OctoInformBeta

OpenAI preprint, September 2026 · §7. Zero avoidance costs information

Residues modulo a finite set of Gaussian integers: The residue code cπc_\pi numbers the N(π)N(\pi) classes modulo a nonzero π\pi, and a finite set FF has weight ∑f∈Flog⁡N(f)\sum_{f \in F} \log N(f)

Residues modulo nonzero moduli have entropy at most their weight.

Hover over a blue or green term in the text, or a node in the graph, to find it in the other. Select it to read its declaration both in Lean and in words.

The Setting

Let ZZ be a random point of ℤ[i]\mathbb{Z}[i] and FF a finite set of Gaussian integers. The residues of ZZ modulo the elements of FF form a random vector. The question is how much information this vector can carry: how large its Shannon entropy can be, given the norms N(a+bi)=a2+b2N(a + bi) = a^2 + b^2 of the moduli.

Residues and Their Codes

The Gaussian integers form a lattice with basis 1,i1, i.✦ For a Gaussian integer π\pi, the set πℤ[i]\pi\mathbb{Z}[i] of multiples of π\pi is a subgroup of ℤ[i]\mathbb{Z}[i].✦ The residue class of a Gaussian integer zz modulo π\pi is its coset z+πℤ[i]z + \pi\mathbb{Z}[i] in the quotient group ℤ[i]/πℤ[i]\mathbb{Z}[i]/\pi\mathbb{Z}[i].✦

The residue code cπ(z)c_\pi(z) of a Gaussian integer zz modulo π\pi is the natural number that labels the residue class of zz modulo π\pi in a fixed numbering of the classes of ℤ[i]\mathbb{Z}[i] modulo π\pi.✦

Entropy of Residue Vectors

A finite probability law μ\mu on a finite set Ω\Omega assigns to each point ω\omega of Ω\Omega a mass μ(ω)≥0\mu(\omega) \ge 0, the masses summing to 11 over Ω\Omega.✦ The entropy H(X)H(X) of a random variable XX on Ω\Omega, with values in any set, is the Shannon entropy of its law under μ\mu, H(X)=−∑xμ(X=x)log⁡μ(X=x),H(X) = -\sum_{x} \mu(X = x) \log \mu(X = x),(1)✦ The sum runs over the finitely many values xx that XX takes, and log\log is the natural logarithm.

For a random variable ZZ on Ω\Omega with values in ℤ[i]\mathbb{Z}[i] and a finite set FF of Gaussian integers, the residue vector cF(Z)c_F(Z) of ZZ modulo FF is the random variable on Ω\Omega whose value at ω\omega is the vector (cf(Z(ω)))f∈F\bigl(c_f(Z(\omega))\bigr)_{f \in F} of residue codes.

The residue weight w(F)w(F) of a finite set FF of Gaussian integers is the sum of the natural logarithms of the norms of its elements: w(F)=∑f∈Flog⁡N(f).w(F) = \sum_{f \in F} \log N(f).(2)✦

The Claim

In the library, two residue vectors arise from a walk z0,z1,…z_0, z_1, \dots, both taken at zt*z_{t^*}, its point at the terminal time t*t^* of the common schedule, difference transitions followed by a uniform offset. The first holds the residues of zt*z_{t^*} modulo the selected factors of the batch of index jj, Gaussian primes πp\pi_p or π‾p\bar\pi_p of norm pp for the primes p≡1(mod⁡4)p \equiv 1 \pmod 4 with 2j≤p<2j+12^j \le p < 2^{j+1}. The second holds its residues modulo a finite set FF of nonzero Gaussian integers of bounded residue weight, which stands for the earlier batches.

Theorem 1. Let π\pi, zz and ww be Gaussian integers. Then cπ(z)=cπ(w)⇔π∣z−w.c_\pi(z) = c_\pi(w) \iff \pi \mid z - w.(3)✦ If π≠0\pi \neq 0, the set of residue codes modulo π\pi is finite, and #{cπ(z):z∈ℤ[i]}=N(π).\#\{c_\pi(z) : z \in \mathbb{Z}[i]\} = N(\pi).(4)✦ Let μ\mu be a finite probability law on a finite set Ω\Omega, let ZZ be a random variable on Ω\Omega with values in ℤ[i]\mathbb{Z}[i], and let FF be a finite set of nonzero Gaussian integers. Then H(cF(Z))≤w(F).H\bigl(c_F(Z)\bigr) \le w(F).(5)✦

In the paper, the result reads in terms of XT′X'_T, the selected residue vector at zt*z_{t^*} of the batch ℬT\mathcal{B}_T of primes p≡1(mod⁡4)p \equiv 1 \pmod 4 in [T,2T][T, 2T], and OT′O'_T, the selected residue vectors of the smaller batches.

What Is Checked

The claim is stated over the residue code cπc_\pi, which numbers the residue classes modulo the multiples πℤ[i]\pi\mathbb{Z}[i] of π\pi, over the residue weight w(F)w(F) of (2), and over the entropy (1) of a random variable under a finite probability law.