THE POPULARITY GAP VSEVOLOD F. LEV AND ILYA D. SHKREDOV Abstract. Suppose that Ais a nite nonempty subset of a cyclic group of either

2025-05-06 0 0 383.97KB 16 页 10玖币
侵权投诉
THE POPULARITY GAP
VSEVOLOD F. LEV AND ILYA D. SHKREDOV
Abstract. Suppose that Ais a finite, nonempty subset of a cyclic group of either
infinite or prime order. We show that if the difference set AAis “not too large”,
then there is a nonzero group element with at least as many as (2 + o(1))|A|2/|AA|
representations as a difference of two elements of A; that is, the second largest number of
representations is, essentially, twice the average. Here the coefficient 2 is best possible.
We also prove continuous and multidimensional versions of this result, and obtain
similar results for sufficiently dense subsets of an arbitrary abelian group.
1. Background and summary of results
A large body of problems, conjectures, and results in additive combinatorics assert, in
different ways, that, normally, the number-of-representations function exhibits an irreg-
ular behaviour. The general framework for this line of research is as follows.
For finite subsets Aand Band an element gof an abelian group, let rA,B(g) :=
|A(gB)|; that is, rA,B (g) is the number of representations of gas a sum of an
element of Aand an element of B. How close to a constant function can the function
rA,B be?
In this paper we consider representations by differences; that is, we are concerned with
the special case where B=A. Notation-wise, we abbreviate rA,Aas rA.
Clearly, the largest value attained by the function rAis |A|. Our principal results
show that for the cyclic groups of either infinite or prime order, the second largest value
attained by rAis, essentially, at least twice the average value.
For a prime p, by Cpwe denote the cyclic group of order p.
Theorem 1. Suppose that pis a prime, ACpis a nonempty subset, and δ(0,1/3).
If
|AA| ≤ min{K|A|,2(p+ 1)/3}, K < |A|δ,
then there is a nonzero element dCpsuch that rA(d)>2K1|A|(1 2δln(2)).
The following integer analog of Theorem 1 is, in fact, its immediate consequence.
Theorem 2. Suppose that Ais a finite, nonempty set of integers, and that δ(0,1/3).
If
|AA| ≤ K|A|, K < |A|δ,
then there is an integer d6= 0 such that rA(d)>2K1|A|(1 2δln(2)).
1
arXiv:2210.09614v1 [math.NT] 18 Oct 2022
2 VSEVOLOD F. LEV AND ILYA D. SHKREDOV
Comparing Theorems 1 and 2, we see that the bounds obtained are not affected by
the “modulo-poverlapping”. Loosely speaking, there is no difference between the group
Cpand the group of integers Zas far as the second largest value of the function rAis
concerned.
The following theorem establishes a similar estimate in terms of the length of the set
Ainstead of the size of its difference set.
Theorem 3. Suppose that L > 1, and that A[1, L]is a nonempty set of integers. If
|AA|<|A|1+δwith δ(0,1/3), then there is an integer d6= 0 such that rA(d)>
|A|2
L(1 2δln(2)).
We note that the assumption |AA|<|A|1+δof Theorem 3 holds trivially if Ais
sufficiently dense in [1, L]; say, |A|>(2L)1/(1+δ)suffices.
Theorem 3 follows readily from Theorem 2 by letting K:= |AA|/|A|and observing
that then 2K1|A|= 2|A|2/|AA|>|A|2/L.
Apart from the factor 1 2δln(2), the lower bounds obtained in Theorems 1–3 are,
essentially, best possible; they are attained, for example, when Ais an interval or an
appropriately sampled random set.
Interestingly, Theorems 1–3 cannot be obtained by a straightforward averaging as the
number of elements dAAwith rA(d) large can have “zero measure”.
Claim 1. For any ε(0,1) there is a finite, nonempty set AZsuch that
nd:rA(d)(1 ε)2|A|2
|AA|o2ε|AA|.
Since the function rA:d7→ |A(A+d)|is generalized by the functions (d1, . . . , dk1)7→
|A(A+d1)···(A+dk1)|(see the next section), it is natural to extend Theorem 1
onto the intersection of k3 translates of the set A. In this direction, we prove the
following multidimensional generalization.
Theorem 4. Suppose that pis a prime, ACpis a nonempty subset, and δ(0,1/(3k))
with an integer k3. If
|AA| ≤ min{K|A|,2(p+ 1)/3}, K < |A|δ,
then there are pairwise distinct, nonzero elements d1, . . . , dk1Cpsuch that
|A(A+d1)∩ ··· ∩ (A+dk1)|>2k1K(k1)|A|(1 3δk2ln(1/kδ)).
From Theorem 4 one can easily derive multidimensional analogs of Theorems 2 and 3;
we omit the details.
Our next result exhibits irregularities in the behaviour of the function rAfor large
subsets Aof an arbitrary finite abelian group.
THE POPULARITY GAP 3
Theorem 5. Suppose that Ais a subset of a finite abelian group Gsuch that |A| ≥ 210
and |AA|= (1 ε)|G|with 0< ε 25. If |AA|≤|A|1+ε/2, then there is a
nonzero element dGsuch that rA(d)|A|2
|AA|1 + 1
8ε.
Finally, we prove a continuous analog of Theorem 3. For a function fL1(R), let
(ff)(x) := ZR
f(t)f(x+t)dt.
Theorem 6. Suppose that fis a real, nonnegative function with supp(f)[0,1]. If fis
not constant on its support then, letting ρ:= kfk2/kfk1, for any 0< δ min{1
2ρ12 ,1
28}
we have
sup
x[1,1]\[δ,δ]
(ff)(x)
kfk2
118 max n8
2δ, ln logρ(1/2δ)
logρ(1/2δ)o.
We note that the matching upper bound kffk≤ kfk2
1is trivial.
The quantity logρ(1/2δ) characterizes the length of the forbidden interval [δ, δ] as
measured against the “scatter” of f. If logρ(1/2δ) is small enough, then we can have
supp(f)[0, δ]; in this case supp(ff)[δ, δ] and there do not exist x /[δ, δ] with
(ff)(x)>0. Theorem 6 assumes logρ(1/2δ)12, and the remainder term is o(1) in
the regime where logρ(1/2δ)→ ∞ and δ0.
Our argument involves two major components: the higher energies technique [SS13]
and a result in the spirit of [L01] establishing an extremal property of the interval subsets
of prime-order groups. The central role is played by the following quantity.
For finite subsets Aand Dof an abelian group and integer k1, let T(k)
D(A) be the
number of k-tuples (a1, . . . , ak)Aksuch that aiajDfor any 1 i, j k. In
particular, T(k)
D(D) is the number of k-tuples (d1, . . . , dk) such that
{0, d1, . . . , dk}−{0, d1, . . . , dk} ⊆ D.
The key result we prove about this quantity shows that in a group of prime order, if
0D= (D), then T(k)
D(A) can only get larger if Aand Dare replaced with the
intervals A, D of sizes |A|=|A|and |D|=|D|, respectively, centered at 0; that is, either
A= [m, m], or A= [(m1), m] with m=b|A|/2cin both cases, and similarly for
D.
Proposition 1. For any prime p5, integer k1, and nonempty subsets A, D Cp
with 0D= (D), we have
T(k)
D(A)T(k)
D(A),
where A, D Cpare intervals with |A|=|A|and |D|=|D|, centered at 0.
As a consequence of Proposition 1 we prove the following, slightly technical, corollary.
4 VSEVOLOD F. LEV AND ILYA D. SHKREDOV
Corollary 1. For any prime p5, integer k3, and subset DCpwith 0D=
(D)and k≤ |D| ≤ 2
3(p+ 1), we have
T(k)
D(D)3k2k1|D|k.
We now turn to the proofs of the results discussed above. In the next section we intro-
duce the basic notation, collect auxiliary tools needed for the proofs, and prove Claim 1.
In Section 3 we prove the key Proposition 1. Theorem 1 is derived from Proposition 1 in
Section 4 where also Corollary 1 is proved; as we have noticed, Theorems 2 and 3 follow
from Theorem 1 and therefore do not require any additional consideration. Theorems 5
and 4 are proved in Sections 5 and 6, respectively. A proof of Theorem 6, along with
a somewhat broader context for the problems studied in this paper, is outlined in the
concluding Section 7.
2. Notation and preliminaries
In this section Aand Dare finite, nonempty subsets of an abelian group G.
Following [SS13] (but using a different notation), for an integer k2 we let
R(k)
A(x1, . . . , xk1) := |A(A+x1)∩ ··· ∩ (A+xk1)|, x1, . . . , xk1G. (1)
This quantity generalizes the function rAwhich is obtained as a particular case k= 2.
Alternatively, R(k)
A(x1, . . . , xk1) is the number of representations of (x1, . . . , xk1) as a
difference of an element of the “diagonal” {(a, . . . , a) : aA}and an element of the
Cartesian power Ak1. Clearly,
X
x1,...,xk1G
R(k)
A(x1, . . . , xk1) = |A|k.(2)
Less obvious is the identity
X
x1,...,xk1GR(k)
A(x1, . . . , xk1)l=X
y1,...,yl1GR(l)
A(y1, . . . , yl1)k(3)
valid for any k, l 2 and any finite subset AG, see [SV12, Lemma 2.8].
The common value of the two sums in (3) is denoted Ek,l(A); thus, Ek,l(A) = El,k(A).
We write
Ek(A) := Ek,2(A) = X
xG|A(A+x)|k.
Let µ(A) := max{rA(d): dAA, d 6= 0}; that is, µ(A) is the second largest value
attained by the function rA. We have
Ek(A) = X
xG|A(A+x)|k≤ |A|k+ (µ(A))k1|A|2.(4)
Recall that in Section 1 we have defined
T(k)
D(A) := |{(a1, . . . , ak)Ak:aiajD, 1i, j k}|, k 1.
摘要:

THEPOPULARITYGAPVSEVOLODF.LEVANDILYAD.SHKREDOVAbstract.SupposethatAisa nite,nonemptysubsetofacyclicgroupofeitherin niteorprimeorder.Weshowthatifthedi erencesetAAisottoolarge",thenthereisanonzerogroupelementwithatleastasmanyas(2+o(1))jAj2=jAAjrepresentationsasadi erenceoftwoelementsofA;thatis,theseco...

展开>> 收起<<
THE POPULARITY GAP VSEVOLOD F. LEV AND ILYA D. SHKREDOV Abstract. Suppose that Ais a nite nonempty subset of a cyclic group of either.pdf

共16页,预览4页

还剩页未读, 继续阅读

声明:本站为文档C2C交易模式,即用户上传的文档直接被用户下载,本站只是中间服务平台,本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间,仅对用户上传内容的表现方式做保护处理,对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私,请立即通知玖贝云文库,我们立即给予删除!

相关推荐

分类:图书资源 价格:10玖币 属性:16 页 大小:383.97KB 格式:PDF 时间:2025-05-06

开通VIP享超值会员特权

  • 多端同步记录
  • 高速下载文档
  • 免费文档工具
  • 分享文档赚钱
  • 每日登录抽奖
  • 优质衍生服务
/ 16
客服
关注