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 Denitions . . . . . . . . . . . . . . . . . . . . . . . . . 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
Denition 1 A metric space is a set X and a mapping d : X X R,
called a metric, which satises:
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)
Denition 2 A sequence {x
n
}

n=1
, of elements of a metric space (X, d) is
said to converge to an element x X if d(x, x
n
)
n
0.
Denition 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 i X \ O is closed
b) x
n
d
x i neighborhood N of x m: n m implies x
n
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 i it is a neighborhood of each of its points.
f ) The union of any number of open sets is open.
g) The intersection of nite number of open sets is open.
h) The union of nite 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 i every convergent
sequence in S has its limit in S, i.e.
{x
n
}

n=1
, x
n
S, x
n
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 | x
n
S : x
n
x}. (1.6)
Denition 4 A subset, Y X, of a metric space (X, d) is called dense if
x X {y
n
}

n=1
, y
n
Y ,: y
n
d
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.
Denition 5 A metric space X is called separable if it has a countable
dense set.
Denition 6 A subset S X of a metric space X is called compact
if every sequence {x
n
} in S contains a convergent subsequence whose limit
belongs to S.
Theorem 5 Compact sets are closed and bounded.
Denition 7 A sequence {x
n
}

n=1
, of elements of a metric space (X, d) is
called a Cauchy sequence if > 0 N: n, m N implies d(x
n
, x
m
) < .
Proposition 1 Any convergent sequence is Cauchy.
Denition 8 A metric space in which all Cauchy sequences converge is
called complete.
4 Functional Analysis
Denition 9 A mapping f : X Y from a metric space (X, d) to a
metric space (Y, ) is called continuous at x if f(x
n
)

f(x) {x
n
}

n=1
,
x
n
X, x
n
d
x, i.e. the image of a convergent sequence converges to the
image of the limit.
Denition 10 A bijection (one-to-one onto mapping) h : X Y from
(X, d) to (Y, ) is called isometry if it preserves the metric, i.e.
(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 nd a
complete metric space (X, d) so that X is isometric to a dense subset of X.
1.1.2 Vector Spaces
Denition 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)
, C, x, y V
v) (x) = ()x (1.12)
vi) ( +)x = x +x (1.13)
vii) (x +y) = x +y (1.14)
viii) 1 x = x (1.15)
A real vector space is dened similarly.
Functional Analysis 5
Examples (Function Spaces). Let R
n
be an open subset of R
n
.
1. P() is the space of all polynomials of n variables as functions on .
2. C() is the space of all continuous complex valued functions on .
3. C
k
() is the space of all complex valued functions with continuous
partial derivatives of order k on .
4. C

() is the space of all innitely dierentiable complex valued (smooth)

functions on .
Example (Sequence Spaces (l
p
-Spaces)). Let p 1. l
p
is the space of
all innite sequences {z
n
}

n=1
of complex numbers such that
_

n=1
|z
n
|
p
_1
p
< . (1.16)
Denition 12 Let V be a complex vector space and let x
1
, . . . x
k
V and

1
, . . . ,
k
C. A vector x =
1
x
1
+
k
x
k
is called a linear combina-
tion of x
1
, . . . x
k
.
Denition 13 A nite collection of vectors x
1
, . . . , x
k
is called linearly
independent if
k

i=1

i
x
i
= 0
i
= 0 , i = 1, 2, . . . , k . (1.17)
An arbitrary colection of vectors B = {x
n
}

n=1
is called linearly independent
if every nite subcollection is linearly independent. A collection of vectors
which is not linearly independent is called linearly dependent.
Denition 14 Let B V be a subset of a vector space V . Then span B is
the set of all nite linear combinations of vectors from B
span B =
_
k

i=1

i
x
i

x
i
B,
i
C, k N
_
. (1.18)
Proposition 3 Let B V be a subset of a vector space V . Then span B is
a subspace of V .
6 Functional Analysis
Denition 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 nite basis in V ,
then V is called nite dimensional vector space. Otherwise V is called
innite dimensional vector space.
Proposition 4 The number of vectors in any basis of a nite dimensional
vector space is the same.
Denition 16 The number of vectors in a basis of a nite dimensional vec-
tor space is called the dimension of V , denoted by dimV .
1.1.3 Normed Linear Spaces
Denition 17 A normed linear space is a vector space, V , over C (or
R) and a mapping || || : V R, called a norm, that satises:
i) ||v|| 0 v V (1.19)
ii) ||v|| = 0 v = 0 (1.20)
iii) ||v|| = || ||v|| v V, C (1.21)
iv) ||v +w|| ||v|| +||w|| v, w V (1.22)
Examples.
1. Norms in R
n
:
||x||
2
=
_
k

i=1
x
2
i
_
1
2
(1.23)
||x||
1
=
n

i=1
|x
i
| (1.24)
||x||

= max
1in
{|x
i
|} (1.25)
2. A norm in C
n
||z|| =
_
k

i=1
|z
i
|
2
_
1
2
(1.26)
Functional Analysis 7
3. Let R
n
be a closed bounded subset of R
n
and dx = dx
1
dx
n
be
a measure in R
n
. Norms in C() can be dened by
||f||

= sup
x
|f(x)| (1.27)
||f||
p
=
__

|f(x)|
p
dx
_1
p
(1.28)
||f||
1
=
_

|f(x)| dx (1.29)
4. A norm in l
p
||z|| =
_
k

i=1
|z
i
|
p
_
1
p
(1.30)
5. A norm in l

||z|| = sup
nN
|z
n
| (1.31)
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 dened as in a metric space in the
induced metric.
Denition 18 A normed linear space is complete if it is complete as a metric
space in the induced metric.
Denition 19 A complete normed linear space is called the Banach space.
Denition 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 satises:
i) T(v +w) = T(v) + T(w), v, w V, , C;
ii) ||T(v)||
V
C||v||
W
for some C 0.
8 Functional Analysis
iii) the number
||T|| = sup
vV,v=0
||T(v)||
W
||v||
V
(1.32)
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
Denition 21 Characteristic function of a set A X is a mapping

A
: X {0, 1} dened by

A
(x) =
_
1, if x A
0, if x A
(1.33)
Denition 22 For a non-zero function f : R
n
R, the set, supp f, of all
points x R
n
for which f(x) = 0 is called the support of f, i.e.
supp f = {x R
n
|f(x) = 0} . (1.34)
Clearly, supp
A
= A.
Denition 23 Let I be a semi-open interval in R
n
dened by
I = {x R
n
| a
k
x
k
< b
k
, k = 1, . . . , n} (1.35)
for some a
k
< b
k
. The measure of the set I is dened to be
(I) = (b
1
a
1
) (b
n
a
n
) . (1.36)
The Lebesgue integral of a characteristic function of the set I is dened
by
_

I
dx = (I) . (1.37)
Functional Analysis 9
Denition 24 A nite linear combination of characteristic functions of semi-
open intervals
f =
N

k=1

I
k
(1.38)
is called a step function.
Denition 25 The Lebesgue integral of a step function is dened by
linearity
_
N

k=1

I
k
dx =
N

k=1

k
(I
k
) . (1.39)
Denition 26 A function f : R
n
R is Lebesgue integrable if a
sequence of step functions {f
k
} such that
f

k=1
f
k
, (1.40)
which means that two conditions are satised
a)

k=1
_
|f
k
| dx < (1.41)
b) f(x) =

k=1
f
k
(x) x R
n
such that

k=1
|f
k
(x)| < . (1.42)
The Lebesgue integral of f is then dened by
_
f dx =

k=1
_
f
k
dx (1.43)
Proposition 6 The space, L
1
(R
n
), of all Lebesgue integrable functions on
R
n
is a vector space and
_
is a linear functional on it.
Theorem 9 a) If f, g L
1
(R
n
) and f g, then
_
fdx
_
gdx.
b) If f L
1
(R
n
), then |f| L
1
(R
n
) and |
_
f dx|
_
|f|dx.
10 Functional Analysis
Theorem 10 If {f
k
} is a sequence of integrable functions and
f

k=1
f
k
, (1.44)
then
_
f =

k=1
_
f
k
, (1.45)
Denition 27 The L
1
-norm in L
1
(R
n
) is dened by
||f|| =
_
|f|dx (1.46)
Denition 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.
Denition 29 The equivalence class of f L
1
(R
n
), denoted by [f], is
the set of all functions equivalent to f.
Remark. Strictly speaking, to make L
1
(R
n
) a normed space one has to
consider instead of functions the classes of equivalent functions.
Denition 30 A set X R
n
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.
Denition 31 Two integrable functions, f, g L
1
(R
n
), are said to be equal
almost everywhere, f = g a.e., if the set of all x R
n
for which f(x) =
g(x) is a null set.
Theorem 12
f = g a.e ||f g|| =
_
|f g| = 0 (1.47)
Theorem 13 The space L
1
(R
n
) is complete.
Functional Analysis 11
1.2 Hilbert Spaces
1.2.1 Geometry of Hilbert Space
Denition 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 satises: x, y, z V, 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, y) = (x, y) (1.51)
v) (x, y) = (y, x)

(1.52)
Denition 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, {x
i
}
N
i=1
, of vectors in V is called an orthonormal set
if (x
i
, x
j
) =
ij
, i.e. (x
i
, x
j
) = 1 if i = j and (x
i
, x
j
) = 0 if i = j.
Theorem 14 Every inner product space is a normed linear space 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 {x
n
}
N
n=1
be an orthonormal set in V . Then x V
||x||
2
=
N

n=1
|(x, x
n
)|
2
+

x
N

n=1
(x, x
n
)x
n

2
(1.53)
Theorem 16 (Bessel inequality) Let V be an inner product space and
{x
n
}
N
n=1
be an orthonormal set in V . Then x V
||x||
2

n=1
|(x, x
n
)|
2
(1.54)
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)
Denition 34 A sequence {x
n
} of vectors in an inner product space V is
called strongly convergent to x V , denoted by x
n
x, if
||x
n
||
n0
0 (1.57)
and weakly convergent to x V , denoted by x
n
w
x, if
(x
n
x, y)
n0
0 y V . (1.58)
Theorem 19
a) x
n
x = x
n
w
x (1.59)
b) x
n
w
x and ||x
n
|| ||x|| = x
n
x (1.60)
Denition 35 A complete inner product space is called a Hilbert space.
Denition 36 A linear transformation U : H
1
H
2
from a Hilbert space
H
1
onto the Hilbert space H
2
is called unitary if if it preserves the inner
product, i.e. x, y H
1
(Ux, Uy)
H
2
= (x, y)
H
1
. (1.61)
Denition 37 Two Hilbert spaces H
1
and H
2
are said to be isomorphic
if there is a unitary linear transformation U from H
1
onto H
2
.
Denition 38 Let H
1
and H
2
be Hilbert spaces. The direct sum H
1
H
2
of Hilbert spaces H
1
and H
2
is the set of ordered pairs z = (x, y) with x H
1
and y H
2
with inner product
(z
1
, z
2
)
H
1
H
2
= (x
1
, x
2
)
H
1
+ (y
1
, y
2
)
H
2
(1.62)
1.2.2 Examples of Hilbert Spaces
1. Finite Dimensional Vectors. C
N
is the space of N-tuples x =
(x
1
, . . . , x
N
) of complex numbers. It is a Hilbert space with the inner product
(x, y) =
N

n=1
x

n
y
n
. (1.63)
Functional Analysis 13
2. Square Summable Sequences of Complex Numbers. l
2
is the
space of sequences of complex numbers x = {x
n
}

n=1
such that

n=1
|x
n
|
2
< . (1.64)
It is a Hilbert space with the inner product
(x, y) =

n=1
x

n
y
n
. (1.65)
3. Square Integrable Functions on R. L
2
(R) is the space of complex
valued functions such that
_
R
|f(x)|
2
dx < . (1.66)
It is a Hilbert space with the inner product
(f, g) =
_
R
f

(x)g(x) dx (1.67)
4. Square Integrable Functions on R
n
. Let be an open set in R
n
(in
particular, can be the whole R
n
). The space L
2
() is the set of complex
valued functions such that
_

|f(x)|
2
dx < , (1.68)
where x = (x
1
, . . . , x
n
) and dx = dx
1
dx
n
. It is a Hilbert space with
the inner product
(f, g) =
_

(x)g(x) dx (1.69)
5. Square Integrable Vector Valued Functions. Let be an open set
in R
n
(in particular, can be the whole R
n
) and V be a nite-dimensional
vector space. The space L
2
(V, ) is the set of vector valued functions f =
(f
1
, . . . , f
N
) on such that
N

i=1
_

|f
i
(x)|
2
dx < . (1.70)
14 Functional Analysis
It is a Hilbert space with the inner product
(f, g) =
N

i=1
_

i
(x)g
i
(x) dx (1.71)
6. Sobolev Spaces. Let be an open set in R
n
(in particular, can
be the whole R
n
) and V a nite-dimensional complex vector space. Let
C
m
(V, ) be the space of complex vector valued functions that have partial
derivatives of all orders less or equal to m. Let = (
1
, . . . ,
n
), N, be
a multiindex of nonnegative integers,
i
0, and let || =
1
+ +
n
.
Dene
D

f =

||
x

1
1
x

n
n
f . (1.72)
Then f C
m
(V, ) i
|D

f
i
(x)| < , || m, i = 1, . . . , N, x . (1.73)
The space H
m
(V, ) is the space of complex vector valued functions such
that D

f L
2
(V, ) , || m, i.e. such that
N

i=1
_

|D

f
i
(x)|
2
dx < , || m. (1.74)
It is a Hilbert space with the inner product
(f, g) =

, ||m
N

i=1
_

(D

f
i
(x))

g
i
(x) dx (1.75)
Remark. More precisely, the Sobolev space H
m
(V, ) is the completion of
the space dened above.
1.2.3 Projection Theorem
Denition 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

), of linear transformations from a Hilbert space

H to H

is a Banach space under the norm

||T|| = sup
||x||
H
=1
||Tx||
H
(1.77)
Denition 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

y
T
H such that x H
T(x) = (y
T
, x), and ||T||
H
= ||y
T
||
H
(1.78)
1.2.4 Orthonormal Bases
Denition 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 = {x

}
A
be an orthonormal basis for a Hilbert space
H. Then y H
y =

A
(x

, y)x

(1.79)
||y||
2
=

A
|(x

, y)|
2
(1.80)
16 Functional Analysis
Denition 42 Let S = {x

}
A
be an orthonormal basis for a Hilbert space
H. The coecients (x

, y) are called the Fourier coecients of y H

with respect to the basis S.
Denition 43 A metric space which has a countable dense subset is said to
be separable.
Theorem 26 A Hilbert space H is separable i it has a countable orthonor-
mal basis S. If S contains nite number, N, of elements, then H is isomor-
phic to C
N
. If S contains countably many elements, then H is isomorphic
to l
2
.
1.2.5 Tensor Products of Hilbert Spaces
Let H
1
and H
2
b Hilbert spaces. For each
1
H
1
and
2
H
2
let
1

2
denote the conjugate bilinear form on H
1
H
2
dened by
(
1

2
)(
1
,
2
) = (
1
,
1
)
H
1
(
2
,
2
)
H
2
(1.81)
where
1
H
1
and
2
H
2
. Let E be the set of nite linear combinations
of such bilinear forms. An inner product on E can be dened by
( , )
E
= (, )
H
1
(, )
H
2
(1.82)
(with , H
1
and , H
2
) and extending by linearity on E.
Denition 44 The tensor product H
1
H
2
of the Hilbert paces H
1
and H
2
is dened to be the completion of E under the inner product dened above.
Theorem 27 Let H
1
and H
2
be Hilbert spaces. If {
k
} and {
l
} are or-
thonormal bases for H
1
and H
2
respectively, then {
k

l
} is anorthonormal
basis for the tensor product H
1
H
2
.
Fock Spaces. Let H be a Hilbert space and let
H
0
= C (1.83)
H
n
= H H
. .
n
(1.84)
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 = L
2
(R), then an element F(H) is a
sequence of functions
= {
o
,
1
(x), (x
1
, x
2
),
3
(x
1
, x
2
, x
3
), . . .} (1.86)
so that
|||| = |
0
|
2
+

n=1
_
R
n
|
n
(x
1
, . . . , x
n
)|
2
dx
1
. . . dx
n
< (1.87)
Let P
n
be the permutation group of n elements and let {
k
} be a basis
for H. Each P
n
denes a permutation
(
k
1

k
n
) =
k
(1)

k
(n)
. (1.88)
By linearity this can be extended to a bounded operator on H
n
, so one can
dene
S
n
=
1
n!

P
n
(1.89)
A
n
=
1
n!

P
n
() (1.90)
where
() =
_
1, if is even
1 if is odd
(1.91)
Finally, the Boson (symmetric) Fock space is dened by
F
s
(H) =

n=0
S
n
H
n
(1.92)
and the Fermion (antisymmetric) Fock space is dened by
F
a
(H) =

n=0
A
n
H
n
(1.93)
In the case H = L
2
(R),
n
S
n
H
n
is a function of n variables symmetric
under any permutations of variables, and
n
A
n
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.
Denition 45 The following are asymptotic estimates
i) f(x) g(x) (x a, x M)
if lim
xa, xM
f(x)
g(x)
= 1 (2.1)
ii) f(x) = o(g(x)) (x a, x M)
if lim
xa, xM
f(x)
g(x)
= 0 (2.2)
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

) (x 0
+
), > 0 . (2.5)
2.
ln x = o(x

) (x ), > 0 . (2.6)
3.
sin z z (z 0) . (2.7)
4.
sin x = O(1) (x R) . (2.8)
5.
n!

2ne
n
n
n
(n ) . (2.9)
Remarks. The relation f(x) = o(g(x)) means that f(x) is innitesimal
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 innitesimal 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
Denition 46 Let
n
: M R, n N, and a be a limit point of M.
Let
n
(x) = 0 in a neighborhood U
n
of a. The the sequence {
n
} is called
asymptotic sequence at x a, x M if n N

n+1
(x) = o(
n
(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 {
n
} be a decreasing sequence of real numbers, i.e.
n
<
n+1
, and
let 0 < /2. Then the sequence
{e

n
z
}, z , | arg z|

2
(2.23)
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) = 0 for x M in some neighborhood of a and {
n
} be an
asymptotic sequence at x a, x M. Then the sequence {f(x)
n
(x)}
is an asymptotic sequence as x a, x M.
3. Let {
n
(x)}, {
n
(x)} be asymptotic sequences as x a, x M. Then
the sequence {
n
(x)
n
(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.
Denition 47 Let {
n
} be an asymptotic sequence as x a, x M. We
say that the function f is expanded in an asymptotic series
f(x)

n=0
a
n

n
(x), (x a, x M), (2.24)
where a
n
are constants, if N 0
R
N
(x) f(x)
N

n=0
a
n

n
(x) = o(
N
(x)), (x a, x M) . (2.25)
This series is called asymptotic expansion of the function f with respect
to the asympotic sequence {
n
}. R
N
(x) is called the rest term of the
asymptotic series.
Remarks
1. The condition R
N
(x) = o(
N
(x)) means, in particular, that
lim
xa
R
N
(x) = 0 for any xed N (2.26)
2. Asymptotic series could diverge. This happens if
lim
N
R
N
(x) = 0 for some xed x (2.27)
3. There are three possibilities:
(a) asymptotic series converges to f(x);
(b) asymptotic series converges to a function g(x) = 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 dierent functions can have the same asymptotic expansion.
For example, f(x) = e
x
and g(x) = e
x
+ e
1/x
have the same asymptotic
expansion with respect to the asymptotic sequence {x
n
}:
e
x
e
x
+ e
1/x

n=0
x
n
n!
, x 0+ (2.28)
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.
Denition 48 Let f : MS R be a function of two variables and a be a
limit point of M and {
n
} be an asymptotic sequence as x a. Let for any
xed y S the function f is expanded in an asymptotic series
f(x, y)

n=0
a
n
(y)
n
(x), (x a, x M) . (2.29)
This asymptotic expansion is called uniform with respect to the parameter
y S, if the relation
R
N
(x, y) f(x, y)
N

n=0
a
n
(y)
n
(x) = o(
N
(x)), (x a, x M) (2.30)
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, dierentiate 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
F() =
_
a
0
f(x, ) dx (2.31)
24 Asymptotic Expansions
where a > 0 and > 0 is a small positive parameter. Here f C

([0, a]
(0,
0
]) is a smooth function for 0 x a, 0 <
0
, with some
0
. Then
the integral converges for > 0. Let f have a singularity when = 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() has a weak
singularity.
This denition is obviously extended for unbounded intervals. Let
F() =
_

a
f(x, ) dx (2.32)
where a > 0 and > 0 is a small parameter. Here f C

([a, ) (0,
0
]) is
a smooth function for a x , 0 <
0
, with some
0
. Let the integral
converge for > 0 and diverge for = 0. If the function g(x) = f(x, 0) is
of power or logarithmic order at x , then we say that the integral F()
has a weak singularity.
2.4.1 Power Singularity on a Bounded Interval
Let a, , R, a, > 0, be some real numbers and > 0 be a small positive
prameter. Let C

[0, a] be a smooth function on [0, a]. We will study

the asymptotics as 0+ of the integrals of the form
F() =
_
a
0
t
1
(t +)

(t) dt . (2.33)
Remarks. The function F is holomorphic in complex plane with a cut
along the negative half-axis. At the point = 0 this function has a singularity
(if > 0 is not integer). The type of this singularity is determined by the
behavior of the function at small t 0.
Standard Integral. One needs the following result. Let and be two
complex numbers such that Re > 0, Re < 0 and Re ( +) < 0. Then
_

0
t
1
(t + 1)

dt =
()( )
()
, (0 < Re < Re ) . (2.34)
Theorem 32 Let C

[0, a]. Let r, > 0 and S

= { C|0 < ||
r, | arg | } be a sector in the complex plane of .
Asymptotic Expansions 25
1. If + is not integer, then
F()

n=0
( +n)( n)
()

(n)
(0)
n!

++n
+

n=0
a
n

n
( 0, S

) (2.35)
2. If + = N is an integer, then
F()

nmax{0,N}
(N +n)
()(N +n )

(n)
(0)
n!

n+N
ln +

n=0
b
n

n
( 0, S

) (2.36)
The coecients a
n
and b
n
depend on the values (t) for 0 t a. The
branch for the functions

and ln is choosen in such a way that

> 0 and
ln is real for > 0.
Examples. In all examples C

([0, a]) is a smooth function bounded

with all its derivatives.
1. Let 0 < a < 1 and
F() =
_
a
0
(t)
t +
dt . (2.37)
Then
F() = (0) ln +O(1), ( 0
+
) (2.38)
2. Let 0 < a < 1 and
F() =
_
a
0
(t)
t
2
+
2
dt . (2.39)
By using
1
t
2
+
2
=
1
2i
_
1
t i

1
t +i
_
(2.40)
we obtain from the previous example
F() = (0)

2
+O(1) ( 0
+
) (2.41)
26 Asymptotic Expansions
3. Let > 1/2 and
F() =
_
a
0
(t)
(t
2
+
2
)

dt . (2.42)
Then
F() = (0)

2
( 1/2)
()

12
+O(
32
)+O(1), ( 0
+
) (2.43)
2.4.2 Power singularity on Unbounded Interval
Standard Integral. To compute the following asymptotics one needs the
following standard integral. Let Re > 0 and Re > 1. Then
_

0
t

e
t

dt =
1

_
+ 1

_
, (Re > 0, Re > 1) . (2.44)
Examples. Let C

([a, )) be a smooth function on [a, ) that has

asymptotic expansion as x
(x)

k=0
a
k
x
k
. (2.45)
1. Let a, > 0, and
F() =
_

a
(x)x

e
x

dx . (2.46)
If < 1, then the integral is not singular as 0
+
. Its asymptotic
expansion can be obtained either by integration by parts or by a change
of variables.
So, let now + 1 > 0 and let N = [ + 1] 0 be the integer part
of + 1. Let us single out the rst N + 1 terms of the asymptotic
expansion in , i.e.
(x) =
N

k=0
a
k
x
k
+R
N
(x) . (2.47)
Since R
N
(x) = O(x
(N+1)
) as x , we have
F() =
N

k=0
a
k
_

a
x
k
e
x

dx +O(1) . (2.48)
Asymptotic Expansions 27
Now by changing the variables and extending the interval to [0, ) we
obtain
F() =
N

k=0
a
k

k+1

_
k + 1

_
+O(1)
_
+O(1)
=
a
0

+1

_
+ 1

_
+O(

) (2.49)
If = 1, then
F() =
a
0

ln +O(1), ( 0
+
) . (2.50)
2. Let a > 0 and let
P(x) = x
n
+ + a
1
x n 1 . (2.51)
Consider the integral
F() =
_

a
(x)x

e
P(x)
dx . (2.52)
If > 1, then the main term of the asymptotics is
F() =
a
0
n

_
+ 1
n
_

+1
n
+O(

n
), ( 0
+
) (2.53)
If = 1, then
F() =
a
0
n
ln + O(1), ( 0
+
) (2.54)
If < 1, then the integral is not singular as 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, : M C a complex valued function and be a large positive
parameter. Consider the integrals of the form
F() =
_
b
a
(x) exp[S(x)] dx . (3.1)
Such integrals are called Laplace integrals. We will study the asymptotics
of the Laplace integrals as .
Lemma 1 Let sup
a<x<b
S(x) = L < and the integral (3.1) converges
absolutely for some
0
> 0. Then
1.
|F()| C|e
L
| (Re
0
) . (3.2)
2. if f, S C(a, b), then F() is holomorphic in the halfplane Re >
0
.
3.1.1 Watson Lemma
Lemma 2 (Watson) Let 0 < a < , > 0, > 0 and let S

be the sector
S

= { C | | arg | /2 } in the complex plane . Let C

([0, a])
and let
() =
_
a
0
(x)x
1
exp(x

) dx (3.3)
29
30 Laplace Method
Then, there is an asymptotic expansion as , S

,
()
1

k=0

(+k)/

_
+k

_

(k)
(0)
k!
(3.4)
Laplace Transform. Let C

(R
+
) be a smooth function on the posi-
tive real axis such that its Laplace transform
L()() =
_

0
(x)e
x
dx (3.5)
converges absolutely for some
0
. Then
L()()

k=0

(k)
(0) (|| , S

) (3.6)
3.1.2 Interior Nondegenerate Maximum Point
Let now S and be smooth functions and the function S have a maximum
at an interior point x
0
of the interval [a, b], i.e. a < x
0
< b. Then S

(x
0
) = 0.
Assume, for simplicity, that S

(x
0
) = 0. Then S

(x
0
) < 0. In other words,
in a neighborhood of x
0
the function S has the following Taylor expansion
S(x) = S(x
0
) + S

(x
0
)
(x x
0
)
2
2
+O((x x
0
)
3
) . (3.7)
Such a point is called nondegenerate critical point.
Then, as the main contribution to the integral comes from a small
neighborhood of x
0
. In this neighborhood the function is almost constant
and can be replaced by its value at x
0
. The terms of order (x x
0
)
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
_

exp
_

2
y
2
_
dy =
_
2

, (Re > 0) , (3.8)

one obtains nally the main term of the asymptotics
F()
1/2

2
S

(x
0
)
(x
0
)e
S(x
0
)
, ( ) (3.9)
One can prove the general theorem.
Laplace Method 31
Theorem 33 Let M = [a, b] and , S C

(M), S has a maximum only at

one point x
0
, a < x
0
< b and S

(x
0
) = 0. Then as , S

there is
asymptotic expansion
F() e
S(x
0
)

k=0
c
k

1/2k
. (3.10)
The coecients c
k
are expressed in terms of derivatives of and S at x
0
.
The theorem can be proved as follows. First, we change the integration
variable
x = x
0
+
1/2
y . (3.11)
So, y is the scaled uctuation from the maximum point x
0
. The interval
of integration should be changed accordingly, so that the maximum point is
now y = 0. Then, we expand both functions S and in Taylor series at x
0
getting
S(x
0
+
1/2
y) = S(x
0
) +
1
2
S

(x
0
)y
2
+

n=3
S
(n)
(x
0
)
n!
y
n

(n2)/2
, (3.12)
(x
0
+
1/2
y) =

n=0

(n)
(x
0
)
n!
y
n

n/2
. (3.13)
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
_

exp
_

2
y
2
_
y
2k+1
dy = 0 , (3.14)
_

exp
_

2
y
2
_
y
2k
dy =
_
k +
1
2
_
_

2
_
k1
, (3.15)
where k is a nonnegative integer and has a positive real part, Re > 0.
Finally, we get a power series in inverse powers of . The coecients c
k
of
the asymptotic expansion are polynomials in the higher derivatives S
(k)
(x
0
),
k 3, and derivatives
(l)
(x
0
), l 0, and involve inverse powers of S

(x
0
).
32 Laplace Method
Stirling Formula
(x + 1) =

2x
x+1/2
e
x
[1 +O(x
1
)], (x ) (3.16)
is obtained by applying the Laplace method to the integral
(x + 1) = x
x+1
_

0
exp[x(ln t t)] dt (3.17)
Stieltjes Transform Let : R
+
C have nite moments
m
n
() =
_

0
t
n
(t) dt < n N. (3.18)
Then the Stieltjes transform of
S()(x) =
_

0
(t)
t +x
dt (3.19)
has asymptotic expansion as x
S()(x)

k=0
(1)
k
m
k
()x
1k
(3.20)
3.1.3 Boundary Maximum Point
Let the function S have a maximum at a boundary point x
0
= a. Let, for
simplicity, S

(a) = 0, i.e. S

(a) < 0, and (a) = 0. Then, as , the

main contribution to the integral comes from the interval [a, a +], where
S(x) = S(a) + (x a)S

(a) +O((x a)
2
) . (3.21)
Now, by replacing the function by its value at a and neglecting nonlinear
terms in S(x), we get
F()
1
(a)
S

(a)
e
S(a)
, ( ) (3.22)
In this way one can prove the following theorem.
Theorem 34 Let M = [a, b], , S C

(M), S has a maximum only at the

point x = a and S

(a) = 0. Then, as , S

, there is asymptotic
expansion
F() e
S(a)

k=0
c
k

1k
. (3.23)
The coecients c
k
are expressed in terms of derivatives of and S at x = a.
Laplace Method 33
Error Function The asymptotic expansion of the (complementary) error
function as x has the form
Erfc x =
_

x
e
t
2
dt
e
x
2
2x

k=0
(1)
k
(2k 1)!!
2
k
x
2k
(x ) . (3.24)
Incomplete Gamma Function The incomplete gamma-function
(a, x) =
_
x
0
t
a1
e
t
dt, (0 < a < , x > 0) (3.25)
has the following asympotic expansion as x
(a, x) = (a) +e
x
x
a1

k=0
(a)
(a k)
x
k
(3.26)
3.2 Background from Analysis
3.2.1 Denitions
1. Let x
j
, j = 1, . . . , n be real numbers and let x be the n-tuple x =
(x
1
, . . . x
n
). The set of all n-tuples of real numbers is denoted by R
n
.
A connected open subset of R
n
is called a domain. The set of
boundary points of is called the boundary of . The union
is called the closure of and is denoted by .
2. Let x, R
n
. Then the scalar product of x and is dened by
x = x
1

1
+ +x
n

n
.
3. On R
n
there is standard Lebesgue measure dx = dx
1
dx
n
.
4. Let
j
, j = 1, . . . , n, be non-negative integers,
j
0. The n-tuple
= (
1
, . . . ,
n
) is called a multi-index. Further, let
|| =
1
+ +
n
(3.27)
! =
1
!
n
! (3.28)

f(x) =

||
f(x)
(x
1
)

1
(x
n
)

n
(3.29)
34 Laplace Method
D

f(x) =
_
1
i

x
1
_

1

_
1
i

x
n
_

n
f(x) (3.30)
Then the Taylor expansion of a smooth function at x
0
can be written
in the form
(x) =

||=0
1
!
[

(x
0
)] (x x
0
)

. (3.31)
5. The boundary is said to be smooth, denoted C

, if in a
neighborhood of any boundary point x
0
it can be locally dened
by an equation x
j
= (x

) with a smooth function .

6. The set of all continuous functions on is denoted by C().
7. The set of all functions with continuous partial derivatives up to order
k on is denoted by C
k
().
8. The set of all functions with continuous partial derivatives up to order
k on is denoted by C
k
().
9. The set of all functions with continuous partial derivatives up to order
k on that vanish in a neighborhood of the boundary is denoted
by C
k
0
().
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 | f(x) = 0} . (3.32)
11. A mapping : , is said to be of class C
k
if C
k
().
12. A one-to-one mapping : of onto is called dieomor-
phism of class C
k
if C
k
() and
1
C
k
().
13. Let
j
, j = 1, . . . , k be some scalar functions on R
n
and be a k-tuple
= (
1
, . . . ,
k
). In other words, : R
n
R
k
. The matrix

x
(x) =
_

i
(x)
x
j
_
, i = 1, . . . , k; j = 1, . . . , n (3.33)
is called the Jacobi matrix.
Laplace Method 35
Theorem 35 (Inverse Function Theorem) Let : R
n
R
n
is of class
C
k
, k 1, in a neighborhood of a point x
0
and det
x
(x
0
) = 0. Then is
local dieomorphism of class C
k
in a neighborhood of the point x
0
.
Theorem 36 (Implicit Function Theorem) Let be a domain in R
2n
,
let F : R
n
be a mapping of class C
k
() and let (x
0
, y
0
) be a point
in such that
F(x
0
, y
0
) = 0, det
y
F(x
0
, y
0
) = 0 . (3.34)
Then in a neighborhood of the point x
0
there is a mapping y = f(x) of class
C
k
such that y
0
= f(x
0
) and
F(x, f(x)) 0 . (3.35)
3.2.2 Morse Lemma
Let S : R be a real valued function of class C
k
on a domain in R
n
with k 2. Let

2
x
S(x) =
_

2
S(x)
x
i
x
j
_
, i, j = 1, . . . , n . (3.36)
Denition 49 1. The point x
0
is called a critical point of the function
S if S(x
0
) = 0
2. A critical point x
0
is called non-degenerate if det
2
x
S(x
0
) = 0.
3. The determinant det
2
x
S(x
0
) is called the Hessian of the function S
at the point x
0
.
Lemma 3 (Morse) Let S : R
n
R and x
0
R
n
be a non-degenerate
critical point of the function S. Let S C

in a neighborhood of the point

x
0
and let
j
= 0, j = 1, . . . , n be the eigenvalues of the matrix
2
x
S(x
0
).
Then there are neighborhoods U and V of the points x
0
and 0 and a smooth
local dieomorphism : V U of class C

such that det

y
(0) = 1 and
S((y)) = S(x
0
) +
1
2
n

j=1

j
(y
j
)
2
. (3.37)
36 Laplace Method
Remark. Nondegenerate critical points are isolated.
3.2.3 Gaussian Integrals
Proposition 7 Let A = (a
ij
be a complex symmetric nondegenerate n n
matrix with the eigenvalues
j
(A), j = 1, . . . , n,. Let Re A 0, which means
that x Re Ax 0 x R
n
, x = 0, or Re
j
(A) 0, j = 1, . . . , n. Then for
> 0, R
n
there holds
_
R
n
exp
_

2
x Ax i x
_
dx
=
_
2

_
n/2
(det A)
1/2
exp
_

1
2
A
1

_
. (3.38)
The branch of

det A is choosen as follows

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

j=1
arg
j
(A), | arg
j
(A)|

2
. (3.40)
By expanding both sides of this equation in Taylor series in we obtain
the following result.
Corollary 1
_
R
n
exp
_

2
x Ax
_
x
i
1
x
i
2k+1
dx = 0 (3.41)
_
R
n
exp
_

2
x Ax
_
x
i
1
x
i
2k
dx
=
_
2

_
n/2
(det A)
1/2
(2)
k
(2k)!
k!
G
(i
1
i
2
G
i
2k1
i
2k
)
. (3.42)
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

+
(A) and

(A) be the numbers of positive and negative eigenvalues of A

and
sgn A =
+

(3.43)
be the signature of the matrix A. Then for > 0, R
n
there holds
_
R
n
exp
_
i

2
x Ax i x
_
dx
=
_
2

_
n/2
| det A|
1/2
exp
_

i
2
A
1
+
i
4
sgn (A)
_
. (3.44)
3.3 Laplace Integrals in Many Dimensions
3.3.1 Interior Maximum Point
Let be a bounded domain in R
n
, S : R, f : C are some functions
on and > 0 be a large positive parameter. We will study the asymptotics
as of the multidimensional Laplace integrals
F() =
_

f(x) exp[S(x)] dx. (3.45)

Let S and f be smooth functions and the function S have a maximum
only at one interior nondegenerate critical point x
0
. Then
x
S(x
0
) = 0
and [det
2
x
S(x
0
)] < 0. Then in a neighborhood of x
0
the function S has the
following Taylor expansion
S(x) = S(x
0
) +
1
2
(x x
0
) [
2
x
S(x
0
)](x x
0
) +O((x x
0
)
3
) . (3.46)
One could also use the Morse Lemma to replace the function S by a quadratic
form. Then as the main contribution to the integral comes from a
small neghborhood of x
0
. 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 x
0
, it can be replaced by its value at x
0
. Then the region
of integration can be extended to the whole R
n
. By using the formula for
the standard Gaussian integral one gets then the leading asymptotics of the
integral F() as
F() exp[S(x
0
)]
_
2

_
n/2
_
det
2
x
S(x
0
)

1/2
f(x
0
) . (3.47)
38 Laplace Method
One can prove the general theorem.
Theorem 37 Let f, S C

() and let x
0
be a nondegenerate critical point
of the function S where it has the only maximum in . Let 0 < < /2.
Then there is asymptotic expansion as in the sector S

= {
C | | arg | /2 }
F() exp[S(x
0
)]
n/2

k=0
a
k

k
. (3.48)
The coecients a
k
are expressed in terms of the derivatives of the functions
f and S at the point x
0
.
The idea of the proof is the same as in the one-dimensional case and goes
as follows. First, we change the integration variables
x
i
= x
i
0
+
1/2
y
i
. (3.49)
So, y is the scaled uctuation from the maximum point x
0
. The interval
of integration should be changed accordingly, so that the maximum point is
now y = 0. Then, we expand both functions S and in Taylor series at x
0
getting
S(x
0
+
1/2
y) = S(x
0
) +
1
2
y [
2
x
S(x
0
)]y +

||=3

(||2)/2
!
[

S(x
0
)] y

,
(3.50)
(x
0
+
1/2
y) =

||=0

||/2
!

(x
0
) y

. (3.51)
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 R
n
and compute the standard Gaus-
sian integrals. Finally, we get a power series in inverse powers of . The
coecients a
k
of the asymptotic expansion are polynomials in the higher
derivatives

S(x
0
), || 3, and derivatives

(x
0
), || 0, and involve
inverse matrices G = [
2
x
S(x
0
)]
1
.
Laplace Method 39
Remark. If x
0
is a degenerate maximum point of the function S, then
the asymptotic expansion as has the form
F() exp[S(x
0
)]
n/2

k=0
N

l=0
a
kl

r
k
(ln )
l
, (3.52)
where N is some positive integer and {r
k
}, r
k
n/2, k N, is a increasing
sequence of nonnegative rational numbers.
The coecients a
k
(and a
kl
) of the asymptotic expansion of the integral
F() are invariants under smooth local dieomorphisms in a neighborhood
of x
0
and play very important role in various applications.
3.3.2 Boundary Maximum Point
Let now S has maximum at the boundary point x
0
. We assume,
for simplicity, that both the boundary and the function S are smooth, i.e.
S C

and C

.
Since the boundary is smooth we can smoothly parametrize it in a neigh-
borhood of x
0
by (n 1) parameters = (
a
), (a = 1, . . . , n 1). Let the
the parametric equations of the boundary be
x
i
= x
i
(), i = 1, . . . , n. (3.53)
Then
T
a
= (T
i
a
) =
_
x
i

a
_
(3.54)
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 we have r > 0. Obviously, r(x()) 0. From this equation
we obtain that the vector
N = (N
i
) =
_
r
x
i
_
(3.56)
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 dened as usual

r
=
n

i=1
x
i
r

x
i
,

a
=
n

i=1
x
i

x
i
(3.57)
The point x
0
is not, in general, a critical point of S, since the normal
derivative of S at x
0
does not have to be equal to zero.
Denition 50 The point x
0
is said to be a nondegenerate boundary
maximum point of S if

r
S(x
0
) = 0 (3.58)
and the (n 1) (n 1) matrix
2

S(x()) is negative denite.

In a neighborhood of a nondegenerate boundary maximum point the func-
tion S has the following Taylor expansion
S(x) = S(x
0
) + [
r
S(x
0
)]r +
1
2
[
2
r
S(x
0
)]r
2
+[
r

S(x
0
)] (
0
)r +
1
2
(
0
) [
2

S(x
0
)](
0
)
+ , (3.59)
up to third order terms in r and (
0
).
Now we replace the integral F() by an integral over a small neighborhood
of x
0
. We change the variables of integration from x
i
, i = 1, . . . , n, to (
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 x
0
. In the remaining
integral we extend the integration to the whole space R
+
R
n1
, i.e. we
integrate over r from 0 to and integrate over the whole tangent plane
at x
0
. These integrals are standard Gaussian integrals and we obtain the
leading asymptotics as
F()
(n+1)/2
(2)
(n1)/2
exp[S(x
0
)]
[
r
S(x
0
)]
1
[det
2

S(x
0
)]
1/2
J(x
0
)f(x
0
) (3.60)
where J(x
0
) 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

() and let S have a maximum only at a non-

degenerate boundary maximum point x
0
. Then as , S

,
F()
(n+1)/2
exp[S(x
0
)]

k=0
a
k

k
(3.61)
3.3.3 Integral Operators with Singular Kernels
Let be a bounded domain in R
n
including the origin, 0 . Let S be a
real valued non-positive function on of class C
2
that has maximum equal
to zero, S(0) = 0, only at a nondegenerate maximum critical point x
0
= 0.
Let K

: C

() C

() be a linear integral operator dened by

(K

f)(x) =
_

2
_
n/2
_

exp[S(x y)]f(y) dy . (3.62)

Let M be a compact subset of . Then
lim

(K

f)(x) = [det
2
x
S(0)]
1/2
f(x) (3.63)
uniformly for x M.
Formally
_

2
_
n/2
exp[S(x y)] [det
2
x
S(0)]
1/2
(x y) (3.64)
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 be a large positive parameter. Consider the integrals of the form
F() =
_
b
a
f(x) exp[iS(x)] dx . (4.1)
The function S is called phase function and such integrals are called
Fourier Integrals. We will study the asymptotics of such integrals.
As the integral F() is small due to rapid oscillations of exp(iS).
Lemma 4 (Riemann-Lebesgue) Let f be an integrable function on the
real line, i.e. f L
1
(R). Then
_
R
f(x)e
ix
dx = o(1), ( ) . (4.2)
Denition 51 1. A point x
0
is called the regular point of the Fourier
integral F() if the functions f and S are smooth in a neighborhood of
x
0
and S

(x
0
) = 0.
43
44 Stationary Phase Method
2. A point x
0
is called the critical point of the integral F() if it is not
a regular point.
3. A critical point x
0
is called isolated critical point if there is a
neighborhood of x
0
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
signicantly to F().
4.1.2 Localization Principle
Lemma 5 Let S C

(R) be smooth function and f C

0
(R) be a smooth
function of compact support. Then as
_
R
f(x) exp[iS(x)] dx = O(

) (4.3)
Remarks.
1. Since the function f has compact support, the integral is, in fact, over
a nite 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 ner 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() have nite number of isolated
critical points. Then as the integral F() is equal to the sum of the
contributions of all critical points up to O(

).
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

(x) = 0 x M. Then as
F()

k=0
(i)
k1
_
1
S

(x)

x
_
k
_
f(x)
S

(x)
_
e
iS(x)

b
a
. (4.4)
The leading asymptotics is
F() = (i)
1
{f(b) exp[iS(b)] f(a) exp[iS(a)]} + O(
2
) (4.5)
The same technique, i.e. integration by parts, applies to the integrals
over an unbounded interval, say,
F() =
_

0
f(x) exp[iS(x)] dx. (4.6)
with some additional conditions that guarantee the converges at as well
as to the integrals of the form
F(x) =
_

x
f(t) exp[iS(t)] dt (4.7)
as x .
46 Stationary Phase Method
Examples
1. Let > 0. Then as
(a)
_

0
e
ix
(1 + x)

dx = i
1
+O(
2
) . (4.8)
(b)
_

0
sin(x)
(1 +x)

dx =
1
+O(
2
) . (4.9)
(c)
_

0
cos(x)
(1 +x)

dx =
2
+O(
3
) . (4.10)
2. Let > 0. Then as x
F(x) =
_

x
t

e
it
dt ie
ix
x

k=0
(k +)
()
(ix)
k
. (4.11)
In particular, the Frenel integral has the asymptotic expansion as x

(x) =
_

x
t

e
it
dt
i
2

e
ix
2
x
1/2

k=0
(1)
k

_
k +
1
2
_
x
k
.
(4.12)
4.1.4 Standard Integrals
Consider the integral
() =
_
a
0
f(x)x
1
e
ix

(4.13)
Lemma 6 (Erdeyi) Let 1, > 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
() =
_
a
0
f(x)x
1
e
ix

k=0
a
k

(+k)/
, (4.14)
Stationary Phase Method 47
where
a
k
=
f
(k)
(0)
k!
1

_
+k

_
exp
_
i
+k

_
(4.15)
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 C

0
(M) be a complex
valued function with compact support in M. Let S have a single isolated
nondegenerate critical point x
0
in M, i.e. S

(x
0
) = 0 and S

(x
0
) = 0. Then
as there is asymptotic expansion of the Fourier integral
F() =
_
b
a
f(x) exp[iS(x)] dx
exp
_
iS(x
0
) + [sgn S

(x
0
)] i

4
_

1/2

k=0

k
. (4.16)
The coecients a
k
are determined in terms of the derivatives of the functions
S and f at x
0
.
The leading asymptotics as is
F() =

2
|S

(x
0
)|

1/2
exp
_
iS(x
0
) + [sgn S

(x
0
)] i

4
_
_
f(x
0
) + O(
1
)
_
.
(4.17)
To prove this theorem one does a change of variables in a suciently
small neghborhood of x
0
and applies the Erdelyi lemma.
4.1.6 Principal Values of Integrals
Let f be a smooth function and consider the integral
_
b
a
f(x)
x
dx (4.18)
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
I() =
_

a
f(x)
x
dx +
_
b

f(x)
x
dx . (4.19)
Denition 52 If the limit of I() as 0+ exists, then it is called the
principal value of the integral I
P
_
b
a
f(x)
x
dx = lim
0
+
__

a
f(x)
x
dx +
_
b

f(x)
x
dx
_
. (4.20)
In this section we consider the asymptotics of the integrals of the form
F() = P
_
R
e
iS(x)
f(x)
dx
x
(4.21)
as .
Lemma 7 Let f C

0
(R) be a smooth function of compact support. Then
as
P
_
R
e
ix
f(x)
dx
x
= if(0) + O(

) . (4.22)
Theorem 42 Let f C

0
(R) be a smooth function of compact support,
S C

(R) be a real valued smooth function and S

(0) = 0. Then as
F() = P
_
R
e
iS(x)
f(x)
dx
x
= [sgn S

(0)] if(0) exp[iS(0)] + O(

) .
(4.23)
Theorem 43 Let f C

0
(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
and S

(0) = 0. Then as there is asymptotic expansion

F() = P
_
R
e
iS(x)
f(x)
dx
x
exp[iS(0)]
1/2

k=0
a
k

k
. (4.24)
Stationary Phase Method 49
The leading asymptotics has the form
F() = exp
_
iS(0) + [sgn S

(0)] i

4
_

2
|S

(0)|

1/2
_

(0)
6S

(0)
f(0) + f

(0) + O(
1
)
_
. (4.25)
4.2 Stationary Phase Method in Many Di-
mensions
Let be a domain in R
n
and f C

0
() be a smooth function of compact
support, S C

() be a real valued smooth function. In this section we

study the asympotics as of the multi-dimensional Fourier integrals
F() =
_

f(x) exp[iS(x)] dx . (4.26)

4.2.1 Nondegenerate Stationary Point
Localization Principle
Lemma 8 Let be a domain in R
n
and f C

0
() be a smooth function
of compact support, S C

() be a real valued smooth function without

stationary points in supp f, i.e.
x
S(x) = 0 for x supp f. Then as
F() = O(

) (4.27)
This lemma is proved by integration by parts.
Denition 53 1. The set S(R
n
) of all smooth functions on R
n
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 L
1
(R
n
) the Fourier transform is
dened by
F(f)() = (2)
n/2
_
R
n
exp(i x ) f(x) dx (4.28)
50 Stationary Phase Method
Proposition 9 1. Fourier transform is a one-to-one onto map (bijection)
F : S(R
n
) S(R
n
), i.e. if f S(R
n
), then F(f) S(R
n
).
2. The inverse Fourier transform is
F
1
(f)(x) = (2)
n/2
_
R
n
exp(i x ) f() d (4.29)
Theorem 44 Let be a nite domain in R
n
, f C

0
() be a smooth
function with compact support in and S C

() be a real valued smooth

function. Let S have a single stationary point x
0
in and let it be non-
degenerate. Then as there is asymptotic expansion
F()
n/2
exp[iS(x
0
)]

k=0
a
k

k
. (4.30)
The coecients a
k
are determined in terms of derivatives of the functions f
and S at x
0
.
The leading asymptotics is
F() =
_
2

_
n/2
exp
_
iS(x
0
) + [sgn
2
x
S(x
0
)] i

4
_

det
2
x
S(x
0
)

1/2
_
f(x
0
) +O(
1
)

. (4.31)
Recall that sgn A =
+
(A)

(A) denotes the signature of a real symmet-

ric nondegenerate matrix A, where

(A) are the number of positive and

negative eigenvalues of A.
4.2.2 Integral Operators with Singular Kernels
Let be a bounded domain in R
n
including the origin, 0 . Let f C

0
()
be a smooth function with compact support and S C

(R
n
) 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 and
2
x
S(0) = 0. Let
K

: C

0
() C

() be a linear integral operator dened by

(K

f)(x) =
_

2
_
n/2
_

exp[iS(x y)]f(y) dy . (4.32)

Stationary Phase Method 51
Then as
(K

f)(x) = exp
_
[sgn
2
x
S(0)] i

4
_

det
2
x
S(0)

1/2
_
f(x) +O(
1
)

(4.33)
uniformly for x . On the other hand, if x , then as
(K

f)(x) = O(

) . (4.34)
52 Stationary Phase Method
Chapter 5
Saddle Point Method
5.1 Saddle Point Method for Laplace Inte-
grals
Let 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 of the Laplace integrals
F() =
_

f(z) exp[S(z)] dz . (5.1)

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 nd a bound for |F()|. For a contour =
0
of nite
length l(
0
) we have, obviously,
|F()| l(
0
) max
z
0
f(z) exp[ Re S(z)] . (5.2)
Now, let be the set of all contours obtained by smooth deformations of the
contour
0
keeping the endpoints xed. Then such an estimate is valid for
any contour , hence,
|F()| inf

_
l() max
z
f(z) exp[ Re S(z)]
_
. (5.3)
53
54 Saddle Point Method
Since we are interested in the limit , we expect that the length of the
contour does not aect 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
signicantly by the large parameter ). Thus we expect an estimate of the
form
|F()| C(, f) inf

_
max
z
exp[Re S(z)]
_
, (5.4)
where C(, f) is a constant that depends on the contour and the function f
but does not depend on . So, we are looking for a point on a given contour
where the maximum of Re S(z) is attained. Then, we look for a contour

where the minimum of this maximum is attained, i.e. we assume that there
exists a contour

where
min

max
z
Re S(z) (5.5)
is attained. Such a contour will be called a minimax contour.
Let z
0

be the only point on the contour

where the maximum of

Re S(z) is attained. Then, we have an estimate
|F()| C(

, f) exp[ Re S(z
0
)] . (5.6)
By deforming the contour of integration to

we obtain
F() =
_

f(z) exp[S(z)] dz . (5.7)

The asymptotics of this integral can be computed by Laplace method.
1. Boundary Point. Let z
0
be an endpoint of

, say, the initial point.

Suppose that S

(z
0
) = 0. Then one can replace the integral F() by an
integral over a small arc with the initial point z
0
. Finally, integrating
by parts gives the leading asymptotics
F() =
1
S

(z
0
)
exp[S(z
0
)]
1
_
f(z
0
) +O(
1
)

. (5.8)
2. Interior Point. Let z
0
be an interior point of the contour

. From
the minimax property of the contour

it follows that the point z

0
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

x
Re S(z
0
) =

y
Re S(z
0
) = 0 . (5.9)
Then from Cauchy-Riemann conditions it follows that S

(z
0
) = 0.
Denition 54 1. A point z
0
C is called a saddle point of the complex
valued function S : C C if S

(z
0
) = 0.
2. A saddle point z
0
is said to be of order n if
S

(z
0
) = = S
(n)
(z
0
) = 0, S
(n+1)
(z
0
) = 0 . (5.10)
3. A rst order saddle point is called simple, i.e. for a simple saddle
point S

(z
0
) = 0.
4. The number Re S(z
0
) is called the height of the saddle point.
To compute the asymptotics at an interior saddle point, we replace the
contour

by a small arc
0

containing the point z

0
. Then we expand the
function S in the Taylor series in the neighborhood of z
0
and neglect the
terms of third order and higher, i.e. we replace S by
S(z) = S(z
0
) +
1
2
S

(z
0
)(z z
0
)
2
+O((z z
0
)
3
) . (5.11)
Finally, by changing the variables and evaluating the integral by Laplace
method we obtain the asymptotics as
F() =

2
S

(z
0
)

1/2
exp[S(z
0
)]
_
f(z
0
) +O(
1
)

. (5.12)
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

that is most suitable for asymptotical estimates. The

analytical part contains then the evaluation of the asymptotics over the con-
tour

.
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

where the minimax

min

max
z
Re S(z) is attained does not exist at all! Next, strictly speak-
ing we need to look for a contour where min

max
z
f(z) exp[Re S(z)] is
attained what makes the problem even more complicated.
Thus, if one can nd the minimax contour, then one can compute the
asymptotics of F() as . Unfortunately, there is no simple algorithm
that would always enable one to nd the minimax contour. Nevertheless,
under certain conditions one can prove that such a contour exists and, in
fact, nd one. We will discuss this point later.
5.1.2 Level Curves of Harmonic Functions
Lemma 9 Let S : C C be holomorphic at z
0
and S

(z
0
) = 0. Then in a
small neighborhood of the point z
0
the arcs of the level curves
Re S(z) = Re S(z
0
), ImS(z) = ImS(z
0
), (5.13)
are analytic curves. These curves are orthogonal at z
0
.
Let (z) = S(z) S(z
0
). Since S

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

u=0
= (
w

1
)(0) ,
(0, v)
v

v=0
= i (
w

1
)(0) (5.14)
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 z
0
be a simple saddle point of the function S. Then in a
small neighborhood of the point z
0
the level curve Re S(z) = Re S(z
0
) consists
of two analytic curves that intersect orthogonally at the point z
0
and separate
the neighborhood of z
0
in four sectors. The signs of the function Re [S(z)
S(z
0
)] in adjacent sectors are dierent.
In a neighborhood of a simple saddle point the there is a one-to-one holo-
morphic function z = (w) such that (0) = z
0
,

(0) = 0, and S((w)) =

S(z
0
) + w
2
. In the complex plane of w the level curve Re S(z) = Re S(z
0
)
takes the form Re w
2
= 0. Its solution consists of two orthogonal lines
w

= (1 i)t with < t < that intersect at the point w = 0. The level
curves z

= (w

) have listed properties.

More generally :
Lemma 11 Let z
0
be a saddle point of the function S of order n. Then
in a small neighborhood of the point z
0
the level curve Re S(z) = Re S(z
0
)
consists of (n +1) analytic curves that intersect at the point z
0
and separate
the neighborhood of z
0
in 2(n +1) sectors with angles (n +1) at the vertex.
The signs of the function Re [S(z) S(z
0
)] in adjacent sectors are dierent.
Denition 55 Let S be a complex valued function and be a simple curve
with the initial point z
0
. The curve is called curve of steepest descent
of the function Re S if ImS(z) = const and Re S(z) < Re S(z
0
) for z ,
z = z
0
. If ImS(z) = const and Re S(z) > Re S(z
0
) for z , z = z
0
, then
the curve is called curve of steepest ascent of the function Re S.
Lemma 12 1. If z
0
is not a saddle point, then there is exactly one curve
of steepest descent.
2. If z
0
is a simple saddle point, then there are 2 curves of steepest descent.
3. If z
0
is a saddle point of order n, then there are (n+1) curves of steepest
descent.
4. In a neighborhood of a saddle point z
0
in each sector in which Re [S(z)
S(z
0
)] > 0 there is exatly one curve of steepest descent.
This is proved by a change of variables in a neighborhood of z
0
.
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

(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 = x
2
y
2
is
hyperbolic paraboloid (saddle).
Denition 56 Two contours
1
and
2
are called equivalent if
_

1
f(z) exp[S(z)] dz =
_

2
f(z) exp[S(z)] dz . (5.15)
Lemma 13 Let S and f be holormorphic functions on a nite contour .
Let the points where max
z
Re S(z) is attained are neither saddle points nor
the endpoints of the contour . Then there is a contour

equivalent to the
contour and such that
max
z

Re S(z) < max

z
Re S(z) . (5.16)
Theorem 45 Let F() be a Laplace integral (5.1) If there exists a contour

such that: i) it is equivalent to the contour and ii) the integral F()
attains the minimax min

max
z
Re S(z) on it. Then among the points
where max
z

Re S(z) is attained there are either endpoints of the contour

or saddle points z
j
such that in aneighborhood of z
j
the contour

goes
through two dierent sectors where Re S(z) < Re S(z
j
).
5.1.3 Analytic Part of Saddle Point Method
In this section we always assume that is a simple smooth (or piece-wise
smooth) curve in the complex plane, which may be nite or innite. The
functions f and S are assumed to be holomorphic on . Also, we assume
that the integral F() converges absolutely.
First of all, we have
Lemma 14 If max
z
Re S(z) C, then
F() = O(e
C
), ( 1) . (5.17)
Saddle Point Method 59
Theorem 46 Let z
0
be the initial endpoint of the curve . Let f and S are
analytic at z
0
, Re S(z
0
) > Re S(z) z , and S

(z
0
) = 0. Then, as
there is asympotic expansion
F() exp[S(z
0
)]

k=0
a
k

k1
(5.18)
where
a
k
=
_

1
S

(z)

z
_
k
_
f(z)
S

(z)
_

z=z
0
(5.19)
The proof is by integration by parts.
Theorem 47 Let z
0
be an interior point of the curve . Let f and S are
analytic at z
0
, Re S(z
0
) > Re S(z) z . Let z
0
be a simple saddle point of
S such that in a neighborhood of z
0
the contour goes through two dierent
sectors where Re S(z) < Re S(z
0
). Then, as there is asympotic
expansion
F() exp[S(z
0
)]

k=0
a
k

k1/2
. (5.20)
The branch of the square root is choosen so that arg
_
S

(z
0
is equal to the
angle between the positive direction of the tangent to the curve at the point
z
0
and the positive direction of the real axis.
In a neighborhood of z
0
there is a mapping z = (w) such that
S((w)) = S(z
0
)
w
2
2
. (5.21)
After this change of variables and deforming the contour to the steepest
descent contour the integral becomes
F() = exp[S(z
0
)]
_

e
w
2
/2
f((w))

(w) dw + O(

) . (5.22)
Since both f and are holomorphic there is a Taylor expansion
f((w)

(w) =

k=0
c
k
w
k
. (5.23)
60 Saddle Point Method
Then the coecients a
k
are easily computed in terms of c
k
a
k
= 2
k+1/2

_
k +
1
2
_
c
2k
. (5.24)
Theorem 48 Let be a nite contour and f and S be holomorphic in a
neighborhood of . Let max
z
ReS(z) be attained at the points z
j
and these
points are either the endpoints of the curve or saddle points such that in
a neighborhood of a saddle points the contour goes through two dierent
sectors where Re S(z) < Re S(z
j
). Then as the integral F() is
asymptotically equal to the sum of contributions of the points z
j
.
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 and ImS(z) =
const on . If has nite number of saddle points, then as |lambda| ,
| arg | /2 < /2, the asymptotic expansion of the integral F() is
equal to the sum of contributions of the saddles points and the endpoints of
the contour.
5.1.4 Examples
1.
_

e
ix
(1 +x
2
)

dx
_
(1 c)e
c

1/2
(2c)

( )
(5.25)
where c =

2 1. There are two saddle points z

1,2
= i(1

2). The
asymptotics is determined by the contribution from the point z
1
.
2.
_
i+
i
e
z
2
(1 +z)
n
dz
_

2
i
n
e
(n1)/2
_
n
2
_
n/2
, (n ) .
(5.26)
Here n > 0 is a positive integer.
Saddle Point Method 61
3. Let a > 0. As
_
ia+
ia
exp(2z
2
4/z) dz
1/6
(6)
1/2
exp(3+ i3

3 ) . (5.27)
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 z
1
= e
i 2/3
.
4.
_
i
i
exp(z
2
+z
2
ln z) dz i

exp
_

1
2
e
21
_
, ( ) .
(5.28)
There are two saddle points z
1
= 0 and z
2
= e
1/2
. The asymptotics
is equal to the contribution from the saddle point z
2
.
62 Saddle Point Method
Notation
Logic
A = B A implies B
A = B A is implied by B
i 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 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

A
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
R
n
vector space of n-tuples of real numbers
C
n
vector space of n-tuples of complex numbers
l
2
space of square summable sequences
l
p
space of sequences summable with p-th power
Normed Linear Spaces
||x|| norm of x
x
n
x (strong) convergence
x
n
w
x weak convergence
Function Spaces
supp f support of f
H G tensor product of H and G
Notation 65
C
0
(R
n
) space of continuous functions with bounded support in
R
n
C() space of continuous functions on
C
k
() space of k-times dierentiable functions on
C

() space of smooth (innitely direntiable) functions on

D(R
n
) space of test functions (Schwartz class)
L
1
() space of integrable functions on
L
2
() space of square integrable functions on
L
p
() space of functions integrable with p-th power on
H
m
() Sobolev spaces
C
0
(V, R
n
) space of continuous vector valued functions with bounded
support in R
n
C
k
(V, ) space of k-times dierentiable vector valued functions on

(V, ) space of smooth vector valued functions on

D(V, R
n
) space of vector valued test functions (Schwartz class)
L
1
(V, ) space of integrable vector valued functions functions on

L
2
(V, ) space of square integrable vector valued functions func-
tions on
L
p
(V, ) space of vector valued functions functions integrable with
p-th power on
H
m
(V, ) Sobolev spaces of vector valued functions
Linear Operators
D

dierential 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
[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. Mikusi nski, 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. Erdelyi, Asymptotic Expansions, (New York: Dover Publications,
1956)
[9] R. B. Dingle, Asymptotic Expansions: Their Derivation and Interpre-
tation, (London: Academic Press, 1973)
67