Skip to content

44a: R_3 = 10 has been known since 2000 (Tarannikov–Kirienko, resilient-function form); Wel2019 Table 2 gives R_8 ≤ 1008 #190

Description

@moffatstudio

While checking prior art for a result related to 44a, I found two items in the literature that the page does not record. They may be worth a line in "Additional comments". I extracted every quote below from the source text myself, not from a summary.

Notation as on the page: $R_d$ is the maximum number of relevant variables of a Boolean function of degree $d$, so $C_{44} = \sup_d R_d/2^d$.

1. $R_3 = 10$ has been known since 2000, in resilient-function language

Tarannikov and Kirienko, Spectral analysis of high order correlation immune functions, IACR ePrint 2000/050, Theorem 11:

p'(4)=p(4)=10

Here $p(k)$ is the least integer such that every $(n-k)$-resilient function on $\mathbb{F}_2^n$ depends nonlinearly on at most $p(k)$ variables.

Why this is $R_3$. Put $f' = f\cdot\chi_{[n]}$ (that is, $f$ XOR parity). Then $\widehat{f'}(S) = \widehat{f}([n]\setminus S)$.

  • By the Xiao–Massey characterisation, $\deg f \le d$ iff $f'$ is $(n-d-1)$-resilient.
  • $x_i$ is irrelevant for $f$ iff $f'$ depends linearly on $x_i$.

Hence $R_d = p(d+1)$. This is Lemma 1 of Krotov and Valyuzhenich, On degree-3 and (n−4)-correlation-immune perfect colorings of n-cubes, Discrete Math. 347(10) (2024) 114138, arXiv:2311.05566. They state the consequence in degree language: a Boolean function of degree at most 3

cannot have more than 10 essential arguments

and they mention "the unique such function with 10 essential arguments". The CHS function $\Xi_3$ attains 10, so $R_3 = 10$.

Two more results from the same literature:

  • Tarannikov, Korolev and Botev, Autocorrelation coefficients and correlation immunity of Boolean functions, ASIACRYPT 2001, LNCS 2248, pp. 460–479, Theorem 6, prove $n \le (k-1)2^{k-2}$. This is exactly the Nisan–Szegedy bound, in resilient form.
  • The same paper notes that $(n-k)$-resilient functions depending nonlinearly on all $n = 3\cdot 2^{k-2}-2$ variables exist. That is the $R_d \ge 3\cdot 2^{d-1}-2$ construction, earlier than [CHS2020].

2. [Wel2019], Table 2, gives $R_8 \le 1008$

arXiv:1903.08214, Table 2, bounds $W(f) = \sum_i 2^{-\deg_i f}$ for $d \le 14$.

  • At $d = 8$ it gives $3.9375$, so $R_8 \le 256 \cdot 3.9375 = 1008 < 8\cdot 2^7$.
  • For $d \le 7$ the table gives $d/2$, which is exactly the Nisan–Szegedy bound.

I re-derived Tables 1 and 2 for $d \le 8$ in exact rational arithmetic, with certificates for every LP verdict, and the values agree.

Disclosure

I am the author of moffatstudio/relevant-variables-degree. It is a preprint, not peer-reviewed, developed with AI assistance. It proves $R_d \le d,2^{d-1}-1$ for every $d \ge 3$, which in view of the above is new only for $4 \le d \le 7$. It also gives a new proof of $R_3 = 10$, with the finite core checked in Lean. An earlier version wrongly presented $R_3 = 10$ as new; that is how I came across this prior art.

I am flagging the prior art here, not asking for our result to be listed. I am happy to open a Markdown-only PR for the comments section if that would help.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions