Photon costs and input rigidity at Gaussian entropy limits
Year: 2026
Type: preprint
Abstract: How many photons are needed to reach an optical entropy limit, and how accurately must that limit be approached to identify the full input state? This article determines both for finite products of one-mode Gaussian channels, with exact total input entropy and a mean-photon budget imposed in the laboratory frame. The comparison includes arbitrary non-Gaussian, entangled and infinite-support normal inputs. Exact entropy and photon allocation. For a channel product \(\Phi\), the full frontier is \[M_\Phi(T,E)=\min_{\rho:\,S(\rho)=T,\,\operatorname{Tr}(N\rho)\le E}S(\Phi\rho).\] An exact finite convex allocation gives this value, physically budgeted Gaussian optimizers and the least photon budget at which the entropy-only limit is attained. Moving a squeezed bath between two links can change that threshold even when their pure-input frontiers and entropy-only limits coincide. Singular limits requiring unbounded squeezing have sharp inverse-energy or logarithmically enhanced photon costs, including leading coefficients, matched corrections, inverse error budgets, joint low-entropy crossovers and actual finite-budget allocations. For a given set of links, specify \(T\) and \(E\) in laboratory coordinates and compare \(E\) with this threshold. Below it, the finite-budget allocation gives the correct benchmark for evaluating an observed output entropy (Sections 2–5). Complete limiting recovery curves. For a fixed singular channel product, let \(\sigma_E\) be its finite-budget optimizer, \(\lambda_E\) the actual photon support price, and \(\eta_E(\rho)=S(\Phi\rho)-M_\Phi(T,E)\). Define the worst-case full trace-norm error by \[\omega_E(x)=\sup_{\substack{S(\rho)=T(E),\ \operatorname{Tr}(N\rho)\le E\\ \eta_E(\rho)\le x\lambda_E/E}}\|\rho-\sigma_E\|_1.\] Along every path \(T(E)\to0\), including zero entropy, and for each fixed finite \(x\ge0\), the article proves \[\lim_{E\to\infty}\omega_E(x)=2\sqrt{1-e^{-4\alpha_{\max}x}},\] where \(\alpha_{\max}\) is the largest leading photon share. This determines the whole limiting error curve, not only the scale at which recovery becomes possible. For fixed positive input entropy, when the dominant modes retain positive thermal occupations, the corresponding curve is a trace-norm optimization over displaced thermal states. The displacement ellipsoid is paid for within the original photon cap while preserving the exact input entropy. Thus approaching the optimal output value can require much less accuracy than identifying its source. To read the curve in reverse, choose a target full trace-norm error \(0<d<2\). Any fixed \(x<-\log(1-d^2/4)/(4\alpha_{\max})\) places the limiting worst error strictly below \(d\); the corresponding output-entropy tolerance is \(x\lambda_E/E\) (Section 8). Local optimality and global guarantees can differ. At a fixed strictly active interior budget, the input optimizer is unique and the optimal local square-root stability coefficient is characterized by one joint thermal and coherence price. Nevertheless, a finite feasible disturbance can have a larger ratio \(\|\rho-\sigma_E\|_1/\sqrt{\eta_E(\rho)}\) than every infinitesimal disturbance at the same channel, entropy and photon budget. An explicit replacement–unitary family proves this strict separation and has a sharp Poisson variational limit. The local comparison already includes all quantum-state perturbations; the separation is not an artefact of restricting it to Gaussian states. Other specified mixed-reference instances have the exact common coefficient \(2/\sqrt{\lambda}\). For a given excess \(u\), first check that it lies in the range certified for the local estimate with the desired slack. A guarantee over a finite interval requires a bound valid on that whole interval (Sections 6–10). Recovery to an optimizer set and changing channel families. Above an attainable regular entropy plateau, the geometry of the original photon cap yields matching square-root, quarter-power and logarithmically modified recovery laws to the entire feasible optimizer set. Two changing channel families have exact recovery thresholds governed respectively by entropy-transfer cost and an unequal-temperature group price. Finite thermal-field approximations and, under specified computable-data conditions, certified small-deficit intervals connect the local theory to explicit bounds. All limiting statements retain their stated channel, entropy and resource regimes. For varying products, Section 11 gives a way to price the same tolerance \(u\): reduce the laboratory photon budget and find how many photons can be withdrawn before the optimum rises by \(u\). At positive input entropy with bounded mode count, a strictly active budget or first attainment of a finite plateau gives full-state error of order \(\min\{1,\sqrt{x_u/v_E}\}\), where \(x_u\) is that withdrawal and \(v_E\) is the narrowest reference variance. Growing products can incur a logarithmic mode penalty; the member-specific criterion distinguishes families for which a bound based on mode count alone loses information. Which photons determine the sharp input error? In the two-temperature loss family, divide the allowed output-entropy excess \(u_k\) by the cold photon price \(a_{{\rm c},k}\). Along every positive path \(x_k=u_k/a_{{\rm c},k}\to0\), the worst full trace-norm error is \((C_*+o(1))\sqrt{x_k}\), where \(1.12615949<C_*<1.12615950\). Counting all cold photons together gives a strictly smaller leading coefficient. To attain the full coefficient with a binary count, mark a subspace of cold modes and record whether one to eight cold photons are present and all lie in that subspace. Choose marks whose one-photon Gibbs weight approaches the optimizing weight \(t_*\). This readout attains the worst leading error even though an individual input can carry coherence invisible to counting (Theorem 17 and Figure 8). Which sources approach that error, and when does the coefficient transfer? Theorem 18 characterizes all near-worst inputs in the full trace norm of the original joint state. Their leading perturbations come from the optimal marked count; the marked subspace and the sign can vary. To apply the coefficient to a different temperature distribution in the same physical loss construction, check that the total cold occupation is one and that every \(n_k r_{i,k}\) lies between fixed positive bounds. Under the original entropy and photon constraints, Theorem 19 then gives the same \(C_*\). Proposition 4 explains why the check matters: at the same numerical entropy and photon cap, a single almost empty, ultracold mode can set the minimum photon price and make the response normalized on that scale tend to zero. Appendices AA–AD give the proofs and finite-mode and finite-budget error bounds. This is an independent companion to The active entropy photon-number inequality: exact minima, energy budgets and input rigidity, doi:10.5281/zenodo.24265503. The present article resolves laboratory-frame phase-sensitive photon allocation and the corresponding finite- and growing-budget input-recovery laws. The integrated 294-page article contains the main text, eight figures, complete proofs and references. Distances are full trace norms, ranging from zero to two. Energy constraints are mean-photon budgets, not hard photon-number cutoffs.
Author: He, Zixuan
Institution: University of Glasgow
Language: English
Subfield: Atomic and Molecular Physics, and Optics
Field: Physics and Astronomy
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7234329479
