See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/266229957

Lecture Notes Methods of Mathematical Physics MATH 536

Article · January 2000

CITATION READS

1 1,566

1 author:

Ivan Avramidi
New Mexico Institute of Mining and Technology
116 PUBLICATIONS   1,784 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Yang Mills theory on spheres. View project

Lecture Notes View project

All content following this page was uploaded by Ivan Avramidi on 15 January 2015.

The user has requested enhancement of the downloaded file.

 Lecture Notes
in Mathematical Physics
Ivan Avramidi
New Mexico Institute of Mining and Technology

Socorro, NM 87801

October 19, 2000

Contents

1 Functional Analysis 1
1.1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.1 Metric Spaces . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.2 Vector Spaces . . . . . . . . . . . . . . . . . . . . . . . 4
1.1.3 Normed Linear Spaces . . . . . . . . . . . . . . . . . . 6
1.1.4 Notes on Lebesgue Integral . . . . . . . . . . . . . . . . 8
1.2 Hilbert Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.2.1 Geometry of Hilbert Space . . . . . . . . . . . . . . . . 11
1.2.2 Examples of Hilbert Spaces . . . . . . . . . . . . . . . 12
1.2.3 Projection Theorem . . . . . . . . . . . . . . . . . . . . 14
1.2.4 Orthonormal Bases . . . . . . . . . . . . . . . . . . . . 15
1.2.5 Tensor Products of Hilbert Spaces . . . . . . . . . . . . 16

2 Asymptotic Expansions 19
2.1 Asymptotic Estimates . . . . . . . . . . . . . . . . . . . . . . 19
2.1.1 Examples . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.1.2 Properties of asymptotic estimates . . . . . . . . . . . 20
2.2 Asymptotic Sequences . . . . . . . . . . . . . . . . . . . . . . 21
2.2.1 Examples . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.2.2 Properties of asymptotic sequences . . . . . . . . . . . 21
2.3 Asymptotic Series . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.4 Asymptotics of Integrals: Weak Singularities . . . . . . . . . . 23
2.4.1 Power Singularity on a Bounded Interval . . . . . . . . 24
2.4.2 Power singularity on Unbounded Interval . . . . . . . . 26

3 Laplace Method 29
3.1 Laplace Integrals in One Dimension . . . . . . . . . . . . . . . 29
3.1.1 Watson Lemma . . . . . . . . . . . . . . . . . . . . . . 29

I
II Contents

3.1.2 Interior Nondegenerate Maximum Point . . . . . . . . 30

3.1.3 Boundary Maximum Point . . . . . . . . . . . . . . . . 32
3.2 Background from Analysis . . . . . . . . . . . . . . . . . . . . 33
3.2.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.2.2 Morse Lemma . . . . . . . . . . . . . . . . . . . . . . . 35
3.2.3 Gaussian Integrals . . . . . . . . . . . . . . . . . . . . 36
3.3 Laplace Integrals in Many Dimensions . . . . . . . . . . . . . 37
3.3.1 Interior Maximum Point . . . . . . . . . . . . . . . . . 37
3.3.2 Boundary Maximum Point . . . . . . . . . . . . . . . . 39
3.3.3 Integral Operators with Singular Kernels . . . . . . . . 41

4 Stationary Phase Method 43

4.1 Stationary Phase Method in One Dimension . . . . . . . . . . 43
4.1.1 Fourier Integrals . . . . . . . . . . . . . . . . . . . . . 43
4.1.2 Localization Principle . . . . . . . . . . . . . . . . . . . 44
4.1.3 Boundary Points . . . . . . . . . . . . . . . . . . . . . 45
4.1.4 Standard Integrals . . . . . . . . . . . . . . . . . . . . 46
4.1.5 Stationary Point . . . . . . . . . . . . . . . . . . . . . 47
4.1.6 Principal Values of Integrals . . . . . . . . . . . . . . . 47
4.2 Stationary Phase Method in Many Dimensions . . . . . . . . . 49
4.2.1 Nondegenerate Stationary Point . . . . . . . . . . . . . 49
4.2.2 Integral Operators with Singular Kernels . . . . . . . . 50

5 Saddle Point Method 53

5.1 Saddle Point Method for Laplace Integrals . . . . . . . . . . . 53
5.1.1 Heuristic Ideas of the Saddle Point Method . . . . . . . 53
5.1.2 Level Curves of Harmonic Functions . . . . . . . . . . 56
5.1.3 Analytic Part of Saddle Point Method . . . . . . . . . 58
5.1.4 Examples . . . . . . . . . . . . . . . . . . . . . . . . . 60

Notation 63

Bibliography 67
Chapter 1

Functional Analysis

1.1 Preliminaries
1.1.1 Metric Spaces
Definition 1 A metric space is a set X and a mapping d : X × X → R,
called a metric, which satisfies:

i) d(x, y) ≥ 0 (1.1)
ii) d(x, y) = 0 ⇐⇒ x=y (1.2)
iii) d(x, y) = d(y, x) (1.3)
iv) d(x, y) ≤ d(x, z) + d(z, y) (1.4)

Definition 2 A sequence {xn }∞

n=1 , of elements of a metric space (X, d) is
n→∞
said to converge to an element x ∈ X if d(x, xn ) −→ 0.

Definition 3 Let (X, d) be a metric space.

a) the set B(y, r) = {x ∈ X | d(x, y) < r} is called the open ball of

radius r about y;

the set B(y, r) = {x ∈ X | d(x, y) ≤ r} is called the closed ball of

radius r about y;

the set S(y, r) = {x ∈ X | d(x, y) = r} is called the sphere of radius

r about y;

1
2 Functional Analysis

b) a set O ⊂ X is called open if ∀y ∈ O ∃r > 0: B(y, r) ⊂ O;

c) a set N ⊂ X is called a neighborhood of y ∈ N if ∃r > 0: B(y, r) ⊂

N;

d) a point x ∈ X is a limit point of a set E ⊂ X if ∀r > 0: B(x, r) ∩

(E \ {x}) = ∅, i.e. if E contains points other than x arbitrarily close
to x;

e) a set F ⊂ X is called closed if F contains all its limit points;

f ) x ∈ G ⊂ X is called an interior point of G if G is a neighborhood

of x.

g) The intersection S of all closed sets containing S ⊂ X is called the

closure of S. The closure of S ⊂ X is the smallest set containg S.

h) The interior of S, S ◦ , is the set of interior points. It is the largest

open set contained in S.

i) The boundary of S is the set ∂S = S \ S ◦ .

Theorem 1 Let (X, d) be a metric space.

a) A set O is open iff X \ O is closed

d
b) xn −→ x iff ∀ neighborhood N of x ∃m: n ≥ m implies xn ∈ N ;

c) the set of interior points of a set is open;

d) the union of a set and all its limit points is closed;

e) a set is open iff it is a neighborhood of each of its points.

f ) The union of any number of open sets is open.

g) The intersection of finite number of open sets is open.

h) The union of finite number of closed sets is closed.

i) The intersection of any number of closed sets is closed.

j) The empty set and the whole space are both open nd closed.
Functional Analysis 3

Theorem 2 A subset S of a metric space X is closed iff every convergent

sequence in S has its limit in S, i.e.

{xn }∞
n=1 , xn ∈ S, xn → x =⇒ x∈S (1.5)

Theorem 3 The closure of a subset S of a metric space X is the set of limits

of all convergent sequences in S,i.e.

S = {x ∈ X | ∃xn ∈ S : xn → x}. (1.6)

Definition 4 A subset, Y ⊂ X, of a metric space (X, d) is called dense if

d
∀x ∈ X ∃ {yn }∞
n=1 , yn ∈ Y ,: yn −→ x.

Theorem 4 Let S be a subset in a metric space X. Then the following

conditions are equivalent:

a) S is dense in X,

b) S = X,

c) every non-empty subset of X contains an element of S.

Definition 5 A metric space X is called separable if it has a countable

dense set.

Definition 6 A subset S ⊂ X of a metric space X is called compact

if every sequence {xn } in S contains a convergent subsequence whose limit
belongs to S.

Theorem 5 Compact sets are closed and bounded.

Definition 7 A sequence {xn }∞

n=1 , of elements of a metric space (X, d) is
called a Cauchy sequence if ∀ > 0 ∃N : n, m ≥ N implies d(xn , xm ) < .

Proposition 1 Any convergent sequence is Cauchy.

Definition 8 A metric space in which all Cauchy sequences converge is

called complete.
4 Functional Analysis

Definition 9 A mapping f : X → Y from a metric space (X, d) to a

ρ
metric space (Y, ρ) is called continuous at x if f (xn ) −→ f (x) ∀{xn }∞
n=1 ,
d
xn ∈ X, xn −→ x, i.e. the image of a convergent sequence converges to the
image of the limit.

Definition 10 A bijection (one-to-one onto mapping) h : X → Y from

(X, d) to (Y, ρ) is called isometry if it preserves the metric, i.e.

ρ(h(x), h(y)) = d(x, y) ∀x, y ∈ X (1.7)

Proposition 2 Any isometry is continuous.

Theorem 6 If (X, d) is an incomplete metric space, it is possible to find a

complete metric space (X, d) so that X is isometric to a dense subset of X.

1.1.2 Vector Spaces

Definition 11 A complex vector space is a nonempty set V with two
operations: + : V × V → V and · : C × V → V that satisfy the following
conditions:

∀x, y, z ∈ V
i) x+y =y+x (1.8)
ii) (x + y) + z = x + (y + z) (1.9)
iii) ∃0 ∈ V : ∀x ∈ V : x + 0 = x (1.10)
iv) ∀x ∈ V ∃(−x) ∈ V : x + (−x) = 0 (1.11)

∀α, β ∈ C, ∀x, y ∈ V
v) α(βx) = (αβ)x (1.12)
vi) (α + β)x = αx + βx (1.13)
vii) α(x + y) = αx + αy (1.14)
viii) 1 · x = x (1.15)

A real vector space is defined similarly.

Functional Analysis 5

Examples (Function Spaces). Let Ω ⊂ Rn be an open subset of Rn .

1. P (Ω) is the space of all polynomials of n variables as functions on Ω.
2. C(Ω) is the space of all continuous complex valued functions on Ω.
3. C k (Ω) is the space of all complex valued functions with continuous
partial derivatives of order k on Ω.
4. C ∞ (Ω) is the space of all infinitely differentiable complex valued (smooth)
functions on Ω.

Example (Sequence Spaces (lp -Spaces)). Let p ≥ 1. lp is the space of

all infinite sequences {zn }∞
n=1 of complex numbers such that

∞
! p1
X
|zn |p < ∞. (1.16)
n=1

Definition 12 Let V be a complex vector space and let x1 , . . . xk ∈ V and

α1 , . . . , αk ∈ C. A vector x = α1 x1 + · · · αk xk is called a linear combina-
tion of x1 , . . . xk .

Definition 13 A finite collection of vectors x1 , . . . , xk is called linearly

independent if
k
X
αi xi = 0 ⇐⇒ αi = 0 , i = 1, 2, . . . , k . (1.17)
i=1

An arbitrary colection of vectors B = {xn }∞ n=1 is called linearly independent

if every finite subcollection is linearly independent. A collection of vectors
which is not linearly independent is called linearly dependent.

Definition 14 Let B ⊂ V be a subset of a vector space V . Then span B is

the set of all finite linear combinations of vectors from B
( k )
X 
span B = αi xi  xi ∈ B, αi ∈ C, k ∈ N . (1.18)

i=1

Proposition 3 Let B ⊂ V be a subset of a vector space V . Then span B is

a subspace of V .
6 Functional Analysis

Definition 15 A set of vectors B ⊂ V is called a basis of V (or a base

of V ) if B is linearly independent and span B = V . If ∃ a finite basis in V ,
then V is called finite dimensional vector space. Otherwise V is called
infinite dimensional vector space.

Proposition 4 The number of vectors in any basis of a finite dimensional

vector space is the same.

Definition 16 The number of vectors in a basis of a finite dimensional vec-

tor space is called the dimension of V , denoted by dim V .

1.1.3 Normed Linear Spaces

Definition 17 A normed linear space is a vector space, V , over C (or
R) and a mapping || · || : V → R, called a norm, that satisfies:

i) ||v|| ≥ 0 ∀v ∈ V (1.19)
ii) ||v|| = 0 ⇐⇒ v=0 (1.20)
iii) ||αv|| = |α| ||v|| ∀v ∈ V, ∀α ∈ C (1.21)
iv) ||v + w|| ≤ ||v|| + ||w|| ∀v, w ∈ V (1.22)

Examples.

1. Norms in Rn :

k
! 12
X
||x||2 = x2i (1.23)
i=1
n
X
||x||1 = |xi | (1.24)
i=1
||x||∞ = max {|xi |} (1.25)
1≤i≤n

2. A norm in Cn
k
! 21
X
||z|| = |zi |2 (1.26)
i=1
Functional Analysis 7

3. Let Ω ⊂ Rn be a closed bounded subset of Rn and dx = dx1 · · · dxn be

a measure in Rn . Norms in C(Ω) can be defined by

||f ||∞ = sup |f (x)| (1.27)

x∈Ω

Z  p1
p
||f ||p = |f (x)| dx (1.28)
Ω
Z
||f ||1 = |f (x)| dx (1.29)
Ω

4. A norm in lp
k
! p1
X
||z|| = |zi |p (1.30)
i=1

5. A norm in l∞
||z|| = sup |zn | (1.31)
n∈N

Proposition 5 A normed linear space (V, || · ||) is a metric space (V, d) with
the induced metric d(v, w) = ||v − w||.

Convergence, open and closed sets, compact sets, dense sets, com-
pleteness, in a normed linear space are defined as in a metric space in the
induced metric.

Definition 18 A normed linear space is complete if it is complete as a metric

space in the induced metric.

Definition 19 A complete normed linear space is called the Banach space.

Definition 20 A bounded linear transformation from a normed linear

space (V, ||·||V ) to a normed linear space (W, ||·||W ) is a mapping T : V → W
that satisfies:

i) T (αv + βw) = αT (v) + βT (w), ∀v, w ∈ V, ∀α, β ∈ C;

ii) ||T (v)||V ≤ C||v||W for some C ≥ 0.

8 Functional Analysis

iii) the number

||T (v)||W
||T || = sup (1.32)
v∈V,v6=0 ||v||V
is called the norm of T .

Theorem 7 Any bounded linear tranformation between two normed linear

spaces is continuous.

Theorem 8 A bounded linear transformation, T : V → W , from a normed

linear space (V, || · ||V ) to a complete normed linear space (W, || · ||W ) can be
uniquely extended to a bounded linear transformation, T , from the completion
V of V to (W, || · ||W ). The extension of T preserves the norm ||T || = ||T ||.

1.1.4 Notes on Lebesgue Integral

Definition 21 Characteristic function of a set A ⊂ X is a mapping
χA : X → {0, 1} defined by

1, if x ∈ A
χA (x) = (1.33)
0, if x 6∈ A

Definition 22 For a non-zero function f : Rn → R, the set, supp f , of all

points x ∈ Rn for which f (x) 6= 0 is called the support of f , i.e.

supp f = {x ∈ Rn |f (x) 6= 0} . (1.34)

Clearly, supp χA = A.

Definition 23 Let I be a semi-open interval in Rn defined by

I = {x ∈ Rn | ak ≤ xk < bk , k = 1, . . . , n} (1.35)

for some ak < bk . The measure of the set I is defined to be

µ(I) = (b1 − a1 ) · · · · (bn − an ) . (1.36)

The Lebesgue integral of a characteristic function of the set I is defined

by Z
χI dx = µ(I) . (1.37)
Functional Analysis 9

Definition 24 A finite linear combination of characteristic functions of semi-

open intervals
XN
f= αk χIk (1.38)
k=1

is called a step function.

Definition 25 The Lebesgue integral of a step function is defined by

linearity
N
Z X N
X
αk χIk dx = αk µ(Ik ) . (1.39)
k=1 k=1

Definition 26 A function f : Rn → R is Lebesgue integrable if ∃ a

sequence of step functions {fk } such that
∞
X
f' fk , (1.40)
k=1

which means that two conditions are satisfied

∞ Z
X
a) |fk | dx < ∞ (1.41)
k=1
∞
X ∞
X
n
b) f (x) = fk (x) ∀x ∈ R such that |fk (x)| < ∞ . (1.42)
k=1 k=1

The Lebesgue integral of f is then defined by

Z ∞ Z
X
f dx = fk dx (1.43)
k=1

1 n
Proposition 6 The space,
R L (R ), of all Lebesgue integrable functions on
n
R is a vector space and is a linear functional on it.

R R
Theorem 9 a) If f, g ∈ L1 (Rn ) and f ≤ g, then f dx ≤ gdx.
R R
b) If f ∈ L1 (Rn ), then |f | ∈ L1 (Rn ) and | f dx| ≤ |f |dx.
10 Functional Analysis

Theorem 10 If {fk } is a sequence of integrable functions and

∞
X
f' fk , (1.44)
k=1

then Z ∞ Z
X
f= fk , (1.45)
k=1

Definition 27 The L1 -norm in L1 (Rn ) is defined by

Z
||f || = |f |dx (1.46)

Definition 28 A function f is called a null function is it is integrable

and ||f || = 0. Two functions f and g are said to be equivalent if f − g is
a null function.
Definition 29 The equivalence class of f ∈ L1 (Rn ), denoted by [f ], is
the set of all functions equivalent to f .

Remark. Strictly speaking, to make L1 (Rn ) a normed space one has to

consider instead of functions the classes of equivalent functions.
Definition 30 A set X ⊂ Rn is called a null set (or a set of measure
zero) if its characteristic function is a null function.
Theorem 11 a) Every countable set is a null set.
b) A countable union of null sets is a null set.
c) Every subset of a null set is a null set.
Definition 31 Two integrable functions, f, g ∈ L1 (Rn ), are said to be equal
almost everywhere, f = g a.e., if the set of all x ∈ Rn for which f (x) 6=
g(x) is a null set.
Theorem 12
Z
f = g a.e ⇐⇒ ||f − g|| = |f − g| = 0 (1.47)

Theorem 13 The space L1 (Rn ) is complete.

Functional Analysis 11

1.2 Hilbert Spaces

1.2.1 Geometry of Hilbert Space
Definition 32 A complex vector space V is called an inner product space
(or a pre-Hilbert space if there is a mapping (·, ·) : V × V → C, called
an inner product, that satisfies: ∀x, y, z ∈ V, ∀α ∈ C:

i) (x, x) ≥ 0 (1.48)
ii) (x, x) = 0 ⇐⇒ x=0 (1.49)
iii) (x, y + z) = (x, y) + (x, z) (1.50)
iv) (x, αy) = α(x, y) (1.51)
v) (x, y) = (y, x)∗ (1.52)

Definition 33 Let V be an inner product pace.

i) Two vectors x, y ∈ V are said to be orthogonal if (x, y) = 0;

ii) A collection, {xi }N i=1 , of vectors in V is called an orthonormal set

if (xi , xj ) = δij , i.e. (xi , xj ) = 1 if i = j and (xi , xj ) = 0 if i 6= j.

Theorem 14pEvery inner product space is a normed linear space p with the
norm ||x|| = (x, x) and a metric space with the metric d(x, y) = (x − y, x − y).

Theorem 15 (Pythagorean Theorem) Let V be an inner product space

and {xn }N
n=1 be an orthonormal set in V . Then ∀x ∈ V
 2
XN  XN 
||x||2 = |(x, xn )|2 + x − (x, xn )xn  (1.53)
 
n=1
   
n=1

Theorem 16 (Bessel inequality) Let V be an inner product space and

{xn }N
n=1 be an orthonormal set in V . Then ∀x ∈ V

N
X
2
||x|| ≥ |(x, xn )|2 (1.54)
n=1

Theorem 17 (Schwarz inequality) Let V be an inner product space. Then

∀x, y ∈ V
|(x, y)| ≤ ||x|| ||y||. (1.55)
12 Functional Analysis

Theorem 18 (Parallelogram Law) Let V be an inner product space. Then

∀x, y ∈ V
||x + y||2 + ||x − y||2 = 2||x||2 + 2||y||2 . (1.56)
Definition 34 A sequence {xn } of vectors in an inner product space V is
called strongly convergent to x ∈ V , denoted by xn → x, if
n→0
||xn || −→ 0 (1.57)
w
and weakly convergent to x ∈ V , denoted by xn → x, if
n→0
(xn − x, y) −→ 0 ∀y ∈ V . (1.58)
Theorem 19
w
a) xn → x =⇒ xn → x (1.59)
w
b) xn → x and ||xn || → ||x|| =⇒ xn → x (1.60)
Definition 35 A complete inner product space is called a Hilbert space.
Definition 36 A linear transformation U : H1 → H2 from a Hilbert space
H1 onto the Hilbert space H2 is called unitary if if it preserves the inner
product, i.e. ∀x, y ∈ H1
(U x, U y)H2 = (x, y)H1 . (1.61)
Definition 37 Two Hilbert spaces H1 and H2 are said to be isomorphic
if there is a unitary linear transformation U from H1 onto H2 .
Definition 38 Let H1 and H2 be Hilbert spaces. The direct sum H1 ⊕ H2
of Hilbert spaces H1 and H2 is the set of ordered pairs z = (x, y) with x ∈ H1
and y ∈ H2 with inner product
(z1 , z2 )H1 ⊕H2 = (x1 , x2 )H1 + (y1 , y2 )H2 (1.62)

1.2.2 Examples of Hilbert Spaces

1. Finite Dimensional Vectors. CN is the space of N -tuples x =
(x1 , . . . , xN ) of complex numbers. It is a Hilbert space with the inner product
N
X
(x, y) = x∗n yn . (1.63)
n=1
Functional Analysis 13

2. Square Summable Sequences of Complex Numbers. l2 is the

space of sequences of complex numbers x = {xn }∞
n=1 such that
∞
X
|xn |2 < ∞ . (1.64)
n=1

It is a Hilbert space with the inner product

∞
X
(x, y) = x∗n yn . (1.65)
n=1

3. Square Integrable Functions on R. L2 (R) is the space of complex

valued functions such that
Z
|f (x)|2 dx < ∞ . (1.66)
R

It is a Hilbert space with the inner product

Z
(f, g) = f ∗ (x)g(x) dx (1.67)
R

4. Square Integrable Functions on Rn . Let Ω be an open set in Rn (in

particular, Ω can be the whole Rn ). The space L2 (Ω) is the set of complex
valued functions such that
Z
|f (x)|2 dx < ∞ , (1.68)
Ω

where x = (x1 , . . . , xn ) ∈ Ω and dx = dx1 · · · dxn . It is a Hilbert space with

the inner product Z
(f, g) = f ∗ (x)g(x) dx (1.69)
Ω

5. Square Integrable Vector Valued Functions. Let Ω be an open set

in Rn (in particular, Ω can be the whole Rn ) and V be a finite-dimensional
vector space. The space L2 (V, Ω) is the set of vector valued functions f =
(f1 , . . . , fN ) on Ω such that
N Z
X
|fi (x)|2 dx < ∞ . (1.70)
i=1 Ω
14 Functional Analysis

It is a Hilbert space with the inner product

N Z
X
(f, g) = fi∗ (x)gi (x) dx (1.71)
i=1 Ω

6. Sobolev Spaces. Let Ω be an open set in Rn (in particular, Ω can

be the whole Rn ) and V a finite-dimensional complex vector space. Let
C m (V, Ω) be the space of complex vector valued functions that have partial
derivatives of all orders less or equal to m. Let α = (α1 , . . . , αn ), α ∈ N, be
a multiindex of nonnegative integers, αi ≥ 0, and let |α| = α1 + · · · + αn .
Define
∂ |α|
Dα f = f. (1.72)
∂xα1 1 · · · ∂xαnn
Then f ∈ C m (V, Ω) iff
|Dα fi (x)| < ∞ ∀α, |α| ≤ m, ∀i = 1, . . . , N, ∀x ∈ Ω . (1.73)
The space H m (V, Ω) is the space of complex vector valued functions such
that Dα f ∈ L2 (V, Ω) ∀α, |α| ≤ m, i.e. such that
N Z
X
|Dα fi (x)|2 dx < ∞ ∀α, |α| ≤ m . (1.74)
i=1 Ω

It is a Hilbert space with the inner product

N Z
X X
(f, g) = (Dα fi (x))∗ Dα gi (x) dx (1.75)
α, |α|≤m i=1 Ω

Remark. More precisely, the Sobolev space H m (V, Ω) is the completion of

the space defined above.

1.2.3 Projection Theorem

Definition 39 Let M be a closed subspace of a Hilbert space H. The set,
M ⊥ , of vectors in H which are orthogonal to M is called the othogonal
complement of M .

Theorem 20 A closed subspace of a Hilbert space and its orthogonal com-

plement are Hilbert spaces.
Functional Analysis 15

Theorem 21 Let M be a closed subspace of a Hilbert space H. Then ∀x ∈ H

∃ a unique element z ∈ M closest to x.

Theorem 22 (Projection Theorem) Let M be a closed subspace of a Hilbert

space H. Then ∀x ∈ H ∃z ∈ M and ∃w ∈ M ⊥ such that x = z + w. That is

H = M ⊕ M⊥ (1.76)

Remark. The set, L(H, H 0 ), of linear transformations from a Hilbert space

H to H 0 is a Banach space under the norm

||T || = sup ||T x||H 0 (1.77)

||x||H =1

Definition 40 The space H ∗ = L(H, C) of linear transformations from a

Hilbert space H to C is called the dual space of H. The elements of H ∗
are called continuous linear functionals.

Theorem 23 (Riesz Lemma) Let H be a Hilbert pace. Then ∀T ∈ H ∗

∃yT ∈ H such that ∀x ∈ H

T (x) = (yT , x), and ||T ||H ∗ = ||yT ||H (1.78)

1.2.4 Orthonormal Bases

Definition 41 Let S be an orthonormal set in a Hilbert pace H. If there is
no other orthonormal set that contains S as a proper subset, then S is called
orthonormal basis (or complete orthonormal system) for H.

Theorem 24 Every Hilbert space has an othonormal basis.

Theorem 25 Let S = {xα }α∈A be an orthonormal basis for a Hilbert space

H. Then ∀y ∈ H
X
y = (xα , y)xα (1.79)
α∈A
X
2
||y|| = |(xα , y)|2 (1.80)
α∈A
16 Functional Analysis

Definition 42 Let S = {xα }α∈A be an orthonormal basis for a Hilbert space

H. The coefficients (xα , y) are called the Fourier coefficients of y ∈ H
with respect to the basis S.

Definition 43 A metric space which has a countable dense subset is said to

be separable.

Theorem 26 A Hilbert space H is separable iff it has a countable orthonor-

mal basis S. If S contains finite number, N , of elements, then H is isomor-
phic to CN . If S contains countably many elements, then H is isomorphic
to l2 .

1.2.5 Tensor Products of Hilbert Spaces

Let H1 and H2 b Hilbert spaces. For each ϕ1 ∈ H1 and ϕ2 ∈ H2 let ϕ1 ⊗ ϕ2
denote the conjugate bilinear form on H1 × H2 defined by

(ϕ1 ⊗ ϕ2 )(ψ1 , ψ2 ) = (ψ1 , ϕ1 )H1 (ψ2 , ϕ2 )H2 (1.81)

where ψ1 ∈ H1 and ψ2 ∈ H2 . Let E be the set of finite linear combinations

of such bilinear forms. An inner product on E can be defined by

(ϕ ⊗ ψ, η ⊗ µ)E = (ϕ, η)H1 (ψ, µ)H2 (1.82)

(with ϕ, η ∈ H1 and ψ, µ ∈ H2 ) and extending by linearity on E.

Definition 44 The tensor product H1 ⊗ H2 of the Hilbert paces H1 and H2

is defined to be the completion of E under the inner product defined above.

Theorem 27 Let H1 and H2 be Hilbert spaces. If {ϕk } and {ψl } are or-
thonormal bases for H1 and H2 respectively, then {ϕk ⊗ ψl } is anorthonormal
basis for the tensor product H1 ⊗ H2 .

Fock Spaces. Let H be a Hilbert space and let

H0 = C (1.83)
Hn = H
| ⊗ ·{z
· · ⊗ H} (1.84)
n
Functional Analysis 17

denote the n-fold tensor product of H. The space

F (H) = ⊕∞
n=0 H
n
(1.85)

is called the Fock space over H. Fock space F (H) is separable if H is

separable. For example, if H = L2 (R), then an element ψ ∈ F (H) is a
sequence of functions

ψ = {ψo , ψ1 (x), ψ(x1 , x2 ), ψ3 (x1 , x2 , x3 ), . . .} (1.86)

so that
∞ Z
X
2
||ψ|| = |ψ0 | + |ψn (x1 , . . . , xn )|2 dx1 . . . dxn < ∞ (1.87)
n=1 Rn

Let Pn be the permutation group of n elements and let {ϕk } be a basis

for H. Each σ ∈ Pn defines a permutation

σ(ϕk1 ⊗ · · · ⊗ ϕkn ) = ϕkσ(1) ⊗ · · · ⊗ ϕkσ(n) . (1.88)

By linearity this can be extended to a bounded operator on H n , so one can

define
1 X
Sn = σ (1.89)
n! σ∈P
n

1 X
An = ε(σ)σ (1.90)
n! σ∈P
n

where 
1, if σ is even
ε(σ) = (1.91)
−1 if σ is odd
Finally, the Boson (symmetric) Fock space is defined by

Fs (H) = ⊕∞
n=0 Sn H
n
(1.92)

and the Fermion (antisymmetric) Fock space is defined by

Fa (H) = ⊕∞
n=0 An H
n
(1.93)

In the case H = L2 (R), ψn ∈ Sn H n is a function of n variables symmetric

under any permutations of variables, and ψn ∈ An H n is a function of n
variables that is odd function under interchanges of any two variables.
18 Functional Analysis
Chapter 2

Asymptotic Expansions

2.1 Asymptotic Estimates

Let M be a set of real or complex numbers with a limit point a. Let f, g :
M → R (or f, g : M → C) be some functions on M .

Definition 45 The following are asymptotic estimates

i) f (x) ∼ g(x) (x → a, x ∈ M )
f (x)
if lim =1 (2.1)
x→a, x∈M g(x)

ii) f (x) = o(g(x)) (x → a, x ∈ M )

f (x)
if lim =0 (2.2)
x→a, x∈M g(x)

iii) f (x) = O(g(x)) (x ∈ M )

if ∃C : |f (x)| ≤ C|g(x)| ∀x ∈ M (2.3)

iv) f (x) = O(g(x)) (x → a, x ∈ M )

if ∃C and a neighborhood U of a such that :
|f (x)| ≤ C|g(x)| ∀x ∈ M ∩ U (2.4)

19
20 Asymptotic Expansions

2.1.1 Examples
1.
ln x = o(x−α ) (x → 0+ ), α > 0 . (2.5)

2.
ln x = o(xα ) (x → ∞), α > 0 . (2.6)

3.
sin z ∼ z (z → 0) . (2.7)

4.
sin x = O(1) (x ∈ R) . (2.8)

5. √
n! ∼ 2πne−n nn (n → ∞) . (2.9)

Remarks. The relation f (x) = o(g(x)) means that f (x) is infinitesimal

with respect to g(x) as x → a. Similarly, f (x) = O(g(x)) means that f (x) is
bounded with respect to g(x) as x → a. In particular, f (x) = o(1), (x → a)
means that f (x) is infinitesimal as x → a and f (x) = O(1), (x → a) means
that f (x) is bounded as x → a.

2.1.2 Properties of asymptotic estimates

There holds (as x → a, x ∈ M ):

o(f (x)) + o(f (x)) = o(f (x)) (2.10)

o(f (x))o(g(x)) = o(f (x)g(x)) (2.11)
o(o(f (x))) = o(f (x)) (2.12)
O(f (x)) + O(f (x)) = O(f (x)) (2.13)
O(f (x))O(g(x)) = O(f (x)g(x)) (2.14)
O(O(f (x))) = O(f (x)) (2.15)
o(f (x)) + O(f (x)) = O(f (x)) (2.16)
o(f (x))O(g(x)) = o(f (x)g(x)) (2.17)
O(o(f (x))) = o(f (x)) (2.18)
o(O(f (x))) = o(f (x)) (2.19)
Asymptotic Expansions 21

2.2 Asymptotic Sequences

Definition 46 Let ϕn : M → R, n ∈ N, and a be a limit point of M .
Let ϕn (x) 6= 0 in a neighborhood Un of a. The the sequence {ϕn } is called
asymptotic sequence at x → a, x ∈ M if ∀n ∈ N

ϕn+1 (x) = o(ϕn (x)) (x → a, x ∈ M ) (2.20)

2.2.1 Examples
1. Power asymptotic sequences

(a)
{(x − a)n }, x→a. (2.21)

(b)
{x−n }, x→∞. (2.22)

2. Let {λn } be a decreasing sequence of real numbers, i.e. λn < λn+1 , and
let 0 < ε ≤ π/2. Then the sequence

π
{eλn z }, z → ∞, | arg z| ≤ −ε (2.23)
2

is an asymptotic sequence.

2.2.2 Properties of asymptotic sequences

1. Any subsequence of an asymptotic sequence is an asymptotic sequence.

2. Let f (x) 6= 0 for x ∈ M in some neighborhood of a and {ϕn } be an

asymptotic sequence at x → a, x ∈ M . Then the sequence {f (x)ϕn (x)}
is an asymptotic sequence as x → a, x ∈ M .

3. Let {ϕn (x)}, {ψn (x)} be asymptotic sequences as x → a, x ∈ M . Then

the sequence {ϕn (x)ψn (x)} is an asymptotic sequence as x → a, x ∈ M .
22 Asymptotic Expansions

2.3 Asymptotic Series

Let f : M → R and a be a limit point of M .

Definition 47 Let {ϕn } be an asymptotic sequence as x → a, x ∈ M . We

say that the function f is expanded in an asymptotic series
∞
X
f (x) ∼ an ϕn (x), (x → a, x ∈ M ), (2.24)
n=0

where an are constants, if ∀N ≥ 0

N
X
RN (x) ≡ f (x) − an ϕn (x) = o(ϕN (x)), (x → a, x ∈ M ) . (2.25)
n=0

This series is called asymptotic expansion of the function f with respect

to the asympotic sequence {ϕn }. RN (x) is called the rest term of the
asymptotic series.

Remarks

1. The condition RN (x) = o(ϕN (x)) means, in particular, that

lim RN (x) = 0 for any fixed N (2.26)

x→a

2. Asymptotic series could diverge. This happens if

lim RN (x) 6= 0 for some fixed x (2.27)

N →∞

3. There are three possibilities:

(a) asymptotic series converges to f (x);

(b) asymptotic series converges to a function g(x) 6= f (x);
(c) asymptotic series diverges.

Theorem 28 Asymptotic expansion of a function with respect to an asymp-

totic sequence is unique.
Asymptotic Expansions 23

Remark. Two different functions can have the same asymptotic expansion.
For example, f (x) = ex and g(x) = ex + e−1/x have the same asymptotic
expansion with respect to the asymptotic sequence {xn }:
∞
x x −1/x
X xn
e ∼e +e ∼ , x → 0+ (2.28)
n=0
n!

Theorem 29 Asymptotic series can be added and multiplied by numbers,

but, cannot be multipied by asymptotic series.
Theorem 30 One can multiply and divide power asymptotic series.
Definition 48 Let f : M × S → R be a function of two variables and a be a
limit point of M and {ϕn } be an asymptotic sequence as x → a. Let for any
fixed y ∈ S the function f is expanded in an asymptotic series
∞
X
f (x, y) ∼ an (y)ϕn (x), (x → a, x ∈ M ) . (2.29)
n=0

This asymptotic expansion is called uniform with respect to the parameter

y ∈ S, if the relation
N
X
RN (x, y) ≡ f (x, y)− an (y)ϕn (x) = o(ϕN (x)), (x → a, x ∈ M ) (2.30)
n=0

is valid uniformly with respect to y ∈ S.

Theorem 31 A uniform asymptotic expansion can be integrated with respect
to the parameter term by term.

Remark. One cannot, in general, differentiate asymptotic series, neither

with respect to x nor with respect to a parameter.

2.4 Asymptotics of Integrals: Weak Singu-

larities
Let us consider the integrals of the form
Z a
F (ε) = f (x, ε) dx (2.31)
0
24 Asymptotic Expansions

where a > 0 and ε > 0 is a small positive parameter. Here f ∈ C ∞ ([0, a] ×

(0, ε0 ]) is a smooth function for 0 ≤ x ≤ a, 0 < ε ≤ ε0 , with some ε0 . Then
the integral converges for ε > 0. Let f have a singularity when ε = 0, i.e.
g(x) = f (x, 0) has a singularity at some 0 ≤ x ≤ a. If this singularity is of
power or logarithmic type then we say that the integral F (ε) has a weak
singularity.
This definition is obviously extended for unbounded intervals. Let
Z ∞
F (ε) = f (x, ε) dx (2.32)
a

where a > 0 and ε > 0 is a small parameter. Here f ∈ C ∞ ([a, ∞) × (0, ε0 ]) is

a smooth function for a ≤ x ≤ ∞, 0 < ε ≤ ε0 , with some ε0 . Let the integral
converge for ε > 0 and diverge for ε = 0. If the function g(x) = f (x, 0) is
of power or logarithmic order at x → ∞, then we say that the integral F (ε)
has a weak singularity.

2.4.1 Power Singularity on a Bounded Interval

Let a, α, β ∈ R, a, β > 0, be some real numbers and ε > 0 be a small positive
prameter. Let ϕ ∈ C ∞ [0, a] be a smooth function on [0, a]. We will study
the asymptotics as ε → 0+ of the integrals of the form
Z a
F (ε) = tβ−1 (t + ε)α ϕ(t) dt . (2.33)
0

Remarks. The function F is holomorphic in complex plane ε with a cut

along the negative half-axis. At the point ε = 0 this function has a singularity
(if α > 0 is not integer). The type of this singularity is determined by the
behavior of the function ϕ at small t ≥ 0.

Standard Integral. One needs the following result. Let α and β be two
complex numbers such that Re β > 0, Re α < 0 and Re (β + α) < 0. Then
Z ∞
Γ(β)Γ(−α − β)
tβ−1 (t + 1)α dt = , (0 < Re β < −Re α) . (2.34)
0 Γ(−α)

Theorem 32 Let ϕ ∈ C ∞ [0, a]. Let r, δ > 0 and Sδ = {ε ∈ C|0 < |ε| ≤
r, | arg ε| ≤ π − δ} be a sector in the complex plane of .
Asymptotic Expansions 25

1. If α + β is not integer, then

∞ ∞
X Γ(β + n)Γ(−α − β − n) ϕ(n) (0) α+β+n
X
F (ε) ∼ ε + an εn
n=0
Γ(−α) n! n=0

(ε → 0, ε ∈ Sδ ) (2.35)

2. If α + β = N is an integer, then
∞ ∞
X Γ(N + n) ϕ(n) (0) n+N X
F (ε) ∼ − ε ln ε + b n εn
Γ(α)Γ(N + n − α) n! n=0
n≥max{0,−N }

(ε → 0, ε ∈ Sδ ) (2.36)

The coefficients an and bn depend on the values ϕ(t) for 0 ≤ t ≤ a. The

branch for the functions εγ and ln ε is choosen in such a way that εγ > 0 and
ln ε is real for ε > 0.

Examples. In all examples ϕ ∈ C ∞ ([0, a]) is a smooth function bounded

with all its derivatives.

1. Let 0 < a < 1 and Z a

ϕ(t)
F (ε) = dt . (2.37)
0 t+ε
Then
F (ε) = −ϕ(0) ln ε + O(1), (ε → 0+ ) (2.38)

2. Let 0 < a < 1 and Z a

ϕ(t)
F (ε) = dt . (2.39)
0 t2 + ε2
By using  
1 1 1 1
2 2
= − (2.40)
t +ε 2iε t − iε t + iε
we obtain from the previous example
π
F (ε) = ϕ(0) + O(1) (ε → 0+ ) (2.41)
2ε
26 Asymptotic Expansions

3. Let α > 1/2 and Z a

ϕ(t)
F (ε) = dt . (2.42)
0 (t2 + ε2 )α
Then
√
π Γ(α − 1/2) 1−2α
F (ε) = ϕ(0) ε +O(ε3−2α )+O(1), (ε → 0+ ) (2.43)
2 Γ(α)

2.4.2 Power singularity on Unbounded Interval

Standard Integral. To compute the following asymptotics one needs the
following standard integral. Let Re β > 0 and Re α > −1. Then
Z ∞  
α −tβ 1 α+1
t e dt = Γ , (Re β > 0, Re α > −1) . (2.44)
0 β β

Examples. Let ϕ ∈ C ∞ ([a, ∞)) be a smooth function on [a, ∞) that has

asymptotic expansion as x → ∞
∞
X
ϕ(x) ∼ ak x−k . (2.45)
k=0

1. Let a, β > 0, and

Z ∞
β
F (ε) = ϕ(x)xα e−εx dx . (2.46)
a

If α < −1, then the integral is not singular as ε → 0+ . Its asymptotic

expansion can be obtained either by integration by parts or by a change
of variables.
So, let now α + 1 > 0 and let N = [α + 1] ≥ 0 be the integer part
of α + 1. Let us single out the first N + 1 terms of the asymptotic
expansion in ϕ, i.e.
N
X
ϕ(x) = ak x−k + RN (x) . (2.47)
k=0

Since RN (x) = O(x−(N +1) ) as x → ∞, we have

N Z ∞
β
X
F (ε) = ak xα−k e−εx dx + O(1) . (2.48)
k=0 a
Asymptotic Expansions 27

Now by changing the variables and extending the interval to [0, ∞) we

obtain
N    
X ak − α−k+1 α − k + 1
F (ε) = ε β Γ + O(1) + O(1)
k=0
β β
 
a0 − α+1 α+1 α
= ε β Γ + O(ε− β ) (2.49)
β β

If α = −1, then
a0
F (ε) = − ln ε + O(1), (ε → 0+ ) . (2.50)
β

2. Let a > 0 and let

P (x) = xn + · · · + a1 x n≥1. (2.51)

Consider the integral

Z ∞
F (ε) = ϕ(x)xα e−εP (x) dx . (2.52)
a

If α > −1, then the main term of the asymptotics is

 
a0 α + 1 − α+1 α
F (ε) = Γ ε n + O(ε− n ), (ε → 0+ ) (2.53)
n n

If α = −1, then
a0
F (ε) = − ln ε + O(1), (ε → 0+ ) (2.54)
n
If α < −1, then the integral is not singular as ε → 0+ . By integration
by parts the integral can be reduced to the cases considered above.
28 Asymptotic Expansions
Chapter 3

Laplace Method

3.1 Laplace Integrals in One Dimension

Let M = [a, b] be a closed bounded interval, S : M → R be a real valued
function, ϕ : M → C a complex valued function and λ be a large positive
parameter. Consider the integrals of the form
Z b
F (λ) = ϕ(x) exp[λS(x)] dx . (3.1)
a

Such integrals are called Laplace integrals. We will study the asymptotics
of the Laplace integrals as λ → ∞.

Lemma 1 Let supa<x<b S(x) = L < ∞ and the integral (3.1) converges
absolutely for some λ0 > 0. Then
1.
|F (λ)| ≤ C|eλL | (Re λ ≥ λ0 ) . (3.2)

2. if f, S ∈ C(a, b), then F (λ) is holomorphic in the halfplane Re λ > λ0 .

3.1.1 Watson Lemma

Lemma 2 (Watson) Let 0 < a < ∞, α > 0, β > 0 and let Sε be the sector
Sε = {λ ∈ C | | arg λ| ≤ π/2 − ε} in the complex plane λ. Let ϕ ∈ C ∞ ([0, a])
and let Z a
Φ(λ) = ϕ(x)xβ−1 exp(−λxα ) dx (3.3)
0

29
30 Laplace Method

Then, there is an asymptotic expansion as λ → ∞, λ ∈ Sε ,

∞
β + k ϕ(k) (0)
 
1 X −(β+k)/α
Φ(λ) ∼ λ Γ (3.4)
α k=0 α k!

Laplace Transform. Let ϕ ∈ C ∞ (R+ ) be a smooth function on the posi-

tive real axis such that its Laplace transform
Z ∞
L(ϕ)(λ) = ϕ(x)e−λx dx (3.5)
0

converges absolutely for some λ0 . Then

∞
X
L(ϕ)(λ) ∼ λ−k ϕ(k) (0) (|λ| → ∞, λ ∈ Sε ) (3.6)
k=0

3.1.2 Interior Nondegenerate Maximum Point

Let now S and ϕ be smooth functions and the function S have a maximum
at an interior point x0 of the interval [a, b], i.e. a < x0 < b. Then S 0 (x0 ) = 0.
Assume, for simplicity, that S 00 (x0 ) 6= 0. Then S 00 (x0 ) < 0. In other words,
in a neighborhood of x0 the function S has the following Taylor expansion
(x − x0 )2
S(x) = S(x0 ) + S 00 (x0 ) + O((x − x0 )3 ) . (3.7)
2
Such a point is called nondegenerate critical point.
Then, as λ → ∞ the main contribution to the integral comes from a small
neighborhood of x0 . In this neighborhood the function ϕ is almost constant
and can be replaced by its value at x0 . The terms of order (x − x0 )3 can be
neglected in the exponent and the remaining integral can be extended to the
whole real line. By using the standard Gaussian integral
Z ∞  α  r
2 2π
exp − y dy = , (Re α > 0) , (3.8)
−∞ 2 α
one obtains finally the main term of the asymptotics
s
2π
F (λ) ∼ λ−1/2 ϕ(x0 )eλS(x0 ) , (λ → ∞) (3.9)
−S 00 (x0 )
One can prove the general theorem.
Laplace Method 31

Theorem 33 Let M = [a, b] and ϕ, S ∈ C ∞ (M ), S has a maximum only at

one point x0 , a < x0 < b and S 00 (x0 ) 6= 0. Then as λ → ∞, λ ∈ Sε there is
asymptotic expansion
∞
X
F (λ) ∼ e λS(x0 )
ck λ−1/2−k . (3.10)
k=0

The coefficients ck are expressed in terms of derivatives of ϕ and S at x0 .

The theorem can be proved as follows. First, we change the integration

variable
x = x0 + λ−1/2 y . (3.11)
So, y is the scaled fluctuation from the maximum point x0 . The interval
of integration should be changed accordingly, so that the maximum point is
now y = 0. Then, we expand both functions S and ϕ in Taylor series at x0
getting
∞
1 X S (n) (x0 ) n −(n−2)/2
λS(x0 + λ−1/2 y) = λS(x0 ) + S 00 (x0 )y 2 + y λ , (3.12)
2 n=3
n!

∞
−1/2
X ϕ(n) (x0 )
ϕ(x0 + λ y) = y n λ−n/2 . (3.13)
n=0
n!

Since the quadratic terms are of order O(1) we leave it in the exponent
and expand the exponent of the rest in a power series. Next, we extend the
integration interval to the whole real line and compute the standard Gaussian
integrals of the form
Z ∞  α 
exp − y 2 y 2k+1 dy = 0 , (3.14)
−∞ 2
Z ∞  
 α 
2 2k 1  α −k−1
exp − y y dy = Γ k + , (3.15)
−∞ 2 2 2
where k is a nonnegative integer and α has a positive real part, Re α > 0.
Finally, we get a power series in inverse powers of λ. The coefficients ck of
the asymptotic expansion are polynomials in the higher derivatives S (k) (x0 ),
k ≥ 3, and derivatives ϕ(l) (x0 ), l ≥ 0, and involve inverse powers of S 00 (x0 ).
32 Laplace Method

Stirling Formula
√
Γ(x + 1) = 2πxx+1/2 e−x [1 + O(x−1 )], (x → ∞) (3.16)
is obtained by applying the Laplace method to the integral
Z ∞
x+1
Γ(x + 1) = x exp[x(ln t − t)] dt (3.17)
0

Stieltjes Transform Let ϕ : R+ → C have finite moments

Z ∞
mn (ϕ) = tn ϕ(t) dt < ∞ ∀n ∈ N. (3.18)
0
Then the Stieltjes transform of ϕ
Z ∞
ϕ(t)
S(ϕ)(x) = dt (3.19)
0 t+x
has asymptotic expansion as x → ∞
∞
X
S(ϕ)(x) ∼ (−1)k mk (ϕ)x−1−k (3.20)
k=0

3.1.3 Boundary Maximum Point

Let the function S have a maximum at a boundary point x0 = a. Let, for
simplicity, S 0 (a) 6= 0, i.e. S 0 (a) < 0, and ϕ(a) 6= 0. Then, as λ → ∞, the
main contribution to the integral comes from the interval [a, a + ], where
S(x) = S(a) + (x − a)S 0 (a) + O((x − a)2 ) . (3.21)
Now, by replacing the function ϕ by its value at a and neglecting nonlinear
terms in S(x), we get
ϕ(a) λS(a)
F (λ) ∼ λ−1 e , (λ → ∞) (3.22)
−S 0 (a)
In this way one can prove the following theorem.
Theorem 34 Let M = [a, b], ϕ, S ∈ C ∞ (M ), S has a maximum only at the
point x = a and S 0 (a) 6= 0. Then, as λ → ∞, λ ∈ Sε , there is asymptotic
expansion
∞
X
F (λ) ∼ eλS(a) ck λ−1−k . (3.23)
k=0
The coefficients ck are expressed in terms of derivatives of ϕ and S at x = a.
Laplace Method 33

Error Function The asymptotic expansion of the (complementary) error

function as x → ∞ has the form
Z ∞ 2 ∞
−t2 e−x X (2k − 1)!! −2k
Erfc x = e dt ∼ (−1)k x (x → ∞) . (3.24)
x 2x k=0 2k

Incomplete Gamma Function The incomplete gamma-function

Z x
γ(a, x) = ta−1 e−t dt, (0 < a < ∞, x > 0) (3.25)
0

has the following asympotic expansion as x → ∞

∞
X Γ(a)
γ(a, x) = Γ(a) + e−x xa−1 x−k (3.26)
k=0
Γ(a − k)

3.2 Background from Analysis

3.2.1 Definitions
1. Let xj , j = 1, . . . , n be real numbers and let x be the n-tuple x =
(x1 , . . . xn ). The set of all n-tuples of real numbers is denoted by Rn .
A connected open subset Ω of Rn is called a domain. The set ∂Ω of
boundary points of Ω is called the boundary of Ω. The union Ω ∪ ∂Ω
is called the closure of Ω and is denoted by Ω.

2. Let x, ξ ∈ Rn . Then the scalar product of x and ξ is defined by

x · ξ = x1 ξ 1 + · · · + xn ξ n .

3. On Rn there is standard Lebesgue measure dx = dx1 · · · dxn .

4. Let αj , j = 1, . . . , n, be non-negative integers, αj ≥ 0. The n-tuple

α = (α1 , . . . , αn ) is called a multi-index. Further, let

|α| = α1 + · · · + αn (3.27)

α! = α1 ! · · · αn ! (3.28)
∂ |α| f (x)
∂ α f (x) = (3.29)
∂(x1 )α1 · · · ∂(xn )αn
34 Laplace Method

 α1  αn
α 1 ∂ 1 ∂
D f (x) = ··· f (x) (3.30)
i ∂x1 i ∂xn
Then the Taylor expansion of a smooth function ϕ at x0 can be written
in the form ∞
X 1 α
ϕ(x) = [∂ ϕ(x0 )] (x − x0 )α . (3.31)
α!
|α|=0

5. The boundary ∂Ω is said to be smooth, denoted ∂Ω ∈ C ∞ , if in a

neighborhood of any boundary point x0 ∈ ∂Ω it can be locally defined
by an equation xj = ϕ(x0 ) with a smooth function ϕ.

6. The set of all continuous functions on Ω is denoted by C(Ω).

7. The set of all functions with continuous partial derivatives up to order

k on Ω is denoted by C k (Ω).

8. The set of all functions with continuous partial derivatives up to order

k on Ω is denoted by C k (Ω).

9. The set of all functions with continuous partial derivatives up to order

k on Ω that vanish in a neighborhood of the boundary ∂Ω is denoted
by C0k (Ω).

10. The closure of the set where a function is not equal to zero is called
the support of the function, denoted by

supp f = {x ∈ Ω | f (x) 6= 0} . (3.32)

11. A mapping ϕ : Ω → Ω, is said to be of class C k if ϕ ∈ C k (Ω).

12. A one-to-one mapping ϕ : Ω → Ω of Ω onto Ω is called diffeomor-

phism of class C k if ϕ ∈ C k (Ω) and ϕ−1 ∈ C k (Ω).

13. Let ϕj , j = 1, . . . , k be some scalar functions on Rn and ϕ be a k-tuple

ϕ = (ϕ1 , . . . , ϕk ). In other words, ϕ : Rn → Rk . The matrix
 i 
∂ϕ (x)
∂x ϕ(x) = , i = 1, . . . , k; j = 1, . . . , n (3.33)
∂xj

is called the Jacobi matrix.

Laplace Method 35

Theorem 35 (Inverse Function Theorem) Let ϕ : Rn → Rn is of class

C k , k ≥ 1, in a neighborhood of a point x0 and det ∂x ϕ(x0 ) 6= 0. Then ϕ is
local diffeomorphism of class C k in a neighborhood of the point x0 .

Theorem 36 (Implicit Function Theorem) Let Ω be a domain in R2n ,

let F : Ω → Rn be a mapping of class C k (Ω) and let (x0 , y0 ) ∈ Ω be a point
in Ω such that

F (x0 , y0 ) = 0, det ∂y F (x0 , y0 ) 6= 0 . (3.34)

Then in a neighborhood of the point x0 there is a mapping y = f (x) of class

C k such that y0 = f (x0 ) and

F (x, f (x)) ≡ 0 . (3.35)

3.2.2 Morse Lemma

Let S : Ω → R be a real valued function of class C k on a domain Ω in Rn
with k ≥ 2. Let
 2 
2 ∂ S(x)
∂x S(x) = , i, j = 1, . . . , n . (3.36)
∂xi ∂xj

Definition 49 1. The point x0 is called a critical point of the function

S if ∂S(x0 ) = 0

2. A critical point x0 is called non-degenerate if det ∂x2 S(x0 ) 6= 0.

3. The determinant det ∂x2 S(x0 ) is called the Hessian of the function S
at the point x0 .

Lemma 3 (Morse) Let S : Rn → R and x0 ∈ Rn be a non-degenerate

critical point of the function S. Let S ∈ C ∞ in a neighborhood of the point
x0 and let µj 6= 0, j = 1, . . . , n be the eigenvalues of the matrix ∂x2 S(x0 ).
Then there are neighborhoods U and V of the points x0 and 0 and a smooth
local diffeomorphism ϕ : V → U of class C ∞ such that det ∂y ϕ(0) = 1 and
n
1X
S(ϕ(y)) = S(x0 ) + µj (y j )2 . (3.37)
2 j=1
36 Laplace Method

Remark. Nondegenerate critical points are isolated.

3.2.3 Gaussian Integrals

Proposition 7 Let A = (aij be a complex symmetric nondegenerate n × n
matrix with the eigenvalues µj (A), j = 1, . . . , n,. Let Re A ≥ 0, which means
that x · Re A x ≥ 0 ∀x ∈ Rn , x 6= 0, or Re µj (A) ≥ 0, j = 1, . . . , n. Then for
λ > 0, ξ ∈ Rn there holds
Z  
λ
exp − x · A x − iξ · x dx
Rn 2
 n/2  
2π −1/2 1 −1
= (det A) exp − ξ · A ξ . (3.38)
λ 2λ
√
The branch of det A is choosen as follows

(det A)−1/2 = | det A|−1/2 exp (−i Ind A) , (3.39)

where n
1X π
Ind A = arg µj (A), | arg µj (A)| ≤ . (3.40)
2 j=1 2

By expanding both sides of this equation in Taylor series in ξ we obtain

the following result.
Corollary 1
Z  
λ
exp − x · A x xi1 · · · xi2k+1 dx = 0 (3.41)
R n 2
Z  
λ
exp − x · A x xi1 · · · xi2k dx
Rn 2
 n/2
2π (2k)! (i1 i2
= (det A)−1/2 (2λ)−k G · · · Gi2k−1 i2k ) . (3.42)
λ k!
Here k is a non-negative integer, G = A−1 , and the round brackets denote
complete symmetrization over all indices included.
An important particular case of the previous formula is when the matrix
A is real.
Laplace Method 37

Proposition 8 Let A be a real symmetric nondegenerate n × n matrix. Let

ν+ (A) and ν− (A) be the numbers of positive and negative eigenvalues of A
and
sgn A = ν+ − ν− (3.43)
be the signature of the matrix A. Then for λ > 0, ξ ∈ Rn there holds
Z  
λ
exp i x · A x − iξ · x dx
Rn 2
 n/2  
2π −1/2 i −1 iπ
= | det A| exp − ξ · A ξ + sgn (A) . (3.44)
λ 2λ 4

3.3 Laplace Integrals in Many Dimensions

3.3.1 Interior Maximum Point
Let Ω be a bounded domain in Rn , S : Ω → R, f : Ω → C are some functions
on Ω and λ > 0 be a large positive parameter. We will study the asymptotics
as λ → ∞ of the multidimensional Laplace integrals
Z
F (λ) = f (x) exp[λS(x)] dx . (3.45)
Ω

Let S and f be smooth functions and the function S have a maximum

only at one interior nondegenerate critical point x0 ∈ Ω. Then ∂x S(x0 ) = 0
and [det ∂x2 S(x0 )] < 0. Then in a neighborhood of x0 the function S has the
following Taylor expansion
1
S(x) = S(x0 ) + (x − x0 ) · [∂x2 S(x0 )](x − x0 ) + O((x − x0 )3 ) . (3.46)
2
One could also use the Morse Lemma to replace the function S by a quadratic
form. Then as λ → ∞ the main contribution to the integral comes from a
small neghborhood of x0 . In this neighborhood the terms of the third order
in the Taylor expansion of S can be neglected. Also, since the function f
is continuous at x0 , it can be replaced by its value at x0 . Then the region
of integration can be extended to the whole Rn . By using the formula for
the standard Gaussian integral one gets then the leading asymptotics of the
integral F (λ) as λ → ∞
 n/2
2π −1/2
− det ∂x2 S(x0 )

F (λ) ∼ exp[λS(x0 )] f (x0 ) . (3.47)
λ
38 Laplace Method

One can prove the general theorem.

Theorem 37 Let f, S ∈ C ∞ (Ω) and let x0 be a nondegenerate critical point

of the function S where it has the only maximum in Ω. Let 0 < ε < π/2.
Then there is asymptotic expansion as λ → ∞ in the sector Sε = {λ ∈
C | | arg λ| ≤ π/2 − ε}

∞
X
−n/2
F (λ) ∼ exp[λS(x0 )]λ ak λ−k . (3.48)
k=0

The coefficients ak are expressed in terms of the derivatives of the functions

f and S at the point x0 .

The idea of the proof is the same as in the one-dimensional case and goes
as follows. First, we change the integration variables

xi = xi0 + λ−1/2 y i . (3.49)

So, y is the scaled fluctuation from the maximum point x0 . The interval
of integration should be changed accordingly, so that the maximum point is
now y = 0. Then, we expand both functions S and ϕ in Taylor series at x0
getting
∞
−1/2 1 2
X λ−(|α|−2)/2 α
λS(x0 + λ y) = λS(x0 ) + y · [∂x S(x0 )]y + [∂ S(x0 )] y α ,
2 α!
|α|=3
(3.50)
∞
X λ−|α|/2 α
ϕ(x0 + λ−1/2 y) = ∂ ϕ(x0 ) y α . (3.51)
α!
|α|=0

Since the quadratic terms are of order O(1) we leave them in the exponent
and expand the exponent of the rest in a power series. Next, we extend
the integration domain to the whole Rn and compute the standard Gaus-
sian integrals. Finally, we get a power series in inverse powers of λ. The
coefficients ak of the asymptotic expansion are polynomials in the higher
derivatives ∂ α S(x0 ), |α| ≥ 3, and derivatives ∂ α ϕ(x0 ), |α| ≥ 0, and involve
inverse matrices G = [∂x2 S(x0 )]−1 .
Laplace Method 39

Remark. If x0 is a degenerate maximum point of the function S, then

the asymptotic expansion as λ → ∞ has the form
∞ X
X N
F (λ) ∼ exp[λS(x0 )]λ−n/2 akl λ−rk (ln λ)l , (3.52)
k=0 l=0

where N is some positive integer and {rk }, rk ≥ n/2, k ∈ N, is a increasing

sequence of nonnegative rational numbers.
The coefficients ak (and akl ) of the asymptotic expansion of the integral
F (λ) are invariants under smooth local diffeomorphisms in a neighborhood
of x0 and play very important role in various applications.

3.3.2 Boundary Maximum Point

Let now S has maximum at the boundary point x0 ∈ ∂Ω. We assume,
for simplicity, that both the boundary and the function S are smooth, i.e.
S ∈ C ∞ and ∂Ω ∈ C ∞ .
Since the boundary is smooth we can smoothly parametrize it in a neigh-
borhood of x0 by (n − 1) parameters ξ = (ξ a ), (a = 1, . . . , n − 1). Let the
the parametric equations of the boundary be
xi = xi (ξ), i = 1, . . . , n . (3.53)
Then
∂xi
 
Ta = (Tai ) = (3.54)
∂ξ a
are tangent vectors to the boundary.
Let r = r(x) be the normal distance to the boundary. Then the equation
of the boundary can be written as
r(x) = 0 . (3.55)
and for x ∈ Ω we have r > 0. Obviously, r(x(ξ)) ≡ 0. From this equation
we obtain that the vector
 
∂r
N = (Ni ) = (3.56)
∂xi
is orthogonal to all tangent vectors and is therefore normal to the boundary.
It can be certainly normalized, since it is nowhere zero. We choose it to be
the inward normal.
40 Laplace Method

The normal and tangential derivatives are defined as usual

n n
X ∂xi ∂ ∂ X ∂xi ∂
∂r = , = (3.57)
i=1
∂r ∂xi ∂ξ a i=1
∂ξ a ∂xi

The point x0 is not, in general, a critical point of S, since the normal

derivative of S at x0 does not have to be equal to zero.

Definition 50 The point x0 is said to be a nondegenerate boundary

maximum point of S if
∂r S(x0 ) 6= 0 (3.58)
and the (n − 1) × (n − 1) matrix ∂ξ2 S(x(ξ)) is negative definite.

In a neighborhood of a nondegenerate boundary maximum point the func-

tion S has the following Taylor expansion
1
S(x) = S(x0 ) + [∂r S(x0 )]r + [∂r2 S(x0 )]r2
2
1
+[∂r ∂ξ S(x0 )] · (ξ − ξ0 )r + (ξ − ξ0 ) · [∂ξ2 S(x0 )](ξ − ξ0 )
2
+··· , (3.59)

up to third order terms in r and (ξ − ξ0 ).

Now we replace the integral F (λ) by an integral over a small neighborhood
of x0 . We change the variables of integration from xi , i = 1, . . . , n, to (ξ a , r),
a = 1, . . . , n − 1, and neglect the terms of third order in the Taylor series.
We also replace the function f by its value at the point x0 . In the remaining
integral we extend the integration to the whole space R+ × Rn−1 , i.e. we
integrate over r from 0 to ∞ and integrate over the whole tangent plane
at x0 . These integrals are standard Gaussian integrals and we obtain the
leading asymptotics as λ → ∞

F (λ) ∼ −λ−(n+1)/2 (2π)(n−1)/2 exp[λS(x0 )]

×[∂r S(x0 )]−1 [− det ∂ξ2 S(x0 )]−1/2 J(x0 )f (x0 ) (3.60)

where J(x0 ) is the Jacobian of change of variables.

The general form of the asymptotic expansion is given by the following
theorem.
Laplace Method 41

Theorem 38 Let f, S ∈ C ∞ (Ω) and let S have a maximum only at a non-

degenerate boundary maximum point x0 ∈ ∂Ω. Then as λ → ∞, λ ∈ Sε ,
∞
X
−(n+1)/2
F (λ) ∼ λ exp[λS(x0 )] ak λ−k (3.61)
k=0

3.3.3 Integral Operators with Singular Kernels

Let Ω be a bounded domain in Rn including the origin, 0 ∈ Ω. Let S be a
real valued non-positive function on Ω of class C 2 that has maximum equal
to zero, S(0) = 0, only at a nondegenerate maximum critical point x0 = 0.
Let Kλ : C ∞ (Ω) → C ∞ (Ω) be a linear integral operator defined by
 n/2 Z
λ
(Kλ f )(x) = exp[λS(x − y)]f (y) dy . (3.62)
2π Ω

Let M be a compact subset of Ω. Then

lim (Kλ f )(x) = [− det ∂x2 S(0)]−1/2 f (x) (3.63)

λ→∞

uniformly for x ∈ M .
Formally
 n/2
λ
exp[λS(x − y)] −→ [− det ∂x2 S(0)]−1/2 δ(x − y) (3.64)
2π
42 Laplace Method
Chapter 4

Stationary Phase Method

4.1 Stationary Phase Method in One Dimen-

sion
4.1.1 Fourier Integrals
Let M = [a, b] be a closed bounded interval, S : M → R be a real valued
nonconstant function, f : M → C be a complex valued nonzero function
and λ be a large positive parameter. Consider the integrals of the form
Z b
F (λ) = f (x) exp[iλS(x)] dx . (4.1)
a

The function S is called phase function and such integrals are called
Fourier Integrals. We will study the asymptotics of such integrals.
As λ → ∞ the integral F (λ) is small due to rapid oscillations of exp(iλS).

Lemma 4 (Riemann-Lebesgue) Let f be an integrable function on the

real line, i.e. f ∈ L1 (R). Then
Z
f (x)eiλx dx = o(1), (λ → ∞) . (4.2)
R

Definition 51 1. A point x0 is called the regular point of the Fourier

integral F (λ) if the functions f and S are smooth in a neighborhood of
x0 and S 0 (x0 ) 6= 0.

43
44 Stationary Phase Method

2. A point x0 is called the critical point of the integral F (λ) if it is not

a regular point.

3. A critical point x0 is called isolated critical point if there is a

neighborhood of x0 that does not contain any other critical points.

4. An interior isolated critical point is called stationary point.

5. The integral over a neighborhood of an isolated critical point that does

not contain other critical points will be called the contribution of
the critical point to the integral.

Clearly the main contribution comes from the critical points since close
to these points the oscillations slow down. As always, we will assume that
functions S and f are smooth, i.e. of class C ∞ (M ). Otherwise, the sin-
gularities of the functions S and f and their derivatives would contribute
significantly to F (λ).

4.1.2 Localization Principle

Lemma 5 Let S ∈ C ∞ (R) be smooth function and f ∈ C0∞ (R) be a smooth
function of compact support. Then as λ → ∞
Z
f (x) exp[iλS(x)] dx = O(λ−∞ ) (4.3)
R

Remarks.

1. Since the function f has compact support, the integral is, in fact, over
a finite interval.

2. This is the main technical lemma for deriving the (power) asympotics
of the Fourier integrals. It means that such integrals can be neglected
in a power asymptotic expansion.

3. The Fourier integrals are in general much more subtle object than the
Laplace integrals. Instead of exponentially decreasing integrand one
has a rapidly oscillating one. This requires much finer estimates and
also much stronger conditions on the phase function S and the inte-
grand f .
Stationary Phase Method 45

Theorem 39 Let the Fourier integral F (λ) have finite number of isolated
critical points. Then as λ → ∞ the integral F (λ) is equal to the sum of the
contributions of all critical points up to O(λ−∞ ).

Thus, the problem reduces to computing the asymptotics of the contribu-

tions of critical points. In a neighborhood of a critical point we can replace
the functions S and f by more simple functions and then compute some
standard integrals.

4.1.3 Boundary Points

If the phase function does not have any stationary points, then by integration
by parts one can easily obtain the asymptotic expansion.

Theorem 40 Let S 0 (x) 6= 0 ∀x ∈ M . Then as λ → ∞

b
∞  k  
X
−k−1 1 ∂ f (x) 
F (λ) ∼ (iλ) eiλS(x)  . (4.4)

k=0
−S 0 (x) ∂x S 0 (x) 
a

The leading asymptotics is

F (λ) = (iλ)−1 {f (b) exp[iλS(b)] − f (a) exp[iλS(a)]} + O(λ−2 ) (4.5)

The same technique, i.e. integration by parts, applies to the integrals

over an unbounded interval, say,
Z ∞
F (λ) = f (x) exp[iλS(x)] dx . (4.6)
0

with some additional conditions that guarantee the converges at ∞ as well

as to the integrals of the form
Z ∞
F (x) = f (t) exp[iS(t)] dt (4.7)
x

as x → ∞.
46 Stationary Phase Method

Examples
1. Let α > 0. Then as λ → ∞

(a)
∞
eiλx
Z
α
dx = iλ−1 + O(λ−2 ) . (4.8)
0 (1 + x)
(b) Z ∞
sin(λx)
dx = λ−1 + O(λ−2 ) . (4.9)
0 (1 + x)α
(c) Z ∞
cos(λx)
dx = αλ−2 + O(λ−3 ) . (4.10)
0 (1 + x)α
2. Let α > 0. Then as x → ∞
Z ∞ ∞
−α it ix −α
X Γ(k + α)
F (x) = t e dt ∼ ie x (ix)−k . (4.11)
x k=0
Γ(α)

In particular, the Frenel integral has the asymptotic expansion as x →

∞
Z ∞ ∞  
−α it i ix2 −1/2 X 1
Φ(x) = t e dt ∼ √ e x k
(−1) Γ k + x−k .
x 2 π k=0
2
(4.12)

4.1.4 Standard Integrals

Consider the integral
Z a
α
Φ(λ) = f (x)xβ−1 eiλx (4.13)
0

Lemma 6 (Erdéyi) Let α ≥ 1, β > 0 and f is a smooth function on a

closed bounded interval [0, a], f ∈ C ∞ ([0, a]), that vanish at x = a with all
its derivatives. Then as λ → ∞
Z a ∞
β−1 iλxα
X
Φ(λ) = f (x)x e ∼ ak λ−(β+k)/α , (4.14)
0 k=0
Stationary Phase Method 47

where
f (k) (0) 1
   
β+k β+k
ak = Γ exp iπ (4.15)
k! α α α

This lemma plays he same role in the stationary phase method as Watson
lemma in the Laplace method.

4.1.5 Stationary Point

Theorem 41 Let M = [a, b] be a closed bounded interval, S ∈ C ∞ (M )
be a smooth real valued nonconstant function, f ∈ C0∞ (M ) be a complex
valued function with compact support in M . Let S have a single isolated
nondegenerate critical point x0 in M , i.e. S 0 (x0 ) = 0 and S 00 (x0 ) 6= 0. Then
as λ → ∞ there is asymptotic expansion of the Fourier integral
Z b
F (λ) = f (x) exp[iλS(x)] dx
a
∞
h
00 π i −1/2 X −k
∼ exp iλS(x0 ) + [sgn S (x0 )] i λ λ . (4.16)
4 k=0

The coefficients ak are determined in terms of the derivatives of the functions

S and f at x0 .

The leading asymptotics as λ → ∞ is

s
2π −1/2
h
00 πi
f (x0 ) + O(λ−1 ) .


F (λ) = 00
λ exp iλS(x 0 ) + [sgn S (x 0 )] i
|S (x0 )| 4
(4.17)
To prove this theorem one does a change of variables in a sufficiently
small neghborhood of x0 and applies the Erdélyi lemma.

4.1.6 Principal Values of Integrals

Let f be a smooth function and consider the integral
Z b
f (x)
dx (4.18)
a x
48 Stationary Phase Method

This integral diverges, in general, at x = 0. One can regularize it by

cutting out a symmetric neighborhood of the singular point
Z −ε Z b
f (x) f (x)
I(ε) = dx + dx . (4.19)
a x ε x

Definition 52 If the limit of I(ε) as ε → 0+ exists, then it is called the

principal value of the integral I
Z b Z −ε Z b 
f (x) f (x) f (x)
P dx = lim+ dx + dx . (4.20)
a x ε→0 a x ε x

In this section we consider the asymptotics of the integrals of the form

Z
dx
F (λ) = P e±iλS(x) f (x) (4.21)
R x
as λ → ∞.

Lemma 7 Let f ∈ C0∞ (R) be a smooth function of compact support. Then

as λ → ∞ Z
dx
P e±iλx f (x) = ±iπf (0) + O(λ−∞ ) . (4.22)
R x

Theorem 42 Let f ∈ C0∞ (R) be a smooth function of compact support,

S ∈ C ∞ (R) be a real valued smooth function and S 0 (0) 6= 0. Then as λ → ∞
Z
dx
F (λ) = P e±iλS(x) f (x) = [sgn S 0 (0)] iπf (0) exp[iλS(0)] + O(λ−∞ ) .
R x
(4.23)

Theorem 43 Let f ∈ C0∞ (R) be a smooth function of compact support,

S ∈ C ∞ (R) be a real valued smooth function. Let x = 0 be the only statinary
point of the function S on supp f , and let it be nondegenerate, i.e. S 0 (0) = 0
and S 00 (0) 6= 0. Then as λ → ∞ there is asymptotic expansion
Z
dx
F (λ) = P e±iλS(x) f (x)
R x
X∞
∼ exp[iλS(0)]λ−1/2 ak λ−k . (4.24)
k=0
Stationary Phase Method 49

The leading asymptotics has the form

s
h πi 2π
F (λ) = exp iλS(0) + [sgn S 00 (0)] i
4 |S 00 (0)|

S 000 (0)
 
−1/2 0 −1
×λ − 00 f (0) + f (0) + O(λ ) . (4.25)
6S (0)

4.2 Stationary Phase Method in Many Di-

mensions
Let Ω be a domain in Rn and f ∈ C0∞ (Ω) be a smooth function of compact
support, S ∈ C ∞ (Ω) be a real valued smooth function. In this section we
study the asympotics as λ → ∞ of the multi-dimensional Fourier integrals
Z
F (λ) = f (x) exp[iλS(x)] dx . (4.26)
Ω

4.2.1 Nondegenerate Stationary Point

Localization Principle
Lemma 8 Let Ω be a domain in Rn and f ∈ C0∞ (Ω) be a smooth function
of compact support, S ∈ C ∞ (Ω) be a real valued smooth function without
stationary points in supp f , i.e. ∂x S(x) 6= 0 for x ∈ supp f . Then as λ → ∞

F (λ) = O(λ−∞ ) (4.27)

This lemma is proved by integration by parts.

Definition 53 1. The set S(Rn ) of all smooth functions on Rn that de-

crease at |x| → ∞ together with all derivatives faster than any power
of |x| is called the Schwartz space.

2. For any integrable function f ∈ L1 (Rn ) the Fourier transform is

defined by
Z
−n/2
F(f )(ξ) = (2π) exp(i x · ξ) f (x) dx (4.28)
Rn
50 Stationary Phase Method

Proposition 9 1. Fourier transform is a one-to-one onto map (bijection)

F : S(R ) → S(Rn ), i.e. if f ∈ S(Rn ), then F(f ) ∈ S(Rn ).
n

2. The inverse Fourier transform is

Z
−1 −n/2
F (f )(x) = (2π) exp(−i x · ξ) f (ξ) dξ (4.29)
Rn

Theorem 44 Let Ω be a finite domain in Rn , f ∈ C0∞ (Ω) be a smooth

function with compact support in Ω and S ∈ C ∞ (Ω) be a real valued smooth
function. Let S have a single stationary point x0 in Ω and let it be non-
degenerate. Then as λ → ∞ there is asymptotic expansion
∞
X
−n/2
F (λ) ∼ λ exp[iλS(x0 )] ak λ−k . (4.30)
k=0

The coefficients ak are determined in terms of derivatives of the functions f

and S at x0 .

The leading asymptotics is

 n/2
2π h πi
F (λ) = exp iλS(x0 ) + [sgn ∂x2 S(x0 )] i
λ 4
−1/2 
f (x0 ) + O(λ−1 ) .

× det ∂x2 S(x0 )

(4.31)

Recall that sgn A = ν+ (A) − ν− (A) denotes the signature of a real symmet-
ric nondegenerate matrix A, where ν± (A) are the number of positive and
negative eigenvalues of A.

4.2.2 Integral Operators with Singular Kernels

Let Ω be a bounded domain in Rn including the origin, 0 ∈ Ω. Let f ∈ C0∞ (Ω)
be a smooth function with compact support and S ∈ C ∞ (Rn ) be a real valued
non-positive smooth function. Let S have a single stationary point x = 0,
and let it to a nondegenerate, i.e. let S(0) = S 0 (0) = 0 and ∂x2 S(0) 6= 0. Let
Kλ : C0∞ (Ω) → C ∞ (Ω) be a linear integral operator defined by
 n/2 Z
λ
(Kλ f )(x) = exp[iλS(x − y)]f (y) dy . (4.32)
2π Ω
Stationary Phase Method 51

Then as λ → ∞
h π i  −1/2 
(Kλ f )(x) = exp [sgn ∂x2 S(0)] i det ∂x2 S(0) f (x) + O(λ−1 ) (4.33)

4
uniformly for x ∈ Ω. On the other hand, if x 6∈ Ω, then as λ → ∞

(Kλ f )(x) = O(λ−∞ ) . (4.34)

52 Stationary Phase Method
Chapter 5

Saddle Point Method

5.1 Saddle Point Method for Laplace Inte-

grals
Let γ be a contour in the complex plane and the functions f and S are
holomorphic in a neighborhood of this contour. In this section we will study
the asymptotics as λ → ∞ of the Laplace integrals
Z
F (λ) = f (z) exp[λ S(z)] dz . (5.1)
γ

5.1.1 Heuristic Ideas of the Saddle Point Method

The idea of the saddle point method (or method of steepest descent)
is to deform the contour in such a way that the main contribution to the
integral comes from a neighborhood of a single point. This is possible since
the functions f and S are holormorphic.
First of all, let us find a bound for |F (λ)|. For a contour γ = γ0 of finite
length l(γ0 ) we have, obviously,
|F (λ)| ≤ l(γ0 ) max f (z) exp[λ Re S(z)] . (5.2)
z∈γ0

Now, let Γ be the set of all contours obtained by smooth deformations of the
contour γ0 keeping the endpoints fixed. Then such an estimate is valid for
any contour γ ∈ Γ, hence,
 
|F (λ)| ≤ inf l(γ) max f (z) exp[λ Re S(z)] . (5.3)
γ∈Γ z∈γ

53
54 Saddle Point Method

Since we are interested in the limit λ → ∞, we expect that the length of the
contour does not affect the accuracy of the estimate. Also, intuitively it is
clear that the behavior of the function S is much more important that that
of the function f (since S is in the exponent and its variations are scaled
significantly by the large parameter λ). Thus we expect an estimate of the
form  
|F (λ)| ≤ C(γ, f ) inf max exp[λRe S(z)] , (5.4)
γ∈Γ z∈γ

where C(γ, f ) is a constant that depends on the contour γ and the function f
but does not depend on λ. So, we are looking for a point on a given contour
γ where the maximum of Re S(z) is attained. Then, we look for a contour γ ∗
where the minimum of this maximum is attained, i.e. we assume that there
exists a contour γ∗ where

min max Re S(z) (5.5)

γ∈Γ z∈γ

is attained. Such a contour will be called a minimax contour.

Let z0 ∈ γ∗ be the only point on the contour γ∗ where the maximum of
Re S(z) is attained. Then, we have an estimate

|F (λ)| ≤ C(γ∗ , f ) exp[λ Re S(z0 )] . (5.6)

By deforming the contour of integration to γ∗ we obtain

Z
F (λ) = f (z) exp[λ S(z)] dz . (5.7)
γ∗

The asymptotics of this integral can be computed by Laplace method.

1. Boundary Point. Let z0 be an endpoint of γ∗ , say, the initial point.

Suppose that S 0 (z0 ) 6= 0. Then one can replace the integral F (λ) by an
integral over a small arc with the initial point z0 . Finally, integrating
by parts gives the leading asymptotics
1
exp[λ S(z0 )] λ−1 f (z0 ) + O(λ−1 ) .
 
F (λ) = 0
(5.8)
−S (z0 )

2. Interior Point. Let z0 be an interior point of the contour γ∗ . From

the minimax property of the contour γ∗ it follows that the point z0 is
Saddle Point Method 55

the saddle point of the function Re S(z). Let z = x + iy. Since the
saddle point is a stationary point, then
∂ ∂
Re S(z0 ) = Re S(z0 ) = 0 . (5.9)
∂x ∂y

Then from Cauchy-Riemann conditions it follows that S 0 (z0 ) = 0.

Definition 54 1. A point z0 ∈ C is called a saddle point of the complex

valued function S : C → C if S 0 (z0 ) = 0.

2. A saddle point z0 is said to be of order n if

S 0 (z0 ) = · · · = S (n) (z0 ) = 0, S (n+1) (z0 ) 6= 0 . (5.10)

3. A first order saddle point is called simple, i.e. for a simple saddle
point S 00 (z0 ) 6= 0.

4. The number Re S(z0 ) is called the height of the saddle point.

To compute the asymptotics at an interior saddle point, we replace the

contour γ∗ by a small arc γ∗0 containing the point z0 . Then we expand the
function S in the Taylor series in the neighborhood of z0 and neglect the
terms of third order and higher, i.e. we replace S by
1
S(z) = S(z0 ) + S 00 (z0 )(z − z0 )2 + O((z − z0 )3 ) . (5.11)
2
Finally, by changing the variables and evaluating the integral by Laplace
method we obtain the asymptotics as λ → ∞
s
2π
λ−1/2 exp[λ S(z0 )] f (z0 ) + O(λ−1 ) .
 
F (λ) = 00
(5.12)
−S (z0 )

The saddle point method consists of two parts: the toplogical part and
the analytical part.
The topological part consists of the deformation of the contour to the
minimax contour γ∗ that is most suitable for asymptotical estimates. The
analytical part contains then the evaluation of the asymptotics over the con-
tour γ∗ .
56 Saddle Point Method

The analytical part is rather straightforward. Here one can apply the
same methods and as in the Laplace method; in many cases one can even
use the same formulas.
The topological part is usually much more complicated since it is a global
problem. It could happen, for example, that a contour γ∗ where the minimax
minγ∈Γ maxz∈γ Re S(z) is attained does not exist at all! Next, strictly speak-
ing we need to look for a contour where minγ∈Γ maxz∈γ f (z) exp[Re S(z)] is
attained what makes the problem even more complicated.
Thus, if one can find the minimax contour, then one can compute the
asymptotics of F (λ) as λ → ∞. Unfortunately, there is no simple algorithm
that would always enable one to find the minimax contour. Nevertheless,
under certain conditions one can prove that such a contour exists and, in
fact, find one. We will discuss this point later.

5.1.2 Level Curves of Harmonic Functions

Lemma 9 Let S : C → C be holomorphic at z0 and S 0 (z0 ) 6= 0. Then in a
small neighborhood of the point z0 the arcs of the level curves

Re S(z) = Re S(z0 ), Im S(z) = Im S(z0 ), (5.13)

are analytic curves. These curves are orthogonal at z0 .

Let ϕ(z) = S(z) − S(z0 ). Since S 0 (z0 ) 6= 0 the function w = ϕ(z) is is a

one-to-one holomorphic, in fact, conformal, mapping of a neghborhood of
the point z = z0 onto a neighborhood of the point w = 0. The inverse
function z = ϕ−1 (w) is holomorphic in a neighborhood of the origin w = 0.
Let w = u + iv and ψ(u, v) ≡ ϕ−1 (w). The arc of the level curve Re S(z) =
Re S(z0 ) is mapped onto an open interval on the imaginary axis. It is defined
by z = ψ(0, v) and is analytic. The same is true for the level curve Im S(z) =
Im S(z0 ). It is defined by z = ψ(u, 0) and is analytic as well. The tangent
vectors to the level curves at z0 are determined by
 
∂ψ(u, 0)  −1 ∂ψ(0, v)
= i (∂w ϕ−1 )(0)

= (∂w ϕ )(0) ,  (5.14)
∂u u=0 ∂v v=0

and are obviously orthogonal. One could also conlude this from the fact that
the map is conformal and, therefore, preserves the angles.
Saddle Point Method 57

Lemma 10 Let z0 be a simple saddle point of the function S. Then in a

small neighborhood of the point z0 the level curve Re S(z) = Re S(z0 ) consists
of two analytic curves that intersect orthogonally at the point z0 and separate
the neighborhood of z0 in four sectors. The signs of the function Re [S(z) −
S(z0 )] in adjacent sectors are different.

In a neighborhood of a simple saddle point the there is a one-to-one holo-

morphic function z = ψ(w) such that ψ(0) = z0 , ψ 0 (0) 6= 0, and S(ψ(w)) =
S(z0 ) + w2 . In the complex plane of w the level curve Re S(z) = Re S(z0 )
takes the form Re w2 = 0. Its solution consists of two orthogonal lines
w± = (1 ± i)t with −ε < t < ε that intersect at the point w = 0. The level
curves z± = ψ(w± ) have listed properties.
More generally :

Lemma 11 Let z0 be a saddle point of the function S of order n. Then

in a small neighborhood of the point z0 the level curve Re S(z) = Re S(z0 )
consists of (n + 1) analytic curves that intersect at the point z0 and separate
the neighborhood of z0 in 2(n + 1) sectors with angles π(n + 1) at the vertex.
The signs of the function Re [S(z) − S(z0 )] in adjacent sectors are different.

Definition 55 Let S be a complex valued function and γ be a simple curve

with the initial point z0 . The curve γ is called curve of steepest descent
of the function Re S if Im S(z) = const and Re S(z) < Re S(z0 ) for z ∈ γ,
z 6= z0 . If Im S(z) = const and Re S(z) > Re S(z0 ) for z ∈ γ, z 6= z0 , then
the curve γ is called curve of steepest ascent of the function Re S.

Lemma 12 1. If z0 is not a saddle point, then there is exactly one curve

of steepest descent.

2. If z0 is a simple saddle point, then there are 2 curves of steepest descent.

3. If z0 is a saddle point of order n, then there are (n+1) curves of steepest

descent.

4. In a neighborhood of a saddle point z0 in each sector in which Re [S(z)−

S(z0 )] > 0 there is exatly one curve of steepest descent.

This is proved by a change of variables in a neighborhood of z0 .

58 Saddle Point Method

Remarks. Let S : D → C be a nonconstant holomorphic function in a

domain D. Let z = x+iy and S(z) = u(x, y)+iv(x, y). Then both u : D → R
and v : D → R are harmonic functions in D. Harmonic functions do not have
maximum or minimum points in the interior of D. They are attained only
at the boundary of the domain D. All critical points of harmonic functions,
i.e. the points where ∂u = ∂v = 0 are saddle points. These are exactly the
points where S 0 (z) = 0. That is why such points are called saddle points
of the function S. In the simplest case S = z 2 the surface u = x2 − y 2 is
hyperbolic paraboloid (saddle).

Definition 56 Two contours γ1 and γ2 are called equivalent if

Z Z
f (z) exp[λ S(z)] dz = f (z) exp[λ S(z)] dz . (5.15)
γ1 γ2

Lemma 13 Let S and f be holormorphic functions on a finite contour γ.

Let the points where maxz∈γ Re S(z) is attained are neither saddle points nor
the endpoints of the contour γ. Then there is a contour γ 0 equivalent to the
contour γ and such that

max0 Re S(z) < max Re S(z) . (5.16)

z∈γ z∈γ

Theorem 45 Let F (λ) be a Laplace integral (5.1) If there exists a contour

γ∗ such that: i) it is equivalent to the contour γ and ii) the integral F (λ)
attains the minimax minγ∈Γ maxz∈γ Re S(z) on it. Then among the points
where maxz∈γ∗ Re S(z) is attained there are either endpoints of the contour
or saddle points zj such that in aneighborhood of zj the contour γ∗ goes
through two different sectors where Re S(z) < Re S(zj ).

5.1.3 Analytic Part of Saddle Point Method

In this section we always assume that γ is a simple smooth (or piece-wise
smooth) curve in the complex plane, which may be finite or infinite. The
functions f and S are assumed to be holomorphic on γ. Also, we assume
that the integral F (λ) converges absolutely.
First of all, we have
Lemma 14 If maxz∈γ Re S(z) ≥ C, then

F (λ) = O(eC λ), (λ ≥ 1) . (5.17)

Saddle Point Method 59

Theorem 46 Let z0 be the initial endpoint of the curve γ. Let f and S are
analytic at z0 , Re S(z0 ) > Re S(z) ∀z ∈ γ, and S 0 (z0 ) 6= 0. Then, as λ → ∞
there is asympotic expansion
∞
X
F (λ) ∼ exp[λS(z0 )] ak λ−k−1 (5.18)
k=0

where  k  
1 ∂ f (z) 
ak = − − 0 (5.19)
S (z) ∂z S 0 (z) 

z=z0

The proof is by integration by parts.

Theorem 47 Let z0 be an interior point of the curve γ. Let f and S are

analytic at z0 , Re S(z0 ) > Re S(z) ∀z ∈ γ. Let z0 be a simple saddle point of
S such that in a neighborhood of z0 the contour γ goes through two different
sectors where Re S(z) < Re S(z0 ). Then, as λ → ∞ there is asympotic
expansion
∞
X
F (λ) ∼ exp[λS(z0 )] ak λ−k−1/2 . (5.20)
k=0
p
The branch of the square root is choosen so that arg −S 00 (z0 is equal to the
angle between the positive direction of the tangent to the curve γ at the point
z0 and the positive direction of the real axis.

In a neighborhood of z0 there is a mapping z = ψ(w) such that

w2
S(ψ(w)) = S(z0 ) − . (5.21)
2
After this change of variables and deforming the contour to the steepest
descent contour the integral becomes
Z ε
2
F (λ) = exp[λS(z0 )] e−λw /2 f (ψ(w))ψ 0 (w) dw + O(λ−∞ ) . (5.22)
ε

Since both f and ψ are holomorphic there is a Taylor expansion

∞
X
0
f (ψ(w)ψ (w) = ck wk . (5.23)
k=0
60 Saddle Point Method

Then the coefficients ak are easily computed in terms of ck

 
k+1/2 1
ak = 2 Γ k+ c2k . (5.24)
2

Theorem 48 Let γ be a finite contour and f and S be holomorphic in a

neighborhood of γ. Let maxz∈γ ReS(z) be attained at the points zj and these
points are either the endpoints of the curve or saddle points such that in
a neighborhood of a saddle points the contour γ goes through two different
sectors where Re S(z) < Re S(zj ). Then as λ → ∞ the integral F (λ) is
asymptotically equal to the sum of contributions of the points zj .

Remark. The integral over a small arc containing a saddle point is called
the contribution of the saddle point to the integral.

Proposition 10 Let f and S be holomorphic functions on γ and Im S(z) =

const on γ. If γ has finite number of saddle points, then as |lambda| → ∞,
| arg λ| ≤ π/2 − ε < π/2, the asymptotic expansion of the integral F (λ) is
equal to the sum of contributions of the saddles points and the endpoints of
the contour.

5.1.4 Examples
1.
Z ∞ p
eiλx (1 + x2 )−λ dx ∼ π(1 − c)e−λc λ−1/2 (2c)−λ (λ → ∞)
−∞
√ √ (5.25)
where c = 2 − 1. There are two saddle points z1,2 = i(−1 ± 2). The
asymptotics is determined by the contribution from the point z1 .

2.
i+∞
Z r
−z 2 −n π −n (n−1)/2  n −n/2
e (1 + z) dz ∼ i e , (n → ∞) .
i−∞ 2 2
(5.26)
Here n > 0 is a positive integer.
Saddle Point Method 61

3. Let a > 0. As λ → ∞
Z ia+∞ √
exp(−2λz 2 −4λ/z) dz ∼ π 1/6 (6λ)−1/2 exp(3λ+ i3 3 λ) . (5.27)
ia−∞

There are three saddle points, which are the roots of the equation
z 3 = 1. The asymptotics is determined by the contribution from the
point z1 = ei 2π/3 .

4.
Z i∞ √
 
2 2 1 2λ−1
exp(−λz + z ln z) dz ∼ i π exp − e , (λ → ∞) .
−i∞ 2
(5.28)
λ−1/2
There are two saddle points z1 = 0 and z2 = e . The asymptotics
is equal to the contribution from the saddle point z2 .
62 Saddle Point Method
Notation

Logic

A =⇒ B A implies B

A ⇐= B A is implied by B

iff if and only if

A ⇐⇒ B A implies B and is implied by B

∀x ∈ X for all x in X

∃x ∈ X there exists an x in X such that

Sets and Functions (Mappings)

x∈X x is an element of the set X

x 6∈ X x is not in X

{x ∈ X | P (x)} the set of elements x of the set X obeying the property

P (x)

A⊂X A is a subset of X

X \A complement of A in X

A closure of set A

X ×Y Cartesian product of X and Y

f :X→Y mapping (function) from X to Y

f (X) range of f

63
64 Notation

χA characteristic function of the set A

∅ empty set

N set of natural numbers (positive integers)

Z set of integer numbers

Q set of rational numbers

R set of real numbers

R+ set of positive real numbers

C set of complex numbers

Vector Spaces

H ⊕G direct sum of H and G

H∗ dual space

Rn vector space of n-tuples of real numbers

Cn vector space of n-tuples of complex numbers

l2 space of square summable sequences

lp space of sequences summable with p-th power

Normed Linear Spaces

||x|| norm of x

xn −→ x (strong) convergence
w
xn −→ x weak convergence

Function Spaces

supp f support of f

H ⊗G tensor product of H and G

Notation 65

C0 (Rn ) space of continuous functions with bounded support in

Rn

C(Ω) space of continuous functions on Ω

C k (Ω) space of k-times differentiable functions on Ω

C ∞ (Ω) space of smooth (infinitely diffrentiable) functions on Ω

D(Rn ) space of test functions (Schwartz class)

L1 (Ω) space of integrable functions on Ω

L2 (Ω) space of square integrable functions on Ω

Lp (Ω) space of functions integrable with p-th power on Ω

H m (Ω) Sobolev spaces

C0 (V, Rn ) space of continuous vector valued functions with bounded

support in Rn

C k (V, Ω) space of k-times differentiable vector valued functions on

Ω

C ∞ (V, Ω) space of smooth vector valued functions on Ω

D(V, Rn ) space of vector valued test functions (Schwartz class)

L1 (V, Ω) space of integrable vector valued functions functions on

Ω

L2 (V, Ω) space of square integrable vector valued functions func-

tions on Ω

Lp (V, Ω) space of vector valued functions functions integrable with

p-th power on Ω

H m (V, Ω) Sobolev spaces of vector valued functions

Linear Operators

Dα differential operator
66 Notation

L(H, G) space of bounded linear transformations from H to G

H ∗ = L(H, C) space of bounded linear functionals (dual space)

Bibliography
View publication stats

[1] M. Reed and B. Simon, Methods of Mathematical Physics. I: Functional

Analysis, (New York: Academic Press, 1972)

[2] R. D. Richtmyer, Principles of Advanced Mathematical Physics, vol. I,

(Berlin: Springer, 1985)

[3] L. Debnath and P. Mikusiński, Introduction to Hilbert Spaces with Ap-

plications, (Boston: Academic Press, 1990)

[4] M. V. Fedoriuk, Asymptotics: Integrals and Series, (Moscow: Nauka,

1987)

[5] E. T. Copson, Asymptotic Expansions, (Cambridge, UK: Cambridge

University Press, 1965)

[6] F. W. J. Olver, Introduction to Asymptotics and Special Functions,

(New York: Academic Press, 1973)

[7] F. W. J. Olver, Asymptotics and Special Functions, (New York: Aca-

demic Press, 1974)

[8] A. Erdélyi, Asymptotic Expansions, (New York: Dover Publications,

1956)

[9] R. B. Dingle, Asymptotic Expansions: Their Derivation and Interpre-

tation, (London: Academic Press, 1973)

67