OctoInformBeta

OpenAI preprint, September 2026 · 5. The ensemble bound and its attainment · Section 5

Operational achievability and converse: Collective decoding achieves every lower nonnegative rate, while the scalar maximum bounds all achievable rates

Operational rates from phase-pair values.

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 damping parameter γ∈[0,1]\gamma\in[0,1] and a stationary excited-state population ν∈[0,1]\nu\in[0,1]. Classical communication through the generalized amplitude-damping channel uses quantum states as signals and a collective measurement on the outputs. The operational question concerns rates with average decoding error tending to zero as the number of channel uses grows.

The computational basis ℬn\mathcal B_n of nn qubits consists of bit strings i:{0,…,n−1}→{0,1}i:\{0,\ldots,n-1\}\to\{0,1\}.✦ An nn-qubit operator is a complex matrix whose rows and columns are indexed by ℬn\mathcal B_n.✦ A density operator PP is a positive semidefinite operator with Tr⁡P=1\operatorname{Tr}P=1.✦

The single-qubit Kraus operators L0,L1,L2,L3L_0,L_1,L_2,L_3 are defined by (1): L0=1−ν(1001−γ),L1=γ(1−ν)(0100),L2=ν(1−γ001),L3=γν(0010).\begin{aligned} L_0&=\sqrt{1-\nu} \begin{pmatrix}1&0\\0&\sqrt{1-\gamma}\end{pmatrix},& L_1&=\sqrt{\gamma(1-\nu)} \begin{pmatrix}0&1\\0&0\end{pmatrix},\\ L_2&=\sqrt{\nu} \begin{pmatrix}\sqrt{1-\gamma}&0\\0&1\end{pmatrix},& L_3&=\sqrt{\gamma\nu} \begin{pmatrix}0&0\\1&0\end{pmatrix}. \end{aligned}(1)✦ For a label r:{0,…,n−1}→{0,1,2,3}r:\{0,\ldots,n-1\}\to\{0,1,2,3\} and computational basis indices i,j∈ℬni,j\in\mathcal B_n, the tensor-product Kraus operator KrK_r and the memoryless channel 𝒩γ,ν⊗n\mathcal N_{\gamma,\nu}^{\otimes n} are defined by (2), where ** denotes the adjoint: Kr(i,j)=∏k=0n−1(Lr(k))i(k),j(k),𝒩γ,ν⊗n(P)=∑rKrPKr*.\begin{aligned} K_r(i,j)&=\prod_{k=0}^{n-1}(L_{r(k)})_{i(k),j(k)},\\ \mathcal N_{\gamma,\nu}^{\otimes n}(P)&=\sum_r K_r P K_r^*. \end{aligned}(2)

Codes and Achievable Rates

A classical communication code CC for nn channel uses has a positive integer message count MM, density operators ρa\rho_a on nn qubits, and a collective output positive operator-valued measure {Da}a=0M−1\{D_a\}_{a=0}^{M-1}, with Da≽0D_a\succeq0 and ∑aDa=I\sum_a D_a=I, where II is the identity operator on the output space.✦ The average decoding error Pe(C)P_{\mathrm e}(C) of a code CC with equally likely messages is the real number defined by (3): Pe(C)=1−1M∑a=0M−1Re⁡Tr⁡(Da𝒩γ,ν⊗n(ρa)).P_{\mathrm e}(C) =1-\frac1M\sum_{a=0}^{M-1} \operatorname{Re}\operatorname{Tr}\!\left( D_a\mathcal N_{\gamma,\nu}^{\otimes n}(\rho_a)\right).(3)✦ A rate RR in bits per channel use is achievable if there is a sequence of codes CnC_n, with message counts MnM_n, satisfying (4): Pe(Cn)→0(n→∞),for every ε>0,R−ε≤ln⁡Mnnln⁡2for all sufficiently large n.\begin{gathered} P_{\mathrm e}(C_n)\longrightarrow0\qquad(n\longrightarrow\infty),\\ \text{for every }\varepsilon>0,\qquad R-\varepsilon\le\frac{\ln M_n}{n\ln2} \quad\text{for all sufficiently large }n. \end{gathered}(4)✦

The Scalar Objective

The binary entropy function hh of a probability q∈[0,1]q\in[0,1] and the qubit entropy function gg of a determinant u∈[0,1/4]u\in[0,1/4] are defined by (5), with 0ln⁡0=00\ln0=0; their values are in nats: h(q)=−qln⁡q−(1−q)ln⁡(1−q),g(u)=h(1+1−4u2).h(q)=-q\ln q-(1-q)\ln(1-q),\qquad g(u)=h\!\left(\frac{1+\sqrt{1-4u}}2\right).(5) For an input excitation population p∈[0,1]p\in[0,1], the phase pair is the equiprobable ensemble of two pure qubit states with excitation population pp and opposite coherences; the output determinant vv of either state and the phase-pair Holevo objective ff are defined by (6): v(γ,ν,p)=γν(1−ν)+γ(1−γ)(p−ν)2,f(γ,ν,p)=h((1−γ)p+γν)−g(v(γ,ν,p)).\begin{aligned} v(\gamma,\nu,p)&=\gamma\nu(1-\nu) +\gamma(1-\gamma)(p-\nu)^2,\\ f(\gamma,\nu,p)&=h((1-\gamma)p+\gamma\nu)-g(v(\gamma,\nu,p)). \end{aligned}(6) The Holevo value in bits is f(γ,ν,p)/ln⁡2f(\gamma,\nu,p)/\ln2. An optimal excitation population p*p_* is a member of [0,1][0,1] satisfying (7): f(γ,ν,q)≤f(γ,ν,p*)for every q∈[0,1].f(\gamma,\nu,q)\le f(\gamma,\nu,p_*)\qquad\text{for every }q\in[0,1].(7)✦

The Claim

Theorem 1 (Operational bounds (8) and (9)). Let γ,ν∈[0,1]\gamma,\nu\in[0,1]. Let p∈[0,1]p\in[0,1], and let rr be a nonnegative real number strictly below f(γ,ν,p)f(\gamma,\nu,p). Then rln⁡2is an achievable rate for 𝒩γ,ν.\frac{r}{\ln2}\quad\text{is an achievable rate for } \mathcal N_{\gamma,\nu}.(8)✦ If p*p_* is an optimal excitation population and RR is an achievable rate for 𝒩γ,ν\mathcal N_{\gamma,\nu}, then R≤f(γ,ν,p*)ln⁡2.R\le\frac{f(\gamma,\nu,p_*)}{\ln2}.(9)✦

What Is Checked

The statements are over physical parameters γ,ν∈[0,1]\gamma,\nu\in[0,1] and sequences of codes for the tensor-power channel in (2), with arbitrary density-operator encodings and collective decoding measurements; the converse uses an optimal excitation population as defined by (7).