{"id":"9115ff6e-5820-4fae-8b65-1a40d4b888b8","arxiv_id":"2608.04564","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"A D-Wave annealer is characterized as a programmable thermal machine whose operating regime can be classified from energy-change statistics and a fitted effective temperature.","lead":"Researchers ran reverse annealing cycles on D-Wave quantum processors and used measured energy changes plus a fitted effective temperature to place lower bounds on heat, work, and entropy production. This adds an energy-efficiency view to quantum annealing, labeling runs as heater, accelerator, refrigerator, or engine-compatible.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The pseudo-likelihood β2 used in Eqs. (7)–(10) is an output effective temperature, not the initial bath temperature; if that identification fails, every regime label and TUR bound loses thermodynamic meaning.","rationale":"The paper's central claim is that the same annealer can realize heater, accelerator, refrigerator, and engine-compatible regimes, diagnosed from the sign structure of cycle-averaged energy exchanges. That diagnosis depends entirely on the bounds in Eqs. (8)–(10), and every one of those bounds contains β2 as the inverse temperature of the environment in the exchange fluctuation theorem. The only estimate of β2 in the paper comes from pseudo-likelihood fitting of the output bitstrings. The manuscript itself repeatedly warns that this is an effective, schedule- and instance-dependent temperature, not a direct measurement of the cryogenic bath. For a single physical environment, the true β2 in Eq. (7) is fixed, but the reported T2 changes with sp, τ, and problem structure, which is strong internal evidence that the fitted parameter is not the Eq. (7) β2. If the identification is wrong, the sign classifications built on the bounds are not thermodynamic statements, and the abstract's central assertion that the device 'can realize multiple thermodynamic behaviours allowed by the TUR bounds' is unsupported. This is a genuine correctness risk, not a disagreement with consensus: the internal consistency of the method is at stake. The concern is addressable, however, by an independent calibration or a synthetic validation with known bath temperature, so the appropriate outcome is conditional acceptance rather than rejection. The reader identified the same weakest assumption, and this stress-test agrees with that assessment.","tokens_in":22286,"tokens_out":9717,"duration_ms":114096,"concrete_test":"Simulate the full protocol with a known single bath at T_true=15 mK (classical Glauber or Lindblad dynamics on a 1D chain with the same schedule and β1 grid), generate 10^4 output bitstrings per (β1, sp), fit β2 by pseudo-likelihood exactly as in Eqs. (11)–(12), and test whether (i) T2 equals T_true to within sampling error for all sp and (ii) the bounds (8)–(10) computed with the fitted β2 actually bracket the simulated ⟨Σ⟩, ⟨Q⟩, and ⟨W⟩. If either fails, the fitted β2 cannot serve as the β2 of Eq. (7) and the thermodynamic regime labels are not certified; the paper would then need to use an independently calibrated fixed bath temperature and propagate its uncertainty.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"Equation (7) is an exchange fluctuation theorem for an initial factorized state in which the environment is in a Gibbs state at inverse temperature β2. The paper never measures the environment; instead β2 is the maximizer of the pseudo-likelihood (Eqs. 11–12) applied to the final output bitstrings. The authors state, in Sec. 2 and Appendix B, that this T2 is an effective temperature of the sampled Ising degrees of freedom, dependent on schedule, freeze-out, calibration, and residual non-equilibrium dynamics, and it varies with sp and with problem instance. A bath temperature, in contrast, is a fixed property of the cryogenic environment. Using the same symbol β2 in Eqs. (8)–(10) silently replaces the initial bath inverse temperature of Eq. (6) with an output-fit parameter. If the output distribution is not Gibbs, or if the fitted parameter differs from the true bath inverse temperature, the claimed lower bounds on Σ, Q, W and every heater/accelerator/refrigerator/engine label in Figs. 3–6 are not consequences of Eq. (7). Appendix A explicitly concedes that the initial ensembles are only checked for low-order stationarity, and the Discussion concedes that a fitted temperature alone does not establish a Gibbs output; these concessions mark the exact spot where the load-bearing assumption sits.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper reports reverse-annealing experiments on D-Wave Advantage and Advantage2 processors, treating each run as a closed thermodynamic cycle. From the measured distribution of the processor energy change ΔE1 and an effective inverse temperature β2 obtained by pseudo-likelihood fitting of the output bitstrings, the authors use an exchange fluctuation theorem and thermodynamic uncertainty relations to derive lower bounds on entropy production, heat, work, and power, and to classify operating regimes as heater, accelerator, refrigerator, or engine. The regime maps are presented as functions of the initial inverse temperature β1 and the reverse-annealing turning point sp, for one-dimensional chains and two-dimensional Pegasus instances, with a companion finite-size scaling analysis. The central claim is that the quantum annealer is a programmable thermal machine whose thermodynamic role can be selected by initialization and schedule and diagnosed from energy statistics alone.","tokens_in":22563,"tokens_out":5800,"duration_ms":65441,"significance":"If the underlying identification of β2 with the environment temperature in the fluctuation theorem were validated, the paper would provide a genuinely useful, device-agnostic methodology for benchmarking quantum annealers on energetic grounds, complementary to solution quality and runtime. The scale of the experiments (multiple D-Wave generations, thousands of qubits, >10^4 cycles per setting), the explicit statement of assumptions in Methods 4.8, the honest use of the label 'engine/accelerator-compatible' in the two-dimensional analysis, and the public data/code repositories are all strengths. However, the central quantitative claims and the four-mode classification rest on an unvalidated identification of a fitted output parameter with the bath temperature of the exchange fluctuation theorem; this makes the significance conditional rather than established.","major_comments":[{"comment":"The β2 appearing in the exchange fluctuation theorem, Eq. (7), is the inverse temperature of the environment in the initial factorized state (6). In the applications, Eqs. (8)–(10), the same symbol β2 is replaced by the maximizer of the pseudo-likelihood (11)–(12) fitted to the final output configurations. This is not a harmless notational choice: the paper itself states in Section 2 and the Discussion that T2 is an effective temperature of the sampled Ising degrees of freedom that depends on freeze-out, calibration, schedule, and residual non-equilibrium dynamics, whereas a bath temperature is a property of the cryogenic environment. Unless the output distribution is exactly Gibbs at the fitted β2 and the environment acts as a single reservoir at that same temperature, the bounds (8)–(10) and every regime label derived from them are not consequences of Eq. (7). The concessions in Appendix A ('these diagnostics do not prove that the full high dimensional distribution is exactly Gibbsian') and the Discussion ('a fitted temperature alone does not establish that the complete output distribution is thermal') mark exactly the unvalidated premise. This issue is load-bearing for all quantitative claims in the paper and needs either an independent validation of β2 (for example, comparison with the vendor's freeze-out effective temperature or with direct cryogenic thermometry) or a systematic reframing of the results as conditional effective-temperature phenomenology rather than thermodynamic bounds.","section":"Methods 4.5–4.6 (Eqs. 6–12)"},{"comment":"Equation (9) is a lower bound on −⟨Q⟩, equivalently an upper bound on ⟨Q⟩. It can certify that Q<0 (the environment gains energy), but it cannot certify Q>0 (the environment loses energy). The refrigerator and engine assignments in Table 2 require, for some of the temperature orderings, a definite sign of ⟨ΔE2⟩ or ⟨W⟩ that is not accessible from Eq. (9) or Eq. (10): Eq. (10) is a lower bound on ⟨W⟩ and cannot certify W<0, which the engine class requires. The body text is careful in the two-dimensional sections to use the combined label 'engine/accelerator-compatible,' but the abstract and Introduction state that refrigerator- and engine-compatible regimes are realized and mapped on the same device (Abstract; Section 1: 'we implement protocols that realize all four behaviours on the same device'). Given the one-sided nature of the available bounds, the data can certify heater and accelerator operation directly, but refrigerator and engine operation are at best compatible with the bounds, not certified. The abstract and the summary sentences in the Discussion should be revised to state precisely which labels are certified by measurements and which are only allowed by the bounds.","section":"Methods 4.6, Eq. (9), Table 2, and Figs. 4–6"},{"comment":"The claim that the per-spin work and power bounds grow logarithmically with system size, and the total bounds as L log L, is based on Fig. 2, which shows four chain lengths without error bars, fit lines, or goodness-of-fit statistics. The text says the data are 'consistent with a logarithmic finite size enhancement,' which is a much weaker statement than the subsequent 'the empirical LlogL law is the expected finite size signature.' If this scaling is presented as a quantitative result, the authors should provide a fit with uncertainties or explicitly label it as a qualitative trend. This is not the central claim of the paper, but as written the section overstates the evidential support.","section":"Section 2, 'Scaling with system size', Eq. (1) and Fig. 2"}],"minor_comments":[{"comment":"The caption reads 'The experiment was performed the D-Wave Advantage6.4 system'; 'performed' should be 'performed on'.","section":"Fig. 3 caption"},{"comment":"The expression for Ep(sp,σ) is hard to read; the factor 'hP 109 2 BGHz(sp)' appears to mean hP × 10^9 × (1/2) × BGHz(sp) but the typesetting is ambiguous. Please rewrite with explicit powers of ten.","section":"Methods 4.6, Eq. (13)"},{"comment":"References [8] and [21] are the same paper, and references [22] and [26] are the same arXiv preprint; please merge or distinguish them.","section":"References"},{"comment":"The phrase 'device-certified, bath model-free lower bounds' is misleading: the bounds are model-free only with respect to the microscopic bath dynamics, but they still require the single-reservoir assumption and the identification of β2 with the bath temperature. Please rephrase to avoid this overstatement.","section":"Methods 4.6"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is potentially interesting but currently sits between a thermodynamic characterization and an operational effective-temperature classification. The decisive issue is the identification of the fitted β2 with the bath temperature in the exchange fluctuation theorem; the authors themselves provide language that undermines this identification. I would ask for either an independent validation of β2 or a substantial softening of the thermodynamic claims. The public data and code are a positive feature, as is the honest 'engine/accelerator-compatible' label in the two-dimensional sections. The abstract and Introduction need to be reconciled with the one-sided certifiability of the bounds."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Let me give you my take on this one. The headline: it's a useful experimental extension of the Buffoni-Campisi accelerator framework, but the central assumption—that the effective temperature fitted to the output distribution is the bath temperature in the fluctuation theorem—is not justified, and that problem propagates into the refrigerator and engine labels.\n\nWhat's genuinely new: they map out heater, accelerator, and (claimed) refrigerator regimes on real D-Wave hardware by scanning preparation temperature and turning point, on both 1D chains and 2D instances, and they repeat it on Advantage and Advantage2. They also report TUR-based power bounds and a finite-size scaling study. The experimental work looks solid: 10^4 cycles per point, open code and data, and convergence checks for the initial Gibbs sampler. The body is often appropriately cautious—they use the label 'engine/accelerator-compatible' in the phase diagrams, which is a good sign.\n\nThe soft spots are real and load-bearing. Equation (7) is an exchange fluctuation theorem for an initial state where the environment is at inverse temperature β2. But β2 is estimated from the final output bitstrings via pseudo-likelihood. The authors themselves stress that this is an effective temperature of the sampled degrees of freedom, dependent on schedule, freeze-out, calibration, and non-equilibrium dynamics—not a measurement of the bath. Replace the bath temperature with an output-fit parameter and the bounds in Eqs (8)–(10) are no longer consequences of the fluctuation theorem. This is conceded, indirectly, in Appendix A (only low-order stationarity is checked) and in the Discussion (a fitted temperature doesn't prove the output is Gibbs). The stress-test note is right to flag this as load-bearing.\n\nSecond, the refrigerator label can't be certified from a lower bound on -⟨Q⟩ (Eq 9). That bound can rule out the environment losing energy in some cases, but you'd need an upper bound to certify ΔE2<0. The abstract says 'refrigerator- and engine-compatible regimes' as if they were demonstrated; the body sometimes hedges, but not systematically.\n\nThese aren't fatal to the paper's utility as a diagnostic tool—the heater/accelerator distinction is on firmer ground, and the energy-accounting view is useful. But the quantitative claims about refrigerator operation need either new data or a careful reframing as 'consistent with refrigerator operation if one assumes the output temperature is the bath temperature.' The absence of error bars on β2 and on the phase boundaries is a minor additional issue but worth fixing.\n\nWho's it for? Anyone benchmarking annealers energetically, or testing fluctuation relations in many-spin systems. It deserves a serious referee and ultimately publication if the identification issue is addressed or the claims are scaled back. Send it to peer review, but expect the referees to push hard on the β2 point.","headline":"A useful experimental extension of the D-Wave thermal machine framework, undercut by identifying a fitted effective temperature with the bath temperature in the fluctuation theorem; refrigerator/engine labels are not certified by the one-sided bounds.","tokens_in":23092,"tokens_out":5017,"would_cite":true,"duration_ms":53598,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A commercial quantum annealer can be described as a programmable thermal machine whose operating mode is selected by the initial thermal ensemble and the reverse-annealing schedule.","keywords":["quantum annealing","D-Wave","thermodynamic uncertainty relations","reverse annealing","thermal machine","entropy production","effective temperature","Ising model"],"falsifier":"Run the same reverse-annealing cycles while independently measuring the heat absorbed by the cryogenic environment (e.g., with calorimetry or power metering at the mixing chamber) and compare with the paper's lower bound on $-\\langle Q\\rangle$: if the measured heat falls below the bound, the identification of the fitted $\\beta_2$ with the fluctuation-theorem temperature fails. A simpler statistical check is to hold out part of the output bitstrings and test whether they are consistent with the Gibbs distribution at the fitted $\\beta_2$; a clear failure would mean the sign-based regime labels are not thermodynamically certified.","tokens_in":22121,"feed_emoji":"⚙️","tokens_out":10981,"duration_ms":136547,"temperature":0.7,"pith_summary":"This paper sets out to show that a commercial quantum annealer is not just an optimizer or sampler but a programmable thermal machine: by choosing the initial inverse temperature $\\beta_1$ and the reverse-annealing turning point $s_p$, the same chip can be made to operate as a heater, an accelerator, a refrigerator, or an engine-compatible device. The analysis uses only the measured change in the programmed Ising energy over a closed cycle, together with an effective environment temperature $\\beta_2$ fitted to the output bitstrings, to place lower bounds on entropy production, heat, work, and power via thermodynamic uncertainty relations. The interest is that these bounds turn the limited energy readout already available on the hardware into an energy-aware characterization that is independent of solution quality and runtime. If the claim holds, thermodynamic mode becomes a practical diagnostic for comparing schedules and hardware, and a bridge between quantum optimization and energy efficiency.","feed_headline":"One D-Wave chip can act as heater, fridge, or engine","feed_subtitle":"Initial state and reverse-annealing schedule set the thermodynamic mode, read off from energy changes alone.","key_machinery":"The argument is carried by the exchange fluctuation theorem, $p(\\Delta E_1,\\Delta E_2)/p(-\\Delta E_1,-\\Delta E_2)=\\exp(\\beta_1\\Delta E_1+\\beta_2\\Delta E_2)$, combined with thermodynamic uncertainty relations that convert the first two moments of the measured processor energy change $\\Delta E_1$ into lower bounds on entropy production, heat, work, and power. The environment inverse temperature $\\beta_2$ is obtained by pseudo-likelihood fitting of the output bitstrings to the programmed Ising Hamiltonian, and the sign structure of the cycle-averaged energy exchanges then assigns the thermodynamic regime. The reverse-annealing schedule keeps the initial and final Hamiltonians identical, which makes the two-point energy change $\\Delta E_1$ directly measurable and closes the thermodynamic cycle.","core_discovery":"The central discovery claimed is that reverse annealing on D-Wave hardware realizes a closed thermodynamic cycle whose operating regime is programmable. With a fixed problem Hamiltonian, changing only the preparation temperature and the turning point moves the device through heater-, accelerator-, refrigerator-, and engine/accelerator-compatible regimes, and these regimes can be read off from the sign structure of the cycle-averaged energy exchanges without a detailed microscopic bath model. The claim is supported by phase diagrams in the $(\\beta_1, s_p)$ plane for one-dimensional chains and two-dimensional Pegasus instances, on two hardware generations, and by the observed scaling of work and power bounds with chain length. The authors further claim that the fitted effective temperature $T_2$ acts as an operational thermometer for the sampled degrees of freedom, and that the thermodynamic mode adds information absent from success probability or runtime alone: it distinguishes driven refinement, net heating, and heat pumping while quantifying their energetic consequences.","pith_inferences":["Editorial inference: reporting success probability, target-energy probability, sample diversity, and time to solution conditional on the thermodynamic mode could reveal whether mode labels predict practical performance; the paper explicitly leaves this conditional analysis as future work.","Editorial inference: folding the chip-level bounds together with programming, readout, control, and refrigeration costs would give an energy-to-solution benchmark for comparing algorithms and hardware.","Editorial inference: the sign-structure classification should transfer to other driven open quantum systems with closed cyclic schedules, making the thermodynamic-mode diagnostic a general tool rather than a D-Wave-specific one.","Editorial inference: replacing the pseudo-likelihood estimate with a direct Gibbsness test on the output distribution would strengthen the claim that the fitted $\\beta_2$ is the thermodynamic temperature, and would show which regions of the phase diagram are robust."],"forward_implications":["The same hardware can be moved between heater, accelerator, refrigerator, and engine/accelerator-compatible regimes by changing only the initial inverse temperature $\\beta_1$ and the turning point $s_p$.","The measured mean processor energy change $\\langle\\Delta E_1\\rangle$ tells whether final configurations improve or worsen the programmed objective on average, so thermodynamic data double as objective-improvement diagnostics.","For sampling, the fitted effective temperature $T_2$ gives an operational measure of how strongly probability concentrates on low-energy configurations, and schedule-conditioned shifts in $T_2$ provide a built-in thermometry diagnostic.","In the one-dimensional critical window, the TUR-based work and power bounds grow as $L\\log L$ with chain length, indicating where dissipation concentrates and how it scales.","Refrigerator operation is not automatically good for optimization; it corresponds to computational improvement only at points where $\\langle\\Delta E_1\\rangle < 0$."],"supporting_citations":[{"why":"Supplies the earlier demonstration that reverse annealing realizes an accelerator thermal machine and provides the heat, work, and dissipation bounds this work extends.","marker":"[12]"},{"why":"Provides the improved entropy-production bound for a quantum annealer that the TUR inequalities build on.","marker":"[13]"},{"why":"Prior D-Wave experiment connecting thermodynamic quantities to computational performance; motivates and anchors the protocol.","marker":"[14]"},{"why":"Establishes the one-dimensional transverse-field Ising critical point used to interpret the mid-anneal signatures and logarithmic scaling.","marker":"[46]"},{"why":"Defines the corrupted biased ferromagnet instance used as the rugged two-dimensional test problem.","marker":"[50]"},{"why":"Review of quantum fluctuation relations supplying the exchange fluctuation theorem underlying Eq. (7).","marker":"[57]"},{"why":"Nonequilibrium fluctuation theorems and counting statistics supporting the exchange fluctuation theorem and second-law form.","marker":"[58]"},{"why":"Derives the dissipation bound on current fluctuations used as the thermodynamic uncertainty relation for the work and heat bounds.","marker":"[60]"},{"why":"Establishes the effective-temperature estimation procedure for quantum annealers used in pseudo-likelihood thermometry.","marker":"[63]"},{"why":"Supplies the inverse-Ising inference method behind the pseudo-likelihood fit that yields $\\beta_2$.","marker":"[64]"}],"fun_headline_variants":["Quantum annealers turn into heaters, fridges, or engines","Reverse annealing sets D-Wave as heater, fridge, or engine","Programmable thermodynamics: D-Wave as fridge, heater, or engine","Same D-Wave chip: heater, fridge, or engine on demand"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"All quantitative claims assume that the inverse temperature $\\beta_2$ fitted from the output bitstrings is the same temperature that enters the exchange fluctuation theorem; if the output distribution is not close to a Gibbs state, or if the cryogenic environment is not a single thermal reservoir, the bounds and every thermodynamic regime label lose their meaning.","fun_headline_variants_meta":{"raw":{"variants":["Quantum annealers turn into heaters, fridges, or engines","Reverse annealing sets D-Wave as heater, fridge, or engine","Programmable thermodynamics: D-Wave as fridge, heater, or engine","Same D-Wave chip: heater, fridge, or engine on demand"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000597,"raw_usage":{"total_tokens":2783,"prompt_tokens":928,"completion_tokens":1855,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":544,"completion_tokens_details":{"reasoning_tokens":1779}},"tokens_in":544,"tokens_out":1855,"duration_ms":13621,"temperature":1.0,"reasoning_tokens":1779,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T21:54:46.773185+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the same reverse-annealing cycles while independently measuring the heat absorbed by the cryogenic environment (e.g., with calorimetry or power metering at the mixing chamber) and compare with the paper's lower bound on $-\\langle Q\\rangle$: if the measured heat falls below the bound, the identification of the fitted $\\beta_2$ with the fluctuation-theorem temperature fails. A simpler statistical check is to hold out part of the output bitstrings and test whether they are consistent with the Gibbs distribution at the fitted $\\beta_2$; a clear failure would mean the sign-based regime labels are not thermodynamically certified.","supporting_citations":[{"cited_title":"& Campisi, M","cited_arxiv_id":null,"evidence_quote":"Supplies the earlier demonstration that reverse annealing realizes an accelerator thermal machine and provides the heat, work, and dissipation bounds this work extends."},{"cited_title":"& Buffoni, L","cited_arxiv_id":null,"evidence_quote":"Provides the improved entropy-production bound for a quantum annealer that the TUR inequalities build on."},{"cited_title":"& Gardas, B","cited_arxiv_id":null,"evidence_quote":"Prior D-Wave experiment connecting thermodynamic quantities to computational performance; motivates and anchors the protocol."},{"cited_title":"The Potential of Quantum Annealing for Rapid Solution Structure Identification","cited_arxiv_id":"1912.01759","evidence_quote":"Defines the corrupted biased ferromagnet instance used as the rugged two-dimensional test problem."},{"cited_title":"& Talkner, P","cited_arxiv_id":null,"evidence_quote":"Review of quantum fluctuation relations supplying the exchange fluctuation theorem underlying Eq. (7)."},{"cited_title":"R., Horowitz, J","cited_arxiv_id":null,"evidence_quote":"Derives the dissipation bound on current fluctuations used as the thermodynamic uncertainty relation for the work and heat bounds."}],"review_version":1}