{"id":"29a02f31-e892-488a-a3ba-224bec122b2a","arxiv_id":"2601.07011","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"Entropy production of correlation-erasing processes in a cavity-QED atom-field system is measured via Bayesian quantum-state estimation, which avoids divergences produced by maximum-likelihood reconstruction.","lead":"A cavity-QED experiment measures the thermodynamic 'entropy production' of processes that erase quantum correlations between an atom and a microwave field, comparing dephasing, full decorrelation, and reset protocols. The authors also show that standard maximum-likelihood state reconstruction produces spurious infinite values, and they implement a Bayesian Monte-Carlo fix.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Backward unitary as described does not implement U_F^†, so measured KLU divergences are not TPM entropy production unless the sequence is recalibrated.","rationale":"The reader's weakest_assumption identified the same load-bearing flaw: the backward evolution is assumed to be U_F† but the described spin-echo-like sequence does not realize it for the Jaynes-Cummings Hamiltonian. I agree this is the most critical point because it undermines the identification of the measured KLU divergence with TPM entropy production for all processes, not just a single one. The reset ideal-value inconsistency is also concerning but is secondary: it affects only the interpretation of the 2-bit ideal for the reset protocol, whereas the backward-unitary error affects the core quantitative claim. The reader's CONDITIONAL verdict is appropriate: the paper should be accepted only after the authors either correct the backward unitary implementation, provide evidence (e.g., calibration data or a corrected derivation) that the sequence effectively inverts U_F, or explicitly reinterpret the measured quantities without relying on the TPM identification. The concrete numerical test I propose would settle whether the concern lands. No ad hominem or theatrical language; this is a technical check of an unverified premise.","tokens_in":13834,"tokens_out":12832,"duration_ms":121215,"concrete_test":"Numerically simulate the ideal sequence U_F σx U_F with H_JC and gt=π/4, acting on |e0⟩; compute fidelity of the final state with |e0⟩ and the KLU divergence D(ρ̃τ||ρ0). If fidelity is not ~1 and divergence is not ~0, the backward unitary is not the inverse. Alternatively, perform the same simulation with the actual coupling profile and pulse calibrations used in the experiment to check whether the backward process reproduces the initial state within experimental error. This would settle whether the TPM identification for the reported entropy production values is valid.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The TPM identification ⟨Σ⟩ = D(ρτ||ρ̃0) = D(ρ0||ρ̃τ) requires the backward evolution to equal the exact inverse of the forward unitary. The paper's implementation (Sec. V) is a first-half π/2 interaction, a central σx pulse on the atom, then a second-half π/2 interaction. For the resonant Jaynes-Cummings Hamiltonian H_JC = ħg(σ+ a + σ- a†), the atomic σx does not anticommute with H_JC; indeed σx H_JC σx = ħg(σ- a + σ+ a†) ≠ -H_JC. Consequently U_F σx U_F is not the identity and U_B ≠ U_F^†. For the initial state |e0⟩ and gt=π/4, the ideal no-environment cycle maps |e0⟩ to a state with zero overlap with |e0⟩ (components |g0⟩, |e1⟩, |g2⟩), so D(ρ̃τ||ρ0) would diverge, not approach zero. The text provides no derivation or calibration showing that the realized pulse sequence acts as U_F^†; the spin-echo analogy fails because σx does not reverse the sign of H_JC. This affects the entropy production values for all three processes, since the final states ρ̃τ are generated by the erroneous backward evolution. Without a corrected backward operation or a clear demonstration that the sequence effectively implements the inverse, the central experimental claim is unsupported.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper reports an experimental cavity-QED study of entropy production for three non-thermal, correlation-erasing processes within the two-point-measurement (TPM) framework: local dephasing, complete decorrelation, and reset (local thermalization to a pure state). The authors measure the Kullback-Leibler-Umegaki divergence D(ρ0||ρ̃τ) between the initial and final states of a forward-backward atom-cavity cycle, with the intermediate process erasing correlations. They report values close to the ideal 0, 1, and 2 bits for the three processes, obtained from full quantum-state tomography. A central methodological contribution is the demonstration that standard maximum-likelihood reconstruction produces spurious divergences of the KL divergence due to rank-deficient states, and the introduction of a Bayesian Monte-Carlo estimator that samples density matrices according to their likelihood, yielding finite and stable estimates.","tokens_in":14098,"tokens_out":19976,"duration_ms":178612,"significance":"If the reported results are correct, this is a valuable experimental test of entropy production for processes that are not thermalization, and a practical recipe for estimating nonlinear, potentially divergent functionals of reconstructed quantum states. The Bayesian estimator is a useful contribution to quantum state reconstruction. The paper also tackles a conceptually important issue: the behavior of the KL divergence for states with different support, and the practical pitfalls in estimating it from experimental data. The experimental setup and data treatment are sophisticated. However, the central quantitative claims rest on the correct implementation of the backward unitary evolution, and that premise is not established in the manuscript as written.","major_comments":[{"comment":"The text states that the backward unitary is U_B = U_F† and describes its implementation as: 'It starts by first applying the σx operation on the atomic qubit state and then activating the resonant π/2 interaction as in U_F.' For the resonant Jaynes-Cummings Hamiltonian H = ħg(σ+ a + σ- a†), σx does not anticommute with H; indeed σx H σx = ħg(σ- a + σ+ a†) ≠ -H. Consequently U_F σx ≠ U_F†. For the ideal no-environment cycle, this sequence maps |e0⟩ to a state with zero overlap with |e0⟩ (e.g., for the π/2 pulse, components |g0⟩, |e1⟩, |g2⟩), so D(ρ̃τ||ρ0) would diverge, not approach zero. The paper provides no derivation, alternative sequence, or calibration showing that the realized sequence effectively acts as U_F†. Since Eq. (1) and the identification of the measured divergences with the TPM entropy production depend critically on U_B = U_F†, this is a load-bearing issue for all three","section":"Section V, 'No environment action' (Eq. 17 and following text)"},{"comment":"Even if the intended sequence is different from the literal description, the manuscript does not provide an experimental verification that the no-environment cycle closes the loop as in Eq. (17). The measured ⟨Σ_id⟩ is reported as 'larger than its ideal zero value' but its magnitude (roughly 0.2 bits in Fig. 4) is not shown to be consistent with the described pulse sequence. A direct test—e.g., full tomography of the state after the backward operation and comparison of its fidelity with ρ0—is essential to validate that U_B = U_F†. Without such a calibration, the deviation of ρ̃τ from ρ0 can be attributed either to the flawed backward operation or to physical imperfections, and the TPM interpretation is not justified.","section":"Section V, 'No environment action'"},{"comment":"The reset process is described as 'local thermalization' with the role of the thermal state played by the initial pure state |0⟩⟨0|. However, Eq. (8) contains the term D(ρ_B||ζ_B); if ζ_B is taken to be the pure state |0⟩⟨0|, the divergence is infinite for any ρ_B with support outside |0⟩, which is the case for the ideal cavity state after forward evolution. The paper avoids this by computing the entropy production through the cycle (D(ρ0||ρ̃τ)), but the connection to Eq. (8) is left unclear. The authors should clarify whether the reset protocol is meant to be a concrete realization of Eq. (8) and, if so, how the finite 2-bit value is obtained.","section":"Section II, Eq. (8) and Section V, 'Local thermalization'"}],"minor_comments":[{"comment":"There is a typographical error: 'ΠB_il' should be 'ΠB_l'.","section":"Eq. (2)"},{"comment":"The phrase 'similar to the spin-echo technique' is misleading because the σx pulse does not reverse the sign of the Jaynes-Cummings interaction. The analogy should be removed or carefully explained.","section":"General"},{"comment":"The Monte-Carlo uncertainty estimates in Fig. 4 are not defined. It would be helpful to state explicitly whether the error bars represent the standard deviation of the posterior distribution of the estimator, the dispersion from experimental repetitions, or the Monte-Carlo sampling error.","section":"Section IV, 'Monte-Carlo approach'"},{"comment":"The sentence 'Possibly, the smaller value of I_deph compared to ⟨Σ_deph⟩ results from some systematic uncertainty in the atom-cavity interaction that is partially compensated in the forward-backward cycle' is vague. If a systematic effect is suspected, it should be quantified or described.","section":"Section V, 'Local dephasing'"},{"comment":"The Metropolis proposal distribution is stated to satisfy detailed balance, but the argument is brief. A short derivation or reference would help.","section":"Appendix A"}],"recommendation":"major_revision","confidential_remarks":"The central issue is the backward evolution. The stress-test concern is valid: the described implementation does not equal U_F†, and the paper provides no calibration. If the actual experiment does realize the inverse (e.g., through a different pulse sequence), the paper needs to state it explicitly and provide data demonstrating that the no-environment cycle returns the state to ρ0. The Bayesian estimator and the experimental data treatment are otherwise sound. I recommend major revision with the requirement to resolve the backward-evolution issue."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know this paper reports real experimental work and one genuinely useful methodological lesson, but its central quantitative claims are shaky on two textually verifiable points.\n\nWhat is genuinely good: the observation that maximum-likelihood state reconstruction produces spurious divergences in the KLU divergence is correct and important for anyone computing entropic quantities from tomography data. The Bayesian Monte-Carlo estimator they build to average nonlinear functionals over posterior-sampled states is a sensible remedy, and the experimental setup is impressive: full tomography of an atom-cavity system at successive stages of the cycle, with careful handling of imperfections. The measured values sitting near 0, 1, and 2 bits for the identity, dephasing, and complete-decorrelation cases give the qualitative picture real weight.\n\nThe soft spots are both load-bearing. First, the reset protocol's claimed ideal value of 2 bits contradicts the paper's own Eq. (8). Equation (8) gives the entropy production of local thermalization as I_AB + D(ρ_B||ζ_B). In the reset protocol, the cavity is reset to the pure vacuum state |0⟩⟨0|, so ζ_B is rank-one. For the reduced cavity state after the forward process—which is a mixture of |0⟩ and |1⟩—D(ρ_B|||0⟩⟨0|) diverges. The paper's own conclusions section discusses exactly this divergence class for pure reference states. So the red line at 2 bits in Fig. 4 is not the ideal value; the ideal value is infinite. The measured finite value near 2 bits can only emerge from imperfections that make the states full-rank, which would need to be stated and derived.\n\nSecond, the backward unitary as described does not implement U_F†. The sequence is a first half of the resonant interaction, a central σx pulse, then a second half. For the Jaynes-Cummings Hamiltonian, σx commutes with H_JC (σx H_JC σx = H_JC, not −H_JC), so U(π/4) σx U(π/4) equals σx U(π/2), not U(−π/2). The no-environment cycle therefore does not return to ρ0, and the identification ⟨Σ⟩ = D(ρ_0||ρ̃_τ) fails. This affects all three measured entropy productions, not just the identity case. The paper gives no calibration showing the realized sequence actually acts as U_F†, and the spin-echo analogy does not work here.\n\nThese are not minor issues. The qualitative message—that erasing quantum correlations carries an entropic cost and that MLE is dangerous for such estimates—is probably robust. But the quantitative claims as stated are not supported. The Bayesian estimator and the MLE-divergence observation could be separated from the flawed cycle implementation and still stand.\n\nWho this is for: researchers in quantum thermodynamics and quantum state estimation. It deserves a serious referee, but the authors need to correct or recalibrate the backward operation, fix the reset ideal-value claim, and likely revise the reported entropy-production numbers. I would not cite the quantitative results until that happens.","headline":"A serious experimental paper whose two load-bearing quantitative claims—the reset ideal value and the backward-unitary implementation—don't hold as written.","tokens_in":14758,"tokens_out":11212,"would_cite":false,"duration_ms":111640,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["42.50.Pq","05.70.Ln"],"model":"deepseek-v4-flash","headline":"The paper reports direct measurement of entropy production for processes that erase quantum correlations, obtaining values close to 0, 1, and 2 bits and showing that standard maximum-likelihood reconstruction spuriously diverges while a Bay","keywords":["entropy production","two-point measurement","Kullback-Leibler divergence","cavity QED","quantum state tomography","decorrelation","quantum coherence","Bayesian estimation"],"falsifier":"Perform full quantum-state tomography on the state after the forward-backward cycle with no intermediate decorrelation; if the reconstructed state is not ρ0 (up to known decoherence), the backward evolution is not the exact inverse and the identification with Eq. (1) fails. A numerical evaluation of the implemented sequence (half-atom-cavity interaction, σx pulse, second half-interaction) under the Jaynes-Cummings Hamiltonian, compared with U_F†, would also settle it.","tokens_in":13592,"feed_emoji":"⚛️","tokens_out":7900,"duration_ms":75190,"temperature":0.7,"pith_summary":"The authors aim to show that entropy production, usually defined for thermalization with a heat bath, can be assessed experimentally for purely informational processes that erase correlations between a two-level atom and a cavity field. Using a two-point measurement scheme, they identify entropy production with the Kullback-Leibler-Umegaki divergence between the states before and after the forward-backward cycle. They demonstrate that local dephasing, complete decorrelation, and reset processes yield measured entropy productions close to the ideal values of 1, 2, and 2 bits respectively. They further show that standard maximum-likelihood reconstruction of the quantum states causes spurious divergences in these entropic quantities, and propose a Bayesian Monte-Carlo estimator that avoids the problem.","feed_headline":"Cavity QED measures entropy cost of erasing correlations","feed_subtitle":"0, 1, 2 bits for identity, dephasing, and full decorrelation of an atom-cavity state.","key_machinery":"The engine of the result is the two-point measurement (TPM) scheme for a forward-backward cycle: a controllable forward unitary U_F, an intermediate irreversible CPTP map E, and a controllable backward unitary U_B = U_F†. The entropy production of E is the Kullback-Leibler-Umegaki divergence D(ρ1||ρ2) = -Tr[ρ1 ln ρ2] - S(ρ1). For the specific decorrelating maps, this divergence reduces to the relative entropy of coherence (local dephasing) and to the mutual information (complete decorrelation), linking irreversibility to information-theoretic correlation measures. To make the quantity experimentally accessible, the authors use a Bayesian Monte-Carlo sampling of density matrices weighted by t","core_discovery":"The central claim is that the entropy production of an irreversible decorrelating process is given by the Kullback-Leibler-Umegaki divergence, ⟨Σ⟩ = D(ρτ || ρ̃0) = D(ρ0 || ρ̃τ), and that this quantity can be measured experimentally. For a resonant π/2 atom-cavity interaction creating a maximally entangled state, local dephasing of one subsystem yields ⟨Σ⟩ = 1 bit (the relative entropy of coherence), complete decorrelation yields ⟨Σ⟩ = 2 bits (the mutual information), and resetting one subsystem to its initial pure state yields ⟨Σ⟩ = 2 bits. The authors implement these processes in a cavity-QED experiment and reconstruct the states by tomography; the measured entropy productions approach the","pith_inferences":["The observed spurious divergences may be a generic artifact for near-pure states in any experiment estimating relative entropies from MLE reconstructions; the Bayesian approach could be widely adopted.","If the backward pulse sequence is not exactly the inverse (which can be checked by full tomography of the identity cycle), the reported values should be read as operational measures of the whole protocol rather than intrinsic properties of the decorrelating maps.","The two-copy simulation used to implement complete decorrelation suggests a general method to realize non-completely-positive operations in the laboratory, which could be exploited beyond thermodynamics."],"forward_implications":["For local dephasing, complete decorrelation, and reset, the measured entropy production approximates the ideal values of 1, 2, and 2 bits, confirming that these information-erasing processes have a quantifiable entropic cost.","The identity cycle yields entropy production near zero, so the protocol can serve as a calibration benchmark for the quality of the forward-backward unitaries.","The Bayesian Monte-Carlo estimator provides a general recipe to compute nonlinear functions of reconstructed quantum states, avoiding the divergent artifacts of maximum-likelihood reconstruction.","In the TPM scheme, entropy production of non-thermal processes becomes experimentally accessible without a heat reservoir, broadening the scope of quantum thermodynamic experiments."],"fun_headline_variants":["Cavity QED measures quantum erasure cost: 1-2 bits","Quantum irreversibility measured: 1 bit for dephasing, 2 for full erasure","Cavity QED weighs entropy cost of erasing correlations","Irreversibility quantified: erasing correlations costs up to 2 bits","Cavity QED puts a number on erasing quantum correlations"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The backward step of the cycle is assumed to exactly undo the forward interaction, so that any measured entropy production is attributed solely to the intermediate decorrelating process.","fun_headline_variants_meta":{"raw":{"variants":["Cavity QED measures quantum erasure cost: 1-2 bits","Quantum irreversibility measured: 1 bit for dephasing, 2 for full erasure","Cavity QED weighs entropy cost of erasing correlations","Irreversibility quantified: erasing correlations costs up to 2 bits","Cavity QED puts a number on erasing quantum correlations"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001311,"raw_usage":{"total_tokens":5143,"prompt_tokens":668,"completion_tokens":4475,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":412,"completion_tokens_details":{"reasoning_tokens":4373}},"tokens_in":412,"tokens_out":4475,"duration_ms":29033,"temperature":1.0,"reasoning_tokens":4373,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T11:16:44.471707+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Perform full quantum-state tomography on the state after the forward-backward cycle with no intermediate decorrelation; if the reconstructed state is not ρ0 (up to known decoherence), the backward evolution is not the exact inverse and the identification with Eq. (1) fails. A numerical evaluation of the implemented sequence (half-atom-cavity interaction, σx pulse, second half-interaction) under the Jaynes-Cummings Hamiltonian, compared with U_F†, would also settle it.","supporting_citations":[],"review_version":1}