Professional Documents
Culture Documents
Szego
Translation by D. Aeppli
Dorothee Aeppli
St. Paul, Minnesota
General rules which could prescribe in detail the most useful disci-
pline of thought are not known to us. Even if such rules could be for-
mulated, they would not be very useful. Rather than knowing the correct
rules of thought theoretically, one must have them assimilated into
one's flesh and blood ready for instant and instinctive use. Therefore,
for the schooling of one's powers of thought only the practice of thinking
is really useful. The independent solving of challenging problems will aid
the reader far more than the aphorisms which follow, although as a
start these can do him no harm.
One should try to understand everything: isolated facts by collating
them with related facts, the newly discovered through its connection
with the already assimilated. the unfamiliar by analogy with the accus-
tomed, special results through generalization, general results by means
of suitable specialization, complex situations by dissecting them into
their constituent parts, and details by comprehending them within a
total picture.
There is a similarity between knowing one's way about a town and
mastering a field of knowledge; from any given point one should be able
to reach any other point 1. One is even better informed if one can imme-
diately take the most convenient and quickest path from the one point
to the other. If one is very well informed indeed, one can even execute
special feats, for example, to carry out a journey by systematically
avoiding certain forbidden paths which are customary-such things
happen in certain axiomatic investigations.
There is an analogy between the task of constructing a well-integrated
body of knowledge from acquaintance with isolated truths and the
building of a wall out of unhewn stones. One must turn each new insight
and each new stone over and over, view it from all sides, attempt to
join it on to the edifice at all possible points, until the new finds its
suitable place in the already established, in such a way that the areas
of contact will be as large as possible and the gaps as small as possible,
until the whole forms one firm structure.
A straight line is determined by two points. Similarly, many a new
result is obtained by means of a kind of linear interpolation between
1 See e.g. problem 92 and the neighboring problems in Part VI, also problem 64
in Part VIII.
VIn Preface to the First German Edition
1 See Part II, Chapter 2; Part IV, Chapter 1, §1; Part V, Chapter 1, §1;
Part VIII, Chapter 1, § 4.
XII Preface to the First German Edition
sessions the easier problems were discussed in the classroom and the
answers were given orally by the students, while the more difficult
problems were answered in writing by an appropriately set deadline.
Important problems serving as paradigms were solved by the instructor.
Within one semester it was possible to cover approximately the material
of one chapter. Several chapters have been tested in this manner and in
part revised on the basis of the experience which we have gained. We
believe that we may in good conscience recommend for the organization
of practice sessions and seminars the method which we have followed:
to pose not isolated problems, but carefully considered connected se-
quences of problems. Nearly all the chapters of this book can be used as
a basis for such instruction. It is obvious that some care should be exer-
cised. Especially, for homework or examinations, it is advisable to
replace some problems by related ones.
A continuous reading of the work, in which immediately after each
problem the solution is also read, can be recommended only to more
experienced readers. On the whole this is not in the spirit of the book.
Nevertheless certain chapters are suitable for such continuous reading
and may be used essentially as a teaching text. However, for this pur-
pose the presentation is rather condensed; it was intended to allow some
time for thinking about the problem between its formulation and its
solution, and about the proposition between its statement and its proof.
Chapter 1
Problem
Numbers Operations with Power Series Prob· Solu-
lem tion
§1 (1-31). Additive Number Theory, Combinatorial Problems, Page Page
and Applications . . . . . . . . . . . . 173
§2 (31.1-43.1). Binomial Coefficients and Related Problems 6 183
§3 (44-49). Differentiation of Power Series 8 185
§4 (50-60). Functional Equations and Power Series 9 186
§5 (60.1-60.11). Gaussian Binomial Coefficients 11 188
§6 (61-64.2). Majorant Series . . . . . 13 189
Chapter 2
Chapter 3
Chapter 4
Miscellaneous Problems
§1 (140-155). Enveloping Series . . . • . . . . . . . . . . . 32 209
§2 (156-185.2). Various Propositions on Real Series and Sequences 36 212
§3 (186-210). Partitions of Sets, Cycles in Permutations 42 224
Contents XV
Part Two
Integration
Chapter 1 Prob- Solu-
Problem lem tion
Numbers The Integral as the Limit of a Sum of Rectangles Page Page
Chapter 2
Inequalities
§1 (69-94). Inequalities. 62 247
§2 (94.1-97). Some Applications of Inequalities 72 255
Chapter 3
Some Properties of Real Functions
§ 1 (98-111). Prope; Integrals _ 75 258
§2 (112-118.1). Improper Integrals 77 261
§3 (119-127). Continuous, Differentiable, Convex Functions 79 264
§4 (128-146). Singular Integrals. Weierstrass' Approximation
Theorem 81 270
Chapter 4
Various Types of Equidistribution
§1 (147-161). Counting Function. Regular Sequences 85 275
§2 (162-165). Criteria of Equidistribution . 87 279
§3 (166-173). Multiples of an Irrational Number. 88 280
§4 (174-184). Distribution of the Digits in a Table of Logarithms
and Related Questions 90 281
§5 (185-194). Other Types of Equidistribution . 92 285
Chapter 5
Functions of Large Numbers
§1 (195-209). Laplace's Method 95 287
§2 (210-217.1). Modifications of the Method 98 291
§3 (218-222). Asymptotic Evaluation of Some Maxima 100 294
§4 (223-226). Minimax and Maximin _ 101 295
Part Three
Functions of One Complex Variable. General Part
Chapter 1
Complex Numbers and Number Sequences
§1 (1-15). Regions and Curves. Working with Complex Vari-
abIes. 103 297
§2 (16-27). Location of the Roots of Algebraic Equations . 105 300
XVI Contents
Problem Prob· Solu-
Numbers lem tion
§3 (28-35). Zeros of Polynomials, Continued. A Theorem of Page Page
Gauss 108 302
§4 (36-43). Sequences of Complex Numbers. 109 305
§5 (44-50). Sequences of Complex Numbers, Continued: Trans-
formation of Sequences . 111 307
§6 (51-54). Rearrangement of Infinite Series 112 311
Chapter 2
Mappings and Vector Fields
§1 (55-59). The Cauchy-Riemann Differential Equations 113 311
§2 (60-84). Some Particular Elementary Mappings 115 313
§3 (85-102). Vector Fields 119 317
Chapter 3
Some Geometrical Aspects of Complex Variables
§1 (103-116). Mappings of the Circle. Curvature and Support
Function . 125 322
§2 (117-123). Mean Values Along a Circle. 127 324
§3 (124-129). Mappings of the Disk. Area . 129 325
§4 (130-144). The Modular Graph. The Maximum Principle 130 326
Chapter 4
Cauchy's Theorem. The Argument Principle
§1 (145-171). Cauchy's Formula . 133 329
§2 (172-178). Poisson's and Jensen's Formulas 138 338
§3 (179-193). The Argument Principle 140 341
§4 (194-206.2). Rouche's Theorem. 142 344
Chapter 5
Sequences of Analytic Functions
§1 (207-229). Lagrange's Series. Applications 145 347
§2 (230-240). The Real Part of a Power Series . 150 354
§3 (241-247). Poles on the Circle of Convergence . 152 357
§4 (248-250). Identically Vanishing Power Series. 153 359
§5 (251-258). Propagation of Convergence 154 360
§6 (259-262). Convergence in Separated Regions . 156 363
§7 (263-265). The Order of Growth of Certain Sequences of Poly-
nomials 157 365
Chapter 6
The Maximum Principle
§1 (266-279). The Maximum Principle of Analytic Functions 157 367
§2 (280-298). Schwarz's Lemma . 160 369
§3 (299-310). Hadamard's Three Circle Theorem . 164 374
§4 (311-321). Harmonic Functions . 165 377
§5 (322-340). The Phragmen-Lindel6f Method 166 379
The following textbooks are quoted repeatedly and they are cited by the name
of the author only or by a particular abbreviation (e.g. Hurwitz-Courant; MPR.):
E. Heeke: Vorlesungen iiber die Theorie der algebraischen Zahlen. New York:
Chelsea Publishing 1948.
E. Hille: Analytic Function Theory, Vol. I: Boston-New York-Chicago-Atlanta-
Dallas-Palo Alto-Toronto-London: Ginn & Co. 1959; Vol. II: Waltham/Mass.-
Toronto-London: Blaisdell Publishing 1962.
A. Hurwitz-R. Courant: Vorlesungen iiber allgemeine Funktionentheorie und ellip-
tische Funktionen, 4th Ed. Berlin-Gottingen-Heidelberg-New York: Springer
1964.
K. Knopp: Theory and Applications of Infinite Series, 2nd Ed. London-Glasgow:
Blackie & Son 1964.
G. Kowalewski: Einfiihrung in die Determinantentheorie, 4th Ed. Berlin: Walter
de Gruyter 1954.
G. P61ya: How to Solve It, 2nd Ed. Princeton: Princeton University Press 1971.
Quoted: HSI.
G. P6lya: Mathematics and Plausible Reasoning, Vols. 1 and 2, 2nd Ed. Princeton:
Princeton University Press 1968. Quoted: MPR.
G. P6lya: Mathematical Discovery, Vols.1 and 2, Cor. Ed. New York: John Wiley &
Sons 1968. Quoted: MD.
E. T. Whittaker and G. N. Watson: A Course of Modern Analysis, 4th Ed. London:
Cambridge University Press 1952.
The following notation and abbreviations are used throughout the book:
Part One
Chapter 1
the same terms but in different order are regarded as different.) The
senes
r
represents a rational function of \!\Thich one?
5. Someone owns a set of eight weights of 1, 1, 2, 5, 10, 10, 20, 50 grams
respectively. In how many different ways can 78 grams be composed of
such weights? (Replacing one weight by an other one of the same value
counts as a different way.)
6. In how many different ways can one weigh 78 grams if the
weights may be placed on both pans of the scales and the same weights
are used as in problem 5?
7. We consider sums of the form
as a product.
8. Let El' E 2 , ••• , Es assume the values -1, 0, 1. Modify problem 7
accordingly. Let Dn denote the number of different sums of value n
Find the factorization of the following expression (function of ()
99
2: DnC.
n= ~-99
15. A set of weights 1,3,9,27,81, ... can be used to weigh any weight
that is a positive integral multiple of a given unit if both pans of the
scales are used, and this can be done in exactly one way.
16. Write
(1 + qC) (1 + qC2 ) (1 + qC") (1 + qt,8) (1 + qC16 ) •••
= ao + a1C+ a2C2 + asC s + ....
= 1- a - b + ab - e + ae + be - abe - d + ....
What is the sign of the n-th term ?
18. Prove the identity
18.1. The first and the third problem considered in the solution of 9
(concerned with An and en respectively) are the extreme cases of a
common generalization which, properly extended, includes also 18.
Formulate such a generalization.
18.2. In a legislative assembly there are 2n 1 seats and three +
parties. In how many different ways can the seats be distributed among
the parties so that no party attains a majority against a coalition of
the other two parties?
19.
1
(1 + C) (1 + C2) (1 + CS) (1 + C") ... = (1 _ C) (1 _ Ca) (1 _ CO) (1 _ C7) ..••
6, 1 + 5, 2 + 4, 1 + 2 + 3,
with odd terms
1 + 5, 3 + 3, 1 + 1 + 1 + 3, 1 + 1 + 1 + 1 + 1 + 1.
4 Operations with Power Series
as a sum of smaller positive integers. Two sums that differ in the order
of terms only are now regarded as different. E.g. only the seven following
sums add up to 4:
1 + 1 + 1 + 1, 1+ 1 + 2, 2 + 2, 1 + 3,
1 + 2 + 1, 3+1.
2 + 1 + 1,
x + 2y = n, 2x + 3y = n - 1, 3x + 4y = n - 2; ... ,
nx + (n + 1) y = 1, (n + 1) x + (n + 2) Y = O.
23. The total number N of non-negative integral solutions of the
following Diophantine equations
x + 2y =n -1, 2x + 3y = n - 3, 3x + 4y = n - 5, '"
is smaller than n + 2; moreover the difference n + 2 - N is equal to
the number of divisors of n +
2 (d. VIII, Chap. 1, § 5).
24. Prove that the total number of non-negative integral solutions
of the following Diophantine equations is n:
x + 4y = 3n - 1, 4x + 9y = on - 4, 9x + 16y = 7n - 9, ...
of
Then we have
. An
lim --= .
n-+= n 1- 1 a l a 2 ... a1(1 - 1)!
-n < X, y, Z < n
that satisfy the condition
-s < X +y + Z < s
where nand s are positive integers. The number of such lattice points is
equal to
+ +
- J
1 1t
2n2 -1 t ) s sin -
sin - 2s 2 -1 t
2:n
( ~-. t
---dt.
. t
-:l sm 2 sm 2
6 Operations with Power Series
f-l denotes any number. If f-l is a positive integer then (~ ) can be inter-
preted as the number of combinations of f-l objects taken r at a time, see
10. In the following problems we assume that n is a non-negative integer.
31.1. Prove that
( : ) = (; =~) + C~ 1)
m several ways, especially with and without a combinatorial inter-
pretation, with and without the binomial theorem.
31.2. Prove that
we have thejdentity
(x + y)* = xnlh + (:) xn~llh ylih + (;) xn~2lh y21h + ... + ynlh.
36 (continued). Prove the following generalization of the multinomial
theorem
0 for n =f= 1,
37. (n ) _ 2(n ) ) _ ... + (_lt~1 n(n) ={
+ 3(n
1 2 3 n 1 for n = 1.
38.
(: ) - ~ C~) + ~ C) - ... + (_lt~1 ~ (:)
1 1 1
=1+ 2 + 3 +"'+-;-,
39. ± (_lt~k22k(n~:~1) =n+1.
i (~ -,x r(:)x (l -
k~O
41. Put
and
t (z ~) y= coY + c1 (z ~) y + ... + cn (z ~) y. n
44. We have
t (z ~)Zk = I(k) Zk.
47. Let I(x) and g(x) be two arbitrary polynomials, assume however
that g(x) does not have any non-negative integral zeroes. The series
Y
=f(O)
g(O)
+ f(lL
g(l)
z +@
g(2)
Z2 +.L@Lz3
g(3)
+ ...
satisfies the differential equation
g(z-~)y
dz =/(z~)_l
dz 1-z .
This is soluble by quadratures.
48. Suppose that I(x) and g(x) are two polynomials relatively prime,
that the degree of g(x) is not smaller than the degree of I(x) and that
g(O) = 0, g(k) =Fe 0 for k = 1, 2, ... The series
54.
3n 2 +n
[q[ < 1.
55.
+ 1 =q3 (1 -
q3
z) (1 - qz) (1 - q2Z) + ....
This function satisfies the functional equation
1 + G(z) - G(qz) =' (1 - qz) (1 - q2Z) (1 - q3Z) ....
58. Find the coefficients of the' power series of the function G(z)
defined in 57:
G(z) = Do + D1z + D2z2 + D3Z3 + ....
59. Prove the identity
n (1 _ an) (1 _ an-I) ... (1 _ an-k-l- I )
L-~k-~- ~-~ - -- = n, n = 1, 2, 3, ...
k~l 1- a
Pt. I, Chap. 1, No. 51-60.3 11
[~J=O.
We suppose initially that q avoids the roots of the denominators (certain
roots of unity). Sometimes we shall find it necessary to emphasize the
dependence on q; then we shall use the more explicit notation [:]q
for [:].
60.1. Show that
Cf.10.
60.2. Show that
k=l
iI (1 + qk-l X ) = i
k=O
[~J t(k-l)/2 Xk.
60.4.
[n+11=[nJ+[
k-J _k k-1
n ]
qn-k+l .
Cn,k,a = Cn,k,k(n-k)-" •
(We may regard henceforth [;J as a polynomial, defined for all values
of q.)
60.6. According to the notation explained above [;Jq' is the ex-
pression which we obtain from [;J by substituting q2 for q. Prove the
identity in z
[2;11 = 1(7)
0 when r odd
when r = 2s,
[2m2s+ 1J -1 -
[2m + 1J
2s + 1 - 1 -
(m)
s .
when n is even. [Call F(q, n) the proposed expression. Then you have to
prove that
F(q, n) = (1 - qn-l) F(q, n - 2), for n > 3.
we call it a "zig-zag path". The "area under the zig-zag path" is included
by the path, the "horizontal" coordinate axis y = 0 and the "vertical"
line x = r.
60.9. The number of zig-zag paths between the street corners (0, 0)
and (r, s) is
60.10. The number of those zig-zag paths between the street corners
(0, 0) and (r, s) the area under which is IX equals cn,r,,,, (notation 60.5).
In order to specify one of the (;) zig-zag paths considered in 60.9
we view in succession the unit segments of which it consists starting from
(0, 0) and we write x or y according as the segment viewed is parallel
to the x-axis or the y-axis. Thus the zig-zag path specified by the sequence
of letters
xxyxyyx
ends at the point (4,3) and the area under it is 4.
Take any two letters in such a sequence; they form an inversion
generated by the zig-zag path if and only if they are different and y
comes before x. Thus, in our above example there are four inversions.
60.11. The number of inversions generated by a zig-zag path equals
the area under the path. (Thus 60.10 determines the number of certain
paths, or letter sequences, having a given number IX of inversions.)
§ 6. Majorant Series
Let
14 Operations with Power Series
a o + a1z
+ a2 z2 + ... + anzn + ... = A(z),
Po + P1z + P2 Z2 + ... + Pn zn + ... = P(z).
If the inequalities
A(z) ~ P(z),
in words: "P(z) is a majorant of A(z)" or "A(z) is a minorant of P(z)".
61. If A(z) ~ P(z) and A*(z) ~ P*(z) then we have also
A(z) + A*(z) ~ P(z) + P*(z)
and
A (z) A *(z) ~ P(z) P*(z).
62. If n is a positive integer we have
63. Put
From
zJ'(z) ~ 1 z +
J(z) 1 - z
deduce the inequalities
n = 1, 2, 3, ...
64. Let al' a2 , ••• , al be positive integers. Prove the following rela-
tion twice, a) by applying, b) without applying, the results of 9:
1 1
~ (1 - z a 1) (1 - z a ') ... (1 - z a I) ~ 1 - za 1 _ za t _ zas _ ... _ zat
64.1. Let Zl' Z2' ... , Zn denote the zeros of the polynomial
zn + a1zn- 1 + a2 zn- 2 + ... + an and define
Sk = z1 + z~ + ... + z!
Pt. I, Chap. 1, No. 61-64.2 • Chap. 2, No. 65 15
for k = 1, 2, 3, ... Assume that
ISkl<l for k=1,2, ... ,n.
Then
Chapter 2
Poo,
Pm' Pu>
P20' P21' P22'
Suppose that the numbers Pn. are non-negative and that the sum of
each row is 1:
Pn. > 0, PnO + Pnl + ... + Pnn = 1 for'P = 0, 1, ... , n; n = 0, 1, 2, ...
We transform any given sequence of numbers so' S1' S2' ... , Sn' •.. into
a new sequence to' t1 , t2 , ••• , tn' ... in setting
°
We have here an important particular case of a linear transformation
of a sequence into a sequence. In defining Pn. = for 'P > n we extend
our triangular array into an infinite square array, the matrix of the trans-
for'mation (Pn.) in which Pn. is the element in the n-th row and 'P-th
column.
16 Linear Transformations of Series. A Theorem of Cesaro
lim tn = S
n-+=
whenever
lim Sn = S.
n-+=
lim
n-+=
Pnv = ° for 'V = 0, 1, 2, ...
(This is a particular case of an important theorem of Toeplitz, see 80,
and III, Chap. 1, § 5.)
67. The existence of lim sn implies
n-+=
hm
. So
~-
+ 51- -
+ -52-+-...- +
--
5"
= hm
.
S
n....:;..= n + 1 n--)-oo n
68. If the sequence PI' P2' ... , Pn' .,. of positive numbers converges
to the positive value P then
n+1_.___
n-+=
lim VPOP1P2 ... Pn = p.
68.1. The numbers ao' ai' a 2 , ••• are positive and
Then lim V~~ exists also and has the same value p.
69. Reduce the computation of l~
n/:!nto the computation of
1
lim
n--.-;..=
(1 + _~)n.
n
70. Let the two given sequences
to the computation of
li (n + 1)'" - n'"
m
n-+co n(X- l '
lim p.. - 0 ·
11m q" = 0.
n .... oo Po + PI + P2 + ... + p.. - , ...... 00 qo + ql + q2 + ... + q,.
Define a new sequence
lim 1'.. = O.
n .... oo 1'0 + 1'1 + 1'2 + ... + I'n
74. Let
p~, PI> P2, ... , P.. , ... ; 10' ql' q2' ... , q,., ...
be defined as in 73, and let
and
If both these limits exist they are equal. (The proposition is of special
interest if lim Sn does not exist. If lim s" exists the proposition is a
n~~ n~=
consequence of 72.)
75. Let (f > O. If the series
is convergent, then
lim (al
n ..... =
+ a2 + a3 + ... + an) n- a = O.
(Series of this kind are called Dirichlet series. Cf. VIII, Chap. 1, § 5.)
76. Assume PI> 0, P2> 0, P3> 0, .. , and that the sequence
PI' P 2, P E, ••• , P n = PI P2 P3 + + + ... +
Pn' ... is divergent, and
lim Pnp;l = O. Then
n ..... =
(Generalization of 1 +~ +
~ + ... + ~ ,. .,." log n.)
77. Let PI' P2' P3' ... , p", ... and ql' q2' Q3' ... , q", ... be two sequences
of positive numbers for which
.
lim
PI + P2 + Pa + ... + P" = IX,
lim ql + q2 + qa + ... + qn = f3 '
n ..... = nP n n-+= nqn
Then
lim PI ql + 2P2 q2 + 3P3 q3 + ... + nPn q"
n ..... = n 2plI qn
lim an = 0
n ..... =
Suppose that all the numbers Pnv are non-negative and that the sum in
v = 0, 1, 2, ... ). Let so' Sl' ... , Sn' ••• form a bounded sequence. Define
a new sequence to' t1 , t2 , ••• , tn' ... by setting
lim
1'-+1-0
(50lPO(t) + 51lP1(t) + S2lP2(t) + ... + S"lP,,(t) + ...) = 5
holds for every convergent sequence 50' S1> S2' .,. for which lim 5" = 5
"-+00
if and only if for each fixed."
lim lP.(t) = O.
1-+1-0
. an
hm b =5.
n-+oo n
Then
and
exists, then
lim (ao + alt + a2t2 + ... + anr + ...) = s.
t-+l-O
2.; bn divergent,
n~O
then
. ao + alt + a 2 t2 + ... + ant n + ."
lim 2 =s,
t-+l-O bO + bIt + b2t + ... + bnt n + ...
lim (1 -
1-+1-0
tt (1,,-lt + 2"-lt2 + 3,.-lt3 + ... + n,,-ltn + "').
90. If 0 < k < 1 and if k converges to 1, then
f
1
dx ~~lo _ i _ [II 202.J
• V(1 - x 2 ) (1 - k 2x 2 ) 2 g 1 - k'
o
91. Let An and En be the numerator and denominator, respectively,
of the n-th convergent of the infinite continued fraction
al al
T1 + j3 + j5 + ... and so T1 + j3 + j5 + ... + 12n -3' a> O.
al al An al al al
En =
Assume that this continued fraction converges. Find its value applying
85 and using the series
E
-T
=
G(x) = 2.; xn.
n~O n.
22 Linear Transformations of Series. A Theorem of Cesaro
Then ao + alt + a 2t 2 + ... + antn + '" converges too for all values of t
and in addition
. ao + alt
+ a2t2 + ... + ant n + ...
lim 2 =S. (Cf. IV 72.)
t..,.= bo + bIt + b2t + ... + bntn + ...
95. If lim Sn = S exists then
n..,.=
. (
hm So
t----'J-oo
+ Sl -1t1 + S2 -2t 1 + ... + Sn n.I"t + .,.
• •
2 n
)
e -t = S.
exists. Define
+a t
+ a 2 2Tt + ... + an n!t + ....
2 n
g(t) = a o l l!
Then
f e-tg(t) dt = s.
o
97. The Bessel function of order 0 is defined as
Jo(x) = 1 -lTi!"2
1 (X)2 + 2!12!"2
(X)4 - '"
(_l)m (X )2m
+ m!,,;tT"2 + ....
Pt. I, Chap. 2, No. 92-97· Chap. 3, No. 98-102 23
We have
Chapter 3
Then
an
lim -= W
n-+oo n
exists; w is finite and we have
wn - 1 < an < wn + 1.
100. If the general term of a series which is neither convergent nor
properly divergent tends to 0 the partial sums are everywhere dense be-
tween their lowest and their highest limit points.
101. Let an> 0, lim an = 0 and the series a l
n..c;.oo
a2 an + + ... + + ...
be divergent. Put a1 + + ... +
a2 an = sn and denote by [sn] the largest
integer < Sn" Find the limit points of the sequence
s1 - [S1]' S2 - [S2J, ... , Sn - [sn], ...
102. Assume that there exists for the sequence t 1 , t2 , ••• , tn' ... a
sequence of positive numbers e1' e2' ... , en' ... , converging to 0, for which
tn+l > tn - en for all n.
Then the numbers t1 , t2 , ••• , tn' ... are everywhere dense between their
lowest and highest limit points.
24 The Structure of Real Sequences and Series
103. Let Yl , Y2 , ••• , Yn , .•• be positive integers, Yl < Y2 < Y 3 ::;:: ••••
n-+-v
VI V2 Vn
1+-;;:-' 2 + V 2 ' ••• , n ' .••
m = 1, 2, 3, ...
satisfying the conditions
lim Im = 0,
m-+= m-+=
Then there are infinitely many subscripts n such that two different kinds
of inequalities hold simultaneously:
InS" > I,,_l sn_1' lnsn > In_2 s ,,_2' ... , Insn > ItS1' [107,108.J
L
110. If the sequence -11,,}, ... ,
L
~
Lm
m
, ... tends to -+ (Xl and if A is
larger than its minimum [105J then there exists a subscript n (or several
Pt. I, Chap. 3, No. 103-112 25
sUbscripts n), n > 1, so that the quotients
L .. - L .. _ s L,.
- - 3 - ' "', n
1"_1'+1 + ... + 1.. _] + I" < A < 1"+1 + 1,,+2 + ... + l,,+v ,
II- - - v
I-' = 1, 2, "', n; 'JI = 1, 2, 3, ....
If A tends to infinity then so does n.
112. Let the sequence 11 , 12 , ••• , 1m> .•• satisfy the two conditions
lim 1m = 0,
m-+= m-+=
and assume 11 > A > O. Then there exists a subscript n, n > 1, such
that the inequalities
1"_1'+1 + ... + 1.. _] + I" > A > 1"+1 + 1.. +2 + ... + 1"+7
II- = = v '
I-' = 1, 2, ... , n; 'JI = 1, 2, 3, ...
hold simultaneously. If A tends to 0, then n tends to infinity.
§ 2. Convergence Exponent
The convergence exponent of the sequence r1 , r2 , r s ' ... , rm' ••. where
o< r1 < r2 < "', lim rm = =, is defined as the number A having the
m-+=
following property:
rIa + r2 a + ria + ... + r;;;a + .. ,
26 The Structure of Real Sequences and Series
converges for (J > A and diverges for (J < A. (For (J = Ait may converge
or diverge.) For (J = 0 the series is divergent, therefore A > O. If the
series does not converge for any (J then A = <Xl.
113. Show that
. logm
lim sup 1 - - =A.
m->-= og rm
114. Let Xl' X 2 ' Xa' ... , X m ' ••• be arbitrary real numbers, xm =l= O.
If there exists a positive distance b such that IXl - xk I > b, 1 < k,
1, k = 1, 2, 3, ... , then the convergence exponent of lXII, IX21, IX31, ... ,
IXm I, ... is at most 1.
115. Let p be larger than the convergence exponent of the sequence
'1> '2' '" Then there exist infinitely many subscripts n for which the
n - 1 inequalities
1
for v = n + 1, n + 2, n + 3, ...
hold simultaneously. [109.J
117. Suppose 0 < '1 < '2 < 'a < .... For what values of X, X > 0,
is the m-th term of the series
r
lim ~=<Xl.
m~oo Sm
Pt. I, Chap. 3, No. 113-121 27
Then arbitrarily large values of nand r can be found for which the
following inequalities hold simultaneously [111J:
k = 0, 1, 2, 3, ...
(At the root of this fact, and of 122, lies the comparison of two power
series
Pm
.... %m
28 The Structure of Real Sequences and Series
k = 0,1,2, ...
= a
[Consider the maximum term of L; bm zm.]
m=O m
§ 4. Subseries
°
Let tl , t2, t3, ... , tn' ... be integers, < tl < t2 < t3 < .... The series
at, + ar , + at, + ... + atn + ... is called a subseries of the series
a l + a2 + a3 + ... + an + ... .
124. From the harmonic series
1 1 1 1
1+"2+3+···+-;-+···
remove all terms that contain the digit 9 in the decimal representation
of the denominator. The resulting subseries is convergent.
125. If all the subseries of a series converge then the series is absolu-
tely convergent.
126. Let k and I denote positive integers. Must the convergent series
a l + a2 + a3 + ... be absolutely convergent if all its subseries of the
Pt. I, Chap. 3, No. 122-131 29
form
ak + a Hl + a k + 2l + a k + 31 + ...
(subscripts in arithmetic progression) converge?
127. Let k and 1be integers, k > 1, 1 > 2. Must the convergent series
a1 + + + ...
a2 a3 be absolutely convergent if all its subseries of the
form
ak + a kl + a kl , + a kl , + ...
(subscripts in geometric progression) are convergent?
128. Let !p(x) denote a polynomial assuming integral values for
integral x, !p(x) = COXl +CIX l - 1 + ...
[VIII, Chap. 2]. Assume that
the degree is 1 :2:: 1 and that the coefficient Co of Xl is positive (co> 0).
The values !p(0), !p(1), !p(2), ... form a generalized arithmetic progression
of order l; since Co > 0 only a finite number of terms of the progression
can be negative. Must a convergent series a 1 a2 a3 converge + + + ...
absolutely if all its subseries whose subscripts form a generalized arith-
metic progression
aq;(O) + aq;(I) + aq;(2) + aq;(3) + .. ,
(omitting the terms with negative subscripts) converge?
129. If the series a 1 a2 + + +' ...
a3 converges absolutely and if
every subseries
1 = 1, 2,3, ...
has the sum 0 then a 1 = a2 = ... = O.
130. Consider the set of points determined by all the subseries of
2 2 2 2
3' + '9 + 27 + ... + 3n + .. ,.
This set is perfect and nowhere dense (closed and dense in itself, but
nowhere dense in the set of all real numbers). (We have to consider all
the subseries, finite and infinite, including the "empty" subseries to
which we attribute the sum 0.)
131. Let the terms of the convergent series
PI + P2 + P3 + ... + Pn + ... = s
satisfy the inequalities
PI > P2 > P3 > "',
o < Pn < Pn+l + Pn+2 + Pn+3 + ... , n = 1, 2, 3, '"
Then it is possible to represent any number (J in the half-closed interval
30 The Structure of Real Sequences and Series
(1 +~)
P p,q = 2 (1 +~) 4 ". (1 +~)2p
Let r 1 , r2 , r3 , " ' , Sv S2' S3' •.• be two sequences of steadily increasing
natural numbers without common terms. Suppose furthermore that all
positive integers appear in one or the other of the two sequences
Pt. I, Chap. 3, No. 132-136 31
(rm < rm +l> Sn < Sn+l> rm ~ Sn for m, n = 1, 2, 3, ... ). The two series
ar, + a" + a" + ... ,
(the "reds" and the "blacks") are complementary subseries of the series
al + a2 + aa + .... Let VI> v2 , Va, ••• be a sequence of integers such that
each natural number 1, 2, 3, ... appears once and only once in it (a
permutation of the natural numbers). The series
to - 00. Provided that all the terms of one of these two subseries are of
the same sign it is possible to obtain an arbitrary sum for the whole
series by shifting the two subseries relatively to each other.
135. It is not possible to accelerate by rearrangement the divergence
of a divergent series with positive monotone decreasing terms.
136. By rearranging the series we can slow down arbitrarily the
divergence of a divergent series with positive terms which tend to 0.
More explicitely: Assume
Pn> 0, lim P..
n->-=
= 0,
such that P" + P" + ... + P'" < Q.. for n = 1,2,3 ...
32 Miscellaneous Problems
139. Suppose Pn > 0, PI > P2 > Pa > ... and that the series
cIP1 + c2P2 + caPa + ... + cnPn + "',
where ck is -lor +1, is convergent. Then
lim (cl
n->-oo
+ c2 + ca + ... + cn) Pn = O.
Chapter 4
Miscellaneous Problems
§ 1. Enveloping Series
We say that the series ao + a l + a2 + ... envelops the number A if
the relations
Pt. I, Chap. 3, No. 137 -139 . Chap. 4, No. 140-145 33
then the terms al> a2 , a3 , ••• have alternating signs and A is enveloped
in the strict sense.
146. Let the function j(x) assume real values for real x, x > R > O.
If jtx) is enveloped for x > R by the real series ao +f!!
x
+ _a;
x
+ a~x + ...
then the numbers al> az, a3 , ••• have alternating signs and the series is
strictly enveloping.
147. Suppo"e that the real valued function j(t) is infinitely often
differentiable for t > 0 and that all its derivatives r)(t) (n = 0, 1, 2, ... )
have decreasing absolute values and converge to 0 for t --+ 00. Then the
integral
J
j(t) cos xt dt
o
is, for real x, strictly enveloped by the series
-
f' (0)
~2- + f'" (0) i' (0) f'II (0)
--~~- - --';6- + ----;:s- - ....
(Example: j(t) = e- t .)
')
1 In 149-155 the terms of the series are complex numbers; they are regarded
as points in the Gaussian plane (complex plane) [Ill 1 et seqq.].
Pt. I, Chap. 4, No. 146-155 35
1(0) +1'(0)
z
+ /,,(0) + /",(0)
Z2 z2
+ ... [147] ;
---+----+
1
z
1
z2
1·3
z5
1·3·5
z?
...
(strictly enveloped if z is real).
153. Suppose that an and bn are arbitrary complex numbers that
a
have the same argument, i.e. bn is real and positive. If at a certain point
z '*' 0 the two series n
w(z) = 2 J~~dt.
o
arctan-
e - 1
2:n- n •
o n=l
162. Let aI' a2 , ••• , an' ... be positive numbers and put
~ ~ : G~ :: t G~ :: t G~ :: y... = (1 - q)2.
165. The series
x X ~,
... + f" (x) + t' (x) + j(x) + Jj(~l) d~l + Jd~l Jj(~2) d~2 + ...
o 0 0
where
VI = sin x, V2 = sin sin x, ... , Vn = sin vn _ l , •••
n = 1, 2, 3, ...
177. Compute
S =
1
cos3 P - 3 cos 3 &p + 31 cos3 3 q; -
2
2
1
33 cos3 33 q; + ....
178. Let an' bn , bn =!= 0, n = 0, 1, 2, 3, ... , be two sequences that
satisfy the conditions:
179. Let
n = 1, 2, 3, ...
be arbitrary functions, Iank I < A for all positive integral values of n
and k and
lim /",(x) = 0 if 0 < x < l.
n-+oo
n = 1, 2, 3, "',
and assume that
(*) an + a12 + a21 + a13 + a22 + a31 + ... + a1n + a2,n-1 + "',
which you obtain by arranging the numbers in the array
Sl + S2 + ... + Sn + ....
(The interesting point is that the absolute convergence of the double
series EEaik is not assumed.)
182. If iX is fixed, iX > 0, and n an integer increasing to 00, then +
2:= I
V'X-I"
(n - a
V )
-2
C"V --
n 1- (n)2
a
--- - •
• ~1 3 ex
80 V + V + V + ... +
2 81 2 82 2 8n V2, n = 0, 1, 2, ...
for n ~ 00. These expressions are well defined for all n. Non-negative
square roots are used throughout.)
184. Every value x of the interval [-2, 2J can be written in the form
or
diverges for I = 5, but converges when I is any odd positive integer differ-
ent from 5.
185.2 (continued). Let the set of all odd positive integers be divided
(arbitrarily) into two complementary subsets C and D (having no ele-
ment in common). Show that there exists a sequence of real numbers
a], a2• ... , an' ... such that the series mentioned converges when I belongs
to C and diverges when 1 belongs to D.
42 Miscellaneous Problems
(Compare 200 with 190,199 with 188, and again 199 with 191. See also
VII 54.2 and VIII 247.1.)
Define S~ as the number of partitions of a set of n elements into k
classes each of which contains more than one element.
I See, e.g., Garrett Birkhoff and Saunders Mac Lane : A Survey of Modern
Algebra, 3rd Ed. New York: Macmillan 1965, p. 137.
44 Miscellaneous Problems
S",,-2 = (n):31t~
3 4'
5"
,,-3
= (n4 ) (n - 2(n -
2) 3)
(1)
= k n
'" '" sn ~
.::.., n! =
.::.., k
Z
e(e -l)w , (2)
n~Ok;;;n
= k n
L: L: s~ wn~ = (1 - z)-W (3)
n~Ok;;;n .
either on the basis of the foregoing [10,190, 200J or independently of the
foregoing.
Pt. I, Chap. 4, No. 201-210 45
(On the other hand, we could regard (1), (2) and (3) as defining G) ,
5; and s;, respectively, and then take them as basis and starting point
for establishing the propositions, especially the combinatorial proposi-
tions, discussed in the section herewith concluded.}
Part Two
Integration
Chapter]
belonging to the subdivision xO' Xl' X 2 ' •.. , X n _ 1 ' x n . Any upper' sum is
always larger (not smaller) than any lower sum, regardless of the suh-
division considered. If there exists only one number which is neither
smaller than any lower sum nor larger than any upper sum, then this
number, denoted by the symbol
b
I
(l
f(x) dx,
is called the definite integral of f(x) over the interval [a, b~ and f(x) IS
Example:
1
t(X) =2'
x
a> O.
1 1 1
2<~<-2-'
xp v-I v xv_ 1
consequently
nx_x nx_x nx_x
~ v .-1 "" v v-I ",,' v v-I
.;..;
v=l
x2
v
< 1'=1
£.. x
v-I
x
11
< v=l
£.. x2
v-I
.
We have
i
v=l
~x--- x:_ 1= i; (!- - xl) = ~ - ~ .
v-I v v=1 v-I v
The number ~ - ~ is therefore larger than any lower sum and smaller
than any upper sum. If we can prove that no other number with this
property exists we can conclude that
b
~-~=J~dx.
a b x2
a
2:nl(
- ---- + ----
r - 1 x'-l x
1
x,-2 x 2
+ ... + - -
x x,-I·
1 1) (x - v
xv-I )
11=1 v v-I v v-I v v-I
and consider the lower sum Ln and the upper sum Un for the function
f(x) = .1.
x
that belong to this subdivision. Show that
1
U 2 =1- 2 + 31 ,
1- -
1
2
+ -31 - -41 + ... +-- 1
2n -. 1
- ~
1
2n
+ ...
6. Consider the infinite sequence whose n-th term is the n-th partial
sum of the series
s~nx + sin 2.-: + ' .. + sin.,:: + ...
1 2 n
(But F(b) - F(a) is not necessarily the only number satisfying this
double inequality for all Land U.)
Pt. II, Chap. 1, No. 3-12 49
J L1 n I < : .
10. Suppose that the function I(x) has a bounded and integrable
derivative in the interval [a, bJ. We write
L1n = JI(x) dx -
a
b b-a"
-n- L: 1( a + v -n-
~=1
b-a)
.
tends to zero like ~ as n -+ 00. More precisely, lim n 2 L1: exists; deter-
n n-+oo
mine its value.
12 (continued). The difference
find its value. Show, in addition, that L1~ > 0 if j'(a) > 0 and I"(x) > 0,
a <x< b.
50 The Integral as the Limit of a Sum of Rectangles
13. We ~Tite
1 1 1 2 2 2
Un = :n + 1 + n + 2 + ... + 2n ' V n = 2n + 1 + 2;;+-3 + ... + 4n ~ 1.
Prove that
lim Un
n-+oo
= log 2, lim Vn = log 2,
n-+oo
. 1
11m n (log 2 - Un)
n-+oo
=-4 ' lim n 2 (log 2 - V n )
n-+oo
= 312'
--.--;;
1
+~
1
+ ... + . 1
(n _ 1) ;. -
2n
-;;- (log 2n +C - log n)
Slll- Slll- Slll----
n n n
is bounded for increasing n; C is Euler's constant [solution 18J.
15.
n 1
- - n+1
--
lim e4 n 2 (11 22 33 ••• nn)-;; = 1.
n-+=
16. Suppose that (X is positive and that Xn is the (only) root of the
equation
1 1 1
2% + % - 1 + % - 2 + ... + x _ n = (X
17. Assume that (X is positive and that <is the (only) root in the
interval (n, (Xl) of the equation
1 2% 2% 2%
-;- + %2 _ 12 + %2 -"22 + ... + %2 _ n2 = (X.
Verify that
lim
n-+oo
Xn - 11 +- e-")
. (, (n-j,- -;-21) ---_- e "
= O. [12.J
18. Suppose that f(x) is differentiable and that /' (x) is monotone
increasing or decreasing to zero as x --3>- (Xl. Then the following limit
exists:
lim (tf(l)
n""""=
+ f(2) + f(3) + ... + f(n - 1) + tf(n) - j fIx) dX) =
1
S.
if'(n) < tf(l) + f(2) + f(3) + ... + f(n -1) + tf(n) - J f(x) dy - s <0
1
Pt. II, Chap. 1, No. 13-19.2 51
lim
n-+=
(1 + -.!..2 + -.!..3 + .. , + -.!..n - log n) = C
--.!..--.!..+
2 3
+-.!..+-.!..+-.!..-
444
1 1 1 1
-5-6-7-8"+
1 1 1 1 1
+9+9+9+9+9-
1 1 1 1 1 1
- 10 - 11 - 12 - 13 - 14 - 15 +
+ ...
Take the terms in the order as you read a book: from left to right and
from top to bottom. This is essential since the series is not absolutely
convergent. Its terms are non increasing in absolute value.
large values. Furthermore let t(x) be properly integrable over each closed
b
subinterval of [a, bJ that does not contain c. Then the integral f t(x) dx
is defined as the limit a
!t(x) dx
a
= l,im
B,B -++0
(7't(x) dx
a
+ /t(x) dX).
c+e'
f
o
t(x) dx = lim
W--+CO
f t(x) dx.
a
One type of improper integral may be transformed into the other type
of improper int~gral by an appropriate substitution.
20. Assume that the function t(x) is monotone on the interval (0, 1).
It need not be bounded at the points x = 0, x = 1, we assume however
1
that the improper integral f t(x) dx exists. Under these conditions
o
23. We put
= = =
~ k"~lZk ~ l{J~lZ1 = ~ anzn , ex, fJ > O.
k=1 1=1 n=1
Then
26. Let the monotone function j(x) be defined on (0, 1). Then the
equation
lim -
1 1:n j (2V- 2_-1) = f1 j(x) dx
n-+oo n 71=1 n 0
Show that this is true also if j(x) is improperly integrable but monotone.
29. If t(x) is monotone for x > 0, lim en = 0,
n-oo
C > 0, en > ~ we find
n
f(e n ) + f( en + ~) + f( en + ~) + ... + f( en + n- 1)
n n n
1
lim
n-+= n
=
0
f j(x) dx,
provided that the integral at right exists and j(x) is finite at x = 1.
31. The T-fundion is defined for ex> 0 (or ffiex > 0) by the integral
=
r(ex) = J e- x x"'-l dx.
o
Using I 89 prove that
'X-I ,
r(ex) = lim _ _ _rl
n-+= ex(ex +
n._
1) ... (ex + n -- 1) ,
ex> o.
32. As it is well known [d. E. T. Whittaker and G. N. Watson,
p. 246J Euler's constant C can be written as an integral
.l
Show that
C= hm
1--+1-0
(1- t) - -(I 1 - I
+---
t
1 - 12
+ -,-----
2
t
1- 1
+ ... +-- In
1 _ tn
+ ..)"
3
3
_log_l_] .
1 - I
33.
t
lim (1 - t) ( - - + --
12
+ --
1 In
+ ... + --- )=
+ ...3
log 2.
1-+1-0 l+t 1+t2 1+t l+tn 3
34.
35. We have
--- V--
lim
1-+1-0
V1 - t (1 + t + t! + t9 + ... + t + ...) = n2 ~
2
lim ~V~
1 - t (1
1 - + 1 - 0 ' "
1 ex
t-.J" t 3" niX
+t +
1 + + ... + t + ...) = - r (-1 '). ,:'<;
36. Compute
.
12!~= t
( 1
+ 12+21 12 + /2+22
21
+ ... + 12 2 t)
+ n + .... 2
J log (1 -
=
2x- 2 cos 2rp + x- 4 ) dx = 2n sin rp,
o
Pt. II, Chap. 1, No. 31-41 55
--
sint
t -
eit _ e- it
2it -n
-
00
.. =1
(
1-
n 2n-
2
t2 )
•
a
39. Compute the integral J log x dx as the limit of an infinite sum
o
of rectangles corresponding to the points of division
.=0 v nn
Solution:
consequently
n, ,>!:....
- 2'
thus [2nJ
v
- 2 [~]v = 1 '.
! ([!] - 2C])dx
1 ~
= lim n~1(_1
n-+= ;*:::1 v + ~
_ __ ~_)
+ v 1
= 1
2 ( :3 - 4
1
+ 5"1 -"""61 + ...) = 2 log 2 - 1 = 0,38629 ...
42 (continued). Compute
li
n~
1
-;;
(nlT + 2nz + :3
n3 + ... + nnn) .
43 (continued). Compute
j 1-X"
1
----dx.
x 1-
[VIII 4.J
o
45. Let O"",(n) denote the sum of the a-th powers of all the divisors
of n (VIII, Chap. 1, § 6) and
[(! - [~-J) XiX is properly integrable over the interval [0, 1J ifa> 0,
respectively. For the last two expressions all the a/s are supposed to be
positive. (More details in Chap. 2.)
48. Suppose that the function I(x) is defined on [a, bJ and properly
integrable over this interval. Define
b-a
<5n =--~.
n
Then
lim fIn + f2n + i 3n + ... + fnn =-= _1_ /I(x) dx,
n ..... oo n b - a·
a
1 b
n b=---;; J logf(x)dx
lim V/lnl2n/3n •.. Inn = e a ,
n ..... oo
n
lim
1 1 1 1
n-+oo
-+-+-+
fIn f2n f3n
... +-
fnn
These three limits are called the arithmetic, geometric and harmonic mean
of the function f(x). In the last two relations the greatest lower bound of f(x)
should be positive. (To improper integrals apply with caution!)
58 The Integral as the Limit of a Su m of f{ectangles
1·Im~-
Vn!
n----7'-CO n
in a way different from 169 and that the limit is equal to the geometric
mean of I(x) = x on the interval [0, 1], i.e. = -.!-.
e
50. Let a and d be positive numbers and call An the arithmetic, and
en the geometric, mean of the numbers a, a + d, a + 2d, ... , a + (n - l)d.
Then we obtain
. Gn 2
hm --=---
n--'7-= An e
Show that
sin (x - ~) = r sin x.
Then we find
J log (1 -
1
+ y2) dx =
2.n
2- 2r cos ~ log (1 - r2).
no
55. Compute
.
hm (n2 + 1)
---~~-----~~--.-
(n2 + 2) •.. (n 2 + n)
n-+= (n 2 - 1) (n2 - 2) ••• (n2 - n) •
Pt. II. Chap. 1. No. 49-60 59
- 1 - -1.)(
-1)(
(1 +<x-1 - 1 -1)(
- 1 - -1)
- .,.
2<x-1 +3<x-1 4/X-1
+ 1) <X
(n + 2) <X
(11 (n+ n) <X
+ 1) <X - 1 + 2) <X - 1
•
(n (n (n + n) <X - 1.
Whence there follows that the product on the left hand side extended
a=1+~
n'
b=1+,t
n'
d=.i..
n
Show that
+ <5)-6-.
,,-{J
lim ~ . a + d • a + 2d ••• a + (n - 1) d = (1
n-+oo b b + d b + 2d b + (n - 1) d
it
§ 7. Multiple Integrals
60. Suppose that the function t(x, y) is the second mixed derivative
of a function F(x, y),
02F(x. Y)
ox oY
= t{x
,y ,
)
in the rectangle R
c < Y < d.
60 The Integral as the Limit of a Sutp. of Rectangles
JJ log Isin (x -
O:;oz:;;y:;on
y) Idx dy = - 2 log 2.
;n2
1 82 84 ···82(n-l)
62. Let f(x, y) be properly integrable over the square 0 < x < 1,
o<y< 1. Show that
1 1
f f f(x,y)dxdy
n n [1 + ~n f (.!!..-n ,~)J
n "
lim = eO 0
"""'00 ,.=1.=1 n
63. Let f(x, y) be properly integrable over the square 0 < x < 1,
o< Y < 1. Compute
. nJ
l~!2 1. 1 + n 2 f
(n nv)J11"
1[ (1n' nV) + f (2n' nV) + ... + tn'
Pt. II, Chap. 1. No. 61-68 61
64. The three-dimensional domain 'l) is defined by the inequalities
-1 < x, y, Z < 1, -(1 <x +y + Z < (1.
dx dy dz = ~ IJt dt.
:i)
65. Let lXI' lX2' ••• , lXp be arbitrary positive numbers. define
f.(z) = l".-l z + 2".-l Z2 + ... + nap-I:!' + "', v = 1,2, ... , P
and
11 (z) Mz) •.• Ip(z) = 2: anz" .
n=l
Show that
a
lim n
n.-+oo n""+""+···+"'p-1
= II'" IX""-lX,,,·-l
1 2
.•• X"'P-1-1
P-1 (1 - x1 - x2 - ••• - XP-1 )"'p-1
n
67. Work out the product IT (1 + Ikn 15n) (same notation as in 48,
k=l
54) as a polynomial of degree n in 15n. Prove that the term containing 15~
r,
converges to the limit
are properly integrable over the interval [a, b]. Then we have
b b
f 11 (xl (h (x) dx f 11 (x) qJ2(X) dx
a a a
b b b
f Mx) qJ] (xl dx f 12 (X) qJ2(X) dx f Mx) qJm(X) dx
a a a
,
I b b I
b - a) ]
(,,) -
qJvn - qJl< ( a + vn - .
Chapter 2
Inequalities
§ 1. Inequalities
Let a 1 , a2 , •.• , an be arbitrary real numbers. Their arithmetic mean
9{(a) is defined as the expression
For @(a) and ~(a) we assume m> O. The three numbers represent
mean values of aI' a2 , ••• , an' The inequalities become equalities only if
all the a/s are equal. The mean values have the following properties:
_1 _@(~)
(lj(a) - a I ' ~(a)
1 = m(l)
a'
m(a + b) = m(a) + m(b), @(ab) = @(a) @(b), log @(a) = ~{ (log a).
Let the function t(x) be defined and properly integrable on the
interval [Xl' X2J. We define the arithmetic mean m(f) of t(x) as
J
X,
dx
f(x)
Xl
[48.J If m denotes the greatest lower and M the least upper bound of
t(x) on [Xl' x2J then
m < m(f) < M, m < @(f) < M, m < ~(f) < M.
It is understood that m > 0 for @(f) and ~(f).
The following relations are obvious:
Let aI' a2 , ••• , an be arbitrary positive numbers which are not all
equal. Then
i.e.
~(a) < @(a) < m(a).
64 Inequalities
1 It is obviously sufficient to prove 2£(a) > m(a). Here is the passage referred to:
"La moyenne gcomctrique entre plusieuys nombyes A, B, C, D, ... [5t touJours
injt'rieure a leur moyenne arithmetique.
Demonstration. - Soit n Ie nombre des !cttres A, B, C, D, '" 11 suffira
de prouver, qu'on a generalement
A+B+C+D+ ...
VA BCD ... < ---_.---
n ____
-----.--
11
( 1)
ABCD···< (
A + B + C'+ D + ...
-
)n (2)
n
Or, en premier lieu, on aura evidemment, pour n = 2,
A + B)2
AB= ( -2--- -(~-) <
A - B 2 (.4- 2+ -B)2
et l'on en conclura, en prenant successivement 11 = 4, n = 8, ... , enfin Il = 2'"
J f(x)
~
Xl
(
tl + t2 + ... + tn) < tp(tl ) + tp(t2) + ... + tp(tn)
q; n n
holds, where the t/s are arbitrary but t. =l= tj for at least one pair i, j.
(
II + t2 + ... + tn) < tp(tl ) + tp(t2) + ... + tp(tn)
q; n - n
for arbitrary points tl , t2 , ••• , tn of the interval.) If instead of the inequality
with < strict inequality is supposed «) then q;(t) is termed strictly
convex. If -q;(t) is convex, cp(t) is termed concave. (A more intuitive but
somewhat clumsier terminology would be "convex from below" for
convex and "convex from above" for concave.) In the sequel we will
consideI bounded convex functions only; these are continuous [d. 124;
110 often useful].
71. Suppose that the function I(x) is properly integrable on the inter-
val [Xl' x 2 ] and m < I(x) < M and, furthermore, that q;(t) is defined and
convex on the interval [m, M]. Then we have the inequality
cp ( X 2 -
1
Xl
x.
J f(x) dx
)< 1
X 2 -Xl
x.
Jcp[f(x)] dx.
%1 %1
66 Inequalities
72. Suppose that the function cp(t) is defined on the interval MJ em,
and that cp"(t) exists and cp"(t) > 0 on em,
M]. In this case cp(t) is strictly
convex. If we have cp"(t) > 0 only, then cp(t) is convex. (A function can
be convex in an interval where its second derivative does not exist at
all points.)
73. The functions
tk (0 < k< 1) and log t
are concave on any positive interval;
tk (k < 0 or k> 1) and t log t
are convex on any positive interval;
log (1 + et ) and Vc + t
2 2 (c> 0)
are everywhere convex.
74. Assume that cp(t) is a convex function defined on M], that em,
P1' PZ ' ••• , Pn are arbitrary positive numbers and that t1, t2 , ••• , tn are
arbitrary points of the interval em,
M]. Then we have the inequality
integrable over [xl> x2 ] and that m <f(x) < M, P(x) > 0 and j'P(x) dx> o.
x,
Let cp(t) denote a convex function defined on the interval m < t < M.
Then we have
J Pix) dx JPix) dx
Xl Xl
76. Suppose that on the interval em, M] the first and the second
derivative of cp(t) exist and that cp" (t) > O. Then we find for positive
P1' P2 ' "',Pn
( P1 ] + P2 t2 + ... + Pntn) < P1rp(t1) + P,!£(~~"-,-~nrp (tn)
t .
cp PI + P2 + ... + Pn - PI + Pz + ... + Pn
There is equality if and only if t1 = t2 = ... = tn'
77. The functions f(x) and P(x) are assumed to be continuous on the
interval [xl> x 2 ], P(x) is strictly positive and m < f(x) < M; the function
cp(t) is defined on the interval em,
M], cp(t) can be differentiated twice
Pt. II, Chap. 2, No. 72-79 67
e
< h +P2 + ... + Pn
Pi + pz + ... + Pn '
al a2 an
p,a,loga, +p,a,loga,+,,· +Pnan10gan
PIal + P2a2 + .. , + Pnan <e p,a,+p,a'+"'+Pnan
PI + P2 + ... + Pn
79. Let f(x) and P(x) be continuous and positive on the interval
[xl> x 2 ]; further, t(x) is not a constant. Then we have the inequalities
x,
I P(x)logf(x)dx
x_,_____
x, x,
I P(x) dx x,
I P(x)dx I P(x) f(x) dx
-"','-------< e x, < "'-'-,----
JX'P(X) dx
j(x) x,
j'P(X) dx
moreover
x,
J P(X)
f(x) logf(x)dx x,
I P(x)f(x)logf(x)dx
x,
x, x,
JPJ!lf(x)
dx I P(x) dx
<X'__ _
j'P(x) f(x) dx
x,
I P(x)f(x)dx
e x, < e x,
J
x, x,
P(X) dx I P(x) dx
x,
f(x)
68 Inequalities
80. Let aI' a2 , .•• , an' bI , b2 , ••. , bn be arbitrary real numbers. Show
that they satisfy the ine-quality
The case of equality arises if and only if the numbers a. and bv are pro-
portional to each other, i.e. 'Aa. + fib. = 0 for v = ], 2, ... , n, 'A2 + fi2 > o.
(Cauchy's inequality.)
81. Suppose that f(x) and g(x) denote two functions that are properly
integrable over the interval [Xl' x 2 J. Then
(Schwarz's inequality.)
81.1. The numbers aI' a2 , •.• , an' bI> b2 , ••• , bn, iX and (3 are positive,
iX + (3 = 1. Then
r r·
and the functions are integrable in the interval [xl> x 2 J. Then
/ [f(x)]" [g(x) JP •.. [h(x) J"' dx < [/I(X) dX]" [/g(X) dX] P... [j'h(X) dX])'
82. Let aI' a2 , ••• , an be arbitrary positive numbers, not all equal.
Then the function
I
(_1_ j'
1
i.e.
@(a + b) > @(a) + @(b).
The relation becomes an equality if and only if a. = J..b., v = 1, 2, ... , n.
85. The functions I(x) and g(x) are properly integrable over the
interval [Xl' x 2 J and strictly positive. Then
1 X, 1 .x z 1 xJ
- - J log[J(x) +g(x) ldx - - logf(x)dx J -- J
logg(x)dx
~-~
e x, > e ~-~
x, +e ~-~
x,
i.e.
@(f + g) > @(f) + @(g).
86. The functions 11 (X), 12 (x), ... , Im(x) are defined on the interval
[Xl' x 2 J, properly integrable and strictly positive (there exists a constant
70 Inequalities
K such that 0< K < f.(x), v = 1, 2, ... , m, Xl < X :=;: x2). Let
PI' Pz, ... , Pm denote arbitrary positive numbers. Then
0j(Pdl + Pzt2 + ... + Pmtm) > P/M(lI) + P2@(l2) + ... + Pm@(lm)'
87. Suppose that the functions t, (x), v = 1, 2, ... , m, are of bounded
variation on the interval [xl> x 2 ] and that PI' P2 ' ••• , Pm are arbitrary
positive numbers. Define
F(x) = PI!] (x) + P2!2(x) + ... + P,!,!~~.
PI+ P2 + ... + Pm
The lengths of the arcs oft1(x), t 2 (x), ... , tm(x) are denoted bylI' 12 , .•. , 1m ,
the length of the arc of F(x) by L. (At a point of discontinuity the jump
must be included in the length of the arc.) Then we have
L _
< PIll + P2 l2 + ... + pn,zm
-c--~--c-----c---:--
- PI + P + ... + Pm
2
o
JP(~) d~
is positive and continuous; furthermore
I 2" I 2"
9" f logF(x)dx 2" f logj(x)dx
e~' 0 >e 0
I.e.
@(F) > @(I).
89. Assume that t(x) is a periodic function with period 2n and that
P(x) is non-negative and properly integrable over the interval [0, 2nJ
and that its integral is positive. If f(x) is of bounded variation then this
is true also for
2"
JP(~) !(~ + x) d~
F(x) = o__ ~ ____ _
2"
o
JP(~) d~
If I, L denote the lengths of the arcs of f(x) and F(x) on [0, 2nJ then I and
L satisfy the inequality
L <I.
Pt. II. Chap. 2. No. 87-93 71
90. The arbitrary numbers al' a 2 • ••• , an and b1 , b2 , ••• , bn are positive.
We define
1
Wl,,(a) = (a~ + a~ + ... + a~)" .
Then
Wl,,(a + b) < or > Wl,,(a) + ilR,,(b) ,
according as ,,> 1 or " < 1. Equality is attained only for a. = Ab.,
v = 1, 2, ... , n, or if ,,= 1. (What does the proposition mean in the
case" = 2?) (Minkowski's inequality.)
91. The function t(x) is defined on [Xl' x2J, properly integrable and
strictly positive. We introduce
1
1<
(ai + a~ + ... + a!l (bi + b~ + ... + b!)
(albl+a2b2+ ••• +anbn)2 s:
(V~: + ffi)2
2 •
The first inequality is identical with SO. The second inequality becomes
an equality if and only if
A B
a b
k=TB n , l=TB n
-+-
a b
-+-
a b
are integers and if k of the numbers a. coincide with a and the remaining
l (= n - k) of the a.'s coincide with A, while the corresponding b:s
coincide with Band b resp.
93. Let a, A, b, B be positive numbers a < A, b < B. If the two
functions t(x) and g(x) are properly integrable over the interval [Xl' x2J
and if a <t(x) < A, b < g(x) < B on [xl> x2J then
The inequality on the right hand side is well known. The inequality on
the left becomes an equality if and only if j(x) is a constant.
94.3 (continued). The surface area of the ellipsoid is larger than the
surface area of the sphere with the same volume, that is
2
E> 4:rl:(abc) 3
unless a = b = c.
95. Let us call the ratio of the electrostatic capacity to the volume
of a conductor its "specific capacity". Show that the specific capacity
of an ellipsoid with three axes is always between the harmonic and the
arithmetic means of the specific capacities of the three spheres whose
radii are equal to the three semiaxes of the ellipsoid.
In analytic terms: we have to prove the inequalities
3 lJ=
:------:-<-
~
abc
k+w+d
<-----;:-----
be + ca + ab 2 0 V(a 2 + u) (b 2 + u) (c 2 + u) 3'
abc
are real, they are all contained in the interval with the endpoints
1
at
n
±n-'II
1 (a2 _ n~
1 -l
a)2 . 2
and assume that this series is convergent for n = 1 (and so also for
n ?:: 1). Prove that
74 Inequalities
and that
. (1 )l/n = y = 11m. -5-n .
lIm -
n-+oo sn n--+= sn+l
(If s1> S2' S3' ... are given, we have here a "perfect" scheme for com-
puting y. ThE' n-th step yields the lower bound s;;l/n and the upper bound
snlsn+l for y, the next step improves both bounds and both bounds
converge to the desired y as n tends to 00.
Observe that
z zG'(z)
sz
1
+SZ2 +SZ3
2 3
+ ... = v::-1
=
""---=-_.-
Y - z G(z) v
where
G(z) = if (1 -~)
v~l Yv
and y is the zero of G(z} nearest to the origin. Compare 197, III342.}
95.4 (continued). Even if Sn is not given for all values of n but only
for n = 1, 2, 4, 8, ... , 2m, ... , we can devise a scheme for computing y.
Prove that
1 1 1 1
- - - -
~«~)2 «~)4
51 52 54
< ... <y< ... «~)4
58
«5 )22
54
<~
52
and that
lim ( ~
5 )l,'n =y.
n--+= S~n
95.5. A wire which forms a closed plane curve C carries a unit electric
current and exerts a force F on a unit magnetic pole in the plane of, and
interior to, C. Given A, the area enclosed by C, prove that F is a mini-
mum when C is a circle and the magnetic pole is at its center.-[Assume
that C is star-shaped with respect to the magnetic pole [III 109J which is
located at the origin of a system of polar coordinates r, cp. Then
=J
:2:t
F d'P.
r
c
Express A.J
96. Assume that
n n
L.: af1V = L.: all' = 1,
1l~1 v~l
and
fl, v = 1, 2, ... , n.
Pt. II. Chap. 2. No. 95.4-97· Chap. 3. No. 98-100 75
Then we have
YIY2 ... Yn > X IX 2 ... x n·
97. Let al • a2 • ••.• a" be positive numbers. M be their arithmetic. G
their geometric mean and let 8 denote a proper fraction. Show that the
inequality
M-G
-G~<8
implies the inequalities
a·
1 + e < ~ < 1 + e', i = 1, 2•... , n
where e and e' denote the only negative and the only positive root res-
pectively of the transcendental equation
(1 + x) e- x = (1 - 8)" .
Chapter 3
§ 1. Proper Integrals
98. We define
g(x) = sin2 nx + sin2 nx cos2 nx + sin2 nx cos4 nx + ...
+ sin 2 nx COS2k nx + ...
and
G(x) = lim g(n! x).
n-+=
Is G(x) integrable?
99. Let the function I(x) (d. also 169. VIn 240) be given by
=1
0 if x is irrational
I(x) ~ q l'f x is rational, X -_ Lq , (P, q) = 1, q > 1.
101. Suppose that j(x) is properly integrable over [a, b] and cp(x)
properly integrable over [a, b + dJ, d > O. Then
b b
lim
6-++0 a
f j(x) cp(x + J) dx = f
a
j(x) cp(x) dx.
102. Let j(x) denote a properly integrable function on [a, b]. There
exist to every positive number c two step-junctions, 1p(x) and P(x), such
that for the entire interval [a, b]
1p(x) s j(x) < P(x)
and
b b
f P(x) dx - f 1p(x) dx < c.
a a
It is even possible to choose 1p(x) and P(x) so that their points of discon-
tinuity are equidistant.
103 (continued). If j(x) is of bounded variation 1p(x) and P(x) may
be chosen so that the total variation of neither exceeds the total varia-
tion of j(x).
104. We define
4[x] - 2[2x] + 1 = s(x).
Then we have (n integer) the limit relation
1
lim
n~oo 0
f j(x) s(nx) dx = 0
for any properly integrable function j(x) on [0, 1]. [Sketch s(nx), VIII 3.J
105. Let j(x) be properly integrable over [a, bJ. Then we can prove
that
b
lim f
n-..-,..oo a
j(x) sin nx dx = o.
106 (conunued). Yet
2
f f
b b
lim j(x) Isin nx I dx = -;- j(x) dx.
n--:'J. OO a a
Suppose that the function j(x) is bounded on the interval [a, bJ and
that this interval is subdivided by the points x O' Xl' X 2 ' ••• , X n - l ' X n '
Pt. II, Chap. 3, No. 101-113 77
whereby
The greatest lower and the least upper bounds of f(x) in the subinterval
[x._I> x.J are denoted by m. and M. respectively. Then M. - m. is
called the oscillation of j(x) on the v-th subinterval. The function j(x) is
properly integrable over [a, bJ if and only if to every pair e, fJ of positive
numbers a subdivision of the interval can be found so that the total
length of the subintervals for which the oscillation is larger than e is
smaller than fJ. [Riemann's criterion, d. l.c. 105, Riemann, p. 226.J
107. The function (: - [: J) XIX is properly integrable over [0, 1J
for IX > o.
108. If j(x) is properly integrable over [a, bJ the points of continuity
of j(x) are everywhere dense on this interval.
109. The function j(x) is properly integrable over [a, bJ. The equation
b
f (t(X))2 dx = 0
a
§ 2. Improper Integrals
J
00
r
114. Determine those pairs of real values IX, (J for which the integral
f
~
[I(x)]", dx
o
converges for IX = 1 but diverges for any real value of IX different from l.
115. The functions
... ,
are properly integrable over any finite interval and they satisfy the
following conditions:
On every finite interval we have lim In(x) = I(x) uniformly in x.
n-+~
J F(x) dx
~
lim
n-+= -00 -00
[Analogous to I 180.J
116. Prove 58 using VI 31.
117. If the Dirichlet series [VIII, Chap. 1, § 5J
a J l- S + a 2- + as3- + ... + ann- + ... =
2
s s s D(s)
Des) res) =
o
where
a1e- Y + a2e- 2Y + aae-SY + ... + ane- ny + ... = P(y).
118. Suppose that I(x) is properly integrable over every finite interval
J \/(x) \ dx exists. Then we have
~
and that
2
J I(x) sin nx dx = J J
+~ +~ +~
lim 0, lim I(x) \sin nx \ dx = - I(x) dx.
n~oo -'X) n-+oo -co 'Jl -00
Pt. II, Chap. 3, No. 114-120 79
118.1. There are rational functions R(x) such that for an arbitrary
function /(x) of the real variable x
= =
J /(R(x)) dx = J /(x) dx
provided that the integral on the right hand side exists. Show that this
property belongs to those, and only to those rational functions that are
of the form
R(x) = e(x-IX -~-~-
x -
•.. -~)
<Xn
<Xl X - <X2 X -
where 13 = +1 or -1, IX, lXI' IX2' ... , IXn are real, and PI' P2"'" P~ are po-
sitive, numbers; n > O. (The case n = 0 must be interpreted as meaning
R(x) = e(x - IX).)
121. Assume that the function f(x) is differentiable on [a, b], but
not a constant and that f(a) = f(b) = O. Then there exists at least one point g
on (a, b) for which
4
(b - a)2 aJ f(x) dx .
b
If'(~) I >----
122. The function f(x) is twice differentiable. Then there exists a
point ~ in (xo - r, Xo + r) where
w= VUV;
assume that f(x) > 0 for a ~ x ~ b.
122.3 (continued). Analogous question for
2UV
w= u:tv;
assume that f(x) > 0 for a < x < b.
123. The numbers Po' PI' P2' ... , Pn' ... are non-negative, at least two
of the P. 's do not vanish. Then the logarithm of the series
Po + hex + P2e2x + .. , + Pn enx + ...
is a function of x, convex on every interval where the series converges.
124. A bounded convex function [po G5J is everywhere continuous
and it is even everywhere differentiable from the left and from the right.
Pt. II, Chap. 3, No. 121-129 81
a
J P.. (t) dt = 1, n = 1, 2, 3, ... ,
lim
n-+oo
(7 a
e
pn(t) dt +z+e; P,,{t) dt) = 0
for all positive values of e for which a < x - e < x e < b (if x = a +
or x = b the first or second integral resp. under the limit sign has to be
omitted.) (Cf. 166, 180, 184.)
82 Some Properties of Real Functions
130. We have
provided that the integral on the left and the limit on the right hand side
exist.
131. If the integral
J fl(t) dt
00
o
converges for A = IX and for A = {3, IX < {3, it converges for IX < A < {3
and it represents a continuous function of A on that interval.
132. We assume that
Pl(X, t), P2(X, t), "', Pn(x, t), ...
are continuous functions of x and t, a < ; < b, and that for each n
b
J Pn(x, t) dt = 1.
a
Let I(t) denote a continuous function. The functions
b
In(x) = J Pn(x, t) I(t) dt, n = 1, 2, 3, ...
a
lie between the minimum and the maximum of I(t) on [a, bJ for any x,
a <x< b; i.e. min I(x) < In(x) < max I(x). Furthermore
a~x~b a~x~b
lim
n-+oo
(7 epn (x, t) dt + JPn(x, t) dt)
a x+e
= 0
lim -~
2"
J I(t)
%-t)2dt
(_ _ sinn--
2_ = I(x) ,
n--,>oo 2nn 0 sin % - t
2
the convergence is uniform for all x.
Pt. II, Chap. 3, No. 130-141 83
a
b
a
b
J P(x) dx - J P(x) dx < e ( !
2",
PCX) dx - !
2", )
P(X) dx < e .
can be written as an integral [31]. Use this fact to prove that T(s) does
not have any zeroes. [T(s + 1) = sT(s) , 142J.
n = 0, 1, 2, ...
1= i (:)
V~O
xV(l - x)n--v = EI + Ell,
where EI refers to the subscripts for which I'll - nx I ::::: n 3/ 4 and Ell to
those for which I'll - nx I > n 3 /4 , n > 1. Then
1
"II 1 --.,
£.. <-"4 n ~.
146. Let f(x) be continuous on [0, 1J. The polynomials Kn(x) converge
uniformly to f(x) on [0, 1]. (New proof of \Veierstrass' theorem, 135.)
Pt. II. Chap. 3. No. 142 -146· Chap. 4. No. 147 -150 85
Chapter 4
N(r)
Iimsup--
1'-+00 r
= l'Imsup-,
n-+=
n
rn
· III
11m . f-
1-+00
N(r)
-
r
= Iin-+oo
. f n
m In - ,
rn
·
11m log N(r) l' log n . flog
IimIn - - = l'1m1n
- -N(r) . flog
- -n.
sup , - - - = 1m sup - - .
r ..... = log r n ..... = log r n r ..... = log r n ..... = log r n
lXi > o.
[logk X = logk_l (log x).] Then L(r) is slowly increasing.
152. If L(r) is slowly increasing then
lim log L(rL = O.
r-+= log r
153. If N(r) denotes the counting function of the sequence
•.• , rn' ... and if
rl' r2 , r3 ,
N(r) '" -IL(r),
where L(r) is slowly increasing, 0 < it < 00, then it is the convergence
exponent of the sequence rl , r2 , r3 , ••• , rn' ...
156. The limit relation in 155 is also valid if t(x) denotes a properly
integrable function on [0, c].
157. Let N(r) denote the counting function of the regular sequence
r 1 , r2 , " ' , rn' ... with convergence exponent it and let IX > O. Then
lim
1
N() 2: (r~ )0:-'< = f1 x-A O:-A
dx = - .
A.
1-+00 r rn~' r 0 ex:
Pt. II, Chap. 4, No. 151-161 87
158 (continued).
lim N()
1
1:
(r~ )-"'-A = J x -"'-A dx = -.
= A.
A
r--+= r 'n>r r 1 ex
159. Assume that N(r) is the counting function of the regular se-
quence r], r r
2 , ••• , n , ... with convergence exponent A, that f(x) is defined
for x> 0 and properly integrable over every finite interval [a, bJ,
0< a < b and that furthermore
if(x) i < X"'-A III a neighbourhood of x = 0
and
if(x) i < X-ex-A III a neighbourhood of x = 00, IX> O.
Show that
. 1
::n;,N(r)
=
~f -;-
(Y )
n
=0
Jf = ( 1\
TI
x fdx.
160. Suppose that the function f(x) is defined and monotone on the
interval (0, 1J and that it satisfies in the neighbourhood of x = 0 the
inequality
IX> O.
The counting function and the convergence exponent of the positive
sequence r1 , r2 , ••• , rn' ... are called N(r) and A resp.; let 0 < A < 00.
Then
rn
The sequence need not be regular.
161. The function f(x) is defined for x > 0, is positive and decreasing
and satisfies the inequalities
f(x) < X"'-A in the neighbourhood of x = 0
f(x) < X-"'-A in the neighbourhood of x = 00, IX> O.
The sequence rrr
1, 2, 3 , ••• , rn .... is defined as in 160. Then
[1116.J
§ 2. Criteria of Equidistrlbution
A sequence of the form
88 Various Types of Equidistribution
is called equidistributed in the interval [0, 1J if all the Xl' X 2 ' X 3 ' ••• , X n ' ••.
are on [0, 1J and if for every function that is properly integrable over
[0, 1J the following equation holds:
. vn(cx,fll
lim - - - = f3 -lX. [102.J
n-+= 1'1,
163. Let [lX, f3J be an arbitrary subinterval of [O,lJ and Sn(lX, f3)
denote the sum of the xv's, l' = 1, 2, , ... , n, that fall into [lX, f3J. A se-
°
quence Xl' X 2 ' X 3 ' ••• , X n ' ••• , < X < 1, is equidistributed if and only if
. sn(cx, fl) fJ2 - cx 2
hm---=---.
2
°
-n-+= n
164. A sequence Xl' X 2 ' X 3 , •.• , X n ' •.• , < Xn :::;; 1, is equidistributed
on [0, 1J if and only if for every positive integer k
. x~+x~+x~+, .. +x~
lim- [137.J
k + 1.
°<
n-+= n
165. A sequence Xl' X 2 ' X 3 ' ••• , X n ' ... , Xn < 1, is equidistributed
on [0, 1J if and only if the two equations
is irrational. [VIn 258.J Prove that the only limit point of the set
n! e - en! eJ, n = 1,2,3, ...
ISzero.
172. Suppose that the polynomial P(x) = a1x a2 x 2 + arxr + + ...
has at least one irrational coefficient. Then the numbers
P(n) - [P(n)], n = 1, 2, 3, ...
have infinitely many limit points.
90 Various Types of Equidistribution
°
are equidistributed on the interval [0, IJ.
175. Suppose that a > 0, < (1 < 1. The sequence
Xn = an" - [anaJ
is equidistributed on the interval [0, IJ.
176. Let a > 0, (1 > 1. The numbers
xn = a (log nt - [a (log ntJ
°
are equidistributed on [0, IJ.
177. For 0< (J < 1, ~ =1= the series
lim
'1--+-00
f(x l ) + f(x 2) + ... + f(xn)
n
=
0
j I(x) K(x, ~) dx
holds provided that n increases to infinity in such a manner that xn --+ t
I
o <~ ~ 1. The function K(x,~) is given by
log q If-HI if 0 < x< ~
q- 1
K(x,~) =
log q
q-l
If-; if ~ <x< 1, q= ella, 0 < ~ < 1;
log q
K(x, 0) = K(x, 1) = -
q-1 If·
covers the interval [f(~ - 0), M + O)J. The statement is true also for
~ = 0 or ~ = 1 if /(1) = frO) and if f(x) is extended so that it becomes a
periodicfunction with period 1. [Then f(l+ 0) =f( +0), f(l - 0)= f( -0).]
183. The seq uence
xn = a (log nt - [a (log nY'], n = 1, 2, 3, ...
is for 0 < (J < 1 everywhere dense on the interval [0, 1J but not equi-
distributed. [176, 179.J
184. Assume that the square roots of the logarithms of the
natural numbers 1, 2, 3, 4, . " are tabulated below each other in an
infinite array. Consider the digits at the i-th decimal place (to the right
of the decimal point) i > 1. There exists no definite probability for the
distribution of the digits 0, 1, 2, ... , H at the i-th decimal place. More
exactly: Let Yg(n) denote the number of integers k among the first n
integers for which Vlog k has the digit g at the j-th decimal place. Tht'n
v (n)
the quotients-f - - , n
n
= 1, 2, 3, ... are everywhere dens<' between 0 and 1.
J J J... Jf(x) , x
. 1 t 1 1 I
hm - f(x 1 (t), x 2 (t), ... , xp(t)) dt = 2 , ..• , xp) dX 1 dx z ... dxp .
/-7= t o o 0 0
186. Let (Xv (X2' f31' f32 be arbitrary constants, 0 ~ (Xl < (X2 < 1,
o ~ f3I < f32 < 1. The conditions
Pt. II, Chap. 4, No. 183-189 93
determine an infinite number of rectangles with sides parallel to the
axes and congruent mod 1, i.e. they are congruent by translations
parallel to the axes through integral lengths. The equations x = a + ()It,
y = b + ().j, a, ()l' b, ()2 constants, t time, define a linear uniform motion.
Let T(t) denote the sum of the time intervals up to time t the moving
point is spending in one of the above mentioned rectangles. In the case
where ()l: ()2 is irrational we can establish the relation
lim
t .... oo
Tt(t) = (1X2 - IXI) (P2 - Pl) .
The numbers
3 n
n'···' -;;-, n = 1, 2, 3, ...
holds for any properly integrable function j(x) on [0, 1J. [VIn 35.J
189. We write down in increasing order all reduced fractions < 1
whose numerators and denominators are among the numbers 1, 2, 3, ... , n:
(Farey series
11 = 0, 1, ... , n; n = 1, 2, 3, ...
Then
,,~~f(Son} + f(sln} + ~2n} + ... + f(snn} = _Z~f(V ~ e-2X')dX.
191. Let
-1 < xvn < 1, 11 = 1, 2, ... , n,
I'
log (1 + X1n)
;.
+ log (1 + X2n)
;.
+ ... + log (1 + X;.nn ) 1;' +
V--
).2 - 1
n~ n = og 2). ,
where the positive value of the square root is considered. [Use 203.J
192 (continued). Let k be any positive integer. Show that
The points xvn are not equidistributed on the interval [-1, 1J but the
values arccos xn are equidistributed on [0, nJ. We may interpret the
interval [-1, 1J as the horizontal diameter of a circle and each point x
as the normal projection of two points of the circumference onto the
diameter. We are facing here an equidistribution on the circumference
but not on the diameter.
Chapter 5
§ 1. Laplace's Method
195. Let PI' P2 , •.. , PI' a], a2 , .•• , aJ be arbitrary positive numbers.
Then
exists and it is equal to the largest among the numbers aI' a2 , ••• , al •
196. Under the same hypotheses as in 195
197. Let t(x) be an arbitrary polynomial whose zeros are all real
and positive and for which
_I'(x) = c
f(x) 0
+ cI x + C x2
2
+ ... + cn
xn ...
'
Show that
198. The two functions <p(x) and I(x) are continuous and positive on
the interval [a, b]. Then
!~~ V/ r(x)[f(x~~ dx
exists and is equal to the maximum of I(x) on [a, bJ.
199. Under the same hypothesis as in 198
b
Jcp(x) [f(x)t+l dx
lim a b = max j(x) .
n->= Jcp(x) [f(x)t dx
a
200. Let k be a positive constant and a < $< b. Show that for
a, b, ~, k fixed and n ~ 00
b
J e-kn(x-~)' d
~-
ll n
a
x r N kn'
201. The functions r(x), h(x) and j(x) = eh(x) are defined on the
finite or infinite interval [a, bJ and satisfy the following conditions:
r
(1) r(x) [f(x) = r(x) enhl,x) is absolutely integrable over [a, b];
n = 0, 1, 2, '"
(2) The function h(x) attains its maximum only at the point $ in
(a, b); moreover, the least upper bound of h(x) is smaller than h(~) on
any closed interval that does not contain ~; there is, furthermore, a
neighbourhood of $ where hlf (x) exists and is continuous; finally hlf (~) < 0,
(3) r(x) is continuous at x = ~, r(~) =f= 0. Then the following asymp-
totic formula holds as n ~ 00 1
a
b
-
= r(~) enh(~) V~ ;!~~)'
.-----~--
1 On the use of such integrals Laplace has this to say: .. _ On est souvent conduit
it des expressions qui contiennent tant de termes et de facteurs, que les substitutions
numeriques y sont impraticables. C'est ce qui a lieu dans les questions de probabilite,
lorsque l'on considere un grand nombre d'evenements. Cependant il importe alors
d'avoir la valeur numerique des formules, pour connaitre avec quelle probabilite
les resultats que les evenements developpent en se multipliant sont indiques. II
importe surtout d'avoir la loi suivant laquelle cette probabilite approche sans cesse
de la certitude qu'elle finirait par atteiwlre, si Ie nombre des evenements devenait
infini. Pour y parvenir, je considerai que les integrales definies de differentielles
multipliees par des facteurs eleves it de grandes puissances, donnaient par l'integra-
tion, des formules composees d'un grand nombre de termcs et de facteurs ...
He adds the following remark on his method of which 201 describes the first
step: .. _ un procede qui fait converger la serie avec d'autant plus de rapic1ite, que
la formule qu'eJle represente est plus compliquee; en sorte qu'il est d'autant plus
exact, qu'il devient plus necessaire . _. (Essai philosophiquc sur les probabilites'
Oeuvres, VoL 7. Paris: Gauthier-Villars 1886, p. XXXVIIL)
Pt. II, Chap. 5, No. 198-207 97
[We consider only a neighbourhood of E and expand h (x) in powers of
(x - E) up to terms of the second order.]
202. Let n be an integer, n -+ + 00. Using the fact that
" "
J2'sin2n x dx = J2'cos2n
o 0
X dx =
1 . 3 ..• (2n - 1) n
2·4···2n
-2 ,
prove that
1 • 3 ..• (2n - 1) 1
2·4 .. ·2n N ynn '
203. We assume that Ais real, A > 1; Pn (x) denotes the n-th Legendre
polynomial. As 'n -+ 00
1 (4 + VA2=1)n+1/2
Pn(A) N _
y2nn ' .
VA2 - 1
eilCOS% = ] o(t) V
• =1
t -+ + 00, v = 0, 1, 2, ...
( nk + 1) N (k - 1t (_k_)nk+I+2'
n Y2nn k - 1
!=
exp ~:
!"
-TX
)
dx C'J V;1--::';T
, ) , -c-"----1
:1(1-.')
(1,' --")
exp -:~ T 1-., .
holds where
A = _1_
(\
Ie
t2
2 dt, B= I~-
1 (f3-- 3+ 2)- e -.,-.
"
():2
V2n 12n
211. We denote by I. a positive proper fraction and hy xn the only
positive root of the transcendental equation
1+ ;! + ;C + ... + :!
2 n
= I.e x [V 42].
Xn = n + ex Vn + f3 +-0(1),
where ex and f3 satisfy the equations:
t'
I
1 =-Q
7= e - dt = I., f3 = ~_-1 ~
3 .
}2n "
212 (Continuation of 201). Let ex denote a real constant. Then, for
n --+ 00,
I
a
<p(X) [f(x)r dx C'J g;(;) [f(;)r '1/2 "1'
I
~"
-nf"(!;)
,-- I e :1 dt
213. The functions g;(x), h(x) and f(x) = eh(x) are defined on the finite
or infinite interval [a, bJ and satisfy the following conditions:
Pt. II, Chap. 5, No. 208-216 99
r
(1) <p(x) [f(x) = <p(x) enh(x) is absolutely integrable over [a, b].
n = 0, 1,2, ...
(2) The value of the function h(x) at a point ~ of (a, b) is larger than
its least upper bound in any closed interval to the left of ~ which does
not contain ~. Moreover there is a neighbourhood of ~ in which h" (x)
exists and is bounded. Finally h' (~) > O.
(3) <p(x) is continuous at x = ~, <p(~) =1= O.
Prove, for n _ 00, the asymptotic formula
<+~o~+£
j n <p(x) [f(x) r dx ~ :.~~) e[3h'«) • n<xh'«) -1 . enh «) ,
a
215. Suppose that n is odd and let -xn denote the only real root
of the equation
X %2 xn
1 +1T+2T+ ... +;T=O [V 74J.
As n _ 00 xn is asymptotically given by
xn = ~n + (X log n + (3 + 0(1),
where ~ is the only real root of the equation e1+<~ = 1 and (X and (3 are
given by
216. Assume that the function g(x) is monotone increasing for positive
x and that
lim g(x) = + 00, lim g(x) = o.
x-++ 00 x-+'+oo .X'
We define
100 Functions of Large Numbers
a a
where the functions h! (x), h2 (x) , ... , hn(x) are positive on (a, b) and attain
their maximum at the same interior point ~. Then we approximate
h.(x) = h.(~) + t h;'(~) (x - ~)2 + ... by h.(~) + t h;'(~) (x _ ~)2
and the integral by
We have supposed that cp(~) =f= 0, moreover h;(~) = 0, (~) < as It:' °
condition for the maximum at the point~, and
-h~ (~) - h~ (~) - ... - h~ (~) = s. The method can be justified m
many instances and it is capable of adaptation and refinement.
217.
. f"
lim
n! 22nc050
n~oo • I(2ne''0 - '0
1) (2ne'
'0
- 2) (2ne' - 3) ••• (2ne' - n)
'0
I df} = 2n.
-"
[Put f} = V~ and recall 59, 115.J
217.1. Analogy to 201 suggests sufficient conditions under which
for n -+ + <Xl
JJcp(x, y) enh(x,y) dx dy ~ cp(~, 1]) enh(~,'7) 2n 2
m n VhXXhyy - h XY
where the partial derivatives of second order hxx' hyy, and hXY are taken
at the point (~, 1]). Give a full statement and a proof.
where a > 1, has the maximum Mn on the interval (n, + =). It can be
approximated by
Mn 1 1
[16.J
Y2n
- (Xl - - - - - " - - ' - - : ; - - , ; 0
n! (a - 1t+1/2'
219. The function
x(x 2 _ 12) (x 2 _ 22) ..• (x 2 _ n 2 ) a-x,
where a > 1, has the maximum Mn on the interval (n, + =). It can be
approximated by
Mn
n!2
(Xl ~
2n
(2 y; )2n+1
a _ l' [17.J
220. We define V% = Qo(x),
V:; (1 - :) (1 - ;) ... (1 - =) = Qn(x) , n = 1, 2, 3, .. ,
The sequence of functions
QI (x) a-x, Q2(X) a-x,
is uniformly bounded for x > 0 if a > 2; it is not uniformly bounded if
0< a < 2.
221. We define x = Po (x),
n = 1, 2, 3, ...
lim
. M
( I n
)n-I< =e.
-a
n-.+oo n.
is a continuous function of x.
102 Functions of Large Numbers
(the minimum of the maxima) refers to the lowest point in the skyline.
*224 (continued). Show that
max min t(x, y)
y x
s min max t(x, y) .
x y
*225 (continued). Which one of the two signs, < and =, is valid in
the example
I(x, y) = 1 - (x - y + 1)2,
a = 0, a' = 2, b = 0, b' = 4?
*226. Add to the assumptions of 223 that t(x, y) > 0. Then
1
lim
n--7-CX)
[I (I[f(x,y)rdy)-l dxl
a b
n
= min max I(x, y).
x y
Part Three
General Part
Chapter 1
I; ~ ;1
is either < 1, or = 1, or > 1, and so the whole plane is divided into three
subsets. Describe them.
7. Let IX, {3 be real; a complex; IX, {3 and a are fixed. Suppose that the
complex variables Zl and Z2 satisfy the relation
IXZ1Z1 + iiZlZ~ + az1z2 + {3Z2Z2 = O.
If 1X{3 - aa < 0 the points Zl
Z2
lie on a circle, possibly a line segment. (The
left hand side of the equation is called a Hermitian form of the variables
Zl and Z2')
8. Let a and b be positive constants and the real variable t signify
time. Describe the curves given by the three equations
+ z(z-1)···(z-n+1)
1·2 ... n "';'(Z)
+ ... = n=O
"' . n
(binomial series for (1 + t)' with t = 1 and complex z) larger than the
absolute value of any other term of this series? n = 0, 1, 2, ...
13. We put
Po(z) = z, Pn(z) = z(l- ::) (1- ~:) (1- ~:) '" (1- ::),
n = 1, 2, 3, ...
For what values of z is \ Pn(z) \ larger than \ Po(z) \' \ PI (z) \"'" \ P n- I (z) \'
\ Pn+1 (z) \' ... ? (Pn(z) is the n-th partial product in the product expansion
of sin ~:.)
:n:
14. We assume that the real functions t(t) and q;(t) are defined on the
interval a < t < b, that t(t) is positive and continuous and q;(t) properly
integrable. Then
Equality holds if and only if the function q;(t) assumes the same value
mod 2n at all its points of continuity.
°
15. Let q;(t) be defined for t > and be properly integrable over any
finite interval. If
f f
00
= Q)
00
e-(t+i<p(t»dt = P, e- 2 (t+i<p(t» dt
o o
then
\4P2 - 2Q\ < 3.
Equality holds if and only if q;(t) assumes the same value mod 2n at
all its points of continuity.
aozon + a1zn-I
O + a zn-2 +
2 0 ••• + an_1 Zo + an = 0 .
(zo is a root of the algebraic equation P(z) = 0.) If zl' Z2' ... , zn are the n
zeros of the polynomial P(z) we can write
P(z) = ao(z - Zl) (z - Z2) •.• (z - zn)
as is proved in algebra.
16. A polynomial of the form
zn - PIzn- 1 - P2Zn-2 - ... - Pn-lz - Pn'
where PI 2 0, P2 > 0, ... , Pn 2 0, PI + P2 + ... + Pn > 0, has just one
positive zero.
17. If Zo is a zero of the polynomial
zn + a1zn- 1 + azzn - 2 + ... + an
then IZo I is not larger than the only positive zero C of the polynomial
zn -laIlzn - l -la z lzn- 2 - ••• -Ian!.
18. Assume an =f= O. The absolute value of none of the zeros of the
polynomial
P(z) = zn + a1zn-l" + azzn-- 2 + ... + an
is smaller than the only positive zero Cof the polynomial
zn + lallz l + la2lzn-2 + ... + lan-llz-Ianl.
n-
19. All the zeros of the polynomial zn + C are on the circle centred at
z = 0 with radius IC Il/n.
*20. Let cl ' c2 ' ••• , Cn be positive numbers and
c1 + c2 + ... + C n < 1.
The absolute values of the zeros of the polynomial
zn + al zn- 1 + a2 z n- 2 + ... + an
are not larger than
-
M -max (~i
CI
J Y 1
a2 1
c2
, ••• , VIiL,,}
Cn
) •
Vn lanl;
3
Vn la 2 1, ~, ••• J
Pt. III, Chap. 1, No. 16-25 107
also they are not larger than the largest of the numbers
k = 1, 2, 3, ... , n;
and they are certainly smaller than the largest of the numbers
k __
2 Vlakl, k = 1, 2, ... , n.
22. Assume
Po > PI > P2 ... > Pn > O.
The polynomial
Po + PIz + P2 z2 + ... + Pnzn
cannot have a zero in the unit disc Iz I < 1.
23. Suppose that all the coefficients PI' P2' •.. , Pn of the polynomial
pozn + PIZn- 1 + ... + Pn-Iz + Pn
are positive. Then the zeros of this polynomial lie in the annulus
IX < Iz I < f3, where IX is the smallest, f3 the largest among the values
... ,
24. Let ao' a], a2, ... , an be digits (if\. the ordinary decimal notation,
that is integers between 0 and 9 inclusively) n > 1, an > 1. Then the
zeros of the polynomial
ao + a1z + a2z2 + ... + an~
are either in the open left half-plane or in the open disk
the boundary whereof is the set of those points at which the line
segment a, b subtends the angle ~. Show that to each point y on
n
the line connecting IX and (3 there exists a point z in tr such that y = P(z).
Then there are points Zv on both sides of any straight line through Z
except when all the z:s lie on that straight line.
We can interpret the numbers m1 , m z' ... , mn as masses fixed at the
points z]> Z2' ••• , zn' Then the point Z defined in 30 is the center of gravity
of this mass distribution. If we consider all such mass distributions at
the points ZI' Z2' .•• , zn the corresponding centers of gravity cover the
interior of a convex polygon, the smallest one containing the points
ZI' Z2' ••• , Zn' The only exception arises when all the points are on a
straight line. Then the centers of gravity fill out the interior of the smal-
lest line segment that contains all the points ZI' Z2' ... , Zn-
31. The derivative P'(z) of P(z) cannot have any zeros outside the
smallest convex polygon that contains all the zeros of P(z) (considered
as points in the complex plane). Those zeros of P'(z) that are not zeros
Pt. III, Chap. 1, No. 26-36 109
of P(z) lie in the interior of the smallest convex polygon (the smallest
line segment) that contains the zeros of P(z).
32. Let Zl' Z2' •••• z,. be arbitrary complex numbers. z" =l= zp for all
I' =l= 11. P..1I = 1. 2•...• n. We consider all the polynomials P(z) that
vanish only at the points Zl' Z2' •••• z,. (having there zeros of arbitrary
order). The set of the zeros of the derivatives P'(z) of all these polynomials
is everywhere dense in the smallest convex polygon that contains
Zl' Z2' •••• z,..
33. Let P(z) denote a polynomial. The zeros of cP'(z) - P(z), c =l= 0
lie in the smallest (infinite) convex polygon that contains the rays
parallel to the vector c starting from the zeros of P(z).
A zero of cP'(z) - P(z) appears on the boundary of this region only
in one of the following two cases: (a) the zero in question is also a zero
of P(z); (b) the region in question degenerates into a ray.
34. Let (h. f!2' •••• (!p be positive numbers, a l • a2 , ••• , a p arbitrary
complex numbers. and let the polynomials A (z) and B(z) of degree p
and p - 1 resp. be related by
B(Z)=~+~+ ... +~.
A (z) z - al z - a2 z - ap
37. Suppose that the numbers Zl' Z2' ... , zn' ••. are all in the half-
plane ffiz > 0 and that the two series
+ Z2 + ... + Zn + ... and zi + z§ + ... + z! + ...
Zl
n = 1, 2, 3, .. ,
fill out the entire circle with radius (1 + (X2)-1/2 and center at the origin.
[The expression in question is closely related to a sum of rectangles.]
41. Find the locus of the limit points of the complex sequence
Zl' z2' ..• , Zn' ... , where
Zn = (1 + +) (1 + ~) (1 + ~) ... (1 + ~).
42. Put
and connect the points Zn-1 and Zn by a straight line. The distance
between these two points is always 1. The polygonal line connecting the
successive points approaches with increasing n an Archimedean spiral;
1'f Zn = rn eiq; n, rn > 0, 0 < ({In - ({In-1 < 2'
n th en
· rn - rn-l
11m
1 . rn 1
=-, hm-=-.
n->-= f{!n - f{!n-l 2 n->-= f{!n 2
it
43. Let t be a fixed real number and put Z = 2neY;; . Then
n->-=
·
11m V n
--
n z(z - 1)(z -
2 z n.,
2) •.• (z - n)
=e -I' .
we transform an arbitrary infinite sequence zO' ZI' ... , Zn' ... into a new
sequence W o' WI' W 2 , ... , W n , •••
n = 0, 1, 2, ...
The triangular array is called convergence preserving if it transforms every
convergent sequence zO' ZI' z2' "', zn' ... into a convergent sequence
W o' WI' W 2 , •.. , W n , ••• (Cf. I, Chap. 2.) The array is convergence preserving
if it fulfills the following condition, consisting of two parts:
(1) lim anv = av exists for all fixed v;
n-+=
'n'
n n
L: an. =
.=0
(1n' L: Ianv I =
.=0
the sequence (10' (11' (12' ... , (1n' ••. is convergent and the sequence
Co' CI , C2 , ... , Cn' ... is bounded. [0. Toeplitz: Prace mat.-fiz. Vol. 22,
pp. 113-119 (1911); H. Steinhaus: Prace mat.-fiz. Vol. 22, pp. 121-134
(1911); T. Kojima: T6hoku Math. J. Vol. 12, pp.291-326 (1917);
1. Schur: J. reine angew. Math. Vol. 151, pp. 79-111 (1921).J
44. Prove the easier part of the above mentioned proposition: If
the conditions (1) and (2) are satisfied the array preserves convergence.
[166, I80.J
45. What conditions must the series U o + u I + u 2 + ... + ~tn + ...
satisfy in order that its Cauchy product [I 34, II 23, VIII, Chap. 1,
§ 5J with any convergent series Vo + VI + v 2 + ... + vn + ... results
in a convergent series
+ (UOVI + uIvO) + (UOV2 + UIV I + u 2VO) + .. .
UOVo
Yn'
turns any convergent series a o + a l + az + ... + an + ... into a con-
vergent series
Yoao + Ylal + Y2a2 + ... + Ynan + ...
if and only if the series
converges.
48. The existence of
lim (u l
n-+oo
+ u 2 + ... + un- l + CUn) = (X
such that those terms of the series z1. + Z2 + ... + Zn + ... that are
+
contained in the sector IX - e < arg Z < IX e constitute an absolutely
divergent subseries for anye > O.
53. If lim Zn = 0 and if the positive real axis is a direction of accu-
mulation of the conditionally convergent series Z1 + Z2 + Z3 + ...
n~=
then there exists a subseries zr, + zr. + zr, + ... the real part of which
diverges to + <Xl and the imaginary part converges to a finite number.
54. If the series Z1 + Z2 + Z3 + ... is convergent, but not absolutely
convergent, any value represented by a point of a certain straight line
can be obtained as the sum of the series rearranged in a suitable order.
[Consider two complementary subseries shifted relatively to each other;
52,53, I 133, I 134.J
Chapter 2
55.3 (continued). Let 'P(x, y) and 1p(x, y) denote functions of the two
real variables x and y having continuous derivatives of first order; they
can also be considered as functions of Zt and v where f' (z) =f= O. Verify that
Idwi2
'Px1px + 'Py'1fJy = ('Pu'1fJu + 'Pv'1fJvl i
dz I
under which the interval [a, bJ is seen from the point z. If possible find
an analytic function the real part of which is 0).
58. Show that for any analytic function f(z) = f(x iy) +
(8~2 + 8~:) \f(x + iy) \2 =+ iy) \2.
4\t'(x
59. Show that for any analytic function of z = x + iy
82 82)
(8x2 + 8y2 log (1 + \f(x + zy)
. \2 _
)- (1
41f'(x+ iy) [2
+ [/(x + iy) (2)2 •
§ 2. Some Particular Elementary Mappings
The Cauchy-Riemann differential equations express the fact that an
analytic function brings about a conformal mapping of the z-plane into
the w-plane. (Preservation of the angles including sense.)
The import of the Cauchy-Riemann differential equations for vector-
fields will be discussed later. Cf. § 3.
60. We consider an orthogonal coordinate system ~, 'Yj, , in three
dimensional space. An arbitrary point (~, 'Yj, ') of the sphere
~2 +'Yj2 + ,2 = 1 (unit sphere) is projected from the point (0, 0, 1)
(north-pole of the sphere) into the plane' = 0 (equatorial plane). Let the
projection be (x, y, 0). Express x + iy in terms of ~, 'Yj, , and ~, 'Yj, , in
terms of x and y. (Stereographic projection.)
61 (continued). Let the point P on the plane' = 0 be the stereo-
graphic projection of the point p' on the unit sphere. A rotation through
the angle 1(; of the unit sphere around the ~-axis moves the point p' to the
point P". This point P" is then projected stereographically into the
point P'" of the ~, 'Yj-plane. Let the point P have the coordinates x, y, 0
and the point P'" the coordinates u, v, O. Express u iv in terms of +
x + iy.
We introduce on the unit sphere the geographic coordinates () and qJ
(longitude and latitude) whereby
-n < () < n,
L
J It' (z) II dz I·
An area A in the z = x + iy-plane is mapped onto an area in the w-plane
of size
JJ It' (z) 12 dx dy.
A
The change in direction of the line element under the mapping w = t(z)
is .equal to arg t:(z). It is called the rotation at the point z and it is deter-
mined up to a multiple of 2n at any point where t'(z) =l= o. The branch
for which -n < arg t' (z) < n is usually adopted.
70. The function w = cos z. z = x + iy. yields a one to one con-
formal mapping of the rectangle
O< <
Xl =
< x2 < "2
X =
:It
•
onto a domain bounded by parts of confocal ellipses and hyperbolas
[67J. Compute the area of this domain.
71. Consider the function w = Z2. What is the locus of those points
at which the linear enlargement equals some given constant? Analogous
question for the rotation.
72. Let a be an increasing positive parameter. Determine the region
onto which the function w = If, z = X + iy, maps the variable square
-a < x < a, -a < y < a. Up to what value of a is this region covered
only once? For what values of a is the image covered exactly n times?
73. We examine the image of the closed disk Iz I < r under the func-
tion w = eZ • Suppose r is continuously increasing. There is on the ray
arg w = IX a point that is covered by the image of the disk growing with
r at least as often as any other point of that ray for all values of r. Where
is this point?
The regular function w = t(z) is called schlicht (or univalent) in the
region m if it does not assume in m any value more than once. E.g. the
function t(z) = Z2 is schlicht in the upper half-plane 3z > 0; the function
V:Z is schlicht in the z-plane cut open along the positive real axis; the
function eZ is schlicht in the horizontal strip -n < 3z < n but not in
any wider horizontal strip [72J etc. A schlicht function w = t(z) estab-
lishes a conformal one to one correspondence between the region mand
a region @) of the w-plane. Very often it is useful to consider the point
118 Mappings and Vector Fields
a1 - Vai - 1 a2 - Va~ - 1
the upper half-plane [79]. At which points of the z-plane is the linear
enlargement equal to ~ ? At which points is the rotation ±~?
82. Map the upper half of the unit disk 1 z 1 < 1, the S'z > 0 onto
w-plane cut along the non-negative real axis in such a way that
Pt. III, Chap. 2, No. 74-84 119
§ 3. Vector Fields
We use a special notation (slightly different from the one used in
other parts of the chapter) to discuss vector fields that are defined by
analytic functions of a complex variable. The independent variable is
denoted by
z = x + iy = reif} ,
x, y, r, f} real, r > o. Let
f= j(z) = cp + i'IjJ = cp(x, y) + i'IjJ(x, y)
be an analytic function of z; cp, 'IjJ real. We write
df .
dZ = w= U - tV,
expresses the fact that the rotation vanishes. (It is obvious that the other
two components of the rotation vanish.) Furthermore the vector field
is a solenoidal vector field: the second Cauchy-Riemann differential
equation
au +~= 0
ax 0)'
shows that the divergence vanishes. (The third term usually appearing
in the expression for the divergence is obviously 0.)
85. Show that
(<p(x, y) is the potential of the vector field; the curves <p(x, y) = const.
are the level lines.) Furthermore
w
i.e. the line integral of the normal component of is equal to the change
of the stream function (flux of force). (The normal to the curve L is
pointing to the right as one moves from Zl to Z2:
- i (cos 'T + i sin r) = sin'T - i cos 'T.)
The vector field generated by an analytic function can be interpreted
as an electrostatic, magnetostatic or gravitational field. The vector
Pt. III, Chap. 2, No. 85-92 121
field may also be interpreted as the field of a steady flow of heat or
electricity. In this case the vector w
is the gradient and as such it is
proportional to the intensity of the flow. Finally the vector field can be
also thought of as the field of the irrotational steady flow of an in-
compressible fluid.
90. The steady flow of an incompressible fluid of constant density e
and variable pressure p, subject to no forces, and moving parallel to the
complex plane is described by the equations 1
u ou +v8u +-.!...8P =0 u~+v~+-.!...op =0
ox oy Q ox ' ex oy (} oy ,
Au+ov_O
ox oy - .
If w = u- tV IS an analytic function the components of the vector
w= u + iv and
P = Po - ; (u 2 + v2 ) (Bernoulli's equation)
potential in z before, and "P' be the value after, describing L exactly once
in the positive sense. Evaluate
"P' -"P.
(Flow of force passing from one condenser plate of the Leyden jar to the
other.) Furthermore find
1 ,
4n ('P - V')
other. These three planes are mapped onto the w-plane with preservation
of the angles but with reversed orientation. The w-plane in particular
is the mirror image of the w-plane with respect to the real axis. The
stream lines and the level lines in the stream plane (z-plane) correspond
to lines parallel to the axes in the potential plane (I-plane). The two
undetermined real constants contained in rp and"P correspond to a trans-
lation of the I-plane.
98. Find an irrotational solenoidal vector field outside the unit circle
w
(I z 1 > 1); ought to be = 1 for z = 00 and tangent to the unit circle.
(A fluid passing a circular pillar; at a considerable distance off the pillar
the flow is uniform.) [By reasons of symmetry the two segments of the
real axis inside the vector field have to be stream lines. We may expect
that the horizontal component of weverywhere points to the right. Find
the mapping of the stream plane onto the potential plane. J
99. At which points of the vector field determined in 98 does the
velocity vector vanish? (Stagnation points.) In how many points of the
w
field does assume the same value? Where is the pressure p minimal,
where is it maximal? [90. J What is the resultant pressure excrted on the
124 Mappings and Vector Fields
pillar? Rotate all the vectors of the field through 90°: Give a physical
interpretation of the vector field so obtained.
100. The figure represents the contours of a vector field determined
by the following conditions: The irrotational and solenoidal field covers
the entire upper, and part of the lower, half-plane; it is symmetric with
respect to the imaginary axis. For z = 00 the velocity Wi becomes -i,
for z = 0 we have Wi = O. The following stream lines are known: the
positive imaginary axis, the two pieces of real axis from z = 0 to z = l
and from z = 0 to z = - l (C to A, A to B and A to D in the figure);
: c= 00
l: l
1
1
I
1
0_1_8
A
Chapter 3
and only if on Iz I = r
f(z)
ffiz J(J) > O.
110. The image of the circle Iz I = r under w = j(z) is convex if and
only if the function w = z!, (z) maps Iz I = r onto a curve that is star-
shaped with respect to the origin.
111. The set of points with respect to which a closed curve is star-
shaped forms a convex set. Prove this purely geometric proposition for
analytic curves with the help of 109.
Let Sf denote a finite convex domain in the w = u + iv-plane. For
a fixed angle rp the expression
u cos rp + v sin rp = ffiweiq;
assumes a certain maximum h(rp) in Sf. The function h(rp) is periodic
with period 2n and is called the support junction of Sf. The straight line
is a line oj support; its normal points away from Sf and forms the angle rp
with the positive u-axis. If Sf extends to infinity these definitions are
modified insofar as a finite maximum exists only in a sector with an
opening < n. The two cases: the infinite strip and the half -plane are
exceptions; then there exist only two lines of support, or only one,
respectively.
112. Find the support function of the convex domain that consists
of the point a = la I eia .
113. The function w = j(z) establishes a one to one correspondence
between the disk Iz I < r and the convex domain Sf. Suppose that j(z)
is regular and!, (z) =1= 0 at a certain point z on the boundary, I z I = r. In
this case a definite line of support (tangent) passes through the boundary
point w = j(z) of Sf. Express the corresponding quantities rp any h(rp)
in terms of j(z).
114. The function w = log (1 + z) maps the disk I z I = 1 onto an
infinite domain contained in the strip - ~ < 3w < ~. Its support
I
-71: < 3w < 71:. Its support function is
cos ffJ log (V cos 2ffJ + Vi cos ffJ)2 + 2 sin ffJ arcsin (Vi sin ffJ)
f 0 < <:rr.
h(ffJ) = or = ffJ = 4"'
. f n < < n.
71: sm ffJ or 4" = ffJ = 2 '
h(ffJ + 71:) = h( -ffJ) = h(ffJ)·
for
for
The functions 1, Z, Z2, Z3, •.• form an orthogonal system on the unit
circle.
118. Let fez) denote a regular function on the disk IzI < r. The arith-
metic mean [II 48J of fez) on the circle IzI = r is defined by
1
"2
J f(re') df} =
2" ' f } .
lim
f(r) + f(rw n) + f(rw;) + ... + f(rw:- 1 )
,
n 0 n~oo n
2"i
where wn = en. Show that
f-noJf(rif})
2"
df} = f(O) .
119. Let the function I(z) be regular and different from zero for
Iz I < r. Prove that the geometric mean of I/(z) I on the circle Iz I = r
1 2"
2" J loglf(reiD)ldD n
e 0 = lim VI/(r) j(rw n ) j(rw~) ... j(rw~-l) I = Ij(O) I.
n-+=
[log j(z) is regular for Izi < r.J
120. The function j(z) is regular in the disk I z I < r and does not
vanish for z = 0; the zeros of j(z) in the disk are Zl' Z2' ... , zn' where each
multiple zero is represented with its multiplicity. Then the geometric
mean of Ij(z) I on the circle Iz I = r is
f- J 2"
loglf(reiD)ldD n
e "0 = [j(O) I ~r_ _ .
IZl Z2 '" Zn I
lj(z) = (Z - Zl) (Z - Z2) ... (Z - Zn) I*(z), j*(z) regular and different
from 0 for Izi < r.J
121. Under the hypothesis of 120 the geometric mean of I/(z) I on
the disk Izl < r,
1 r2.n
----.- f J logl/(QeiD)IQdQdD
g(r) = e"r 0 0
w = t{z) = ~
~ a" z" = ... + a-"z-" + a-,,+1z-,,+1 + ... + a-1 Z-1
fI.=-oo
w = t(z) = ~
~
an z" = ... + a-"z-n + a-,,+1z-n+1 + ... + a -1 Z-l
n=-oo
Lis
n=-oo
o
•
r
4 fJ(Ql de =
0
(2 •
rI1
2n
the curve L in the w-plane, a distribution such that arcs which correspond
to each other under the mapping w = p(z) carry the same mass. The
mass distribution defined in this way on L has a certain center of gravity
~ (conformal center of gravity of L). We find
~ = co·
= 2
2 1 12 2
( ~+~+~+ 12 ... ... +_n "2,, + ... )
1a 12__
nr 1 2 3 n+l .
131. Let y(z) denote the angle between the x, y-plane and the tangen-
tial plane of the modular graph at z, If(z) 12. Then
(PP. is the distance between the points P and p.) assumes its maximum
on the boundary.
138. Let f(z) be regular, single-valued and non-vanishing in the
domain 'II. If J(z) is not a constant, iJ(z) i can assume its minimum only
at boundary points of 'II.
139. We assume that the given points PI' P2 ' •.• , P n are all inside a
circle of radius Rand P is moving along this circle. Then
n ==---===-_-::
Vpp I .pp2 ... PPn
(the geometric mean of the n distances PP.) attains a maximum> R
and a minimum < R unless all the P:s coincide with the center of the
circle.
140 (continued). The same statement as in 139 is true for the maxi-
mum of the arithmetic mean
P PI + ~ + ... + PPn
n
Chapter 4
W = en,
2"i
Zv = awv , ,. = Zv_l + Zv
'ov 2'
'JI = 1, 2, ... , n; Zo = zn' a fixed, a =l= 0.
Compute the sum
Zl - Zo + Zz - Zl + Za - Zz + .,. + Zn - zn_l
C1 C2 Ca Cn '
¢1~z'
along the ellipse Z = x + iy, x2 - xy + y2 + X + y = O.
148. Show that
f
2"
xd/} n
x 2 + sin2 /} = 2Vl + x2
when x> o.
o
149. Prove the formula
f
2"
(1+2COSD)nCOsnDd{}= 2n (1-1'- Vl-21'-31'2)n
o
1 - I' - 21' cos f) V1 _ 21' _ 31'2 21'2 ,
1
-1<r<3' n=0,1,2, ...
150. Evaluate the curvilinear integral
,+, (1 - x 2 - y2) Y dx (1 + + +
x 2 y2) X dy
';f 1 + 2x2 - 2y2 + (X2 + y2)2
along an ellipse with the foci (0, -1) and (0, 1).
134 Cauchy's Theorem. The Argument Principle
J x'-le- ix dx = r(s) e 2
o
152. The equation
J= e -xilcoslX . (il'
SIn x SIn IX
)
x n dX = - 1r(n-+-1) . (n + Slll--
1)
--.
.'X
o fl fl II
f. -ds
a+i=
1 p"'s
J(IX) = - .
2m ~ S2
a-icc
along the straight line s = a + it, - <Xl < t < <Xl, parallel to the imag-
inary axis converges absolutely for all real values of IX. The integral
turns out to be
if IX ~ 0,
if IX > 0.
156. We call ft(t) the largest term of the series
t t2 tn
1+-+-+···+-+···
°
1! 2! n!
Assume A > and let z be the only positive root of the equation
A - z - e- z = 0.
Then
[It(t) e-).t dt = : .
o
[
1 + = sin ; ei(n+t-t)u
=
Jl
10
if n< t < n + 1, 1
n _[ - - - z i - --du
if t< n or t > n + 1.
Pt. III, Chap. 4, No. 151-161 135
Pn(X) = -
i
n -1
f (x + V--
1
x2 -it·
d
(Xex_
Vi - ex 2
-
t:
2nl,
f -r
~ -
C
ee-dC
Z
= E(z)
L
defines a function E(z) for points z on the left hand side of L. Show that
E(z) is an entire function which assumes real values for real z.
159 (continued). We find
2~
nt L
f ee cdC = 1.
160 (continued). The function
is bounded.
161. We define
2Z
----,------:c----,--
z(z - 1) (z - 2) ••• (z - n)
= fn (z) .
136 Cauchy's Theorem. The Argument Principle
Show that
~fn{z) dz
lim /z/=2n = i,
n-+oo ~ Ifn{z) II dz I
/z/=2n
where the integrals are computed along the positively oriented circles.
[II 217.]
162. Assume that t(z) is regular in the disk Iz I < r and different from
oon the circle Iz I =r. The largest value of ~z ~;~;~ on Iz I = r is at least
equal to the number of zeros of t(z) in Iz I < r.
163. Let Zl' Z2' ... , zn be arbitrary but distinct complex numbers
and L be a closed continuous curve without double points, enclosing all
the z;s. The function t(z) is supposed to be regular inside of, and on, L.
Then
165. Assume that the entire function F(z) satisfies the inequality
I
iO )
o
Let k > -1. Then
1 ~ when k is an integer,
J :i'1(x) dx
I
1 < 1
o ---- when k is not an integer.
21sin knl'
169. Let IX > -2. The quadratic form of the infinitely many real
variables Xl' x 2 , xa' ...
xx
L; L; ;.,..
00 00
;'~1 ,..~1 ;. + It + IX
is bounded, i.e. there exists a constant M independent of n such that
i i X;.X,..
I;'~1 ,..~1 ). + I' + IX
I<M
whenever the variables xl' X 2 ' ••• , xn satisfy the condition
xi + x~ + ... + x~ = 1; n = 1, 2, 3, ... We may choose M =n if IX is
an integer and M = ISIn
. nIXn I if IX is not an integer.
170. We assume that the functions 11 (z), 12 (z), ... , In (Z) , ... are regular
in the open region mand that they converge uniformly to the function
138 Cauchy's Theorem. The Argument Principle
t(z) in any closed domain inside ffi. Then the limit function t(z) is regular
in ffi.
171. The complex function t(z) = u(x, y) + iv(x, y) of the real
variables x and y is defined and continuous in a region ffi of the z = x + iy
plane. Moreover we assume that the integral
~ t(z) dz
along any circle inside ffi vanishes. Then j(z) is an analytic function of
the complex variable z, regular in the entire region ffi. [Compute the
variation of the area integral
F,(z) = JJ j(z + ~ + i'Yj) d~ d'Yj
~2+t}2~r2
f- ]"j(e
:no
i {}) df} = 1(0) .
174. Suppose that the function I(z) is regular and bounded in the
half-plane ffiz ;::::: O. Then, for x > 0,
By excluding the e-disks and the connecting paths we reduce the disk
Iz I < 1 to a simply connected region me' Compute the integral f log!(Zl dz
along the positively oriented boundary of me']
176. The function t(z) is meromorphic in the disk Iz I < R, regular
°
and different from on the circle Iz I = R and it has inside the circle
the zeros a p a2 , ••• , am and the poles b1 , b2 , ••• , bn , counted with correct
multiplicities. If the point z = rei (), r < R, is neither a zero nor a pole
we have
I R2 - a",z I I R2 - b.z I
RI- .~ log
m n
log It(z) I + "'~1 log I (a", _ z) (b. - z)R
1 2", '9 R2 _ 1'2
= 2n J
o
log It(Re' ) I R2 _ 2Rrcos (@ _ f}) + rzde.
177. The function t(z) is meromorphic in the half-plane mz > 0,
regular and non-zero on its boundary and it has inside the half-plane
the zeros ap a2 , ••• , am and the poles bl , b2 , ••• , bn , counted with correct
multiplicity. If t(z) is regular at infinity (but also under weaker condi-
tions on its behaviour at infinity) and if t(z) is regular and non-zero at
z = x +
iy, x > 0, we have
IZ+ii i Iz+b I
+ ~ log I~ _ a~l-
m n
log It(z) I ~ log I' z - b'
",=1 I '" .=1 •
1 +00 _
=-
n
J log It(i?]) Id arctan~.
-00 x
140 Cauchy's Theorem. The Argument Principle
1: ( ;- -
1 rp)
R2 cos {}" =
1
;rr;R
+i-
J U(R, {}) cos {} d{}
r<rp<R P _"-
2
+ 2~ +X(R),
r
185. Let 0 < a 1 < az < ... < an' Then the trigonometric polynomi.al
ao + a 1 cos f} + a2 cos 2f} + ... + an cos nf}
has 2n distinct zeros in the interval 0 ~ {J < 27T 2n. [22.] [Consequently it has
real zeros only, VI 14.]
186. The function j(z) is meromorphic in the interior of the curve L
and regular on the curve L. If \a \ is larger than the maximum of \j(z) \
on L then j(z) assumes the value a inside L just as often as it has poles
there.
187. The function
assumes any value w with positive real part once and only once in the
half-strip ffiz > 0, - ! < ,3z < !.
188. Suppose that j(z) is regular in the closed disk \z \ <r and
univalent on the ci.rcle \z \ = r, that is it assumes there no value more
than once. Then the image of \z \ = r under w = j(z) has the same orien-
tation as \z \ = r and the function j(z) is univalent (schlicht) also in the
disk \z\ < r.
z ;\:"2
defines an entire function which does not vanish on the boundary of the
annulus
and has exactly one zero inside the annulus, n = 1, 2, ... [Examine
the maximum term on the circle 1z I = I a 12n , I 117.]
201 (continuation of 170). Let e; denote the set of all zeros of all
the functions In(z), n = 1, 2, 3, ... in ffi. If the limit function I(z) does
not vanish identically its zeros in ffi are identical with the limit points
of e: in ffi. (The term "limit point" is used here to mean a point an
arbitrary fixed neighborhood of which contains at least one zero of In (z)
for all sufficiently large n.)
202. The functions
are schlicht in the unit disk 1 z 1 < 1 and converge in any smaller disk
Iz I < r < 1 uniformly to a not everywhere constant limit function I(z).
Then the function I(z) is schlicht in the unit disk Iz I < 1.
203. Let gl (z), g2 (z), ... , gn (z), ... be entire functions which have real
zeros only. If
lim gn(z) = g(z)
n-+=
uniformly in any finite domain, the entire function g(z) can have only
real zeros.
204. Assume
a >0, d> O.
144 Cauchy's Theorem. The Argument Principle
taken in this order, and N for the number of those zeros of the function
Pt. III, Chap. 4, No. 205-206.2 145
Chapter 5
The equation
Z
W = <p(z) ,
f(z)
1 - wq/(z)
213 (continued). Note the cases (3 = 0,1,2, -1, t and derive 209,
210 by taking the limit in 211, 212.
214. Evaluate the sum of the power series
1 + Y, (n + iXt w~.
n~l 'I'll
1 + 2w + :Jw2 + 2w3 + w4
1 + 3w + 6w + 7w + 6w + 3w + w
2 3 4 5 6
n
(1) P (x) = _1_ d (x 2 _ 1t (Legendre's polynomials);
n 2nn! dx n
(3) IX> -1
(generalized Laguerre's polynomials).
(Cf. VI 84, VI 98, VI 99. The generating function of the Legendre poly-
nomials is the series
3x 2 - 1
=1+xw+ 2
w2 + ... ,
148 Sequences of Analytic Functions
(4)
[212, 216J.
221. The four formulas mentioned in 220 hold for any polynomial
F(z). (In this case the series are obviously finite.)
222. The formulas (1), (2) given in 220 are also valid for any rational
function F(z) if the real part of z is larger than the real part of any of the
finite poles of F(z); formula (1) requires the additional condition that
F(z) be regular for z = 0, 1, 2,3, ... -Do the formulas (3) and (4) hold
for rational but not entire functions?
Pt. III, Chap. 5, No. 220-229 149
k=-oo k=-oo
225. Define
F(z) = ao + 2a·1z + 2a2 z2 + ... + 2an zn + ...
and show that
1
V1 + 4t
F( 1 + 21 - V1 +_ 41)
\ 2t
= a
0
+ Ll 2a
-1
t + Ll4a -2 [2 + .,. +
226. Define
F(z) = 2a 1z + 2a2z2 + 2a3z3 + ... + 2anzn + "', a_ n = -an
and show that
~F(1+2t--V1+4t)
t 2t -_ a1 _ a_I +(A2
LI ao
_A2
LI a - 2
)t
227.
IJ (1 +
= _Z(l ____ ~) _
n (n +
1) -
sin nz
nz(1 - z) .
[
d_n-'.(n_---cx)_-_n_-_l_c_os_x]
n > ,
° n = 0, 1, 2, ...
dx x~o
150 Sequences of Analytic Functions
o 0
holds for 0 < r < R provided that it holds for r = O.
235. The function
j(z) = ~ alz +
anzn + ... + + ...
is assumed to be regular and to have positive real part in the open disk
I z [ < 1. Then Ian [ < 1, n = 1, 2, ... In none of these inequalities can we
replace 1 by a smaller number.
236. Suppose that the function
+
j(z) = ao a1 z +
a2 z2 + ... +
anzn + ...
is regular and that lRj(z) < A
on the disk Iz I < R. Then the inequality
n=-=
converges on the annulus 0 < Iz I < 00 (sphere from which two points
have been removed) and that z = 0 and z = 00 are essential isolated
singular points. The maximum of the real part of 1J!(z) on the circle
Iz I = r is denoted by A (r). A (r) increases to 00 faster than any power of
r as r --+ 00 and faster than any power of ~ as r --+ O. Precisely
r
2
The factor - cannot be replaced by a smaller one.
n
Interpret this proposition geometrically.
239. Let the function j(z) = au al z + anzn + ... +
be regular + ...
in the disk Iz I < R and denote by D(I) the largest oscillation of j(z)
for Iz I < R, i.e. D(I) is the least upper bound of Ij(Zl) - j(Z2) I for
IZII < R, IZ21 < R. Then we have
1
lall R < "2 D (I).
The factor t can not be replaced by a smaller one. Interpret this propo-
sition geometrically.
240. Let the function j(z) satisfy the following conditions:
(1) j(z) is regular, Ij(z) I< Iz - s I <r;
M in the disk
(2) j(z) does not vanish in the closed half-disk Iz- s I<r, \R(z- s) > 0;
(3) j(z) has in the disk Iz - s I :::;: i r the zeros cl ' c2' ... , Ct. Then
f'(s) 2 1VI 1
-\R--- < -log ~- -
j(5) = r [j(5) [ A=l
.l'I \R S--~
- C •
A
that converges inside the unit disk. Then the Dirichlet series, assumed
convergent for certain values of s,
D(s) = al 1- s + a 2 2- s + ... + ann- + ... s
+ an (/;"s + e- V;;s) + .. .
converges in the infinite strip
-h<9ls<h
of the s-plane. The convergence is absolute and uniform in any closed
interior strip -h + e < ffis < h - e, e > 0; there the series defines an
analytic function CP(s). The function CP(s) vanishes identically only if
all the coefficients ao' aI' a2 , ••• , an' ... vanish. [Compute
1 a+ioo e- us
F(u) = -2'
7ft.
J
a-loa
CP(s) -S2 ds, 0 < a < h, U> 0; 155.]
converges inside the unit disk and that I(z) and all its derivatives tend
to 0 as z tends to 1 on the real axis, i.e. lim I(n)(z) = 0, n = 0, 1,2, ...
z=x-+l
Then there are two possibilities:
(1) I(z) vanishes identically;
(2) z = 1 is a singular point for I(z).
154 Sequences of Analytic Functions
If the power series converges in a disk larger than the unit disk, i.e. if
. log Ian I
hm sup --- -.. -
n---:;~= n
< U,
we are necessarily dealing with the first casco The weaker condition
log Ia I
lim sup -- =n_ < 0
n--:>-= lin
admits the same conclusion. [Construct the function 1>(5) of 248,J
250. The proposition 249 is not valid if the condition on the coeffi-
. . log Ian I
Clents ao' aI' a2, .. " an' ... of j(z), namely lim sup --/_ < 0, is replaced
n-+= Vn
by
. log Ian I
hm SUp - - -
n--+= nf-l
< 0,
[Put
j(z) = J e-xf.'COSf.'7t sin (xf.' sin .un) e--x(l-z) dx; 153, II 222,J
o
§ 5. Propagation of Convergence
The following examples show that the convergence of sequences of
analytic functions is often "contagious".
251. If the series
g(z) + g' (z) + g" (z) + ... + g(n) (z) + ...
converges at one single point at which g(z) is regular then g(z) is an entire
function and the series converges at every point. The convergence is
uniform in any finite domain of the z-plane.
252. If the sequence
Ig'(z) I, VIg"(z) I,
is bounded at a single point of the z-plane then g(z) is an entire function
and the sequence stays bounded at all the points of the z-plane, It even
has the same limit superior at all the points.
253. Let
Cn'
be two infinite sequences, the second one being arbitrary, the first one
such that an =l= 0, am =l= an when m =l= n, m, n = 0, 1, 2, ... and that
1
~+~+~+
ao a1
1
az
1
... +~+
an
1
...
Pt. III, Chap. 5, No. 250~256 155
Q2n(n - 1) = Cn_I'
°
converges uniformly for all positive values of x; it represents the function
e"'x for < x < 1 and a dilferent analytic function for 1 < x < 00.
261. The sequence of functions 11 (z), 12 (z), ... , In(z), ...
[1nJ 1 +"2"2+
[nJ ... +-;;n
z [nJ z z
n === 1, 2, 3, ... ,
fn(z) =
n
(1'-1 + 2z-1 + ... + n z -II , '
converges uniformly in any finite domain that does not contain the
imaginary axis.
262. Let
IX> 0, f3 > 0, 1X+f3=1,
and put
q;(Z) = IXZ + f3-.
1
z
The sequence of iterated functions
q;(z) , q;[q;(Z)J, q;{q;[q;(z)]},
converges to +1, when ffiz> 0, to -1 when ffiz < °and diverges when
ffiz = 0.
Pt. III, Chap. 5, No. 258-265 157
§ 7. The Order of Growth of Certain Sequences of Polynomials
263. Let h(q;) denote the support function of the infinite convex
domain considered in 114. The sequence 11 (z), Mz), ... , In(z), ...
I ()= V:;-(z - 1) (.l' - 2) •• , (z - n + 1) (3 - n) -'''('1')
n = 1, 2, 3, ... ,
n z n! e,
where z = reUp, is bounded in the entire half-plane ffiz > 0 [12, II 220J ;
h(q;) is the smallest function of the angle q; that keeps the sequence
bounded.
264. Let h(q;) denote the support function of the convex domain
considered in 115. The sequence 11 (z), Mz), ... , In(z), .. ,
Chapter 6
the moduli of the values are related that the function assumes in
different parts of its domain of existence.
Let the function I(z) be regular in the circle 1 z 1 < R; the maximum
of its absolute value on the circle 1 z 1 = r, r < R, i3 denoted by M(r).
266. The maximum of I/(z) Ion the disk Jzl :s;: r is M(r).
267. The maximum M(r) is monotone increasing with runless I(z)
is a constant.
268. We assume that the function I(z) is regular in the simply connected
region Izl > R. M(r) denotes the maximum of I/(z) 1 on the circle
1z 1 = r, r > R. Then M(r) is also the maximum of I/(z) 1 for 1 z 1 ~ rand
M(r) is monotone decreasing unless I(z) is a con::;tant.
269. Let I(z) denote a polynomial of degree n; then
v=o
C) w-vJ III a suitable domain,
lim j(z) = 0
Izl-+=
m
278. We denote by M a positive constant and by a connected open
region. The analytic function f(z) is assumed to have the following
properties:
(1) j(z) is regular at every point of m;
(2) j(z) is single-valued in m;
(3) each boundary point of m
has for every positive number B a
neighbourhood such that at every point z of m in this neighbourhood
the inequality
Ij(z) 1< M + B
is satisfied.
160 The Maximum Principle
hold uniformly in a sector ex ~ {} ::; {J, ex < (J. Then j(z) vanishes iden-
tically.
§ 2. Schwarz's Lemma
280. The function t(z) is assumed to be regular and Ij(z) I < 1 in
the disk Iz I < 1. If j(O) = 0 either the stricter inequality Ij(z) I < Iz I
holds for z =1= 0 or t(z) = eiaz, ex = real.
281. We denote by z = cp(C) and w = V'(C) two schlicht maps of
the unit disk ICI < 1 into the regions at and e; of the z- and w-plane
resp.; the images of the origin C= 0 are the points z = zo and w = Wo
resp. In addition let 0 < (! < 1 and rand 5 be the images of the disk
ICI < (! under the above mentioned maps. We assume that w = t(z) is
a regular analytic function in at the range of which belongs to e; and
that j(zo) = Woo Then t(z) assumes in the subdomain r of at values that
belong to the subdomain 5 of e;. These values are in the interior of 5
unless j(z) is a schlicht function that maps the region at onto the region e;.
282. Let j(z) be regular and Ij(z) I < 1 for Iz I < 1. Then
1 -- [flO) [2 I
Ij(z) - j(O) I < Iz 11=l/(0) I~ , 0 < z I < 1.
M(r) < M(O) + R2~ r [A(R) - A(O)] < ~ ~ >il1(0) + R2~ r A(R).
285. We assume that I(z) is regular, non-zero and bounded for
Izl < R. Then
R-r 2r
M(r) < M(O)R+r M(R)R+r ,
1(0) ~ ~ :~t < I/(z) I </(O)! ~ 1:\ ' 0< Izl < 1.
There is equality only if
1 + ei!Xz
I(z) = Wo
1 - e z
i,,' Wo, (X real, Wo > O.
288. Let I(z) be a regular function with Im/(z) I < 1 for Iz I < 1. If
1(0) = 0 the stronger inequality
289. Suppose that the function I(z) is regular for Iz I < R and that
L1 denotes the oscillation of its real part in the disk, i.e
Iffi/(Zl) - ffi/(Z2) I < L1
for Izll < R, Iz21 < R. Then the oscillation of its real part in the smaller
disk Izi < r, r < R, is
4.1 r
Iffi/(Zl) -- ffi/(Z2) I ;;:: - n arctan -R ' IZl! ;;:: r, IZ21 ;;:: r.
real axis and I(a) = b, b real. Then f' (a) is real and positive and we have
the inequality
~ ~_1_ > ~ __ ~1~_~ for ~z > o.
b - j(z) = (a - z) f(a)
294. Let the function I(z) be regular, have the zeros Z1' Z2' ... , zn
and let I(z) be bounded, I/(z) 1 :::::::: M, in the unit disk 1 z 1 < 1. Then the
stronger inequality
I
/(z) I <
-
i z -
11 -
:1 . 1 -
Z1 Z
z - _Z2 •••
Z2Z
_~- :'n 1 M
1- znz I
holds for all z with ffiz> O. We have equality either at every point or
at no point of the right half-plane ffiz > O.
296. A function that is meromorphic in a closed disk and of constant
absolute value on the boundary circle is a rational function; in fact it is,
up to a constant factor, the product of linear fractional functions that
map the disk in question either onto the interior or the exterior of the
unit circle.
297. We assume that the function I(z) is regular and bounded in
the disk IZI < 1 and vanishes at the points Zl' Z2' Z3' ... Then
(the sum of the distances of the zeros from the unit circle) is finite or
else I(z) O.
298. We assume that the function I(z) is regular and bounded in
the half-plane ffiz> 0 and vanishes at the points Zl' Z2' Z3' ... outside
the unit disk in the half-plane, i.e. IZn I> 1, ffiz n > 0, n = 1, 2, 3, ...
Then the sum of the series
(PP. is the distance between P and P.) of the point P assumes its
maximum in any domain on the boundary. (Generalization of 137.)
302. We assume that the functions Il(z), fz(z), ... ,In(z) are regular
and single-valued in the domain'll. Let PI' P2' ... , Pn denote positive
numbers. The function
denotes the geometric mean of It(Z) 1 on the circle 1z 1 = 1', l' < R. The
function @(1') is monotone increasing with l' and a convex function of
log l' (in the wide sense).
308. The function t(z) is supposed to be regular for 1z I < R and not
a constant. We put
2,.
1(1') 1
=-2
no
J It(1'ii}) 1dO., 1'<R.
The function 1(1') is monotone increasing with l' and log 1(1') is a convex
function of log 1'. [299,304.]
309. Let t(z) be regular and not a constant in the disk 1z 1 < R.
The function w = t(z) maps the circle 1z 1 = 1', l' < R, in the z-plane
onto a curve in the w-plane with length l(1'). The ratio 12(1') is monotone
nl'
increasing with 1'.
310. We suppose that t(z) is regular and not constant for 1z 1 < Rand
that P is a positive number. We define
1'<R.
The function 1p(1') is monotone increasing with l' and log 1p(1') is a convex
function oflog 1'. (Cf. 306, 308, 307, 267 and 304, for an analogous case see
IV 19.)
§ 4. Harmonic Functions
311. An analytic function that is regular in the closed disk ~ cannot
assume real values at all the boundary points of ~ except if if is a real
constant.
166 The Maximum Principle
on the two bounding rays {} = -lX and {} = lX because 0 < AlX < ~ .
On the circular arc Iz I = r, -lX < {} < lX, we have
Ie'A I = e,A COSA{} ?: e,AcosA<X •
in the sector 0 < {} < ~, e > 0; take the branch of z-" that maps the
positive real axis onto itself. We see that in the above mentioned sector
It(Z) e-1}Ze-e. - ih
4 ..
I A ( An)
I = It(Z) e-1}ZI e-er cos A8- 4
(1) there exist two constants A and B, A > 0, B> 0, such that
on the entire half-plane
I/(z) 1< Ae B1z1 ;
(2) there exist two constants C and y, C > 0, y > °such that for
r2:0
1/(±ir) I < Ce- yr .
A function satisfying these conditions must vanish identically. [Examine
the function I(z) e-PZ10g(Z+l).J
°
328. The function sin ;rr;z is the smallest function that is analytic for
)Hz > and that vanishes at the points z = 0, 1, 2, 3, ... More precisely,
the following proposition holds:
We assume that the function I(z) is analytic in the half-plane )Hz > 0
and that it satisfies the conditions:
(1) there exist two constants A, B, A > 0, B > 0, such that for
)Hz> °
I/(z) I< Ae B1z1 ;
(2) there exist two constants C and y, C > 0, y > °such that for
r>O
1/(±ir) I < Ce(Jr-y)r;
(3) I(z) has the zeros 0, 1, 2, ... , n, ...
Such a function vanishes identically.
329. Let w(x) be a positive function of the positive variable x that
increases with x and tends to + 00 as x increases to + 00. A function
I(z), regular in the half-plane )Hz > 0, that satisfies the inequality
I/(z) I 2: ew(lzl)lzl for )Hz > °
does not exist.
330. Suppose that the function I(z) is regular at any finite point of
the sector IX ::;;: f}:S f3, bounded by 1, I/(z) I < 1, on the two rays f} = IX
and f} = f3 and that, furthermore, there exists a positive constant 0
such that
an ro such that
"
fJ-x
If(reil}) I < eer for r > ro
in the above mentioned sector then the stronger inequality
If(z) I ~ 1
holds in the entire sector. [Method of 325.J
332. The fundion g(z) is assumed to be an entire function, M(r) be
the maximum of Ig(z) I on the circle Iz I = r. If
lim log_~i1 =
(---+= Vr
°
then g(z) cannot be bounded along any ray. [E.g. g(z) is not bounded
along the nega ti ve real axis. J
333. Suppose that the function f(z) is not a constant and that it is
regular in the half-strip @ defined by the inequalities
x> 0,
If there exist two constants A and a, A > 0, °< a < 1, such that in @
x:2::: 0, z= x + iy
must satisfy the inequality
If(x + iy) I < ew(x)e X
holds everywhere in lR. (Only the case M' > M is interesting.) This
modification of the hypothesis does not change the conclusion of 278
that under those conditions It(z) I <M, It(z) 1< M respectively. [In
the case where the point at infinity belongs to lR and all the boundary
points of lR lie in the disk Iz I < r we examine the comparison function
n (z -
n
(2rt zvl- 1 .]
336. The domain '1) is supposed to lie in the half-plane Sz > 0; the
boundary contains a finite number of segments of the real axis; Q denotes
the sum of the angles under which the segments are seen from an inner
point C of '1). Assume that t(z) is regular and single-valued in the interior
of '1) and continuous on the boundary of '1), that It(z) I < A at the real
boundary points, It(z) I < a on the remaining boundary of '1), 0 < a < A.
Then
Q Q
- 1--
It(C)I<A"a " [57.J
Part One
*1. [Cf. HSI, pp. 238, 252-253, ex. 20.] AlOO = 292 [2].
*2. [For an intuitive solution see G. P6Iya:Amer. Math. Monthly Vol. 63,
pp. 689-697 (1956). Cf. MD, Vol. 1, p. 97, ex. 3.84.]
~ AnCn = (1
(1 + C5 + CIO + Cl5 + ... + C5Y + ...)
(1 + ClO + C20 + C30 + ... + ClO• + ... )
(1 + C25 + C50 + C75 + ... + C25u + ... )
(1 + C50 + CIOO + Cl50 + ... + C50V + ...)
1
(1- C)-l (1- C5)-1 (1- ClO)-1 (1- C25 )-1 (1- C50)-I.
()9
8. 2: Dn(n = ((-I + 1 + ()2 ((-2 + 1 + (2) ((-5 + 1 + (5)
-99
(t,-10+ 1 + (10)2 (t,-20 + I + (0) (C 50 -I- 1 + (,0).
n~
.-.-------~-------- A (n.
(1 ~ ,a'l (1 ~ ,a,) .. , (1 ~ tl) n~O n
The "first weighing problem" [5J (all the weights on one pan) :
therefore:
-(P)-
Cn - P!
n - n! (n - P) ! •
~1_
(1 - C)p
= 1 +(-P)
1
(-Cl + ... + (-P)(-ct + "',
n
thus
A = P(P+1) ... (p+n-1)=(p+n-1)
n 1.2 ... n P-1'
=, -, 3
-1' - 1 -3'
, - 1
-~
- 1
1 ,a _ . . ,
9
-an'3 + - 1
n
,ran _ 1
1
=, -_._-
-N ~.
"3 n +1
, -
-
1
1
3n + 1 _ 1
N=-~-.
2
16. an = aFn , where Fn is the number of digits 1 in the binary repre-
sentation of n (its expansion in powers of 2).
17. [E. Catalan, Problem: Nouv. Corresp. Math. Vol. 6, p.143
(1880). Solved by E. Cesaro: Nouv. Corresp. Math. Vol. G, p. 276 (1880).J
The series in question results from the expansion into a power series of
Particular cases:
En = en when PI = P2 = '" = PI = 1,
En = An when PI = P2 = ... = PI = 00,
Ea+n = Dn when PI = P2 = ... = 2 and a l + a + ... + a
2 l = a.
To encompass 18 we admit l = 00 (properly interpreted). We have
PI = P2 = ... = 9, a l = 1, a2 = 10, aa = 100, ... in the case 18.
18.2. Particular case of 18.1; l = 3, al = a2 = a3 = 1, PI = P2 = P3 = n,
E 2n +1 = en: 3) _ 3 (n ~ 2) = (n ~ 1).
*19. [Euler, l.c. 9.J First solution: According to solution 14 we
have
(1 + C) (1 + C2 ) (1 + ~ ) (1 + C8 ) ••• = 1 ~ C'
(1 + C3 ) (1 + C6 ) (1 + C12 ) (1 + C24 ) ••• = 1 ~ C3 '
(1 + C5 ) (1 + ClO) (1 + C20) (1 + ~O) ••• = 1 ~ CS • etc.
Second solution:
K(C) = 1)
n=1
Third solution:
n~l
n= (1 - 52n ) = n= __1 _
n= (1 + en) = n~l (1 - en) n~l (1 _ e 2n - 1 ) .
or in
23. [E. Catalan: Nouv. AnnIs Math. Ser. 3, Vol. 1, p.528 (1882).
Solved by E. Cesaro: Nouv. AnnIs Math. Ser. 3, Vol. 2, p. 380 (1883).J
e
The number of solutions is equal to the coefficient of n - 1 in the expansion
Pt. I, Solutions 20-25 179
of
=
L: - - v =
c .(1) C+ .(2) C+ ... + .(v) cv + ... ,
v=l 1 - C
where .(v) denotes the number of divisors of v [VIn 74].
24. [E. Cesaro, Problem: Mathesis (Gand) Vol. 2, p.208 (1882).]
The number of solutions is equal to the absolute term (C-free) of the
infinite sum
+ ...
?"vll-(2J1+1)n
L: L:
00 00
(v+1)' =
~ v' (C-(2 V+1)n +C-(2V+1)(n-1) +C-(2 V+1)(n-2)
v=l(l-C)(l-C ) .'=1
~~~2-'-:-' -;;;(1- ~- C) I •
Since aI' a 2, ... , al do not have a common iactor the denominator of the
other terms is of degree I - I at most. The statement follows from this
[solution III242J. Cf. 28. Notice also that the I-dimensional volume
characterized by the inequalities
Xl > 0, x 2 > 0, X3 ~ 0, ... , Xl ~ 0, alx l + a2x 2 + a3x 3 + ... + alxl :s;: n
in I-dimensional space is
n
and
Aab-a-b=O,
An=A n- [n]
- ab
ab
+ [nbJ·
a
Yet
n - [!'ab_J ab < n - (!'ab.- - 1) ab = ab
and so, by (1),
A n- [~]ab = 0 or 1, which proves 26.
:182 Infinite Series and Infinite Sequences
expansion of
i.e. equal to
30. [G. P61ya: Math. Ann. Vol. 74, p.204 (1913).J The number of
lattice points is equal to the sum of the coefficients of C- s, (-s+ \ ... ,
1, ... , (s-l, C in the expansion of
ave
k
The following relation between the series 2: and its coefficients
av holds in general: v~ - k
and
2m+l 2m+l . 2m + 1
m v ( 2 _( 2
sm ---- -- t
2: ( = -- - - --- - 2
"~-m c- t - (t . t
sm 2
31. [Cf. Ch. Hermite, Problem: Nouv. AnnIs Math. Ser. 2, Vol. 7,
p. 336 (1868). Solved by V. Schlegel: Nouv. AnnIs Math. Ser. 2, Vol. 8,
p. 91 (1869).J Since z = n - X - Y we have X + y < n, x> 0, y> 0;
also x < n - x, y - x ::;: n - X - Y ~ x + y.
Consequently the number of solutions in question is equal to the
number of solutions of the ~nequalities
n n
2-x~y::;:2' y> 0,
Pt. I, Solutions 28-38 183
31.1. (1) You are one of n persons forming an assembly which elects
a committee consisting of r of its members. There are (:) possible
committees [10], namely (: ~ ;) to which you belong and (n ~ 1) to
which you do not belong.
(2) (1 + xt = (1 + xt- 1 (1 + x). Use the binomial theorem and
then consider the coefficients of x' on both sides.
(3) Straightforward verification if you use the usual expression of the
binomial coefficients.
31.2. Pass from r - 1 to r by mathematical induction; use 31.1.
32. Compare the coefficients of zn in the identity
(1 + zt (1 + zt = (1 + z)2n.
33. Compare the coefficients of z2n in the identity
34. Evident.
34.1. From
there follows
therefore
Apply 34.
36. Cf. 35.
37. Differentiate the identity
and put x = l.
38. [Problem from Ed. Times. Cf. Mathesis (Gand) Ser. 2, Vol. 1,
p. 104 (1891). Solved by Greenstreet et al.: Mathesis (Gand) Ser. 2,
184 Infinite Series and. Infinite Sequences
J Jl1-=- :n dx = r(1 + x + x
1 1 1
1 - (~ - xt dx = 2 + ... + xn - 1 ) dx
o 0 0
1 1 1
= I +2+3+ ... +-;;-.
39. The general term of the left hand side sum is the coefficient of
x 2n +l in HI + r-
2xt+k+l (-x 2 k • Therefore we have to consider the
2n
coefficient of x +l in
n (1 + 2 )n+l ( 2)n+l
+
2:! (1 2xt+k+l (_x 2t- k = ! (1 2xt+l x +
- 2 -x - .
k~O (1 + x)
or in the power series of !(I + 2x)2n+2 (1 + x) -2. Division leads to a
polynomial of degree 2n, the remainder is HI - (2n 2) (2x + 2)]. +
The coefficient of x 2n +l in
1 1 2n + 2
-------
2 (1 + X)2 1+x
is equal to -!(2n + 2) + 2n + 2 = n + 1.
40. Decompose the sum into three terms according to
(v - nIX)2 =n2IX2 - (2nIX - 1) v + v(v - 1).
The formula
and its first and second derivatives with respect to p furnish the three
terms if p is replaced by x and q by 1 - x. Cf. II 144.
41. It is sufficient to consider n > p and
cp x
( ) = xIx - 1) ••• (x -
P!
p + 1) = (X)
P , 1p(x) = -21P (XP).
\
We have now
i (:) G)
'~P
= P! (:!~ p)! i .~p
(: =::) G) 2n-
= p = 2n1p(n) ,
45. [G. Darboux, Problem: Nouv. AnnIs Math. Ser. 2, Vol. 7, p. 138
(1868).J We deduce from 44
= f(k) d
1; - - I Zk = I (z -) eZ = eZg(z) ,
k=O k. dz
Therefore the coefficients of In{z) = a~n)z + akn )z2 + ... + a~!)zn are
linked by the relations
y = 1, 2, ... , n + 1; o -- a(n)
a(n) n+l -- 0 .
This together with 11 {z) = z concludes the proof. The value of In(l) is
determined by
= 1 + al + a2 (1 + al ) + a3 (1 + al ) (1 + a2 )
+ a4 (1 + al ) (1 + a2 ) (1 + a3) + ...
= (1 + a l ) (1 + a + a (1 + all (1 + a
2) 3 2)
+ a4 (1 + al ) (1 + a2 ) (1 + a3 ) + ...
= (1 + a l ) (1 + il 2 ) (1 + a3 )
+ a4 (1 + al ) (1 + a2 ) (1 + a3 ) + "', etc.
58.
D o =G(O) =
•
-
1 - q
q
+--q2
1 - q2
+---+
qa
1 - q3
... +---qn
1 __ qn
+ ....'
applying 50 and 57 we find
=
G(z) - G(qz) = 2: Anzn, G(qz) - G(q2 Z) = 2: Anqnzn,
n~l n~l
=
G(q2z) - G(lz) = 2: A nq 2nzn, ... ,
n~l
hence addition of the first m equations and taking the limit m --+ 00
yield
= A
G(z) - G(O) = 2: ---"--n zn.
n~ll - q
Introducing (1 - g) n = y, 1 - Ln = g we get
i; ~ (1 -
k~l k
eY)k = -y, [181.J
60. If we regard as known the sum of the series (ct, 59) we have
1 1 +z
f(z) = 2z- 10g 1=-;
and the solution is straightforward. Without supposing such knowledge
we equate the coefficients of z2n on both sides and so obtain
hO
n
L: ( W- k
-
2
2k+1
2k
(nn +- kk) -_ 2n 1+ 1.
2n for n, g2 for g,
then
zg-2 n + 1 for x
Pt. I, Solutions 60-61 189
and multiply finally by
3 2n-l
qn'z-n = _~ . ~ ... _q_ _
Z Z Z
1J + L: (_1)k [n -= 1J qn-k
n-l n
= L: (_1)k[n -
k=O k k=l k 1
= L:
n-l
(_1)k (1 _ qn-k-l) [n - 1J
k=O k
= L:
n-2
(_l)k (1 _ qn-l) [n - 2J Q.E.D.
k=O k
Therefore
1'(z) 1.. 2
- . - - <; - -
f(z) z 1 .- z
leads to
j(z) 1 j(z) 1 __; ~n I
log - - ~ log - - - . _. ~ ---- - .::... n"
2 (l-z)2' 2 (1 - 2)2 n~1
where we put a = aI' a2 , ••. ,al and multiply [61]. The second part
follows from
(1 - za , _ za, _ •.• _ zam-I) (1 _ z"m)
-----------,--- - , - - -- - -----
~ zn exp ( -1 Isn+ll
+ -2z21 + ... + -nzn1 + ---'------,-- + ... )
Z (n + 1) zn+l
1 tn - S1 < (N + 1) 2M 4(N : 1) M
Po = 1, PI = ( ~ y, P2 = ( ~ y,
then we have
(n + l)n+1
POP1P2 ... Pn = (n + 1)!
.
11m (n + 1)'" - n'"
=
1.1m (1 + lin)" - 1'"
=
(d-xex) =iXi
n--+oo na- 1 n--+= lin dx x=l '
we obtain
. 1",-1 + 2",-1 + ... + (n + 1)",-1 . (n + 1),,-1 1
l i m - - - = hm --- ----- -- =- .
n-+= (n + 1)" n-+= (n + 1)" - n'" IX
P = --_ Pn - v < _ Pn - v _
nv Po + PI + P2 + ... + Pn = Po + PI + P2 + ... + Pn - v
and
P = Pn - . Qv < Pn - v _ _ __
nv Po Qn +h Qn-1 + ... + Pn Qo = Po + h + ... + Pn - v
We obtain [d. 74J
74. We write
SOP n + slPn-1 + ... + snP O soqn + sl qn-1 + ... + snqO
~n = Po + h + ... + Pn ' qn = qo + q1 + ... + qn
00 00 0:) 00
= =
2: Pn~nzn = 2: Pk Zk 2: s/, 2: Qnqn zn = 2: qk zk 2: s/,
n~O k~O I~O n~O k~O I~O
2: Rntn zn = 2: Pki 2: q/ 2: sm zm
n~O k~O I~O m~O
75. Put
tn = (a 1 + a2 + .. , + an) n- rJ ,
converges to 0 [66].
76. [E. Cesaro: Nouv. AnnIs Math. Ser. 3, Vol. 9, pp.353-367
(1890).J According to 70 the limit is equal to
Pn n p-l p- 1
·
11m l'
=lm Pn n .
n-+= log Pn - log Pn- 1 n-+= -log (1 - PnP;;l)
n n
77. [1. Schur.J Set 2: Pv = P n, 2: qv = Qn and assume (3 > O.
v~l v~l
qn rs; f£.,
n
q > 0; i.e. [76J Qn rs; q log n -+ 00: contradiction. Also
lim P n = lim npn = 00. If IX> 0 the same argument as for Qn can be
n--+oo n-+=
194 Infinite Series and Infinite Sequences
1.e.
Apply 70:
an ---i'- 0 and
a1 + a2 + '" + an ~ nan :::; K for n = 1, 2, 3, ...
For a given m find n such that an < tam' From
K > a1 + a2 + ... + am ~ man + (am+! + ... + an)
~ (n - m) an ;;::0. m(a m - an) ~ tma m
follows that a 1 + a 2 + ... + am <::::: K + ma m ;::: 3K for m = 1, 2, 3, ...
We are dealing here with a transformation under special conditions; 66
in itself is no help.
79. Contains 65 as a particular case; the proof is the same. If Pkl = 0
for l> k the matrix is called triangular (d. 65, 66) or more specifically
lower triangular. If Pkl = 0 for 1< k the matrix is termed upper tri-
angular.
80. Contains 66 as a special case. Proof analogous.
81. With sn = nCn + (n + 1) cn _, 1 + ... we can write
tn = ~ S
n n
+ (-~
n + 1
-~)
n
S
n+1
+ (_~_
n +2 n + 1 n+_
_1 . ) S + .... ?
1:~ (1 - ex)
n-k
y
k
~
=
b1y
I (1 - lX)n+! =
=----- ~ al(ex + y) I
k~O I~O 1 - (IX + y) I~O
=
= (1 - exr+ 1 ~ Sl(ex + y)l.
The coefficient of yn on the left hand side is tn and on the right hand side
We have
b tV
CPv(t) < v9 •
bo + bIt + b2t- + ... + bntn
The right hand side converges to a value smaller than s as t -+ 1. The
proposition holds also if the radius of convergence is e instead of 1, e > O.
196 Infinite Series and Infinite Sequences
J: t n
n~O
. aO + a l + ... + an
-7 hm -- --- = s.
n-+= 1
87. [G. Frobenius: ]. reme angew. Math. Vol. 89, pp.262-264
(1880).J It follows from the hypothesis that n-l an is bounded, therefore
= =
J: (a o + a 1 + ... + an) t n J: (so + 51 + ... + sn) tn
n~O n~O
. So + 51 + ... + 5 n
-)- hm -= s.
n-+= n + 1
88. Multiplying numerator and denominator with the geometrical
series we obtain
=
J: (a o + a 1 + ... + an) t n
n~O
=
2: (bo + b1 + ... + bn ) t n
n~O
. a o + a 1 + ... + an
-7 hm - - - - - - - - -
n-+= bo + b1 + ... + bn
= s.
n = 0, 1, 2, ... ;
An
n";J 2n-1
1 ( t) V-;;- _ e-V-;;-
.
lim -
n-+oo
An
En
=
.
n-+oo
.
n En _1_
F'"2
hm - - - - = lim --~
1--+1-0 G' (~)
= Va _e-c= __
eVa + e-
-c=
Va •
n! 2n-1 2
The power series for G' ( ~) is divergent for t = 1 because all its coeffi-
cients are non-negative and lim G'
1--+1-0
(-2t ) = 00.
b
o
+ b1 + b2 + ... + bn = ((1 +n n) = (a + 1) (a +n!2) ... (a + n) <Xl bna
'
b> 0.
[Solution 89. ] According to 75 we obtain
198 Infinite Series and Infinite Sequences
·
11m (1, - t)'JjO)
.-
= '
'----,
2.,
(I ~; --,
nJ - 2
[']'
"-
11 I
)
tH = •
lIm
[1'11 ] - 2
.-' -" - _- -. - 2 ~ -
[I; I'] < O.
/-,.]-0 n-,,=l \ 2 n 'o-oo.~ ~) I 2Ji ---r 1
'2'-4"-6"'--27',
Th e l'Iml't'IS -
:\ '4
(t'z ,- 1)
'2" 5 i
. -;;
2
V;
. . . 2nt 1" }
' as
211 ,""':. -;'
rII 202J. I'll
94. The statement 85 is true not only for t -+ 1 but also for t - -3>- =.
Cf. 84. The sum of the series
bo -+ bit + bi'- + ... + bnt" + '"
increases to infinity as t -+ = because b n > 0,
1· ·
95. ApplCatJon of 94: an =-"n. Sn b
n =--,"
12.
1
iBorel':.; summation,
Knopp, p. 471.J
96. We write sn = ao + a l + ... + an' s 1 = O. Then using partial
integration for the subtrahend we get [95J
J :- 'g(x) dx
o
= ~ ~--=--=~
....
"~o
l'!
JI e-"x" dx = ....-5'
0 "~O
S,
,
JI(X
0 '
v!
V
_
(1'
X'
+
~ 1) I, e
1 ) -,1 dx
The series
Applying 66 with
So = 0,
Iw - - I< I- - -
am
m -
a 2m
2m
am \
m
+ \ a4m
-4m - a-2m I+ ... <-+-+
2m
1
2m
1
4m
... = m-1 .
Another proof: From
a",+n + :. < (am + 1) + (an + 1),
we conclude by 98 that
. an +1 1- a
lim _ _n=_w,
lim--=w,
n-+oo n ft.-+oo n
where -w, being a lower bound, cannot be + 00, and so w is not - 00.
100. [L. Fejer: C. R. Acad. Sci. (Paris) Ser. A-B, Vol. 142, pp. 501-
503 (1906).J The proof will show that it is sufficient to discuss the case
of bounded partial sums SI' S2' s3' ... , Sn' ... Put lim inf sn = m and
n->=
· sup
11m Sn . a posItIve
= M , 1 IS . . .mteger, 1 > 2 an d -II/I---
- m = -"
u. Divl·de
n~= I
the number line into 1 intervals by the points
m + 0, m + 20, ... , M - 20, M - 0,
°
-00, 00.
Choose N such that [sn - sn+l[ < for n > N. Let, furthermore, sn , '
n l > N, be in the first (infinite) interval, sn,' n 2 > n l , in the last (infinite)
interval. In each of the 1 - 2 intervals of length there will be at least
one point sn,+k (0 < k < n 2 - n l ). A similar argument holds if the
°
sequence does not "slowly increase" but is "slowly decreasing".
101. [Cf. G. Szeg6, Problem: Arch. Math. Phys. Ser. 3, Vol. 23, p. 361
(1914). Solved by P. Veress: Arch. Math. Phys. Ser. 3, Vol. 25, p.88
(1917).J The interval (0, 1). Cf. 102.
102. [G. P6lya: Rend. Circ. Mat. Palermo Vol. 34, pp. 108-109
(1912).J There are subsequences with arbitrarily high subscripts
t n" t n, +!, ... , tn" that descend arbitrarily slowly from the lim sup to
the lim inf of the sequence. The detailed proof follows the lines of solu-
tion 100.
103.
Vn Vn+ 1
+ Vn + 1 + Vn+l
-------
n n
104. Let s1' S2' S3' ... , sn' ... , lim sn = S, be the sequence in question.
n-+=
Choose S
~
anywhere in the interval S - 21 , S +~,
2
and more generally
S
Vn
in the interval S -~,
2n
S + 2n
~; VI < V2 < V3 < .... The terms of the
series S + (s - S ) + (s - S ) + ... are not larger than the terms
of [SVl [ v~ (~ ~ : )v~ (: : !)'~ ....
105. Only finitely many terms of the sequence are below a certain
fixed number. Among finitely many numbers there is a smallest one.
106. If the Weierstrass least upper and greatest lower bound of the
sequence coincide there is nothing to prove. If they are different then at
least one of them is different from the limit of the sequence. This bound
is equal to the largest or smallest term of the sequence.
Pt. I, Solutions 100-110 201
107. The smallest among the numbers ll' 12, la • ...• lm (m given) is
called 'YJ. 'YJ> o. According to the hypothesis there are terms of the
sequence that are smaller than 'YJ. Let n be the smallest index for which
In < 'YJ. Then we have
n>m; In<ll.ln<l2 •..•• ln<ln_l.
108. Apply 105 to the sequence l;;l.l;;ll' l;;~2' ...
109. [G. P6lya: Math. Ann. Vol. 88. pp. 170-171 (1923).] We call
lm an "outstanding term" of the sequence if lm is larger than all the
following terms. According to the hypothesis and to 108 there are in-
finitely many outstanding terms:
In,. In•• In,, ...• In, > In. > In. > ....
If l. is not oustanding it lies between two consecutive outstanding terms
(for v > n1 ), i.e. nr _ 1 < v<
nr • We find successively Inr -1 < In r •
Inr -2 < Inr •...• l. < Inr • consequently
(*)
From this we conclude
lim sup In sn
1-+00 ' 1
= + 00.
Otherwise In r Snr and consequently. according to (*). the sequence
would be bounded, contrary to the hypothesis. Apply now
lIS!> l2s2' •••
107 to the sequence
l ns-1 Sn.
-1
' •••
taining these rays and determine the line of support 1 of slope A. We call
(n, Ln) the corner (or one of the corners) through which this line of
support passes. The lines connecting (n, Ln) with the points to the left
(in the convex domain) have a slope smaller than A, and with the points
to the right (in the convex domain) a slope larger than A.
111. [Cf. G. P6lya, Problem: Arch. Math. Phys. Ser. 3, Vol. 24,
p. 282 (1916}.J Write
m = 1, 2, 3, ... , Lo = 0 [110J.
Since Ll - A < 0 the difference Lo - 0 cannot be the minimum mentioned
in the solution of 110. Then In+! 2: A, therefore In+!' and consequently
n, increases to infinity simultaneously with A.
112. Put II + l2 + ... + lm = Lw m = 1, 2, 3, .,., Lo = o. We find
L - mA
lim -"'-~ = -A [67].
1»4-00 m
The sequence
Lo - 0, Lm - mA, ...
tends to - (Xl. Let the maximum be Ln - nA. The inequalities in
question are satisfied for this subscript n. There are in the sequence
L o' L l , ... , L m , ••• infinitely many terms larger than all preceding terms
[107J. Let Ls be one of them.
lS_l
-------
+ ls
2 s
are all positive. If A is smaller than their minimum the subscript n
belonging to A is > s. - The points (n, Ln) are to be enclosed in an
infinite polygon convex from above.
113. Set lim sup ]log m = S. Then we have a} S 2: JL This is obvious
m--+oo og rm
for S = (Xl. If S is finite we find for 8 > 0 and for large enough m
log m < (S + 8) log rm. Therefore
$+26
"" m
£.J r
-$ -26 converges,
m~l
1 By line of support of a closed set \Ill we mean a line that contains at least one
point of ~J1 but such that one of the open half-planes determined by this line does
not contain any point of m. The intersection of the closed half-planes defined by all
possible lines of support containing 9J1 is a convex domain Sf, the smallest that
contains m (the convex hull of m). Every line of support of m
is a line of support
of Sf and vice versa. These concepts can be easily adapted to the case where the
ideal point (point at infinity) belongs to the set m, as is the case here. Cf. III,
Chap. 3, §1.
Pt. I, Solutions 111-119 203
if
204 Infinite Series and Infinite Sequences
120. If for a certain x a term Pnxn is larger than all preceding terms,
i.e. if for a certain value of x all the inequalities
hold, then this remains true also for any larger value of x.
121. Let m be arbitrary and choose x so that Pmxm is the maximum
term. Then
On the other hand Pm(ee)m is bounded for m -+ <Xl, 0< e < 1. Conse-
quently
122. [l.c.110.] Suppose that for a certain positive value z the central
= a.
subscript of the series 2: bm zm is n [121] and that y is a value for which
m~O m
a
--f zm.
=
be the successive values of the central subscript of the series 2:
m=O m
Assume that the term with the specific subscript k is the maximum
term in the interval (Ck - 1 , Ck ) and that Yk is the value for which bkyk
becomes the maximum term of the series 2: bmym. The method used
m~O
for the solution 122 associates these values Yl> Y2' ... with values of x
that belong to the intervals
124. [A.]. Kempner: Amer. Math. Monthly Vol. 21, pp. 48-50.J
All the non-negative integers between 0 = 00 ... 000 and 10m - 1 =
99 ... 999 that are written with the nine digits 0, 1, 2, ... , 8 are obtained
by lining up these nine digits in all possible ways m at a time. Thus we
get a total of 9m numbers. Let rn be the n-th non-negative integer that
is written without the figure 9. If lom-I - 1 < rn < 10m - 1 then
n :::::: 9m • Therefore
· log n < log 9
1lmsup--_--< 1
n-+= log r n - log 10
[113J.
9 - 1 9 9 9 9 (9 9
+10 1+ + 100 +"') =
2 - 3 - 2 2
+~ + ... =
\
-1- 8 10 80.
125. Consider the two sub series which contain all positive and all
negative terms, respectively.
126. [K. Knopp: ]. reine angew. Math. Vol. 142, pp.292-293
(1913).J No. Example: Let bl + b2 + ba + ... be convergent and
Ib11 + Ibzl + Ib3 1+ ... be divergent. Put
The inequality p(tm) < tP(m) < p(tm + 1) determines the integer tm
completely. Collecting the terms that belong to the same bm we trans-
t - t
form the series a<p(l) + a<p(2) + ... into the series m~l lP{mT -1P7~~ 1) bm·
al < L: lanl,
n=Pm+1
Pn + Pn+l + ... + Pn+v = Pn,v, lim Pn,v=Pn, n = 1, 2,3, ... ,v= 0,1,2, ...
V~=
Pt. I, Solutions 129-132.2 207
Assume that Pn, is the first term for which Pn, < a. Either there exists
a VI such that Pn"v, < a, P n"v,+l ~ a, VI ~ 0 or P n, < a. In the second
case we have P n, = (f because P n, > Pn,-1 2" a (for n 1 = 1 this means
PI = S > a), i.e. a may be represented as an infinite subseries. In the
first case we determine the first term Pn, with n z > n 1 + VI' Pn"v, + Pn, < a.
Either there exists a 1'2 with Pn"v, + Pn"v, < a, Pn"" + P n"v,+1 > a,
1'2 ;::;; 0 or Pn"v, + P n, < a. In the second case we have Pn"v, + P n, = a,
because Pn"v, + P n, 2" Pn"v, + Pn,-1 ~ a (n2 > n 1 + 1\ + 1, because
Pn"v, + Pn,+v,+1 = P n"v,+1 > a) i.e. a may again be written as an
infinite subseries. If this procedure never terminates (if the first case
occurs at every step) then a = P n "" + Pn"v, + Pn"v, + ....
132. From the relations
Pn = Pn+l + Pn+2 + Pn+3 + ...
Pn+l = Pn+2 + Pn+3 + .,.
1 1 1
we gather Pn = 2Pn+l' thus PI =2' P2 =4"' ... , Pn = 2 n ' ... The
representation by infinite binary fractions is unique.
132.1. [G. P61ya, Problem: Amer. Math. Monthly Vol. 51, pp. 533-
534 (1944). Solution Amer. Math. Monthly Vol. 53, pp. 279-282 (1946).J
Define
Then
(1 + ~) (1 + ~) ... (1 + 2~) = R" ,
( 1 - ~)
3
(1 - ~) '" (1 _ _1_ )
5 2n + 1
= ~
Rn'
1 1
Rn N n 2 2n -2 by II 202
:m (p-)2
p_
mq
N
q
proof can be derived from the expression of Euler's constant given III
the solution of II 18.
133. Insertion of appropriate vanishing terms into the two comple-
mentary subseries reduces the proposition to the termwise addition of
two convergent series.
*134. We assume that all the terms of the divergent subseries
+
aT, aT, aT, + + ...
are non-negative. Then the complementary sub-
series as, + + + ...
as, as, will be such that to each positive 8 there
corresponds an integer N so that
aSm +aSm+l +···+aSn <8
provided that n > m > N. After this remark the
proof essentially
coincides with the well known usual proof for Riemann's theorem on
the rearrangement of the terms of conditionally convergent series.
[Knopp, pp. 318; d. also W. Threlfall: Math. Z. Vol. 24, pp. 212-214
(1926).J
135. From PI > P2 > Pa > ... , 0 < ml < m2 < ma < ... follows
that
.=1 .=1
Pt. I, Solutions 133-140 209
i 8vppl =
I v=m+l
I i
[(Ev - Em) - (E._l - Em)] ppl
p=m+l
x 2n + 1 0
'"x
= (-It+1
x 2n + 1
f [/(2n+1)(t) - 1(2n+l) (t + ~)
x
+ 1(2n+1) (t + 211:)
x
o
- ...Jsin xt dt.
j (2n+1) (0)
Obviously Rn has the same sign as (-1r+1 x 2n +2 • Apply 144.
148. The sum of the first 2n -'- 2 terms is
~ + (_l)n
3 3. 2n - 2 '
and that t is real and positive. Then tz-l lies on S). Therefore
fz e f
~=~ 1 oo~
152. e 2 2 dt = - e 2z e-tdt.
c
--i ~ °we can apply 151 (141 resp.).
Z
Since ffi Z -
[151,153J.
155. [Cf. Cauchy, l.c. 154.J Its Maclaurin series envelops arc tan z
for ffi Z2 ~ 0, Z =+= 0. This has been proved in 143 for real z; we use the
same formula for the complex z in question.
156. It is sufficient to consider <p(x) = a o + a1 . We get
x
Hence the series converges for Iao I < 1 and diverges for Iao I > 1.
Assume ao = 1 and (for the sake of simplicity) p(n) > 0, n = 1, 2, 3, ...
Then we get [157J
logp(l) p(2) ... p(n) = a1 logn + b + en, i.e. p(l) p(2}··· p(n) = eH8n na"
lim en = 0, b is an n-free constant. Consequently we have convergence
n-+~
m(x)
T
-
= 2 - C = 1 - - - - -2
x 2! x
(X (X2 1
+ ...
convergent for eX > 1, and eX = log 2; log 4, log 8, ... are> 1.
160. J 1 1
ex10g-X dx = ~ "1
~ 1
Jxn
1 ( 1 )n
log- dx.
o n=O n. 0 x
Substitute xn +1 = e- Y •
tn_1 < tn' n = 2, 3, 4, ... For positive x we have x 2 < 1 + x if and only
if x < t(l + V5), i.e. if x is smaller than the positive root of the equation
x 2 - x - 1 = O. Hence t! = 1 + tn_ 1 < 1 + tn' tn_1 < tn < t(l + V5),
n = 2, 3, 4, ... and lim tn = t exists, 0 < t < t(l + V5), t2 = 1 + t,
i.e. t = H1 + V5). W-:~proceed similarly in the case of the continued
fraction where the recursion formula is
n = 2, 3, 4, ...
214 Infinite Series and Infinite Sequences
162. [G. P6lya, Problem: Arch. Math. Phys. Ser. 3, Vol. 24, p.84
(1916). Solved by G. Szeg6: Arch. Math. Phys. Ser. 3, Vol. 25, pp. 88-
89 (1917).J If (for the sake of simplicity for Y > 1) log log a. < Y log 2,
a. < e2', then tn < -Ve2 + Ve 4 + ... + Ve 2n < e1+/"5 [161J. If, however,
an> e
~
,fJ > 2, then tn > e
fin . - If an < 1, then, of course, - -n-
~~~
163. We prove
where
164. [Jacobi, l.c. 53, § 52, Corollarium: Werke, Vol. 1, pp. 200-201.J
Write 1- q = ao' 1 + qm = am> m= 1,2,4,8,16, ... , then the n + 1-th
partial product is
n·
(a 1 a2 a4 aS ••• a2 n)2
fPn (x )
_ xn
- n! ' 1jJ
() _ (X) _ :-~- 1) (x - 2) ••• (x - n
x - n - n!
+ 1) ' n-
_ 1 2 3
, , , ...
167.
log a = 1 -!-p
n
--
n
1 + 0(1)
+ ! + _._-
12n
-n3 '
2
thus
log a
n+1
- log a -
n - (n +!!)- (np + i) + 0 (~)
n3 '
r
P< 0 this is true already for n > 1 as can easily be verified by expanding
(1 + ~ with help of the binomial formula.
169. We wnte an =
. (1
1 + ~-
1+z- ; the hrst
)nH -(--i-)t . factor decreases
1 +"
[168J; the square of the second factor is 1 +- - + n(n
2x - 1
n + 1
---.x
+ 1)
2
The
condition x ~ ! is therefore sufficient. Now expand
log an = 2n ( 2n
1
+1 + 3(2n 1+ 1)3 + 5(2n 1+ 1)5 + ...)
+ 2 [2n ~x + ~ (2n'~Xr + ~ (2n:xY + ... J
_ 2n
- 2n + 1
2x
+ 2n + x + 12n +
1 2
0 (1)
n
3 '
216 Infinite Series and Infinite Sequences
·
Slllce Iog an - Iog. a n + 1 = 4x4n- ---
2
2+ 0 ( -;;31) t he cand'ltIon
. x = 21.IS a Iso
>
necessary.
170. [Cf. problem No. 1098, Nouv. AnnIs Math. Ser. 2, Vol. 11, p. 480
(1872). Solved by C. Moreau; Nouv. AnnIs Math. Ser. 2, Vol. 13, p. 61
(1874).J The first inequality means
[
f'(X) =-~- _log~>~x_ -~
x+2
1+ 2
x x+2
1+ 2
x
+1= 0, 1(0) = 0.]
The second inequality is equivalent to
[169J.
171. [1. Schur.J The number e lies in the second quarter of the interval
because
n = 1, 2, 3, ...
1 1 ( 1 +-1)(1 +-,
+-< 1)-1
4n n 2n
r V~: r r
From the binomial expansion we derive that
thussin"x> V
n+IX ,n> N. Consequentlyforalle < V'S, liminfV-;sin"x~e
C
n~
i.e. :c: Vi If e > V'S choose m so large that sinm x < V- c -. In a similar
N+1
wayasmt . h C f'Irst I desm
caseweconcu . c.
m+l VN+2 smm+2 <-~-,
C
x<~="
etc.
VN+3
174. The sequence vn is decreasing, Vn > 0, therefore lim v" = v
n-+oo
exists; v = I(v) implies v = O. Consequently it is sufficient to prove the
proposition for small x. Let b' be fixed, b' > b. For sufficiently small x
we have
x - axk < I(x) < x - a~ + b' Xl
and when n is sufficiently large, n > N(c),
1
en -k=! - a en
(l)k
-k-l > e(n + 1)-k=! 1
or
en -k-l
1 _
( l)k + b' (1)1
a en - k - l en -k=! < e(n + 1)
1 -k-l
depending on whether
-1 -]
°
imply that the sequence Un is steadily decreasing in the first case, Un > 0,
and increasing in the second case, un < 0. We have lim un = It =
because n--+oo
eU - 1
U ~ log-'-- for U ~ 0.
u
. f ormu Ia eUn - 1 -- uneUn +1 , 1 1
Th e recurSlOn - -1
" 2 .3, ... . lds
Yle
178. [1. Schur, Problem: Arch. Math. Phys. Ser. 3, Vol. 27, p. 162
°
(1918). Cf. O. Szasz: Sber. Berlin Math. Ges. Vol. 21, pp. 25-29 (1922).J
If B > is so small that 1 q 1 + B < r then there exists a constant A
independent of 11 and v such that
bn_.1 <
IIb.:'! A (I q 1+ B,
). v = 0, 1, ... , 11; 11 = 0, 1, 2, ...
For 11 > m we obtain
Cn
b - f(q) =
m
L: a.
(bb- -
n-v v
q
) + L:n a. bn-v
b- -
n-,
~ a.q .
v
i.e. arbitrarily small with m- 1 • Choose m so large that these two terms
are smaller than B. For fixed m choose 11 so that the first term becomes
absolutely smaller than B.
179. [Special case of an important proposition in function theory by
Vitali. Cf. E. Lindelof: Bull. Soc. Math. France Vol. 41, p. 171 (1913).J
We show only that lim anI = 0. (Then form x-I fn(x) - anI' etc.) Assume
B > ° and x so small that
n-+=
°< A _x_ ..
1 - x
< B. Then we have
Pt. I, Solutions 177 -181.1 219
For fixed x choose n so large that Ifn(x) I < eX.
180. We have lak I <A k , k = 0, 1, 2, ... , therefore ~ a k is conver-
gent. Moreover k~O
Assume 10 > 0. Choose m large enough to render the last term smaller
than e. Having fixed m we select n so large that lank - a k I < ~1 '
m+
k = 0,1, ... , m. Then we get
ISn - sl < 210.
3;' n"-l(n + kyx-1 [(n + k)" - n"J -2 = 3;' k12 ffJ (:)
k~l-n k~l-n
for n ---+ (Xl; the term with sUbscript k '- 0 must be omitted. The function
ffJ(x) is defined by
m(x) = (1 + X),,-l (__ _X___ )2;
'f' (1 + x)" - 1
ffJ(x) is continuous for -1 < x < (Xl if we set ffJ(O) = IX- 2 ; ffJ(x) = 1 if
IX = 1; otherwiseffJ(x) '"'"' xl-a for X---+(Xl, ffJ(x) '"'"' (1 + X)'-l for x ---+ -l.
If IX = 1 the value of the limit follows immediately from
Hence the remaining part of the series (from -in to 00) has£'M Ik 1- 1 - "
as a majorant. [180.J
183. We have
V +V +V + .~ + V < V + V +V + ...
2 2 2 2 2 2 2 = 2,
therefore
an = Co ~ c1 V2 + c2 V2 + ... '+ V2 Cn
4sin2 (: i
v~O
COC1:"cV)_2=_2COS(~
2
ibOt'l.:.Fv)
v~o 2
c1 c2 ... c
= -2 cos ( ---:-
n
2
+-n2 v~l
1: ----
n
2"
v)
1 - 88 ···8
o 1 n
2 00 2
:=~--n--;
n~O 2
therefore
n = 0, 1, 2, ... ,
2({J gl gq-2 1 0 0
n = go + "2 + ... + 2q-2 + 2q~T + 2q + 2H1 + ...
gl gq-2 0 1 1
= go +"2 + ... + 2q-2 + 2q - 1 + 2q + 2q+1 + ....
In this case co' c1' ... , cq _2 are, as before, uniquely determined, cq = -1,
cq +! = cq +2 = ... = 1, cq_1 may be chosen -lor +1. We have there-
fore
x = 2 cos qJ = Co V+ V+
2 c1 2 c2 -V 2 + ... + c V2 . q _2
According to 183 any number written in the above way has to be of the
type 2 cos P.... n, p, q integers, 0 < p < '2:'. If q = 1 we have to modify
2q
slightly: the numbers go' ... ' gq-2' co' ... , cq _2 do not exist, qJ = ; ,
x= o.
185. The sequence gn is periodic beyond a certain term if and only
if this is true of the sequence Cn.
Pt. I. Solution 185-185.2 223
185.1. [For precursory heuristic considerations see MD, Vol. 2,
pp. 49-50 and 171.] We seek a solution of the system of 2 equations
u+2v+ 3w=0
u + 8v + 27w = 0
with integral values for the three unknowns. We find
u = 5, v = -4, w = 1.
We note that
u + 32v + 243w = 120.
We define, for m = 1, 2, 3, ... ,
= _2m- 1/ 5 ,
= 3m- 1/ 5
and
Then
1> 5.
185.2. [G. P6lya, Problem: Amer. Math. Monthly Vol. 51, p.593
(1944). Solved by N. ]. Fine: Amer. Math. Monthly Vol. 53, pp. 283-
284 (1946).] (1) The case where D consists of just one odd number can be
*
settled by an easily visible extension of solution 185.1. (Why is
u + 32v + 243w O? Give a reason avoiding numerical computation.)
(2) If D consists of a finite number, say h, of different odd numbers,
construct by (1) the corresponding sequences
each of which yields a divergent series just for one required exponent
224 Infinite Series and Infinite Sequences
where the positive numbers PI' P2' PS' ... , decrease sufficiently rapidly.
*186. List concretely the different possibilities for small n, use 187
for all n ~ 3.
n\k 2 3 4 5 6 7 8
1
2 1
3 3 1
4 7 6
5 15 25 10
6 31 90 65 15
7 1 63 301 350 140 21
8 1 127 966 1701 1050 266 28
and this proves our assertion for k if it was assumed for k - 1. The case
k = 1 is easy.
Pt. I, Solutions 186-194 225
*189. The right-hand side in 188, decomposed into partial fractions,
equals
1 1 (-1l- 1 1 (_i)k
k! z - k - 1! (k -lj! z - k + 1 + ... + (k -1) ! 1! z - 1 + ~.
Expand in powers of Z-l and consider the coefficient of z-n-1. For a
combinatorial proof see VIII 22.1.
*190. From 189, since
ee Z -1=e1 ( 1+-h+~!
, .2z
+~3!
3=
+".
)
n\k 2 3 5 6 7 8
1
2 1 1
3 2 3
4 6 11 6
5 24 50 35 10
6 120 274 225 8.~ 15
7 720 1764 1624 n.~ 175 21
8 5040 13068 1:H32 6769 1960 ;~22 2H
*200.
(1 _ t) -x = i; x(x + 1) (x + 2) ,••• (x + n - 1) tn
n=O n.
= e-x1og(1-t)
= xk 1 )k
= k~ kT (log 1 - t .
We have first used the binomial expansion, then 199. Consider the coef-
ficient of Xk.
*201. List concretely the different possibilities for small n, use 202
for all n ~ 4.
n\k 2 3 4
0
2 0
3 0
4 3 0
5 10 0
6 25 15 0
7 56 105 0
8 119 490 105
snn-a = n 1) sa+l
(a + 1
+ (a +n 2) sa+2
2
+ ... + (n2a) S2a.
a
o 2 3 4 5 6 7 8
1 2 5 15 52 203 877 4140
o 4 11 41 162 715
[34J.
We can give an analogous proof for (2) by using the well known fact
that the number of partitions of a set containing 11, elements into kl
subsets each containing 1 element, k2 subsets each containg 2 elements,
Pt. I, Solutions 204-210 229
where, of course,
1kl + 2k2 + 3k3 + ... = n.
To rework 186-210 starting from 210 is left to the reader as a
research project. Just one hint: If we let I denote one of the expressions
(1), (2) and (3), then
oj oj oj
oz - wi = I, w ow' z-
oz
which involves
respectively.
For additional material on the subject treated in the section just
concluded see V 62.1, VII 54.2, VIII 22.1, 22.2, 22.3, 58.3, 247.1, AI 191.1,
191.2.
Part Two
Integration
1.
2.
3.
n
2: hea +(v-1 l\
v~l
b-a
where h = xv - xv-I = ----
n
a 1 - e nh
lim he - - = hm h - - - = e - e ,
• . i - ea b a
n .... =· 1 _ eh n .... = eh - 1
because
lim
h ....O
e h ~ = (dedx
h
-
X
)
X~O
= 1.
4.
n v-I n v-I
'" aq (q - 1) '" aq (q - 1)
k.J • ' .:;..; v-I'
.~1 aq v~l aq
li -. We have [3J
n -
x Ib
with q =-~ =
x._1 a
lim n
n .... =
(1/r a
b _ 1) = lim e~ - 1
n .... = log b - log a
(log b -log a) = log b -log a.
n
Pt. II, Solutions 9-14 233
The intergal of the linear term over the interval
[a + (1/ - 1) b: a, a+ 'JIb: a] vanishes. Cf. 10.
12. We have [11]
b-a
J
a+ 2.. +1
+~
0=1 b-a
J [/(x) - I (a + 2'J1 :n~ a1) ] dx
a+(2.-1)2 .. +1
b-a
,,+ 2.. +1
- J (x - a) f'(~o) dx
" b-a
.. a+(!o+1)2,,+1
+I
~=1
! J b-a
(x - a - 2'J1:n~a1Y I"(~~) dx.
a+(2~-1)2"+1'
= - (n
n 1
"-1(-1- - -1- - -1)
2-+-+ .. ,+- +nI n
2
- -1 n)
n - 1 ~=1 sin 'lin 'lin (n - 'II) n n
n n n
=2n
n
[logn +C +O·(~)J
L n
+n [l(-.
0
1
510 nx
_ ~_
nx
1
n(1 - x)
) dX+O(~)]
n
2n
= - (log n
n
+ C) - 2n
-n log -2
n
+ 0(1),
with the help of 9.
234 Integration
15. [E. Cesaro, Problem: Nouv. AnnIs Math. Ser. 3, Vol. 17, p. 112
(1888). Solved hy G. P6lya: NOllV. AnnIs Math. Ser. 4, Vol. 11, pp. 37:3--
381 (1!111).J Put f(x) = x log x, a = 0, b = 1 in 10. Although the hypo-
thesis of 10 is not completely satisfied the conclusion of 10 remains
valid.
I
1
Xn<$«n+~)fJ·
nP'((n+t)fJ)--'>-- J
' 1
dx
((3-X)2·
o
1
yields
t/(I) + f(2) + f(3) + ... + f(n - 1) + tf(n) - F(n)
= -1J[f'('YJ1) - /'($1) + /'('YJ2) - /'($2) + ... + /'('YJn-l) - /'($n I)J·
The series
Pt. II, Solutions 1- 8 231
*5. Set 1 +2
1
+,f1 + .. , +n1 =Hn' Then
1 n 1 1 1 1
L
n
=-n~l
Y'-----=~+--+ ...
n+l n+2
+~
n+n
y
v~ 1 +-
n
= H 2n - Hn = H 2n - 2(tHn)
1 1 1 1 1
=1 +2+"3+4"+"'+2n-l +:2;
2 2 2
2 4 2n
= 1--21 +---
1
3
1
4
1
+ ... +----.
2n - 1
1
2n
Moreover
1 1 1
Un = n + ~ + 1 + ... + 2n _ 1
1 1 1
=Ln +---2
n n =Ln +-2
n .
6.
sin ----'!.-- sin 2 _:rr: sin n _:rr:) "
. :rr: n+1
hm - - ( ------
n-->= n +1 ~_
n+l
+-----
2 __:rr:__
n+l
+ ... +----
n_:rr:_
f ·-dx>O.
=o sin x
x
n+1 n+l n+l
(VI 25.)
7. The mean value theorem implies
n
F(b) - F(a) = ~ [F(xJ - F(xv_1)J
n
need not be true. [ef. V. Volterra: Giorn. Mat. Battaglini Vol. 19,
p. 335 (1881).J
8. Let k -
n
1 <
-
~ < !!.....
n
Then
n
232 Integration
and
-n
9. [Cf. G. P6lya: Arch. Math. Phys. Ser. 3, Vol. 26, p. 198 (1917).J
The least upper bound of the expression
Ij(X1) - j(xo) I + Ij(x 2 ) - j(x 1 ) I + ... + Ij(xn) - j(xn _ 1 ) I
for all possible subdivisions of the interval [a" bJ is called the total
variation of the function j(x) on [a, bJ (same notation as on p.46).
Functions of finite total variation are also called functions of "bounded
variation" .
10.
b-a
a+v-
I
n
n
-L1n=~ (a+v?:_a-x)j'($v)dX,
.~l b-a
a+(v-l)n
b-a b-a
where a + (v - 1) - - < $
n'
< a + v---
n
, thus
-21 (b--n-a)2 n
~
k.:
< -L1 n==2
mv== _ a)2
< -1 (b--
n
n
~
£..J
M v'.
~=l v=l
My and m. denote the least upper and the greatest lower bound of j' (x)
in the v-th subinterval. We obtain
b - a
lim n L1n = - 2- [f(a) - j(b)].
n--+=
11.
because
j(x) - j ( a + (2v - -~ = ( x - a - (2v - 1) b-a)
1) b-a) ~
X j '( a + (2v - 1) -
b - -a)
2n
+ -21 ( x - b - -a)2 j"(
a - (2v - 1) -2n
n.
v
Pt. II, Solutions 15-19.2 235
is convergent because the terms have alternating signs and their absolute
values converge monotonically to O. In the case of I' (x) < 0 we find
-2n1- 1
-.
8n"
<1 11
-1 + -2 + -3 + .. , + -n1 - log n - C < -2n1 .
For j(x) = -log x we find
log n! = (n + t) log n - n +1- S + IOn.
where s is a constant and 0 < IOn < 8~' Stirling's formula [205J states
that 1 - s = log V2n.
19. Cf. solution 18. The sum mentioned there.
implies
log 2 = lim (H2n - Hn).
n~~
n-+~
1 1 1 1 1
- - - 2 -"3 -"4 -"5 - .. , - n2
1 1 1 3 1 2n - 1
=T-2-"3+"4-"5- ... +--;;;-2-'
236 Integration
which coincides with the partial sum of the proposed series that has
1 + + + ... +
2 3 (2n - 1) terms. It still must be shown that the
limit of this subsequence of partial sums is the sum of the series. (Easy.)
20. [Cf. e.g. 1.c. 9. J Let f(x) be monotone increasing. [Otherwise
consider -f(x).J Then
1~ 1n (n - 1)
1( -;;1 ) +1( -;;2 ) +... +1-1'1- 1
f
o
f(x) dx <
n
.----- <
1
f f(x) dx.
n
examine the two terms separately J. Let E > 0, 'Y} be chosen such that
1-rl
f(x) dx < E. Then we have
1 [(l~1»)nJ 1-'1
lim -
n-+= n
L; rp(~) f(~) =
v=l n n
f
0
rp(x) f(x) dx.
If, on the other hand, M denotes the least upper bound of Irp(x) I we can
write
I~ • nil
~[(1-'1)nJ+1
rpC) f (:) I < ~ ni
.~[(1~'1)nl+1
f (:) < M / f(x) dx
1-1)
~ ME .
22. In 20 set f(x) = X"-l.
23.
an = l"-l(n - 1)/i~1 + 2"-1(n - 2)/i- 1 + ... + (n - 1),,-1 1/i-1
f
n-1 1
= n"+fi~l L; ~ (~)a-1 (1 _ ~)/i-1 C'V n"+fi- 1 X,,-l (1 - X)/i-1 dx .
• ~1 n n n 0
Cf.20.
24.
1 1
f(x) =x - -1 - .
- x
25. Let f(x) be monotone decreasing and finite for x = 1. The in-
equality [20J
and a fortiori
2
2n-l
J
M
I(x) dx +
2n-1
1
M
7 f(x) dx < e
±f v2: 1) < 20
~ .=1 r
1
f(x) dx
2n-1
+
M
7
1
f(x) dx.
Similarly
[~] 1
lim ~ L; f(2V -
n-+oo n 71=1 n
1) = J f(x) dx.
0
28.
n-l [~] n-l
~ L; (-1)"-1 I(~) =~ L; f(2V - 1) -~ L; f(~).
n .=1 n n >=1 n n .=1 n
sign. Assume that I(x) is positive and monotone decreasing. Then we have
~+~ ~
J I(x} dx < h(/(h) + f(2h) + ... + I(mh)) < J f(x) dx,
h 0
and so for m ~ 00
00 00 00
The condition that I(x) be monotone is essential only for large x and in
the neighborhood of x = o.
238 Integration
00 -nh
h 2.; enh = log 1 -h.
n=1 1 - e
~
r=
e-x
---dx=
1 + e- x
f -1-=log2.
+
1
dy
y
o 0
o
f x
xe- _ x dx
1- e
= J~ x ( i
0 n=1
e- nx ) dx = i
n=1
:2.
We can also argue in the following manner: we have [VIII 49, VIII 65J
= n =
1: nO< _ t-n = 1: O'o«t) tn,
,,=1 ~ - t n=1
1 = =
1 _ t 1: O'o«n) r = 1: (0'0«1) + 0'0«2) + ... + O"o«n)) r.
n=l n=l
37. The proposition follows from 30: t(x) = log (1 + x-"), h = t '"
Transformation of variables and partial integration lead to
r
1
--
j'log (1 + x-"') dx = •
~
~
+U
du = _n_ .
. n
o 0 sm~
38. Apply 30 with t(x) = log (1 - 2X--2 cos 2q! + .r4). Write
n . sin t
t = II eUP, use the formula for -t- and square the absolute value on
both sides:
II (1 -
~ h2 [ T
2", Slntp
. 2", . ]
2 cos 2rp
~ + ?;4J;4) =
1
4,t2 e + e
- "Slntp
- 2 cos (h
2n
cos q;) .
39.
a
J log X dx < L; (aqn -
~
~ k-l ~
= 2knGrn-~(~r·
For more details see e.g. Jordan: Cours d'Analyse, Vol. 2, 3rd Ed. Paris:
Gauthier-Villars 1913, pp. 218-221.
41. [For problems 41-47 d. G.P6lya: Arch. Math. Phys. Ser. 3,
Vol. 26, pp. 196-201 (1917).]
42.
-n
holds; i.e. the operations of taking the limit and computing the mean
value can be interchanged [44].
43. [Cf. Cesaro, Problem: Nouv. AnnIs lVlath. Ser. 3, Vol. 2, p. 239
(1883).]
n 1
lim 1
n-+oo n
L; (1
S'=l
- [-'V~J :) = 1- J' [~ ] x dx <5
I
= . J' ([-.x1] - [1-
bm
n-+oo • ~'r
- IX J) dx = bm
. > (1- - ---
n-1
n-;..co.-..; V
1)
+V (X
1 .~1
f'l - -, dx.
1
= 1 -- ~
l+cx
+ ~2 - -2+cx
~ + ... =
• ~
XX
o
If = ! we obtain 41.
IX
45. [G. P61ya, l.c. 41, pp. 199-200.] We assume at first that IX> 1.
Then
1
f [~J f (IX + I)
1 n
(IX + I) x
x'" dx = i;_ n x' dx = 1 + _1 _ +_1_
2,,+1 3,,+1
+ ...
o n -1 I
n-+ 1
= '(IX + I).
Pt. II, Solutions 43-50 241
1
1 n
i
0=1
(0. - E.) = i
.=1
(-1)"-1 GJ
n
= ~ (_1),-1 ([: ] - :) +n ~
.=1
n
.=1
(-ll v-I
The first sum on the left hand side divided by n converges to 0 as n --* <Xl
[28].
48. Consequence of the definition of the definite integral.
1
49. Special case of 20 for I(x) = log x. With regard to f log x dx
d~ 0
11 ~ 0~ c + 2 ... c + n - 1
rn n n n
[29J.
c n-1
-+--
n 2n
242 Integration
2n
51. An = --1' Moreover
n+
= II (0_~~)n+l~2v
v~l + n 1 '
n
because L: (n + 1 - 2v) = O. Making appropriate use of 20 we obtain
implies
Iln/2n ... Inn = (rn - 1)2.
If r = 1, Inn must be omitted (it vanishes); notice 20.
53. [G. Szeg6, Problem: Arch. Math. Phys. Ser. 3, Vol. 25, p.196
(1917). Solved by J. Mahrenholz: Arch. Math. Phys. Ser. 3, Vol. 28,
pp. 79-80 (1920).J According to the hypothesis e~ix(ei" - r) is real, i.e.
the arguments of eix and ei" - r are equal or differ by n. Since ~ is the
number closest to x with this property eix and ei" - r have the same
argument, which means that eio is the point of intersection of the ray
from r parallel to the vector eix with the unit circle. If, therefore,
o <x < n and ~' is the argument that belongs so to x + n as ~ to x,
then ei ", r, and eiO' are on the same line. Thus we obtain by elementary
geometry
[eW - r [2 [ei " - r [2 = (1 - 2r cos ~' +~) (1 - 2r cos ~ + r2) = (1 _ ~)2 ,
hence
1
2"-
no
f"
[log (1 - 2r cos ~ + r2) + log (1 - 2r cos~' + r 2)J dx
1 "
= -2 flog (1 - r 2)2 dx = log (1 - r2) •
no
Pt. II, Solutions 51- 57 243
The sum on the right hand side converges to an integral. Cf. 67. Other
subdivisions of the interval [a, bJ may be considered instead of the sub-
division by points in arithmetic progression, and we can choose any
point in a given subinterval as the point where we evaluate the function.
There is an obvious analogy between the integral as a limit of sums and
the limit of products considered.
55. According to 54:
1
V 1 J xdx
+ -;; -;;
.
hm
n-+oo
n
1-
n
=1
1
1 -
v 1 =-1--
- - _ J xdx
eO
=e.
nne 0
56. The product in question is
1 . 3 . 5 ... (2n - 1) IX n . 2n
f(;- + l)--ex - 1J [(n +
2) ex -'lf~~-~ (2nex =-1)
_ __0_j-~~. _jn_+2)~__ ... (n + n) IX
- (n + 1) IX - 1 (n + 2) IX - 1 (n + n) IX - 1
1
1
-+ ----1---- [54J.
1
"
f dx
l+x
e 0
m _m_-_1 ... m - (v - m - 1)
m + 1m +2 In + (v - m)
2m) 22m
Notice that ( m _N Vmn - [202J, and that, furthermore,
( 2m v+ 1 ) ( 2m )
e:) .
m+1 v-I
c::: : :)- -v-
59.
iI 1 2n - vl 2
z- v
= iI (2n - = iln ______~ __ .___ _
V)2
t · 8nv t
4n 2 4nv cos ~ + v .=1 I + sin2 --=-
0=1
Vn Vn
~=1 - 2
(2n - V)2 2
J ~_x~2_dx
1
n -
1 (2-x)'
Nil 8 2NeO .
0=11 + -~-!.......
(2n - V)2 4n
thus
'" n
L1 2p(X, Y) = 1: 1: L12p(X, y).
R 1'=1.=1 Rl'v
O,l,1:
..
,n-l
::..
2 I . k) I
log sin (J1t - ~ =
2
~ log n -
(1)
1 - - ~ log 2.
2
i<k n2 Inn 2n n 2
Argument 20!
62. Cf.54.
63. We denote the expression in question by lIn" Then the inequality
used in solution 54 yields for sufficiently large n
Cauchy's inequality [SO] provides an upper bound for the right hand
side:
64. [Cf. G. P6lya: Math. Ann. Vol. 74, pp. 204-208 (1913}.J Sub-
divide the space by the three sequences of planes
5 3 3 5
x, y, z = ... , - 2n ' - 2n' - 2n ' 2n ' 2n ' 2n' ...
thus
k = 1, 2, ... , p.
Introducing
F(z) = 2: n rx .+ rx ,+".+rxp-lzn
n=1
we obtain
F(t) ~ F(IXI + + ... + IXp) (1 -
IX2 t)-(rx.+rx,+".+rxp) ;
1
J J '" J I(x
b b b
= p! l) !(X2 ) '" I(xp) dX l dX 2 ••• dxp.
a a a
Moreover [I 62J
n (1 +
n
v~l
zlvn~n) ~ (1 + zM~nt ~ ezMonn = ezM(b-a).
+ ;! ! I(x) dx
P( b )P
+ ... =
z
ea
J f(x)dx
This new solution of 54 illustrates well the limit operations which lead
to Fredholm's solution of integral equations. Conversely 67 can be
deduced from 54 with the help of I 179.
68. Multiplication by rows leads to a determinant P of order m with
the general term
n b
E 1~~CfJ'::,) ~ b :
v=1
a
a
J IA(X) CfJI'(x) dx; A,/h = 1, 2, .... , m.
76. [0. Holder: Nachr. Akad. Wiss. Gottingen 1889, p. 38.J We put
Pltl + P2t2 + '" + Pntn = M,
PI + P2 + ... + Pn
then
thus
n {t - M}2
P{P{tl} + P2CP{t 2 } + ... + pncp{tn} :£ p. - ' - - - cp"{Tv}
PI+ Pz + .,. + P n
= cp(M) + .=1 : >cp(M) ,
:£ p•
• =1
a b l )
<L; ( iX.: +f3~ +"'+A~
=iX+f3+···+A=1.
81.4. From 81.3 by a passage to the limit or by analogy. Analogy
may be better: It may allow us to discuss the case of equality.
82. Let t =l= 0 and introduce a~ = A" Y = 1, 2, ... , n. Because of 78
we have
21p'(t) Al log Al + Azlog A2 + ... + An log An
t -- =--- ------- - -------- ----------------- --
1p(t) A] + A2 + ... + An
_
Al+ A2 + ... + A
log--------- ____n > O.
n
We find
'IjJ(- <Xl) = min (a), 11'(-1) = ~(a), 11'(0) = @(a), 11'(1) = m(a) ,
11'( + <Xl) = max (a).
Thus we have a new proof of the proposition on the relation between the
arithmetic, geometric and harmonic means.
83. Assume t =l= 0 and set [f(x)]t = F(x). Proposition 79 implies
lJI'(t)
t2 -
j'F(x) log F(x) dx
= -X,- -
lJI(t) x,
---log - x,
- - jF(x)dx
x. - Xl
)
:2 O. (1
J F(x) dx
--
x,
x,
(M - c) (.;-~-;-f <
2 I
P(t) < M,
Pt. II, Solutions 81.1-87 251
i.e. 1[1(00) = lim 1[I(t) = M; 1[1(- 00) is found in a similar way. This
1-+00
proposition contains therefore a new proof for 69.
84. [H. Minkowski; d. e.g. Hardy, Littlewood and P6lya, Lc. 69,
p. 21.] First proof: We assume 0 <t < 1 and define
1
n
n
qJ(t) = eta. + (1 - t) b.] n . Then [80]
0=1
cp"(t)
cp(t) = n 2
1(n.~ tap
a. - b.
+ (1 - t) b.
)2 -;-.~
1" (
- tap
a. - b.
+ (1 - t) b.
)2
< 0,
unless a. = Ab., 'II = 1, 2, ... , n.
Second proof: With log b. - log a. = t. the inequality becomes
cp"(t) _
cp(t) -
(_l_J
x2 - Xl x,
x
, f(x) - g(x)
tf(x) + (1 - t) g(x)
dX)2
__l_J (
x2 -Xl
x,
tf(x)
f(x) - g(x)
+(1 - t)g(x)
)2 dx::;:_0.
x,
V(x(·) -
n
lk = least upper bound of J; X(v-1))2 + [A(x(V)) - MX(·-1))]2
.=1
for all possible subdivisions of the interval [Xl' X 2 ]
(Xl = X(O) < X(l) < X(2) < ... < X(n) = X 2 ).
252 Integration
Since V -+ t
C2 Z is convex [73J we get for an arbitrary subdivision
PI + Pz + ... + Pm - v~l i: h
k~l
88. Put
,-
2"
n
p~nl = f p(~) d~, V = 1, 2, ... , n;
(v-II 2"
n
89. Using the same notation as in 88 we establish that lim Fn(x) =F(x)
uniformly in x, 0 <x< 2n: n-+=
I 2"
2: + x)] d~.1
i
2"
< maX(P)i [v~ 1/((v -1) ~ +~ +x) -I((v -1) 2: +~)IJ d~.
Thus
IFn(x) - F(x) I < 2:_ :ax~ V,
I P(~) d~
o
where V denotes the total variation of I(x) on [0, 2nJ. Since the length I
of the arc of I( (v - 1)~. + x) ,0 < x <2n, is the same for each v,
Pt. II, 'Solutions 88-92 253
the length L~ of the arc of Fn(x) can be estimated by Ln < 1. Besides,
the limit relation
2: V(x(a) + [F(x(a))
s
- x(a-l))2 - F(x(a-l))]2
a=l
2: V(x(a) + [Fn(x(a)) -
s
= lim - x(a-l))2 F n(x(a-l))]2 < 1
n",,*oo ~=l
We may also assume that not all the a~'s are equal, nor all the b.'s, i. e.
anb l - alb n = (an - al)b l + al(b l - bn) > o.
If n > 2 the numbers u I ' u2 , ••. , un-I' VI' V 2, ... , Vn_ l are defined by the
equations
We find U v > 0, Vv > 0 and avbv > u~albl + v~anbn [80J; 2t~ = 0 if and
only if a~ = a~+l = .,. = an' bv = bv+l = ... = bn and v,. = 1. In a
similar way Vv = 0 implies u~ = 1, etc. If u~ > 0, v. > 0 then
a~b~ > uva l bI + vvanbn. Thus the expression in question is
(v bl
2 Vpa l b]qanbn
an Val b--';' )2
--+
a l bn
--
anb l
- 2
a~ = t( Xl +v-
X2-Xl)
n- - - ,
bv =g ( Xl +V---
X 2 - X1 )
n- - , v = 1,2, ... , n.
r
Then
(17a=): =
(17<) T Rr
Pt. II, Solutions 93-94.1 255
Whereas several foregoing problems of this chapter were arranged
in pairs, each problem about sums followed by a companion problem
about integrals, the present problem has no such companion. (Why is
that so? Whereas in former cases, e.g. 81.3, multiplying ~ach E by ~
n
leaves the relation in question unchanged, this is not so in the present
case.)
94. [G. P6lya, Problem: Arch. Math. Phys. Ser. 3, Vol. 28, p.174
(1920).J Assume that I(x) is not constant. We show that the quadratic
form
1
Q(x, y) = J I(t) [(2a + 1) t2ax2 + 2(a + b + 1) t~+bxy + (2b + 1) t2by2J dt
o
= Ax2 + 2Bxy + Cy2
is indefinite, i.e. AC - B2 < o. Integration by parts leads to
f t I(t) dt
1
k
= k + 1: I(t) )1
( tk + 1
0 -
f 1
k
tk+1
+ 1 dl(t) = k
/(1)
+1- f 1
tk + 1
k + 1 dl(t) , k>O,
o 0 0
94.2. By symmetry
< ~- (a 2 + b2 + e2).
To obtain the last line work backwards and use that c < a, e ~ b.
(3) The more elementary 94.1 yields better estimates when c is near
O. For other approximations to the area of the ellipsoid see G. P6lya:
Publicaciones del Instituto de Mat. Rosario Vol. 5, pp.51-61 (1943).
94.3. Lower bound in 94.2 and the theorem of the means.
95. [G. P6lya, Problem: Arch. Math. Phys. Ser. 3, Vol. 26, p.65
(1917). Solved by G. Szeg6: Arch. Math. Phys. Ser. 3, Vol. 28, pp. 81-82
(1920).J [For a generalization see G. P6lya. and M. Schiffer: Journal
d'Analyse math. Vol. 3, p. 323 (1953/54).J
95.1. Consider the polynomial in u
w = (a 2 + u) (b 2 + u) (c 2 + u).
By the theorem of the means
w
- 3
1
<"3
1(1 1 + ~~;
a + u +b +
2
1) = 3w1 du
2 U
dw
and so
= 1 = 1 [1]00 - 31
1
2 f w -"2 dU<21 0f
o
w
-"6 w'
3w du= -w
-"6
U~O = (abc).
Obviously
S2" = Ey;ny.-" < y-n s".
The result is useful in certain applications of Graeffe's method, see l.c.
95.3.
95.5. [See Hardy, Littlewood and P6lya, l.c. 69, p. 163, theor. 218.]
2"
2A = f r2 dq;.
o
1(: t (1d~ f3 (1 f3
By 81.2
2n = 3
(r2)1/3 dq; < r 2dq; = p 2/3 (2A )1 / 3.
96. [Cf. 1. Schur: Sber. Berlin. Math. Ges. Vol. 22, pp.16-17
(1923).J Suppose that xl' x2 ' .•• , xn are positive. Since log x is concave
on any positive interval
log Y", ~ a",l log Xl + a",2 log x 2 + ... + a",n log x n ' fl = 1, 2, ... , n.
j(x+ It) = ~;
q
~q
converges to 0 as h -» O. If x = 1..-, rational, j(x) =
q
-~
q
,
100. If we call Q. the oscillation of j(x) in the interval [X.~l' xvJ and
if !q;(x) 1< M we get
I.~j(yv) q;(rJ.) (xv - X'~l) -v~~(rJv)q;(rJv) (xv - xv_I) I < Mv~~Dv(Xv -xv~l)'
The last sum converges to O.
101. Let n denote a positive integer, t5 < b -n a and Dv the oscillation
b _ a n~l b - a
J !q;(x + t5) -
b
cp(x) Idx < - :;;,-- L: (Q. + Q'-i-J) + 2M-n- ·
a v=l
Pt. II, Solutions 96-105 259
102. Construct the lower sum L and the upper sum U that belong to
the subdivision a = Xo < Xl < ... < xn = b (as described on p. 46):
n n
U = L; Mv(xv - xv-I)' L = L; mv(xv - Xv-I)'
.~l v~1
a
J P(x) dx = U, J 1J!(X) dx =
a
L-
The only condition imposed on the subdividing points is that the maximal
length of the subintervals [xv_I' XvJ, 11 = 1, 2, ... , n, converges to 0
as n increases. Therefore these points can be chosen equidistant, forming
an arithmetic progression. The functions P(x) and 1J!(x) constructed in
the described way are continuous on the right; they could be defined
continuous on the left instead.
103. Define P(x) and 1J!(x) as in solution 102. Then the total variation
of P(x) is
Both are not larger than the total variation of t(x) because t(x) assumes
on [xv_I' xvJ values which are arbitrarily close to Mv and m,.
104. Let 11 be an integer, 11 = 1, 2, ... , n; in the first half of the
interval [11 ~~, :J we have s(nx) = +1, in thesecondhalfs(nx) = -1.
Thus
I
1 + Y) - t(1I ~~ + Y + 2~)}dY.
o 0
of Fourier's Series, 2nd Ed., Vol. II. New York: Dover Publication 1957,
pp. 514-515.J We may choose a = 0, b = 2n.
2n
J I(x) sin nx dx
°
=jsinnyv~ {/((1'_1)2: +y)_/((1'_1)2: +Y+:)}dY.
The absolute value of the expression in the curly brackets is smaller than
the oscillation of I(x) on [(v - 1) ~;- , v 2:J .
106. [L. Fejer: ]. reine angew. Math. Vol. 138, p.27 (1910).J We
may choose a = 0, b = 2n.
2" 2n
v- v-
2n n n n n
J I(x) Isin nx I dx = ~ J I(x) Isin nx I dx = ~ I,'n J Isin nx Idx,
o v~l(v_l)2" v~1 2n
(v-l)n
n
where Ivn denotes a value between the least upper and the greatest
lower bound of I(x) on [(v - 1) 2: ,v 2~_].
107. If the points of discontinuity of the bounded function I(x) have
only finitely many accumulation points on [a, bJ it is possible to find
finitely many intervals of arbitrary small length outside of which I(x)
is continuous. Such a function is therefore integrable. The function in
.
questIOn h as t h e pomts
. 0 f d'IscontmUlty
.. 1 ' - 1 , ... , - 1 , ... ; an d
x = -2
°
3 n
also x = if iX = 0.
108. According to Riemann's criterion it is possible to find in an
arbitrary subinterval [a o' boJ of [a, bJ a smaller subinterval [aI' bIJ,
b1 - a l < ~'Li~~' in which the oscillation of I(x) is smaller than ~ .
Iterating this procedure we obtain a sequence of intervals [an' bnJ,
n = 1, 2, ... , bn - an < ""--~~
n
, on which the oscillation of I(x) is smaller
2
than in . The point iX = lim an = lim bn lies in [ao' boJ and I(x) is conti-
2 n~= n~=
nuous at iX.
109. Assume that I(~) = ° at each point ~ of continuity of I(x),
a < ~ < b.
b n
a
JI(X)2 dx = lim ~ f(~.)2 (xv - X,_I)'
n-+= v=l
Pt. II, Solutions 106-114 261
where xv_1 < ~v< x. and the maximal length of the subintervals
[Xv_I' Xv] converges to 0 as n -+ 00. According to 108 ~v can be chosen
such that f(x) is continuous at X = ~., and so f(~.) = O. Let, on the other
hand, be f(~) =1= 0 at the point of continuity ~, a < ~ < b. Then we have
for 15, 15 > 0 and sufficiently small, f(X)2 > /(;)2 whenever IX - ~ I < 15,
and therefore
b ;+"
Jf(X)2 dx > J j(X)2 dx > !5j(~)2 > O.
a ;-"
110. Assume that e, 'Y} are given, e, 'Y} > 0 and that 15 is such that
I!P(YI) - !P(Y2) I < e whenever IYl - Y21 < 15. Since f(x) is integrable, a
subdivision of [a, b] can be found for which the total length of the sub-
intervals where the oscillation of f(x) is > 15 is < 'Y}. On the other sub-
intervals the oscillation of !P [f(x)] is at most e.
111. [Cf. C. Caratheodory: Vorlesungen tiber reelle Funktionen. Leip-
zig and Berlin: B. G. Teubner 1918, pp. 379-380.] Let f(x) be defined
as in 99 and G(x) as in 98 and
( 1 for Y = 0
!p(y) =10 for'y ~ O.
2% 2", a+1
J Caj(C) dC < j(X) J ca dC = ~+1 f(x) 2 a +~ ~ .
'" '"
In the case of a = -1 the last factors, both positive, must be replaced
by log 2.
113. Change the variable of integration in 112: substitute ~ for x.
x
Or prove the statement directly in a similar manner as 112.
114. The integral over [0, e] exists if and only if the integral of
x" (1 - x; ).
.,6 or of x"e-
t ",P+2 converges, i.e. certainly for {3 < -2; for
{3 > -2 if and only if (X> -1. The integral over [£0, oo), £0 > 0, is
convergent for {3 < 0 if and only if (X < -1. If {3 > 0 and n an integer
262 Integration
f Icos x l(nn)f3 dx
o
and ilcosxl[(n+l)"Ji1 dx,
o
fJ
which increase like n 2 [202]. For the integral to converge we must have
<X - ~ < -1. Combining all these results we find that the integral in
question is convergent if and only if either <X < -], (3 < - 2 or if
-1<<x< ~-1.
114.1. Let aI' a 2, a 3, ... be a sequence of positive numbers such that
~ ~
l
Set for n = 1, 2, 3, ...
a for n - 1 ::;; x < n - a!,
n
I(x) = 1 for
n - a! ::;; x < n.
an
Observe that
n n
J I(x) dx <
n-l
2an,
n-l
J [/(x)J' dx > t a: + a!-"'.
115. The limit function is properly integrable over every finite inter-
w w
val [-w, wJ and lim J In (x) dx = J I(x)
n-.,..oo -w --w
dx. Since I/(x) I < F(x) the
J I(x) dx exists.
~
integral
We have
V-
2: (:) = 2~
-"V"
,,'In
f _(cos V; 2 rcos Anx dx.
In 115 we put:
n
117. Put ~ a.e-·Y = P n (y). Then
.=1
2
s
o
According to 115it is sufficient to findF(x) such that IP,,(y) yS-11 < F(y),
J F(y) dy exists. Partial summation yields
co
whereby
o
"-1
P,,{y) = ~ D.(a) [vae-'Y - (v + l)a e-(V+1)Y] + naD,,(a) e-"Y .
• =1
264 Integration
The first two integrals on the right converge to 0 as w ---* 00, the third
integral converges to 0 as n ---* 00, w fixed [105J. The proof of the second
part of the problem runs similarly:
118.1. [G. P6lya, Problem: Jber. deutsch. Math. Verein. Vol. 40,
2. Abt., p. 81 (1931). Solved by G. Szeg6: Jber. deutsch. Math. Verein.
Vol. 43,2. Abt., pp. 17-20 (1934).J
119. (1) Reduction to two functions of two variables possible: Let
tp(x, y) make different values u = tp(x, y) correspond to different pairs
of numbers x, y (e.g. a one to one correspondence between the xy-plane
and the number line u). For a given function !(x, y, z), 1p(u, z) is then
found in the following way:
If u* does not belong to the range of tp(x, y), 1p(u*, z) is chosen arbi-
trarily, e.g. 1p(u*, z) = l.
If u* does belong to the range of tp(x, y), it corresponds to a unique
pair x*,y*, thus u* =tp(x*,y*) ;in this case we choose 1p(u*, z) = !(x*, y*, z).
Then it is true for all x, y, z: 1p(tp(x, y), z) = !(x, y, z).
(2) Representation impossible, e.g. for the function
!(x, y, z) = yz + zx + xy.
Arbitrarily many pairs Xl' Yl; x 2 ' Y2; ... ; xn' Yn exist so that a
given continuous function tp(x, y) assumes the same value for all of
them. (They are points of the same "level line".) If tp(x, y) = const. the
statement is evident. If tp(x, y) is not a constant and if e.g. tp(x', y') = 1,
tp(x"', y''') = 3 there exists on each circular arc connecting x', y' with
x"', y'" an intermediate point x", y" for which tp(x", y") = 2.
Pt. II. Solutions l1S-U9a 266
If I{x, Y, z) = VJ{rp{X, y), z) as proposed, rp(x, y) continuous. there are
arbitrarily many pairs xl> Y1 ; x2 ' Y2; ... ; X". y" for which
l(x1, Y1' z) = l(x2, Y2' z) = ... = I(x", Y", z)
identically in z. If
*
(1) The relation I(x, Y, z) = rp{VJ[x(x, y), z], z} for all values X; Y, z
for which I'Y = z + X 0 implies
:z (~) = ! (::::~:) = 0,
Ix 'f/!x 0
i Iy
0 Xy 1= O.
10 'f/!z Xz I
We can assume that 'f/!x =1= 0, Xy =1= O. If 'f/!x =f= 0 and Xy =f= 0 we may write
'Pz Xz
-- == ~7), -=-u
Xy ,
'Px
Iytt + Ixv + Iz = O.
x +
y Xz
x - y Xy
The right hand side depends on y and z = - y only: contradiction.
(3) Differentiation of I(x, y, z) = CfJ{'f/![X(x, y), z], x} implies
a < ~ < x, x < 'fJ < b. It is not possible that both inequalities hold
identically because such a function would cease to be differentiable at
a+b
x = - - - . Thus
2
a+b
J I(x) dx < M J (x -
b
a a
2 .
a) dx + M J (b -
b
a+b
x) dx = M -T-
(b a)2
.
-2-
122. [W. Blaschke; Problem: Arch. Math. Phys. Ser. 3, Vol. 25,
p. 273 (1917).J According to Taylor's theorem we can write for x ~ Xo
f J J ~~
xo+r xo+r xo+r
[I(x) - I(xo) ] dx = (x -2 x o)2 t" (x) dx = I" (~) XO)2 dx,
xo-r xo-r XO-1
because the function t" (x), being a derivative, assumes all the values
between its greatest lower and its least upper bound.
122.1. The function I(x) is linear; method of 122.2.
122.2. Introduce the abbreviations
I(a) = A, I(x) = X, I(b) = B,
and take for (u, v) first (a, x), then (x, b), then (a, b):
(x - a) VAX + (b - x) VXB = (b - a) VAB,
268 Integration
and hence
1
I(x) = X = (ex + d)2 '
where c and d are appropriate constants. Convince yourself by performing
an elementary integration that a function of this form satisfies the
imposed condition.
122.3. By the method of 122.2
1
I(x) = Vre~+ d! .
123. First proof: Suppose that the series L: Pnenx = I(x) is con-
n~O
(
Xl + X2 + ... + Xn ) < 'P(X 1) + 'P(X 2) + ... + 'P(Xn )
cp n = n
(*)
t > q:>(x) - ~x - mol > 'P(x) - q:>~x - no) .
Pt. II, Solutions 122.3-126 269
Hence for m = 1
G -ntp(X) > qJ(x + <5) - qJ(x) > qJ(x) - qJ(x - <5) > tp(x) n- G.
I
-6
n
Since qJ(x) is continuous:
tp(x+ 6) - tp(x) > tp(x + 6'), - tp(x)
6 - 6 0
> tp(x) - tp(x - 6)
, > tp(x) - tp(x - 6)
<u <u.
51.1 51.
= 6' = 6
As <5 -+ 0 the first term converges to a limit l+ and the last to a limit L,
l+>L
125. [G. P6lya, Problem: Arch. Math. Phys. Ser. 3, Vol. 24, p. 283
(1916).J The values which y = I(x) assumes on an interval of length l
fill a closed interval of length L:
L = max I/(xl ) - I(x~ I = max III'(X) dxl < l max II' (x) I·
Let e > O. About each point x where I' (x) = 0 construct the largest
interval for which If'(x) I < e. We denote the length of the interval by
lx. The values I(x) is assuming on such an interval cover at most an in-
terval of length (,e. The points of M (y = I(x) with I'(x) = 0, a < x <b)
are enclosed in countably many intervals of total length < e(b - a). This
proof shows in addition that the set M has measure o.
126. [U.Dini; d. C.CaratModory, l.c. 111, pp.176-177; Hille,
Vol. II, p. 78.J It is sufficient to consider the following case:
11 (x) > 12 (x) > ... > In(x) > ... , In(x) continuous, lim In (x) = 0,
"->-00
o< x < 1. If the convergence were not uniform, infinitely many points
x" would exist for which In (Xn) > a > 0, 0 <X" < 1, a independent of n.
Let e
denote an accumulation point of the x,,'s and m be such that
270 Integration
x+e
The absolute value of the first term on the right is <0 J Pn (t)
x-€
dt ~ 0
whenever If(t) - f(x) ! < 0 for x - 8 < t < x + 8. The absolute values
of the other two terms are smaller than
11
8
a
J Pn(t) f(t) dt - f(x) = J Pn(t) f(t) dt + J Pn(t) f(t) dt + aJ Pn(t) dt
x-e X+£-1]
b
+ J Pn(t) dt-~ 0
x+e
follows that the condition is necessary because all the terms are positive.
Pt. II, Solutions 127 -133 271
J e-et dt =
",
hence for w -+ 00
1
0 0 0
e-et1p(t) dt = e Je-et (11p(T) dT) dt.
[1 - (x - t)2t
-~-------.- -
1
J [1 - (x - t)2t dt
o
For 0 < e < x< 1- e
X-e -e
-e
j(l - t r dt "-' V1l-,,-,~·~·~
2
n 1 3 5
... 2n~+ 1..
272 Integration
134. [L. Fejer: Math. Ann. Vol. 58, pp. 51-69 (1904).J Apply 132:
It'1(x) - P(x) 1< 'YJ, then t(x) ::;: t'1(x) - 'YJ < P(x)
Pt. II, Solutions 134-140 273
and
b b b b
J P(X) dx - J I(x) dx < 1J(b - a) + J I~(x) dx - J I(x) dx.
a a a a
thus
b b
J [I(X)J2 dx <
a
max I/(x) - P(x)
a~x~b
I' J I/(x) 1dx.
a
[135,109.J
o
J I(x) cos
2"
. nx dx, n =
sm
0, 1, 2, ...
139. Determine P(x) as in 137 and assume I/(x) 1 <M. Then [d.
solution 138J
< Me.
b b
J [I(X)J2 dx :S:: M J [I(x) -
a a
P(x)J dx
140. Let I(x) not vanish identically and not change sign more than
JI(x) dx =
b
n - 1 times. According to the first condition, 0, there exist
a
numbers Xl' X2 ' ... , Xk, a < Xl < X2 < ... < xk < b, with the following
property: I(x) does not vanish identically in any of the subintervals
(a, Xl)' (Xl' x2 ), ... , (Xk- l , xk), (Xk' b); I(x) does not change sign in any of
these subintervals (V, Chap. 1, § 2) but the signs alternate for consecutive
subintervals. Therefore I(x) (x - Xl) (X - x2) ••• (X - xk) has the same
sign for the entire interval (a, b) and does not vanish identically. Accord-
ing to the hypothesis we would have in the case k < n - 1
b
J I(x) (x - Xl) (X - x2 ) ••• (X - xk ) dx = 0
a
+ (k - JIP(x) e-(k-ko)"dx.
00
Je-nxx
o
a- l cos (t log x) dx = ffi Je-nxx
0
s - 1 dx = ffi r~)
n
= 0, n = 1,2,3, ...
n " ( )n
Kn(x)=2;en(:)xV(l-xr-V= en x+l-x =
1 [1+ (1
i' -Ix .
) ]n
• ~o
Pt. II~ Solutions 141-148 275
145. We have [140]
For k = 1 the value of the limit is obvious; if k > 1 we have for suffi-
ciently large r
logk (2r) logk r2 logk -1 (2 log r)
1< -logk
-- <- - = --.----
r logk r logk_1 (log r)
[154].
so that
lim inf and lim sup of the middle term lie therefore [155J between
o
f1p(x+) dx and f
0
p(J)
dx, which differ from f t(x+) dx by less
0
than s.
157. According to 147 we have
_1_
I'n~r
N(r) (N(r)
N(r) ~_ (rnr)"'--' = /-'" /.-ex
+ (J.. _ IX) J' N(t) t-A+ex-1 dt )
0
For
r-"'L(cr) t"'-1
L(r) c,
j dt < r-'"
L(r) 0
j L(t) tx - 1 dt < r-" j tex - 1 dt = ~ ;0 <X
lim inf and lim sup of the middle term lie therefore between !--=~ and ~
where c is arbitrarily small. <X <X
-1
- '" (r~)-"'-). = -1 + (<X +N(r) J N(t) t-"'-).-l dt
A) "'+-' =
Y
N(r)"::" Y,
~>, ,
159. [As to 159-161 d. G. P61ya: Math. Ann. Vol. 88, pp. 173-177
(1923).J Generalization of 155-158. As in 156 bound j(x) by two func-
tions 'IJ'(x) and P(x) that coincide with _x",-i. and x"'-i. respectively on
[0, b), with j(x) on [b, w) and with _x-"'-i. and x-",-i. respectively on
[w, =). We have [157J
and [158J
f-l = 1, 2, ... , n - 1.
f-l = 1, 2, ... , n - 1.
---~j(~)
N(r)
~ _1_ jr(!!:)~]
r , - n - 1 n
> _ n-
- n - 1 p
j- j(x})dx,
n
The integral /
o
l x*) dx is a continuous function of f3 [131]. Similar
arguments apply if I(x) is increasing. Replace I(x) by -/(x).
161. Let 0 < iX < A < f3. By making use of 1116 we establish simi-
larly as in 160 the existence of arbitrarily large n for which
1
N(rn)
=
k~ I
(r) 1 ( 1)
r: < [I x~ dx + /
= ( 1)
I x7f dx.
1 1
The first and the last expression converge to f lj!(x) dx and f P(x) dx
o 0
1
resp. and both are arbitrarily close to J I(x) dpc. The weaker conditions
o
. vn(O,{l)
hm - - -
n---;.-oo n
=fJ, O<fJ<l
or
O<fJ<l
are also sufficient. It will even do if one of these conditions is satisfied
for a set of fJ-values everywhere dense on (0, 1).
163. The condition is obviously necessary. As in 162 the interval
may be open, half-open or closed. We replace the functions lj!(x) and
280 Integration
lJI(x) in 102, if a > 0 by piecewise linear functions for which the extensions
of the single line segments pass through the origin (y = kx). As in 162
we now can establish (*) for a properly integrable function on [O,lJ
that vanishes in an interval containing the origin, e.g.
. vn ((J,l)
hm - - = 1 - p, 0< p< 1 [solution 162].
n--+= n
provided that the limit on the right hand side exists. For
an = (n() - [noJ) e2ninct
this limit becomes, according to 166,
= J xe-,,,qx dx =
o
1 9' 1
----c- •
2::nq
169. [E. Steinitz, Problem: Arch. Math. Phys. Ser. 3, Vol. 19, p. 361
(1912). Solved by G. P61ya: Arch. Math. Phys. Ser. 3, Vol. 21, p. 290
Pt. II, Solutions 164-174 281
(1913).] The limit is the function t(x) defined in 99. For irrational x d.
166; easier in the case of rational x.
170. We obtain 10"0 - [10"0] by multiplying the decimal fraction
by 10" and omitting all the digits to the left of the decimal point. Let
IX = 1X11X2 ••• IXk represent a finite decimal fraction. Choose n so that
10"0 - [10"0] starts with the digits lXI' 1X2' ••• , IXk and that r zeros follow.
Then
1 10"0 - [10"0] - IX 1 < 10!+r.
x
X--
x
above that I' (x) -+ 0 as X -+ 00; furthermore ~ -+ 1, 8 fixed or
y(x + e)
bounded, as X increases.
Let 0 < IX < 1. It is sufficient to prove [162J that
m-l
J: (N(k + <X) - N(k)) + N(m + An) - N(m)
k=l , m = [g(n)J, An = min (xn' IX)
N(m +x n )
y(m + <X)
2y(m)
- y(m) + _1_
y(m) x.
j (y(x + IX) _ y(x)) dx + 0 ('Y'(m~)
y(m)
m
= y(:) J y'(~) dx + 0(1),
x.
lim
n-+oo
Sn
non
= JIsin 2nx I dx = ~ .
Thus (so = 0)
=
n-l
L:
• =1
s. -
(1 vQ (v
1)
- - - - - +--+ +00 .
+ 1)2
sn
ne
Pt. II, Solutions 175-180 283
178. [J. Franel: Vjschr. Naturf. Ges. Zurich Vol. 62, p. 295 (1917).J
Write in decimal notation
V-n-- c(n)
·123
c(n)c(n)c(n) • ••
.
(We exlude the case where all the c;n)'s are equal to 9 for f larger than a
certain given index.) In 175 we put <1 = !, a = 10j - 1 ; then
T n - [10 j -
xn -_ . \l"i-lV- 1 V-J
n -- 0. ,(n) (n) (n)
[j c j + 1 Cj + 2 ···,
1.e.
•
cj(n) = [ 10]
xn • We f'III d Cj(n) -- g 1'f an d on1y 1'f 10
g <
= Xn + 1 . Th·IS
< g-W
means
g+1
10 1
=J g
dx = 10·
10
J K(x,~) dx as
IX
is continuous in the interval [0, 1J, cp(O) = <p(1); cp(~) is constant if and
only if I(x) assumes the same value at all its points of continuity (differ-
entiate). As a -» 00, q converges to 1, as a -» 0, q tends to infinity. We find
accordingly
1 1
lim
q-+= 0
JI(x) K(x, ~) dx = J I(x) dx. 0
284 Integration
1
As a increases to infinity J(a; I) is reduced to a point, J I(x) dx; the
o
distribution is almost uniform for large a's [176J. If 0 < ;< 1 we find
in addition 1
J
lim I(x) K(x, ;) dx = 1(;)
q-+oo 0
10i - 1
In 179 and 180 set a = log 10' Then
= lOi- 1 Log n - [loi- I LognJ = O. e(n) e(n) e(n) •••
er
XII
J J+l J+2 '
i.e. = [10xn J. Thus we are concerned with the range of the continuous
function g+l
10
cp(;) = J K(x,;) dx
g
10
log 10
on the interval [0, IJ; K(x,;) is defined as in 179 and q = = 10101 - j e1oi - 1
182. Let [solution 174J
m=(g(n)], x=g(t), t=y(x), N(x) = (y(x)].
Because of the conditions (1)--;(4) the function y(x) is continuously
differentiable for x ~ g(l), monotone increasing to infinity as x -+ <Xl,
moreover y' (x) -+ <Xl, y'(X) -+ <Xl, whence y(x - e) -+ 0 as x -+ <Xl and also
y(X) y(x)
n N( m
Now suppose that ~ is a jump point of f(x) anci that again 1xn - ~ 1< ~ .
Then
_ 2M N(m -±-l~ + ,u(o) < !(x 1) + !(x 2) + ... + !(x n)
n n
< 2M N (m + ~ -
n
0) + M(o);
where ,u(o) and M(o) denote the greatest lower and the least upper bound
of f(x) in the interval 1 x - ~ 1 < o. These inequalities imply that the
limit points lie between f(~ - 0) and f(~ + 0). Now we prove that the
limit points cover the entire interval between f(~ - 0) and f(~ + 0):
Since f(x) is integrable there exist points of continuity of f(x) arbitrarily
close to H108]. Let~' and~" be two such points, 0 < < ~ < ~" < 1 and r
let xn ' and xn" (n', nil integers) be two sequences such that xn' -+~' and
xn" -+ ~", [g(n')J = [g(n")J. We set f(x 1 ) + f(x 2 ) + ... + f(xn ) = nFn'
thus F n , -+ f(~'), F n " -+ fW')· Moreover we have for each n
Fn !(xn+l) \ 2M
1
Fn - Fn+l
1
= \n + 1 - -;-+1 < n + l'
The sequence F n" Fn'+l' F n'+2' ... , F n" is "slowly increasing" or "slowly
decreasing" in the sense of solution I 100. It is easy to adapt the proof
to the case ~ = 0 or ~ = l.
183. Special case of 182: g(t) = a (log t)".
184. The proposition follows from 182, 183 for
. -- 1()1-1 --
g(t) = 10J - 1 VLog t = ·V-=~ Vlog t, f(x) = 1 for [10xJ = g, otherwise
log 10
f(x) = O. Cf. 178, 181.
185. [H. Weyl, l.c. 162, pp. 319-320.J It is sufficient to prove the
propos1·t1·onf0 rf( Xl' x 2 ' ••• , xp ) = e2"i(k ,x,+k x +···+kpxp) ,were
h 2k l' k 2' ... ' kP
2
are integers at least one of which does not vanish [165J. Introduce
k 1 a1 + k 2 a2 + ... + kpap = a, k/)l + ki)2 + ... + k/J p = e;
then
-1t ft e
0
2 "i(a + el) dt =
2~m 1
e2 "ia !..---.- - -+ O.
2m(Jt
187. [Cf. D. Konig and A. Sziics: Rend. Circ. Mat. Palermo Vol. 36,
pp. 79-83 (1913).J We may assume that the motion in question takes
place in the square 0 < x <t, 0 < Y < t. By reflection in the lines
x = t, y = t three domains besides f are obtained. The four domains
form a subdomain f* of the unit square 0 < x < 1, 0 < Y < 1. Trans-
lating f* by an integer number of units parallel to the axes we construct
an infinite number of domains as in 186. The original zigzag motion in
the square 0 < x < t, 0 < Y s;: t can be replaced by a rectilinear motion.
Special case of 185: p = 2, l(x1> x 2 ) = 1 if x1> x 2 is in f*, otherwise
l(x1> x 2 ) = 0 in the unit square.
188. [G. P6lya: Nachr. Akad. Wiss. Gottingen 1918, pp. 28-29.J
We put 'tjJ(y) = e2"ik y , k positive integer, in VIn 35. Then we obtain,
using the same notation, g(n) = 0 for n > k. I.e.
I (r,~I'tjJ (~) I = it'n~t (;) g(t) I < Ig(l) I + Ig(2) I + ... + Ig(k) I
for any value of n [165J; and rp(n) -+ 00 d. VIn 264.
189. Using the notation of 188 we set in I70
X
1
= -},' (-1) k-I (k
ank = - - -k - -1 I " cosk {}d{} ) A =
k + .....+-xnn
- -xIn + x2n 2
k n no"
Pt. II, Solutions 187-199 287
193. [Cf. G. Szeg6: Math. termeszettud. Ert. Vol. 36, p. 531 (1918).J
Follows from 192 in analogy to 164.
194. Special case of 193: t(x) = 1 jf IX < X <(3, otherwise t(x) = 0
in the interval [-1, 1J.
195. We can assume a 1 > a2 > ... > at. Then
we find
Use 195, 196. (95.3, somewhat amplified, so that it applies also to finite
sums, is much more informative. See also III 242.)
198. Let t(x) attain its maximum at x =~, a < ~ < b, 8 > 0, 0 be
positive and so small that
then [81J
1! < I n - l I n +1' n = 1, 2, 3, ...
In+l
The sequence - I- is therefore monotone increasing. The value of the
n
limit follows from 198 and 168.1.
200. By introducing the new variable Vkn(x - .;) = t we transform
the integral into
1 l/lm(b-;)
Vkn ,/_ J e
-t'dt
.
-,kn(;-a)
= <p(~') f
~H ~(x-~l'h"W'l
e2 dx + O(lXn);
;-~
~ - 6 < ~" < ~ + 6. The first term on the right hand side lies between
203.
+ 1) = nn+1 f
00
Ckn+ I) = :
I\n+ ~(;~n~~ ~ 1+ -1) ~ ("k n~ ;,Y 1'(-;+ ~(;~n: ~ + 1) n
~ (_k _)1 (nll)nk
II - 1e
(!.. . )n (.,,-_)nk- nJ/~-~~2;n~_
n nil - n 2nn . 2n(nll - n) .
t'
t~ 1
r x' 1 (_~)tx dx.
,t
t' Jx'
/-1
1 ( : r dx
interval [ 0, +]' i.e. fix) ~ J/ 2e < eon [0, ~ ] . Apply 201 to the integral
r;
i x
-'- (1
1-,- -
+ x) , .- I - C\x) d x.
-
(X
a === ~
11
, ) === (X), rr ()
x :::=
1h()
, X == (1+x)x-1-<C\X
C\:--- ,
; = 0,
1
209. Substitute e-1r (1 + x)for x. There results
[1 +O(~)] l
lin ",n-i+pn- 1
= V~ [e- X(l +x)rdx+O(Vne-!n2e ).
-n~(1+6x)-4
The factor e 4 is of the form 1 + O(n- 1 +4 6 ). The integral be-
comes therefore
We have
",n-!+pn- 1 -n~ n~
Since J e 2 dx is of order n- l the O-term of e 3 yields a contri-
-'1
bution of O(n-1+6e). Hence
1/-
V ;n [1 + O(n-1+4e)] ",n-t+pn- 1
J
-'1
x'
e -n"2 (1 + n X;) dx + O(n-1+6e)
1 ",+p n - l -~ ( 1 ",+pn-l __~
=-~
V2n
J
_ne
e 2
X3
3Vn
)
1 +~ dx +O(n-1+6e)=-=-
V2n
J
-00
e 1l(1+~)
X3
3Vn
X dx + O(n--1+6e)
1 '" -~ 1 ",+pn-! _~ 1 ",+pn-! _~X3
= V- Je 2 dx + V- J e 2 dx -~- f
+ V2nn e 2 - dx
2n -co 2n IX -00 3
+ O(n-1+6e).
292 Integration
B' = B + ---
1 ce - "- = -1- ce - ~)- > 0. S'mce A = 1 - A1 the last in-
V2:n V2:n
equality is impossible. In a similar way we show that
where a is real, k> 0, a, k fixed. 212 does not completely imply 210.
213. By a similar argument as in 201 we find that the integral in
question is
<+en
I I
~+en
C"V cp(x) enh(x) dx = cp(f) enh(~) en[h(x) -hEm dx,
~-6 ~-6
Herein hlf W') is bounded and n(x - ~)2 < nrJ! = n- t , furthermore
~+6n
Je n"'(~)(x-<) d _ en6n "'(~) - e
-n'1
n
"'(~) e(Ildogn+/l)"'(;)
x - nh'(~) N nh'(~)
~-'1n
[205.213].
a e n
= ---- J e1-x+ ...- nx X
-n n+1 =( g(nx) )n
dx
n n! 0
into four parts corresponding to the intervals (0, e), (e, 1- e), (1- e, 1 + e),
(1 + e, 00) where e does not depend on n, 0 < e < t, e < y and so small
that in the first interval xe 1 - X + G6 < 1. In the second and fourth interval
xe1 - x < 1; choose 15 = !5(e) so small that we have evenxe1 - x +<,x < 1 and
n so large, n > N = N(e), that g(nx)
nx
< 15 and ne> 1. Then, except on
the third interval, the integrand is O((r), 0 < () < 1, () independent of x
and n, () = ()(e). The mean value theorem of integrals [in addition note
205] implies that
1+6 _
J
-n n+1
an = eK(n~) _e_n_l _ (e1 - x xt dx + O(Vn()n). 1 - e < ~ < 1 + e.
n. 1-6
Hence
log an = g(n~) + 0(1).
g(n) g(n)
294 Integration
n! 22n . 1.. n
(2;-1)(2n-2)(2n-3) ... nV; rv;;
-71Vn
2n( cos
2
=- -])
'In IJ
v-·l
n
----ix
2n . _ v
Vn-
--
:
·dx.
2n e - v.
Apply 115. The limit of the integrand is e- x ' [59J, the convergence is
uniform on any finite interval as a supplement to the proof of 59 shows.
For an appropriate F(x) in the sense of 115 d. G. P6lya: Nachr. Akad.
Wiss. G6ttingen 1920, pp. G-7.-For the generalized Laplace formula d.
also R. v. Mises: Math. Z. Vol. 4, p. 9 (1919).
217.1. The real-valued functions f{J(x, y) and h(x, y) are defined in the
(bounded or unbounded) region at and satisfy there the following con-
ditions:
(1) f{J(x, y) enh(x,y) is absolutely integrabl~ in at for n = 0, 1, 2, ...
(2) The function h(x, y) attains its maximum in at at a single point
~, rJ, and in the region at' that remains when we exclude from at a closed
neighbourhood of ~, rJ the upper bound of h(x, y) is less than h(~, rJ).
Moreover, the second partial derivatives hxx' hXY and hyy exist and are
continuous in a certain neighbourhood of ~, rJ and at the point ~, rJ
n! n! Vx n
. n
-- +---
1
2xn
1
xn - 1
+--~ + ... +--= loga.
1
xn - 2
1
xn - n
n
= 1, lim - - = aft. Conse-
n-..+oo nl-'
Xn
In view elf the first operation (the inner one, written on the right)
296 Integration
*225.
1 - (y - 1)2 when '\I > 2,
max I(x, y) = 1, min I(x, y) ={ --
y x 1 - (3 - y)2 when y< 2.
min max I(x, y) = 1, max min I(x, y) = 0
x y y x
1. Z + Z = 2x, Z - z
= 2iy, zz = r2.
2. The open right half-plane; the closed right half-plane; the open
horizontal strip bounded by the lines y = a and y = b parallel to the
x-axis; the closed sector between the two rays which form the angles (X
and f3 resp. with the positive x-axis; the imaginary axis; the circle with
center Zo and radius R; the open disk and the closed disk resp. with center
Zo and radius R; the closed annulus between the two circles with radii
R and R' and centred at the origin; the circle with center at Z =~ and
ra d'lUS"2'
R
3. The ellipse, and the domain bounded by the ellipse, with foci
z = a and z = b and the semimajor axis k if Ia - b I S k. (If Ia - b I = k
the ellipse degenerates into a segment.) If k < Ia - b Ino point z satisfies
the condition.
4. Let ZI and Z2 denote the two roots of the equation Z2 + az + b = O.
The region in question is the interior of the curve Iz - zlll z - z21 = R2
with "foci" ZI and Z2' The curve is the locus of all points for which the
product of the distances to ZI and Z2 is constant, equal to R2. It consists
of two pieces for R < !z1 -2 z
2
! and of one piece for R > !z1 -2 z.! If
- .
= [d 2r _
dt 2
r (d{})2] ei/)
dt
+ ~er !:...
i/}
dt. dt
(r2 d{}) •
1< 1:, 1< ii~ I< ... < i(~;:T,1 < I~~! > i~~+:),1 > ... ,
are satisfied; i.e. in ffin the absolute value of the n-th term is larger
than the absolute value of any other term. On the common border of
ffin and ffin+1 the absolute values of the n-th and the n + I-st term are
equal, all the others are smaller. In general: let
:zl
iT >1, ·lz-
- 2 l>1""'1Iz-n+111>1, 'Iz-nl <1, Iz-n-1 1<1, ...
l
n n+1 n+2 1
i.e. the absolute value of the n-th term is larger than any other in ffi n.
On the common boundary of ffin and ffin+1 the n-th and the n + 1-st
terms have the same absolute value, all the other terms have smaller
absolute values. (At z = -1 all the terms have the absolute value 1.)
The regions ffin together with their boundaries cover the entire half-
plane ffiz> -1, including z = -1. If ffiz < -1, z =F -1, the absolute
values increase monotonically, there does not exist a largest term.
13. The lemniscates
n = 1, 2, 3, ...
are tangent to each other and to the straight lines ffiz = ±3z at the
point z = O. They intersect the real axis at the points ±n Vi. .2n+1
contains .2n' The inequalities
1Po(z) 1< 1PI (z) 1< 1P 2 (z) 1< ... < IPn- 1 (z) 1< IPn(z) 1> IPn+1 (z) 1>"',
are satisfied in ffi n, i.e. IP n(z) 1 is in ffin larger than the absolute value of
any other partial product. On the common boundary of ffin and ffin+1
the absolute values of the n-th and n + l-st partial products are equal
300 Functions of One Complex Variable
and larger than any other. (At z = 0 all vanish.) The mn's together with
their boundaries fill out the region consisting of the two angles
n n 3n 5n
- "4 < arg z < "4 ' -4 < arg z < 4
and the point z = o. Outside this region and this point Pn(z) increases
with n monotonically.
b
14. We put {} = arg f f(t) ei<p(t) dt. Then
a
If b
f(t) ei<p(t) dt = e- ifJ f
b
f(t)ei<p(t) dt = f
b
f(t) cos [<pet) - {}J dt < f
b
jet) dt,
Ia I a a a
except if <pet) - {} (mod 2n) at all points of continuity of <p(t).
15. [K. Lowner: Math. Ann. Vol. 89, p. 120 (1923).J It is sufficient
to prove m( 4p2 - 2Q) < 3. [Replace <pet) by <pet) + : where
{} = arg (4p2 - 2Q), d. solution 14.J Now we have [II 81J
=
f
- 2 e- 2t cos 2<p(t) dt <
o
If m( 4p2 - 2Q) = 3 then cos 2 <pet) = 1 and cos <pet) has the same sign
at all the points of continuity of <p(t), i.e. <pet) _ 0 or <pet) __ n (mod 2n).
16. The function P1z- 1 P2Z-2 + + ... +
Pnz-n is monotone decreas-
ing from 00 to 0 as z is positive and increasing; therefore it assumes the
value 1 at exactly one positive point C. Consequently
zn - P1Z n - 1 - P2Zn-2 - ••• - Pn > 0 or < 0
according as z > Cor z < c.
17. We have
Izo In = lalz~-l + a2z~-2 + ... + an I
< la11Izol 1+ la2 1lzol 2 + ... + lanl,
n -- n-
9
? 1- Iz l2 -I~l
The last expression is ? 0 if Iz I ~ r, where r is the positive root of the
equation r2 - r = 9, r = i ((1 + V37), 3 < r < 4. The polynomial
corresponding to the number 109, 9 + z2, has the roots +3i.
302 Functions of One Complex Variable
n n 11 - zv x I·
n n
Ix - Zv I =
Such an equation can hold only for ~ x I = 1. Assume I x I < 1: then [5J
Ix - Zv I < ! 1 - zv x I for allY, thus the first product is smaller than the
second. By the same token it is impossible that Ix I > 1. We argue
analogously in the case of P(z) + yP*(z), Iy I = 1.
27. [M. Fekete.J Define y = AP(a) flP(b) , 0 < A < t, A fl = 1. + +
If all the zeros of P(z) - y = ao(z - Zj) (z - Z2) ... (z - zn) were outside
the domain in question, the inequality
:n; a - Zv :n;
- -n < arg ._-.
b - Zv
< -,
n
'I' = 1, 2, ... , n
would hold, thus
P(a) - y P(a) - y
-n < arg --- --- - < n
P(b) - Y ,
in contradiction to - -
P(b) -
--
Y
28. We assume that the straight line mentioned is the imaginary
axis and ffizv > 0 for ally (this can always be achieved by multiplication
with a suitable eia ); then we have also ffi _1_ > 0 and
Zv
P'(z)
P(z)
= i; _1 __ = °
v~l z - Zv '
i.e. = 1
+= 1 1
+ ... +-=.::...=...-= 0,
thus
Z = m l Zl + m 2z2 + ... + mnzn, ml + m 2 + ... + mn = 1,
where the v-th "mass" mv is proportional to [ 1 [2' V = 1, 2, ... , n.
z-z v
(1)
P'(Z)
----
P(z)
- -1 = ----
c
1
Z - Zl
+ - 1-- + ... + ------
Z -- Z2
1
zn
Z -
- --1 = C
O.
Introducing
1
m =
n Iz--- --
Zn 12 ,
(2)
The first term pn the right hand side of the equation represents the center
of gravity of a certain mass distribution at the points Z1' Z2' ••• , zn'
that means a point inside the smallest convex polygon containing all
the points zv. The second term represents a vector parallel to the vector
c. Hence the statement follows. -Cf. V 114.
Pt. III, Solutions 33- 36 305
34. [T. J. Stieltjes: Acta Math. Vol. 6, pp. 321-326 (1885); G. P6lya:
C. R. Acad. Sci. (Paris), Vol. 155, p. 767 1-769 (1912).] Let zv,
y = 1, 2, ... , n denote the zeros of P(z) and assume A (zJ =F O. Then
P' (zJ =F 0 because otherwise the differential equation for P(z) would
imply that P" (zv) = 0 and repeated differentiation would show that
P(z) is identically zero. The equation
implies [31 J that Zv lies in the interior of the smallest convex polygon
that contains the points Z1' Z2' •.• , zv-l' zv+1> ... , Zn' aI' a z' ... , ap (on the
line segment that contains all these points). Consider now the smallest
convex polygon that encloses Zl' Z2' ... , z,,' aI' a z' ... , ap- Only the a:s
and no Zv different from the zeros of A (z) can lie on the polygon.
35. [L. J. W. V. Jensen: Acta Math. Vol. 36, p. 190 (1913); J. v. Sz.
Nagy: Jber. deutsch. Math. Verein. Vol. 31, p.239-240 (1922).J We
denote the zeros of I(z) by z1> Z2' •.. , zn and assume
-----
1
Z -
+ -----
1
Z1
+ ... + --._--
Z -
1
Z2
=
Z - Zn
0, Z =F -z, Z =F zV' Y = 1, 2, ... , n.
XS
n
---+---
z-z z-z.
(1 1) =0.
y=l v Ii
The formula
Z = X
.
+ zy, Zo = Xo
+ zYo,
. ~
~
(_1_ + _1_) __ 2Y y~ - (x - XO)2 - y2
1
Z - Zo Z - Zo (z - zo) (z - Zo) 12 '
shows that the above equation can not hold when Z is outside all the
circles described.
x
36. Writing zn = xn + iYn we find Iz" - ~n_.
cos ex
Therefore the con-
1 :::;;:
n
38. Example: zn = e 2 7'linO (log (n + 1)) - \ () irrational, 1: eZ:nivkO IS
39. We assume that all the numbers Zn are different from zero and
that they are arranged according to increasing magnitude,
tl < IZI I < 1
z21 < 1zal < .... We enclose each zV' v = 1, 2, ... , m, in a
circle with center Zv and radius ~ . These circles have no common inner
points and are completely contained in 1Z 1 < 1 zm I + ~ . Therefore
mn °42 < n (I zm [ + 2° )2, i.e. lim sup m-'>=
_1~g_1n
log
-s;: 2
IZm I -
[I 113].
= nia i: (~)ia ~.
v=1 n n
The first factor is everywhere dense on the unit circle [I 101J; the second
is a sum of rectangles and converges to
J Xia dx =
1
-~--.
o 1 + iiX
42.
r! = 1Zn 12 = (1 + 4-) (1 + !) ... (1 + :) = : . : ... n ~ 1= n + I,
1 -1)
I zn+l-zn -1-~I-l
1
-I'~
Vn + 1 1- lPn -
- arctan _ 1 ~n(1/1+
, rn-rn-l_V~-V;-
lPn-l
V n
V;-
1 1
=2"-Sn+····
Hence rn ~ t ({In [I 70].
43. According to II 59 and II 202 the absolu te value of the expression
in question converges to e- I •• Therefore it is sufficient to prove
. t 2
t n n sm V;-
2n sin V;- log 2 - ~~ arctan - - - - t- = 0(1),
2n cos V-:- - v
where - ; < arctan x < ; . Let n > t2, then [I 142]
v:-
I 2 . t
n sm < 2 V;-Itl < 2 Vn It I = O(n-l).
2n cos -V~ -v - 2n (1 _ ::) - v - n - t2
x3
Thus arctan x = x - 3" + ... can be replaced by x. The statement
now follows from
IX = ao + al + a2 + ... + an + ...
and
308 Functions of One Complex Variable
that
n n n n
1- 1- T= 1(1 _ 1) > [2 > 1.
Pt. III, Solutions 45-49 309
In order that the sum of the absolute values of the coefficients in the n-th
row be bounded the same must hold for lUI I + lu2 1+ ... + luv I, that
is, u l + u 2 + ... + un + .. , is absolutely convergent. Hence the validity
of the other conditions follows, and so the absolute convergence of the
series u l + u 2 + ... + 1tn + ... is the desired necessary and sufficient
condition.
47. [R. Dedekind, d. Knopp, p. 315; Hadamard: Acta Math. Vol. 27,
pp. 177 -183 (1903).-Cf. 1. Schur, l.c. 45, pp. 104-105.J Put
convergent. This is the case if and only if Ic -: 1_1 < 1, i.e. if and only if
me> t.
49. [I. Schur: Math. Ann. Vol. 74, pp. 453-456 (1913.)J The case
e = -k, k positive integer, can be excluded from the start; example:
U o + U I + ... + Un
Un = k _ 1 ' Un + e = 0 for k > 2, un = log (n + 1)
n )
n + 1
(
310 Functions of One Complex Variable
U o + u l -1- ••• + U
fork=l [I 69J. Thecasec=Oisobvious. Weputu +c------------'"
n It + 1
= Zn' un = Wn and, as on p. 111,
(n + 1) zn - nZn_1 = (n + 1) wn - nW n_ 1 + ew n, n = 1, 2, 3, ...
n-I
_
_
-
F(n + c +_12
F(n +" 1) (n + 1) zn _
e......
11=1
~ r(v ~-t...ll
r(l' 1) Zv + r(c + ')~) ZO'
l.e.
n+1 r(n+1) n-I r (v+c+1)
W
n
=
n +c+
Z
1 n
-e-···-------
F(n + c + 2) v~O
---------Z.
T(v + 1) v
2:
exists if and only if ffie > -1. Assume ffie > -1 and put
u o = u I = u 2 ='" = _!.-.- then Zn = W = - J - - thus
1' n
l+c' l+c'
I
--+ Ie I A B
o
f x dx =
I
Y ~
1
lAB
-
+y
[II 22].
The desired necessary and sufficient condition is therefore ffie > -1.
50. Set an = iXn +
if3n' iXn' f3n real. The relation
Pt. III, Solutions 50-55.1 311
lim (IXI
n->-oo
+ IXZ + ... + IXn) n-<J = lim (f3l
n-+oo
+ f3z + ... + f3n) n-<J = O.
hence
51. Whenever the four subseries consisting of the terms in the four
quadrants (ffiz 2 0, 3z 2: 0, etc.) converge, the series converges abso-
lutely.
52. By successive bisection: Assume that all the terms z, ' z, , ... lie
in the sector {}1 < arg z < {}2 and that Iz,,! Iz" I +
diverges. + .. .'
Com;truct the two subseries with terms in {}1 < arg z < 1}1 ; 1}z and in
1}1 ; 1}2 < arg z < {}2 resp. At least one of the two is divergent.
53. Choose a finite number of terms zm = xm + iYm from each of the
successive sectors
54. [More on this topic: P. Levy: Nouv. AnnIs Math. Ser. 4, Vol. 5,
pp. 506-511 (1905); E. Steinitz: J. reine angew. Math. Vol. 143, pp.128-
175 (1913).J Let the direction of the positive real axis be the direction
of accumulation (which can always be arranged by multiplication with
.
a suitable ei"') and z, , + z, + ... be the subseries chosen in 53; apply
1134 to the real part, 1133 to the imaginary part.
55. Z = x + iy, z2 = x 2 - y2 + 2ixy are analytic, Izl = VX2 + y2
z
and = x - iy however are not analytic.
55.1. Cauchy-Riemann equations.
312 Functions of One Complex Variable
images of the lines v = const. All these parabolas have the common axis
y = 0 and the focus x = Y = O. Two orthogonal parabolas pass through
every point z, z =1= 0, of the z-plane.
67. Let z = x + iy, W = u + iv. Then
. ei(x+iy) + e -i(x+iy) eY + e- y e- Y - eY .
u+w= 2 thus u = - - 2 - - cos x, v = ----2- SIll x;
u2 v2
The lines x = const. are mapped onto the hyperbolas cos2 x - sin2-;· = I,
. .
and the hnes y
~
= const. onto the ellipses------- + ------------
~
= 1.
(~~ e-Yf (!=;~ eYr
They have the Gommon foci W = -I, W = 1. The two families of curves
are perpendicular to each other (confocal conics).
68. Elimination of v and u respectively from x = u eU cos v and +
y = v + eU sin v leads to
x - u = eU cos (y ~Ve2u - (x - U)2) , y - v = ex-(y-v)cotv sin v.
The line v = 0 is transformed 'into y = 0 and the line v = n into the
twice covered line segment y = n, - 00 < x < -1.
69. The image of the square is bounded by the two rays arg w = 10
and arg w = -10 and the two circles 1wi = ea +. and 1wi = ea - •• The
area is therefore s(e2a + 2' - e 2a - 2.). The ratio in question is
70.
Because of
1sin (x + iy) 12 = sin (x + iy) sin (x - iy) = -i cos 2x + He 2Y + e- 2Y )
the integral is
X2 -
8
Xl (e2Y, _ e2Y1_ e-2y, + e-2Y,) _ Y2 -
4
Yl (sin 2x - sin 2x )
2 1 •
°
72. When < a ;;; n the image of the square is the simply covered
region bounded by the two circles with radii ea and e- a and the two rays
arg w = -a and arg w = a. When a > n either part or all of the region
is covered several times. If a = nn the image is covered exactly n times,
except certain points of the real axis that are covered only n - 1 times.
73. The intersection of the ray arg w = IX with the circle Iwi = 1;
observe the vertical line segment within the disk IZ I < r that intersects
as many of the parallels Sz = IX + 2kn, k = 0, ±1, ±2, ... as possible:
it lies on ffiz = 0, i.e. on the image of 1 wi = 1 (VIII 16; N(r, a, IX) assumes
its largest possible value for r, IX fixed if log a = 0.)
74. Assume ZI =1= Z2' 1zil < 1, Iz21 < 1. Then
z~ + 2Z2 + 3 - (zi + 2Z1 + 3) = (zz - Zl) (Z2 + ZI + 2) =1= 0.
°
75. If Z = rei/}, r> 0, < {} < n, then w = Re iO = r2e2i/}. R = r2,
() = 2{}, and so R > 0, °
< () < 2n. If, on the other hand, Rand () are
given r, and {} are completely determined.
76. The function in question is schlicht on the closed unit disk
IZ I < 1, where furthermore Iwi ::;;: 1 [5J. The inverse function is
Z
a+ e-i"w
1 + ae-U"w
=---~--,
The circles Iz I = r, 0 < r < 1, are mapped onto confocal ellipses with
1 1
-+r ----r
semiaxes T and ~, the common foci are w = +1, w = -1. The
rays f} = const. are transformed into confocal hyperbolas with the same
foci, w = +1, W = -1. The two families of conics are orthogonal to
each other. If I z 1 = 1, z = ei!J, then w = cosf}. Consequently, if zdescribes
the unit circle then w describes the segment -1 < w :::;;: 1 twice.
80. The function w = kz + k~ , 0 < krl < kr2 < 1 maps the annulus
in question onto the region bounded by the two ellipses with foci w = - 2
and w = 2 and major semiaxes krl + ~klr
1
and kr2 + ~klr respectively.
2
Put
k =a 1 - V~ = a2-V~.
r1 r2
1
z+-
81. The function w = - ~ , z = reiD maps the upper half of the
unit disk into the upper half-plane [79J:
1
-+r --r
w = - r
- 2 - cos
f}. r
+ . f}
~ - 2 - SIll ; ~w > 0 for 0 < f} < n.
jz12 ~II = t
I 2
for the points z = x + iy for which
Ix2 - y2 - 1 + 2ixy 12 = Ix2- y2 + 2ixy 12, I.e. x 2 - y2 = t.
They define an equilateral hyperbola which intersects the real axis at
the points z = ± V~ . The rotation becomes
1
Z2 -- 1 n
arg - - 2 -- =±2
when ffi ~ = 1, i.e. r2 = cos 2f}. These points lie on the lemniscate
Z
Iz - v~-II z + V~ I= ~ .
Pt. III, Solutions 79- 89 317
1
z +--
82. The auxiliary functionC= - ~ maps the region in question
onto the upper half-plane SC > 0 [81J. The origin z = 0 corresponds to
C= ex:>, Z = i to C= 0 and z = ± 1 to C= + 1. The function w = ;2
transforms the upper half-plane SC > 0, note 75, into the w-plane cut
open along the non-negative real axis. The point C= ex:> corresponds to
w = 0, C= 0 to w = ex:>, C= ±1 to w = 1. The mapping function
having the required properties is therefore given by
W = (- - 1
2)2 = (1
4z2
+ Z2)2 •
Z +-z
The images of z = + 1 are both at w = 1 but on different sides of the cut.
83. arg w = f3 ~ IX (arg z - IX), i.e. 0 < arg w < 2n.
"
84. The first auxiliary function C= (e- ia z)f3- a maps the circular
sector onto the upper half of the disk ICI < 1. The second auxiliary
1
C +~-
function s = - ___
C [81 J maps the disk onto the upper half-plane
2
Ss > O. Apply 78.
85. Follows from
u - iv = uX
: [rp(x, y) + i1jJ(x, y)] = ~:
2 uy
[rp(x, y) + i1jJ(x, y)J
by separation of real and imaginary parts.
86. They are the images in the z-plane of the lines 'iR1 = const. and
SI = const. parallel to the axes in the I-plane under the conformal mapp-
ing I = I(z).
87. Follows from 85 by virtue of the Cauchy-Riemann differential
equation ~u
uX
+ dy
~v = O. Also the function 1jJ(x, y) satisfies Laplace's
differential equation.
88. With u cos i + v sin i = 'iR(u - iv) eir the integral in question
becomes
= 'iR Jiz
L
eir ds = 'iR Jiz dz = 'iR [f(Z2) -
L
I(Zl)]
where dz denotes the directed line element with modulus ds and argu-
ment i.
89. u sin i - V cos i = S(u - iv) eir ; d. 88.
318 Functions of One Complex Variab!e
91. The vector if; = ~ eiff forms the angle {} with the positive real
r
axis, its modulus is ~. The functions in question are up to a constant
r
The level lines are concentric circles around the origin, the stream lines
are rays perpendicular to these circles.
92.
W = Z2
2i .
-=t, thus t(z) = z log Z + l'
z- 1
"p = log IZz ~ 11 I ' -
Z -
cp = arg Z + 11 .
The level lines are circles through the points z -" -1 and z = +1, the
stream lines are circles too, namely the ones with respect to which the
points z = -1 and z = +1 are mirror images of each other (circles of
Apollonius).
Pt. III, Solutions 90-97 319
95. We are looking for the points z for which [93]
-~-~-···-~=O
i.e.
the positive numbers AI' .1.2 , ... , An are proportional to the intensities of
the currents. Cf. 31, in particular the first solution given.
96. The vector field is generated by an analytic function j(z) for
which ffij = const. on the given ellipses. The function [80J
provided that a1 > a2 and the positive roots are chosen. The problem is
now reduced to 91 : w = ~ defines a vector field in the Z-plane for which
the concentric circles around the origin Z = 0 are level lines, I.e.
ffi f a: = const. Consequently the same is true for IZ I = const.
f dZ ~dz-
dz Z -
-f V dz
z2-4
along the given ellipses in the z-plane, i.e.
W = - Vz2-4
1 ,j(z) = log (z - V
Z2 - 4) •
The stream lines are confocal hyperbolas, the level lines confocal ellipses
with foci - 2 and 2. The relations between the potentials are
The capacity is
1
99. The stagnation points are z = ± 1; iii assumes the same values at
each pair of points that are symmetric with respect to the origin. There-
fore the resultant total pressure on the pillar vanishes [d. 90]. The
pressure is minimal or maximal when 11 - -\-:
I Z I
is maximal or minimal
resp., i.e. for z = +i and z = ± 1 resp. Rotation of all the vectors through
90° generates a field of force which admits the following interpretation:
A homogeneous electrostatic field is disturbed by a circular, insulated
wire perpendicular to the direction of the field. (The most simple ex-
ample of electrostatic influence.)
100. [G. Kirchhoff: Vorlesungen liber Mechanik, 4th Ed. 1897,
pp. 303-307; A. Sommerfeld: Mechanics of Deformable Bodies. New
York: Academic Press 1950, pp. 215-217.J Supplementary continuity
condition: the boundary of the wake (the stagnant water) stretches
to infinity where Iwi = 1; since Iw I is constant on the entire boundary
it has to be equal to 1. The direction of iii is known along the boundary
segments AB, AD (barrier) and the magnitude of iii is known along the
boundary lines BC, DC (along the wake), iii is completely known at the
four points A, B, C, D. Noticing that either the direction or the magni-
tude of iii is constant on the respective parts of the boundary we find a
half-circle in the iii-plane as image of the boundary of the field of flow.
We fix the constant contained in j [po 123J so that j = 0 corresponds
to the stagnation point z = O. Then the left and right "banks" of the
positive real axis of the I-plane correspond to the streamlines A BC and
ADC, respectively. The I-plane cut along the positive real axis corres-
ponds to the whole field of flow; it is not possible that only a subregion
Pt. III, Solutions 98-101 321
z iii w I
A 0 0 0 0
B I 1 1
C 00 -i 00
D -/ -1 -1 1
region lies to the left right left left
C
Velocity plane (iV-plane)
W
DAB
(w-plane)
A
o
B
-----~------~--
z= J
o
dl
w
= 2 Vi + Vi V1 - I + arcsin VI,
z= x + iy = 2 Vi + ~ - i [I F ~- log (Vi + VI - 1)] .
The first formula for z is to be used when 0 < t < 1, the second when
t > 1 (positive roots) ; it furnishes the boundary of the wake:
322 Functions of One Complex Variable
I
=!(1-w2 )-=!4V1-ld
dJ 1 -- V-I=n.
o W 0
103. Put z = rei!). Since the sign of the angular velocity is positive, {}
is increasing, i.e. z describes the circle in the positive sense. We have
~~ = iz and the velocity vector in question is
ry(z) _ dJ(z) dz _ .1'( )
df} - dz df} - tZ z.
104. Let w denote the angle through which the vector w (radius
vector) has to be rotated in the positive sense to fall into the direction
of the vector iz/,(z) (tangential vector, 103). Then the distance in question
is given by
"" izj'(z)
• <\5 J(z)· ffizj' (z) f[Z)
[/(z) [ SIll W = [/(z) [ I!![~)_I = -Izj'(~···
J(z)
105. The amplitude of the vector in question is 3 log I(z). The angular
velocity is therefore, with z = rei/},
.!:... C'.< I I() - C'.< d logJJzl dz _ 0;<j'(1· _ roJ'jz)
df} 0 og ~ - 0 dz df} - v f(z) tZ - i11..< J(z) .
de "" d ..log
_ zj'(z)
... _dz
1 df}
d~310g izj'(z) <\5_
dz
-
df}
-(! dS lizj'(z) I --Izf'(z) I
liii
The curvature is positive or negative according as the velocity vector
turns in the positive or the negative sense.
Pt. III, Solutions 102-113 323
107.
I concave
I convex
to the left + -
to the right - +
/,(z) /,(z)
ffiz j(z) -=-;, > 0, ffiz j(z) _ b > 0,
I.e.
'iftz!' (z) j(z) - a > 0, 'iftz!,(z) j(z) - b > 0.
114.
iff
z e2 #
cp = Slog 1+; = Slog 2--&- = 2 ' [113J.
cos-
2
115. For ~ < cp ;::::; 3: the support function is identical with the one
of the point ni [112]. Outside this sector we find on the basis of 113
~1 2z ~l 2e # 2
CP=,J
V1 + V2 cos #
og·~~=,J og~=-.
z2 2
121. For I(z) =1= Oin Izl::;;: rthe means are equal: g(r) =@(r) = 1/(0) 1
[119]. If I(z) is the product I(z) = 11(z) 12(Z) of two functions 11(z) and
Mz) that are regular for 1 z 1 ;::::; r, its mean is equal to the product of
the corresponding means. It is therefore sufficient to examine the parti-
cular case I(z) = (z - zo), 1Zo 1 < r. According to II 52
1 2"
- f loglre iff -zoldff
@(r) = e2" 0 = max (r, 1zo)1 = r,
2 r 1 Izol'
-,
r 0
J max(loge,loglzoilede logr-- +----,
2 2r
=e =e
122. [Cf. M. A. Parseval: Mem. par. divers savants Vol. 1, pp. 639-
648 (1805); A. Gutzmer: Math. Ann. Vol. 32, pp. 596-600 (1888).J
Pt. III, Solutions 114-127 325
o(u,-.!'l
o(x, y)
= ou OV
ox oy
_ ou OV
oy ox
= (OU)2
ox
+ (OV)2
ox
= 1o(u +~ iv) 12 =
ox
II'(z) 12,
=
= n L: n 1an 12 (R2n - r2n) •
n=-oo
I
or to the right of the bounding oriented arc. It coincides with the sign of
sin OJ introduced in solution 104. Thus the area is
4 f !~'?)
a \;,
de = 271: i;
n=l
[an [2 r2n [122].
j
o
r
0
I/(ee iIJ ) 12 e de df} = 271:
0
j (i; 1an 12 e2n +1) de,
n~O
where z = eei'~.
131. [J. L. W. V. Jensen: Acta Math. Vol. 3G, p. 195 (1912).J
~ tan2 y =
4 ax + V OV)2
(u ou ax + (u ayau + v oyaV)2 = (u2 + v2) [(81£)2
ax + (aV)2]
ox .
132. Let Zo be a point with horizontal tangent plane. According to
131 we have either I(zo) = °
or I(zo) =l= 0, f' (zo) = 0. As to the first case
°
one has to note that an analytic function has only isolated zeros. In the
second case we consider f' (zo) = f" (zo) = ... = 1(1-1) (zo) = 0, 1(1) (zo) =l=
[l > 2, saddle point of order l - 1J, thus
has a regular branch for Iz I < R, furthermore I(z) =1= 0 for Iz I < R.
According to 135 and 138 1/(0) I = R must lie between the maximum
and the minimum of Ifez) I on Iz I = R. We have for Iz I = R [5J
n __________________
I/(z)I=V!z-zlllz-z21"""lz- znl·
The only exception occurs in the case fez) = const., i.e. if
zl = Z2 = ".. = zn = O.
328 Functions of One Complex Variable
1 1 = 1 1 R-5 1
> If + 4R3 - 1: -R,,"':i =
k~3
If + 4R3(R -- 1) > R •
142. If I(z) oF 0 everywhere inside the level line the absolute value
I/(z) 1 attains its maximum and its minimum on the boundary according
to 135 and 138. This implies that I/(z) 1 must be constant in the interior,
Pt. III, Solutions 140-146 329
*'
144. The theorem is not valid for I(z) = const. Thus we may assume
j(zo) O. If there is a saddle point on the circle IZ I = r the projection
of at least one of the sectors with points above the saddle point mentioned
in 132 protrudes into the interior of IZ I < r. Therefore Zo cannot be a
saddle point, i.e. t:(zo) O. *'
We put j(z) = wand consider the image of the circle Iz I = r in the
*'
w-plane. The point farthest from w = 0 is Wo = j(zo), the curve has a
definite tangent at Wo because t' (zo) O. The tangent, that is the vector
izol'(zo) [103], is perpendicular to the vector Wo = j(zo) (obvious for
geometric reasons). Hence iZf~~;o) is purely imaginary. In the neigh-
bourhood of Wo the side of the image curve turned towards the origin
corresponds. according to the hypothesis, to the side of the circl~ turned
towards the origin. Therefore iZit::~o) must be positive imaginary.
145. [ef. A. Pringsheim: Sber. bayer. Akad. Wiss. 1920, p.145;
1921. p. 255.J
n a(w. _ 00.- 1 ) n wi _ w- t 3t
2 J; 1 = 2 J; t t = 2ni tan - --? 2:ni .
• =1 a(w· + 00·- J 0=1 00 + 00- n
IZI - I + /Z2 -
Zo zll +I Z3 - z21 + ... + IZn - zn_ll
is the length of an inscribed polygon, thus < l. The points Zl' Z2' ... , z'"
can be chosen so that for a given 15, 15 > 0, and sufficiently large n.
330 Functions of One Complex Variable
consequently
J I' J
2 n n
4i
~ sf}1'n-2"f}' =
:;:-2 +x._d-
L
ix df}
sin 2 f} ~ =
df}
sin f}
A:. ~2 +-2dz
-ix = J' 2z;;-1 '
o -n -n
with z = eifJ • The integral is taken along the circle z = 1 which includes 1 1
the integral is taken along the circle = 1. If 0, -1 < r < 3'1 '
I: - 11 >
1 Z 1 r ~
2'"
J xs-1e- x dx + wSi f
ins w
J x'-le- ix dx =
co
eis{}-wei {} dO. - e2 O.
o 0 0
Let w converge to + 00. Then the first integral converges to r(s) and
the third to the integral in question. The modulus of the second integral is
= - x n sin x dx.
n 0
o
This integral converges for -1 < ffts < 1 and the right hand side is
regular for the same s values. Hence the formula holds for -1 < ffts < 1.
332 Functions of One Complex Variable
o
J exp ( - nf~--=-Iog x) sin ((~~g~ 2
(log:r)2 +;n2 (log X)2 + n
+!!) xndx = 0, n= 0, 1, 2, ...
154.
+~ += +=
XI'+l J e-tt-' cos xt dt = xl' J sin xt . /If'-le-tl' dt = 0' J eizP-~z dz,
o 0 0
j"
a+iT
as
1
-2' e
-2 ds, T>a,
;n! s
a-iT
r
II (t)
t
= n!
--
n
for n < t :::;; n + 1,
thus (the interchange of summation and integration can be justified
by various arguments)
u ~
_ ~ JOO]OO sin~e2 e
-IA
e
-itu
~
00 n inu
'>' _t_e_
nl
dt d
u
u=-oo 1=0 n=O
= ~
2m
JJ
+00 +00 .
e'u_ 1 et(-A-iu+eiU )
u
dt du
-00 0
=-. 1
2m -00
J'
+00
u(J.
.
e'u - 1
+
.
iu - e'u)
duo
This integral is equal to the sum of the residues in the upper half-plane
iu - 1 .
because the integral of . along the half-circle u = re,fJ,
u(J. + iu - e'U)
o < f} < n, converges to 0 as r tends to infinity. The only pole in the
upper half-plane is u = iz [196], the corresponding residue is.!... (Cf.
z
215, IV 55.)
157. [Dirichlet: J. reine angew. Math. Vol. 17, p. 35 (1837); Mehler:
Math. Ann. Vol. 5, p. 141 (1872).] Assume -1 < x < 1, x = cos f},
0< f} < n. Then [cf. G. Szeg6: Orthogonal Pelynomials, ;Srd Ed. 1967,
pp.86-90]
Pn(cosf}) = -2·
1 f V1-2zcosff+z2
zn
dz,
m
where the (positively oriented) path of integration encloses the two
singular points eifJ and e- ifJ . We may contract the path to a straight line
334 Functions of One Complex Variable
segment from ei & to e- i & (Laplace's formula) or to the arc -{} < arg z :S {}
of the unit circle (Dirichlet-Mehler formula). (In both cases we should
proceed carefully because the integrand becomes infinite at the endpoints,
although only of the order t.) When z goes around the singular point
the integrand changes only its sign. Thus
j.
1
P (cos {}) = ~ (cos {} + iIX sin{}t i sin {} dlX
n 2m. V1 - 2 (cos {} +
iIX sin {}) cos {} (cos {} + +
iIX sin {})2
J
-J
& . .
2 e,nt . ie,t dt
= 2ni _& V1 _ 2e it cos {} + iit .
entire plane.
159. The integral along L = Lo has the same value as the integral
along La' iX ~ O. Since eez + in = eez-in = e-ex the contributions of the
horizontal parts of La cancel each other. The vertical segment of La
supplies
n .
2n f ee"+'Y dy.
This value is independent of iX, hence =1, as can be seen for iX -+ - 00.
E(z) = - ~
z
+~ ~
2nz
Z2
f(-C i
+~)
r - z
;
ee' dC.
Evaluate the integral not along L but along L', an inner parallel curve
to L at the distance 0 (boundary of the region lRz > 0, -n + 0 < 3z < n-o)
whereby 0 < 0 < ; . Note that the real integral
J ee-e$cos6 df;
o
converges.
Pt. III, Solutions 158-165 335
(2) Let z be in the rectangle -1 < ffiz < 0, -n < 3z < n. Accord-
ing to the residue theorem we have
_e-e'
JC-z de - J_e_.
.c
C-z
de = 2nie ez .
L L_l
Consequently
E(z) = eeZ - -1 1 J( -.,r + --
+ -1 -. C 2
)
eeC dr.,
z 2m Z2 C- Z
L_l
= J'" IIl2ne'
-1<
22ncosl}
----~d{}~-
n'l}
2n
n!
-vi
[II 217].
v~l
= 2~i z= reil}.
Izl~r 0 0
d(G(Z)) i; e{-l)nGcn+/)_~)
- dz cos ez = -:=_= (ez - (n + t) n)2 •
~( G(z) ) _
dz cos ez -n=o
(-lr Gi; (n +ei)~) ( e (!)
[ez - (n + i) n]2 + [ez + (n + i) n]2
and integrate
G(z) =
cos ez
_ i; -1
n=O ( )
nG (n +e1) n) (ez 1 1)
- (n + 1) n + ez + (n + t) n .
1/ f(x) dx I< ~ .
If k is an integer we replace fez) by zkf(z); if k is not an integer the second
formula in 167 is used.
Pt. III, Solutions 166-171 337
and
1 1 n n 1 1 n n xx
JJl'f(x) dx = 2n
o
~ ~ x;.x,. J ~+"+"-1 dx
;'=1 ,.=1 0
= 2n ~ ~;. + ;. ~
;'=1 ,.=1 II-
IX •
fn(z) = -2
1 . J; ;n{C) dC.
mJ'~-z
L
Therefore [po 113] Fr(z) is analytic as well as r- 2 F r(z) and finally [170J
Instead of all the circles we could consider all the curves that are similar
to a given closed curve without double points and similarly situated.
172. To show that the difference
2n 2n
f f(e ifJ ) df} - 2nf(O) = f [f(e ifJ ) - f(re ifJ )] df}, 0 S r< 1 [118J
o 0
vanishes we prove that this expression can be made arbitrarily small with
1 - r. We take each point of discontinuity as the center of an open disk
of radius E so that different disks have no common point. Removing these
"small" disks from the unit disk we obtain a domain in which f(z) is
uniformly continuous. vVe now split the integral into two parts: The first
part consists of the integral along 'the arcs inside the above mentioned
disks, the second part consists of the integral along the remaining piece
of the circle. The first part can be made arbitrarily small with E [f(z)
is bounded] and then, once E is fixed, the second part can be made
arbitrarily small when 1 - r is sufficiently small.
173. Cf. 174. Cf. also 231, Hurwitz-Courant, p. 327, and E. Hille,
Vol. II, p. 361.
174. It is convenient to consider the following general situation:
Let 'Il be a simply connected domain in the C-plane and assume that
the mapping 7p(C) = Z is conformal in the interior of 'Il, sufficiently
continuous on the boundary and that it establishes a one to one rela-
tionship between 'Il and the disk IZ I < 1; let the point C = z correspond
to Z = O. We have C = 7p-l(Z) where 7p-l denotes the inverse function of
7p. We "transplant" the function ftC), which is regular in 'Il, from the C-
plane to the Z-plane by defining
the integral is taken along the positively oriented boundary of '1). 173
and 174 are particular cases:
!p'(C) dC (R2 - y2) de
'1): \C\<R, hP(Cr- R2 _ 2Rr cos (13 -_-c{}0c")-+---;;Cy2 '
C= i'fj, z = x + iy.
175. [J. L. W. V. Jensen: Acta Math. Vol. 22, pp. 359-364 (1899).
°
Cf. E. Hille, Vol. II, pp.189-190.J The integrand is a single valued
function on me; me contains the point r = if e is sufficiently small
(hypothesis) and then the value of the integral is 2ni log 1(0). We denote
by iX)' iX2, ... , iXm , f3i' ... , f3n the endpoints of the paths that connect the
e-circles around aI' a2, ... , am' b), ... , bn with the unit circle \z \ = 1
(\iX) \ = ... = \iXm\ = \f31\ = .,. = \f3n \ = 1). The loop starting at iXw
turning around a" along the e-circle and ending at iX" contributes, as
e --+ 0,
°
(1922).J Same notation as in 174. Consider a function I(C) that is mero-
morphic in '1), and different from and (X) on the boundary of '1) and at
the inner point z; inside '1) the zeros are a» a2, ... , am and the poles
340 Functions of One Complex Variable
i7log
,,~l
rTa--)1 -
,1p",
i:
v~ 1
log -I
1p
-(h-, = if> log [f(C) [2~;r;;(~)- -
v 1p -
log [f(z) [,
the integral taken along the boundary of '1). The signs on the left hand
side are in evidence: [1p(a) [ < 1, [1p(b v) [ < 1; 2~;~~)- is positive. Suppose
that ftC) is variable and [f(z) [ is fixed; then the essence of the formula
can be roughly expressed as follows: zeros in the interior increase,
poles in the interior decrease the moduli of the boundary values.
176, 177 are a consequence of Jensen's general formula as 173 and
174 follow from the generalized formula given in 174.-Also the proof
of 120 can be suitably generalized [174}. -The condition that fez) be
different from 0 on the boundary can always be weakened [120J, that
j(z) be regular on the boundary can be weakened in many cases.
178. [Cf. F. Nevanlinna: C. R. Acad. Sci. (Paris) Ser. A-B, Vol. 175,
p. 676 (1922); T. Carleman: Ark. Mat., Astron. Fys. Vol. 17, No.9, p. 5
(1923).J Put a circle of radius s, s sufficiently small, around a.u and
connect the s-circle with the half-circle [z [ = R, - ; ;:::;: arg z ~ ;
so that the connecting paths do not intersect. Removal of the s-circles
and the connecting paths reduces '1) to a simply connected region ffiE ;
its boundary is the appropriate path of integration. The loop around
a.u starting and ending at Z,u' [z" I = R, contributes
"
+ --
2 . 2 cos {}
_ { log f(Re"'}) - R - df} + 1+)-R) (-
(r 1
y2
1)
+ R2 log f(ty) dy
.
The formula is so simple because the differential (:2 + ~2) diZ is always
real on the half-circle Iz I = R, l)tz > 0 and the function ~ - ;2 is
purely imaginary.
179. We introduce ao = lao I eiy =f= 0, z - z. = r.eif{!v, thus
shows that w describes the right half of the ellipse centred at the origin
and with semi-axes e7lX - e- 7lX and e7lX + e~nx as z moves on the line
alz = x, -1 < Jz < t·
Therefore, if Wo is an arbitrary point in the right half-plane \Rw > 0,
x can be chosen so large that the following situation is met: As z moves
on the boundary of the rectangular region
means of w =
o
can be obtained by rotating through 45°, and reflecting and rotating,
respectively, the right half of the Cornu spiral, described by Sommerfeld,
Pt. III, Solutions 185-192 343
l.c. The image of the bounding circular arc is near the point w =1/ ~
[IV 189J. We choose r so small that the image of the arc z = re iIJ ,
- : <{} < : ' intersects the images of the radii only at the points that
.7t .:n:
1.- -1.-
correspond to z = re 4 and z = re 4. We now consider the image of
the boundary of the above mentioned sector outside the circle Iz I = r.
It has the winding number O. Hence w =F 0 when - 4n < arg z <2Jl .
Different proof in V 178.
190. The integral
2:i f ~~t!~
is an integer. Therefore it cannot change if the path of integration is
deformed continuously. The curves in question can be transformed into
each other by a continuous deformation.
191. Let j(z) =1= 0, then j(z) =F 0 on L. The function
log j(z) = log R + if) is regula at every point of L. Now [clear when OV
and os coincide with ox and oy, respectively, d. p. 113]
8 log R 8e
8v as'
where 1;; denotes the differentiation in the direction of the outer normal
and ~ in the direction of the positive tangent at the point z of L. The
derivative on the left hand side is positive [maximum principleJ hence
the derivative on the right hand side has to be positive too. The image
of L is a circle (possibly described several times in the same sense).
192. [B. Riemann: Werke. Leipzig: B. G. Teubner 1876, pp.106-
107; H. M. Macdonald: Proc. Lond. Math. Soc. Vol. 29, pp.576-577
(1898); d. also G. N. Watson: Proc. Lond. Math. Soc. Ser. 2, Vol. 15,
pp. 227-242 (1916); Whittaker and Watson, p. 121.J We assume that
j'(z) = Rieie dd~ =F 0 on L (same notation as in 191). Let the winding
numbers of the curves described by j(z) and j' (z) be denoted by Wand
W' resp. (W is the winding number of a circle which may be described
several times in the same sense). Thus 2n(W' - W) is equal to the change
de
of the argument of dz as z describes the curve L. Let ds = Idz I be the
line element of L, ~~ = ~~ ~.. The first factor is always real, therefore
only the change of the argument of the second factor is important. The
argument of ~; = I~; I is equal to the negative value of the argument of
344 Functions of One Complex Variable
dz. The change in direction of dz, i.e. the change in direction of the tangent
vector along a closed curve without double points, is 2n; this can be
easily seen in the case of a polygon. Hence W' - W = -1. If f' (z)
vanishes on L we consider a level line inside but sufficiently close to L.
The geometric formulation of the proposition can also be verified directly;
d. H. M. Macdonald, l.c., or MPR, Vol. 1, pp. 163-164.
193. The function fez) is not a constant because f' (z) =1= 0; fez) has
exactly one zero in the interior of 'Il [192J. Therefore w = fez) describes
a (circular) curve with the winding number 1 as z describes the boundary
of 'Il. The winding number of the path described by fez) - Wo is 1 or 0
depending on whether IWo I < or IWo I > the constant modulus of fez)
on the boundary of 'Il.
194. [E. Rouche: J. de l'Ec. Pol. Vol. 39, p. 217 (1862).] We may
assume that fez) and qJ(z) are regular on L because fez) and fez) + qJ(z)
are different from 0 and If(z) I > Icp(z) I sufficiently close to L. The func-
tion 1 + ~~;~ has on L a positive real part and so, as z describes L,
the change of its argument is equal to O. Furthermore fez) + qJ(z) =
fez) (1 + ~~;~) thus [181 Jthe image of Lunder fez} has the same winding
number as the one under fez) + qJ(z).
195. Special case of 194: fez) = zi-', qJ(z) = -1, L is the unit circle.
Since zei.-· increases with z on 0 < z < 1 from 0 to ei.-l > 1 there is a
root on this segment.
196. If z = iy, y real, we have lit - iYI > it > 1 = le- iY I. If Izl is
sufficiently large, ffiz > 0, we have lit - z I > 1 > Ie-' I. Moreover
it - z = 0 has the root z = it in the right half-plane. Special case of 194:
fez) = it - z, qJ(z) = -e-', L a sufficiently large half-circle in ffiz > 0
and its closing diameter. The only root is real because the modular graph
is symmetric with respect to the vertical plane through the real axis.
197. [Cf. G. Julia: J. Math. Pures Appl. Ser. 8, Vol. 1, p. 63 (1918).]
Special case of 194: on the unit circle Iz I = 1 > Ifez) I.
198. [Example of a general theorem of G. Julia: Ann. Sci. Ecole
Norm. Sup. (Paris) Ser. 3, Vol. 36, pp. 104-108 (1919).J Let Rn be the
rectangle with the four corners n ± I ± id, n integer, d fixed, d> 0
and "'Tite z = x +iy = reifJ. Because of 1155 we have on the boundary
of Rn as n_oo
log IF(z) I C-.:> (x - !) log r - yO - x,
so that the minimum of IF(z) I on Rn tends to 00 as n - 00. On the other
hand we find Isin nz I > c on the boundary of R,. where c is independent
Pt. III, Solutions 193-200 345
of nand c > O. Hence the minimum of
~1 1= Isin nZilr(l _
1 r(z)j z) I
n
converges on the boundary of R_ n to + ex:> as n --+ ex:>. Thus we have for
arbitrary a and sufficiently large n
o - 11'(1)
i() /) cos z converges to 0 as z --+ ex:>. [To begin with prove this for
the strip -n < ffiz < n.J With the exception of finitely many zeros
zF(z) has, therefore, in any disk the same number of zeros as the function
1(0) - 1(1) cos z [194J, that is 1. This zero is necessarily real in the case
1/(1) I > 1/(0) I where the disk is cut in half by the real axis; the non-real
zeros of the functions that assume real values for real z appear in pairs
[solution 196J.
200. The term zna-n' assumes its role as maximum term of the
series
1 +~ +~.~3 +~.~.~
a a a
3 a a a·
+ ...
on the circle Iz I = Ia 2n - 1 and abandons it on the circle Iz I = Ia 2n +1
1 1
n
On the circle \z \= \a \2 the corresponding terms of the two subseries on
the right hand side have the same absolute value, thus
I!'(~)_ ~~~nr:-n~1 < 2 (_~ +~. _1_ +~. _1_. _~_ + ...)
i zna~n' I lal !al lal 3 lal lal 3 lal"
2 1 2 Ia 12
< r;r G~f·3 = lal 3 - 1 < 1,
because the only positive root of the equation Z3 - 2Z2 - 1 = 0 is
smaller than 2.5 [16]. Hence F(z) has in the disk IZ \ < \a 2n the same 1
number of zeros as zna-n', namely n. The disk Izl < lal 2n - 2 contains, by
virtue of the same proof, n - 1 zeros. -Cf. V 176.
201. [A. Hurwitz: Math. Ann. Vol. 33, pp.246-266 (1889).J The
closed disk D has a as its center, lies completely in ffi and contains no
other zeros of I(z) than possibly a. We have \/(z) \ > Iln(z) -/(z) Ion the
boundary of D when n is sufficiently large. Apply 194, In(z) - I(z) = q;(z).
More generally: Each subdomain of ffi on the boundary of which there
are no zeros of I(z) contains exactly the same number of zeros of In (z)
as of I(z) if n is sufficiently large. Important for the applications!
202. The limit function is regular in the unit disk \z \ < 1 [170J.
Assume that I(Zl) = I(Z2) for Zl =l= Z2' \Zl\ < 1, \Z2\ < 1; consider the
sequence In(z) -In(zl)' n = 1, 2, ... which converges to I(z) -/(Zl).
In a disk that has its center at z2' lies completely inside the unit circle
and does not contain zl> In(z) -In(zj) would have to vanish for sufficiently
large n [201 J: contradiction.
203. [170,201.J
204. In the case where a and d are integers the proposition is proved
in the same way as proposition 185 because the zeros of the polynomial
aoza +
a1za+d +a2za+2d + ... +
anza+nd lie in the disk \z \ <1 [23].
If a and d are rational z has to be replaced by a suitable multiple of z.
If a and d are irrational approximate these two constants by rational
numbers and apply 203.
205. [G. P6lya: Math. Z. Vol. 2, p. 354 (1918).J We have [II 21J
o
JI(t) cos zt dt =
1
lim 2:
n-}
n->oov=l
1 ' )
-I (~ cos ~ Z
n \n n
[185,203J.
206. Counter-example
1
In(z) = Z2 +n' n = 1, 2, 3, ... ; 'Il: \z\ < 2; a = -1, b= +1.
206.1. [G. P6lya, Problem: Jber. deutsch. Math. Verein. Vol. 34,
2. Abt., p. 97 (1925). Solved by R. Jungen: Jber. deutsch. Math. Verein.
Pt. III, Solutions 201-208 347
i.e. the formula to be proved for the function j'(z) rp(z). The family of
admissiblefunctions/(z) is identical with the family of functions j'(z) rp(z),
where j' (z) is the delivative of an admissible function, because rp(O) =F O.
Thus 207 leads to Lagrange's formula (L), p. 145, by integration.
208.
integrated along a circle around the center C= 0 and for w so small that
1C1 > 1wrp(C) 1 along the path of integration. Then the path of integration
encloses the same number of zeros of C - wrp(C) as of C [194J, i.e. exactly
one. Denoting this single zero by z we further find
_1 J. f(~) d~ ~_
2ni 'Y ~ - wtp(~) 1 - wtp'(z) •
348 Functions of One Complex Variable
(1- V~)'"
2w
= 1 + IX;;
n=1
(ex + 2n -
n - 1
1) wn n
;
(1V;:;-~2
1 +: + ~
)2'" = (ex + ; - 1)
= 1 + IXn~' 1 :. n -
n
n=On
;; J n+1
tn, e- lt dt =
n.
i: J
n=OO
1
(n +,ex)n e-).(n+",) dIX,
'If'
Pt. III, Solutions 209-219 349
i(n+~t e-An=~
n=O
n! 1 - z'
is uniform for 0 < eX < 1, and z is given by the equation ze- z = e-\
Iz I < 1 [214, 209J. Thus the integral in question becomes
=f 1
e"(Z-A) deX
1 - z
= e Z- A -
(1 - z) (z - J.)
1 = _1_.
J. - z
o
We verify immediately that C=A - zsatisfies the equation A-C-e-c=O;
A - z is real and positive.
216. [Regarding 216-218, 225, 226, d. G. P6lya: Enseignement
Math. Vol. 22, pp. 38-47 (1922).J We put !p(z) = (1 + z)fJ, I(z) = (1 + z)"
and apply 207:
= (~ + fJn) n (1 +
z)" x"'+l
n~ n W = 1 _ wfJ.(l +
z)fJ- 1 = (1 - fJ.) x + fJ. '
where 1 + z = x and x is the root of the equation 211 which is algebraic
for rational (J.
217. [Cf. L. Euler: Opuscula analytica, Vol.l. Petropoli 1783,
pp. 48-62.J We introduce !p(z) = 1 + z + Z2, I(z) = 1 and apply 207:
1 - w - Vi - 2w - 3w2
Z = . 2w '
218. The sum of the terms in the k-th column to the left of the middle
column is [216J
=
On the right hand side we find (1 - ~)'" (1 + ~)fJ 2: p::,fJ)(~) w", whereas
w = q;fZ) implies
Z=-
1 - ~w - Vi-=- 2~w + w 2
2"+fJ /
V1 - - (l-w+ll-2~w+w2)-a(l+w+VI-2~w+w2)-fJ.
2~w +w 2
~w ¢" --"---
z=1_w' J(z)=---"e w - 1 , l-wrp'(z)=I-w,
(1 - w)
therefore
= ~w
2: LI")(~) wn = ~_1~ ew - 1 .
n~O n (1 _ w)"H
+ 1)
+ z(z - 1) ... (z - n
n! . -w
n
+ ... ;
where w = e S
- 1, valid for I eS - 11 < 1, z arbitrary.
(2) s = eS - 1 - ~ (e S
2
- 1)2 + ~3 (e S _ 1)3 - ...
+ (_I)"-l~(eS - I t
n
+ "', les -11 < 1.
hence the series converges whenever Ises +1 I < 1; at any rate the formula
holds in a neighbourhood of s = o.
(4) in 212 put: x = eS • (3 = t. thus w = eS / z - e- s/z , iX = z,
esz = 1 + 17 : (z :
n~l
~ ~ 1) (e~ _e- f r
Replace s by -s and add the two formulas:
(3 -.1
- 2' W -- est'!. - e- s/2 ,iX-
- Z - .1.
2'
~-
2
e- sz = ;, ~ (z + m
..:.., 2 2m -
-
1
1) (
S
e2 _ e 2S)2m-Z (eS _ e-S).
m~l
The sum of the last two formulas produces the desired expansion of eSz •
We remember Stirling's formula and whte sin nz as an infinite product
to obtain
3_ (z + m - 1) ( e~2
S S
- "2
)2m zsin nz (.sm-is2 )2m m - 2
3
.
2m 2m - 1 .
- e '" - -V--:;;-
,.
,
this shows convergence for I sin i; I < 1; the formula is certainly correct
in a neighbourhood of s = o.
221. [With respect to the formulas (2) and (3) d. N. H. Abel:
Oeuvres, Vol. 2, Nouvelle edition. Christiania: Gmndahl & Son 1881,
pp. 72-73.J Expand in 220, F(z) = eSz , both sides in ascending powers
of s and compare the coefficients of :! . k
We have
222. It is sufficient to prove (1) and (2) for F(z) = (z - W)-l; for
other rational functions differentiate with respect to w, split them
up into partial fractions and apply 221. The formulas (1) and (2) hold
for F(z) = esz if s is real and negative. [Solution 220.J Multiplication
with e- sw ds and integration over - 00 < s < 0 yield (1) and (2) for
352 Functions of One Complex Variable
(1)
_1_ _ _ ~ _ i;
z(z - 1) ... (z - nL __
z - w - W n=O w(w - 1) ... (w - n - 1) ,
1 = n!
(2) -- (z - W)2 = -n~ (z - w) (z - w + 1) ... (~ - w + n :+-1)'
We obtain (2) by replacing in (1) wand z by w - z - 1 and -1 respec-
tively. Formula (1) is derived by a limit process from the identity
1 1 z z(z - 1)
w -- ~ = -;;; + ;'-C;;;=1) + w(w - 1) (w - -2) + ...
z(z - 1) ... (z - n + 1) z z - 1 z - n 1
+ w(w - 1) (w - 2) •.• (w - n) + -;;; ;;-=--{ •.. w-~~ w--~ z
which is easily proved by complete induction. If wand z are different
from one another and from 0, 1, 2, 3, ... the remainder is [d. II 31J
(-z)(-z + 1) ... (-z + n) n-Wn! n
W--i;
+ 1) ... (-w +
- - -
-w(-w n) w-z
r(-w) n W - z
('0----
F( -z) w - z·
The formulas (3) and (4) cannot be valid in a non-empty region for
non-entire rational functions: Otherwise the partial sums of order nand
2n resp. would converge uniformly in any finite domain according to 255
and 254, thus F(z) would be regular everywhere: contradiction.
223. The identity in question is purely formal; it represents the
combination of the infinitely many equations that define the quantities
Lln ak , k = 0, ±1, ±2, ...
224. Since the expansion of --'- does not contain a constant term
1 +t
the coefficient tn in the expansion of 1 ~ t F(itt) depends on ao' aI' ... , an
only. Therefore we can restrict ourselves to the case where F(z) is a poly-
nomial. Every polynomial however can be written as a linear combination
of the special polynomials (1 - z)m, m = 0, 1, 2, ... ; besides we have for
two sequences ak and bk and two constants c1 and c2 the linear relation
Lln(c 1 a k + c 2 bk ) = c1 Lln ak + c2 Llnbk • Consequently it is sufficient to prove
the statement for F(z) = (1 - z)m, m = 0, 1, 2, ... In this case Llnao is
Pt. III, Solutions 223-229 353
(1 - Z)2" i
k=-oo
akt' = (1 _ z)2" F(z) +2F (Z-1) = (1 _ z)2"+m 1 + (-21)"'Z-m
(-1)"+1 ~
=
n+1
(2n + nm+ 1).
Set p = 2, IX = m, W = -t in solution 213.
*227.
sin :n;x _
:n;x (1 - x) -
iI (1 + .!...-)n (1 _ n_ +
n=l
x ) _
1 -
iI (1 + n(n + X))
,,=1
x(l -
1) •
228. [1. Schur.] By carrying out the multiplication in the last infinite
product in 227.
229. Apply (L), p. 145, to
. 1 0
f
2" {}
:J/(Re'O) = 2n U(R,8) cot -- ~ d8,
ex
d C'l!
d{}:Jt(Re iO ) = - 2n 1f 2n
U(R,8) SIn
- 0)-2 d8 < o.
(. {}-2·
"
Pt. III, Solutions 230- 237 355
2~ 2: r2n(b; + C;),
o n~1
2n
J [V(r, #)]2 d# =
~
2~ C~ + ~ 2: r2n(b; + C;).
o .. ",,1
235. [Cf. C. Caratheodory: Rend. Circ. Mat. Palermo Vol. 32,
pp. 193-217 (1911).] In the notation of 230 we have R = 1, ao = t and
an = n
nro
J U(r, #) e-,n{} d#,
1 2n .
Ian I <n
1
nro
J U(r, #) d# =
2n 1
n'
r
j(z) = ~ ~+~ = ~
2 1 - z 2
+ z + + ... + zn + ....
Z2
the entire function 2: anzn on the circle Iz I = r; then we have for r> 1
~
238. [E. Landau: Arch. Math. Phys. Ser. 3, Vol. 11, pp.32-34,
(1907); d. F. Schottky: J. reine angew. Math. Vol. 117, pp. 225-253
(1897). J It is sufficient to prove the inequality <
_;rr; Iffiall
_2 L1 (I); for R
if IX is a real constant the largest oscillation of the real part of f(i lX z) is
also L1(1) and the inequality IffieilXalIR<~L1(1)
_;rr; for all IX implies
Jail R < ! L1(1).-Let A denote the arithmetic mean of the upper and
lower bounds of ffif(z) in the disk Iz I < R. We find I)Rf(z) - A I < tL1 (I)
for Iz I < R. Besides [230J
2"
nra 1 = f [)Rf(re iO )] e- iO df}, 0 <r< R,
o
thus
LI
f f
2" 2"
nr)Ra) = [)Rf(re iO ) - A] cosf}df}, nr I)RaIl < ~) Icosf} Idf} = 2L1(1).
o 0
Let r converge to R.-For R = 1
f( z) = 2i1 I og 11 -+ zz = fZ
.
+ 3i z
3
+ 5i i' + ... ,
we have
L1(1) = ~, lall = 1.
In geometrical terms the theorem reads as follows: A disk is mapped
conformally but not necessarily univalently onto a region. The width
of this region in any direction is at least equal to -~ times the product
of the radius of the disk and the linear enlargement at the center of the
disk. In the above mentioned special case the image of the disk is the
vertical strip - ~- < )Rf(z) < ~ .
239. [E. Landau, O. Toeplitz: Arch. Math. Phys. Ser. 3, VoL 11,
pp. 302-307 (1907).J The least upper bound of If(z) - f( -z) I on
Izl < R is denoted by D*(I), then D*(I) < D(I}. Let 0 < r < R.
f
2"
4nra l = [f(re~O) - f( -reiD) J e-if} df}
o
implies 4nr Iall < D*(I) . 2n < D(I) . 2~; let r converge to R. If f(z) is
linear, f(z) = ao + alz, then D(I) = 2 Ia l I' R. The proposition admits
the following geometrical interpretation: A disk is mapped conformally
but not necessarily schlicht onto a region. The maximum distance of
two boundary points (diameter) of this region is at least equal to the
product of the diameter of the disk and the linear enlargement at the
center. In the special case mentioned the image is an open disk.
Pt. III, Solutions 238- 242 357
240. [Cf. E. Landau: Math. Z. VoL 20,pp. 99-100 (1924).J We derive
from 232 by differentiation and by setting z = 0 [117J
r
1(0) 1'=1 \CI' r. nr 0
m (1
= ~ ffi - -
(;1'
--;J -
2
-log -I I
r) + -2 M
11(0) I,·
1'=1 CI' r~ r CI' r
[l20J. According to the condition (2) we have fficl' < 0; thus for", > l
1 CI') r2_lc1'12 1
ffi ( - - - = ffi-<O.
cl' y2 r2 cl'
For '" <:::: l we have
cl'
ffi ( - - - 2 log -
y) <--
ICI'I
+ 2 log ICI'I
- < O.
r [CI'I - r r
The inequality x + 2 log x <.0 holds for 0 < x < : because the left
hand side increases with x and e2/ 3 ( : y"< 1, i.e. 8e < 27.
241. The power series in question can be written in the form
The power series P(z) 2: anzn = 2: bnzn can be made to satisfy the
n=O n=O
conditions of 242 by an appropriate choice of the polynomial
+ zq-l . Ibnl
1bn+ 1 I---+e implies
lim sup
n-+oo
VIb I n = b <-~,
(!
i
n=O
IXnZn rational with exactly k poles on the
a
lim (1 - z) (ao + a1z + ... + anzn + ...) = lim ~ = 0
z-+l -0 10-+= 1
F( )
u
= ;. -~"-
..:::,., 2 .
1O=o:n;~.
f.
a+ioo _
es{l/n-u)
S
_
+2 e-s(\"n+u) d
s.
a-loa
250. The function j(z) described in the problem has the following
properties
(1) j(z) is regular for I z I < 1 because the integral converges absolutely
when ffiz < 1; we have
f
= n
an = e~(x+x/LcOS/L")sin (x/L sin,un) -~dx, n = 0,1,2, ... ;
o .
(2) an cannot vanish for n = 0, 1, 2, ... , because
Ie~(x+x/LCOS/L") sin (x/L sin ,un) I < e- x [solution 153J;
(3) for Iz I < 1, the derivatives are
given in solution 153, instead of e~xltcos/Ln sin (xl' sin ,un) shows in a
log Ian! loglanl
similar way that in 249 -V-- cannot be replaced even by (logn) 2 -~/ •
n In
251. [G. P6lya, Problem: Arch. Math. Phys. Ser. 3, Vol. 25, p. 337
(1917). Solved by H. Priifer, K. Scholl: Arch. Math. Phys. Ser. 3, Vol. 28,
p. 177 (1920).J We assume
Pt. III, Solutions 249-254 361
and that the series in question converges for Z= a, i.e. that
Co + c1 + c2 + '" + cn + ... converges, thus
Icm+k + cm+k+1 + ... + cm+k+n I < e for m sufficiently large, k,
n = 0, 1, 2, 3, ... Then
(A + t+ 1 (A + )n+2
«A+et+ 1~ IZ-Zol+ 2~ IZ- ZoI2+ ...
= (A + et e(A+e)lz-zol.
n __
Hence lim sup VI g(n) (z) I < A = lim sup Vfg(n) (zo) I.
n~~
where Yn and (In are constants. Let a and b be the two points of conver-
gence. The two series 1: An and 1: En' (Yn a + (In) Pn(a) = An'
n=O n=O
362 Functions of One Complex Variable
;,IPn(z) _ Pn_l(Z~II_
::lI Pn(iX)
i
IIPn- 1 (Z) ~2=_-iXn221'
Pn _ 1(iX) - n=] Pn- 1(iX) ~
n. n=O
product can be expanded into a power series:
( 1- na)-n+1 (Z)n-1
1- n = ea - z (A' A" + ... ),
1+ n + 112
A', A", ... depend on a and Z but not on n. The series with the general
(1 - = n~:-:A:) + ...
converges absolutely, therefore
Co 1: an aZ ( 1 - na)-n+] ( 1 - nz)n-1 .
+ n=l
00
g,,(z) = e-u,,(x,y) -iv" (x,y) , Z = x + iy, and I,,(z) = go(z) g1 (z) ••• g,,(z) .
implies
1 + i
,,=1
Oi(Oi + ~)"-l w"
n.
364 Functions of One Complex Variable
has the radius of convergence e-1 [214J; therefore the series in question
converges in any connected region of the x-plane where Ixe- X I < e- 1
and there it represents an analytic function. The interval 0 :S x < 1 as
well as the interval 1 < x < 00 can be imbedded in such a region, one
in m!> the other in m2 , but the intervals cannot both be imbedded in the
same region. According to 210 the infinite sum is equal to eO'" if x is
sufficiently small, thus it is equal to eIXx for x in Suppose that mI'
1 < x < 00 and that x' is defined by xe- X = x' e- x ', 0 < x' < 1. The
series in question stays the same, therefore its sum is e!Xx' =1= e!XX. - The
series converges also for x = 1 [220, (3)], its sum is eO< according to Abel's
theorem [I 86J.
261.
. . V#1 ( [ : ] - :) ( : dx
11m In(z) = lim - - - - - - - - 1 = 1 -- - -- +1
n-+= n->-= .;,
.:..,- (V )Z-1 -1 J x =-1 dx
v~l n n 0
+ 1)-1 C(z + 1)
= z(z [II 45].
lim In(z) = 1.
n .... =
262. [G. P6lya, Problem: Arch. Math. Phys. Ser. B, Vol. 25, p. 337
(1917). Solved by H. Prtifer: Arch. Math. Phys. Ser. 3, Vol. 28, pp. 179-
180 (1920).J We put
z - 1 cp(z) - 1
;+1 =C, ----=1IJ(C)
cpt:;) +1 ' '
Pt. III. Solutions 261-263 365
then
q:>[q:>(z)1 - 1 q:>{q:>[q:>(z)]} - 1 { }
q:>[q:>(z)]+ 1 = 1I'[1I'(C)J, q:>{q:>[q:>(z)J} + 1 = "P "P["P(C)J , •.. ,
'+IX-fJ
"P(C) = C1 + (IX - fJ) ,
and the statement becomes: the sequence
"P(C) , "P["P(C)J, "P{"P["P(C)J}, ...
°
converges to if ICI < 1 and converges to CX) if ICI > 1, converges to 1
or diverges if ICI = 1.
Let ICI = r, r < 1 and denote by M(r) the maximum of
I 1 '+IX-fJ
+ (;X-=7ilt
I
for \C\ < r; we find M(r) < 1 [5J and monotone increasing with r [267].
The inequalities
\"P(C) I < rM(r) , I"P["P(C)J \ < I"P(C) I M[rM(r)J < r[M(r)J2,
I"P{"P["P(C)]} I < \1I'["P(C)J \ M{r[M(r)J2} < r[M(r)]3,
follow from the definition of "P(C). We reason analogously if ICI > 1.
Ass ume finally ICI = 1. Then
l1l'(C) \ = \"P[1I'(C) \ = \"P{"P["P(C)]} I = ... = 1. If the sequence in question
converges to a limit point Co. ICo I = 1, we have
r
<'0 = Co 1 +
'0
+ fJ
fJ) '0' I.e. Co = l.
0; -
(0; _
Z = iy. y> V-
0, \ iy Pn(iy) 1- =
0
Y 1
(
+ y2)
T
(1 + y2) ( y2) <-i;-
'4 ... 1 +
sininy
n2
2n
For fixed Z we have
p~ (-It Pn(z) n Z
= F(ll~ z)'
Consequently the sequence is bounded. e.g. in the half-disk ffiz > 0,
Iz I ::::; 1. In the crescent
Iz - n I :2: n, Iz - n - 11 < n +1
the absolute value of Pn(z) is larger than the moduli of
Pn_ 1 (z), P n_ 2 (z), ... , Pn+1(z), ... [12J. On the inside border of the
366 Functions of One Complex Variable
P (z)
n -
Vzr(z) I-I y';- z'-i e- z Ie'P(z,n)
r(z - n) r(n + 1) I -I (z _ n)z-n- t e-z+nnnH e- n 1
= 1
(
z n
Z _ )ZIIZ-n-
- n In+ t e'P (z,n) = I . i'l" 1 e'P(z,n ) ,
(2 cos cpe-''I'),
where 1p(z, n) remains bounded for all pairs of values z, n. [For a proof of
Stirling's formula to the extent as it is used here d. e.g. Whittaker and
Watson, pp. 248-253.] On the outside border of the crescent the same
estimate holds for Pn+1 (z) and Pn(z) because 1 P n +! (z) 1 = I Pn(z) I on
Iz - n - 11 = n + 1. The modulus of Pn(z) increases as z moves from
the inner to the outer border of the crescent on a ray from the origin with
slope cp, 0 < cp < ; , because all the factors (z - 1), (z - 2), ... , (z - n)
are increasing as can be seen by a geometric consideration.
264. [Cf. N. E. N6rlund, I.c. 254, p. 214.] On the inside border of the
crescent cut out by two lemniscates
2n2 cos 2cp < r2 < 2(n + 1)2 cos 2cp
but in the exterior of the unit circle we have the relation
1 2
+ 2e cos cp + 1) + e2 e+ +ecosrp
2e cos-q, +1
2
- "2 log (e
=ffi(log~ + 1_:::_
1 -)= 0,
(;+1 (;+1
Pt. III, Solutions 264- 271 367
= I1 -w_1 at(1 -
wi
w).
and M (r) is the maximum of Ij(C-l) I on the circle Ie I =...!.. [266, 267J.
r
269. The functionf(z) is regular in the "punctured" plane Iz I > 0,
zn
including the point z = 00 [268].
270. [So Bernstein: Communic. Soc. Math. Charkow, Ser. 2, Vol. 14
(1914); M. Riesz: Acta Math. Vol. 40, p.337 (1916).J Apply 268 to
e- 1(' +2 '-') [79J. We find for e =
n r, r > 1, r = a + b,
272. We may assume without loss of generality that 1(0) is real and
1(0) > 0; put I(z) = I(e/fj) = U(e, 0) + iV(e, 0), then
The upper bound for 1(0) is reached if V(e, 0) = 0 for 0 ::;: 0 < 2n, hence
I(z) - 1(0) [230J.
273. [l34.J
274. The function I(z) = :~:~ is regular for Iz I < 1. It is also
regular on the unit circle, where I/(z) I = 1 unless 1p(z) = O. If Zo is a zero
of 1p(z) it is also a zero of !p(z) and with the same multiplicity [otherwise
z = Zo were a zero or a pole of I(z) which is impossible: at other points
of the unit circle, arbitrarily close to zo, we have I/(z) 1= 1J. We drop
the common factors of !p(z) and 1p(z) and so obtain the regular function
I(z) which is different from zero in the closed disk Iz I < 1 and whose
modulus is equal to 1 on the unit circle, I/(z) I = 1 for Iz I = 1, therefore
[138J I(z) c, Ic I = 1. Since !p(0) and 1p(0) are real and positive we have
c = 1.
275. The absolute value I/(z) I is a real continuous function in 'Il;
it assumes therefore its maximum in 'Il. This is impossible at an inner
point if I(z) is not a constant [134].
276. [ef. E. Lindelof: Acta Soc. Sc. Fennicae, Vol. 46, No.4, p. 6
(1915).J A rotation through 2:rv around a point' maps the domain onto
n
the domain 'Il, and the set 58 onto the set 58" 'J! = 0, 1, 2, ... , n - 1,
580 = 58, 'Ilo = 'Il. The intersection ~ (largest common subdomain) of
the domains 'Ilo' 'Ill' ... , 'Iln - 1 contains' as an inner point. Those inner
points of ~ that can be connected with ; by a continuous curve in the
interior of ~ form a region ~*. The boundary of ~* consists, according
to the hypothesis and the construction, of certain points of the sets
580 , 58], ... , 5Bn _]. The absolute value of the function 1(' + (z - ') w-')
is < A at all the boundary points of ~* [275J and < a at those boundary
points of ~* that belong to 58,. The absolute value of the function
278. [Cf. E. Lindel6f, l.c. 276. J Let R denote the least upper bound
of If(z) I in lR. There exists at least one point P in lR or on the boundary
of lR so that in the intersection of lR with a sufficiently small disk around
P the least upper bound of If(z) I is equal to R.
If there is no such point P in lR then If(z) I < R in lR, but then there
exists a boundary point P with the required property. According to
condition (3) we have R < M, i.e. If(z) 1< Min lR.
In case there exists at least one point P, z = zo' of the described type,
then If(zo) I = R. Along a sufficiently small circle around Zo we have
If(z) I < R, thus, according to 134, f(z) - constant.
279. [Po Fatou: Acta Math. Vol. 30, p. 395 (1905).J Put w = e2 "i/n.
If n is sufficiently large then
the convergence is uniform in the unit disk, 0 < {) < 2n. The function
f(z) f(wz) f(w 2z) .• ·f(wn - 1 z) vanishes identically according to 278. [This
proposition is not an immediate consequence of 275.J
280. [R. A. Schwarz: Gesammelte mathematische Abhandlungen,
Vol. 2. Berlin: Springer 1890, pp. 110-111. J Apply 278 to the function
j(z) which is regular in the disk Iz I < 1.
z
281. [E. LindelOf: Acta Soc. Sc. Fennicae, Vol. 35, No.7 (1908).
Concerning problems 282-289 d. P. Koebe: Math. Z. Vol. 6, p.52
(1920), where also ample bibliography is provided.J We denote by
1,; = 1p-1(W) the inverse function of w = 1p(1,;). The function
370 Functions of One Complex Variable
satisfies the conditions of 280. Hence IF(C) I < e for ICI < e; there is
equality only if F(C) = eilXC, eX real. The inequality states that the points
F(C) lie in the disk IC I < e, that means that the points "P[F(C)] = 1[g;(C)]
lie in the domain §; z = g;(C) represents an arbitrary value in r. In the
extreme case we have "P(eiIXC) = 1[g;(C)J, i.e. I(z) = "P[eilXg;-I(z)J, where
C = g;-l(Z) is the inverse function of z = g;(C), eX is real. This is the most
general function that maps ffi univalently onto 6 and z = Zo onto
w = Wo [IV 86].
282. [C. Caratheodory: Math. Ann. Vol. 72, p.l07 (1912).J Apply
281 to the following special case: ffi: the disk Iz I < 1; 6: the disk
Iwl < 1; Zo = 0, Wo = 1(0),
The sub domain r is the disk Iz I ;£: e, r is the image of ICI ;£: e under the
function w = "P(C), thus § is also a disk. The points of § satisfy the relation
Iw - w 1= ICJJ1-=-Lwo I < e ~~
2
) [wo l2
o 11 + woC I = 1 ~ ! Wo Ie'
The inequality becomes an equality only if I(z) = "P [ei(Xg;-1 (z) J = "P(ei'z) =
=
+ WO"
eiIXz
---~ , m whIch case 11 + woe
- iIX·
z I = 1 - IWo II z I, l.e.
.
1 + woe"":;
arg z = arg Wo - eX + n.
283. Apply 281 to the following special case: ffi: disk Iz I < R;
6: half-plane ffiw < A(R); Zo = 0, Wo = 1(0), ffiwo = A(O),
C
wo+[wo~2A(R)JC
g;(C) = RC, "P(C) = - - - - 1 ~C-- = Wo + [wo+wo-
-
2A(R)] 1 ~t·
1~ e 2e
=T+e ffiwo + 1 +eA(R).
It is an equality only if I(z) = "P [ei"g;-I (z)] = 1p (e iX ~).
284. The following relation [solution 283J holds
Iwl < IWol + [2A(R) - Wo - wo] 1-~ e = M(O) +l~e [A(R) - A (0)],
which is a weaker statement than 236.
285. Apply 283 to log I(z) :
ffi log I(z) = log I/(z) I < log M(r).
Pt. III, Solutions 282- 289 371
286. Proposition 285 implies
2 1 - 1•1
Iln(z) 12 < 1/,,(0) I 1+1'1
For Izi <~ the exponent is 2 ~ ~ 1:1 > 1,consequently Iln(z) 12< IIn (0) I·
28T. Use 281: ffi: the disk Izi < 1; 6: the half-plane ffiw> 0;
Zo = 0, Wo = 1(0) > 0,
1+'
IP(C} = C, 1p(C) = Wo 1 - ,.
t: the disk IZ I < e. i3: the disk whose boundary circle intersects the real
axis orthogonally at the points wo } := ~ and wo ! ~ : .The radius of this
circle is Wo 1 2{! 2' The points of i3 satisfy the inequalities
-(!
t is the disk IZ I < e· The points ICI< e are mapped by! := ~ onto the disk
that intersects the real axis orthogonally at the points 11 + {! and 11 - e.
-(! +e
This disk lies completely in the sector whose bisector is the real axis
and whose center angle is 2 arctan 1 2e
-e
2 = 4 arctan e. Therefore £) lies
completely in the strip Iffiw I -
<~n arctan e. Besides we have in £)
2 e 2 + iC 2 1 + ie t.
+ in log
2
w = 1p(i;) = in log - - -inw~- = Wo inwo
1 - ie 2 C 1 - ie 2 C
The image of I i; I = e in the w-plane is convex [318J, moreover it is
symmetric with respect to the real axis because 'Ip(i;) is real for real i;-
values. Thus ffiw assumes its maximum and its minimum for real i; = ±e.
The width in the horizontal direction is therefore
l - i e - -2-e
inwo ( inw, )
2 l+ie--:i- e 2
w + -:- log
o 'in
- - c - ...... - -
~~Wo
w
0
+ 'tn
-:-log - -.-. ....1-nwo
----
1 - ie 2 e 1 + ie 2 e
ecos----
nwo e cos ...nwo
- )
= .!.( arctan ---_~__ + arctan ___ 2_ .
n 1 + e sin ~!£Jo 1 - e sin ~wo
2 2
We have /(1) = 1, ffi/(z) > o. Apply 281 by identifying ffi with ;:t, c.:::>
with the right half-plane, Zo with 1, Wo with 1. The functions z = cp(i;),
w = 1p(i;) are supposed to be normed in such a way that they transform
real i;-values into real z- and w-values; in addition let cp(l) = =, 1p(I) = =
[IV 119]. Hence
1p(i;) =
1 + t.
1 _ , .
Assume x > 1 and x = cp(e), 0 < e < 1. According to 281 the range
of j(z) in the w-plane is contained in the image of the disk Ii; I ;;:;: e under
the mapping w = 1p(i;). The image of the disk is a disk whose points are
at a distance of at most 1p(e) from the origin; i.e. I/(x) I <1p(e). Hence
we can define h(x) = 1p[cp-l(X)].
Pt. III, Solutions 290- 294 373
291. [K. Lowner.J According to 280 we have I/(z) I < Izl for Izl <
°
1;
hence for positive z, < z < 1,
I~- J(z) I > 1.
I 1- z 1-
By taking the limit z -+ 1 we find
11'(1) I > 1.
Assume that arg I' (1) = IX. A sufficiently small vector with the direction
{}, ; < {} < 3; , attached to z = 1, points towards the inside of the
unit circle and is mapped by the function w = I(z), 1'(1) =1= 0, onto an
analytic curve segment through the point w = 1 with the direction
{} + IX. Since I/(z) I < 1 for Iz I < 1 the direction is restricted by the
condition
~<{}+iX<3n
2 2
for any admissible {}. Thi.s is possible for iX = only. Cf. 144. °
292. Let Zo be fixed, IZo I <.1, I(zo) = wo, IWo I < 1. Choose the
constants 10 and 'YJ so that 10 11· - :"0 = 'YJ 1 - ~o = 1, 1101 = It71 = 1. Put
-zo 1 - Wo
z- z W - Wo -W
s-l-_.J!..=Z, 'YJ 1 - wow - .
- ZOZ
The function W = F(Z) defined with the help of w = f(z) satisfies the
hypothesis of 291. Therefore
293. [G. Julia: Acta Math. Vol. 42, p. 349 (1920).J Let Zo be fixed,
~zo >
0, I(zo) = wo' ~wo > 0. Choose the constants 10 and 'YJ so that
a-z b-w
e-~='YJ-b
a-zo
_0= 1, lsi
-W o
= I'YJI = 1. Define
ez - ~o = Z, 'YJ W - ~o = W.
z - Zo W - Wo
Then [291, 292J
1 < (dW) = (dW) I' (a) (dE)· = 1)(wo '- wo) I' (a) (a - zo):_
- dZ z';;i , .. dw w=o dZ Z=1 (b - WO)2 e(zo - zo) •
1 - Z 1z 1 ~ Z2Z 1 - Z nZ
294. The function I(z) - - - . . ... - - - is regular in the
z - Z1 Z - Z2 z - Zn
disk Iz I < 1, its absolute value is smaller than M 10, e > 0, sufficiently +
close to any boundary point [5]. Apply 278.-Different proof by careful
application of 176.
374 Functions of One Complex Variable
-+ -+
295. The function fez) ~.~ . ~ ... _n__ is regular in the half-
Z+Z
Zl - Z Zz - Z Zn - Z
converges.
298. Let lX be real, lX> 1, f(lX) =1= o. Proposition 295 implies that
the product it
v~l
II Z. -
Zv + IX
IX Idoes not diverge to O. Consequently the series
.~
00 (
1
I Zv
-I z. +
- IX
IX
12)
= ~
00
F: r +
41X Z. Eo
2z.z. > (1
41X 00 1
+ IX)Z.~ ~ z.
converges.
299. [T. Carleman; d. also P. Csillag: Mat. phys. lap. Vol. 26,
pp. 74-80 (1917).J Let Zo be an inner point of 'll and put If.(zo) 1= e.fv(zo)'
'11= 1, 2, ... , n. (In the case f.(zo) = 0 we choose e. = 1.) The function
F(z) = edl(z) + e2f 2 (z) + ... + enln(z)
is regular and single-valued in'll; F(z) assumes its largest absolute value
at a boundary point Zl of'll. Hence
P(Zl) > IF(Zl) I ~ IF(zo) I = p(zo)·
Pt. III, Solutions 295-302 375
300. It is more convenient to prove the statement as follows than
to refine the proof of 299: Let Zo be an inner point of '1) and the disk
1Z - Zo 1 < r be inside '1). Addition of the inequalities [272]
1 2"
(*) I/v(zo) 1 < 211: f I/v(zo + re ili ) 1 dtJ., 11 = 1, 2, ... , n
o
yields
1 2"
p(zo) < 211: f
o
p(Zo + reili ) dtJ..
If one of the summands (*) satisfies a strict inequality the sum satisfies
a strict inequality; i.e. if at least one of the lv's is not a constant the
maximum cannot be attained at an inner point ZOo
301. [G. Szego, Problem: Jber. deutsch. Math. Verein. Vol. 32,
2. Abt., p. 16 (1923).] It is sufficient to prove that pep) attains its
maximum on the boundary of any plane domain '1) whereby the points
PI' P 2 ' ••• , P n are not necessarily in the same plane. We introduce an
orthogonal coordinate system in '1), x, y; Z = x + iy. Now we have to
show: A function of the form
n (Iz -
n
av 12 + b;) I,
where av are arbitrary complex, and bv real, constants, 11 = 1, 2, ... , n,
attains its maximum on the boundary of any domain of the z-plane.
Work out the product [299].
302. Let Zo be an inner point of '1), where some of the functions,
called il-'(z) , do not vanish and other functions, called Iv(z), do vanish.
(One or the other type may be absent.) Let the radius r be so small that
the disk 1Z - Zo 1 < r lies completely inside '1) and does not contain any
other zero of the functions Iv(z) but ZOo The functions II-'(z)PI-' are regular
in this disk. According to 299 there exists a point zl> 1Zl - Zo 1 = r such
that
l: I/I-'(Zl) IPI-' > l: I/I-'(zo) [PI-'.
I-' I-'
Obviously
i.e. P(Zl) > p(zo); in fact P(Zl) > p(zo) if at least one II-' (z) is not a constant
or, in the other case, if at least one fv(z) does not vanish identically. Since
'1) is closed and p(z) continuous there exists a point in '1) where the func-
tion p(z) assumes its maximum: it cannot be an inner point except in the
particular case mentioned in the problem.
376 Functions of One Complex Variable
303. Since :I) is closed the function Ij(z) I, which is single-valued and
continuous, attains its maximum in :I). Proposition 134 shows that this
cannot happen at an inner point of :I) except in the case where j(z) is a
constant.
304. [J. Hadamard: Bull. Soc. Math. France Vol. 24, p. 186 (1896);
O. Blumenthal: Jber. deutsch. Math. Verein. Vol. 16, p. 108 (1907);
G. Faber: Math. Ann. Vol. 63, p. 549 (1907). Hurwitz-Courant, pp. 429-
430; E. Hille, Vol. II, pp. 410-411.J The function z"j(z) is not single-
valued in the annulus r 1 < 1z 1 < r3 , its modulus however is. Hence the
maximum of Iz"'j(z) I is either r~ M(rl) or r~M(r3) [303]. Choose IX so that
(*) r~M(rl) = r~M(r3)'
where Pn > 0 and at least two Pn's are non-zero [II 123].
307. Assume that I(z) is not a constant and that none of the zeros
z1' Z2' ... , zn of j(z) in the disk Iz I < r coincides with z = 0 (for simplicity's
sake assume also 1(0) = 1). Then we have [120J
log @(r) = n log r - loglzll - log IZ21 - .•• - log IZn I.
Hence the graph of log @(r), as a function of log r, consists in a sequence
of straight pieces with monotone increasing slopes. The change of slope
for log r = log ro is caused by the appearance of additional zeros on the
circle z = roo The increase in slope is equal to the number of such zeros
1 1
The function F(z) is regular in the disk Iz I < "3 and its absolute value
teaches the maximum on the boundary, say, at the point "seifJ•• Hence
1("2) = F("2) :;:;; IF("sifJ ,) I :;:;; 1("s),
that means 1(,,) is not decreasing. Determine the real number IX by the
equation
,,~1("1) = r;1("s)'
The absolute value of the function ZX F(z), which is regular in the annulus
"1< Izl < "3' is single-valued. Hence [303]
1'21("2) = "2'F("2) < max IZXF(z) I <"~1("1) = "31 ("3),
r,:a;lzl:a;r.
from which the convexity property of 1(,,) follows [304].
2",
309. l(,,) = J II' ("eifJ ) I"df} [308].
o
310. Define e2",iv/., = wv ' v = 1, 2, ... , n. Let 0 < < "1 "2 < R. There
exists [302] a point '''2eifJ., on the circle Izl = "2 such that
1 ., 1 .,
n v=1
~ Ij("1 Wv) IP < n ~ Ij("2Wi fJ, ) IP •
.=1
As n -t 00 this inequality becomes
1p ("1) < 1p ("2)'
Assume 0 < "1 < "2 < "3 < R, IX real. The functions
IX IX IX
consequently
1 2n .
/U(XO' Yo) / < 2n J /u(xo + r cos f}, Yo + r
o
SIll f}) / df}.
where the sign depends on whether u(xo' Yo) :::::: 0 or It(XO' Yo) < O. The
integrand must vanish identically, i.e. u(x, y) cannot change sign on the
given circle (u(x, y) possibly becomes 0 at some points).
313. Suppose that the point xO' Yo at which the maximum is reached
is an interior point of 'Il. Choose r so small that the disk with radius r
and center xo, Yo lies in the interior of 'Il. The equation
1
J [u(xo' Yo) -
2"
2-
no
u(xo + r cos f}, Yo + r sin f})] df} = 0 [solution 312]
implies
u(xo' Yo) - u(xo + r cos f}, Yo+ r sin f}) = 0, 0 < f} < 2n,
I.e. [311] u(x, y) =
const.
314. Follows from 313.
315. log /z - Zl/ + log /z - Z2/ + ... + log /z - zn / = mlog P(Z) ,
which is the potential of the system of forces in question, is a harmonic
function and as such it does not have a maximum nor a minimum at a
regular point. Stability would require a minimum of potential energy.
316. Remove the finitely many exceptional points from 'Il byenclos-
ing them in circles so small that these disks have no points in common
and do not contain any point at which the function reaches its maximum
in 'Il. Apply 313 to the remaining domain.
317. According to 188 the orientation of the image of the circle
/z/ = R is preserved. Therefore 109 can be applied. The harmonic func-
tion
mJ(z)_
J(z)
which is regular in the disk /z / < R [f(z) being schlicht has the only and
simple zero z = 0] is positive for /z / = R. Hence it is positive on any
smaller concentric circle /z / = r < R [313]. The images of the circles
/z/ = r are star-shaped with respect to the origin according to 109.
Pt. III, Solutions 313-322 379
mz j"(z)
lTzl + 1,
which is regular on the disk 1 z 1 < R [/' (z) =!= 0 because fez) is schlicht J
is positive for 1 z 1 = R. Hence it is positive on any smaller concentric
circle Izl = r < R. According to 108 the images of the circles Izl = r
are convex. The statement is now proved in the case where the inner
circle is concentric to the disk 1z 1 < R.
Let the inner circle lie anywhere inside the disk Izl < R. We build
up w = f(z) , the given function, by combining two functions: a linear
mapping ~ = l(z) of the disk 1 z I < R onto itself whereby the given inner
circle is transformed into one with center at the origin ~ = 0 [77J and a
second one w = g(~) = f(l-l(~)); in fact, W = fez) = g(l(z)). By considering
W = g(~) we reduce the problem to the previously discussed special case.
[313,304J.
321. [Cf. A. Walther, l.c. 320.J (1) The three circle theorem proved
in 320 holds also for harmonic functions that have ih the disk I z 1 < R
finitely or infinitely many isolated singularities provided their accumu-
lation points are not interior points of the disk and that, in addition,
the function tends to - 00 as z approaches a singular point [316J.
Apply this generalized three circle theorem to the function
323. The arcs may be replaced by any continuous curves that cross
the sector connecting the two rays. Their minimal distance from the
origin must increase beyond all bounds.
324. All the estimates of solution 322 remain valid in the domain
r
bounded by 1 and 2• r
325. Cf. solution p. 168.
326. The function e""f(z) satisfies the conditions (1), (2) and the modi-
fied condition (3) of 325 [final remark in the proofJ however large w.
Consequently
If(z) I< e- wx for ffiz = x > o.
Let w increase to infinity.
327. Put arctan ~1 = 1p. Then
x+
z log(z + 1) = (r cos {} + ir sin {}) [1 log (r 2 + 2r cos {} + 1) + i1pJ,
n
hence for - 2" < {} n
< 2" ' r > 1,
satisfies the conditions (1), (2) and (3) of 326 with <i = o. - Instead of
quoting 326, 325 we could apply the ideas developed in those solutions
to the function
(with A' > A) is best established first outside and then inside the circles
Iz - nl =!, n = 0,1,2, ... [solution 165].
Pt. III, Solutions 323 - 332 381
Second solution: 178 and the fact that the terms on the left hand
side are positive imply
~
n(lIi - Ik) < 2~1fn(1
n2 (12 -
1) [2 log C + 2{:rt -
n2 y) e] de +C , I
~-l 1
where C and C' denote constants. The left hand side is <Xl log n, the right
hand side <Xl ~ - Y log n: contradiction. This method can be generalized
~
unchanged.
331. In the special case (X = - ; , (J = ; the statement is weaker
than 325. The proof involves the function
/ .fJ+a:n;:n; )
t(z) exp \ - "Ie -, fJ-a "2 :I-a .
Start by proving with the help of 330 that the maximum of the modulus
of t(z) on the bisector is not larger than 1 [325J.
332. Assume Ig( -r) I < C: apply 331 to the function g~l in the
sector -:rt < {} < :rt; this leads to Ig(z) I < C in the entire plane, thus
g(z) is a constant.
382 Functions of One Complex Variable
z e -ny-;-e(-iz)i.
g()
This function is analytic in the open sector 0 < f} < n and continuous
ini the closed sector 0 <f} < n. Cf. 325.-Also 325 can be applied to
C- Ig(Z2).- The function SiV~ illustrates how strong the theorem is.
333. Let a < b < 1. The absolute value of the test function ee bz
! eb(x+iy)
i e
I'
== eebxcos by ;
=b 2 1 A
thus on the boundary of 1R it is > e 2 > 1. Let l> b _ a log n
f: cos b'i
Then we find on the boundary of the rectangle 0 :S: x < l, - ~ ::::::: y < ~
the inequality
bz I
i j(z)
1
e-ee < 1.
where Xl is chosen so that sw(xIl > 1., On the boundary of this rectangle
we have
zl
ee I <e,
thus e,g. ee < e: contradiction. The role of the function ee z is explained
by 290 and 187.
335. It is sufficient to consider the case hinted at in the problem
[linear transformation]. Take S > 0; the function
<p(z) = j(z) iI ( ~r
.~1
z z. r
Pt. III, Solutions 333- 339 383
is regular in the intersection ffi, of ffi and the disk Iz I < r and its modulus
is single-valued in ffir [303]. The maximum of \j(z) \ on the circle Iz I = r
is denoted by M(r). Taking into account the condition on j(z) on the
boundary of ffi [278J we find that in fir
q;(z) < max [M, M(r)J.
Hence for 8 ---+ 0
\j(z) \ <max [M, M(r)J.
If, in particular, r' is also admissible [see hintJ and r' < r, then we have
M(r') ::;;: max [M, M(r)J.
On the other hand [268J
M(r) < M(r').
Hence the alternative: Either M(r) > M, thus M(r) = M(r') , j(z) = const.
[268] or M(r) < M, Ij(z) I < M, even Ij(z) I < M [278].
336. We assume that n boundary sections of St: lie on the real axis.
From a variable point z in the upper half-plane they are seen under the
angles wI> w2 ' ••• , wn . Determine a regular and analytic function q;v(z)
in the half-plane ~z > 0 so that nffiq;v(z) = w. [57J and put
if>(z) = a . ( ~ )",,(Z) +'I',(z) +... H'n(Z).
The function j(z) if>(Z)-l is regular and bounded in the interior, < 1 on
the boundary, of 'Il with the possible exception of 2n boundary points
[335J.
337. The function j(eU ) is single-valued, regular and bounded in the
half-plane ffiu < O. We have \j(eU ) I < 1 at all the boundary points of this
half-plane except at the boundary point u = 00. [335.J
338. If z = 00 wt're not a boundary point we would have [135
sufficesJ necessarily Ig(z) I ::;;: k in ffi: contradiction. Helke z = 00 is a
boundary point of ffi. If g(z) were bounded in ffi we could use 335 (only
excluded point z = 00) which would lead to Ig(z) I < k in ffi.
339. There exists, according to the hypothesis, a constant M, M > 0,
such that Ij(z) 1< M in the region ffi bounded by Tl and T 2 • Choose
R > 1 and so large that Ij(z) \ < 8 on Tl and T2 outside the circle Iz 1= R.
Take the branch of log z that is real for positive z. This branch is regular
and single-valued in ffi, where \log z I < log Iz I + n provided that
Iz I > R. The inequality
1M (log R + n) + 8 (log z + n)1 > M (log R + n) + 8 (log Izl + n)
384 Functions of One Complex Variable
holds and for Iz I > R both addends on the right hand side are positive.
Thus we obtain for Izl = R, z in ffi,
--- t(z) I < log R+ n M = 1
IM(log R + n)log_z ___ -
+ e(log z + n) M(log R + n) .
If Iz I > R, z on Fl or on F2 we find
log z I log Iz I + n 1
IM(log R + n) + e(log z + n) t(z) ,i < e(log Iz I + n)
e-
- .
Hence we have, by virtue of 335, at each point z in ffi but outside the
circle Iz I = R
It(z) I < Ie(log z ±-nL± ~~og R ~ 'I' .
log z
The right hand side becomes smaller than 2E when Iz I gets sufficiently
large.
340. Assume, if possible, a =F b and consider two disks DI and D2
in the w-plane, with no points in common, centered at a and b respec-
. 1y. 0 utSI·de t he two d·IS k s t he expreSSIOn
tlve . I(
I w - --2- - - - 2 - - a+b)2 (a-b)21
has a positive minimum = E. Proposition 339 applied to the function
- b)2 Imp
(t( z) - -a-+2-b)2 - (a--2- . 1·Ies t h at the absolute value of thIS. functron
.
, ,
Author Index
Enestr6m, G. 301
Bagnera, J. 208
Euler, L. 50, 51, 54, 57, 174, 177, 178,
Bendixson, J. 361
186, 234, 240, 243, 348, 349
Bernoulli, D. 121, 124
Bernoulli, Jakob 35
Faber, G. 376
Bernstein, S. 275, 367
Farey, J. 93
Bessel, F. W. 22, 97
Fatou, P. 369
Bieberbach, L. 325
Fejer, L. 200, 253, 260, 272, 281, 303,
Biehler, Ch. 302
337
Blaschke, W. 251, 267, 374
Fekete, M. 198, 302, 359
Blumenthal, O. 376
Fine, N. J. 228
Bohr, H. 372
Fourier, J. 83, 253, 273
Borel, E. 157, 198, 332
Franel, J. 234, 275, 283, 284
Brauer, A. 304
Frank, Ph. 326
Brauer, R. 304
Fredholm, 1. 247
Briggs, H. 92, 284
Frobenius, G. 196
Biirmann, H. 146
Fujiwara, M. 304
Burkhardt, H. 288
Gauss, C. F. 11, 186, 187, 303
Cantor, G. 206 Goursa1;, E. 313
Caratheodory, C. 261, 269, 355, 370 Greenstreet,E. 183
Carleman, T. 195, 325, 337, 340, 374 Gutzmer, A. 324
Carlson, F. 380
Catalan, E. 176, 178 Haar, A. 206
Cauchy, A. L. 64, 68, 111, 113, 119, 133, Hadamard, J. 164, 166, 309, 376
146, 221, 245, 248, 312, 317, 326, 330, Hamburger, H. 332, 360
337 Hansen, P. A. 97
Cesaro, E. 15, 20, 176, 178, 179, 185, Hardy, G. H. 179, 195, 208, 247, 251,
193, 208, 217, 234, 240 257, 332, 359, 376
Cornu 142, 342 Harnack, A. 363
Cramer, H. 286 Heeke, E. 280
Csillag, P. 287, 374 Hermite, Ch. 104, 180, 182,288,302,332
386 Author Index
Soft and hard cover editions are available for each volume up to vol. 14, hard cover
only from Vol. 15