{"id":"62eb3a3d-f6ee-403a-b62a-df29200b32cb","arxiv_id":"2507.23637","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For the stochastic heat equation on the torus with coefficients growing like u|log u|^A near zero (A<1 for drift, A<1/4 for noise), a unique global strictly positive mild solution exists.","lead":"This paper proves that a stochastic heat equation whose drift and noise coefficients become very steep near zero still has a unique global solution that stays positive, provided the coefficients grow at most like u times a power of |log u|. This matters because such logarithmic coefficients arise in wave and relativistic models, and previously only partial cases were understood.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified to the central Theorem 3.1; the stopping-time proof uses the stated A1<1 and A2<1/4 exactly, and I find no load-bearing gap.","rationale":"I focused on the proof of Theorem 3.1, the paper's central claim. The stopping-time estimate (3.11) indeed depends on A1 < 1 and 4A2 < 1, but these are explicit hypotheses of the theorem, not an unstated assumption; the reader's 'weakest assumption' is therefore not a load-bearing objection to the stated result. I checked the potentially confusing definition of the stopping times T_k: as written, T_k uses u_{\\epsilon(k)}, while the strong Markov step invokes u_{\\epsilon(n)}. This is resolvable because for k ≤ n and before the process drops below e^{-k}, all relevant values lie above e^{-n}, where u_{\\epsilon(k)} and u_{\\epsilon(n)} have identical coefficients; hence T_k is also the first hitting time of e^{-k} for the fixed process u_{\\epsilon(n)}. The later comparison, scaling, and Kolmogorov-continuity steps in (3.9)-(3.11) are internally consistent. I therefore do not find a load-bearing gap in the central existence/uniqueness argument. I do agree with the reader's secondary concerns: the abstract claims strict positivity without the Theorem 3.1 hypothesis inf u0 > 0, and Theorem 4.1's condition (4.1) fails for its own stated example b(z) = -z log(1/z), since |b(z)|/z = log(1/z) increases as z → 0. These are real issues in the paper's presentation and extensions, but they do not change the main theorem's verdict.","tokens_in":22790,"tokens_out":45636,"duration_ms":475352,"concrete_test":"Re-derive Section 3, Step 1-1 with T_k defined as the first time the fixed process u_{\\epsilon(n)} (with \\epsilon(n)=e^{-n}) drops below e^{-k}, and verify that the conditional probability estimate (3.10) follows from the strong Markov property applied to u_{\\epsilon(n)}; if this derivation is valid, the central stopping-time argument is sound.","verdict_should_be":"UNCHANGED","load_bearing_attack":"I do not find a load-bearing concern in the proof of Theorem 3.1. The critical estimate (3.11) requires A1 < 1 and 4A2 < 1, which are precisely the theorem's assumptions; this is a quantitative boundary, not a hidden flaw. The chaining argument can be read coherently with a single fixed process u_{\\epsilon(n)}: for k ≤ n the processes u_{\\epsilon(k)} and u_{\\epsilon(n)} coincide above e^{-n}, so T_k is also the first hitting time of e^{-k} for u_{\\epsilon(n)}, and the strong Markov application in (3.10) is justified. The main theorem's remaining steps (moment bounds, localization, comparison, limit) are standard and consistent. Non-central issues remain: the abstract overstates strict positivity without the inf u0 > 0 condition, and Theorem 4.1's condition (4.1) is incompatible with its stated example b(z) = -z log(1/z). These do not undermine the central claim but should be corrected.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the one-dimensional stochastic heat equation on the torus with drift b and diffusion coefficient σ whose Lipschitz constants may blow up as the argument approaches zero. Under growth conditions |b(z)/z| = O((log(1/z))^{A1}) and |σ(z)/z| = O((log(1/z))^{A2}) for z near zero, with A1<1 and A2<1/4, the authors construct a global mild solution by truncating the coefficients away from a small cutoff and proving, via a stopping-time estimate, that the truncated solutions do not hit the cutoff before any fixed time. They claim existence, uniqueness, and strict positivity when the initial condition is bounded away from zero, and they also propose extensions to the critical case A1=1 and to superlinear growth at infinity. The central stopping-time estimate is the main technical contribution.","tokens_in":23039,"tokens_out":18605,"duration_ms":192779,"significance":"If Theorem 3.1 is correct, it is a genuine advance: it extends well-posedness for the stochastic heat equation to drift and diffusion coefficients that are only locally Lipschitz away from zero, it removes the monotonicity assumption on σ(z)/z used in the recent work of Han-Kim-Yi, and it permits a nonzero drift term. The proof is a derivation from the stated assumptions using standard external tools (Walsh integral, Kotelenez comparison, heat-kernel estimates); there are no fitted parameters, and the central claim does not reduce to an earlier result. The stopping-time probability estimate is a real technical novelty. The main issues are concentrated in the abstract, in a reduction step for small positive initial data, and in the critical-case extension, which is internally inconsistent as stated.","major_comments":[{"comment":"The abstract states that a unique global mild solution that remains strictly positive is established under an initial condition that is only nonnegative and not identically zero. Theorem 3.1(1), however, proves strict positivity and uniqueness only under the additional condition inf_{z∈T} u0(z) > 0, stated in (3.3). The abstract should either include this hypothesis or the theorem must be strengthened; as written, the abstract overclaims the main result.","section":"Abstract and Theorem 3.1(1)"},{"comment":"In Step 3, the proof approximates a general nonnegative initial condition by u0,n = u0 + 1/n and says that Step 1 applies to these initial data. However, Step 1 (specifically Step 1-1) was proved only under the assumption inf u0 ≥ 1. The approximating data u0 + 1/n have infimum 1/n, which is below 1 for all large n, so the proof does not cover the intermediate case 0 < inf u0 < 1. A scaling argument or an analogous stopping-time proof for arbitrary positive lower bound is needed; as written, the reduction is incomplete.","section":"Theorem 3.1, Step 3 (Section 3, after Remark 3.2)"},{"comment":"The theorem's motivating example b(z) = -z log(1/z) does not satisfy condition (4.1): for this function, |b(z)|/z = log(1/z) is strictly increasing as z ↓ 0, so for every δ ∈ (0,1) and z < δ one has |b(z)|/z > |b(δ)|/δ, violating (4.1). In addition, under (4.1) the modified coefficient b^{(α)} defined in (4.3) is nonincreasing in α when θ_b = +1 and nondecreasing when θ_b = -1, opposite to the monotonicity claim in the proof of Step 1; the comparison direction must be reversed. The theorem as stated is internally inconsistent with its stated example.","section":"Theorem 4.1, condition (4.1) and Step 1"}],"minor_comments":[{"comment":"The condition 2√p Lσ (π/√κ)^{1/2} < 1/8 requires κ > (16√π√p Lσ)^4 = 65536 π^2 p^2 Lσ^4, not the stated κ > 216π^2 p^2 Lσ^4. The proof of the explicit constant in (2.2) is therefore not valid as written; replacing 216 by a sufficiently large constant repairs the argument without affecting the qualitative results.","section":"Proposition 2.3, after (2.8)"},{"comment":"The displayed equality for E[u(t,x)^p] drops the term E[u(t,x)^p 1_{t>τ_ϵ}] without justification as ϵ → 0. A Fatou argument applied to u(t,x)^p 1_{t<τ_ϵ}, together with the uniform moment bounds for u~_ϵ, would give the claimed finiteness, but the current line is not rigorous.","section":"Proof of Theorem 3.1, Step 1-1, the proof of (3.2)"},{"comment":"The suprema in (3.10) involve |V^{(k+1)}(s,x) - V^{(k+1)}(0,x)|, but (3.11) writes |V^{(k+1)}(s,x) - V^{(k)}(0,x)|; also the event in (3.11) is labeled {T_m ≤ T} after the proof was tracking T_{2m}. These notational inconsistencies should be cleaned up.","section":"Equations (3.10) and (3.11)"},{"comment":"The hypothesis 'u0 > 0' should be stated precisely as inf_{x∈T} u0(x) > 0, since uniqueness in Theorem 3.1 requires the condition (3.3).","section":"Theorem 4.2"},{"comment":"The notation 'T_k := inf_{0≤s≤t} {s > T_{k-1}, ...}' is confusing because t is already used as the fixed horizon; it should read 'T_k := inf{s > T_{k-1} : inf_{x∈T} u_{ϵ(k)}(s,x) ≤ e^{-k}}'.","section":"Definition of T_k in Step 1-1"},{"comment":"The statement that u^{(M)} is a strong Markov process is used in the induction on the time intervals but is not justified or cited; a reference or a short argument should be provided.","section":"Proof of Theorem 4.2"}],"recommendation":"major_revision","confidential_remarks":"The central stopping-time argument for A1<1 and A2<1/4 appears sound and is the real contribution of the paper. The main obstacles are the missing reduction for initial data with small positive lower bound, the abstract overclaim, and the internally inconsistent critical-case theorem. These are fixable within the manuscript's scope, so I do not recommend rejection. The high number of self-citations is noticeable but not inappropriate, since the cited works provide auxiliary estimates rather than the main result."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The main theorem (Theorem 3.1) is a genuine advance, and the proof appears sound. But the paper needs statement-level fixes: the abstract overclaims, and the extension section has a genuine inconsistency.\n\nWhat is new: the result extends Han–Kim–Yi [11] by allowing a nonzero drift and dropping the monotonicity assumption on σ(z)/z, and extends Shang–Zhang [29] to space-time white noise. The coefficient class near zero is natural, and the stopping-time estimate (3.11) is the technical heart. Section 2's moment bounds and Hölder regularity are careful and keep the dependence on Lipschitz constants explicit, which the stopping-time argument needs.\n\nWhere it is soft: First, the abstract says 'unique global mild solution that remains strictly positive' without mentioning that Theorem 3.1 needs inf u0 > 0 for both uniqueness and strict positivity. For general nonnegative u0, the constructed solution is only nonnegative. That is a real overclaim. Second, Theorem 4.1's condition (4.1) is incompatible with its own example b(z) = −z log(1/z). For that b, |b(z)|/z = log(1/z) is larger for z < δ than at δ, so the inequality in (4.1) is reversed. The monotonicity argument in the proof appears to want the reverse inequality; as written, the theorem and proof disagree. This is in an extension section, so it does not undermine Theorem 3.1, but it must be fixed. Third, the proof of Theorem 4.2 constructs a solution and shows it stays finite, but uniqueness is asserted without proof. Smaller, but still a statement-level gap.\n\nOn the stress-test: I agree that the chaining in the proof of Theorem 3.1 can be read coherently; A1 < 1 and 4A2 < 1 are exactly used in (3.11), not hidden assumptions.\n\nWho it is for: SPDE researchers working on non-Lipschitz coefficients, positivity, and support properties. It deserves a serious referee. The central result is solid enough that these issues are fixable in revision.","headline":"The main well-posedness theorem is real and the proof is sound, but the abstract and Theorem 4.1 both overstate or contradict what is actually proved.","tokens_in":23535,"tokens_out":8186,"would_cite":true,"duration_ms":73609,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60H15","35R60"],"pacs":[],"model":"deepseek-v4-flash","headline":"Logarithmic singularities at zero in drift and noise still admit a unique global positive mild solution of the stochastic heat equation.","keywords":["non-locally Lipschitz coefficients","stochastic heat equation","space-time white noise","torus","logarithmic nonlinearity","strict positivity","mild solution","stopping-time localization"],"falsifier":"Run the construction for a coefficient pair inside the stated range, say $b(z)=z(\\log(1/z))^{0.9}$ and $\\sigma(z)=z(\\log(1/z))^{0.24}$ near zero with a bounded Hölder initial condition, and compute the cutoff-crossing probabilities in (3.8) as the cutoff shrinks. The theorem predicts these probabilities converge to zero and the limit in (3.14) solves the mild equation (1.2) with finite moments of every order; a violation for any pair inside the stated range would falsify Theorem 3.1.","tokens_in":22636,"feed_emoji":"🔥","tokens_out":19769,"duration_ms":193956,"temperature":0.7,"pith_summary":"The paper proves that singularities at zero do not prevent well-posedness of the stochastic heat equation, as long as the singularity is logarithmic. It shows global existence, uniqueness, and strict positivity of the mild solution on the torus driven by space-time white noise when the drift satisfies $|b(z)| = O(z(\\log(1/z))^{A_1})$ with $A_1 < 1$ and the diffusion coefficient satisfies $|\\sigma(z)| = O(z(\\log(1/z))^{A_2})$ with $A_2 < 1/4$ near $z = 0$. These coefficients are not Lipschitz at zero: their Lipschitz constants diverge. The result matters because such logarithmic nonlinearities occur in nonlinear wave mechanics, and earlier theory either required the noise coefficient to be locally Lipschitz at zero or excluded a drift term. The same stopping-time construction also handles the critical drift exponent $A_1 = 1$ under an extra sign/monotonicity condition and permits superlinear growth at infinity.","feed_headline":"Stochastic heat equation survives logarithmic blow-up at zero","feed_subtitle":"Drift and noise can be singular at zero, yet a unique positive solution still exists.","key_machinery":"The central machinery is a stopping-time localization scheme, driven by estimate (3.11). For each cutoff $\\epsilon$, the coefficients are replaced by $b_\\epsilon(z)=b(\\epsilon)z/\\epsilon$ and $\\sigma_\\epsilon(z)=\\sigma(\\epsilon)z/\\epsilon$ on $(0,\\epsilon]$, making them globally Lipschitz with growth constants $O((\\log(1/\\epsilon))^{A_1})$ and $O((\\log(1/\\epsilon))^{A_2})$; the approximating equations have unique global solutions. The proof tracks the stopping time at which a rescaled solution with initial value $e^{-k}$ first dips to $e^{-(k+1)}$, bounding this probability through the Hölder modulus of a process $V^{(k+1)}$ via a moment inequality and a continuity argument. The load-bearing chain of estimates ends in (3.11): with $H=L_b+p^2L_\\sigma^4$, the probability that any ladder is crossed within the allotted time is bounded by $\\binom{2m}{m}\\exp(\\cdots)$, and because $A_1<1$ and $4A_2<1$ the exponent is dominated by $-\\beta\\lambda p m\\log m/4$, forcing the crossing times to diverge almost surely.","core_discovery":"On the paper's own terms, the central discovery is Theorem 3.1: for nonnegative bounded Hölder continuous initial data, the stochastic heat equation $\\partial_t u = \\frac12 \\partial_x^2 u + b(u) + \\sigma(u)\\dot W$ on $\\mathbb{T}=[0,1]$ has a global mild solution, nonnegative, with $\\sup_{t\\le T}\\sup_x E[u(t,x)^p]<\\infty$ for every $p\\ge 2$; and if $\\inf_x u_0(x)>0$, the solution is unique and strictly positive in $C([0,\\infty)\\times\\mathbb{T};\\mathbb{R})$. The solution is built by modifying $b$ and $\\sigma$ below a cutoff $\\epsilon$ so they become globally Lipschitz, then showing the first passage below $\\epsilon$ does not occur before a fixed time with probability tending to one. A comparison principle extends the construction to merely nonnegative initial data by approximating $u_0+1/n$ and taking a monotone limit. The quantitative engine is estimate (3.11), whose decay forces the cutoff-crossing times to diverge almost surely under exactly the assumptions $A_1<1$ and $4A_2<1$.","pith_inferences":["Extension — The paper leaves the sharpness of $A_2<1/4$ open; testing $\\sigma(z)=z(\\log(1/z))^{1/4}$ would show whether the threshold is real or only a limitation of the proof.","Extension — Because the proof relies on heat-kernel estimates on the torus rather than on the specific structure of the noise, the same localization scheme may carry over to colored noise or fractional Laplacians, though the authors do not claim this.","Extension — Strict positivity of the solution means it never touches zero at finite times; that suggests pathwise separation-from-zero and support properties that could be useful in comparison or population-model arguments, a direction the paper does not pursue."],"forward_implications":["For every bounded Hölder initial condition, the constructed solution is nonnegative, exists for all times, and has finite moments of every order uniformly on compact time intervals.","If the initial condition is bounded away from zero, the global solution is unique and strictly positive at every position and time.","With the same noise coefficient, smaller drift and smaller initial data yield a pointwise smaller solution almost surely; in particular the comparison principle covers the solutions built by localization even without the strict-positivity assumption.","The critical drift case $A_1=1$ admits a global solution for sign-definite $b$ satisfying (4.1), for example $b(z)=-z\\log(1/z)$.","Allowing superlinear growth at infinity as well, the same method yields a unique global solution for coefficients with $b(u)=O(u\\log u)$ and $\\sigma(u)=O(u(\\log u)^{1/4})$ as $u\\to\\infty$."],"supporting_citations":[{"why":"It supplies the stopping-time and moment framework for global solutions with superlinear drift and multiplicative noise, which the paper adapts to the zero-singularity regime and to growth at infinity.","marker":"[6]"},{"why":"It is the closest prior well-posedness result for a diffusion coefficient with a logarithmic blow-up at zero and no drift; the paper relaxes its monotonicity assumption and adds a drift term.","marker":"[11]"},{"why":"It provides the weak comparison theorem used to paste the localized solutions and to prove monotonicity in initial data and drift.","marker":"[14]"},{"why":"It supplies the heat-kernel increment estimates and semigroup bounds that drive the regularity and moment estimates.","marker":"[4]"},{"why":"It defines the stochastic integral against space-time white noise and the notion of random-field mild solution used throughout.","marker":"[30]"},{"why":"It establishes that a superlinear drift failing an integral growth condition can force finite-time blow-up, motivating the singular-zero regime studied here.","marker":"[7]"},{"why":"It gives an earlier existence result for a logarithmic drift under Brownian forcing; the paper extends that setting to space-time white noise and non-Lipschitz diffusion at zero.","marker":"[29]"},{"why":"It provides the standard Picard-iteration moment framework used in the preliminary estimates of Section 2.","marker":"[13]"}],"fun_headline_variants":["Heat equation survives log blow-up: unique positive solution","Positive heat solution despite log-singular drift and noise","Overcoming log blow-up: positive heat solution proven","Log blow-up at zero: heat equation finds unique solution"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that near zero the noise coefficient grows no faster than $u\\,(\\log(1/u))^{1/4}$ and the drift grows no faster than $u\\log(1/u)$; at or beyond that growth the proof's key probability bound no longer shrinks to zero, and the paper supplies no alternative argument.","fun_headline_variants_meta":{"raw":{"variants":["Heat equation survives log blow-up: unique positive solution","Positive heat solution despite log-singular drift and noise","Overcoming log blow-up: positive heat solution proven","Log blow-up at zero: heat equation finds unique solution"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000597,"raw_usage":{"total_tokens":2809,"prompt_tokens":979,"completion_tokens":1830,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":595,"completion_tokens_details":{"reasoning_tokens":1765}},"tokens_in":595,"tokens_out":1830,"duration_ms":15905,"temperature":1.0,"reasoning_tokens":1765,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T10:37:30.810856+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the construction for a coefficient pair inside the stated range, say $b(z)=z(\\log(1/z))^{0.9}$ and $\\sigma(z)=z(\\log(1/z))^{0.24}$ near zero with a bounded Hölder initial condition, and compute the cutoff-crossing probabilities in (3.8) as the cutoff shrinks. The theorem predicts these probabilities converge to zero and the limit in (3.14) solves the mild equation (1.2) with finite moments of every order; a violation for any pair inside the stated range would falsify Theorem 3.1.","supporting_citations":[{"cited_title":"On the support of solutions to nonlinear stochastic heat equations","cited_arxiv_id":"2407.06827","evidence_quote":"It is the closest prior well-posedness result for a diffusion coefficient with a logarithmic blow-up at zero and no drift; the paper relaxes its monotonicity assumption and adds a drift term."},{"cited_title":"Theory Related Fields 93 (1992), no","cited_arxiv_id":null,"evidence_quote":"It provides the weak comparison theorem used to paste the localized solutions and to prove monotonicity in initial data and drift."},{"cited_title":"Parabolic Anderson model with colored noise on torus","cited_arxiv_id":"2308.10802","evidence_quote":"It supplies the heat-kernel increment estimates and semigroup bounds that drive the regularity and moment estimates."},{"cited_title":"Walsh, An introduction to stochastic partial differential equations , ´Ecole d’´ et´ e de probabilit´ es de Saint-Flour, XIV—1984, 1986, pp","cited_arxiv_id":null,"evidence_quote":"It defines the stochastic integral against space-time white noise and the notion of random-field mild solution used throughout."},{"cited_title":"Differential Equations 313 (2022), 85–121, https://doi.org/10.1016/j.jde.2021.12.033","cited_arxiv_id":null,"evidence_quote":"It gives an earlier existence result for a logarithmic drift under Brownian forcing; the paper extends that setting to space-time white noise and non-Lipschitz diffusion at zero."},{"cited_title":"119, Published for the Conference Board of the Mathematical Sciences, Washington, DC; by the American Mathematical Society, Providence, RI, 2014","cited_arxiv_id":null,"evidence_quote":"It provides the standard Picard-iteration moment framework used in the preliminary estimates of Section 2."}],"review_version":1}