OctoInformBeta

OpenAI preprint, September 2026 · 1. Introduction · Section 1.1

Unrestricted Holevo information and operational capacity: Finite ensembles and collective decoding at each block length, with vanishing-error rates measured in bits

Holevo information and capacity for unrestricted inputs.

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]. A signal across several uses may be mixed or entangled, and the receiver may measure all output qubits jointly.

The space of nn qubits is ℋn=(ℂ2)⊗n\mathcal H_n=(\mathbb C^2)^{\otimes n}, with computational basis indexed by bit strings i∈{0,1}ni\in\{0,1\}^n.✦ An nn-qubit operator is a complex matrix whose rows and columns have these indices.✦ A density operator PP on ℋn\mathcal H_n is a positive semidefinite operator of trace one; write 𝒟(ℋn)\mathcal D(\mathcal H_n) for the set in (1). 𝒟(ℋn)={P∈End⁡(ℋn):P≽0,Tr⁡P=1}.\mathcal D(\mathcal H_n) =\{P\in\operatorname{End}(\mathcal H_n):P\succeq0,\ \operatorname{Tr}P=1\}.(1)✦ For a complex matrix PP indexed by a finite set, the von Neumann entropy S(P)S(P) is the real part of the trace obtained by applying x↦−xln⁡xx\mapsto-x\ln x to PP through spectral functional calculus, with 0ln⁡0=00\ln0=0, and is measured in nats. For a density operator PP on ℋn\mathcal H_n it is given by (2). S(P)=−Tr⁡(Pln⁡P).S(P)=-\operatorname{Tr}(P\ln P).(2)✦

The single-qubit Kraus operators K0,K1,K2,K3K_0,K_1,K_2,K_3 are the matrices in (3), in the basis |0⟩,|1⟩\lvert0\rangle,\lvert1\rangle. K0=1−ν(1001−γ),K1=γ(1−ν)(0100),K2=ν(1−γ001),K3=γν(0010).\begin{aligned} K_0&=\sqrt{1-\nu} \begin{pmatrix}1&0\\0&\sqrt{1-\gamma}\end{pmatrix}, &K_1&=\sqrt{\gamma(1-\nu)} \begin{pmatrix}0&1\\0&0\end{pmatrix},\\ K_2&=\sqrt{\nu} \begin{pmatrix}\sqrt{1-\gamma}&0\\0&1\end{pmatrix}, &K_3&=\sqrt{\gamma\nu} \begin{pmatrix}0&0\\1&0\end{pmatrix}. \end{aligned}(3)✦ For r=(r1,…,rn)∈{0,1,2,3}nr=(r_1,\ldots,r_n)\in\{0,1,2,3\}^n, the block Kraus operator KrK_r on ℋn\mathcal H_n is the tensor product in (4). Kr=⨂k=1nKrk.K_r=\bigotimes_{k=1}^{n}K_{r_k}.(4)✦ The memoryless channel on nn qubits is the tensor power 𝒩γ,ν⊗n\mathcal N_{\gamma,\nu}^{\otimes n} defined by (5). 𝒩γ,ν⊗n(P)=∑r∈{0,1,2,3}nKrPKr*.\mathcal N_{\gamma,\nu}^{\otimes n}(P) =\sum_{r\in\{0,1,2,3\}^n}K_rPK_r^*.(5)✦

The Definition

Definition 1. A finite state ensemble EE on ℋn\mathcal H_n is a family of LL density operators ρa\rho_a with probabilities waw_a, subject to (6). E={(wa,ρa)}a=1L,wa≥0,∑a=1Lwa=1,ρa∈𝒟(ℋn).E=\{(w_a,\rho_a)\}_{a=1}^{L},\qquad w_a\geq0,\qquad\sum_{a=1}^{L}w_a=1,\qquad \rho_a\in\mathcal D(\mathcal H_n).(6)✦ The output Holevo information χ(E;𝒩γ,ν⊗n)\chi(E;\mathcal N_{\gamma,\nu}^{\otimes n}) of a finite state ensemble EE on ℋn\mathcal H_n is the entropy difference in (7), measured in bits. χ(E;𝒩γ,ν⊗n)=S(∑a=1Lwa𝒩γ,ν⊗n(ρa))−∑a=1LwaS(𝒩γ,ν⊗n(ρa))ln⁡2.\chi(E;\mathcal N_{\gamma,\nu}^{\otimes n}) =\frac{ S\!\left(\sum_{a=1}^{L}w_a \mathcal N_{\gamma,\nu}^{\otimes n}(\rho_a)\right) -\sum_{a=1}^{L}w_aS\!\left( \mathcal N_{\gamma,\nu}^{\otimes n}(\rho_a)\right) }{\ln2}.(7) The unrestricted block Holevo capacity χn(γ,ν)\chi_n(\gamma,\nu) is the supremum in (8) over finite ensembles on ℋn\mathcal H_n. χn(γ,ν)=supEχ(E;𝒩γ,ν⊗n).\chi_n(\gamma,\nu) =\sup_{E}\chi(E;\mathcal N_{\gamma,\nu}^{\otimes n}).(8)✦

A classical communication code CC of block length nn consists of MM messages, encoding states ρa\rho_a and a decoding positive operator-valued measurement {Da}\{D_a\} on ℋn\mathcal H_n, with the conditions in (9). C=(M,{ρa}a=1M,{Da}a=1M),M∈ℕ,M≥1,ρa∈𝒟(ℋn),Da≽0,∑a=1MDa=Iℋn.\begin{gathered} C=(M,\{\rho_a\}_{a=1}^{M},\{D_a\}_{a=1}^{M}),\qquad M\in\mathbb N,\quad M\geq1,\\ \rho_a\in\mathcal D(\mathcal H_n),\qquad D_a\succeq0,\qquad\sum_{a=1}^{M}D_a=I_{\mathcal H_n}. \end{gathered}(9)✦ The average decoding error eγ,ν(C)e_{\gamma,\nu}(C) of a classical communication code CC for 𝒩γ,ν\mathcal N_{\gamma,\nu} is the probability of a decoding error for uniformly distributed messages, given by (10). eγ,ν(C)=1−1M∑a=1MRe⁡Tr⁡[Da𝒩γ,ν⊗n(ρa)].e_{\gamma,\nu}(C) =1-\frac1M\sum_{a=1}^{M}\operatorname{Re}\operatorname{Tr} \!\left[D_a\mathcal N_{\gamma,\nu}^{\otimes n}(\rho_a)\right].(10)✦

A real rate RR is achievable for 𝒩γ,ν\mathcal N_{\gamma,\nu} if there is a sequence of codes CnC_n, one for each block length, with message counts MnM_n, satisfying (11); rates are measured in bits per channel use. limn→∞eγ,ν(Cn)=0,ln⁡Mnnln⁡2≥R−εfor every ε>0 and all sufficiently large n.\begin{gathered} \lim_{n\to\infty}e_{\gamma,\nu}(C_n)=0,\\ \frac{\ln M_n}{n\ln2}\geq R-\varepsilon \quad\text{for every $\varepsilon>0$ and all sufficiently large $n$.} \end{gathered}(11)✦ The unassisted classical capacity C(γ,ν)C(\gamma,\nu) is the supremum of achievable real rates in (12). C(γ,ν)=sup⁡{R∈ℝ:R is achievable for 𝒩γ,ν}.C(\gamma,\nu)=\sup\{R\in\mathbb R: R\text{ is achievable for }\mathcal N_{\gamma,\nu}\}.(12)✦

What Is Checked

The optimizations in (8) and (12) range over finite ensembles and sequences of block codes for the fixed channel, with arbitrary density operators on ℋn\mathcal H_n and joint positive operator-valued measurements.