Method of the Steepest Descent

Z
I () = f (z)e (z) dz, (z) C
C
(2)

I () is a complex integral, and we want to find its asymptotic

behaviour as .

We use the analyticity of the integrand to deform the path

C into a new path C 0 has constant imaginary part. Then we
apply Laplaces method.
I The directional derivative df /ds of f (x, y ) R along the
direction n f , where f = (f /x, f /y ).
is given by n
I When n = f / |f |, df /ds is maximum: f points in the
direction of the most rapid change of f .
I f is perpendicular to contours of constant f (x, y ) (level
curves).
I Let z = x + iy . (z) = u(x, y ) + iv (x, y ) is analytic if and
only if
u v u v
= and = . (3)
x y y x

Cauchy-Riemann equations. They imply uv = 0! .

I If(z) is analytic

d(z) u v
= +i .
dz x x
Consider the function (z) = iz 2 ,

(z) = u(x, y ) + iv (x, y ) = 2xy + i(x 2 y 2 ). (4)

Because of the Cauchy-Riemann equations u is tangent to

the curves x 2 y 2 = const. and v is tangent to the curves
2xy = const.
 2 2

1 1

0 0

-1 -1

-2 -2
-2 -1 0 1 2 -2 -1 0 1 2

The contour plot of The contour plot of

u(x, y ) = 2xy v (x, y ) = x 2 y 2

The contour plots of the real and imaginary part of (z) = iz 2 .

Regions with higher values of u(x, y ) and v (x, y ) are lighter.
 2

-1

-2
-2 -1 0 1 2

The contour curves 2xy = const. and x2 y2 = const.

Let us come back to the integral

Z
I () = f (z)e (z) dz, (z) C
C
(6)

Deform C into a path where the imaginary part of (z),

v (x, y ) = v0 , where v0 is a constant.I () then becomes
Z
I () = e iv0 f (z)e u(x,y ) dz (7)
C0

The main contribution to I () comes where u is largest on a

steepest descent path, i.e. u = 0
At these points
u = v = 0. (8)
Since (z) is analytic we have

u v
0 (z) = +i = 0. (9)
x x

u(x, y ) and v (x, y ) are harmonic functions, therefore such

points are saddle points! They are not maxima or minima

2u 2u
u = + =0 (10)
x 2 y 2
2v 2v
v = + =0 (11)
x 2 y 2
Steepest ascent-descent paths of (z) = iz 2 passing through the
saddle point z = 0. The red curves are the steepest ascent, while
the blue ones are the steepest descent
I Let 0 (z0 ) = 0 and = (z) (z0 ) with |z z0 | = small.
I Denote

= u+iv , z = zz0 = e i , 00 (z0 ) = ae i , , a > 0.

I u = Re and v = Im .
I Then,

00 (z0 ) 2 a i(+2)
(z z0 )2 + O(z 3 )=


(z) = e 1 + O () .
2 2!

Find the directions (i.e. tangents) through z0 of:

1. Steepest descent curves constant v or v = 0;
2. Steepest ascent curves constant v or v = 0;
3. Curves of constant u or u = 0
Recall: z = x + iy , z0 = x0 + y0 and u(x0 , y0 ) = v (x0 , y0 ) = 0
1. Steepest descent curves:

v = 0 = = u R and u < 0.

Therefore, for small enough

2 2u
2
= 2 = e i(+2) = 1 = +2 = odd multiple of
a a
Thus,
3
1 = + and 5 = + = 1 +
2 2 2 2
2. Steepest ascent curves:

v = 0 = = u R and u > 0.

Therefore, for small enough

2 2u
= 2 = e i(+2) = 1 = +2 = even multiple of
2 a a
Thus,

3 = and 7 = + = 3 +
2 2
3. Curves of constant u(x, y ):

u = 0 = = iv iR dont care about sign of v

Therefore, for small enough

2 2iv
= 2 = e i(+2) = i +2 = odd multiple of
2 a a 2
Thus,
3
2 = + , 4 = +
2 4 2 4
5 7
6 = + , 8 = +
2 4 2 4
I Recall that our goal is the asymptotics of
Z
I () = f (z)e (z) dz, z .
C

I We have
   
 (z)   (u(x,y )+iv (x,y )) 
e  = e  = e u(x,y )

1. Def 1: Regions where u(x, y ) > u(x0 , y0 ) are called hills.

2. Def 2: Regions where u(x, y ) < u(x0 , y0 ) are called valleys.

I Recall that 0 (z0 ) = 0, 00 (z0 ) = ae i and z z0 = e i

I The line of constant u through z0 are
3
2 = + , 4 = +
2 4 2 4
5 7
6 = + , 8 = +
2 4 2 4
I Since e i = cos + i sin , for any line through z0
2 2u
2
= 2 = cos ( + 2)
a a
I In the region with 2 < < 4
3
< + 2 < = u < 0 (valley)
2 2
I In the region with 4 < < 6
3 5
< + 2 < = u > 0 (hill)
2 2
I Recall
2 2u
2
= 2 = cos ( + 2)
a a
I In the region with 6 < < 8
5 7
< + 2 < = u < 0 (valley)
2 2
I In the region with 8 < < 2
7 9
< + 2 < = u > 0 (hill)
2 2
I Lets go back to (z) = iz 2
I Then

(z) = u(x, y ) + iv (x, y ) = 2xy + i(x 2 y 2 ).

I

0 (0) 00 (0) = 2i = = .
2
Steepest ascent-descent paths of (z) = iz 2 passing through the
saddle point z = 0. The red curves are the steepest ascent, while
the blue ones are the steepest descent. C2 , C4 , C6 and C8 are
curves u(x, y ) = const.
I Region between C2 and C4 : valley
I Region between C4 and C6 : hill
I Region between C6 and C8 : valley
I Region between C8 and C2 : hill

Two consecutive curves u(x, y ) = const. meet at angle

/2. The region in between is either a valley or a hill. The
bisector is the tangent to a curve of steepest descent or
ascent, respectively.
Consider
(z) = i cosh z. (12)
Use the formula
cosh(x + iy ) = cosh x cos y + i sinh x sin y . (13)
(z) becomes

(z) = sinh x sin y + i cosh x cos y

We have a saddle point at the origin:

(z)0 (0) = i sinh z = 0 and 00 (0) = i cosh z = i. (14)
At the origin the steepest ascent-descent curves are given by

cosh x cos y = 1
Steepest ascent-descent paths of (z) = i cosh z passing through
the saddle point z = 0. The red curves are the steepest ascent,
while the blue ones are the steepest descent. C2 , C4 , C6 and C8
are curves u(x, y ) = const.
I Region between C2 and C4 : valley
I Region between C4 and C6 : hill
I Region between C6 and C8 : valley
I Region between C8 and C2 : hill
Theorem:
Let (z) be analytic in z0 and

d q  d n 
 
= 0, q = 1, 2, . . . , n 1, = ae i , a > 0.
dz q z=z0 dz n z=z0
(15)
If z z0 = e i , then the directions of steepest descent, steepest
ascent, and constant u are given by the table below.

Direction of
Steepest descent n + (2p + 1) n p = 0, 1, . . . , n 1
Steepest ascent n + 2p
n
p = 0, 1, . . . , n 1
Constant u n + p + 21 n p = 0, 1, . . . , 2n 1
I Lets go back to
Z
I () = f (z)e (z) dz, (16)
C

I Suppose that:
1. 0 (z0 ) = 0 and 00 (z0 ) 6= 0
2. The contour C can be deformed into a steepest descent C 0
through z0 .
I Recall that (z) = u(x, u) + iv (x, y ).
I On C 0 , v = 0.
I The integral
Z
iv (x0 ,y0 )
I () = e f (z)e u(x,y ) dz, (17)
C0

is a Laplace type integral.

I We expect that the leading order contribution to I () will
come from a small neighbourhood of z0 along C 0 .
I Therefore,
 
Z z0 + (zz )2
(z0 )+ 2!0 00 (z0 )
I () f (z0 )e dz, (18)
z0
I C0 steepest descent implies
00 (z0 )
(z z0 )2 = u < 0 z C 0
2!
I We then set r
00 (z0 )
= (z z0 )2 ,
2!
I which leads to
Z
f (z0 )e (z0 ) 2
I () p e d
00
(z0 )/2
s (19)
2
f (z0 )e (z0 ) ,
00 (z0 )
Method of the steepest descent:
1. Identify the possible critical points of the integrand. These are
the end points of integration, singular points of f (t) and (t),
and saddle points of (t).
2. Determine the paths of steepest descent from each of the
critical points.
3. Justify, via Cauchys integral theorem, the deformation of the
original contour of integration C onto one or more of the
steepest descent found in (2).
4. Determine the asymptotic expansion of the integrals that arise
as a result of the deformation in (3). (Note that each integral
is of Laplaces type.)
Example:

We want to find the full asymptotic expansion of

Z 1
2
I () = e iz dz. (21)
0

The leading order behaviour of this integral can be found by using

the stationary phase method, obtaining

r
1 i
I () e4, (22)
2
 We now want to deform the interval C0 = [0, 1] into a steep-
est descent path

Recall that

(z) = u(x, y ) + iv (x, y ) = xy + i(x 2 y 2 ). (23)

The unique saddle point of (z) is the origin. There, the steepest
ascent-descent curves are given by

v (x, y ) = x 2 y 2 = 0 (24)
2 2
x =y (25)
x =y x = y . (26)

The steepest descent path is x = y , i.e. z = (1+i)x.

Note that iz 2 = 2x 2 .
At z = 1 we have

v (x, y ) = x 2 y 2 = x 2 = 1 (27)

On the branch of hyperbola

p
x = y 2 + 1, y 0 (28)

v (x, y ) = 1. In complex notation

p
z= y 2 + 1 + iy , y 0. (30)

(30) is a steepest descent curve, in fact

p
(z) = iz 2 = 2y y 2 + 1 + i (31)
 1.5

0.5

0.5 1 1.5 2
-0.5

-1

-1.5

Steepest ascent-descent paths of (z) = iz 2 passing through the

point z = 1. The red curve is the steepest ascent, while the blue
one is the steepest descent.
 1.75

1.5

1.25

0.75

0.5

0.25

0.5 1 1.5 2
R1 2
Steepest descent curve for 0 e it dt

I Green curve, C1 , z = (1 + i)x, 0 x < L;

I blue curve, C2 , z = x + iL, L x < L2 + 1;
p
I red curve, C3 , z = y 2 + 1 + iy , 0 < y L.
Because of Cauchys theorem
Z 1 Z Z Z
iz 2 iz 2 iz 2 2
e dz = e dz + e dz e iz dz. (32)
0 C1 C2 C3

As L , the length of C2 tends to zero (C1 is an asymptote of

C3 ): p
lim L2 + 1 L = 0. (33)
L

Therefore, as L

Z 1 Z Z
2 2 2
e iz dz = e iz dz e iz dz. (35)
0 C1 C3
Using the substitution z = (1 + i)x, the integral over C1 gives
Z Z r
iz 2 2x 2 1 i
e dz = (1 + i) e dx = e4. (36)
C1 0 2

This is the leading order term

R which we obtained using the
iz 2
stationary phase method. C3 e dz gives higher order
contributions.
2
The trick now is to transform C3 e iz dz into an integral of the
R

form
Z
f (s)e s ds (38)
0

then apply Watsons lemma.

We now make the substitution

iz 2 = i s

p
With this change of variable, s = 2y y 2 + 1 is real and and
satisfy 0 < s < along C3 . Since z = (1 + is)1/2 , so
dz/ds = i(1 + is)1/2 /2

e s
Z Z
iz 2 1
e dz = ie i ds. (40)
C3 2 0 1 + is
Now, using the expansion
1


1/2 1 X n+
(1 + is) = 2
(is)n , |s| < 1 (41)
n!
n=0

and applying Watsons lemma with = 0 and = 1, we have

1


n n + 2
Z
iz 2 i i
X
e dz e (i) , .
C3 2 n+1
n=0
(43)

Here we have used the fact that (n + 1) = n!.

Finally, we have
Z 1
2
I () = e iz dz
0
(44)
n + 12
r

i i i
X
n
e4 e (i) , .
2 n+1
n=0
Example:

A standard integral representation for J0 () is given by

Z /2
1
J0 () = cos ( cos ) d, (45)
/2

which can be transformed into

Z i/2
1
J0 () = Re e i cosh z dz (46)
i i/2

and finally into

Z i/2+
1
J0 () = Re e i cosh z dz (48)
i i/2
The previous transformation can be done because the two integrals
Z i/2
1
e i cosh z dz (49a)
i i/2
(49b)

and
Z i/2+
1
e i cosh z dz (49c)
i i/2

converge and cancel each other (check this!).

 1.5

0.5

-10 -5 5 10

-0.5

-1

-1.5

The path of integration of the integral

R i/2+ i cosh z
i/2 e dz. We denote this path by C0 .
Now we have
(z) = i cosh z. (50)
Use the formula
cosh(x + iy ) = cosh x cos y + i sinh x sin y . (51)
(z) becomes

(z) = sinh x sin y + i cosh x cos y

We have a saddle point at the origin:

(z)0 (0) = i sinh z = 0 and 00 (0) = i cosh z = i. (52)
At the origin the steepest ascent-descent curves are given by

cosh x cos y = 1
 4 4

2 2

0 0

-2 -2

-4 -4

-4 -2 0 2 4 -4 -2 0 2 4

The contour plot of The contour plot of

u(x, y ) = sinh x sin y v (x, y ) = cosh x cos y

The contour plots of the real and imaginary part of (z) = i cosh z.
Regions with higher values of u(x, y ) and v (x, y ) are lighter.
Steepest ascent-descent paths of (z) = i cosh z passing through
the saddle point z = 0. The red curves are the steepest ascent,
while the blue ones are the steepest descent. We denote the latter
path by C1 .
 17.5
-4 -2 2 4
-2.5 15

-5 12.5

-7.5 10

-10 7.5

-12.5 5

-15 2.5

-17.5 -4 -2 2 4

u(x, y ) along the steepest u(x, y ) along the steepest

descent path ascent path
cosh x cos y = 1, xy 0 cosh x cos y = 1, xy 0

The behaviour of the real part of (z) = i cosh z along the steepest
descent-ascent paths passing through the saddle point z = 0.
Next sep:
We deform our original path of integration C0 into the steepest
descent path C1 and approximate it at the origin with the straight
line z = (1 + i)s

Using the expansion

x2
cosh x = 1 + + O(x 4 ) (53)
2!
we have that cosh[(1 + i)s] 1 + is 2 , as s 0. Therefore
Z
1
J0 () = Re e i cosh z dz
i C1
(54)
(1 + i) is 2
Z
Re e ds, .
i

We now make the substitution s = . We get
Z
(1 i) i 2
J0 () Re e e d

(1 i) i 2
Z
Re e e d (55)

r
2 (1 i) i
Re e , .
2

Now, (1 i)/ 2 = e i/4 . Therefore

r
2  
J0 () cos , .
4
(57)
Let us evaluate the full asymptotic expansion of J0 () as .
Along C1 , (z) = i cosh z = u(z) + i. Therefore.

u(z) = i cosh z i.

Moreover, along C1 , u as z ( i/2). We then use

u as variable of integration.

du = i sinh zdz (58)

du
dz = , (59)
u 2 2iu

where we have used the identity cosh2 z sinh2 z = 1.

Replace u with uand insert it into
Z
1
J0 () = Re e i cosh z dz. (60)
i C1

We obtain

e ii/4 Z e u
J0 () = Re 2 q
du (62)
0 u 1 iu
2
 R
The previous integral has the form 0 f (u)e u du. Moreover

iu 1/2

 n
1 X n + 21
 
1 iu
1 = , |u| < 2 (63)
u 2 n! u 2
n=0

Therefore, we can apply Watsons lemma with = 1/2 and

= 1:


2  n
e ii/4 X n + 12 i
J0 () Re 2 3/2
,
n! 2
n=0
(65)
We can rearrange the previous expression and write

r
2
J0 () = [() cos ( /4) + () sin ( /4)] ,

(67)

where
 1

2
X n 2n + 2
() (1) ,
n=0
(2n)! (2)2n

and

2
2n + 32

X n+1
() (1) ,
n=0
(2n + 1)! (2)2n+1
Example:

An integral representation for J () is given by

Z
1
J () = e (sinh zz) dz (70)
2i C

where C is a Sommerfeld contour which begins at + i and

ends at + + i
 3

1 2 3 4
-1

-2

-3

e (sinh zz) dz
R
Limits of Sommerfeld contours for the integral C
The function (z) = sinh z z has a saddle point of order three at
the origin:

0 (z) = cosh z 1, 00 (z) = sinh z and 000 (z) = cosh z, (71)

Therefore (0) = 1, 0 (0) = 0, 00 (0) = 0, 000 (0) = 1.

Using the formula

sinh(x + iy ) = sinh x cos y + i cosh x sin y , (72)

we obtain

(z) = u(x, y )+iv (x, y ) = sinh x cos y x +i cosh x sin y y ,

The steepest ascent-descent paths are given by the equation

cosh x sin y y = c.

At z = 0, c = 0. The above equation has then the solutions

y =0

(the x axis) and

x = cosh1 (y / sin y )
 4 4

2 2

0 0

-2 -2

-4 -4

-4 -2 0 2 4 -4 -2 0 2 4

The contour plot of The contour plot of

u(x, y ) = sinh x sin y x v (x, y ) = cosh x cos y y

The contour plots of the real and imaginary part of

(z) = sinh z z. Regions with higher values of u(x, y ) and
v (x, y ) are lighter.
 3

-4 -2 2 4

-1

-2

-3

Steepest ascent-descent paths of (z) = sinh z z passing

through the saddle point z = 0. The red curves are the steepest
ascent, while the blue ones are the steepest descent.
 0.075

0.05

0.025

-2 -1 1 2
-0.025

-0.05

-0.075

The real part of (z) = sinh z z along the steepest

descent-ascent path y = 0.
 3.5
-2 -1 1 2
-0.5 3

-1 2.5

-1.5 2

-2 1.5

-2.5 1

-3 0.5

-3.5
-2 -1 1 2

u(x, y ) along the steepest u(x, y ) along the steepest

descent path ascent path
cosh x sin y y = 0, x 0 cosh x sin y y = 0, x 0

The behaviour of the real part of (z) = sinh z z along the

steepest descent-ascent paths passing through the saddle point
z = 0.
The slopes at the origin of our steepest decent path, C 0 are /3
and /3. At the origin we approximate such a path by the two
straight lines

i i
z = re 3 and z = re 3 .

We then use the power expansion

z3
sinh z = z + + O(z 5 ), z 0. (73)
3!
Therefore, along C 0 we have

r 3 i r3
sinh z z e , r 0 (74a)
6 6
r 3 r3
sinh z z e i , r 0 (74b)
6 6
 Z
J () = e (sinh zz) dz
C0
Z 0 Z
1 3
i 3 r6 1 r3
e e dr + e i 3 e 6 dr (75)
2i 2i 0
Z
e i 3 e i 3 r 3
e 6 dr , .
2i 0
3
By making the substitution = r6 , we have

1 6 1/3 1/31
 
dr = d. (76)
3
Therefore
  Z 3
sin (/3) 6 1/3 6 1/31
J () e d
3 0 (77)
Z
3
2/3 1/3 1/31 e d, .
6 0
Finally, since Z  
1/31 1
e d = (78)
0 3

3 13


J () 2/3 1/3 , . (80)
6
Example:

One solution of the Airy equation

y 00 (x) xy (x) = 0

has the following integral representation:

Z
1 3
Ai() = e 3 d (82)
2i C

where C is a Sommerfeld contour which originates at e 2/3

and terminates e 2/3
 20

10

-10 -8 -6 -4 -2

-10

-20

z3
e z 3 dz
R
Limits of Sommerfeld contours for the integral C
Now, in this case ( ) = : No saddle points!

We therefore rescale and set

= 1/2 z. (83)

Then, our integral becomes

1/2 3
Z  
3/2 z z3
Ai() = e dz (85)
2i C

The limits of the Sommerfeld contours remain the same.

Now, we have
z3
(z) = z , 0 (z) = 1 z 2 and 00 (z) = 2z. (86)
3

Therefore (z) has saddle points at z = 1.

Now
z3
 3   3 
x 2 y 2
z = u(x, y )+iv (x, y ) = + x + xy +i + y yx .
3 3 3
Therefore at z = 1

v (x, y ) = y 3 /3 + y yx 2 = 0
 10 10

5 5

0 0

-5 -5

-10 -10
-10 -5 0 5 10 -10 -5 0 5 10

The contour plot of The contour plot of

u(x, y ) = x 3 /3 + x + xy 2 v (x, y ) = y 3 /3 + y yx 2

The contour plots of the real and imaginary part of

(z) = z z 3 /3. Regions with higher values of u(x, y ) and v (x, y )
are lighter.
The curves defined by the equation y 3 /3 + y yx 2 = 0 are given by

y = 0, (87a)
x 2 y 2 /3 = 1, x 1 (87b)
2 2
x y /3 = 1, x 1 (87c)

I On y = 0, u(x, y ) = x 3 /3 + x,
I on the other two curves u(x, y ) = 8x 3 /3 2x.
 10

-6 -4 -2 2 4 6

-5

-10

Steepest ascent-descent paths of (z) = z z 3 /3 passing through

the saddle points z = 1 and z = 1. The red curves are the
steepest ascent, while the blue ones are the steepest descent.
 3
0.6
2
0.4
1
0.2
-2 -1 1 2
-1 -2 -1 1 2
-0.2
-2

-3 -0.4

-0.6

The function
The function
u(x, y ) = x 3 /3 + x + xy 2
u(x, y ) = x 3 /3 + x + xy 2
along the path
along the path y = 0
y 2 /3 + 1 x 2 = 0

The behaviour of the real part of z z 3 /3 along the steepest

descent-ascent paths passing through the saddle points z = 1
and z = 1
The steepest descent path, which is also a Sommerfeld contour for
the integral,
3
Z  
3/2 z z3
e dz (88)
C
is

x 2 y 2 /3 = 1, x 1

The tangent to this hyperbola at x = 1 is the straight line

x = 1. We then make the substitution

z = 1 + is. (89)

Then (z) becomes

z3 2
z = s 2 + O(s 3 ). (90)
3 3
Therefore, setting, as usual, 2 = 3/2 s 2
3/4
1/4 2 3/2
Z
2
Ai() e 3 e d
2 3/4
(91)

1/4
Z
23 3/2 2
e e d, .
2

Finally,

1/4 2 3/2
Ai() e 3 , (93)
2