Wiki
Wiki

Research notes on every problem, and a library of the papers behind them. Built from the open erdos repository.

Updated


Source. Theorem 1.3, p. 5, with its proof in Section 6, pp. 17--20, of Ben Green and Imre Z. Ruzsa, On the arithmetic Kakeya conjecture of Katz and Tao, arXiv:1712.02108 (2017); the edition read is named on the source card.

Statement

Fp∞\mathbb F_p^\infty is the vector space over Fp\mathbb F_p of countably infinite dimension, and H\mathbf H is Shannon entropy.

Theorem 1.3 (p. 5). Let XX and YY be Fp∞\mathbb F_p^\infty-valued random variables, each taking only finitely many values. Then

H(X−Y)≤(1+O(1log⁡p))sup⁡r∈Fp∪{∞}∖{−1}H(X+rY),\mathbf H(X-Y)\le\Bigl(1+O\bigl(\tfrac{1}{\log p}\bigr)\Bigr) \sup_{r\in\mathbb F_p\cup\{\infty\}\setminus\{-1\}}\mathbf H(X+rY),

where the constant in the O(⋅)O(\cdot) is absolute.

This is the finite field variant of the paper's Conjecture 2 (see Theorem 1.1). The paper says the O(1/log⁡p)O(1/\log p) term is best possible (p. 5); the example on pp. 19--20, with X=(a+b,ab)X=(a+b,ab) and Y=(a+b′,ab′)Y=(a+b',ab') for independent uniform a,b,b′∈Fpa,b,b'\in\mathbb F_p, has H(X−Y)=2log⁡p+O(log⁡p/p)\mathbf H(X-Y)=2\log p+O(\log p/p) and H(X+rY)≤2log⁡p−log⁡2+O(log⁡p/p)\mathbf H(X+rY)\le2\log p-\log2+O(\log p/p) for r≠−1r\ne-1. The remark introducing that example names Theorem 1.2 [sic]; it concerns Theorem 1.3.

Proof pointer

Section 6, pp. 17--20. Assuming $\mathbf H(X-Y)\ge(1+\varepsilon)\sup\mathbf H(X+rY)$, a tensor-power construction gives large finite B⊂Fp∞×Fp∞B\subset\mathbb F_p^\infty\times\mathbb F_p^\infty with #π−1(B)≥sup⁡r≠−1(#πr(B))1+ε/2\#\pi_{-1}(B)\ge\sup_{r\ne-1}(\#\pi_r(B))^{1+\varepsilon/2}; the lines through the pairs of BB form a set AA of size at most psup⁡#πr(B)p\sup\#\pi_r(B) containing a line with common difference dd for every nonzero d∈π−1(B)d\in\pi_{-1}(B), so N=#π−1(B)−1N=\#\pi_{-1}(B)-1 directions. Proposition 6.1 (p. 18), proved by a random projection to Fpn\mathbb F_p^n, covering by translates (Corollary A.3, p. 20) and a lower bound for finite field Kakeya sets (the paper's reference [6]), gives #A≫pN1−log⁡2/log⁡p−o(1)\#A\gg_pN^{1-\log2/\log p-o(1)} for such sets, forcing ε=O(1/log⁡p)\varepsilon=O(1/\log p).

Read depth

Claims checked: the statement, the remark on sharpness and the outline of Section 6 were read on the print; the finite field Kakeya bound cited as reference [6] was not read. Nothing here is independently reviewed.

Dependencies

External input: the finite field Kakeya bound of the paper's reference [6].

Bears on

No Erdős problem in the corpus.