{"id":"963deb62-dd2a-4a5f-b23b-698d438673ce","arxiv_id":"2607.23902","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A physical system computes autonomously only when an internal readout state selects the next operation; by that criterion the walking-droplet system is a wave-memory machine, not a closed autonomous computer.","lead":"This paper proposes a formal test for when a physical system with memory is actually computing on its own, rather than being interpreted or controlled from outside. Applying this test to the 'walking droplet' experiment, it concludes the droplet stores, reads, and erases its own wave memory, but is not an autonomous computer because the erase signal is externally imposed.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Closure criterion under-specifies O and X_Y: with internal σ as readout and F_± as operations, the walker satisfies Eq. (30), so the 'externally imposed π-shift' classification rests on an unformalized choice of operation set.","rationale":"The paper is a coherent conceptual framework, and the formal statements from Sec. V onward are internally consistent. The reader's CONDITIONAL verdict is appropriate. My concern sharpens the reader's weakest-assumption point: the problem is not merely that 'autonomy' is stipulated; it is that the stipulated condition depends on unformalized choices of operation set O and readout subsystem X_Y. The phase-state counterexample shows that Eq. (30) alone does not separate autonomous operation selection from ordinary state-dependent dynamics. This does not refute the framework's usefulness, but it means the walker classification is not established by the equation alone. I do not recommend changing the verdict because the paper already acknowledges the criterion is minimal, and the concern can be addressed by adding a causal non-vacuity condition in revision. The concrete test would settle whether the bug is real by exhibiting the admissible closure decomposition or forcing the needed axiom.","tokens_in":47057,"tokens_out":8725,"duration_ms":103159,"concrete_test":"Analytic check: instantiate Eq. (30) using the paper's own stroboscopic model with X_Y={σ_n}, O={F_+,F_-}, and C(σ_n)=F_{σ_n}. Direct substitution of Eqs. (1)-(5) shows that after the external phase flip, the closure equation holds at every subsequent step and |Im(C)|=2. If this holds, the Sec. IX claim that the walker is not closed fails under an admissible reading of Criterion 1. Then attempt to formulate the minimal extra axiom that excludes this counterexample while preserving the intended distinction; if no such axiom can be stated in physical terms, Criterion 1 is under-specified and the walker classification should be revised.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The load-bearing step is not the coarse-graining commutativity (Eq. 19) or the robustness conditions, but the closure criterion itself: Eq. (30) uses objects O, X_Y, C that are stipulated without formal independence or non-vacuity conditions. The paper's own stroboscopic model contains an internal phase state σ_n. After the one-time external π-shift, σ_n remains part of X. Define O = {F_+, F_-}, where F_σ is the stroboscopic map of Eqs. (1)-(5) with fixed bounce phase σ, and define C(σ_n)=F_{σ_n}. Then for every step after the flip, x_{n+1}=F_{C(σ_n)}(x_n), with |Im(C)|=2, satisfying Eq. (30) and the nontriviality requirement of A5.7. The paper's contrary classification in Sec. IX depends on refusing to count F_± as operations and on excluding σ_n from X_Y. That refusal is not derived from any formal condition in Criterion 1. Equivalently, extending X to include the forcing apparatus and controller makes the 'external' π-shift internal, again satisfying closure. Thus the central result is boundary-relative and operation-set-relative as stated. The paper needs a causal-independence axiom: e.g., the actuator executing C(y_n) must be a physical channel distinct from the variable that reports y_n, and the selected operation must not be a mere re-labeling of the readout variable's own dynamics. Without such an axiom, 'closed autonomous' is not a physical invariant but a representational choice.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops a conceptual hierarchy for autonomous physical computation: physical memory, transition-preserving computation via a commuting coarse-graining Π∘F=G∘Π, robust symbolization via separated basins and noise tolerance, and closed autonomous computation via an internal readout state selecting the next physical operation, x_{n+1}=F_{C(y_n)}(x_n). The framework is applied to the Perrard–Fort–Couder walking-droplet system, which is modeled as a stroboscopic reservoir with wave-field memory. The paper concludes that the walker is a wave-memory physical machine with writing, storage, reading, feedback, finite-time reversal, and externally triggered erasure, but not a closed autonomous physical computer, because the erasing π-shift is externally imposed. A categorical reformulation and several constructive designs for closing the loop (autonomous eraser, branch, coupled walkers) are also presented.","tokens_in":47447,"tokens_out":9576,"duration_ms":106462,"significance":"The paper addresses a genuine foundational question: when does physical memory become autonomous computation? Its main strengths are the explicit formalization of computation as quotient-functorial transition preservation, the clean separation of memory, transition preservation, robustness, and closure, and the self-critical application to the walker. Proposition 1 is elementary and correct, and the classification of the walker follows from the stipulated definitions rather than from fitted parameters or circular empirical claims. The paper also makes a useful design point: physical readout states must be coupled back into operation selection. However, the central closure criterion is stipulated, and, as discussed below, it is under-specified in a way that makes the walker classification boundary-relative. If the authors add a formal causal-independence condition, the framework would be substantially more convincing.","major_comments":[{"comment":"The closure criterion is under-specified. No formal restrictions are imposed on the operation set O, the readout subsystem X_Y, or the decomposition X=X_R×X_Y×X_Z. After the one-time external π-shift in the stroboscopic model, take X_Y={+1,−1} as the σ-coordinate, O={F_+,F_-} with F_σ the map (1)–(5) with the phase held at σ, and C(σ_n)=F_{σ_n}. Then x_{n+1}=F_{C(σ_n)}(x_n) and |Im(C)|=2, so Eq. (30) and the A5.7 non-vacuity condition hold at every subsequent step. This contradicts the Sec. IX assertion that no internal readout state satisfies C(y_n)=o_π. The contradiction is resolved only by the unformalized choice to exclude F_± from O and σ_n from X_Y. The paper needs an explicit causal-independence axiom: the actuator executing C(y_n) must be a distinct physical channel from the variable reporting y_n, and C(y_n) must not be a mere re-labeling of the readout's own update law. Without","section":"§VII, Eq. (30); §A5.7; §IX"},{"comment":"The positive classification 'wave-memory physical machine with genuine Turing-like primitives' is not tied to any formal notion of 'Turing-like'. The paper lists strict Turing-machine requirements (finite alphabet, internal states, transition table, robust encoding) in §XI.B and A7.12, and admits the walker lacks them, yet the classification table credits the walker with 'Turing-like primitives'. Since this phrase appears in the main result, it should either be defined precisely (e.g., a finite set of operations that can compose, under a stated encoding, to Turing-complete transitions) or replaced by neutral language such as 'read/write/store/erase primitives'. As written, the positive claim is stronger than the formalism supports.","section":"§IX; §XI.B; Table A7.13"},{"comment":"The constructive closed-loop architectures are presented as the main payoff of the closure criterion, but no quantitative demonstration is given that they satisfy the paper's own robust symbolization conditions: basin separation (22), transition stability (24)–(25), and operation-selection reliability (A36). For example, the autonomous eraser uses hysteresis thresholds Θ_E and Δ, but no noise scale, barrier height, or coupling strength is estimated; the claim that this device 'would test the closure criterion directly' is therefore a design sketch rather than a supported prediction. The authors should either label these explicitly as open modeling proposals or provide elementary feasibility estimates.","section":"§X, Eqs. (50)–(54)"}],"minor_comments":[{"comment":"The sentence 'categorical treatments of information such as ?' contains a missing citation/reference placeholder. This should be completed before publication.","section":"§VIII"},{"comment":"The classification table in the appendix has no caption and is not referenced from the main text; the reader encounters it only after the prose classification. Add a numbered caption and a cross-reference.","section":"§A7.13"},{"comment":"The distinction between the loose operational sense and the strict computability-theoretic sense of 'Turing machine' is useful, but the loose sense is never formalized. The term 'Turing-like primitives' should be defined in the same section where this distinction is drawn.","section":"§XI.B"},{"comment":"The symbol U is used both for a domain U⊆X and for an abstract operation set U. This creates avoidable confusion in the appendix; rename one of them.","section":"§A3.8 and §A5.4"},{"comment":"The paper is extremely long, and the appendices repeat much of the main text nearly verbatim. Condensing the appendices to definitions, proofs, and tables not already in the main text would improve readability without changing content.","section":"General"}],"recommendation":"major_revision","confidential_remarks":"The paper is a conceptual/foundational contribution rather than an experimental or computational study. The main central claim is defensible under a stipulated definition, but the under-specification of the closure criterion is sufficiently load-bearing that the current version should not appear without revision. The author should either add a formal causal-independence/non-vacuity condition to Criterion 1 or explicitly weaken the classification to 'with respect to a stipulated operation/readout decomposition'. The self-citations are relevant and not excessive. I do not see grounds for rejection: Proposition 1 is correct, the framework is clearly presented, and the proposed designs are testable in principle. With the suggested formalization, the paper could be a useful contribution to the physical-computation literature."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"This paper gives the reservoir-computing and neuromorphic communities a cleaner way to ask when memory plus feedback becomes autonomous computation. The stroboscopic separation of the walker into write, store, read, feedback, and externally triggered erase operations is genuinely helpful, and the coarse-grained transition-preservation criterion with its robust basin conditions is a reasonable, non-vacuous way to define physical computation. The quotient-functor formulation is correct as far as it goes, and the walker classification as a wave-memory machine with real primitives but not a closed autonomous computer is appropriately cautious, especially given the paper's own pushback against loose 'Turing machine' talk.\n\nThe soft spot is the closure criterion itself. Eq. (30) and Criterion 1 define O and X_Y only loosely. The stress-test worry is fair: after the one-time external pi-shift, you can take sigma_n as the readout and F_+ / F_- as operations, and then the walker satisfies Eq. (30) with |Im(C)| = 2. The paper's contrary classification depends on refusing to count F_+/- as operations and on excluding sigma_n from X_Y, but nothing in Criterion 1 or Appendix A5.7 licenses that refusal. Extending X to include the forcing apparatus makes the 'external' pi-shift internal in the same trivial way. The central result is therefore boundary-relative and operation-set-relative as stated. The fix is not deep: add an axiom that the actuator executing C(y_n) be a physical channel distinct from the variable that reports y_n, and that the selected operation not be a mere re-labeling of the readout variable's own dynamics. Without that, the criterion is too permissive.\n\nA secondary issue is the phrase 'Turing-like primitives.' Writing, storage, reading, feedback, and erasure are real physical primitives, but they are not tied to any formal Turing-machine operations, and the paper overstates what is demonstrated. The robust-basin conditions are stated but not instantiated for the walker or any proposed extension; the constructive architectures are design principles, not experiments. That is fine for a conceptual paper, but it should be labeled as such.\n\nWho gets value from this: philosophers and theorists working on physical computation, and experimentalists who want a sharper language for when a reservoir is computing on its own rather than being read out externally. It deserves a serious referee, but it needs revision before publication: tighten the closure definition with a causal-independence condition, tone down 'Turing-like,' and either instantiate the basin conditions on a concrete example or explicitly leave that as future work.","headline":"Useful conceptual framework with a real soft spot: the closure criterion as written is too permissive to support the walker classification without an added causal-distinctness axiom.","tokens_in":47891,"tokens_out":1960,"would_cite":true,"duration_ms":23176,"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":"The paper claims that autonomous physical computation requires an internal physical readout state to select each next operation, and that the wave-particle walker, despite its memory and feedback, fails this closure test.","keywords":["autonomous computation","closure criterion","physical computation","reservoir computing","wave-memory","walking droplet","coarse-graining","neuromorphic"],"falsifier":"The decisive test is the paper's own 'autonomous eraser': couple a physical detector of wave energy to the forcing phase so the system triggers its own π-shift. If this modified walker is still classified as not closed, or the original externally triggered walker is classified as closed, the necessity of the closure criterion is refuted.","tokens_in":46929,"feed_emoji":"🌊","tokens_out":5656,"duration_ms":53036,"temperature":0.7,"pith_summary":"The paper establishes a sharp boundary between physical memory, transition-preserving computation, and autonomous physical computation. It argues that memory and feedback are not enough; a system computes autonomously only when a physical readout state inside the system selects the next operation. Applying this to the walking-droplet system, the paper shows that the walker genuinely writes, stores, reads, and erases its own wave memory, but its erasing phase shift is externally imposed, so it does not qualify as a closed autonomous physical computer. The criterion turns this classification into a design principle for building closed physical computers from wave-memory reservoirs.","feed_headline":"Autonomy in physical computation requires internal readout selection","feed_subtitle":"Wave-memory droplet stores, reads, and erases its past, but fails closure because its erasing flip is externally imposed.","key_machinery":"The load-bearing object is the closure criterion for autonomous physical computation: the closed-loop map Φ = Act ∘ ⟨C ∘ p_Y, id_X⟩, which internalizes morphism selection so that the next physical process is chosen by the system's own readout state rather than by an external agent. This is coupled with a coarse-grained transition-preservation condition (Π ∘ F = G ∘ Π) that distinguishes genuine computation from arbitrary observer labeling.","core_discovery":"The paper's central claim is that a physical system computes autonomously exactly when its dynamics takes the form x_{n+1} = F_{C(y_n)}(x_n), where y_n is an internal physical readout state and C maps that state to a physical operation. Under this closure criterion, the wave-particle walker is a wave-memory machine with genuine Turing-like primitives—writing, storage, reading, feedback, and erasure—but not a closed autonomous physical computer, because the phase shift that produces erasure is imposed externally rather than selected by an internal readout.","pith_inferences":["If the closure criterion is accepted, many neuromorphic and reservoir-computing devices described as 'computing' are technically externally read dynamical systems; their computational status depends on whether readouts are internalized.","The criterion suggests an operational benchmark for cortical traveling waves: they contribute to computation only if read out by downstream circuits that in turn select subsequent wave-generating operations.","The paper's distinction implies a three-way experimental comparison—external phase schedule, observer readout without feedback, and physical readout with feedback—as a practical test for autonomy in any proposed physical computer.","One could extend the criterion to define a hierarchy of autonomous physical computation in which the readout basin itself is learned or tuned, connecting closure with trainability and adaptation."],"forward_implications":["The wave-particle walker is classified as a wave-memory physical machine with Turing-like primitives, not a closed autonomous Turing machine.","Standard reservoir computing, in which a readout is trained externally and not fed back into the reservoir, does not satisfy closure unless the readout is physically internalized.","Physical memory and feedback do not by themselves imply computation; robust coarse-grained transition preservation is a separate requirement that must be demonstrated.","Closure is independent of computational power: a closed system can compute nothing of interest, and nontrivial computation requires additional structure such as a finite alphabet and programmable transitions.","The framework yields concrete designs for closing the wave-memory loop: multistable readout traps, thresholded wave detectors, boundary-controlled Faraday baths, and coupled-walker controllers."],"fun_headline_variants":["Physical autonomy: readout must select the next operation","Wave-particle walker: Turing primitives but not autonomous","Closure criterion: internal readout is key to autonomous computing","Droplet machine lacks closure: erasing flip is externally imposed","Autonomy in physical computation demands internal readout coupling"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The definition of autonomy as internal operation selection is a stipulated criterion, not a derived theorem; if autonomy is instead defined as self-organization or homeostatic closure, the walker's classification could change.","fun_headline_variants_meta":{"raw":{"variants":["Physical autonomy: readout must select the next operation","Wave-particle walker: Turing primitives but not autonomous","Closure criterion: internal readout is key to autonomous computing","Droplet machine lacks closure: erasing flip is externally imposed","Autonomy in physical computation demands internal readout coupling"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000193,"raw_usage":{"total_tokens":1176,"prompt_tokens":726,"completion_tokens":450,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":470,"completion_tokens_details":{"reasoning_tokens":368}},"tokens_in":470,"tokens_out":450,"duration_ms":4504,"temperature":1.0,"reasoning_tokens":368,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-31T23:33:24.434473+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"The decisive test is the paper's own 'autonomous eraser': couple a physical detector of wave energy to the forcing phase so the system triggers its own π-shift. If this modified walker is still classified as not closed, or the original externally triggered walker is classified as closed, the necessity of the closure criterion is refuted.","supporting_citations":[],"review_version":1}