{"id":"d3378511-b890-4a81-93b0-a1601c2455f4","arxiv_id":"2501.13593","paper_version":1,"verdict":"UNVERDICTED","confidence":"HIGH","novelty_score":2.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"The paper reviews experimental information engines and argues they have advanced beyond proof of concept into active, many-body, inertial, and optimally controlled regimes.","lead":"This paper is a perspective reviewing experimental information engines that use measurement and feedback to extract work from thermal fluctuations. It argues the field has moved beyond proof of concept and sketches new directions in active matter, many-body systems, inertia, and optimal control.","discovery_kind":"review","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Unsupported claim that ideal-gas many-body Szilard work per measurement decreases with N; Section III.A's active-matter motivation may rest on an incorrect baseline.","rationale":"The paper is a perspective whose central assertion, as identified by the reader, is that experimental information engines have moved beyond proof of concept and that the field is ready for new directions in active matter, many-body systems, inertia, and optimal control. The reader's weakest assumption was the general reliability of cited experimental realizations. I find a more specific and more internally checkable soft spot: the quantitative baseline used to motivate the active-matter and many-body direction. Section III.A's contrast between ideal-gas and active-particle many-body Szilard engines is presented as a monotonic decrease of work per measurement with particle number for ideal gases, but no calculation, citation, or protocol specification is given. Elementary scaling suggests the claim is at least protocol-dependent and possibly wrong: the typical imbalance grows as sqrt(N), and quasistatic demixing yields constant typical work, while a constant-load protocol could give a decrease only if the load is not rescaled. Since this contrast is the paper's primary argument that giant number fluctuations make active many-body engines a promising new direction, an error here would undercut a key part of the 'beyond proof of concept' narrative. A targeted simulation or re-analysis of the original data in ref [18] would settle the issue. The verdict is moved to CONDITIONAL rather than REJECT because the paper remains a useful review, but its active-matter claim should be corrected or explicitly qualified with the protocol details.","tokens_in":15814,"tokens_out":14003,"duration_ms":129349,"concrete_test":"Simulate or analytically compute the average extracted work per measurement for an ideal-gas Szilard engine with N particles under the protocol of ref [18]: measure n_L, insert a partition, attach a constant mass, and let the partition move to force balance; then repeat with a per-cycle optimized mass. Compute W(N) for N = 10^2 to 10^6 and compare with the active-particle data in ref [18]. If W(N) is non-monotonic or increases with N under optimized load, the Section III.A claim is incorrect and the active-matter advantage needs re-baselining.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"Section III.A states that for an ideal-gas many-body Szilard engine, 'as the number of molecules increases, relatively small imbalances in the density, that result in even smaller work extraction, become more likely. Consequently, the average work extracted per measurement decreases as the number of molecules grows.' This claim is load-bearing: it frames the many-body active engine [18] as a qualitatively new direction enabled by giant number fluctuations. No derivation or citation accompanies it, and the scaling is not obviously correct. With N ideal-gas particles, the typical population imbalance after a left/right measurement scales as ΔN ~ sqrt(N), while the quasistatic demixing work is approximately k_B T [n_L ln(2n_L/N) + n_R ln(2n_R/N)]. Expanding for n_L, n_R = N/2 ± ΔN/2 gives W ≈ k_B T (ΔN)^2/N, which is O(1) for typical fluctuations. If the protocol instead uses a constant load and allows a small displacement, the displacement scales as ΔN/N, so W may decrease with N for a fixed load; however, the paper does not specify the load protocol, and under a per-cycle optimized load W does not decrease. The perspective thus asserts a monotonic decrease without specifying the feedback protocol or load scaling, and the claimed contrast with active particles is not established.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript is a perspective on experimental information engines. It argues that the field has matured beyond proof-of-concept demonstrations, classifies existing experiments into phase-splitting (Szilard-type) and conditional-feedback engines, reviews the thermodynamic framework (Eqs. 1-4), and proposes three future directions: active and many-body working substances, inertial effects, and optimal control. The paper does not present new derivations or original experimental data; its contribution is a synthesis and a research agenda.","tokens_in":16038,"tokens_out":8531,"duration_ms":74601,"significance":"If its assessments are accurate, the paper provides a useful roadmap for experimentalists and theorists working on information engines. It correctly reproduces the standard generalized Second Law and information-work relations, and it brings together a diverse set of experiments across microscopic and macroscopic scales. The division into phase-splitting and conditional engines is a helpful organizing principle, and the proposed future directions are timely. However, two specific technical claims used to motivate the new directions are either unsupported or inconsistent with the cited literature: the asserted N-dependence of the ideal-gas many-body Szilard engine in Section III.A, and the asserted monotonic decrease of power with sampling time for overdamped engines in Section III.B. These issues do not invalidate the entire perspective but need to be corrected before the roadmap can be fully endorsed.","major_comments":[{"comment":"The statement that for an ideal-gas many-body Szilard engine 'the average work extracted per measurement decreases as the number of molecules grows' is stated without derivation or citation, and as a general claim it is not correct. For a protocol in which the partition is inserted after measuring n_L and n_R and then moved quasistatically to the equal-pressure position, the work per cycle is approximately k_B T [n_L ln(2n_L/N) + n_R ln(2n_R/N)] ≈ k_B T (ΔN)^2/N, with ΔN = n_L - n_R. Since typical fluctuations scale as ΔN ∼ √N, this work is O(k_B T) and does not decrease with N in the large-N limit. A decreasing work with N is obtained only under a specific load protocol (e.g., a fixed load with displacement proportional to ΔN/N), which the paper does not specify. Because this claim is used to motivate the many-body active engine as a qualitatively new regime, the protocol and derivation must be stated, or the claim must be revised.","section":"Section III.A"},{"comment":"The claim that 'in the overdamped case, the power monotonically decreases as τ increases [104] while in the underdamped case, power is maximal for a finite τ' is inconsistent with the discussion in Section II.D, where an optimal measurement rate is said to exist for overdamped colloidal engines [37,42], and with the cited experiment [21] that maximizes power of an information engine. If the authors intend a specific protocol or a specific definition of τ, that must be stated; otherwise the contrast between overdamped and underdamped regimes is not supported by the cited literature.","section":"Section III.B"}],"minor_comments":[{"comment":"The sentence 'The operation of an information engine requires real-time measurements... spans various length scales' has a missing subject; 'spans' should be 'span' or the sentence should be rephrased.","section":"Section II.C"},{"comment":"There is a typo in 'inertial effects natuarally arise' – 'natuarally' should be 'naturally'.","section":"Section III.B"},{"comment":"The phrase 'the the Israel Science Foundation' contains a duplicated article and should be corrected.","section":"Acknowledgments"},{"comment":"'Reversely' should be 'Conversely'.","section":"Section II.D"},{"comment":"The notation Σ_i W_i is used without a clear definition of the index i; please specify what the sum runs over (e.g., the different work contributions in a cycle).","section":"Equation (4)"},{"comment":"The phrase 'which also releases the constrain of erasing the memory' should read 'constraint'.","section":"Footnote [10]"}],"recommendation":"major_revision","confidential_remarks":"This is a perspective rather than a research article, so the bar for novel derivations is appropriately lower. The most original claim—that many-body active information engines constitute a qualitatively new regime—rests in part on ref. [18] by the same group, and the scaling argument in Section III.A should be checked carefully. The issues I raise are local and fixable; I do not see a basis for rejection, but the unsupported scaling claim and the overdamped-power contradiction should be addressed in a revision."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Rough take: this is a review/perspective, not a new-results paper, but it's a solid and readable survey of where experimental information engines have gone since the first proof-of-concept demonstrations. The split into phase-splitting (Szilard-type) and conditional (phase-contracting) engines is a useful organizing device, even if the authors themselves note that the boundary is debatable. The thermodynamics section correctly reproduces the Sagawa-Ueda bound, the unavailable-information correction, and the efficiency definition, and the survey of experiments covers the main lines: colloids, DNA, single-electron boxes, granular gas, bristle bots, and micro-cantilevers.\n\nThe main soft spot is in Section III.A. The paper claims that for an ideal-gas many-body Szilard engine, average work per measurement decreases as the number of molecules grows. No derivation or citation is given, and the claim is not obviously correct. With typical binomial fluctuations, ΔN ~ sqrt(N), the quasistatic demixing work is about k_B T (ΔN)^2/N, which is O(1) — it does not decrease with N. If the protocol uses a fixed load and a small partition displacement, the extracted work may indeed fall with N, but the paper doesn't specify the load protocol. This claim frames the active-particle engine as qualitatively new because of giant number fluctuations, so it's load-bearing. A referee should ask the authors to either derive the scaling under a specified protocol or soften the statement.\n\nThe other caveat is standard for a perspective: the paper trusts the cited experiments. That's fine for a review, but it means the 'beyond proof of concept' narrative rests on the reliability of those measurements.\n\nOverall, this is a useful overview for someone entering the field or for a researcher wanting a compact map of recent experiments and open directions. It deserves a serious referee, not a desk reject, but the scaling issue should be fixed before publication. I'd probably cite it as a secondary source, though the active-matter motivation needs to be cleaned up.","headline":"A competent and useful review of experimental information engines, but the ideal-gas many-body scaling claim in Sec. III.A is unsupported and likely wrong under standard protocols.","tokens_in":16542,"tokens_out":2257,"would_cite":true,"duration_ms":20691,"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":"Experimental information engines have moved from proof of concept to quantitative machines that now reach into active matter, many-body systems, and inertial dynamics.","keywords":["information engines","Maxwell's demon","Szilard engine","stochastic thermodynamics","measurement-based feedback","active matter","nonequilibrium thermodynamics","optimal control"],"falsifier":"A high-precision calorimetric or force measurement of any of these information engines that extracts more work than $k_B T(I - I_u)$ permits, with $I$ and $I_u$ measured independently, would falsify the generalized Second Law on which the perspective's assessment of the field rests; alternatively, a reanalysis showing that the reported work values in the cited experiments vanish once feedback delay and measurement noise are accounted for would undermine the 'beyond proof of concept' claim.","tokens_in":15647,"feed_emoji":"⚙️","tokens_out":6756,"duration_ms":55436,"temperature":0.7,"pith_summary":"This perspective argues that experimental information engines have matured from thought experiments and first demonstrations into a field that tests quantitative thermodynamic bounds and explores new working substances. The authors review realizations across scales, from colloidal particles and DNA hairpins in optical traps to nanoscale single-electron boxes and macroscopic bristle-bot collectives, and sort them into phase-space-splitting and conditional-feedback classes. They present the generalized Second Law $\\Delta W \\geq \\Delta F - k_B T (I - I_u)$ as the framework that lets experiments quantify information-to-work conversion, with $I$ the information gathered by measurement and $I_u$ the information made unavailable by irreversibility. The paper's main forward-looking claim is that active, many-body, inertial, and optimally controlled information engines are the next stage of the field.","feed_headline":"Information engines move past proof of concept","feed_subtitle":"Colloidal, DNA, electronic, and active-matter engines now test a generalized second law.","key_machinery":"The carrying device is the measurement–feedback–erasure cycle analyzed through the generalized Second Law for feedback control, $\\Delta W \\geq \\Delta F - k_B T(I - I_u)$, with $I = \\sum_m p(m) D_{\\mathrm{KL}}(\\rho(x|m) \\| \\rho_{\\mathrm{eq}})$ the information acquired by measurement and $I_u$ the unavailable information quantifying irreversibility of the protocol. This inequality, following from the Sagawa–Ueda generalized Jarzynski equality, supplies the quantitative bound that every experimental realization is compared against, and the unavailable-information term is what lets experiments diagnosed as irreversible still be assessed. The paper also uses a classification of engines into phase-space splitting versus conditional phase-space contracting, since the two operation modes have different symmetry-breaking and reversibility signatures.","core_discovery":"The central claim is that information engines are no longer only proof-of-concept demonstrations of Maxwell's demon. According to the paper, modern experimental engines confirm the generalized Second Law, extract measurable work and power, and are now being built around non-equilibrium working substances: self-propelled particles, many-particle active systems, and underdamped micro-cantilevers. A key supporting assertion is the distinction between phase-space-splitting engines (Szilard-type) and conditional phase-space-contracting engines, because the two classes differ in what information does and in their reversibility properties. On this basis the authors argue that the next advances will come from giant number fluctuations in active matter, inertial resonances at finite measurement intervals, and optimal control protocols, including learning-based protocols, for feedback and erasure.","pith_inferences":["One step the authors leave implicit: if the same generalized bound governs active and inertial engines, those platforms become precision testbeds for quantifying how much information is destroyed by feedback delay, measurement noise, and non-quasistatic driving.","The active-matter results suggest a concrete bio-inspired extension: molecular machines operating in the cytoplasm may be understood as information engines harvesting non-equilibrium fluctuations, which could be tested by measuring whether their work output tracks the measured information term $I$ under controlled noise.","The optimal-control and reinforcement-learning discussion points toward autonomous information engines that infer their own parameters from measurement histories; a testable extension would be to run a colloidal engine with an online learning controller and compare its extracted work against the optimized theoretical protocol.","Because most experiments dissipate extracted energy as heat rather than coupling to external loads, a practical extension would be to build mesoscopic engines that do mechanical or electrical work on a load, testing whether the generalized bound survives under continuous loading."],"forward_implications":["If the generalized Second Law bound is correct, every irreversible information engine has a tight upper bound on extractable work, so reported efficiencies can be compared on a common scale across colloidal, electronic, DNA, and active-matter platforms.","With active working substances, giant number fluctuations make many-body Szilard engine work per measurement increase with particle number, opposite to the ideal-gas trend, so larger active collectives should give proportionally more extractable work.","For inertial (underdamped) engines, power output is non-monotonic in the measurement interval and can resonate with relaxation times, so finite-time feedback can be tuned to maximize power rather than degraded by faster sampling.","Optimal control and model-free learning can in principle design feedback and erasure protocols that approach the generalized bound at finite time, raising both power and information-to-work efficiency."],"supporting_citations":[{"why":"Supplies the generalized Jarzynski equality and the information-dependent Second Law inequality that defines the thermodynamic resource.","marker":"[7]"},{"why":"Introduces unavailable information $I_u$ and the generalized bound Eq. (3) that the paper uses to assess irreversible engines.","marker":"[9]"},{"why":"First experimental demonstration of information-to-energy conversion, validating the generalized Jarzynski equality in a colloidal system.","marker":"[11]"},{"why":"Colloidal phase-space-splitting engine; used to connect the energetics of symmetry breaking with information thermodynamics.","marker":"[12]"},{"why":"Lossless Brownian information engine realizing conditional feedback; experimental test of the tight bound on extractable work.","marker":"[13]"},{"why":"DNA-hairpin continuous Maxwell demon showing large work extraction and approaching the Landauer limit.","marker":"[16]"},{"why":"Many-body active Szilard engine with bristle-bots; supplies the giant-number-fluctuation work enhancement claim.","marker":"[18]"},{"why":"Colloidal information engine against gravity; supplies measurement-rate tuning, power maximization, and the inertial/underdamped discussion.","marker":"[21]"},{"why":"Micro-cantilever discrete-sampling information engine; supplies inertial effects and finite-$τ$ power resonances.","marker":"[40]"},{"why":"First-passage information engine; experimental verification of the generalized Second Law with unavailable information.","marker":"[41]"}],"fun_headline_variants":["Information engines now extract real work and power","Maxwell's demon enters the real world","Generalized second law passes info engine tests","Active matter and inertia expand info engine horizons"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The narrative that the field has moved beyond proof of concept rests on the reliability of the cited experiments: each must genuinely implement the described measurement-feedback loop, and its reported work and efficiency values must faithfully reflect the stochastic-thermodynamic quantities they claim to measure.","fun_headline_variants_meta":{"raw":{"variants":["Information engines now extract real work and power","Maxwell's demon enters the real world","Generalized second law passes info engine tests","Active matter and inertia expand info engine horizons"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000418,"raw_usage":{"total_tokens":2080,"prompt_tokens":801,"completion_tokens":1279,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":417,"completion_tokens_details":{"reasoning_tokens":1224}},"tokens_in":417,"tokens_out":1279,"duration_ms":9884,"temperature":1.0,"reasoning_tokens":1224,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T15:46:22.309477+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A high-precision calorimetric or force measurement of any of these information engines that extracts more work than $k_B T(I - I_u)$ permits, with $I$ and $I_u$ measured independently, would falsify the generalized Second Law on which the perspective's assessment of the field rests; alternatively, a reanalysis showing that the reported work values in the cited experiments vanish once feedback delay and measurement noise are accounted for would undermine the 'beyond proof of concept' claim.","supporting_citations":[{"cited_title":"Archambault, C","cited_arxiv_id":null,"evidence_quote":"Micro-cantilever discrete-sampling information engine; supplies inertial effects and finite-$τ$ power resonances."}],"review_version":1}