{"id":"2f6bdbbf-bf15-4500-ab6c-c497d2033bf2","arxiv_id":"2501.00547","paper_version":2,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"The Shannon entropy of a Bayesian observer tracking a chaotic system on a Cayley tree grows linearly with reduced rate |v| below a critical measurement strength, saturates above it, and shows universal sqrt(t) growth exactly at the transition.","lead":"A new solvable model shows that an observer tracking a chaotic system through measurements faces a phase transition: below a critical measurement strength, uncertainty keeps growing, while above it, uncertainty stays bounded. The transition point matches the directed polymer freezing transition, and the entropy growth is described by an exact universal scaling function confirmed by numerics.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The transition and scaling function are derived correctly for the tree model, but the load-bearing premise that a chaotic system's uncertainty growth is a Cayley tree with independent per-site measurements remains untested against real chaotic dynamics.","rationale":"The reader's weakest assumption is the tree representation of the chaotic uncertainty growth, and I agree that this is the load-bearing premise. The derivation of Eq. (12) and the scaling function (13)-(14) is internally sound: the identity (9) is exact, the one-point term in Appendix A is correct, and the transition point v=0 coincides with the DP freezing condition by the calculation in Appendix I.2. The only potentially uncontrolled step is replacing the nonlinearity by a wall in the critical scaling (Appendix F); however, the authors verify the resulting scaling function against the full nonlinear equation (8) and the full PDE (11) in Fig. 3, and the inset shows the difference decreasing with time. I therefore do not find an internal inconsistency that would change the verdict. The residual concern is external validity: a real chaotic system is not a Cayley tree, and the paper's own final paragraph concedes the tree is only qualitative for d>2. This does not invalidate the solvable model, but it means the title's 'state estimation of chaotic systems' should be understood as a minimal-model statement. The proposed numerical test on a concrete chaotic map would settle whether the predicted transition survives beyond the tree geometry. Since the paper is explicit about this limitation, the reader's ACCEPT remains appropriate.","tokens_in":19840,"tokens_out":26495,"duration_ms":255038,"concrete_test":"Run a Bayesian state-estimation simulation on a concrete chaotic system (e.g., the Lorenz-63 flow or the tent map) with a fine partition and Gaussian measurement noise. Compute the Shannon entropy of the posterior over long times for several noise levels, extract the entropy growth rate s and the transition point, and compare with Eq. (12), using the system's finite-time Lyapunov exponent lambda and the KL divergence of the measurement model. Also test the critical scaling <S_t> approximately (sigma^2/v) S(v sqrt(t/(2 sigma^2))) at the predicted v=0. If the transition point or rate differs, the Cayley-tree representation omits essential correlations; if it matches, the model is validated as a minimal description.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The mathematical argument for Eqs. (6), (12)-(14) is internally consistent: the mapping to the DP via Eq. (4), the exact identity Eq. (9), and the KPP analysis of Eq. (8) all hold for the tree model, and the numerics (Figs. 2-3) support the critical scaling. The load-bearing premise is the modeling step in Fig. 1: representing the growth of uncertainty in a chaotic system as a Cayley tree with branching ratio K=e^{lambda*Delta t} and treating measurements at each site as independent with distributions P1/P0. Everything downstream depends on this geometry and on the absence of correlations between branches. Real chaotic systems have correlated nearby trajectories, a folded phase-space structure, and measurement outcomes that are not independent across phase-space cells; the tree is only a qualitative high-dimensional caricature, as the authors note for d>2. If the true geometry differs, the transition point v=0 and the rate s=|v| may not describe actual Bayesian state estimation. This is a scope limitation rather than an internal error, but it is the single most load-bearing assumption for the title's claim about chaotic systems.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces a solvable model of Bayesian state estimation for a chaotic system, represented as a directed random walk on a Cayley tree with branching ratio K=e^{λΔt}. At each time step, the observer measures every site independently with outcome distributions P1 (occupied) and P0 (empty). The authors map the average over the true trajectory and measurement outcomes to a P0-average reweighted by the directed polymer (DP) partition function, and then study the Shannon entropy of the observer's posterior distribution. They derive an exact growth-rate transition: s = lim_{t→∞} ⟨S_t⟩/t equals |v| for v≤0 and 0 for v>0, where v=D_KL(P1||P0)/Δt−λ. Near the transition they obtain a universal scaling function S(η) for the subleading growth, Eq. (14). Numerical solution of the exact recursion Eq. (8) and Monte Carlo simulations of the particle and posterior support the predictions. The paper presents the results as a toy model and discusses its connection to the freezing transition of the directed polymer.","tokens_in":20091,"tokens_out":11421,"duration_ms":121878,"significance":"If accepted, this is a valuable exactly solvable example of a classical measurement-induced phase transition, with a clean mapping to the directed polymer on a Cayley tree in the rare-event regime. The derivation is careful and largely self-contained: Eq. (4) is an exact reweighting identity, Eq. (9) is proved in Appendix B, and the scaling function Eqs. (13)–(14) is derived without adjustable parameters and checked against numerics. The paper also establishes that the transition location coincides with the DP freezing point while the critical properties differ, which is a physically interesting distinction. The main limitation, acknowledged by the authors in the Conclusions, is that the Cayley-tree representation with independent per-site measurements is a modeling assumption rather than a derived property of generic chaotic systems; the authors describe it as qualitative for sufficiently high dimension. I regard this as a scope limitation, not an internal inconsistency: the mathematical claims are claims about the tree model, and the broader 'chaotic systems' title should be read through that lens.","major_comments":[],"minor_comments":[{"comment":"The first sentence of the abstract and the title state the result for 'chaotic systems', but all proofs concern the Cayley-tree model with independent per-site measurements. Since the Conclusions already call the model a toy model and note that the tree is only a qualitative description for d>2, I recommend making this scope explicit in the abstract as well, e.g. by saying 'for a tree model of a chaotic system'.","section":"Abstract and Conclusions"},{"comment":"The citation 'following [65]' at the start of Appendix I appears to be incorrect: if the numbering follows the main text, [65] is a reference on classical many-body information dynamics, not the Derrida–Spohn traveling-wave analysis. The intended reference is likely [68] or the footnote [72] of the main text.","section":"Appendix I"},{"comment":"The caption writes 'D(P1||P0)=Var(P1||P0)/2' without defining D; the main text uses D_KL(P1||P0). Please use a consistent notation in the caption.","section":"Figure 2 caption"},{"comment":"For v<0, the truncation protocol in the Monte Carlo simulations causes the entropy to saturate at a value proportional to the cutoff τ*, as shown in Fig. S1. The main text's comparison of ⟨S_t⟩ to the linear growth |v|t implicitly relies on subtracting this truncation artifact; it would be helpful to state this explicitly in the main text where the numerical agreement for v<0 is discussed.","section":"Supplemental Fig. S1 and text after Eq. (12)"}],"recommendation":"minor_revision","confidential_remarks":"This is a solid theoretical Letter in scope for cond-mat.stat-mech. The central derivation is sound for the tree model, and the only substantive caveat is the modeling premise for real chaotic systems, which the authors already acknowledge. I recommend minor revision rather than major revision because the requested changes are local and concern scope wording and small presentation issues; no load-bearing mathematical step needs reworking."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should read this one, and if you edit, send it out. For the tree model it delivers a genuinely new result: the observer's entropy rate is exactly the negative part of the control parameter v, and the critical regime has an exact universal scaling function that matches both the functional-equation solution and Monte Carlo. The derivation is careful and largely self-contained. Eq. (4) is a real identity, Eq. (9) is proven in Appendix B rather than waved through, and the wall approximation used for the critical regime is transparent. I also like that they keep the n→1 object (Z ln Z) distinct from the n→0 free-energy front; that distinction is easy to botch and they handle it cleanly. The numerics are honest, and the agreement between the two simulation protocols and the recursion solution is convincing.\n\nThe soft spot is not the math but the opening premise. The Cayley tree with independent per-site measurements is a caricature of a chaotic system. Real chaotic dynamics have correlated nearby trajectories, folded phase space, and measurement outcomes that are not independent across cells. The paper acknowledges this, saying the tree is only a qualitative description for sufficiently high dimension, but that caveat sits in the conclusion while the title and abstract claim the transition for chaotic systems. This is a scope limitation rather than an internal error: everything downstream of the tree assumption is derived correctly and tested against the model. The Monte Carlo truncation in the v<0 phase limits direct verification there, but the exact recursion fills the gap, so I would not call that a load-bearing flaw.\n\nOn the circularity worry: yes, v is defined with the subtracted Lyapunov exponent, so the transition location is built into the observable. But the entropy rate coefficient, the saturation, and the universal scaling function are derived and then reproduced by numerics without fitting. That is not circular in any damaging sense.\n\nWho should read it: the monitored-dynamics community, people working on classical state estimation, and directed polymer theorists. It is a clean example of a classical MIPT without post-selection, and the connection to DP freezing is likely to be cited. It deserves a serious referee. My recommendation is to engage with it, with the main request being a more disciplined statement about what the tree model does and does not say about actual chaotic systems.","headline":"A clean, mostly self-contained mapping of Bayesian state estimation on a tree to directed polymer freezing, with exact scaling; the tree premise for real chaos is the only genuinely soft spot.","tokens_in":20570,"tokens_out":1903,"would_cite":true,"duration_ms":20286,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["82B26","82B44","37D45"],"pacs":["05.40.-a","05.45.-a"],"model":"deepseek-v4-flash","headline":"This paper claims that the Shannon entropy of a Bayesian observer tracking a chaotic system grows at a rate equal to the negative part of a single control parameter, with a universal square-root-time scaling at the transition.","keywords":["measurement-induced phase transition","state estimation","chaotic systems","Bayesian inference","directed polymer","Cayley tree","Shannon entropy","Lyapunov exponent"],"falsifier":"Simulate the same Bayesian estimation protocol on a deterministic chaotic system that is not a tree, such as a one- or two-dimensional coupled map lattice, and measure the Shannon entropy of the filter's posterior as a function of time for measurement strengths above the threshold $v>0$. The paper's claim predicts a vanishing entropy growth rate and, at $v=0$, the universal $\\sqrt{t}$ growth; observing linear entropy growth for all measurement strengths, or a critical scaling that depends on the details of $P_0$ and $P_1$, would show the transition is an artefact of the tree geometry.","tokens_in":19617,"feed_emoji":"🎯","tokens_out":12202,"duration_ms":108098,"temperature":0.7,"pith_summary":"The paper introduces a solvable model of a measurement-induced phase transition (MIPT) in a classical chaotic system, with no quantum mechanics and no postselection. In the model, the chaotic trajectory is a directed walk on a Cayley tree with branching ratio $K=e^{\\lambda \\Delta t}$, and at each time step the observer performs independent noisy measurements of every site, updating a Bayesian posterior distribution. The paper claims that the long-time growth of the Shannon entropy of this posterior is governed by a single control parameter $v = D_{\\mathrm{KL}}(P_1\\|P_0)/\\Delta t - \\lambda$: the growth rate is $|v|$ for $v \\le 0$ and $0$ for $v > 0$. Exactly at the threshold the entropy grows as $\\sqrt{t}$, with a universal scaling function that the authors derive in closed form and verify numerically. Because the observer's measurements do not perturb the classical state, the predicted transition is in principle directly observable, and it pinpoints the same critical location as directed-polymer freezing while having different critical properties.","feed_headline":"Bayesian observer tames chaos only past a sharp measurement threshold","feed_subtitle":"The uncertainty growth rate drops to zero exactly when measurement precision beats the Lyapunov exponent.","key_machinery":"The central object is the mapping from Bayesian filtering on the tree to the directed polymer (DP) partition function. Unnormalized weights $z_j^{(\\tau)}$ update multiplicatively, $z_j^{(\\tau+1)} = B(a_j^{(\\tau+1)})\\, z_{\\lceil j/K\\rceil}^{(\\tau)}$ with $B(a)=P_1(a)/(K P_0(a))$, so the normalization $Z^{(\\tau)}$ is exactly the partition function of a DP on a Cayley tree. The observer's average Shannon entropy is then the difference of two expectations expressible through $\\langle Z^{(\\tau)} \\ln Z^{(\\tau)}\\rangle_0$, and the self-similarity of the tree gives the closed recursion $G_{\\tau+1}(y) = \\langle G_\\tau(y - \\ln B(a))^K\\rangle_0$ for the Laplace transform $G_\\tau(y) = \\langle \\exp(-e^{-y} Z^{(\\tau)})\\rangle_0$. In the continuum limit this recursion becomes a KPP (Kolmogorov–Petrovsky–Piskunov) reaction-diffusion equation for the traveling wave, and the entropy-facing object $u_t(y) = e^y(1-G_\\tau(y))$ obeys a drifted-diffusion equation whose drift is $v$; at criticality the nonlinear term acts as a reflecting wall at the origin, and the resulting diffusion-process calculation yields the universal scaling function $S(\\eta)$. This machinery is what carries the exact rate (12) and the scaling form (13)–(14).","core_discovery":"On the paper's own terms, the central result is the exact entropy growth rate of Eq. (12): $s = \\lim_{t\\to\\infty}\\langle S_t\\rangle / t = |v|$ for $v \\le 0$ and $s=0$ for $v>0$, with $v = D_{\\mathrm{KL}}(P_1\\|P_0)/\\Delta t - \\lambda$ measuring the balance between the information content of each measurement and the chaotic spreading of trajectories. In the chaotic or weak-measurement phase the observer's uncertainty still grows exponentially but with a reduced effective Lyapunov exponent $|v|$; in the strong-measurement phase the estimated distribution localizes onto an $\\mathcal{O}(1)$ number of sites and the entropy saturates. In the critical window the paper provides the exact universal scaling form $\\langle S_t\\rangle \\simeq (\\sigma^2/v)\\, S(v\\sqrt{t/(2\\sigma^2)})$ with $S(\\eta) = (\\tfrac12+\\eta^2)\\operatorname{erf}\\eta - \\eta^2 + (\\eta/\\sqrt{\\pi}) e^{-\\eta^2}$, which at $v=0$ reduces to $\\langle S_t\\rangle \\simeq \\sigma\\sqrt{2t/\\pi}$. The transition point coincides with the freezing transition of the directed polymer on the Cayley tree, but the MIPT is dominated by rare polymer configurations and its critical properties differ, as the numerical solutions of the exact recursion confirm.","pith_inferences":["An extension the paper leaves implicit: if the same Bayes-reweighting logic holds on other geometries, the coincidence with directed-polymer freezing suggests that in high-dimensional coupled chaotic systems the filtering transition should sit at the same condition $D_{\\mathrm{KL}}/\\Delta t = \\lambda$, with the tree appearing as the effective mean-field geometry.","The paper's scaling function gives a ready-made finite-time crossover prediction: for $v$ near zero, the entropy should deviate from its asymptotic linear or constant behavior on the time scale $t_v \\sim |v|^{-2}$, which is directly measurable in particle-filter or Kalman-type estimators before their exponential cost becomes prohibitive.","A practical design implication left implicit by the authors is that only $D_{\\mathrm{KL}}(P_1\\|P_0)$ per measurement matters for the phase boundary; redesigning observables to increase this divergence at fixed measurement cost is equivalent to tuning the Lyapunov exponent downward, so smart observable selection and stronger chaos are on the same footing.","In the quantum analogue, Born's rule replaces Bayes' rule and the measurement back-action cannot be ignored; this framework suggests that a similar control parameter combining Lyapunov growth with measurement information would set the entanglement phase boundary, a connection the paper does not pursue."],"forward_implications":["A Bayesian observer of a tree-like chaotic system gains a sharp threshold: for measurement strengths below $v=0$ (i.e., $D_{\\mathrm{KL}}/\\Delta t < \\lambda$) the estimated distribution continues to spread exponentially, while above the threshold it localizes and the entropy rate is exactly zero.","In the weak-measurement phase the entropy rate is exactly $\\lambda - D_{\\mathrm{KL}}/\\Delta t$, so each unit of Kullback–Leibler information per unit time reduces the effective Lyapunov exponent linearly until it vanishes.","At the critical point the entropy grows like $\\sigma \\sqrt{2t/\\pi}$, and the full crossover from linear growth to saturation in time is described by the single universal function $S(\\eta)$, with the same curve in discrete- and continuous-time versions of the model.","The transition location coincides with the directed-polymer freezing transition on the tree, showing that a classical state-estimation transition and a known disordered-system transition can share a critical point while belonging to different universality classes.","Because the measurements are classical, no postselection is required, making this MIPT prediction testable without the exponential overhead that complicates quantum MIPT experiments."],"supporting_citations":[{"why":"Supplies the self-similarity recursion and KPP traveling-wave analysis for directed polymers on the Cayley tree that underpin Eq. (8).","marker":"[68]"},{"why":"Establishes a two-phase picture for the same tree setup when the exact trajectory is known, which this paper extends to the observer-only Bayesian protocol.","marker":"[67]"},{"why":"Provides the one-dimensional continuous-measurement walker result that this tree model contrasts with, where no such transition is expected.","marker":"[66]"},{"why":"Supplies the rare-event regime needed because Bayes-updated measurement outcomes are correlated, making Eq. (4) a rare-event average.","marker":"[69]"},{"why":"Provides the directed-polymer background used to argue the tree gives a qualitative description for sufficiently high spatial dimension.","marker":"[73]"},{"why":"Supports the high-temperature self-averaging properties of directed polymers that justify the tree description in high dimension.","marker":"[74]"}],"fun_headline_variants":["Sharp measurement threshold halts chaos-driven entropy growth","Chaos estimation exhibits phase transition at critical measurement rate","Directed polymer freezing pinpoints chaos measurement transition","Strong measurements bound uncertainty in chaotic state estimation","Measurement-induced transition curbs chaos entropy growth"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that a chaotic system's uncertainty spread is exactly a tree with branches multiplying like $e^{\\lambda \\Delta t}$, and that the observer independently measures every branch tip at every step; the paper itself notes the tree is only a qualitative model for real chaotic systems at sufficiently high spatial dimension.","fun_headline_variants_meta":{"raw":{"variants":["Sharp measurement threshold halts chaos-driven entropy growth","Chaos estimation exhibits phase transition at critical measurement rate","Directed polymer freezing pinpoints chaos measurement transition","Strong measurements bound uncertainty in chaotic state estimation","Measurement-induced transition curbs chaos entropy growth"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000662,"raw_usage":{"total_tokens":3062,"prompt_tokens":1017,"completion_tokens":2045,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":633,"completion_tokens_details":{"reasoning_tokens":1976}},"tokens_in":633,"tokens_out":2045,"duration_ms":16821,"temperature":1.0,"reasoning_tokens":1976,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T22:48:23.822005+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Simulate the same Bayesian estimation protocol on a deterministic chaotic system that is not a tree, such as a one- or two-dimensional coupled map lattice, and measure the Shannon entropy of the filter's posterior as a function of time for measurement strengths above the threshold $v>0$. The paper's claim predicts a vanishing entropy growth rate and, at $v=0$, the universal $\\sqrt{t}$ growth; observing linear entropy growth for all measurement strengths, or a critical scaling that depends on the details of $P_0$ and $P_1$, would show the transition is an artefact of the tree geometry.","supporting_citations":[{"cited_title":"The planted directed polymer: inferring a random walk from noisy images","cited_arxiv_id":"2404.07263","evidence_quote":"Supplies the self-similarity recursion and KPP traveling-wave analysis for directed polymers on the Cayley tree that underpin Eq. (8)."},{"cited_title":"Jin and D","cited_arxiv_id":null,"evidence_quote":"Establishes a two-phase picture for the same tree setup when the exact trajectory is known, which this paper extends to the observer-only Bayesian protocol."},{"cited_title":"Pizzi, D","cited_arxiv_id":null,"evidence_quote":"Provides the one-dimensional continuous-measurement walker result that this tree model contrasts with, where no such transition is expected."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the directed-polymer background used to argue the tree gives a qualitative description for sufficiently high spatial dimension."},{"cited_title":"Halpin-Healy and Y.-C","cited_arxiv_id":null,"evidence_quote":"Supports the high-temperature self-averaging properties of directed polymers that justify the tree description in high dimension."}],"review_version":1}