{"id":"031fc353-98f3-4f57-a12b-569a1331c92f","arxiv_id":"2412.14712","paper_version":1,"verdict":"REJECT","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"high","formal_verification":"none","parameter_count":3,"one_line_summary":"For transient-dimension directed polymers in time-correlated Markovian random fields, the paper asserts the law of large numbers and a weak-disorder localization regime, with proofs that have significant gaps.","lead":"A mathematics preprint studies directed polymers in random fields that are correlated in both space and time, a more realistic setting than the usual independent randomness. It claims to prove a law of large numbers, existence of free energies, and a localization-delocalization transition, but central proof steps are incomplete.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 2.3's LLN limit is defined as lim_{L→∞} γ_L/β_L, but existence of this deterministic limit is never proved; the diagonal limit L→∞ after the block argument is therefore unsupported.","rationale":"The reader's weakest_assumption identifies exactly the most load-bearing gap: Theorem 2.3 defines ℓ as lim_{L→∞} γ_L/β_L without proving that this limit exists. I reviewed Section 4, especially the lines around (4.4)–(4.5) and the final paragraph of Lemma 4.2. The fixed-L block argument is plausible: Lemma 4.1 plus the conditional coupling from Lemma 3.2 does give convergence of the block averages to γ_L/β_L for each L, and Lemma 3.1 gives positive normalized block lengths. However, nothing in the argument controls γ_L/β_L as L varies. The quantity γ_L involves an expectation of a sum of ω-values along a random-walk path killed at the random time τ_1^{(L)}; the mixing condition controls conditional dependence between past and future blocks, not the drift of these block means in L. Since the theorem's conclusion is exactly the existence of a deterministic ℓ, this is not a mere technical annoyance but an unproved premise of the central claim. The same missing limit is also used when the paper later interpolates between τ_{k_N}^{(L)} and N to convert block limits into the original time-average limit. I do not adjust the reader's REJECT verdict: the concern supports rejection, and the paper should either prove the limit exists under (TC)/(TCG) or state it as an explicit assumption.","tokens_in":31301,"tokens_out":5269,"duration_ms":47594,"concrete_test":"Prove or disprove convergence of γ_L/β_L by deriving an increment bound of the form |γ_{L+1}/β_{L+1} − γ_L/β_L| ≤ C e^{-cL}, using the cone-mixing estimates (3.5)–(3.6), Lemma 3.1, and the exponential moment condition (2.1). If such a bound holds, insert it before Eq. (4.5) and supply the missing diagonal argument to complete Theorem 2.3. As a sanity check, simulate the block means for a simple stationary time-correlated field (e.g. a Gaussian AR(1) field on N×Z^d with the same TC parameters) for L = 5, ..., 15 and verify that γ_L/β_L stabilizes to a constant independent of L; if the ratios oscillate, the proof's diagonal limit is false rather than merely omitted.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section 4, Eq. (4.5), shows only that for each fixed L the block-averaged moving average converges a.s. to γ_L/β_L, with an error that vanishes as L→∞. The proof then writes ℓ = lim_{L→∞} γ_L/β_L and uses this as the limit for the original time average, but no argument establishes that this deterministic limit exists. Here γ_L := EE^S_0[4^{-L}∑_{k=1}^{τ_1^{(L)}} η_{k,S_k}] is a time- and environment-averaged sum over a block of random length of order 4^L, while β_L := E_Q[4^{-L}τ_1^{(L)}] is the normalized expected block length. A priori the ratio γ_L/β_L could oscillate in L: Lemma 3.2 gives exponentially small conditional perturbation between successive blocks for fixed L, and Lemma 3.1 gives uniform L^p bounds on the normalized block length, but neither controls the large-L evolution of γ_L itself. The statement 'there exists a deterministic limit ℓ' is precisely the missing conclusion. Moreover, even if the limit exists, the proof still needs a diagonal argument to pass from the fixed-L, N→∞ limits in (4.5) to the L,N→∞ double limit; the manuscript does not supply uniform-in-L errors for the original time averages. Because the LLN is used as the starting point for the block interpolation in the proof of Theorem 2.5 (β_* > 0), this gap is load-bearing: without lim_L γ_L/β_L, neither the LLN nor the small-β localization claim is established by the manuscript.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies directed polymers in a Markovian random field on N×Z^d satisfying one of two time-correlation conditions, (TC) or (TCG). It claims a law of large numbers for the polymer moving average (Theorem 2.3), existence of quenched and annealed free energies with differentiability of the annealed free energy (Theorem 2.4), a positive-temperature localization regime with β*>0 (Theorem 2.5), and entropy-type criteria for delocalization and 0<β*<∞ (Theorem 2.6). The proof strategy introduces auxiliary fields and regeneration times to isolate blocks of the polymer path and uses cone estimates to control the time-correlated environment.","tokens_in":31645,"tokens_out":12368,"duration_ms":108521,"significance":"The topic is timely and the auxiliary-field/block decomposition is an original technical idea; if the results were fully established, they would be a substantial extension of directed-polymer results beyond independent environments. The paper does not fit parameters to data, and the main claims are stated as falsifiable mathematical theorems. However, several load-bearing steps are incomplete or incoherent as written, so the manuscript in its current form does not establish its central claims.","major_comments":[{"comment":"The proof of Theorem 2.3 defines the deterministic limit as ℓ := lim_{L→∞} γ_L/β_L, but no argument is given for the existence of this limit; Lemmas 3.1 and 3.2 only control each fixed-L block and do not control the evolution of γ_L as L varies, and the estimate (4.5) is only for fixed L and n→∞. Passing from the fixed-L limits to the original time average also requires a uniform-in-L diagonal argument, which is absent. Since Theorem 2.5 later uses the LLN for τ_n^{(L)}/n and the block interpolation, this gap is load-bearing for the central localization claim.","section":"Section 4, after Eq. (4.5)"},{"comment":"The definition of ω† is not meaningful as written, and the FKG step is not justified; the measure with density (W^{β,ω†}_N)^{-1} e^{β∑ω†}/E[Z^{β,ω†}_N] is not shown to be monotone in ω, so the claimed E[∂_β log W_N] ≤ 0 does not follow. Consequently the non-increasing property of β↦ρ(β)−λ(β) in Theorem 2.4, and hence the standard definition of β*, is not established.","section":"Section 5, Step IV, around Eq. (5.5)"},{"comment":"The criteria for β*<∞ require ℏ=ess sup ω_{1,0} finite with P(ω_{1,0}=ℏ)>0, but the Gaussian example immediately after Theorem 2.6 has unbounded Gaussian ω_{1,0}, so the claimed 0<β*<∞ for that example does not follow from the stated theorem. The proof of Theorem 2.6 also contains a sign mismatch: the displayed hypothesis uses βλ′(β)−λ(β)>−K(S)H(S_1), whereas the subsequent derivative condition requires βλ′(β)−λ(β)>K(S)H(S_1).","section":"Section 7, Lemma 7.1 and Theorem 2.6"}],"minor_comments":[{"comment":"The definition of C(k,x,γ,ζ) has |vec z−vec x|^2 on the right-hand side, which is dimensionally inconsistent with the linear left-hand side; this appears to be a typo and should be corrected or explained.","section":"Section 3, cone definition"},{"comment":"The abstract states that smoothness of limiting free energies is proved at all temperature, but Theorem 2.4 asserts differentiability only for the annealed free energy λ, not for the quenched free energy ρ; the wording should be adjusted.","section":"Abstract and Theorem 2.4"},{"comment":"The displayed chain in Lemma 7.1 should use liminf and limsup explicitly; as written, the repeated use of lim over β→∞ is ambiguous.","section":"Section 7, Lemma 7.1"}],"recommendation":"reject","confidential_remarks":"The manuscript has original content and the topic is suitable for the journal, but the central proofs have unresolved gaps: the LLN limit is defined without a proof of existence or a diagonal argument, the monotonicity proof of ρ−λ rests on an undefined field and an unsupported FKG claim, and the Gaussian example is not covered by the stated delocalization criteria. These are load-bearing issues rather than presentation problems, so I recommend rejection of the current version."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Let me give you my read on arXiv:2412.14712.\n\nThe paper tries to carry the directed polymer program over to time-correlated random fields. The genuinely new piece is the auxiliary-field (cone-mixing) construction, adapted from Comets–Zeitouni and Guerra Aguilar, applied to polymers for the first time. If the proofs worked, Theorem 2.4 (free energies exist and are smooth) and Theorem 2.5 (beta* > 0) would be real progress. The appendix has substantial, serious technical work: the Khas'minskii-type bounds and the martingale arguments are not placeholders.\n\nThe soft spots are not minor. The law of large numbers in Theorem 2.3 defines its limit as lim_{L→∞} gamma_L/beta_L, but the existence of that deterministic limit is never proved. Section 4 only shows that for each fixed L the block averages converge to gamma_L/beta_L with an error vanishing in L; it then jumps to \"the desired law of large numbers\" with ell = lim gamma_L/beta_L. The stress-test note is right: without a proof that gamma_L/beta_L converges, or a diagonal argument with uniform errors, Theorem 2.3 is unsupported. This is a main theorem, not a technicality.\n\nTheorem 2.4's monotonicity step is also shaky. The field omega-dagger is defined as an integral of omega over \"N\\{n}×Z^d\", which is not a well-posed object, and the FKG step is asserted rather than demonstrated. Lemma 7.1 explicitly omits the upper-bound proof. The Gaussian example claimed to satisfy the conditions is stated without verification. These are real gaps in load-bearing places.\n\nWhere I'd push back on the reader's take: the stress-test claim that Theorem 2.5 depends on the missing LLN seems overstated. The proof of Theorem 2.5 uses the law of large numbers for the renewal times tau_n^(L) under Q, which is independent block lengths, not the moving-average LLN. So the localization result might survive even if Theorem 2.3 is repaired or removed. But that doesn't save the paper; the LLN is still advertised as a main result and it is not proved.\n\nWhat's the bottom line? This is a serious attempt by someone who knows the relevant literature and has a promising tool. The gaps are fixable in principle, but they are gaps. I'd send it back for major revision rather than accept it. A referee who knows the area should engage with it; desk rejection would be too harsh.\n\nMy recommendation: reject in current form, but invite a resubmission with the LLN fixed and the monotonicity/delocalization arguments tightened.","headline":"A promising but unfinished extension of directed polymer results to time-correlated fields; the main LLN is unproven because its claimed limit is never shown to exist.","tokens_in":32212,"tokens_out":3488,"would_cite":false,"duration_ms":29765,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60K35","82B44","60F10","60G42"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper proves that a high-dimensional directed polymer in a time-correlated random field obeys a law of large numbers, has well-defined free energies at all temperatures, and localizes only beyond a positive critical inverse…","keywords":["directed polymer","time-correlated random field","law of large numbers","quenched free energy","annealed free energy","localization-delocalization transition","Markovian random field","cone-mixing condition"],"falsifier":"Compute $\\gamma_L := E E^S_0[4^{-L}\\sum_{k=1}^{\\tau_1^{(L)}}\\eta_{k,S_k}]$ and $\\beta_L := E_Q[4^{-L}\\tau_1^{(L)}]$ for the paper's Gaussian example with covariance $\\exp(-\\|x-y\\|_1)G(x,y)$, and check whether $\\gamma_L/\\beta_L$ converges as $L\\to\\infty$; if two subsequences give different limits, the deterministic $\\ell$ in Theorem 2.3 does not exist. A Monte-Carlo simulation of the polymer path showing that $N^{-1}\\sum_{k=1}^N\\omega_{k,S_k}$ fails to concentrate across independent environment samples as $N$ grows would likewise disprove the law of large numbers.","tokens_in":31036,"feed_emoji":"🧬","tokens_out":8440,"duration_ms":60308,"temperature":0.7,"pith_summary":"This paper studies a directed polymer chain in a random environment that varies with time as well as position, under two exponential-mixing conditions on the field. It claims that in transient dimensions ($d\\ge 3$), for every inverse temperature $\\beta\\ge 0$, the time-averaged environment seen along the polymer path obeys a law of large numbers and converges to a deterministic constant. It further claims that the quenched and annealed free energies exist at all temperatures, that the annealed free energy is differentiable, and that the difference $\\rho(\\beta)-\\lambda(\\beta)$ is continuous and non-increasing, so a critical $\\beta^*$ is well defined. The main new result is that $\\beta^*>0$: weak disorder always survives for small positive $\\beta$, and under an entropy-type condition one gets $0<\\beta^*<\\infty$, meaning a genuine delocalized-localized transition. This matters because time-correlated disorder is substantially harder than the i.i.d. case and its localization picture was previously unresolved.","feed_headline":"Polymer averages converge in time-correlated random fields","feed_subtitle":"Free energies exist at all temperatures and a positive critical β marks the localization regime.","key_machinery":"The load-bearing construction is a pair of auxiliary random fields plus a block-regeneration scheme that disconnects the time correlation. For an independent field $\\epsilon_n\\in\\{-1,0,1\\}$, the paper sets $\\eta_{n,z}=2\\omega_{n,z}\\mathbf{1}_{\\{\\epsilon_n=0\\}}$, so that $E_Q[\\eta_{n,z}]=\\omega_{n,z}$, and $\\xi^{(l)}_{n,z}=-\\beta l\\mathbf{1}_{\\{\\epsilon_n=\\pm1\\}}+\\log(2e^{\\beta\\omega_{n,z}}-e^{-\\beta l})\\mathbf{1}_{\\{\\epsilon_n=0\\}}$, so that $E_Q[Z^{1,\\xi^{(l)}}_N]=Z^{\\beta,\\omega^{(l)}}_N$ for the truncated field $\\omega^{(l)}=\\max\\{\\omega_{n,z},-l\\}$. Random times $\\tau_n^{(L)}$, defined by long runs of $+1$ followed by $-1$ or $0$, split the path into blocks whose conditional laws are shown, via cone-mixing estimates, to be within $e^{-gtL}$ in total variation of a fixed law. This yields the law of large numbers. For localization, the normalized block partition function $H_{n,\\beta}$ is exhibited as a nonnegative martingale; an $L^2$ estimate, a zero-one law, and comparison with $L_{n,\\beta}$ give $\\rho(\\beta)=\\lambda(\\beta)$ for small $\\beta$.","core_discovery":"On the paper's own terms, the discovery is that time-correlated disorder does not destroy the basic thermodynamic and path-level structure of directed polymers. Under either the time-correlated condition (TC) or Guo's time-correlated condition (TCG), and for a transient reference random walk, the paper proves that for every fixed $\\beta\\ge 0$ the moving average $N^{-1}\\sum_{k=1}^N\\omega_{k,S_k}$ converges $P\\otimes P^S_0$-almost surely to a deterministic constant $\\ell$; that the quenched free energy $\\rho(\\beta)$ and annealed free energy $\\lambda(\\beta)$ exist for all $\\beta\\ge 0$, with $\\lambda$ differentiable on $[0,\\infty)$ and $\\rho-\\lambda$ continuous and non-increasing; and that the critical inverse temperature $\\beta^*=\\inf\\{\\beta>0:\\rho(\\beta)<\\lambda(\\beta)\\}$ is strictly positive. The paper also gives sufficient conditions under which $\\beta^*=\\infty$ and under which $0<\\beta^*<\\infty$, and it exhibits a Gaussian field with covariance $\\exp(-\\|x-y\\|_1)G(x,y)$ that realizes the latter case.","pith_inferences":["Editorial inference: one could test numerically whether $\\gamma_L/\\beta_L$ converges for the paper's Gaussian example, since the proof defines $\\ell$ as that limit; an explicit computation showing oscillatory behavior would pinpoint exactly where the law of large numbers needs an additional hypothesis.","Editorial inference: the block-regeneration construction is not tied to nearest-neighbor walks and could be extended to $\\alpha$-stable long-range walks by replacing the single space-time cone with a countable superposition of cones, yielding analogous LLN and free-energy statements for L\\'evy directed polymers.","Editorial inference: the constants $\\kappa_1,\\kappa_2,K'$ in the localization criteria are crude, and the paper itself notes that $K'$ vanishes in the i.i.d. limit; sharpening them might produce a closed-form threshold for $\\beta^*$ in terms of single-site exponential moments and walk entropy."],"forward_implications":["For fixed $\\beta\\ge 0$, the almost-sure limit $\\ell=\\lim_{N\\to\\infty}N^{-1}\\sum_{k=1}^N\\omega_{k,S_k}$ makes the environment self-averaging along the polymer path despite temporal correlations.","The free energies $\\rho(\\beta)$ and $\\lambda(\\beta)$ exist for every $\\beta\\ge 0$, with $\\lambda$ differentiable and $\\rho-\\lambda$ continuous and non-increasing, so the weak-disorder/strong-disorder dichotomy is governed by a single critical value $\\beta^*$.","The inequality $\\beta^*>0$ guarantees that in transient dimensions small disorder is always in the delocalized regime under either (TC) or (TCG).","If $\\lim_{\\beta\\nearrow\\infty}\\Lambda(\\beta)