OctoInformBeta

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

Lemma 4.2: damping-power specialization: Normalized pure inputs on n+1n+1 qubits, for a nonnegative integer nn, under one fixed generalized amplitude-damping channel

A recursive entropy bound for amplitude damping.

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

Consider repeated uses of a generalized amplitude-damping channel with damping γ∈[0,1]\gamma\in[0,1] and stationary excited-state population ν∈[0,1]\nu\in[0,1]. Separating the first qubit of a pure input gives two amplitude branches on the remaining qubits.

The computational basis ℬm\mathcal B_m of mm qubits is the set of bit strings in (1), where Fin⁡(m)={0,…,m−1}\operatorname{Fin}(m)=\{0,\ldots,m-1\}.✦ An mm-qubit operator is a complex matrix with rows and columns indexed by ℬm\mathcal B_m.✦ ℬm={i:Fin⁡(m)→Fin⁡(2)}.\mathcal B_m=\{i:\operatorname{Fin}(m)\to\operatorname{Fin}(2)\}.(1)✦

The single-qubit Kraus operators K0(1),…,K3(1)K^{(1)}_0,\ldots,K^{(1)}_3 are the matrices in (2). K0(1)=1−ν(1001−γ),K1(1)=γ(1−ν)(0100),K2(1)=ν(1−γ001),K3(1)=γν(0010).\begin{aligned} K^{(1)}_0&=\sqrt{1-\nu} \begin{pmatrix}1&0\\0&\sqrt{1-\gamma}\end{pmatrix}, &K^{(1)}_1&=\sqrt{\gamma(1-\nu)} \begin{pmatrix}0&1\\0&0\end{pmatrix},\\ K^{(1)}_2&=\sqrt{\nu} \begin{pmatrix}\sqrt{1-\gamma}&0\\0&1\end{pmatrix}, &K^{(1)}_3&=\sqrt{\gamma\nu} \begin{pmatrix}0&0\\1&0\end{pmatrix}. \end{aligned}(2)✦

For a label r:Fin⁡(m)→Fin⁡(4)r:\operatorname{Fin}(m)\to\operatorname{Fin}(4), the tensor product Kraus operator KrK_r is the mm-qubit operator with entries given by (3). Kr(i,j)=∏k=0m−1Kr(k)(1)(i(k),j(k)).K_r(i,j)=\prod_{k=0}^{m-1}K^{(1)}_{r(k)}\bigl(i(k),j(k)\bigr).(3)✦ The memoryless channel 𝒜γ,ν⊗m\mathcal A_{\gamma,\nu}^{\otimes m} is the map on mm-qubit operators defined in (4), with ** denoting the adjoint. 𝒜γ,ν⊗m(P)=∑r:Fin⁡(m)→Fin⁡(4)KrPKr*.\mathcal A_{\gamma,\nu}^{\otimes m}(P) =\sum_{r:\operatorname{Fin}(m)\to\operatorname{Fin}(4)}K_r P K_r^*.(4)✦

For complex amplitudes φ\varphi on a finite basis ℐ\mathcal I, the squared norm M(φ)M(\varphi) and pure state operator Pφ=|φ⟩⟨φ|P_\varphi=|\varphi\rangle\langle\varphi| are defined by (5). M(φ)=∑i∈ℐ|φ(i)|2,Pφ(i,j)=φ(i)φ(j)¯.M(\varphi)=\sum_{i\in\mathcal I}|\varphi(i)|^2, \qquad P_\varphi(i,j)=\varphi(i)\overline{\varphi(j)}.(5) Fix a reference element i*∈ℐi_*\in\mathcal I. The reference basis vector e*e_* and normalized amplitude vector φˆ\widehat\varphi for complex amplitudes φ\varphi on the finite basis ℐ\mathcal I are the vectors on ℐ\mathcal I given by (6). e*(i)={1,i=i*,0,i≠i*,φˆ={e*,M(φ)=0,φ/M(φ),M(φ)≠0.e_*(i)=\begin{cases}1,&i=i_*,\\0,&i\ne i_*,\end{cases} \qquad \widehat\varphi= \begin{cases} e_*,&M(\varphi)=0,\\ \varphi/\sqrt{M(\varphi)},&M(\varphi)\ne0. \end{cases}(6)

The branch amplitudes ψb\psi_b of an (n+1)(n+1)-qubit amplitude vector ψ\psi, for b∈{0,1}b\in\{0,1\}, are its restrictions in (7); (b,i)(b,i) denotes the bit string with first bit bb and tail ii. ψb(i)=ψ(b,i),i∈ℬn.\psi_b(i)=\psi(b,i),\qquad i\in\mathcal B_n.(7)✦ The excitation population pj(φ)p_j(\varphi) at site j∈Fin⁡(m)j\in\operatorname{Fin}(m) of amplitudes φ\varphi on ℬm\mathcal B_m is the sum in (8). pj(φ)=∑i∈ℬmi(j)=1|φ(i)|2.p_j(\varphi)=\sum_{\substack{i\in\mathcal B_m\\i(j)=1}}|\varphi(i)|^2.(8)✦

The entropy S(P)S(P) of a finite complex square matrix PP is the real trace of the spectral functional calculus for η(x)=−xln⁡x\eta(x)=-x\ln x, as in (9).✦ Here 0ln⁡0=00\ln0=0, and entropies are measured in nats. S(P)=Re⁡Tr⁡η(P),S(ρ)=−Tr⁡(ρln⁡ρ)for a density matrix ρ.S(P)=\operatorname{Re}\operatorname{Tr}\eta(P), \qquad S(\rho)=-\operatorname{Tr}(\rho\ln\rho) \quad\text{for a density matrix }\rho.(9)✦ The qubit entropy function gg and output determinant vv are defined by (10), where hh is binary entropy with natural logarithms. h(q)=−qln⁡q−(1−q)ln⁡(1−q),g(u)=h(1+1−4u2),v(γ,ν,p)=γν(1−ν)+γ(1−γ)(p−ν)2.\begin{aligned} h(q)&=-q\ln q-(1-q)\ln(1-q),\\ g(u)&=h\!\left(\frac{1+\sqrt{1-4u}}2\right),\\ v(\gamma,\nu,p)&=\gamma\nu(1-\nu) +\gamma(1-\gamma)(p-\nu)^2. \end{aligned}(10)

The Claim

Theorem 1. Let γ,ν∈[0,1]\gamma,\nu\in[0,1], let nn be a nonnegative integer, and let ψ\psi be amplitudes on ℬn+1\mathcal B_{n+1} with M(ψ)=1M(\psi)=1. Put p=p0(ψ)p=p_0(\psi) and wb=M(ψb)w_b=M(\psi_b) for b∈{0,1}b\in\{0,1\}. The output entropies satisfy (11). S(𝒜γ,ν⊗(n+1)(Pψ))≥g(v(γ,ν,p))+∑b=01wbS(𝒜γ,ν⊗n(Pψbˆ)).S\!\left(\mathcal A_{\gamma,\nu}^{\otimes(n+1)}(P_\psi)\right) \ge g\bigl(v(\gamma,\nu,p)\bigr) +\sum_{b=0}^{1}w_b\, S\!\left(\mathcal A_{\gamma,\nu}^{\otimes n} (P_{\widehat{\psi_b}})\right).(11)✦

In the paper, the result reads: for γ,ν∈[0,1]\gamma,\nu\in[0,1], put 𝒜=𝒜γ,ν\mathcal A=\mathcal A_{\gamma,\nu} and e(p)=g(v(γ,ν,p))e(p)=g(v(\gamma,\nu,p)); if 𝒩:ℒ(Hin)→ℒ(Hout)\mathcal N:\mathcal L(H_{\mathrm{in}})\to\mathcal L(H_{\mathrm{out}}) is a completely positive trace-preserving map between operators on nonzero finite-dimensional complex Hilbert spaces, where ℒ(H)\mathcal L(H) denotes the space of linear operators on HH, and χ=1−p|0⟩⊗χ0+p|1⟩⊗χ1\chi=\sqrt{1-p}\,|0\rangle\otimes\chi_0+\sqrt{p}\,|1\rangle\otimes\chi_1 is a unit vector in ℂ2⊗Hin\mathbb C^2\otimes H_{\mathrm{in}}, with p∈[0,1]p\in[0,1] and unit branch vectors chosen arbitrarily at zero weight, then S((𝒜⊗𝒩)(Pχ))≥e(p)+(1−p)S(𝒩(Pχ0))+pS(𝒩(Pχ1))S((\mathcal A\otimes\mathcal N)(P_\chi))\ge e(p)+(1-p)S(\mathcal N(P_{\chi_0}))+pS(\mathcal N(P_{\chi_1})).

What Is Checked

The claim is stated over pure inputs of squared norm one and powers of a fixed generalized amplitude-damping channel at parameters γ,ν∈[0,1]\gamma,\nu\in[0,1], with branch normalization as in (6).