OctoInformBeta

OpenAI preprint, September 2026 · 2. An entropy inequality with one zero block · Section 2.3

Lemma 2.4: Positive homogeneity and concavity on the x=1x=1 slice, where xx is the first weight

The scalar deficit is superadditive on nonnegative triples.

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

A block entropy comparison involves the entropy contributions of three nonnegative weights and those of two block eigenvalues. The scalar entropy deficit measures their difference in nats.

The block eigenvalue functions λ+(x,y,z)\lambda_+(x,y,z) and λ−(x,y,z)\lambda_-(x,y,z) of nonnegative real weights x,y,zx,y,z are defined by (1), with the plus sign giving the larger eigenvalue and the minus sign the smaller one: λ±(x,y,z)=x+y+z±(x+y+z)2−4yz2.\lambda_\pm(x,y,z) = \frac{x+y+z\pm\sqrt{(x+y+z)^2-4yz}}{2}.(1)

The entropy contribution η\eta is the real-valued function of a nonnegative weight tt defined by η(t)=−tln⁡t\eta(t)=-t\ln t, with 0ln⁡0=00\ln0=0. The scalar entropy deficit F:ℝ3→ℝF:\mathbb{R}^3\to\mathbb{R} is the real-valued function defined by (2): F(x,y,z)=η(x)+η(y)+η(z)−η(λ+(x,y,z))−η(λ−(x,y,z)).F(x,y,z) = \eta(x)+\eta(y)+\eta(z) -\eta\bigl(\lambda_+(x,y,z)\bigr) -\eta\bigl(\lambda_-(x,y,z)\bigr).(2)✦

The Claim

Theorem 1. For nonnegative real numbers a,x,y,za,x,y,z, F(ax,ay,az)=aF(x,y,z).F(ax,ay,az)=aF(x,y,z).(3)✦ For nonnegative real numbers y0,y1,z0,z1y_0,y_1,z_0,z_1 and θ∈[0,1]\theta\in[0,1], F(1,θy0+(1−θ)y1,θz0+(1−θ)z1)≥θF(1,y0,z0)+(1−θ)F(1,y1,z1).\begin{aligned} &F\bigl(1,\theta y_0+(1-\theta)y_1, \theta z_0+(1-\theta)z_1\bigr) \\ &\qquad\ge \theta F(1,y_0,z_0)+(1-\theta)F(1,y_1,z_1). \end{aligned}(4)✦ Let II be a finite index set and let (xi,yi,zi)(x_i,y_i,z_i) be a nonnegative real triple for each i∈Ii\in I. Then ∑i∈IF(xi,yi,zi)≤F(∑i∈Ixi,∑i∈Iyi,∑i∈Izi).\sum_{i\in I}F(x_i,y_i,z_i) \le F\left(\sum_{i\in I}x_i,\sum_{i\in I}y_i,\sum_{i\in I}z_i\right).(5)✦

In the paper, the result reads: FF is positively homogeneous of degree one and concave on the probability simplex {(x,y,z)∈[0,∞)3:x+y+z=1}\{(x,y,z)\in[0,\infty)^3:x+y+z=1\}, and ∑i∈IF(xi,yi,zi)≤F(∑i∈Ixi,∑i∈Iyi,∑i∈Izi)\sum_{i\in I}F(x_i,y_i,z_i)\le F(\sum_{i\in I}x_i,\sum_{i\in I}y_i,\sum_{i\in I}z_i) for every finite collection of nonnegative triples.

What Is Checked

The statements (3), (4), and (5) concern the scalar entropy deficit in (2) on nonnegative real weights, with arbitrary finite index sets in (5).