{"id":"58a36782-ee6a-4831-a806-95e5f062fa78","arxiv_id":"2509.06957","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"The author contends that Landauer's principle is not fundamental because the thermodynamic cost of erasure depends on the observer's knowledge, and he sketches quasistatic erasure counterexamples.","lead":"This essay argues that thermodynamics is observer-dependent because energy and entropy are tied to what we know, and that Landauer's erasure principle is not a fundamental law of physics. It chronicles the history from Joule to Shannon to Landauer, and offers a concrete objection: erasing an unknown bit can be thermodynamically reversible, with a sketched counterexample for known bits.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Section V.F.2's counterexample is not a closed quasistatic erasure: resetting the tilt to a usable memory either restores the 50/50 equilibrium (no erasure) or requires a non-quasistatic step, so the asserted 'as slow as desired' protocol cannot satisfy the standard Landauer erasure definition.","rationale":"The reader's weakest_assumption is in the right place: Section V.F.2 is the only constructive counterexample that would actually erase a bit, as opposed to V.E's erase-plus-restore cycle. I agree with the conditional verdict, and no independent evidence (experiment, code, or explicit potential) is provided. My stress-test sharpens the worry: the missing protocol is not just an exposition gap. A quasistatic isothermal process with the control parameter returned to its initial value cannot end in a different equilibrium distribution; if the final distribution is the metastable all-0 state, the last step cannot be quasistatic. The paper's own Section V.A defines a data-bit as an object with two distinguishable stationary configurations, so V.F.1's 'same thermodynamic state, two logical states' is in tension with that definition. The historical and philosophical narrative has some value, and the known/unknown distinction has partial support in the literature, but the central law-invalidating claim rests on the unproven counterexample. Therefore the reader's CONDITIONAL is appropriate: acceptance should require a fully specified closed-cycle protocol, verified by simulation or an analytic free-energy calculation. If that test fails, the claim should be rejected. I do not change the verdict because the reader already flagged the same weak point.","tokens_in":27951,"tokens_out":14714,"duration_ms":184755,"concrete_test":"Fix an explicit double-well potential U(x) with a high central barrier and a tilt protocol lambda(t) satisfying lambda(0)=lambda(tau)=0; simulate the overdamped Langevin equation for an ensemble initially at equilibrium in the two wells. For tau approaching infinity, measure the final probability p0 in the 0 well and the net heat Q dissipated over the full cycle. The paper's claim requires p0 approaching 1 with Q approaching 0. If instead p0 approaching 1 implies Q approaching kT ln2, or Q approaching 0 leaves p0 approaching 1/2, the V.F.2 counterexample fails. If one only tests the non-cyclic protocol, the heat cost of resetting lambda to 0 must be included.","verdict_should_be":"UNCHANGED","load_bearing_attack":"To invalidate Landauer, the paper needs a complete erasure cycle: starting from a symmetric double-well memory in the unknown state (50/50), ending in state 0 with the control (the tilt) returned to its initial value, and with zero dissipated work in the quasistatic limit. Section V.F.2 and Fig. 10 only assert such a protocol ('The motion ... can be as slow as desired') but do not specify U(x), lambda(t), or the final reset. The omission is not merely a detail. For a genuinely quasistatic isothermal process, the system is in equilibrium at every instant, so the final equilibrium distribution is determined by the final Hamiltonian. If the tilt is reset, the final Hamiltonian is the original symmetric one and the equilibrium populations are again 50/50; the bit is not erased. If the tilt is not reset, the memory is no longer in its standard reusable configuration, and the cost of restoring the tilt must be included; the free-energy balance of the closed cycle then gives the usual kT ln2 (or the equivalent entropy bookkeeping). The Section V.E cycle (Figs. 8-9) has the same structure: steps 3-4 restore the bit to random, so the net operation is not an erasure. Thus the central counterexample needs a concrete closed-cycle model before the claim 'Landauer is not a law' can be evaluated; the current text does not provide one.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper is a history-and-philosophy essay that traces the role of the observer in the energy-information link from classical thermodynamics through statistical mechanics, Shannon information theory, and Landauer's principle. Its central thesis is that thermodynamics is inherently observer-dependent, that information is a non-material abstraction stored in material data-bits, and that Landauer's principle is not a fundamental law of physics. The concrete invalidation claim is made in Section V: for an unknown initial bit value, the two-step Landauer erasure followed by a restore cycle is said to be thermodynamically reversible (Section V.E), and a single externally controlled time-varying field acting on a particle in a fixed repulsive potential is said to allow quasistatic erasure of either known or unknown bits (Section V.F.2, Fig. 10). The paper also critiques information-mass equivalence arguments.","tokens_in":28297,"tokens_out":9087,"duration_ms":115515,"significance":"If the invalidation of Landauer's principle were established, the paper would make a significant contribution to a long-standing debate in the foundations of thermodynamics and computation. The essay has genuine strengths: the historical narrative is readable, the free-expansion and Gibbs-paradox examples are presented clearly, the distinction between a piece of information and its data-bit support (Section V.A) is useful, and the observation that logical irreversibility and thermodynamic irreversibility are not the same thing (Section V.B) is supported by Maroney and Bennett. The paper also correctly identifies the shift of dissipation location in Brillouin versus Landauer formulations. However, the central counterexample is only sketched; no explicit potential, control schedule, work-heat bookkeeping, or closed-cycle proof is given. The significance of the paper is therefore conditional on the authors supplying a concrete, self-contained model of a zero-cost erasure cycle.","major_comments":[{"comment":"The decisive counterexample is asserted rather than demonstrated. The text states that 'the motion of the particle follows the direction of maximum potential descent and can be as slow as desired,' but it does not specify the potential U(x), the time-dependent external field, the control schedule lambda(t), or the final reset procedure. This is load-bearing: in a genuinely quasistatic isothermal process the system is in equilibrium at every instant, so if the control is reset to the initial symmetric configuration the final equilibrium distribution is again 50/50 and the bit is not erased; if the control is not reset, the memory is not in its standard reusable configuration and the work and heat needed to restore the control must be included in the cycle balance. The paper needs a complete closed-cycle model with explicit U(x) and lambda(t), together with a proof that the protocol ends in state 0 with the control restored and with dissipated work going to zero in the quasistatic limit, before the claim 'Landauer is not a law' can be evaluated.","section":"V.F.2 and Fig. 10"},{"comment":"The 'reversible Landauer erasure for an unknown value' is not an erasure. The protocol's steps 3 and 4 restore the data-bit to a random unknown value, so the complete cycle maps the initial 50/50 distribution to itself; it does not set the bit to state 0. A zero-cost cycle that erases a bit and then recreates it does not show that erasure itself has zero cost. The text concedes this when it says 'the thermodynamic cycle is not yet closed' and adds restore steps. To refute Landauer, the paper must exhibit a protocol that starts from an unknown bit (0 or 1) and ends with a known 0, with the control returned to its initial value, and with dissipated work below kT ln2 in the quasistatic limit. As written, Section V.E establishes only that a particular no-op cycle is reversible, which is not the same as reversible erasure.","section":"V.E and Figs. 8-9"},{"comment":"The alternative erasure counterexamples are delegated to the author's own prior papers without a self-contained derivation. Section V.F.1 says 'counter-examples to the generality of Landauer erasure have been proposed [90, 91]' and then describes the scheme in one sentence; Section V.F.2's scheme is also a sketch. Since the conclusion that Landauer's principle is not a fundamental law rests entirely on these counterexamples, the manuscript should present the actual model, potential, and thermodynamic bookkeeping in enough detail for the reader to check the claim, or explicitly state the assumptions and provide a verifiable supplement. Citations to the author's own earlier work are not sufficient for a claim that overturns a widely accepted principle.","section":"V.F.1 and references [87, 90, 91]"}],"minor_comments":[{"comment":"The free-expansion example is used as a foundational illustration of observer dependence, but the manuscript does not engage with the standard objection that thermodynamics defines macroscopic states by the values of state variables, not by the observer's state of knowledge. A sentence acknowledging this alternative reading would help the reader locate the paper's operationalist commitments.","section":"I.C.1"},{"comment":"The claim that the usual resolutions of the second Gibbs paradox 'benefit from the cancellation of two approximations' and that the exact Stirling formula changes the result is delegated to reference [55] without a derivation; since this point is not central to the main argument, a brief statement of the corrected result would improve readability.","section":"II.D"},{"comment":"The caption says 'the same time-varying tilt of the surface relative to the field' but does not define the parameter that is time-dependent or how it is reset. Clarifying the control schedule would make the figure self-contained.","section":"Fig. 10 caption"},{"comment":"The sentence 'since m(x)=n=2^log2 n, x requires log2(n) bits to be recorded' is notationally confusing; it should say that the outcome x is encoded by the integer n, whose binary representation requires roughly log2(n) bits.","section":"III.A"},{"comment":"The statement 'Both must be erased so that the information is irretrievably destroyed' should be qualified as 'both must be erased and no other copy may exist elsewhere,' since the previous paragraph already makes this point but the sentence as written could be read as claiming that two copies are always both necessary.","section":"V.A"},{"comment":"There is a typo in the van Kampen quotation, 'The choise,' and the paper uses 'textquote' markup artifacts in the abstract; these should be cleaned up, and the reference list should be checked for consistency of author names and formatting.","section":"References"}],"recommendation":"major_revision","confidential_remarks":"The paper is a broad historical and philosophical essay with a strong opinionated voice, and it is likely to interest readers of physics.hist-ph. My main concern is that the central technical claim is not yet supported by a concrete model. The author's own prior papers are cited for the decisive counterexamples, but the current manuscript does not provide enough detail for a referee or reader to verify the claimed zero-cost closed-cycle erasure. I would encourage the editor to seek a revised version in which either the counterexample is fully specified (potential, control schedule, closed-cycle balance) or the paper's conclusion is softened to a claim about the observer-dependence and implementation-dependence of erasure costs, which the known/unknown distinction can support. The self-citation pattern is heavy but not improper given the specialized topic; a self-contained derivation would remove any concern."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear colleague,\n\nThe paper is a readable chronicle of the observer's role in thermodynamics, and the historical half earns its keep. The free-expansion and Gibbs-paradox examples are crisp, the known/unknown distinction is real, and the author correctly notes Bennett's own caveat that erasing unknown random data can be thermodynamically reversible. The critique of the 'information has mass' literature is also fair: copies don't multiply information, and old Szilard-cycle data isn't stored energy. If the paper were only that, it would be a useful essay.\n\nThe problem is the central invalidation of Landauer's principle. The decisive pieces in Section V don't close the cycle. In V.E, the 'reversible erasure' of an unknown bit is actually erasure to 0 followed by expansion back to S and randomization; the net operation is not an erasure, and the erase step itself still costs T ln2. In V.F.2, the tilt protocol is asserted but never specified: no U(x), no lambda(t), no proof that the two initial states merge quasistatically at zero dissipation. The stress-test objection lands. If the control is reset, the final Hamiltonian is symmetric and the equilibrium distribution is 50/50, so nothing has been erased; if the control is not reset, the memory is no longer a usable symmetric bit and the reset cost has to be counted.\n\nThe paper leans heavily on the author's own prior papers for the counterexamples [87,90,91], and those are not reproduced or independently verified here. Maroney gives independent support for the known/unknown point, but not for the stronger claim that a controlled quasistatic erasure can beat the bound.\n\nNet: I don't think the paper establishes that Landauer's principle is not a law. The philosophical and historical narrative is worth a serious referee, and the challenge to Landauer is important enough that a referee should look at it carefully. But the current version should not be accepted: the invalidation claim needs either a fully specified closed-cycle protocol or a withdrawal to the weaker, better-supported claim that Landauer's bound is observer-relative in the specific known/unknown sense. I'd send it out, with the expectation that the referee asks for that.","headline":"A useful historical essay on the observer in thermodynamics, but the central claim that Landauer's principle is not a law is not backed by a closed-cycle counterexample.","tokens_in":28751,"tokens_out":5560,"would_cite":false,"duration_ms":59302,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["05.70.-a","89.70.Cf"],"model":"deepseek-v4-flash","headline":"This paper argues that Landauer's principle—erasing one bit of data costs at least $T\\ln 2$ of heat—is not a fundamental law, because erasure can be thermodynamically reversible under the right knowledge and control.","keywords":["Landauer principle","thermodynamic reversibility","observer dependence","Shannon entropy","Maxwell's demon","Gibbs paradox","information is physical","data-bit versus information"],"falsifier":"For the paper's Figure 10 setup, write an explicit potential and drive schedule, then measure or simulate the dissipated heat as the drive speed tends to zero. If for either initial bit value the quasistatic heat fails to go to zero—or if the two trajectories cannot merge without passing through an unstable point where control is lost—the counterexample fails and the Landauer bound survives. A numerical simulation of overdamped Langevin dynamics with the explicit landscape would settle the question.","tokens_in":27709,"feed_emoji":"🔥","tokens_out":11041,"duration_ms":110844,"temperature":0.7,"pith_summary":"The paper tells the history of the link between energy and information as an oscillation between two views: that objective reality is what exists without an observer, and that it is what an observer can interact with. It argues that thermodynamics has always been observer-dependent, that statistical mechanics tried to escape this but failed, and that Shannon information theory and then Landauer's principle were further attempts to make the link look intrinsic. The paper's own contribution is the last step: it claims Landauer's principle is not a fundamental law. The reason is that erasing a data-bit need not cost a minimum heat of $T\\ln 2$; whether the erasure is thermodynamically reversible depends on whether the initial bit value is known and on the control protocol used, and at least one protocol erases known or unknown bits quasistatically. If right, the celebrated energy cost of forgetting is an observer- and implementation-dependent statement, not a law of physics.","feed_headline":"Landauer's erasure bound is not a law of physics, paper argues","feed_subtitle":"Erasing a bit can be heat-free when the value is unknown or when a single time-varying field steers both initial states to 0.","key_machinery":"The central mechanism is the distinction between a data-bit and a piece of information. A data-bit is a material system with two distinguishable stationary states; a piece of information is an abstract value that is not localized in any copy, so erasing one data-bit does not itself destroy information unless no copy remains. On top of this, the argument uses the Clausius inequality as the criterion of thermodynamic reversibility, applied to concrete implementations of a bit: the particle-in-a-box used by Landauer, and the paper's alternative of a particle on a fixed repulsive topographic relief driven by a single time-varying external field. In Landauer's derivation, erasure is free expansion followed by isothermal compression, and the merging of two paths is treated as necessarily uncontrolled; the paper's counterexample replaces that step by a controlled merging in which two initial trajectories can be quasistatically steered to the same final state. The known/unknown distinction does the final work: the first step of Landauer erasure is reversible for an unknown bit value and irreversible for a known one, so no intrinsic minimum heat cost follows for erasure as such.","core_discovery":"On the paper's own terms, the central claim is that Landauer's erasure principle is not a fundamental law of physics. The author distinguishes a piece of information, an abstract value that can be copied and is not located in any physical support, from a data-bit, a material system whose erasure is a thermodynamic process. Judged by the Clausius inequality, the thermodynamic reversibility of erasure depends on whether the initial value is known to the observer and on which control protocol is used, not on the fact that the logical operation 'set to 0' is non-injective. For an unknown initial value, the two-step Landauer erasure is thermodynamically reversible when the full restore cycle is included; for a known value it is irreversible, but an alternative procedure—a single time-varying external field acting on a particle in a fixed potential landscape—can erase either value quasistatically. Hence the inequality $W_{\\mathrm{erase}}\\ge T\\ln 2$ does not express an intrinsic energy cost of forgetting; it characterizes a particular class of procedures and a particular observer. The same data-bit/information distinction is used to reject the information-mass equivalence that some authors present as a consequence of Landauer's principle.","pith_inferences":["Editorial inference: if the paper is right, existing measurements reported as tests of the Landauer bound may actually be measuring a chosen protocol plus the experimenter's state of knowledge, so the same physical memory could show a $T\\ln 2$ cost in one arrangement and zero in another.","Editorial inference: a direct next step would be to build an explicit topographic-relief potential and drive schedule—for example with a colloidal particle in an optical or magnetic trap—and check in the slow-drive limit that the dissipated heat approaches zero for both initial bit values.","Editorial inference: taken further, the data-bit/information distinction suggests that 'destroying information' is a global property of the whole memory (all copies gone), while local operations on a single data-bit can always be made reversible; this reframes the thermodynamics of computation away from logical non-injectivity and toward actual dissipative mechanisms."],"forward_implications":["The bound $W_{\\mathrm{erase}}\\ge T\\ln 2$ would be demoted from a universal law to a statement about specific erasure protocols, valid when the protocol and the observer's knowledge make the process irreversible.","The information-mass equivalence, which is derived from Landauer erasure in the literature, would lose its foundation, since erasure need not dissipate and a data-bit need not store information.","Brillouin's negentropy principle would remain standing: acquiring one bit of information still costs at least $T\\ln 2$ of work through the Clausius inequality, but the cost is a property of the whole acquisition cycle, not of the erasure step.","Any claimed experimental measurement of the Landauer bound would have to specify the initial knowledge about the bit and the full control protocol before the result can be attributed to 'erasure' rather than to the chosen implementation."],"supporting_citations":[{"why":"Supplies the original Landauer derivation of the erasure bound and the claim that logical irreversibility implies thermodynamic irreversibility, the target of the invalidation.","marker":"[73]"},{"why":"Contains the acknowledged nuance that erasing random data can be thermodynamically reversible, on which the known/unknown distinction in Section V.E relies.","marker":"[76]"},{"why":"Provides the analysis of thermodynamic reversibility of Landauer erasure for unknown initial values.","marker":"[89]"},{"why":"Introduces the boolean-logical versus thermodynamic irreversibility distinction and the topographic-relief implementation used in the counterexample.","marker":"[87]"},{"why":"Presents the earlier concrete objection that incomplete thermodynamic states can allow quasistatic erasure.","marker":"[90]"},{"why":"Argues against information having mass, used to invalidate the information-mass equivalence derived from Landauer's principle.","marker":"[91]"},{"why":"States Brillouin's negentropy principle, the second-law-based alternative that remains valid and must be distinguished from Landauer's claim.","marker":"[66]"},{"why":"Foundation of Shannon information theory, which the paper uses to define the quantity of information at stake.","marker":"[61]"}],"fun_headline_variants":["Landauer's principle isn't a fundamental law, new paper claims","Erasure heat cost depends on observer, not just logic","Bit erasure can be free if value unknown, paper says","Thermodynamic cost of forgetting is protocol-dependent","Landauer's bound fails as a universal law, argues paper"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that a single smoothly varied external field, acting on a particle in a fixed repulsive potential, can bring both possible starting positions to the same final position along paths whose heat dissipation can be made arbitrarily small by slowing the drive.","fun_headline_variants_meta":{"raw":{"variants":["Landauer's principle isn't a fundamental law, new paper claims","Erasure heat cost depends on observer, not just logic","Bit erasure can be free if value unknown, paper says","Thermodynamic cost of forgetting is protocol-dependent","Landauer's bound fails as a universal law, argues paper"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000416,"raw_usage":{"total_tokens":2236,"prompt_tokens":1121,"completion_tokens":1115,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":737,"completion_tokens_details":{"reasoning_tokens":1031}},"tokens_in":737,"tokens_out":1115,"duration_ms":121132,"temperature":1.0,"reasoning_tokens":1031,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T14:02:10.093362+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"For the paper's Figure 10 setup, write an explicit potential and drive schedule, then measure or simulate the dissipated heat as the drive speed tends to zero. If for either initial bit value the quasistatic heat fails to go to zero—or if the two trajectories cannot merge without passing through an unstable point where control is lost—the counterexample fails and the Landauer bound survives. A numerical simulation of overdamped Langevin dynamics with the explicit landscape would settle the question.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the original Landauer derivation of the erasure bound and the claim that logical irreversibility implies thermodynamic irreversibility, the target of the invalidation."},{"cited_title":"Frenkel, Why colloidal systems can be described by statistical mechanics: some not very original comments on the Gibbs paradox, Molecular Physics112, 2325 (2014)","cited_arxiv_id":null,"evidence_quote":"Contains the acknowledged nuance that erasing random data can be thermodynamically reversible, on which the known/unknown distinction in Section V.E relies."},{"cited_title":"Brillouin, The negentropy principle of information, Journal of Applied Physics24, 1152 (1953)","cited_arxiv_id":null,"evidence_quote":"Argues against information having mass, used to invalidate the information-mass equivalence derived from Landauer's principle."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"States Brillouin's negentropy principle, the second-law-based alternative that remains valid and must be distinguished from Landauer's claim."},{"cited_title":"Boltzmann,Lectures on gas theory(Dover ed., New York, NY, USA, 1964)","cited_arxiv_id":null,"evidence_quote":"Foundation of Shannon information theory, which the paper uses to define the quantity of information at stake."}],"review_version":1}