{"id":"74deb549-001c-428e-aad3-6feb397756e7","arxiv_id":"2608.11488","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"Non-Markovian bath memory modifies the standard heat-exchange fluctuation theorem by a factor built from stochastic free energies and mutual information, and a qubit example shows enhanced reverse heat flow.","lead":"This paper derives a modified heat-exchange fluctuation theorem for baths with memory, using a quantum collisional model in which neighboring auxiliary particles interact before each heat-exchange collision. The correction is built from memory-induced free-energy and information terms, and a qubit example shows that memory makes heat flow against the temperature gradient more likely.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Eqs. (9)–(12) mix labels of fresh and old auxiliary units, so the central XFT derivation does not follow as written; C_dir_M-M=1 is asserted from a false thermal assumption instead of the actual cancellation.","rationale":"The reader's weakest assumption identifies a genuine gap: the paper drops Δf and ΔI in the Markovian limit by claiming the states are thermal, which is false for the steady-state pair σ_ss = U_AB(γ_A⊗γ_B)U_AB†. However, the underlying problem is broader: the index conventions in Eqs. (9)–(12) are inconsistent, so the derivation of the modified XFT cannot be verified as written. A corrected derivation (with q_{mα} for the old pair) shows that C_dir_M-M=1 actually arises from the old pair being a spectator, not from thermality; the paper never provides that argument. This is load-bearing because the recovery of the standard JW-XFT is a central advertised result, and the correction factors in Eq. (15) are the main new quantities. The concern is addressable by rewriting the derivation with consistent labels and stating the correct cancellation mechanism. Since this is a major but fixable presentation flaw rather than a demonstrated wrong result, the reader's CONDITIONAL verdict remains appropriate; the numerical agreement claimed in Figs. 3 and 4 suggests the underlying formulas may be correct once the notation is repaired. The lack of a general proof for the 'enhancement' claim is noted but is not the primary issue, as the paper frames it as an illustrative example.","tokens_in":17092,"tokens_out":19671,"duration_ms":157503,"concrete_test":"Re-derive Eqs. (12)–(15) with consistent subsystem labels: set q_{mα} = ⟨m α|σ_ss|m α⟩ for the old pair (A_{n-1},B_{n-1}) and p_i^{βA}p_μ^{βB} for the fresh pair (A_n,B_n). For U_AA=U_BB=I, show that the transition amplitude imposes n=m and ξ=α, so q_{mα} cancels in the forward/reverse ratio and C_dir_M-M=1 identically, without assuming q is thermal. Additionally, numerically evaluate the right-hand side of Eq. (15) for a qubit example with U_AA=I using the paper's own mixed labels; if C_dir_M-M deviates from 1, the stated recovery of the JW-XFT fails.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central derivation is not internally consistent. In Eqs. (9)–(11), q_{iα}^{AB} and q_{jξ}^{AB} are diagonal elements of the steady state σ_ss^{AB} of the old auxiliary pair A_{n-1}B_{n-1}, but the text states that i_A and α_B label the nth (fresh) units while m_A and μ_B label the (n−1)-th units. Thus q_{iα} refers to a hybrid pair (A_n, B_{n-1}) that is never a physical subsystem; the old pair should be labeled (m, α). Similarly, in Eq. (12), Δf^A_{ji} uses indices j,i for A_n, not for the old A_{n-1}, so the free-energy changes of the old units never appear. Consequently, the ratio in Eq. (12) and the correction factors in Eq. (15) do not follow from the stated forward and reverse probabilities. The paper justifies C_dir_M-M=1 by claiming that the relevant states are thermal so Δf and ΔI vanish. But for U_AA=U_BB=I, the steady state is σ_ss = U_AB(γ_A⊗γ_B)U_AB†, which is not a thermal product; the correct reason the correction factor vanishes is that the old pair is left unchanged by the unitary (n=m, ξ=α), so its contributions cancel. This cancellation is not stated or derived. As written, the advertised recovery of the standard JW-XFT in the Markovian limit is an assertion rather than a consequence of the derivation.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a generalization of the Jarzynski–Wójcik exchange fluctuation theorem to non-Markovian baths using a quantum collisional model in which intra-bath collisions between successive auxiliary units generate memory. The central claim is a modified fluctuation relation P(Q)/P_B(-Q) = e^{ΔβQ} C, where C is expressed through conditional averages of stochastic free-energy changes and stochastic mutual-information differences of the auxiliary steady states. The authors treat both direct bath–bath interactions and probe-mediated heat exchange, state that the Markovian limit recovers the standard JW-XFT, and support the result with numerical examples using qubit auxiliaries, reporting that increasing bath memory enhances the relative probability of reverse heat transfer.","tokens_in":17427,"tokens_out":29254,"duration_ms":248973,"significance":"If the derivation were correct as written, the paper would provide a useful microscopic bridge between environmental memory and heat-exchange fluctuation statistics, with explicit, computable correction factors. The use of a transparent two-point-measurement protocol and the inclusion of concrete qubit numerics are strengths, as are the explicit formulas for the correction factors and the stated recovery of the standard XFT in the Markovian limit. However, the central derivation contains index-labeling inconsistencies and unjustified Markovian-limit simplifications, so the advertised result is not established in the present form. The core idea is plausible and the issues appear fixable, but the load-bearing formulas need to be rederived and the numerical checks repeated with corrected indices.","major_comments":[{"comment":"The indices of the old and fresh auxiliary units are interchanged in the central derivation. The text states that i_A and μ_B label the nth (fresh) units while m_A and α_B label the (n−1)-th (old) units, but Eq. (9) writes the TPM initial probability as p^{βA}_m p^{βB}_μ q^{AB}_{iα}. Since σ_ss is the state of the old pair, the correct probability is p^{βA}_i p^{βB}_μ q^{AB}_{mα}. Consequently, Eq. (12) should involve Δf^A_{nm}, Δf^B_{ξα}, and ΔI_{nξ,mα}, not Δf^A_{ji}, Δf^B_{ξα}, and ΔI_{jξ,iα}. As written, Eqs. (10)–(12) do not follow from the stated forward and reverse trajectories, and the central XFT derivation is internally inconsistent.","section":"Sec. III B, Eqs. (9)–(12)"},{"comment":"The Markovian-limit formulas are not justified by the text's thermal-state argument. With U_AA = U_BB = I, the old pair before the next collision is σ_ss = U_AB(γ_A ⊗ γ_B)U_AB†, which is generically neither thermal nor product; the correct reason C_dir_M-M = 1 is that the old pair is left unchanged (n = m and ξ = α), so all free-energy and mutual-information differences vanish identically. When only one bath is Markovian, the corresponding old index is fixed but the other is not, and the mutual-information difference ΔI_{mξ,mα} is generically nonzero; the formula C_dir_M-NM in Eq. (15) therefore omits a factor e^{−ΔI_{mξ,mα}}. The analogous issue affects the probe-mediated factors in Eq. (21), where the M–M and M–NM cases omit nonvanishing mutual-information terms involving the probe and the unchanged old subsystem. The derivation must be redone with the corrected indices and the missing terms included unless additional product-structure assumptions are stated and proved.","section":"Sec. III B, Eq. (15); Sec. III C, Eq. (21)"},{"comment":"The reverse Hamiltonian is defined as ~H(t) = ΘH(t2−t)Θ^{-1} with no minus sign. For a time-reversal-invariant interaction such as H_{+-}, the unitary obtained from the reversed collision sequence is not ΘU†Θ^{-1}: with U = U_AB(U_AA ⊗ U_BB), sequence reversal gives (U_AA ⊗ U_BB)U_AB, while U† = (U_AA† ⊗ U_BB†)U_AB†. The correct time-reversed Hamiltonian is ~H(t) = −ΘH(t2−t)Θ^{-1}; the missing minus sign is load-bearing for the reverse probability in Eq. (11). There is also a sign-convention mismatch between the example in Sec. III B, which uses U_AA = e^{+iκ_A H_AA}, and Appendix A, which uses U_AA = e^{−iκ_A H_AA}; this should be harmonized.","section":"Appendix A, Eqs. (A3)–(A5)"}],"minor_comments":[{"comment":"The first sentence of the Conclusion contains a typo: 'fluctuation theroem' should be 'fluctuation theorem'.","section":"Sec. IV"},{"comment":"The notation Jχ ∈ {NM, M} is never defined; please state explicitly that it denotes whether each bath is Markovian or non-Markovian.","section":"Sec. III B, Eq. (14)"},{"comment":"The Jensen-step leading to Δβ⟨Q⟩ + ⟨ln C⟩ ≥ 0 uses the fact that the conditional averages in C are positive; it would be helpful to state this positivity explicitly.","section":"Sec. III B, Eq. (8)"},{"comment":"The captions describe the plotted quantity as the left-hand side of Eq. (14) or Eq. (20), while the dashed line is e^{Δβω}; since the right-hand side is e^{Δβω}C with C ≠ 1 in general, the dashed line is the Markovian prediction rather than the right-hand side using the plotted parameters, and this should be clarified.","section":"Figs. 3 and 4"},{"comment":"The numerical agreement in Figs. 3 and 4 is reported as a validation of Eq. (14) or Eq. (20), but the plotted equality checks the implemented formulas against the directly computed ratio; because the derivation leading to those formulas has index inconsistencies, the numerics should be repeated after the indices and missing terms are corrected.","section":"Sec. III B, Eq. (14)"}],"recommendation":"major_revision","confidential_remarks":"The paper fits the scope of the journal and the intended generalization is of interest, but the central derivation as written contains fixable yet load-bearing errors: incorrect index labeling in the TPM probabilities, an unjustified thermal-state assumption in the Markovian limit, and a sign error in the time-reversal construction. The numerical results do not resolve these because they are consistency checks of the same formulas. I recommend major revision rather than rejection, provided the authors rework Secs. III B, III C, and Appendix A with the corrected indices and rerun the examples."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here is my honest read. The idea is right and the final formulas are probably correct, but the derivation as printed has a genuine labeling bug and a wrong Markovian-limit argument. I would not trust the paper until both are fixed.\n\nWhat is new: the authors set up a collisional model in which intra-bath interactions between successive auxiliaries generate non-Markovian memory, and they derive a modified Jarzynski–Wójcik exchange fluctuation theorem for both direct and probe-mediated heat exchange. The correction factor, written as a conditional average of exponentials of stochastic free-energy and mutual-information differences, is a natural extension of results for correlated initial states (Refs. [42,84,85]), but applied to a specific steady-state collisional construction that I have not seen before. The qubit example is concrete and the numerics show the ratio P(Q)/P_B(−Q) decreasing with intra-bath coupling, which is a plausible signature of enhanced reverse heat flow.\n\nThe soft spots are real. In Eq. (9), the probability q_{iα} is written as a diagonal element of the steady state of the old auxiliary pair, but the indices i and α refer to the fresh nth A unit and the (n−1)th B unit respectively. That object is not a physical density-matrix element. The same index mix-up propagates into Eq. (11) and into the free-energy differences, so the ratio in Eq. (12) does not follow as printed. And the Markovian limit C=1 is justified by claiming the old auxiliary pair is thermal; that is false. When U_AA = U_BB = I, the old pair is in the state U_AB(γ_A⊗γ_B)U_AB†, which is generally not a thermal product. The actual reason the correction factor vanishes is that the old pair is left untouched by the unitary, so its indices cancel and its energy changes are zero. I believe the intended derivation goes through once the labels are fixed and the cancellation argument is substituted for the thermal claim, and the final correction-factor formulas are consistent with that fix. The claim about enhanced reverse heat flow, however, is demonstrated only for the qubit example; the paper does not prove a general monotonic relation between non-Markovianity and the correction factor, so that statement should be scaled back.\n\nWho should read this: people working on fluctuation theorems in collisional models and non-Markovian quantum thermodynamics. It deserves a serious referee rather than a desk reject, but my recommendation is major revision. The authors need to rewrite the derivation with consistent indexing, replace the thermal assumption with the actual cancellation, and separate proven results from numerical illustrations.","headline":"A promising generalization of the JW exchange fluctuation theorem to non-Markovian collisional baths, but the central derivation has an index-labeling error and a wrong Markovian-limit argument that must be fixed.","tokens_in":17931,"tokens_out":8274,"would_cite":false,"duration_ms":85498,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Bath memory changes the exchange fluctuation theorem by a correction factor built from stochastic free-energy and mutual-information terms.","keywords":["exchange fluctuation theorem","non-Markovian baths","quantum collisional model","heat exchange statistics","stochastic free energy","mutual information","two-point measurement","quantum thermodynamics"],"falsifier":"Evaluate the left- and right-hand sides of Eq. (14) for two qubit baths with $U_{AA}=U_{BB}=I$, using the actual steady-state joint distribution $q^{AB}_{i\\alpha}=\\langle i\\alpha|\\sigma^{\\rm ss}_{AB}|i\\alpha\\rangle$ in the two-point-measurement probabilities without imposing thermal marginals; if $P(Q)/P_B(-Q)$ differs from $e^{\\Delta\\beta Q}$, the Markovian-limit claim $C^{\\rm dir}_{\\rm M-M}=1$ fails. A direct check is to compute the marginal $\\mathrm{Tr}_B[U_{AB}(\\gamma_A\\otimes\\gamma_B)U_{AB}^\\dagger]$: if it is not the Gibbs state at $\\beta_A$, the free-energy term for a memoryless bath $A$ is not zero.","tokens_in":16878,"feed_emoji":"🔥","tokens_out":12765,"duration_ms":108487,"temperature":0.7,"pith_summary":"This paper asks how the statistics of heat exchange between two thermal baths change when the baths retain memory of their past interactions. In a quantum collisional model where successive bath auxiliaries collide with each other before the heat-exchanging collision, the authors derive an exchange fluctuation theorem of the form $P(Q)/P_B(-Q) = e^{\\Delta\\beta Q} C$, with $C$ built from conditional averages of stochastic free-energy changes and stochastic mutual information. When both baths are memoryless, $C=1$ and the standard Jarzynski–Wójcik relation is recovered. For qubit auxiliaries, increasing the intra-bath coupling—and with it the non-Markovianity—decreases the probability ratio, making heat transfer against the temperature gradient relatively more probable. This is a microscopic link between environmental memory and the statistics of heat fluctuation.","feed_headline":"Bath memory boosts odds of heat flowing against the gradient","feed_subtitle":"The standard quantum exchange fluctuation relation gains a correction factor that vanishes for memoryless baths.","key_machinery":"The load-bearing object is the steady state $\\sigma^{\\rm ss}_{AB}$ of the colliding auxiliary pair, combined with the two-point measurement (TPM) protocol applied at each collision step. Memory is generated by intra-bath unitaries $U_{AA}$ and $U_{BB}$ that couple the $(n-1)$-th and $n$-th auxiliaries of each bath before the inter-bath unitary $U_{AB}$; the steady state is the fixed point of the resulting completely positive trace-preserving map. Writing the forward and backward trajectory probabilities in terms of $\\sigma^{\\rm ss}_{AB}$ and thermal Gibbs factors produces $e^{\\Delta\\beta Q}C$, where $C$ is a conditional average over trajectories at fixed heat $Q$. The stochastic free energy $f_k = E_k - \\beta^{-1}(-\\ln q_k)$ captures the athermality of the measured auxiliary states, and the stochastic mutual information $I_{xy}=\\ln(q_{xy}/q_x q_y)$ captures their correlations; the Markovian limit is the statement that both vanish.","core_discovery":"The central claim is that bath memory modifies the exchange fluctuation theorem through a trajectory-dependent factor rather than destroying the relation. For direct bath–bath interaction, the forward/backward heat-probability ratio is $P^{\\rm dir}(Q)/P^{\\rm dir}_B(-Q) = e^{\\Delta\\beta Q} C^{\\rm dir}_{J_A-J_B}$, where the non-Markovian–non-Markovian correction is $C^{\\rm dir}_{\\rm NM-NM} = [\\langle e^{\\beta_A \\Delta f^q_A} e^{\\beta_B \\Delta f^q_B} e^{\\Delta I^q_{AB}}\\rangle_Q]^{-1}$; with only one non-Markovian bath the mutual-information term drops out, and with two Markovian baths $C^{\\rm dir}_{\\rm M-M}=1$. The probe-mediated version has the same structure but keeps a stochastic free-energy term for the probe that survives even when both baths are Markovian. The correction is therefore a conditional average, at fixed exchanged heat $Q$, of exponentials of stochastic free-energy differences and of the stochastic mutual information accumulated by the auxiliary pair. In the qubit example, the ratio decreases as intra-bath coupling grows, so reverse heat flow becomes relatively more likely.","pith_inferences":["The measured ratio $P(Q)/P_B(-Q)$ could be used as an empirical probe of bath memory: a deviation from $e^{\\Delta\\beta Q}$ constrains the conditional average of stochastic free-energy and mutual-information changes, giving an information-theoretic readout of the auxiliary pair's state.","One can conjecture that the modified bound $\\Delta\\beta\\langle Q\\rangle + \\langle\\ln C\\rangle \\ge 0$ extends beyond collisional models to any heat exchange between correlated reservoirs; if true, it would connect environment memory to entropy-production statistics in a model-independent way.","The qubit numerics explore only a parameter regime where $C<1$; mapping when $C>1$ across coupling strengths and temperatures would reveal whether bath memory can also suppress reverse heat flow, a question the paper leaves open.","Because $C$ depends on the steady state of the colliding pair, tuning the intra-bath coupling could act as a control knob on the effective thermodynamic force between baths, with possible use in heat diodes or memory-assisted thermal machines."],"forward_implications":["For memoryless baths the generalized relation collapses to the standard Jarzynski–Wójcik exchange fluctuation theorem, so the standard result is exactly the $C=1$ special case.","The integral form yields the modified second-law bound $\\Delta\\beta\\langle Q\\rangle + \\langle\\ln C\\rangle \\ge 0$, shifting the usual Clausius inequality by the average logarithm of the correction factor.","In the qubit example, increasing the intra-bath coupling lowers $P(Q)/P_B(-Q)$, so bath memory makes heat transfer against the temperature gradient relatively more probable.","The probe-mediated protocol keeps a nontrivial correction even when both baths are Markovian, because the probe's non-equilibrium steady state contributes a stochastic free-energy term.","The derivation is independent of the specific interaction Hamiltonians, relying only on global energy conservation and the collisional structure, so it applies to both direct and probe-mediated scenarios."],"supporting_citations":[{"why":"Supplies the standard exchange fluctuation theorem $p(q)/p(-q)=e^{\\Delta\\beta q}$ that the paper generalizes.","marker":"[23]"},{"why":"Provides the quantum collisional model framework of repeated system-auxiliary interactions used throughout.","marker":"[59]"},{"why":"Shows how interactions between consecutive bath auxiliaries generate non-Markovian dynamics, the memory mechanism studied here.","marker":"[61]"},{"why":"Gives the reverse-trajectory construction for collisional charge exchange that the paper adapts to heat exchange with memory.","marker":"[54]"},{"why":"Derives joint fluctuation theorems for sequential heat exchange in Markovian collisional settings, the baseline the non-Markovian case extends.","marker":"[78]"},{"why":"Defines the Breuer–Laine–Piilo non-Markovianity measure used to quantify bath memory in the qubit examples.","marker":"[73]"}],"fun_headline_variants":["Bath memory tilts heat flow odds against the gradient","Non-Markovian bath memory favors reverse heat exchange","Memory in baths flips heat statistics toward the anti-gradient","Quantum exchange fluctuation theorem gets memory boost","Memory makes heat flow uphill more likely"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The argument assumes that when one or both baths are made memoryless, the previously collided auxiliary pair is thermal and uncorrelated, so its stochastic free-energy and mutual-information terms drop out of the correction factor; the model's steady state for that pair does not generally have thermal marginals, so the assumption is doing load-bearing work.","fun_headline_variants_meta":{"raw":{"variants":["Bath memory tilts heat flow odds against the gradient","Non-Markovian bath memory favors reverse heat exchange","Memory in baths flips heat statistics toward the anti-gradient","Quantum exchange fluctuation theorem gets memory boost","Memory makes heat flow uphill more likely"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000226,"raw_usage":{"total_tokens":1465,"prompt_tokens":942,"completion_tokens":523,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":558,"completion_tokens_details":{"reasoning_tokens":450}},"tokens_in":558,"tokens_out":523,"duration_ms":5074,"temperature":1.0,"reasoning_tokens":450,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T14:13:28.819168+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Evaluate the left- and right-hand sides of Eq. (14) for two qubit baths with $U_{AA}=U_{BB}=I$, using the actual steady-state joint distribution $q^{AB}_{i\\alpha}=\\langle i\\alpha|\\sigma^{\\rm ss}_{AB}|i\\alpha\\rangle$ in the two-point-measurement probabilities without imposing thermal marginals; if $P(Q)/P_B(-Q)$ differs from $e^{\\Delta\\beta Q}$, the Markovian-limit claim $C^{\\rm dir}_{\\rm M-M}=1$ fails. A direct check is to compute the marginal $\\mathrm{Tr}_B[U_{AB}(\\gamma_A\\otimes\\gamma_B)U_{AB}^\\dagger]$: if it is not the Gibbs state at $\\beta_A$, the free-energy term for a memoryless bath $A$ is not zero.","supporting_citations":[{"cited_title":"Jarzynski and D","cited_arxiv_id":null,"evidence_quote":"Supplies the standard exchange fluctuation theorem $p(q)/p(-q)=e^{\\Delta\\beta q}$ that the paper generalizes."},{"cited_title":"Ciccarello, G","cited_arxiv_id":null,"evidence_quote":"Shows how interactions between consecutive bath auxiliaries generate non-Markovian dynamics, the memory mechanism studied here."},{"cited_title":"Scandi and G","cited_arxiv_id":null,"evidence_quote":"Gives the reverse-trajectory construction for collisional charge exchange that the paper adapts to heat exchange with memory."},{"cited_title":"Santos, A","cited_arxiv_id":null,"evidence_quote":"Derives joint fluctuation theorems for sequential heat exchange in Markovian collisional settings, the baseline the non-Markovian case extends."}],"review_version":1}