{"id":"2a1d78a0-8784-43ff-b523-9498c543df79","arxiv_id":"1908.04062","paper_version":1,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For highly unstable potentials, the long-time distribution of trajectories conditioned to never diverge is light-tailed, while the quasi-stationary distribution of trajectories surviving to a fixed time is heavy-tailed with an exponent set by the potential's divergent term.","lead":"This paper studies Brownian particles in steep, unstable potentials where most trajectories fly off to infinity in finite time. It shows that two natural ways of conditioning on non-diverging trajectories give very different position statistics: one is tightly localized, the other has a power-law tail.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Spectral-gap proof in Sec. 3 is internally mis-signed and 'U confining' is asserted, not shown, for the finite-MFPT class; central general claim is conditional.","rationale":"The reader's weakest assumption is exactly the spectral-gap premise, and I agree it is load-bearing. My pass sharpens it: Eqs. (18a)-(18b) have a sign error, so the proof of the spectral identification is not valid as typeset; and the step from finite-MFPT to confining U is asserted rather than derived. These are correctness risks for the 'all' claim, not for the explicit monomial results. I verified the rest of the chain: Eq. (24) follows from normalization (25)-(26); Eqs. (46)-(48) and (62)-(63) follow from the asymptotic form (47); Eq. (50) is a legitimate integration by parts. The paper has no data or fitting, and the machinery (Q-process, Doob h-transform) is standard. With the sign corrected and either a proof of U confining under the Sec. 2 definition or an explicit restriction of the central claim to potentials with monomial unstable tails, the dichotomy is sound. Hence conditional acceptance rather than flat accept; no rejection is warranted.","tokens_in":14238,"tokens_out":18479,"duration_ms":190180,"concrete_test":"Re-derive Eq. (18a) explicitly: compute e^{βV/2} L† (e^{−βV/2} f) for a Schwartz test function f and compare with −D f'' + U f; the correct identity should read e^{βV/2} L† e^{−βV/2} = −H, not H. Then check whether the defining condition of Sec. 2 implies U(x) → +∞: construct a potential with finite MFPT but U not confining (e.g., V with V''/(V')² unbounded on a sequence while ∫ dx/V' < ∞); if such a potential exists, the general claim must be restricted to potentials with confining U (e.g., monomial tails).","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central dichotomy for the full class of 'highly unstable potentials' (finite mean first-passage time, Sec. 2) depends on the existence of a positive spectral gap Δ = λ1 − λ0, because Eqs. (16)-(17), the Q-process propagator (31), and both tail asymptotics (48) and (62) require the slowest eigenmode to dominate with exponential accuracy. The argument in Sec. 3 has two problems. First, the displayed similarity transformation is mis-stated: acting on a smooth test function gives e^{βV/2} L† e^{−βV/2} = −(−D∂² + U), not (−D∂² + U) as written in Eq. (18a); the same sign is missing in Eq. (18b). Taken literally, Eq. (18a) together with HΨ_n = λ_nΨ_n (Eq. 20) would make the Fokker-Planck eigenvalues +λ_n, contradicting Eq. (6). The correct statement is H = −e^{βV/2} L† e^{−βV/2} = −e^{−βV/2} L e^{βV/2}. Second, the assertion 'U(x) is always confining for highly unstable V(x)' is not proved from the defining finite-MFPT condition; finite MFPT does not by itself rule out oscillatory potentials for which U has infinitely many deep negative wells and the Schrödinger spectrum is not purely discrete. Thus, as written, the text does not establish the spectral gap that the central 'for all highly unstable potentials' claim requires. The monomial results are unaffected once the sign is corrected.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies one-dimensional overdamped Brownian motion in potentials that decrease to -∞ so rapidly that the particle reaches infinity in finite time ('highly unstable' potentials). Using the spectral decomposition of the Fokker-Planck operator, it analyzes two conditional ensembles: the Q-process (trajectories conditioned to never diverge) and the quasi-stationary distribution (trajectories conditioned to survive up to the observation time). It derives explicit tail asymptotics: the Q-process limit distribution is light-tailed with exponential decay e^{βV} on the unstable side, while the quasi-stationary distribution is heavy-tailed with a power law ~ |x|^{-(n-1)} for monomial potentials V ~ -|x|^n with n > 2. The paper also derives an exact identity for the generalized partition function, an exact formula for the mean squared effective force, and a scaling law for that force in terms of μ, kBT, and n. The central claim is that the two natural conditional ensembles have fundamentally different statistical properties for all highly unstable systems.","tokens_in":14430,"tokens_out":8232,"duration_ms":76922,"significance":"If correct, the dichotomy established here is conceptually important and practically useful: it distinguishes two natural conditional ensembles in unstable stochastic dynamics and gives explicit, parameter-free asymptotic predictions. The derivations are elegant and largely self-contained: the identity (24) follows from eigenfunction normalization, the tail asymptotics follow from the adjoint eigenvalue equation, and the mean-squared-force result (50) is an exact integration-by-parts identity. The paper provides falsifiable predictions, such as the power-law exponent n-1 and the scaling (54), which are checkable in the colloidal experiments mentioned in the introduction. The connection to Q-process theory and Doob's h-transform is well explained. The monomial results are clean and convincing; the main weakness is the generality of the spectral-gap claim.","major_comments":[{"comment":"The similarity transformation is mis-signed. Acting on a smooth test function yields e^{βV/2} L† e^{-βV/2} = -(-D∂² + U), not (-D∂² + U) as written; the same sign error is present in Eq. (18b). The correct statement is H = -e^{βV/2} L† e^{-βV/2} = -e^{-βV/2} L e^{βV/2} = -D∂² + U. As written, Eq. (18a) combined with Eq. (20) (H Ψ_n = λ_n Ψ_n) would imply Fokker-Planck eigenvalues +λ_n, contradicting Eq. (6). This is a local sign error, but it appears in the proof of the discrete spectrum and should be corrected.","section":"Sec. 3, Eqs. (18a)-(18c)"},{"comment":"The assertion that U(x) is confining for all highly unstable potentials is not proved from the defining finite-mean-first-passage-time condition in Sec. 2. Finite MFPT does not by itself rule out oscillatory potentials for which U(x) has infinitely many deep negative wells and the associated Schrödinger operator has a non-discrete spectrum. Consequently, the discrete-spectrum claim (5), which is load-bearing for the propagator asymptotics (16)-(17), the Q-process propagator (31), and the tail results (48) and (62), is not established for the full class as defined. The monomial potentials (for which U ~ |x|^{2n-2}) are fine, but the 'for all' claims in the abstract and Sec. 7 need either a rigorous proof of discrete spectrum under the stated definition or a more precisely restricted class of potentials.","section":"Sec. 3, Eq. (19) and the sentence 'U(x) is always confining for all highly unstable potentials'"},{"comment":"The tail expression Q_st(x) ~ λ0 γ / V'(x) is written without an absolute value. For a potential that decreases to -∞ as x → -∞, V'(x) is negative on that side, which would make the right-hand side negative. The subsequent monomial formula in Eq. (63) correctly uses |x|^{n-1}. Please replace V'(x) by |V'(x)| (or explicitly state that V'(x) is taken in the direction where it is positive).","section":"Sec. 6, Eq. (62)"}],"minor_comments":[{"comment":"This equation has the same sign issue as Eq. (18a); both should be corrected as described in the major comment.","section":"Sec. 3, Eq. (18b)"},{"comment":"The cross-reference to '(eq:Qsts0)' is a broken LaTeX label; it should refer to Eq. (59).","section":"Sec. 6, after Eq. (59)"},{"comment":"'Chapmann-Kolmogorov' should be 'Chapman-Kolmogorov', and 'absorbed Marov chains' should be 'absorbed Markov chains'.","section":"Sec. 4 and Sec. 6"},{"comment":"The caption labels both the left and right axes of panel (d) with '(d)'; it should read '(d) Left axis: ...' and '(d) Right axis: ...'.","section":"Figure 1 caption"}],"recommendation":"major_revision","confidential_remarks":"The sign error in Eq. (18) appears to be a straightforward typographical slip, and the monomial results of the paper survive once the sign is corrected. The substantive issue is the unproven claim that U(x) is confining for all potentials satisfying the finite-MFPT definition; I think this can be fixed either by proving the discrete spectrum under a slightly more restrictive but physically natural condition, or by explicitly restricting the general claims. I would not reject the paper on this basis, as the central dichotomy is well supported for the monomial class and the mathematical framework is sound once the transformation sign is fixed."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear Colleague,\n\nThe paper by Ryabov, Holubec, and Berestneva generalizes the spectral machinery they used for the cubic potential to all monomial unstable potentials V(x) ~ -|x|^n with n>2. The main result is a clean dichotomy: the Q-process limit distribution is light-tailed with exponential decay on the unstable side, while the quasi-stationary distribution is heavy-tailed with power-law exponent n-1. That contrast is the real news, and I think it will hold up.\n\nThe derivations are mostly solid. The identity (24) is a neat consequence of the normalization of left and right eigenfunctions. The tail exponents follow from the adjoint eigenvalue equation without any fitted constants. The mean-squared-force result (50) is an exact integration-by-parts identity, and the scaling exponents check out. The paper is honest about these being formal results for a class defined by finite mean first-passage time.\n\nThat said, there are two soft spots. First, the similarity transformation in Sec. 3 is mis-signed: as written, Eq. (18a) would give the opposite sign for the Hamiltonian relative to Eq. (20). The correct transformation is H = - e^{βV/2} L^† e^{-βV/2}. This is a typo, and it does not affect the physics, but it should be fixed because the printed equations are inconsistent.\n\nSecond, and more substantively, the claim that the spectrum is discrete for all \"highly unstable potentials\" is not actually proved. The definition in Sec. 2 is finite MFPT, and the authors assert that U(x) is confining for this entire class. That may be true for the monomial potentials they analyze, but finite MFPT alone does not rule out oscillatory potentials for which U has infinitely many deep wells and the Schrödinger operator has a collapsed gap or extra essential spectrum. The general claim in the abstract is therefore conditional on a spectral-gap assumption that is argued but not established for the full class. Reframing the paper as applying to the monomial family, or to potentials where the gap is checked, would remove the overreach.\n\nThe reader's report had this about right; the stress-test note correctly identifies the sign error and the unproved assertion. My own read is that the sign error is a slip and the spectral-gap issue is a proof-sketch gap rather than a fatal flaw. For the stated examples the results are correct, and the heavy/light dichotomy is a genuine contribution.\n\nWho should read this: anyone working on Q-processes, quasi-stationary distributions, or unstable stochastic dynamics in optical traps. It deserves a serious referee, but it needs minor revision: correct the sign, tone down the \"for all\" claim, and either prove the gap or state the class as those potentials for which U is confining.\n\nRecommendation: send it to peer review, with the expectation of minor revision.","headline":"Generalizes the authors' earlier Q-process results for the cubic potential to all monomial unstable potentials; the light/heavy-tailed dichotomy is the key new result, but the spectral-gap proof for the full class is sketched rather than rigorous and the similarity transformation has a sign typo.","tokens_in":15070,"tokens_out":6483,"would_cite":true,"duration_ms":59127,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60J60","60J70","82C31"],"pacs":["05.40.-a","02.50.Ey"],"model":"deepseek-v4-flash","headline":"For the same unstable potential, 'never diverging' trajectories are exponential-tailed while 'alive at a fixed time' trajectories are power-law-tailed.","keywords":["unstable potentials","Q-process","quasi-stationary distribution","Brownian dynamics","h-transform","heavy tails","Fokker-Planck spectrum","effective potentials"],"falsifier":"Compute the two lowest eigenvalues of the transformed Hamiltonian for a family of monomial potentials $V(x)=-\\mu|x|^n/n$ with $n$ decreasing toward $2$. If for any $n>2$ with finite mean first-passage time the gap $\\lambda_1-\\lambda_0$ vanishes, or the spectrum becomes continuous, then the claimed exponential tail for $\\pi_{\\mathrm{st}}$ and power-law tail for $Q_{\\mathrm{st}}$ cannot hold in that regime.","tokens_in":13923,"feed_emoji":"🎲","tokens_out":19385,"duration_ms":180782,"temperature":0.7,"pith_summary":"The paper studies overdamped Brownian motion in potentials that fall toward minus infinity so steeply that typical trajectories escape to infinity in finite time. It asks what happens if one keeps only the non-escaping trajectories, and it claims that the answer depends sharply on how the conditioning is done. Conditioning on divergence only in the infinitely distant future (the Q-process) yields a localized, light-tailed position law, while conditioning on mere survival up to a fixed late time (the quasi-stationary distribution) yields a heavy-tailed law whose power-law exponent is set by the leading unstable term $|x|^n$, $n>2$. Both distributions reduce to the ordinary thermal equilibrium distribution for stable potentials, so the paper reads them as two rival generalizations of equilibrium statistics to unstable dynamics, with opposite tail behavior.","feed_headline":"Unstable-potential paths split: exponential vs power-law tails","feed_subtitle":"Never-diverge conditioning gives exponential tails; survive-to-time conditioning gives power-law tails.","key_machinery":"The argument is carried by transforming the non-Hermitian Fokker-Planck operator into a Hermitian Hamiltonian $\\hat H = \\mathrm{e}^{\\beta V/2}\\hat L^\\dagger \\mathrm{e}^{-\\beta V/2} = -D\\partial_x^2 + U(x)$ with $U(x)=[V'(x)/\\gamma]^2/(4D)-V''(x)/(2\\gamma)$. Since $U$ is confining whenever $V$ is highly unstable, the spectrum is discrete, and the ground state wave function $\\Psi_0$ encodes the two relevant eigenfunctions through $p_0=Z^{-1/2}\\mathrm{e}^{-\\beta V/2}\\Psi_0$ and $s_0=\\sqrt{Z}\\,\\mathrm{e}^{\\beta V/2}\\Psi_0$. Substituting these relations into the two normalization conditions yields the identity that turns normalization into tail information, and differentiating the left-eigenfunction equation converts the same identity into the asymptotic formulas for both tails. The h-transform then generates an effective statistical force $F_s=2k_BT\\,s_0'/s_0$ that confines non-diverging trajectories; its mean-square value is $4\\gamma k_BT\\lambda_0$.","core_discovery":"For potentials $V(x)\\sim -|x|^n$ with $n>2$, the paper establishes a dichotomy between the two natural conditional ensembles. The limit distribution of the Q-process is $\\pi_{\\mathrm{st}}(x)=s_0(x)p_0(x)=\\Psi_0^2(x)$, where $p_0$ and $s_0$ are the right and left eigenfunctions of the slowest Fokker-Planck mode and $\\Psi_0$ is the ground state of the transformed Hamiltonian; on the unstable side it decays as $\\pi_{\\mathrm{st}}(x)\\sim Z[\\lambda_0\\gamma/V'(x)]^2 e^{\\beta V(x)}$, an exponential tail. The quasi-stationary distribution is $Q_{\\mathrm{st}}(x)=p_0(x)$ and decays on the same side as $Q_{\\mathrm{st}}(x)\\sim \\lambda_0\\gamma/V'(x)$, which for monomial potentials means the power law $|x|^{-(n-1)}$. The two formulas are forced by a single identity, $\\int e^{-\\beta V(x)}s_0(x)\\,dx = \\int e^{-\\beta V(x)}s_0^2(x)\\,dx$, that follows from the simultaneous normalization of the two eigenfunctions, so the exponential-versus-power-law split is not an accident of a particular potential.","pith_inferences":["A direct experimental test would compare two conditioning protocols on colloid trajectories in a cubic optical trap—conditioning on survival to a fixed observation time versus conditioning on eventual divergence—and check that the log-log slope of the left tail is $-(n-1)$ only in the first protocol.","The paper's dependence on the order of limits suggests that the crossover from the localized to the heavy-tailed law is observable: an ensemble conditioned at a finite final time should pass from $\\pi_{\\mathrm{st}}$ to $Q_{\\mathrm{st}}$ on a time scale set by $1/(\\lambda_1-\\lambda_0)$.","The equal-area identity $\\int e^{-\\beta V}s_0=\\int e^{-\\beta V}s_0^2$ may be a useful design principle in higher-dimensional unstable systems: any variational approximation to $s_0$ would immediately yield both tails without a full spectral solution.","As $n\\to 2^+$, the effective force weakens, and the paper itself raises the question of when the Q-process ceases to be stationary; one could test this at potentials just above $n=2$ and look for the disappearance of a normalizable $\\pi_{\\mathrm{st}}$."],"forward_implications":["For monomial unstable potentials, the quasi-stationary left tail is $|x|^{-(n-1)}$, so for a cubic trap no integer moments exist and moment-based averages of the surviving ensemble are undefined.","The Q-process distribution is localized with an exponential left tail and a logarithmic effective barrier that is impenetrable for $n>2$, so never-diverging trajectories effectively avoid the unstable region.","Both stationary laws and the generalized partition function depend on the friction constant $\\gamma$ through $s_0$, meaning they are kinetic quantities, not purely thermodynamic equilibrium distributions.","The mean-square effective force obeys $\\langle F_s^2\\rangle=4\\gamma k_BT\\lambda_0$, which for monomial potentials scales as $\\mu^{1/n}(k_BT)^{(n-1)/n}$; stronger instability and higher temperature both strengthen the confining force.","On the stable right side, both distributions approach the ordinary tail $\\mathrm{e}^{-\\beta V(x)}/Z$, so the two conditionings differ visibly only on the unstable side."],"supporting_citations":[{"why":"Defines highly unstable potentials through finite mean first-passage time and supplies the cubic-potential analysis the paper starts from.","marker":"[24]"},{"why":"Gives the one-dimensional mean first-passage formula and the Hermitian transformation used to establish the discrete spectrum.","marker":"[27]"},{"why":"Supplies the spectral expansion and the transformation of the Fokker-Planck operator into Hermitian form.","marker":"[29]"},{"why":"Provides the eigenfunction expansion of the propagator on which the long-time asymptotics and both limit formulas rest.","marker":"[28]"},{"why":"Provides the general theory of conditioned Markov processes used to identify the Q-process ensemble.","marker":"[6]"},{"why":"Supplies the quasi-stationary distribution framework used for the survival-conditioned limit.","marker":"[7]"},{"why":"Supplies the h-transform that turns the conditioned process into a Markov process with an effective drift.","marker":"[8]"}],"fun_headline_variants":["Never-diverge vs survive: exponential vs power-law tails","Unstable-potential paths: two tails, exponential and power-law","Conditioning choice dictates tail shape in unstable potentials","Exponential or power-law: unstable potentials branch on conditioning","Two conditionings, two tail laws in unstable potentials"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"Everything rests on the assumption that, in the long-time limit, one slowest decay mode is cleanly separated from all faster modes; the paper's Section 3 argues for such a spectral gap via the transformed Hamiltonian, but a complete proof for every potential satisfying its defining finite-divergence condition is not given.","fun_headline_variants_meta":{"raw":{"variants":["Never-diverge vs survive: exponential vs power-law tails","Unstable-potential paths: two tails, exponential and power-law","Conditioning choice dictates tail shape in unstable potentials","Exponential or power-law: unstable potentials branch on conditioning","Two conditionings, two tail laws in unstable potentials"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00095,"raw_usage":{"total_tokens":4086,"prompt_tokens":1013,"completion_tokens":3073,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":629,"completion_tokens_details":{"reasoning_tokens":2992}},"tokens_in":629,"tokens_out":3073,"duration_ms":21466,"temperature":1.0,"reasoning_tokens":2992,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T13:53:49.058643+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the two lowest eigenvalues of the transformed Hamiltonian for a family of monomial potentials $V(x)=-\\mu|x|^n/n$ with $n$ decreasing toward $2$. If for any $n>2$ with finite mean first-passage time the gap $\\lambda_1-\\lambda_0$ vanishes, or the spectrum becomes continuous, then the claimed exponential tail for $\\pi_{\\mathrm{st}}$ and power-law tail for $Q_{\\mathrm{st}}$ cannot hold in that regime.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Defines highly unstable potentials through finite mean first-passage time and supplies the cubic-potential analysis the paper starts from."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the one-dimensional mean first-passage formula and the Hermitian transformation used to establish the discrete spectrum."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the spectral expansion and the transformation of the Fokker-Planck operator into Hermitian form."},{"cited_title":"Henri Poincar´ e16 2005 URL https://doi.org/10.1007/ s00023-014-0375-8","cited_arxiv_id":null,"evidence_quote":"Provides the general theory of conditioned Markov processes used to identify the Q-process ensemble."}],"review_version":1}