Supplements To The Exercises in Chapters 1-7 of Walter Rudin's
Supplements To The Exercises in Chapters 1-7 of Walter Rudin's
Supplements To The Exercises in Chapters 1-7 of Walter Rudin's
This packet contains both additional exercises relating to the material in Chapters 1-7 of Rudin, and
information on Rudin’s exercises for those chapters. For each exercise of either type, I give a title (an idea
borrowed from Kelley’s General Topology), an estimate of its difficulty, notes on its dependence on other
exercises if any, and sometimes further comments or hints.
Numbering. I have given numbers to the sections in each chapter of Rudin, in general taking each of
his capitalized headings to begin a new numbered section, though in a small number of cases I have
inserted one or two additional section-divisions between Rudin’s headings. My exercises are referred to by
boldfaced symbols showing the chapter and section, followed by a colon and an exercise-number; e.g.,
under section 1.4 you will find Exercises 1.4:1, 1.4:2, etc.. Rudin puts his exercises at the ends of the
chapters; in these notes I abbreviate ‘‘Chapter M, Rudin’s Exercise N ’’ to M : R N. However, I list both
my exercises and his under the relevant section.
It could be argued that by listing Rudin’s exercises by section I am effectively telling the student where
to look for the material to be used in solving the exercise, which the student should really do for his or her
self. However, I think that the advantage of this work of classification, in showing student and instructor
which exercises are appropriate to attempt or to assign after a given section has been covered, outweighs
that disadvantage. Similarly, I hope that the clarifications and comments I make concerning many of
Rudin’s exercises will serve more to prevent wasted time than to lessen the challenge of the exercises.
Difficulty-codes. My estimate of the difficulty of each exercise is shown by a code d: 1 to d: 5. Codes
d: 1 to d: 3 indicate exercises that it would be appropriate to assign in a non-honors class as ‘‘easier’’,
‘‘typical’’, and ‘‘more difficult’’ problems; d: 2 to d: 4 would have the same roles in an honors course,
while d: 5 indicates the sort of exercise that might be used as an extra-credit ‘‘challenge problem’’ in an
honors course. If an exercise consists of several parts of notably different difficulties, I may write
something like d: 2, 2, 4 to indicate that parts (a) and (b) have difficulty 2, while part (c) has difficulty 4.
However, you shouldn’t put too much faith in my estimates – I have only used a small fraction of these
exercises in teaching, and in other cases my guesses as to difficulty are very uncertain. (Even my sense of
what level of difficulty should get a given code has probably been inconsistent. I am inclined to rate a
problem that looks straightforward to me d: 1; but then I may remember students coming to office hours for
hints on a problem that looked similarly straightforward, and change that to d: 2.)
The difficulty of an exercise is not the same as the amount of work it involves – a long series of straightforward manipulations
can have a low level of difficulty, but involve a lot of work. I discovered how to quantify the latter some years ago, in an
unfortunate semester when I had to do my own grading for the basic graduate algebra course. Before grading each exercise, I
listed the steps I would look for if the student gave the expected proof, and assigned each step one point (with particularly simple
or complicated steps given 12 or 112 points). Now for years, I had asked students to turn in weekly feedback on the time their
study and homework for the course took them; but my success in giving assignments that kept the average time in the appropriate
range (about 13 hours per week on top of the 3 hours in class) had been erratic; the time often ended up far too high. That
Semester, I found empirically that a 25-point assignment regular kept the time quite close to the desired value.
I would like to similarly assign point-values to each exercise here, from which it should be possible to similarly calibrate
assignments. But I don’t have the time to do this at present.
Dependencies. After the title and difficulty-code, I note in some cases that the exercise depends on
some other exercise, writing ‘‘>’’ to mean ‘‘must be done after ...’’.
Comments on Rudin’s exercises. For some of Rudin’s exercises I have given, after the above data,
notes clarifying, motivating, or suggesting how to approach the problem. (These always refer the exercise
listed immediately above the comment; if other exercises are mentioned, they are referred to by number.)
True/False questions. In most sections (starting with §1.2) the exercises I give begin with one
numbered ‘‘0’’, and consisting of one or more True/False questions, with answers shown at the bottom of
the next page. Students can use these to check whether they have correctly understood and absorbed the
definitions, results, and examples in the section. No difficulty-codes are given for True/False questions. I
tried to write them to check for the most elementary things that students typically get confused on, such as
the difference between a statement and its converse, and order of quantification, and for the awareness of
what Rudin’s various counterexamples show. Hence these questions should, in theory, require no original
thought; i.e., they should be ‘‘d: 0’’ relative to the classification described above. But occasionally, either
I did not see a good way to give such a question, or I was, for better or worse, inspired with a question
that tested the student’s understanding of a result via a not-quite-trivial application of it.
Terminology and Notation. I have followed Rudin’s notation and terminology very closely, e.g. using
R for the field of real numbers, J for the set of positive integers, and ‘‘at most countable’’ to describe a
set of cardinality ! "0 . But on a few points I have diverged from his notation: I distinguish between
sequences (si ) and sets {si } rather than writing {si } for both, and I use # rather than $ for
inclusion. I also occasionally use the symbols % and &, since it seems worthwhile to familiarize the
student with them.
Advice to the student. An exercise may only require you to use the definitions in the relevant section of
Rudin, or it may require for its proof some results proved there, or an argument using the same method of
proof as some result proved there. So in approaching each problem, first see whether the result becomes
reasonably straightforward when all the relevant definitions are noted, and also ask yourself whether the
statement you are to prove is related to the conclusion of any of the theorems in the section, and if so,
whether that theorem can be applied as it stands, or whether a modification of the proof can give the result
you need. (Occasionally, a result listed under a given section may require only material from earlier
sections, but is placed there because it throws light on the ideas of the section.)
Unless the contrary is stated, solutions to homework problems are expected to contain proofs, even if
the problems are not so worded. In particular, if a question asks whether something is (always) true, an
affirmative answer requires a proof that it is always true, while a negative answer requires an example of a
case where it fails. Likewise, if an exercise or part of an exercise says ‘‘Show that this result fails if
such-and-such condition is deleted’’, what you must give is an example which satisfies all the hypotheses
of the result except the deleted one, and for which the conclusion of the result fails. (I am not counting the
true/false questions under ‘‘homework problems’’ in this remark, since they are not intended to be handed
in; but when using these to check yourself on the material in a given section, you should be able to justify
with a proof or counterexample every answer that is not simply a statement taken from the book.)
From time to time students in the class ask ‘‘Can we use results from other courses in our homework?’’
The answer is, in general, ‘‘No.’’ Rudin shows how the material of lower division calculus can be
developed, essentially from scratch, in a rigorous fashion. Hence to call on material you have seen
developed in the loose fashion of your earlier courses would defeat the purpose. Of course, there are
certain compromises: As Rudin says, he will assume the basic properties of integers and rational numbers,
so you have to do so too. Moreover, once one has developed rigorously the familiar laws of differentiation
and integration (a minor aspect of the material of this course), the application of these is not essentially
different from what you learned in calculus, so it is probably not essential to state explicitly in homework
for later sections which of those laws you are using at every step. When in doubt on such matters, ask
your instructor.
Unfinished business. I have a large list of notes on errata to Rudin, unclear points, proofs that could be
done more nicely, etc., which I want to write up as a companion to this collection of exercises, when I
have time. For an earlier version, see http://www.math.berkeley.edu/~gbergman/ug.hndts/Rudin_notes.ps.
As mentioned in the paragraph in small print on the preceding page, I would like to complement the
‘‘difficulty ratings’’ that I give each exercise with ‘‘amount-of-work ratings’’. I would also like to
complement the dependency notes with reverse-dependency notes, marking exercises which later exercises
depend on, since this can be relevant to an instructor’s decision on which exercises to assign. This will
require a bit of macro-writing, to insure that consistency is maintained as exercises are added and moved
around, and hence change their numbering. On a much more minor matter, I want to rewrite the page-
header macro so that the top of each page will show the section(s) of Rudin to which the material on the
page applies.
I am grateful to Charles Pugh for giving me comments on an early draft of this packet. I would
welcome further comments and corrections on any of this material.
George Bergman
Department of Mathematics
University of California
Berkeley, CA 94720-3840
gbergman@math.berkeley.edu
July 2001, December 2003, May 2006, December 2006
(b) Prove that for any example with the properties described in (a) (not just the example you have given),
the least upper bound of E in S1 must be different from the least upper bound of E in S 3 .
(c) Can there exist an example with the properties asked for in (a) such that E = S1 ? (If your answer is
yes, you must show this by giving such an example. If your answer is no, you must prove it impossible.)
1.2:7. A simpler formula characterizing l.u.b.’s. (d: 2)
Let S be an ordered set, E a subset of S, and x an element of S.
If one translates the statement ‘‘x is the least upper bound of E ’’ directly into symbols, one gets
((% y )E ) x * y) 1 ((% z )S ) ((% y )E ) z * y) / z * x ).
This leads one to wonder whether there is any simpler way to express this property.
Prove, in fact, that x is the least upper bound of E if and only if
(% y )S ) (y < x 2 ((& z )E ) (z > y))).
1.2:8. Some explicit sup’s and inf’s. (d: 2)
(a) Prove that inf {x + y + z x, y, z )R, 0 < x < y < z } = 0.
(b) Determine the values of each of the following. If a set is not bounded on the appropriate side, answer
‘‘undefined’’. No proofs need be handed in; but of course you should reason out your answers to your
own satisfaction.
a = inf {x + y + z x, y, z )R, 1 < x < y < z }. d = sup {x + y + z x, y, z )R, 1 < x < y < z }.
b = inf {x + y – z x, y, z )R, 1 < x < y < z }. e = sup {x + y – 2z x, y, z )R, 1 < x < y < z }.
c = inf {x – y + z x, y, z )R, 1 < x < y < z }.
(d) If a subset E of the real numbers is bounded above, and x = sup E, then x )E.
(e) If a subset E of the real numbers has a largest element, x (i.e., if there exists an element x )E
which is greater than every other element of E), then x = sup E.
(f) If E is a subset of R, and s is a real number such that s > x for all x )E, then x = sup E.
1.4:1. Some explicit sup’s and inf’s. (d: 2)
(a) Prove that inf {x + y + z x, y, z )R, 0 < x < y < z } = 0.
(b) Determine the values of each of the following. If a set is not bounded on the appropriate side, answer
‘‘undefined’’. No proofs need be handed in; but of course you should reason out your answers to your
own satisfaction.
a = inf {x + y + z x, y, z )R, 1 < x < y < z }. d = sup {x + y + z x, y, z )R, 1 < x < y < z }.
b = inf {x + y – z x, y, z )R, 1 < x < y < z }. e = sup {x + y – 2z x, y, z )R, 1 < x < y < z }.
c = inf {x – y + z x, y, z )R, 1 < x < y < z }.
1.4:2. Details on decimal expansions of real numbers. (d: 3)
This exercise gives some of the details skipped over in Rudin’s sketch of the decimal expansion of real
numbers.
In parts (a) and (b) below, let x be a positive real number, and let n 0 , n1 , ... , nk , ... be constructed as
in Rudin’s 1.22 (p.11).
(a) Prove that for all nonnegative integers k, one has 0 ! x – 3 ki= 0 ni 10 –i < 10 –k, and that for all
positive integers k, 0 ! nk < 10.
We would like to conclude from the former inequality that x is the least upper bound of
ni 10 –i k * 0 }. However, in this course we want to prove our results, and to prove this, we need a
{ 3 ki= 0
fact about the numbers 10 –k. This is obtained in the next part:
(b) For any real number c > 1, show that {c k k * 0 } is not bounded above. (Hint: Write c = 1+ h
and note that c n * 1 + nh. Then what?) Deduce that the greatest lower bound of {c –k k * 0 } is 0.
Taking c = 10, show that this together with the result of (a) implies that the least upper bound of
{ 3 ki= 0 ni 10 –i k * 0 } is x.
In the remaining two parts, let m 0 be any integer, and m1 , m 2 , ... any nonnegative integers < 10.
(c) Show that { 3 ki= 0 mi 10 –i k * 0 } is bounded above. (Suggestion: Show m 0 + 1 is an upper
bound.) Thus this set will have a least upper bound, which we may call x.
(d) Let x be as in part (c), and n 0 , n1 , n 2 , ... be constructed from this x as in (a) and (b). Show that
if there are infinitely many values of k such that nk 4 9, then nk = mk for all k. (Why is the
restriction on cases where nk = 9 needed?)
The remaining three exercises in this section go beyond the subject of the text, and examine the
relationship of R with other ordered fields which do or do not satisfy the archimedean property. As their
difficulty-numbers show, these should probably not be assigned in a non-honors course, though students
whose curiosity is piqued by these questions might find them interesting to think about.
1.4:3. Uniqueness of the ordered field of real numbers. (d: 5)
On p.21, Rudin mentions, but does not prove, that any two ordered fields with the least-upper-bound
property are isomorphic. This exercise will sketch how that fact can be proved. For the benefit of students
who have not had a course in Abstract Algebra, I begin with some observations generally included in that
course (next paragraph and part (a) below).
If F is any field, let us define an element nF )F for each integer n as follows: Let 0 F and 1 F
be the elements of F called ‘‘0’’ and ‘‘1’’ in conditions (A4) and (M4) of the definition of a field. (We
add the subscript F to avoid confusion with elements 0, 1 )Z .) For n * 1, once nF is defined we
recursively define (n +1) F = nF +1 F ; in this way nF is defined for all nonnegative integers. Finally, for
negative integers n we define nF = – ( – n) F . (Note that in that expression, the ‘‘inner’’ minus is applied
in Z , the ‘‘outer’’ minus in F.)
(a) Show that under the above definitions, we have (m + n) F = mF + nF and (m n) F = mF nF for all
m, n )Z .
(b) Show that if F is an ordered field, then we also have mF < nF 2 m < n for all m, n )Z . Deduce
that in this case, the map n . nF is one-to-one.
The results of (a) and the first sentence of (b) above are expressed by saying that the map n . nF
‘‘respects’’ the operations of addition and multiplication and the order relation ‘‘<’’.
(c) Show that if F is an ordered field, and if for every rational number r = m ⁄ n (m, n )Z, n 4 0) we
define rF = mF ⁄ nF )F, then r . rF is a well-defined one-to-one map Q . F, which continues to
respect addition, multiplication, and ordering (i.e., satisfies (r + s) F = rF + sF , (r s) F = rF sF , and
rF < sF 2 r < s for all r, s )Q). Thus, Q is isomorphic as an ordered field to a certain subfield of F.
(The statement that the above map is ‘‘well-defined’’ means that the definition is consistent, in the
sense that if we write a rational number r in two different ways, r = m ⁄ n = m( ⁄ n( (m, m( n, n( )Z )
then the two candidate values for rF , namely mF ⁄ nF and m(F ⁄ n(F , are the same. We have to prove
such a ‘‘well-definedness’’ result whenever we give a definition that depends on a choice of how to write
something.)
(Remark: We constructed the map Z . F of (a) without assuming F ordered. Could we have done
the same with the above map Q . F ? No. The trouble is that for some choices of F, the map Z . F
would not have been one-to-one, hence starting with a rational number r = m ⁄ n, we might have found
that nF = 0 F even though n 4 0, and then mF ⁄ nF would not be defined.)
We will call an ordered field F archimedean if for all x, y )F with x > 0 F , there exists a positive
integer n such that nF x > y. Note that by the proof of Theorem 1.20(a), every ordered field with the
least-upper-bound property is archimedean. If F is an archimedean ordered field, then for every x )F let
us define Cx = {r )Q rF < x}. (This set describes how the element x )F ‘‘cuts’’ Q in two; thus it is
called ‘‘the cut in Q induced by the element x )F ’’.)
(d) Let F be an archimedean ordered field, and K an ordered field with the least-upper-bound property
Answers to True/False question 1.4:0. (a) F. (b) F. (c) T. (d) F. (e) T. (f) F.
-8-
(hence also archimedean). Let us define a map f : F . K by setting f (x) = sup {rK r )Cx } for each
x )F. Show that f is a well-defined one-to-one map which respects addition, multiplication, and ordering.
(Note that the statements f (r + s) = f (r) + f (s) etc. must now be proved for all r, s in F, not just in
Q.) Show, moreover, that f is the only map having these properties. In other words, F is isomorphic as
an ordered field, by a unique isomorphism, to a subfield of K.
(e) Deduce that if two ordered fields F and K both have the least-upper-bound property, then they are
isomorphic as an ordered fields.
(Hint: For such F and K, step (d) gives maps f : F . K and k : K . F which respect the field
operations and the ordering. Hence the composite maps f k : K . K and k f : F . F also have these
properties. Now the identity maps id K : K . K and id F : F . F, defined by id K (x) = x (x )K ) and
id F (x) = x (x )F ) also respect the field operations and the ordering. Apply the uniqueness statement
of (d) to each of these cases, and deduce that f and k are inverse to one another.)
Further remarks : It is not hard to write down order-theoretic conditions that a subset C of Q must
satisfy to arise as a cut Cx in the above situation. If we define a ‘‘cut in Q ’’ abstractly as a subset
C # Q satisfying these conditions, then we can show that if F is any ordered field with the least-upper-
bound property, the set of elements of F must be in one-to-one correspondence with the set of all cuts in
Q. If we know that there exists such a field F, this gives a precise description of its elements. If we do
not, it suggests that we could construct such a field by defining it to have one element xC corresponding
to each cut C # Q, and defining addition, subtraction, and an ordering on the resulting set, {xC C is a
cut in Q }. After doing so, we might note that the symbol ‘‘x’’ is a superfluous place-holder, so the
operations and ordering could just as well be defined on {C C is a cut in Q }. This is precisely what
Rudin will do, though without the above motivation, in the Appendix to Chapter 1.
1.4:4. Properties of ordered fields properly containing R. (d: 4)
Suppose F is an ordered field properly containing R.
(a) Show that for every element + )F which does not belong to R, either (i) + is greater than all
elements of R, (ii) + is less than all elements of R, (iii) there is a greatest element a )R that is
< + , but no least element of R that is > + , or (iv) there is a least element that is > + , but no
greatest element of R that is < + .
(b) Show that there will in fact exist infinitely many elements + of F satisfying (i), infinitely many
satisfying (ii), infinitely many as in (iii) for each a )R, and infinitely many as in (iv) for each a )R.
(Hint to get you started: There must be at least one element satisfying one of these conditions. Think
about how the operation of multiplicative inverse will behave on an element satisfying (i), respectively on
an element satisfying (iii) with a = 0.)
In particular, from the existence of elements satisfying (i), we see that F will be non-archimedean.
(c) Show that for every + )F not lying in R there exists 0 > + such that no element x )R satisfies
+ < x < 0 . (This shows that R is not dense in F.)
1.4:5. Constructing a non-archimedean ordered field. (d: 4)
We will indicate here how to construct a non-archimedean ordered field F containing the field R of
real numbers.
The elements of F will be the rational functions in a variable x, that is, expressions p(x) ⁄ q(x)
where p(x) and q(x) are polynomials with coefficients in R, and q is not the zero polynomial.
Unfortunately, though expressions p(x) ⁄ q(x) are called rational ‘‘functions’’, they are not in general
functions on the whole real line, since they are undefined at points x where the denominator is zero. We
may consider each such expression as a function on the subset of the real line where its denominator is
nonzero (consisting of all but finitely many real numbers); but we then encounter another problem: We
want to consider rational functions such as (x 2 – 1) ⁄ (x – 1) and (x + 1) ⁄1 as the same; but they are not,
strictly, since they have different domains.
There are technical ways of handling this, based on defining a rational function to be an ‘‘equivalence
-9-
class’’ of such partial functions under the relation of agreeing on the intersections of their domains, or as
an equivalence classes of pairs ( p(x), q(x)) under an appropriate equivalence relation. Since the subject
of equivalence classes is not part of the material in Rudin, I will not go into the technicalities here, but will
simply say that we will consider two rational functions to ‘‘be the same’’ if they can be obtained from one
another by multiplying and dividing numerator and denominator by equal factors, equivalently, if they
agree wherever they are both defined; and will take for granted that the set of these elements form a field
(denoted R(x) by algebraists). We can now begin.
(a) Show that if q is a polynomial, then either q is ‘‘eventually positive’’ in the sense that
(& B )R) (% r )R) (r > B / q(r) > 0)
or q is ‘‘eventually negative’’, i.e.,
(& B )R) (% r )R) (r > B / q(r) < 0).
or q = 0. (Hint: look at the sign of the coefficient of the highest power of x in q(x).)
(b) Deduce that if f is a rational function, then likewise either
(& B )R) (% r )R) (r > B / f (r) > 0),
or
(& B )R) (% r )R) (r > B / f (r) < 0).
or r = 0. Again, let us say in the first two cases that f is ‘‘eventually positive’’, respectively ‘‘eventually
negative’’.
Given rational functions f and f (, let us write f < f ( if f ( – f is eventually positive.
(c) Show that the above relation ‘‘<’’ makes the field of rational functions an ordered field F.
We shall regard the real numbers as forming a subfield of F, consisting of the constant rational
functions r ⁄ 1 (r )R). In particular, the sets of integers and of rational numbers also become subsets
of F.
(d) Show that in F, the polynomial x (i.e., the rational function x ⁄ 1, which is the function f given
by f (r) = r) is > n for all integers n. Thus, F is not archimedean.
We note a consequence:
(e) Deduce that the element 1 ⁄ x )F is positive, but is less than all positive rational numbers, hence less
than all positive real numbers. (Thus, ‘‘from the point of view of F ’’, the field of real numbers has a
‘‘gap’’ between 0 and the positive real numbers. It similarly has ‘‘gaps’’ between every real number and
all the numbers above or below it.)
1.4:6. A smoother approach to the archimedean property. (d: 2)
Let F be an ordered field.
(a) Suppose A is a subset of F which has a least upper bound, + )F, and x is an element of F.
Show that {a+x | a )A} has a least upper bound, namely + + x.
(b) Suppose x is an element of F, and we let A = {nx | n )Z}. Show that {a+x | a )A} = A.
(c) Combining the results of (a) and (b), deduce that if x is a nonzero element of F, then the set {nx |
n )Z} cannot have a least upper bound in F.
(d) Now suppose x is a positive element, and that F has the least upper bound property. Deduce
from (c) that {nx | n )Z} is not bounded above; deduce from this that {nx | n )J} is not bounded above,
where J denotes the set of positive integers, and deduce from this the statement of Theorem 1.20(a).
1.4:7. An induction-like principle for the real numbers. (d: 1, 2, 2)
Parts (a) and (b) below will show that some first attempts at formulating analogs of the principle of
mathematical induction with real numbers in place of integers do not work. Part (c) gives a form of the
principle that is valid.
Let symbols x, y, etc. denote nonnegative real numbers. Suppose that for each x, a statement P (x)
about that number is given, and consider the following four conditions:
- 10 -
note two changes: First, though as before we have (c a) · b = c (a · b), we now have a · (c b) = c- (a · b).
Secondly, the dot product is no longer commutative. Rather, b · a = a · b .
With these facts in mind, repeat the calculation of the preceding exercise for vectors a and b of
complex numbers. You will get an inequality close to the one you first got in that exercise, except that in
place of a · b, you will have the expression 12 (a · b + a · b ) = Re(a · b). To get around this problem,
verify that for every complex number z there exists a complex number 6 with | 6 | = 1 such that
6 z = | z |. Choosing such a 6 for z = a · b, verify that on putting 6 a in place of a in your inequality,
you get the Schwarz inequality. Again, give a separate quick argument for the case where a or b is
zero.
1.7:4. The Schwarz inequality via the quadratic formula. (d: 3)
The method of proving the real Schwarz inequality given in 1.7:2 requires us to remember one trick,
‘‘Take the dot product of a |b| – |a| b with itself, and use its nonnegativity.’’ One can also prove the
result with a slightly different trick: ‘‘Let t be a real variable, regard the dot product of a + t b with
itself as a real quadratic function of the real variable t, and use its nonnegativity.’’
Namely, expand that dot product in the form a t 2 + b t + c, note why the coefficients a, b and c are
real, and recall that a function of that form has a change of sign if the discriminant b 2 – 4 a c is positive.
Conclude that the discriminant must here be ! 0. Verify that that conclusion yields the real case of the
Schwarz inequality (this time without special treatment of the situation where a or b is zero). Again,
you can get the complex Schwarz inequality by applying the ideas of 1.7:3 to this argument.
The only difficulty is that since we are developing the properties of R from scratch, we should not
assume without proof the above property of the discriminant! So for completeness, you should first prove
that property. This is not too difficult. Namely, assuming the discriminant is positive, check by
computation that if a 4 0, the quadratic formula you learned in High School leads to a factorization of
a t 2 + b t + c, and that this results in a change in sign. The case a = 0 can be dealt with by hand.
Answers to True/False question 1.6:0. (a) T. (b) T. (c) F. Answer to True/False question 1.7:0. (a) T.
- 13 -
Answers to True/False question 2.1:0. (a) F. (b) T. (c) F. (d) T. (e) T. (f) F. (g) F. (h) T. (i) F. (j) F. (k) T.
(l) F. (m) T.
- 17 -
Answers to True/False question 2.2:0. (a) T. (b) F. (c) T. (d) T. (e) T. (f) F. (g) T. (h) F. (i) T. (j) F. (k) T.
(l) F. (m) T. (n) F. (o) T.
- 18 -
metric d.
(b) Will the same be true of the function d ( ( given by d ( ((x, y) = d(x, y) 2 ?
(c) For what functions f from the nonnegative real numbers to the nonnegative real numbers is it true
that for every metric d on a set X, the function d f defined by d f(x, y) = f (d(x, y)) is also a metric
on X ?
2.2:12. A non-closed set has no largest closed subset. (d: 2)
Let E be a subset of a metric space X. Theorem 2.27(c) shows that even if E is not closed, there is
a smallest closed subset of X containing E; i.e., a closed subset which contains E and is contained in
all closed subsets which contain E. However –
(a) Show that if E is not closed, then there does not exist a largest closed subset contained in E. (Hint:
If F is any closed subset contained in E, show that by bringing in one more point one can get a larger
closed subset, still contained in E.)
(b) Exercise 2: R 9(c) (p.43) shows that there is always a largest open subset of X contained in E. Will
there in general exist a smallest open subset of X containing E ?
2.2:13. Reconciling Rudin’s two uses of ‘‘dense subset’’. (d: 2)
On p.9, in the sentence following Theorem 1.20, Rudin implicitly defines a subset E # R to be
‘‘dense’’ if it has the property
(i) For all x, y )R with x < y )R, there exists p )E such that x < p < y.
On the other hand, on p.32, Definition 2.18(j), a subset E of a general metric space X is defined to
be ‘‘dense’’ if
(ii) Every point of X is a limit point of E or a point of E.
Prove that these uses of the word are consistent, by showing that a subset E # R satisfies (i) if and
only if it satisfies (ii).
2.2:14. The n-adic metric on Z. (d: 2)
Let n > 1 be a fixed integer.
For any nonzero integer s, let en (s) be the largest integer a such that s is divisible by n a . Now
define a function dn on pairs of integers by letting dn (s, t) = n –en (s – t) if s 4 t, and letting it be 0 if
s = t. Show that dn is a metric on Z . In fact, in place of condition (c) of Definition 2.15, prove
(c() dn ( p, q) ! max( dn ( p, r), dn (r, q)),
and then show that (c()/(c).
(When n is a prime number p, the metric dp is important in number theory, where it is called the
‘‘p-adic metric’’ on Z . Condition (c(), known as ‘‘the ultrametric inequality’’, is not satisfied by most
metric spaces; in particular it does not hold in R.)
2.2:15. Iterated limit sets. (d: 4)
If E is a subset of a metric space X, let us (in this exercise) write L(E ) for the set of all limit
points of E. We shall write L 2(E ) for L(L(E )), L 3(E ) for L(L 2(E )), etc.; L 0(E ) will denote E
itself.
Show that for every positive integer n there exists a subset E of some metric space X such that
L n–1(E ) 4 :, but L n(E ) = :. (Note: This can be done using X = R, but you may, if you prefer, give
a construction in some other metric space.)
2.2:16. Limit points described in terms of closures. (d: 1)
Let X be a metric space, E a subset of X, and p a point of X. Show that p is a limit point of
E if and only if p )E – { p}.
- 19 -
2.2:17. {q )X : d( p, q) ! r } . (d: 1, 2)
Let X be a metric space. For every point p of X and positive real number r, let
Vr ( p) = {q )X : d( p, q) ! r }.
(a) Show that Vr ( p) is closed, and that
Nr ( p) # Nr ( p) # Vr ( p).
(b) Give four examples of a metric space X, a point p, and a positive real number r, which together
exhibit all four possible combinations of equality and inequality in the above displayed line; i.e., a case
where all three sets are equal, a case where there is equality at the first ‘‘#’’ but not at the second; a case
where there is equality at the second but not at the first, and a case where all three sets are distinct.
2.2:18. Not every finite metric space embeds in an R k. (d: 1, 3, 4)
Let X be a 4-element set {w, x, y, z}, and let d be the metric on X under which the distance from
w to each of the other points is 1, and the distance between any two of those points is 2.
(a) Verify that the above conditions do indeed determine a metric on X.
(b) Show that no function f of X into a space R k is distance-preserving, i.e., satisfies d ( f ( p), f (q)) =
d ( p, q) for all p, q )X.
(c) The above example has the property that every 3-point subset of X can be embedded (mapped by a
distance-preserving map) into a space R k for some k, but the whole 4-point space cannot be so
embedded for any k. Can you find a 5-point metric space, every 4-point subset of which can be so
embedded but such that the whole 5-point space cannot?
2.2:19. An infinite metric space has uncountably many open sets. (d: 4, 2)
Let X be an infinite metric space.
(a) Show that X has a countable family of pairwise disjoint neighborhoods; i.e., that there exist points pi
and positive real numbers ri (i )J ) such that whenever i 4 j, we have Nr ( pi ) - Nr ( pj ) = :.
i j
(Remark: Looking at the case where X is the subset {1 ⁄ n : n )J } , {0} # R, you will find that if
you take any of the pi equal to 0, you can’t complete the construction. Suggested step to get around
this problem: Prove that given any infinite subset E # X and any two distinct points x and y of X, at
least one of x and y has a neighborhood that misses infinitely many points of E; and if you take a
neighborhood of any smaller radius, there will be infinitely many points of E not in its closure.)
(b) Deduce that X has uncountably many open sets. (Hint: Associate to every sequence (si ) of 0’s and
1’s the union of those Nr ( pi ) for which si = 1.)
i
2.2:20. Iterated interior and closure operations. (d: 1,3,2,1,4. > 2: R 9)
Let E be a subset of a metric space X. We shall examine here how many different sets we can get
by successively applying the closure and interior operators to E.
(a) Show that E = E, and that (E °)° = E °. (Hint: For the case of closure, make use of
Theorem 2.2.7. For the case of interior, you may assume the assertions of 2: R 9.)
It follows that if we start with a set and apply some sequence of closure and interior operators to it
(e.g., take the closure of the interior of the interior of the closure of the closure of E ), the application of
one or the other of these operators more than once in succession gives nothing more than applying it once
(e.g., the set just described is simply the closure of the interior of the closure of E ); so anything we can
get, we can get at least as simply by applying closure and interior alternately. This could still, in principle,
lead to infinitely many different sets; but the next result limits further the distinct sets we can get.
(b) Show that (E°) ° = E °.
(c) Deduce from (b) that (E °) ° = E °. Again you may assume the results of 2: R 9.
(d) Deduce from (b) and (c) that starting with E and applying closure and interior operators, one can get
at most 7 distinct sets (counting E itself).
(e) Show by example that a certain subset E # R does indeed yield 7 distinct sets under these operations.
- 20 -
(Some of you might find it easier to keep track of the order of operations if you write cl(E ) in place
of E and int(E ) in place of E °. As examples showing both the advantages and disadvantages of this
notation, the equations of (b) and (c) would become
int(cl(int(cl(E )))) = int(cl(E )) and cl(int(cl(int(E )))) = cl(int(E )).
If you decide to use this notation, do so consistently throughout this exercise.)
2.2:21. Iterated closures and complements. (d: 1,1,3,3. > 2.2:20)
This exercise continues the ideas of 2.2:20. As in that exercise, let E be a subset of a metric
space X.
(a) Show that E ° can be obtained by applying to E some combination of the operations of closure and
complement. (You may assume 2: R 9(d ).)
(b) Deduce that every subset of X that can be obtained from E using a sequence of the operations
closure and interior can also be obtained using a sequence of the operations closure and complement.
(c) Deduce from (b) and the results of 2.2:20 that starting with E and applying closure and complement
operators, one can get at most 14 distinct sets (counting E itself).
(e) Show by example that a certain subset E # R does indeed yield 14 distinct sets under these
operations.
(f) Does there exist a nonempty metric space X and a subset E such that some set obtained from E
using the operator of closure and an even number of applications of the complement operation is equal to a
set obtained from E using the operator of closure and an odd number of applications of the complement
operation?
2.2:22. Some questions on relative closures and interiors. (d: 2)
Suppose X is a metric space and Y # X a subset. For any E # Y, let us write cl X (E ) for the
closure of E in X, and cl Y (E ) for the closure of E relative to Y ; and similarly, int X (E ) and
int Y (E ) for the interior of E in X and relative to Y respectively.
Determine which of the following statements are true whenever X and Y are as above, and E and
F are two subsets of Y :
(a) cl Y (E ) = cl Y (F ) / cl X (E ) = cl X (F ).
(b) cl X (E ) = cl X (F ) / cl Y (E ) = cl Y (F ).
(c) int Y (E ) = int Y (F ) / int X (E ) = int X (F ).
(d) int X (E ) = int X (F ) / int Y (E ) = int Y (F ).
In any case(s) where the assertion is true, you should give a proof; the easiest way is show that the sets
on the right can be constructed from those on the left in a way that doesn’t depend on E or F. In the
case(s) where the assertion is false, you should give a counterexample.
open covering. It would be interesting to find a proof that does not use 2: R 26.)
2.3:5. Redundant coverings of compact sets. (d: 2)
Let K be a compact subset of a metric space, and {G+ }+ )I a covering of K with the property that
every p )K belongs to at least two sets in this covering. Show that {G+ }+ )I has a finite subcovering
with the same property.
2.3:6. R is closed in any metric space. (d: 3)
Suppose X is a metric space having the real line R as a subspace; i.e., such that R is a subset of X,
and the metric on R induced by that of X is the standard metric d(r, s) = | r – s |. Show that R is
closed in X. (Hint: Points close to each other in R belong to a compact subset.)
(b) Deduce Theorem 2.43 from part (a) above with the help of Theorem 2.41 and 2.2:10.
2.4:3. Two metrics on the Cantor set. (d: 2, 4)
Recall that the Cantor set is constructed in 2.44 as the intersection of a chain of sets E1 ; E 2 ; E 3 ;
... , where each set En is a union of 2 n intervals in [0,1]. Let us say that two points p, q of the
Cantor set are ‘‘together at the n th stage’’ if p and q belong to the same interval in En . Let us define
t ( p, q) to be the greatest integer n such that p and q are together at the nth stage, or + 5 if p = q.
(a) Show that the function d( p, q) = 2 – t ( p, q) (defined to be 0 if p = q) is a metric on the Cantor set,
and satisfies the ultrametric inequality (mentioned earlier in 2.2:14):
d(p, q) ! max(d(p, r), d(r, q)).
(b) Show that the same subsets of the Cantor set are open with respect to this metric d as with respect to
the ordinary distance-function | x – y | of R.
3.1:1. A convergent sequence together with the point it approaches form a compact set. (d: 2)
Let X be a metric space.
(a) Show that if ( pi ) is a sequence in X which converges to a point p, then the set { p} , { pi i =
1, 2, ... } is compact.
(b) Deduce that a subset S # X is closed if and only if for every compact subset E # X, the
intersection S - E is also compact.
Answers to True/False question 3.1:0. (a) F. (b) F. (c) T. (d) T. (e) T. (f) F. (g) T.
- 27 -
points of (ak ), and for each j ){1, ... , r} we write Ej for the set of subsequential limit points of
(an ), then E = E1 , ... , Er . Conclude that if X = R, then
j, k
lim sup k . 5 ak = max (lim sup k . 5 an , ... , lim sup k . 5 an ).
1, k r, k
The next part shows that things are very different for infinite unions of subsequences:
(c) Let (ak ) be a sequence in a metric space, let E be its set of subsequential limit points, and let x be
any point of E. Show that (ak ) can be written as the union of countably many subsequences (an ),
1, k
(an ), ... such that each subsequence (an ) converges to x. Moreover, show that we can take these
2, k i, k
subsequences to be ‘‘disjoint’’, in the sense that for i 4 j, {ni,1 , ni,2 , ... } - {nj,1 , nj,2 , ... } = :.
(Hint: Choose a subsequence (an ) of (ak ) that converges to x. Define the subsequences (an )
k i, k
so that each one contains infinitely many terms from (an ) and at most one term not belonging to it.)
k
Deduce that if X = R, then lim sup k . 5 ak is not determined by the numbers lim sup k . 5 an .
j, k
3.2:5. [A revised version of the exercise previously having this number is now 7.1:3.]
3.2:6. Some properties of subsequential limit sets. (d: 3, 3, 4, 5, 5, 4, 5)
In (a)-(f) below, let X be a metric space, (sn ) a sequence in X such that lim n . 5 d(sn , sn+1 ) = 0,
and E the subsequential limit set of (sn ).
(a) Show that if X = R, then E is connected.
(b) Show that for every 8 > 0, all but finitely many natural numbers n have the property that there
exists p )E with d(sn , e) < 8.
(c) Show that if X is compact, then E is connected.
(d) Give an example where X = R 2 and E is not connected. (By Rudin’s definition, the empty set is
connected, so your example must have E 4 :.)
(e) Show that if X = R k and E is not connected, then it contains an unbounded connected subset.
(f) Show that if X = R k and (sn ) is not convergent, then E is perfect.
(g) Show that if X is a connected compact metric space, then there exists a sequence (sn ) with
lim n . 5 d(sn , sn+1 ) = 0 whose subsequential limit set is all of X.
(c) Show that if {G+ } does not have a finite subcover, we can continue this process indefinitely, getting,
in particular, a sequence of points (xn ) such that any two points of this sequence are at distance at least
c apart. Show that such a sequence cannot have a convergent subsequence, contradicting our hypothesis.
The above contradiction shows that {G+ } must have a finite subcover, hence that X is indeed
compact.
(c) Verify the statement on products in the three cases (c1 ) a and b both finite, (c2 ) a finite and
positive, b = + 5 , and (c3 ) a = b = + 5 . (Again, one case you can quote from Rudin.)
(d) Sketch an argument showing that all cases where a b is defined can be deduced from these three.
Similar results are true for quotients a ⁄ b. I omit these for brevity.
(e) Show, conversely, that in every case where a + b, respectively a b is undefined, there exist
sequences of real numbers (xn ) and (yn ) such that xn . a and yn . b, but such that xn + yn ,
respectively xn yn does not approach any extended real number. Again, I recommend first doing some
‘‘key’’ cases, and then getting the remaining cases from these.
(f) Also give an example, for some pair of extended real numbers a and b such that a + b is not
defined, of sequences of real numbers (xn ), (yn ), (xn(), (yn() such that xn . a, yn . b, xn( . a,
yn( . b, but such that xn + yn and xn( + yn( approach different extended real numbers.
(Similar examples exist for all cases where a + b or a b is undefined, but the preceding parts give
enough exercise in passing from key cases to general cases.)
3.4:4. Every real sequence has a subsequence that approaches some extended real. (d: 1)
Show that every sequence of real numbers has either a convergent subsequence, or a subsequence which
. + 5 , or a subsequence which . – 5 .
3.4:5. An infinite cube as k-dimensional analog of the extended real line. (d: 1, 3. > 3.4:3, 3.4:4)
As Rudin shows us, it is often useful to regard the real line R as a subset of the extended real line
R , { – 5 , + 5 }. However, there is no one natural analog of this construction for Euclidean space R k .
This exercise and the next discuss two distinct extensions of R k .
The one we shall consider in this exercise is the set (R , { – 5 , + 5 }) k of all k-tuples of extended real
numbers; i.e., strings a = (a1 , ... , ak ) in which each ai is either a real number or – 5 or + 5 . Given a
sequence (x n ) of elements of R k and a point a )(R , { – 5 , + 5 }) k, let us write x n . a if for each
i ){1,, ... , k } one has xn,i . ai , where xn,i denotes the i th component of x n . (Note that like Rudin’s
limits in the extended real line, this concept of limit has not been defined in terms of a metric on our set.)
(a) Show that every sequence (x n ) of elements of R k has a subsequence which approaches some point
of (R , { – 5 , + 5 }) k. (Hint: Apply 3.4:4 to first coordinates, getting a subsequence of (x n ), then to
second coordinates of the terms of that subsequence, etc..)
(b) Given two points a, b )(R , { – 5 , + 5 }) k, one would like to define elements a + b )
(R , { – 5 , + 5 }) k and a · b )R , { – 5 , + 5 } in a way that respects limits of sequences; that is, so that if
(x n ) and (y n ) are sequences in R k such that x n . a, y n . b, then x n + y n . a + b and
x n · y n . a · b.
Determine for what pairs of elements a and b there exist elements ‘‘a + b’’, respectively ‘‘a · b’’,
with these properties. Prove in these cases that the desired properties hold, and in all other cases that there
exist sequences (x n ) and (y n ) which approach the indicated points of (R , { – 5 , + 5 }) k, but such that
(x n + y n ) does not approach any point of (R , { – 5 , + 5 }) k, respectively such that (x n · y n ) does not
approach any element of R , { – 5 , + 5 }. You may assume the result of 3.4:3 above, even if you did not
do it.
State also how and under what conditions the product a b of an extended real number a and an
element b )(R , { – 5 , + 5 }) k can be defined so as to respect limits of sequences. (The verification is so
close to that of the case of the dot product a · b above that I won’t ask you to give it.)
3.4:6. An infinite ball as k-dimensional analog of the extended real line. (d: 2, 3, 2, 4, 4. > 3.4:3, 3.4:4)
Here is another way of defining ‘‘infinite limits’’ of sequences in R k .
Suppose a is a point in R k satisfying |a| = 1, i.e., lying on the sphere of radius 1 centered at 0.
If (x n ) is a sequence in R k , let us write x n . a 5 if |x n | . + 5 and x n ⁄ |x n | . a. (In the latter
limit statement we must drop any terms for which x n = 0; but if |x n | . + 5 there can be only finitely
many such n, so this is no real problem.)
(a) Show that every sequence (x n ) in R k has a subsequence which either approaches a point of R k or
approaches some a 5 in the sense defined above. (Hint: Show that (x n ⁄ |x n |) is bounded.)
(b) Show that if for every a satisfying |a| = 1, we define a 5 + a 5 = a 5 , then this operation
respects limits of sequences, in the sense that if x n . a 5 and y n . a 5 , then x n + y n . a 5 .
(c) Show similarly that for a as above and p )R k , the definition p + a 5 = a 5 respects limits of
sequences.
(d) On the other hand, if a, b )R k are distinct points satisfying |a| = |b| = 1, show that one cannot
define a 5 + b 5 in a way that will respect limits of sequences. Namely, show that for every such pair
of points a and b there exist sequences (x n ) and (y n ) satisfying x n . a 5 , y n . b 5 , such that
x n + y n does not approach c 5 for any c, nor any point of R k . (Suggestion: Get an example where
the odd-subscripted terms x 2n+1 + y 2n+1 approach a 5 and the even-subscripted terms x 2n + y 2n
approach b 5 .)
(e) Determine similarly the conditions under which the dot product of two ‘‘infinite points’’ (a 5 ) · (b 5 ),
and the dot product of a ‘‘finite’’ and an ‘‘infinite’’ point, (a 5 ) · p, can be defined in a manner that
respects limits of sequences.
3.4:7. Convergence in the sense of 3.4:5 and in the sense of 3.4:6 are almost independent. (d: 4, 4, 2.
> 3.4:5, 3.4:6)
(a) Show by examples that a divergent sequence in R k can approach a point of (R , { – 5 , + 5 }) k in the
sense of 3.4:5 above, but not approach any a 5 in the sense of 3.4:6, and similarly that such a sequence
can approach a point a 5 in the sense of the latter exercise, but not approach any point of
(R , { – 5 , + 5 }) k in the sense of the former exercise.
(b) Which points p of (R , { – 5 , + 5 }) k have the property that every sequence in R k which
approaches p in the sense of 3.4:5 approaches a point a 5 in the sense of 3.4:6? Which points a 5
have the property that every sequence in R k which approaches a 5 in the sense of 3.4:5 approaches a
point (R , { – 5 , + 5 }) k in the sense of 3.4:6?
(c) Show that every unbounded sequence in R k has a subsequence which both approaches a point of
(R , { – 5 , + 5 }) k in the sense of the former exercise and approaches a point a 5 in the sense of the
latter.
3.4:8. Incomparable sets of natural numbers. (d: 2, 5)
This exercise has little to do directly with Real Analysis, except for getting one thinking about the
operation ‘‘lim sup’’; but I find it interesting.
For every set S of natural numbers and every natural number n, let
in(S, n) = the number of natural numbers m < n such that m )S,
out(S, n) = the number of natural numbers m < n such that m )S.
So, out(S, n) = n – in(S, n).
(a) Show that if T is a subset of the natural numbers and S is a proper subset of T, then
lim sup n . 5 (in(S, n) – out(S, n)) < lim sup n . 5 (in(T, n) – out(T, n)), unless both of these are + 5 or
both are – 5 .
Now let A be the set of all subsets of the natural numbers such that
lim sup n . 5 (in(S, n) – out(S, n)) = 0.
Thus, we see from (a) that no member of A is a proper subset of any other member of A.
(b) Show that every set S of natural numbers either contains (as a subset) some member of A, or is
contained (as a subset) in some member of A.
- 32 -
an < 1. (Suggestion: Let the sequence have the form 1 + c1 , 1 – c1 , 1 + c 2 , 1 – c 2 , ... , 1 + cn , 1 – cn , ...
where cn )(0, 1), and examine conditions for the above sum and product respectively to converge.)
3.6:3. The Cantor set, and the 2 -adic metric on Z. (d: 4. > 2.2:14, 2.4:3, 3: R 19)
The result of Rudin’s exercise 3: R 19 is equivalent to saying that the Cantor set consists of precisely
those real numbers which can be written in base 3 using only 0’s and 2’s to the right of the decimal point,
and nothing to the left. (For instance, the largest element of the Cantor set, namely 1, can be written
.2222... in base 3, just as it can be written .9999.... in base 10.) Now consider the following function f
from the nonnegative integers to the Cantor set:
Given a nonnegative integer n, write n in base 2 (binary) notation, then reverse the order of digits,
precede the resulting symbol by a decimal point, change all the ‘‘1’’s to ‘‘2’’s, and regard the result as an
expression for a real number f (n) in base 3. For example, taking n = 6, which in binary notation is
110, our construction gives the number whose base-3 expression is .022. This is 8 ⁄ 27, so f (6) = 8 ⁄ 27.
Show that for positive integers m and n, one has d 2 (m, n) = d ( f (m), f (n)), where d 2 is defined
as in 2.2:14, and d as in 2.4:3. Show also that f (J) is dense in the Cantor set.
3.6:4. The space 2 of square-summable sequences. (d: 2, 2, 1, 3)
Let 2 (pronounced ‘‘little-ell-2’’) denote the set of all sequences x = (x1 , x 2 , ... , xn , ... ) of real
numbers such that 2
35n =1 | xn | converges.
Such a sequence is called square-summable. (I used absolute-value signs above to put the definition in a
form applicable to complex sequences as well, but we will only consider real sequences in this exercise.)
For x ) 2, we define 2 12
|x| = (3 5n =1 | xn | ) .
In this exercise, sums and scalar multiples of sequences of real numbers will be defined by the same
rules used for sums and scalar multiples of elements of R k in Definition 1.36 (p.16). A general technique
recommended for most of the steps below is to use Theorem 1.37 (p.16) to estimate partial sums.
(a) Show that if x, y ) 2 and + )R, then + x and x + y are also in 2
, and that the series
35n =1 xn yn converges. (Its sum is denoted x · y.)
2
(b) Show that if we define d (x, y) = |x – y|, then this definition makes a metric space.
2
(c) For each positive integer n, let e n ) be the sequence whose n th component is 1, while all other
components are 0. Show that the sequence e1 , e 2 , ... in 2 has the property that for each m, the m th
coordinates of the terms of the sequence converge to 0, but the sequence itself is not convergent.
2
(d) Show that is a complete metric space.
Remark: The space 2 is an example of what is known as a Hilbert space, that is, a complete inner
product space. (‘‘Complete’’ in the sense of this course; ‘‘inner product space’’ in the sense of
Math 110.) All finite-dimensional inner product spaces are Hilbert spaces; it is the infinite-dimensional
ones, such as 2, that make the subject particularly interesting. Hilbert spaces over both the real and
complex numbers are studied; I would have used complex sequences above, except that Rudin only states
Theorem 1.37 for tuples of real numbers.
Another construction of a Hilbert space, using square-integrable functions rather than square-summable
sequences (and based on Lebesgue integration, hence not within the scope of Math 104) is studied in the
very last section of Rudin (Definition 11.34 on p.326; cf. last sentence before the exercises on p.332).
From the results of that section, in particular the formula (107), one can show that the Hilbert space of
square-summable sequences and the Hilbert space of square-integrable functions are isomorphic.
3.6:5. A divergence result. (d: 3)
On p.55, Rudin defines what it means for a sequence of real numbers to satisfy sn . + 5 or
sn . – 5 , noting that a sequence which satisfies one of those conditions is still not said to converge.
Answers to True/False question 3.5:0. (a) T. (b) F. Answers to True/False question 3.6:0. (a) T. (b) F.
- 34 -
(Thus, a nonconvergent sequence of real numbers either approaches + 5 or approaches – 5 or does not
approach any extended real number.) On p.59, the equation 3 an = s is defined to mean sn . s, where
(sn ) is the sequence of partial sums. Combining these conventions we have definitions of the conditions
3 an = + 5 and 3 an = – 5 .
Suppose now that (an ) and (bn ) are sequences of real numbers satisfying an ! bn for all n. Show
that if 3 an does not converge, and is not – 5 , then 3 bn does not converge.
positive terms that approaches 0, and that it decays more rapidly than (an ), but that (Bn ) still
approaches + 5 , hence that 3 bn is still divergent.
(Hint: Here and in part (b), the relation Bn = An2 is more easily used in the form An = Bn2.)
1
(c) Suppose 3 an is a convergent series of positive terms. Show that there is a decreasing sequence
(An ) of positive real numbers with lim n . 5 An = 0 such that An – An+1 = an for all n. Again let
Bn = An2 and now let bn = Bn – Bn+1 . Show that (Bn ) also approaches 0, hence that 3 bn
1
(ii) If (an ) and (bn ) are decreasing sequence of positive terms such that bn = an for all k,
k k
then the series 3 an converges if and only if 3 bn does.
(iii) The set of ratios nk+1 ⁄ nk is bounded.
(iv) (nk ) has a subsequence (nm ) such that the set of ratios (nm – nm ) ⁄ (nm – nm ) is
k k+1 k k k –1
bounded.
Suggestion: Prove (i)/(ii)/(iii)/(iv)/(i). The second implication is the hardest; I suggest proving
it in contrapositive form, using the idea of 3.7:2. For the third implication, to get the subsequence (nm ),
k
suppose mk has been chosen; then let mk+1 be the least integer such that nm * 2 nm .
k+1 k
3.7:5. Series that can be tested for convergence by looking at the terms a 2k
. (d: 2. > 3.7:2)
2
The fallacious argument discussed in 3.7:2(b) can in fact be made to work, if we just add an additional
condition to our series.
Namely, show that if (an ) is a sequence of real numbers satisfying the stronger condition a1 *
k
2 a 2 * ... * n an * ... , then 3 an converges if and only if 3 2 k 2 2 a 2 k does. Deduce that if (bn ) is
2
another sequence with the same property, and for all k, a 2 k = b 2 k , then 3 an converges if and only if
2 2
3 bn does.
3.7:6. The series of distances associated with a convergent sequence. (d: 2)
Suppose ( pn ) is a sequence in a metric space X which converges to a point p.
(a) Show that d ( p1 , p) ! 3 5 n =1 d ( pn , pn+1 ).
(b) Show that for every 8 > 0 there exists a subsequence ( pn ) with n1 = 1 such that
k
3 k d ( pn , pn ) < d ( p1 , p) + 8 .
k k+1
Now suppose ( pn ) is an arbitrary sequence of points in X.
(c) Show that if the series 3 5
n =1 d ( pn , pn+1 ) converges, then ( pn ) is a Cauchy sequence; but that the
converse is not true. (Hint for the last part: Look for a counterexample in a metric space we are very
familiar with.)
the series, is given in part (b), and is relatively easy to state and prove if we again allow ourselves to
assume basic properties of the logarithm function. The analog of the ratio test, based on considering the
relation between successive terms, is noted in part (e). Like the ratio test, it is generally easier to use than
the other criterion, but cannot handle series whose terms don’t decrease regularly. It is, unfortunately, hard
to prove without the use of differentiation, which we have not yet defined. (The one way I see that one
could prove it at this stage is by taking a rational number j ⁄ k between p and 1, and using inequalities
obtained from the binomial theorem for exponents j and k.) However, I will state both tests here for
their interest.
(a) If p is any positive real number, and we let an = n – p , show that – lim n . 5 (log an ) ⁄ (log n) = p.
(b) Let (an ) be any sequence of positive terms, and let p = – lim sup (log an ) ⁄ (log n). Show that if
p > 1, then 3 an converges.
(c) What happens when we apply this test to a series which the root test shows to converge?
(d) For p again any positive real number and an = n – p , show that lim n . 5 n (1 – (an+1 ⁄ an )) = p.
(e) Let (an ) be any sequence of positive terms, and let p = lim inf n (1 – (an+1 ⁄ an )). Show that if
p > 1, then 3 an converges.
(f) What happens when we apply this test to a series which the ratio test shows to converge?
3.9:2. A slightly modified ratio test. (d: 3,1)
(a) Show that a series 3 an converges if for some positive integer c one has
lim sup n . 5 |an+c ⁄ an | < 1.
(b) Show that this condition is satisfied for c = 2 by both the series of Examples 3.35 (p.67).
Answers to True/False question 3.11:0. (a) F. (b) T. Answer to True/False question 3.12:0. (a) F.
- 40 -
(You should give a careful ‘‘8 ’’-proof, unless you prove it using some previous result in Rudin.)
(b) Obtain the analogous result relating 3 5 5
n = 0 an and 3 k = 0 (a 3k + a 3k +1 + a 3k + 2 ). Rather than
repeating the whole proof, just indicate carefully what changes need to be made in the proof of (a).
(c) Use (b) above to show that the series (23) on p.76 of Rudin converges. (First describe that series
precisely.)
(d) Obtain a different proof of the convergence of Rudin’s series (23) by applying Theorem 3.42 with
(an ) the sequence ‘‘1, 1, – 2, 1, 1, – 2, 1, 1, – 2, ...’’. (You should figure out what (bn ) is to be.)
(e) Show that the result of (a) becomes false if the condition lim n . 5 an = 0 is removed.
3.14:2. Which series have convergent rearrangements? (d: 3)
Find a simple criterion for a series 3 an of real numbers to have the property there exists a
rearrangement 3 ak which converges.
n
(The answer to this question is fairly easy; the answer to the corresponding question for series of
complex numbers, equivalently, of points of R 2, is much more difficult to state and prove.)
3.14:3. Which rearrangements of terms don’t affect convergence of any series? (d: 1, 3, 4, 5)
We saw in Theorem 3.54 that rearranging the terms of a non-absolutely convergent series can change
its behavior drastically. But not all rearrangements can have such effects. Parts (a)-(c) below show that
the effects of certain rearrangements are limited in one way or another. Part (d), which is much harder,
gives a general criterion for when this happens.
(a) If (an ) is a sequence, consider the rearrangement gotten by interchanging successive pairs a 2m and
a 2m +1 . Clearly, this can be written (ak ), where (kn ) is the sequence of integers 2, 1, 4, 3, 6, 5, ... ,
n
defined by k 2m –1 = 2m, k 2m = 2m –1 (m = 1, 2, ... ).
Show that if 3 an or 3 ak is a convergent series, then so is the other, and 3 ak = 3 an .
n n
(b) If (an ) is a sequence, consider the rearrangement gotten by breaking it into blocks whose lengths are
successive powers of 2, i.e., (a i , ... , a i +1 ), and within each such block, collecting all the even-
2 2 –1
subscripted terms before the odd-subscripted ones, but otherwise preserving their order. This
rearrangement can be written (ak ), for an appropriate sequence (kn ) of integers, whose first 16 terms
n
are as follows. (I use extra space to make visible the separation into ‘‘blocks’’.)
1, 2, 3, 4, 6, 5, 7, 8, 10, 12, 14, 9, 11, 13, 15, 16, ... .
i
Explicitly, for 0 ! i and 0 ! j < 2 we have k 2 i +2j = 2 +j, and (if i > 0) k 2 i +2j +1 = 2 i +2 i –1 + j.
i
Find a convergent series 3 an such that 3 ak diverges.
n
(c) On the other hand, show that for (kn ) as in part (b) above, if 3 ak converges then 3 an also
n
converges and 3 ak = 3 an .
n
(It follows that if we write ( jn ) for the permutation inverse to (kn ), i.e., for each n let jn be the
unique integer such that kj = n , then the sequence ( jn ) has the opposite properties: if 3 an
n
converges then so does 3 aj , and 3 aj = 3 an , but there exists a divergent series 3 an such that
n n
3 aj converges.)
n
(d) Now let (kn ) be an arbitrary sequence of positive integers in which each positive integer occurs once
and only once. (This could be described in Rudin’s language as a rearrangement of the sequence of
positive integers; or, in different terminology, as a permutation of the positive integers.) For every positive
integer N, let us define the mixing number mix((kn ), N ) to be the largest integer M for which there
exist M integers n1 , n 2 , ... nM which are alternately ! N and > N (i.e., such that if ni ! N then
ni+1 > N and vice versa), and such that kn < kn < ... < kn . In other words, if we color each positive
1 2 M
integer m red or blue according to whether it shows up before or after the N th comma in the sequence
Answer to True/False question 3.13:0. (a) F. Answers to True/False question 3.14:0. (a) T. (b) T. (c) F.
- 41 -
k1 , k 2 , k 3 , ... , then mix((kn ), N ) denotes the number of same-color blocks into which the set of all
positive integers is divided. Since only N integers are colored red, this number is at most 2N +1; in
particular, it is finite.
If you think through the example you constructed for part (b), you should find that it implicitly used the
fact that for the sequence of integers occurring there, mix((kn ), N ) assumed arbitrarily large values, while
the positive result of (c) was based on the fact that mix(( jn ), N ) never goes higher than 4.
In fact, prove that the following conditions on a rearrangement (kn ) of the positive integers are
equivalent:
(i) mix((kn ), N ) is bounded as a function of N.
(ii) For every convergent series 3 an of real numbers, the series 3 ak also converges.
n
Show moreover that if these conditions hold, then for every convergent series 3 an of real numbers,
3 ak = 3 an .
n
3.14:4. Rearrangements and the root test. (d: 3)
Find a series which converges by the root test, and a rearrangement of this series for which the root test
gives no information.
However, note a reason why, in a situation of this sort, the rearranged series must still converge.
3.14:5. The subsequential limit set of a rearranged convergent series. (d: 1)
Let 3 an , + and 0 be as in the hypothesis of Theorem 3.54, and let 3 a(n and (s(n ) be as in the
conclusion thereof. Assuming the result of 3.2:6(a) (even if you haven’t proven it), deduce that the
subsequential limit set of (s(n ) is the interval [+ , 0 ].
3.14:6. Rearrangements with alternating signs. (d: 5)
Show that if (an ) is a non-absolutely convergent series, and + is a real number, then there exists a
rearrangement (ak ) of (an ) such that for all even n, ak * 0, for all odd n, ak ! 0, and 3 ak = + .
n n n n
(One could even get the partial sums to have lim sup any + and lim inf any 0 such that
– 5 ! + ! 0 ! + 5 , as in Theorem 3.54. But proving that would just be more work, without a really
different idea.)
Chapter 4. Continuity.
point of E. Show that if for some r > 0, the set of real numbers f (E - Nr ( p)) is bounded, and if
lim x . p g(x) = 0, then lim x . p f (x) g(x) = 0. (In words, ‘‘If as x . p, one function remains bounded
and the other approaches 0, then their product approaches 0.’’)
4.1:2. Limit points of functions that don’t necessarily approach a limit. (d: 2)
In Chapter 3, we saw that even if a sequence of points of a metric space did not approach a limit, we
could still look at its ‘‘subsequential limit points’’. We can associate a similar set of points to a function
between metric spaces:
(a) If f : E . Y is a function from a subset of a metric space X to another metric space, and p is a
limit point of E in X, show that the following three subsets of Y are the same:
(i) {y )Y (% 8 > 0) (& x )E ) 0 < d(x, p) < 8 and d( f (x), y) < 8 }.
(ii) The set of all points lim n . 5 f (xn ) such that (xn ) is a sequence converging to p in E – { p}
and ( f (xn )) converges in Y.
(iii) The union, over all sequences (xn ) converging to p in E – { p}, of the set of subsequential
limit points of ( f (xn )) in Y.
(b) Show that the set described in three equivalent ways in (a) is closed.
4.1:3. Analog of the Cauchy criterion for functions between metric spaces. (d: 2, 1)
(a) Suppose f : E . Y is a function from a subset E of a metric space X to a complete metric space
Y, and let p )X be a limit-point of E. Formulate and prove a necessary and sufficient condition for
lim x . p f (x) to exist, analogous to the condition ‘‘(sn ) is Cauchy’’ for the limit of a sequence in Y to
exist.
(Since ‘‘complete metric space’’ is defined in terms of the convergence of Cauchy sequences, your
proof will have to relate the behavior of functions to the behavior of sequences. But the actual formulation
of your criterion should be as an ‘‘8 -7 ’’ condition, not one about sequences.)
(b) Show that if the assumption that Y is complete is deleted from (a), the condition you have obtained is
still necessary, but no longer sufficient, for lim x . p f (x) to exist.
(b) Deduce that X is connected if and only if every continuous function X . {0, 1} is constant.
(c) Show likewise that X is connected if and only if every continuous function X . Z is constant.
4.2:4. The archimedean property for powers of a continuous increasing function. (d: 4 or 2, 3)
(a) Suppose f : R . R is a continuous function such that f (x) > x for all x. For any positive integer
n, let f n denote the n-fold composite f ° f ° ... ° f. Show that for all x, y )R, there exists a positive
integer n such that f n (x) > y.
(Part (a) above has difficulty d: 4 if you have not done Exercise 1.4:7, d: 2 if you have. Suggestion on
how to use that exercise: For each real number x, consider the statement P (t) about a nonnegative real
number t saying that for some positive integer n, f n (x) > x + t.)
(b) Does the conclusion of part (a) remain true if the assumption that f is continuous is removed?
4.2:5. Fixed points for continuous increasing functions. (d: 2)
Let E be a closed bounded subset of R, and f : E . E a continuous strictly monotone increasing
function, i.e., a continuous function such that for all x, y )E, x < y / f (x) < f (y). Take any x 0 )E,
and define x1 , x 2 , x 3 , ... )E by the rule xn+1 = f (xn ).
(a) Show that if x1 > x 0 , then xn+1 > xn for all n, and sketch why the corresponding implications
hold with ‘‘>’’ replaced by ‘‘=’’ and by ‘‘<’’.
(b) Show that lim n . 5 xn exists, and is a fixed point of f ; i.e., that denoting this limit by x we have
f (x) = x.
(c) Show that if x1 > x 0 , then the point x found in (b) is the least fixed point of f which is > x 0 .
State the corresponding characterizations of x in the cases x1 = x 0 and x1 < x 0 . (You do not have to
write out proofs for these cases.)
(Remarks: The same result can be proved if E is the extended real line. Exercise 4.2:4 is related to
that fact, though it is not a special case since the function there is not assumed monotone. Another related
exercise is 1.2:4.)
4.2:6. The descendants of an amoeba. (d: 2. > 4.2:5)
Let t )[0, 1]. Suppose we have a population of amoebas such that every hour on the hour, each
amoeba in the population splits in two, and such that over the course of each hour, each amoeba has
probability t of surviving, and 1 – t of dying, with the survival of different amoebas in the population
independent of one another. Suppose we start with a single living amoeba at hour 0 (the result of a
division that has just taken place, so that it will divide again at hour 1 if it survives till then), and for
n = 0, 1, 2, ... let pn denote the probability that at least one of its descendants will be alive at hour n.
(We count it as one of its own descendants; thus, p 0 = 1.) Since this number depends on the constant t
as well as n, let us write it, more precisely, as pn (t).
(a) Find a formula for computing pn+1 (t) from pn (t).
(Remark: The hard way to approach this question is to look ahead to hour n and consider the number
of descendants alive at that time, and whether any will survive during the next hour. The easy way is to
note that the original amoeba has probability t of surviving to hour 1, and that if it does so, it will divide
and each of the resulting amoebas will have, independently, probability pn (t) of having at least one living
descendant n hours later. Recall also the principle: If two independent events have probabilities x and
y of occurring, then the probability that both will occur is xy. To determine the probability that one or
both will occur, note that this is the probability that they will not both fail to occur. The probability that
they will both fail to occur is, by the preceding principle, (1– x) (1– y), so the probability that they will
not is 1 – (1– x) (1– y).)
(b) Does lim n . 5 pn (t) exist? If so, determine its value. (Suggestion: Use 4.2:5.)
Tangential remarks: The assumption of synchronized reproduction is, of course, an absurd
Answers to True/False question 4.2:0. (a) T. (b) T. (c) T. (d) F. (e) F. (f) T. (g) F. (h) T. (i) T.
- 45 -
simplification; and one can also see that conditions can’t remain constant when n . 5 , since an
expanding population will run out of food and space. Nevertheless, the result you will obtain above is
probably a reasonable first approximation to the probability that a simple organism entering a new habitat
will succeed in proliferating rather than dying out.
Actually, I didn’t come upon this question by thinking about amoebas. Rather, I was considering the
algebraic situation where one has a binary operation *, not necessarily associative, on a set X, and one
wants to study expressions in which * connects an arbitrary number of elements, for instance, (a *b),
((a * (b *c)) * (d *e)), etc.. To study ‘‘statistical’’ properties of such expressions, I wanted to assume them
generated in a random way. In the expressions in question, the elements a, b, c etc. were to appear in
fixed order, so when generating the expressions, one could use a place-holder instead of these letters,
and work with symbols like (( * ( * )) * ( * )). I decided that the mathematically simplest sort of
random generation would be to start with a single symbol , and execute a series of steps, at each of
which every would either be replaced by ( * ), with some probability t, or else be declared
‘‘final’’, and undergo no more changes after that. When all squares have become ‘‘final’’, one has a
randomly generated expression. But can one expect all squares to eventually become ‘‘final’’? I worked
out the probability that this would happen as a function of t – and found that the resulting computation
was an example of something it will be useful to know when we reach Chapter 7 of Rudin. So I
reformulated the problem as a more concrete question about amoebas, and have introduced it here in
preparation for that later chapter.
Some years later, I learned of still another way of looking at this computation. Consider a system of
passageways, beginning at an initial point, where each passageway ends by branching into two further
passageways. Suppose each passageway has probability t of being open, and 1– t of being blocked.
Then the above problem concerns the probability that this system has an infinite unblocked route from the
initial point. This is a simple case of the subject of percolation theory, which takes its name from the case
where the passageways are pores in a solid. Closer to real percolation problems, and more difficult to
analyze, are cases where the passageways form infinite ‘‘checkerboard arrays’’ in 2 or more dimensions.
4.2:7. When can one compose limit-statements? (d: 1; 3)
The principle ‘‘If as x . p, f (x) . q, and if as y . q, g(y) . r, then as x . p, g( f (x)) . r’’
may appear ‘‘obvious’’. But the following example shows that it is false.
(a) Let f : R . R be defined by f (x) = x sin x –1 if x 4 0, f (0) = 0, and let g : R . R be defined by
g(x) = 0 if x 4 0, g(0) = 1. Show that lim x . 0 f (x) = 0 and lim x . 0 g(x) = 0, but that it is not true
that lim x . 0 g( f ((x))) = 0.
Why is the apparently obvious principle that we started with false? Because ‘‘. q’’ has slightly
different meanings in the two statements ‘‘as x . p, f (x) . q’’ and ‘‘as y . q, g(y) . r’’! In the
first it means that f (x) takes on values that become arbitrarily close to q; but in the second, it means
that y takes on values getting arbitrarily close to q, other than the value q itself. In the example
above, we see that as x . p the function y = f (x) has the first of these properties, but not the second.
Obviously, it would be good to know conditions under which we can correctly ‘‘compose’’ limit
statements. Necessary and sufficient conditions are given in
(b) Suppose f is a function from a subset E of a metric space X to a metric space Y, and g is a
function from a subset F of Y which contains f (E ), to a metric space Z . Let p be a limit point of
E in X, q a limit point of F in Y, and r a point of Z , such that lim x . p f (x) = q and
lim y . q g(x) = r. Show that the following conditions are then equivalent:
(i) lim x . p g( f ((x))) = r.
(ii) Either q )F and g is continuous at q, or there exists 8 > 0 such that q )f (N8 (x) - E ) (i.e.,
the function f does not take on the value q at points arbitrarily close to p).
4.2:8. Two different meanings of ‘‘neighborhood’’ have similar properties. (d: 3)
What Rudin calls a ‘‘neighborhood’’ of a point p of a metric space is nowadays more often called an
- 46 -
‘‘open ball’’ about the point. Topologists instead generally define a neighborhood of p to mean any set
E such that p is an interior point of E . In this exercise, let us call this concept a ‘‘topologist’s
neighborhood’’, and let the unmodified word have Rudin’s meaning. We shall see that the topologist’s
concept is as convenient as Rudin’s for some typical uses of the concept.
(a) Show that if p is a point of a metric space X, then a subset E # X is a ‘‘topologist’s
neighborhood’’ of p if and only if it contains a neighborhood of p in Rudin’s sense.
(b) If f : X . Y is a map between metric spaces and p a point of X, show that the following
conditions are equivalent:
(i) The map f is continuous at p.
(ii) The inverse image under f of every neighborhood of f ( p) contains a neighborhood of p.
(iii) The inverse image under f of every ‘‘topologist’s neighborhood’’ of f ( p) is a ‘‘topologist’s
neighborhood’’ of p .
(c) State similarly how the concepts of a limit-point of a set and the limit of a sequence can be formulated
in terms of the topologist’s concept of neighborhood. (No proofs asked for.)
4.2:9. At-most-countably-many-to-one continuous maps, and perfect sets. (d: 3. > 2.4:2)
Let us call a function f : X . Y ‘‘at most countably many to one’’ if for every y )Y, the set
–1
f (y) # X is at most countable. Suppose f is a continuous, at most countably many to one map of
metric spaces. Show that for every compact perfect subset E # X, f (E ) is a perfect subset of Y.
(This exercise might be given with one of the next two sections, since the above result nicely parallels
the results of those sections on how continuous maps behave on compact sets and connected sets.)
4.2:10. k-dimensional space-filling curves. (d: 2, 1, 3)
In 7: R 14 (p.168), Rudin will show how to construct a continuous map > from [0,1] onto [0,1] 2 =
{ (x, y) x, y )[0,1]}. This is called a ‘‘space-filling curve’’ because a continuous function from [0,1]
into the plane can be thought of as a parametrized curve, and this curve fills up all the space inside the
square [0,1] 2. Assuming this result, we shall show here, using only the methods of this section, that there
must also exist curves filling higher-dimensional boxes, and even something that can be looked at as an
infinite-dimensional space-filling curve.
Given a continuous map > from [0,1] onto [0,1] 2, let us write >(t) = (a(t), b(t)) for each
t )[0,1]. (Rudin writes (x(t), y(t)), but I will be using x and y for real numbers.) Thus, a and b
are continuous functions [0,1] . [0,1] such that for every point (x, y) )[0,1] 2 there exists t )[0,1] such
that (a(t), b(t)) = (x, y).
(a) Assuming such functions a and b given, deduce that for every point (x, y, z) )[0,1] 3 there exists
t )[0,1] such that (a(t), a(b(t)), b(b(t))) = (x, y, z). Conclude that the function taking t to
(a(t), a(b(t)), b(b(t))) is a continuous function from [0,1] onto [0,1] 3 (a 3-dimensional space-filling
curve).
(b) Generalize the argument of part (a) to show that if we write b i for the i-fold composite function
b ° b ° ... ° b (with i factors), then for every n * 2 the function [0,1] . [0,1] n taking t to
(a(t), a(b(t)), ... , a(b n–2(t)), b n–1(t)) is continuous and onto (an ‘‘n-dimensional space-filling curve’’).
(Remark: In Rudin, for f a real-valued function, f n usually means the function defined by f n (x) =
f (x) n. Hence that notation holds in this exercise packet unless the contrary is stated. In this exercise, I
have stated the contrary!)
(c) Deduce that for every sequence (xn ) of elements of [0,1], there exists t )[0,1] such that
a(b n–1(t)) = xn for all n * 1.
(Recall that a sequence (xn ) involves infinitely many terms, x1 , x 2 , ... . We understand b 0 to
denote the identity map of [0,1], given by b 0(t) = t for all t. Thus, the n = 1 and n = 2 cases of the
above equation are a(t) = x1 and a(b(t)) = x 2 .)
Hint: Deduce from part (b) that for every N, the set of t for which the first N of the above
- 47 -
done 4: R 3, by showing that the function taking x )X to d ( f (x), g(x)) )R is continuous, and noting that
E is the zero-set of this function.)
(b) Show that (a) implies the result of 4: R 3.
(c) Deduce the main result of 4: R 4 – the final non-parenthetical statement, beginning ‘‘If g( p) = f ( p) ...’’
– from (a).
(d) Assuming the results of 4: R 20 (whether or not you have done that exercise), prove the following
converse to (a) above: For every closed subset E of a metric space X, there exist continuous functions
f and g from X to another metric space Y such that {x )X | f (x) = g(x)} = E.
4.2:16. Functions which approach a limit everywhere. (d: 3)
Suppose f : X . Y is a function between metric spaces (not assumed continuous) such that for every
p )X, lim x. p f (x) exists. Define g : X . Y by g( p) = lim x. p f (x) for all p )X. Show that g is
continuous.
(There are several ways this result can be strengthened. The above assumption that the limit of f is
defined at every p )X necessitates that X have no isolated points. One can weaken this to say that the
limit of f exists at every non-isolated point of X, defining g as above at those points, while making it
agree with f at the isolated points; one can then establish the same conclusion as above. One can also
merely assume f to be defined on a dense subset E of X; this still allows it to have a limit at every
non-isolated point of X, so that one can define g on all of X as above, and again get the same
conclusion.)
(c) Let f be a continuous real-valued function on a metric space X. If there exists a point p )X such
that f ( p) = sup x )X f (x), then X is compact.
(d) The function f : (0 , 1) . R defined by f (x) = 1 ⁄ x is uniformly continuous.
(e) The function f : (1, + 5 ) . R defined by f (x) = 1 ⁄ x is uniformly continuous.
(f) The function f : (0, + 5 ) . R defined by f (x) = 1 ⁄ x is uniformly continuous.
(g) If f : R . R is uniformly continuous, then f assumes a maximum value, i.e., there exists a real
number a such that f (a) = sup x )R f (x).
(h) If X is a noncompact metric space, then there exists a real-valued function f on X which is
uniformly continuous but not continuous.
(i) A subset E # R is compact if and only if every continuous real-valued function on E is bounded.
4.3:1. On what metric spaces is every function continuous, respectively uniformly continuous? (d: 3)
(a) What metric spaces X have the property that every function from X to any metric space Y is
continuous?
(b) What metric spaces X have the property that every function from X to any metric space Y is
uniformly continuous?
4.3:2. Continuous periodic functions on R are uniformly continuous. (d: 2)
(a) Let c be a positive real number, and f : R . X a continuous function from R to a metric space X
which has period c, i.e., such that f (r + c) = f (r) for all r )R. Show that f is uniformly continuous.
So, for instance, the function sin x is uniformly continuous. However
(b) Show that the function sin (x 2) is not uniformly continuous. (Rudin has not developed the sine
function as of this point. However, all you need to know to do (b) is that sin x is a nonconstant periodic
continuous real-valued function on R.)
4.3:3. A continuous map on R k takes bounded sets to bounded sets. (d: 2)
Let X be a metric space and f : R k . X a continuous function. Show that if E is a bounded subset
of R k , then f (E ) is a bounded subset of X.
4.3:4. A shrinking map on a compact space has a fixed point. (d: 4, 2, 2, 3, 4, 4)
(a) Suppose K is a compact metric space, and f : K . K is a map with the property that for every pair
of distinct points x, y )K one has d( f (x), f (y)) < d(x, y). Show that there exists a unique point p )K
such that f (p) = p.
(The remaining parts, though of interest, can be omitted without detracting from the interest or
challenge of the above problem.)
(b) Show by example that if the ‘‘<’’ is weakened to ‘‘!’’ in part (a), the map need not have any fixed
point.
(c) Show by examples that the result of (a) is also false if we put any of the noncompact metric spaces
[0,1), [0, + 5 ) or R in place of the compact space K.
(d) Deduce from (a) that for K a compact metric space, there cannot exist a continuous function f :
K . K such that for all distinct points x, y )K one has d( f (x), f (y)) > d(x, y).
(e) Show that in the situation of (a), for every x )K one has lim n. 5 fn (x) = p.
(f) Suppose K is a compact subset of a metric space X, and g : K . X is a map such that g(K ) ? K,
and such for every pair of distinct points x, y )K one has d(g(x), g(y)) > d(x, y). Show that there exists
a unique point p )K such that g( p) = p.
(For some further related results, see 4.3:8.)
4.3:5. Uniform ‘‘either/or continuity’’ of two or more functions. (d: 3)
(a) Let X, Y and Z be metric spaces, with X compact, and let f : X . Y, g : X . Z be functions.
We shall not assume f and g are continuous, but let us assume that for each p )X, at least one of f
- 50 -
and g is continuous at p. Show that for every 8 > 0 there exists a 7 > 0 such that whenever
d( p, q) < 7 in X, one has either d( f (p), f (q)) < 8 or d(g (p), g(q)) < 8 .
(b) Does a similar result hold for an infinite family of maps fi : X . Yi (i )I ) ? More generally, if we
have such a family of maps, and also an arbitrary family of positive real numbers 8 i (i )I ), can we find a
single 7 such that whenever d( p, q) < 7 in X, there exists some i )I such that d( fi (p), fi (q)) < 8 i ?
4.3:6. A weak condition that implies boundedness. (d: 3,1,2)
(a) Suppose E is a subset of a compact metric space X, and f is a function from E to a metric space
Y which satisfies the following condition much weaker than uniform continuity: For some 8 > 0 there
exists 7 > 0 such that all p, q )X that satisfy dX ( p, q) < 7 also satisfy dY ( f ( p), f (q)) < 8 . Show that f
must be bounded. (Note: f is not assumed continuous. Hint: Construct a certain open covering of E.)
(b) Deduce 4: R 8 from (a).
(c) Find a bounded (but, necessarily, non-compact) metric space X and a uniformly continuous function
f from X to another metric space Y such that f is not bounded.
4.3:7. More on functions which approach a limit everywhere. (d: 4)
Suppose f : X . Y is a function between metric spaces (not assumed continuous) such that for every
p )X, lim x. p f (x) exists. (This condition was considered in 4.2:16, but the present exercise does not
depend on that one.) Let D # X be the set of points where f is discontinuous. Show that for every
compact subset K # X, the set D - K is countable.
(Suggestion: For p )X, let w( p) = lim8 . 0 diam( f (N8 ( p))). Show that for every positive constant
c, if { p )X | w( p) > c} had a limit point q )X, then lim x. q f (x) could not exist, a contradiction.
Deduce that for each positive integer n there are only finitely many p )K with w( p) > 1 ⁄ n.)
4.3:8. Weakly shrinking maps on compact spaces. (d: 5)
(a) Suppose K is a compact metric space, and f : K . K is a map with the property that for all x, y )K
one has d( f (x), f (y)) ! d(x, y). (Cf. 4.3:4.) Show that f is surjective if and only if for all x, y )K one
has d( f (x), f (y)) = d(x, y).
(b) Deduce from (a) that if K is a compact metric space and g : K . K a continuous map with the
property that for all x, y )K one has d( f (x), f (y)) * d(x, y), then for all x, y one has d( f (x), f (y)) =
d(x, y).
Answers to True/False question 4.3:0. (a) T. (b) T. (c) F. (d) F. (e) T. (f) F. (g) F. (h) F. (i) T.
- 51 -
4.4:2. The closed half-line is not like the whole line. (d: 3)
(a) Show that there is no one-to-one continuous map from [0, + 5 ) onto R.
(b) Show that there is no one-to-one continuous map from R onto [0, + 5 ).
(c) Deduce from (a) and (b) the corresponding results with [0, 1) in place of [0, + 5 ).
(Exercise 4.2:13 showed that in some ways [0, 1) was like the real line. This exercise shows that this
similarity only goes so far.)
4.4:3. A space-filling curve cannot be one-to-one. (d: 4)
In some previous exercises we have discussed properties of space-filling curves. (See in particular first
paragraph of 4.2:10.) We shall show here that such a curve cannot be one-to-one.
(a) Let n > 1. Show that for every point p of [0,1] n, the subset obtained by removing p from [0,1] n
is still connected. On the other hand, point to a result in Rudin showing that [0,1] does not have this
property.
(b) Show how the existence of a one-to-one continuous map from [0,1] onto [0,1] n, together with the
above pair of facts, would lead to a contradiction. (Hint: What do you know about one-to-one continuous
maps of one compact metric space onto another?)
4.4:4. Continuous functions on disconnected sets. (d: 2, 1, 3)
In 2.5:1(a) we saw that a metric space X is connected if and only if the only subsets of X that are
both open and closed are X and :.
(a) Deduce from that result that X is disconnected if and only if there exists a family of nonempty open
subsets Ei # X (i )I ) such that X = , i )I Ei , the sets Ei are pairwise disjoint (i.e., i 4 j /
Ei - Ej = :), and I has more than one element. (Suggestion: Think first about the case where I has
two elements.)
(b) Show that if X and the Ei satisfy the conditions shown in (a), then a function f from X to a
metric space Y is continuous if and only if for each i )I, the restriction f of to Ei is continuous.
(The restriction of f to Ei means the function f | E : Ei . Y defined by f | E (x) = f (x) for all x )Ei .)
i i
(c) Show, conversely, that if (Ei ) i )I is a family of pairwise disjoint subsets of X whose union is all of
X, and such that every function f from X to any other metric space Y such that the restriction of f to
each Ei is continuous is itself continuous, then all Ei are open.
(f) Suppose f, g : R . R are functions such that lim x . +5 f (x) = + 5 and lim x . +5 g(x) = – 5 .
Then lim x . +5 f (x) + g(x) exists.
4.7:1. A function on R that has limits at ± 5 is uniformly continuous. (d: 2)
Let f : R . R be a continuous function such that lim x . +5 f (x) and lim x . – 5 f (x) both exist.
Show that f is uniformly continuous.
(Note that, following Rudin’s conventions, the statement that the limits both exist means that they are
real numbers. If f approached + 5 , as x approached + 5 , for instance, Rudin would write
‘‘f (x) . + 5 as x . + 5 ’’, but he would not write ‘‘lim x . +5 f (x) = + 5 ’’, and would not say that
lim x . +5 f (x) exists.)
4.7:2. A continuous map asymptotic to a uniformly continuous map is uniformly continuous. (d: 2, 3)
For what we want to prove, we first need to strengthen Theorem 4.19:
(a) Let f : X . Y be a continuous map between metric spaces, and K a compact subset of X. Prove
that for every 8 > 0 there exists a 7 > 0 such that for all p )K and x )X with d ( p, x) < 7 , one has
d ( f ( p), f (x)) < 8 . (If you wish, you may simply note, in precise detail, changes in the proof of Theorem
4.19 that will yield the above result.)
(b) If f, g : X . Y are two maps between metric spaces, let us say f and g are ‘‘asymptotic’’ to one
another if for every 8 > 0, the set {x )X d( f (x), g(x)) * 8 } is contained in a compact subset of X.
Show that any continuous map which is asymptotic to a uniformly continuous map is uniformly continuous.
Then show that the result of 4.7:1 above follows from this result.
4.7:3. Cauchy criteria for limits of functions. (d: 2)
(a) Suppose f is a complex-valued function on a neighborhood of + 5 , (c, + 5 ). Show that
lim x . +5 f (x) exists if and only if only if for every 8 > 0 there exists a real number M * c such that
for all x, y )(M, + 5 ) one has | f (x) – f (y)| < 8 .
State a similar criterion for a complex-valued function defined on a neighborhood of – 5 to have a
limit as x . – 5 . (Since the proof is almost identical, I don’t ask you to write it out.)
(b) Suppose f is a complex-valued function defined on a subset E of a metric space X, and let p be a
limit point of E in X. State and prove an analogous ‘‘Cauchy criterion’’ for lim x . p f (x) to exist.
(c) Show by example(s) that one or more of the above criteria fail if ‘‘complex-valued function’’ is
replaced by ‘‘function into a metric space Y ’’. For which metric spaces Y will those results hold?
Chapter 5. Differentiation.
Answers to True/False question 4.5:0. (a) F. (b) T. (c) T. (d) F. Answers to True/False question 4.6:0. (a) T.
(b) F.
- 54 -
assigned here. It might also be assigned with section 5.3, in view of its relation to the topic of that section.
One could at this point likewise look at the first question asked in 5: R 21, about getting an f which is
differentiable; but since the remaining parts of the question refer to higher-order differentiability, I have left
that question to section 5.5.
Exercise not in Rudin:
5.1:0. Say whether each of the following statements is true or false.
(a) If a bounded function R . R is continuous, then f ((t) exists for all real numbers t.
(b) If f is a differentiable real-valued function on [0, 2] and lim t . 1 f (t) = 7, then f (1) = 7.
(c) If f and g are real-valued functions on [a, b], then at any point where f and g are both
differentiable, f +g is also differentiable.
(d) If f and g are real-valued functions on [a, b], then at any point where f and f +g are both
differentiable, g is also differentiable.
Answers to True/False question 4.7:0. (a) F. (b) T. (c) T. (d) F. (e) F. (f) F.
- 55 -
5.2:3. Functions that are differences between two increasing functions. (d: 2, 1, 3)
If g1 and g 2 are increasing functions on an interval, then their difference, g1 – g 2 , need not be
either increasing or decreasing. It is natural to ask how wide a class of functions can be written as such a
difference. Parts (a) and (b) below show easy classes of examples which can and which cannot be so
written; part (c) concerns a more subtle case.
(a) Show that if f is a differentiable function on [a, b] whose derivative is zero at only finitely many
points, then f can indeed be written as the difference of two increasing functions.
(b) Show that if f is an unbounded (hence, necessarily, discontinuous) function on [a, b], then it cannot
be written as the difference of two increasing functions.
(c) Let f be the function on [0, 1] such that f (x) = x 2 sin x –2 if x 4 0, while f (0) = 0. By the same
reasoning as in Example 5.6(b) (p.106), f (x) is everywhere differentiable. (This is immediate, so you are
not asked to prove it.) However, show that f cannot be written as the difference g1 – g 2 of two
increasing functions. (Suggestion: Consider how much f decreases on each interval
– 12 – 12
[(2 @ n) , (2 @ (n – 4))
1 ]. Sum over n. What can you conclude about g 2 ?)
5.2:4. A mean-value theorem with possibly infinite end-points. (d: 3)
Suppose – 5 ! a < b ! + 5 , and f is a differentiable function on (a, b) such that lim x . a f (x) =
lim x . b f (x). (Note that by Rudin’s conventions, writing this entails that the two limits exist, and are real
numbers, not ± 5 . On the other hand, the beginning of the above sentence indicates that a and b
themselves may be ± 5 .)
Show that there exists c )(a, b) such that f ((c) = 0.
5.2:5. A condition for f (a+) to exist. (d: 2)
Let f be a differentiable function on (a, b). Show that if f ( is bounded, then lim x . a f (x) exists.
(Suggestion: Use 4.1:3(a) above.)
nth derivative is zero at exactly n points, but the second does not. (To get a feel for the problem, you
might begin by sketching for yourself the functions and their first two or three derivatives. You will be
able to prove that exact equality does not hold for one of the functions by finding > n zeroes of f (n)(x)
for a particular value of n. In proving that it does hold for the other two, you may use the fact that a
polynomial equation of degree d, ad x d + ... + a1 x + a 0 = 0, has at most d solutions.)
(d) If one looks for other functions f which satisfy the conditions of (a) and whose nth derivative is zero
at only n points for every n, one easily sees that this is true of the trivial variants of the above two
2
examples, A ⁄ ((Bx + C ) 2 + 1) and A e – (Bx + C ) , for all real numbers A, B, C with A and B
nonzero. In each case, this is a 3-parameter family of functions. Can you find any functions with these
properties that does not belong to one of these families? Any families of such functions with more than
three parameters? If so, how large can you make the number of parameters? (By an n-parameter family
let us understand a set of functions fA , A , ... , A (x) depending on real numbers A1 , A2 , ... , An such
1 2 n
that if (A1 , A2 , ... , An ) 4 (A1(, A2(, ... , An( ), then the corresponding functions are distinct.)
5.5:2. The relation between boundedness of f ( and uniform continuity of f. (d: 1, 2, 3, 4)
Let f : R . R be a differentiable function.
(a) Show that if f ( is bounded, then f is uniformly continuous.
(b) Show by example that the converse is not true. (Suggestion: Find a function that is uniformly
continuous by 4.7:1, but whose derivative is unbounded because the function wiggles rapidly for large
values of x.)
(c) Show that if an example such as you are asked for in (b) is twice differentiable, then its second
derivative must also be unbounded. Equivalently (in view of (a)), show that a twice differentiable function
R . R whose second derivative is bounded is uniformly continuous if and only if its first derivative is
bounded.
(d) Can one strengthen (c) to say for every integer n > 1 that if f : R . R is an n times differentiable
function which is uniformly continuous, but such that f ( is unbounded, then f ( (, f (3), ... , f (n) must all
be unbounded?
5.5:3. Lagrange interpolation. (d: 3)
(a) Let f be a function on [a, b] which is n times differentiable, i.e., such that f (n) exists, and
suppose there are at least n +1 points x on [a, b] such that f (x) = 0. Show that there is at least one
point f on [a, b] such that f (n)(x) = 0.
(Suggestion if you have trouble getting started: Draw a sketch of such an f for n = 3, and see how
many points x in your picture satisfy f ((x) = 0. Can you prove there must be that many, or come up
with a picture in which there are fewer? Once you can prove something about the number of zeroes of f (,
see whether you can get from this a conclusion about the number of zeroes of f ( (, and so on.)
(b) Let f be any function on [a, b] and let x 0 , ... , xn be distinct points of [a, b]. Show that there
exists a polynomial p(x) of degree ! n such that p(xi ) = f (xi ) for i = 0, ... , n.
(Hint: If you have found a polynomial that agrees with f at x 0 , ... , xi –1 , show that by adding a
scalar multiple of (x – x 0 ) ... (x – xi –1 ) you can get a polynomial that agrees with f at x 0 , ... , xi .)
(c) Let f and p be as in part (b). Because p is a polynomial of degree ! n, its nth derivative p (n)
is a constant c. Show with the help of (a) above that if f is n times differentiable, then for some
x )[a, b], f (n)(x) = c.
(d) Let f be as in part (b), and let q be the polynomial of degree ! n –1 which agrees with f at the n
points x 0 , ... , xn –1 . Then the number E = f (xn ) – q(xn ) represents the error resulting when we use this
polynomial to approximate f at the n+1st point xn ; also, this number E determines what multiple of
(x – x 0 ) ... (x – xn –1 ) must be added to q as in the hint to part (b) to make equality at xn hold. Write
down the formula that describes the polynomial of degree ! n agreeing with f at all n +1 points in
terms of q and E, take its nth derivative, and apply the result of (c). Conclude that if we know a bound
on f (n) on [a, b], then we can bound the error arising when we use q to approximate f at xn .
(Remark: The polynomial q is called a ‘‘Lagrange interpolation polynomial’’ for f, and the result
of (d) is the ‘‘remainder formula for Lagrange interpolation’’.)
(a) Show that for every polynomial p(x), as x . + 5 we have p(x) e –x . 0. (Hint: Use L’Hospital’s
rule and induction on the degree of p.)
–2
(b) Sketch an argument showing from this that for every polynomial p(x), lim x . 0 p(x –1) e –x = 0.
(I say ‘‘sketch’’ because, although Rudin proves in Theorem 4.7 that for continuous functions f and g
with appropriate domains and codomains, the composite function g f is also continuous, he does not
develop general results on limits of composite functions; in particular limits at infinity; and although it is
not too difficult to prove such results, it would be time-consuming to have you do so here. So simply
sketch how these functions behave, assuming that in this example, limits of composites behaves as one
would expect. Note that ‘‘x . 0’’ involves values both above and below 0.)
–2
We can now give the example. Let f : R . R be defined by f (x) = e –x if x 4 0, and f (0) = 0.
(c) Show that on R – {0}, f has derivatives f (n) for all nonnegative integers n, and that for each n,
–2
f (n)(x) has the form pn (x –1) e – x for some polynomial pn (x).
(d) Show that for every nonnegative integer n, f (n)(0) is also defined, and equals 0. Now compute the
Taylor polynomials shown in display (23) on p.110 for + = 0. Show that these converge as n . 5 , but
that the limit is not the original function f (t).
–2
(e) As a variant of the above, suppose we define a function g by g(x) = e –x if x > 0, and g(0) = 0
if x ! 0. Show that g is also infinitely differentiable, and is given by its Taylor series for all negative x,
but not for any positive x.
5.6:2. Lagrange interpolation with multiplicities. (d: 4. > 5.5:3)
Suppose f is a function on [a, b], and for some x )[a, b] and n * 0, f is m –1 times differentiable
at x, and f (x) = f ((x) = ... = f (m –1)(x) = 0. Then one says that f ‘‘has a zero of multiplicity at least
m’’ at x. (For instance, this is the behavior of a polynomial p(t) that is divisible by (x – t) m.) If there
are points x1 , ... , xr in [a, b] and positive integers m1 , ... , mr such that f has a zero of multiplicity at
least m1 at x1 , a zero of multiplicity at least m 2 at x 2 , etc., then writing n = m1 + ... + mr , we say
that ‘‘f has at least n zeroes on [a, b], counting multiplicities’’.
Prove the analog 5.5:3(a) with the condition that f be zero at at least n +1 distinct points replaced by
the condition that it have at least n +1 zeroes counting multiplicities, and use this to get results similarly
generalizing parts (b), (c) and (d) of that exercise. Show that Taylor’s Theorem (Theorem 5.15, p.110) is a
case of the analog of part (d) of that exercise.
I’ve made a ‘‘section’’ out of the first two pages of this chapter, which discusses the Riemann integral,
because the definition of the Riemann-Stieltjes integral involves many concepts difficult for the students,
and those two pages ‘‘ease one into’’ the subject. I break in two the remainder of Rudin’s first section of
this chapter, and likewise Rudin’s second section, because each is long and contains a number of diverse
concepts. After that, the sections below coincide with Rudin’s.
6.1. The Riemann integral (beginning of Rudin’s section DEFINITION AND EXISTENCE OF THE
INTEGRAL). (pp.120-121)
Relevant exercises in Rudin:
6: R 2. The only continuous positive function with integral 0 is the zero function. (d: 2)
Rudin ends by contrasting this with 6: R 1; but that exercise is stated for the Riemann-Stieltjes integral,
- 62 -
and so can’t be given after just covering these two pages. Hence I give as 6.1:1 below a version of that
exercise which refers only to the Riemann integral, and can be assigned in conjunction with this exercise.
6: R 4. The function that is 1 at rationals and 0 at irrationals is not Riemann integrable. (d: 1)
Exercise not in Rudin:
6.1:1. A function that is zero except at one point is integrable, with integral zero. (d: 1)
Let f be a function on an interval [a, b] which is zero everywhere except at one point c )(a, b).
b
Prove that f ) , and that B f (x) dx = 0.
a
6.2. The Riemann-Stieltjes integral (middle of Rudin’s section DEFINITION AND EXISTENCE OF
THE INTEGRAL). (pp.122-125)
Relevant exercises in Rudin:
6: R 1. Riemann-Stieltjes integrability of a function zero except at one point. (d: 1)
6: R 3. Integration with respect to three step functions: one right continuous, one left continuous, and one
that splits the difference. (d: 2)
I would add to this exercise two more parts:
(e) Given an interval [a, b] and a point c )[a, b], let 0 j,a,b,c ( j = 1, 2) be the functions on [a, b]
which equal 0 on [a, c), 1 on (c, b], and are 0 or 1 at c depending on whether j is 1 or 2.
(So for j = 1, 2, Rudin’s 0 j is 0 j, –1,1, 0 in this notation.) State the results analogous to Rudin’s (a)
b
and (b) for integrals B f (x) d 0 j,a,b,c (x). Do not hand in proofs. But think things through carefully; you
a
will be graded on the correctness of your answers.
The cases c = a and c = b are slightly different from the general case, so I recommend that you first
state the results for c )(a, b), then say how they must be modified for those cases.
(f) Assuming the results of part (e), deduce that a function f on an interval [a, b] is integrable with
respect to every increasing function + if and only if it is continuous.
Exercises not in Rudin:
6.2:0. Say whether each of the following statements is true or false.
(a) In the equation U (P, f, + ) = 3 ni=1 Mi C + i (used in Rudin’s definition of the Riemann-Stieltjes
integral), Mi denotes f (xi ).
(b) In the same equation, n denotes the number of intervals [xi – 1 , xi ] into which the partition P
divides [a, b].
(c) If a partition P* contains more points than a partition P, then P* is a refinement of P.
(d) If P* is a refinement of the partition P of the interval [a, b], then for every bounded function f
and increasing function + on [a, b], U (P*, f, + ) – L (P*, f, + ) ! U (P, f, + ) – L (P, f, + ).
(e) If f is a bounded function and + an increasing function on [a, b], and if there exist partitions P1 ,
P2 , ... of [a, b] such that for each i, U (Pi+1 , f, + ) – L (Pi+1 , f, + ) ! 12 (U (Pi , f, + ) – L (Pi , f, + )), then
f ) (+ ).
(f) If f is a bounded function and + an increasing function on [a, b], and if P1 # P2 # ... are a
–
sequence of partitions, each a refinement of the one before, then inf n = 1, 2, ... (U (Pn , f, + )) = B f d + .
6.2:1. Riemann-Stieltjes integrability is symmetric. (d: 1)
If + and 0 are increasing functions on [a, b], show that the following conditions are equivalent:
(i) + ) (0 ). (ii) 0 ) (+ ). (iii) For every 8 > 0 there exists a partition P of [a, b] such that
3 C + i C 0 i < 8 , where C + i and C 0 i are defined with respect to the partition P as in Definition 6.2
(p.122).
6.2:2. The Riemann integral as a limit over partitions with mesh approaching 0. (d: 3)
If P = {x 0 , ... , xn } is a partition of [a, b], let || P || = sup C xi ; i.e., the maximum gap between
successive points of the partition, often called the ‘‘mesh’’ of the partition. In some texts one sees the
Riemann integral defined as a limit over partitions whose mesh approaches 0. This exercise will show that
such a definition is equivalent to the one given in Rudin. We first need an observation.
(a) Suppose f is a function on [a, b] and M a constant such that | f (x)| ! M for all x. Show that if
P is a partition of [a, b], and P* a refinement obtained by adding a single point to P, then U(P, f ) –
U(P*, f ) ! 2 M || P ||, and similarly that L(P*, f ) – L(P, f ) ! 2 M || P ||.
(b) Show that the following conditions on a bounded function f on [a, b] are equivalent:
(i) f ) .
(ii) For every 8 > 0 there exists a 7 > 0 such that for every partition P of [a, b] of mesh < 7 ,
one has U(P, f ) – L(P, f ) < 8 .
Suggestion for (i)/(ii): Choose by (i) a partition P0 for which the difference between the upper and
lower sums is < 8 ⁄ 2; then show using part (a) that for any partition P of sufficiently small mesh,
U(P , P0 , f ) – L(P , P0 , f ) differs from U(P, f ) – L(P, f ) by < 8 ⁄ 2.
6.2:3. The Riemann-Stieltjes integral as a limit over partitions with ‘‘+ -mesh’’ approaching 0. (d: 2, 4.
> 4.6:1, 6.2:2)
Let + be an increasing function on [a, b]. We would like to get a translation, similar to that of the
preceding exercise, for the property of integrability with respect to + . One might think that the logical
analog of the mesh used there would be the ‘‘+ -mesh’’ || P ||+ defined as sup i C + i . But with this
definition, if + has discontinuities one could never get a partition P with + -mesh smaller than the
largest step by which + jumps. So let us instead define the + -mesh by
|| P ||+ = sup i (+ (xi –1 +) – + (xi –)).
(a) Show that for every 8 > 0 there exist partitions of [a, b] of + -mesh < 8 . (Suggestion: Use 4.6:1.)
Show moreover that such a partition must contain every point x such that + (x+) – + (x –) > 8 .
(b) Show that the following conditions on a function f on [a, b] are equivalent:
(i) f ) (+ ).
(ii) For every 8 > 0 there exists a 7 > 0 such that for every partition P of [a, b] of + -mesh < 7 ,
one has U(P, f, + ) – L(P, f, + ) < 8 .
This is not as easy as part (b) of the preceding exercise. The difficulty is that the effect of inserting an
additional point in a partition will depend on whether it is inserted in an interval that has a ‘‘large’’
discontinuity of + ; if so, then one has to take into account the behavior of f near that discontinuity, and
not just general bounds on the absolute value of f .
6.3. Conditions for integrability (end of Rudin’s section DEFINITION AND EXISTENCE OF THE
INTEGRAL). (pp.125-127)
Relevant exercises in Rudin:
6: R 6. Discontinuities limited to the Cantor set can’t interfere with Riemann integrability. (d: 3)
6: R 7. Improper integrals of the first kind. (d: 2)
6: R 8. Improper integrals of the first kind, and the integral test for convergence of series. (d: 3.
> 4.7:3(a))
The result of this exercise is important, but the exercise is probably not good to give as homework
since most students should be able to find proofs, of varying quality, in their lower division calculus texts.
One way to get around this would be to hand out a sketchy proof taken from such a text, and ask students
to justify specified steps using results from Rudin.
Answers to True/False question 6.2:0. (a) F. (b) T. (c) F. (d) T. (e) T. (f) F.
- 64 -
Exercise 4.7:3(a) supplies a tool that Rudin neglected to develop for showing existence of limits at
infinity, which is needed for this exercise.
Exercises not in Rudin:
6.3:0. Say whether the following statement is true or false.
(a) If | f (x)| ! | g(x)| for all x )[a, b], and g ) (+ ), then f ) (+ ).
6.3:1. The obvious formula for B 1 d + . (d: 1)
Show (from the definition of the integral) that for any increasing function + on an interval [a, b],
b
one has B 1 d + = + (b) – + (a). (This should probably have been made part of Theorem 6.12.)
a
6.3:2. Functions whose values are ‘‘mostly’’ zero have integral zero. (d: 2)
(a) Let f : [0,1] . R be the function of 4: R 18 (p.100) which takes the value 0 at all irrationals, and the
value 1 ⁄ n at a rational number whose expression in lowest terms is m ⁄ n. Show that for every
1
continuous increasing function + on [0,1] one has f ) (+ ), and B f d + = 0. (You may do this by
0
doing part (b) below, if you choose.)
(b) Show that the same conclusion is true of any function f on [0,1] with the property that for every
8 > 0, the set {x )[0,1] | f (x)| * 8 } is finite. Show that the function f of part (a) has that property.
6.3:3. An + which does all its increasing on the Cantor set. (d: 4, 3, 2, 1, 3, 1)
Let P denote the Cantor set, and let us define a function + 0 on the complement of P in [0,1] as
follows. For all x in the segment (1 ⁄ 3, 2 ⁄ 3), i.e., the segment one deletes at the first step in constructing
the Cantor set, let + 0 (x) = 1 ⁄ 2. At all points of the segments (1 ⁄ 9, 2 ⁄ 9) and (7 ⁄ 9, 8 ⁄ 9), the two
segments one deletes at the second stage in the construction of the Cantor set, let + 0 (x) have the values
1 ⁄ 4 and 3 ⁄ 4 respectively. Similarly, on the 2 n –1 segments deleted at the nth stage of the construction
of the Cantor set, let + 0 have the constant values 1 ⁄ 2 n, 3 ⁄ 2 n, ... , (2 n–1) ⁄ 2 n (each one half-way
between the values previously assigned on the two surrounding segments; or in the case of the first and last
of these segments, half-way between the value at the adjacent previously assigned segment and the value
0, respectively 1). Now
(a) Show that + 0 can be extended to a function + on all of [0,1] so as to give a continuous increasing
function. (Suggestion: Show that the set on which we have defined + 0 is dense in [0,1], and that
+ 0 (x –) and + 0 (x+) are defined and equal for all x )[0,1], and deduce the result using these facts.)
(The function + can also be constructed as follows: Given x )[0,1], write it in base 3 notation. If it
has any digit ‘‘1’’, change the first such digit to a ‘‘2’’, and all digits after it to ‘‘000...’’. In the resulting
string of 0’s and 2’s, change all 2’s to 1’s, and regard the result as an expression for + (x) in base 2. But
if you use this description, you must prove it equivalent to the one stated above.)
(b) For + as in part (a), show that if f is a continuous real-valued function on [0,1] which is zero on
all points of the Cantor set, then B f d + = 0.
(c) Give an example of a function f as in (b) which is not the zero function.
(d) Deduce from (b) that if f and g are two continuous real-valued functions on [0,1] which agree on
all points of the Cantor set, then B f d + = B g d + .
(e) Let f : [0,1] . R be the function such that f (x) = 0 for all x in the Cantor set, and f (x) = 1 for
all other x. Show that f ) (+ ).
(f) Why does the result of (e) not contradict the result of (b) above? Why does not it contradict the result
of 6: R 6 (p.138)?
6.3:4. An integrable function of an integrable function need not be integrable. (d: 2)
Show by example that, in contrast to Theorem 6.11, if f and D are Riemann-integrable functions,
their composite D ° f need not be. Suggestion: Let f be as in 4: R 18, and choose D to be discontinuous
at the real number which f most often approaches. (Examples are even known where f is continuous;
but they are more difficult to describe.)
- 65 -
6.3:5. Condition for an increasing function to be Riemann-Stieltjes integrable. (d: 4. > 4.3:5(a))
Prove that f ) (+ ) if f is increasing and is continuous at every point where + is discontinuous.
(Suggestion: Combine the ideas of the proofs of Theorems 6.8 and 6.9, using 4.3:5(a). Incidentally, the
idea of the first displayed formula in the proof of Theorem 6.9 is that the numbers C + i are all small.
You won’t be able to get a precise formula like that one in your proof, but you will want to use the same
idea.)
6.3:6. A condition for mutual integrability of + and 0 . (d: 4. > 6.2:1)
Show that the three conditions of 6.2:1 are also equivalent to: (iv) For every x )[a, b), either + (x) =
+ (x+) or 0 (x) = 0 (x+), and for every x )(a, b], either + (x) = + (x –) or 0 (x) = 0 (x –). (Suggestion:
assuming (iv) holds, let M = (+ (b) + 0 (b)) – (+ (a) + 0 (a)). Given 8 , find a partition such that in every
interval of the partition, either C + i or C 0 i is < 8 ⁄ 2 M. Deduce that C + i C 0 i ! (C + i + C 0 i ) (8 ⁄ 2 M ),
and sum these inequalities.)
6.3:7. ‘‘Since 8 is arbitrary ...’’ (d: 2)
(a) Find the fallacy in the following argument:
‘‘Theorem’’ Every function f : X . Y between metric spaces is continuous.
Proof Given any 8 > 0 and any x )X, take 7 = 1, and consider a point y 4 x of X satisfying
d (x, y) < 7 . Choose any C > d ( f (x), f (y)) ⁄ 8 . Multiplying this inequality by 8 , we get
d ( f (x), f (y)) < C 8 .
But since 8 was arbitrary, we can choose it to make this as small as we like, proving continuity.
(b) The ends of the proofs of Theorem 3.50 (pp.74-75) and Theorem 6.11 (p.127) use a similar argument
that since 8 is arbitrary, an expression having 8 as a factor can be made arbitrarily small, yielding a
desired conclusion. Why do those proofs not suffer from the same fallacy as above?
6.4. Basic properties (beginning of Rudin’s section PROPERTIES OF THE INTEGRAL). (pp.128-
130)
Relevant exercises in Rudin:
2 3
6: R 5. Does f ) or f ) imply f ) ? (d: 2)
6: R 10. Hölder’s inequality. (d: 3, 2, 4, 3)
The only way I can see to do part (a) is quite roundabout: Let u p = s, q = t, 1 ⁄ p = a, and hence,
by the first displayed equation, 1 ⁄ q = 1 – a. Turn the resulting inequality into an inequality concerning
s ⁄ t, and write s ⁄ t = r. The resulting inequality will be true for r = 1; prove it for general r using the
Mean Value Theorem, assuming the standard formula for the derivative of x a . (Without the above hint, I
would rate that part at least d: 4.)
Hint for part (c) in the case where neither of the integrals on the right is zero: Use a scalar
multiplication to reduce to the case where those integrals are 1, and apply (b).
I don’t know what method Rudin had in mind for the case where one or both of those integrals is zero.
It could be proved using results from Chapter 11, but these are not available yet. One method that will
work is to approximate f and/or g by functions for which the integrals in question are nonzero, and
show that the inequality for those approximating functions implies the same inequality for the limit
functions. Another proof of this case of (c) for the special case p = q = 2 is given in 6.4:2 below, and an
alternative development of the rest of part (c) for those values of p and q in 6.4:3.
Part (d ), of course, depends on 6: R 7 and 6: R 8. (I didn’t indicate this above because parts (a)-(c),
which form the core of the exercise, do not.)
6: R 11. The triangle inequality for the L 2 norm. (d: 1. > 6: R 10(c), or 6.4:2 and 6.4:3)
The ‘‘Schwarz inequality’’ that Rudin refers to here is not the result of that name which we saw in
Theorem 1.35, but the version of that result with integration replacing summation referred to in part (c) of
the preceding exercise. As the difficulty-rating indicates, this exercise is easy – assuming the difficult
exercise 6: R 10 discussed above, or the somewhat easier substitute exercises given below. Incidentally,
|| u ||2 is known as ‘‘the L 2 norm of u’’; hence the titles I have given this and the next exercise.
6: R 12. An integrable function is L 2 -approximable by a continuous function. (d: 3. > 6: R 11)
Exercises not in Rudin:
6.4:0. Say whether the following statement is true or false.
(a) If f ) (+ ) on an interval [a, b], then f ) (+ ) on every subinterval [c, d] # [a, b].
6.4:1. Integration with respect to + and + –1 . (d: 1)
A real-valued function + on a set E of real numbers is said to be strictly increasing if for all
x, y )E, x < y / + (x) < + (y). Clearly, such a function is one-to-one. If + is a strictly increasing
continuous function on an interval [a, b], the Intermediate Value Theorem shows that + is onto
[+ (a), + (b)], hence is a bijection from [a, b] to [+ (a), + (b)], hence it has an inverse function, + –1:
[+ (a), + (b)] . [a, b].
For such an + , show that for every real-valued function f on [a, b] we have
b + (b) –1
Ba f (x) d + (x) = B+ (a) f (+ (x)) dx,
in the sense that if either side is defined, so is the other, and they are then equal.
(This result is related to Theorem 6.19; but that is outside of section 6.4, so you can’t use it here.)
6.4:2. Functions that behave like zero under Riemann-Stieltjes integration. (d: 3)
b
(a) Suppose that f ) (+ ) on [a, b], and that B f (x) 2 d + (x) = 0. Prove that for all g ) (+ ) one also
b a
has B f (x) g(x) d + (x) = 0.
a
Suggestion: Fixing g, verify on general principles that for all real numbers t,
b 2
Ba ( f (x) + t g(x)) d + (x) * 0, and that as a function of t, this integral has a local minimum at t = 0.
Now expand the integral in terms of the integrals of f 2, f g and g 2, and draw a conclusion.
(b) Deduce that if f ) (+ ) on [a, b], and f (x) * 0 for all x, then the following conditions are
b b b
equivalent: (i) B f (x) d + (x) = 0. (ii) B f (x) 2 d + (x) = 0. (iii) B f (x) g(x) d + (x) = 0 for all g ) (+ ).
a a a
(Hint: Can you write f as the square of a function in (+ ) ?)
(c) Show that if from part (b) we delete the assumption that f (x) * 0 for all x, then the equivalence
still holds, with f (x) replaced by | f (x)| in statement (i), but not in statements (ii) and (iii). (Hint: f =
( | f | + f ) ⁄ 2 – ( | f | – f ) ⁄ 2.)
6.4:3. The Schwarz inequality for integrals. (d: 3. > 6.4:2)
The calculations by which we proved the Schwarz inequality for n-tuples of real numbers in 1.7:2 can
be mimicked using integrals instead of sums. Given an increasing function + , the analog of the dot
b
product for f, g ) (+ ) is Ba f g d + . Use the method of that exercise to obtain a Schwarz inequality for
such functions (namely, the bottom display on p.139 with p = q = 2), assuming the integral of the square
of each function is nonzero.
For vectors, the case of ‘‘zero norm’’ created no difficulty, because vectors of zero norm were zero;
however, the analogous statement for integrals is not true (cf. 6: R 1, p.138). Show, however, that that case
can be handled with the help of 6.4:2 above.
6.4:4. Description of (+ + 0 ). (d: 2)
Show that if + and 0 are increasing functions on [a, b], then (+ + 0 ) = (+ ) - (0 ).
6.4:5. Extending the Riemann-Stieltjes integral to the case of non-increasing + . (d: 3)
The definition of the Riemann-Stieltjes integral B f d + requires that + be an increasing function. In
this exercise we will see that, having defined such integrals for increasing + , we can extend the definition
in a natural way to the wider class of all functions that can be written as differences of increasing
functions. (Exercise 5.2:3 looked at that class of functions; but the result proved there is not needed for
this exercise.)
Suppose that f and E are functions on [a, b], and that we can express E as
- 67 -
E = +1 – + 2 ,
where +1 and + 2 are increasing functions, and f is integrable with respect to both +1 and + 2 .
Show that if we define
B f d E = B f d +1 – B f d + 2
then B f d E is well-defined, independent of our choice of decomposition of E as +1 – + 2 .
Thus, what you must prove is that if E can also be written E = 01 – 0 2 , where 01 and 0 2 are
increasing and f is integrable with respect to both 01 and 0 2 , then
B f d +1 – B f d + 2 = B f d 01 – B f d 0 2 .
(Suggestion: abbreviating B f d +1 – B f d + 2 to I (+1 , + 2 ), show that I (+1 , + 2 ) = I (+1 + 0 2 , + 2 + 0 2 )
= I (+ 2 + 01 , + 2 + 0 2 ) = I (01 , 0 2 ).)
6.4:6. Converse to Theorem 6.12 (c). (d: 2)
Show that if a < c < b, and + is a monotonically increasing function on [a, b], and f is a
function on [a, b] such that f ) (+ ) both on [a, c] and on [c, b], then f ) (+ ) on [a, b].
Thus the equality
c b b
B f d + + B f d + = B f d +,
a c a
which Rudin proves under the hypotheses of Theorem 6.12(c), also holds in this situation.
6.4:7. Integrable functions can have infinitely many big jumps. (d: 1, 3)
Let s1 < s 2 < ... < sn < ... be elements of an interval [a, b] such that s1 = a and lim n . 5 sn = b.
Let f : [a, b] . [0,1] be defined by the condition that for x )[si , si+1 ), f (x) = (x – si ) ⁄ (si+1 – si ),
while f (b) = 0.
(a) At what points x is f discontinuous, and at each of these points, what are the values of f (x),
f (x – ), and f (x+) if the latter two are defined? (No proof required for this part.)
(b) Assuming the properties noted in part (a), show that f ) .
The above example is generalized in the next exercise.
6.4:8. Integrability piece by piece. (d: 3)
Let s1 < s 2 < ... < sn < ... be elements of an interval [a, b] such that s1 = a and lim n . 5 sn = b.
Let + be an increasing function on [a, b], and let f be any bounded real-valued function on [a, b].
For each positive integer i, let fi denote the restriction of f to [si , si+1 ], and + i the restriction of +
to that interval.
Prove that f ) (+ ) if and only if for every i, fi ) (+ i ), and that when this holds, we have
b s
i =1 ( Bs f d + ),
Ba f d + = 3 5
i+1
i
with the sum on the right absolutely convergent. (The ith summand of that sum could, of course, be
s
written more formally as B i+1 fi d + i . )
si
6.4:9. Functions with only countably many discontinuities are integrable. (d: 2, 3)
Let us begin by proving explicitly an ‘‘intuitively obvious’’ general fact that we will need.
(a) Suppose an interval [a, b] is covered by finitely many open neighborhoods, N8 ( p1 ), ... , N8 ( pm ).
1 m
Show that there is a partition x 0 , ... , xn of [a, b] such that every subinterval [xi–1 , xi ] is contained in
(at least) one of the neighborhoods N8 ( pj ). Show further that in this situation, given any subset
i
S # {1, ... , m}, if we let T denote the set of indices i ) {1, ... , n} such that [xi–1 , xi ] is contained in
N8 ( pj ) for some j )S, then 3 i )T Cxi ! 3 j )S 2 8 j .
1
(Suggestion: Let y1 < ... < yk be those boundary points of the neighborhoods N8 ( pi ) – i.e., points
i
of the form pi – 8 i or pi + 8 i – that lie in (a, b), arranged in increasing order, and let y 0 = a,
yk+1 = b. Use the partition y 0 < (y 0 + y1 ) ⁄ 2 < (y1 + y 2 ) ⁄ 2 < . . . < (yk –1 + yk ) ⁄ 2 < (yk + yk+1 ) ⁄ 2 < yk+1 ,
verifying that it has the required properties.)
6.5. Step functions, differentiable + , and change of variables (end of Rudin’s section PROPERTIES
OF THE INTEGRAL). (pp.130-133)
Relevant exercises in Rudin: None
Exercises not in Rudin:
6.5:0. Say whether each of the following statements is true or false.
(a) There exists an increasing function + on [0,1] such that for every continuous function f on that
interval, B f d + = 3 2 – n f (1 ⁄ n).
1
0
(b) There exists an increasing function + on [0,1] such that for every continuous function f on that
interval, B f d + = 3 (1 ⁄ n) f (2 – n ).
1
0
(c) If + increases monotonically on [a, b] and is differentiable, then + ( ) on [a, b].
1
12
(d) If f ) on [0,1], then B f (x) dx = B f (2x) d(2x).
0 0
6.5:1. Getting the Fundamental Theorem of Calculus from Theorem 6.17. (d: 2. > 6.3:1)
This exercise obtains of one of the main results of the next section from a result in this section.
Let f ) on [a, b], and suppose there exists a differentiable function F on [a, b] such that F ( = f.
(a) Show that if f (x) * 0 for all x )[a, b], then one can apply Theorem 6.17 with the constant function
1 as the ‘‘f ’’ of that theorem, and F as the + of that theorem, and that 6.3:1 above is applicable to one
side of the formula you get. Show the formula that results from applying 6.3:1 to that side.
(b) Deduce that the same formula holds if f can be written f1 – f2 , where each of f1 and f2 is
nonnegative, is Riemann-integrable, and is the derivative of a differentiable function.
(c) Show that any f which is Riemann-integrable and is the derivative of a differentiable function can in
fact be written f = f1 – f2 as in (b). (Hint: Take f2 to be an appropriate constant function.) Conclude
that the formula you got in (a) is applicable to any such f .
6.5:2. Strengthening Theorem 6.16. (d: 2)
Show that Theorem 6.16 (p.130) remains valid if the assumption that f is continuous on [a, b] is
replaced by the condition that it is bounded, and is continuous at each of the points sn . Specifically –
We see that no continuity condition is needed until display (24). Say why that display is valid under
the condition stated above. Then modify the remainder of the proof (in which Rudin uses integrability of f
with respect to + 2 , which under his assumptions follows from Theorem 6.8) to conclude that (23) holds
with the upper, respectively the lower integral in place of the left-hand side. Conclude that f ) (+ ) and
that (23) holds.
6.5:3. Weaker conditions for integrability. (d: 3. > 6.5:2)
Let J(x) be defined like I(x) at the bottom of p.129, except that J(0) = 1. (Equivalently, J(x) =
1 – I(– x) for all x.)
(a) Show that every increasing function + (x) on an interval [a, b] can be written
+ (x) = 315 cn I(x – sn ) + 315 dn J(x – sn ) + 0 (x),
where 3 cn and 3 dn are convergent series of nonnegative real numbers, (sn ) is a sequence of distinct
points of [a, b], and 0 is a continuous increasing function on [a, b].
(Hint: If + is discontinuous at infinitely many points, can you write the set of these points as { sn
n = 1, 2, ... } ? If so, what should the cn and dn be to make the above equation plausible? Once you
have found these, to prove that + (x) – 3 cn I(x – sn ) – 3 dn J(x – sn ) is increasing, approximate this
function using finite partial sums in place of the infinite sums, and prove each of the resulting functions
increasing. Finally, verify continuity.)
(b) Taking for granted that the version of Theorem 6.16 described in 6.5:2 is also valid with J in place
of I, deduce that Theorem 6.9 remains valid if the assumption that + is continuous on [a, b] is
replaced by the assumption that it is continuous at every point where f is discontinuous.
(The proof of the result gotten by replacing ‘‘I ’’ with ‘‘J ’’ in 6.5:2 is virtually identical to the original
- 70 -
proof of that exercise. It can also be gotten using the next exercise.)
6.5:4. Integration with respect to d ( – + ( – x)). (d: 2)
Suppose f ) (+ ) on [a, b]. We would like to be able to apply Theorem 6.19 in the case D (x) = – x,
–b b
and deduce that B f (– x) d + (– x) = Ba f (x) d + (x). Unfortunately, + (– x) is not an increasing function
–a
of x, and – a is not < – b, so the left-hand integral has not been defined. However, – + (– x) is an
increasing function, so
(a) Show that –a b
B– b f (– x) d (– + (– x)) = Ba f (x) d + (x).
(Remark: Rudin doesn’t like showing the variable in integrations. To satisfy his preference, we could
let = : R . R (= = rho, for ‘‘reversal’’) be defined by =(x) = – x, and write the above equation
B ( f ° = ) d ( = ° + ° = ) = B f d + .)
(b) From part (a) and Theorem 6.19, deduce that if, in the first line of that theorem we change
‘‘increasing’’ to ‘‘decreasing’’, then we get a formula like (32), but with d 0 changed to d( – 0 ).
ends of the interval; then get the general case by writing any f ) (+ ) - ( 0 ) as a difference of two
nonnegative-valued functions in (+ ) - ( 0 ). Cf. the method of passing from part 6.5:1(b) to (c).)
(b) Show that given continuous functions u and on [a, b], there exists a continuous function w of
two variables such that for any f ) on [a, b], one has
x t x
Bt =a u(t) Bs =a (s) f (s) ds dt = Bt =a w(x, t) f (t) dt.
Hint: Prove the case where u and are nonnegative-valued using (a).
6.6:6. A characterization of the Riemann-Stieltjes integral. (d: 2, 4, 4)
Let + be an increasing function on [a, b], and f any bounded function on that interval. Given x1 ,
x 2 with a ! x1 < x 2 ! b, let us use C F as an abbreviation for F(x 2 ) – F(x1 ) and C + as an
abbreviation for + (x 2 ) – + (x1 ).
x
(a) Show that if f ) (+ ) and we define F(x) = B f (t) d + (t), then
t=a
(i) F(a) = 0,
(ii) For all x1 , x 2 with a ! x1 < x 2 ! b, we have C F ) [(C + ) (inf f (x)), (C + ) (sup f (x))], where the
inf and sup are over all x )[x1 , x 2 ].
(We could write the conclusion of (ii) more suggestively as C F ⁄ C + )[ inf f (x), sup f (x)], were it not
that C + might be zero for some choices of x1 and x 2 .)
(b) Show that F is the unique function satisfying (i) and (ii).
(c) Show that for any bounded function f on [a, b], if there is a unique function F on [a, b]
satisfying these conditions, then f ) (+ ). –x
Suggestion: Show that for any bounded f, the upper and lower integrals F+ (x) = B f (t) d + (t) and
x t=a
F– (x) = B t = a f (t) d + (t) each satisfy the indicated conditions, and that any function F satisfying those
–
conditions satisfies F– (x) ! F(x) ! F+ (x) for all x )[a, b].
6.6:7. A change-of-variables result. (d: 3. > 6.6:6)
Suppose + is a monotonically increasing real-valued function on [a, b] and f, g are continuous
real-valued functions on that interval, with g nonnegative-valued. Prove
x t x
Bt =a f (t) d Bs =a g(s) d+ (s) = B
t =a
f (t) g(t) d+ (t).
In parts (b) and (c) below, assume f is indeed integrable with respect to + , and let 8 be a positive
real number, and P = {x 0 , ... , xn } a particular partition making the above inequality hold.
(b) Show that for every choice of points ti )[xi –1 , xi ] (i = 1, ... , n), we have
b
( 3 ni=1 f(ti ) C + i ) – ( B f d + ) < 8 .
a
(c) Show that every refinement P* of P also satisfies the inequality of (a), and hence the condition
of (b).
In (c) and (d), point out why your examples have the indicated properties. You do not have to give
formal proofs.
7.1:3. Iterated limits and diagonal limits. (d: 2)
Suppose X is a metric space, and for all positive integers m and n we have an element sm,n )X,
such that for each m, the limit lim n . 5 sm,n exists, and such that these limits approach a limit,
lim m . 5 lim n . 5 sm,n = p.
Show that there exists a sequence of positive integers N1 , N 2 , ... such that for every sequence of
positive integers n1 , n 2 , ... satisfying nm * Nm one has lim m . 5 sm,n = p.
m
Suggestion: For each m, choose Nm so that for all r * Nm one has d (sm,r , lim n . 5 sm, n ) <
1 ⁄ m (noting why such an Nm exists).
7.1:4. The ‘‘1 at rationals’’ function isn’t a pointwise limit of continuous functions. (d: 5)
In Example 7.4, Rudin showed that the function f on R which has value 1 at every rational number
and 0 at every irrational is a pointwise limit of pointwise limits of continuous functions.
Show, however, that this function f is not itself a pointwise limit of any sequence ( fn ) of continuous
functions.
7.1:5. Variants of the preceding exercise. (d: 5, 5)
Suppose f is a function from a complete metric space X to a metric space Y, and suppose Y has
points y 0 , y1 such that the subsets f –1(y 0 ) and f –1(y1 ) are both dense in Y.
(a) Show that f is not a pointwise limit of continuous functions X . Y.
(b) Show that in the above situation, if f maps every point of X either to y 0 or to y1 (as in 7.1:4),
and if f –1(y1 ) is countable, then f cannot even be expressed as a pointwise limit of functions X . Y
that are continuous at all points of f –1(y1 ); but that it can be expressed as a pointwise limit of functions
X . Y that are continuous at all points of f –1(y 0 ).
7: R 24. An isometric embedding of any metric space in a complete metric space. (d: 3)
Exercises not in Rudin:
7.3:0. Say whether each of the following statements is true or false.
(a) If ( fn ) is a sequence of real-valued functions on a metric space X which converges uniformly to a
function f, and if f is continuous, then at least one of the functions fn is continuous.
(b) If ( fn ) is a sequence of continuous real-valued functions on a compact metric space K which
converges pointwise to a continuous function f, and if for each x )K and each n, fn (x) ! fn+1 (x),
then fn . f uniformly on K.
(c) The function f defined by f (x) = sin (x 2) belongs to (R).
(d) If ( fn ) is a sequence of continuous bounded functions on a compact metric space K which
converges pointwise to a continuous bounded function f, then fn . f in the metric space (K ).
(e) If ( fn ) is a sequence of continuous bounded functions on a compact metric space K which
converges uniformly to a function f, then fn . f in the metric space (K ).
(f) If X is a metric space, then every Cauchy sequence in (X ) converges.
(g) If K is a compact metric space, then (K ) is compact.
7.3:1. The amoeba meets uniform convergence. (d: 2. > 4.2:6, 7: R 2)
For each positive integer n, let pn : [0, 1] . R be the function so named in Exercise 4.2:6, i.e., the
probability that an amoeba that divides every hour will have at least one descendent after n hours,
expressed in terms of the probability that an amoeba survives for one hour.
(a) With the help of the result of 4.2:6(a), show that for every n, pn is given by a polynomial in t, and
that if we write p5 (t) for lim n . 5 pn (t), the convergence of the polynomials pn (t) to the function
p5 (t) is uniform on [0, 1].
Deduce that the convergence of the polynomials t pn (t) to the function t p5 (t) is likewise uniform
on that set. The fact that t p5 (t) is the limit of a uniformly convergent sequence of polynomial functions
on [0, 1] is what we will use in the remaining parts.
(b) From the above statement and the result of 4.2:6(b), deduce using a change of variables that the
function [ –1, 1] . R taking t to max(0, t ) is the limit of a uniformly convergent sequence of
polynomial functions on that interval, and that the same is true of the function [ –1, 1] . R taking t to
max(0, – t ). Adding these functions, deduce that the same is true of the absolute value function on [ –1, 1].
(c) Deduce that the same is true of the absolute value function on [ – a, a] for any positive real
number a.
(d) Show that by subtracting constants from the polynomials arising in (c), one can represent the absolute
value function on [ – a, a] as the uniform limit of a sequence of polynomials each of which has the
value 0 at 0.
Remark: Rudin gets the result of (d) in another way in 7: R 23. As he notes, this allows one to save a
good bit of work in the last section of Chapter 7. I hope the word-problem about amoebas has provided an
entertaining journey to this useful fact.
7.3:2. A version of Theorem 7.13 for functions between metric spaces. (d: 2)
Suppose K is a compact metric space and ( fn ) a sequence of continuous functions from K to a
metric space Y, which converges pointwise to a continuous function f . Suppose further that for all x )K
and all n, d( fn+1 (x), f (x)) ! d( fn (x), f (x)). Show that fn . f uniformly on K . (Suggestion: Either
adapt the method of proof of Theorem 7.13 (p.150), or use that theorem.)
Show that Theorem 7.13 is implied by the above result.
Answers to True/False question 7.2:0. (a) T. (b) F. (c) F. (d) F. (e) T. (f) T. (g) F. (h) T.
- 78 -
7.3:3. If ( fn ) converges uniformly to a bounded function, then most fn have a common bound. (d: 1)
Let ( fn ) be a sequence of real- or complex-valued functions on a set E, which converges uniformly
to a function f. Show that if f is bounded, then there exist a constant M and an integer N such that
for all n > N and all x )E, | fn (x)| ! M.
Show by example, however, that there may be values of n for which fn is unbounded.
7.3:4. A convergent series of nonnegative functions on a compact space converges uniformly. (d: 1)
Show that if ( fn ) is a sequence of nonnegative-valued continuous functions on a compact metric space
K, and the series 3 fn converges pointwise to a continuous function, then it converges uniformly. (Hint:
This follows easily from a result in this section.)
7.3:5. Subsets of (E ) determined by limit-conditions are closed. (d: 3, 1)
(a) Show that { f ) (R) lim x . + 5 f (x) exists} is a closed subset of (R).
(b) Suppose E is a subset of a metric space X and p )X is a limit point of E. Deduce from
Theorem 7.11 that { f ) (E ) lim x . p f (x) exists} is a closed subset of (E ).
7.3:6. Uniform continuity, and continuity of the translation-map. (d: 2)
Let f ) (R), and for each c )R, let fc be defined by fc (x) = f (x+c). Show that the map h :
R . (R) given by h(c) = fc is continuous if and only if f is uniformly continuous.
7.3:7. Discontinuities of uniform limits of discontinuous functions. (d: 2, 2, 2, 3)
Let ( fn ) be a sequence of real- or complex-valued functions on R which converges uniformly to a
function f.
(a) Show that if each fn has at most countably many discontinuities, then the same is true of f .
(b) If each fn has only finitely many discontinuities, must the same be true of f ?
(c) Show that if each fn has no discontinuities of the second kind, then the same is true of f .
(d) If each fn has no discontinuities of the first kind, must the same be true of f ?
7.3:8. Examples showing the need for the hypotheses of Theorem 7.13. (d: 2)
Let us examine the need for the various hypotheses in Theorem 7.13 (p.150).
(a) Indicate which examples given by Rudin show that the theorem becomes false (i) if the word
‘‘continuous’’ is dropped from hypothesis (b) of the theorem but kept in hypothesis (a); (ii) if
condition (c) of the theorem is dropped, and (iii) if the requirement that K be compact is dropped.
(b) Give an example showing that if the word ‘‘continuous’’ is deleted from hypothesis (a) of the
theorem, but kept in hypothesis (b), the statement also becomes false.
(c) Give an example showing that if the hypothesis that K is compact is replaced by the hypothesis that
all fi are bounded and uniformly continuous, the statement is still false.
7.3:9. A second uniformity in Theorem 7.11. (d: 2)
In Theorem 7.11, p.149, the condition that the convergence of the fn be uniform concerns ‘‘uniformity
in t ’’, i.e., it says that for every 8 there is an N independent of t with the appropriate property.
Prove that under the conditions of the theorem, the convergence of the fn (t) to the values An is
‘‘uniform in n’’, in the sense that for every 8 > 0 there exists a 7 > 0 (independent of n) such that for
every n and every t )E with d(t, x) < 7 , one has | fn (t) – An | < 8 .
Suggestion: Choose N such that for n * N, | fn (t) – fN (t) | < 8 ⁄ 2 for all t ; then find a 7 such
that when d(t, x) < 7 and n ){1, ... , N }, one has | fn (t) – An | < 8 ⁄ 2.
7.3:10. More on lim fn (xn ). (d: 3. > 7: R 9)
As noted in my comment on 7: R 9, the last sentence thereof can be taken ask whether it is true that if a
sequence ( fn ) of continuous functions on a metric space E and a continuous function f on E have the
property that for every sequence of points (xn ) in E which approach a limit x, one has
Answers to True/False question 7.3:0. (a) F. (b) T. (c) T. (d) F. (e) T. (f) T. (g) F.
- 79 -
lim n . 5 fn (xn ) = f (x), then ( fn ) must converge uniformly to f. Prove that this is so if E is assumed
compact.
7.3:11. Uniform convergence is convergence in a metric even for unbounded functions. (d: 2)
Rudin’s observation on p.151, lines 4-6, that uniform convergence is equivalent to convergence in the
metric defined on the top line of that page, is necessarily limited to bounded functions, since the supremum
of the absolute value of an unbounded function is infinite. Show, however, that if we define d ( f, g) =
min( || f – g ||, 1), then this function gives a metric on the space of all real- or complex-valued functions on
any set E, such that uniform convergence of such functions is equivalent to convergence in this metric.
(Remark: Although this metric is of interest for the above reason, it has the disadvantage of not
satisfying the law d (cf, cg) = c d ( f, g) for all positive real numbers c, which is of importance in the
study of vector spaces of functions, and which is satisfied by the metric || f – g|| on (X ).)
7.3:12. Pointwise convergence is not convergence in any metric. (d: 3. > 7.1:3, 4.2:10, 7: R 14)
In contrast to Rudin’s observation that uniform convergence of functions in (X ) is equivalent to
convergence in the metric d ( f, g) = || f – g||, we shall show here that there is in general no metric d on
(X ) such that convergence with respect to d is equivalent to pointwise convergence of functions. We
shall do this by showing that pointwise convergence does not have the property proved in 7.1:3 for
convergence in a metric space, even in the case where all the elements called pm in that exercise are the
same. Specifically, we will construct functions gm,n ) ([0,1]) such that for each m, lim n . 5 gm,n =
0, but such that there is no sequence of integers (nk ) such that lim k . 5 gk,n = 0.
k
(Such an example will be obtained here relatively quickly, assuming the earlier exercise 4.2:10, which
was in turn based on Rudin’s quite challenging 7: R 14. In 7.3:17 below, we will obtain an example with
the same properties with only a little more work, and without relying on more difficult exercises.)
To construct the gm,n recall that in 4.2:10(c) we found (assuming the result of 7: R 14) a sequence of
functions, which we will denote (cn ), such that for every sequence (xi ) of points of [0,1], there exists
a point t )[0,1] such that cn (t) = xn for all n. (What we are calling cn was, in the notation of that
exercise, a ° b n .) Recall also that in Example 7.21 (p.156), Rudin gives a sequence of functions fn :
[0,1] . [0,1] converging pointwise to 0, but such that each fn takes the value 1 at some point.
Using the functions cn and fn just named, let us, for all m, n * 1, define gm,n = fn ° cm .
(a) Show that for each m, (gm,n ) is a sequence of continuous functions [0,1] . [0,1] which converges
pointwise to the zero function as n . 5 .
(b) Show, however, that given any sequence of positive integers (nk ), one can find t )[0,1] such that
for all k, gk,n (t) = 1.
k
(c) Deduce that there is no sequence of positive integers (nk ) such that the sequence of functions gk, n
k
converges pointwise to the zero function.
(d) Conclude with the help of 7.1:3 that there is no metric on ([0,1]) such that pointwise convergence
of functions in [0,1] is equivalent to convergence in this metric.
7.3:13. Another property of the above example. (d: 4. > 7.3:12 or 7.3:17)
Let us use the construction of the preceding exercise, or the similar construction of 7.3:17 to show
another way that pointwise convergence in ([0,1]) is unlike convergence in a metric space.
Let (gm,n ) be any family of continuous functions on [0,1] having the properties asserted in
7.3:12 (a) and (b), or in 7.3:17 (a) and (c). Let be the set of functions {gm,n + 1 ⁄ m m, n )J }, and
let be the set of pointwise limits in ([0,1]) of all pointwise convergent sequences of functions in
. Show that there exists a sequence of functions in which is pointwise convergent to a function in
([0,1]) that does not lie in . Why could this not happen if pointwise convergence were convergence
with respect to a metric?
7.3:14. But pointwise convergence on a countable set is pointwise convergence in a metric. (d: 3.
> 4.2:14)
- 80 -
Let E be a countable set, and the space of all real- or complex-valued functions on E (or any
subset thereof). Show that there exists a metric d on such that convergence in d is equivalent to
uniform convergence of functions. (Suggestion: Use the idea of 4.2:14.)
7.3:15. Uniform convergence expressed in terms of uniform continuity. (d: 1, 2)
Let E be a set, ( fn ) a sequence of complex-valued functions on E, and f another complex-valued
function on E. We shall show that the sequence ( fn ) converges (respectively, converges uniformly) to f
if and only if a certain function F on a certain metric space X is continuous (respectively, uniformly
continuous).
Namely, let S = {1 ⁄ n | n )J } , {0} # R (where, following Rudin, we are using J for the positive
integers), let X = E × S, meaning the set of ordered pairs ( p, s) with p )E, s )S, and let us make X a
metric space by defining the distance d(( p, s), ( p( , s()) to be | s – s(| if p = p(, and to be 1 otherwise.
(a) Verify that this function is indeed a metric.
Let us now define F : X . C by letting F (( p,1 ⁄ n)) = fn ( p), and F (( p, 0)) = f ( p).
(b) Show that fn . f pointwise if and only if F is continuous, and that fn . f uniformly if and only
if F is uniformly continuous.
The above exercise puts in visualizable form a way in which the concept of uniform convergence is
parallel to that of uniform continuity. I wondered whether I could similarly find a construction whereby
the uniform continuity of any function on a metric space could be expressed as the uniform convergence of
a sequence of functions on a set. The next exercise gives two such constructions; the one in (a) is simple,
but somewhat hoaky; the one in (b) is more natural, but also more complicated. (Actually, the one in (a)
can be looked at as a simplified version of the one in (b).)
7.3:16. Uniform continuity expressed in terms of uniform convergence. (d: 2)
Let X be a metric space, and F : X . C any function.
(a) Let E denote the set X × X of all ordered pairs ( p, q) with p, q )X, and let us define a sequence
( fn ) of functions on E and another function f on E as follows: For each n, and each ( p, q) )E, let
fn ( p, q) = F ( p) if d( p, q) < 1 ⁄ n, and F (q) otherwise. Let f ( p, q) = F ( q). Show that whatever the
choice of F, we will have fn . f pointwise, and that this convergence will be uniform if and only if F
is uniformly continuous.
(b) Let E ( denote the set of those pairs (( pn ), q) such that ( pn ) is a sequence in X, q is a point of
X, and for all n, d( pn , q) < 1 ⁄ n. For each m > 0 let fm ((( pn ), q)) = F ( pm ), and let f ((( pn ), q)) =
F ( q). Show that fn . f pointwise if and only if F is continuous, and that this convergence is uniform
if and only if F is uniformly continuous.
7.3:17. Another example showing pointwise convergence is not convergence in any metric. (d: 3. > 7.1:3)
As in 7.3:12 above, we shall construct here a family of functions gm,n whose properties with respect
to pointwise convergence would contradict 7.1:3 if pointwise convergence were convergence with respect
to a metric; but this time our construction will be self-contained.
For every pair of positive integers m and n, let gm,n ) ([0,1]) be the function such that for each
nonnegative integer k < 2 m, the values of f on the subinterval [k 2 – m , (k+1) 2 – m ] # [0,1] are
determined by the formulas
gm,n (k 2 – m + t 2 – m–n ) = t for 0 ! t ! 1,
gm,n (k 2 – m + t 2 – m–n ) = 2 – t for 1 ! t ! 2,
gm,n (x) = 0 for x )[k 2 – m + 2 · 2 – m – n, (k+1) 2 – m ].
(To get a feel for this definition, you might graph g 0, n for the first few values of n (I wrote
‘‘m > 0’’ above so that the sequences in this exercise would all be indexed by positive integers, but the
definition makes sense for m = 0 as well, and gives the simplest picture), and then go to m =1, m = 2, to
see how the behavior varies with m.)
(a) For each m, show that the sequence of functions gm,n (n = 1, 2, ... ) converges pointwise to the zero
- 81 -
function on [0,1].
Since the sequence of zero functions in turn converges pointwise to the zero function on [0,1], if
pointwise convergence were convergence in a metric on ([0,1]), 7.1:3 would imply the existence of a
sequence of positive integers N1 , N 2 , ... such that for any positive integers n1 , n 2 , ... with nm * Nm
for all m, the sequence of functions gm,n converged pointwise to 0. To prove the contrary, we will
m
need a description of places where our functions gm,n take on values that are not close to 0.
(b) Prove that if m and n are positive integers, and x )[0,1] has the property that the digits in the
binary expression of x with place-value 2 – m – j are 0 for j = 1, ... , n, while the digit with place-value
2 – m – n –1 is 1, then gm,n (x) * 12.
(c) Show with the help of (b) that for any sequence of positive integers n1 , n 2 , ... , there exists x )[0,1]
such that gm,n * 12 for infinitely many values of m.
m
(d) Deduce that there is no sequence n1 , n 2 , ... such that the sequence of functions gm,n converges
m
pointwise to 0. Conclude using 7.1:3 that pointwise convergence on ([0,1]) is not equivalent to
convergence in any metric d on that set.
7.3:18. Uniform limits of uniformly continuous functions. (d: 2)
Show that if ( fn ) is a sequence of complex-valued functions on a metric space X, each of which is
uniformly continuous, and if fn . f uniformly, then f is also uniformly continuous.
7.3:19. Locally uniform convergence. (d: 1, 2, 2, 2)
Let us say that a sequence ( fn ) of complex-valued functions on a metric space X converges locally
uniformly to a function f if for every x )X and every 8 > 0, there exists a 7 > 0 and a positive integer
N such that for every n * N, and every y with d (x, y) < 7 , one has | fn (y) – f (y)| ! 8 .
(a) Show that uniform convergence implies locally uniform convergence, and locally uniform convergence
implies pointwise convergence.
(b) Show by examples that neither of the two preceding implications is reversible. (Looking at part (d)
below may help you see what is needed in one of these examples.)
(c) Show that Theorem 7.13 remains true if we delete the assumption that K is compact, while
weakening the conclusion by inserting the word ‘‘locally’’ before ‘‘uniform’’ on the last line.
(d) Show that on a compact metric space, locally uniform convergence is equivalent to uniform
convergence. Deduce Theorem 7.13 from this and part (c) above.
(e) Prove the result of 7.3:10 without the assumption of compactness, but with the conclusion of uniform
convergence weakened to locally uniform convergence. With the help of the first statement of (d) above,
deduce from this the result of 7.3:10 as stated.
7.3:20. Locally uniform convergence and composition of functions. (d: 3)
Suppose that m : X . Y is a continuous function between metric spaces. Recall that if f is a
function on Y, then f ° m denotes the function on X defined by ( f ° m)(x) = f (m(x)). Recall also the
definition of ‘‘locally uniform’’ convergence given in the preceding exercise.
(a) Show that if f and fn (n * 1) are continuous complex-valued functions on Y, and if fn . f
locally uniformly, then fn ° m . f ° m locally uniformly.
(b) Show by example that the result of (a) becomes false if the word ‘‘locally’’ is deleted in both places.
(c) Show, on the other hand, that the statement shown to be false in (b) becomes true again if the
assumption on m : X . Y is strengthened from ‘‘continuous’’ to ‘‘uniformly continuous’’.
- 82 -
Answer to True/False question 7.4:0. (a) T. Answers to True/False question 7.5:0. (a) F. (b) F. (c) T. (d) F.
- 84 -
exists. A few lines later, where after defining C n (t) he adds ‘‘except ...’’, he means that for t = xi , one
defines C n (t) = 0. (The preceding definition is not applicable when t = xi because fn((xi ) is generally
undefined; so the arbitrary value 0 is used to make C n (t) defined.)
7: R 26. The corresponding result for a system of differential equations. (d: 3. > 7: R 25)
Rudin’s hint says ‘‘Use the vector-valued version of Theorem 7.25’’. He means that you should prove
such a version and use it. The vector-valued result is not hard to prove from the theorem as given.
Exercises not in Rudin:
7.6:0. Say whether each of the following statements is true or false.
(a) The set of functions {x n n = 1, 2, ... } on [–1, 1] is uniformly bounded.
n
(b) The set of functions {x n = 1, 2, ... } on [0, 2] is uniformly bounded.
(c) Every uniformly bounded family of continuous functions on a compact set is equicontinuous.
(d) If is a family of differentiable functions on R such that the set { f ( f) } is uniformly
bounded, then is equicontinuous.
7.6:1. When do boundedness at one point and equicontinuity imply pointwise boundedness?
(d: 4, 1, 4, 2, 2)
(a) Show that the following two conditions on a metric space X are equivalent:
(i) Every equicontinuous family of functions on X whose values at one point are bounded is
pointwise bounded. (I.e., if is an equicontinuous family of functions on X, and if there exists
x )X such that { f (x) | f ) } is bounded, then { f (y) | f ) } is bounded for every y )X.)
(ii) For every two points x, y )X, and every 7 > 0, there exist points x 0 , ... , xn such that x = x 0 ,
y = xn , and d(xi–1 , xi ) < 7 for i = 1, ... , n –1.
(b) Show that X = [0,1] , [2, 3] does not satisfy the above equivalent conditions (i) and (ii).
(c) Show that every connected metric space satisfies the above equivalent conditions.
(d) Show that if a metric space X satisfies the above equivalent conditions, then so does every dense
subset of X.
(e) Use the results of (c) and (d) to find an open subset of R which satisfies the equivalent conditions
of (a), but which is not connected.
7.6:2. Proving Example 7.20 without using a result from Chapter 11. (d: 2)
In Example 7.20, Rudin shows that the sequence of functions fn = sin nx on the interval [0, 2 @ ] has
no pointwise convergent subsequence, but to do so, he calls on a result in Chapter 11. Below, you will
prove this result using only material from Chapters 1-4, and the facts that sin x is a continuous function
which takes on the value 1 at x = @ ⁄ 2 and the value –1 at x = 3@ ⁄ 2, and satisfies sin (x + 2 @ ) = sin x
for all x. When we speak of an ‘‘interval [a, b]’’ below, we shall understand this to entail a < b.
(a) Show that for any interval [a, b] # R there exists an integer N such that for all n > N the function
sin nx takes on both the values 1 and –1 at points of [a, b].
(b) Deduce that for any infinite set S of positive integers and any interval [a, b], there exists an n )S,
and subintervals [a( , b(] and [a( ( , b( (] of [a, b], such that the function sin nx has value everywhere
* 12 on [a( , b(], and has value everywhere ! – 12 on [a( ( , b( (].
(c) Deduce from (b) that for any infinite set S of positive integers one can find a point x and a sequence
n1 < n 2 < ... in S such that the real numbers sin n1 x, sin n 2 x, ... are alternately * 12 and ! – 12.
(d) Conclude that the sequence of functions sin nx (n = 1, 2, ... ) has no pointwise convergent
subsequence.
- 85 -
7.7. The Weierstrass Theorem, and a corollary (beginning of Rudin’s section THE STONE-
WEIERSTRASS THEOREM). (pp.159-161)
Relevant exercises in Rudin:
7: R 20. A continuous function on [0,1] is determined by its moments. (d: 2)
The integral on the left-hand side of the displayed equation of this exercise is called the nth moment of
the function f. Add to this exercise ‘‘Deduce that if g and h are continuous functions on [0,1] such
that for every n, the nth moments of g and of h are the same, then g = h’’. (This is the meaning of
the title I have given this exercise.) In Rudin’s Hint for the exercise, f 2(x) in the integral should be
changed to | f (x)| 2.
7: R 22. Every integrable function is L 2 -approximable by polynomials. (d: 3. > 6: R 12)
Suggestion: Use the exercise Rudin refers to parenthetically at the end, together with a result from this
section.
7: R 23. An explicit algorithm for uniformly approximating | x | by polynomials. (d: 3)
This gives a direct proof of Corollary 7.27, which is the only consequence of the Weierstrass Theorem
that Rudin will use in proving the Stone-Weierstrass Theorem. (As noted above, 7.3:1 gives another proof
of this result.)
Exercises not in Rudin:
7.7:0. Say whether the following statement is true or false.
(a) A complex-valued function on [–1, 1] is continuous if and only if it can be written as the uniform
limit of a sequence of polynomial functions.
7.7:1. The Weierstrass Theorem fails for functions on the whole line. (d: 2)
(a) Show that the only polynomials which, as functions on R, are bounded, are the constant functions.
(Suggestion: Use results of Chapter 3.)
(b) Deduce that, in contrast with Theorem 7.26, if a sequence of polynomials Pn converges uniformly on
the whole real line R to a function f, then f is itself a polynomial.
7.7:2. A modified Weierstrass theorem that does work for functions on all of R. (d: 2)
Suppose f is a continuous complex-valued function on the real line. Show that there exists a sequence
of polynomials Pn such that for each finite interval [a, b], the polynomials Pn converge uniformly to
f on [a, b].
7.8. Algebras of Functions, Uniform Closure, and Separation of Points (middle of Rudin’s section
THE STONE-WEIERSTRASS THEOREM). (pp.161-162)
Relevant exercises in Rudin: None
Exercises not in Rudin:
7.8:0. Say whether each of the following statements is true or false.
(a) For every metric space X, (X ) is an algebra of functions on X.
(b) If and are uniformly closed algebras of functions on a set E, then - is also a
uniformly closed algebra of functions on E.
(c) The set of all monotonically increasing real-valued functions + on an interval [a, b] is an algebra.
(d) The set of all monotonic real-valued functions + (increasing and decreasing) on an interval [a, b] is
an algebra.
(e) If a family of functions on a set E separates points, then so does every family of functions on E
containing .
(f) If a family of functions on a set E separates points, then so does every subset of .
(g) If a family of functions on a set E vanishes at no point of E, then so does every family of
functions on E containing .
(h) If a family of functions on a set E vanishes at no point of E, then the same is true of every
subset of .
7.8:1. Equicontinuous algebras are mostly uninteresting. (d: 2, 1, 2, 4)
This exercise will show that ‘‘equicontinuous’’ is not, in general, an interesting condition to impose on
algebras of functions, though as (b) shows, there are some nontrivial examples.
(a) Show that there is no equicontinuous algebra of real-valued functions on [0,1] which separates points.
(b) Show that the algebra of all real-valued functions on Z is equicontinuous and separates points.
(c) Let X # { 1 ⁄ n n = 1, 2, 3, ... }. Does there exist an equicontinuous algebra of real-valued functions
on X which separates points?
(d) Give a simple characterization of the class of metric spaces X such that there exists an
equicontinuous algebra of real-valued functions on X which separates points.
7.8:2. Transporting algebras of functions from one metric space to another. (d: 1, 1, 2, 3, 3)
Throughout this exercise, let m : X . Y be a continuous map of metric spaces. Recall that if f is a
function on Y, then f ° m denotes the function on X defined by ( f ° m)(x) = f (m(x)).
(a) Show that if ( fn ) is a sequence of functions on Y that converges uniformly to a function f, then
the sequence of functions ( fn ° m) on X converges uniformly to f ° m.
In the remaining parts, let be an algebra of continuous functions on X, and let be an algebra
of continuous functions on Y. Let m *( ) = { f ° m f ) }, and let m ( ) denote the set of all
*
continuous functions f on Y such that f ° m ) . To avoid confusion with complex conjugation, let us
use cl for uniform closure; e.g., cl( ) will denote the uniform closure of .
(b) Show that m *( ) is an algebra of continuous functions on X, and that m ( ) is an algebra of
*
continuous functions on Y.
(c) Show that m *(cl( )) # cl(m *( )) and that m (cl( )) ? cl(m ( )).
* *
(d) Show that the reverses of the above two inequalities do not hold in general. (Suggestions: For the
first inequality, let m be the inclusion map [0,1] . R, i.e., the map defined by m(x) = x, and the
algebra of all polynomial functions on R. For the second inequality, let X = [–1,1], Y = [0,1], m(x) =
max(0, x), and let be the algebra of all polynomial functions on [–1,1]. You may assume 7.7:1
whether or not you did it, and you may assume the fact that the only polynomial p(x) such that the
equation p(x) = 0 has infinitely many solutions is the zero polynomial.)
(e) What implications, if any, hold between the statements ‘‘ is uniformly closed’’ and ‘‘m ( ) is
*
uniformly closed’’? Between ‘‘ is uniformly closed’’ and ‘‘m *( ) is uniformly closed’’?
(There are four possible implications – one each way for each of the indicated pairs of statements. For
full credit you need to give, for each of these four implications, either a proof that it is true, or an example
showing that it is false. In doing this, you may assume any of the previous parts of this exercise, whether
you did them or not.)
7.8:3. The uniform closure of the algebra of Laurent polynomials. (d: 4. > 7.7:1)
A Laurent polynomial means a function which can be written in the form f (x) = 3 nN= – N an z n ,
where N * 0 and a–N , ... , aN are constants. (An example is z –2 + 3 z –1 – 7 + z 2 + 5 z 3, which we can
get by taking N = 3 and letting a– 3 = 0, a– 2 = 1, ... a 3 = 5.) A Laurent polynomial can be evaluated
at any nonzero value of x; in particular, given any subset E of R not containing the point 0, the
Laurent polynomials with real coefficients yield an algebra of real-valued continuous functions on E.
n n
Given a Laurent polynomial 3 N n =– N an z , let us call 3N
n = 0 an z its ‘‘polynomial part’’ and
n
3 n–1
= – N an z its ‘‘negative-exponent part’’.
(a) Let E be the half-open interval (0,1], and suppose that ( fk ) is a sequence of Laurent polynomials
which, regarded as a sequence of functions on E, converges uniformly. Show that if we write each fk
as gk + hk , where gk is its polynomial part and hk its negative-exponent part, then all but finitely
many of the functions hk are equal, and the sequence of functions (gk ) converges uniformly.
(b) Deduce that the uniform closure of the algebra of Laurent polynomial functions on (0,1] consists of
all functions which can be written as the sum of a continuous function and a ‘‘negative-exponent Laurent
polynomial’’, i.e., a function of the form 3 n–1= – N an z n .
(c) Deduce that the uniform closure of the algebra of Laurent polynomial functions on (0,1] is not an
algebra. This shows that a certain word cannot be omitted from the statement of Theorem 7.29 – which
word?
Answers to True/False question 7.8:0. (a) T. (b) T. (c) F. (d) F. (e) T. (f) F. (g) T. (h) F.
- 88 -
you need to write down. In each case of the opposite sort, give an example of a function not in the
uniform closure of i , and state at least one hypothesis of the Stone-Weierstrass Theorem which fails to
hold for that set. If the condition that fails is that i be an algebra, state one of the properties defining
an algebra which is not satisfied by i . For your own sake, you should also be able to show that the
function you give is not in the uniform closure; but you are not asked for this verification in your
homework.
(a) The set 1 of all complex-valued polynomial functions f on [0, 1] which satisfy f (0) = f (1).
(b) The set 2 of all real-valued polynomial functions f on [0, 1] that satisfy f (( 2) = 0.
1
(c) The set 3 of all continuous real-valued functions f on R such that lim x . +5 f (x) exists.
(d) The set 4 of all complex-valued polynomial functions f on [0, 1] that satisfy f (1) = f (0).
(e) The set 5 of all real-valued continuous functions f on [0, 1] that satisfy f (0) + f ( 2) + f (1) = 0.
1
(f) The set 6 of all functions on [0,1] of the form p(x), where p is a real-valued polynomial which
is divisible by x –1.
(g) The set 7 of all functions on [0,1] of the form p(x), where p is a real-valued polynomial
which is divisible by x – 2.
(h) The set 8 of all continuous real-valued functions f on [0, 1] that satisfy
(& 8 > 0) (% x )[0, 8 ]) f (x) = f (0).
7.9:2. The hypothesis of the Stone-Weierstrass Theorem is sufficient but not necessary. (d: 2)
Give an example of a set of real-valued functions on a metric space K which does not satisfy all
the hypotheses of Theorem 7.32, but such that the uniform closure of does consist of all continuous
real-valued functions on K.
7.9:3. An even or odd function is uniformly approximable by even or odd polynomials. (d: 2)
(a) Let be the algebra of all even polynomial functions on [–1, 1] (polynomial functions p
satisfying p(– x) = p(x)). Show that the uniform closure of consists of all even continuous functions.
(The easier direction is ‘‘#’’. Suggestion for ‘‘?’’: Apply the Stone-Weierstrass Theorem to the
restrictions of these functions to [0,1].)
(b) Let be the set of all odd polynomial functions on [–1, 1] (polynomial functions p satisfying
p(– x) = – p(x)). Show that the uniform closure of consists of all odd continuous functions.
(Suggestion for ‘‘?’’: Approximate such a function f by polynomials pn , break each pn into the sum
of an odd and an even polynomial, and show that the odd summands also approximate f.)
7.9:4. Functions with value zero at 0 are uniformly approximable by polynomials with value zero at 0.
(d: 2)
Let be the algebra of all polynomial functions p on [–1, 1] satisfying p(0) = 0. Show that the
uniform closure of consists of all continuous functions f satisfying f (0) = 0. (As in the previous
exercise, the easier direction is ‘‘#’’. Suggestion for ‘‘?’’: See Rudin’s proof of Corollary 7.27.)
7.9:5. Intersections of uniform closures, and uniform closures of intersections. (d: 4, 2)
Let be the set of restrictions to [0,1] of even polynomial functions, i.e., polynomial functions
satisfying p(– x) = p(x), and let be the set of restrictions to [0,1] of polynomial functions satisfying
p(2– x) = p(x).
(a) Show that - consists only of constant functions.
(b) Let us write cl( ) instead of for the uniform closure of , to avoid confusion with complex
conjugation. Show that cl( ) = cl( ) = ([0,1]). Deduce that cl( ) - cl( ) 4 cl( - ).
7.9:6. Uniform closures of algebras of continuous real functions not satisfying the hypotheses of the
Stone-Weierstrass Theorem. (d: 1, 2)
Answers to True/False question 7.9:0. (a) T. (b) F. (c) F. (d) T. (e) T. (f) F. (g) T.
- 89 -
Let be an algebra of continuous real-valued functions on a compact set K, but let us not assume
that separates points or vanishes at no point of K. Rather, given any continuous real-valued function
f on K, let us say that f ‘‘separates no points not separated by ’’ if for all x, y )K, we have
((% h ) ) h(x) = h(y)) / f (x) = f (y),
and let us say that f ‘‘vanishes wherever vanishes’’ if for all x )K, we have
((% h ) ) h(x) = 0) / f (x) = 0.
(You might find these conditions easier to think about in contrapositive form:
f (x) 4 f (y) / (& h ) ) h(x) 4 h(y), respectively, f (x) 4 0 / (& h ) ) h(x) 4 0.)
Then I claim that
The uniform closure of consists of all continuous real-valued functions f that
separate no points not separated by , and vanish wherever vanishes.
(a) Show (by arguments and/or quoting results from Rudin) that all functions f in the uniform closure of
are indeed continuous, separate no points not separated by , and vanish wherever vanishes.
To prove the converse, suppose f is a continuous real-valued function on K which separates no
points not separated by and vanishes wherever vanishes. We must show that f is uniformly
approximable by members of . This can be done by a small change in the proof of Theorem 7.32.
Steps 1 and 2 of that proof do not use anything about separating points or not vanishing, and so need
no change. In the statements of Steps 3 and 4, the only change needed is to add, after ‘‘a real function f,
continuous on K ’’, the words ‘‘which separates no points not separated by , and vanishes wherever
vanishes’’. All assertions in the proofs of those two steps then become true, except for the first
sentence of the proof of Step 3, which is used to justify the second sentence. So –
(b) Prove that second sentence (the one beginning ‘‘Hence’’ and ending with display (55)) under the
above hypotheses. (You will need to consider different cases, depending on whether separates x and
y, and whether it vanishes on one or both of these points.)
This completes the proof of the result stated in italics above.
(If you are careful, you will see the need for one small condition in the definition of an algebra that
Rudin accidentally omitted: That it be nonempty, i.e., contain at least one function. Assume this.)
7.9:7. The Stone-Weierstrass theorem fails for non-self-adjoint algebras of complex-valued functions.
(d: 3)
Let K be any compact subset of the complex plane which contains the origin 0 and the unit circle
{z )C | |z| = 1} (for instance, the unit disk D = {z )C | |z| ! 1}), and let be the algebra of
N n
polynomial functions on K, i.e., functions f of the form f (z) = 3 n =0 an z where a 0 , ... , aN )C.
(a) Show that separates points of K and vanishes nowhere on K.
(b) Show that every f ) satisfies
2@
f (0) = 1
B
2@ 0 f (e it ) dt .
(Hint: First verify this for each of the functions z n . In doing so, you may assume that exponentials of
imaginary numbers, defined by formula (32) on p.112, satisfy the familiar formula (e z ) n = e nz .)
(c) Show that if ( fn ) is a sequence of continuous functions on K which each satisfy the equation
of (b), and if this sequence converges uniformly to a function f, then f also satisfies that equation.
(d) Deduce that the uniform closure of the algebra is not the whole algebra (K ). (This
phenomenon, in particular, the above integral formula, can be regarded as a tip of the iceberg of the subject
of Complex Analysis.)
(e) Deduce from parts (a) and (b) the result of 7: R 21.