5
$\begingroup$

All sets are assumed to be finite subsets of the integers.

The additive energy of two sets $E(A,B)$ is defined as the number of solutions to $a+b=a'+b'$ with $a,a'\in A$ and $b,b'\in B$. The well-known Balog-Szemeredi-Gowers theorem states that there is an absolute constant $C>0$ such that (assuming $\lvert A\rvert=\lvert B\rvert$ for convenience) if $E(A,B)=K^{-1}\lvert A\rvert^3$ then there exist $A'\subset A$ and $B'\subset B$ such that

$$ \lvert A'\rvert\gg K^{-C}\lvert A\rvert\textrm{ and }\lvert B'\rvert\gg K^{-C}\lvert B\rvert $$ and $$ \lvert A'+B'\rvert\ll K^C\lvert A\rvert.$$

When $A=B$ we can qualitatively strengthen this conclusion to allow $A'=B'$ (see the recent paper by Schoen 'New bounds in Balog-Szemeredi-Gowers theorem' at http://www.staff.amu.edu.pl/~schoen/remark-B-S-G.pdf)

My question is whether we can obtain a similar strengthening when $B=d\cdot A$ for some $d\neq 0$ (here $d\cdot A$ denotes the dilate of $A$, the set $\{ da : a\in A\}$). That is, does there exist an absolute constant $C>0$ such that if $E(A,d\cdot A)=K^{-1}\lvert A\rvert^3$ then there exists $A'\subset A$ such that

$$ \lvert A'\rvert\gg K^{-C}\lvert A\rvert$$ and $$ \lvert A'+d\cdot A'\rvert\ll K^C\lvert A\rvert.$$

Note it is important that $C$ and the implied constants must be independent of $d$, else this would follow from the above and the usual sumset inequalities.

Adapting Schoen's argument, the best I have been able to do is to prove that there exists $s$ and $A'\subset A\cap(s-d\cdot A)$ such that $\lvert A'\rvert\gg K^{-C}\lvert A\rvert$ and $$ \lvert A'-A'\rvert \ll K^c\lvert A\rvert.$$

This is qualitatively stronger than a direct application of the original BSG theorem, but still falls short.

Also, if one actually has $E(A,d\cdot A)\geq K^{-1}\lvert A\rvert^3$ for all $d\in D$ then using arguments of Bourgain I can show that there exists $D'\subset D$ with $\lvert D'\rvert\gg K^{-C}\lvert D\rvert$ and $A'\subset A$ with $\lvert A'\rvert\gg K^{-C}\lvert A\rvert$ such that for all $d\in D'$ we have $$ \lvert A'+d\cdot A'\rvert \ll K^C\lvert A\rvert.$$

In my case, with $\lvert D\rvert=2$, however, this again falls short.

There are by now quite a few arguments for proving BSG-type statements, however, so hopefully someone knows of one which can be adapted for my question!

$\endgroup$
1
  • 1
    $\begingroup$ Should it be $B=d\cdot A$ rather than $A=d\cdot A$? $\endgroup$ Mar 24, 2014 at 3:45

1 Answer 1

2
$\begingroup$

Tomasz Schoen has observed in an email that the answer to my question is no, since there is a function $f(d)\to\infty$ as $d\to\infty$ such that $\lvert A+d\cdot A\rvert\geq f(d)\lvert A\rvert$ for all $A\subset \mathbb{Z}$. In particular, Boris Bukh in 'Sums of dilates' showed that roughly $f(d)=d+1$ is permissible.

To see that this leads to a counter example, consider the set $A=[1,N]\cup d\cdot [1,N]$ where $N$ is very large and d is even larger, depending on $N$. Then

$$E(A,d\cdot A)\geq E(d\cdot [1,N],d\cdot [1,N])=E([1,N],[1,N])\gg N^3\gg \lvert A\rvert^3,$$

so if my conjecture were true then there'd be some $A'\subset A$ such that $\lvert A'\rvert\gg N$ and $\lvert A'+d\cdot A'\rvert\ll N$, which contradicts the mentioned result of Bukh for $d$ sufficiently large.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.