0% found this document useful (0 votes)
77 views8 pages

Oscillatory Integrals

The document discusses oscillatory integrals, which arise in Fourier analysis. It explores methods for estimating the decay of oscillatory integrals as their parameter approaches infinity. These include the principle of non-stationary phase, Van der Corput's lemma, and the method of stationary phase. Van der Corput's lemma provides bounds for oscillatory integrals over intervals where the phase function has no critical points or where all critical points are non-degenerate.

Uploaded by

Liviu Ignat
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
77 views8 pages

Oscillatory Integrals

The document discusses oscillatory integrals, which arise in Fourier analysis. It explores methods for estimating the decay of oscillatory integrals as their parameter approaches infinity. These include the principle of non-stationary phase, Van der Corput's lemma, and the method of stationary phase. Van der Corput's lemma provides bounds for oscillatory integrals over intervals where the phase function has no critical points or where all critical points are non-degenerate.

Uploaded by

Liviu Ignat
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 8

OSCILLATORY INTEGRALS

OPEN PROBLEMS IN NUMBER THEORY


SPRING 2018, TEL AVIV UNIVERSITY

ZEÉV RUDNICK

Contents
1. Oscillatory integrals 1
1.1. The principle of NON-stationary phase 1
1.2. Van der Corput’s Lemma 2
1.3. An asymptotic expansion: The method of stationary phase 6
1.4. An application: The Fourier transform of the unit disk 6

1. Oscillatory integrals
R∞
The Fourier transform fb(k) = −∞ f (x)e−2πikx dx is an example of an oscillatory
integral. The general form that we take is
Z ∞
I(λ) := A(x)eiλφ(x) dx
−∞

where the phase function φ(x) is assumed to be real, and the amplitude A(x) is
assumed to be compactly supported, or at least in the Schwartz space S.
We will need to analyze a number of such oscillatory integrals
R and in particular
understand their decay at infinity. The trivial bound is O( |A(x)|dx), and we want
to have some cancellation as λ → ∞.

1.1. The principle of NON-stationary phase. Suppose first that A is smooth,


and either
1) |φ0 | ≥ 1
or
2) if A is compactly supported we may simply assume that φ0 6= 0 has no critical
points on the support of A.
Then as λ → +∞,
1
I(λ) N N , ∀N ≥ 1
λ
(the implied constants depend on A and φ).

Date: April 12, 2018.


1
2 ZEÉV RUDNICK

Proof. The proof is by repeated integration by parts. Lets take the case that A is
compactly supported. Define differential operators acting on functions supported
in supp(A):
1 d d 1
L= 0 , LT = − ◦
iφ (x) dx dx iφ0 (x)
(this makes sense since φ0 6= 0 on the support of A). Integration by parts shows
that for any f, g ∈ S,
Z ∞ Z ∞
(Lf )(x)g(x)dx = f (x)(LT g)(x)dx.
−∞ −∞
iλφ(x)
Moreover a computation shows that L(e ) = λeiλφ(x) , so that
LN (eiλφ(x) ) = λN eiλφ(x) , ∀N ≥ 1.
Now we have
Z ∞ Z ∞ Z ∞
iλφ(x) 1 N iλφ(x) 1
I(λ) = e A(x)dx = N
L (e )A(x)dx = N eiλφ(x) (LT )N (A)(x)dx
−∞ −∞ λ λ −∞
and taking absolute values gives
Z ∞
1 CN
|I(λ)| ≤ |(LT )N (A)(x)|dx =
λN −∞ λN
R∞
with CN = −∞ |(L ) (A)(x)|dx < ∞ since (LT )N (A) is also smooth and com-
T N

pactly supported. 
1.2. Van der Corput’s Lemma. In the above, it was crucial to have A smooth
in order to integrate by parts. When this does not hold, the result may fail. We
will need the case A = 1[a,b] , when
Z b
I(λ) = eiλφ(x) dx
a
We first assume that φ has no stationary (critical) points on [a, b]:
Proposition 1.1 (Van der Corput’s Lemma 1). Assume that φ is smooth, that
φ0 6= 0 on [a, b], and that φ0 is monotonic. Then
Z b 4 1
eiλφ(x) dx ≤ · .


min |φ 0 (x)| |λ|
a x∈[a,b]

Proof. Since φ0 is continuous and nonzero on [a, b], we have


c := min |φ0 (x)| > 0.
x∈[a,b]

Integrating by parts, we have


Z b Z b
1
I(λ) = e iλφ(x)
dx = eiλφ(x) iλφ0 (x) ·
0 (x)
dx
a a iλφ
b Z b
1 iλφ(x) d
 1 
= eiλφ(x) − e dx

iλφ0 (x)) a dx iλφ0 (x)

a
and therefore (WLOG take λ > 0)
1 eiλφ(b) eiλφ(a) 1 b d n 1 o
Z
|I(λ)| ≤ 0 − 0 +
λ φ (b) φ (a) λ a dx φ0 (x)
dx.
OSCILLATORY INTEGRALS 3

Since |φ0 | ≥ c on [a, b], we have


eiλφ(b) eiλφ(a) 1 1 2
− 0 ≤ 0 + ≤ .

0
φ (b) φ (a) |φ (b)| |φ0 (a)| c
0 0
For the nintegral,
o since φ is monotonic (and nonzero), we have 1/φ monotonic,
d 1
so that dx φ0 (x) has a fixed sign. Therefore
Z b d n 1 o Z b d n 1 o 1 1 2
= dx = − ≤

dx φ0 (x) 0
a dx φ (x) φ0 (b) φ0 (a) c
dx
a

as before. Thus we find


4/c
|I(λ)| ≤ .
λ


We now turn to study the case when the phase function φ has critical (stationary)
points. Assume that all critical points are non-degenerate (φ00 (x) 6= 0 if φ0 (x) = 0).
By subdividing the interval1, we can assume that either case 1 holds (no critical
points) or φ00 6= 0 on the entire interval [a, b]. In that case we have
Proposition 1.2 (Van der Corput’s Lemma 2). Let φ be real valued and smooth
on [a, b], with φ00 6= 0 on [a, b]. Then
Z b 8 1
eiλφ(x) dx ≤ p ·p .

minx∈[a,b] |φ00 (x)|

a |λ|
Proof. By replacing φ by φ/ minx∈[a,b] |φ00 (x)| and λ by minx∈[a,b] |φ00 (x)| · λ, we
may assume that |φ00 (x)| ≥ 1, and WLOG take φ00 ≥ 1.
Let c ∈ [a, b] be a point where the first derivative |φ0 | is minimal: |φ0 (c)| ≤ |φ0 (x)|
for all x ∈ [a, b]. Since the second derivative is non-vanishing, it cannot be the case
that c is an interior local minimum/maximum of φ0 , and hence either φ0 (c) = 0 or
c is one of the endpoints a, b.
Assume first that φ0 (c) = 0, as in Figure 1. Then outside the interval (c−δ, c+δ),

0.5 1.0 1.5 2.0

-1

Figure 1. A local minimum of |φ0 | where φ0 (c) = 0.

1We assume that φ has only finitely many critical points in [a, b], which would be the case if
it was real analytic.
4 ZEÉV RUDNICK

we have |φ0 (x)| ≥ δ, because we assume φ00 (x) ≥ 1 for all x ∈ [a, b]. Indeed, if
c + δ ≤ x ≤ b then
Z x Z x
φ0 (x) = φ0 (x) − φ0 (c) = φ00 (t)dt ≥ 1dt = x − c ≥ δ
c c
with a similar argument if a ≤ x ≤ c − δ. Now divide the range of integration [a, b]
into three subintervals
Z b Z c−δ Z c+δ Z b
eiλφ(x) dx = + +
a a c−δ c+δ

On the interval [a, c−δ], use |φ0 | ≥ δ and the fact that φ0 is monotonic increasing
(since φ00 ≥ 1 > 0) to invoke Proposition 1.1 giving
Z c−δ 4 1 4 1
eiλφ(x) dx ≤ · ≤ ·

0
| min(|φ (x)| : x ∈ [a, c − δ]) λ δ λ
a
Rb
with the same bound holding for c+δ .
On the middle interval (c − δ, c + δ), just estimate trivially |eiλφ(x) | ≤ 1 giving
Z c+δ
eiλφ(x) dx ≤ 2δ


c−δ
Thus we find
Z b 8
eiλφ(x) dx ≤ + 2δ.

δλ

a

Taking δ = 2/ λ gives
Z b 8
eiλφ(x) dx ≤ √ .


a λ
It remains to deal with the case that c is one of the endpoints, say c = a and
φ0 (a) 6= 0, say φ0 (x) ≥ φ0 (a) > 0 for all x ∈ [a, b]. Then as before φ0 (x) ≥ δ for
x ∈ [a + δ, b] since φ00 ≥ 1, and then the previous argument shows that
Z b Z a+δ Z b 4 4
iλφ(x)
e dx ≤ 1dx + eiλφ(x) dx ≤ δ + ≤√

δλ

a a a+δ λ

on taking δ = 2/ λ. 
Remark: A similar result holds for the case of degenerate critical points, if we
assume that for some k ≥ 2, we have φ(k) 6= 0 on [a, b]:
Proposition 1.3. There is an absolute constant ck > 0 so that for all smooth, real
valued φ with φ(k) =
6 0 on [a, b],
Z b ck 1
eiλφ(x) dx ≤ · .

|φ (k) (x)|)1/k |λ|1/k

a (min x∈[a,b]
m
z 2m
P∞
Exercise 1. The Bessel function J0 (z) = m=0 (−1)

(m!)2 2 admits an integral
representation Z π
1
J0 (z) = e−iz sin t dt
2π −π

Show that as z → +∞, J0 (z)  1/ z.
We now include an amplitude
OSCILLATORY INTEGRALS 5

Corollary 1.4. Assume that A(x) ∈ C 1 [a, b] is differentiable, that φ is smooth and
real valued, and that φ(k) 6= 0 on [a, b] (and if k = 1 also that φ0 is monotonic).
Then
Z b ck  Z b  1
0
A(x)eiλφ(x) dx ≤ · |A(b)| + |A (t)|dt .

(minx∈[a,b] |φ(k) (x)|)1/k |λ|1/k

a a

Proof. For notational simplicity we treat the case k = 2. Denote


Z t
Jλ (t) = eiλφ(x) dx
a
iλφ(t)
So that Jλ (t) = e . Integrating by parts, we have
Z b b Z b
A(x)eiλφ(x) dx = A(t)Jλ (t) − A0 (t)Jλ (t)dt

a a a

Using our results for A ≡ 1 for Jλ (t), we have


b Z b
8 1
A(t)Jλ (t) = A(b) eiλφ(x) dx ≤ |A(b)| p ·p

a a minx∈[a,b] |φ00 (x)| |λ|
and
Z b Z b
8 1
|A0 (t)||Jλ (t)|dt ≤ |A0 (t)| p 00
· p dt
a a minx∈[a,b] |φ (x)| |λ|


Corollary 1.5. Assume that the amplitude A and the phase function φ are smooth,
and that φ has finitely many critical points, all of them non-degenerate. Then
1
I(λ)  √ , λ → +∞
λ
the implied constant depending on A and φ.
Proof. To use van der Corput’s Lemma, we use a smooth partition of unity to write
X
1[0,2π] = ψj
j

where ψj are smooth, the support of each contains at most one of the critical points,
and when the support does contain a critical point x0 (at which φ00 (x0 ) 6= 0), we
take the support sufficiently small so that φ00 6= 0 on all of supp ψj , while the
remaining ψj are supported away from the critical points. Hence we can write
X Z 2π
I(λ) = Ij (λ), Ij (λ) = ψj (t)A(t)eiλφ(t) dt.
j 0

To bound the integrals Ij (γ) where the support of ψj does not include any critical
points, we use the principle of NON-stationary phase, with a smooth amplitude and
a phase function so that |φ0 | ≥ cj > 0 where cj = min(|φ0 (x)| : x ∈ supp ψj ) to
bound
1
Ij (λ)  N , ∀N ≥ 1.
|ξ|
6 ZEÉV RUDNICK

For j such that supp ψj contains a critical point (unique by assumption), we can
Corollary 1.4 with k = 2, since we have taken the support of such ψj so that φ00 6= 0
on supp ψj . Hence for such j we have
1
Ij (λ)  √ .
λ
Altogether we have proven I(λ)  λ−1/2 . 

1.3. An asymptotic expansion: The method of stationary phase. It is


possible to go from upper bounds to asymptotic expansions. I will quote a typical
result (we will not explicitly use this).
Theorem 1.6. Assume that the amplitude A ∈ Cc∞ (R) is smooth and compactly
supported, and that the phase function (real valued and smooth) has a single critical
point x0 ∈ supp A, which is non-degenerate: φ0 (x0 ) = 0, φ00 (x0 ) 6= 0. Then
s
π 00 2π eiλφ(x0 )
I(λ) ∼ ei 4 sign(φ (x0 )) A(x0 ) · √ , as λ → +∞.
|φ00 (x0 )| λ
1.4. An application: The Fourier transform of the unit disk. We want to
bound the Fourier transform of the unit disk B(0, 1) ⊂ R2 : If χ(x) = 1, |x| ≤ 1,
and is zero otherwise, the Fourier transform is (we have dropped the factor of −2π
from the exponent)
Z
χ
b(ξ) = eihξ,xi dx, ξ ∈ R2 .
|x|≤1

Proposition 1.7.
1
b(ξ) 
χ .
1 + |ξ|3/2
Proof. We convert the 2-dimensional integral to a one-dimensional integral, to
which we can apply the van der Corput bound, by using Green’s theorem. Recall
that Green’s theorem says that for a bounded planar domain D, with a piecewise
smooth boundary ∂D, we have for A, B ∈ C 2 (R2 ),
Z  I
∂B ∂A 
− dx ∧ dy = Adx + Bdy
D ∂x ∂y ∂D

where the line integral over the boundary ∂D is taken counterclockwise.


For us, D = B(0, 1) is the unit disk, with boundary ∂D = S 1 the unit circle,
and we want to find A, B so that
∂B ∂A
− = ei(ax+by) , ξ = (a, b) 6= 0.
∂x ∂y
A solution is to take
ibei(ax+by) −iaei(ax+by)
A= , B= .
|ξ|2 |ξ|2
Hence we find Z
i
χ
b(ξ) = 2 ei(ax+by) (bdx − ady).
|ξ| ∂D
OSCILLATORY INTEGRALS 7

Parameterizing ∂D = S 1 by arc-length: γ(t) = (cos t, sin t), which runs counter-


clockwise if t runs from 0 to 2π, we obtain
Z 2π
i i
χ
b(ξ) = 2 ei(a cos t+b sin t) (−b sin t − a cos t)dt = I(|ξ|)
|ξ| 0 |ξ|
where Z 2π
I(λ) = Aξ (t)eiλφξ (t) dt
0
is an oscillatory integral, with amplitude
ξ⊥ −a cos t − b sin t
Aξ (t) = hγ̇(t), i= √
|ξ| a 2 + b2
where ξ ⊥ = (b, −a), and phase function
ξ a cos t + b sin t
φξ (t) = hγ(t), i= √ .
|ξ| a2 + b2
We want to apply our estimates on oscillatory integrals to this case. Note that
both the amplitude and the phase function depend on ξ/|ξ|, so it is important to
make sure that our estimates our uniform in ξ/|ξ| ∈ S 1 .
The phase function has two critical points φ0 (t) = hγ̇(t), ξ/|ξ|i = 0, when γ̇(t) ⊥
ξ, i.e. when γ̇(t) = ±ξ ⊥ /|ξ| (note γ̇ is a unit vector), with ξ ⊥ = (b, −a), say at t0
and therefore also at t0 + π.
We claim that these critical points are non-degenerate: The second derivative of
the phase function is
φ00 (t0 ) = hγ̈(t0 ), ξi.
Now further note that hγ̈, γ̇i = 0 (which follows by direct computation or better
yet by differentiating |γ̇|2 = 1, which is arc-length parameterization condition), and
since also ξ ⊥ γ̇(t0 ), we must have
ξ
γ̈(t0 ) = ±κ
|ξ|
where κ = |γ̈(t0 )|; note that here |γ̈| ≡ 1 so κ = 1 and hence2
ξ ξ ξ
φ00 (t0 ) = hγ̈(t0 ), i = ±h , i = ±1 6= 0.
|ξ| |ξ| |ξ|
To use van der Corput’s Lemma, we use a smooth partition of unity to write
1[0,2π] = ψ1 + ψ2 + ψ3
where ψj are smooth, ψ1 is supported in say (t0 − 0.1, t0 + 0.1), ψ1 is supported
in (t0 + π − 0.1, t0 + π + 0.1) and ψ3 is supported away from the critical points
t0 , t0 + π. This gives
I(λ) = I1 + I2 + I3
with Z 2π
Ij (λ) = ψj (t)A(t)eiλφ(t) dt.
0

2For general D, |γ̈| = κ will be the curvature of ∂D; if we assume that the boundary ∂D
has nowhere vanishing curvature, then this computation shows that the critical points are all
non-degenerate.
8 ZEÉV RUDNICK

For the integrals I1 , I2 we can use van der Corput’s Lemma to deduce that
Z 2π
8 
0
 1
|I1 (|ξ|)| ≤ 0 · |A(t0 + 0.1)| + |A (t)|dt .
(minx∈[t0 −0.1,t0 +0.1] |φ‘ (x)|)1/2 0 |ξ|1/2
We have
ξ ξ
|A(t)| = |hγ̇(t), i| ≤ |γ̇| · | =1
|ξ| |ξ|
and likewise |A0 (t)| ≤ 1. Since φ00 (t0 ) = ±1, we have
min(|φ00 (t)| : t0 − 0.1 < t < t0 + 0.1) > c0
for some positive constant c0 . Hence we deduce from van der Corput’s Lemma that
there is some C > 0, independent of ξ, so that
C
|I1 (|ξ|)|, |I2 (|ξ|)| ≤ 1/2 .
|ξ|
To bound the third integral I3 (γ), we use the principle of NON-stationary phase,
with a smooth amplitude and a phase function so that |φ0 | ≥ c3 > 0 (uniformly in
ξ) to bound
1
I3  N , ∀N ≥ 1.
|ξ|
Altogether we obtain that uniformly in |ξ| ≥ 1,
c4
I(λ) ≤ √
λ
and hence
c4
|b
χ(ξ)| ≤ 3/2 .
|ξ|

Exercise 2. Show that in dimension 3, we have an estimate for the Fourier trans-
form of the unit ball
Z
1
eihx,ξi d3 x  2
, ξ ∈ R3 .
|x|≤1 1 + |ξ|
Hint: Here we can directly evaluate the Fourier transform in elementary terms!

You might also like