{"id":"e4708386-5618-47a9-afd1-5ea8ecef788e","arxiv_id":"2602.18321","paper_version":2,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"For Markov networks with fixed transition timescales, conservative driving is proven to dissipate at most twice the minimal possible amount, and a ring-with-barrier model shows nonconservative driving can beat it by about a third.","lead":"Moving a discrete network between two states with minimal energy loss generally needs nonconservative, cycle-driving forces, but this paper proves that a conservative protocol always loses at most a factor of two in dissipation versus the true optimum. It also builds a ring-with-barrier model where nonconservative driving beats conservative driving by roughly 30%, suggesting such gains are generic when constraints are tight.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Missing convexity proof for ρ is the load-bearing gap behind Eq. (9); the abstract additionally overclaims a 4/3 bound absent from the body.","rationale":"The reader's weakest-assumption identification is correct: the proof of Eq. (9) depends on the unproved convexity of ρ, and without it the stationarity condition Eq. (13) is not sufficient to identify a global minimizer. However, I verified (and the paper could easily show) that convexity holds: the per-edge cost in Eq. (8) has second derivative positive, and the edge currents are affine functions of the cycle-current variables, so ρ is convex. Thus the concern is a rigor gap rather than a mathematical flaw in the factor-two theorem. The more serious public-facing issue is the mismatch between the abstract's 4/3 claim and the body's actual factor-two theorem; the body itself declares stronger bounds open. Since the reader's CONDITIONAL verdict already accounts for these issues, no verdict change is needed, but the paper must supply the convexity proof and reconcile the abstract with the proven result before acceptance.","tokens_in":11267,"tokens_out":16286,"duration_ms":139447,"concrete_test":"Verify ρ-convexity by computing the Hessian of ρ with respect to cycle currents on a two-cycle network (e.g., Fig. 3a). If it is positive semidefinite for all currents, Eq. (13) gives a global minimum and the factor-two proof is complete. To settle the abstract mismatch, perform a full-text search for '4/3' and 'saturat'; if no derivation appears, the abstract must be corrected to the factor-two statement proven in the body.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Equation (9) rests on Appendix B's chain σ* ≤ σ_c ≤ 2ρ* ≤ 2σ*, which requires x_c, the minimizer of ρ, to be a global minimum of ρ. The paper states 'Like σ, ρ is a convex measure' (Main result, after Eq. (8)) but never proves convexity; the derivation in 'Conservative protocols minimize ρ' only computes the stationary condition ∂_{j_C}ρ = Σ F_ij (Eq. (13)). If ρ were not convex, this condition could be a saddle, and the identification ρ* = ρ(x_c) would fail. In fact the convexity is true: for each edge the per-current cost has a positive second derivative, and cycle currents enter affinely, so ρ is convex in cycle currents. But the paper omits the argument, leaving the central theorem not fully self-contained. Separately, the abstract supplied with the paper promises a 4/3 bound, a saturating example, and a weaker bound for general load-sharing parametrizations; the body proves only factor two, and the Discussion explicitly says a tighter bound is an open problem. The advertised central claim is therefore unsupported.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies finite-state Markov jump processes with fixed symmetric transition rates κ_ij(t) and optimizes the entropy production over the antisymmetric forces A_ij(t). Its main theorem (Eq. (9)) claims that for any prescribed time evolution, the best conservative protocol (zero cycle affinities) has an entropy production rate σ_cons satisfying σ* ≤ σ_cons ≤ 2σ*, where σ* is the true minimum, and hence total dissipation is within a factor of two. The proof introduces an auxiliary cost ρ (Eq. (8)) with σ/2 ≤ ρ ≤ σ; the ρ-minimizer is shown (Appendix C) to satisfy the conservative condition Σ F_ij = 0, and Appendix B converts the pointwise cost bound into the main inequality. A unicyclic example with an energy barrier is analyzed numerically, giving σ*/σ_cons ≈ 0.76. The supplied abstract promises a stronger 4/3 bound and saturation, which the body does not prove.","tokens_in":11440,"tokens_out":21067,"duration_ms":180270,"significance":"If the factor-two theorem is correct, it is a clean, topology-independent statement: giving up nonconservative driving costs at most a factor of two in dissipation, regardless of network size or distance from equilibrium. This usefully complements Ref. [28] and explains why conservative protocols are near-optimal. The proof strategy, based on the auxiliary cost ρ and the sandwich in Appendix B, is elegant and machine-checkable in principle. The paper also includes a concrete model illustrating an O(1) improvement. However, as submitted, the advertised stronger claims in the abstract are unsupported, and a key convexity step in the proof is omitted.","major_comments":[{"comment":"The supplied abstract claims a stronger result than the body proves: it states that a conservative protocol exists whose dissipation is at most 4/3 times the optimum, with a saturating energy-barrier example, and that a weaker numerical factor is proved for general load-sharing parametrizations. The main text proves only σ_cons ≤ 2σ* (Eq. (9)); the full-text abstract itself says 'at most twice', and the Discussion explicitly leaves a tighter bound as an open problem. No 4/3 theorem, saturating example, or load-sharing bound appears in the body. The advertised central claim is therefore unsupported and the abstract must be corrected (or the missing results proved).","section":"Abstract; Main result (Eq. (9))"},{"comment":"The sandwich in Appendix B, σ* ≤ σ_c ≤ 2ρ* ≤ 2σ* (Eq. (18)), requires x_c to be a global minimizer of ρ over the allowed cycle currents. The text asserts 'Like σ, ρ is a convex measure' but gives no proof, and Appendix C only derives the stationary condition ∂_{j_C}ρ = Σ_{(ij)∈C} F_ij (Eq. (13)). Without convexity of ρ in the cycle currents, stationarity need not identify the global minimum, so the identification ρ* = ρ(x_c) that feeds the theorem could fail. The convexity is elementary (the per-edge cost has positive second derivative in the edge current, and cycle currents enter affinely), but it must be stated before Eq. (9) is used.","section":"Main result, after Eq. (8); Appendix C"},{"comment":"The abstract additionally promises a 'weaker numerical factor for a more general class of rate parametrizations with different load-sharing factors'. The body contains no such theorem: the Discussion merely says the parametrization 'can, e.g., be generalized' and does not state a bound. Either prove the load-sharing generalization or remove the claim from the abstract.","section":"Discussion and outlook"}],"minor_comments":[{"comment":"The full-text abstract (factor two) differs from the metadata abstract (4/3); synchronize the versions after the content is corrected.","section":"Abstract"},{"comment":"The inequality C(x) ≥ (x/2)sinh(x/2) used for σ/2 ≤ ρ is stated without proof; a one-line verification would help the reader.","section":"Main result"},{"comment":"'parametrization adapted in this work' should be 'adopted'.","section":"Discussion and outlook"},{"comment":"The 'optimized J' is not defined in the caption; specify the optimization criterion in the caption or main text.","section":"Fig. 2(b)"},{"comment":"σ_c is not explicitly identified with σ_cons of the main text; define x_c as the conservative configuration.","section":"Appendix B"},{"comment":"Eq. (22): the bound |A*_C|/2 ≤ N follows because the sum has N terms each with |tanh| ≤ 1; the intermediate step is worth spelling out.","section":"Appendix D"}],"recommendation":"major_revision","confidential_remarks":"The mathematical core of the body (factor-two bound) appears sound and the ρ-sandwich is a nice technique. The submission as provided, however, has a serious mismatch between the supplied abstract (4/3 bound, saturating example, load-sharing generalization) and the full text (factor two, no saturation, open tighter bound). This is likely a versioning error, but as it stands the advertised claims are unsupported. I recommend major revision: add the missing convexity proof, correct the abstract, and ensure the metadata abstract matches the full-text abstract."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things up front. This paper proves a genuine factor-two bound: for a Markov jump process with fixed symmetric transition rates, the best conservative protocol dissipates at most twice the true optimum, independent of network size or topology. The proof mechanism—an auxiliary cost ρ with σ/2 ≤ ρ ≤ σ whose minimizer is conservative—is elegant and the core chain in Appendices B and C checks out. Second, the version we saw includes an abstract claiming a sharper 4/3 bound and a saturating example, but the body only proves factor two and explicitly leaves tighter constants as an open problem. That mismatch must be fixed before publication, whichever direction it resolves.\n\nWhat's new: the sandwich bound is the first universal constant-factor handle on the gap between conservative and nonconservative driving in this setting. The barrier example is genuinely illustrative, showing an order-one improvement (~32%) and explaining how optimal driving balances current across the barrier with bulk transport while conservative driving avoids the barrier. The derivation of ∂_{j_C} ρ = Σ F_ij is correct.\n\nSoft spots, in proportion. First, the abstract issue above: if that 4/3 sentence is on the arXiv page, it's a real overclaim; if it's an artifact of our copy, the submitted abstract still needs to match the proof. Second, the load-bearing assertion that ρ is convex is not proven. The sentence 'Like σ, ρ is a convex measure' is doing real work: the critical point from Appendix C must be a global minimum for the main inequality. Convexity is elementary (C''(x)>0), but the paper doesn't show it, and deferring to an upcoming publication is not appropriate for a load-bearing step. Third, the numeric value of the ~32% gap comes from a calculation the authors themselves say becomes increasingly unstable at large N; the qualitative point stands, but the quantitative claim should be flagged.\n\nNothing here is a fatal flaw. The result is real, the proof is mostly self-contained, and the credit to Remlein-Seifert is accurate. Fix the abstract, add the one-line convexity argument, and this is a solid letter. It deserves a serious referee; a good referee will ask exactly for these revisions.","headline":"A clean factor-two bound on conservative vs. optimal driving, worth serious refereeing, but the abstract overclaims a 4/3 bound and the convexity proof of ρ is missing.","tokens_in":12141,"tokens_out":5156,"would_cite":true,"duration_ms":66874,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["05.70.Ln","05.40.-a"],"model":"deepseek-v4-flash","headline":"For a Markov jump process (a system hopping between discrete states), driving with conservative forces alone costs at most twice the minimal entropy production — on any network, at any distance from equilibrium.","keywords":["stochastic thermodynamics","entropy production","optimal transport","Markov jump process","nonconservative driving","cycle currents","conservative forcing","near-optimality"],"falsifier":"Fix any network with at least one cycle, fix the symmetric rate factors κ_ij, and prescribe the target probability current; numerically minimize entropy production over all forces and over conservative forces only. Any instance with σ_cons > 2σ* (σ*/σ_cons < 1/2) refutes the theorem; none should exist if the proof holds. For the abstract's sharper claim, a single instance with σ*/σ_cons < 3/4 — the ring example's 0.76 does not reach it — would refute the factor-4/3 statement without touching the factor-two result.","tokens_in":10947,"feed_emoji":"⚙️","tokens_out":25715,"duration_ms":224498,"temperature":0.7,"pith_summary":"This paper asks how much dissipation is lost when a controller of a Markov jump process is restricted to conservative (potential-derived) forces rather than the full set of allowed forces, given that each transition's timescale is fixed and cannot be adjusted. Earlier work showed that in this setting the truly optimal protocol generally requires nonconservative cycle forces, so the question is whether the much simpler conservative protocols remain competitive. The paper proves they do: at every instant, on any network topology, the best conservative protocol's entropy production rate is at most twice the true minimum, and the same factor-two bound holds for the total dissipation of a full protocol. It then constructs a ring network with one large energy barrier where the optimal protocol beats the conservative one by about 32%, and argues that the nonconservative advantage grows when constraints leave fewer adjustable degrees of freedom. The engine is a second cost function that brackets the entropy production rate from below and above and, unlike entropy production itself, is minimized exactly by conservative forces.","feed_headline":"Give up nonconservative forces, pay at most twice the dissipation","feed_subtitle":"Why it matters: the simpler, cheaper controllers are never far from the theoretical best.","key_machinery":"The load-bearing device is the auxiliary cost function ρ = Σ_{i,j} ω_ij C(F_ij) with C(x) = x sinh(x/2) − 2[cosh(x/2) − 1], where ω_ij = κ_ij √(p_i p_j) is the timescale of the transition between neighboring states and F_ij the thermodynamic force on that edge. It does two things at once: it brackets the entropy production rate pointwise, σ/2 ≤ ρ ≤ σ, and — unlike σ — its minimizer is conservative, since differentiating with respect to a cycle current j_C yields ∂_{j_C} ρ = Σ_{(ij)∈C} F_ij = 0, exactly the condition that the cycle affinity vanish. The proof (Appendix B) is a generic sandwich: given two cost functions on the same domain with σ/2 ≤ ρ ≤ σ, their minima satisfy σ* ≤ σ_c ≤ 2ρ* ≤","core_discovery":"For any Markov network with fixed, non-optimizable transition timescales, the cheapest conservative protocol for the same task costs at most twice the minimal entropy production — σ* ≤ σ_cons ≤ 2σ* and S_cons/2 ≤ S* ≤ S_cons — on any network. The engine is an auxiliary cost ρ (Eq. 8), σ/2 ≤ ρ ≤ σ, that — unlike σ — is minimized by conservative forces, its stationarity condition per cycle current being exactly vanishing cycle affinity (Eq. 13); a sandwich argument transfers the pointwise inequality to the minima. In the worked example, a unicyclic network with one large energy barrier, the optimal protocol balances barrier crossing with a bulk cycle current, a ~32% gain over conservative driv","pith_inferences":["Gap in the proof: the paper asserts that ρ is convex (one sentence before Eq. 8) but never demonstrates it, and the factor-two bound needs the stationary point of Eq. (13) to be the global minimizer of ρ. The gap looks fillable — each edge contribution C(F) has second derivative cosh(F/2)/2 + F·sinh(F/2)/4 > 0, and the currents feasible at fixed ṗ form an affine set — but as written the proof lea","The abstract's stronger claims (factor 4/3, a saturating example, and a proven weaker-factor generalization to other load-sharing parametrizations) do not match the body, which proves the factor of two and lists those generalizations as future work. A numerical search over multi-cycle networks with heterogeneous κ would show whether ratios below the ring example's 0.76 exist and whether 1/2 is eve","The sandwich template is portable: any control problem with two cost functions c ≤ c' ≤ Kc, where the simpler subclass happens to minimize c', inherits a factor-K near-optimality theorem; each new rate parametrization would need its own convexity check.","Read as a design principle, the factor-two ceiling says conservative controllers are the right first attempt: measure the dissipation gap, and invest in implementing cyclic driving only when the gap approaches the ceiling. Conversely, coarse-grained models of network-driven machines that exclude cycle currents could systematically overestimate achievable efficiency by up to a factor of two."],"forward_implications":["For any discrete network whose transition timescales are individually fixed, a controller restricted to conservative driving is guaranteed to stay within a factor of two of minimal dissipation, independent of network size, topology, number of cycles, and distance from equilibrium.","Near equilibrium the optimal and conservative protocols coincide — their difference first appears at order F³ — so the extra power of nonconservative cycle forces only becomes significant far from equilibrium.","In the ring-with-barrier model, optimal nonconservative driving beats the best conservative protocol by roughly 32% for large networks, because it splits probability flow between the direct route across the barrier and the bulk of the ring, while the conservative protocol mostly avoids the barrier.","The large-N limit of that transport problem does not reduce to the continuum Langevin optimal-transport problem where conservative forces are optimal: the optimal cycle affinity grows linearly with N, so the nonconservative advantage survives the thermodynamic limit.","Tighter constraints — fewer optimizable force degrees of freedom — amplify the value of the extra degrees of freedom that nonconservative forces provide, suggesting optimal nonconservative driving is the generic case in strongly constrained real systems."],"fun_headline_variants":["Conservative driving: never more than 33% above optimal dissipation","Nonconservative forces only yield up to 25% savings over conservative","Simpler control is near-optimal: max 4/3 dissipation","Choose conservative driving: within 33% of the best"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The factor-two theorem rests on two premises the text asserts but does not prove — that the auxiliary cost ρ is convex, so the configuration with all cycle affinities zero is its true global minimizer, and that optimizing cycle currents decouples instant by instant — since the sandwich bound fails if either is false, even though the convexity is in fact elementary to verify.","fun_headline_variants_meta":{"raw":{"variants":["Conservative driving: never more than 33% above optimal dissipation","Nonconservative forces only yield up to 25% savings over conservative","Simpler control is near-optimal: max 4/3 dissipation","Choose conservative driving: within 33% of the best"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000857,"raw_usage":{"total_tokens":3554,"prompt_tokens":735,"completion_tokens":2819,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":479,"completion_tokens_details":{"reasoning_tokens":2745}},"tokens_in":479,"tokens_out":2819,"duration_ms":22771,"temperature":1.0,"reasoning_tokens":2745,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-02T21:58:06.582714+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Fix any network with at least one cycle, fix the symmetric rate factors κ_ij, and prescribe the target probability current; numerically minimize entropy production over all forces and over conservative forces only. Any instance with σ_cons > 2σ* (σ*/σ_cons < 1/2) refutes the theorem; none should exist if the proof holds. For the abstract's sharper claim, a single instance with σ*/σ_cons < 3/4 — the ring example's 0.76 does not reach it — would refute the factor-4/3 statement without touching the factor-two result.","supporting_citations":[],"review_version":1}