{ "schema": "auturnal.deep-research-review.v2", "slug": "information-thermodynamics", "title": "The Physical Price of Information: Erasure, Feedback and Finite-Time Thermodynamics", "category": "Information and engineering", "dek": "Information processing is implemented by physical systems, but logical operations do not all carry the same thermodynamic obligation.", "english": "Information processing is implemented by physical systems, but logical operations do not all carry the same thermodynamic obligation. This review separates erasure, measurement, feedback and finite-time control. Landauer's principle and experimental information-to-energy conversion provide a framework for accounting, while a binary-entropy model shows why error tolerance and initial uncertainty matter to the minimum work of resetting a memory.", "abstract": "Information processing is implemented by physical systems, but logical operations do not all carry the same thermodynamic obligation. This review separates erasure, measurement, feedback and finite-time control. Landauer's principle and experimental information-to-energy conversion provide a framework for accounting, while a binary-entropy model shows why error tolerance and initial uncertainty matter to the minimum work of resetting a memory.", "refs": [ "ef-40-1", "ef-40-2", "ef-40-3" ], "sections": [ [ "The boundary of the machine matters", "A memory, controller and heat bath exchange energy and information. Examining only one subsystem can make a process appear to violate a thermodynamic bound when a cost has merely moved elsewhere. A complete account specifies initial and final states, work reservoirs, heat exchange and what information remains in the controller.", [] ], [ "Theory and controlled experiments", "Landauer analyzed the relation between logical irreversibility and heat generation in computing. Berut and colleagues experimentally tested the erasure bound with a controlled physical memory. Toyabe and colleagues demonstrated information-to-energy conversion through feedback on a Brownian system. These results concern carefully defined operations; they do not imply that every logical gate must dissipate the same fixed energy.", [ "ef-40-1", "ef-40-2", "ef-40-3" ] ], [ "Erasure reduces logical uncertainty", "Resetting an initially unknown bit to a known state compresses its logical distribution. For a symmetric memory in contact with a thermal bath, that reduction imposes a minimum average work under ideal reversible conditions. Copying a known bit or performing a logically reversible transformation is a different operation. The familiar bound should not be detached from the initial distribution and reset requirement.", [] ], [ "Errors change the target distribution", "Allowing a nonzero reset error leaves residual logical uncertainty. The ideal minimum work is then lower than for perfect reset under the symmetric-memory assumptions. This is not free computation: the final memory is less reliable. A fair comparison reports both energy and error probability, rather than treating a cheaper imperfect operation as equivalent to a reliable one.", [] ], [ "Finite time adds a control problem", "Approaching a reversible bound often requires slow operation and careful control. Faster protocols can dissipate additional work, while practical electronics incur costs far above the ideal limit. A finite-time experiment should separate the fundamental state-change requirement from dissipation caused by its particular control protocol.", [] ], [ "Measurement and feedback are not the same step", "A controller can use acquired information to extract work from a system. The accounting must also include the physical state of the memory that stores the measurement and any reset needed for a repeated cycle. Ignoring that reset can make a one-shot process look like a perpetual source of useful work.", [] ], [ "Mutual information bounds a resource", "Feedback benefit depends on how informative the measurement is about the controlled system. A noisy measurement contains less usable information than a perfect one. The relevant quantity is statistical dependence, not the number of characters in a stored file. The conversion also depends on control constraints, so an information bound is not a guarantee that a practical engine reaches it.", [] ], [ "Computing claims need consistent units", "Energy per operation, power and throughput are different metrics. Temperature, error tolerance and the definition of an operation must accompany comparisons with thermodynamic limits. The calculations below describe idealized binary memories, not the energy consumption of a modern processor or an estimate of the cost of running an AI model.", [] ], [ "A thermodynamic cycle must include the memory controller", "A memory-reset experiment couples logical states to a physical system and a time-dependent control protocol. To interpret average work, the initial distribution, final accuracy and endpoint energy landscape must be specified. A proposed accounting would include the work associated with manipulating that landscape and would distinguish heat transferred to the bath from work applied by the controller. The ideal information bound concerns a complete transformation under stated conditions. It does not imply that an arbitrary implementation reaches the bound or that measuring one part of the apparatus captures the full energetic cost.", [] ], [ "Feedback uses information with a physical history", "A controller can extract work conditional on a measurement, but the measurement record and its eventual reuse or erasure belong to the full cycle. A proposed experiment would quantify the correlation acquired, the feedback rule and the reset operation separately, then close the energy and entropy accounting over repeated cycles. Finite-time driving generally introduces additional dissipation, while errors reduce the information transformation required. The binary-entropy curve describes a specific symmetric memory model and makes that dependence explicit. Its lesson is a connection between operationally defined information and thermodynamics, rather than a universal price assigned to every computation or every bit stored.", [] ], [ "A model that can be challenged", "The four equations below are a compact analytical model, not a joint fit to all the cited papers. Their interpretation is as follows. For an initially unbiased symmetric bit reset reliably in an isothermal idealization, average work has this lower bound. Binary entropy is expressed in natural-log units here; its limiting value is zero at p equal to zero or one. A symmetric initially unbiased memory reset with error epsilon retains entropy and changes the ideal work requirement; physical free-energy asymmetry would modify the expression. An ideal information-assisted work bound uses mutual information I in natural-log units under appropriate cyclic and feedback assumptions.\nThe observational target is work, heat and information change over a defined operation. The model deliberately isolates a tractable mechanism; its purpose is to reveal which measurements would distinguish that mechanism from an alternative. Parameter values in the figures are declared illustrative inputs. A curve produced by those values is a mathematical consequence, not an estimate of an unmeasured biological or physical quantity.", [] ], [ "A discriminating validation design", "The primary comparison should test energy accounting that closes for the memory, controller and bath. The unit of independent replication is independent control cycle with a specified initial distribution; repeated readings from the same unit improve precision but do not create independent replication. A strong validation would hold out different reset errors and protocol durations before choosing thresholds or fitting nuisance corrections. In particular, unaccounted controller states and finite-time dissipation must be measured, constrained or varied in a sensitivity analysis. Report both the principal estimate and how it changes when that assumption is relaxed. A model that succeeds only after choosing the most favorable exclusion rule has answered a different question from one evaluated with a locked rule.", [] ], [ "Worked numerical analysis", "At 300 kelvin, kBT ln(2) is approximately 2.871 times ten to the minus twenty-one joules using the exact Boltzmann constant. In the symmetric-memory model, allowing reset error 0.1 leaves binary entropy about 0.3251 in natural-log units, reducing the ideal work requirement to about 53.1% of the perfect-reset value. The figures show temperature scaling and the reliability tradeoff; they are not measurements of a computing device.\nThese calculations can be reconstructed from the companion Python file and its parameter table. Axis labels retain the model's units, and the caption distinguishes the quantities varied from those held fixed. The numerical examples were selected to expose a dependency, not to reproduce the appearance of a published figure. A useful independent check is to evaluate a limiting case and then recompute an intermediate point without the plotting code. Agreement validates the arithmetic within the declared model; it does not establish that nature follows that model.", [] ], [ "Evidence, transfer and next steps", "The scope of the conclusion is limited by unaccounted controller states and finite-time dissipation. Transfer to different reset errors and protocol durations requires evidence beyond fitting the original observations. A future replication should preserve raw measurements, calibration records, exclusions and a machine-readable link between each result and the corresponding analysis. The evidence table below identifies what each selected primary study contributes. The comparison is a focused narrative synthesis, not a systematic review or a claim that all subsequent evidence has been surveyed. The proposed study has not been performed here. Its value is that a negative result would constrain an explicit explanation rather than merely lower a benchmark score.", [] ] ], "equations": [ { "tex": "W_{erase}\\geq k_BT\\ln2", "explanation": "For an initially unbiased symmetric bit reset reliably in an isothermal idealization, average work has this lower bound." }, { "tex": "H_b(p)=-p\\ln p-(1-p)\\ln(1-p)", "explanation": "Binary entropy is expressed in natural-log units here; its limiting value is zero at p equal to zero or one." }, { "tex": "W_{min}\\geq k_BT[\\ln2-H_b(\\epsilon)]", "explanation": "A symmetric initially unbiased memory reset with error epsilon retains entropy and changes the ideal work requirement; physical free-energy asymmetry would modify the expression." }, { "tex": "W_{extract}\\leq k_BT I", "explanation": "An ideal information-assisted work bound uses mutual information I in natural-log units under appropriate cyclic and feedback assumptions." } ], "table": { "headers": [ "Evidence or design choice", "What it identifies", "What remains unresolved" ], "rows": [ [ "Source 1", "Landauer related logical irreversibility to thermodynamic cost.", "The bound depends on the physical operation and state distribution." ], [ "Source 2", "A controlled memory experiment tested the erasure-work limit.", "Finite-time implementation and experimental uncertainty remain relevant." ], [ "Source 3", "Feedback on a Brownian system converted information into free-energy gain.", "Controller memory and cycle closure are essential to complete accounting." ], [ "Primary observable", "work, heat and information change over a defined operation", "unaccounted controller states and finite-time dissipation" ], [ "Independent test", "energy accounting that closes for the memory, controller and bath", "Requires new measurements" ], [ "Transfer set", "different reset errors and protocol durations", "Not validated by the constructed numerical example" ] ] }, "protocol": [ "Specify work, heat and information change over a defined operation before fitting.", "Retain independent identifiers for each independent control cycle with a specified initial distribution.", "Separate different reset errors and protocol durations from model development.", "Measure or bound unaccounted controller states and finite-time dissipation.", "Record units, calibration and the observation operator.", "Compare the primary model with a named rival explanation.", "Prespecify the test: energy accounting that closes for the memory, controller and bath.", "Archive raw data, exclusions, parameter files and executable analysis." ], "limitations": [ "The bound depends on the physical operation and state distribution.", "Finite-time implementation and experimental uncertainty remain relevant.", "Controller memory and cycle closure are essential to complete accounting.", "Companion figures are constructed analytical examples, not new experimental results.", "Transfer to different reset errors and protocol durations has not been demonstrated by this review." ], "evidence": [ { "source": "ef-40-1", "result": "Landauer related logical irreversibility to thermodynamic cost.", "boundary": "The bound depends on the physical operation and state distribution.", "locator": "Abstract / selected primary-source sections; see linked bibliographic record." }, { "source": "ef-40-2", "result": "A controlled memory experiment tested the erasure-work limit.", "boundary": "Finite-time implementation and experimental uncertainty remain relevant.", "locator": "Abstract / selected primary-source sections; see linked bibliographic record." }, { "source": "ef-40-3", "result": "Feedback on a Brownian system converted information into free-energy gain.", "boundary": "Controller memory and cycle closure are essential to complete accounting.", "locator": "Abstract / selected primary-source sections; see linked bibliographic record." } ], "research_questions": [ "Which physical states must be included to close a repeated feedback cycle?", "How does finite-time control change the work-error tradeoff?", "When does an asymmetric memory invalidate the simple binary bound?" ], "diagram": [ "Logical distribution", "Physical memory and bath", "Measurement and control", "Work and entropy accounting" ], "keywords": [ "Information and engineering", "information-thermodynamics", "measurement", "identifiability" ], "kind": "ai_assisted_research_guide", "author": "Codex (AI)", "original_language": "en", "created_date": "2026-10-22", "published_date": "2026-10-22T13:02:56+00:00", "publication_status": "published_commissioned_review", "version": 5, "previous_version": "/guides/information-thermodynamics/v4/", "analytic": "information-thermodynamics", "peer_reviewed": false, "journal_publication": false, "disclosure": { "authorship": "Source-based research review and original mathematical analysis written by Codex (AI). Cited experiments belong to the cited research teams.", "figures": "Concept cover created with the built-in ImageGen tool. Quantitative figures are explicit constructed analyses, with code and numerical data; they are not measurements from the cited papers.", "review": "This local draft has no independent peer-review or journal-acceptance record." }, "revision_note": "First English local draft; upload and publication deferred at the owner's request.", "review_scope": "A focused narrative analysis of the named primary studies, not a systematic literature search or a claim to comprehensive coverage through 2026.", "section_kinds": [ "author_analysis_or_verification_proposal", "source_based_review", "author_analysis_or_verification_proposal", "author_analysis_or_verification_proposal", "author_analysis_or_verification_proposal", "author_analysis_or_verification_proposal", "author_analysis_or_verification_proposal", "author_analysis_or_verification_proposal", "author_analysis_or_verification_proposal", "author_analysis_or_verification_proposal", "author_analysis_or_verification_proposal", "author_analysis_or_verification_proposal", "author_analysis_or_verification_proposal", "author_analysis_or_verification_proposal" ], "source_slug": "information-thermodynamics", "source_revision": "41dc5ea7bf1db9868e2ee5b121f7021c823cc018", "source_article_sha256": "d7c3167f4f67ee8f323273d405905855184f06661965c09863a1e1acb05987cd", "sources": [ { "id": "ef-40-1", "doi": "10.1147/rd.53.0183", "url": "https://doi.org/10.1147/rd.53.0183", "title": "Irreversibility and Heat Generation in the Computing Process", "authors": [ { "given": "R.", "family": "Landauer" } ], "journal": "IBM Journal of Research and Development", "year": 1961, "article_slug": "information-thermodynamics", "verified_date": "2026-10-22", "verification": "DOI/title/authors/journal metadata checked against Crossref. Narrative claims were scoped to accessible primary publisher or author records, abstracts and selected text; full-text and supplementary-methods coverage is not exhaustive.", "source_article_slug": "information-thermodynamics" }, { "id": "ef-40-2", "doi": "10.1038/nature10872", "url": "https://doi.org/10.1038/nature10872", "title": "Experimental verification of Landauer’s principle linking information and thermodynamics", "authors": [ { "given": "Antoine", "family": "Bérut" }, { "given": "Artak", "family": "Arakelyan" }, { "given": "Artyom", "family": "Petrosyan" }, { "given": "Sergio", "family": "Ciliberto" }, { "given": "Raoul", "family": "Dillenschneider" }, { "given": "Eric", "family": "Lutz" } ], "journal": "Nature", "year": 2012, "article_slug": "information-thermodynamics", "verified_date": "2026-10-22", "verification": "DOI/title/authors/journal metadata checked against Crossref. Narrative claims were scoped to accessible primary publisher or author records, abstracts and selected text; full-text and supplementary-methods coverage is not exhaustive.", "source_article_slug": "information-thermodynamics" }, { "id": "ef-40-3", "doi": "10.1038/nphys1821", "url": "https://doi.org/10.1038/nphys1821", "title": "Experimental demonstration of information-to-energy conversion and validation of the generalized Jarzynski equality", "authors": [ { "given": "Shoichi", "family": "Toyabe" }, { "given": "Takahiro", "family": "Sagawa" }, { "given": "Masahito", "family": "Ueda" }, { "given": "Eiro", "family": "Muneyuki" }, { "given": "Masaki", "family": "Sano" } ], "journal": "Nature Physics", "year": 2010, "article_slug": "information-thermodynamics", "verified_date": "2026-10-22", "verification": "DOI/title/authors/journal metadata checked against Crossref. Narrative claims were scoped to accessible primary publisher or author records, abstracts and selected text; full-text and supplementary-methods coverage is not exhaustive.", "source_article_slug": "information-thermodynamics" } ], "original_source_url": "/guides/assets/v9-english-frontiers-articles-information-thermodynamics-json-d7c3167f4f67ee8f.json", "pdf": { "document_id": "157b1f5e-3b1e-53b6-8b61-77a2b826efc2", "version_id": "07feb5c4-7ae5-5f38-991c-fa990614eb4f", "media_type": "application/pdf", "authentication_required": true, "download_audited": true }, "presentation_change": "Complete version-bound translated reading views and selectable authenticated original/translated PDFs; original scientific text and evidence retained.", "source_locale": "en", "source_version_id": "ffa1a68a-83aa-5e83-aad2-d42fb4b82ca1", "translations": [ { "locale": "zh", "translation_id": "4e732c66-1bad-5af5-b944-ca1503bea8af", "translated_content_sha256": "7ce435a0926dc89e2c17c149cda1fb0d20a4bb7d8e931dd8772071b5b07d1e7a", "sha256": "67d60ebeed43ce167f9c2289e35416aa890d24b66eb7f3277fa394036dc38002", "byte_length": 854154 }, { "locale": "es", "translation_id": "6bdb5273-effa-503e-8dd0-d281ff394e83", "translated_content_sha256": "32930c665f7446f8abc0b7e328fee9bcc33a7a9a8f75ee81d5c95a7df0de3ef4", "sha256": "d5db0cabf635f983fb14b26778c8199ac3f2751858f9ee3c5286696a7bf84019", "byte_length": 627446 }, { "locale": "ja", "translation_id": "bbabe3e9-18ea-5d62-89c3-b4166e823cde", "translated_content_sha256": "9538edf307d2a61674a6268685721155156113c2a4c8f8d26a018b3b715d467f", "sha256": "430228f28890346ef28c490ea14bdf19f2c911f26a8d3e0ed2d57e58f65bfd4b", "byte_length": 814737 }, { "locale": "ar", "translation_id": "08847039-d747-5581-9dc3-9347f3063de0", "translated_content_sha256": "f8b32dded33ebaacf2831bdfaf4687fe2c3748f74a986c1707fa1204d5be3714", "sha256": "975825a5c4ebe989984a6f042995c50b911c629d6afb9f02e0d62aaba4e9bd78", "byte_length": 611114 } ] }