New York Journal of Mathematics

New York J. Math. 6 (2000) 153–225.

Green’s Functions for Elliptic and Parabolic

Equations with Random Coefficients

Joseph G. Conlon and Ali Naddaf

Abstract. This paper is concerned with linear uniformly elliptic and par-
abolic partial differential equations in divergence form. It is assumed that
the coefficients of the equations are random variables, constant in time. The
Green’s functions for the equations are then random variables. Regularity
properties for expectation values of Green’s functions are obtained. In par-
ticular, it is shown that the expectation value is a continuously differentiable
function whose derivatives are bounded by the corresponding derivatives of
the heat equation. Similar results are obtained for the related finite difference
equations.

Contents
1. Introduction 153
2. Proof of Theorem 1.6 159
3. Proof of Theorem 1.5 174
4. Proof of Theorem 1.4—Diagonal Case 192
5. Proof of Theorem 1.4—Off Diagonal case 204
6. Proof of Theorem 1.2 215
References 224

1. Introduction
Let (Ω, F, µ) be a probability space and a : Ω → Rd(d+1)/2 be a bounded mea-
surable function from Ω to the space of symmetric d × d matrices. We assume that
there are positive constants Λ, λ such that
(1.1) λId ≤ a(ω) ≤ ΛId , ω ∈ Ω,
in the sense of quadratic forms, where Id is the identity matrix in d dimensions. We
assume that Rd acts on Ω by translation operators τx : Ω → Ω, x ∈ Rd , which are
measure preserving and satisfy the properties τx τy = τx+y , τ0 = identity, x, y ∈ Rd .
We assume also that the function from Rd × Ω to Ω defined by (x, ω) → τx ω,

Received May 25, 2000.

Mathematics Subject Classification. 35R60, 60J75.
Key words and phrases. Green’s functions, diffusions, random environments.

ISSN 1076-9803/00

153
154 Joseph G. Conlon and Ali Naddaf

x ∈ Rd , ω ∈ Ω, is measurable. It follows that with probability 1 the function

a(x, ω) = a(τx ω), x ∈ Rd , is a Lebesgue measurable function from Rd to d × d
matrices.
Consider now for ω ∈ Ω such that a(x, ω) is a measurable function of x ∈ Rd ,
the parabolic equation

d  
∂u ∂ ∂
(1.2) = ai,j (x, ω) u(x, t, ω) , x ∈ Rd , t > 0,
∂t i,j=1
∂xi ∂xj

u(x, 0, ω) = f (x, ω), x ∈ Rd .

It is well known that the solution of this initial value problem can be written as

u(x, t, ω) = Ga (x, y, t, ω)f (y, ω)dy ,
Rd

where Ga (x, y, t, ω) is the Green’s function, and Ga is measurable in (x, y, t, ω).

Evidently Ga is a positive function which satisfies

(1.3) Ga (x, y, t, ω)dy = 1.
Rd

It also follows from the work of Aronson [1] (see also [5]) that there is a constant
C(d, λ, Λ) depending only on dimension d and the uniform ellipticity constants λ, Λ
of (1.1) such that
 
C(d, λ, Λ) −|x − y|2
(1.4) 0 ≤ Ga (x, y, t, ω) ≤ exp .
td/2 C(d, λ, Λ)t
In this paper we shall be concerned with  the expectation value of Ga over Ω.
Denoting expectation value on Ω by we define the function Ga (x, t), x ∈
Rd , t > 0 by
 
Ga (x, 0, t, ·) = Ga (x, t) .

Using the fact that τx τy = τx+y , x, y ∈ Rd , we see from the uniqueness of solutions
to (1.2) that
Ga (x, y, t, ω) = Ga (x − y, 0, t, τy ω),
whence the measure preserving property of the operator τy yields the identity,
 
Ga (x, y, t, ·) = Ga (x − y, t) .

From (1.3), (1.4) we have


Ga (x, t)dx = 1, t > 0,
Rd
 
C(d, λ, Λ) −|x|2
(1.5) 0 ≤ Ga (x, t) ≤ exp , x ∈ Rd , t > 0.
td/2 C(d, λ, Λ)t
In general one cannot say anything about the smoothness properties of the function
Ga (x, y, t, ω). We shall, however, be able to prove here that Ga (x, t) is a C 1 function
of (x, t), x ∈ Rd , t > 0.
 Green’s Functions for Equations with Random Coefficients 155

Theorem 1.1. Ga (x, t) is a C 1 function of (x, t), x ∈ Rd , t > 0. There is a

constant C(d, λ, Λ), depending only on d, λ, Λ such that
 
∂Ga −|x|2
(x, t) ≤ C(d, λ, Λ) exp ,
∂t td/2 + 1 C(d, λ, Λ)t
 
∂Ga −|x|2
(x, t) ≤ C(d, λ, Λ) exp .
∂xi td/2 + 1/2 C(d, λ, Λ)t
The Aronson inequality (1.5) shows us that Ga (x, t) is bounded by the kernel
of the heat equation. Theorem 1.1 proves that corresponding inequalities hold for
the first derivatives of Ga (x, t). We cannot use our methods to prove existence of
second derivatives of Ga (x, t) in the space variable x. In fact we are inclined to
believe that second space derivatives do not in general exist in a pointwise sense.
As well as the parabolic problem (1.2) we also consider the corresponding elliptic
problem,

d  
∂ ∂u
(1.6) − ai,j (x, ω) (x, ω) = f (x, ω), x ∈ Rd .
i,j=1
∂xi ∂xj

If d ≥ 3 then the solution of (1.6) can be written as


u(x, ω) = Ga (x, y, ω)f (y, ω)dy,
Rd

where Ga (x, y, ω) is the Green’s function and is measurable in (x, y, ω). It follows
again by Aronson’s work that there is a constant C(d, λ, Λ), depending only on
d, λ, Λ, such that

(1.7) 0 ≤ Ga (x, y, ω) ≤ C(d, λ, Λ)/|x − y|d−2 , d ≥ 3.

Again we consider the expectation of the Green’s function, Ga (x), defined by

 
Ga (x, y, ·) = Ga (x − y).

It follows from (1.7) that

0 ≤ Ga (x) ≤ C(d, λ, Λ)/|x|d−2 , d ≥ 3.

Theorem 1.2. Suppose d ≥ 3. Then Ga (x) is a C 1 function of x for x

= 0. There
is a constant C(d, λ, Λ) depending only on d, λ, Λ, such that

∂Ga C(d, λ, Λ)

∂xi (x) ≤ |x|d−1 , x
= 0.

We can also derive estimates on the Fourier transforms of Ga (x, t) and Ga (x).
For a function f : Rd → C we define its Fourier transform fˆ by

fˆ(ξ) = f (x)eix·ξ dx, ξ ∈ Rd .
Rd

Evidently from the equation before (1.5) we have that |Ĝa (ξ, t)| ≤ 1
156 Joseph G. Conlon and Ali Naddaf

Theorem 1.3. The function Ĝa (ξ, t) is continuous for ξ ∈ Rd , t > 0, and differen-
tiable with respect to t. Let δ satisfy 0 ≤ δ < 1. Then there is a constant C(δ, λ, Λ)
depending only on δ, λ, Λ, such that
C(δ, λ, Λ)
|Ĝa (ξ, t)| ≤ ,
[1 + |ξ|2 t]δ

∂ Ĝ C(δ, λ, Λ)|ξ|2
a
(ξ, t) ≤ ,
∂t [1 + |ξ|2 t]1+δ
where |ξ| denotes the Euclidean norm of ξ ∈ Rd .
Remark 1.1. Note that the dimension d does not enter in the constant C(δ, λ, Λ).
Also, our method of proof breaks down if we take δ → 1.
In this paper we shall be mostly concerned with a discrete version of the parabolic
and elliptic problems (1.2), (1.6). Then Theorems 1.1, 1.2, 1.3 can be obtained as
a continuum limit of our results on the discrete problem. In the discrete problem
we assume Zd acts on Ω by translation operators τx : Ω → Ω, x ∈ Zd , which
are measure preserving and satisfy the properties τx τy = τx+y , τ0 = identity. For
functions g : Zd → R we define the discrete derivative ∇i g of g in the i th direction
to be
∇i g(x) = g(x + ei ) − g(x), x ∈ Zd ,
where ei ∈ Zd is the element with entry 1 in the i th position and 0 in other
positions. The formal adjoint of ∇i is given by ∇∗i , where
∇∗i g(x) = g(x − ei ) − g(x), x ∈ Zd .
The discrete version of the problem (1.2) that we shall be interested in is given by

d
∂u
(1.8) =− ∇∗i [aij (τx ω)∇j u(x, t, ω)] , x ∈ Zd , t > 0,
∂t i,j=1

u(x, 0, ω) = f (x, ω), x ∈ Zd .

The solution of (1.8) can be written as


u(x, t, ω) = Ga (x, y, t, ω)f (y, ω),
y∈Zd

where Ga (x, y, t, ω) is the discrete Green’s function. As in the continuous case, Ga

is a positive function which satisfies


Ga (x, y, t, ω) = 1.
y∈Zd

It also follows from the work of Carlen et al [3] that there is a constant C(d, λ, Λ)
depending only on d, λ, Λ such that
 
C(d, λ, Λ) min{|x − y|, |x − y|2 /t}
0 ≤ Ga (x, y, t, ω) ≤ exp − .
1 + td/2 C(d, λ, Λ)
Now let Ga (x, t), x ∈ Zd , t > 0, be the expectation of the Green’s function,
 
(1.9) Ga (x, y, t, ·) = Ga (x − y, t).
 Green’s Functions for Equations with Random Coefficients 157

Then we have


Ga (x, t) = 1, t>0,
x∈Zd
 
C(d, λ, Λ) min{|x|, |x|2 /t}
(1.10) Ga (x, t) ≤ exp − , x ∈ Zd , t > 0 .
1 + td/2 C(d, λ, Λ)
The discrete version of Theorem 1.1 which we shall prove is given by the following:
Theorem 1.4. Ga (x, t), x ∈ Zd , t > 0 is differentiable in t. There is a constant
C(d, λ, Λ), depending only on d, λ, Λ such that
 
∂Ga 2
(x, t) ≤ C(d, λ, Λ) exp − min{|x|, |x| /t} ,
∂t 1 + td/2 + 1 C(d, λ, Λ)
 
C(d, λ, Λ) min{|x|, |x|2 /t}
|∇i Ga (x, t)| ≤ exp − .
1 + td/2 + 1/2 C(d, λ, Λ)
Let δ satisfy 0 ≤ δ < 1. Then there is a constant C(δ, d, λ, Λ) depending only on
δ, d, λ, Λ such that
 
C(δ, d, λ, Λ) min{|x|, |x|2 /t}
(1.11) |∇i ∇j Ga (x, t)| ≤ exp − .
1 + t(d+1+δ)/2 C(d, λ, Λ)
Remark 1.2. As in Theorem 1.1, Theorem 1.4 shows that first derivatives of
Ga (x, t) are bounded by corresponding heat equation quantities. It also shows
that second space derivatives are almost similarly bounded. We cannot put δ = 1
in (1.11) since the constant C(δ, d, λ, Λ) diverges as δ → 1.
The elliptic problem corresponding to (1.8) is given by
d


(1.12) ∇∗i [ai,j (τx ω)∇j u(x, ω)] = f (x, ω), x ∈ Zd .
i,j=1

If d ≥ 3 then the solution of (1.12) can be written as

u(x, ω) = Ga (x, y, ω)f (y, ω),
y∈Zd

where Ga (x, y, ω) is the discrete Green’s function. It follows from Carlen et al [3]
that there is a constant C(d, λ, Λ) depending only on d, λ, Λ such that
(1.13) 0 ≤ Ga (x, y, ω) ≤ C(d, λ, Λ)/[1 + |x − y|d−2 ], d ≥ 3.
Letting Ga (x) be the expectation of the Green’s function,
 
Ga (x, y, ·) = Ga (x − y),

it follows from (1.13) that

(1.14) 0 ≤ Ga (x) ≤ C(d, λ, Λ)/[1 + |x|d−2 ], d ≥ 3.
We shall prove a discrete version of Theorem 1.2 as follows:
Theorem 1.5. Suppose d ≥ 3. Then there is a constant C(d, λ, Λ), depending only
on d, λ, Λ such that
|∇i Ga (x)| ≤ C(d, λ, Λ)/[1 + |x|d−1 ], x ∈ Zd .
158 Joseph G. Conlon and Ali Naddaf

Let δ satisfy 0 ≤ δ < 1. Then there is a constant C(δ, d, λ, Λ) depending only on

δ, d, λ, Λ such that

|∇i ∇j Ga (x)| ≤ C(δ, d, λ, Λ)/[1 + |x|d−1+δ ], x ∈ Zd .

Remark 1.3. As in Theorem 1.4 our estimates on the second derivatives of Ga (x)
diverge as δ → 1.

Next we turn to the discrete version of Theorem 1.3. For a function f : Zd → C

we define its Fourier transform fˆ by


fˆ(ξ) = f (x)eix·ξ , ξ ∈ Rd .
x∈Zd

For 1 ≤ k ≤ d, ξ ∈ Rd , let ek (ξ) = 1 − eiek ·ξ and e(ξ) be the vector e(ξ) =

(e1 (ξ), . . . , ed (ξ)). Let Ĝa (ξ, t) be the Fourier transform of the function Ga (x, t), x ∈
Zd , t > 0, defined by (1.9). From the equation before (1.10) it is clear that
|Ĝa (ξ, t)| ≤ 1.

Theorem 1.6. The function Ĝa (ξ, t) is continuous for ξ ∈ Rd and differentiable
for t > 0. Let δ satisfy 0 ≤ δ < 1. Then there is a constant C(δ, λ, Λ) depending
only on δ, λ, Λ, such that
C(δ, λ, Λ)
|Ĝa (ξ, t)| ≤ ,
[1 + |e(ξ)|2 t]δ
∂ Ĝa C(δ, λ, Λ)|e(ξ)|2
| (ξ, t)| ≤ ,
∂t [1 + |e(ξ)|2 t]1+δ

where |e(ξ)| denotes the Euclidean norm of e(ξ) ∈ Cd .

In order to prove Theorems 1.1–1.6 we use a representation for the Fourier trans-
form of the expectation of the Green’s function for the elliptic problem (1.12), which
was obtained in [4] . This in turn gives us a formula for the Laplace transform of the
function Ĝa (ξ, t) of Theorem 1.6. We can prove Theorem 1.6 then by estimating
the inverse Laplace transform. In order to prove Theorems 1.4, 1.5 we need to use
interpolation theory, in particular the Hunt Interpolation Theorem [10]. Thus we
prove that Ĝa (ξ, t) is in a weak Lp space which will then imply pointwise bounds
on the Fourier inverse. We shall prove here Theorems 1.4–1.6 in detail. In the
final section we shall show how to generalize the proof of Theorem 1.5 to prove
Theorem 1.2. The proofs of Theorems 1.1 and 1.3 are left to the interested reader.
We would like to thank Jana Björn and Vladimir Maz’ya for help with the proof of
Lemma 2.6.
There is already a large body of literature on the problem of homogenization of
solutions of elliptic and parabolic equations with random coefficients, [4] [6] [7] [8]
[11]. These results prove in a certain sense that, asympotically, the lowest frequency
components of the functions Ga (x) and Ga (x, t) are the same as the corresponding
quantities for a constant coefficient equation. The constant depends on the random
matrix a(·). The problem of homogenization in a periodic medium has also been
studied [2] [11], and similar results been obtained.
 Green’s Functions for Equations with Random Coefficients 159

2. Proof of Theorem 1.6

Let Ĝa (ξ, t), ξ ∈ Rd , t > 0, be the function in Theorem 1.6. Our first goal will
be to obtain a formula for the Laplace transform of Ĝa (ξ, t), which we denote by
Ĝa (ξ, η),
 ∞
Ĝa (ξ, η) = dt e−ηt Ĝa (ξ, t) , Re(η) > 0 .
0

It is evident that Ĝa (ξ, η) is the Fourier transform of the expectation of the Green’s
function for the elliptic problem,
d


(2.1) ηu(x, ω) + ∇∗i [ai,j (τx ω)∇j u(x, ω)] = f (x, ω) , x ∈ Zd .
i,j=1

In [4] we derived a formula for this. To do that we defined operators ∂i , 1 ≤ i ≤ d,

on functions ψ : Ω → C by ∂i ψ(ω) = ψ(τei ω) − ψ(ω), with corresponding adjoint
operators ∂i∗ , 1 ≤ i ≤ d, defined by ∂i∗ ψ(ω) = ψ(τ−ei ω) − ψ(ω). Hence for ξ ∈ Rd
we may define an operator Lξ on functions ψ : Ω → C by
d


Lξ ψ(ω) = P eiξ·(ei −ej ) [∂i∗ + ei (−ξ)] ai,j (ω) [∂j + ej (ξ)] ψ(ω) ,
i,j=1

where P is the projection orthogonal to the constant function and ej (ξ) is defined
just before the statement of Theorem 1.6. Note that Lξ takes a function ψ to a
function Lξ ψ satisfying Lξ ψ = 0. Now for 1 ≤ k ≤ d, ξ ∈ Rd , Re(η) > 0, let
ψk (ξ, η, ω) be the solution to the equation,
d



(2.2) [Lξ + η]ψk (ξ, η, ω) + eiej ·ξ ∂j∗ + ej (−ξ) [ak,j (ω) − ak,j (·)] = 0 .
j=1

Then we may define a d × d matrix q(ξ, η) by,

 d



−iej ·ξ
(2.3) qk,k
(ξ, η) = ak,k
(·) + ak,j (·)e [∂j + ej (ξ)] ψk
(ξ, η, ·) .
j=1

The function Ĝa (ξ, η) is then given by the formula,

1
(2.4) Ĝa (ξ, η) = , ξ ∈ Rd , Re(η) > 0 .
η + e(ξ)q(ξ, η)e(−ξ)
We actually established the formula (2.4) in [4] when η is real and positive. In that
case q(ξ, η) is a d × d Hermitian matrix bounded below in the quadratic form sense
by λId . It follows that Ĝa (ξ, η) is finite for all positive η. We wish to establish
this for all η satisfying Re(η) > 0. We can in fact argue this from (2.1). Suppose
the function on the RHS of (2.1) is a function of x only, f (x, ω) = f (x). Then the
Fourier transform û(ξ, ω) of the solution to (2.1) satisfies the equation,

û(ξ, ·) = Ĝa (ξ, η)fˆ(ξ), ξ ∈ Rd .

160 Joseph G. Conlon and Ali Naddaf

If we multiply (2.1) by u(x, ω), and sum with respect to x, we have by the Plancherel
Theorem,
 
|η|2 2
|û(ξ, ω)| dξ ≤ |fˆ(ξ)|2 dξ .
[−π,π]d [−π,π]d

Since fˆ(ξ) is an arbitrary function it follows that |Ĝa (ξ, η)| ≤ 1/|η|. We improve
this inequality in the following:
Lemma 2.1. Suppose Re(η) > 0 and ξ ∈ Rd . Let ρ = (ρ1 , . . . , ρd ) ∈ Cd . Then
(2.5) Re[η + ρ̄q(ξ, η)ρ] ≥ Re(η) + λ|ρ|2 ,
(2.6) Im(η)Im[ρ̄q(ξ, η)ρ] ≥ 0.
Proof. From (2.2), (2.3) we have that


d


qk,k
(ξ, η) = ai,j (·) δk,i + eiei ·ξ [∂i + ei (−ξ)]ψk (−ξ, η, ·)
i,j=1

  
δk
,j + e−iej ·ξ [∂j + ej (ξ)]ψk
(ξ, η, ·) + η ψk (−ξ, η, ·)ψk
(ξ, η, ·) .
Thus we have


d  
(2.7) ρ̄q(ξ, η)ρ = ai,j (·) ρ̄i + eiei ·ξ [∂i + ei (−ξ)]ϕ(ξ, η̄, ·)
i,j=1

  
ρj + e−iej ·ξ [∂j + ej (ξ)]ϕ(ξ, η, ·) + η ϕ(ξ, η̄, ·)ϕ(ξ, η, ·) ,
where
d


(2.8) ϕ(ξ, η, ·) = ρk ψk (ξ, η, ·) .
k=1
Evidently we have that
d

d


(2.9) [Lξ + η]ϕ(ξ, η, ω) + ρk eiej ·ξ [∂j∗ + ej (−ξ)] [ak,j (ω) − ak,j (·)] = 0 .
k=1 j=1

It follows from the last equation that

[Lξ + η]ϕ(ξ, η, ·) = [Lξ + η̄]ϕ(ξ, η̄, ·),
whence
(2.10) [Lξ + Re(η)][ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·)] = −i Im(η)[ϕ(ξ, η, ·) + ϕ(ξ, η̄, ·)] .
Hence
 
(2.11) [ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·)][Lξ + Re(η)][ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·)]
 
= −i Im(η) [ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·)][ϕ(ξ, η, ·) + ϕ(ξ, η̄, ·)] .
Observe that since the LHS of (2.11) is real, the quantity
 
[ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·)][ϕ(ξ, η, ·) + ϕ(ξ, η̄, ·)]
is pure imaginary.
 Green’s Functions for Equations with Random Coefficients 161

Next for 1 ≤ j ≤ d, let us put

1
Aj = ρj + e−iej ·ξ [∂j + ej (ξ)] {ϕ(ξ, η, ·) + ϕ(ξ, η̄, ·)},
2
1
Bj = e−iej ·ξ [∂j + ej (ξ)] {ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·)}.
2
Then


d   
ρ̄q(ξ, η)ρ = ai,j (·)[Āi − B̄i ][Aj + Bj ] + η ϕ(ξ, η̄, ·)ϕ(ξ, η, ·) .
i,j=1

We can decompose this sum into real and imaginary parts. Thus



d  

d  

d 
ai,j (·)[Āi − B̄i ][Aj + Bj ] = ai,j (·)Āi Aj − ai,j (·)B̄i Bj
i,j=1 i,j=1 i,j=1


d 
+ 2i Im ai,j (·)Āi Bj .
i,j=1

Evidently the first two terms on the RHS of the last equation are real while the
third term is pure imaginary. We also have that
 
ϕ(ξ, η̄, ·)ϕ(ξ, η, ·)

1  
= [ϕ(ξ, η̄, ·) + ϕ(ξ, η, ·)] + [ϕ(ξ, η̄, ·) − ϕ(ξ, η, ·)]
4

{ϕ(ξ, η, ·) + ϕ(ξ, η̄, ·)] + [ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·)]}
   
1 2 1 2
= |ϕ(ξ, η, ·) + ϕ(ξ, η̄, ·)| − |ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·)|
4 4
 
i
− [ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·)][Lξ + Re(η)][ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·)] ,
2Im(η)

where we have used (2.11). Observe that the first two terms on the RHS of the last
equation are real while the third term is pure imaginary. Hence
 
η ϕ(ξ, η̄, ·)ϕ(ξ, η, ·) =
Re(η)   Re(η)  
|ϕ(ξ, η, ·) + ϕ(ξ, η̄, ·)|2 − |ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·)|2
4 4
1  
+ ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·)][Lξ + Re(η)][ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·)
2
Im(η)   Im(η)  
+i |ϕ(ξ, η, ·) + ϕ(ξ, η̄, ·)|2 − i |ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·)|2
4 4
i Re(η)  
− [ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·)][Lξ + Re(η)][ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·) .
2 Im(η)
162 Joseph G. Conlon and Ali Naddaf

We conclude then from the last four equations that

(2.12)


d  

d 
Re[ ρ̄q(ξ, η)ρ] = ai,j (·)Āi Aj − ai,j (·)B̄i Bj
i,j=1 i,j=1
Re(η)   Re(η)  
+ |ϕ(ξ, η, ·) + ϕ(ξ, η̄, ·)|2 − |ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·)|2
4 4
1  
+ [ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·)][Lξ + Re(η)][ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·) ,
2
(2.13)


d  Im(η)  
Im[ρ̄q(ξ, η)ρ] = 2 Im ai,j (·)Āi Bj + |ϕ(ξ, η, ·) + ϕ(ξ, η̄, ·)|2
i,j=1
4
Im(η)   Re(η)
− |ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·)|2 − ·
4 2 Im(η)
 
[ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·)][Lξ + Re(η)][ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·) .

We can simplify the expression on the RHS of (2.12) by observing that



d  1 
ai,j (·)B̄i Bj = [ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·)]Lξ [ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·)] .
i,j=1
4

Hence we obtain the expression,



d  Re(η)  
(2.14) Re[ρ̄q(ξ, η)ρ] = ai,j (·)Āi Aj + |ϕ(ξ, η, ·) + ϕ(ξ, η̄, ·)|2
i,j=1
4
1  
+ [ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·)][Lξ + Re(η)][ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·)] .
4

Now all the terms on the RHS of the last expression are nonnegative, and from
Jensen’s inequality,



d 
ai,j (·)Āi Aj ≥ λ |ρ|2 .
i,j=1

The inequality in (2.5) follows from this.

To prove the inequality (2.6) we need to rewrite the RHS of (2.13). We have
now



d  

d 
ai,j (·)Āi Bj = ai,j (·)ρ̄i Bj
i,j=1 i,j=1
1  
+ [ϕ(ξ, η, ·) + ϕ(ξ, η̄, ·)]Lξ [ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·) .
4
 Green’s Functions for Equations with Random Coefficients 163

From (2.10) we have that

  
ϕ(ξ, η, ·) + ϕ(ξ, η̄, ·) Lξ [ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·)]
 
= −i Im(η) |ϕ(ξ, η, ·) + ϕ(ξ, η̄, ·)|2
  
− Re(η) ϕ(ξ, η, ·) + ϕ(ξ, η̄, ·) [ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·)]
 
= −i Im(η) |ϕ(ξ, η, ·) + ϕ(ξ, η̄, ·)|2
Re(η)   
+i ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·) [Lξ + Re(η)] [ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·)] ,
Im(η)
where we have used (2.11). We also have that


d 
ai,j (·)ρ̄i Bj
i,j=1

1 
 
d


= ρi eiej ·ξ ∂j∗ + ej (−ξ) ai,j (·) [ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·)]
2 i,j=1
1 
=− {[Lξ + η] ϕ(ξ, ·) + [Lξ + η̄] ϕ(ξ, η̄, ·)} [ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·)]
4
1 
=− {[Lξ + Re(η)] [ϕ(ξ, η, ·) + ϕ(ξ, η̄, ·)]} [ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·)]
4
i  
+ Im(η) |ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·)|2
4
1   
=− ϕ(ξ, η, ·) + ϕ(ξ, η̄, ·) [Lξ + Re(η)] [ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·)]
4
i  
+ Im(η) |ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·)|2
4
i
   
= Im(η) |ϕ(ξ, η, ·) + ϕ(ξ, η̄, ·)|2 + |ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·)|2 .
4
It follows now from (2.13) and the last three identities that
1  
(2.15) Im[ρ̄q(ξ, η)ρ] = Im(η) |ϕ(ξ, η, ·) + ϕ(ξ, η̄, ·)|2
4
1  
+ Im(η) |ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·)|2 .
4
The inequality (2.6) follows.

Let us denote by Ĝa (ξ, t), t > 0 the inverse Laplace transform of Ĝa (ξ, η). Thus
from (2.4) we have
 N
1 eηt
(2.16) Ĝa (ξ, t) = lim d[Im(η)].
N →∞ 2π −N η + e(ξ)q(ξ, η)e(−ξ)

It follows from Lemma 2.1 that, provided Re(η) > 0, the integral in (2.16) over the
finite interval −N < Im(η) < N exists for all N . We need then to show that the
limit as N → ∞ in (2.16) exists.
Lemma 2.2. Suppose Re(η) > 0 and ξ ∈ Rd . Then the limit in (2.16) as N → ∞
exists and is independent of Re(η) > 0.
164 Joseph G. Conlon and Ali Naddaf

Proof. We first note that for any ρ ∈ Cd , ρ̄q(ξ, η)ρ and ρ̄q(ξ, η̄)ρ are complex
conjugates. This follows easily from (2.7). We conclude from this that

(2.17)
 N 
1 eηt 1 N eηt
d[Im(η)] = Re d[Im(η)]
2π −N η + e(ξ)q(ξ, η)e(−ξ) π 0 η + e(ξ)q(ξ, η)e(−ξ)
1  N  N 
= exp[Re(η)t] h(ξ, η) cos[Im(η)t]d[Im(η)] + k(ξ, η) sin[Im(η)t]d[Im(η)] ,
π 0 0

where

(2.18) h(ξ, η) = Re [η + e(ξ)q(ξ, η)e(−ξ)] |η + e(ξ)q(ξ, η)e(−ξ)|2 ,

k(ξ, η) = Im [η + e(ξ)q(ξ, η)e(−ξ)] |η + e(ξ)q(ξ, η)e(−ξ)|2 .

We show there is a constant Cλ,Λ , depending only on λ, Λ such that

 ∞
(2.19) |h(ξ, η)|d[Im(η)] ≤ Cλ,Λ , Re(η) > 0.
0

To see this, observe from (2.14), (2.15) that

Re (η) + Θ
(2.20) |h(ξ, η)| ≤ , where
[{·}2 + (Im η)2 ]
    
d 1
i,j=1 ai,j (·)Āi Aj + 4 ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·) Lξ [ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·)]
Θ=  ,
1 + 14 |ϕ(ξ, η, ·) + ϕ(ξ, η̄, ·)|2  + 41 |ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·)|2 

and the quantity {·} in the first line of (2.20) is the same as the one in the second.
It is easy to see that

(2.21)


d  1   
ai,j (·)Āi Aj + ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·) Lξ [ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·)] ≥ λ|e(ξ)|2 .
i,j=1
4

We can also obtain an upper bound using the fact that



d  1   
ai,j (·)Āi Aj + ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·) Lξ [ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·)]
i,j=1
4


d  1   
≤2 ai,j (·)ei (ξ)ej (ξ) + ϕ(ξ, η, ·) + ϕ(ξ, η̄, ·) Lξ [ϕ(ξ, η, ·) + ϕ(ξ, η̄, ·)]
i,j=1
2
1   
+ ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·) Lξ [ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·)]
4
1   
≤ 2Λ|e(ξ)|2 + ϕ(ξ, η, ·) + ϕ(ξ, η̄, ·) Lξ [ϕ(ξ, η, ·) + ϕ(ξ, η̄, ·)]
4
1  1 
+ ϕ(ξ, η, ·)Lξ ϕ(ξ, η, ·) + ϕ(ξ, η̄, ·)Lξ ϕ(ξ, η̄, ·) .
2 2
 Green’s Functions for Equations with Random Coefficients 165

We see from (2.9) that

 
ϕ(ξ, η, ·)Lξ ϕ(ξ, η, ·) ≤ Λ|e(ξ)|2 ,
 
ϕ(ξ, η̄, ·)Lξ ϕ(ξ, η̄, ·) ≤ Λ|e(ξ)|2 ,
  
ϕ(ξ, η, ·) + ϕ(ξ, η̄, ·) Lξ [ϕ(ξ, η, ·) + ϕ(ξ, η̄, ·)] ≤ 4Λ|e(ξ)|2 .
Hence we obtain an upper bound
(2.22)

d  1   
ai,j (·)Āi Aj + ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·) Lξ [ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·)] ≤ 4Λ|e(ξ)|2 .
i,j=1
4

Observe that the upper and lower bounds (2.21), (2.22) are comparable for all
Re(η) > 0, ξ ∈ Rd .
Next we need to find upper and lower bounds on the quantity,
1  1 
|ϕ(ξ, η, ·) + ϕ(ξ, η̄, ·)|2 + |ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·)|2
4 4
1  1 
= |ϕ(ξ, η, ·)|2 + |ϕ(ξ, η̄, ·)|2 .
2 2
Evidently zero is a trivial lower bound. To get an upper bound we use (2.9) again.
We have from (2.9) that
   1/2
|η| |ϕ(ξ, η, ·)|2 ≤ ϕ(ξ, η, ·)Lξ ϕ(ξ, η, ·)


d 1/2
ai,j (·)ei (ξ)ej (ξ) ≤ Λ|e(ξ)|2 ,
i,j=1

whence
 
|ϕ(ξ, η, ·)|2 ≤ Λ|e(ξ)|2 /|η|.
We conclude then that
1  1 
(2.23) |ϕ(ξ, η, ·) + ϕ(ξ, η̄, ·)|2 + |ϕ(ξ, η, ·) − ϕ(ξ, η̄, ·)|2 ≤ Λ|e(ξ)|2 /|η|.
4 4
We use this last inequality together with (2.21), (2.22) to prove (2.19). First
note from (2.5) that


|h(ξ, η)| ≤ 1/ Re(η) + λ|e(ξ)|2 , Re(η) > 0, ξ ∈ Rd .
We also have using (2.20),(2.21), (2.22),(2.23) that
 
 1 2 
(2.24) |h(ξ, η)| ≤ Re(η) + 4Λ|e(ξ)|2 / Im(η)2 + Re(η) + λ|e(ξ)|2 ,
2
Re(η) > 0, |η| ≥ Λ|e(ξ)|2 , ξ ∈ Rd .
We then have
 ∞  Λ|e(ξ)|2
d[Im(η)]
|h(ξ, η)|d[Im(η)] ≤
0 0 Re(η) + λ|e(ξ)|2
 ∞
Re(η) + 4Λ|e(ξ)|2
+ d[Im(η)] ≤ Cλ,Λ ,
0 Im(η) + [Re(η) + 12 λ|e(ξ)|2 ]2
2
166 Joseph G. Conlon and Ali Naddaf

where Cλ,Λ depends only on λ, Λ, whence (2.19) follows.

Next we wish to show that the function k(ξ, η) of (2.18) satisfies
(2.25) |∂k(ξ, η)/∂[Im(η)]| ≤ 1/|Im(η)|2 , Re(η) > 0, ξ ∈ Rd .
In view of the analyticity in η of q(ξ, η) we have
∂ 1
∂k(ξ, η)/∂[Im(η)] = −Re
∂η η + e(ξ)q(ξ, η)e(−ξ)
1 + e(ξ)[∂q(ξ, η)/∂η]e(−ξ)
= Re .
[η + e(ξ)q(ξ, η)e(−ξ)]2
Let us denote now by ψ(ξ, η, ω) the function ϕ(ξ, η, ω) of (2.8) with ρ = e(−ξ). It
is easy to see then from (2.9) that

(2.26) ϕ(ξ, η̄, ω) = ψ(−ξ, η, ω) .

It follows now from (2.7) and (2.9) that
 
e(ξ)[∂q(ξ, η)/∂η]e(−ξ) = ψ(−ξ, η, ·)ψ(ξ, η, ·)
 ∂ψ   ∂ψ 
+η (−ξ, η, ·)ψ(ξ, η, ·) + η ψ(−ξ, η, ·) (ξ, η, ·)
∂η ∂η
 ∂ψ   ∂ψ 
+ ψ(−ξ, η, ·)Lξ (ξ, η, ·) + (−ξ, η, ·)Lξ ψ(ξ, η, ·)
∂η ∂η
 ∂ψ 
− (−ξ, η, ·)[Lξ + η]ψ(ξ, η, ·)
∂η
 ∂ψ 
− ψ(−ξ, η, ·)[Lξ + η] (ξ, η, ·)
∂η
 
= ψ(−ξ, η, ·)ψ(ξ, η, ·)
Hence,

 1 + ψ(−ξ, η, ·)ψ(ξ, η, ·) 1

(2.27) |∂k(ξ, η) ∂[Im(η)]| ≤ ≤ ,
[η + e(ξ)q(ξ, η)e(−ξ)]2 Im(η)2

from (2.15).
We can use (2.19), (2.27) to show the limit in (2.16) exists. In fact from (2.19)
it follows that
 N
lim h(ξ, η) cos[Im(η)t]d[Im(η)]
N →∞ 0

exists. We also have that

 N 
1 N ∂k(ξ, η)  
k(ξ, η) sin[Im(η)t]d[Im(η)] = − 1 − cos[Im(η)t] d[Im(η)]
0 t 0 ∂[Im(η)]
1
+ k(ξ, Re(η) + iN ){1 − cos[N t]}.
t
From (2.6) we have that
|k(ξ, Re(η) + iN ){1 − cos[N t]}| ≤ 2/N.
 Green’s Functions for Equations with Random Coefficients 167

From (2.27) we have that


1 ∞ ∂k(ξ, η)  
(2.28) ∂[Im(η)] 1 − cos[Im(η)t] d[Im(η)] ≤ C,
t 0
for some universal constant C. Hence the limit,
 N
lim k(ξ, η) sin[Im(η)t]d[Im(η)]
N →∞ 0
exists, whence the result holds.

Lemma 2.3. Let Ĝa (ξ, t) be defined by (2.16). Then Ĝa (ξ, t) is a continuous
bounded function. Furthermore, for any δ, 0 ≤ δ < 1, there is a constant C(δ, λ, Λ),
depending only on δ, λ, Λ, such that

(2.29) |Ĝa (ξ, t)| ≤ C(δ, λ, Λ) [1 + |e(ξ)|2 t]δ .

Proof. Consider the integral on the LHS of (2.28). We can obtain an improvement
on the estimate of (2.28) by improving the estimate of (2.27). We have now from
(2.27), (2.14), (2.15), and (2.23) that
1 + Λ|e(ξ)|2 /|η|
(2.30) |∂k(ξ, η)/∂[Im(η)]| ≤ .
λ2 |e(ξ)|4 + |Im(η)|2
Assume now that |e(ξ)|2 t > 2 and write the integral on the LHS of (2.28) as a sum,
 1/t  |e(ξ)|2  ∞
1 1 1
+ + .
t 0 t 1/t t |e(ξ)|2
We have now from (2.28), (2.30) that for 0 ≤ δ < 1,
 1/t  1/t  δ  1−δ
1 1 1+Λ 1 − cos[Im(η)t]
≤ d[Im(η)]
t 0 t 0 λ2 |e(ξ)|2 Im(η) Im(η)2

 1/t  δ
1−2δ 1+Λ
≤ 10t d[Im(η)] ≤ C(δ, λ, Λ)/[|e(ξ)|2 t]δ ,
0 λ2 |e(ξ)|2 Im(η)
for some constant C(δ, λ, Λ) depending only on δ, λ, Λ. Next we have
 |e(ξ)|2  |e(ξ)|2
1 1 1+Λ
≤ d[Im(η)]
t 1/t t 1/t λ2 |e(ξ)|2 Im(η)

1+Λ log[|e(ξ)|2 t]
= .
λ2 |e(ξ)|2 t
Finally, we have
 ∞  ∞
1 1 d[Im(η)] 1
≤ 2
≤ .
t |e(ξ)|2 t |e(ξ)|2 Im(η) |e(ξ)|2 t
We conclude therefore that the integral on the LHS of (2.28) is bounded by
C(δ, λ, Λ)
,
[1 + |e(ξ)|2 t]δ
168 Joseph G. Conlon and Ali Naddaf

for any δ, 0 ≤ δ < 1.

Next we need to estimate as N → ∞ the integral
 N 
1 N ∂h(ξ, η)
h(ξ, η) cos[Im(η)t] d[Im(η)] = − sin[Im(η)t] d[Im(η)]
0 t 0 ∂[Im(η)]
1
+ h(ξ, Re(η) + iN ) sin[N t].
t
It is clear from (2.6) that limN →∞ h(ξ, Re(η) + iN ) = 0. We have now that
∂h(ξ, η) ∂ 1
(2.31) = − Im
∂[Im(η)] ∂η η + e(ξ)q(ξ, η)e(−ξ)
1 + e(ξ)[∂q(ξ, η)/∂η]e(−ξ)
= Im .
[η + e(ξ)q(ξ, η)e(−ξ)]2
Hence the estimates (2.27), (2.30) on the derivative k(ξ, η) apply equally to the
derivative of h. We therefore write the integral
(2.32)
  1/t  |e(ξ)|2  ∞
1 ∞ ∂h(ξ, η)
| sin[Im(η)t]|d[Im(η)] = 1 +
1
+
1
.
t ∂[Im(η)] t 0 t 1/t t |e(ξ)|2
0

as a sum just as before. We have now from (2.30),

 1/t   
1 1 1/t 1+Λ
≤ | sin[Im(η)t]|d[Im(η)]
t 0 t 0 λ2 |e(ξ)|2 Im(η)
≤ Cλ,Λ /[|e(ξ)|2 t],

where Cλ,Λ depends only on λ, Λ. The other integrals on the RHS of (2.32) are
estimated just as for the corresponding integrals in k. We conclude that
 ∞

h(ξ, η) cos[Im(η)t]|d[Im(η)] ≤ C(δ, λ, Λ)/[1 + |e(ξ)|2 t]δ ,

0

for any δ, 0 ≤ δ < 1. It follows now from (2.17) that (2.29) holds for any δ, 0 ≤
δ < 1.

Lemma 2.4. The function Ĝa (ξ, t), ξ ∈ Rd is t differentiable for t > 0. For any
δ, 0 ≤ δ < 1, there is a constant C(δ, λ, Λ) depending only on δ, λ, Λ such that

∂ Ĝ (ξ, t) C(δ, λ, Λ)
a
≤ , t > 0.
∂t t[1 + |e(ξ)|2 t]δ

Proof. From Lemma 2.2 we have that

 ∞
(2.33) π exp[−Re(η)t]Ĝa (ξ, t) = h(ξ, η) cos[Im(η)t]d[Im(η)]
0
 ∞
1 ∂k(ξ, η)
− {1 − cos[Im(η)t]} d[Im(η)].
t 0 ∂[Im(η)]
 Green’s Functions for Equations with Random Coefficients 169

We consider the first term on the RHS of (2.33). Evidently, for finite N , we have
(2.34)
 N  N
∂
h(ξ, η) cos[Im(η)t]d[Im(η)] = − h(ξ, η) Im(η) sin[Im(η)t]d[Im(η)]
∂t 0 0

1 N ∂
= {h(ξ, η)Im(η)} {1 − cos[Im(η)t]} d[Im(η)]
t 0 ∂[Im(η)]
1
− h(ξ, Re(η) + iN )N {1 − cos[N t]}.
t
It is clear from the inequality (2.24) that
lim h(ξ, Re(η) + iN )N = 0.
N →∞

We have already seen from (2.24) that

 ∞
(2.35) |h(ξ, η)|d[Im(η)] ≤ Cλ,Λ ,
0
for some constant Cλ,Λ depending only on λ, Λ. We shall show now that we also
have
 ∞
∂h(ξ, η)
(2.36)
∂[Im(η)] |Im(η)|d[Im(η)] ≤ Cλ,Λ .
0

To see this we use (2.31) to obtain

∂h(ξ, η) 1 + ψ(ξ, η, ·)ψ(−ξ, η, ·)
(2.37) = Im .
∂[Im(η)] [η + e(ξ)q(ξ, η)e(−ξ)]2
For any complex number a + ib it is clear that
1 −2ab
Im = 2 ,
(a + ib)2 (a + b2 )2
whence
 
1
Im ≤ min 1 , 2|a| .
(a + ib)2 a2 |b|3
We conclude then that
 
1 1 8Λ|e(ξ)|2
Im ≤ min , .
[η + e(ξ)q(ξ, η)e(−ξ)]2 λ2 |e(ξ)|4 |Im(η)|3
It follows that
(2.38)
 ∞  |e(ξ)|2  ∞
1
Im(η) Im d[Im(η)] ≤ + ≤ Cλ,Λ .
[η + e(ξ)q(ξ, η)e(−ξ)]2
0 0 |e(ξ)|2

We also have that

 
ψ(ξ, η, ·)ψ(−ξ, η, ·) Λ Λ|e(ξ)|2
≤ min , .
[η + e(ξ)q(ξ, η)e(−ξ)]2 λ2 |e(ξ)|2 |Im(η)| |Im(η)|3
We conclude that
 ∞
ψ(ξ, η, ·)ψ(−ξ, η, ·)
(2.39)
Im(η) d[Im(η)] ≤ Cλ,Λ .
0 [η + e(ξ)q(ξ, η)e(−ξ)]2
170 Joseph G. Conlon and Ali Naddaf

The inequality (2.36) follows from (2.38), (2.39). It follows from (2.34), (2.35),
(2.36) that the first integral on the RHS of (2.33) is differentiable with respect to t
for t > 0 and
 ∞
∂
(2.40) h(ξ, η) cos[Im(η)t]d[Im(η)t] =
∂t 0
  
1 ∞ ∂
{h(ξ, η) Im(η)} 1 − cos[Im(η)t] d[Im(η)].
t 0 ∂[Im(η)]
Furthermore, there is the inequality,
 ∞
∂
h(ξ, η) cos[Im(η)t]d[Im(η)] ≤ Cλ,Λ /t.
∂t
0

Next we wish to improve this inequality to

 ∞
∂ Cλ,Λ,δ
(2.41) h(ξ, η) cos[Im(η)t]d[Im(η)] ≤ .
∂t t[1 + |e(ξ)|2 t]δ
0

To do this we integrate by parts on the RHS of (2.40) to obtain

 ∞
∂
(2.42) h(ξ, η) cos[Im(η)t]d[Im(η)] =
∂t 0
  
1 ∞ 2∂h(ξ, η) ∂ 2 h(ξ, η)
+ Im(η) sin[Im(η)t]d[Im(η)].
t2 0 ∂[Im(η)] ∂[Im(η)]2
We have already seen in Lemma 2.3 that

1 ∞ ∂h(ξ, η) Cλ,Λ,δ
| sin[Im(η)t]|d[Im(η)] ≤ .
t 0 ∂[Im(η)] [|e(ξ)|2 t]δ
for any δ, 0 ≤ δ < 1, where we assume |e(ξ)|2 t > 1. The inequality (2.41) will
follow therefore if we can show that
(2.43)

1 ∞ ∂ 2 h(ξ, η) Cλ,Λ,δ
|Im(η)| sin[Im(η)t]d[Im(η)] ≤ , |e(ξ)|2 t > 1, 0 ≤ δ < 1.
t 0 ∂[Im(η)]2 [|e(ξ)|2 t]δ
To prove this we use the fact that
2
∂ h(ξ, η) ∂ 2 1
(2.44) ≤
∂[Im(η)]2 ∂η 2 η + e(ξ)q(ξ, η)e(−ξ) ,

∂2 1 ∂ 1 + ψ(−ξ, η, ·)ψ(ξ, η, ·)

=−
∂η 2 η + e(ξ)q(ξ, η)e(−ξ) ∂η [η + e(ξ)q(ξ, η)e(−ξ)]2
2{1 + ψ(−ξ, η, ·)ψ(ξ, η, ·)}2
=
[η + e(ξ)q(ξ, η)e(−ξ)]3
[[∂ψ(−ξ, η, ·)/∂η]ψ(ξ, η, ·) + ψ(−ξ, η, ·)[∂ψ(ξ, η, ·)/∂η]]
− .
[η + e(ξ)q(ξ, η)e(−ξ)]2
Observe now that similarly to (2.27), (2.30) we have that
 
|1 + ψ(−ξ, η, ·)ψ(ξ, η, ·) |2 1 2 2Λ2
(2.45) ≤ min , + 3 .
|η + e(ξ)q(ξ, η)e(−ξ)|3 |Im(η)|3 λ3 |e(ξ)|6 λ |e(ξ)|2 |η|2
 Green’s Functions for Equations with Random Coefficients 171

We can conclude from this last inequality just like we argued in Lemma 2.3 that

1 ∞ |1 + ψ(−ξ, η, ·)ψ(ξ, η, ·) |2 Cλ,Λ,δ
|Im(η)|| sin[Im(η)t]|d[Im(η)] ≤ ,
t 0 |η + e(ξ)q(ξ, η)e(−ξ)|3 [|e(ξ)|2 t]δ
for 0 ≤ δ < 1, |e(ξ)|2 t > 1.
Next from (2.9) we see that ∂ψ(ξ, η, ·)/∂η satisfies the equation,
∂ψ(ξ, η, ·)
(2.46) [Lξ + η] + ψ(ξ, η, ·) = 0.
∂η
From this equation and the Schwarz inequality we easily conclude that
   
|∂ψ(ξ, η, ·)/∂η|2 ≤ |η|−2 |ψ(ξ, η, ·)|2 .
It follows then that
 
| ψ(−ξ, η, ·)[∂ψ(ξ, η, ·)/∂η] |2 1 Λ
(2.47) ≤ min , .
|η + e(ξ)q(ξ, η)e(−ξ)|2 |Im(η)|3 λ2 |e(ξ)|2 |η|2
Since this inequality is similar to (2.45) we conclude that (2.43) holds.
We have proved now that (2.41) holds. To complete the proof of the lemma we
need to obtain a similar estimate for the second integral on the RHS of (2.33). To
see this observe that we can readily conclude that the integral is differentiable in t
and

∂ 1 ∞ ∂k(ξ, η)
(2.48) − {1 − cos[Im(η)t]} d[Im(η)]
∂t t 0 ∂[Im(η)]
 ∞
2 ∂k(ξ, η)
= 2 {1 − cos[Im(η)t]} d[Im(η)]
t 0 ∂[Im(η)]
 ∞ 2
1 ∂ k(ξ, η)
+ 2 Im(η) {1 − cos[Im(η)t]} d[Im(η)].
t 0 ∂[Im(η)]2
We have already seen in Lemma 2.3 that

1 ∞ ∂k(ξ, η) Cλ,Λ,δ
{1 − cos[Im(η)t]}d[Im(η)] ≤ ,
t2 0 ∂[Im(η)] t[1 + |e(ξ)|2 t]δ
for any δ, 0 ≤ δ < 1. Hence we need to concern ourselves with the second integral
on the RHS of (2.48). Now it is clear that ∂ 2 k(ξ, η)/∂[Im(η)]2 satisfies the same
estimates we have just established for ∂ 2 h(ξ, η)/∂[Im(η)]2 . It follows in particular
that

1 ∞ ∂ 2 k(ξ, η)
|Im(η)|{1 − cos[Im(η)t]}d[Im(η)]
t2 0 ∂[Im(η)]2

1 ∞ 1 − cos[Im(η)t] C
≤ 2 2
d[Im(η)] ≤ ,
t 0 [Im(η)] t
for some universal constant C. Arguing as in Lemma 2.3 we also have that

1 ∞ ∂ 2 k(ξ, η)
|Im(η)|{1 − cos[Im(η)t]}d[Im(η)]
t2 0 ∂[Im(η)]2
 1/t  |e(ξ)|2  ∞
Cλ,Λ,δ
≤ + + ≤ 2 t]δ
,
0 1/t |e(ξ)| 2 t[|e(ξ)|
172 Joseph G. Conlon and Ali Naddaf

for 0 ≤ δ < 1, |e(ξ)|2 t > 1. The last three inequalities then give us the same
estimate on the derivative of the second integral on the RHS of (2.33) as we have
already obtained for the first integral.

The estimate for ∂ Ĝa (ξ, t)/∂t in Lemma 2.4 diverges as t → 0. We rectify this
in the following:
Lemma 2.5. There is a constant Cλ,Λ depending only on λ, Λ such that

∂ Ĝ (ξ, t)
a
≤ Cλ,Λ |e(ξ)|2 , ξ ∈ Rd , t > 0.
∂t

Proof. To bound the derivative of the first integral on the RHS of (2.33) it is
sufficient to show that

1 ∞
(2.49) |h(ξ, η)| {1 − cos[Im(η)t]} d[Im(η)] ≤ Cλ,Λ |e(ξ)|2 ,
t 0

1 ∞ ∂h(ξ, η)
|Im(η)| {1 − cos[Im(η)t]} d[Im(η)] ≤ Cλ,Λ |e(ξ)|2 .
t 0 ∂[Im(η)]
We have now from (2.24) that
|h(ξ, η)| ≤ 4Λ|e(ξ)|2 /|Im(η)|2 , Re(η) = 0,
and from the inequalities before (2.39) that
|∂h(ξ, η)/∂[Im(η)]| ≤ 9Λ|e(ξ)|2 /|Im(η)|3 , Re(η) = 0.
The inequalities (2.49) follow from these last two inequalities.
The derivative of the second integral is given by the RHS of (2.48). Using the
identity
  
1 N 2∂k(ξ, η) ∂ 2 k(ξ, η)
+ Im(η) {1 − cos[Im(η)t]}d[Im(η)]
t2 0 ∂[Im(η)] ∂[Im(η)]2
 N
= k(ξ, η) Im(η) cos[Im(η)t]}d[Im(η)]
0
 
∂k(ξ, Re(η) + iN ) {1 − cos[N t]}
+ k(ξ, Re(η) + iN ) + N
∂[Im(η)] t2
N k(ξ, Re(η) + iN ) sin N t
− ,
t
we see that the derivative of the second integral on the RHS of (2.33) is also given
by the formula
 πm/t
(2.50) lim k(ξ, η) Im(η) cos[Im(η)t]d[Im(η)],
m→∞ 0

where the limit in (2.50) is taken for integer m → ∞. Writing η+e(ξ)q(ξ, η)e(−ξ) =
a(η) + ib(η) we see from (2.18) that
 
Im(η)b(η) Im(η) a(η)2
Im(η)k(ξ, η) = = 1− .
a(η)2 + b(η)2 b(η) a(η)2 + b(η)2
 Green’s Functions for Equations with Random Coefficients 173

Observe now that just as before we have

 ∞  ∞
|Im(η)| a(η)2 a(η)2
2 2
d[Im(η)] ≤ d[Im(η)]
0 |b(η)| a(η) + b(η) 0 a(η)2 + b(η)2
 |e(ξ)|2  ∞
≤ + ≤ Cλ,Λ |e(ξ)|2 ,
0 |e(ξ)|2

for a constant Cλ,Λ depending only on λ, Λ. We also have from (2.15) that
 
Im(η) 1  1 
= 1/ 1 + |ψ(ξ, η, ·)|2 + |ψ(−ξ, η, ·)|2
b(η) 2 2
1
 2
 1 
2 |ψ(ξ, η, ·)| + 2 |ψ(−ξ, η, ·)|2
= 1− .
1 + 12 |ψ(ξ, η, ·)|2  + 12 |ψ(−ξ, η, ·)|2 
It follows therefore from (2.50) that the result will be complete if we can show that
 ∞
 
(2.51) |ψ(ξ, η, ·)|2 d[Im(η)] ≤ Cλ,Λ |e(ξ)|2 ,
0

with a similar inequality

 for ψ(−ξ, η, ·). Note that (2.51) does not follow from the
bound |ψ(ξ, η, ·)|2 ≤ Λ|e(ξ)|2 /|η| which we have already established. In view of
(2.8) the inequality (2.51) is a consequence of the following lemma.

Lemma 2.6. Let ϕ(ξ, η, ·) be the function defined by (2.8). Then there is the in-
equality,
 ∞
 
|ϕ(ξ, η, ·)|2 d[Im(η)] ≤ 2πΛ|ρ|2 .
0

Proof. Let ϕ(t, ξ, ·), t > 0, be the solution to the initial value problem,
∂ϕ(t, ξ, ·)
(2.52) + Lξ ϕ(t, ξ, ·) = 0, t > 0,
∂t
d


d

∗ 
ϕ(0, ξ, ·) + ρk ∂j + ej (−ξ) eiej ·ξ [ak,j (·) − ak,j (·)] = 0.
k=1 j=1

It is clear that
 ∞
ϕ(ξ, η, ·) = e−ηt ϕ(t, ξ, ·)dt, Re(η) > 0.
0
Hence the Plancherel Theorem yields
 ∞  ∞
 2
  
|ϕ(ξ, η, ·)| d[Im(η)] ≤ 2π ϕ(t, ξ, ·)|2 dt.
0 0

We can estimate the RHS of this last equation by defining a function Φ(t, ξ, ·), t > 0,
which satisfies the equation,
Lξ Φ(t, ξ, ·) = ϕ(t, ξ, ·), t > 0.
Hence (2.52) yields
 ∂ϕ(t, ξ, ·)   
(2.53) Φ(t, ξ, ·) + |ϕ(t, ξ, ·)|2 = 0, t > 0.
∂t
174 Joseph G. Conlon and Ali Naddaf

Observe next that

 ∂ϕ(t, ξ, ·)  1 ∂  
Φ(t, ξ, ·) = Φ(t, ξ, ·)Lξ Φ(t, ξ, ·) .
∂t 2 ∂t
Integrating (2.53) w.r. to t and using the positivity of Lξ we conclude that for
any δ > 0, there is the inequality,
 ∞
  1 
ϕ(t, ξ, ·)|2 dt ≤ Φ(δ, ξ, ·)Lξ Φ(δ, ξ, ·) .
δ 2
We have now from (2.52) that
 
Φ(0, ξ, ·)Lξ Φ(0, ξ, ·) ≤ Λ|ρ|2 .

The result follows now on letting δ → 0.

3. Proof of Theorem 1.5

For η real and positive, let q(ξ, η) be the d × d matrix defined in (2.3). We shall
show that the function Ga (x) of Theorem 1.5 is given by,

1
(3.1) Ga (x) = lim dξe−ix·ξ /e(ξ)q(ξ, η)e(−ξ), x ∈ Zd .
η→0 (2π)d [−π,π]d

In view of the fact that q(ξ, η) ≥ λId , we see from the following lemma that the
limit (3.1) exists if d ≥ 3.
Lemma 3.1. The limit limη→0 q(ξ, η) exists for all ξ ∈ Rd .

Proof. For η > 0, x ∈ Zd , let Gη (x) be the Green’s function satisfying the equation
d


∇∗i ∇i Gη (x) + ηGη (x) = δ(x), x ∈ Zd ,
i=1

where δ(x) is the Kronecker δ function. Now for ϕ ∈ L2 (Ω) there is a unique
solution ψ ∈ L2 (Ω) to the equation,
d


[∂i∗ + ei (−ξ)] [∂i + ei (ξ)] ψ(ω) + ηψ(ω) = ϕ(ω), ω ∈ Ω ,
i=1

which can be written as

(3.2) ψ(ω) = Gη (x)e−ix·ξ ϕ(τx ω), ω ∈ Ω.
x∈Zd

Observe the RHS of (3.2) is square integrable since Gη (x) decreases exponentially
as |x| → ∞. Now for 1 ≤ k, k  ≤ d we define operators Tk,k
,η,ξ by Tk,k
,η,ξ (ϕ) =
e−iek ·ξ [∂k + ek (ξ)]ψ, where ψ is the solution to the equation,
d


(3.3) [∂i∗ + ei (−ξ)] [∂i + ei (ξ)] ψ(ω) + ηψ(ω)
i=1
= eiek
·ξ [∂k∗
+ ek
(−ξ)] ϕ(ω), ω ∈ Ω.
 Green’s Functions for Equations with Random Coefficients 175

From (3.2) we see that

(3.4) Tk,k
,η,ξ ϕ(ω) = ∇∗k ∇k
Gη (x)e−ix·ξ ϕ(τx ω), ω ∈ Ω.
x∈Zd

Since ∇∗k ∇k
Gη (x) is exponentially decreasing as |x| → ∞ it follows that Tk,k
,η,ξ
is a bounded operator on L2 (Ω). Observe that the projection operator P on L2 (Ω)
orthogonal to the constant function commutes with Tk,k
,η,ξ . It follows from (3.3)
that Tk,k
,η,ξ  ≤ 1, independent of η as η → 0. We wish to show that there is an
operator Tk,k
,0,ξ on L2 (Ω) with Tk,k
,0,ξ  ≤ 1 such that
(3.5) lim Tk,k
,η,ξ ϕ − Tk,k
,0,ξ ϕ = 0, ϕ ∈ L2 (Ω).
η→0

We follow the argument used to prove the von Neumann Ergodic Theorem [9].
Thus if ϕ ∈ L2 (Ω) satisfies [∂k∗
+ ek
(−ξ)]ϕ = 0, then Tk,k
,η,ξ ϕ = 0. Thus we set
Tk,k
,0,ξ ϕ = 0 for ϕ in the null space of [∂k∗
+ek
(−ξ)]. Now the range of [∂k
+ek
(ξ)]
is dense in the subspace of L2 (Ω) orthogonal to the null space of [∂k∗
+ ek
(−ξ)]. If
ϕ = e−iek
·ξ [∂k
+ ek
(ξ)]ψ with ψ ∈ L2 (Ω) then


Tk,k
,η,ξ ϕ(ω) = ∇∗k
∇∗k ∇k
Gη (x)e−ix·ξ ψ(τx ω), ω ∈ Ω.
x∈Zd

It is clear from this representation that if we take

Tk,k
,0,ξ ϕ(ω) = ∇∗k
∇∗k ∇k
G0 (x)e−ix·ξ ψ(τx ω), ω ∈ Ω ,
x∈Zd

then Tk,k
,0,ξ (ϕ) ∈ L2 (Ω) and (3.5) holds. Thus Tk,k
,0,ξ is defined on a dense
subspace of L2 (Ω) and Tk,k
,0,ξ  ≤ 1. If follows easily that one can extend the
definition of Tk,k
,0,ξ to all of L2 (Ω) and (3.5) holds.
Suppose now b : Ω → Rd(d+1)/2 is a bounded measurable function from Ω to the
space of symmetric d × d matrices. We define b to be

d d


b = sup | bi,j (ω)λi λj | : λ2i = 1, ω ∈ Ω .
i,j=1 i=1

Next let H(Ω) = {ϕ = (ϕ1 , . . . , ϕd ) : ϕi ∈ L2 (Ω), 1 ≤ i ≤ d} be the Hilbert space

with norm ϕ2 = ϕ1 2 + · · · + ϕd 2 , ϕ = (ϕ1 , . . . , ϕd ). We define an operator
Tb,η,ξ on H(Ω) by
d





Tb,η,ξ ϕ(·) k = Tk,i,η,ξ bi,j (·)ϕj (·) , 1 ≤ k ≤ d.
i,j=1

Evidently,


Tb,η,ξ ϕ(·) k = e−iek ·ξ [∂k + ek (ξ)]ψ(·), 1 ≤ k ≤ d,
where ψ(·) satisfies the equation,
d

d



∗ 


∂r + er (−ξ) ∂r + er (ξ) ψ(ω) + ηψ(ω) = eiei ·ξ [∂i∗ + ei (−ξ)] bi,j (·)ϕj (·) .
r=1 i,j=1
176 Joseph G. Conlon and Ali Naddaf

If follows that Tb,η,ξ is a bounded operator on H(Ω) with norm Tb,η,ξ  ≤ b. In
view of (3.5) there exists a bounded operator Tb,0,ξ on H(Ω) such that Tb,0,ξ  ≤
b and
lim Tb,η,ξ ϕ − Tb,0,ξ ϕ = 0, ϕ ∈ H(Ω) .
η→0


Let us take now b(·) = ΛId − a(·) /Λ, whence b < 1. Let ψk (ξ, η, ·) be the
function satisfying (2.2). Define Ψk (ξ, η, ·) to be
 
Ψk (ξ, η, ·) = e−ie1 ·ξ [∂1 + e1 (ξ)]ψk (ξ, η, ·), . . . , e−ied ·ξ [∂d + ed (ξ)]ψk (ξ, η, ·) .
Then Ψk (ξ, η, ·) ∈ H(Ω) and satisfies the equation,
d
1

 
(3.6) Ψk (ξ, η, ω) − P Tb,η/Λ,ξ Ψk (ξ, η, ω) + Tj,η/Λ,ξ ak,j (ω) − ak,j (·) = 0,
Λ j=1

where for 1 ≤ j ≤ d, Tj,η,ξ is a bounded operator from L2 (Ω) to H(Ω) defined by

Tj,η,ξ (ϕ) = (T1,j,η,ξ ϕ, . . . , Td,j,η,ξ ϕ). Writing
Ψk (ξ, η, ·) = (Ψk,1 (ξ, η, ·), . . . , Ψk,d (ξ, η, ·)),
we have from (2.3) that
 d


(3.7) q k,k
(ξ, η) = a k,k
(·) + ak,j (·)Ψk
,j (ξ, η, ·) .
j=1

It is clear now that limη→0 q(ξ, η) exists.

Our next goal is to assert some differentiability properties of q(ξ, η) in ξ which
are uniform as η → 0.
Lemma 3.2. Suppose η > 0, 1 ≤ k ≤ d, ϕ ∈ L2 (Ω) and b(·) a random symmetric
matrix satisfying b < 1. For ξ ∈ Rd let Ψ(ξ, η, ·) be the solution to the equation,
(3.8) (I − P Tb,η,ξ )Ψ(ξ, η, ·) = Tk,η,ξ ϕ(·).
Then Ψ(ξ, η, ·), regarded as a function of ξ ∈ Rd to H(Ω) is differentiable. The
derivative of Ψ(ξ, η, ·) is given by
(3.9)
 
∂Ψ ∂
(ξ, ·) = (I − P Tb,η,ξ )−1 Tk,η,ξ ϕ(·)
∂ξj ∂ξj
 
∂
+ (I − P Tb,η,ξ )−1 P Tb,η,ξ (I − P Tb,η,ξ )−1 Tk,η,ξ ϕ(·), 1 ≤ j ≤ d.
∂ξj

Proof. Consider first the k  th component of Tk,η,ξ ϕ(·), which is

Tk
,k,η,ξ ϕ(·) = ∇∗k
∇k Gη (x)e−ix·ξ ϕ(τx ·).
x∈Zd

Since Gη (x) decreases exponentially as |x| → ∞ it is clear that Tk
,k,η,ξ ϕ(·), re-
garded as a mapping from Rd to L2 (Ω) is differentiable and
∂

(3.10) Tk
,k,η,ξ ϕ(·) = − ixj ∇∗k
∇k Gη (x)eix·ξ ϕ(τx ·).
∂ξj d x∈Z
 Green’s Functions for Equations with Random Coefficients 177

We regard (3.10) as the definition of the operator ∂/∂ξj Tk
,k,η,ξ which is clearly a
bounded operator on L2 (Ω). Similarly we can define ∂/∂ξj Tb,η,ξ by
 
d  
∂ ∂
(3.11) Tb,η,ξ ψ(·) = Tr,i,η,ξ bi,j (·)ψj (·) ,



∂ξj r

∂ξj
i,j =1

ψ = (ψ1 , . . . , ψd ) ∈ H(Ω), 1 ≤ r ≤ d. Again it is clear that ∂/∂ξj Tb,η,ξ is a

bounded operator on H(Ω). From (3.8) Ψ(ξ, η, ·) is given by the Neumann series,
∞


Ψ(ξ, η, ·) = (P Tb,η,ξ )n Tk,η,ξ ϕ(·),
n=0

which converges since b < 1. Formally the derivative of Ψ(ξ, η, ·) is given by

∞
∂ ∂
(3.12) Ψ(ξ, η, ·) = (P Tb,η,ξ )n Tk,η,ξ ϕ(·)
∂ξj n=0
∂ξ j
∞

∂
+ (P Tb,η,ξ )n (P Tb,η,ξ )(P Tb,η,ξ )m Tk,η,ξ ϕ(·).
n,m=0
∂ξj

Since the RHS of (3.12) converges it is easy to see that Ψ(ξ, η, ·), regarded as a
mapping from Rd to H(Ω) is differentiable and the derivative is given by (3.12).
Finally observe that the RHS of (3.12) is the same as the RHS of (3.9).


d
For 2 ≤ p < ∞ let Lp (Ω, − π, π ) be the space of functions ψ(ξ, ω), ξ ∈

d
− π, π , ω ∈ Ω such that ψp < ∞, where

 p/2
ψpp = dξ |ψ(ξ, ·)|2 .
[−π,π]d

d
Suppose now f : − π, π → C is a smooth periodic function. The Fourier
transform f of f is given by

 1
f (x) = f (ξ)e−ix·ξ dξ, x ∈ Zd .
(2π)d [−π,π]d

Since f is rapidly decreasing we can define for ϕ ∈ L2 (Ω) an operator Tϕ by

d
(3.13) Tϕ (f )(ξ, ω) = f(x)e−ix·ξ ϕ(τx ω), ξ ∈ − π, π , ω ∈ Ω .
x∈Zd

d
Evidently Tϕ (f ) ∈ L∞ (Ω, − π, π ).
Lemma 3.3. Suppose 2 ≤ p ≤ ∞ and ϕ ∈ L2 (Ω). Then the operator Tϕ extends

d
to a bounded operator from Lp ( − π, π ) to Lp (Ω, [−π, π]d )and the norm of Tϕ
satisfies Tϕ  ≤ ϕ.

Proof. Now by Bochner’s Theorem [9] there is a positive measure dµϕ on [−π, π]d
such that
  
ϕ(τx ·)ϕ(τy ·) = ei(x−y)·ζ dµϕ (ζ),
[−π,π]d
178 Joseph G. Conlon and Ali Naddaf

whence

 2

(3.14) |Tϕ (f )(ξ, ·)| = |f (ζ − ξ)|2 dµϕ (ζ).
[−π,π]d

Since [−π,π]d dµϕ (ζ) = ϕ2 the fact that Tϕ  ≤ ϕ follows immediately from
(3.14) for p = 2, p = ∞. The fact that Tϕ  ≤ ϕ when 2 < p < ∞ can be
obtained by an application of Holder’s inequality.

Next for 2 ≤ p < ∞ let Hp (Ω, [−π, π]d ) be the space of functions ψ(ξ, ω), ξ ∈
[−π, π]d with ψ(ξ, ·) ∈ H(Ω) such that ψp < ∞, where

ψpp = dξψ(ξ, ·)p .
[−π,π]d

We define an operator for ϕ ∈ L2 (Ω), Tϕ,b,η by

 
(3.15) Tϕ,b,η (f )(ξ, ·) = f(x)e−ix·ξ τx b(·)(I − P Tb,η,ξ )−1 Tk,η,ξ ϕ(·) .
x∈Zd

It is clear that if f is smooth then Tϕ,b,η (f ) is in H∞ (Ω, [−π, π]d ) provided b < 1.
Lemma 3.4. Suppose b < 1. Then,
(a) If ϕ ∈ L2 (Ω), Tϕ,b,η extends to a bounded operator from L∞ ([−π, π]d ) to
H∞ (Ω, [−π, π]d ). The norm of Tϕ,b,η satisfies
bϕ
Tϕ,b,η  ≤ .
(1 − b)
(b) If ϕ ∈ L∞ (Ω), Tϕ,b,η extends to a bounded operator from L2 ([−π, π]d ) to
H2 (Ω, [−π, π]d ). The norm of Tϕ,b,η satisfies
√
dbϕ∞
Tϕ,b,η  ≤ .
(1 − b)2

Proof. To prove (a) observe that (3.14) implies

Tϕ,b,η (f )(ξ)2 ≤ f 2∞ b(·)(I − P Tb,η,ξ )−1 Tk,η,ξ ϕ(·))2
f 2∞ b2 ϕ2
≤ .
(1 − b)2
To prove (b) we consider the integral
(3.16)



dξ f(x)e−ix·ξ τx [b(·)(P Tb,η,ξ )m Tk,η,ξ ϕ] 2 = (2π)d Θ1 2
[−π,π]d
x∈Zd r∈Zd


= (2π)d Θ2 2 ,
r∈Zd

where Θ1 is given by

 m+1
! 
f(x1 )b(τx1 ·) ∇∗ ∇Gη (xj )P b(τx1 +···+xj ·) ∇∗ ∇k Gη (xm+2 )ϕ(τr ·)
x1 +···+xm+2 =r j=2
 Green’s Functions for Equations with Random Coefficients 179

and Θ2 is given by

 m+1
! 
f(y1 )b(τy1 ·) ∇∗ ∇Gη (yj − yj−1 )P b(τyj ·) ∇∗ ∇k Gη (r − ym+1 )ϕ(τr ·).
y1 ,...,ym+1 j=2

Observe now that

 m+1
! 

(3.17) f (y1 )b(τy1 ·) ∗
∇ ∇Gη (yj − yj−1 )P b(τyj ·) ∇∗ ∇k Gη (r − ym+1 )ϕ(τr ·)
j=2
 m+1
! 

= f (y1 )b(τy1 ·) ∗
∇ ∇Gη (yj − yj−1 )b(τyj ·) ∇∗ ∇k Gη (r − ym+1 )ϕ(τr ·)
j=2
m

!
n 
− f(y1 )b(τy1 ·) ∇∗ ∇Gη (yj − yj−1 )b(τyj ·)
n=1 j=2
  m+1
! 
∇∗ ∇Gη (yn+1 − yn )b(τyn ·) ∇∗ ∇Gη (yj − yj−1 )P b(τyj ·)
j=n+2

∗
∇ ∇k Gη (r − ym+1 )ϕ(τr ·) .

Next let M be the space of complex d × d matrices and L2 (Zd , M) the set of
functions A : Zd → M. We can make L2 (Zd , M) into a Hilbert space by defining
the norm of A to be


A2M = (2π)d T r(A∗ (x)A(x)).
x∈Zd

We can also define an operator Tη on L2 (Zd , M)by

Tη A(x) = A(y)∇∗ ∇Gη (x − y), x ∈ Zd .
y

It is easy to see that Tη is bounded on L2 (Zd , M) and Tη  ≤ 1. For n = 1, . . . , m+1,

ω ∈ Ω, yn ∈ Zd , let us define An (yn , ω) ∈ M by

!n 
An (yn , ω) = 
f (y1 )b(τy1 ω) ∗
∇ ∇Gη (yj − yj−1 )b(τyj ω) .
y1 ,...yn−1 j=2

It follows from the fact that Tη  ≤ 1 that for any fixed ω ∈ Ω the function
An (·, ω) ∈ L2 (Zd , M) and
√
(3.18) An (·, ω)M ≤ bn fId M = dbn f 2 .
Recall now that P is the projection operator, P ψ(·) = ψ(·) − ψ , ψ ∈ L2 (Ω). If
we introduce the notation P ∗ as ψ(·)P ∗ = P ψ(·) then
  m+1
! 
∇∗ ∇Gη (yn+1 − yn )b(τyn ·) ∇∗ ∇Gη (yj − yj−1 )P b(τyj ·)
j=n+2

∗
∇ ∇k Gη (r − ym+1 )ϕ(τr ·)
180 Joseph G. Conlon and Ali Naddaf

is equal to
  m+1
! 
∇∗ ∇Gη (yn+1 − yn )b(τyn ·) ∇∗ ∇Gη (yj − yj−1 )P ∗ b(τyj ·)
j=n+2

∇∗ ∇k Gη (r − ym+1 )ϕ(τr ·) .

Let us denote now by L2 (Zd × Ω, M) the set of functions A : Zd × Ω → M with

norm,


A2M,ran = (2π)d T r(A∗ (x, ·)A(x, ·)) .
x∈Zd

For n = 1, . . . , m define an operator Tn on this space by

Tn A(ym+1 , ·) = A(yn , ·)∇∗ ∇Gη (yn+1 − yn )
yn ,...,ym
 m+1
! 
b(τyn ·) ∇∗ ∇Gη (yj − yj−1 )P ∗ b(τyj ·) .
j=n+2

We can see as before that Tn is a bounded operator on L2 (Zd × Ω, M) and

(3.19) Tn  ≤ bm+1−n .
Observe now from (3.17) that Θ2 (the expression inside the norm in the last line
of (3.16)) is given by

 m+1
! 

f (y1 )b(τy1 ·) ∇ ∇Gη (yj −yj−1 )P b(τyj ·) ∇∗ ∇k Gη (r−ym+1 )ϕ(τr ·)
∗

y1 ,...,ym+1 j=2
m


 
= [Tη Am+1 (r, ·)ek ] ϕ(τr ·) − Tη Tn Aran
n (r, ·)ek ϕ(τr ·) ,
n=1

where Aran
n denotes that An (x, ω), x ∈ Zd , ω ∈ Ω is to be regarded as a function
of x only, with parameter ω, on which Tn acts. We have now

 
(2π)d  [Tη Am+1 (r, ·)ek ] ϕ(τr ·)2 ≤ ϕ2∞ Am+1 (·, ω)2M
r∈Zd

≤ ϕ2∞ db2(m+1) f 22 ,

where we have used (3.18). Similarly we have


   
(2π)d  Tη Tn Aran
n (r, ·)ek ϕ(τr ·) 2 ≤ ϕ22 Tη Tn Aran 2
n (ω)M,ran ,
r∈Zd

where ω denotes the random parameter for Aran n . The expectation is then to be
taken with respect to this parameter. If we use now (3.18), (3.19) we have


  2
(2π)d  Tη Tn Aran 2
n (r, ·)ek ϕ(τr ·)  ≤ ϕ2 db
2(m+1)
f 22 .
r∈Zd
 Green’s Functions for Equations with Random Coefficients 181

We conclude therefore from (3.16) that



dξ f(x)eix·ξ τx b(·)(P Tb,η,ξ )m Tk,η,ξ ϕ 2

[−π,π]d
x∈Zd

≤ d(m + 1) ϕ2∞ b2(m+1) f 22 , m ≥ 0.

It follows that
 1/2 ∞


√ √
dξTϕ,b,η (f )(ξ)2 ≤ dϕ∞ f 2 m + 1 bm+1
[−π,π]d m=0
√
≤ dϕ∞ f 2 b/(1 − b)2 .


It follows from Lemma 3.3 and the Riesz-Thorin Interpolation Theorem [10] that
if ϕ ∈ L∞ (Ω) and b < 1 then Tϕ,b,η is a bounded operator from Lp ([−π, π]d ) to
Hp (Ω, [−π, π]d ), 2 ≤ p ≤ ∞ and the norm of Tϕ,b,η satisfies the inequality,
Cd bϕ∞
Tϕ,b,η  ≤ ,
(1 − b)2
where Cd is a constant depending only on d. Consider next the weak spaces
Lpw ([−π, π]d ) and Hwp
(Ω, [−π, π]d ), 2 < p < ∞. Thus f ∈ Lpw ([−π, π]d ) if for
all α > 0 there is the inequality,
(3.20) meas{ξ ∈ [−π, π]d : |f (ξ)| > α} ≤ C p /αp .
The weak Lp norm of f, f p,w is then the minimum constant C such that (3.20)
p
holds for all α > 0. Similarly ψ(ξ, ω) ∈ Hw (Ω, [−π, π]d ) if for all α > 0 there is the
inequality,
(3.21) meas{ξ ∈ [−π, π]d : ψ(ξ, ·) > α} ≤ C p /αp .
The weak Lp norm of ψ, ψp,w is again the minimum constant C such that (3.21)
holds. Lemmas 3.3, 3.4 and Hunt’s Interpolation Theorem [10] then imply the
following:
Lemma 3.5. Suppose 2 < p < ∞. Then
(a) There is a constant Cp depending only on p such that Tϕ is a bounded operator
from Lpw ([−π, π]d ) to Lpw (Ω, [−π, π]d ) and Tϕ  ≤ Cp ϕ.
(b) There is a constant Cp,d depending only on p and d such that Tϕ,b,η is a
bounded operator from Lpw ([−π, π]d ) to Hw p
(Ω, [−π, π]d ) and
Cp,d bϕ∞
Tϕ,b,η  ≤ .
(1 − b)2
We can use Lemma 3.5 to obtain bounds on the first two derivatives with respect
to ξ of the function q(ξ, η) defined by (3.7).
Lemma 3.6. Let d = 3, η > 0, 1 ≤ k, k  ≤ d. Then qk,k
(ξ, η) is a C ∞ function
of ξ ∈ Rd and for any i, j, 1 ≤ i, j ≤ d the function ∂qk,k
/∂ξi ∈ L3w ([−π, π]d ) and
3/2
∂ 2 qk,k
/∂ξi ∂ξj ∈ Lw ([−π, π]d ). Further, there is a constant Cλ,Λ , depending only
on λ, Λ such that
∂qk,k
/∂ξi 3,w ≤ Cλ,Λ , ∂ 2 qk,k
/∂ξi ∂ξj 3/2,w ≤ Cλ,Λ .
182 Joseph G. Conlon and Ali Naddaf

Proof. Observe that the function Ψk
occurring in (3.7) satisfies (3.6) and hence
corresponds to the function Ψ of Lemma 3.2 with ϕ ∈ L∞ (Ω). Observe next from
(3.10) that
∂
Tk
,k,η,ξ ϕ = Tϕ (f ),
∂ξi

where Tϕ is defined by (3.13) and the function f is

∂ −iek ·ξ  η (ξ)].
(3.22) f (ξ) = [e − 1)(eiek
·ξ − 1)G
∂ξi

Clearly f ∈ L3w ([−π, π]d ) and f 3,w ≤ C, where C is universal. Consider now the
formula (3.9) for the derivative of Ψ. For the first term on the RHS of (3.9) we
have

∂ 1 ∂
(I − P Tb,η,ξ )−1 ( Tk,η,ξ ) ϕ(·) ≤  Tk,η,ξ ϕ(·).
∂ξi 1 − b ∂ξi

It follows from Lemma 3.5 (a) that

∂
Tk,η,ξ ϕ(·) ∈ L3w (Ω, [−π, π]3 ) ,
∂ξi

∂
 Tk,η,ξ ϕ(·)3,w ≤ CΛ ,
∂ξi

for some universal constant C, since ϕ is bounded by a constant times Λ. Hence
the first term on the RHS of (3.9) is in L3w (Ω, [−π, π]3 ) with norm bounded by a
constant depending only on λ, Λ. Similarly the second term on the RHS of (3.9) is
bounded by
 
1 ∂
 Tb,η,ξ (I − P Tb,η,ξ )−1 Tk,η,ξ ϕ .
1 − b ∂ξi

It follows from (3.11),(3.15) that

 
∂
Tb,η,ξ (I − P Tb,η,ξ )−1 Tk,η,ξ ϕ = Tϕ,b,η (f ),
∂ξi

where Tϕ,b,η is like the operator (3.15) but acts on matrix valued functions f (ξ) =
[fi,j (ξ)], ξ ∈ [−π, π]d . The functions fi,j (ξ) are similar to (3.22) and hence are in
L3w ([−π, π]3 ). It follows by the argument of Lemma 3.4 and Lemma 3.5 (b) that
the second term on the RHS of (3.9) is in L3w (Ω, [−π, π]3 ) with norm bounded by a
constant depending only on λ, Λ. We conclude that ∂qk,k
/∂ξi ∈ L3w ([−π, π]d ) with
norm bounded by a constant depending only on λ, Λ.
Next we turn to the second derivative, ∂ 2 qk,k
/∂ξi ∂ξj . To estimate this we need
a formula for the second derivative of the function Ψ of Lemma 3.2. One can see
 Green’s Functions for Equations with Random Coefficients 183

from (3.9) that

∂2Ψ  −1  ∂ 
(ξ, ·) = I − P Tb,η,ξ P Tb,η,ξ
∂ξi ∂ξj ∂ξj
 −1  ∂   −1  ∂ 2 
I − P Tb,η,ξ Tk,η,ξ ϕ(·) + I − P Tb,η,ξ Tk,η,ξ ϕ(·)
∂ξi ∂ξi ∂ξj
 −1  ∂  −1  ∂ 
+ I − P Tb,η,ξ P Tb,η,ξ I − P Tb,η,ξ Tk,η,ξ ϕ(·)
∂ξi ∂ξj
 −1  ∂  −1  ∂  −1
+ I−P Tb,η,ξ P Tb,η,ξ I−P Tb,η,ξ P Tb,η,ξ I−P Tb,η,ξ Tk,η,ξ ϕ(·)
∂ξj ∂ξi
 −1  ∂2  −1
+ I − P Tb,η,ξ P Tb,η,ξ I − P Tb,η,ξ Tk,η,ξ ϕ(·).
∂ξi ∂ξj
 
Let ϕ ∈ H∞ (Ω) and consider the expectation value ϕ (·)∂ 2 Ψ(ξ, η, ·)/∂ξi ∂ξj .
From above this is a sum of five terms. The first term is given by
 ∂ ∂ 
ϕ (·)(I − P Tb,η,ξ )−1 (P Tb,η,ξ )(I − P Tb,η,ξ )−1 ( Tk,η,ξ )ϕ(·)
∂ξj ∂ξi
 ∂  ∂ 
= (P Tb,η,ξ )(I − P Tb,η,ξ )−1 ϕ (·) (I − P Tb,η,ξ )−1 ( Tk,η,ξ )ϕ(·) ,
∂ξj ∂ξi
whence we have
 ∂ ∂ 

ϕ (·)(I − P Tb,η,ξ )−1 (P Tb,η,ξ )(I − P Tb,η,ξ )−1 ( Tk,η,ξ )ϕ(·)
∂ξj ∂ξi
∂ ∂
≤ (P Tb,η,ξ )(I − P Tb,η,ξ )−1 ϕ (·) ( Tk,η,ξ )ϕ(·)/(1 − b).
∂ξj ∂ξi
We have already seen that both
 
∂ ∂
 P Tb,η,ξ (I − P Tb,η,ξ )−1 ϕ (·), ( Tk,η,ξ ) ϕ(·)
∂ξj ∂ξi
 
are in L3w ([−π, π]3 ). We conclude that the first term in ϕ (·)∂ 2 Ψ(ξ, η, ·)/∂ξi ∂ξj
3/2
is in Lw ([−π, π]3 ) with norm depending only on λ, Λ. Consider now the second
term. To estimate this observe that if Tϕ is given by (3.13) then


(3.23) Tϕ (f g)(ξ, ω) = f(x)eix·ξ Tϕ (g)(τx ω) .
x∈Zd

We have from (3.22) that

∂2
Tk
,k,η,ξ ϕ = Tϕ (f g),
∂ξi ∂ξj
3/2
where f g ∈ Lw ([−π, π]3 ) , whence we can choose f, g so that f, g ∈ L3w ([−π, π]3 ).
With this choice of f, g and usingthe formula (3.23) we can argue as for the first
term of ϕ (·)∂ 2 Ψ(ξ, η, ·)/∂ξi ∂ξj to conclude that the second term is also in
3/2
Lw ([−π, π]3 )with norm depending only  on λ, Λ. One can estimate the other
three terms of ϕ (·)∂ 2 Ψ(ξ, η, ·)/∂ξi ∂ξj similarly.

184 Joseph G. Conlon and Ali Naddaf

Proposition 3.1. Let d = 3. Then the function Ga (x) defined by (3.1) satisfies
the inequalities
(3.24) 0 ≤ Ga (x) ≤ Cλ,Λ [1 + |x|]2−d , x ∈ Zd ,
(3.25) |∇Ga (x)| ≤ Cλ,Λ log [2 + |x|]/(1 + |x|)2 , x ∈ Zd .

The constant Cλ,Λ depends only on λ, Λ.

Proof. For η > 0 let Ga,η (x) be defined by


1
Ga,η (x) = dξe−ix·ξ /e(ξ)q(ξ, η)e(−ξ), x ∈ Zd .
(2π)d [−π,π]d

Then by Lemma 3.1 it follows that limη→0 Ga,η (x) = Ga (x). It will be sufficient
therefore for us to obtain bounds on Ga,η (x), ∇Ga,η (x) which are uniform as η → 0.
Since q(ξ, η) ≥ λId , ξ ∈ [−π, π]d , we clearly have Ga,η (x) ≤ Cd , where Cd is a
constant depending only on d ≥ 3. To obtain the decay in (3.24) we write
 
(3.26) Ga,η (x) = + ,
|ξ|<γ/|x| |ξ|>γ/|x|

where γ is a parameter, 1 ≤ γ ≤ 2. Evidently there is a constant C such that


dξ ≤ C/|x|.
|ξ|<γ/|x|

To bound the second integral in (3.26) we integrate by parts. Thus

   
1 1 dξ ∂ −ix·ξ
(3.27) = − e
|ξ|>γ/|x| ix1 (2π)d |ξ|>γ/|x| e(ξ)q(ξ, η)e(−ξ) ∂ξ1
 −ix·ξ
−1 1 e ∂
= dξ [e(ξ)q(ξ, η)e(−ξ)]
ix1 (2π)d |ξ|>γ/|x| [e(ξ)q(ξ, η)e(−ξ)]2 ∂ξ1

1 1 e−ix·ξ ξ1
+ dξ ,
ix1 (2π)d |ξ|=γ/|x| [e(ξ)q(ξ, η)e(−ξ)]|ξ|

where we have assumed wlog that |x1 | = max[|x1 |, . . . , |xd |]. Evidently the surface
integral on the RHS of the last expression is bounded by C/|x|. We estimate the
volume integral by integrating by parts again. Thus,
(3.28)

e−ix·ξ ∂

dξ e(ξ)q(ξ, η)e(−ξ)
|ξ|>γ/|x| [e(ξ)q(ξ, η)e(−ξ)]2 ∂ξ1
  
1 1 ∂
 ∂ −ix·ξ
= dξ e(ξ)q(ξ, η)e(−ξ) − e
ix1 |ξ|>γ/|x| [e(ξ)q(ξ, η)e(−ξ)]2 ∂ξ1 ∂ξ1
  
1 ∂ 1 ∂

= dξe−ix·ξ e(ξ)q(ξ, η)e(−ξ)
ix1 |ξ|>γ/|x| ∂ξ1 [e(ξ)q(ξ, η)e(−ξ)]2 ∂ξ1

1 ξ1 ∂

+ dξe−ix·ξ 2
e(ξ)q(ξ, η)e(−ξ) .
ix1 |ξ|=γ/|x| [e(ξ)q(ξ, η)e(−ξ)] |ξ| ∂ξ1
 Green’s Functions for Equations with Random Coefficients 185

We divide the surface integral in the last expression into three parts corresponding
to the three terms in the expression,

∂
 ∂e(ξ) ∂q
e(ξ)q(ξ, η)e(−ξ) = q(ξ, η)e(−ξ) + e(ξ) (ξ, η)e(−ξ)
∂ξ1 ∂ξ1 ∂ξ1
∂e(−ξ)
+ e(ξ)q(ξ, η) .
∂ξ1

The surface integral corresponding to the first and third terms in this expansion is
bounded by

C Λdξ
≤ C  Λ/λ2 ,
|x| |ξ|=γ/|x| λ2 |ξ|3

where C, C  are universal constants. The surface integral corresponding to the

middle term is bounded by

C 1
(3.29) ∂q(ξ, η)/∂ξ1 dξ,
|x| |ξ|=γ/|x| λ |ξ|2
2

for some universal constant C. To bound this we use the well known fact that if
f ∈ Lpw ([−π, π]3 ), 1 < p < ∞, then for any measurable set E one has

(3.30) |f |dξ ≤ Cp f p,w m(E)1−1/p ,
E

where the constant Cp depends only on p. If we average the expression (3.29) over
γ, 1 < γ < 2, then we have from Lemma 3.6 that
 2 
C 1
dγ ∂q(ξ, η)/∂ξ1 dξ
1 |x| |ξ|=γ/|x| λ2 |ξ|2

C  |x|2
≤ dξ∂q(ξ, η)/∂ξ1  ≤ Cλ,Λ ,
λ2 |x|−1 <|ξ|<2|x|−1

where we have used (3.30) with p = 3.

Next we consider the volume integral on the RHS of (3.28). For any γ, 1 ≤ γ ≤ 2,
this is bounded by
 
C(λ, Λ)
(3.31) dξ |ξ|−4 + |ξ|−3 ∂q(ξ, η)/∂ξ1 
|x| |ξ|>1/|x|

+ |ξ|−2 [∂ 2 q(ξ, η)/∂ξ12  + ∂q(ξ, η)/∂ξ1 2 ] ,

for some constant C(λ, Λ) depending only on λ, Λ. Evidently there is a constant

C  (λ, Λ) such that

C(λ, Λ) dξ
4
≤ C  (λ, Λ).
|x| |ξ|>1/|x| |ξ|
186 Joseph G. Conlon and Ali Naddaf

Also we have

C(λ, Λ) dξ
3
∂q(ξ, η)/∂ξ1 
|x| |ξ|>1/|x| |ξ|
∞ 
C(λ, Λ)

= dξ
|x| n=0 2n+1 /|x|>|ξ|>2n /|x|
∞


2 −3n
≤ C(λ, Λ)|x| 2 ∂q(ξ, η)/∂ξ1  dξ
2 n+1 /|x|>|ξ|>2n /|x|
n=0
∞

≤ C  (λ, Λ)|x|
2 −3n n
2 2 /|x|]2 ≤ C  (λ, Λ) ,
n=0

where we have used Lemma 3.6 and (3.30). We can similarly bound the third
term in (3.31) using Lemma 3.6 and (3.30). We conclude that (3.31) is bounded
by a constant depending only on λ, Λ. If we put this inequality together with the
previous inequalities we obtain (3.24).
The proof of (3.25) is similar. For any unit vector n ∈ Rd we have

1
(3.32) n.∇Ga,η (x) = dξe−ix·ξ [−n · e(−ξ)]/e(ξ)q(ξ, η)e(−ξ) .
(2π)d [−π,π]d

We do a decomposition similar to (3.26) and it is clear that


≤ C/|x|2 .
|ξ|<γ/|x|

For the {|ξ| > γ/|x|} integral we do a decomposition analogous to (3.27). It is clear
the surface integral which appears is bounded by C/|x|2 . For the other integral we
do a separate integration by parts as in (3.28). The average of the corresponding
surface integral over γ, 1 ≤ γ ≤ 2, is bounded by C(λ, Λ)/|x|. The volume integral
is bounded analogously to (3.31) by

C(λ, Λ) 
dξ |ξ|−3 + |ξ|−2 ∂q(ξ, η)/∂ξ1 
|x| |ξ|>1/|x|

+ |ξ|−1 [∂ 2 q(ξ, η)/∂ξ12  + ∂q(ξ, η)/∂ξ1 2 ] .

Arguing as before we see this is bounded by C  (λ, Λ)log[1 + |x|]/|x|. Putting this
inequality together with the previous inequalities yields (3.25)

Proposition 3.1 gives an improvement of the estimate (1.14) when d = 3. We

wish now to obtain a corresponding improvement for all d ≥ 3. To do this we need
generalizations of Lemmas 3.4–3.6 appropriate for all d ≥ 3. Let A ∈ M, the space
of complex d × d matrices. The norm of A is defined to be

A2 = T r(A∗ A).

Similarly if A : Ω → M is a random function we define A(·) by

A(·)2 = T r(A∗ (·)A(·)) .

 Green’s Functions for Equations with Random Coefficients 187

Consider now functions A : [−π, π]d → M. For 2 ≤ p ≤ ∞ we can define the space
Lp ([−π, π]d , M) with norm Ap given by

App = A(ξ)p dξ.
[−π,π]d

Similarly we can consider functions A : [−π, π]d × Ω → M and associate with them
spaces Lp ([−π, π]d × Ω, M) with norm Ap given by

p
Ap = A(ξ, ·)p dξ.
[−π,π]d

We define an operator which generalizes the operator given in (3.15). For n =

1, 2, . . . the operator Tn,ϕ,b,η acts on n functions Ar : [−π, π]d → M, 1 ≤ r ≤ n.
The resulting quantity Tn,ϕ,b,η (A1 , . . . An ) is a function from [−π, π]d × Ω → M.
Specifically we define

(3.33) Tn,ϕ,b,η (A1 , . . . , An )(ξ, ·) =

!
n−1  
Ar (xr )e−ixr ·ξ τx P b(·)(I − P Tb,η,ξ )−1 A n (xn )e−ixn ·ξ τx ϕ(·).
r n

x1 ,...,xn ∈Zd r=1

Evidently the operators (3.13),(3.15) correspond to the cases n = 1 and n = 2 in

(3.33).

Lemma 3.7. Suppose ∞ ≥ p1 , . . . , pn , p ≥ 2 and p11 + · · · + p1n = p1 . Then if Ar ∈

Lpr ([−π, π]d , M), 1 ≤ r ≤ n, the function Tn,ϕ,b,η (A1 , . . . , An ) ∈ Lp ([−π, π]d ×
Ω, M) and
"n
bn−1 ϕ∞ r=1 Ar pr
(3.34) Tn,ϕ,b,η (A1 , . . . An )p ≤ .
(1 − b)2n

Proof. Consider first the case n = 2. If A1 , A2 ∈ L∞ ([−π, π]d , M) then it is clear

that T2,ϕ,b,η (A1 , A2 ) ∈ L∞ ([−π, π]d × Ω, M) and

bϕ2 A1 ∞ A2 ∞

T2,ϕ,b,η (A1 , A2 )∞ ≤ .
(1 − b)

If A2 ∈ L∞ , A1 ∈ L2 then we can see by the argument of Lemma 3.4 that

T2,ϕ,b,η (A1 , A2 ) ∈ L2 and

bϕ∞ A2 ∞ A1 2

(3.35) T2,ϕ,b,η (A1 , A2 )2 ≤ .
(1 − b)2

It follows therefore by interpolation theory that if A2 ∈ L∞ , A1 ∈ Lp , p ≥ 2, then

T2,ϕ,b,η (A1 , A2 ) ∈ Lp and

bϕ∞ A2 ∞ A1 p

(3.36) T2,ϕ,b,η (A1 , A2 )p ≤ .
(1 − b)2
188 Joseph G. Conlon and Ali Naddaf

Suppose next that A1 ∈ Lp1 , A2 ∈ Lp2 and 1/p1 + 1/p2 = 1/2. We have now that
 1/2
2
(3.37) dξ T2,ϕ,b,η (A1 , A2 )(ξ, ·)
[−π,π]d
∞ 



≤ dξ 1 (x1 )e−ix1 ·ξ τx
A 1
m=0 [−π,π]d
x1 ,x2 ∈Zd
1/2
2 (x2 )e−ix2 ·ξ τx ϕ(·)2
P b(·)(P Tb,η,ξ ) A m
.
2

Observe now that as in (3.16) one has



dξ 1 (x1 )e−ix1 ·ξ τx b(·)(Tb,η,ξ )m A
A 2 (x2 )e−ix2 ·ξ τx ϕ(·)2
1 2
[−π,π]d
x1 ,x2 ,∈Zd

 m+1
! 
= (2π) d
 
A1 (y1 )b(τy1 ·) ∗
∇ ∇Gη (yj − yj−1 )b(τyj ·)
r∈Zd y1 ,...,ym+1 j=2
2 (r − ym+1 )ϕ(τr ·)2
A


 m+1
! 
≤ (2π)d ϕ2∞ || 1 (y1 )b(τy ·)
A ∇ ∗
∇G (y
η j − y j−1 )b(τ yj ·)
1

r∈Zd y1 ,...,ym+1 j=2

A2 (r − ym+1 )|| 2

 

= ϕ∞2
dξ 
A1 (x1 )e−ix1 ·ξ
τx1 b(·)(Tb,η,ξ ) Id A2 (ξ)2
m
[−π,π]d
x1 ∈Zd


≤ ϕ2∞ dξ 1 (x1 )e−ix1 ·ξ τx b(·)(Tb,η,ξ )m Id ||2 ||A2 (ξ)2
A 1
[−π,π]d
x1 ∈Zd

2/p1
≤ ϕ2∞ dξ 1 (x1 )e−ix1 ·ξ τx b(·)(Tb,η,ξ )m Id p1
A A2 2p2 .
1
[−π,π]d
x1 ∈Zd

We have already seen from interpolation theory that


1/p1
dξ 1 (x1 )e−ix1 ·ξ τx b(·)(Tb,η,ξ )m Id p1
A ≤ bm+1 A1 p1 .
1
[−π,π]d
x1 ∈Zd

We conclude therefore that



dξ 1 (x1 )e−ix1 ·ξ τx b(·)(Tb,η,ξ )m A
A 2 (x2 )e−ix2 ·ξ τx ϕ(·)2
1 2
[−π,π]d
x1 ∈Zd

≤ b2m+2 ϕ2∞ A1 2p1 A2 2p2 .

Arguing as in Lemma 3.4 we also see that


dξ A1 (x1 )e−ix1 ·ξ τx P b(·)(P Tb,η,ξ )m A
2 (x2 )e−ix2 ·ξ τx ϕ(·)2
1 2
[−π,π]d
x1 ,x2 ∈Zd

≤ (m + 1)2 b2m+2 ϕ2∞ A1 2p1 A2 2p2 .

 Green’s Functions for Equations with Random Coefficients 189

It follows now from (3.37) that

bϕ∞ A1 p1 A2 p2
(3.38) T2,ϕ,b,η (A1 , A2 )2 ≤ ,
(1 − b)2
provided 1/p1 + 1/p2 = 1/2. We can now use the inequalities (3.36),(3.38) to do a
further interpolation. Thus we have if p ≥ 2 and 1/p1 + 1/p2 = 1/p then
bϕ∞ A1 p1 A2 p2
T2,ϕ,b,η (A1 , A2 )p ≤ .
(1 − b)2
This proves the result for n = 2.
To deal with n = 3 we subdivide into 7 cases:
(a) 1/p1 = 0, 1/p2 = 0, 1/p3 = 0,
(b) 1/p1 = 1/2, 1/p2 = 0, 1/p3 = 0,
(c) 1/p1 < 1/2, 1/p2 = 0, 1/p3 = 0,
(d) 1/p1 + 1/p2 = 1/2, 1/p3 = 0,
(e) 1/p1 + 1/p2 < 1/2, 1/p3 = 0,
(f) 1/p1 + 1/p2 + 1/p3 = 1/2,
(g) 1/p1 + 1/p2 + 1/p3 < 1/2.
For (a) it is easy to see that
b2 ϕ2 A1 ∞ A2 ∞ A3 ∞
T3,ϕ,b,η (A1 , A2 , A3 )∞ ≤ .
(1 − b)2
For (b) we use the argument of Lemma 3.4 to conclude
b2 ϕ∞ A1 2 A2 ∞ A3 ∞
T3,ϕ,b,η (A1 , A2 , A3 )2 ≤ .
(1 − b)4
The Riesz-Thorin Theorem applied to (a), (b) yield for (c) the inequality,
b2 ϕ∞ A1 p1 A2 ∞ A3 ∞
T3,ϕ,b,η (A1 , A2 , A3 )p ≤ ,
(1 − b)4
where p = p1 . For (d) we use the argument to obtain (3.38) to conclude that
b2 ϕ∞ A1 p1 A2 p2 A3 ∞
T3,ϕ,b,η (A1 , A2 , A3 )2 ≤ .
(1 − b)4
Now the Riesz-Thorin Theorem applied to (c) and (d) yield for (e) the inequality
b2 ϕ∞ A1 p1 A2 p2 A3 ∞
T3,ϕ,b,η (A1 , A2 , A3 )p ≤ ,
(1 − b)4
where 1/p = 1/p1 + 1/p2 . To obtain an inequality for (f) we use the argument to
obtain (3.38). This reduces us to the case dealt with in (e). Hence we can use the
inequality for (e) to obtain the bound
b2 ϕ∞ A1 p1 A2 p2 A3 p3
T3,ϕ,b,η (A1 , A2 , A3 )2 ≤ .
(1 − b)4
Finally the Riesz-Thorin Theorem applied to (e) and (f) yields the inequality (3.34)
with n = 3 for (g). Since it is clear we can generalize the method for n = 3 to all
n, the result follows.

190 Joseph G. Conlon and Ali Naddaf

Lemma 3.8. Suppose ∞ > p1 , . . . , pn , p > 2 and p11 + · · · + p1n = p1 . Then if Ar ∈

Lpwr ([−π, π]d , M), 1 ≤ r ≤ n, the function Tn,ϕ,b,η (A1 , . . . , An ) ∈ Lpw ([−π, π]d ×
Ω, M) and
"n
Cbn−1 ϕ∞ r=1 Ar pr ,w
(3.39) Tn,ϕ,b,η (A1 , . . . , An )p,w ≤ ,
(1 − b)2n
where the constant C depends only on p1 , . . . pn .

Proof. We use Lemma 3.7 and Hunt’s Interpolation Theorem [10]. Suppose n = 2.
Now from Lemma 3.7 we know that for a given p1 , 2 ≤ p1 ≤ ∞, then
bϕ∞ A1 p1 A2 ∞
T2,ϕ,b,η (A1 , A2 )p1 ≤ ,
(1 − b)2
bϕ∞ A1 p1 A2 q1
T2,ϕ,b,η (A1 , A2 )2 ≤ ,
(1 − b)2
where 1/p1 + 1/q1 = 1/2. Hence the Hunt Theorem implies that
C(p1 , p2 )bϕ∞ A1 p1 A2 p2 ,w
(3.40) T2,ϕ,b,η (A1 , A2 )p,w ≤ ,
(1 − b)2
provided 1/p1 + 1/p2 = 1/p > 1/2, p2 < ∞, and C(p1 , p2 ) is a constant depending
only on p1 , p2 . Suppose now that A2 is fixed with A2 p2 ,w finite for some p2 , 2 <
p2 < ∞. We consider the mapping
A1 → T2,ϕ,b,η (A1 , A2 ).
We see from (3.40) that this maps L∞ to Lpw2 . For ε > 0 let p1 (ε) satisfy 1/p1 (ε) +
p(ε)
1/p2 = 1/2 + ε = 1/p(ε). From (3.40) we also see that it maps Lp1 (ε) to Lw . It
follows again from interpolation theory that for any p1 , p1 (ε) < p1 < ∞ it maps
Lpw1 to Lpw where 1/p1 + 1/p2 = 1/p. The inequality (3.39) for n = 2 follows from
this. It is clear this method can be generalised to all n.

Lemma 3.9. Let d ≥ 3, η > 0, 1 ≤ k, k  ≤ d. Then qk,k
(ξ, η) is a C ∞ function

of ξ ∈ [−π, π]d . Further, let α = (α1 , . . . , αd ), where αi ≥ 0, 1 ≤ i ≤ d, and |α| =
"d d/|α|
α1 + · · · + αd < d. Then the function i=1 ( ∂ξ∂ i )αi qk,k
(ξ, η) is in Lw ([−π, π]d ),
"d
and  i=1 ( ∂ξ∂ i )αi qk,k
d/|α|,w ≤ Cλ,Λ,d , where the constant Cλ,Λ,d depends only on
λ, Λ, d.

Proof. We argue as in Lemma 3.6. It is easy to see that the function Ψ of

Lemma 3.2 has the property that
!d
∂ αi
( ) Ψ(ξ, η, ·)
i=1
∂ξi

is a sum of terms Tn,ϕ,b,η (A1 , . . . , An )(ξ, ·)ek , where 1 ≤ n ≤ |α|, Ar ∈ Lpwr , 1 ≤

r ≤ n, and 1/p1 + · · · + 1/pn = |α|/d. The result follows now from Lemma 3.8 if
|α| < d/2. To deal with the case d/2 ≤ |α| < d, we argue exactly as for the d = 3
case with |α| = 2.

 Green’s Functions for Equations with Random Coefficients 191

Proposition 3.2. Let d ≥ 3. Then the function Ga (x) defined by (3.1) satisfies
the inequalities
0 ≤ Ga (x) ≤ Cλ,Λ,d [1 + |x|]2−d , x ∈ Zd ,
|∇Ga (x)| ≤ Cλ,Λ,d log[2 + |x|](1 + |x|)1−d , x ∈ Zd ,
where the constant Cλ,Λ,d depends only on λ, Λ, d.

Proof. We argue as in Proposition 3.1, using Lemma 3.9.

Proposition 3.2 gives an alternative derivation of the inequality (1.14). We can
extend the argument in the proposition to obtain Theorem 1.5. To do this we need
the following improvement of Lemma 3.9.
Lemma 3.10. Let qk,k
(ξ, η) be the function of Lemma 3.9 and |α| = d−1. Suppose
ρ ∈ Rd , |ρ| ≤ 1. Then for any ε, 0 < ε < 1 the function
!d
∂ αi

( ) qk,k
(ξ + ρ, η) − qk,k
(ξ, η) /|ρ|1−ε
i=1
∂ξi
d/(d−ε)
is in Lw ([−π, π]d ), and
# d #
#! ∂ α
 #
# ( 1−ε #
) qk,k
(ξ + ρ, η) − qk,k
(ξ, η) /|ρ| #
i
≤ Cλ,Λ,d," ,
# ∂ξi
i=1 d/(d−ε),w

where the constant Cλ,Λ,d," depends only on λ, Λ, d, 4.

Before proving Lemma 3.10 we first show how Theorem 1.5 follows from it.
Proof of Theorem 1.5. We shall confine ourselves to the case d = 3 since the
argument for d > 3 is similar. Consider the representation (3.32) for n.∇Ga,η (x).
We have now,
   
1 1 −ix·ξ ∂ −n · e(−ξ)
= dξe
|ξ|>γ/|x| ix1 (2π)d |ξ|>γ/|x| ∂ξ1 e(ξ)q(ξ, η)e(−ξ)

1 1 n · e(−ξ) ξ1
− dξe−ix·ξ .
ix1 (2π)d |ξ|=γ/|x| e(ξ)q(ξ, η)e(−ξ) |ξ|
Evidently one has a bound for the surface integral,



≤ C/|x|2 .
|ξ|=γ/|x|
If we integrate by parts again in the volume integral we have
  
1 1 −ix·ξ ∂ −n · e(−ξ)
dξe
ix1 (2π)d |ξ|>γ/|x| ∂ξ1 e(ξ)q(ξ, η)e(−ξ)
  2  
1 1 −ix·ξ ∂ n · e(−ξ)
= 2 dξe
x1 (2π)d |ξ|>γ/|x| ∂ξ1 e(ξ)q(ξ, η)e(−ξ)
  
1 1 −ix·ξ ∂ n · e(−ξ) ξ1
+ 2 d
dξe .
x1 (2π) |ξ|=γ/|x| ∂ξ1 e(ξ)q(ξ, η)e(−ξ) |ξ|
Since q(ξ, η) is bounded and ∂q(ξ, η)/∂ξi is in L3w it follows that the average of the
surface integral over γ, 1 < γ < 2, is bounded by C/|x|2 for some constant C. To
192 Joseph G. Conlon and Ali Naddaf

bound the volume integral let ρ ∈ R3 be such that e−ix·ρ = −1 and ρ has minimal
magnitude. Then |ρ| ≤ 10/|x| and
  2  
1 1 −ix·ξ ∂ n · e(−ξ)
dξe
x21 (2π)d |ξ|>γ/|x| ∂ξ1 e(ξ)q(ξ, η)e(−ξ)

1 1
= 2 dξe−ix·ξ
2x1 (2π)d |ξ|>100/|x|
 2  
∂ n · e(−ξ) n · e(−ξ − ρ)
−
∂ξ1 e(ξ)q(ξ, η)e(−ξ) e(ξ + ρ)q(ξ + ρ, η)e(−ξ − ρ)

1
+ 2 R(ξ)e−ix·ξ dξ ,
x1 1/|x|<|ξ|<200/|x|
3/2
where R(ξ) is the remainder term. Using the fact that ∂q/∂ξ1 ∈ L3w , ∂ 2 q/∂ξ12 ∈ Lw
we can easily see that

1
|R(ξ)|dξ ≤ C/|x|2
x21 1/|x|<ξ<200/|x|
for some constant C. We can bound the first term above using Lemma 3.9 and
Lemma 3.10. Evidently we will get a term

C|ρ|1−ε
 ∂ 2

1−ε dξ
q(ξ + ρ, η) − q(ξ, η) /|ρ| .
|x|2 |ξ|>100/|x| ∂ξ1 |ξ|
From Lemma 3.10 this is bounded by
∞  ∞
C|ρ|1−ε
C  |ρ|1−ε 1−ε
−n(1−ε) 10C 
2
≤ 2
|x| 2 ≤ ,
|x| n=0 2
n+1 /|x|>|ξ|>2n /|x| |x| n=0
|x|2
for some constant C  . Other terms are bounded using Lemma 3.9. We have proved
the first inequality of Theorem 1.5. The second inequality of the theorem for d = 3
is proved similarly.

Proof of Lemma 3.10. The argument follows the same lines as in Lemma 3.9.
The main point to observe is that if f (ξ) is given by
!d  αi  
∂  η (ξ) ,
f (ξ) = ek (−ξ)ek
(ξ)G
i=1
∂ξi

with |α| < d then for any ε, 0 < ε < 1, the function [f (ξ + ρ) − f (ξ)]/|ρ|1−ε
d/(1+|α|−ε)
is in Lw and there is a constant Cd,ε depending only on d, ε such that
[f (ξ + ρ) − f (ξ)]/|ρ|1−ε d/(1+|α|−ε),w ≤ Cd,ε , provided |ρ| ≤ 1 .

4. Proof of Theorem 1.4—Diagonal Case

Here we shall prove the inequalities of Theorem 1.4, but without the exponential
falloff term. We shall call this the diagonal case. First we show that Theorem 1.6
already gives us the diagonal case of the inequality (1.10) in dimension d = 1.
Corollary 4.1. The function Ga (x, t) satisfies the inequality
√
(4.1) 0 ≤ Ga (x, t) ≤ C(λ, Λ)/[1 + t], if d = 1,
 Green’s Functions for Equations with Random Coefficients 193

where the constant C(λ, Λ) depends only on λ, Λ. If d > 1, it satisfies an inequality,

(4.2) 0 ≤ Ga (x, t) ≤ Cε (d, λ, Λ)/[1 + t1−ε ],
where ε can be any number, 0 < ε < 1, and Cε (d, λ, Λ) is a constant depending only
on ε, d, λ, Λ.

Proof. We have

1
Ga (x, t) ≤ |Ĝa (ξ, t)|dξ
(2π)d [−π,π]d
 
1 1
≤ dξ + dξ .
(2π)d |ξ|<1/√t (2π)d |ξ|>1/√t

Since Ĝa (ξ, t) is bounded on [−π, π]d it follows that


1
(4.3) dξ ≤ C(d, λ, Λ)/[1 + td/2 ].
(2π)d |ξ|<1/√t
√
The integral over {|ξ| > 1/ t} is nonzero only if t > 1/π 2 d. In that case one has
from Theorem 1.6,
 
1 dξ
dξ ≤ C(δ, λ, Λ) √ (ξ 2 t)δ ,
(2π)d |ξ|>1/√t |ξ|>1/ t

for any δ satisfying 0 ≤ δ < 1. For d = 1 we have on taking δ > 1/2 an inequality

dξ C
√ (ξ 2 t)δ ≤ √ .
|ξ|>1/ t t
The inequality (4.1) follows from this and (4.3). For d > 1 and any p > d/2 we
have
 $ %1/p
dξ d(1−1/p) dξ
√ (ξ 2 t)δ ≤ (2π) √ (ξ 2 t)δp ≤ Cp /td/2p ,
|ξ|>1/ t |ξ|>1/ t

where the constant Cp depends only on p and we have chosen δ to satisfy 1 >
δ > d/2p. The inequality (4.2) follows now from this last inequality and (4.3) on
choosing p to satisfy d/2p = 1 − ε.

We can similarly use Theorem 1.6 to obtain estimates on the t derivative of
Ga (x, t).
Corollary 4.2. The function Ga (x, t) is differentiable w.r. to t for t > 0 and the
derivative satisfies the inequality

∂Ga (x, t)
≤ C(λ, Λ)/[1 + t3/2 ] , if d = 1 ,
∂t

where the constant C(λ, Λ) depends only on λ, Λ. If d > 1, it satisfies an inequality

∂Ga (x, t)
≤ Cε (d, λ, Λ)/[1 + t2−ε ] ,
∂t

where ε can be any number, 0 < ε < 1, and Cε (d, λ, Λ) is a constant depending only
on ε, d, λ, Λ.
194 Joseph G. Conlon and Ali Naddaf

Observe now that Corollary 4.1 almost obtains the diagonal case of the inequality
(1.10) for d = 2. In order to obtain this inequality for d = 2 we shall have to use
the methods of Section 3.
Lemma 4.1. For d = 2, there is a constant C(λ, Λ) depending only on λ, Λ such
that
0 ≤ Ga (x, t) ≤ C(λ, Λ)/[1 + t] , t>0.

Proof. We shall use the notation of Section 2, in particular the functions h, k

defined by (2.18). It is easy to see that
(4.4)
 ∞  ∞
1 ∂ 2 h(ξ, η)
h(ξ, η) cos[Im(η)t]d[Im(η)] = 2 {1 − cos[Im(η)t]}d[Im(η)].
0 t 0 ∂[Im(η)]2
We have now from (2.44), (2.45), (2.47) that
 ∞ 2
∂ h(ξ, η) 2
d[Im(η)] < .
∂[Im(η)]2 |e(ξ)|4
|e(ξ)|2

Hence if t > 1, then

  ∞

h(ξ, η) cos[Im(η)t]d[Im(η)] dξ

[−π,π]d 0
  
C 2 −4
= √ + √ ≤ t + t2 √ |e(ξ)| dξ
|e(ξ)|<1/ t |e(ξ)|>1/ t |e(ξ)|>1/ t
 
1 |e(ξ)|2 ∂ 2 h(ξ, η)

+ √ dξ {1 − cos[Im(η)t]}d[Im(η)] ,
|e(ξ)|>1/ t t2 0 ∂[Im(η)]2

where we have used the fact that the LHS of (4.4) is bounded by a universal
constant. Evidently the first two terms on the RHS of the last inequality are
bounded by C/t, so we concentrate on the third term.
In view of (2.44) and the inequalities following it we have
 
1 |e(ξ)|2 ∂ 2 h(ξ, η)

√ dξ {1 − cos[Im(η)t]}d[Im(η)] ≤
|e(ξ)|>1/ t t2 0 ∂[Im(η)]2
  |e(ξ)|2  
Cλ,Λ 1 |ψ(ξ, η, ·)|2
+ Cλ,Λ √ dξ 2 {1 − cos[Im(η)t]}d[Im(η)] ,
t |e(ξ)|>1/ t t 0 |e(ξ)|4 Im(η)
for some constant Cλ,Λ depending only on λ, Λ. Now for Re(η) > 0 and 2 < p < ∞
let hη,p (ξ), ξ ∈ [−π, π]2 , be the function
$ 2 %1/2

 
1/2 − 1/p 2
hη,p (ξ) = |Im(η)| |ψk (ξ, η, ·)| ,
k=1

where ψk (ξ, η, ·) is given by (2.2). We shall show that hη,p ∈ Lpw ([−π, π]2 ) and there
is a constant Cp,λ,Λ depending only on p, λ, Λ such that
(4.5) hη,p p,w ≤ Cp,λ,Λ , Re(η) > 0, 2 < p < ∞.
 Green’s Functions for Equations with Random Coefficients 195

Observe now from (2.8),(2.26) that

 2  
1 |e(ξ)| |ψ(ξ, η, ·)|2
{1 − cos[Im(η)t]}d[Im(η)]
t2 0 |e(ξ)|4 Im(η)
 |e(ξ)|2
1 hη,p (ξ)2 {1 − cos[Im(η)t]}
≤ 1+1/p 2 1−1/p
d[Im(η)]
t 0 |e(ξ)| Im(η) [Im(η)t]1−1/p
 |e(ξ)|2
C hη,p (ξ)2
≤ 1+1/p d[Im(η)],
t 0 |e(ξ)|2 Im(η)1−1/p
for some universal constant C. Next we have that
  |e(ξ)|2
1 hη,p (ξ)2
(4.6) √ dξ 1+1/p
d[Im(η)] ≤
|e(ξ)|>1/ t t 0 |e(ξ|2 Im(η)1−1/p
∞  2n+2 /t 
C
−2n 2 d[Im(η)] 2
2 √ hη,p (ξ) dξ .
t1/p n=0 0 Im(η) 1−1/p
|e(ξ)|<2 n+1 / t

From (3.30) it follows that


√ hη,p (ξ)2 dξ ≤ Cp hη,p 2p,w 22n(1−2/p) /t1−2/p .
|e(ξ)|<2n+1 / t

If we use the inequality (4.5) we see from the last inequality that the RHS of (4.6)
is bounded by Cp,λ,Λ /t for some constant Cp,λ,Λ depending only on p, λ, Λ. We
conclude that if (4.5) holds then
  ∞


dξ h(ξ, η) cos[Im(η)t]d[Im(η)] ≤ Cλ,Λ /t, t > 0,
[−π,π]2 0

for some constant Cλ,Λ depending only on λ, Λ.

To prove (4.5) note that ψk (ξ, η, ·) is a sum of terms P Tϕ,b,η (f ) where Tϕ,b,η is
the operator (3.15), ϕ is an entry of the matrix a(·) and f is the Fourier transform
of ∇j Gη (x), x ∈ Zd , 1 ≤ j ≤ d. Hence
|e(ξ)| 1 1
|f (ξ)| ≤ ≤ &  , ξ ∈ [−π, π]2 ,
|e(ξ)|2 + |Im(η)| 1−2/p |e(ξ)|2/p
2 |Im(η)|

whence f ∈ Lpw ([−π, π]2 ) with norm bounded by a constant times |Im(η)|1/p−1/2 .
The inequality (4.5) follows now from Lemma 3.5.
Next, observe that for finite N ,
 N
−k(ξ, Re(η) + iN ) cos N t
k(ξ, η) sin[Im(η)t]d[Im(η)] =
0 t
 N
1 ∂k(ξ, η)
+ cos[Im(η)t]d[Im(η)],
t 0 ∂[Im(η)]
where we have used the fact that k(ξ, η) = 0 if Im(η) = 0. Integrating again by
parts and letting N → ∞ we conclude that
 N 
−1 ∞ ∂ 2 k(ξ, η)
lim k(ξ, η) sin[Im(η)t]d[Im(η)] = 2 sin[Im(η)t]d[Im(η)].
N →∞ 0 t 0 ∂[Im(η)]2
196 Joseph G. Conlon and Ali Naddaf

Since the estimates on ∂ 2 k(ξ, η)/∂[Im(η)]2 are similar to those on ∂ 2 h(ξ, η)/∂[Im(η)]2
we can argue as previously to conclude that
 
N

dξ lim k(ξ, η) sin[Im(η)t]d[Im(η)] ≤ Cλ,Λ /t, t > 0,
[−π,π]2 N →∞ 0
for some constant Cλ,Λ depending only on λ, Λ.

We can similarly sharpen the result of Corollary 4.2 when d = 2.
Lemma 4.2. For d = 2, there is a constant C(λ, Λ) depending only on λ, Λ such
that

∂Ga (x, t)
≤ C(λ, Λ)/[1 + t2 ] , t > 0 .
∂t

Proof. Integrating by parts in (2.42) we have that

 ∞
∂
h(ξ, η) cos[Im(η)t]d[Im(η)] =
∂t 0
  
−1 N ∂ 2 h(ξ, η) ∂ 3 h(ξ, η)
lim 3 3 + Im(η) {1 − cos[Im(η)t]}d[Im(η)] .
N →∞ t 0 ∂[Im(η)]2 ∂[Im(η)]3
We can compute ∂ 3 h(ξ, η)/∂[Im(η)]3 from (2.44). It is clear we can derive similar
estimates on Im(η)∂ 3 h(ξ, η)/∂[Im(η)]3 to the ones on ∂ 2 h(ξ, η)/∂[Im(η)]2 which we
used in the proof of Lemma 4.1. We conclude therefore that there is a constant
Cλ,Λ depending only on λ, Λ such that
  ∞
∂

dξ h(ξ, η) cos[Im(η)t]d[Im(η)] ≤ Cλ,Λ /t2 , t > 1 .
2
[−π,π] ∂t 0

From (2.48) we have that


∂ −1 ∞ ∂k(ξ, η)
{1 − cos[Im(η)t]}d[Im(η)]
∂t t 0 ∂[Im(η)]
    
1 ∞ ∂ ∂k(ξ, η) ∂k(ξ, η)
= 2 Im(η) + {1 − cos[Im(η)t]}d[Im(η)]
t 0 ∂[Im(η)] ∂[Im(η)] ∂[Im(η)]
    
−1 N ∂ ∂k(ξ, η) ∂k(ξ, η)
= lim 2 Im(η) + cos[Im(η)t]d[Im(η)],
N →∞ t 0 ∂[Im(η)] ∂[Im(η)] ∂[Im(η)]
where we have used the fact that k(ξ, η) = 0 if η > 0 is real. Integrating now by
parts we conclude that

∂ −1 ∞ ∂k(ξ, η)
{1 − cos[Im(η)t]}d[Im(η)] =
∂t t 0 ∂[Im(η)]
  
1 N ∂ 2 k(ξ, η) ∂ 3 k(ξ, η)
lim 3 + Im(η) sin[Im(η)t]d[Im(η)] .
N →∞ t3 0 ∂[Im(η)]2 ∂[Im(η)]3
We can then argue just as for the integral in h that
 
∂ −1 ∞ ∂k(ξ, η)

dξ {1 − cos[Im(η)t]}d[Im(η)] ≤ Cλ,Λ /t2 , t > 1 ,
[−π,π]2 ∂t t 0 ∂[Im(η)]
for a constant Cλ,Λ depending only on λ, Λ.

 Green’s Functions for Equations with Random Coefficients 197

So far we have not obtained diagonal estimates, even in dimension 1, on spatial

derivatives of Ga (x, t), which correspond to the estimates of Theorem 1.4. Next we
shall prove these estimates for the case d = 2. The method readily extends to the
case d = 1.
Lemma 4.3. For d = 2 there is a constant C(λ, Λ) depending only on λ, Λ such
that
|∇i Ga (x, t)| ≤ C(λ, Λ)/[1 + t3/2 ] .
Let δ satisfy 0 ≤ δ < 1. Then there is a constant C(δ, λ, Λ) depending only on
δ, λ, Λ such that
|∇i ∇j Ga (x, t)| ≤ C(δ, λ, Λ)/[1 + t(3+δ)/2 ] .

Proof. If we use the fact that for 0 ≤ δ < 1, |ej (−ξ)| ≤ 21−δ |ej (−ξ)|δ , then we see
from the proof of Lemma 4.1 that it is sufficient to show that for t > 1,
(4.7)
 
1 |e(ξ)|2 ∂ 2 h(ξ, η) C
1+δ
√ dξ|e(−ξ)| 2
t 0 ∂[Im(η)]2
{1 − cos[Im(η)t]}d[Im(η)] ≤ (3+δ)/2 .
t
|e(ξ)|>1/ t

This follows by the argument of Lemma 4.1 if we can choose hη,p ∈ Lpw with
p < 2/(1 + δ). We cannot do this since hη,p ∈ Lpw only for p > 2. To get around this
3/2
we can argue similarly to Lemma 3.6 in the proof that ∂ 2 qk,k
/∂ξi ∂ξj  ∈ Lw .
For Re(η) > 0 and 2 < p < ∞, let gη,p (ξ), ξ ∈ [−π, π]2 be the function


2 1/2
1−1/p
gη,p (ξ) = |Im(η)| | ψk (−ξ, η, ·)[∂ψk (ξ, η, ·)/∂η] |

.
k,k
=1

Then from Section 3 we see that gη,p ∈ Lpw ([−π, π]2 ) and there is a constant Cp,λ,Λ
depending only on p, λ, Λ such that
gη,p p,w ≤ Cp,λ,Λ , Re(η) > 0, 2 < p < ∞ .
Observe now that the contribution of the last term on the RHS of (2.44) to the
integral on the LHS of (4.7) is bounded by a constant times
  2
1 |e(ξ)| gη,p (ξ)2
√ dξ d[Im(η)] .
|e(ξ)|>1/ t t2 0 |e(ξ)|1−δ Im(η)2−2/p
Arguing as in Lemma 4.1 we see that this is bounded by the RHS of (4.7) provided
δ < 1. Note that as δ → 1 the estimate diverges. Since we can make similar
estimates for the other terms on the RHS of (2.44), the result follows.

We wish to extend Lemmas 4.1, 4.2, 4.3 to d ≥ 3. To do this we need the

following lemma.
Lemma 4.4. The functions h(ξ, η), k(ξ, η), Re(η) > 0, have the property that
∂ m h(ξ, η)/∂[Im(η)]m = 0, m odd, η > 0 real,
∂ m k(ξ, η)/∂[Im(η)]m = 0, m even, η > 0 real.
198 Joseph G. Conlon and Ali Naddaf

Proof. Note that we can see ∂h(ξ, η)/∂[Im(η)] = 0 for real η > 0 from (2.37) if we
use (2.15) and the fact that ψ(−ξ, η, ·) = ψ(ξ, η̄, ·). More generally the result follows
from the fact that the function g(η) = h(ξ, η) + ik(ξ, η) is analytic for Re(η) > 0
and real when η is real. In fact from the Cauchy-Riemann equations we have that
 m 
∂ m g(η) m ∂ h(ξ, η) ∂ m k(ξ, η)
= (−i) + i , m = 0, 1, 2, . . . . .
∂η m ∂[Im(η)]m ∂[Im(η)]m
Evidently the LHS of this identity is real for all η > 0 real.

Lemma 4.5. For d ≥ 3 there is a constant C(λ, Λ, d) depending only on λ, Λ, d
such that
0 ≤ Ga (x, t) ≤ C(λ, Λ, d)/[1 + td/2 ], t > 0.

Proof. From Lemma 4.4 we see that

 ∞
(4.8) h(ξ, η) cos[Im(η)t]d[Im(η)] ,
0
is, for d odd, one of the integrals,
 ∞ (d+1)/2
1 ∂ h(ξ, η)
(4.9) ± (d+1)/2 (d+1)/2
sin[Im(η)t]d[Im(η)]
t 0 ∂[Im(η)]
 ∞ (d+1)/2
1 ∂ h(ξ, η)
± (d+1)/2 {1 − cos[Im(η)t]}d[Im(η)].
t 0 ∂[Im(η)](d+1)/2
For d even it is one of the integrals,
 ∞ d/2+1
1 ∂ h(ξ, η)
(4.10) ± d/2+1 d/2+1
sin[Im(η)t]d[Im(η)]
t 0 ∂[Im(η)]
 ∞ d/2+1
1 ∂ h(ξ, η)
± d/2+1 d/2+1
{1 − cos[Im(η)t]}d[Im(η)].
t 0 ∂[Im(η)]
We can see from (2.44) that for m = 1, 2, . . . , the derivative ∂ m h(ξ, η)/∂[Im(η)]m
is bounded in absolute value by a sum of terms
 r 
∂ ψ(±ξ,η,·) 2 αr /2
m−1
! ∂η r
(4.11)  ,
m+1− m−1
r=1 rαr
r=0 |η + e(ξ)q(ξ, η)e(−ξ)|
where the αr , r = 0, . . . , m − 1, are nonnegative integers satisfying the inequality
m−1


(4.12) (2r + 1)αr ≤ 2m.
r=0

By differentiating (2.46) sufficiently often and using Cauchy-Schwarz we obtain the

inequality,
 ∂ r ψ(±ξ, η, ·) 2  (r!)2  

≤ |ψ(±ξ, η, ·)|2 , r = 0, 1, 2, . . . .
∂η r |η|2r
Observe next that (4.12) implies that
m−1

m−1
1

m+1− rαr > αr .
r=1
2 r=0
 Green’s Functions for Equations with Random Coefficients 199

It follows therefore from (2.15) that there is a constant Cm depending only on m

such that (4.11) is bounded by Cm /|Im(η)|m+1 . Hence,
 ∞ m
∂ h(ξ, η)
d[Im(η)] ≤ Cm ,
m |e(ξ)|2m
|e(ξ)|2 ∂[Im(η)]

for some constant Cm depending only on m.

Let us assume now that d is odd and that the first integral in (4.9) is the correct
representation of (4.8). Then, following the argument of Lemma 4.1, we have that
  ∞

h(ξ, η) cos[Im(η)t]d[Im(η)] dξ
d

[−π,π] 0
 
= √ + √
|e(ξ)|<1/ t |e(ξ)|>1/ t

Cd,λ,Λ Cd −d−1
≤ d/2 + (d+1)/2 √ |e(ξ)| dξ
t t |e(ξ)|>1/ t
  |e(ξ)|2 (d+1)/2
1 ∂ h(ξ, η)

+ √ dξ sin[Im(η)t]d[Im(η)] .
|e(ξ)|>1/ t t(d+1)/2 0 ∂[Im(η)](d+1)/2

The first two terms on the RHS of the last inequality are bounded by Cd,λ,Λ /td/2
for some constant Cd,λ,Λ depending only on d, λ, Λ. We are left therefore to deal
with the final term.
Consider the simplest case of (4.11) when αr = 0, r > 0 and α0 ≤ 2m. Then
(4.11) is bounded by
 α0 /2 2(m+1)
|ψ(±ξ, η, ·)|2 /λ |e(ξ)|2(m+1) .
We shall show that
(4.13)
  |e(ξ)|2  α0 /2
1 |ψ(ξ, η, ·)|2

√ dξ sin[Im(η)t]d[Im(η)] ≤ Cd,λ,Λ /td/2 ,
|e(ξ)|>1/ t t(d+1)/2 0 |e(ξ)|d+3
provided α0 ≤ d + 1. To do this we define for d ≤ p < ∞ the function hη,p (ξ),
ξ ∈ [−π, π]d , by


d 
  1/2
(4.14) hη,p (ξ) = |[Im(η)]|1/2 − d/2p |ψk (ξ, η, ·)|2 ,
k=1

where ψk (ξ, η, ·) is given by (2.2). We can see similarly to the argument of Lemma 4.1
that hη,p ∈ Lpw ([−π, π]d ) and that there is a constant Cp,λ,Λ,d depending on p, λ, Λ, d
such that
hη,p p,w ≤ Cp,λ,Λ,d , Re(η) > 0, d ≤ p < ∞ .
We have now that
  |e(ξ)|2  α0 /2
1 |ψ(ξ, η, ·)|2

√ dξ sin[Im(η)t]d[Im(η)]
|e(ξ)|>1/ t t (d+1)/2
0 |e(ξ)| d+3
  |e(ξ)|2
1 hη,p (ξ)α0 d[Im(η)]
≤ √ dξ d+3−α0 [Im(η)]α0 (1/2 − d/2p)
.
|e(ξ)|>1/ t t (d+1)/2
0 |e(ξ)|
200 Joseph G. Conlon and Ali Naddaf

Since α0 ≤ d + 1 we can find p > d + 1 such that α0 (1/2 − d/2p) < 1. The proof of
the inequality (4.13) follows now exactly the same lines as in (4.6).
Next we consider the general case of (4.11). To do this we define for r =
0, 1, 2, . . . , Re(η) > 0 and p satisfying d/(2r + 1) ≤ p < ∞, functions hη,p,r (ξ)
by


d  r 1/2
r+1/2 − d/2p ∂ ψk (ξ, η, ·) 2
hη,p,r (ξ) = |Im(η)| .
∂η r
k=1
d
Observe that hη,p,0 (ξ) is the function (4.14). We shall show that for p > max[2, 2r+1 ],
p d
the function hη,p,r ∈ Lw ([−π, π] ) and there is a constant Cp,r,λ,Λ,d depending only
on p, r, λ, Λ, d such that
(4.15) hη,p,r p,w ≤ Cp,r,λ,Λ,d , Re(η) > 0, max[2, d/(2r + 1)] < p < ∞.
To prove this note that ∂ r ψk (ξ, η, ·)/∂η r is a sum of terms P Tn,ϕ,b,η (A1 , . . . , An )(ξ, ·),
where Tn,ϕ,b,η is the operator (3.33) and ϕ is an entry of the matrix a(·). Further-
more, there are positive integers r1 , . . . , rn such that r1 + · · · + rn = r + 1 and

A1 (ξ) ≤ |e(ξ)| [|e(ξ)|2 + |Im(η)|]r1 ,

Aj (ξ) ≤ |e(ξ)|2 [|e(ξ)|2 + |Im(η)|]rj +1 , 2 ≤ j ≤ n.
Suppose now p1 , . . . , pn are positive numbers satisfying
d d
(4.16) p1 ≥ , pj ≥ , 2 ≤ j ≤ n.
2r1 − 1 2rj
p
If also pj > 1, j = 1, . . . , n, then we see that Aj ∈ Lwj ([−π, π]d ), 1 ≤ j ≤ n, and
A1 p1 ,w ≤ Cd,r /|Im(η)|r1 −1/2−d/2p1 ,
Aj pj ,w ≤ Cd,r /|Im(η)|rj −d/2pj , 2 ≤ j ≤ n,
where Cd,r is a constant depending only on d, r. Note now that since
n
2r1 − 1
2rj 2r + 1
+ = ,
d j=2
d d

if p satisfies p > max[2, d/(2r + 1)], it is possible to choose p1 , . . . , pn satisfying

pj > 2, 1 ≤ j ≤ n, the inequalities (4.16) and the identity
1 1 1
+ ··· + = .
p1 pn p
The inequality (4.15) follows now from Lemma 3.7.
For the general case of (4.11) we need to show that
  |e(ξ)|2 "(d−1)/2
1 r=0 hη,qr ,r (ξ)αr
(4.17) √ dξ 
|e(ξ)|>1/ t t(d+1)/2 0 (d−1)/2
|e(ξ)|d+3− r=0 (2r+1)αr
d[Im(η)] Cd,λ,Λ
 (d−1)/2 ≤ d/2 ,
[Im(η)] r=0 (r+1/2−d/2qr )αr t
where the qr , r = 0, . . . , (d − 1)/2 are restricted to satisfy
(4.18) qr > max[2, d/(2r + 1)], r = 0, . . . , (d − 1)/2.
 Green’s Functions for Equations with Random Coefficients 201

Define now q by
(d−1)/2
1
αr
(4.19) = .
q r=0
qr

It follows then from (4.12) that

(d−1)/2    

1 d 1  d
r+ − αr = d +1− ,
r=0
2 2qr 2 q

where d is an integer satisfying −1 ≤ d ≤ d. Observe that the RHS of the last



inequality is strictly less than 1 if d ≤ 1 or d > 1, q < d/(d − 1). Suppose now
qr , r = 0, . . . , (d − 1)/2 and q satisfy (4.18), (4.19), with q > 1. Then it follows
from (4.15) that
(d−1)/2
!
hη,qr ,r (ξ)αr ∈ Lqw ([−π, π]d ) ,
r=0

with norm bounded by a constant depending only on λ, Λ, d q, qr , r = 0, . . . (d−1)/2.

If we can also arrange that for d > 1, q satisfies q < d/(d − 1) then we can see that
(4.17) holds by arguing just as we did in Lemma 4.1. Evidently it is possible to
choose the qr , r = 0, . . . , (d−1)/2, such that qr > d/(2r +1) and 1 < q < d/(d −1).
It is not possible, however, to satisfy the condition qr > 2, r = 0, . . . , (d − 1)/2 in
general.
To deal with this problem we need to use a sharper estimate on the derivative
∂ m h(ξ, η)/∂[Im(η)]m than sums of terms of the form (4.11). For r, j satisfying
0 ≤ j ≤ r, 0 ≤ r ≤ m − 1, let αr,j be nonnegative integers satisfying the inequality
m−1


r
(4.20) (1 + r + j)αr,j ≤ m.
r=0 j=0

If we define αr for 0 ≤ r ≤ m − 1, by
r

m−1


(4.21) αr = αr,j + αj,r ,
j=0 j=r

then we see from (4.20) that αr defined by (4.21) satisfies (4.12). It is also easy to
see that ∂ m h(ξ, η)/∂[Im(η)]m is bounded in absolute value by a sum of terms,
"m−1 "r  ∂ r ψ(±ξ,η,·) ∂ j ψ(±ξ,η,·)  αr,j
r=0 j=0 ∂η r ∂η j
(4.22)  m−1 ,
|η + e(ξ) q(ξ, η)e(−ξ)|m+1− r=1 rαr
where the αr,j satisfy (4.20) and the αr are defined by (4.21). Evidently the Schwarz
inequality implies that (4.22) is bounded by (4.11). We define for r, j = 0, 1, 2, . . . ,
Re(η) > 0 and p satisfying d/(r + j + 1) ≤ p < ∞, functions hη,p,r,j (ξ) by


d  ∂ r ψ (ξ, η, ·) ∂ j ψ
(ξ, η, ·)  1/2
k k
hη,p,r,j (ξ) = |Im(η)|(r+j+1−d/p)/2 r j .


∂η ∂η
k,k =1
202 Joseph G. Conlon and Ali Naddaf

We shall show that for p > max[2, d/(r +j +1)], the function hη,p,r,j ∈ Lpw ([−π, π]d )
and there is a constant Cp,r,j,λ,Λ,d depending only on p, r, j, λ, Λ, d such that
(4.23) hη,p,r,j p,w ≤ Cp,r,j,λ,Λ,d , Re(η) > 0, max[2, d/(r + j + 1)] < p < ∞.
In fact it is easy to see that (4.23) follows by arguing exactly as we did in Lemma 3.9
for the case d/2 ≤ |α| < d.
On replacing (4.11) by (4.22) the inequality (4.17) gets replaced by
  |e(ξ)|2 "(d−1)/2 "r 2αr,j
1 r=0 j=0 hη,qr,j ,r,j (ξ)
(4.24) √ dξ 
|e(ξ)|>1/ t t(d+1)/2 0 (d−1)/2
|e(ξ)|d+3− r=0 (2r+1)αr
d[Im(η)] Cd,λ,Λ
 (d−1)/2  r ≤ d/2 ,
[Im(η)] r=0 j=0 (r+j+1−d/q r,j )α r,j t

where the αr,j , αr satisfy (4.20), (4.21). Let q be defined

(d−1)/2 r
1

2αr,j
(4.25) = .
q r=0 j=0
qr,j

Then from (4.25) it follows that if qr,j > max[2, d/(r + j + 1)], 0 ≤ j ≤ r, 0 ≤ r ≤
(d − 1)/2, and q > 1 then the function
(d−1)/2 r
! !
hη,qr,j ,r,j (ξ)2αr,j ∈ Lqw ([−π, π]d ) ,
r=0 j=0

with norm bounded by a constant depending only on λ, Λ, d and the qr,j . We have
now from (4.20), (4.25) that
(d−1)/2 r


(d + 1) d
(r + j + 1 − d/qr,j )αr,j = − ,
r=0 j=0
2 2q

where d is an integer satisfying −1 ≤ d ≤ d. As before, the RHS of the last
inequality is strictly less than 1 if d ≤ 1 or d > 1, q < d/(d − 1). Hence
if q < d/(d − 1) the power of Im(η) on the LHS of (4.24) is strictly less than
1 and hence integrable. Finally, observe that in (4.22) one has αr,j = 0 unless
r + j ≤ m − 1 = (d − 1)/2. Note that if r + j < (d − 1)/2 then d/(r + j + 1) > 2.
On the other hand if αr,j
= 0 for some (r, j) with r + j = m − 1 then αr,j = 1 and
αr
,j
= 0 for (r , j  )
= (r, j). In that case d = d and (4.25) becomes 1/q = 2/qr,j ,
whence the condition q < d/(d − 1) becomes qr,j < 2d/(d − 1), so we may still
choose qr,j > 2. Thus (4.24) holds on appropriate choice of the qr,j .
The proof of the lemma for d odd is complete if we make the observation that
from Lemma 4.4 one has

1 ∞ ∂k(ξ, η)
{1 − cos[Im(η)t]}d[Im(η)] =
t 0 ∂[Im(η)]
 ∞ (d+1)/2
1 ∂ k(ξ, η)
± (d+1)/2 (d+1)/2
sin[Im(η)t]d[Im(η)]
t 0 ∂[Im(η)]
 ∞ (d+1)/2
1 ∂ k(ξ, η)
± (d+1)/2 (d+1)/2
{1 − cos[Im(η)t]}d[Im(η)],
t 0 ∂[Im(η)]
 Green’s Functions for Equations with Random Coefficients 203

depending on the value of d. Since the estimates on ∂ m k(ξ, η)/∂[Im(η)]m are the
same as those on ∂ m h(ξ, η)/∂[Im(η)]m the argument proceeds as before. The case
for d even is similar.

Lemma 4.6. For d ≥ 3 there is a constant C(λ, Λ, d) depending only on λ, Λ, d

such that

∂Ga (x, t)
≤ C(λ, Λ, d)/[1 + t1+d/2 ], t > 0.
∂t

Proof. Suppose d is odd and the first integral in (4.9) is the appropriate represen-
tation for this value of d. Then we have that
 ∞
∂
h(ξ, η) cos[Im(η)t]d[Im(η)]
∂t 0
is a sum of the terms,
 ∞ (d+1)/2
(d + 1) ∂ h(ξ, η)
± (d+3)/2
sin[Im(η)t]d[Im(η)],
2t 0 ∂[Im(η)](d+1)/2
 ∞ (d+1)/2
1 ∂ h(ξ, η)
± Im(η) cos[Im(η)t]d[Im(η)].
t(d+1)/2 0 ∂[Im(η)](d+1)/2
Evidently the first term is bounded by C(λ, Λ, d)/[1 + t1+d/2 ] by Lemma 4.5. On
integration by parts we can write the second term as a sum of two integrals,
 ∞ (d+1)/2
1 ∂ h(ξ, η)
± (d+3)/2 sin[Im(η)t]d[Im(η)],
t 0 ∂[Im(η)](d+1)/2
 ∞ (d+3)/2
1 ∂ h(ξ, η)
± (d+3)/2 Im(η) sin[Im(η)t]d[Im(η)].
t 0 ∂[Im(η)](d+3)/2
Again Lemma 4.5 implies that the first integral is bounded by C(λ, Λ, d)/[1+t1+d/2 ],
so we are left to deal with the second integral. Using the method of Lemma 4.1 we
see that we are left to deal with
  |e(ξ)|2 (d+3)/2
1 ∂ h(ξ, η)

√ dξ Im(η) sin[Im(η)t]d[Im(η)] .
|e(ξ)|>1/ t t (d+3)/2
0 ∂[Im(η)] (d+3)/2

Arguing exactly as in Lemma 4.5 we see this integral is bounded by C(λ, Λ, d)/[1 +
t1+d/2 ]. We can similarly bound the corresponding integral in k(ξ, η) and hence the
result follows.

Lemma 4.7. For d ≥ 3 there is a constant C(λ, Λ, d) depending only on λ, Λ, d

such that
|∇i Ga (x, t)| ≤ C(λ, Λ, d)/[1 + t(d+1)/2 ] .
Let δ satisfy 0 ≤ δ < 1. Then there is a constant C(δ, λ, Λ, d) depending only on
δ, λ, Λ, d such that
|∇i ∇j Ga (x, t)| ≤ C(δ, λ, Λ, d)/[1 + t(d+1+δ)/2 ] .

Proof. Same as for Lemma 4.6.

204 Joseph G. Conlon and Ali Naddaf

5. Proof of Theorem 1.4—Off Diagonal case

Here we shall complete the proof of Theorem 1.4. Thus we need to establish the
exponential falloff in (1.10) and in the inequalities of Theorem 1.4. Evidently (1.10)
implies that the periodic function Ĝa (ξ, t), ξ ∈ Rd , can be analytically continued
to Cd . Our goal will be to establish analyticity properties of Ĝa (ξ, t) and derive
from them the inequality (1.10) and Theorem 1.4.

Lemma 5.1. Suppose ε satisfies 0 < ε < 1. Then the periodic function Ĝa (ξ, t), ξ ∈
Rd , can be analytically continued to the strip {ξ ∈ Cd : |Im(ξ)| < ε}. There are
constants C1 (λ, Λ, d), C2 (λ, Λ, d) depending only on λ, Λ, d such that

|Ĝa (ξ, t)| ≤ C1 (λ, Λ, d) exp[C2 (λ, Λ, d)ε2 t], |Im(ξ)| < ε, t > 0 .

Proof. We have already seen in Section 2 that the matrix q(ξ, η) of (2.3) is defined
for all ξ ∈ Rd , Re(η) > 0, is continuous in (ξ, η) and analytic in η for fixed ξ. We
shall show now that for any δ > 0 there exists a constant C(λ, Λ, d, δ) > 0 depending
only on λ, Λ, d, δ such that if Re(η) > C(λ, Λ, d, δ)ε2 then q(ξ, η), ξ ∈ Rd , can be
analytically continued to the strip {ξ ∈ Cd : |Im(ξ)| < ε} and

(5.1) q(ξ, η) − q(Re(ξ), η) < δ, |Im(ξ)| < ε, Re(η) > C(λ, Λ, d, δ)ε2 .

In view of (3.6), (3.7) this will follow if we can show that for any δ > 0 there exists
a constant C(d, δ) > 0 depending only on d, δ, such that if Re(η) > C(d, δ)ε2 ,
then the operator Tk,k
,η,ξ of (3.4), which is bounded on L2 (Ω) for ξ ∈ Rd , extends
analytically to a bounded operator on L2 (Ω) for |Im(ξ)| < ε and

(5.2) Tk,k
,η,ξ − Tk,k
,η,Re(ξ)  < δ, |Im(ξ)| < ε, Re(η) > C(d, δ)ε2 .

To prove (5.2) observe that the Green’s function Gη (x) satisfies an inequality


∗ Cd exp − g(Re(η))|x|
|∇k ∇k
Gη (x)| ≤ , x ∈ Zd ,
[1 + |x|]d
where g(z), z > 0, is the function
' c √
d z, 0 < z < 1,
g(z) =
cd log(1 + z), z ≥ 1.
Here Cd , cd are positive constants depending only on d. It follows in particular
that there is a constant C1 (d), depending only on d, such that if 0 < ε < 1 and
Re(η) > C1 (d)ε2 , then the function ∇∗k ∇k
Gη (x)eix·ξ decreases exponentially in x
as |x| → ∞, provided |Im(ξ)| < ε. It follows from (3.4) that if Re(η) > C1 (d)ε2 then
the bounded operators Tk,k
,η,ξ on L2 (Ω), ξ ∈ Rd , extend analytically to bounded
operators on L2 (Ω) provided ξ ∈ Cd satisfies |Im(ξ)| < ε.
To prove (5.2) we use Bochner’s Theorem [9]. Thus for any ϕ ∈ L2 (Ω) there is
a positive finite measure dµϕ on [−π, π]d such that
  
ϕ(τx ·)ϕ(τy ·) = ei(x−y)·ζ dµϕ (ζ).
[−π,π]d
 Green’s Functions for Equations with Random Coefficients 205

Hence,

2
Tk,k
,η,ξ ϕ−Tk,k
,η,Re(ξ) ϕ = dµϕ (ζ)
[−π,π]d

  2

∇∗k ∇k
Gη (x) exp{ix · (ζ + ξ)} − exp{ix · (ζ + Re(ξ))}
x∈Zd


= dµϕ (ζ) ek (ζ + ξ)ek
(−ζ − ξ)Ĝη (ζ + ξ)
[−π,π]d
2

− ek (ζ + Re(ξ))ek
(−ζ − Re(ξ))Ĝη (ζ + Re(ξ)) .
We have now that
ek (ζ + ξ)ek
(−ζ − ξ)
ek (ζ + ξ)ek
(−ζ − ξ)Ĝη (ζ + ξ) = d .
j=1 ej (ζ + ξ)ej (−ζ − ξ) + η
Observe that
|ek (ζ + ξ) − ek (ζ + Re(ξ))| ≤ eε − 1 < 2ε, ζ ∈ [−π, π]d , |Imξ| < ε < 1.
Hence if C1 (d) > 24d and Re(η) > C1 (d)ε2 then

d
d 1

(5.3)
ej (ζ + ξ)ej (−ζ − ξ) + η ≥ ej (ζ + Re(ξ))ej (−ζ − Re(ξ)) + η
j=1
2 j=1
1
≥ C1 (d)ε2 , ζ ∈ [−π, π]d , |Imξ| < ε.
2
Similarly we see that


ek (ζ + ξ)ek
(−ζ − ξ) − ek (ζ + Re(ξ))ek
(−ζ − Re(ξ))

d 1/2
≤ 1000ε2 + 100ε ej (ζ + Re(ξ))ej (−ζ − Re(ξ)) , ζ ∈ [−π, π]d , |Imξ| < ε.
j=1

We conclude therefore that the integrand in the dµϕ integral is bounded as

ek (ζ + ξ)ek
(−ζ − ξ)Ĝη (ζ + ξ) − ek (ζ + Re(ξ))ek
(−ζ − Re(ξ))Ĝη (ζ + Re(ξ))
 1/2
d
1000ε2 + 100ε j=1 ej (ζ + Re(ξ))ej (−ζ − Re(ξ))
≤ d
| j=1 ej (ζ + ξ)ej (−ζ − ξ) + η|
 1/2 
2
√ d 2
|ek (ζ + Re(ξ))||ek
(−ζ − Re(ξ))| 4dε + 2ε d j=1 |ej (ζ + Re(ξ))|
+ d d .
| j=1 ej (ζ + ξ)ej (−ζ − ξ) + η|| j=1 ej (ζ + Re(ξ))ej (−ζ − Re(ξ)) + η|
We can see from (5.3) that the expression on the RHS of the last inequality is
bounded by
√
2000 200 4d 2 d
+& + +& .
C1 (d) C1 (d) C1 (d) C1 (d)
Evidently this last expression can be made smaller than δ by choosing C1 (d) suffi-
ciently large, whence (5.2) follows.
206 Joseph G. Conlon and Ali Naddaf

We have shown that (5.1) holds. Next consider the function h(ξ, η) defined by
(2.18) for ξ ∈ Rd , Re(η) > 0. Furthermore, (2.19) holds. We show now that there
is a constant Cλ,Λ,d depending only on λ, Λ, d such that if Re(η) > Cλ,Λ,d ε2 then
h(ξ, η) may be analytically continued to the strip {ξ ∈ Cd : |Im(ξ)| < ε} and
 ∞
(5.4) |h(ξ, η)|d[Im(η)] ≤ Cλ,Λ,d , |Im(ξ)| < ε .
0

To do this we need to rewrite the identities (2.14), (2.15) in such a way that they
extend analytically in ξ from ξ ∈ Rd to the strip |Im(ξ)| < ε. First consider (2.14).
We define a function Aj (ξ, η, ω) for ξ ∈ Rd , Re(η) > 0, ω ∈ Ω, by
1
Aj (ξ, η, ·) = ej (−ξ) + e−iej ·ξ [∂j + ej (ξ)] {ψ(ξ, η, ·) + ψ(ξ, η̄, ·)} ,
2
where ψ(ξ, η, ·) is defined just before (2.26). It is easy to see that the complex
conjugate Aj (ξ, η, ·) = Aj (−ξ, η, ·). We conclude from this and (2.14), (2.26) that


d 
(5.5) Re[e(ξ)q(ξ, η)e(−ξ)] = ai,j (·)Ai (−ξ, η, ·)Aj (ξ, η, ·)
i,j=1
Re(η)
+ [ψ(ξ, η, ·) + ψ(ξ, η̄, ·)] [ψ(−ξ, η, ·) + ψ(−ξ, η̄, ·)]
4
1  
+ [ψ(−ξ, η̄.·) − ψ(−ξ, η, ·)] [Lξ + Re(η)] [ψ(ξ, η, ·) − ψ(ξ, η̄, ·)] .
4
Similarly we have from (2.15) that
1
(5.6) Im[e(ξ)q(ξ, η)e(−ξ)] = Im(η) ψ(ξ, η, ·)ψ(−ξ, η̄, ·)
2
1
+ Im(η) ψ(−ξ, η, ·)ψ(ξ, η̄, ·) .
2
We write now h(ξ, η) as

(5.7)
Re(η) + Re[e(ξ)q(ξ, η)e(−ξ)]
h(ξ, η) = ,
[Re(η) + Re[e(ξ)q(ξ, η)e(−ξ)]]2 + [Im(η) + Im[e(ξ)q(ξ, η)e(−ξ)]]2

and use the expressions (5.5), (5.6) to analytically continue h(ξ, η), ξ ∈ Rd , to
complex ξ ∈ Cd . Note now that it follows from our proof of (5.1) that for any δ > 0
there exists a constant C(λ, Λ, d, δ) > 0 depending only on λ, Λ, d, δ such that if
Re(η) > C(λ, Λ, d, δ)ε2 then the function ψk (ξ, η, ·) of (2.2) from Rd to L2 (Ω) can
be analytically continued to the strip {ξ ∈ Rd : |Im(ξ)| < ε} and

(5.8)
 2 
−iej ·ξ
e [∂j + ej (ξ)]ψk (ξ, η, ·) − e−iej ·Re(ξ) [∂j + ej (Re(ξ))]ψk (Re(ξ), η, ·) ≤ δ,
 
|η| |ψk (ξ, η, ·) − ψk (Re(ξ), η, ·)|2 ≤ δ,
1 ≤ j, k ≤ d, |Im(ξ)| < ε, Re(η) > C(λ, Λ, d, δ)ε2 .
 Green’s Functions for Equations with Random Coefficients 207

It is easy to see from this and (5.5), (5.6), (5.7) that there is a constant C(λ, Λ, d)
depending only on λ, Λ, d such that
 
1 Re(η) + 4Λ|e(Re(ξ))|2
(5.9) |h(ξ, η)| ≤ 2 min , ,
Re(η) + λ|e(Re(ξ))|2 [Im(η)]2
|Im(ξ)| < ε, Re(η) > C(λ, Λ, d, δ)ε2 .
It is clear now from this last inequality that h(ξ, η) as defined by (5.7), (5.6), (5.5)
can be analytically continued to |Im(ξ)| < ε and that (5.4) holds.
Next consider the function k(ξ, η) defined by (2.18). We shall show that there is
a constant Cλ,Λ,d depending only on λ, Λ, d such that the function ∂k(ξ, η)/∂[Im(ξ)]
can be analytically continued to the strip |Im(ξ)| < ε, provided Re(η) > Cλ,Λ,d ε2 .
Furthermore there is the inequality
2
(5.10) |∂k(ξ, η)/∂[Im(η)]| ≤ , |Im(ξ)| < ε, Re(η) > Cλ,Λ,d ε2 .
|Im(η)|2
To see this we use the identity,
 
∂k(ξ, η) 1+ < ψ(−ξ, η, ·)ψ(ξ, η, ·) >
= Re , ξ ∈ Rd .
∂[Im(η)] [η + e(ξ)q(ξ, η)e(−ξ)]2
The inequality (5.10) follows now from the proof of (2.27) and the inequalities (5.8).
The proof of the lemma is completed by using the representation (2.17) with
Re(η) = Cλ,Λ,d ε2 and the inequalities (5.4), (5.10).

Corollary 5.1. There are constants C1 (d, λ, Λ) and C2 (d, λ, Λ) > 0 depending only
on d, λ, Λ such that


0 ≤ Ga (x, t) ≤ C1 (d, λ, Λ) exp −C2 (d, λ, Λ) min{|x|, |x|2 /t} , x ∈ Zd , t > 0.

Proof. Follows from Lemma 5.1 on writing


1
Ga (x, t) = Ĝa (ξ, t)e−iξ·x d[Re(ξ)] ,
(2π)d [−π,π]d
and deforming the contour of integration to ξ ∈ Cd with |Im(ξ)| = min[1, |x|/2C2 t],
where C2 = C2 (λ, Λ, d) is the constant in the statement of Lemma 5.1.

Next we extend Lemma 2.3 to ξ ∈ Cd .
Lemma 5.2. Let ε satisfy 0 ≤ ε < 1 and Ĝa (ξ, t), ξ ∈ Cd , |Im(ξ)| < ε the
function of Lemma 5.1. Then for any δ, 0 < δ < 1, there is a constant C1 (λ, Λ, d, δ)
depending only λ, Λ, d, δ and a constant C2 (λ, Λ, d) depending only on λ, Λ, d such
that
C1 (λ, Λ, d, δ)
|Ĝa (ξ, t)| ≤ , exp[C2 (λ, Λ, d)ε2 t], |Im(ξ)| < ε, t > 0.
[1 + |e(Re(ξ))|2 t]δ

Proof. Observe that one may choose Cλ,Λ,d sufficiently large, depending only on
λ, Λ, d such that both (5.10) holds and the inequality
2(1 + Λ)
(5.11) |∂k(ξ, η)/∂[Im(η)]| ≤ , |Im(ξ)| < ε, Re(η) > Cλ,Λ,d ε2 .
λ2 |e(Re(ξ))|2 |η|
The inequality (5.11) is analogous to (2.30) and is proved using (5.8), We con-
clude then from (5.10), (5.8) just as we did in Lemma 2.3 that the LHS of (2.28)
208 Joseph G. Conlon and Ali Naddaf

is bounded by C1 (λ, Λ, d, δ)/[1 + |e(Re(ξ))|2 t]δ if |Im(ξ)| < ε provided Re(η) >
Cλ,Λ,d ε2 . Since we can do similar estimates on ∂h(ξ, η)/∂[Im(η)] the result follows
just as in Lemma 2.3.

Corollary 5.2. The function Ga (x, t) satisfies the inequality
C1 (λ, Λ)

0 ≤ Ga (x, t) ≤ √ exp −C2 (λ, Λ) min{|x|, |x|2 /t} , if d = 1,
(1 + t)
where the constants C1 (λ, Λ) and C2 (λ, Λ) depend only on λ, Λ. For d > 1 it
satisfies an inequality,
C1 (λ, Λ, d, δ)

0 ≤ Ga (x, t) ≤ δ
exp −C2 (λ, Λ, d) min{|x|, |x|2 /t} ,
(1 + t )
for any δ, 0 ≤ δ < 1. The constant C1 (λ, Λ, d, δ) depends only on λ, Λ, d, δ and
C2 (λ, Λ, d) only on λ, Λ, d.

Proof. Same as for Corollary 4.1 on using the method of proof of Corollary 5.1
and Lemma 5.2.

Next we generalize Lemma 2.4.
Lemma 5.3. Let ε satisfy 0 < ε < 1 and Ĝa (ξ, t), ξ ∈ Cd , |Im(ξ)| < ε, be the
function of Lemma 5.1. Then Ĝa (ξ, t) is differentiable for t > 0. For any δ,
0 ≤ δ < 1, there is a constant C1 (λ, Λ, d, δ) depending only λ, Λ, d, δ and a constant
C2 (λ, Λ, d) depending only on λ, Λ such that

∂ Ĝ (ξ, t) C1 (λ, Λ, d, δ)
a
≤ exp[C2 (λ, Λ, d)ε2 t] , |Im(ξ)| < ε , t > 0 .
∂t t[1 + |e(Re(ξ))|2 t]δ

Proof. In analogy to the proof of the inequality (2.41) we see that there are con-
stants C1 (λ, Λ, d) > 0 and C2 (λ, Λ, d) > 0 such that
(5.12)
 
C1 (λ, Λ, d) 1 Re(η) + |e(Re(ξ))|2
|∂h(ξ, η)/∂[Im(η)]| ≤ min , ,
|Im(η)| Re(η) + |e(Re(ξ))|2 |Im(η)|2
|Im(ξ)| < ε, Re(η) > C2 (λ, Λ, d)ε2 .
It follows then just as in Lemma 2.4 that
 ∞
∂
(5.13) h(ξ, η) cos[Im(η)t]d[Im(η)] ≤ Cλ,Λ,d /t, |Im(ξ)| < ε .
∂t
0
We can also see that there are constants C1 (λ, Λ, d) > 0 and C2 (λ, Λ, d) > 0 such
that
2  
∂ h(ξ, η) C1 (λ, Λ, d) 1 1
(5.14) ≤ min , ,
∂[Im(η)]2 |Im(η)|2 Re(η) + |e(Re(ξ))|2 |Im(η)|

|Im(η)| < ε , Re(η) > C2 (λ, Λ, d)ε2 .

We conclude from this last inequality and (5.13) that for any δ > 0 there is a
constant C(λ, Λ, d, δ) such that
 ∞
∂ C(λ, Λ, d, δ)
h(ξ, η) cos[Im(η)t]d[Im(η)] ≤ , |Im(ξ)| < ε,
∂t t[1 + |e(Re(ξ))|2 t]δ
0
 Green’s Functions for Equations with Random Coefficients 209

provided Re(η) > C2 (λ, Λ, d)ε2 . Arguing similarly we see also that

∂ 1 ∞ ∂k(ξ, η) C(λ, Λ, d, δ)
{1 − cos[Im(η)t]}d[Im(η)] ≤
∂t t ∂[Im(η)] t[1 + |e(Re(ξ))|2 t]δ ,
0
|Im(ξ)| < ε, Re(η) > C(λ, Λ, d)ε2 .
The result of the lemma follows from these last two inequalities and

∂
exp[Re(η)t] ≤ 1 exp[2Re(η)t].
∂t t


Corollary 5.3. The function Ga (x, t) is differentiable with respect to t for t > 0
and satisfies the inequality

∂Ga (x, t)

≤ C1 (λ,√
Λ)
exp −C2 (λ, Λ) min{|x|, |x|2 /t} , if d = 1,
∂t t(1 + t)
where the constants C1 (λ, Λ) and C2 (λ, Λ) depend only on λ, Λ. For d > 1 it
satisfies an inequality,

∂Ga (x, t) C1 (λ, Λ, d, δ)

≤ exp −C2 (λ, Λ, d) min{|x|, |x|2 /t} ,
∂t δ
t[1 + t ]
for any δ, 0 ≤ δ < 1. The constant C1 (λ, Λ, d, δ) depends only λ, Λ, d, δ and
C2 (λ, Λ, d) only on λ, Λ, d.

Proof. Same as for Corollary 5.2 on using Lemma 5.3.

Lemma 5.4. There are constants C1 (λ, Λ, d) and C2 (λ, Λ, d) depending only on
λ, Λ, d such that

∂ Ĝ (ξ, t)
a
≤ C1 (λ, Λ, d)[|e(Re(ξ))|2 + ε2 ] exp[C2 (λ, Λ, d)ε2 t] , |Im(ξ)| < ε, t > 0.
∂t

Proof. In analogy to the inequalities (2.49) we have from (5.9), (5.12) the inequal-
ities
(5.15)

1 ∞
|h(ξ, η)|{1 − cos[Im(η)t]}d[Im(η)] ≤ C1 (λ, Λ, d) [|e(Re(ξ))|2 + Re(η)],
t 0

1 ∞ ∂h(ξ, η) 2
t ∂[Im(η)] |Im(η)|{1 − cos[Im(η)t]}d[Im(η)] ≤ C1 (λ, Λ, d)[|e(Re(ξ))| + Re(η)],
0

provided ξ ∈ Cd and η ∈ C satisfy

(5.16) |Im(ξ)| < ε, Re(η) > C2 (λ, Λ, d)ε2 ,
where C1 (λ, Λ, d) and C2 (λ, Λ, d) depend only on λ, Λ, d.
Next we need to deal with the integral (2.50) in k(ξ, η). Observe first from (5.8)
that there are constants C1 (λ, Λ, d) and C2 (λ, Λ, d) depending only on λ, Λ, d such
that if (5.16) holds then
|Im(η)||k(ξ, η)| ≤ C1 (λ, Λ, d), |Im(η)| ≤ Re(η) + |e(Re(ξ))|2 .
210 Joseph G. Conlon and Ali Naddaf

We can write k(ξ, η) as in Lemma 2.5 to be

b(ξ, η)
k(ξ, η) = ,
a(ξ, η)2 + b(ξ, η)2
where a(ξ, η) is the sum of Re(η) and the RHS of (5.5) whence b(ξ, η) is the sum
of Im(η) and the RHS of (5.6). We have now from (5.8) that
1
|b(ξ, η) − b(Re(ξ), η)| ≤ |Im(η)|, |Im(η)| > Re(η) + |e(Re(ξ))|2 ,
10
provided (5.16) holds and the constant C2 (λ, Λ, d) is sufficiently large. We also have
that
1
|a(ξ, η) − a(Re(ξ), η)| ≤ a(Re(ξ), η),
10
provided (5.16) holds and C2 (λ, Λ, d) is sufficiently large. It follows from these last
two inequalities that
 ∞
|Im(η)| |a(ξ, η)|2
2 2
d[Im(η)] ≤ C(λ, Λ, d)[Re(η)+|e(Re(ξ))|2 ]
Re(η)+|e(Re(ξ))|2 |b(ξ, η)| |a(ξ, η) + b(ξ, η) |

for some constant C(λ, Λ, d), provided (5.16) holds with sufficiently large constant
C2 (λ, Λ, d). Now the following lemma implies that
 ∞
 
|ψ(ξ, η, ·)|2 d[Im(η)] ≤ C(λ, Λ, d)[|e(Re(ξ))|2 + ε2 ] ,
0

for some constant C(λ, Λ, d) depending only on λ, Λ, d, provided (5.16) holds with
sufficiently large C2 (λ, Λ, d). We conclude then from these last inequalities that
(5.17)

πm/t

lim k(ξ, η)Im(η) cos[Im(η)t]d[Im(η)] ≤ C(λ, Λ, d)[Re(η) + |e(Re(ξ))|2 ] ,
m→∞ 0

for some constant C(λ, Λ, d) provided (5.16) holds. The Lemma follows now from
(5.15), (5.17).

Lemma 5.5. Let ψk (ξ, η, ·) be the function defined by (2.2), and 0 < ε < 1. There
is a constant C1 (λ, Λ, d) depending on λ, Λ, d such that if Re(η) > C1 (λ, Λ, d)ε2 then
ψk (ξ, η, ·), regarded as a mapping from Rd to L2 (Ω), can be analytically continued
to {ξ ∈ Cd : |Im(ξ)| < ε}. Furthermore, there is a constant C(λ, Λ, d) depending
only on λ, Λ, d such that
 ∞ 
|ψk (ξ, η, ·)|2 d[Im(η)] ≤ C2 (λ, Λ, d), Re(η) > C1 (λ, Λ, d)ε2 , |Im(η)| < ε.
0

Proof. We proceed as in Lemma 2.6. Let ψk (t, ξ, ·), t > 0, be the solution to the
initial value problem,
∂ψk (t, ξ, ·)
(5.18) + [Lξ + Re(η)]ψk (t, ξ, ·) = 0, t > 0,
∂t
d


ψk (0, ξ, ·) + [∂j∗ + ej (−ξ)]eiej ·ξ [ak,j (·) − ak,j (·)] = 0.
j=1
 Green’s Functions for Equations with Random Coefficients 211

It is clear that (5.18) is soluble for ξ ∈ Rd , Re(η) > 0, and

 ∞
(5.19) ψk (ξ, η, ·) = e−i Im(η)t ψk (t, ξ, ·)dt.
0
2
Taking Re(η) > C1 (λ, Λ, d)ε for sufficiently large C1 (λ, Λ, d) it is clear that one
can solve (5.18) for ξ ∈ Cd with |Im(ξ)| < ε and the resulting function is analytic
in ξ. Evidently the corresponding function ψk (ξ, η, ·) defined by (5.19) is analytic
in ξ for |Im(ξ)| < ε. Furthermore, from the Plancherel Theorem we have that
 ∞   ∞ 
2
|ψk (ξ, η, ·)| d[Im(η)] ≤ 2π |ψk (t, ξ, ·)|2 dt.
0 0
d
Now for ξ ∈ C let L∗ξ
be the adjoint of Lξ acting on L2 (Ω). Thus L∗ξ = Lξ̄ where
¯
ξ is the complex conjugate of ξ. Let ϕk (t, ξ, ·) be the solution of the equation
[L∗ξ + Re(η)]ϕk (t, ξ, ·) = ψk (t, ξ, ·), t > 0, |Im(ξ)| < ε.
It follows from (5.18) that
 
∂ψk (t, ξ, ·)  
(5.20) ϕk (t, ξ, ·) + |ψk (t, ξ, ·)|2 = 0, t > 0.
∂t
We also have that
(5.21)
     
∂ψk (t, ξ, ·) ∂ϕk ∂ ϕ̄k
ϕk (t, ξ, ·) = Re [Lξ + Re(η)]ϕk = Re [Lξ + Re(η)]ϕk
∂t ∂t ∂t
   
∂ϕk ∂ ϕ̄k ∗
= Re ([Lξ + Re(η)] ϕk + Re [Lξ − Lξ ]ϕk .
∂t ∂t
We conclude that
    1  ∂ ϕ̄ 
∂ψk 1 ∂ k ∗
ϕ̄k = Re [Lξ + Re(η)]ϕk ϕk + Re [Lξ − Lξ ]ϕk .
∂t 2 ∂t 2 ∂t
Observe now that
 
∂ ϕ̄k
[Lξ − L∗ξ ]ϕk =
∂t
 
− ψ̄k [L∗ξ + Re(η)][Lξ + Re(η)]−1 [Lξ − L∗ξ ][L∗ξ + Re(η)]−1 ψk .
Hence if C1 (λ, Λ, d) is sufficiently large one has
 
∂ ϕ̄k  
∗ 2
∂t [Lξ − Lξ ]ϕk ≤ |ψk (t, ξ, ·)| , t > 0, |Im(ξ)| < ε.
Putting this inequality together with (5.20), (5.21) we have that
1 ∂   1 
Re [Lξ + Re(η)]ϕk ϕk + |ψk (t, ξ, ·)|2 ≤ 0, |Im(ξ)| < ε, t > 0.
2 ∂t 2
Integrating this inequality with respect to t we conclude
 ∞  
 
|ψk (t, ξ, ·)|2 dt ≤ Re ϕk (0, ξ, ·)ψk (0, ξ, ·) .
0
Arguing as in Lemma 2.6 we see that the RHS of this last inequality is bounded by
a constant C(λ, Λ, d) depending only on λ, Λ, d.

212 Joseph G. Conlon and Ali Naddaf

Corollary 5.4. There are constants C1 (λ, Λ, d) and C2 (λ, Λ, d) > 0 depending only
on d, λ, Λ such that

∂Ga (x, t)

≤ C1 (d, λ, Λ) exp −C2 (d, λ, Λ) min{|x|, |x|2 /t} , x ∈ Zd , t > 0.
∂t

Proof. Same as for Corollary 5.1 on using Lemma 5.4.

Corollary 5.2 proves (1.10) for d = 1. Following the argument of Lemma 4.1 we
shall use our methods to prove (1.10) for d = 2.
Lemma 5.6. For d = 2 there are positive constants C1 (λ, Λ), C2 (λ, Λ) depending
only on λ, Λ such that
C1 (λ, Λ)

0 ≤ Ga (x, t) ≤ exp −C2 (λ, Λ) min{|x|, |x|2 /t} , x ∈ Z2 , t > 0.
1+t

Proof. It will be sufficient to show that there are constants C1 (λ, Λ), C2 (λ, Λ)
such that

C1 (λ, Λ)
(5.22) |Ĝa (ξ, t)|d[Re(ξ)] ≤ exp[C2 (λ, Λ)ε2 t], |Im(ξ)| < ε, t > 0.
[−π,π]2 1+t
In view of Lemma 5.1 it will be sufficient to prove (5.22) for t ≥ 1. It is also evident
from Lemma 5.1 that

C1 (λ, Λ)
√ |Ĝa (ξ, t)|d[Re(ξ)] ≤ exp[C2 (λ, Λ)ε2 t], |Im(ξ)| < ε, t > 1,
|e(Re(ξ))|<1/ t t
whence we are left to show that
(5.23)

C1 (λ, Λ)
√ |Ĝa (ξ, t)|d[Re(ξ)] ≤ exp[C2 (λ, Λ)ε2 t], |Im(ξ)| < ε, t > 1.
|e(Re(ξ))|>1/ t t
Now the integral on the LHS of (5.23) is a sum of an integral in h(ξ, η) and k(ξ, η).
We first consider the integral in h(ξ, η). Following the argument of Lemma 4.1 and
using (5.14) we see that it is sufficient to show that
 
1 |e(Re(ξ))|2 ∂ 2 h(ξ, η)

√ d[Re(ξ)] t2 ∂[Im(η)] 2
{1 − cos[Im(η)t]}d[Im(η)]

|e(Re(ξ))|>1/ t 0

C1 (λ, Λ)
≤ , |Im(ξ)| < ε, Re(η) > C2 (λ, Λ)ε2 ,
t
for sufficiently large constant C2 (λ, Λ) depending only on λ, Λ. Again, arguing as
in Lemma 4.1, we see it is sufficient to show that
(5.24)
  |e(Re(ξ))|2
1 < |ψ(ξ, η, ·)|2 >
√ d[Re(ξ)] t2 |e(Re(ξ))|4 Im(η)
{1 − cos[Im(η)t]}d[Im(η)]
|e(Re(ξ))|>1/ t 0
C1 (λ, Λ)
≤ , |Im(ξ)| < ε, Re(η) > C2 (λ, Λ)ε2 .
t
 Green’s Functions for Equations with Random Coefficients 213

Define for 2 < p < ∞ a function hη,p (ξ) by



2 
1/2 − 1/p
 2
 2
hη,p (ξ) = |Im(η)| |ψk (ξ, η, ·)| , ξ ∈ Cd , |Im(ξ)| < ε.
k=1
2
Suppose now Im(ξ) ∈ R is fixed and regard h2,p (ξ) as a function of Re(ξ) ∈
[−π, π]2 . Then one can see just as in Lemma 4.1 that if |Im(ξ)| < ε and Re(η) >
C2 (λ, Λ)ε2 for sufficiently large constant C2 (λ, Λ) depending only on λ, Λ, then
hη,p ∈ Lpw ([−π, π]2 ) and there is a constant Cp,λ,Λ such that
hη,p p,w ≤ Cp,λ,Λ , |Im(ξ)| < ε, Re(η) > C2 (λ, Λ)ε2 .
The inequality (5.24) follows from this last inequality just as in Lemma 4.1. Hence
we have found an appropriate bound for the contribution to the LHS of (5.23)
from the integral in h(ξ, η). The contribution from the integral in k(ξ, η) can be
estimated similarly.

We can similarly generalize Lemma 4.2 to obtain the following.
Lemma 5.7. For d = 2 there are positive constants C1 (λ, Λ), C2 (λ, Λ) depending
only on λ, Λ such that

∂Ga (x, t) C1 (λ, Λ)

≤ exp −C2 (λ, Λ) min{|x|, |x|2 /t} , x ∈ Z2 , t > 0.
∂t (1 + t2 )
Next we wish to consider derivatives of Ga (x, t) with respect to x ∈ Zd . If d = 1
then it is clear from Corollary 5.2 that
C1 (λ, Λ)

|∇i Ga (x, t)| ≤ √ exp −C2 (λ, Λ) min{|x|, |x|2 /t} , x ∈ Z2 , t > 0.
(1 + t)
We can use Lemma 5.2 to obtain an improvement on this inequality.
Lemma 5.8. Suppose d = 1 and 0 ≤ δ < 1. Then there exist constants C1 (λ, Λ, δ)
depending only on λ, Λ, δ and C2 (λ, Λ) depending only on λ, Λ such that
C1 (λ, Λ, δ)

|∇i Ga (x, t)| ≤ δ
exp −C2 (λ, Λ) min{|x|, |x|2 /t} , x ∈ Z2 , t > 0.
(1 + t )

Proof. The result follows from Lemma 5.2 and the observation that
 π
|e(Re(ξ))| C(δ  , δ)
2 δ
d[Re(ξ)] ≤ ,
−π [1 + |e(Re(ξ))| t] 1 + tδ
for any δ  , δ satisfying 1/2 < δ < δ  < 1, where C(δ  , δ) is a constant depending
only on δ  , δ.

We can improve Lemma 5.8 by using the techniques developed in Section 3.
Lemma 5.9. Suppose d = 1 and 0 ≤ δ < 1. Then there are constants C1 (λ, Λ),
C2 (λ, Λ) depending only on λ, Λ and a constant C3 (λ, Λ, δ) depending on λ, Λ, δ
such that
C1 (λ, Λ)

(5.25) |∇i Ga (x, t)| ≤ exp −C2 (λ, Λ) min{|x|, |x|2 /t} ,
(1 + t)
(5.26)
C3 (λ, Λ, δ)

|∇i ∇j Ga (x, t)| ≤ exp −C2 (λ, Λ) min{|x|, |x|2 /t} , x ∈ Z2 , t > 0.
[1 + t1+δ/2 ]
214 Joseph G. Conlon and Ali Naddaf

Proof. It will be sufficient to show that for any δ, 0 ≤ δ < 1, there are constants
C3 (λ, Λ, δ) and C2 (λ, Λ) such that
 π
C3 (λ, Λ, δ)
(5.27) |e(Re(ξ))|1+δ |Ĝa (ξ, t)|d[Re(ξ)] ≤ 1+δ/2
exp[C2 (λ, Λ)ε2 t],
−π 1 + t

for |Im(ξ)| < ε, t > 1. Now from Lemma 5.1 we have


1+δ C3 (λ, Λ)
√ |e(Re(ξ))| |Ĝa (ξ, t)|d[Re(ξ)] ≤ exp[C2 (λ, Λ)ε2 t],
|e(Re(ξ))|<1/ t 1 + t1+δ/2

for |Im(ξ)| < ε, so we are left to prove

(5.28)

C3 (λ, Λ, δ)
√ |e(Re(ξ))|1+δ |Ĝa (ξ, t)|d[Re(ξ)] ≤ exp[C2 (λ, Λ)ε2 t],
|e(Re(ξ))|>1/ t t1+δ/2

for |Im(ξ)| < ε, t > 1. Proceeding now as in Lemma 5.6 we write the integral on the
LHS of (5.28) as an integral in h(ξ, η) and an integral in k(ξ, η). We first consider
the integral in h(ξ, η). If we use (5.14) we see that it is sufficient to show that

(5.29) √ d[Re(ξ)]|e(Re(ξ))|1+δ
|e(Re(ξ))|>1/ t

1 |e(Re(ξ))|2 ∂ 2 h(ξ, η)

2 {1 − cos[Im(η)t]}d[Im(η)]
t 0 ∂[Im(η)]2
C3 (λ, Λ, δ)
≤ , |Im(ξ)| < ε, Re(η) > C2 (λ, Λ)ε2 , 0 ≤ δ < 1,
t1+δ/2
for sufficiently large C2 (λ, Λ) depending only on λ, Λ. Arguing as in Lemma 5.6 we
see that to prove (5.29) it is sufficient to show that

(5.30) √ d[Re(ξ)]|e(Re(ξ))|1+δ
|e(Re(ξ))|>1/ t
 |e(Re(ξ))|2
1 < |ψ(ξ, η, ·)|2 >
{1 − cos[Im(η)t]}d[Im(η)]
t2 0 |e(Re(ξ))|4 Im(η)
C3 (λ, Λ, δ)
≤ , |Im(ξ)| < ε, Re(η) > C2 (λ, Λ)ε2 , 0 ≤ δ < 1.
t1+δ/2
Define for 2 < p < ∞ a function hη,p (ξ) by
 1/2
hη,p (ξ) = |Im(η)|1/2−1/2p |ψ1 (ξ, η, ·)|2 , ξ ∈ Cd , |Im(ξ)| < ε.

We fix now Im(ξ) ∈ R and regard hη,p (ξ) as a function of Re(ξ) ∈ [−π, π]. Then
one can see just as in Lemma 5.6 that if |Im(ξ)| < ε and Re(η) > C2 (λ, Λ)ε2 for
sufficiently large constant C2 (λ, Λ) depending only on λ, Λ, then hη,p ∈ Lpw ([−π, π])
and there is a constant Cp,λ,Λ such that

hη,p p,w ≤ Cp,λ,Λ , |Im(ξ)| < ε, Re(η) > C2 (λ, Λ)ε2 .

 Green’s Functions for Equations with Random Coefficients 215

Arguing as in Lemma 4.1 we see that (5.30) holds if

  |e(Re(ξ))|2
1 hη,p (ξ)2
√ d[Re(ξ)] d[Im(η)]
|e(Re(ξ))|>1/ t |e(Re(ξ))|1−δ t1/2+2/p 0 [Im(η)]1/2+1/p
C3 (λ, Λ, δ)
≤ , t > 1.
t1+δ/2
The LHS of this last expression is bounded by

∞  √ 1−δ  22n+2 /t 
C t d[Im(η)]
hη,p (ξ)2 d[Re(ξ)]
t1/2+2/p n=0 2n 0 [Im(η)]1/2+1/p |e(ξ))|<2n+1 /√t
∞  √ 1−δ  2n+2 1/2 − 1/p  n+1 1 − 2/p
Cp,λ,Λ
t 2 2 C(λ, Λ, δ)
≤ 1/2+2/p n
√ ≤ ,
t n=0
2 t t t1+δ/2

provided we choose p to satisfy 2 < p < 4/(1 + δ). Hence the contribution to the
LHS of (5.28) from h(ξ, η) is bounded appropriately. Since we can similarly estimate
the contribution from k(ξ, η) we have proved (5.28) and hence (5.27). Now (5.25)
follows by taking δ = 0 in (5.27), and (5.26) by taking δ close to 1.

We have proven Theorem 1.4 for d = 1. It is clear by now that we can use the
methods developed in Section 4 to extend the results of Lemmas 5.6, 5.7, 5.9 to all
dimensions d ≥ 1.

6. Proof of Theorem 1.2

For η > 0, x ∈ Rd , let Gη (x) be the Green’s function which satisfies the equation,
d

∂ 2 Gη (x)
− + ηGη (x) = δ(x), x ∈ Rd ,
i=1
∂x2i

where δ(x) is the Dirac δ function. Analogously to (3.4) we define an operator

Tk,k
,η,ξ on L2 (Ω) by

∂ 2 Gη (x) −ix·ξ
(6.1) Tk,k
,η,ξ ϕ(ω) = − dx e ϕ(τx ω), ω ∈ Ω.
Rd ∂xk ∂xk

Evidently Tk,k
,η,ξ is a bounded operator on L2 (Ω). Just as in Section 3 we can
define operators Tb,η,ξ and Tj,η,ξ , j = 1, . . . , d associated with the operators (6.1).
We then have the following.
Lemma 6.1. Let Tb,η,ξ and Tj,η,ξ , j = 1, . . . , d, be the operators associated with
(6.1). Then if b < 1 the equation (3.6) has a unique solution Ψk (ξ, η, ·) ∈
H(Ω), ξ ∈ Rd , η > 0, which satisfies an inequality,
Ψk (ξ, η, ·) ≤ C(λ, Λ, d)/[1 − b], k = 1, . . . , d,
where the constant C(λ, Λ, d) depends only on λ, Λ, d. The function (ξ, η) →
Ψk (ξ, η, ·) from Rd × R+ to H(Ω) is continuous.
Next we put b(·) = [ΛId − a(·)]/Λ and define a d × d matrix q(ξ, η) by (3.7),
where Ψk (ξ, η, ·), k = 1, 2, . . . , d are the functions of Lemma 6.1.
216 Joseph G. Conlon and Ali Naddaf

Lemma 6.2. For ξ ∈ Rd , η > 0, the matrix q(ξ, η) is Hermitian and is a contin-
uous function of (ξ, η). Furthermore, there is the inequality
(6.2) λId ≤ q(ξ, η) ≤ ΛId .
Proof. We use the operators Tk,k
,η,ξ of (6.1) to define an operator Tη,ξ on H(Ω).
Thus if ϕ = (ϕ1 , . . . , ϕd ) ∈ H(Ω) then
d


(Tη,ξ ϕ)k = Tk,k
,η,ξ ϕk
.
k
=1

It is easy to see that Tη,ξ is a bounded self-adjoint operator on H(Ω) which is

nonnegative definite and has norm Tη,ξ  ≤ 1. We can see this by using Bochner’s
Theorem as in Lemma 3.3. Now for ρ ∈ Cd , we put Ψρ to be
d


Ψρ = ρk Ψk ,
k=1

where Ψk is the solution of (3.6). Then from (3.7) we have that

ρ̄q(ξ, η)ρ = ρ̄a(·)ρ + ρ̄a(·)Ψρ (ξ, η, ·) .
It is also clear from (3.6) that Ψρ satisfies
1
(6.3) Ψρ (ξ, η, ω) − P Tb,η/Λ, ξ Ψρ (ξ, η, ω) +Tη/Λ, ξ [a(ω)ρ− < a(·)ρ >] = 0.
Λ
To obtain the upper bound in (6.2) we observe that
 
(6.4) ρ̄a(·)Ψρ (ξ, η, ·) ≤ 0.
To prove (6.4) we generate Ψρ from (6.3) by a Neumann series. The first order
approximation to the solution is then
1
Ψρ (ξ, η, ·)  − Tη/Λ, ξ [a(ω)ρ − a(·)ρ] .
Λ
Putting this approximation into (6.4) yields
  
1
ρa(·) − Tη/Λ, ξ [a(ω)ρ − a(·)ρ
Λ
1 
=− ρ[a(·) − a]Tη/Λ, ξ [a(·) − a]ρ ≤ 0,
Λ
since Tη/Λ,ξ is nonnegative definite. One can similarly argue that each term in the
Neumann series makes a negative contribution to (6.4).
To prove the lower bound in (6.2) we use the fact that
2
Tη,ξ = Tη,ξ [1 − ηAη,ξ ],
where Aη,ξ is the bounded operator on L2 (Ω) defined by

Aη,ξ φ(ω) = dxGη (x)e−ix·ξ φ(τx ω).
Rd
Note that Aη,ξ is self-adjoint, nonnegative definite, and commutes with Tη,ξ . It
follows now from (6.3) that
 η 
Tη/Λ, ξ Ψρ = 1 − Aη/Λ, ξ Ψρ .
Λ
 Green’s Functions for Equations with Random Coefficients 217

Hence if we multiply (6.3) by Ψ̄ρ and take the expectation value we have that
  η  
Ψρ (ξ, η, ·)2 − Ψρ (ξ, η, ·)b(·)Ψρ (ξ, η, ·) + Ψρ (ξ, η, ·)Aη/Λ, ξ b(·)Ψρ (ξ, η, ·)
Λ
1  η  
+ Ψρ (ξ, η, ·) 1 − Aη/Λ, ξ [a(·)ρ − a ρ] = 0.
Λ Λ
If we define the operator K by
η
K = 1 − Aη/Λ, ξ ,
Λ
then the previous equation can be written as
  1 
Ψρ (ξ, η, ·)2 = KΨρ (ξ, η, ·)b(·)Ψρ (ξ, η, ·) − KΨρ (ξ, η, ·)[a(·) − λId ]ρ .
Λ
Applying the Schwarz inequality to this last equation, we obtain
1  1 
Ψρ (ξ, η, ·)2 ≤ KΨρ (ξ, η, ·)b(·)KΨρ (ξ, η, ·) + Ψρ (ξ, η, ·)b(·)Ψρ (ξ, η, ·)
2 2
1   1
+ KΨρ (ξ, η, ·)[a(·) − λId ]KΨρ (ξ, η, ·) + ρ̄[a(·) − λId ]ρ .
2Λ 2Λ
Observing now that K is also nonnegative definite and bounded above by the
identity, we see from this last inequality that
 
Ψρ (ξ, η, ·)a(·)Ψρ (ξ, η, ·) ≤ ρ̄[a(·) − λId ]ρ .

The lower bound in (6.2) follows now from the Schwarz inequality on writing
ρ̄a(·)Ψρ (ξ, η, ·) = ρ̄[a(·) − λId ]Ψρ (ξ, η, ·) .


We have defined the functions Ψk (ξ, η, ·), k = 1, . . . , d corresponding to the
solutions of (3.6). Next we wish to define functions ψk (ξ, η, ·) corresponding to
the solutions of (2.2). To do this we consider an equation adjoint to (3.6). Since
∗ ∗
Tb,η,ξ = Tη,ξ b(·), the adjoint Tb,η,ξ of Tb,η,ξ is Tb,η,ξ = b(·)Tη,ξ . For k = 1, . . . , d
∗
let Ψk (ξ, η, ω) ∈ H(Ω) be the solution to the equation,
1
(6.5) Ψ∗k (ξ, η, ω) − P Tb,η/Λ,
∗ ∗
ξ Ψk (ξ, η, ω) +
[ak (ω) − ak (·)] = 0,
Λ
where ak (ω) is the k th column vector of the matrix a(ω). Just as in Lemma 6.1
we see that Ψ∗k regarded as a mapping from Rd × R+ to H(Ω) is continuous. We
also define an operator Sη,ξ : H(Ω) → L2 (Ω) by

d 
∂Gη (x) −ix·ξ
(6.6) Sη,ξ ϕ(ω) = dx e ϕk
(τx ω),

R d ∂xk

k =1

where ϕ = (ϕ1 , . . . , ϕd ) ∈ H(Ω). Evidently Sη,ξ is a bounded operator. We define

the functions ψk (ξ, η, ω), k = 1, . . . , d, then by
(6.7) ψk (ξ, η, ω) = Sη/Λ, ξ Ψ∗k (ξ, η, ω), ω ∈ Ω,
where Ψ∗k (ξ, η, ω) is the solution to (6.5). It is easy to see that there is a constant
C(λ, Λ, d) depending only on λ, Λ, d, such that
√
(6.8) ψk (ξ, η, ·) ≤ C(λ, Λ, d)/ η .
218 Joseph G. Conlon and Ali Naddaf

Let us define Ĝa (ξ, η) similarly to (2.4) by

1
(6.9) Ĝa (ξ, η) = , ξ ∈ Rd , η > 0,
η + ξq(ξ, η)ξ
where the matrix q(ξ, η) is as in Lemma 6.2. Suppose now f : Rd → R is a C ∞
function with compact support and Fourier transform fˆ(ξ). We define v̂(ξ, η, ω) to
be
 d


v̂(ξ, η, ω) = 1 − i ξk ψk (ξ, η, ω) Ĝa (ξ, η)fˆ(ξ).
k=1

Let η > 0 be fixed. Then v̂(ξ, η, ω), regarded as a function of ξ ∈ Rd to L2 (Ω), is

continuous and rapidly decreasing. Hence the Fourier inverse

1
v(x, η, ω) = dξe−ix·ξ v̂(ξ, η, ω), x ∈ Rd ,
(2π)d Rd
regarded as a mapping of x ∈ Rd to L2 (Ω) is C ∞ . In particular it follows that
v(x, η, ω), regarded as a function of (x, ω) ∈ Rd × Ω to C, is measurable and in
L2 (Rd × Ω). Define now the function u(x, η, ω) = v(x, η, τx ω). It is clear that
u(x, η, ω), regarded as a function of (x, ω) ∈ Rd × Ω to C is measurable and in
L2 (Rd × Ω), with the same norm as v.
Lemma 6.3. With probability one the function u(x, η, ·), x ∈ Rd , is in L2 (Rd ) and
its distributional gradient ∇u(x, η, ·) is also in L2 (Rd ). Furthermore u(x, η, ·) is a
weak solution of the equation

d  
∂ ∂u
(6.10) − ai,j (τx ·) (x, η, ·) + ηu(x, η, ·) = f (x), x ∈ Rd .
i,j=1
∂xi ∂xj

Proof. Since u(x, η, ω) ∈ L2 (Rd × Ω), it follows that with probability one u(x, η, ·),
x ∈ Rd is in L2 (Rd ). To see that the distributional gradient of u(x, η, ·) is also in
L2 (Rd ) with probability one, we shall establish a formula for ∇u(x, η, ·). To do this
we define for any C ∞ function of compact support g : Rd → C, an operator Ag,ξ
on L2 (Ω) by

Ag,ξ ϕ(ω) = dxg(x)e−ix·ξ ϕ(τx ω), ϕ ∈ L2 (Ω).
Rd

Evidently Ag,ξ is a bounded operator on L2 (Ω). Suppose now Ψk (ξ, η, ·) ∈ H(Ω) is

the function of Lemma 6.1 with components Ψk = (Ψk,1 , . . . , Ψk,d ) and ψk (ξ, η, ·)
is given by (6.7). Then
(6.11) A∇j g,ξ ψk (ξ, η, ·) = −Ag,ξ Ψk,j (ξ, η, ·),

where ∇j g is the j th partial of g. To see that (6.11) holds, observe that if Sη,ξ is
the operator of (6.6) then for ϕ ∈ H(Ω),
d


(6.12) A∇j g,ξ Sη,ξ ϕ = −Ag,ξ Tj,k
,η,ξ ϕk
,
k
=1
 Green’s Functions for Equations with Random Coefficients 219

where ϕ = (ϕ1 , . . . , ϕd ) and Tj,k
,η,ξ are the operators (6.1). The identity (6.11)
follows now from (6.12) by observing that Ψk (ξ, η, ·) = Tη/Λ, ξ Ψ∗k (ξ, η, ·) . We have
now that
 
(6.13) − dx∇j g(x)u(x, η, ·) = − dx∇j g(x)v(x, η, τx ·)
Rd Rd

1
=− dξA∇j g,ξ v̂(ξ, η, ·)
(2π)d Rd

1
= dξAg,ξ v̂j (ξ, η, ·),
(2π)d Rd
where
 d


(6.14) v̂j (ξ, η, ω) = −i ξj + ξk Ψk,j (ξ, η, ω) Ĝa (ξ, η)fˆ(ξ).
k=1

Now we put,

1
vj (x, η, ω) = dξe−ix·ξ v̂j (ξ, η, ω).
(2π)d Rd

It is clear that vj (x, η, ω), regarded as a function of (x, ω), is in L2 (Rd × Ω), whence
vj (x, η, τx ω) is also in L2 (Rd × Ω). It follows now from (6.13) that the function
∇j u(x, η, ω) = vj (x, η, τx ω) is in L2 (Rd ) with probability 1 in ω and is the distri-
butional derivative of u(x, η, ω).
Next we wish to show that with probability 1, u(x, η, ·) is a weak solution of the
equation (6.10). To do that we need to observe that for any ϕ ∈ H(Ω) and C ∞
function g : Rd → C of compact support, one has
d

d


(6.15) A∇j g,ξ T j,k
η,ξ ϕ k
= −ηAg,ξ Sη,ξ ϕ + A∇k
g,ξ ϕk
.
j,k
=1 k
=1

We have now, for any C ∞ function g : Rd → R with compact support,

(6.16)
 d

 d


dx ∇i g(x)ai,j (τx ·)∇j u(x, η, ·) = dx ∇i g(x)ai,j (τx ·)vj (x, η, τx ·)
Rd i,j=1 Rd i,j=1
 d


1
= dξ A∇i g,ξ [ai,j (·)v̂j (ξ, η, ·)] .
(2π)d Rd i,j=1

Observe next that for any k, 1 ≤ k ≤ d,

d

d


A∇i g,ξ [ai,j (·)Ψk,j (ξ, η, ·)] = Λ A∇i g,ξ Ψk,i (ξ, η, ·)
i,j=1 i=1
d


−Λ A∇i g,ξ [bi,j (·)Ψk,j (ξ, η, ·)] .
i=1
220 Joseph G. Conlon and Ali Naddaf

If we use the fact that Ψk = Tη/Λ,ξ Ψ∗k and (6.15), then we see that
d


d
η
A∇i g,ξ Ψk,i (ξ, η, ·) = − Ag,ξ Sη/Λ,ξ Ψ∗k (ξ, η, ·) + A∇i g,ξ Ψ∗k,i (ξ, η, ·).
i,j=1
Λ i=1

We also have from (6.5) that

b(ω)Ψk (ξ, η, ω) = b(ω)Tη/Λ,ξ Ψ∗k (ξ, η, ω)
∗ ∗
= P Tb,η/Λ, ξ Ψk (ξ, η, ω) + b(·)Ψk (ξ, η, ·)
1
= Ψ∗k (ξ, η, ω) + [ak (ω) − ak (·)] + b(·)Ψk (ξ, η, ·)
Λ
1 1
= Ψ∗k (ξ, η, ω) + ak (ω) − ak (·) + a(·)Ψk (ξ, η, ·) .
Λ Λ
It follows now from the last three equations that
d

d


A∇i g,ξ [ai,j (·)Ψk,j (ξ, η, ·)] = −ηAg,ξ ψk (ξ, η, ·) − A∇j g,ξ ak,j (·)
i,j=1 j=1
d


+ iĝ(ξ) ξj qj,k (ξ, η).
j=1

Hence from (6.14), (6.16) we have that

 d


dx ∇i g(x)ai,j (τx ·)∇j u(x, η, ·)
Rd i,j=1
 
η 1
=− dξA g,ξ v̂(ξ, η, ·) + dξ[η + ξq(ξ, η)ξ]Ĝa (ξ, η)fˆ(ξ)ĝ(ξ)
(2π)d Rd (2π)d Rd
 
= −η dxg(x)u(x, η, ·) + dxg(x)f (x),
Rd Rd

where we have used (6.9). The result follows from this last equation.

Next, let Ga (x, y, η, ·), x, y ∈ Rd be the Green’s function for the equation (6.10).
It follows easily now from Lemma 6.3 that if Ga (x, η) is the Fourier inverse of the
function Ĝa (ξ, η) of (6.9), then
 
Ga (x − y, η) = Ga (x, y, η, ·) .

We can now use the methods of Section 3 to estimate Ga (x, η). We shall restrict
ourselves to the case d = 3 since the method generalizes to all d ≥ 3. Evidently
one can generalize Lemma 3.6 to obtain:
Lemma 6.4. Let d = 3, η > 0, 1 ≤ k, k  ≤ d. Then qk,k
(ξ, η) is a C ∞ function
of ξ ∈ Rd and for any i, j, 1 ≤ i, j ≤ d, the function ∂qk,k
/∂ξi ∈ L3w (Rd ) and
3/2
∂ 2 qk,k
/∂ξi ∂ξj ∈ Lw (Rd ). Further, there is a constant Cλ,Λ , depending only on
λ, Λ such that
∂qk,k
/∂ξi 3,w ≤ Cλ,Λ , ∂ 2 qk,k
/∂ξi ∂ξj 3/2, w ≤ Cλ,Λ .
 Green’s Functions for Equations with Random Coefficients 221

We can use Lemma 6.4 to estimate Ga (x, η). To do this suppose ϕ : Rd → R is

a C ∞ function of compact support satisfying

ϕ(y)dy = 1.
Rd
d
For R > 0 we put ϕR (y) = R ϕ(Ry). Evidently the Fourier transform of ϕR is a
rapidly decreasing function and ϕ̂R (ξ) = ϕ̂(ξ/R), where ϕ̂(0) = 1. We define now
the function Ga (x, η) by

1 ϕ̂R (ξ)e−ix·ξ
(6.17) Ga (x, η) = lim d
dξ .
R→∞ (2π) Rd [η + ξq(ξ, η)ξ]
Lemma 6.5. Let d = 3. The function Ga (x, η) defined by (6.17) is continuous
in (x, η), x ∈ Rd \{0}, η > 0. The limit Ga (x) = limη→0 Ga (x, η) exists for all
x ∈ Rd \{0} and Ga (x) is a continuous function of x, x
= 0. Further, there is a
constant Cλ,Λ depending only on λ, Λ such that
(6.18) 0 ≤ Ga (x, η) ≤ Cλ,Λ /|x|, x ∈ Rd \{0}, η > 0.
Proof. We argue as in Proposition 3.1. Thus for γ satisfying 1 ≤ γ ≤ 2, we write
 
(6.19) Ga (x, η) = lim + lim .
R→∞ |ξ|<γ/|x| R→∞ |ξ|>γ/|x|

It is clear that
 
1 e−ix·ξ
lim = dξ .
R→∞ |ξ|<γ/|x| (2π)d |ξ|<γ/|x| [η + ξq(ξ, η)ξ]
To evaluate the limit as R → ∞ in the second integral on the RHS of (6.19) we
integrate by parts. Thus for fixed R > 0, assuming x1
= 0,
   
1 1 −ix·ξ ∂ ϕ̂(ξ/R)
(6.20) = dξe .
|ξ|>γ/|x| ix1 (2π)d |ξ|>γ/|x| ∂ξ1 η + ξq(ξ, η)ξ

1 1 e−ix·ξ ϕ̂(ξ/R)ξ1
+ d
dξ .
ix1 (2π) |ξ|=γ/|x| [η + ξq(ξ, η)ξ]|ξ|
Evidently for the surface integral in the last expression one has
 
1 1 e−ix·ξ ξ1
lim = dξ .
R→∞ |ξ|=γ/|x| ix1 (2π)d |ξ|=γ/|x| [η + ξq(ξ, η)ξ]|ξ|
To evaluate the limit of the volume integral in (6.20) as R → ∞, we need to
integrate by parts again. Thus, for the integral over {|ξ| > γ/|x|} on the RHS of
(6.20) one has
  2
 
−1 1 −ix·ξ ∂ ϕ̂(ξ/R)
(6.21) = 2 dξe
|ξ|>γ/|x| x1 (2π)d |ξ|>γ/|x| ∂ξ12 η + ξq(ξ, η)ξ
  
1 1 −ix·ξ ∂ ϕ̂(ξ/R) ξ1
− 2 dξe .
x1 (2π)d |ξ|=γ/|x| ∂ξ1 η + ξq(ξ, η)ξ |ξ|
In view of Lemma 6.4 it follows that the limit of the volume integral on the RHS
of (6.21) is given by
  2
 
−1 1 −ix·ξ ∂ 1
lim = 2 dξe .
R→∞ |ξ|>γ/|x| x1 (2π)d |ξ|>γ/|x| ∂ξ12 η + ξq(ξ, η)ξ
222 Joseph G. Conlon and Ali Naddaf

We can similarly see that the limit of the average of the surface integral on the
RHS of (6.21) is given by


lim Av =
R→∞ |ξ|=γ/|x|
1≤γ≤2
'    (
−1 1 −ix·ξ ∂ 1 ξ1
Av dξe .
x21 (2π)d |ξ|=γ/|x| ∂ξ1 η + ξq(ξ, η)ξ |ξ|
1≤γ≤2

We have therefore established a formula for the function Ga (x, η). It easily follows
from this that Ga (x, η) is continuous in (x, η), x ∈ R3 \{0}, η > 0. To show that
the function Ga (x) = lim Ga (x, η) exists we observe that q(ξ, η) converges as η → 0
η→0
to a function q(ξ, 0). This follows from the fact that the operators Tk,k
,η,ξ of (6.1)
converge strongly as η → 0 to bounded operators Tk,k
,0,ξ on L2 (Ω). One can prove
this last fact by using Bochner’s Theorem. Suppose p satisfies 1 < p < 3. Then
∂qk,k
(ξ, η)/∂ξi can be written as a sum,

∂qk,k
(ξ, η)/∂ξi = fk,k
,i (ξ, η) + gk,k
,i (ξ, η).

The function fk,k
,i (ξ, η) ∈ L3w (Rd ) and converges in L3w (Rd ) as η → 0 to the
distributional derivative ∂qk,k
(ξ, 0)/∂ξi of the function ∂qk,k
(ξ, 0). The function
gk,k
,i (ξ, η) ∈ Lp (Rd ) and converges as η → 0 in Lp (Rd ) to 0. This follows by
writing

η[|ξ|2 + η]−1 = f (ξ, η) + g(ξ, η),

where f ∈ L∞ (Rd ) f (·, η)∞ ≤ η 1/2 and g ∈ L∞ (Rd ) g(·, η)∞ ≤ 1, g(ξ, η) = 0
if |ξ| > η 1/4 . One can also make a similar statement about convergence of the
derivative ∂ 2 qk,k
(ξ, η)/∂ξi ∂ξj as η → 0. We conclude that one can take the limit
as η → 0 in the integral formula we have established for Ga (x, η). In view of
Lemmas 6.2 and 6.4 the limiting function Ga (x) is also continuous for x
= 0.
Finally the inequality (6.18) follows by exactly the same argument as we used in
Proposition 3.1.

We can complete the proof of Theorem 1.2 by applying the argument for the
proof of Theorem 1.5 at the end of Section 3.

Lemma 6.6. Let d = 3 and Ga (x) be the function defined in Lemma 6.5. Then
Ga (x) is a C 1 function for x
= 0 and there is a constant Cλ,Λ , depending only on
λ, Λ such that

∂Ga (x) Cλ,Λ
(6.22)
∂xi ≤ |x|2 , x
= 0, i = 1, 2, 3.
 Green’s Functions for Equations with Random Coefficients 223

Proof. Let r > 0 and suppose x ∈ R3 satisfies 10r < |x| < 20r. From Lemma 6.5
we have that
 2 
1 e−ix·ξ
(6.23) Ga (x) = dγ dξ
(2π)3 1 |ξ|<γ/r ξq(ξ, 0)ξ
 2 
1 1 e−ix·ξ ξ1
+ 3
dγ dξ
ix1 (2π) 1 |ξ|=γ/r ξq(ξ, 0)ξ |ξ|
 2   
1 1 −ix·ξ ∂ 1 ξ1
− 2 3
dγ dξe
x1 (2π) 1 |ξ|=γ/r ∂ξ1 ξq(ξ, 0)ξ |ξ|
 2  2
 
1 1 −ix·ξ ∂ 1
− 2 dγ dξe ,
x1 (2π)3 1 |ξ|>γ/r ∂ξ12 ξq(ξ, 0)ξ

where q(ξ, 0) is defined in Lemma 6.5. Let Ha (x) be the final integral on the RHS of
(6.23). Then it follows from Lemmas 6.2 and 6.4 that if x1 ∼ |x| then Ga (x)−Ha (x)
is a C 1 function and

∂ Cλ,Λ
[G (x) − H (x)] ≤ , x
= 0, j = 1, 2, 3.
∂xj a a |x|2
To show the differentiability of Ha (x) we expand
  
3 2
∂2 1 −2q1,1 (ξ, 0) 8
(6.24) = + Re q1,j (ξ, 0)ξj ,
∂ξ12 ξq(ξ, 0)ξ [ξq(ξ, 0)ξ]2 [ξq(ξ, 0)ξ]3 j=1

plus terms involving derivatives of q(ξ, 0). The contribution of the first term on the
RHS of (6.24) to Ha (x) is given by
 2 
2 1 q1,1 (ξ, 0)
(6.25) dγ dξe−ix·ξ
x21 (2π)3 1 |ξ|>γ/r [ξq(ξ, 0)ξ]2
 2 
2 1 q1,1 (ξ, 0) ξ1
= 3 3
dγ dξe−ix·ξ
ix1 (2π) 1 |ξ|=γ/r [ξq(ξ, 0)ξ]2 |ξ|
 2   
2 1 −ix·ξ ∂ q1,1 (ξ, 0)
+ 3 dγ dξe .
ix1 (2π)3 1 |ξ|>γ/r ∂ξ1 (ξq(ξ, 0)ξ)2

Using the fact that ∂q1,1 /∂ξ1 ∈ L3w (R3 ), it is easy to see that the RHS of (6.25) is
a C 1 function of x, x
= 0, and its derivative is bounded by the RHS of (6.22). The
same argument can be used to estimate the contribution to Ha (x) from all terms
on the RHS of (6.24) except the term involving the second derivative of q(ξ, 0). The
contribution to Ha (x) from this term is given by Ka (x)/x21 , where
 2 
1 1 ξ∂ 2 q(ξ, 0)ξ
Ka (x) = dγ dξe−ix·ξ .
3
(2π) 1 |ξ|>γ/r [ξq(ξ, 0)ξ]2 ∂ξ12

We can see now just as in Lemma 3.10 that for any ρ ∈ R3 , the function
3/(3−ε)
[∂ 2 q(ξ + ρ, 0)/∂ξ12 − ∂ 2 q(ξ, 0)/∂ξ12 ]/|ρ|1−ε is in Lw (R3 ) and

(6.26) [∂ 2 q(ξ + ρ, 0)/∂ξ12 − ∂ 2 q(ξ, 0)/∂ξ 2 ]/|ρ|1−ε 3/(3−ε),w ≤ Cλ,Λ,ε ,

224 Joseph G. Conlon and Ali Naddaf

where ε is any number satisfying 0 < ε < 1. For h ∈ R3 we write

 2 
1
(6.27) Ka (x + h) − Ka (x) = dγ dξe−ix·ξ [gh (ξ) − gh (ξ + ρ)]
(2π)3 1 |ξ|>γ/r

f (ξ)
+ dξe−ix·ξ [e−ih·ξ − 1] ,
1/4r<|ξ|<4/r |ξ|2

where
1 ξ∂ 2 q(ξ, 0)ξ
g(ξ) = , gh (ξ) = [e−ih·ξ − 1]g(ξ) ,
2[ξq(ξ, 0)ξ]2 ∂ξ12
3/2
f ∈ Lw (R3 ) and ρ ∈ R3 has the property that x · ρ = π. In view of (6.26) and the
3/2
fact that f ∈ Lw (R3 ) it follows from the Dominated Convergence Theorem that
one can take the limit in (6.27) as h → 0 to obtain that Ka (x) is differentiable in
x and
 2 
∂Ka (x) −i
= dγ dξe−ix·ξ {ξj [g(ξ) − g(ξ + ρ)] − ρj g(ξ + ρ)}
∂xj (2π)3 1 |ξ|>γ/r

f (ξ)
− i dξe−ix·ξ ξj 2 .
1/4r<|ξ|<4/r |ξ|

We can see from this last expression that ∂Ka (x)/∂xj is continuous in x and also
|∂Ka (x)/∂xj | ≤ Cλ,Λ .

References
[1] D. G. Aronson, Non-negative solutions of linear parabolic equations, Ann. Sci. Norm. Sup.
Pisa (3) 22 (1968), 607–694, MR 55 #8553, Zbl 182.13802.
[2] K. Astala and M. Miettinen, On Quasiconformal Mappings and 2-Dimensional G-Closure
Problems, Arch. Rational Mech. Anal. 143 (1998), 207–240, MR 2000f:74068, Zbl 912.65106.
[3] E. A. Carlen, S. Kusuoka and D. Stroock, Upper bounds for symmetric Markov transition
functions, Ann. Inst. H. Poincaré 23 (1987), 245–87, MR 88i:35066, Zbl 634.60066.
[4] J. Conlon and A. Naddaf, On homogenisation of elliptic equations with random coefficients,
Electronic Journal of Probability 5 (2000), paper 9, 1–58, CMP 1 768 843.
[5] E. B. Davies, Heat Kernels and Spectral Theory, Cambridge Tracts in Mathematics 92, Cam-
bridge University Press, Cambridge-New York, 1989, MR 90e:35123, Zbl 699.35006.
[6] A. Fannjiang and G. Papanicolaou, Diffusion in Turbulence, Prob. Theory Relat. Fields 105
(1996), 279–334, MR 98d:60156, Zbl 847.60062.
[7] C. Landim, S. Olla and H. T. Yau, Convection-diffusion equation with space-time ergodic
random flow, Probab. Theory Relat. Fields 112 (1998), 203–220, MR 99j:35084, Zbl 914.60070.
[8] G. Papanicolaou and S. Varadhan, Boundary value problems with rapidly oscillating random
coefficients, Random Fields, Vol. 2 of Coll. Math. Soc. Janos Bolya, no. 27, North Holland,
Amsterdam, 1981, pp. 835–873, MR 84h:82005, Zbl 499.60059.
[9] M. Reed and B. Simon, Methods of Modern Mathematical Physics II: Fourier Analysis, Self-
Adjointness, Academic Press, New York-London, 1975, MR 58 #12429b, Zbl 308.47002.
[10] E. Stein and G. Weiss, Introduction to Fourier Analysis on Euclidean Spaces, Princeton
Mathematical Series, No. 32, Princeton University Press, Princeton, N.J., 1971, MR 46 #4102,
Zbl 232.42007.
[11] V. Zhikov, S. Kozlov and O. Oleinik, Homogenization of Differential Operators and Integral
Functionals, Springer Verlag, Berlin, 1994, MR 96h:35003b.
 Green’s Functions for Equations with Random Coefficients 225

Department of Mathematics, University of Michigan, Ann Arbor, MI 48109

[email protected] http://www.math.lsa.umich.edu/˜conlon/

Department of Mathematics, University of Michigan, Ann Arbor, MI 48109

[email protected]

This paper is available via http://nyjm.albany.edu:8000/j/2000/6-10.html.