SOME RESULTS IN ANALYSIS

MIGUEL A. LERMA

Abstract. This is a list of selected mathematical results placed here to have them handy.

Contents

1. Introduction 3

2. Inequalities 3

2.1. Inequalities 3

3. Some Identities 5

3.1. Infinite Sums 5

3.2. Infinite Products 6

4. Functions, Measure, Integration 6

4.1. Lebesgue Integral 6

4.2. Various Definitions and Properties of Functions 7

4.3. Henstock-Kurzweil Integral 9

4.4. Summability 12

5. Fourier Series 13

5.1. Definition and properties 13

5.2. Some Theorems 15

5.3. Some Kernels 17

6. Integral Transforms 19

6.1. Fourier Transform 19

6.2. Laplace Transform 22

1
7. Complex Analysis 24

7.1. Some Basic Theorems 24

7.2. Harmonic Analysis 26

7.3. H p Spaces 29

8. Some Special Functions 35

8.1. Bernoulli periodic functions 35

8.2. Polylogarithms 36

9. Summation Formulas 38

9.1. The Euler-Maclaurin Summation Formula 38

9.2. The Poisson Summation Formula 39

10. Miscelanea 41

10.1. Various Results 41

References 43

2
 1. Introduction

This is a list of selected mathematical results placed here to have them handy.

2. Inequalities

2.1. Inequalities.

2.1.1. Abel. If u1 ≥ u2 ≥ u3 ≥ · · · ≥ up ≥ 0, then

p
X k
X
(2.1) an un ≤ u1 max an .
1≤k≤p
n=1 n=1

2.1.2. Bessel.
n
X
(2.2) |u|2 ≥ |u · xk |2 ,
k=1
where {xk }k=1,...,n is orthogonal normalized, i.e., xi · xj = δij (Kronecker delta) for i, j =
1, 2, . . . , n.

2.1.3. Cauchy. (Hölder for p = q = 2.)

(2.3) |u · v| ≤ kuk2 kvk2 .

n 2 n 2 n 2
X X X
(2.4) ai bi ≤ ai bi .
i=1 i=1 i=1

2.1.4. Chebyshev.

h(x − µ)2 i σ2
(2.5) P (x − µ)2 ≥ k 2 ≤ = ,
k2 k2
where P (S) = probability of event S, and hf i = mean value of f , µ = hxi = mean value of
x, σ 2 = variance of x.

2.1.5. Hadamard.
2
a11 a12 ··· a1n  
n n
a21 a22 ··· a2n Y X
(2.6) .. .. .. ≤  |aij |2 
. . ··· . i=1 j=1
an1 an2 ··· ann

1 1
2.1.6. Hausdorff-Young. If 1 < p < 2, p + q = 1, then

(2.7) q 1/2q kfbkq ≤ p1/2p kf kp .

3
2.1.7. Hölder. If p > 1 and 1/p + 1/q = 1 then

(2.8) |u · v| ≤ kukp kvkq .

Particular cases:
n n
!1/p n
!1/q
X X X
p p
(2.9) ai bi ≤ |ai | |bi | ,
i=1 i=1 i=1

Z Z 1/p Z 1/q
p q
(2.10) f g dµ ≤ |f | dµ |g| dµ .
Ω Ω Ω

2.1.8. Jensen. If µ(Ω) = 1, f : Ω → (a, b) in L1 (µ), ϕ convex on (a, b), then

Z  Z
(2.11) ϕ f dµ ≤ (ϕ ◦ f ) dµ .
Ω Ω
Pn
If ϕ is convex on (a, b), x1 , x2 , . . . , xn ∈ (a, b), λi ≥ 0 (i = 1, 2, . . . , n), i=1 λi = 1, then
n n
!
X X
(2.12) ϕ λi xi ≤ λi ϕ(xi ) .
i=1 i=1

If ai > 0 (i = 1, 2, . . . , n), s > t > 0, then

n
!1/s n
!1/t
X X
s
(2.13) ai ≤ ati .
i=1 i=1

2.1.9. Markov. If a > 0 and the random variable x takes only non negative values, then
hxi
(2.14) P (x ≥ a) ≤ ,
a
where hxi = mean value of x.

2.1.10. Minkowski. If p ≥ 1 then

(2.15) ku + vkp ≤ kukp + kvkp ,

n
!1/p n
!1/p n
!1/p
X X X
(2.16) |ai + bi |p ≤ |ai |p + |bi |p ,
i=1 i=1 i=1

Z 1/p Z 1/p Z 1/p

p p p
(2.17) |f + g| dµ ≤ |f | dµ + |g| dµ .
Ω Ω Ω

4
2.1.11. Newton. Let pr be the average value of the nr terms comprising the rth elementary

symmetric function br of a set of numbers a1 , . . . , an (0 ≤ r ≤ n), i.e.,

Yn Xn
(−1)k bk (a1 , . . . , an ) xn−k , pr = br / nr .


(x − ai ) =
i=1 k=0
Then:
(2.18) pr−1 pr+1 ≤ p2r .

2.1.12. Schwarz. (Hölder with p = q = 2)

(2.19) |u · v| ≤ kuk2 kvk2 ,
n 2 n
! n !
X X X
(2.20) ai bi ≤ |ai |2 |bi |2 ,
i=1 i=1 i=1
Z 2 Z  Z 
(2.21) f g dµ ≤ |f |2 dµ |g|2 dµ .
Ω Ω Ω

2.1.13. Young’s Theorem. Assume p, q, r ∈ [1, ∞] and

1 1 1
= + − 1.
r p q
If f ∈ Lp and g ∈ Lq , then their convolution f ∗ g exists and belongs to Lr . Moreover:
kf ∗ gkr ≤ kf kp kgkq .

2.1.14. Young. If f is continuous, strictly increasing for x ≥ 0, f (0) = 0, g = f −1 , then for

a, b ≥ 0:
Z a Z b
(2.22) ab ≤ f (x) dx + g(y) dy .
0 0
The equality holds for b = f (a).

3. Some Identities

3.1. Infinite Sums.

∞ N
1 X 2z X 1
(3.1) π cot πz = + = lim .
z z 2 − n2 N →∞ z−n
n=1 n=−N

∞
 π 2 X 1
(3.2) =
sin πz n=−∞
(z − n)2

∞
( )
 π 2 X 1 1
(3.3) = +
cos πz 1 2 1 2



n=0 z−n− 2 z+n+ 2

5
 ∞
1 X m sin mx sin {α(π − x)}
(3.4) 2 2
= .
π m=−∞ m − α sin απ

∞
1 X cos mx cos {α(π − x)}
(3.5) 2 2
=− .
π m=−∞ m − α α sin απ

∞
2 X cos {(2m + 1)x} cos { α2 (π − x)} cos {α(π − x)}
(3.6) = − .
π (2m + 1)2 − α2 2α sin απ 2 α sin απ
m=0

3.2. Infinite Products.

∞  N 
z2

sin πz Y Y z
(3.7) = 1 − 2 = lim 1− .
πz n N →∞ n
n=1 n=−N
n6=0

∞ N
! !
Y z2 Y z
(3.8) cos πz = 1− = lim 1− .
n=0
(n + 21 )2 N →∞
n=−N
n+ 1
2

∞ 
z2

sin {π(z − α)}  zY
(3.9) = 1− 1−
− sin (πα) α (n + α)2
n=1
N  
Y z
= lim 1− .
N →∞ n+α
n=−N

∞
!
cos {π(z − α)} Y z2
(3.10) = 1−
cos (πα)
n=0
(n + 12 + α)2
N
!
Y z
= lim 1− 1 .
N →∞
n=−N
n+ 2 +α

4. Functions, Measure, Integration

4.1. Lebesgue Integral.

Theorem 4.1 (Lebesgue’s Monotone Convergence [LMCT]).

Let fn : X → [0, ∞] be a nondecreasing sequence of measurable functions. Then their limit

is measurable and Z Z
lim fn dµ = lim fn dµ .
X n→∞ n→∞ X

6
Corollary 4.2. If fn : X → [0, ∞] is measurable for every n,
Z X ∞ ∞ Z
X
fn dµ = fn dµ .
X n=0 n=0 X

Theorem 4.3 (Fatou’s Lemma). If fn : X → [0, ∞] is measurable for every n,

Z Z
lim inf fn dµ ≤ lim inf fn dµ .
X n→∞ n→∞ X

Theorem 4.4 (Lebesgue’s Dominated Convergence [LDCT]).

Let fn : X → C be measurable, f (x) = limn→∞ fn (x), and |fn (x)| ≤ g(x) for some function
g ∈ L1 (X). Then f ∈ L1 (X) and
Z
lim |fn − f | dµ = 0 ,
n→∞ X

Z Z
lim fn dµ = f dµ .
n→∞ X X

Theorem 4.5. Let X, Y be σ-finite measure spaces.1

(1) (Tonelli) If f : X × Y → [0, ∞] is measurable then for a.e. y ∈ Y the function

x 7→ f (x, y) is measurable and
Z Z  Z
f (x, y) dx dy = f (x, y) dxdy .
Y X X×Y

(2) (Fubini) If f is in L1 (X ×R Y ) then for a.e. y ∈ Y the function x 7→ f (x, y) is in

L1 (X), the function y 7→ X f (x, y) dx is in L1 (Y ) and
Z Z  Z
f (x, y) dx dy = f (x, y) dxdy .
Y X X×Y

(The integrals are respect to the corresponding measures in X, Y or the product space
X × Y .)

4.1.1. Lp and lp spaces.

• Lp (X) = set of functions f : X → C such that X |f |p dµ < ∞, where

R R
is the
Lebesgue integral respect to the measure µ in the measure space X.
X∞
• lp = set of complex sequences {cn }n∈N such that |cn |p < ∞.
n=0

4.2. Various Definitions and Properties of Functions.

1A measure space is σ-finite if it is a countable union of sets of finite measure.

7
4.2.1. Variation. The variation of a function f : [a, b] → Rn over a compact interval I =
[a, b] is

(4.1) VI (f ) = Vab (f ) =
( n )
X
sup |f (xi ) − f (xi+1 )| : a = x0 ≤ x1 ≤ · · · ≤ xn = b
i=0

It is of bounded variation if Vab (f ) < ∞. Notation: BV (I) = set of functions of bounded

variation in I.

A function f : [a, b] → R has bounded variation iff it is the difference of two nondecreasing
functions.

Over non compact (including unbounded) intervals the definition can be generalized like
this:
Vab (f ) = lim Vf (a0 , b0 )
a0 →a+
b0 →b−

A function is locally of bounded variation if it is of bounded variation over every compact

interval in its domain.

4.2.2. Absolutely Continuous. A function f : [a, b] → R is absolutely continuous if for every

 > 0 therePis a δ such that for any Pnon-overlapping collection of subintervals [xi , yi ] ⊆ [a, b]
such that m (y
i=1 i − x i ) < δ, then m
i=1 |f (yi ) − f (xi )| < ε.

Notation: C(I) = set of continuous functions in I. AC(I) = set of absolutely continuous

functions in I.

Any absolutely continuous function is continuous and has bounded variation.

A function is absolutely continuous iff it is an indefinite integral of a Lebesgue-integrable

function.

4.2.3. Singular Function. A singular function is a function whose derivative is zero a.e.

4.2.4. Jump Function. An elementary nondecreasing (resp. nonincreasing) jump function

is a function of the form 
A if x < x0

h(x) = B if x = x0

C if x > x0


where A ≤ B ≤ C (resp. A ≥ B ≥ C). We also assume A = 6 C. A jump function is

a function than can be expressed as an absolutely convergent series of elementary jump
functions:
∞
X
(4.2) f (x) = hn (x) ,
n=0

8
A jump function is the difference of two nondecreasing jump functions, hence it is locally
of bounded variation.

A step function is a sum of finitely many elementary jump functions.

4.2.5. Regulated Function. A regulated function is a function that has all of its one-sided
limits at all points of its domain. A function is regulated iff it is the uniform limit of step
functions. All functions of bounded variation are regulated, but the converse is not true.

4.2.6. Lebesgue Decomposition Theorem. Every function of bounded variation f : [a, b] → R

can be expressed in the form:
(4.3) f = fa + fs + fj ,
where fa is absolutely continuous, fs is continuous singular, and fj is a jump function.
These functions are unique to within additive constants.

If f is nondecreasing (resp. nonincreasing), fa , fs and fj can be chosen to be nonde-

creasing (resp. nonincreasing). In this case the function is absolutely continuous iff
Z b
f 0 (x) dx = f (a) − f (b) .
a

4.2.7. Absolutely Rintegrable. A function f : R → C is absolutely integrable, or in L1 (R), if f

∞
is integrable and −∞ |f (x)| dx < ∞.

4.2.8. Locally integrable. A function f is locally integrable, or in L1loc , if at every point of its
domain there is a neighborhood where f is integrable and its integral is finite.

4.2.9. Convex. A function f : (a, b) → R is convex if for every x, y ∈ (a, b), 0 ≤ λ ≤ 1:

f ((1 − λ)x + λy) ≤ (1 − λ)f (x) + λf (y) .

4.3. Henstock-Kurzweil Integral.

4.3.1. Partitions. Given a nondegenerate closed interval I = [a, b], a partition or division
of I a finite collection P = {Ii | i = 1, . . . , n} of intervals Ii = [xi−1 , xi ] with a = x0 ≤ x1 ≤
x2 ≤ · · · ≤ xn = b. A tagged partition of I is a collection Ṗ = {(Ii , ti ) | i = 1, . . . , n}, where
P = {Ii | i = 1, . . . , n} is a partition of I and ti ∈ Ii for each i.

4.3.2. Riemann Sums. Given a function f : I = [a, b] → R and a tagged partition Ṗ =

{(Ii , ti ) | i = 1, . . . , n} of I, the Riemann sum of f corresponding to Ṗ is
n
X
S(f ; Ṗ) = f (ti ) l(Ii ) ,
i=1
where l(Ii ) = b − a is the length of Ii .

9
4.3.3. Riemann Integral. The Riemann integral of a function f : I = [a, b] → R, is the
number
Z b
(4.4) f (x) dx = lim S(f ; Ṗ) ,
a |Ṗ|→0

where |Ṗ| = max l(Ii ). In other words, for every ε > 0 there is a δ > 0 such that if
Ii in Ṗ
Ṗ = {(Ii , ti ) | i = 1, . . . , n} is a tagged partition of I verifying l(Ii ) ≤ δ for i = 1, . . . , n,
then
Z b
f (x) dx − S(f ; Ṗ) ≤ ε .
a

4.3.4. Gauges. A gauge on I = [a, b] is a function δ : I → (0, ∞). A tagged partition

Ṗ = {(Ii , ti ) | i = 1, . . . , n} of I is δ-fine if Ii ⊆ [ti − δ(ti ), ti + δ(ti )] for all I = 1, . . . , n.

4.3.5. Henstock-Kurzweil Integral. (See [7, 2]). The Henstock-Kurzweil integral (or gauge
integral, or generalized Riemann integral ) of a function f : I = [a, b] → R, is the number
Rb
a f (x) dx verifying that for every ε > 0 there is gauge δ : I → (0, ∞) such that if Ṗ is a
δ-fine tagged partition of I, then
Z b
f (x) dx − S(f ; Ṗ) ≤ ε .
a

Note: The Riemann integral coincides with the Henstock-Kurzweil integral restricted to
constant gauges.

Both the Riemann and the Henstock-Kurzweil integrals can easily be generalized to
functions f : I = [a, b] → Rn

The Henstock-Kurzweil integral can be generalized to functions defined on unbounded

intervals by allowing a and b to be infinity, and omitting in the Riemann sums the terms
corresponding to unbounded subintervals. Alternatively define f (∞) = f (−∞) = 0 and
adopt the conventions (±∞) · 0 = 0 · (±∞) = 0, (±∞) + x = x + (±∞) = ±∞, (±∞) · x =
x · (±∞) = ±∞ or ∓∞ depending on whether x > 0 or x < 0.

4.3.6. Properties of the Integral. Here f, g : R → R (although some properties are obviously
generalizable to vector-valued functions). The integral is assumed to be Henstock-Kurzweil
unless stated otherwise.

(1) Linearity.
Z Z Z
(f + g) = f + g
I I I
Z Z
cf = c f
I I

10
 (2) Interval additivity.
Z b Z c Z b
f= f+ f
a a c
(3) Squeeze Theorem. f is integrable over I if and only if for every ε > 0 there are two
integrable functions ψε , ϕε integrable over I, such that ψε (x) ≤ f (x) ≤ ϕε (x) for
every x ∈ I and
Z
(ψε − ϕε ) ≤ ε .
I
Rb
(4) Change of Variable. Let τ : [c, d] → [a, b] be continuous and monotone. If a f dα
Rb
exists then a f ◦ τ d(α ◦ τ ) exists and
Z b Z τ (d)
(4.5) f ◦ τ d(α ◦ τ ) = f dα .
a τ (c)

Theorem 4.6 (Fundamental Theorem of Calculus). Given f : [a, b] → R, if f is integrable

over I and Z x
F (x) = f (u) du ,
a
then F is continuous on I and F 0 (x) = f (x) for a.e. x in [a, b].

If f has a right (left) hand limit at x ∈ [a, b), then

F (x + h) − F (x)
f (x±) = lim .
h→0± h
Theorem 4.7. A function f is Lebesgue-integrable iff f and |f | are Henstock-Kurzweil-
integrable.

4.3.7. Stieltjes Integral. A Stieltjes integral of a function f respect to another function ϕ,

represented
Z b
(4.6) f (x) dϕ(x) ,
a

is defined as in the previous paragraphs replacing Riemann sums with Riemann-Stieltjes

sums:
n
X
Σ(f, ϕ; Ṗ) = f (ti ) [ϕ(b) − ϕ(a)] .
i=1
It this way it is possible to define the Riemann-Stieltjes integral, and the Henstock-Kurzweil-
Stieltjes (or generalized Riemann-Stieltjes) integral.

In the following theorem we use the notation

∆− (f ; c) = f (c) − f (c−)
∆+ (f ; c) = f (c+) − f (c)
∆(f ; c) = f (c+) − f (c−)

11
Theorem 4.8 (Integration by Parts for the Henstock-Kurzweil-Stieltjes Integral). Suppose
that one of f , ϕ is a function of bounded variation and the other one is a regulated function
Rb Rb
over [a, b]. Then both a f (x) dϕ(x) and a ϕ(x) df (x) exist and
Z b Z b
(4.7) f (x) dϕ(x) = f (b)ϕ(b) − f (a)ϕ(a) − ϕ(x) df (x) + S ,
a a

where
X
(4.8) S= {∆− (f ; cn ) ∆− (ϕ; cn ) − ∆+ (f ; cn ) ∆+ (ϕ; cn )}
n
X
= {f (cn ) ∆(ϕ; cn ) + ϕ(cn ) ∆(f ; cn ) − ∆(f ϕ; cn )} .
n

Here the cn ’s are the points at which f and ϕ are simultaneously discontinuous from the
left or from the right (see [7, chap. 7].)

4.4. Summability.

∞
X
4.4.1. Cesàro Summability. A series an is summable (C, 1) to A if
n=1

N N  
1 X X n−1
A = lim Sn = lim 1− an ,
N →∞ N N →∞ N
n=1 n=1
n
X
where Sn = ai .
i=1

∞
X
A series an is summable (C, k) to A if
n=1

(k)
Tn
A = lim n+k−1

,
n→∞
k
(k)
where Tn is defined recursively in the following way:
n
X
Tn(0) = Sn = ai
i=1
n
(j−1)
X
Tn(j) = Ti (j > 0)
i=1

Equivalently:
N
1 X
N +k−n


A = lim N +k−1

k an .
N →∞
k n=1

12
 ∞
X
4.4.2. Hölder Summability. A series an is summable (H, 1) to A if
n=1

N N  
1 X X n−1
A = lim Sn = lim 1− an ,
N →∞ N N →∞ N
n=1 n=1

n
X
where Sn = ai .
i=1

∞
X
A series an is summable (H, k) to A if
n=1

A = lim Un(k) ,
N →∞

(k)
where Un is defined recursively in the following way:
n
X
Un(0) = Sn = ai
i=1
n
1 X (j−1)
Un(j) = Ui (j > 0)
n
i=1

Equivalently:
N
X
A = lim cn an .
N →∞
n=1

where
X 1
cn =
i1 i2 . . . ik−1 N
n≤i0 ≤i1 ≤···≤ik−1 ≤N
X i1 − n + 1
=
i1 i2 . . . ik
n≤i1 ≤···≤ik−1 ≤ik =N

5. Fourier Series

5.1. Definition and properties. Note: There are two approaches to Fourier series, de-
pending on whether we are interested in studying 1-periodic functions, defined on

(5.1) T1 = R/Z ,

or 2π-periodic functions, defined on

(5.2) T2π = R/2πZ .

13
Both T1 and T2π are endowed with a uniform measure such that their total measure is 1,
so if f is an integrable function defined in either T1 or T2π we have
Z Z 1
(5.3) f (x) dx = f (x) dx ,
T1 0
Z Z 2π
1
(5.4) f (x) dx = f (x) dx .
T2π 2π 0

The second approach is perhaps the most common, but here we will use the first one
because it provides simpler and more symmetric formulas, so some of the usual functions
and classical kernels in Fourier analysis will be replaced with the ones obtained by the
variable change x → 2πx.

Given f : T1 → C, we define the nth Fourier coefficient of f :

Z
(5.5) f (n) =
b f (x) e−2πinx dx .
T1

The Fourier series of f is2

∞
X
(5.6) s[f ](x) = fb(n) e2πinx = lim sN [f ](x) ,
N →∞
n=−∞

where
N
X
(5.7) sN [f ](x) = fb(n) e2πinx
n=−N

The Cesáro means of the Fourier series are

N N  
1 X X |n|
(5.8) σN [f ](x) = sk [f ](x) = 1− fb(n) e2πinx .
N +1 N +1
k=0 n=−N

The concept of Fourier series can be extended to distribution functions in the following
way. Let F : R → C be a function such that F (x + 1) − F (x) = F (1) − F (0) for every
x ∈ R. The Fourier-Stieltjes series of F , or Fourier series of dF is defined
∞
X
(5.9) s[dF ](x) = c (n) e2πinx ,
dF
n=−∞

where
Z 1
(5.10) dF
c (n) = e−2πinx dF (x)
0

2When both limits in a sum are infinite we assume that the index approaches infinity in a symmetric
way:
X∞ XT
f (k) = lim f (k) .
T →∞
k=−∞ k=−T

14
is the nth Fourier-Stieltjes coefficient of F or Fourier coefficient of dF —the integral being
of Stieltjes type.

For instance the periodic Dirac delta corresponds to the distribution function F (x) =
bxc = integer part of x. Its Fourier coefficients are so:
dF
c (n) = 1 for every n ,
and its Fourier series is
∞
X
s[dF ](x) = e2πinx .
n=−∞

5.1.1. Properties.

(1) Linearity.
\
af +bg = afb + bb
g.
(2) Translation. If (Ty f )(x) = f (x + y) (translated of f by y), then:
2πiy
Td
y f (n) = f (n) e
b .
(3) Modulation. If m ∈ Z and (Em f )(x) = e2πimx f (x), then

m f (n) = f (n − m) .
[
E b
(4) Complex Conjugation.
fb(n) = fb(−n) .
(5) Differentiation. If f is differentiable over T1 then:
fb0 (n) = 2πinfb(n) .

5.1.2. Convolution. The convolution of two functions f, g : T1 → C is

Z
(5.11) (f ∗ g)(x) = f (u)g(x − u) du .
T1

If f, g ∈ L1 (T1 ) then f[
∗ g ∈ L1 (T1 ) and
(5.12) ∗ g = fb b
f[ g.

Note:
sN [f ] = f ∗ DN ,
and
N
1 X
σN [f ] = sk [f ] = f ∗ FN ,
N +1
k=0
where DN and FN are respectively the Dirichlet and Fejér kernels (see bellow).

5.2. Some Theorems.

15
5.2.1. Inversion Theorems.
Proposition 5.1. If cn is nonincreasing and cn → 0 then the series
∞
X
cn e2πinx
n=−∞
converges uniformly on compact subsets of T1 \ {0}.
∞
X
Proposition 5.2. If |cn | < ∞ then the series
n=−∞
∞
X
f (x) = cn e2πinx
n=−∞

converges uniformly on T1 (so f is continuous) and cn = fb(n).

∞
X
Theorem 5.3 (Fourier Inversion Theorem). Assume f ∈ L1 (T1 ) and |fb(n)| < ∞.
n=−∞
Then s[f ] is continuous and f (x) = s[f ](x) a.e. If f is continuous then f (x) = s[f ](x) for
every x.
Theorem 5.4. If f : T1 → C is bounded and verifies the Dirichlet conditions:

(1) f is continuous except possibly for a finite number of jump discontinuities,

(2) f has a finite number of maxima and minima,

then (in Cesáro sense)

∞
X
f (x) = fb(n) e2πinx ,
n=−∞
provided that f (x) = 12 {f (x+) + f (x−)}.

Note that the summation converges in Cesàro sense, i.e.:

N  
X |n|
f (x) = lim 1− fb(n) e2πinx .
N →∞ N +1
n=−N

Theorem 5.5. If f ∈ L1 (T1 ), f (x) = 12 {f (x+) + f (x−)}, and for some constant C
|f (y) − f (x±)| ≤ ±C(y − x)
for all y in a neighborhood of x, then
N
X
f (x) = lim fb(n) e2πinx .
N →∞
n=−N

Corollary 5.6. If f satisfies a Lipschitz condition on T1 then its Fourier series converges
uniformly to f .
Theorem 5.7 (Dirichlet-Jordan Test). Suppose f ∈ BV (T1 ). Then

(1) s[f ](x) = 12 {f (x+) + f (x−)} for every x ∈ T1 .

16
 (2) If f is continuous in a closed interval I, s[f ] converges uniformly in I.

Proof. See [11, thm. 8.1]. 

Theorem 5.8 (Cesáro Sumability of Fourier Series). The Cesàro means of the Fourier
series of f converge to f in the following ways.

(1) f ∈ C(T) ⇒ σN [f ] → f uniformly.

(2) f ∈ L1 (T) and f continuous at x0 ⇒ σN [f ](x0 ) → f (x0 ).
(3) f ∈ Lp (T) ⇒ σN [f ] → f in Lp (T) for 1 ≤ p < ∞.

5.2.2. Other Theorems.

Theorem 5.9. The mapping f ↔ gb is a Hilbert space isomorphism between L2 (T1 ) and
l2 (Z). In particular:

∞
X
• (Riesz-Fischer Theorem) If |cn |2 < ∞ then there is an f ∈ L2 (T1 ) such that
n=0
Z
cn = f (x) e−2πinx dx .
T1

• (Parseval Theorem) If f, g ∈ L2 (T1 ) then

X∞ Z
f (n)b
b g (n) = f (x)g(x) dx .
n=−∞ T1

Also, if f ∈ L2 (T1 ) then lim kf − sN [f ]k2 = 0.

N →∞

5.3. Some Kernels.

5.3.1. Dirichlet Kernel.

N
X
2πinx sin {2π(N + 12 )x}
DN (x) = e = .
sin πx
n=−N

Note:
T
X
DN (x) = N lim D{N (x + n)} ,
T →∞
n=−T

where
Z 1
sin 2πx
D(x) = e2πixt dt =
−1 πx
is the non-periodic Dirichlet Kernel.

17
5.3.2. Fejér Kernel.
N
1 X
FN (x) = Dk (x)
N +1
k=0
N  
X |n|
= 1− e2πinx
N +1
n=−N

sin {π(N + 1)x} 2

 
1
= .
N +1 sin πx
Note:
∞
X
FN (x) = (N + 1) K{(N + 1)(x + n)} ,
n=−∞
where
Z 1  2
2πixt sin πx
K(x) = (1 − |t|) e dt =
−1 πx
is the non-periodic Fejér Kernel.

5.3.3. Poisson Kernel. For 0 < r < 1, the 2π-periodic Poisson kernel Pr : T2π → C is
∞
X ∞
X
(5.13) Pr (θ) = r|n| einθ = 1 + 2 rn cos nθ
n=−∞ n=1
r2 1 + reiθ
 
1−
= =< .
1 − 2r cos θ + r2 1 − reiθ
The 1-periodic version of the Poisson kernel is
∞
X
(5.14) Pr(1-per) (α) = Pr (2πα) = r|n| e2πinα .
n=−∞

In the upper half plane H(+) = {z ∈ C | =(z) > 0} the Poisson kernel is defined
 
(+) i 1 y
(5.15) P (x + yi) = < = Py(+) (x) = ,
πz π x + y2
2

where z = x + yi. Also:

Z ∞ Z ∞
1 −y|t| ixt
(5.16) Py(+) (x) = e e dt = e−2πy|t| e2πixt dt .
2π −∞ −∞

Another relation with the periodic Poisson kernel is

∞
1 X
P (+) θ + 2πn + i log 1r


(5.17) Pr (θ) =
2π n=−∞
X  
= P (+) −i log {reiθ }
all branches
of log

18
or
∞
X
Pr(1-per) (α) = P (+) α + n + i
log 1r


(5.18) 2π
n=−∞
 
X
(+) i
= P − log {re2πiα } ,
2π
all branches
of log

where α = θ/2π.

5.3.4. Conjugate Poisson Kernel.

For 0 < r < 1, the conjugate Poisson kernel Qr : T2π → C is

∞
X ∞
X
(5.19) Qr (θ) = −i sgn (n) r|n| einθ = 2 rn sin nθ
n=−∞ n=1
reiθ
 
2r sin θ 1+
= == .
1 − 2r cos θ + r2 1 − reiθ

6. Integral Transforms

6.1. Fourier Transform. Given a function f : R → C, the Fourier transform of f is

defined in the following way:
Z ∞
(6.1) F[f (x)](t) = f (t) =
b f (x) e−2πixt dx .
−∞
The inverse Fourier transform of F : R → C is
Z ∞
−1
(6.2) F [F (t)](x) = F (t) e2πixt dt .
−∞

Theorem 6.1 (Riemann-Lebesgue Lemma). If f ∈ L1 (R), then its Fourier transform fb(t)
is defined everywhere, uniformly continuous, and lim fb(t) = 0.
t→±∞

Theorem 6.2 (Inversion Theorem). Let f : R → C be a locally integrable function, of

bounded variation in a neighborhood of a point x0 , and such that
1
f (x0 ) = lim {f (x0 + h) + f (x0 − h)} .
2 h→0
If f satisfies either one of the following conditions:

(1) f ∈ L1 (R),
(2) x →
7 f (x)/(1 + |x|) is in L1 (R), and the Fourier integral of f at x0 converges
uniformly on every finite interval of t,

then
Z T
(6.3) f (x0 ) = lim fb(t) e2πix0 t dt .
T →∞ −T

19
6.1.1. Properties of the Fourier Transform.

(1) Linearity.
F[af (x) + bg(x)](t) = afb(t) + bb
g (t) .
(2) Scaling. if a 6= 0,
1 b s 
F[f (ax)](t) = f .
|a| a
(3) Shifting.
F[f (x − b)](t) = fb(t) e−2πibt .
(4) Complex conjugation.
F[f (x)](t) = fb(−t) .
(5) Modulation.
F[f (x) e2πiax ](t) = fb(t − a) .
(6) Differentiation. If f (k) ∈ L1 (R) for 0 ≤ k ≤ n and lim f (k) (t) = 0 for 0 ≤ k ≤ n−1,
|t|→∞
then
F[f (n) (x)](t) = (2πit)n fb(t) .
(7) Product by powers. If x 7→ xk f (x) is in L1 (R), then
(−1)k b(k)
F[xk f (x)](t) = f (t) .
(2πi)n
(8) Convolution. If f, g ∈ L1 (R) then f[ ∗ g ∈ L1 (R), and
F[f g] = fb ∗ gb , F[f ∗ g] = fb b
g.
Here Z ∞
(f ∗ g)(x) = f (u) g(x − u) du .
−∞

6.1.2. Parseval Theorem. If f ∈ L1 (R) ∩ L2 (R) then fb ∈ L1 (R) ∩ L2 (R) and

Z ∞ Z ∞
2
(6.4) |f (x)| dx = |fb(t)|2 dt
−∞ −∞

6.1.3. Extension to L2 (R).

Theorem 6.3 (Plancheret Theorem). (See [9].) Given f ∈ L2 (R) there is an fb ∈ L2 (R)
such that

(1) If f ∈ L1 (R) ∩ L2 (R) then fb is the usual Fourier transform of f .

(2) For every f ∈ L2 (R), kfbk2 = kf k2 .
(3) The mapping f 7→ fb is a Hilbert space isomorphism of L2 (R) onto L2 (R).
(4) If
Z T Z T
−2πixt
(f )T (t) =
b f (x) e dx , fT (x) = fb(t) e2πixt dt ,
−T −T
then
lim kfb − (fb)T k2 = 0 and lim kf − fT k2 = 0 .
T →∞ T →∞

20
Theorem 6.4 (Paley-Wiener Theorem). Suppose f ∈ L2 (R). Then f is the Fourier trans-
form of a function vanishing outside [−σ, σ] if and only if f is the restriction to R of an
entire function of exponential type3 2πσ (see [10, III.4, p. 108]).
Theorem 6.5 (Shannon’s Sampling Theorem). If f : R → R is a function band-limited to
[−σ, σ], i.e., Z σ
f (x) = F (t) e2πixt dt
−σ
with F ∈ L2 ([−σ, σ]), then it can be reconstructed from its values at the points xk = k/2σ,
k ∈ Z, via the formula
∞
X D(σ(x − xk ))
(6.5) f (x) = f (xk )
2
k=−∞
∞
sin (2πσx) X f (xk )
= (−1)k ,
2πσ (x − xk )
k=−∞
where D(x) = sin 2πx/πx is the non-periodic Dirichlet kernel (see bellow), with the series
absolutely convergent and uniformly convergent on compact sets.

Partial Proof. Since fb(t) = F (t) is supported in [−σ, σ], for |t| < σ we have:
X∞
F (t) = F (t + 2σn) .
n=−∞
So, using the Poisson Summation formula:
X∞ Z ∞
F (t) = F (t + 2σu) e2πiuk du
k=−∞ −∞
∞ Z ∞
X 1
= F (s) e2πi(s−t)k/2σ ds (s = t + 2σu)
2σ −∞
k=−∞
∞ ∞
e−2πitk/2σ
X Z
= F (s) e2πisk/2σ ds
2σ
k=−∞ | −∞ {z }
f (k/2σ)
∞
X
k e−2πitk/2σ
= f ( 2σ ) ,
2σ
k=−∞
hence: Z σ
f (x) = F (t) e2πixt dt
−σ
∞ σ
e2πit(x−k/2σ)
X Z
k
= f ( 2σ ) dt
−σ 2σ
k=−∞
∞ k
X
k D(σ(x − 2σ ))
= f ( 2σ ) .
2
k=−∞

3Section 7.1.2.

21
 

6.1.4. Some Fourier Transforms.

(1) Non-periodic Dirichlet Kernel.


1
 if |t| < 1
sin 2πx 1
D(x) = ⇒ D(t)
b = if |t| = 1
πx 2
0 if |t| > 1


(2) Non-periodic Fejér Kernel.

sin πx 2
 
K(x) = ⇒ K(t)
b = (1 − |t|) , (|t| < 1) .
πx
(3) Poisson Kernel in the Upper Half-Plane (y > 0).
1 y [ (+)
Py(+) (x) = 2 2
⇒ Py (t) = e−2πy|t| .
πx +y
(4) Conjugate Poisson Kernel in the Upper Half-Plane (y > 0).
1 x [(+)
Q(+)
y (x) = ⇒ Qy (t) = −i sgn (t) e−2πy|t| .
π x + y2
2

(5) Gaussian.
1 (x−µ)2
Gµ,σ (x) = √ e− 2σ2 ⇒ G
d µ,σ (t) = G0, 2π (t) e
−2πiµt
.
σ 2π σ

6.1.5. More Fourier Transforms.

1
(1) F[sgn(x)](t) = .
πit
1 4
(2) F[log |x|](t) = − .
2|t|
p 1 −2πx|t|
(3) fx (y) = log x2 + y 2 ⇒ fbx (t) = − e (x > 0) .
2|t|
y i
(4) fx (y) = arctan ⇒ fbx (t) = − e−2πx|t| (x > 0) .
x 2t
p y
(5) fx (y) = log(x + iy) = log x2 + y 2 + i arctan ⇒
  x
1 1
fbx (t) = − + e−2πx|t| (x > 0) .
2|t| 2t

6.2. Laplace Transform. The Laplace transform of a function f (t) is defined:

Z ∞
(6.6) F (s) = L[f (t)](s) = f (t)e−st dt .
0
4F[log |x| e−λ|x| ](t) → − 1 as λ → 0+ .
2|t|

22
Theorem 6.6 (Existence of the Laplace transform). If f is a locally integrable
R∞ function
on [0, ∞) and of (real) exponential type γ, then the Laplace integral of f , 0 f (t) e−st dt,
converges for <(s) > γ, and converges uniformly for <(s) ≥ γ1 > γ.

Note: A real function f is said to be of (real) exponential type if

|f (x)| ≤ M eγx , for all x ≥ x0
for some constants M , x0 > 0 and real γ.

6.2.1. Properties of the Laplace Transform.

Here we denote F (s) = L[f (t)](s), G(s) = L[g(t)](s).

(1) Linearity.
L[af (t) + bg(t)](s) = aF (s) + bG(s) .
(2) Dilation. if a > 0,
1 s
L[f (at)](s) = F .
a a
(3) Translation.
L[f (t − a)H(t − a)](s) = e−as F (s) ,
where H is the Heaviside function.
(4) Multiplication by exponential functions.
L[eat f (t)](s) = F (s − a) .
(5) Differentiation. If f is an n differentiable function, f (k) (k = 0, 1, . . . , n − 1) is
of (real) exponential type, limt→0+ f (k) (t) = f (k) (0+ ) exists, and f (n) is locally
integrable on [0, ∞), then its Laplace transform exists and

L[f (n) (t)](s) = sn F (s) − sn−1 f (0) − sn−2 f 0 (0) − · · · − f (n−1) (0) .
(6) Integration.
Z t 
F (s)
L f (x) dx (s) = ,
0 s
if the transform exists.
(7) Multiplication by powers of t. Let f be a locally integrable function whose Laplace
integral converges absolutely and uniformly for <(s) > σ. Then F (s) is analytic in
<(s) > σ, and for n = 0, 1, 2, . . . , <(s) > σ:
d n
 
L[t f (t)](s) = −
n
F (s) ,
ds
 n   n
d d
L t f (t) (s) = − s F (s) ,
dt ds
where − ds s F (s) = − ds (sF (s)) = −F (s) − sF 0 (s).
d d

23
 (8) Division by t. If f is a locally integrable function of (real) exponential type, then
  Z ∞
f (t)
L (s) = F (u) du .
t s

if the transform exists. Also:

1 ∞
Z t  Z
f (x)
L dx (s) = F (u) du .
0 x s s
(9) Periodic functions. If f is a locally integrable function that is periodic with period
T , then:
Z T
1
L[f (t)](s) = f (t) e−st dt .
1 − e−T s 0
(10) Convolution. If f, g : [0, ∞) → R are locally integrable on [0, ∞) and their Laplace
integrals converge absolutely in some half-plane <(x) > α, then their convolution
f ∗ g is locally integrable on [0, ∞), and is continuous if either f or g is continuous.
Additionally it has a Laplace transform give by:
L[(f ∗ g)(t)](s) = F (s) G(s) .
Here Z ∞
(f ∗ g)(x) = f (u) g(x − u) du .
0

6.2.2. Inversion Formula.

Theorem 6.7. Let t0 a real number and f : R → C a function such that

(1) f (t) = 0 for every t < 0,

(2) f (t0 ) = 21 {f (t0 −) + f (t0 +)},
(3) is locally integrable,
(4) is of bounded variation in a neighborhood
R∞ of t0 ,
(5) the Laplace integral of f , F (s) = 0 f (t) e−st dt, converges absolutely on the line
<(s) = c.

Then
Z c+iT
1
(6.7) f (t0 ) = lim PV F (s) est0 ds ,
T →∞ 2πi c−iT

In particular if f is differentiable on (0, ∞) then the inversion formula holds for every
t0 > 0.

7. Complex Analysis

7.1. Some Basic Theorems.

24
7.1.1. Some Definitions and Notations.

A domain or region in the complex plane is a non-empty subset of C that is open and
connected.

A path in the complex plane is a continuous function γ : [a, b] → C. Its image or trajectory
is represented |γ|. A cycle σ is a finite sequence of closed, piecewise smooths paths. Its
winding number about z ∈ C \ |σ| is
Z
1 dξ
n(σ, z) = .
2πi σ ξ − z
A cycle in and open set Ω is said to be homologous to zero if n(σ, z) = 0 for every z ∈ C \ Ω.

Given a function f analytic in the punctured disk ∆∗ (z0 , r) = {z ∈ C : 0 < |z − z0 | < r}

and with an isolated singularity at z0 , it can be represented in ∆∗ (z0 , r) by a Laurent series
X∞
f (z) = an (z − z0 )n .
n=−∞
The coefficient a−1 is called residue of f at z0 , and represented Res(z0 , f ).

An analytic function defined on the whole C is called entire.

7.1.2. Exponential Type. An entire function F : C → C is said to be of exponential type τ if

log |F (z)|
lim sup =τ < ∞
|z|→∞ |z|

7.1.3. Semicontinuity. A function f with values in the real or extended real line is lower
semicontinuous if {x | f (x) > x0 } is open for every real x0 . A function f with values in the
real or extended real line is upper semicontinuous if {x | f (x) < x0 } is open for every real
x0 .

7.1.4. Some Theorems.

Theorem 7.1 (Liouville’s Theorem). Every Bounded entire function is constant.
Theorem 7.2 (Maximum Principle). If f is analytic in a domain D and |f | attains its
maximum in D, then f is constant.
Corollary 7.3. If f is analytic in a bounded domain D and continuous in D then |f |
attains its maximum at some point on the boundary of D.
Theorem 7.4. If fn are analytic in the domain D and fn → f uniformly on compact
subsets of D then f is analytic in D and fn0 → f 0 uniformly on compact subsets of D.
Theorem 7.5 (Cauchy’s Integral Formula). Suppose that f is analytic in an open set Ω
and that σ is a cycle in Ω which is homologous to zero in this set. Then
Z
1 f (ξ) dξ
(7.1) n(σ, z)f (z) =
2πi σ ξ − z
for every z ∈ Ω \ |σ|.

25
Theorem 7.6 (Residue Theorem). Suppose that f is analytic modulo isolated singularities
in an open set Ω, that E is the singular set of f in Ω, and that σ is a cycle in Ω \ E which
is homologous to zero in Ω. Then
Z X
(7.2) f (z) dz = 2πi n(σ, z) Res(z, f ) .
σ z∈E

Theorem 7.7 (Paley-Wiener Theorems).

(1) Suppose f is analytic in the upper half-plane and

Z ∞
sup |f (x + iy)|2 dx = C < ∞ .
0<y<∞ −∞

Then there exists an F ∈ L2 (0, ∞) such that

Z ∞
f (z) = F (t) e2πitz dt , =(z) > 0
0
and Z ∞
|F (t)|2 dt = C .
0
(2) Suppose that f is an entire function such that
|f (z)| ≤ Ce2πA|z|
for some positive constants A, C and for all z, and
Z ∞
|f (x)|2 dx < ∞ .
−∞

Then there exists an F ∈ L2 ([−A, A]) such that

Z A
f (z) = F (t)e2πitz dt
−A
for all z.

Proof. See [9, 19.2–19.3]. 

7.2. Harmonic Analysis. Here we use the notations of section 5 for T1 and T2π . Also we
denote the unit disk U = ∆(0, 1).

7.2.1. Harmonic Functions. Let D be a complex domain. A function u : D → R is harmonic

if it verifies Laplace’s equation
∆u = uxx + uyy = 0 .
Given u harmonic in a domain D, a harmonic conjugate is function v : D → R such that
u + vi is analytic in D. Two harmonic conjugates of a given function differ in a constant.
Theorem 7.8. Let D be a complex domain. Every harmonic function in D has a harmonic
conjugate if and only if D is simply connected.

26
7.2.2. Mean Value Property. A continuous function u : Ω → R has the Mean Value Property
in the open set Ω if for each z ∈ Ω there is a ρ > 0 such that and
Z Z
iθ
(7.3) u(z) = u(z + re ) dθ = u(z + re2πiα ) dα .
T2π T1

for every 0 < r < ρ (the disk ∆(z, ρ) = {z 0 ∈ C : |z 0 − z| < ρ} is supposed to be contained
in Ω.)

Every harmonic function in an open set Ω has the mean value property with any ρ such
that the disk ∆(z, ρ) is contained in Ω. The converse also is true:

Theorem 7.9. A continuous real function u defined in an open set Ω is harmonic in Ω if

and only if it has the mean value property in Ω.

7.2.3. Functions Harmonic in Annuli.

Theorem 7.10. Suppose a function u is harmonic in the annulus D = {z ∈ C | a <

|z − z0 | < b}, where 0 ≤ a < b ≤ ∞. Then there exist constants c and d such that
Z
u(z0 + reiθ ) dθ = c log r + d .
T2π

7.2.4. Subharmonic Functions. If a function verifies

Z
(7.4) u(z) ≤ u(z + reiθ ) dθ
T2π

in place of (7.3), it is called subharmonic. The precise definition is as follows ([9, def. 17.1]):

Definition 7.11. A function u : Ω → [−∞, ∞) defined in an open set Ω ⊆ C is said to be

subharmonic if:

(1) u is upper
Z semicontinuous in Ω.
(2) u(z) ≤ u(z + reiθ ) dθ for every closed disk ∆(z, r) ⊂ Ω.
T2π
(3) None of the integrals above is −∞.

A subharmonic function always lies below any harmonic function with which it shares
the same boundary values.

A twice continuously differentiable function w : Ω → R in an open set Ω is subharmonic

in Ω if and only if ∆w ≥ 0 throughout Ω.

Theorem 7.12. If u is harmonic (subharmonic) in a domain D and attains its maximum

in D, then u is constant.

Corollary 7.13. If u is harmonic (subharmonic) in a bounded domain D and continuous

in D then u attains its maximum at some point on the boundary of D.

27
7.2.5. Poisson Integrals. Here we use the following notations (z = reiθ ):

(1) Convolution. Given f, g : T → C, where T = T1 or T = T2π , the convolution of f

and g is: Z
f ∗ g(x) = f (t)g(x − t) dt .
T
(2) Cauchy Kernel.
1
C(z) =
1−z
1
Cr (θ) = .
1 − reiθ
(3) H Kernel.
1+z
H(z) = 2C(z) − C(0) =
1−z
1 + reiθ
Hr (θ) = 2Cr (θ) − C0 (θ) = .
1 − reiθ
(4) Poisson Kernel.
P (z) = < {H(z)} = C(z) + C(z) − C(0)
1 − r2
Pr (θ) = Cr (θ) + C r (θ) − C0 (θ) = .
1 − 2r cos θ + r2
(5) Conjugate Poisson Kernel.

Q(z) = = {H(z)} = −i C(z) − C(z)
2r sin θ
Qr (θ) = .
1 − 2r cos θ + r2
(6) Poisson Integral. If h : T2π → R is integrable, its Poisson integral is
Z
P [h](reiθ ) = h(t) Pr (θ − t) dt = h ∗ Pr (θ) .
T2π

The main result is the following:

Theorem 7.14. Suppose u is harmonic in the unit disk U and continuous on U . Then u is
the Poisson integral of its restriction to ∂U , and is the real part of the holomorphic function
eit + z
Z
(7.5) f (z) = it − z
u(eit ) dt .
T2π e

Assume u, v : U → R, v(0) = 0, f = u + iv is holomorphic in U and continuous in U .

Then (we use the notation fr (θ) = f (reiθ ), etc.):

(1) fr = f1 ∗ Cr (Cauchy’s Integral Formula.)

(2) fr = u1 ∗ Hr .
(3) ur = u1 ∗ Pr (Poisson’s Integral Formula.)
(4) vr = u1 ∗ Qr .

28
7.2.6. Other results.
Theorem 7.15 (Harnack’s Theorem). Let {un } be a sequence of harmonic functions in a
domain D.

(1) If un → u uniformly on compact subsets of D, then u is harmonic in D.

(2) If the sequence is nondecreasing then either {un } converges uniformly on compact
subsets of D or un (z) → ∞ for every z ∈ D.

7.3. H p Spaces. Here we use the notation:

H(D) = {f : D → C | f is analytic in D}.

h(D) = {f : D → R | f is harmonic in D}.
U = ∆(0, 1) (unit disk).

7.3.1. Integral Means. For f : U → C we define ([3]):

Z 1/p
iθ p
(1) Mp (r, f ) = |f (re )| dθ , 0 < p < ∞.
T2π

(2) M∞ (r, f ) = max |f (reiθ )|.

0≤θ<2π
Z 
+ iθ
(3) M0 (r, f ) = exp log |f (re )| dθ .
T2π

(4) kf kp = sup Mp (r, f ), 0 ≤ p ≤ ∞.

0≤r<1

(5) H p = {f ∈ H(U ) : kf kp < ∞}.

(6) hp = {f ∈ h(U ) : kf kp < ∞}.

Note: p ≤ q ⇒ H q ⊆ H p .

For p = 0, N = H 0 is the Nevanlinna class of functions of bounded characteristic. If

u ∈ H(U ), then u ∈ N if and only if u is the quotient of two bounded analytic functions.
Proposition 7.16. A function f ∈ H(U ) is the Poisson integral of a function ϕ ∈ Lp (T2π )
iff f ∈ H p .

7.3.2. Poisson-Stieltjes Integrals. A Poisson-Stieltjes integral is a function in U of the form:

Z
iθ iθ
(7.6) u(re ) = P [dϕ](re ) = Pr (θ − t) dϕ(t) ,
T2π
where Pr (θ) is the Poisson kernel, ϕ : T2π → C is of bounded variation, and the integral is
of Stieltjes type.5
Z Z x0 +2π
5 1
f (x) dϕ(x) = f (x) dϕ(x) for any real x0 .
T2π 2π x0

29
Theorem 7.17. The following three classes of functions in U are identical:

(1) Poisson-Stieltjes integrals.

(2) Differences of two positive harmonic functions.
(3) h1 .
Theorem 7.18. Assume ϕ : T2π → C is of bounded variation and the following symmetric
derivative exists:
ϕ(θ0 + t) − ϕ(θ0 − t)
Dϕ(θ0 ) = lim .
t→0 2t
Let u be the Poisson-Stieltjes integral u(z) = P [ϕ](z). Then u(z) → Dϕ(θ0 ) as z → eiθ0
along any path not tangent to the unit circle.

7.3.3. Harmonic Conjugation. The harmonic conjugate of a function u ∈ h(U ) is another

function v such that uP+ iv is analytic in U . It is “normalized” if v(0) = 0. Note: if
f (z) = u(z) + iv(z) = ∞ n
n=0 cn z , cn = an − ibn , then:
∞
X
u(z) = rn (an cos nθ + bn sin nθ) ,
n=0
(7.7) ∞
X
v(z) = rn (−bn cos nθ + an sin nθ) .
n=0

Also:

∞
X ∞
X
(7.8) u(z) = c0n rn eniθ , v(z) = c00n rn eniθ ,
n=−∞ n=−∞

where c00 = <(c0 ), c000= =(c0 ), and

( (
0 cn /2 (n > 0) cn /2i (n > 0)
(7.9) cn = , c00n = .
cn /2 (n < 0) −cn /2i (n < 0)

The relation between the coefficients of u and v for n 6= 0 is:

(7.10) c00n = −i sgn(n)c0n .
Theorem 7.19. Let v be the harmonic conjugate of u ∈ h(U ).

(1) (M. Riesz) For 1 < p < ∞, u ∈ hp ⇒ v ∈ hp . Furthermore there is a constant Ap

depending only on p such that
Mp (r, v) ≤ Ap Mp (r, u), 0≤r<1
for all u ∈ hp .
(2) (Kolmogorov) If u ∈ h1 then v ∈ hp for every 0 < p < 1. Furthermore there is a
constant Bp depending only on p such that
Mp (r, v) ≤ Bp Mp (r, u), 0≤r<1
for all u ∈ h1 .

30
 (3) If both u, v ∈ h1 then the function ϕ such that u(z) = P [ϕ](z) is absolutely contin-
uous.

7.3.4. H p Spaces Over General Domains. Here D is a simply connected domain with at
least two boundary points.

(1) kf k∞ = sup |f (z)|.

z∈D

(2) H ∞ (D) = {f ∈ H(D) : kf k∞ < ∞}.

(3) H p (D) = {f ∈ H(D) : |f |p has a harmonic majorant in D} (0 < p < ∞). In this
case the norm can be defined as kf kp = u(z0 )1/p , where z0 ∈ D is fixed and u is the
least harmonic majorant of |f |p .

(4) A function f analytic in D belongs to the class E p (D) if there exists a sequence
of rectifiable Jordan curves Cn in D tending to ∂D (Cn eventually surrounds each
compact subdomain of D) such that
Z
|f (z)|p |dz| ≤ M < ∞ .
Cn
Note: for the unit disk, E p (U ) = H p (U ).

7.3.5. H p Spaces On the Upper Half-Plane. (See [3]).

Here H(+) = {z ∈ C : =(z) > 0} represents the upper half-plane; H(−) = {z ∈ C :

=(z) < 0} is the lower half-plane.

For the upper half-plane we have that E p (H(+) ) ⊂ H p (H(+) ) properly. On the other
hand for 0 < p < ∞ we have E p (H(+) ) = Hp (H(+) ),

(7.11) Hp (H(+) ) = {f ∈ H(H(+) ) | ∀y > 0, fy ∈ Lp (R); sup Mp (y, f ) < ∞} ,

0<y<∞

where fy (x) = f (x + iy) and

Mp (y, f ) = kfy kp .
Theorem 7.20. (Poisson Integral in Hp (H(+) ).)

If f ∈ Hp (H(+) ), 1 ≤ p ≤ ∞, then
Z ∞
(7.12) f (x + iy) = P (+) ((x − t) + iy)f (t) dt ,
−∞
where
1 y
P (+) (x + iy) =
π x + y2
2

is the Poisson kernel for the upper half-plane.

Conversely, if h ∈ Lp (R), 1 ≤ p ≤ ∞, and

Z ∞
f (x + iy) = P (+) ((x − t) + iy)h(t) dt
−∞

31
is analytic in H(+) , then f ∈ Hp (H(+) ), and its boundary function

f0 (x) = lim f (x + iy)

y→0+

verifies f0 (x) = h(x) a.e.

Proposition 7.21. If f ∈ Hp , 0 < p < ∞, then

lim kfy kp = kf0 kp .

y→0+

Theorem 7.22. (Cauchy Integral in Hp (H(+) ).)

If f ∈ Hp (H(+) ), 1 ≤ p < ∞, then

Z ∞
1 f (t)
(7.13) f (z) = dt , =(z) > 0 ;
2πi −∞ t−z
and the integral vanishes for =(z) < 0.

Conversely, if h ∈ Lp (R), 1 ≤ p < ∞, and

Z ∞
1 h(t)
dt ≡ 0 , =(z) < 0 ,
2πi −∞ t − z

then for =(z) > 0 the integral defines a function f ∈ Hp (H(+) ), and its boundary function

f0 (x) = lim f (x + iy)

y→0+

verifies f0 (x) = h(x) a.e.

Proposition 7.23. [4, ex. 8.10] Every function f ∈ L2 (R) is uniquely expressible in the
form f = f(+) + f(−) , f(+) ∈ H2 (H(+) ), f(−) ∈ H2 (H(−) ), where
Z ∞
1 f (x)
f(+) (w) = dx =(w) > 0
2πi −∞ x − w
Z ∞
1 f (x)
f(−) (w) = − dx =(w) < 0 .
2πi −∞ x − w

2x
Example: For f (x) = we have
x2 +1
Z ∞ 2x
1 x2 +1 1
f(+) (w) = dx = =(w) > 0
2πi −∞ x −w w+i
Z ∞ 2x
1 x2 +1 1
f(−) (w) = − dx = =(w) < 0 ,
2πi −∞ x−w w−i
hence
2x 1 1
= + .
x2 + 1 x+i x−i

32
7.3.6. H p Spaces On the Right Half-Plane. (See [4]).

Here H(
) = {z ∈ C : <(z) > 0} represents the right half-plane. Also

(7.14) Hp (H(
) ) = {f ∈ H(H(
) ) | ∀x > 0, fx ∈ Lp (R); sup kfx kp < ∞} ,

0<x<∞
where fy (x) = f (x + iy).

The Poisson kernel and its conjugate for the right half-plane are (z = x + iy, x > 0):
 
(
) 1 1 x
(7.15) P (z) = < =
πz π x2 + y 2
Z ∞
(
)
= Px (y) = e−2πx|t| e2πiyt dt ,
−∞
 
(
) 1 1 −y
(7.16) Q (z) = = =
πz π x2 + y 2
Z ∞
= Q(
)
x (y) = i sgn(t) e−2πx|t| e2πiyt dt .
−∞

(
)
The relation between Pr and Px is as follows. Consider the following linear fractional
map from the right half-plane to the unit disk:6
z−1
(7.17) w(z) =
z+1
iθ
Assume w(iy) = e . Then
dθ dy
=− ,
2π π(1 + y 2 )
so the normalized Lebesgue measure dθ/2π on the circle corresponds to the Cauchy proba-
bility measure dy/π(1 + y 2 ) on the imaginary axis.

If g : ∂U → C is measurable and f (iy) = g(eiθ ) = g(w(iy)) then g is Lebesgue-integrable

if and only if f is integrable respect to the measure dy/π(1 + y 2 ), and
Z Z ∞
iθ 1
g(e ) dθ = f (iy) · dy .
T2π −∞ π(1 + y 2 )
0
On the other hand, writing w(x + iy) = reiθ , w(iy 0 ) = eiθ :
2
Pr (θ − θ0 ) = π (1 + y 0 ) Px(
) (y − y 0 ) ,
so the Poisson formula for the right half-plane becomes as shown in the following result:
Theorem 7.24. Let h be a measurable function on the imaginary axis which is integrable
respect to the measure (1 + t2 )−1 dt. Define f in the right half-plane by
1 ∞
Z
x
(7.18) f (x + iy) = h(it) 2 dt
π −∞ x + (y − t)2
Z ∞
= h(it) Px(
) (y − t) dt .
−∞

6A similar analysis works for P (+) using w(z) = z−i

instead.
x z+i

33
Then f is harmonic and has non-tangential limits which exist and agree with h at almost
every point of the imaginary axis.
Theorem 7.25. Assume p ≥ 1 and h ∈ Lp (R). Let f be the harmonic function in the right
half-plane defined by
Z ∞
x
f (x + iy) = h(t) 2 dt = h ∗ Px(
) (y) = fx (y) .
−∞ x + (y − t)2
Then:

(1) fx ∈ Lp (R) for every x > 0.

(2) The Lp -norms kfx kp are bounded for x > 0. In fact, kfx kp is a decreasing function
of x for x > 0.
(3) kfx − hkp → 0 as x → 0+ .
(4) f (ξ) → 0 uniformly as ξ → ∞ inside any fixed half-plane <(ξ) ≥ δ > 0.

7.3.7. Kernels in the Right Half-Plane. We have already seen by looking at the effects
of the linear fractional map (7.17) that the Poisson kernel for the unit circle becomes
P (
) (z) = <(1/πz) in the right half-plane. On the other hand the equation
dθ0
Z 2π
iθ 0 1
g(re ) = g(eiθ ) 0
0 1 − rei(θ−θ ) 2π
is transformed into the equation
∞
dy 0
Z
1
f (x + iy) = f (iy 0 ) iy 0 −1
,
−∞
x+iy−1
− π(1 + iy 0 )2
x+iy+1 iy 0 +1
z−1
where f (z) = g(w(z)) = g( z+1 ). Hence the Cauchy kernel for the unit circle becomes
1 1
(7.19) Cx(
) (y, y 0 ) = iy 0 −1
x+iy−1
− π(1 + iy 0 )2
x+iy+1 iy 0 +1

in the right half-plane. The H kernel then becomes:

iy 0
 
(
) 0 (
) 0 1 1
(7.20) 2Cx (y, y ) − C1 (0, y ) = −
π x + (y − y 0 )i 1 + y 02
iy 0
= Hx(
) (y − y 0 ) − ,
π(1 + y 02 )
where
1
(7.21) H (
) (z) = P (
) (z) + iQ(
) (z) =
Z ∞ πz
= Hx(
) (y) = 2 e−2π(x+iy)t dt .
0
So the H kernel for the unit circle becomes what we have called the H kernel for the right
half-plane corrected with the term −iy 0 /π(1 + y 02 ).

We know that an analytic function in the unit circle that is real at zero can be recovered
from the boudary values of its real part by convolution with the H kernel. The corresponding

34
result for an analytic function f (z) = u(z) + iv(z) in the right half-plane such that v(1) = 0
can be expressed with the following equation:
Z ∞  
1 it
(7.22) f (x + iy) = u(it) − dt
−∞ π{x + (y − t)i} π(1 + t2 )
The integral converges as long as t 7→ u(it)/(1 + t2 ) is in L1 (R). If t 7→ u(it)/(1 + |t|) is in
L1 (R) then the equation
Z ∞
1
(7.23) f (x + iy) = u(it) dt
−∞ π{x + (y − t)i}
Z ∞
= u(it) Hx (y − t) dt
−∞

yields essentially the same function but normalized so that

v(∞) = lim ={f (x + iy)} = 0 .
x→∞

In such case: Z ∞
it
v(1) = u(it) dt .
−∞ π(1 + t2 )

8. Some Special Functions

8.1. Bernoulli periodic functions. The Bernoulli polynomials B∗n (x) can be defined in
various ways.7 The following are two of them ([8, ch. 1], [1]):

(1) By a generating function:

∞
t ext X
∗ tn
(8.1) = B n (x)
et − 1 n!
n=0

(2) By the following recursive formulas (n ≥ 1):

(8.2) B∗0 (x) = 1 ,
(8.3) B∗0 ∗
n (x) = n Bn−1 (x) ,
Z 1
(8.4) B∗n (x) dx = 0 .
0

The first Bernoulli polynomials are:

B∗0 (x) = 1
B∗1 (x) = x − 1
2
B∗2 (x) = x2 − x + 1
6
B∗3 (x) = x3 − 23 x2 + 12 x

7Here we use the notation B∗ for the Bernoulli polynomials, and reserve the notation B for the Bernoulli
n n
periodic functions.

35
 The Bernoulli numbers are Bn = B∗n (0), and the Bernoulli periodic functions are usually
defined Bn (x) = B∗n (hxi). However here we normalize B1 defining B1 (k) = 0 instead of
−1/2 for k integer, so that B1 coincides with the normalized sawtooth function:
(
hxi − 21 if x 6∈ Z
(8.5) B1 (x) =
0 if x ∈ Z,
where hxi = x − bxc = fractional part of x, bxc = integer part of x (greatest integer
≤ x). Also we will leave B0 (k) undefined for k integer—in
P∞ fact B0 should be defined as the
distribution B0 (x) = 1 − δper (x), where δper (x) = k=−∞ δ(x − k) is the periodic Dirac’s
delta.

8.1.1. Properties of the Bernoulli Periodic Functions. (n ≥ 1):

1. B1 (x) = sawtooth function (eq. 8.5).

2. B0n (x) = n Bn−1 (x) for n > 2 or x 6∈ Z.

Z 1
3. Bn (x) dx = 0.
0

8.1.2. Fourier expansions. The Fourier expansion for the Bernoulli periodic functions is
(n ≥ 1):
∞
n! X e2πikx
(8.6) Bn (x) = − ,
(2πi)n kn
k=−∞
k6=0
so:

0 if k = 0,
(8.7) B
b n (k) = n!
− otherwise.
(2πik)n

This result also holds in the distributional sense for n = 0.

8.2. Polylogarithms. The Bernoulli periodic functions appear naturally in expressions

involving polylogarithms, together with the so called Clausen functions (see [6]):
∞
X cos(kθ)
(8.8) Cl2n−1 (θ) = ,
k 2n−1
k=1
∞
X sin(kθ)
(8.9) Cl2n (θ) = ,
k 2n
k=1
for n ≥ 1.

To be more precise, the polylogarithms can be defined by the series:

∞
X zk
(8.10) Lin (z) =
kn
k=1

36
for n ≥ 0, |z| < 1, or by the following recursive relations:
z
(8.11) Li0 (z) = ,
1−z
Z z
Lin−1 (ξ)
(8.12) Lin (z) = dξ (n ≥ 1),
0 ξ
in C \ [1, ∞). Note that Li1 (z) = − log(1 − z) is the usual logarithm. Li2 (z) is the diloga-
rithm.8

A generating function is
∞ ∞
z e(t+1)u
Z X
(8.13) u 2
du = Lin (z) tn .
0 (e − z)
n=0

The Bernoulli periodic functions and the Clausen functions are related to the polyloga-
rithms in the following way:
2in!
(8.14) − Lin (e2πix ) = An (x) + i Bn (x),
(2πi)n
for x 6∈ Z, where
n+1 2n!
(8.15) An (x) = (−1)b 2
c
Cln (2πx).
(2π)n
We will call the An (x) conjugate Bernoulli periodic functions. The first ones are A0 (x) =
cot πx, A1 (x) = π2 log (2 | sin πx|), . . .

The series (8.10) converges for |z| = 1 if n ≥ 2.

For n = 1 both Li1 (e2πix ) and Cl1 (2πx) diverge at x = 0, but

∞
X sin 2πkx
(8.16) − πi B1 (x) = Li1 (e2πix ) − Cl1 (2πx) = i ,
k
k=1
and the series becomes zero for x = 0, so our definition B1 (0) = 0 allows (8.16) to hold also
for x = 0.

For n = 0, x 6∈ Z, we easily compute

i 1
(8.17) Cl0 (x) = −i Li0 (e2πix ) − = cot(πx).
2 2
Also by definition Cl0 (k) = 0 for k ∈ Z. Hence,
(8.18) ={Li0 (e2πix )} = Cl0 (x)
for every x ∈ R.

Finally we observe that for y > 0

(Z )
x
1
Li0 (e2πi(u+yi) ) du = − arg e2πy − e2πix ,

(8.19) <
− 1 2π
2

8For some authors the dilogarithm is Li (1 − z).

2

37
which tends to − 12 B1 (x) as y → 0+ for every x ∈ R, hence
n o 1 1 1
(8.20) lim < Li0 (e2πi(x+yi) ) = − B0 (x) = − + δper (x)
y→0+ 2 2 2
(where δper is the periodic Dirac’s delta) in the distributional sense.

We also note that the Bernoulli periodic functions and their conjugates have harmonic
extensions to the upper half plane, given by the formula:
2in!
(8.21) − Lin (e2πiz ) = An (z) + i Bn (z),
(2πi)n
for =(z) > 0.

9. Summation Formulas

9.1. The Euler-Maclaurin Summation Formula.

Rb
Theorem 9.1. Let f : [a, b] → C be q times differentiable, a |f (q) (x)| dx < ∞. Then for
1 ≤ m ≤ q:
X 0 Z b
f (n) = f (x) dx
a≤n≤b a
m
X (−1)k  
(9.1) + Bk (b) f (k−1) (b) − Bk (a) f (k−1) (a)
k!
k=1
(−1)m+1 b
Z
+ Bm (x) f (m) (x) dx,
m! a
X 0
where f (k) for a < b represents a summation modified by taking only half of f (k)
a≤k≤b
when k = a or k = b.

Proof. (See [5]) We have

X 0
Z b
f (n) = f (x) d(x − B1 (x))
a≤k≤b a
(9.2) Z b Z b
= f (x) dx − f (x) d B1 (x)
a a
Next, integrate by parts successively the last integral on the right hand side of (9.2). 

9.1.1. Sum of Powers. As an example of application of the Euler-Maclaurin summation

formula, we give the sum of the first m rth powers:
m
X
S(m, r) = nr = 1r + 2r + 3r + · · · + mr .
n=1

38
Here f (x) = xr , so f (k) (x) = r!x(r−k) /(r − k)! for k = 0, 1, . . . , r, f (k) (x) = 0 for k > r, and
X 0 Z m
nr = xr dx
0≤n≤m 0

r+1
X (−1)k  
+ Bk (m) f (k−1) (m) − Bk (0) f (k−1) (0)
k!
k=1
r+1
mr+1 X (−1)k r! Br+1 (0)
= + Bk (0) mr−k+1 −
r+1 k! (r − k + 1)! r+1
k=1
r+1
mr+1 1 X r+1 Br+1 (0)
(−1)k Bk (0) mr−k+1 −


= + k
r+1 r+1 r+1
k=1

Hence
X 0 mr
S(m, r) = nr +
2
0≤n≤m
( r+1 )
1 X 
(−1)k r+1 r−k+1


= k Bk m − Br+1
r+1
k=0
where Bk are the Bernoulli numbers B0 = 1, B1 = −1/2, Bk = Bk (0) for k > 1.

9.2. The Poisson Summation Formula. Here f represents a function f : R → C.

Theorem 9.2. If f is absolutely integrable over R then
N
X
fper (x) = lim f (x + n)
N →∞
n=−N

exists for a.e. x and is periodic: fper (x + 1) = fper (x).

Furthermore:
Z 1
(9.3) fb(k) = fper (x) e−2πikx dx .
0

i.e., fb(k) = fbper (k) (the Fourier transform fb(k) of f coincides with the k-th Fourier coeffi-
cient of fper .)

Proof. We have (LMCT):

Z 1 N
X N Z
X n+1
lim |f (x + n)| dx = lim |f (x)| dx
0 N →∞ n=−N N →∞
n=−N n
Z N +1
= lim |f (x)| dx
N →∞ −N
Z ∞
= |f (x)| dx < ∞ ,
−∞

39
so the series converges absolutely at almost every x. Also, if the series converges at x, then
fper (x + 1) − fper (x) = lim {f (x + N + 1) − f (x − N )} = 0 ,
N →∞
so fper is periodic. Furthermore, since
N
X N
X
−2πikx
f (x + n) e ≤ lim |f (x + n)| ,
N →∞
n=−N n=−N
and the right hand side being integrable in T1 , by the LDCT we have:
Z 1
fper (k) =
b fper (x) e−2πikx dx
0
Z 1 N
X
= lim f (x + n) e−2πikx dx
0 N →∞ n=−N

N Z
X n+1
= lim f (x) e−2πikx dx
N →∞
n=−N n
Z N +1
= lim f (x) e−2πikx dx
N →∞ −N
Z ∞
= f (x) e−2πikx dx
−∞
= fb(k) .

Theorem 9.3 (Poisson Summation Formula). If f is absolutely integrable over R, of
bounded variation and normalized in the sense that for every x,
1
f (x) = lim {f (x + h) + f (x − h)} ,
2 h→0
then
∞
X X T
(9.4) f (x + n) = lim fb(k) e2πikx .
T →∞
n=−∞ k=−T

The following is a formulation covering the case in which f is not absolutely integrable
(see [11, II.13]).
Theorem 9.4. Assume that

(1) f is integrable over every finite interval;

Z n+1
(2) lim |f (x)| dx = 0 ;
n→±∞ n
(3) the function
Z n+1
∗
f (x) = f (x) − f (u) du for n ≤ x < n + 1, (n ∈ Z)
n
is absolutely integrable over R.

40
 Then
Z M
(9.5) fb(k) = lim f (x) e−2πikx dx ,
M →∞ −M

exists for every k ∈ Z \ {0}, and if g ∗ is the function

( N )
X Z N
∗
g (x) = lim f (n + x) − f (u) du
N →∞ −N
n=−N

then g ∗ (x) exists for a.e. x, is periodic: g ∗ (x + 1) = g ∗ (x), and the k-th Fourier coefficient
of g ∗ is
Z 1 (
∗ −2πikx 0 if k = 0
gb∗ (k) = g (x) e dx = b
0 f (k) if k ∈ Z \ {0}

10. Miscelanea

10.1. Various Results.

10.1.1. Here we want to justify the following expression, which can be considered as a
formal application of the Poisson Summation Formula:
Proposition 10.1. For x, y ∈ R, x > 0:
( N
N + 21
Z )
X
(10.1) lim log {x + (y + n)i} − log {x + (y + u)i} du
N →∞ −N − 12
n=−N
T  
X 1 1
= lim − + e−2πx|k| e2πiyk
T →∞ 2|k| 2k
k=−T
k6=0
∞ −2πk(x+iy)
X e
=−
k
k=1
 
= log 1 − e−2π(x+iy) .

Proof. We define
N
X Z N + 12
SN (x + iy) = log {x + (y + n)i} − log {x + (y + u)i} du .
n=−N −N − 12

Using the Euler-Maclaurin summation formula:

1
Z N+ 2
i
(10.2) SN (x + iy) = B1 (u) du ,
1
−N − 2 x + (y + u)i

41
where B1 (u) = hui − 12 for u ∈ / Z, hui = u − buc, and we have used B1 (−N − 21 ) =
1
B1 (N + 2 ) = 0 for N integer. Also, using one more step of the Euler-Maclaurin summation
formula, or integrating (10.2) by parts:
( )
1 i i
(10.3) SN (x + iy) = B2 ( 12 ) −
2 x + i(y + N + 12 ) x + i(y − N − 12 )
1
1 N+ 2
Z
1
+ B2 (u) du ,
2 −N − 1 {x + i(y + u)}2
2

where B2 (u) = hxi2 − hxi + 16 .

First, (10.3) shows that the sequence converges:

S(x + iy) = lim SN (x + iy)
N →∞
1 ∞
Z
1
= B2 (u) du .
2 −∞ {x + i(y + u)}2
From here we also get:
lim S(x + iy) = 0 .
x→∞

Next changing t = −u and taking the limit as N → ∞ in (10.2), we get formally:

Z ∞
1
S(x + iy) = i B1 (−t) dt
−∞ x + (y − t)i
Z ∞
= iπ B1 (−t) Hx(
) (y − t) dt ,
−∞
(
) (
) (
)
where Hx (y) = Px (y) + iQx (y)
= 1/π(x + iy) is the H kernel for the right half-plane.
Unfortunately the integral does not converge absolutely, so we rewrite it like this:
Z ∞  
1 1 it
S(x + iy) = B1 (−t) − dt
πi −∞ π{x + (y − t)i} π(1 + t2 )
Z N+ 1
2 it
+ lim B1 (−t) dt .
N →∞ −N − 1 π(1 + t2 )
2

Since t 7→ B1 (t)/(1 + t2 )
is in L1 (R),
the first integral converges absolutely and equals
an analytic function in the right half-plane such that the boundary values of its real part
coincide with B1 (−y) almost everywhere and is real at z = 1 (sec. 7.3.7). The last term,
being the difference of two converging sequences, converges and amounts to adding an
imaginary constant to that function.

Next, using the harmonic extension of B1 to the upper half-plane we have:

 
i −2πz
B1 (iz) = < L1 (e ) , <(z) > 0 ,
π
where L1 (z) = − log(1 − z) (z ∈ C \ [1, ∞)) is the first polylogarithm. Hence
1 i
S(x + iy) = L1 (e−2πz ) + iC ,
πi π

42
where C is a real constant. Since L1 (e−2π(x+iy) ) → 0 as x → ∞ the constant must be zero,
so:
S(x + iy) = −L1 (e−2π(x+iy) )
 
−2π(x+iy)
= log 1 − e
for x > 0.

References
[1] Tom M. Apostol. Introduction to Analytic Number Theory. Springer-Verlag, New York, 1976.
[2] Robert G. Bartle. A Modern Theory of Integration, volume 32 of Graduate Studies in Mathematics.
American Mathematical Society, Providence, Rhode Island, 2002.
[3] Peter L. Duren. Theory of H p Spaces. Dover, 2000.
[4] Kenneth Hoffman. Banach Spaces of Analytic Functions. Prentice-Hall, Inc., Englewood Cliffs, N.J.,
1962.
[5] Ralph P. Boas, Jr. Partial sums of infinite series, and how they grow. Amer. Math. Monthly, 84:237–258,
1977.
[6] Leonard Lewin. Polylogartihms and Associated Functions. North Holland, 1981.
[7] Robert M. McLeod. The Generalized Riemann Integral, volume 20 of The Carus Mathematical Mono-
graphs. The Mathematical Association of America, 1980.
[8] Hans Rademacher. Topics in Analytic Number Theory. Springer-Verlag, 1973.
[9] Walter Rudin. Real and Complex Analysis. McGraw-Hill, Inc., 1987.
[10] Elias M. Stein and Guido Weiss. Introduction to Fourier Analysis on Euclidean Spaces. Princeton
University Press, Princeton, NJ, 1971.
[11] Antoni Zygmund. Trigonometric Series, volume I and II. Cambridge University Press, second edition,
1968.

43