Indexed metadata

Linearity bounds for APN functions

Christof Beierle

Source record

Source: arXiv

Published: Sep 28, 2026

arXiv: 2609.35689

Open original source ↗

Source abstract

For n≥5n\ge5, let F ⁣:F2n→F2nF\colon\mathbb{F}_2^n\to \mathbb{F}_2^n be almost perfect nonlinear and write N=2nN=2^n. It is proven that the linearity L(F)\mathcal{L}(F) of FF, i.e., the largest absolute Walsh coefficient of a nonzero component, is at most N−10N-10 in even dimension and at most N−6N-6 in odd dimension. This improves the general upper bound of N−6N-6 in even dimension and N−4N-4 in odd dimension. It is further proven that, for each fixed kk, the kk-th largest absolute Walsh coefficient among nonzero components, counted with multiplicity, is at most (1+Ok(2−n))N/k(1+O_k(2^{-n}))N/\sqrt{k} as n→∞n\to\infty. The second and fourth largest coefficients are at most 2⌊N/3⌋2\lfloor N/3\rfloor and N/2N/2, respectively. Finally, a bound on the linearity in terms of the number qq of nonplateaued nonzero components is derived. In odd dimension, for q>0q>0, we have L(F)2≤N(1+q(N−1))\mathcal{L}(F)^2\le N(1+\sqrt{q(N-1)}), so that L(F)/N→1\mathcal{L}(F)/N\to1 implies q/N→1q/N\to1. In even dimension, for every 1/2≤C<11/2\le C<1, the condition q≤(4C−C2−1)N/4+1q\le(4C-C^2-1)N/4+1 implies L(F)≤CN\mathcal{L}(F)\le CN. In particular, q≤3N/16+2q\le3N/16+2 implies L(F)≤N/2\mathcal{L}(F)\le N/2.

Evidence graph

No public relationships recorded yet.

Integrity note: This page is a factual metadata record created by deterministic ingestion. It is not a claim that the work moves a mathematical frontier or has been independently verified.