OctoInformBeta

OpenAI preprint, September 2026 · 4. From the thermal inequality to channel outputs · Section 4

Lemma 4.1: For generalized amplitude damping with damping and stationary excited-state population in [0,1][0,1]

Pure-output entropy is convex in population.

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

Fix a generalized amplitude-damping channel with damping parameter γ∈[0,1]\gamma\in[0,1] and stationary excited-state population ν∈[0,1]\nu\in[0,1]. For a pure qubit input, let p∈[0,1]p\in[0,1] denote its excitation population.

The binary entropy hh is the function on [0,1][0,1] defined by h(q)=−qln⁡q−(1−q)ln⁡(1−q),0ln⁡0=0.h(q)=-q\ln q-(1-q)\ln(1-q),\qquad 0\ln0=0.(1) The qubit entropy function gg at determinant uu is defined using (1) by g(u)=h(1+1−4u2).g(u)=h\!\left(\frac{1+\sqrt{1-4u}}{2}\right).(2)✦ These entropies are in nats.

The output determinant vv for the channel parameters γ,ν\gamma,\nu and pure-input population pp is the function v(γ,ν,p)=γν(1−ν)+γ(1−γ)(p−ν)2.v(\gamma,\nu,p) =\gamma\nu(1-\nu)+\gamma(1-\gamma)(p-\nu)^2.(3)✦ The pure-output entropy is g(v(γ,ν,p))g(v(\gamma,\nu,p)), with gg and vv defined in (2) and (3).

The Claim

Theorem 1. Let γ,ν∈[0,1]\gamma,\nu\in[0,1]. The pure-output entropy is convex as a function of the input excitation population: for p0,p1∈[0,1]p_0,p_1\in[0,1] and t∈[0,1]t\in[0,1], g(v(γ,ν,(1−t)p0+tp1))≤(1−t)g(v(γ,ν,p0))+tg(v(γ,ν,p1)).g\!\left(v\bigl(\gamma,\nu,(1-t)p_0+tp_1\bigr)\right) \leq (1-t)g\!\left(v(\gamma,\nu,p_0)\right) +t g\!\left(v(\gamma,\nu,p_1)\right).(4)✦

What Is Checked

The claim in (4) is stated over the real functions gg and vv in (2) and (3), with fixed γ,ν∈[0,1]\gamma,\nu\in[0,1] and input populations in [0,1][0,1].