Notes of Szemer´edi’s Theorem for

length 4
1 Some notations
In this paper, C will denote various large absolute constants, and c will
denote various small absolute constants.
[a, b] denotes the integers from a to b inclusive. Similarly define the
[a, b), etc. If E is a set, [E[ will denote the cardinality of E. N will be
0-0
a large integer, and 0 < δ < 1 will be a parameter, and ε = ε(δ) > 0 be
a quantity depending continuously on δ.
Let M = M(N) be a slowly growing function of N such that ε
−C
<
M < N
c
; one could take for instance M = log(N), although one can
get away with a much larger M actually. The exact choice of M is only
important if one is interested in the dependence of N on δ to be chosen
later.
For any function f on Z
N
, define the Fourier transform
ˆ
f by
ˆ
f :=

m∈Z
N
f(m) exp(2πimn/N),
and define the Wiener norm |f|
A
∞ by
|f|
A
∞ := |
ˆ
f|
l

(Z
N
)
= sup
n∈Z
N
[
ˆ
f(n)[,
we also define the modified A

norm |f|
˙
A

by
|f|
˙
A

:= |
ˆ
f|
l

(Z
N
−{0})
.
We write Y X if Y is not of the form o(X).
0-1
2 The main idea of the proof
Szemer´edi asserts that for any integers k, if N is sufficiently large de-
pending on δ and k, then every set E ⊂ [0, N] with [E[ δN must
contain a non-trivial arithmetic progression of length k.
We are going to deal with the case k = 4.
The main idea is to show by induction. Let S be the set of all
0 < δ < 1 such that the Szemer´edi theorem hold for k = 4. It is
easy to see that all the δ > 3/4 belong to S, since there is at least one
of a, a + 1, a + 2, a + 3 lies in A, furthermore, we shall prove that if
δ +ε(δ) ∈ S then δ ∈ S.
Assume these, we can give the proof of the theorem.
Proof. Consider the infimum δ
0
of S, now claim δ
0
= 0. If not, δ
0
> 0,
and since δ
0
+ε(δ
0
) ∈ S, we get δ
0
∈ S by induction. By the continuity
of ε, we get
δ +ε(δ) → δ
0
+ε(δ
0
), δ → δ
0
−.
0-2
Therefore there exists δ
1
< δ
0
such that
δ
1
+ε(δ
1
) = δ
0
,
we get δ
1
∈ S by induction, contradict to the infimum of δ
0
.The proof is
done.
Fix δ, let N be a prime sufficiently large depending on δ and ε, and
let E ⊂ [0, N) be such that [E[ δN, by refining E if necessary we may
assume that
[E[ = δN (1)
we can assume that
[E ∩ P[ (δ +ε)[P[ (2)
for all arithmetic progressions P ⊂ [0, N) of length M by induction.
This is the only place where we shall use the induction hypothesis.
0-3
3 Proof of the induction above
Lemma 3.1. We have
[E ∩ P[ δ[P[ +O(ε[P[) (3)
for all Z
N
arithmetic progression P of length [P[ M
3
Proof. Let r = 0 be the spacing of an arithmetic progression P in Z/NZ.
(i) if 0 < [r[ < N/M. Then we can partition P into the disjoint union of
genuine arithmetic progressions of the same step size r, simply by making
a cut every time the progression wraps around the end of 1, , N.
Except possibly for the first and last progression, each of the genuine
progressions has length at least M, and so the density of A in those
genuine progressions is at most δ +ε by (2). Adding up all these density
estimates for those genuine progressions of length at least M, we obtain
the bound
[E ∩ P[ (δ +ε)[P[ + 2M.
0-4
(ii) Assume r is arbitrary. Consider the first O(M) multiples of r in Z/NZ.
By the pigeonhole principle, two of them must be < N/M apart, thus
we can find a non-zero j = O(M) such that [jr mod N[ < N/M. We
can then partition P into [j[ disjoint progressions, all of length ∼ [P[/[j[,
with step size jr mod N. Applying the previous estimate to all of these
progressions and adding up, we obtain
[E ∩ P[ (δ +ε)[P[ + 2[j[M (δ +ε)[P[ +O(M
2
),
and the claim follows if [P[ is large enough.
We can show that (1) and (3) will imply
|1
E
|
˙
A

= O(εN), (4)
and

k∈Z
N
|1
E∩(E+k)
|
˙
A

= O(εN
2
). (5)
0-5
Assume (4) and (5) hold, we show that E contains ∼ δ
4
N
2
+ O(ε
1
4
N
2
)
genuine arithmetic progression of length 4, and the induction follows.
It suffices to show that

a,r∈Z
N
1
E
(a)1
E
(a +r)1
E
(a + 2r)1
E
(a + 3r) ∼ δ
4
N
2
+O(ε
1
4
N
2
) (6)
where E

= E ∩ [0.4N, 0.6N], since the r = 0 contribution is negligible
if N is sufficiently large. From (1) and (3), we see that
[E`E

[ = [E ∩ (0.6N, 1.4N)[ (δ +ε)0.8N
hence we have
[E

[ = δ

N
for some δ

∼ δ. Write
f(a) =

r∈Z
N
1
E
(a +r)1
E
(a + 2r)1
E
(a + 3r).
It suffices to show that

a∈E

(f(a) −δ

δ
2
N) = O(ε
1
4
N
2
).
0-6
By Cauchy-Schwarz, it suffices to show that

a∈E

(f(a) −δ

δ
2
N)
2
= O(ε
1
2
N
3
).
We need to show that

a∈Z
N
(f(a) −δ

δ
2
N)
2
= O(ε
1
2
N
3
).
Lemma 3.2. For any a, b, c ∈ Z
N
−¦0¦ we have

k,l,m∈Z
N
:ak+bl+cm=0
f
1
(k)f
2
(l)f
3
(m) |f
1
|
2
|f
2
|
2
|f
3
|
A
∞ (7)
Similarly for permutations.
Proof. We can rewrite the left hand side of (7) as
1
N

n∈Z
N
ˆ
f
1
(an)
ˆ
f
2
(bn)
ˆ
f
3
(cn),
0-7
which can then be estimated using Cauchy-Schwarz by
1
N
|
ˆ
f
1
|
2
|
ˆ
f
2
|
2
|f
3
|
A
∞.
The claim then follows from Plancherel’s theorem.
Corollary 3.3. For any a, b, c ∈ Z
N
`¦0¦, we have

k,l,m∈Z
N
:ak+bl+cm=0
f
1
(k)f
2
(l)f
3
(m) = N
2
E(f
1
)E(f
2
)E(f
3
)+O(|f
1
|
2
|f
2
|
2
|f
3
|
˙
A

).
(8)
Similarly for permutations.
Proof. Note that
|f|
˙
A

= |f −E(f)|
A
∞. (9)
Then the corollary can be got from Lemma 3.2, (9)and the identity

k,l,m∈Z
N
:ak+bl+cm=0
(f
1
(k) −E(f
1
))(f
2
(l) −E(f
2
))(f
3
(m) −E(f
3
))
=

k,l,m∈Z
N
:ak+bl+cm=0
f
1
(k)f
2
(l)f
3
(m) −N
2
E(f
1
)E(f
2
)E(f
3
)
0-8
.
Proposition 3.4. For any F ⊂ Z
N
, we have

u∈Z
N
[[F ∩ (E +u)[ −δ[F[[ = O(εN
2
). (10)
Proof. It suffices to show

u∈A
([F ∩ (E +u)[ −δ[F[) = O(εN
2
)
for arbitrary A, F ⊂ Z
N
. From (4) and Lemma 3.2, we see:
[

u∈A
([F ∩ (E +u)[ −δ[F[)[ = [

u,k∈Z
N
1
A
(u)1
F
(k)(1
E
(k −u) −δ)[
|1
A
|
2
|1
F
|
2
|1
E
−δ|
A

N|1
E
|
˙
A

= O(εN
2
),
and the claim follows.
0-9
Remark: In Lemma 3.4, if we take F = E, we get

u∈Z
N
[[E ∩ (E +u)[ −δ
2
N[ = O(εN
2
).
Hence we get for all but at most O(ε
1
2
N) values of u such that
[E ∩ (E +u)[ = δ
2
N +O(ε
1
2
N). (11)
By Lemma 3.4, take F = 2E = ¦2e : e ∈ E¦, we get
0-10

a∈Z
N
(f(a) −δ

δ
2
N) =

a∈Z
N
(

r∈Z
N
1
E
(a +r)1
E
(a + 2r)1
E
(a + 3r) −δ

δ
2
N)
=

b∈Z
N
(

r∈Z
N
1
E
(b)1
E
(b +r)1
E
(b + 2r) −δ

δ
2
N)
=

b∈Z
N

r∈Z
N
1
E
(b)1
E
(b +r)1
E
(b + 2r) −δ

δ
2
N
2
=

b∈E

(

r∈Z
N
1
E
(b +r)1
E
(b + 2r) −δ
2
N)
=

b∈E

(

u∈Z
N
1
E
(u)1
E
(2u −b) −δ
2
N)
=

b∈E

([2E ∩ (E +b)[ −δ
2
N)
= O(εN
2
).
0-11
Hence we have

a∈Z
N
f(a) = δ

δ
2
N
2
+O(εN
2
).
Therefore

a∈Z
N
(f(a) −δ

δ
2
N)
2
= O(ε
1
2
N
3
)
⇐⇒

a∈Z
N
f
2
(a) −2δ

δ
2
N

a∈Z
N
f(a) +δ
2
δ
4
N
3
= O(ε
1
2
N
3
)
⇐⇒

a∈Z
N
f
2
(a) = δ
2
δ
4
N
3
+O(ε
1
2
N
3
)
0-12
By Corollary 3.3 and (5),

a∈Z
N
f
2
(a)
=

a∈Z
N

r∈Z
N

s∈Z
N
1
E
(a +r)1
E
(a +s)1
E
(a + 2r)1
E
(a + 2s)1
E
(a + 3r)1
E
(a + 3s)
=

a∈Z
N

u∈Z
N

r∈Z
N
1
E

∩(E

+u)
(a +r)1
E∩(E+2u)
(a + 2r)1
E∩(E+3u)
(a + 3r)
=

u∈Z
N

b,r∈Z
N
1
E

∩(E

+u)
(b)1
E∩(E+2u)
(b +r)1
E∩(E+3u)
(b + 2r)
=

u∈Z
N
N
−1
[E

∩ (E

+u)[[E ∩ (E + 2u)[[E ∩ (E + 3u)[ +O(N

u∈Z
N
|1
E∩(E+3u)
|
˙
A

)
=N
−1

u∈Z
N
[E

∩ (E

+u)[[E ∩ (E + 2u)[[E ∩ (E + 3u)[ +O(εN
3
)
Hence it suffices to show

u∈Z
N
[E

∩ (E

+u)[[E ∩ (E + 2u)[[E ∩ (E + 3u)[ = δ
2
δ
4
N
4
+O(ε
1
2
N
4
)
0-13
By the remark after the lemma 3.4,
[E ∩ (E + 2u)[ = δ
2
N +O(ε
1
2
N),
[E ∩ (E + 3u)[ = δ
2
N +O(ε
1
2
N)
for all but at most O(ε
1
2
N) values of u. Therefore

u∈Z
N
[E

∩ (E

+u)[[E ∩ (E + 2u)[[E ∩ (E + 3u)[
=

u∈Z
N
[E

∩ (E

+u)[(δ
2
N +O(ε
1
2
N))
2
+O(ε
1
2
N
4
)
=(δ
4
N
2
+O(ε
1
2
N
2
))

u∈Z
N
[E

∩ (E

+u)[ +O(ε
1
2
N
4
)
=(δ
4
N
2
+O(ε
1
2
N
2
))[E

[
2
+O(ε
1
2
N
4
)
=(δ
4
N
2
+O(ε
1
2
N
2
))δ
2
N
2
+O(ε
1
2
N
4
)

4
δ
2
N
4
+O(ε
1
2
N
4
).
That’s the end of the proof, all the left is to show (4) and (5).
0-14
3.1 The proof of (4)
Proposition 3.5. We have

s∈Z
N
[

j∈E∩(P
0
+s)
exp(2πiφ
s
(j)/N)[ = O(εN[P
0
[) (12)
for any Z
N
arithmetic progression P
0
of length M
C
and any non-
constant quadratic functions φ
0
, φ
1
, , φ
N−1
with integer coefficients.
Assume this proposition, we show that (4) is the corollary of it.
Corollary 3.6. |1
E
|
˙
A

= O(εN).
Proof : Assume Proposition 3.5, we have
[

s∈Z
N

j∈E∩(P
0
+s)
exp(2πiφ
s
(j)/N)[ = O(εN[P
0
[).
0-15
For a fixed 1 n N −1, take φ
s
(j) = nj, we get:
O(εN[P
0
[) = [

s∈Z
N

j∈E∩(P
0
+s)
exp(2πinj/N)[
= [

s∈Z
N

j∈P
0
+s
1
E
(j) exp(2πinj/N)[
= [[P
0
[

j∈Z
N
1
E
(j) exp(2πinj/N)[
= [[P
0
[
ˆ
1
E
(n)[,
and therefore we get (4) since n is non-zero. The lemma is done.
We shall prove Proposition 3.5. Fix the functions φ
0
, φ
1
, , φ
N−1
and the Z
N
arithmetic progression P
0
, we can reduce to the case when
P
0
is the arithmetic progression [0, r) for some M
C
< r < N since N is
prime.
Definition 3.7. We say Z
N
arithmetic progressions P
1
, P
2
, , P
n
evenly
cover Z
N
if the function

n
j=1
1
P
j
is a constant on Z
N
.
0-16
Lemma 3.8. Let P
1
, P
2
, , P
n
be Z
N
arithmetic progression of length
M
3
which evenly cover Z
N
. Then we have
n

j=1
[[E ∩ P
j
[ −δ[P
j
[[ = O(ε
n

j=1
[P
j
[). (13)
Proof. From (3), we have
[E ∩ P
j
[ −δ[P
j
[ = O(ε[P
j
[),
if [E ∩ P
j
[ −δ[P
j
[ 0. Write f =

n
j=1
1
P
j
, then
n

j=1
([E ∩ P
j
[ −δ[P
j
[) =

x∈E
f −δ
n

j=1
[P
j
[ =
[E[
N
N

k=1
f −
[E[
N
n

j=1
[P
j
[ = 0
with the identity
n

j=1
[P
j
[ =
n

j=1
N

k=1
1
P
j
(k) =
N

k=1
n

j=1
1
P
j
(k) =
N

k=1
f.
0-17
Hence,
n

j=1
[[E ∩ P
j
[ −δ[P
j
[[ = 2
n

j=1
([E ∩ P
j
[ −δ[P
i
[)
+
= O(ε
n

j=1
[P
i
[).
Lemma 3.9. For every s ∈ Z
N
, we can partition P
0
+s = [0, r) +s into
arithmetic progressions P
s,1
, , P
s,n
s
of length [P
s,j
[ M
3
such that
for each 1 j n
s
, the function exp(2πiφ
s
()/N) is within o(1) of a
constant on P
s,j
, j = 1, , n
s
.
Assume this lemma for the moment and complete the proof of (12).
Proof of (12): Firstly we observe that when s ranges over Z
N
and
j ranges from 1 to n
s
, the progressions P
s,j
evenly cover Z
N
. By Lemma
3.8, we have

s∈Z
N
n
s

j=1
[[E ∩ P
s,j
[ −δ[P
s,j
[[ = O(ε

s∈Z
N
[P
s,j
[) = O(εN[P
0
[). (14)
0-18
By Lemma 3.9, we have

k∈E∩P
s,j
exp(2πiφ
s
(k)/N)
=
[E ∩ P
s,j
[
[P
s,j
[

k∈P
s,j
exp(2πiφ
s
(k)/N) +o([P
s,j
[)
=
[E ∩ P
s,j
[ −δ[P
s,j
[ +δ[P
s,j
[
[P
s,j
[

k∈P
s,j
exp(2πiφ
s
(k)/N) +o([P
s,j
[)

k∈P
s,j
exp(2πiφ
s
(k)/N) +O([[E ∩ P
s,j
[ −δ[P
s,j
[[) +o([P
s,j
[).
Summing this in j and taking absolute value, we get
0-19
[

k∈E∩(P
0
+s)
exp(2πiφ
s
(k)/N)[
=δ[

k∈P
0
+s
exp(2πiφ
s
(k)/N)[ +O(
n
s

j=1
[[E ∩ P
s,j
[ −δ[P
s,j
[[) +o([P
0
[).
Summing the above equation in s and using (14) we have

s∈Z
N
[

k∈E∩(P
0
+s)
exp(2πiφ
s
(k)/N)[

s∈Z
N
[

k∈P
0
+s
exp(2πiφ
s
(k)/N)[ +O(

s∈Z
N
n
s

j=1
[[E ∩ P
s,j
[ −δ[P
s,j
[[) +o(N[P
0
[)

s∈Z
N
[

k∈P
0
+s
exp(2πiφ
s
(k)/N)[ +O(εN[P
0
[)
From standard exponential sum estimates and the non-constancy of φ
s
0-20
we observe that

k∈P
0
+s
exp(2πiφ
s
(k)/N) = o([P
0
[). (15)
and (12) follows from that.
It remains to prove Lemma 3.9. Fix s, without loss of generality we
can assume that s = 0. It suffices to partition [0, r) into Z
N
arithmetic
progression P
0,j
of length ∼ r
c
for some small c such that
[[
φ
0
(k

) −φ
0
(k)
N
[[ = o(1)
for all k, k

∈ P
0,j
. Let d
j
be the step size and k
j
the initial element of
P
0,j
. Assume φ
0
(k) = A
2
k
2
+A
1
k +A
0
, then
φ
0
(k
j
+ad
j
) −φ
0
(k
j
+bd
j
) = (a
2
−b
2
)A
2
d
2
j
+ (a −b)d
j
(2A
2
k
j
+A
1
).
It suffices to find d
j
such that
[[
a
2
A
2
d
2
j
N
[[ = o(1),
0-21
[[
(2A
2
k
j
+A
1
)d
j
a
N
[[ = o(1)
hold for all [a[ r
c
. It thus suffices to obtain the conditions
[[
A
2
d
2
j
N
[[, [[
(2A
2
k
j
+A
1
)d
j
N
[[ = O(r
−3c
). (16)
By a quantitative version of Weyl’s theorem (Weyl: Let k 2,
t 2
2
32k
2
, N t and a ∈ Z
N
. Then there exists p t such that
[p
k
a/N[ t
−1/k2
k+1
. Here we take k = 2, t = r
c
7
.)
We can find d r
c
7
such that
[[
A
2
d
2
N
[[ = o(r
−100c
).
Fix this d and partition [0, r) into arithmetic progressions P
1
, P
2

of step d and length ∼ r
5c
. For each such progression P
t
, let k
t
be an
element of it. By pigeonhole principle we can find an integer m
t
= O(r
3c
)
such that
[[
(2A
2
k
t
+A
1
)m
t
d
N
[[ = O(r
−3c
).
0-22
Now partition each P
t
into arithmetic progressions P
0,j
of step m
t
d and
length ∼ r
5c
/m
t
M
3
. The first condition of (16) is clear. To verify
the latter we observe that
k
j
= k
t
+ud
for some integer u = O(r
5c
), and thus
[[
(2A
2
k
j
+A
1
)m
t
d
N
[[ = [[
(2A
2
k
t
+A
1
)m
t
d
N
+
2A
2
um
t
d
2
N
[[
[[
(2A
2
k
t
+A
1
)m
t
d
N
[[ +[[
2A
2
um
t
d
2
N
[[
= O(r
−3c
) +O([[
A
2
d
2
N
[[r
5c
r
3c
)
= O(r
−3c
)
as desired.
0-23
3.2 The proof of (5)
Lemma 3.10. For any Z
N
arithmetic progression P of length M
C
<
[P[ = o(N) and (λ, µ) ∈ Z
N
Z
N
−¦(0, 0)¦ we have

k∈P
[
ˆ
1
E∩(E+k)
(λk +µ)[
2
= O(εN
2
[P[). (17)
Assume this lemma, we can now prove (5). Suppose for contradiction
that (5) failed, we get ∀c > 0, ∀N
0
> 0, ∃N > N
0
such that

k∈Z
N
|1
E∩(E+k)
|
˙
A

2cεN
2
.
For an arbitrary fixed c and N. We can find a set B ⊂ Z
N
of size
[B[ cεN such that
[[1
E∩(E+k)
[[
˙
A

cεN.
In fact, let B be the set of all such k that
[[1
E∩(E+k)
[[
˙
A

cεN.
0-24
Hence we have
2cεN
2

k∈Z
N
|1
E∩(E+k)
|
˙
A

=

||1
E∩(E+k)
||
˙
A

<cεN
|1
E∩(E+k)
|
˙
A

+

||1
E∩(E+k)
||
˙
A

cεN
|1
E∩(E+k)
|
˙
A

k∈Z
N
cεN +N

||1
E∩(E+k)
||
˙
A

cεN
1
= cεN
2
+N[B[.
Therefore [B[ cεN.
We can therefore find a function φ : B → Z
N
−¦0¦ such that
[
ˆ
1
E∩(E+k)
(φ(k))[ cεN (18)
for all k ∈ B. Fix B, φ, and let Ω denote the graph of φ:
Ω := ¦(k, φ(k)) : k ∈ B¦.
Ω has considerable arithmetic structure:
0-25
Lemma 3.11. We have
¦(a, b, c, d) ∈ Ω
4
: a +b = c +d¦ (cε)
12
N
3
Proof. Square (18) and use the identity
[
ˆ
1
A
(n)[
2
=

u∈Z
N

s∈A∩(A+u)
exp(2πiun/N) (19)
for all A ⊂ Z
N
and n ∈ Z
N
, we have

u∈Z
N

s∈E∩(E+k)∩(E+u)∩(E+k+u)
exp(2πiuφ(k)/N) c
2
ε
2
N
2
.
Summing this over k ∈ B,
c
3
ε
3
N
3

k∈B

u∈Z
N

s∈E∩(E+k)∩(E+u)∩(E+k+u)
exp(2πiuφ(k)/N)
=

u∈Z
N

k∈Z
N

s∈E∩(E+k)∩(E+u)∩(E+k+u)
1
B
(k) exp(2πiuφ(k)/N)
=

u∈Z
N

k∈Z
N

s∈Z
N
1
E∩(E+u)
(s)1
E∩(E+u)
(s −k)1
B
(k) exp(2πiuφ(k)/N)
0-26
By Lemma 3.2, we have
c
3
ε
3
N
3

k∈B

u∈Z
N

s∈E∩(E+k)∩(E+u)∩(E+k+u)
exp(2πiuφ(k)/N)

u∈Z
N
[[1
E∩(E+u)
[[
2
[[1
E∩(E+u)
[[
2
[[1
B
() exp(2πiuφ()/N)[[
A

N

u∈Z
N
[[1
B
() exp(2πiuφ()/N)[[
A
∞.
Hence,

u∈Z
N
[[1
B
() exp(2πiuφ()/N)[[
A
∞ c
3
ε
3
N
2
.
0-27
Since

u∈Z
N
[[1
B
() exp(2πiuφ()/N)[[
A
∞ c
3
ε
3
N
2
=

u∈Z
N
max
m∈Z
N
[

s∈Z
N
[[1
B
(s) exp(2πiuφ(s)/N) exp(2πims/N)[
=

u∈Z
N
max
m∈Z
N
[

(s,t)∈Z
N
×Z
N
1

(s, t) exp(2πi(ut +ms)/N)[
=

u∈Z
N
max
m∈Z
N
[
ˆ
1

(m, u)[,
we have

u∈Z
N
sup
m∈Z
N
[
ˆ
1

(m, u)[ c
3
ε
3
N
2
.
From H¨older, we thus have

u∈Z
N
sup
m∈Z
N
[
ˆ
1

(m, u)[
4
(cε)
12
N
5
0-28
and hence

u,m∈Z
N
[
ˆ
1

(m, u)[
4
(cε)
12
N
5
which by Plancherel becomes
[[1

∗ 1

[[ (cε)
12
N
3
as desired.
With this lemma and use Balog -Szemer´edi theorem which asserts
that we can find a subset Ω

⊂ Ω with [Ω

[ > c
ε
N and
[Ω

+ Ω

[ C
ε
N.
Remark. Balog-Szemer´edi theorem: Let A = ¦a
1
, a
2
, , a
n
¦ be a
finite set of integers and if ¦(a
i
, a
j
, a
k
, a
l
) ∈ A
4
[a
i
+a
j
= a
k
+a
l
¦ cN
3
,
then there are positive integers c
1
and c
2
and a subset A

⊂ A such that
[A

[ c
1
N and [A

+A

[ c
2
N.
Apply Freiman’s theorem to conclude that Ω

is contained in a proper
multi-dimensional Z
N
Z
N
arithmetic progression P
1
+P
2
+ +P
d
= Q
0-29
of dimension d
ε
< 1 and cardinality at most C

ε
N. Assume [P
1
[ [P
i
[
for all 1 i d.
Claim that there exists a ∈ Z
N
Z
N
such that
[(a +P
1
) ∩ Ω[ c
ε
C
−1
ε
[P
1
[,
or else, we have
[Ω ∩ Q[

a∈P
2
+···+P
d
[(a +P
1
) ∩ Q[

a∈P
2
+···+P
d
c
ε
C
−1
ε
[P
1
[
= c
ε
C
−1
ε
[P
1
[[P
2
[ [P
d
[
c
ε
C
−1
ε
C

ε
N
= c
ε
N.
However observe that Ω

⊂ Ω ∩ Q we get [Ω ∩ Q[ [Ω

[ > c
ε
N, which
is a contradiction. Fix such a a, since Ω is a graph, a +P
1
can not be a
0-30
vertical line segment, hence there exists a Z
N
arithmetic progression P
0
of length N
1
d
ε
, and

k∈P
0
[
ˆ
1
E∩(E+k)
(φ(k))[
2
=

k∈P
0
[
ˆ
1
E∩(E+k)
(λk +µ)[
2
c
2
ε
2
N
2
[P
0
[
contradict to Lemma 3.10.
We have already shown (5) , all the remains is to prove lemma 3.10.
Proof of lemma 3.10: In the case λ = 0. From (19) and rewrite
the left hand side of (17) as

k∈P

u∈Z
N

s∈E∩(E+k)∩(E+u)∩(E+k+u)
exp(2πiu(λk +µ)/N).
Using the identity

u∈Z
N
f(u) =
1
[P[

u∈Z
N

l∈P
f(u +l),
0-31
[

u∈Z
N

l∈P
f(u +l) =

k∈Z
N
f(k)

u∈k−P
1 = [P[

k∈Z
N
f(k)]
we get

k∈P
[
ˆ
1
E∩(E+k)
(λk +µ)[
2
=
1
[P[

k∈P

u∈Z
N

l∈P

s∈E∩(E+k)∩(E+u+l)∩(E+k+u+l)
exp(2πi(u +l)(λk +µ)/N)
=
1
[P[

s∈E

u∈Z
N

k∈P

l∈P
1
E+k
(s)1
E+u+l
(s)1
E+k+u+l
(s) exp(2πi(u +l)(λk +µ)/N)
=
1
[P[

s∈E

u∈Z
N

k∈Z
N

l∈Z
N
1
(s−E)∩P
(k)1
(s−u−E)∩P
(l)1
(s−u−E)∩(P+P)
(k +l) exp(2πi(u +l)(λk +µ)/N)

1
[P[

s∈E,u∈Z
N
[

k∈Z
N
,l∈Z
N
f
1,s
(k)f
2,s−u
(l)f
3,s−u
(k +l)[

1
[P[

s∈E,u∈Z
N
[[f
1,s
[[
A
∞[[f
2,s−u
[[
2
[[f
3,s−u
[[
2
,
0-32
where
f
1,s
(k) = 1
(s−E)∩P
(k) exp(2πiφ
1
(k)/N),
f
2,s−u
(l) = 1
(s−u−E)∩P
(l) exp(2πiφ
2
(l)/N),
f
3,s−u
(k +l) = 1
(s−u−E)∩(P+P)
(k +l) exp(2πiφ
3
(k +l)/N),
φ
1
(k), φ
2
(l), φ
3
(k +l) are some non-affine quadratic functions such that
φ
1
(k) +φ
2
(l) +φ
3
(k +l) = (u +l)(λk +µ).
Since
[[f
2,s−u
[[
2
2
= [

l∈P∩(s−u−E)
exp(4πiφ
2
(l)/N)[ [P[
and
[[f
3,s−u
[[
2
2
[P +P[ 2[P[
0-33
we get

k∈P
[
ˆ
1
E∩E+k
(λk +µ)[
2

1
[P[

s∈E

u∈Z
N
[P[[[f
1,s
[[
A
∞ = N

s∈E
[[f
1,s
[[
A
∞.
It suffices to show

s∈E
[[f
1,s
[[
A
∞ = O(εN[P[).
From proposition 3.5, we see

s∈E
[[f
1,s
[[
A

=

s∈E
[

m∈Z
N
f
1,s
(m) exp(2πimn(s)/N)[

s∈Z
N
[

m∈(s−E)∩P
exp(2πi(φ
1
(m) +mn(s))/N)[
=

s∈Z
N
[

j∈(s−P)∩E
exp(2πi(φ
1
(s −j) + (s −j)n(s))/N)[
=O(εN[P[)
0-34
since φ
1
(s−)+(s−)n(s) is a non-constant quadratic function for every
s.
In the case λ = 0 and µ = 0, By a rescaling, we may assume that
the progression P = [0, r). It then suffices to show that

m∈Z
N
ψ(
m
r
)[
ˆ
1
E∩(E+m)
(µ)[
2
= O(εN
2
r)
for some suitable real even bump function ψ. We can rewrite the left
hand side as
1
N
2

n,n

∈Z
N
r
ˆ
ψ(
r(n −n

)
N
)
ˆ
1
E
(n)
ˆ
1
E
(µ −n)
ˆ
1
E
(n

)
ˆ
1
E
(µ −n

)
where
ˆ
ψ is the Fourier transform of ψ on R. From (4) we have
ˆ
1
E
(n

)
ˆ
1
E
(µ −n

) = O(εN
2
)
and so it suffices to show that

n,n

∈Z
N
[
ˆ
ψ(
r(n −n

)
N
)[[
ˆ
1
E
(n)[[
ˆ
1
E
(µ −n)[ = O(N
2
)
0-35
But this follows from Schur’s test, Plancherel, and the rapid decay of
ˆ
ψ.
0-36