)2, z(0) Solve these differential equations: (a) y'=y2 (b) y' cos x cos y (c) x2 + y2y' 0 (d) y'=(y2-l)(x2-l) (e) y"=xy' (f) y'=(\+x2)y (g) xy2 + (l-x)/ 0 (h) y' e*+" (i) y' sin(x + y) + sin(x y) Solve these differential equations: (a) y' + xy cos x, y(0) 0 1 (b) y' cos x + y sin x tan x, y(0) (c) y' + xy x2,y(0)=0 1 (d) e"y' + e*y e-,x, y(0) l (e) y'=ye-",y(l) Solve these differential equations: (a) y'" 2y" + y' 2y 0, y(0) 0, y'(0) 0, y"(0) 1 y'(0) 0 (b) y" 2/ y 0, y(0) (c) y'" (1 + 3i)y" + (3i- 2)y' + y 0, y(0) 0, y'(0) y"(0)=0 =
=
=
=
=
20.
,
=
=
=
=
=
=
-
=
21.
=
=
=
=
=
=
=
=
22.
-
=
-
=
=
=
-
=
=
=
1
=
,
3.5
=
=
-
=
1
,
Existence Theorems
In this section
we
differential
shall state and prove the basic existence theorem for method is due to Picard and is that of
equations. The successive approximations. (Recall solution to the equation y' cy.) ordinary
how
we
found, in Section 2.10, the
=
The first theorem is about first-order the method of successive
equations. approximations.
We shall first illustrate
Example 33. Successive approximations. of
y>
=ex + y
y(0)
=
0
There is
one
and
only
one
solution
3.5
if/(x)
Now
/(x)=
solves this
equation,
ffV)dt= \\e'
then
by integration we
we
fV
=
see
that
+
Tf
0(0]
dt
f;
that
=
Newton's method sequence
/0
sequence.
T/0
=
T2f0 T3f0 T%
=
=
=
we
is, / is ...,
=
T(T/0)
f(2e<
=
T(T2f0)
=
of T.
point
T"f0
,
...
Let
.
for/0
,
According
to
the limit of the
as
us
say f0
compute =
0.
this
Then
ex-l
=
j\3e<
dt
1)
-
2
-
-
=
t)
2ex
dt
2
-
=
3ex
x
-
-
3
-
2x
-
%-
4ex-4-3x-2X--%x2
T5/0
fixed
a
We may choose any function
fe' dt
that
continuous functions by
on
should be able to find T
TfQ, T(Tf0),
,
see
267
f(ty]dt
+
Thus if T is the transformation defined
Tg(x)
Existence Theorems
x4
x3
5^-5-4x-3y-2--x2
T%
=
nex-n-(n- l)x -(n-2)x""1
xj
--(n-j)
77-
(n-1)!
y!
but if We can't tell yet that this sequence of functions converges, better a picture: replace ex by its Taylor expansion we can get co
T"fo
=
Y-*
n
1
we
y/
ln--Y(n-j)Jj=o j!
j=o -l =
I
X1j
XJ
In
-
(n -y)]
-
+n
-
n-1 =
"^
X XJ
x1
^ (jyj + ;!
(3.40)
268
3
As
Ordinary Differential Equations
lim
the last
oo
->
n
T%
-^
=
y
n->oo
in
sum
=
o
(3.40) tends
to zero, and we obtain
xex
=
1)!
U
given problem! Now we would like unique. This is easy, because it is easy
Indeed xex solves the
to show
that the solution is
to
that T is
a
Tf(x)
Tg(x)
so
-
contraction
fV
=
+
'o
in the interval
|x|
verify
:
<
f(t) -e'- g(t))
\,
dt
=
f(/(f)
-
Jo
g(t)) dt
say
\Tf-T9\<>\\f-9\ Thus if
T/
=
/
and
g ||, which is
||/ Now, the
most
Tg
=
g,
we
obtain
possible only if/= g
general
differential
=
i||/- g\\
>
\\Tf- Tg\\
=
0.
equation of first
order that
we
shall
consider is
v'
=
where F is
F(x, y) a
(3.41)
real-valued function defined in
in the
a
(a, b) plane. A solution is a function neighborhood of a with these properties f(a) If/is
a
=
6
/'(x)
solution, it is
Tg(x)
=
a
=
fixed
neighborhood of the point =/(x) defined for x in a
y
F(x,f(x)) point
of the transformation
fV(f, ^(O) dt + b
(3.42)
"a
The fixed
point will be found by the method of successive approximations : /0 anything, fx T/0 f2 Tflt and in general / T/n_x. In order to that this a limit has and the fixed guarantee sequence point is unique, we must guarantee the hypothesis of the fixed point theorem. More precisely, we must know enough about the function F in order to guarantee that the transformation defined by (3.42) is a contraction on the space of continuous =
=
=
,
=
3.5
functions
shows)
on
269
suitable interval about a. It suffices (as the proof below condition is satisfied.
a
if the
Existence Theorems
following
Definition 4. D in
Let F be a function of two variables x, y in the domain R"+m (x ranges in R" and y in Rm). F is Lipschitz in y if there is a
constant M such that
\F(x, yt)
-
F(x, y2)\
for all yl,y2 such that
Notice that since
<
(x, yt)
(1
+
M\yt
and
y2\
-
(x, y2)
x)(l +y)_1
are
in D.
is not
Lipschitz near y= -1, apply Example 26. of the generality we need for Picard's theorem. Notice that if n m F will be Lipschitz if the partial derivative dF/dy exists and is bounded. by the mean value theorem (along the line x constant)
we
Picard's theorem; and in fact it does not hold as we saw in We have allowed x, y to range through many variables because
cannot
=
=
1,
For
=
dF
F(x, yt)
and thus
we can
Now let
order k is
(x, OCfi
=
dy
-
y2)
yt
<
<
y2
take the M of Definition 4 to be the bound of
higher
order
equations.
dF/dy. equation
A differential
a
of
in the form
F(x,y,y',y",...,yW)
=
where F is
F(x, y2)
turn to
us
given
/k)
R*+1.
-
function defined in
A solution is
a
a
(3.43)
neighborhood
of
(a, b0, =/(x)
A:- times differentiable function y
..
.,
/Jt-i) in
with these
properties
f(a)
=
fm(x)
b0 ,f'(a) =
=
bu... jC-'Xa)
F(x,/(x),/'(x),
.
.
.
=
bk.it
,f(k^\x))
(3.43) with the given initial conditions by successive but the method is not transparent. However, the problem does reduce to the first-order case by means of a great idea. First, we We would like to solve
approximations,
illustrate.
270
3
Ordinary Differential Equations
Example 34.
y"=2y'-y,y(0)
We introduce
Then the
y' z'
y(0)
2z
=
z(0)
y
-
=
0
=
1
Thus, what
which is first order.
of the vector differential
(y, z)'
(z,
=
2z
z
and
require
that
y'
=
z.
seek is the vector-valued solution
we
equation
(y(0), z(0))
y)
-
0,y'(0)=l.
is reduced to the system
given equation
z
=
=
unknown function
a new
(0, 1)
=
by integration and thus solve by successive approximations. Precisely, the solution is the fixed point of the transformation (defined on pairs of functions) : This
we
T(f, g)(x) Let
rewrite
can
ffo(0, 2g(t)
=
compute
us
(/o,0) (fi,9i) (f2 92)
=
=
=
,
(A g3)
=
,
ih,9d
T(f0,g0)
=
T(fu gt)
=
T(f2 92)
\3!
/
1 +
and that
approximations.
(x,2x+l)
=
(x2 + x, f x2 + 2x + 1) (i*3 + x2 + x, f x3 + \x2
*3
+
*2
2
5x4 + *'
x""1
7T7 + 7
\(n-l)!
^T, +
(n-2)!
lim/
This then is the first order.
of the successive
~4\
2 +
=
"
"
+
x
+ X,
3 +
general
+ 2x +
1)
2
%1
+ 2X + X
form of (/
,
\
)
gn) is
\
, -
X*
3
not hard to surmise that the
x" -,
(0, 1)
T{f3,g3)
/x4
now
dt +
(0,l)
=
It is
some
,
=
f(t))
-
'o
.
.
.
/
xex.
typical means of reducing the higher order equation to Equation (3.43), we introduce new unknown functions
Given the
3.5
y0
,
yi,
yk-
,
y'o
=
J^
y\
=
y2
1
and
Existence Theorems
271
replace (3.43) by the first-order system
=F{x,y0,yl, ...y^)
yk-i
.
y0(a)
=
b0, yi(a)
=
bl,..., yk-i(a)
=
bk.Y
Now the
general existence theorem for &th-order equations falls directly first-order equations for systems. The beauty of this trick is that Picard's theorem is no harder for systems and consists merely of verifying that the appropriate transformation defined by an integral on vector-valued functions is a contraction, so the fixed point theorem applies. Here, then, are the fundamental existence and uniqueness theorems for ordinary differential equations. out of the theorem for
Theorem
3.3. (Fundamental Existence and Uniqueness Theorem) Let (a, b) be a point in R x R", and F an Revalued Lipschitz function defined in a neighborhood of (a, b). There is an s > 0 and a unique continuously differenti able R" valued function f defined on (a- e,a + e) such that ((a) b and f'(x) F(x, f(x))for all x in (a s, a + s). =
=
-
Proof. The idea behind the proof is to change the given problem to a problem involving integration. In fact, by the fundamental theorem of calculus, our desired function is that function f such that
f(x)=b+
for all
[C((a
points e,
a
x near a.
+
7T(x)
f F(f, f (f)) rff
e))]" (the
=
b+
That is, space of
we
seek
n-tuples
a
fixed
point of the
function T defined
of continuous functions
on
(a
e,
a
+
on
e))
\\(t,f(t))dt
Because F is Lipschitz, we can choose e so that T is a contraction. We shall of refer to the distance between functions introduced in Chapter 2. First, since F is Lipschitz in a neighborhood of (a, b), there is an M and some
course
rectangle
B centered at
(a, b) such that
|F(x,y),F(x',y')l<M|y-y'|
272
3
Ordinary Differential Equations
(x, y), (x\ y') in that rectangle. In particular, F is bounded on that rectangle byii:= |F(a, b)| + Me0. Let e < fi0tf"\ M~l/2, e0. Let X be the set of n-tuples b || < e0 If of continuous functions f on the interval (a e, a + e) such that || f feX, then for all f e (a e, a + e), (t, f (f)) is in B and F is defined on B, so the for all
.
transformation
7T(x)
=
f
b+
F(t,f(t))dt
"a
is well defined
on
We
X.
verify
now
that it is
a
contraction
on
Let f e X.
X.
Then
H7T(x)-b||<
f
Jfl
|F(f, f (f)) <7f
||rf-b||<e0,sorf6^ralso.
Let f, g
e
X.
\\Tf(x)
7g(x)||
-
<
f
J a
|F(f, f(f))
<m\
-
F(f, g(f)| dt
llf-gll*
<M|x-a| ||f-g||<Me!|f-g||
a
contraction,
so
by
the fixed
point theorem
it has
a
unique fixed point
f.
We have
f(x)
so
=
b+
f
F(f,f(f))rff
by the fundamental theorem of calculus f is continuously differentiable (because b, f '(x) F(x, f (x)) for all x e (a e, a + e). right-hand side is so), and f (a)
the
=
Certain remarks
on
=
this theorem
are
necessary.
First of all, the
differential equation of first order is of the form F(y', y, x) 0, noty' The question arises: when can we rewrite the relation F(y', y, x) =
=
=
general F(y, x). 0 in the
theorem, for in this case we will know that solutions exist. 0 for one of its of explicitly solving an equation H(u, v)
form of Picard's
This
question,
variables,
say
u
=
(so
that there is
a
function
G(v)
such that
H(u, v)
=
0 if and
3.5
only
if
u
=
will be discussed further in
G(v)),
Theorem 2.16 that
we
exists, for each y0 f '(x)
f(x0) The
=
=
answer
from
(Recall
a
e
R",
for all
y0
for
general,
asserts the existence of local solutions.
/
x
R", / any interval in R,
function defined
a
F(x, f(x))
is in
7.
Chapter
273
condition for functions F of two real variables: that this is the general condition.)
have
dF/dy # 0. We shall see Secondly, Picard's theorem only Supposing that F(x, y) is defined in if there
Existence Theorems
on
we can
ask
all of /such that
/
x 6
Siven ^o^ For
no.
the function
example,
F(x, y)
=
y2
is
certainly Lipschitz in any rectangle, so local solutions always exist. But we already know that if y' =y2, y must be of the form (c x)"1 for some constant
c.
Thus, if
we
impose
an
initial condition
/(x0)
=
c0
,
the
(local)
solution is
/W
=
(I
On any interval
Exercise interval
+ x0
x)
-
which the solution exists it is given by this formula (see Thus there is no solution to this initial value problem in any
on
19(a)). containing the point x0
+
l/c0
.
higher order and the reduction to systems of Let us represent a point of R1+<*+1>" by coordinates first order. R". y^ where x is a real number and the yt range through (x, y0 We
now
turn to
equations
of
Revalued
(a0 b0 ,...,bk) e R1+<*+1>" and let F be bk). There is an Lipschitz function defined in a neighborhood of (a0,b0, Revalued function e > 0 and a unique (k + l)-times continuously differentiable that such + defined on(a-e,a e) Theorem 3.4.
Let
an
,
...,
f(a)
=
b0
fii\a0)^bi
l
F(x,f(x),f'(x),...,f(k\x))=fk+1)(x) Proof. Consider (a, b0,..., bk) by
in the R(t+1 '"-valued function G defined
G(x, y0,...,y.)= (yt,
>
*->
pix,
Vo
,
.
,
y
a
neighborhood
of
274
3
Ordinary Differential Equations
Clearly, G is Lipschitz wherever unique function g defined in (a
g(a)
=
By Theorem 3.3, there is
F is. e,
a
+
e) taking
an e
> 0
and
a
values in ,R<*+1>" such that
(b0,...,bk)
(3.44) g'(x)
=
G(x, g(x))
gk)'(x) (gi(x), Writing g =(g0, ...,gk) we have g,(a)=b, and (g0 gk-i(x), F(x, g0,..., gk)). Thus, splitting this into coordinates, gl =g,+i,0<,i
,
=
=
=
.
.
.
.
,
=
.
.
,
=
.
,
Thus
g0(a)
=
gX\a)
b0
=
b,
l
and
g0k+lKx)
=
F(x, g0(x), g6(x),
...,
g0k\x))
problem. The uniqueness follows immediately, for if /is a solu original problem then clearly (/,/', ...,/(") solves (3.44), but the solution of that is unique.
which solves
tion of
our
our
PROBLEMS 23. Let
h0,
...,
/ and suppose f is
(
=
hk-i be infinitely differentiable functions
+
the interval
0
Show that f also must be infinitely differentiable.
y
on
solution of
a
A*-.y< +
kf(h,-i + h',)ya) + h0 y
=
(Hint: Any solution of
0
1=1
solution of the first equation. {/} is a sequence of bounded functions in C(I) such that /n-i II < C, where 2 C, < oo, then the sequence {/} converges to a
is also
a
24. Prove: If
ll/n
continuous function. 25. The differential
ponding
y(0) y(0)
equation y" + y
to the initial conditions
=
l
y'(0)=0
=
0
y'(0)
=
l
=
0 has
unique solutions
corres
3.6
Linear
Differential Equations
275
respectively. Let C, S be these two functions. Prove : l (a) C2 + 52 -S (b) S' C, C' (c) S(2x) 2S(x)C(x) (d) e" C(x) + iS(x) Of course, the reader will recognize that C(x) cos x and Six) sin x and thus these equations should follow. However, the intention here is to verify these equations on the basis only of the defining differential equation. 26. Sometimes it is of value to find a linear differential equation which has as its space of solutions the vector space spanned by n given functions. We find an equation of nth order by substituting the n functions in the 0. +goy equation y(n) + ^-iy<""1) -I For example, suppose we want to find the linear equation whose solution =
=
=
=
=
=
=
=
is
set
the
of
span
y" + gy' + hy
=
and
x
sin
0 and substitute
x
We
x.
and sin
try
a
second-order
equation
x:
g + hx=0
sin
We
x
+g
can
cos x
x
x cos
equations : x
g(x)
x cos x
sin x
0
=
x
=
Thus the differential
(sin
x
solve these linear sin
hix)
+ h sin
x)y"
equation y'x
sin
=
sin
x
-.
sin x
x cos x
is
x
+ y sin
x
=
0
Find the linear differential equation whose solution set is the vector space spanned by the given set of functions.
(a) (b) (c) (d) (e) (f )
3.6
x,
x2, x3
+ nx ex, e", ea
xex, exp(x2) sin x, cos x, tan x sin x, cos x x,
ex, tan
Linear Differential
x
x
Equations
The most important and best understood class of differential equations We and its derivatives. are those which are linear in the unknown function now
give
the definition of this class.
276
3
Ordinary Differential Equations
Definition 5. order k is on
Let / be
an
interval in R.
A linear differential
operator of
transformation from the space of A>times differentiable functions /to the space of continuous functions on /of the form a
L(/)
=
/(k)+I1"i/(0
(3.45)
;=o
where
h0
,
.
.
.
,
hk_i
are
given continuous functions
Notice that the coefficient of the
highest order
on
/.
term is 1
More
.
generally,
it could be any function hk In this case, if hk is never zero on /, we could divide by hk and obtain the form (3.45). If hk sometimes has the value zero, .
then the
theory
to
be
presented
A transformation of the type L (/ + g )
=
L(f)
+
here will fail
(3.45)
L(g)
(see
Problem
is linear, in the
L(cf)
=
31).
sense
that
cL(f)
It follows that the collection of functions which get mapped into zero by L, {/ L(f) 0}, the kernel of L, is a vector space of functions. We shall K(L) =
now
=
show that this is
a
fc-dimensional vector space.
First of all, the equation /(/) g, for defined on the interval / has a solution /
given continuous function g the whole interval, which is determined initial conditions uniquely by given f(a) f>0, ...,/(*"1)(fl) h-iIn other words, in this case, Picard's theorem is more than local ; it gives a solution on the whole interval. We shall verify this fact below (in Pro position 9). Thus, we can state : a
=
on
=
=
Proposition 6. Let I be an interval in R, a el, and L a linear differential operator of order k defined on I. (i) if g is continuous on I and b0, ...,bk_l are any real numbers, there is a unique Ck function f defined on I such that
f(a) (ii)
=
b0 f'(a)
The space
,
=
bu... J^Xa)
K(L) of solution
on
I
=
ofLf=
6k_i
0 is
a vector
space
of dimension k.
Proof. (i) will follow immediately from Proposition 9 below according to the same procedure as in the preceding section for reducing a th-order equation to a firstorder system.
3.6
(ii)
Linear
Let Ea be the transformation from
E*(f)
=
Kit)
Differential Equations
to Rk defined
277
by evaluation
at
a:
(f(a),f'(a),...,f-lXa))
By the existence and uniqueness theorem, E is one-to-one and
Thus
onto.
K(L)
also has dimension k.
Let
us
reconsider
L(/)
=
briefly
case
of constant coefficient linear operators :
/W+Z1i/( =
(
We associate to L the
PL(X)
the
=
(3.46)
0
polynomial
Xk+t^aiXi i=0
(called the characteristic polynomial of L). stitution of f(x) erx, that if PL(r) 0, then
already seen, by sub K(L). Now if PL has k distinct roots rx,...,rk, then all of the functions exp(rxx), exp(yvO Since K(L) has are in K(L), as well as all linear combinations of these. dimension k, these exponential functions form a basis for K(L) and every solution L(f) 0 is of the form =
=
We have
e is in
.
.
.
,
=
At exp(rtx)
+
+
Ak exr>(rkx)
by the initial conditions. In case PL 2X + 1), the X2 does not have k distinct roots (for example, PL(X) in the this discussion We shall situation is more complicated. complete of the discuss next chapter, where we shall also question factoring polynomials. where the
Aj
are
to be determined
=
-
Examples 35. Solve
ym y'(0) =1, y"(0) has the roots
+ =
3y" 1.
conditions:
A+B+C=0
4B + C
=
=
2y'
0,-2,-1.
A + Be~2x + Ce~x.
-2B-C
+
1
l
=
0 with the initial conditions
The characteristic Thus the
polynomial, X3
general
We solve for A, B, C
+
y(0)
3X2
=
0,
+ 2X
solution is of the form
by substituting the initial
278
3
Ordinary Differential Equations
Solving, we f(x) 2 + e~2x =
Linear We
Systems
now
find
A
2, B
=
=
1,
C
3,
=
the
so
solution is
3e~x.
-
with Constant
Coefficients
turn to the solution of
First, let
with constant coefficients.
systems of linear differential equations us try to see an example through to the
end. 36. Consider the system
x[
=
x'2
=
xx + x2
xt(0)
x2
x2(0)
Xj
=
a
=
b
(3.47)
According to the fundamental theorem we can approach by successive approximations using the transformation
T(xi(0, x2(f))
=
f (xj(0 + x2(0, xx(f)
It is convenient to
(3.48)
use
matrix notation.
-l)x
*()
Equation (3.48)
becomes
=
Tx(0=j^{ _jWr)dT Now, =
x1
=
X2
=
we
*o
+ x0
successively approximate
Xn
|L _1jx0dT + x0
=
L
_1j<x0
/0[(l -l)(l _l)0 X0]dT +
dt +
(a, b)
Thus, writing
becomes
x=\l
Xn
x2(f))
-
Jo
+
+ X0
^
a
solution
(3.48)
3.6
Linear
=G -!) yx+(! _!)*<>+ i\3 13
ii
-lj
=
ll x
n2 12
ii +
-1.)
ll
3!
279
x
a +
2!
Differential Equations
n
(l -l)'
+
/*
=
to the fundamental theorem the series converges to the
According solution
k
tk-\
>=['+!. C -H,
(3.49)
c0
The formula
(3.49) represents the solution in the sense that it describes computing approximations to the pair of functions xx(t), x2(0- (The question of measuring the accuracy of those approxima tions is important; we shall return to those questions in Chapter 5.) However, we have not obtained formulas for the functions individually. That is not really surprising since the functions are given by an interdependent relation (3.47). By analogy with the series for ex, we defined the exponential of a a
way of
matrix
as
exp(M)
=
eM
=
/ +
f^
Then
x(f) We
now
we can
=
expfL
state a
(3.50)
k\
fc=i
write the solution to
(3.47)
as
_1jx0
proposition
which summarizes this discussion for
general
linear first-order systems.
Proposition 7. Consider unknown functions :
x'(t)
=
Mx(0
the linear
x(0)
=
x0
first-order system of
n
equations
in
n
3
280 where
x
Ordinary Differential Equations
=
(xu
x(t)
=
7x(f )
.
.
,
x) and
M is
an n x n
matrix.
The solution is given
by
eMtx0 find, by successive approximations, the fixed point of
We
Proof.
.
f
=
Jo
Mx(f )
dt + x0
We obtain
Xo
=
Xi
=
Xo
MfXo +
Xo
(Mt)2 -
=
x2
Xo
+ MfXo +
x0
(Mt)-1
/(Mt)"
A
By the fundamental theorem the sequence of vector-valued functions to the
solution of the
x(0
=
Although
[
we
know there is
{x}
1+
I
x converges
But the limit of x is
(Mt)"'
given by
(3.51)
T-
questioned the convergence of the series (3.50), we problem. For, by the fundamental theorem the sequence so the series in (3.51) must converge. Finally, eM is just
have not
no
converges, 1.
eM'atf
given differential equations.
=
exponential of a matrix is not an easy thing to do; ordinarily just work with the series and approximate solutions. However, cases we can obtain explicit formulas for the solution.
Finding
the
it is best to in certain
Examples (Eigenvectors) 37.
Suppose
(di
\
the matrix M is
diagonal.
Then
3.6 and the
xi
equations
dxxu x'2
=
=
d2x2,...,x'
in
Thus,
(dx
not
The solutions
expOAOXifO),
=
Xi
Differential Equations
281
are
However, this system is
equations.
Linear
.
particular,
.
.
,
xn
/expOA)
'dj
\
dx
system
a
at
all, but just
n
independent
are
=
exp040xn(0)
that
we see
0\
=
0 \
'exp(4,)/
0
38.
Suppose that the vector of initial conditions x0 is an eigenvector Mx0 Xx0 for some X. Then M2x0 X2x0, ...,M"x0 X"x0 so we can compute the solution explicitly, of M:
=
=
=
,
x(f)
eM'x0
=
(/\
=
Ii
+
n
=
oo
^)x0 / n
=
!
n
f)n
_i_ V x0 + 2j r x0 ~i n!
_
x0 +
e
=
-}!
i n
M"x0
tx
x0
This
computation leads us to speculate as to the existence and quantity of eigenvectors of the (n x n)-matrix M. In general, this is a difficult quest and still does not lead to a complete explicit solution of the differential equation x' Mx. However, if there is a basis of R" of eigenvectors of M, then we can give a complete explicit solution. =
Proposition 8. Suppose Vj v are independent eigenvectors of the x n)-matrix M, with Xn, respectively. Then the equation eigenvalues Xlt x' as follows. Write x0 can be solved Mx, x(0) c1v1 + explicitly x0 The solution is c"v
(n
...,
=
=
=
.
x(t) Proof.
=
c1 exp(A1f)v1
+ c"
-I
We compute the series
Mxo
=
M 2x0
=
M'xo
=
IVKc'vi -I
Mfc^v, + c'A/vi -|
(3.51) directly:
+ Cv) +
exp(A f)v
=
c'AtVt -|
c"Av)
h c"Av
=
c'A,2?! +
h
c"Av + c"A 2v
282
3
Ordinary Differential Equations
Thus
Mktk\
/
tk
I
"
\
"("-(,+.?1Tr)^-J?,4+,?^M^) I y=i
tkXk
I
n =
"
\
+ I Vy 4*0 -rfk! \ / fc=i
^
I
=
exp(l, OVy
y=i
Examples 39. Consider the system of differential
equations
*-(:Si)" >-(,J) We find the
eigenvalues
of equations
as
and
eigenvectors corresponding to this system
in Section 1 .7.
Let
=(-l i) Then X
=
det(M
-
1, 2 of this
XT) X2 -3X polynomial. =
+ 2.
The
The eigenvalues are the roots eigenvalue 1 has an eigenspace the
kernel of
m-.=(:26 J) (1,2) spans the kernel. Similarly, the vector (1, 3) eigenvector of M with eigenvalue 2 since it is in the kernel of
The vector
is
an
--G \) The
general
solution of the
given
differential
This vector has the initial conditions Our initial conditions
are
(4, 12),
x(0)
=
equation
cx + c2,
so we can
is
y(0)
=
2ct
+
3c2
.
solve for ci,c2:c1= 0,
3.6
c2
4.
=
x(f)
=
Thus,
4e2t
we
y(0
obtain the
=
Linear
283
Differential Equations
explicit solution
12e2'
40.
*'=(_i/4 !) x()=(S) eigenvalues are 1/2, 3/2, and they have eigenvectors (2, 1), (2, 1), respectively. Thus the general solution is
The
x(0
=
cie"2(_J) c2e3"2(^ +
Substituting
in
initial conditions,
our
we
obtain these
equations
for
c-i, c2:
3
=
3
=
2ct
+
2c2
-Ci + c2
The solutions
given
3
x
are
ct
=
-3/4,
c2
Thus the solution of the
=4/9.
system is
=
"
3,/2/2\ t,i( 2W4 S +
4
l-l)
9
ll)
=
/-3/2e"2 I 3/4e"2
+ +
8/9e3"2\
4/9e3'/3)
41.
*-(_! !)" *-(!)
,3H)
X)2 + 1 0, so X 1 + /'. (1 The root 1=1 + / gives eigenvector (1, -i), and for the root Now our initial con X 1 / we obtain the eigenvector (1, +0thus we obtain the and + ditions are (1,0) (1, -i)/2 (1, +i)/2, The
equation
for X here turns out to be the
=
-
=
solution
-
=
=
284
3
Ordinary Differential Equations
There is
easier way to solve this equation, and that consists in that the matrix is of the form
an
recognizing
r>) and represents
replace
z'O)
(1
=
complex number: in our case system (3.53) by the single equation
our
a
i)z(t)
-
z(0)
by substituting z(f) z(f)
=
y(0
same as
=
Re
z(0
=
Im
z(t)
=
-
=
42. Find the
x'=
x(t)
+
(e{
-.
(3.53), l
"
'>'
+
(e(1-f)'
-
Thus,
we can
1
iy(t).
This has the solution
/I
-3
3\
3
-5
3
\6
-6
4/
thus M has the
eigenvalue
M-/II=
Thus the
of course,
e( 1 + ')
e(1+i)')
general solution
of
x
Now, after computation,
Two
/.
e(1-'"
which is the
x(f)
=
=
1
eigenvalues
we
find
det(M
-
XI)
=
(-2
-
A)2(4
X),
-
2, 4.
2:
(3
-3
3\
3
-3
3
\6
-6
6/
corresponding eigenvectors lie in the plane x y independent vectors in this plane are (1, 2, 1), (0, 1, 1).
+
z
=
0.
3.6
eigenvalue
M-AI
The x
-
Linear
285
4:
=
corresponding eigenvectors 0, which is spanned by ( 1
y
Differential Equations
lie 1
=
,
,
on
2).
the line -x-y + z 0, Thus, the general solution is =
43.
Hi) This matrix is
it has
spanning set of eigenvectors. We (1, 1, 0), (0, 1, 1) have the already Example has the 1, 1, eigenvalue (1, 1) eigenvalue a- The general solution is
symmetric,
so
found them in
a
9:
The initial condition is
(-M-M-i)*(i) Thus the solution is
XjO)
=
2e' + e2'
x2(t)
=
-
4e' + e2t
x3(t)
=
2e' + e2'
44.
Hi !)* x(0)=(?)
&54>
286
3
Ordinary Differential Equations
X)2 0, so we obtain only equation for the eigenvalues is (1 has the X=\. This eigenvector (1,0). Thus we eigenvalue, know one solution of the general equation
The
=
-
one
X(o
=*<(;)
However, this does
satisfy the given conditions. equation without further study of
not
to
solve this
and that is
generally
proceed
first
row
x'(t) and
=
=
we
after
y,
y(0)
=
by observing
This has the solution
1.
the
matrix,
In the present case we can that the second row of (3.54) is
difficult search.
a
avoid such difficulties
just y'
We cannot
y(t)
=
e'.
Then the
is
x(t)
+ e'
x(0)
=
0
know how to solve this
equation : x(0
=
te'.
Thus
our
sought-
pair is (te1, e').
example the solutions are not linear combinations exponentials, but admit polynomial factors. Only when there is a basis of eigenvectors are the solutions linear combinations of exponentials ; when there are too few eigenvectors, we must expect more complicated coeffic Notice that in the last
of
ients.
There is
a
theorem that any solution of a first-order linear system a combination of exponentials with polynomial
with constant coefficients is factors.
This theorem follows from the Jordan canonical form of
shall not go into it here. We conclude this section with the
a
matrix;
we
theorem from which
Proposition
6
proof
was
of the
global version
of Picard's
obtained.
Proposition 9. (Global Version of Picard's Theorem) Let I be an interval Suppose F is a continuous R"-valued function defined on I x R" which satisfies this strong Lipschitz property : there is a constant K>0 such that for allyuy2eR" in R.
sup{| Fix, yi) Then the system
y' has
a
=
-
F(x, y2) | :
x e
1}
<
K\\7l
-
y2
1|
of n equations
F(x, y)
y(c)
unique solution for
=
a
any initial condition
a at c e
I.
(3.55)
Linear
3.6
We cannot
Proof.
simply
use
the fixed
287
Differential Equations
point theorem, for the transformation
T
defined by
7T(x)
is not
a
=
+
a
contraction
| F(f, fit)) on
theless the successive continuous functions
fo(x)
d(x)
=
=
dt
the space of functions continuous works.
on
the interval /.
Define
approximations procedure
a
Never
sequence
{f} of
/ by induction :
on
a
a
j Fit, f0(f ))
+
f(x)=a+
f
dt
F(f,f-i(r))rff
The sequence {f} converges in C(I). besides (3.55) we also have ||F(x, a) ||,
By making K larger we may assume that We prove by induction that
< K.
|x)-.-.(x)|^-^-|x-c|" (i)
=
1
I fi(x)
(ii)
-
fo(x)|
=
f F(f, a)
\
dt < K
dt
=
K \x
-
c\
nl=>n
lx)-t-i(x)|
=
f
J
[F(f, f_i(f))
-
F(f, f-2(r))]
*
c
From
(3.56),
we
obtain
[K(b-a)]" |f-f-ilU<
n\
288
3
Ordinary Differential Equations
Since the series
2 [Kib
a)f/n ! converges, it follows that {f} is a Cauchy sequence C(/) such that f -> f. Since T is continuous on C(7), Tf->Tf. But rf f+i, so f^Tf also, thus Tf f. Since f is a fixed point of T we conclude as in Picard's theorem that f solves our problem. Now the fixed point theorem asserts the uniqueness of our fixed point, and we seem to have lost that. But we can regain it on /, because locally we have Picard's theorem. uniqueness, by Suppose g is another solution of the problem; we f on /. For this purpose we may assume that the point c at have to show that g which the initial condition is given is one of the end points of /. Let
in
C(7)
so
there is
f
an
e
=
=
=
R
=
sup{r
/
f (x)
:
=
g(x) for all
x
<
r}
a g(c), cis in the set on the right. Also, b is an upper bound for this set, We have to show that R=b. If R < b, then the least upper bound R exists. the differential equation is defined in a neighborhood of R. By Picard's theorem,
Since f (c)
=
=
so
there is
equation y' F(x, y) has a unique solution in f But both f and g, when con (R). e) y(R) sidered as functions on (R s, R + e), are such solutions. (Notice g(7?) f (R) e, R + e), so R + e is in the set above, and R by continuity.) Thus, f g on (7? is not an upper bound. Thus the assumption R < b is contradicted, so R b and the proposition is proved. (R
an
s, R +
e
> 0
such that the
with initial condition
=
=
=
=
=
EXERCISES 23. Find the
(a)
y2
(b)
(c)
general solution of these systems of equations 2y2 2y2 + 4yi
y'i =4yi yi
=
=
yi -y2
y2
=
ay i + y2
yi
=yi + y2 + yi
y2
y'i
=
ayi + y2
=
ayi + y3
24. Find the solution of these initial value
problems 1. (a) The system in 23(a) with initial condition yi(0) 1, y2(0) (b) The system in 23(c) with initial conditions Vi(0) =y2(0) =0, =
25.
yi
1 *(0) 1 y2 yi + y2 *2(0) l 3yi (d) y'i y3 yM 1 y'2 yi + 2y2 y3 *(0) -l y'i 2yi 2y3 *(0) Find the general solution of the equation
(c)
=yi
+ y2
=
=
=
=
=
=
=
(b)
the matrix in Example 10. the matrices in Exercise 10.
(c)
the matrices in Exercise 1 1
(a)
=
=
by:
=
x'
=
Mx, where M is given
3.7
(d)
/
o
-1
2
3
-2
Second-Order Linear Equations
-3\
(g)
/
2
0^
0
/-2300
3
I
11/
(e)
/
7\
(h)
(J I)
m
4
3
289
0
0
T
2
PROBLEMS 27.
Suppose M
=
(a/)
that the solutions of x'
(Hint:
M"
is
an n x n
Mx
=
are
matrix such that all
polynomials
a/ =0ifi< j. of degree at
Show most
n
0.)
=
28. Show that
exp(M')
=
(eM)'.
29. Show that if M is skew-symmetric (M' -M), then eM(eM)' =1. For such a matrix the rows form an orthonormal basis : A matrix A with the =
property AA'
3.7
=
I is thus called
Second-Order Linear
The most
y" Thus the In this
+
cr1(x)y'
we
rotation.
Equations
equation : +
techniques
section,
a
type of equation arising from physical problems is the
comon
second-order linear
orthogonal, and represents
a0(x)y for
shall
=
(3.57)
g(x)
solving such equations
assume
that
we
know
have been well
one
developed.
solution of the associated
homogeneous equation y"
+
ai(x)y'
+
a0(x)y
=
0
(3.58)
and show how to find the general solution of (3.57). The question of finding this first solution is of course difficult, and further discussion will be postponed until Chapter 5. The technique involved in finding the general solution consists in substituting candidates involving the given solution and a new to reduce the complication in the unknown and
function,
given equation.
thereby attempting
290
3
Ordinary Differential Equations
theory of the first-order equation: y' h(x)y g(x). homogeneous equation is easily solved of variables : by separation f(x) exp(Jx h) is a solution of y' hy 0. to find the solution of the given equation we substitute y Now, z/ general where z is some new unknown function. From/' + hf= 0, we obtain In order to motivate this +
discussion, let
us
recall the
The
=
=
=
=
y'
=
9
hy
+
=
z'f+ z(f'
+
hf)
=
z'f
by integration: z J /" V + cNow, the second-order homogeneous equation (3.58) has two independent Let us try to find another solutions. By assumption we know one, call itfx by substituting y zfx. The new equation in z is
Thus
z
=f~1g,
so z
is found
=
.
=
y"
aiy'
+
+ a0y
=
=
This
find
equation
z(x)
c
=
+
2z'f\
z"/1
+
(2/'1+a1/1)z'
is linear in z' and thus
We have
z.
zf'[
z"fx
+
we can
+
ax(z'fx
+
zf\)
+ a0
zfx
0
=
(3.59)
solve for z' and then
integrate
to
result
as a
f f(t)-2
exp
(- Q
dt
Examples 45. The
We
y'
now
=
equation x2y"
z'x + z,
x2y"
xy' y 0 has the by substituting y(x)
+
=
find another solution
y"
+
xy'
+
(3z')x2
-y
z"x + 2z',
=
=
z"x3
+
solution =
z(x)x.
y(x)
=
x.
We have
so
2z'x2
+
z'x2
+
xz
-
zx
=
0
or
z"x3
=
0
Dividing by x2 we have z"x + 3z' 0, which we separation of variables: z Cxx~2 + C2. We =
=
and thus the second
46.
xy"
-
sin x2 is
y'
+
4x3y
a
solution,
solution of
=
0
y
=
zx
=
x"1,
can can
is found.
solve for z' take
z
=
by
x~2,
Second-Order Linear Equations
3.7
We substitute y
z"x sin x2 +
=
z'(4x2
sin x2 and obtain this differential
z
x2
cos
sin
-
x2)
equation
291
for
z
0
=
Thus
_=
-4xcot x2 +-
z'
x
Integrating, In z'
obtain
csc(x2)
2 In
=
we
x
+ C
again,
we
+ In
or
z'
=
Cxx esc2 x2
Integrating
once
second solution
might Now that
have
we
can
guessed
have
a
find
be chosen the
at
z
the homogeneous equation, we our cue from the first-order case,
as/i,/2 y(x) If
we
Now, =
we
Let
consider
zMhix)
+
a
+
=
C2
cos
Thus, the x2 (which we .
independent solutions for general equation (3.57). Taking
technique for finding
homogeneous equation.
Cx cot x2 x2 sin x2
beginning).
return to the
the
=
cot
as
a
we
try
us
refer
two
combination of the solutions of to
these
two
solutions of (3.58)
function of the form
(3-6)
z2(x)/2(x)
into (3.57) compute y' and y" and substitute
we
will get
a
totally
un
two unknown functions z z2 intelligible equation of second order in the of course, a pair of equations. What we need, to find two unknown functions, is We notice, first of all, that From where is the second equation to come? functions zt, z2 even 1 the the formula (3.60) does not uniquely determine so that (3.60) if For, z z2 are found we know the sought after function y. subtract and gf from z2 to r gives the solution y, then we may add gf2 condition another seek thus We another making (3.60) valid. .
,
pair obtaining (preferably involving derivatives) which functions zx, z2 Differentiating (3.60),
will we
serve to
uniquely identify
the
obtain
.
y'ix)
=
z,(x)/;(x)
+
z2(x)/2(x)
+
z\ix)hix)
+
z2(x)f2(x)
(3.61)
292
3
Ordinary Differential Equations
Equations (3.60) and (3.61) will give a pair of linear functions in zx(x) and z2(x) if the sum z'x(x]fx(x) + z2(x)/2(x) vanishes. This pair of equations (if noncollinear) will then identify zx(x), z2(x) in terms of y(x), y'(x). Thus, if that condition is satisfied we know that zx, z2 are uniquely determined by the solution y. Turning the argument around, we impose the condition
*'i/i
+
(3.62)
zi/2=0
hope now that, together with this condition, the given differential equa explicitly determine zx,z2. (In fact it will do so theoretically, since Equation (3.57) determines the solution y which in turn determines zx, z2 in the presence of the condition (3.62).) Let us try our idea on Example 45. and
tion will
Example
x2y" + xy' y x2. We have the two solutions x,
47. Solve
=
We consider y
z'1x
+
Now let
=
y''
=
and
x"1 of the homogeneous equation.
impose
the condition
z2x"1=0
(3.63)
substitute this information into the
us
the presence of
y'
1
zxx + z2 x"
=
(3.63),
-z2x"2 z'x z'2x~2
we
given equation.
In
have
zx
2z2x'3
+
Then
x2
=
=
x2z\
x2y" x2z\ -
z2
Now the
+
xy'
y
z2 + =
2z2x_1
z2x_1
+ xzx
ZjX
z2x_1
x2
pair
(3.64) of linear
Equations (3.63), (3.64)
can
be solved
by
Cramer's rule:
X3
1
-X =
-2x
z2
2
Integrating, we find general solution is y
=
zl/l
+
^2/2
=
=
-2x
that zx
\x
-1
=
-
0-x3
=
,
x2
2
x/2
+ cx, z2
+ cxx + c2
X
=
x2/6
+ c2, and
so
the
Second-Order Linear Equations
3.7
Now, it
not an accident that in this case the
was
equations
293
turned out to be
of linear first-order equations: this is always the case. We shall now describe the technique in general. Supposing \!na\. fx,f2 are two independent
a
pair
solutions of the homogeneous Equation (3.58) we try a function y We impose the condition + z2/2 as solution of (3.57).
zi/i
z'2/2
+
=
=
zxfx
(3.65)
0
Then
y'
y" Thus
=
=
Zlf'l
+
Z2/2
z'J'x
+
z'2f'2
+
zxn
z2f2
+
becomes
(3.57)
z'i/i
Z2/2
+
+ z2
a
+
1/2
Zl/l
+
+ ziao
z2/2
/1
+
ZlOl/i
+ z2 o
fi
=
9
or
z'i/i
+
(3-66)
z'2fi=g
vanishing since fx,f2 solve (3.57). linear Equations (3.65), (3.66) by Cramer's rule.
the rest of the terms
-fiQ
, _
Zl~/i/2-/2/i and these
can
=
/i/'2-/2/i
in order to find the general solution. One functions is the denominator. If it ever vanishes, these down. break will discussion whole In fact, our
may not be
integrable.
Fortunately,
we can
verify
once
and for all that this function is
The function
=
fx(x)f'2(x)
-
f'x(x)f2(x)
is called the Wronskian of the
W
=
hf\
of
integrated
be
apparent problem
W(x)
pair
/iff
,
'
We solve the
-
fih
pair/j f2 ,
=
det(g] /?$) Notice that
nonzero.
294
3
Ordinary Differential Equations
Since fx,f2 solve W +
(3.57),
easily check
we can
that
axW=0
and thus
W(x)
W(x0)
=
Thus if W is
exp(- fax(t) dt)
nonzero at one
point,
it is
never zero.
W is
the vectors
nonzero at
(fx(x0), f{(x0)) and (/2(x0), /2'(x0)) are independent; guaranteed if the functions^ and/2 are independent.
x0 if
this is
Examples 48. Solve to
y"
that
see
+
is
x
Xy'-y
find another solution
is z"x +
(2
Z
+
x2)z'
=
=
x,
y(0)
solution of the
a
by substituting
0.
0,
y'(0) 0. It is homogeneous equation.
=
y
=
=
zx.
easy We
The equation for
z
Thus
X
so
z'
Cx"2exp(-^
=
Thus
we
may take
as
the second solution *2\
y(x)
=
xz(x)
Now let
us
=
zi
+ z2
+
-
refer to the
by substituting z'xx
j t~2 exp( -) dt
x
X(j)(x)
z'2(x(b'(x)
y
=
+
=
integral by cb(x). We solve the given equation z2x
zxx +
0
=
1
3. 7
Since
>'(x)
=
x"2
Second-Order Linear Equations
exp(-x2/2),
we
295
find, by Cramer's rule,
-x(p(x)
exp(-x2/2)
exp(-x2/2)
or
zi{x)=f0-texp{-2-)[sy2exp{-2~)d\ z2(x)
The
exp(yj
=
integrals defining
nevertheless define
y(x)
-x
=
+
This
It
we are
zx
are
not
function.
expressible
in closed form, but
dx
dt
xexp(y)J"0r2exp(~T)dT solving second-order equations is called variation of applied to higher order linear equations. Suppose differential equation :
for
can
be
such
y(n)
I atix)yll>
a
+ y
=
(3.67)
g(x)
Suppose we have somehow found n independent solutions fx, ...,/ homogeneous equation. Then we try a solution y
=
zi/i
+ z
+
.
.
,
z'i/i
zi/i
+
---
+
z;/;
+
---
+
z;/;'
zi/r2)
+
--
+
of the
fn
As in the second-order case, the solution will zx, z if we impose the conditions .
they
Thus the solution is
fj exp(- ^) [jV> exp(- l)
given +
a
.
technique
parameters.
dt
=
o
=
o
z;/r2)=o
uniquely determine the
functions
296
3
Ordinary Differential Equations
In the presence of these
zi/S,_1) We
-"
+
+
conditions, (3.67) becomes
z;/i"-1)
solve this system
can
as a
system of linear equations and then find the as in the second-order case, this system is
solvable since the determinant is
fx, ...,/)
The y
(called
the Wronskian of the
n
functions
never zero.
\x3y"'
49. Solve
=
0
Just
by integration.
zx, ...,z
=
x2y"
-
2xy'
+
2y
-
=
x5(x2
-
9).
(3.68)
homogeneous equation has the solutions x, x2, x3. z2x2 + z3x3. We impose these conditions:
Thus
we
try
zxx +
z\x z\
+
+
z'2x2
2z2 x
z'3x3
=
0
3z'3 x2
=
0
+
+
In the presence of these conditions z2 +
6z3
x5(x2
=
we
compute (3.68)
to
be
9)
-
The matrix of this system is
tx
X2
x3
1
2x
3x2
1
6
\0
\
/
which has the determinant -2x3 + 6x2
rule,
we must
"
,
""
'
=
-2x2(x
-
3).
have
x5(x2-9)-x4 -2x2(x-3)
,
Z2
=
'
-x5(x2~9)-2x3 -2x2(x 3) -
x5(x2 9) x2 -2x2(x-3) -
,
23
After
_ ~
integration
we can
x9
y(x)
=
zr
3
rx7
express the
general solution
as
3x8 +
^r
x6
+X1
+ c,
I
7+2"H
+
+c2
Thus, by Cramer's
3.7
Second-Order Linear Equations
297
EXERCISES 26. Show that the general solution of
y" + y=f can
be expressed
y(t)
=
ci cos f
+
as
c2
sin f +
Jo
sin(f
t)/(t)
-
dr
27. Find the general solution of
y"
+
y'
=
x
28. Find the general solution of: (a) y" 4y 1 =
-
(b) (c)
y'"-y' x2 y" + 3y' + 2y
(d)
+ y"-^Lry' 1
(e)
x2y"
=
=
sin
x
^y
x
4xy' + 6y
=
=
0
1
x
x3 + x2
29. Find the solution of
x2y"
-
2y
=
2x2
30. Find the
y(0)
=
1
y'(0)
=
1
general solution of
y" + xe'y' -exy=0 31. Find the solution of
l
y(0)=0
y'(0)
30. A differential
equation of
the form
e"*y" + xy'-y
=
1
=
PROBLEMS
akx"yw
+
ak-i
where the a,'s y
=
x\
xk-1y('"1) ^ are
constants
1-
<*ixy' +
can
be
a0 y
=
0
easily solved.
Try the substitution
You should obtain
xs[akis)is -l)---(s-k) + ak.iis)is
-
1)
(s
-
k+
1) + h
ais
+ tfo]
=
0
298
3
Ordinary Differential Equations
Thus
we
need only find the k roots of the polynomial in brackets. general solution of these differential equations:
Find the
(a) (b) (c) id)
x2y" 2xy' + y x2y" 3xy' 3y x2y" + 4xy' + 3y x2y"-xy' + y
0
=
-
-
=
=
0
=
xs
0
31. Solve the second-order 2x2 system of
equations
*-+(-! jm; 0'iHint: Go
3.8
to the first-order 4x4
system by adding the equations y' =z.)
Summary
An Revalued function defined in
neighborhood
a
of x0 in R is called
differentiable at x0 if
lim/(x0 + t0 -/(x0) f->0
exists.
This limit is denoted
/ in R its
image
is
a curve
f'(x0).
If
/ is differentiable on an interval through /(x0) spanned by /'(x0) If h is differentiable in a neighbor
The line
in R".
is the tangent line to the curve at /(x0). hood of the curve and has a relative maximum
g
are
We
can
deduce the
differentiable functions defined in
(or minimum) x
0.
=
of h
subject
for which there exists
g(x)
=
0
a
Vn(x)
on
the
curve
following principle
at
f(x0),
then
from this.
If
domain in R", then the maximum to the restraint g(x) 0 is attained at those points a
=
X such that
=
XVg(x)
If h, gx, ...,gk are differentiable in R", and h has a maximum (or minimum) Xk subject to the restraints gx(x) 0, gk(x) 0 at x0, there exists Xx =
=
.
.
.
.
,
..,
such that
<7,(x0)
=
0,
.
.
.
,
gk(x0)
=
0
V(x0)
=
X.Vg^Xo)
Suppose / is an Revalued function defined on (fc-times continuously differentiable) if /', ...,/(t)
+
+
XkWgk(x0)
the interval /. all exist and
are
/
is Ck
contin-
3.8
If/ is
uous.
fix) where
such
a
/(x0)
=
e(x x0) is If/has derivatives
function
+
we
'"'/"W,.. 1 '! (x ~
i^-
-
'
limM*(x~Xo)* k\
,., + six x0); "'
by Mk orders, and
of all
299
have Taylor's expansion about any x0
bounded
-
Summary
=
"
-
x
max{|/(*'(/) I
e
/:
' ,(x-x0)
xj ""'
:
/
(fc+1)! between x0 and x}.
0
=
k->OD
then/can
be
expanded in
/(x0)+I
=
n=i
A differential x, y, y ',...,
/("(x),
equation
^-^(x-xo)" n!
equation of order k is
y(M. is
a relation involving a function of ^-times differentiable function / such that this after the substitution y =/(x), y' =/'(x), y(,
If there is
relation holds for all a
x
a
.
solution of the differential
say / of order k is
we
infinite Taylor expansion:
f(')(r )
to
/(x)
an
a
.
=
.
,
A linear differential
equation.
relation of the form
k- 1
y(t> +'I X ai(x)y(0 ;=i
+
a0{x)y
=
g(x)
(3.69)
where the functions a-, and # are (at least) continuous on an interval /. If jsO, the equation is called homogeneous. The space of solutions of the homogeneous equation is a vector space of dimension k. Equation (3.69) has a solution on / uniquely determined by the initial conditions
/(x0)
=
has
a
=
Fix,
unique
condition
on
=
*i,...,/('"n(xo)=flt-i
(3-7)
of the form
Any equation
y(k)
/'(*o)
o
y,
ylk~u)
y'
solution
subject
to
the initial conditions (3.70) under this
F:
(i) Fis defined and continuous in a neighborhood (ii) F is Lipschitz: there is an M such that
of
(x0,
a0,
|F(x,y1,y'1,...,yri>)-F(x,y2,y'2,...,yri,)l <
Mi\)\
-
y2\
+
\y\
-
y'i\
+
+
\y\k~u- ^""D
.
.
.
,
ak-x).
300
3
Ordinary Differential Equations
Techniques for Solution 1
.
Successive
y'
=
/0
{/} =
y(x0)
F(x, y),
is solvable if F is sequence
Lipschitz
defined
as
=
a0
near
x0
fi(x)=
equation
The solution
.
can
be
approximated by
a
follows :
any continuous
/i(x)=
The
approximations.
function,
fF(t,f0(t))dt
+ a0
fF(t,fx(t))dt
+ a0
Jx0
Jx0
fn(x)=CF(t,fn-x(t))dt + a0 2.
If
Separation of variables.
\g-1(y)dy \f(x)dx =
implicitly
determines y
as a
then the
y' =f(x)g(y),
equation
+ C
function of The
3. First-order linear equations.
x.
homogeneous equation
y'+/y=0 by separation of variables: y cexp( |/). The equation y' +fy g can be reduced by the substitution y z exp( J/). The result ing equation in z is solved by separation of variables. 4. Constant coefficient linear equations. The characteristic polynomial can
be solved
=
=
=
of the differential
equation
y(t) + ^_1y(fc-1)+---
+
a1y'
+ a0y
=
+ axX polynomial Xk + ak_x X*"1 + polynomial, then erx is a solution of (3.71).
is the
in
5. First-order linear systems. Let A be n unknown functions y y): (yx, =
.
.
.
,
(3.71)
0
+ a0
an n x n
.
If
r
is
matrix.
a
root of this
The
equation
3.8
y'
Ay
=
y(0)
has the solution
eM
=
y(f)
exp(M)
=
n=i
If y0 is
The exponential of
a
matrix is defined
by
?L
/ +
=
301
y0
eA'y0.
=
Summary
n!
eigenvector with eigenvalue X, then the solution is y(f) basis yx, y of eigenvectors of M, with eigenvalues Al5 respectively, then the general solution is an
cx
In
a
a
.
exp(Xxt)yx
general,
we
must
.
.
,
+
+ c
allow
y" find
a
equation y"
+
ax(x)y'
=
+
Suppose fx,f2
ax(x)y'
by the substitution
zi/i
+
a0(x)y
+
y
=
zxfx
=
+
knowing
one
solution.
.
,
Suppose fx
0
second, by substituting in z'.
.
is
(3.72) =
y
are
zfx.
This produces a linear first-order Then we solve
solutions of (3.72).
g(x) z2f2
(3.73) .
z2/2=0
In the presence of the condition
(3.74)
Equation (3.73) becomes
zi/i
+
4/2
=
(3.75)
0
The linear Equations (3.74), found by integration.
(3.75) can
be solved for zx, z2 and then zx, z2
are
FURTHER READING E. A.
.
Xn
polynomial coefficients.
a0(x)y
+
.
exp(Xn t)y
6. Second-order linear equations, solution of
we
ex'y0
=
If R" has
Coddington, An Introduction to Ordinary Differential Equations, Prentice-Hall, Englewood Cliffs, N.J., 1961. An elementary book on differential equations which goes more deeply into the material of this chapter. M. Tennenbaum and H. Pollard, Ordinary Differential Equations, Harper
302
3
Ordinary Differential Equations
and Row, New York, 1963. This is a thorough treatment of the subject of differential equations. Many special techniques and applications are
exposed. F. Brauer and J. A.
Nohel, Qualitative theory of Ordinary Differential York, 1967. This book studies the theory of
New
Equations, Benjamin,
systems of differential equations, and in particular the behavior of
sets of
solutions. L. Loomis and S.
Sternberg, Advanced Calculus, Addison-Wesley, Reading, a very modern approach to the subject. It goes
This is
Mass., 1968.
into the fundamental theorem.
thoroughly
MISCELLANEOUS PROBLEMS 32. Show that if M is
<Mx, x>
=
skew-symmetric matrix (M' M), then symmetric and thus has a basis of Conclude that, considered as a matrix over C, M also has a
0 for all
eigenvectors.
a
=
Show that M2 is
x.
iHint : M2 A (M + V A)(M eigenvector of M2 with eigenvalue A, then either x is
basis of eigenvectors. an
with
eigenvalue V A,
V
=
or
(M
V
A)x
is
an
an
eigenvector of
A).)
Thus if x is
eigenvector of M
M with
eigenvalue
-Vx. 33. Let / be any linear transformation. Compute the gradient of 1 is attained at
an
eigenvalue
of T'T.
34. Show that if T is whose
major
axes are
35. Find the
of
points p0
a symmetric matrix, 7X{||x||2 1}) is an ellipsoid length equal to the eigenvalues of T. e {(x, y) e R2 : xy 1}, pi e {(x, y) e R2 : y + x2 0} =
=
=
which minimize the distance between these two
curves.
36. Minimize and maximize the volume of
a box with given surface area. point on the ellipse [x2 + iy2 1} which is closest to (|, 0). Find the furthest point from (, 0). 38. Find the point on the ellipse {x2 + iy2 1} which is closest to the circle of radius centered at (I, i). 39. Suppose {} is a bounded sequence. Define fix) 2= i a x". Show that /is infinitely differentiable in the interval (1, 1), and nla
37. Find the
=
=
=
=
fM(0). 40. Let / be a twice continuously differentiable function defined in a neighborhood N of (0, 0) in R2. Show that there is a function e defined in N such that lim e(p) 0 and =
P-.0
fix, y) =/(0) 41.
+
8 (0, 0)x 8 (0, 0)y +
+ e(x,
y) \\(x, y)\\
Using Taylor's theorem, we can derive the exponential function in Suppose that / is a function with the property that
yet another way.
3.8
fix) =/(x) for all Taylor expansion *
f(x)=
Then
x.
1
I ~X"+t(x) =on!
/<"(*) =/(*)
for all x,
Summary so
/
must
303 have the
xk+1
(A:+l)!
for all k.
Because of the estimate on ek it remains bounded as k -> oo, so should expect /to be the limit of the polynomials Pt(x) 2'=o (l/!)x". We already know, from the theory of Chapter 2 that the lim Pkix) exists ,
we
=
for all
Noticing
x.
that Pk
=
/,_
prove
that/(x)
=
lim k~"
have the
42. With
a
little bit of
patience,
you should be able to find
/'(0)
=
Redoes indeed
property/' =/.
0 and /<"(x) + fix)
a =
and in the
same
function / defined 0 for all x.
way
on
as
in Exercise 41,
R such that
/(0)
=
l'
43. (a) Suppose that /is C on [-R,R] and /(0) =/'(0) /('_I)(0) =0. Then there is a continuous function g such that /(f) fgit), and<7(0) (l//c!)/<(0). (b) Suppose that/is C on [- R R]. Show that there is acontinuous =
=
=
=
function g such that
+ tW) f(t)=2 r-^t' l\ 1
=
0
44. Change the conditions in Problem 18 as follows: The ratio of horse population to total population is constant and only the eggs hatched in horses produce mature insects. Derive the differential equations now governing the population growth. 45. Suppose now we have an insect which has a natural death rate of d, The per insect per year and which lays h eggs per insect per year in the air. egg hatches if it lands on a horse and the hatching causes the horse's death. Assuming birth and death rates bH dH for the horse and a probability k that a given egg will land on a given horse, now find the differential equations of population. Let a z represents a force field on the plane. 46. Suppose f(z) in case the is i, the motion Describe be at 1 at time 0. velocity particle (1 + 0/2, (1 i)/2. 47. We assume that a particle generates a force field directed toward the particle and of strength equal to the inverse of the square of the distance 0 there are particles at rest at points pi, to the particle. At time t p* in R3. Let f,(f ) be the position at time t of the particle originally at p, What is the differential equation the function (fi, f) must satisfy? 48. Suppose a river deposits water in a lake at the rate of v gal/day. We Suppose may assume that v is a periodic function of time with period 365. Finally, two pumps pump water out at the constant rates of wt, w2 gal/day. ,
=
-
=
.
.
.
,
.
.
.
.
,
304
3
Ordinary Differential Equations
evaporates out of the lake at a rate of kit ) gal/day/ft2, where k is also periodic with period 365. We may assume that the area of the lake is proportional to W213, where Wit ) is the volume of the water in the lake at water
Write the differential equation W must satisfy. Suppose a missile A is moving in a straight line with constant velocity A tracking missile B of constant speed s0 is always pointed toward the v0 missile A Find the differential equation of motion of the tracking missile B. 50. Suppose we have the same situation as in Problem 49, but this time the speed of B is proportional to the distance between A and B. Find the equation of motion of B. 51. A falling body actually experiences a drag due to air resistance which is proportional to its velocity. Suppose a body of 100 tons is dropped from a plane 5 miles high ; and this constant of proportionality (which depends of course on the shape of the body) is 20. How long will it take for the body to reach the ground ?
day
f
.
49.
.
.
52. Two chemicals A, B in solution combine to create chemical C accord C. ing to the equation 2A + B Suppose the rate of the formation of C is proportional to the product of the amounts of A and B present and inversely proportional to the amount of C present. Find the differential equation governing the formation of C, assuming initial amounts A0 B0 of chemicals -
,
A,B. 53.
10, B0 Suppose in the above problem, A0 5, and the proportion How long will it take for the reaction to complete? =
=
constant is 1.
54. If two bodies A B of different temperatures come in contact with each rate of change of temperature is proportional to the difference in ,
other, the
temperature (the proportion constant depends on the bodies). TA TB are the temperatures of A, B, respectively, we have
Thus if
,
T'A
T'B
=
=
kAiTA kiTB
TB) -
TA)
Find the formula for TA
kA 4, kB kA =2,kB 55. In Problem 54,
(a) (b)
What is it in
=
=
=
,
TB with these data :
5, 7^(0) i, TAi0)
as f
->
oo
=
=
100, TB(0) 120, r(0)
(a),
case
-
=
=
=
=
=
=
=
.
-
-
-
0.
=
50.
the bodies tend to
(b), in general ? 56. Solve these differential equations : (a) y<4) 3y" + 2y 0. 2e". (b) y" + 3/ + 2y cos x. (c) y' sin y + cos x cos y (d) (x2 + l)y'-2xy x2 + l. (e) xy' + 3y x~2 sin x. x' + ax b sin f. (f ) (g) y" xey (h) y<4> y<3> y(2) y' 2y 0. case
=
-
=
a common
temperature.
3.8
(i) (j)
(k)
Summary
305
ay" + by' + cy 0. 1 + y\ y'(l + x2) x' + y' 2x. x'-y' 3y. =
=
=
=
M
,-(_? {),.
57. Solve these initial value problems: (a) y" 3/ + 2y e3*, y(0) 0, y'(0) -
(b) (c)
(*> (e) (f ) (g) (h)
=
xy' + 3y
y<4>
=
5. x3, y(0) 3y<2> + 2y 0, y(0)
-
=
=
1
.
=
=
=
1
,
y'(0)
=
0, y"(0)
0,
y '"(0)
=
0
r-(] l)y,y(o) (\). r-(_93 83)y>y(o) Q. =
=
e*yf + Xy'-y gx; ^Q) Q 0> y(()) x2y" + 3xy' + y 0, y(0) 1 y'(0) 1 x2y" + 4xy' + 2y x1 y(0) 1, y'(0) =
=
=
=
=
=
,
=
=
.
=
,
58. Show that if all the entries of the matrix M series
n
=
0.
are
less than 1, then the
2M" =
0
Show that the limit is (I converges. M)_1. 59. Use the idea of the preceding problem to -
two
decimals, (a) A
approximate A"1
where
/ =
1
0
(0.07
0.91
\0.14
()
/ A=i
0.08\ 0.11
-0.03
1.13/
0.98
0.01
-0.12
-0.13
1.18
0
-0.03 N -0.1
0.02
-0.02
1.01
0
0.11
-0.11
0.13
1
to within
4
Chapter
CURVES
Force Fields
According
to Newton's laws of
is
accelerate
third law
F
velocity unless it according to Newton's
motion, a particle will subjected to forces.
line at constant
=
move
in
In that
a
straight
case
it will
(4.1)
ma
In this chapter we shall study the motion m is the mass of the particle. particles subjected to variable forces. That is, we must allow the possi bility that the force applied to the particle depends upon its position (as in gravitation) or even upon time (in the case of a variable electromagnet). This gives rise to the notion of a field of force. A field of force will be given in this way: at time t and position x a particle of unit mass will experience Thus for each t0 we have associated a vector F(x, f0) to a force F(x, t). each point x. We can illustrate this as in Figure 4.1. Now, we have seen that a particle of unit mass situated at x0 at time t 0, with a velocity v0 at t 0 will follow the path of motion determined by the given field of force as the solution of the differential equation
where
of
=
=
f"(0 f '(0)
f(0)
=
F(f((), 0
=
v0
=
x0 306
Force Fields
Figure
path of motion equation.
The
this
is
a curve
307
4.1
in space
given by the function
f which solves
Examples Suppose a particle moves around the unit circle in according to the function 1.
f(0
=
(cos
t, sin
f'0)
f"(0
we
=
=
plane
(4.2)
f)
What force field would account for this motion? twice
the
Differentiating
find that
cost)
(-sin
t,
(-cos
t, -sin
0
=
-
f (0
particle is accelerating magnitude (see Figure 4.2). This Thus the
toward the motion
can
origin
with constant
be accounted for
by
the force field
F(z,t)= In t a
-z
fact, in the presence of this field, if
=
0
orthogonal
to its
circle centered at the
ential
f"(t) f(0)
equation
=
=
-fit)
zo,f'(0)
=
izo
position origin.
a
particle
has
a
velocity
at
time
vector, then it will continue to move in We can see this by solving the differ
308
4
Curves
Figure 4.2
The solution of this
f(z)
=
z0 e"
which is 2. z
=
equation
z0(cos
just (4.2)
t, sin
with z0
is
f)
=
1.
Suppose we are given in space a force field directed toward the magnitude the distance from the z axis (Figure 4.3).
axis with
F(x.y,z) =-(x,y,0)
(Jf.y.z)
Figure 4.3
Fluid Flows Here
again the force field is independent
F(x,y, If
z,
of time and is
309
given by
f)= ~(x,y,0)
particle is at (1, 0, 0) with an initial velocity of (0, 1, a), what is path of motion? We must solve this differential equation for three unknown functions f(f) (x(t), y(t), z(t)) a
its
=
f"(0
(x"(0, y"(t), z"(t))
=
x(0)
1, y(0)
=
x'(0)
=
0, y'(0)
=
The solution is
f(0
=
(cos
Thus, if a is
positive,
0, z(0)
=
=
1, z'(0)
easily
f, sin t,
=
=
(x(f), X0, 0)
0 =
a
found to be
at)
0, the path of motion is
the
circle, of slope
path
a, and if
a <
circle in the
plane z 0. If a upward spiral lying over the unit the 0, path followed is a downward spiral
of motion is
a
=
an
(Figure 4.4). If
given a time-independent R2, R3, graph sufficiently many values of the field, it seems to be a broken line picture of a family of curves. In fact, there is a family of curves which fits the picture in this sense: there is a curve through each point x which is tangent to the vector F(x) at that point. These curves are called the lines of force of the field and are found by solving the differential equation 3.
Time-independent fields.
force field
f '(f)
f(0)
=
=
on a
domain in
or
we
are
and
we
F(f(0) x0
The solution of this differential
passing through
the
point
x0
equation
describes the line of force
.
Fluid Flows of field of vectors arises in many other ways besides as force fields. Such an example which gives rise to a field is that of a fluid in motion in a certain domain in R3. There are various ways of describing 0, we that flow. First of idealize, by assuming that at the time t The
general notion
all,
=
may
310
Curves
4
a
a >
=
0
a< 0
0
Figure
4.4
Then we can describe the flow by a particle at each point x0 in R3. describing the motion of each particle. The particle which is at x0 at time t 0 follows a certain path which is given by a function f(x0 t). The equations of motion are thus there is
=
,
x
=
f(xo,0
position x at time t of the particle originally at x0 is f(x0 0particles are neither created nor destroyed; this amounts to f(x0 0 is one-to-one and onto, and asking that, for each t the function x0 Precisely, We
the
assume
,
that
->
thus
can
be inverted.
x0
for
some
time
-
=
we can
,
also write
function
1 was
So
(b(x, t).
Precisely,
the
original position
of the
particle
at
x
at
Fluid Flows 4.
Suppose
a
the vertical axis. in
gas is rising at constant speed, and spiraling around Thus the motion of particle is a helix as described
2.
Example
We do best to express this motion in cylindrical be the (complex) coordinates in the plane (z reie) the height off the plane. Thus the path of motion described
coordinates: Let and
311
w
z
=
the gas is
by
(z,w)
=
(zoeu,at
Thus the
time
t
+
w0)
(4.3)
particle originally at (z0 w0) will be can certainly invert these equations : ,
at z0
e",
w0 + at at
We
.
(z0, w0)
=
(ze-it, w-at)
(4.4)
Now, another
Let v(x, t) way to describe a fluid flow is by its velocity. velocity of the particle which is at position x at time t The field v is called the velocity field of the flow. We can find the equations of motion from the velocity field by solving the appropriate differential equation. For the function f (x0 0 describes the motion of the particle originally at x0 The velocity of this particle at time t is f '(x0 0 and its position is f (x0 t).
be the
.
,
.
,
Thus
we
must have
f'(xo,0 f (x0
This
,
,
0)
equation
x0
can
5. Let
flow
v(f(xo,0, 0
=
=
us
be solved find the
equations
(z',w')
=
uniquely.
are
velocity field of the gas flow in Example 4. The (4.3). The velocity of the particle originally at x0 is
(iz0e",a)
velocity field we must write this as a original position. We can inversion (4.4), obtaining as velocity field
To find the
function of
at time t, rather than
do this
the
y(z, w)
=
by
position
means
of
(iz, a)
6. Suppose a fluid on the plane is spiraling (Figure 4.5) according to this equation of flow
in toward the
origin
312
4
Curves
Figure 4.5 Here the
particle at time t 1 moves toward the origin so that its argument is proportional to time elapsed, and its distance from the origin is inversely proportional to time. Then
*
=
=
iz0eil
z0eu
Thus the
velocity
v(z,t)=
(--)z
The
I
field is
angular velocity is
decreases
as
1\
F-'=(,")Z(0
-
time goes
thus constant whereas the radial
7. Suppose now a fluid spiraled velocity field was time independent,
v(z) The
/'(0 /(0)
=
(i
=
=
of motion
0-i)/0) z0
in toward the for
example,
l)z
equations
velocity
on.
are
the solutions of
origin
so
that its
4.1
This
/(z)
Parametrization
of Curves
313
gives =
e("1 + i)(
=
e-'ei'
In this case the distance from the time (Figure 4.6).
origin decreases exponentially
with
We shall make a study of the geometry of paths of motion of single particles and fluid flows, or families of motions, in this chapter. This study is a con tinuation of analytic geometry, and begins the subject of differential geometry.
Figure 4.6
4.1
A
Parametrization of Curves
curve
in R" is
a
one-dimensional subset T of R".
be put into one-to-one correspondence with We make this notion a little more precise. set T can
a
This
means
line, in
a
that the
smooth way.
Definition 1. The image in R" of an interval under a continuously differ entiable one-to-one function with a nowhere vanishing derivative is called a C1 curve. If the function is A>times continuously differentiable we shall call
this
curve a
of the
curve.
Ck
curve.
The
particular
function is called
a
parametrization
314
Curves
4
Examples 8. The unit circle in R2 is T:
z(f)
t, sin
parametrization:
teR
t)
(4.5)
( sin t, cos t) is never zero (the sine and cosine simultaneously zero), this is a good parametrization. could also parametrize the unit circle in this way :
Since
z'O)
never
We
z(t)
(cos
=
It has this
a curve.
(t,(l-t2)1'2)
=
but this
(4.6)
parametrization fails
is not differentiable there.
at t
=
other parts of the circle.
(0
=
(4.6)
does not
parametrize of these failings
parametrize
z(0
=
the circle in the
care
cos t
Another
z(f)
=
where
z(t)
For
plane.
be written
=
=
example,
+ i sin t
curve
use
the
so on.
complex notation to parametrization of
describe
curves
the circle
(4.5)
as
=
is the
eu
spiral :
ect c
is
some
complex
number.
ett,eibt
or, in
polar notation,
r(t)
e"'
=
right half-plane,
of the lower semicircle, and
9. It is often convenient to
z(0
the
0, -(l-*2)1'2)
in the can
cover
is,
t2)1'2, 0
-
will
takes
That
f2)1/2
+1, since the function (1
Notice that
the whole circle, but only the upper semicircle. Both can be alleviated by introducing parametrizations which
z(0
are
=
0(0
=
bt
z
=
rew
Writing
c
=
a
+
ib, this becomes
4.1
Thus the modulus of
varies
z
is linear in t
(see Figures
10. The
T:
x(t)
x'0) is
(sin t,
=
called
a
=
.
cos
right
(cos
never
11
curve
t,
circular
(4.7)
exponentially 4.8)
315
of Curves
with /, and the argument
4.7 and
0
t, sin t,
zero,
Parametrization
is
(4.7) helix, is pictured in Figure 4.9.
Since
1) a
valid
parametrization
The intersection of two
of the
curve.
cylinders with different axes is a curve cylinders are both of radius 1 and
the
(see Figure 4.10). Suppose one, Cx, has as axis the y axis, and Then Cx has the equation x2
+
and
y2
+
z2
=
the other,
C2 has ,
as
axis the
x
axis.
(4.8)
1
C2 has the equation z2
=
(4.9)
1
z(t) =eu',Rea >0
Figure 4.7
316
4
Curves
zU)
=
e"',Rea
Figure
<
0
4.8
The intersection is, of course, the set of points where both can be written x2 1 z2, y2 1 z2. thus parametrize at least part of the curve by hold and thus
x
=
f(0
(l-z2)1'2 =
=
y
=
(l-z2)1/2
((i-2)1/2,0-f2)1/2,0
Figure
4.9
-
or
=
-
equations We
can
4.1
Parametrization
317
of Curves
Figure 4.10 Other parts will be found
f(0 f(0
on
this theme :
(-(l-f2)1/2,(l-f2)1/2,0 (M,(l-*2)1/2)
=
=
and
by variations
so on.
simpler parametrization
A
Then
we
fi(0
=
f2(0
=
is found
by
the substitution
have the two distinct branches of the intersection
(cos,
f,
(cos,
t,
cos -
t, sin
cos t ,
x
=
cos f.
given by
t)
sin
t)
Implicitly Defined Curves In the situation of the above
the
example, (4.9). Equations (4.8)
implicitly by are given a collection
and
of
equations
such
as
say that the curve is given More often than not, when we
we
these,
we can
determine, just by
318
4
Curves
working with them,
whether or not they do implicitly define a curve. Never theless, the theoretical question remains: under what conditions can the set defined by a collection of equations be parametrized as a curve ? We
have
answered this
already
restate the conclusion
as a
question
fact about
in R2 in Theorem 2.14.
We shall
curves.
Proposition 1. Suppose that F is a differentiable real-valued function defined in a neighborhood of (a0 b0) and F(a0, b0) 0 but dF(a0 b0) # 0. =
,
,
Then the set
{(x,y)eN:F(x,y) is
a curve
in
some
=
0}
(4.10)
neighborhood N of(a0 b0). ,
Proof. Since dFia0 b0) = 0, then either (SF/8x)(a0 b0) = 0 or i8F/Sy)iao b0) Suppose the latter. Then, according to Theorem 2.16, there is an e > 0 and a differentiable function g defined on the interval (a0 0 e, a0 + e) such that Fix, y) if and only if y Let /: (a 0 gix). In particular, gia0) b0 e, a0 + e) ->- R2 be defined by /(f) (f, git)). Then / parametrizes the set (4.10) near (a0 b0), and clearly f'(t) (1, g'it )) ^ 0. If instead (dF/8x)ia0 b0) # 0 we can give the same argument merely by changing the roles of x and y. ,
,
,
= 0.
=
=
=
.
=
,
=
,
In
dimensions the situation is
higher
describe it in R3. borhood
of a point
{p: Fiji) is
-
p0
,
little
more complicated. We shall differentiable functions defined in a neigh and VF(p0), VG(p0) are independent, then the set
If F,G
F(p0)
=
are
a
two
0, G(p)
-
through p0. The verification of this fact is
G(p0)
=
0}
(4.1 1)
a curve
basically
another
of the fixed
point assume that algebra. F(Vo) 0 G(p0). Since the vectors VF(p0), VG(p0) are independent, we can change coordinates in R3 so that That VF(p0) E2 and VG(p0) E3 is, with respect to the new coordinates (x, y, z), dF/dx(p0) 0, dF/dy(p0) 1, 1. Now let dF/dz(pQ) 0 and dG/dx(p0) 0, dG/dy(j,0) 0, dG/dz(p0) Po (xo yo zo) ; fr x near x0 we want to show that there are uniquely theorem, complicated by =
some more
linear
use
We first
=
=
=
.
=
=
=
=
=
=
=
>
,
determined y,
such that
0
G(x,
Newton's
method,
F(x, Following
z
y,
z)
=
y,
we
z)
=
0
ask to find the fixed
point
in the y,
z
plane
4.1
of Curves
Parametrization
319
of the transformation
T(y, z)
dF
/
Our conditions
VF(p0)
borhood of p0 such that
T is
,
dG
[y + Yy iv*)
=
a
T(g(x), h(x))
Fix>
y>
E2 VG(p0)
=
=
,
contraction.
Z^>z
+
Tz
(x'
(Po)
\
y'
z))
E3 will guarantee that in some neigh are unique y g(x), z h(x)
Thus there
=
=
(g(x), h(x))
=
or
F(x, g(x), h(x)) Thus the function
=
/(0
0
=
G(x, g(x), h(x))
=
0, g(t), h(t)) parametrizes
the set
(4.11)
as a curve.
Examples
plane is the set ex+y y a curve? Let 1). Since dF/dx F(x, y) ex+y y. Then VF(x, y) (ex+y, ex+y 0 is never zero, this is everywhere a curve and the equation ex+y y determines x as a function of y implicitly. dF/dy is zero when 0 The only point on the curve where ex+y 0. x + y y and x + y a function as to find is ( 1, 1), so at that point we cannot expect y 12. At what
points
in the
=
=
-
=
-
=
=
=
of
x
=
x.
Notice, that even though we cannot explicitly determine the function =f(y) given implicitly by ex+y y, we can find its derivative. For =
exp(/(y) so
upon
+
y)
-
y
=
0
differentiating
we
have 1
y)(f'(y)
+
D
exp[-(/(y)
+
^)]-l
exp(/(y)
+
-
=
0
or
/'(y)
=
13.
F(x, y)
=
x
sin xy
-
cos
y
(4.12)
320
4
Curves
WF(x, y) If
x >
of
x.
(sin
=
xy + xy
cos
1, dF/dy(x, y) # 0,
Differentiating (4.12)
sin xy +
cos(xy)(y
x
xy')
+
xy,
x2
cos
(4.12)
so
xy + sin
defines y
with respect to +
y'
sin y
=
y) implicitly
x we
as a
function
find
0
or
sin xy + xy =
y
:
sin
14. VF
F(x,
y +
y,
VF and VG
x3
3x2y
=
These
=
y2, G(x, y, z)
when
=
(yz,
=
xyz + ez.
xz, xy +
e2)
0
2y
xy + ez
become
and
0
+
dependent
equations
xy + ez
x3y
VG
xz
yz
xy
xy
2y, 0)
+
are
+
cos
x
z)
(3x2y, x3
=
cos
2
=
y
x3
or
3x2y
=
0
The first
=
x3
pair
+
has
2y no
solutions, and the second pair
amounts to
x
=
0
0. But the set F(x, y, z) and y 0 never intersects 0, G(x, y, z) this plane, so everywhere on that set F and G are independent. Thus =
{(x, is
y,
=
z): F(x,
a curve
y,
z)
=
G(x,
y,
z)
=
=
0}
in R3.
Comparison of Parametrizations that a given curve admits many parametrizations, and advantage to be able to single out a best possible one. In the study of the motion of particles there is a distinguished parameter, that of time. But as far as the geometric study is concerned we can take any parametrization we care to, the only criterion being that of convenience.
Now,
we
have
it would be to
seen
our
4.1
Geometrically,
a
is that of length
as
Before
most convenient
measured from
considering
how to compare
see
Parametrization
parameter,
or
321
of Curves along
measure,
the
curve
fixed
point. particular parametrization by arc length, let us first two different parametrizations. Suppose T is a curve, a
the
parametrized by x
=
a
f(t)
continuously differentiable function with nonzero derivative defined [a, /J] and taking values on the interval \a, b], then the composed function / a also parametrizes T. That is, we can write T as the image of
If
is
a
a
the interval
on
x
If
t
increases
sense
*
<700=/(
=
does, then these
as f
of direction
the
along
two
curve
parametrizations
T.
This
sense
determine the
same
of direction is called
We know from calculus that the necessary and sufficient condition for t, x to increase simultaneously along the curve is that a' > 0 We shall say that x is an orientation-preserving on the interval [a, j8]. orientation.
parameter if this condition is satisfied, and otherwise x is orientation reversing. On the other hand, if we started out with two different parametrizations of
a curve
r:x=/(0
then there must exist
point of
=
function
a
(4.13)
flf(t) a
precisely corresponds correspondence to
of T The
t.
x
or
relating one
For each
the two parameters.
value of t and
precisely
one
value
T-0(T)=/(t)-*f defines the function function of
t
and
Notice that, the chain rule
we
given
a.
have
We shall
g{x)
the two
=
verify below
and
point
same
in the
g'(x) and/'O) same
is
a
differentiable
so
that t
=
o(x),
we
have
by
(4-14) are
collinear when t, x are the same points, > 0, that is, when g,f induce the
direction when a'
orientation along T.
a
f(o(x)).
parametrizations,
g'(r)=f'(o(x))-o'(x) Thus the vectors
that
322
Curves
4
Definition 2.
Let T be
a curve
parametrized by x =f(t), a
The
unit tangent vector to T at /(f) is the vector
no
rS l/'0)l
=
By the above remarks
parametrization
orientation.
For if
(since
a' >
we
have the two
/VOOVO) l/VO)) r'OOl
_
determine the
x
we
parametrizations (4.13), then by (4.14)
0)
/'0) Ifftol when f,
that the unit tangent vector is the same no choose so long as it induces the same
we see
matter what
same
fit) l/'(0l
_
point
of T.
Examples 15. Consider the unit
z
=
circle, given parametrically by
e"
Then z'
=
Notice :
ie",
we
which is
have T
a
iz,
unit vector, soJ"= ieu. so that the tangent vector is
orthogonal to position vector. More generally, consider the spiral z eat, where a is a complex number. Then z' aeat, so the tangent vector is exp f(Im a + arg a)t. Notice that the angle between the tangent vector and the position =
the
=
=
vector is
arg T
arg
Thus T,
z
z
=
we
ft
have
(l,2t,3t2)
a
make the
curve
(t, t2, t3)
=
arg
always
16. For the
x
=
same
in space
angle.
given by
4.1
Parametrization of Curves
323
so
T(0=(i
8f2l+9fr2(1'2t'3f2)
+
Now, here is the verification of the fact that related
by
continuously differentiable
a
2.
Proposition
Let T be
a
curve,
two
parametrizations
are
function.
and
f: [a, b]->r,
g:
[a, /?]
->
T two
parametrizations of T. Then there is a continuously differentiable function
andf(t)
=
g(o-l(t)), for
all
t e
[a, 6].
Proof. Let t g [a, /?]. Since / maps [a, b] one-to-one onto V there is precisely t. Then a is a well-defined te[a, b] such that fit)=gir). Define
=
=
9(ri) =/(o-(t0) =/(a(r2)) =^(r2) Since g is one-to-one a
fit)
maps =
gir).
[a, /3]
we
Clearly,
must have
then
t
ti=t2.
For if t
[a, 6].
onto
=
6
[a, 6] there is
a
point
t g
[a, /?] such that
o-(t).
We
now have only to verify that a is a continuously differentiable function. Let [a, ft] and f0 o-(t0). Now / is a differentiable function at f0 and /'(f0) = 0. Let / (/,...,/) in coordinates. There is a / such that />(/<>) ^ 0. Then / is a real-valued continuously differentiable function of a real variable and since /'j(fo) ^ 0, it is invertible. That is, there is a function /; defined on the /? is also continuously differ t for f near f0 range of/ near f0 such that hifit)) entiable. Now, since /(o-(t)) =#(t), we have /j(
=
=
=
.
^(t)=(/(o-(t)) Since
and /}
are
=
(/Io^)(t)
continuously differentiable
so
is
<j.
The
proposition is
proven.
requirement that the derivative of the parametrization is nonzero we would in general not have such a good relationship between different parametrizations. Notice that by the same argument the inverse mapping
=
(ff-1)'W0)-ff'(0
=
i
324
4
Curves
is also
never zero. If it is always positive, a is an increasing function always negative o is a decreasing function of t. Notice that if / g are two parametrizations of a curve and they do reverse orientation, then they will become compatible simply by negating one of the param eters. Thus if /is not compatible with g, then/: \_-b, a] C defined by f(t) =/( 0 certainly is. T is a parametrization of a curve we shall call f(a) the left If/: la, b~\ end point of T and f(b) the right end point. so
cr'O)
of t; if
-
->
The
Line
Tangent
Now, let T be at x0
is the
in R", and x0 a point through x0 which best
a curve
straight
line
shall show that this is the line
through
T.
The tangent line to T approximates the curve. We on
the tangent vector and is
given by
this
equation x
x0 +
=
tT(x0)
t e R
The tangent line at x0 can be x0 and nearby points xx
through
be that line. to xx
Now
-
x0
L(xx)
the
limiting position of lines (Figure 4.11). Let L(xx) Then Z,(xj) is the set of all vectors originating at x0 and parallel Let /give a parametrization of T so that x0 /(f0), xx f(tx). the set of points x such that x x0 is parallel to computed on
T,
as
as
xx
->
x0
=
.
is
=
xi-x0=/01)-/0o) But that is the
same as
the set of points
x
such that
fih) -fjtp) h-t0
Figure
4.11
x
-
x0 is
parallel
to
Parametrization
4.1
Now Xt
fj
-
f0 is
/'00),
x0 is the
->
as
same as t x
Thus
/'(f0)-
L0u)
->
10
of Curves
325
and the limit of the difference quotient as through x0 and parallel to
tends t0 tne lme
desired.
Examples 17. Consider the helix
f(0
(a
=
t,
cos
a
sin t,
(Figure 4.9), given by
the
parametrization
bt)
Then
f'O) f is
(
=
a
T(0 w
a
sin t,
a cos
t,
b)
positive parametrization
=
r-2
(a2
+
T2TT72
b2)1'2
(~fl
sin *'
if
we
a cos
take for the unit tangent
'
b>>
(see Figure 4.12). 18. A
f(0
=
(e'
damped cos
helix
t, e' sin t,
(Figure 4.13) parametrized by bt)
Thus
f'O)
=
(e'(cos
t
-
sin
t), e'(sin
f + cos
Figure 4.12
t), b)
(4-15)
326
4
Curves
Figure so we can
T(0
take
as
the tangent vector
=
(2e2t
+
b2)1'2
Notice that the
4.13
^' (-CS
'
Sin
~
^' e'(sin
* + C0S
^' ^
('4*16^
the unit
sphere swept out by the tangent (Figure 4.14), and that the functions (4.15) and (4.16) give two different parametrizations of this curve. If we consider the parameter as t, then the moving point described by (4.15) has no tangential acceleration, whereas in (4.16) it is acceler ating exponentially. is the
same
curve on
for both helices
"
19. A different helix is this
f(0
Here
T(0
(cos
=
we
t, sin t,
take
as
(1
+
e2()i/2
(Figure 4.15):
e')
tangent
=
one
(~sin
vector
'>
cos
'>
e<)
"
4.1
Parametrization
of Curves
327
Figure 4.14 This
(Figure 4.16). to the
equator
as t
+
->
as t
again
->
(Notice
oo.
1
z(T(0)
-
oo
is
a
helix
on
and winds
the unit
rapidly
that
?1
as t
oo
-
=
(l
+
-2t\l/2
e~2<)
*
0
as t
-
Figure 4.15
-oo)
sphere
which tends
around the north
pole
328
4
Curves
Figure 20. The intersection of a
x2
y2
+
z2
(x-i)2
+
y2=i
+
=
x(0)
=
and
z(6) is z(9) is
curve
lying
(1, 0, 0)
at
cross
We shall
i
+
above the xy
the the
(2)
y(9)
=
\
shall restrict attention to the
plane.
Let
angle
us
9
first as
parametrize
shown in the
point on the unit sphere lying above (x(6), y(9), 0), positive square root of 1 (x(0))2 (y(8))2, which is -
.
we can
9
parametrize this
Then
-
1
-
-
=Sm2
+
=
we
sin 9
curve
f(0)=(- -cos0,^sin0,sin^ f '(6)
cylinder (Figure 4.17)
parameter the
use as
i cos 9
/l-cos0\1/2
Thus,
a
Then
figure.
thus
sphere and
1
In order to avoid the
part of the this curve.
4.16
sin 6,
cos
9,
cos
-1
with the function
4.1
and
take
we can
as
Parametrization
of Curves
329
tangent line
\1^2/
9\
(2 JTc^j {-^e^os9,oos^
(4.17)
Notice that this does not
parametrize T at the point (1, 0, 0), since point corresponds parametric values 0, 2n. In fact, T is not a curve at the point (1, 0, 0) since it does not have a unique tangent line: the limiting position to (4.17) as x-> (1, 0, 0) is either (0,1, 1)/V2or(0, 1, -1)/V2!
this
to both
EXERCISES 1. Find
x2 +
a
parametrization for the
iy2 + z2
with the
x2 + z2
=
curve
of intersection of the
ellipsoid
1
cylinder
=
1
paraboliod z=x2 + y2 with the 1. y2 + z2 3. At what points is the set defined (in polar coordinates in R2) by 1 a curve? Find a parametrization of the curve. r(l + a cos 6) 2. Parametrize the intersection of the
unit
sphere
x2 +
=
=
Figure 4.17
330
4
Curves
Figure 4. Consider the
r
4.18
family of cardiods (Figure 4.18)
(l+c)_1(l +CCOS0)
=
(a) Describe the behavior of this family as c ranges between 0 and + oo 1, c 2, calculate the unit tangent vector to the curve as a (b) For c .
=
=
function of 6. 5. What is the tangent vector to the for 6
=
1,2,
curve r
a cos
bd ?
Graph the curve
5,V2.
6. Calculate the tangent lines to the
(a)
f(f)=(e-'cosf, e-'sinf)
(b)
f(x)
(c)
f (f )
=
=
(X'Sinx) (e''7+-rsin') (x,sin-|
I e',
,
following curves : at (1,0).
at
(1,0).
sin f I
at
(1
,
1
,
0).
at (2a, 0, 0). x2+y2 + z2=4a2,ix-a)2 + y2=a2 f it) at (0, 1, 0). it, cos f, sin f) at (1,0,1). f(f)=(f2,l-f2,f) Find the tangent line at the origin for these curves. (a) ex+v-y-\=0 (b) cos xy y + 1 exy2 cos xy =0 (c) x2 + y3z + sin z 0 1 x2 + y2 + z2 x + y + z (d) exp(sin (xy + z))
(d) (e) (f) 7.
=
=
=
=
=
=
PROBLEMS 1
A snail
deposits calcium at the leading edge of its shell in a direction a fixed angle with the ray from the snail's center to the leading edge. Show that this hypothesis explains the spiral form of a snail's shell. 2. Graph the curve r (1 + d2Y\2 + 62) and compute its tangent .
which makes
=
vector.
4.2
Arc
331
Length
Graph the curve in R3 given in spherical coordinates by r ', 6 t, Graph the curve on the unit sphere made by the tangent vector of the given curve. 3.
Arc
4.2
=
Length Let T be
Definition 3. f:
[a, b~]
and
->
f(b0)
=
e'.
=
z
T.
Let
a <
oriented
an
a0 <
b0
<
b.
curve
be the least upper bound of all
to
positively parametrized by length of T between f(a0)
Define the sums
(4.18)
I llfft) -*('-. over
all choices of
=
a0
t0
<
tx
points
10
<
<
,
tk
.
.
=
.
,
tk such that
b0
description. Approximate (Figure 4.19) the curve joining a succession of points along T between a0 by a of the lengths of the line segments is less than, sum Then the and b0 and approximates the length of the curve. Now, if the points t, and tt-t are very close, then the vector f(t,) f(tt-x) is approximately equal to we get a sum in this If we (4.18) replace f'O.X'i *i-i)-
This definition has this "
broken line
"
.
-
-
(4.19)
El|f(*,)ll('i-'i-i) which is
a
Riemann
sum
approximating
the
integral
/.&0
f
J
l|f'(OII
(4.20)
dt
fin
t
Figure 4.19
b
332
4
Curves
(4.18) to (4.19) admit a small error by large, we have no hold on the error between (4.18) and (4.19). Nevertheless, we can, by being very careful, justify that substitution and deduce that the limit of the lengths of the approxi mating line segment curves is the integral (4.20). Of course, the substitutions
taking us
from
term but since k may be very
term
Proposition 3. Let T be a curve parametrized by f : [_a, b~\ length ofT between f(a0) and f(b0) is given by the integral (4.20).
->
F.
The
Proof. We will use the fundamental theorem of calculus to show this. Let sit) be the length of T between f (a0) and f (f). We shall show that s is a differen tiable function of f, and /(f) ||f '(f) ||. If S0 is any sum like (4.19) approximating the Fix f0 > a0 and consider a f > f0 length of r between f (a0) and f (f0), then S0 + ||f (f ) f (fo) II is a sum like (4.19) for the length between f (a0) and f (f ). Thus =
.
-
S0+\\f(t)-f(to)\\<s(t) Taking the
least upper bound
over
all such So
,
we
obtain the inequality
(4.21)
5(f0)+||f(f)-f(fo)ll<s(f)
(4.19) corresponding to a partition of the interval one of the points in this partition. For if not, add it to the given partition, and get a still larger sum. Let f0 < fi < t be the points of the partition between f0 and f. Then
Now, [a0 t ]. ,
we can
< tk
=
suppose S is
S
like
a sum
We may suppose that t0 is
=
S0+
l!f(f,)-f0,-i)ll
i=i
where So is
a sum
s<sit0) +
^s(t0)+
corresponding
to the interval
[a0 t0]. ,
Thus
2\\fit>)-Hti-M 2
f'
f'O) dt
"'i-l '
<^s(t0) + 2 < i =
f
-Vi
||f '(Oil dt<s(t0) +
Since this is true for all such
sums
*0)^^o)+f'l|f'(OII*
S,
we
f ||f '(f)|| dt Jt0
have
(4.22)
4.2 From
and
inequalities (4.21) ||f(f)-f(f0)ll
(4.22),
s(t)-s(t0)
<
As t->t0, both the left and
Thus
for f0 between
5
a0
Now, if T is
1 <
right
Length
333
obtain
f
, , tto
t-t0
||f'(fo)ll.
we
Arc
llf '0)11 dt
(4.23)
ends converge, since f is differentiable, to Since this is valid
is differentiable at f0, and s'(t0) ||f'(f0)!l. and b0 we have the desired conclusion. =
,
parametrized by f : [a, b~\ T, we can consider arc T. along Precisely, let s(t) be the length of the piece of length T from /(a) to /(f). Then, from the above proposition, a curve
->
function
as a
5(f)
=
J'||f'(0ll
dt
parametrize T by arc length, and it induces the same orientation original parametrization. Thus g(s), for every on T of distance s from a: g(s(t)) s is the point f(0- If L 's the length of Notice that T parametrizes T. T from a to b, g: [0, L] Since
s'(t)
||f0)11
=
>
0,
as
we can
the
=
-*
f'O)
=
=
so
g'0(0) s'(t)
g'O(0)-llf'(0ll
that
8,(!(,,)=If|i=T(,) Thus
g'OO
is the unit tangent to T at
g(s).
Examples 21. The circle x2 +
x
=
a cos
9
y
=
a
y2
sin 9
Thus
f(9)
f'(0)
=
=
(a
cos
9,
a
sin
9)
a(-sin0,cos0)
||f'(0)||=a
=
a2.
Parametrize this circle by
334
4
Curves
Thus
s
=
x
re
s(9)
=
ad9
a9
=
Jo
parametrization according to
The
ing
length is given by
arc
s
=
I
g(s)
I
=
s
a cos
,
a sin
(\ -sin-,
cos-
given by
substitut
given by
1
af
a
(a
=
is thus
-
22. Consider the helix of
f(0
length
s\1
.
-
The unit tangent vector is
T(s)=
arc
a9.
=
t,
cos
a
sin t,
Example
17
given by
bt)
Then
f'O)
sin f,
(-a
=
\\f'(t)\\=(a2 Thus
g(s)
s
=
=
(a
+
a cos
t,
bt)
b2)112
s(t) =((a2
cos
(fl2
+
b2)ll2)t,
+Sb2)X/2
,
a
and the
sin
(<j2
arc
+Sfo2)1/2
length parametrization
,
(fl2
+
fc2)1/2
The tangent vector is 1
T()
=
(fl2
23. The
+
b2)l/i(-a
curve
of
Sin
*>
Example
C0S
fc)
20 has the
f(0)=^ ^cos0,-sin0,sin-j +
f>
parametrization
s)
is
4.2
and
we
Arc
Length
335
find
||f'(0)||=-i=(3 2cos0)i2 +
2sJ2
so
5(0)
=
-^= f (3
2^2 Jo
+ 2 cos
deb
and the unit tangent vector is
given
as
T(0)=(3r!o70)1/2(-sin0'cos0'co4) Equations of Motion Now
shall consider in greater detail the equations of a particle in Suppose a particle moves through R" along the path given by
we
motion.
The
at time t is
x'(f), and the acceleration is x"(f). These describing the instantaneous change in the motion (direction and magnitude) of the particle. The speed of the particle is the rate at which the distance covered changes, and thus is the time deri vative, ds/dt, of arc length. As we have seen above, this is the magnitude of the velocity. Thus x
=
are
x(f).
velocity
vector-valued functions
.
dx
.
velocity
dt
Now, it is instinctive tangent
to
a
d2x acceleration
=
=
-=
dt2
(4.24)
=
dt
decompose
to the curve, and
dx
ds
.
speed
=
dt
the acceleration vector into
component orthogonal
a
component
to the curve.
We write
_
=
aTi + aN IN
where T is the tangent vector and N is a unit vector orthogonal to T and lying in the plane spanned by the velocity and acceleration vectors. N is called the principal normal to the curve of motion, aT is the tangential acceleration of the
particle,
and aN is the normal acceleration.
We
now
show how to
336
4
Curves
compute these components of the acceleration. dx
Differentiate the
equation
ds T
=
dt
dt
obtaining d2x
d2s
ds dT
l?~dT2
+Jtli
_
Now
(4-25)
dT/dt is orthogonal to T,
=
since T is
a
unit vector.
Differentiate
1
We then have
we can
N
(Of
+
2
-
=
0
dT/ds
(4.26) dT/dt
(d 27)
-
course, the differentiation in
length as well as time.) Let k path of motion. Then dT/ds
=
take for the normal vector the unit vector in the direction
dT/dt
=
=
=
(4.25) could have been with respect to arc || dT/ds || This is called the curvature of the .
=
dh
dsdTds
dt2
Ttls It
jcN, and (4.25) becomes
_ ~
dt2 .
Thus the
d2x
.
acceleration
=
^
d2s d2^ (ds\: /ds\2 N T + k\
dr2T+idt)\
=
tangential acceleration is the rate of change of the speed, and proportional to the curvature, or bending, of
normal acceleration is
the the
curve.
d2s aT
=
dT2
a
=
/ds\2
{Tt)K
(4-28)
4.2
Arc
337
Length
Examples 24.
Suppose a particle moves along according to these equations x
=
t- 1
y
=
2t
-
the
parabola
y
=
1
x2
t2
Then
x
=
(t-l,2t-t2)
l-O.Ki-0) d^X -(0,-2) dt
2
particle is determined by a downward vertical acceleration of constant magnitude (perhaps due to gravity) (see Figure 4.20). The speed of the particle is
Thus the motion of the
dx =
It Thus mum
(1
+
we see
=
(1
+
4(1
-
speed is decreasing until time t trajectory), and then increases.
that the
path -
(4-29)
t)2)1/2
of the
height
vector to the
T
4(1
of motion is
OT1/20> 2(1
-
0)
Figure 4.20
=
1
(the maxi
The tangent
338
Curves
4
and
so
(4.30)
-[l+4(l2-,ff"0(1-"'-1) The normal vector is the unit vector in this direction : N
(1
=
+
4(1
-
f)2)"1/2(2(l
-
t),-l)
Now
dT
dT /ds
ds
dt/dt
(l+4(l-02)2
(2(1-0,-1)
2
N
[1
+
4(1
02]3/2
-
Thus the curvature of the
path
of motion is
2 K
~
(1 And
+
4(1
-
02)3/2
finally d2s
aT
4(1
-
tds\2
0
_
""
=
dt2
The
+
4(1
t2))3/2
length of the trajectory from dx
f Jo
(1
-
dt
x
=
-
"
\dt) 1 to
x
2 K
~
(1 =
4(1
-
f)2)1/2
+ 1 is
dt=\Jq [1 + 4(1 -f)2l1/2 dt
(Rotation) (Figure 4.21). Suppose now that according to the equations
25.
+
a
particle
rotates
around the unit circle x
=
cos(e')
y
=
sin(e')
Then x
=
(cos(e'), sin(e'))
dx =
It
(4.31)
e'( sin(e'), cos(e'))
dht. -j-%
dt
=
e'(-sin(e'), cos(e'))
-
e2'(cos(e'), sin(e'))
(4.32)
4.2
Arc
339
Length
Figure 4.21 Now are
T
we
the tangent and normal
=
(
Thus
(4.31)
d2x =
di
can
to
N
sin(e'), cos(e'))
-
from
already know, just
be written
the
=
-
geometric considerations,
path
(cos(e'), sin(e'))
as
e'T + e2tN
Thus the normal acceleration is the square of the tion. From (4.31) we read
ds_
dx
dt~
dt
=
tangential
accelera
e'
thus s
=
e'
and the curvature of the unit circle is 1
Notice, that x
=
what
of motion:
any motion
on
the unit circle
(cos(/(0), sin(/(0)
.
can
be written in the form
340
Curves
4
Figure 4.22 where
f(t) represents
of the unit circle is 1,
acceleration
The
arc
=
length
as a
function of time.
Since the curvature
obtain for any circular motion
we
d2s (ds\2^ I N T + I -j
dt2
\dt]
tangential acceleration
is the rate of
acceleration is the square of the
change of speed, and the normal
speed.
us consider the motion of an object down a slide (see The slide will be represented by the curve T. Let 4.22). Figure z z(f) x(f) + z'yO) be the equation of motion of the particle. The acceleration is z"(t) ; according to Newton's laws
26. Now let
=
=
mz"
=
where
F
m
is the
mass
of the
object,
and F is the
sum
of the forces
One such force is the force due to
on the object. gravity which is mg, where g is the gravitational field. The other force is the restraining force due to the curve. This force acts in a direction normal to the curve, and has undetermined magnitude. (That is, its
acting
magnitude <j>N, where
is determined
a
have
mz"
=
mg +
<j>N
only by
the
object.)
Let
scalar and N is the normal to the
us
call this force
curve.
Thus
we
4.2
Arc
Length
341
Now, since we know the path of motion, we need only determine the tangential acceleration aT By Equation (4.28), we have .
dh =
dt
2
aT
=
=
(4.33)
where T is the tangent of the
parametrized by then the
curve.
If
we
consider the
curve
as
length: z=f(s) is the equation of the curve, tangent vector is f'(s). Then Equation (4.33) becomes arc
dh =
dt
2
<9, f(s)>
and the
speed
can
be found
with initial conditions
as
the solution to this differential
s(0) s'(0) For specific examples, let us first line (Figure 4.23) with equation z(s)
=
where
Tit)
=
=
=
consider the
i +
s0
0 o
is constant, and the force due to
=
a
+ ib is
a
unit vector in the third
Figure 4.23
equation
0. curve
to be
a
straight
quadrant (b > 0). Then gravity is ig. The speed
342
4
Curves
Figure 4.24 is thus found
as
the solution of the differential
equation
d2s 2
s(0)
=
s'(0)
=
Thus z
{-ig,a
=
s(t)
Z(t)
27.
z(s)
sin
i
s
=
(gbt2)/2
-gb
and the
equation
of motion is
(\gbt2)^
-
Suppose
=
ib}
0
=
=
=
+
+
the
now
curve
is
a
semicircle
(Figure 4.24)
/ cos s
Then
T(s)
=
and the
i sin
cos s
is the solution of the differential
speed
d2s ~2
s(0)
s
equation
.
=
=
ig,
\
s'(0)
cos 5
i sin
5>
=
g
sm 5
0
=
Rotating Plates 28. We
referring
can
describe the motion of
to the
center of the
angle plate be
as a
rotating
chosen
this line makes at time t with
flat circular
plate by through the at time t 0 and let 0(f) be the angle its original position. Then a point at a
function of time. =
Let
a
line
4.2 z0 at time f z
z"(0
=
iz0 0 V<"
Thus the
343
of motion
The acceleration of
z0(9')2eie^
-
tangential
(4.34)
acceleration is
|zo|0"
and the radial acceleration
z0(9')2. If there is
so
that the
provide is
path
velocity is iz09'e'm, so its speed is |zo|0'. point is found by differentiating further :
the
is
0 follows the
Length
z0em,)
=
Its
=
Arc
object of mass object will follow an
this force.
zo(0')2,
so
even
m on
the
plate,
a
force
the motion of the
mz"(t) is required plate. Friction may
Notice that the central component of this force no angular acceleration, friction must
if there is
do its job. The further the object is from the center, or the faster the plate spins, the greater the force required. It is this principle which explains the centrifuge, which settles precipitates in solution by spinning the fluid. 29.
Suppose
now
we
have
angular velocity, (Figure 4.25). Assuming constant
a
curved circular
and there is
a
there is
friction,
no
ball of
plate spinning at in the plate
mass m we
can
describe the
motion of the ball in terms of the initial data.
spherical coordinates r, 0, z in R3, equation z f(r). In Figure 4.25 given by section of the plate. Let Let
us use
the
r
=
r(t)
be the of the
9
=
=
0(f)
z
=
so
that the
we
depict
the
angular velocity
a
plate is planar
z(t)
equations of motion of the ball, and let a be plate. Since there is no friction, the ball
Figure 4.25
rotates as does the
344
4
Curves
plate, then 0 0(0 a'- Since plate we must have z(f) /(r(0), =
=
=
x(t)
=
Thus
for all t.
we
on
the
have
(r(t)e"",f(r(t))) equation of motion, and we must find, using Newton's laws, r(t). Now the acceleration is
the
as
the ball is constrained to lie
the function
x"
=
((r"
a2r
-
+
2ir')ei, f'(r')2
+
(4.35)
f'r")
-(0, g) we have a force g be the gravitational field, g that which restrains the is another There due to force, gravity. mg motion to the profile of the plate. This acts in a direction normal Let cbN denote this to the plate and has undetermined magnitude. Letting
There is
force. of
=
plate
a
third force
acting
on
the ball, due to the rotation tangential to the circle on
and the direction of this force is
which the ball lies.
We shall denote this force
by
C.
Then, by
Newton's laws
=
(4.36)
mx"
equate coordinates.
us
it lies in the normal to
Now, since N is normal
plane through the the curve z f(r). =
z
axis and the ball
Thus N
=
to the
surface,
(the plane) (nxe"*', n2) and rz
and is
-=-(f(r)y1 i
(since n2/nx
is the
slope
of the line
perpendicular to
the
curve z
=/(r)).
Since C is tangent to the circle on which the ball lies, C (ceix', 0). The magnitude c of C is yet to be determined. Finally, g is vertical, =
Using (4.35) and substituting equations as a result:
(0, g).
so
g
we
have these three
(bnx c
=
=
these values in
a2r)
m(r"
2ar'm
=
-mg +
Thus, eliminating
(1
=
(4.36)
r(t) is
+
a
solution of the differential
f(r)2)r"
=
a2r
-
f(r)f"(r)(r')2
-
we
find that
equation
f'(r)g
(4.37)
4.2
For what kind of
down,
or
up
have r'
=
r"
a
released
0,
(4.37)
=
so
Length
345
will it be true that the ball will not move We must no matter what its position?
plate
once
Arc
becomes
<x2r=f'(r)g Solving,
we
obtain /(r)
=
a
plate is a paraboloid angular velocity so that it
Thus if the
(a2/2g)r2.
of revolution, we can rotate it at will have this property.
suitable
Suppose we are given a field of force in space, and the initial of position and velocity of a particle. Then we can find the path 30.
example, suppose the force field is the particle are x, and the initial position and velocity of F(x) of this solution the Then the path of motion is given by x0 v0 differential equation: motion of that
For
particle.
=
.
,
/"(0
/(0)
-At)
=
=
x0
/'(0)
=
v0
We know the solution; it is
f(t)
=
cos t
x0 + sin f
v0
path of the particle is an ellipse in the plane determined by If x0 , v 0 are orthogonal the major and minor the vectors x0 , v0 axis have lengths |x0|, \v0\ (see Figure 4.26). The velocity vector is Thus the
.
/'(t)
=
and the
-sin
f
speed
x0 +
is the
cos t
v0
length of this
vector.
X"
Figure 4.26
346
4
Curves
Suppose we have a force field in the plane which is of the same magnitude as the position vector, but orthogonal to it. Using complex variables on the plane, the force field is given by 31
.
F(z) Let
iz
=
iz
or
it is the former.
us assume
z0 and
velocity
v0
f'V)
i/(0
/(0)
=
Then,
.
=
Suppose
a
particle has initial position by solving
the motion is found
f(P)
zo
=
v0
The solution is of the form
f(t)
=
Aea' + Be-'"
where
a
y/i
=
=
(1
i)/~j2.
+
We solve for
A,
B
by substituting
the
initial conditions,
/(0)
=
f'(0)
z0
=
A + B
=
=
v0
a(A
B)
-
Thus
f{t)
Suppose f(t) For
=
z0
=
=
1
2-(e*<
large positive z
=
,
=
i'0
+
e< +
0.
z +
fttP
e-
Then
e-<)
t, the second term is
negligible,
and the
curve
is very close to
fe"'
which
we know is an outgoing counterclockwise spiral. For large negative t, the second term e~"' is dominant and that gives an incoming clockwise spiral. Thus the particle comes spiraling in from outer space and then at
time came.
t
=
0 pauses for
a
breath and then goes
(See Figure 4.27.)
racing
back from whence it
4.2
Arc
Length
347
Figure 4.27
EXERCISES 8. Find
arc
length
as a
function of the parameter for each of the following
curves.
(a) (b) (c)
r(l + r
=
a cos
1 + 2
The
6)
cos
curves
=
1
9
in Exercises
9. Parametrize these
curves
and 7(a). length, and find the curvature
6(a)(b)(d)(f ),
according to
arc
and normal.
(a) (b) (c)
x2 +
y2
=
1 x2 + z2
curves
The
curve
10. Find
the
1
=
,
The
in Exercise
in Exercise
normal
and
.
8(a)(b).
(d) The curve of Example 22. (e) The curve of Example 23.
6(a)(e).
tangential accelerations for these planar
motions:
(a) (b) 1 1
Find the
.
space
zit) x(f)
i)t (c) z(t) (1 + 2 cos t)e" exp(l t + e1' f^,y(f)=f3 (d) zit) normal and tangential accelerations of these
=
=
-
=
=
:
(a) (b)
x(f) x(t)
=
=
(f, sin f, sin t) (e', e~',t2)
(c)
x(f)
=
f(sin t,
cos
f,
1)
motions in
348
4
Curves
PROBLEMS 4. The
graph of a differentiable function
Find the curvature 5. The
as a
function of
y
=/(x)
is
a curve
in the plane.
x.
graph of a differentiable Revalued function y =/(x),
z
=
gix) is
a
in space. Find its curvature as a function of x. 6. A skier has to negotiate a series of hills whose
curve
profile is the curve (Figure 4.28). There are three forces acting on the skier: that due to gravity, the restraining force of the hills, and a force due to friction which is proportional to his velocity. Find the differential equation describing his motion. 7. I shot an arrow into the sky at an initial velocity of 80 feet/second and at an angle of tt/3 with the horizontal. The gravitational field is vertical downward with a magnitude of 32 feet/second2 The air drags the arrow with a force of 0.05 times its velocity. Find the equation of motion, and the curvature of the curve of motion (the arrow weighs one pound). 8. In Example 26, let k be the curvature of the slide. Show that the magnitude of the constraining force due to the slide is/= ids/dt)2K
y
=
e'x
cos x
=
=
=
Figure 4.28
4.3
Local Geometry
349
of Curves
Figure 4.29 10.
Suppose
l)2 + z2 l, by rotating the curve (x (The surface is, in cylindrical coordinates, 1, Figure 4.29). A cyclist cycling around the track tends to
a race
track is formed
1
(r
z'
l)2 +
=
z
=
axis.
ride up the bank as he goes faster. Explain that. 11. Water is at rest in a very large sink when a stopper is removed in the
bottom center of the sink.
An idealization of the
The water accelerates toward the hole.
follows.
particle of
water
are
due to
gravity and the
ensuing motion is as acting on each
The forces
mass
of the fluid itself.
The
field due to the former is (0, 0, g) and the field due to the latter operates Find as if the particle were on an inclined plane with vertex at the hole. the resultant force field. Find the differential
equation giving the
rate of
rotation around the hole. a ball of unit mass over a hill whose profile is the curve l. What minimum initial speed is x 1 to x from x2) y=exp( required to ensure that the ball maneuvers this hill? 13. Suppose we are given in space a force field which is directed toward Find the origin and so that its component in the z direction is always 1. 0 at the point the path of motion of a particle which is at rest at time t (1,1,1).
12. We must send
=
=
=
4.3
Local Geometry of Curves
physical problems discussed, that the higher-order parametrizing a curve have some significance. In this section we will discuss the higher-order invariants of a curve ; that is, those concepts which depend only on the geometry and not on the particular We have seen, from the
derivatives of
a
parametrization.
function
350
4
Curves
Let r be a curve in R". For purposes of simplicity, we shall take T to be parametrized by arc length by x x{s). If T is twice differentiable, the tangent vector TOO 1 for x'00 is a differentiable function. Since (T(s), (Ts)> all s, we obtain through differentiation 2(T(s), T'(s)} 0. Thus at any point T' is orthogonal to T. =
=
=
=
Definition 4.
The normal line to T at x0 x(s0) is the line through x0 parallel to the vector T'(j0). The osculating (or tangent) plane to T at x0 is the plane spanned by, the tangent and normal lines. The name osculating plane is quite descriptive. This plane osculates in the following sense. =
and
Proposition 4. Let x0,xx, x2 be three points on the curve Y. If they are noncollinear, they determine a plane. This plane has the osculating plane as limiting position, as xx, x2 tend to x0 .
In order to determine the
Proof.
it suffices to find two
independent
limiting position of the plane through
vectors which
are
limits of vectors
on
x0
,
Xi, x2
the variable
plane. The easiest way to do this is to refer to the Taylor expansion of the arc length parametrization. Suppose/: (a, b)^T parametrizes T with respect to arc For simplicity we may assume x0 is the origin, 0. x0 length, and/(0) According =
.
to Theorem 4. 1
fis) where lim
=
eis)
we can
write
W)s + T'i0)s2 + eis)s2 =
(4.38)
0.
5-.0
Let x0
is
,
Xl=f(si), X2=fis2). Since x0=/(0)=0, the plane tt(si,s2) through plane spanned by the vectors /OO, /(s2). Now, for each su si TX?,). -rrisi,s2). Now
xi, x2 is the
on
Ji-1/(*i) Letting
st
=
7X0) + T\0)si + eis)si2
-^0, that
says that lim *r
V00
7"(0) is
=
on
the
limiting plane. Now,
to
find another vector on the limiting plane, we take an appropriate combination of f(si),f(s2) so as to dispose of the T(0)s term in the Taylor expansion (4.38). Thus, we consider
s2
We
are
tend to
f(si)
-
sif(s2)
interested in zero.
7"(0)
Let
T'(0)is2 si2
finding
us
2s3 +
=
some
take the
(25)
4s3
Sis22) + eisi)s2 Si2
-
eis2)sis22
vector of this form which has
special -
-
case Si
Eis) 2s3
=
=
2s2
=
2s3iT'iO)
a
limit
2s; (4.39) becomes +
2ei2s)
-
sis))
(4.39) as
su s2
Local Geometry
4.3 Thus
T'iO) + 2ei2s that T'iO) is by 7X0) and 7"(0).
-
we see
A few remarks defined. vector is
line has
In
eis)) the
on
are
351
is on the plane spanned by fis) and /(2s). Letting s - 0, limiting plane. Thus the limiting plane is indeed spanned
in order.
particular,
of Curves
If
if the
0, then the osculating plane is not a straight line, then the tangent plane which is closest to Y, so a straight
T'(0)
constant, and there is
no
=
T is
curve
osculating plane anywhere. Conversely, if T and T' are always collinear along Y, then T must be a straight line (Problem 14). Now, in the case where T' and T are collinear at the point in question, but not always collinear, it may happen that the plane through x0, x^ x2 of Proposition 3 has a limiting position as xx, x2 x0 and it may not (see Problem 14). In the former case we shall consider the normal plane as defined by the limiting position, and in the latter case, we shall say that the normal plane does not exist. Generally speaking, such cases are pathological, and we shall exclude no
-*
,
them from further discussion. Observe that for For N
on
Y in
R2,
the
osculating plane
Then the normal vector N varies
will form
and the vectors
R2 along the no
in
is
(of course) just R2.
we
(see Figure 4.30). curve
curves
define the normal vector to Y at x0 as that unit vector R2, the normal line so that the sense of rotation T -> N is counterclockwise
curves
uniquely
curve
Definition 5.
save
Let Y be
normal vector to Y is
natural
(T, N) (called the moving frame).
a
"
continuously along
the
orthonormal basis for
In R" for
n >
2 there is
normal vector, and thus we leave the that it should vary continuously along Y.
determined choice for
choice undetermined
"
a
a
a
twice differentiable oriented
choice of unit vector
Figure 4.30
on
curve
in R".
The
the normal line which varies
352
4
Curves
continuously along Y. The curvature such that T'(s) k(s)N(s) along Y.
of Y is the scalar function of s,
k(s),
=
Examples 32. The circle in R2
x(s)
a cos
=
\
,
cos
-
\
,
orthogonal
/
I
=
sin
-
,
=
z(0)
z'(0)
=
=
(l
eVa
+
-
so
)
[cos(s/a), sin(s/a)1/a
33. The spiral r parametrization is =
and counterclockwise from T
a)
a
T'(s)
to
s\
s
cos
\
the circle of radius
z
-
af
a
The normal is
Then
s\
s
sin
=
N(s)
a)
a
I
T(s)
-I
a sin
-
(Figure 4.31)
=
a
=
=
N(s)/a,
so
the curvature
of
is a"1.
ee (in polar coordinates) (Figure 4.32).
e(1 + i)(,
Oe(1+i)e
Figure 4.31
The
4.3
Local
Geometry of Curves
353
T()
t/4
N()
Figure 4.32 SO
ee
ds
Te
-*m-Tl
Thus the tangent vector is 1 + i
T(9)-
e>8
The normal is
dl
=
dldl d0 ds
ds
=
_
gi( + */4)
N(0)
=
iew+lvJ2e-e
Thus the curvature is Here is
plane
a
e'<9+ 3*/4>.
proposition
=
Now,
J2e~ V<9+3*':/4)
given by k(9)
which
gives
an
=
yJ2e~e.
interpretation of curvature in the easily computable. It says that
and sometimes makes the curvature
354
Curves
4
the curvature is the rate of rotation of the
moving
frame with respect to
arc
length. Proposition 5. Let Y be a given plane curve. The curvature ofY of rotation of the tangent with respect to arc length, that is,
k(s)
=
js
r(s)e"(,) in polar coordinates. T(s) ?'<"< +*'2>, and Then NO)
JPT1
Since T
'
is
a
unit vector,
J
_(gI(J)\
=
iQ'eW
_
ds
ds
k(s)
=
=
=
Thus
(arg TOO)
Let
Proof. 1. r(s)
is the rate
__
Q'eH6)+nl2
0'(s).
=
Examples 34. The helix
f(0
=
has
(a
=
t,
length
arc
TO)
cos
(a2
+
sin t,
a
s
=
(a2
bt) +
b2)ll2t,
and tangent vector
b2y1/2(-a sin(a2
+
b2)~1/2s,
a
cos(a2
b2)~1(-a cos(a2
+
b2y1/2s,
-a
+
b2y1/2s, b)
Thus
T(s)
(a2
=
Thus N
=
+
(-cos
t, sin t,
0)
sin(a2
+
b2y1/2s, 0)
and
a k
=
a' + b2 Observe that the normal line to the helix
axis of the helix. 35. Consider the
x(0
=
(cos
curve
t, sin t, sin
30
(Figure 4.33)
always points
toward the
4.3
Local
Geometry of Curves
355
Figure 4.33 Then
x'O)
(-sin t,
=
cos
t,
ds
||x'(OI|
=
-
T(f)
=
(1
=
+ 9
(1
+ 9
cos2
cos
3f)
cos2 3f)1/2
30"1/2(-sin
t,
cos
f,
cos
3f)
Computing
dT_dTaj_ ds
dt ds =
-(1
+ 9
cos2 3f)"2(10
and the curvature is the
cos
t, sin 9f,3
length of this
cos
3f
cos
2f + sin 3f
sin2f)
vector.
Now, let us make one final remark about a curve in the plane. It is completely determined, up to Euclidean motions, by its curvature. Thus, for example, the only curve of constant curvature is a circle. This is, as we shall see,
an
easy consequence of Picard's existence theorem for differential
equations. Theorem 4.1.
Let
k(s)
be
a
continuous function
the origin. There is a curve Y whose another curve parametrized by arc length curvature, then
a
curvature on
ofs in some interval I about function is k(s). If Y' is
the interval I which has the
Euclidean motion will move Y'
onto
Y.
same
356
4
Curves First
Proof. curvature.
Euclidean
shall
verify the uniqueness. Let r be a curve with the given x(s) be its arc length parametrization. We may apply a motion (translation and rotation) so that x(0) is the origin and T(0) is we
Let
x
the vector Ei.
=
Now
we
show there is only
one curve
with these
The
properties.
proof depends on the observation that the normal is rigidly attached to the tangent; that is, its motion along the curve is completely determined by the tangent. In /e'e<9+"/2) fact, writing T(s)=ems\ we have N(i) =e'<8(s+"'2). Thus N' 0V<9+") T. Now the system of differential equations =
=
=
T(s) has T is
k(s)N(s)
=
N(s)
only one solution subject unique, so
x(s)
f
=
Jo
(4.40)
-k(s)T(s)
=
to the initial conditions
T(0)
=
Ei, N(0)
=
E2
Thus
.
da
T(
uniquely determined by the given conditions. Thus there is only one r given curvature. We now turn to the question of the existence of a plane curve with given curva ture. Again, by the fundamental theorem on differential equations, there exists a solution of the system (4.40) subject to the initial conditions T(0) E2 Ei, N(0) If (T(s), N(s)) is the solution, then
is also
with the
=
=
.
x
defines
=
x(s)
=
'o
T(ct) da
plane curve r. k(s)N(s), so k(s) show that x'(s) T(s) is x"(s)
We must show that
a
=
=
Now, let f (s)
f (0)
=
f '(s) N'(s)
=
=
-
=
-
iE2
=
=
N(0)
Ei
=
=
is
arc
-
=
iT(s). 1E1
-K-k(s)T(s))
=
=
-
as
so
T(s) has
constant
we
must
E2
kCs)N(s)
i/c(i)N(j) =-k(s)T(s) By the uniqueness, T
=
=
For then
Then
Thus T, N also solve the given initial value problem. N. Thus N /T, so N T. It follows that
N
length along r. 5 is arc length
unit vector.
iN(s), N(s)
-iN'(j) f'T'Cr)
a
j
To show that
is the curvature.
=
2
length.
=
2k(sKT(s), N(s)>
Since T(0) =Ei, it is
a
=
0
unit vector.
=
T,
4.3
Local Geometry
357
of Curves
PROBLEMS 14. Show that if T :
collinear, 15. (a) x
=
x x(s) is a curve in R3 and T(s), T(s) straight line. be given by
then T is Let r
=
T(0), T'(0) origin. (b) Let
\x3
.
*W
=
=
collinear, but
are
T has
an
osculating plane
at the
ifx<0
(o
ifx>0
Show that the
x
everywhere
(x, x3, x3)
Show that
,
are
a
curve
V
given by
(x, -g(x),g(-x))
does
have
osculating plane at the origin. the sphere x2 + y2 + z2 1. Show that r is an arc of a great (i.e., diametric) circle if and only if the normal to T is always collinear with the position vector. 17. Show that a curve is a straight line if all its tangents are parallel. 18. Three noncollinear points in R2 determine a circle. If, for the purposes of this exercise, we consider a straight line as a circle (of infinite radius) we may assert that any three points determine a circle. Suppose T is a curve in R through p0 Following the kind of reasoning on pages 324 and 325, define the osculating circle to T at p0 and find its equation in terms not
16. Let r be
an
a curve on
=
2
.
of
a
parametrization
of r.
19. The radius of the
osculating circle is called the radius of
curvature.
Show that it is k~\ 20. If the a
straight
osculating circle
to F is
always
a
straight line, deduce that
T is
line.
osculating circle at a general point of an ellipse. osculating circle at a general point of a parabola. that if the osculating circle to a curve is always a circle of
21. Find the 22. Find the 23. Show
radius R, the
curve
is
a
circle of radius R.
given parametrically by Suppose Show that the curvature is given by T is
24.
k
=
x'y"
-
y'x"
=
25. Show that
arc
length by
x
=
x(s),
y
=
[{x")2 + (y")2]"2
a curve
in the
plane of
constant curvature is
a
circle.
y(s).
358
4
Curves
Figure 4.34
26.
V:
Suppose
x=t(s) is
a curve
distance between f (s) and f (s + circle. 27.
Suppose f is
t)
is
with this property: for every /, the s. Show that T is a
independent of
nonnegative function of a real variable with the graph of /between 0 and x is proportional to the arc length of that graph. Find the curve. 28. Find the curve V with the property that at any point p the angle between the tangent to r at p and the tangent to the ellipse property that the
E: x2 +
2y2
=
a
area
under the
l
at the
point of intersection of E with the ray through p is constant. x t(s) be a planar curve. Suppose we have a string along T with one end point at x0 If we unwind the string tautly and without stretching, the end point will follow a curve E, called an evolute of T (Figure 4.34). If 5 measures arc length from x f(0), the curve E is f (s) + sf'(s). Find the evolutes to (a) the unit circle parametrized by x ell + n', (c) the parabola y (b) the spiral z x2, (d) an ellipse. 30. If we rotate a cylinder of water about its axis, the surface of the water does not remain a plane. What shape does it take and why ? 29. Let T:
=
.
=
=
=
=
4.4
4.4
Curves in
Curves in
359
Space
Space
T is a curve in space. Let x0 e I\ and suppose T and N are the and normal T at A third unit vector orthogonal to both T to tangent x0 and N will serve to provide a natural frame within which to discuss the
Suppose
.
behavior of the is chosen
curve
near
x0
that the
.
T
This vector B, called the binormal to the -> N -> B forms a right-handed frame (see
triple Figure 4.35). In this section we trihedron along the curve, much as plane curves. curve
so
shall we
use
this frame, called the moving
used the tangent and normal to
study
The three vectors T, N, B determine three planes : the tangent {or osculating) plane is spanned by T and N, the normal plane is spanned by N and B, and the
plane spanned by of the
T and B is called the
Now the
rectifying plane.
cur
have seen, the rate of rotation, with respect to arc length, of the tangent line in the osculating plane. In three dimensions there is another important intrinsic function on the curve. Since B is a vature
unit vector
=
0,
curve
on we
Since T"
=
is,
as we
Y,
0.
=
Thus B' lies in the
osculating plane.
Since
have
+
kN,
=
=
0
k
=
0, thus also
be collinear with N.
Figure 4.35
=
0
so
B' must
360
4
Curves
Definition 6.
B'
=
The torsion
of
i
a
curve
T is that function such that
tN.
-
The torsion measures the torque, that is, the twisting of the osculating plane about the tangent line. That is, since the binormal is orthogonal to the osculating plane, the change in the binormal reflects adequately the change in the osculating plane. The Taylor development of the binormal in a neighborhood of a point x0 x(0) is =
B(s)
=
B(0)
t(0)N(0>
-
+
b(s)
(considering only first-order terms) the binormal at x(s) has moved t(0) s toward the normal. Thus if t(0) > 0, the osculating plane has twisted in the right-handed sense about the tangent line. At a point where t 0, the osculating plane pauses ; it may or may not change its direction of rotation about the tangent line. If x 0, the osculating plane remains fixed along the curve; it follows that the curve lies on this plane. Thus
=
=
t
Let F be
Proposition 6. 0 along T.
a curve
in
R3.
T is
a
plane
curve
if and only if
=
If T is
Proof.
a plane curve, let n be the plane containing T. The tangent always lie on n, so the binormal is always the unit vector orthogonal
and normal to T to n.
Thus the binormal is constant,
On the other
hand,
r
=
0.
so
B'
Let
=
0, thus
t
=
0.
x(s) be the parametrization of T by arc length. Since t 0, B' 0, so B is constant along T. If for some s0 x(so)is not on the plane through x(0) and orthogonal to B, then suppose
x
=
=
=
,
<xC$o)-x(0),B>^0 Let
6(s)
since B
6(0)
=
be the function =
B(j)
0, 6(s0)
x(0), B>. Then 6'(s)
for all =
(4.41)
s
<x(s)
-
=
and is
The fundamental formulas of space curve theory We can now easily derive them.
N', B' with T, N, B. Theorem 4.2. T'=
(Frenet-Serret Formula) kN
N'=-kT B'
=
+ tB -
tN
are
those
relating 1",
Curves in
4.4 The first and the third
Proof. N is N
=
unit vector,
a
<xT +
=
a
p
,8b; 0.
=
=
we
are
0,
=
=
just the definitions of k, t, respectively. so N' lies in the rectifying plane.
j8
-k,
=
But that follows from
t.
361
Space
Since
Write
=
0,
For
=
=
-
-
=
=
-k
-(-T)
=
t
Examples 36. The circular helix:
x(t)
(a
=
cos
We have
T(,)
t,
sin t,
a
bt)
already computed
that
s
=
ct, where
c
=
(a2
+
b2)112,
and
l(-asinQ,flCos(^)
=
kN(s)
=
-5
I
a
cosl-J,
a
sinl-l,
01
thus
K
B
=
^N=-(cosQ,sing),o)
=
T
_TN
thus
=
t
N
x
B'
=
=
=
-
I
-tsinl-J, b cosl-J,
-a
J
iI(-bcosQ,^sing),0)
b/c2.
37. Let C be a curve in the xy plane, and let T be a curve of constant slope lying over the curve C (see Figure 4.36). Thus if T is the tangent Let b be that constant. Then T has the to T,
parametrization x(0
=
where
WO, y(t), 6(0)
(x(t), y(t)) parametrizes
C.
We may
assume
the parameter
362
4
Curves
Figure is
length along
arc
x'
=
C.
4.36
Then
(x',y',b)
so
ds =
-
Thus
T
||x'|| s
=
=
(1
x')2 +
+
{y'f
b2)1/2t
=
(1
+
^1,2
+
b2)1'2
and the
=
(1
tangent
+
b2)1'2
to T is
(*'. /.
Thus
kN
=
T'
=
(TTW7l(^/'.0)
Now if kc is the curvature of C, since ( /, x') is its normal vector, so
(x\y")=Kc(-y',x')
(*', /)
is its tangent vector,
4.4
Curves in
363
Space
Thus
Kq
kN=
{1
+
b2yl2(-y',x',o)
so
K
N
{1+Cb2)m
=
=
(-/,x',0)
Then
B
=
T
N
x
(1
+
^2)i/2 i-bx',-
(l
+
b2)i/2(-bx',-by',l)
=
=
by', (x')2
+
(y'f)
Differentiating, 1 ~TN
B'
=
fojc
=
(i
+
b2Y'2
{-bx"> -by"> 0)
=
(i
+
b4>2 {~y'' x'' 0)
Thus
bKc
(1
+
b2)1'2
Local Behavior
of a
Curve
now make a close study of the local behavior of a curve relative moving trihedron. Let T be a sufficiently differentiate curve, par a < s < a. We may perform a ametrized by arc length by x x(s), Euclidean transformation so that x(0) 0, T(0) E^ N(0) E2 B(0) E3 Expanding \(s) in a Taylor series, we obtain
We shall
to the
=
=
=
=
=
,
s2
x(s)
=
x(0)
+
x'(0)s
+
x"(0)
-
2
s3 +
x'"(0)
-
+
6
e(s3)
(4.42)
Now
x'
=
T, x"
=
kN, x*
=
jc'N + kN'
=
k'N +
k(-kT
.
+
tB)
364
4
Curves
Evaluating these x(s) In
=
at
sEi
zero
+
and
substituting into (4.42),
s2E2
we
+ E2 Ei +. 2ooo -
-
-
obtain
E3
+
e(s3)
coordinates,
x
=
s(s3)
+
s
6
y
z
=
'iL s2
+ t
s3
=
6
s3
+
o
+
(53)
e(53)
Thus for small values of s, the the equations
given
looks like the cubic
curve
curve
given
by
KX
y
Figure
=
z
r
4.37 is
a
=
picture
,
x3
of this
-
zz
4 3
curve
for
X
,
'3
K
jc>
# 0 the curve always passes through its but lies on one side of its rectifying plane. as kx
Figure 4.37
Notice that, so long osculating and normal planes,
0,
x >
0.
4.5.
Varying
a
Curve in the Plane
365
Now, just as the curvature determines plane curves up to a Euclidean motion, space curves are so determined by the curvature and torsion. The
proof of this fact is by the same kind of application of Picard's existence and uniqueness theorem as we used in the case of the plane. We shall leave the verification to the interested reader.
Theorem 4.3 is
a
curve
space
interval
of
Given continuous functions f, g defined in an interval I there Y: x \(s) given parametrically by arc length in some sub=
I such that
K(s)=f(s) Y is
x(s)
=
g(s)
up to Euclidean motions in
unique
R3.
PROBLEMS 31. Show that
through
a
a curve
in R3 is
plane
a
curve
if all its tangent
planes
pass
given point.
32. Show that 33. Let T be
a curve
a curve
in R3 is
a
plane
curve
if its binormal is constant.
in the plane and let y be the intersection of the cylinder
T with the cone x2 + y2 Find the curvature of T in terms z, z > 0. of that of y. What is the torsion of y? 34. Let T be a curve in space, and y its projection onto the xy plane. What is the relation between the curvature and torsion of T and the curva =
over
ture of
y?
Suppose that T is the intersection of the surface z y2 in R3, with the plane ax + by 0. What is the curvature of r at the origin ? x2 + 2y2 with the plane 36. Let T be the intersection of the surface z 35.
=
=
=
ax
+
by
=
0.
What
37. Let T be
are
given
the curvature and torsion of T ?
in R3
by
x
=
Show that
the tangent lines to T. gonal to those tangent lines is
x
4.5
=
x(s) + (c
Varying
a
s)T(s)
for
x(s).
Let S be the surface swept out
a curve on
which is
by everywhere ortho
given by some
constant
c
Curve in the Plane
A family of curves in the plane is a collection of curves {Yc}, as c range through some set, usually of n-tuples of numbers. It is to be understood that the curves of the family vary smoothly ; although we shall not make this idea precise. For example, if x(t, c) are functions defined for real t and c lying
366
4
Curves
in
set
S, then the equations
some
x
value of More
(4.43)
y(t, c)
=
y
each curve in the family is found by fixing a Equations (4.43) as the explicit form of the family. often, a curve is determined by a relation between x, y and a family be given by an equation
determine
could
x(t, c),
=
a
of
family
curves:
We refer to
c.
F(x, y,c)
=
0
(4.44)
which, for fixed c gives the relation determining a curve. We refer to (4.44) Since it does not refer to any particular as the implicit form of the family. of the individual members, this form is particularly useful. parametrization The "constant" c which picks out the member of the family usually ranges through some set in R" : in which case we refer to the family ((4.43), (4.44)) as an n-parameter family of curves.
Examples 38. A
ax
+
by
straight line in +
c
the
plane
is
given by
the
equation
0
=
(4.45)
Thus the set of all
straight lines is given by (4.45) implicitly as a 3-parameter family of curves. If instead, we write down the slopeintercept form of a straight line, y
mx
=
then
we
+ b
(4.46)
exhibit this
family explicitly
as a
2-parameter family
of curves.
and consider the
family of (x(t),
39. Let
x
x(t)
=
be the
y
equation
tangent lines
y(t))
y
=
=
y(t) of
a curve
to Y.
The
Y in the
equation
plane,
of the line tangent to Y at
is
^
+
y'(t)
x\t)(X
~
X(0)
(4"47)
4.5 This is the
Varying
a
Curve in the Plane
explicit form then of a 1 -parameter family.
367
(The parameter
is*.) 40. Consider the
x
=
The
cos t
=
y
family
case
sin
where Y is the circle
t
of tangent lines to Y is
given by
the
equation
cos t
y
=
t sin t
This
y
sin
simplifies x
=
(x
cos
0
(4.48)
to
cot t +
esc
t
can make this appear even more palatable by the parameter of the family. Letting c cot t so becomes + -(1 c2)1/2/c, (4.48)
We as
=
,
XC
a
-
(i
+
1 -parameter
c2yi2
family
taking we
find
cot t esc t
=
(4.49) of lines.
41. Suppose a hoop is rolling along a horizontal line (see Figure 4.38). This collection of positions of the hoop forms a 1-parameter family of circles where the point of tangency with the horizontal (the
Figure
4.38
4
Curves
Figure
axis) is taken family is thus x
(x-c)2
+
42. The
is
a
to be the
(y-l)2 family
1 -parameter
4.39
parameter.
The
of circles tangent to both the x axis and the y axis family of curves (Figure 4.39). We take for the
parameter the point of tangency of the
-
It is
c)2
+
easily
(y- r)2 seen
considerations.
for the
l
=
is the radius of the cth circle, then the
(x
implicit equation
=
curve
equation
with the of the
x
axis.
family
is
If
r
clearly
r2
that
r
Thus
=
c; this follows from
the
elementary geometric family is implicitly described by this
equation (x
-
c)2
+
(y- c)2
=
c2
(4.50)
43. The family of circles of radius 1 tangent to the parabola y x2 (Figure 4.40). We can take as the parameter the x coordinate c of the points of tangency. The center of the circle is on the line per pendicular to the parabola at (c, c2). Thus if (r, s) are the coordinates =
4.5 of the center of the cth
s-c2
(r
These
+
(s- c2)2
equations
=
C
Curve in the Plane
have
we
have the solution
+
_
S~C
Thus the
1
2
(l+4c2)1/2
(1+4C2)1'2
implicit equation
for this
family
of circles is
(x-c-(TT^)2 (^c2 (JTi?F)2 +
44. Let T be to
T.
369
1
2c T
a
-h(r-c)
=
c)2
-
circle,
Varying
If T is
a curve
given
+
in the
as a
We seek the
plane.
function of
1
=
arc
family of tangents length by x \(s), then the =
lines
g(w)
=
x(s)
form the
+
uT(s)
family
(4.51)
of tangents to Y with
s as
Figure 4.40
Suppose now family at every point. particular tangent line
parameter.
that y is a curve which is orthogonal to this If h(s) is the point of intersection of y with the
370
4
Curves
x(s), then y is parametrized by x h(s). h(s) is then of the form (4.51) with a particular choice u(s) of u. Writing then h(s) x(s) + u(s)T(s), and differentiating, we obtain
(4.51)
at
=
=
h'(s)
(1
=
Since
h'(s)
is tangent to y and thus, 1 u'(s) 0. Thus
=
x(s)
+
family given by
=
eis
These
The
+
are
to the
=
(s+ c)T(j)
The
x
by assumption, orthogonal to T, u(s) s + c. So the family of tangent lines to Y is given by
=
orthogonal
curves
is
u(s)T(s)
+
must have
we
x
u'(s))T(s)
-
i(s
of
+
just
(4.52)
curves
c)eis
=
orthogonal
[1
+
i(s
+
to
the tangents to the circle
z
=
e"
c)]e's
the evolutes of the circle.
Differential Equation of a Family
A differential
V'
=
determines
equation
Fix, y) a
1 -parameter
family
of curves, if the function Fis decent enough. c there is a unique solution of the initial
For, under such conditions, for each value
problem y'
=
F(x, y)
The solution
y(x0)
=
c
be written y =f(x, c), which can be considered as either explicit, implicit form of the family. Now, it is usually true that a 1 -parameter family of curves is the family of solutions of some differential equation, and we would often like to find that differential equation. the
can
or
Suppose, for example, that y f(x, c) is the equation of a given 1-parameter family. If y y(x) is one particular curve (i.e., y(x) =/(*, c0) for some fixed c0), then these two equations must hold =
=
y
=
f(x,c)
y'
=
dJ-(x,c)
4.5 for
value of
Varying
a
Curve in the Plane
371
(i.e., c c0). It may be possible to eliminate the param equations, thus obtaining a relation between x, y, y' which must be satisfied; this is the differential equation of the family. For it is a differential equation which must be valid for each member of the family, and this is a differential equation which determines the family. More generally, suppose the family is given implicitly by some
eter
c
F(x, y,c) If
x
is
a c
=
c
=
from these two
x(t),
y
0
=
y(t) parametrizes
=
of the
one
curves
in the
family,
then there
such that
F(x(t), y(t), c) identically in
=
(4.53)
Differentiating
t.
now
with respect to t,
we
have
dF
dF ox
0
(x,
c)x'
y,
+
oy
(x,
y,
c)y'
=
0
(4.54)
we can eliminate c from Equations (4.53) and (4.54), the result will be a relation between x, y, x', y' which must be satisfied for each curve in the family and thus is the differential equation of the family. Of course, if x is the parameter along the curve, and y y(x) is its equation, (4.54) becomes
If
=
8l(x,y,c) + d-f(x,y,c)^ dx oy
=
0
(4.55)
ox
Examples 45. Consider the
y2
-
ex
=
-
c
=
with respect to
-
parabolas (Figure 4.41)
lyy'x
x
(considering
0
Thus the differential
y2
of
0
Differentiating
2yy'
family
=
0
equation
of the
family
is
y
as a
function of
x),
372
4
Curves
Figure 4.41 or,
y
excluding 2y'x
-
=
the
curve
family y (as already know). equation =
cex is
we
=
exp(-
0,
=
0
46. The
y
y
x
The
given by family y
the differential =
ecx is
equation y' y given by the differential =
j
(Clairaut's Equation). Let y f(x) give a curve in the plane, and consider the family of lines tangent to that curve. That family is given implicitly (taking the x coordinate of the point of tangency as the parameter) by this equation, 47.
y=f(x)
=
+
f'(c)(x-c)
Now, upon differentiation
/ =/'(c)
(4.56) we
find
(4.57)
4.5 To say that
(4.57)
y'. Then, equation y
=
we can
amounts to
y'x
where
upon
eliminate that
saying eliminating
Varying c
from the
we can
we
a
Curve in the Plane
373
pair of Equations (4.56) and (4.57) for c as a function of as differential equation, the
solve
obtain
h{y')
+
(4.58) the
h(y') represents y'.
considered
expression f(c) f'(c)c,
as
a
function of
Thus
Equation (4.58), known
of the differential solutions
y
equation
of
as a
Clairauts'
family
equation,
is the
form
general
of lines tangent to
a
curve.
Its
are
=
+
ex
h(c)
Notice that the
given curve y=/(x) also solves Equation (4.58) (because (4.56) and (4.57) which hold under the substitution y f(x)). singular solution of the equation.
it is derived from It is called the
48. The
implicit y
=
c2
=
family
y
+
2c(x
c)
-
y
=
x2 has the
we
=
lex
obtain
-
c2
y' 2c. Thus c \y'', so equation of the family, =
=
we can
eliminate
to obtain this differential
=
/x-Ky')2
49. The
family implicitly by
xc
y
Then
y'
=
yx
of lines tangent to the circle x2 +
(1-c2)1'2
=
c, so the Clairaut
(1 y
parabola
form
Differentiating, c
of lines tangent to the
-
(/fy2 r
equation
of the
family
is
y2
=
1 is
given
374
4
Curves
Family Orthogonal
to a Given
Family
given family of curves. We propose to find a family G of curves everywhere orthogonal to F. Thus, if p is a point in the plane, and Y is the curve in F through p with tangent T1; and y is the 0. curve in G through p with tangent T2 we must have
a
=
a(x, y)x'
+
b(x, y)y'
=
(4.59)
0
Thus, since (x', y') is the tangent field with
to
(a(x,y), b(x,y)) (for
differential *'
equation
for the
family
=
we
0
must have
by (4.59)).
T2 collinear Thus
the
G is
y'
(4.60)
=
a(x, y)
F,
b(x, y)
Figure 4.42
4.5 51. Find the
xy
=
Varying
family orthogonal
a
Curve in the Plane
to the
family
of
family
is
+
375
hyperbolas
c
The differential
equation
of this
yx'
xy'
=
0.
Thus the
differential equation of the orthogonal family is x'
y'
y
x
xx'
or
is
yy'
=
0 which
integrates
to
x2
y2
-
52. The family orthogonal to the family given by the differential equation
1
V'
y
-2x
(here
x
is the parameter,
2,y2 +
*
=
y
2xx' +
=
This
1).
of parabolas in
integrates
Example
=
45
to
(4.61) family
which makes
The differential
yy'
x'
c.
c
53. Find the
(4.61).
so
=
equation
an
of the
angle of n/4 with family (4.61) is
the
family
0
The
family orthogonal to this family has tangent collinear with 2x + iy, family we seek has tangent collinear with this vector rotated by n/4. Thus the tangent field is collinear with ei("/4)(2x + iy), or, what is the same, (1 + i)(2x + iy) 2x y + i(2x + y). Thus, the differential equation is thus the
=
2x
y
2x + y
Envelopes we have been studying have the property that there (or curves) which is not a member of the family but bounds the family (see Figures 4.39-4.41). Similarly, for a family of lines tangent to a
Many of the families
is
a curve
376
4
Curves
given curve, an envelope.
the
First of all, x
=
c, y
exists
=
curve
bounds the
We want to
c, y
we can
find
a
bounding
curve
is called
for families.
families do not admit
some =
Such
family.
how to find
envelopes envelopes. Clearly, x2 + c do not admit envelopes. However, if it by the present techniques. see
the families an
envelope
Definition 7. Let F be a family of curves in the plane. A curve Y is an envelope for the family F, if through every point p in r there goes a curve in F which is
tangent
Suppose
that
F(x, y,c)
a
to Y at p.
family
is
given implicitly by
0
=
and that the
curve Y: y =f(x) is an envelope of this family. Then, for every there is a x0 c(x0) such that the curve C corresponding to F(x, y, c(x0)) 0 is tangent to Y at (x, f(x0)). Thus we must have =
F(x0,f(x0), c(xo)) and since the
=
0
(4.62)
C has the tangent direction
curve
(l,/'(*o))>
we
must
have, by
(4.54), dF
dF
+ -fa (*o /(*o)> c(x0)) y (*o f(x0), c(x0))/'(x0) ,
,
Differentiating (4.62) dF
-dx
with respect to x0
8F +
we
=
0
(4.63)
also find
dF +
Comparing (4.63)
and
(4.64)
<4-64>
=
TyfiX) TcC'(X) we
have
as a
result
dF dc Thus if
(x0 f(x0), c(x0)) c'(x0)
(x, y)
,
is
on
the
=
evenlope Y,
0
(4.65)
there is
dF
F(x,y,c)
=
0 oc
(x,y,c)
=
0
a c
such that
4.5
and
we
eliminate
can
of Y.
equation
x
also hold
3F
^
=
a
Curve in the Plane
from this
c
Notice that from
F(x, y,c)
Varying
0
pair of equations to (4.64), the equations
obtain
an
377
implicit
dF +
_
dx
dy
/
=
0
Y. Eliminating c from equation of the family, so differential equation.
this
on
differential
pair we obtain once again envelope must also satisfy
the
the
this
Examples 54. Find the
(x of
-
c)2
+
(y
Example 2(x
c)
of
-
c)2
+
-2(x
c)
-
or
-
family
1
(4.66)
x
=
obtain
c we
42.
=
l)2
(y
of the
=
1,
c)
-
+ y
=
(-j)2
+
this in
(-*)2
=
=
2c
(4.67)
we
obtain
(*+>')2
or
2xy
=
0
Thus the
envelopes
y
=
2,
y
=
0.
(4.67)
c
Substituting
or
c2
or
x
to find
family
We must eliminate
2(y
c
c
envelopes
(y- c)2
Example
=
of the
We differentiate with respect to
0
55. Find the
(x
l)2
41.
=
Eliminating
-
envelopes
are x
=
0,
y
=
0.
c
from this
equation
and
4
378
Curves
56. Find the
y
=
x2 sin
of the
envelopes
ex
Differentiation with respect 0
=
ex2
family
to
c
yields
cos ex
or
n c
=
0,
cx
=
gives y 0 which fails as an envelope. x2 (Figure 4.43). envelopes y n/2, 3tt/2 yields
The condition ex
=
=
Xy'
This is
y
=
c
=
0
=
the
57. Find the
y
2>n
-,,...
ex
+
a
(1
+
envelope
of the
family given by
{y')2)
Clairaut
+ 1 +
=
equation
and has the solution
c2
Figure 4.43
But
4.5
Differentiation with respect to
0
=
x
+ 2c
or
c
=
Varying
a
379
Curve in the Plane
yields
c
-
2
Thus the
envelope
3x2 y
=
of this
family
is the
curve
1
-+l
EXERCISES
equations for these families of curves: l (c) xec" 0 (d) x sin y + c sin x 0 xy
12. Find the differential
(a) (b) (e) (f )
xyc
=
sin xy
l
=
a cos
=
=
1
yec('+ 1 sin(x + y+c) + cos(x + y + c) 13. Find the implicit form of the family given by these differential equations : l (c) (y')2 + ^ (a) xy'-yx'=0 l (d) y + y'x + siny=0 (b) x' + yy' \y (e) y'(sec x tan x) 14. Find the implicit form and the differential equation of the family of =
=
=
=
=
circles with center
on
the y axis and tangent to the
x
axis.
family of ellipses with foci at (-1, 0), (0, 1). Find the family of curves orthogonal to the family in
15. Find the 16.
Exercise
14;
Exercise 15. 17. Find the
family orthogonal
to the families of Exercises
12(a), (b),
(f), 13(b), (d). 18. Find the
family making
an
angle of tt/3 with the family of
Exercises
12(a), (b), (c). 19. Find the envelopes of the families of Exercises 20. Find the envelopes of these families: (c) 13e (a) y sin(x-c)2
12(a), (b), (c), (d), (e).
=
(d) family of cardiods sin ad. The family r 136
(b) (e) (f )
The
y r
=
=
e*sincx
(1
+
c)_1(l
+
c cos
6).
=
PROBLEMS 38. Find the
orthogonal
to
family of evolutes of the parabola
y
=
x2.
Find the
family
this family of evolutes.
cee. 39. Find the family orthogonal to the family of spirals r a tall feet building slips originally leaning against 40. A ladder 10 of curves which are the trajectories of the the Find family 4.44). =
(Figure points on
the ladder.
380
4
Curves
\\\\\\\\\\\\\\\\\\\\\\\\\V\\\VVVX\V
Figure 41
Find the
.
family
of
4.44
trajectories of the points
horizontal plane. 42. A line segment of length 2 has its
ball
rolling
on
the circumference of
a
on a
of
Find the
trajectories parabola (Figure 4.45). 43. A ball of unit
points
mass
x0
on
endpoints
on
the segment
is at the end of
a
the
as
parabola
y
=
x1
it slides along the
string of unit length attached
vertical bar rotating at constant angular velocity. Find the 0 of motion of the ball assuming its position and velocity at time t path Find the trajectory of any point on to be (1, 0, 0), (1, 0, 1), respectively. to the
top of
a
=
the
string.
44. Find the
length
with
family of curves swept endpoints along the curve
out xy
by the midpoints of bars of given 1 in the first quadrant.
=
Figure 4.45
4.6
Vector Fields and Fluid Flows
We have
come across
vector fields several times
now mass
of
a
study noninterreacting particles.
want to
already : the gradient of a
field of forces, are all vector fields. We such fields in connection with fluid flows : motions of a
function, the gravitational field,
4.6
Vector Fields and Fluid Flows
381
Figure 4.46 A vector field is
in R", a
a
a
function which
vector field defined
on
assigns
R", usually considered
vector in
on
D in R" is
U, but interpreted pictorially
as
as
to each
point in a given domain given point. Thus,
based at the
nothing more than in Figure 4.46.
an
Revalued function
Examples body in
58. A
Suppose
there is
a
space sets up a field of gravitational attraction. body of unit mass situated at the origin. According
to Newton's laws another
body
at the
origin squared.
distance
with
body of unit a
force
gravitational
vector field defined on
or, in
,
rectangular
_
to
field of
R3
-
a body {0} by
given
the inverse of the at "2
situated at
a point p by a (see Figure 4.47). the origin is the
coordinates
(x,
.
is attracted to the
represent this attraction origin and of length ||p ||
We
vector directed toward the
Thus the
mass
proportional
y,
z)
v(x,y,z)--(x2 + y>2 + z2)3/2 family of curves, we tangents to the family (Figure 4.48). gents to the family of circles x2 + y2 59. Given
a
may consider the field of unit In particular the field of tan =
c2 is defined
on
R2
-
{0, 0},
4
Curves
Figure 4.47 and is
given by
}
The R
-
(x2
family
+
y2f'2
of unit tangents to the
[0, 0} by
T(x, y)
family
(x, y)
=
(x2
+
y2)1'2
Figure 4.48
of rays is defined
on
4.6
If we it
are
given
a vector
field tangent to a Suppose then that
field
is
v
v on a
domain D in R", the
a
in D such that
v(x) is tangent to Y parametrizes the curve Y. so we must have f '(0 an(l v(f(0) collinear. of the differential equation
curve
at each
function which
f then f
of the
'
(0
arise: Is
point
x on
Let f be
Y.
a
Then f '(0 is tangent to Y at f(0 In particular then, if f is a solution
v(f(0)
=
parametrizes a curve tangent preceding section dx
,
to the
given field.
In the
terminology
,
--v(x) is the
questions
383
of curves, and if so, can we discover the curves ? given vector field in the domain D, and T is a
family
a
Vector Fields and Fluid Flows
0
=
(parametric) differential equation of
the
family of
curves
tangent
to
the vector field. 60.
Suppose v(x, y)
differential
X'
x(0)
=
y(0)
x0
=
Xoe'
We
can
y-cx2
y
(4.68)
2y
=
The solution is x
(x, 2y). (Figure 4.49.) Then the family of field v is given parametrically by this
equation: y'
x
=
=
to the vector
tangent
curves
=
y0
given by =
(4.69)
y0e2t
write this
family
of curves
implicitly
as
=0
(taking the constant of parabolas.
c as
y0 Xo
2).
Thus the
family
we
Another way to find the implicit equation of the one equation in (4.68) by the other:
dy
_
dx
This
dyjdt
curve
2y _
dx/dt
x
we can
solve
directly by separation
seek is
of variables.
a
system
is to divide
384
4
Curves
-^
Figure
4.49
4.6 61. Let
dy
v(x, y)
dyjdt
x
=
(x
y=
Now let
us
equation
is
y=x + y
or
general solution + cex
-(x+ l)
consider
of fluid motion
Then the differential
1).
385
+ y
Tx=jxidr~r which has the
4- y,
Vector Fields and Fluid Flows
a
fluid in motion in
written
a
domain D in R".
The
equations
0 We suppose that at time t there is a particle of fluid at each point x0 in D. The position of that particle The equation of motion at the subsequent time t is denoted by <j)(x0 t). are
follows.
as
=
,
then is
x
For
a
=
fixed x0
,
is at x0 at time xo
(4-7)
=
the
curve
/
0.
=
described
Thus
we are
of the
by (4.69) is the path assuming that
particle which
(4.71)
no two particles can ever occupy the same position Then for each t, the function
It is also assumed that at the same time.
thus
can
the f
=
,
always 0 position of the particle
x
=
,
at the time t
~St
only
at time t such that
x
if
xo
=
Given the fluid motion described 1 0 is the vector field
Definition 8.
velocity
if and
>(x0 0
at
by Equation (4.70) its
=
t=t0
situated at the
point
x
=
0(xo *o)>
If tne vector field is
independent
of
time, we say that the fluid motion is a steady flow. Thus the velocity field of a flow at time t0 and
v(x, t0)
of the
particle
dent of the time,
river of constant
which is at
x
at that time.
point x is the velocity velocity is indepen is steady. The flow in a
If the
particular particle, the flow volume is determined by the shape of the river bed, or
the
and thus
386
4
Curves
tends to be
steady, whereas the flow of clouds in the sky is time dependent. steady, then the path lines (the curves described by (4.70)) are the curves of the family tangent to the velocity field. If the velocity field is time dependent, then these tangent families (called the lines of force) vary with time and have little to do with the paths of individual particles. This is easy to see. Suppose the flow If the flow is
x
=
(4-72)
velocity field v(x, t). equation
has the
Then the
path lines (4.72)
are
the solutions of
the differential
^
x(0)
v(x(0,0
=
The lines of force at time
t
=
=
r0
(4.73)
x0
are
the solutions of the
equation
dx -
These
=
v(x(r), t0)
the
are
same
all t, that is, if and
x(0)
differential
only
=
(4.74)
x0
equations if and only steady.
if
v(x, 0
=
v(x, r0)
for
if the flow is
Examples 62. Consider the flow in R2
x
=
x0e'
=
y
given by Equation (4.67):
y0e2'
(4.75)
Then
x*=x0e'
y'
=
2y0e2'
(4.76)
Thus the
velocity at time t of the particle originally at (x0 y0) is (x0 e', 2y0 e2'). To find the velocity field we must solve (4.75) for x0 y0 in terms of x, t and substitute. Thus (4.76) becomes ,
,
x'
=
so
the
x
y'
=
velocity
2y
field is
v(x, y)
=
(x, 2y)
and the flow is
steady.
Vector Fields and Fluid Flows
4.6
63. Consider the flow in R3
x
x'
=x0 + 1 =
1
y
y'
Thus the
\(x,y, z)
=
=
z'
2r
velocity =
y0 + t2 =
387
given by z
=
z0 +
t3
3i2
field
(l,2t, 3f2)
independent of position but is time dependent. In fact, the path lines are independent of position and are just translates of the twisted cubic (Figure 4.50). It is as if all of space were being rigidly trans lated along the line curve y x2,z x3. Notice that since the t time field at t0 is a constant field, the lines of any given velocity force are straight lines. is
=
=
=
64.
x
=
x0 + t, y
=
y0(l
+
t),
z
=
8x =
~di so
the
(1, y0,z0e')
velocity
field is
v(x,y,z)=^l,^-^,zj
Figure 4.50
z0e'.
Then
388
4
Curves
the flow is not
The lines of force at time t
steady.
=
t0
are
the solutions
of
x'
=
so
x
1
is the
=
? l + f0
y'=
y
=
=
y0
explyj
quite different
65. The flow is
x
z
=
family
x0 + t
which is
z'
x0e'
y
from the
z
family
z0e'
=
of
path
lines.
given by
y0e~'
=
x0(e'
+
e~')
-
z
=
z0
e2t
-
x0(e'
-
e2t) (4.77)
x'
=
x0e'
y' =-y0e~'
+
x0e' z'
x0e~'
+
2z0e2t-x0(et-e2t)
=
(4.78)
The Equations (4.77) are linear in x0 y0 z0 so we may solve for these in terms of x, y, z. Doing so, and substituting the result in '(4.78), we obtain the velocity field of the flow, ,
v(x,
y,
z)
=
(x,
2x
This flow is time It is
y,2z
+
,
,
x)
independent,
or
steady.
immediate consequence of the uniqueness assertion of Picard's a flow is completely determined by its velocity field. For the flow equation is the solution of the initial value problem (4.73), which is an
theorem that
unique. Notice also that the existence part of Picard's theorem asserts that always is a flow associated with a given velocity field (which is sufficiently smooth). The last remark we care to make at this time (we shall continue the study of fluid flows in Chapter 8) is that in the case of a steady flow, the particles follow one another along a fixed family of paths (whereas in general each particle determines its own path). These are of course the lines of force. there
4.6 What
show is this : If two
we must
at different times
there
are
s0
(of course),
,
xt occupy the same
same
paths.
position
That is, if
>(xi, 'i)
=
>
curves
T0 : are
they
x0
follow the
^ such that
,
particles
then
389
Vector Fields and Fluid Flows
x
=
(j>(x0 ,0
I\ :
x
=
the same, except for parametrization. The following proposition proves more. It makes explicit the relation between the two parametriza-
this, and tions.
Proposition 7. (x0 s0), (xu st), ,
(j)(x0 s0)
Suppose
=
x
<j)(x0,t)
describes
a
steady flow.
Iffor
some
have
we
(j)(xu st)
=
,
then
(l>(xo,So In
particular,
+
xt
t)
=
=
(j)(x0
,
s0
-
+
t)
for all
t
(4.79)
s^).
proof is simply that the two functions in (4.79) solve the same problem. Let v(x) be the velocity field of the flow (by assumption v is independent). Consider these functions
Proof.
The
initial value time
f(0
=
0(xo ,50
+ /)
We have
f0(0)
=
fi(0)
Since
^(xo,0=v(#Xo,r)) ot
fx(t)
=
#xi, s,
+
t)
(4.80)
390
4
Curves
for all x, t,
have
we
S
dia
(r)
-
=
,
*>
+ f)
=
y(
,
so
+ 0)
=
v(fo(0)
dU
(0 Thus f0 same
,
=
v(f.(0)
fi solve the
value at 0.
Planetary
same
Thus f0
first-order differential
=
equation
and by
(4.80) have the
fi identically.
Motion
chapter with a study of the classical equations of planetary study first requires these simplifications. We assume all action is in a plane, and that the only force acting is that due to the sun's gravitational field. These simplifications approximate the true situation with For the other forces acting on the body are gravitational enormous accuracy. forces due to other celestial bodies, which are either too far away or too small, relative to the sun, to make a substantial contribution. According to Newton's laws, the acceleration of a body due to the gravitational field is proportional to the field. The motion is thus completely determined by this force and an initial position and velocity. For if s s(t) is the equation of motion, then s is the solution of an initial value problem ; We conclude this
motion.
This
=
s(0)
=
s'(0)
s0
=
s"(t)
=
v0
kF(s(t))
where F is the
given
force field.
Our purpose here is to describe the motion of planets in terms of an observed position and velocity. If we locate the sun at the origin, then the
gravitation
Thus,
we
force field is
must
z(0) z'(0) w
given (in complex notation) by
explicitly
=
z0
=
v0
solve this system
(4.81)
wois
4.6
Vector Fields and Fluid Flows
The best way to solve this is by z(t) r(t)e,ei'\ Then differentiating, =
z'
=
z-
and
r'ew
=
=
r"eie
+
+
2i9'r'eie
+
id"rew
(d')2r
r9")
i(26'r'
+
+
which reduces to this system
r"
(6')2r
-
The second
either &'
natives.
^
=
In
r
or
(4.83)
-
% r2
(4.84)
obtain
=
^ r
(equating
real and
20V + r6"
=
0
imaginary parts) :
(4.85)
6-=(ln6'y t)
0' is
one case
=
reads
=
=
0
(e')2reie '
-
r
equation
2(lnr)'
so
=
Write
becomes
we
-
coordinates.
i8"reie -(9')2reie
Multiplying through by e~'e, r"
polar
(4.82)
2ie'r'eie
+
Equation (4.81) Z"
of
have
iO're'e
+
r"eie
means we
391
proportional
to
r~2.
We have then these two alter
9 is constant, in which case the planet approaches the line. In the other case, the planet rotates around the
a straight angular velocity inversely proportional to the square of the distance from the sun (the closer it is to the sun the faster it rotates around it). Notice constant is possible, so that an also that the solution r constant, 9' The motion of constant angular velocity. admissible path is that of circular the circle gets larger. angular velocity decreases as h, We proceed now to the full solution of (4.84). We already have 9'r2 From (4.84), we obtain a constant (determined by the initial conditions). sun
along
sun
at an
=
=
=
pie
1
i
r2
h
h
392 Thus
4
Curves
we can
'
z
=
Where C
peim is
+
=
arbitrary
an
i9'rew
=
Multiply through by B'r
to obtain
1h e'e + C
=
r'eie
integrate
i eie ie
e
-+p sin(co
again using 9'r2 h
=
rl-
+ p
we
have
imaginary parts :
9)
-
=
Now, using (4.82)
pe"
+
and equate
h
Once
constant.
h,
obtain this
we
implicit