OctoInformBeta

OpenAI preprint, September 2026 · §2. The scalar rate and its curvature

Lemma 2.1: The entropy deficit has a positive power series on the full interval of biases

Quadratic bounds for binary entropy.

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

For a binary coordinate of bias v∈[−1,1]v\in[-1,1], the entropy deficit measures the shortfall from the entropy of an unbiased bit. Its comparison with v2v^2 gives quadratic control of the entropy as the bias varies.

The constant LL is the natural logarithm of 22: L=log⁡2.L=\log 2.(1)✦

The binary entropy deficit ψ:ℝ→ℝ\psi:\mathbb R\to\mathbb R is defined by (2), with 0log⁡0=00\log 0=0: ψ(v)=1+v2log⁡(1+v)+1−v2log⁡(1−v).\psi(v)=\frac{1+v}{2}\log(1+v)+\frac{1-v}{2}\log(1-v).(2) The binary entropy H:ℝ→ℝH:\mathbb R\to\mathbb R in nats is defined by (3): H(v)=L−ψ(v).H(v)=L-\psi(v).(3)

The deficit coefficients ckc_k, for integers k≥0k\ge0, are the positive real numbers in (4): ck=1(2k+2)(2k+1).c_k=\frac{1}{(2k+2)(2k+1)}.(4)✦

The Claim

Theorem 1. Let v∈ℝv\in\mathbb R satisfy |v|≤1|v|\le1. Then v22≤ψ(v),\frac{v^2}{2}\le\psi(v),(5)✦ ψ(v)≤Lv2,\psi(v)\le Lv^2,(6)✦ L(1−v2)≤H(v).L(1-v^2)\le H(v).(7)✦ The following series converge to the indicated values: ∑k=0∞ckv2(k+1)=ψ(v),\sum_{k=0}^{\infty}c_kv^{2(k+1)}=\psi(v),(8)✦ ∑k=0∞ck=L.\sum_{k=0}^{\infty}c_k=L.(9)✦

In the paper, the result reads: for |v|≤1|v|\le1, v2/2≤ψ(v)≤Lv2v^2/2\le\psi(v)\le Lv^2 and H(v)≥L(1−v2)H(v)\ge L(1-v^2), the quotient ψ(v)/v2\psi(v)/v^2 is strictly increasing on 0<v<10<v<1, and 2/3<L<25/36<7/102/3<L<25/36<7/10 and 1/2<M=4L2/3<2/31/2<M=4L^2/3<2/3.

What Is Checked

The bounds (5), (6) and (7) and the series identity (8) are stated over real biases v∈[−1,1]v\in[-1,1] with their entropy H(v)H(v) and deficit ψ(v)\psi(v), while (9) concerns the real coefficient sequence (ck)k≥0(c_k)_{k\ge0}.