REVIEW 4 major objections 5 minor 11 references
On the well-posedness of SPDEs with locally Lipschitz coefficients
T0 review · 4 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash
Pith's one-line read This paper proves that the stochastic heat equation on the whole line with locally Lipschitz, linearly growing coefficients has a unique random-field solution under the slow-growth condition Assumption 1.3.
desk verdict Genuinely new existence theorem for the SHE on R with locally Lipschitz coefficients, but the uniqueness proof is an outline that currently does not close. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The engine is the moment bound of Proposition 2.1: $E|u^N(t,x)|^k \le 4^k(\|u_0\|_\infty+1)^k e^{128 L_\sigma^4 k^3 t}$ for the solution of the equation truncated at level $e^N$, obtained by combining the heat kernel, the Burkholder–Davis–Gundy inequality, and the choice $\beta=128k^2L_\sigma^4$. The bound is then converted, via Chebyshev's inequality with $k=c\sqrt N$, into the tail estimate $P(|u^{N+1}(t,x)|\ge e^N)\le \exp(-N^{3/2}/(64L_\sigma^2\sqrt t))$ of Proposition 2.4. These two estimates make the series $\sum_N \sup_{t,x}\|u^{N+1}(t,x)-u^N(t,x)\|_k$ converge: on the event $\{|u^{N+1}|\le e^N\}$ the difference is controlled by the local Lipschitz constant $L_{N,\sigma}$, while on its complement the tail estimate provides an exponentially small factor that beats any power of $N$.
What would settle it
Choose coefficients at the boundary of Assumption 1.3, e.g. with $L_{N,\sigma}=c N^{3/8}$ and $L_{N,b}=c L_{N,\sigma}^4$, and compute or bound $P(|u^{N+1}(t,x)|\ge e^N)$ for the truncated solutions; if this tail decays strictly slower than $\exp(-c' N^{3/2})$ for large $N$, then the series in (3.1) diverges and the proof's central convergence step fails.
Extended reading notes
Core claim
On the paper's own terms, the central discovery is Theorem 1.4: under Assumption 1.3 — $L_{N,\sigma}=o(N^{3/8})$, $L_{N,b}/L_{N,\sigma}^4=O(1)$ when $\sigma$ is unbounded, and analogous conditions when $\sigma$ is bounded — the SPDE $\partial_t u = \tfrac12 \partial_x^2 u + b(u) + \sigma(u) \dot W$ with bounded measurable initial data has a unique random-field solution with $\sup_{t\in[0,T]}\sup_{x\in\mathbb{R}} E|u(t,x)|^k < \infty$ for every $T>0$ and $k\ge1$. The proof constructs the solution as the $L^k$-limit of solutions to truncated equations and shows that the differences $\|u^{N+1}-u^N\|_k$ sum to a finite series; the limit is then shown to satisfy the mild equation. The novelty is that the truncation is pointwise in $(t,x)$, with tail probabilities of the truncated solution doing the work that stopping times do on bounded domains.
Load-bearing premise
The load-bearing premise is that the moment bound of Proposition 2.1, $E|u^N(t,x)|^k\le 4^k(\|u_0\|_\infty+1)^k e^{128L_\sigma^4k^3t}$, holds with exactly these constants; if the true growth in $k$ were any worse, the tail estimate would decay more slowly than $e^{-N^{3/2}}$ and the series of truncation differences would not converge.
Editorial extensions
If this is right
- The class of admissible coefficients includes oscillatory diffusions such as $\sigma(x)=\sin(1000(1+|x|)^{1/4})$, which are nowhere near globally Lipschitz; for these, any drift with Lip$_n(b)=o(n)$ is allowed.
- The proof works verbatim for time-dependent coefficients $b(t,x),\sigma(t,x)$ with uniform-in-time growth and oscillation bounds, so the result applies to non-autonomous heat equations.
- The same truncation-and-tail scheme yields well-posedness for the stochastic wave equation, as the authors state.
- When $\sigma$ is bounded, Assumption 1.3 becomes much milder: $\sigma$ may have Lip$_n(\sigma)=o(\sqrt n)$ and the drift may have Lip$_n(b)=o(n)$, substantially enlarging the tractable regime.
- The convergence in the series (3.1) is strong enough to give $L^k$-continuity and a predictable version of the limit, so the constructed object is a genuine random-field solution, not merely a weak solution.
Reading between the lines
- The sharp form of the Burkholder–Davis–Gundy inequality is what pins the moment exponent at $k^3$; with a coarser constant the tail decay would fall below $e^{-N^{3/2}}$ and the series (3.1) would not converge, so the method is sensitive to this constant.
- The paper's Remark 1.8 sketches a route to improving $o(N^{3/8})$ to $o(N^{2/3})$ by an alternative choice of the parameter $k(N)$; carrying this out would directly expand the class of admissible coefficients.
- The same exponential-truncation ladder with pointwise tail estimates should apply to other semilinear SPDEs on unbounded domains whenever the truncated solutions admit moment bounds whose growth in $k$ is at most cubic.
- Because the construction is purely $L^k$-based and never invokes a stopping time, it suggests that on the line the only obstruction to well-posedness is the growth of the local Lipschitz constants, not the unboundedness of the spatial domain.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper studies the stochastic heat equation on the real line driven by space-time white noise, with drift and diffusion coefficients that are locally Lipschitz and of at most linear growth. The main result, Theorem 1.4, asserts that under the rate conditions of Assumption 1.3 on the local Lipschitz constants, the equation has a unique random-field solution with finite moments of all orders. The method truncates the coefficients at levels exp(N), derives uniform moment bounds for the truncated solutions (Propositions 2.1 and 2.3), converts them into tail estimates (Proposition 2.4), and then proves summability of the L^k-norms of the successive truncation differences, yielding a limiting solution. Uniqueness is addressed by applying the same estimates to the difference of two solutions.
Significance. If the proof is completed, the paper makes a substantial contribution: it gives the first well-posedness result on the unbounded spatial domain R for locally Lipschitz coefficients, a setting in which the classical stopping-time argument fails because the solution is expected to be unbounded. The pointwise truncation mechanism and the explicit moment/tail estimates are natural and potentially useful for related equations. The result is, however, conditional on Assumption 1.3, which imposes logarithmic rate conditions on the local Lipschitz constants; it is therefore not a proof of well-posedness under local Lipschitz and linear growth alone, as the opening sentence of the abstract might suggest. The paper also contains a clear, detailed existence argument with explicit constants, which is a strength.
major comments (4)
- [3 (Proof of uniqueness)] The concluding inference of the uniqueness proof is not justified as written. The preceding estimate controls the weighted norm N_{c,beta_N,T}(u-v) with beta_N = 16 A_0^4 c^2 L_{N,sigma}^4, which tends to infinity with N. Smallness of the weighted norm does not imply smallness of the unweighted sup-norm, because e^{-beta_N t} tends to 0 exponentially. The statement that 'sum_{N=1}^infty sup_{t,x} ||u(t,x)-v(t,x)||_2 < infinity, and since the summand does not depend on N, it must be zero' therefore does not follow from the displayed estimates. The argument can likely be repaired by applying Lemma 2.5, exactly as in the existence proof, to convert the weighted bound into a bound of the form sup_{t<=T,x} ||u-v||_c <= C(T) exp(512 L_sigma^4 c^2 T - N^{3/2}/(128 c L_sigma^2 sqrt(T))), and then letting N tend to infinity. This conversion is load-bearing and needs to be written out.
- [2, Eq. (2.7)] There is a mismatch in the Burkholder-Davis-Gundy inequality as displayed: Eq. (2.7) bounds the stochastic integral by an integrand involving ||sigma^N(...)||_{2k}^2, whereas the standard BDG inequality for Walsh integrals (as used, for example, in the cited lecture notes) gives ||f||_k^2, not ||f||_{2k}^2. With the subscript 2k, the subsequent step replacing ||u^N(s,y)||_{2k}^2 by N_{k,beta}(u^N)^2 e^{2 beta s} is invalid; at best one obtains a bound involving N_{2k,beta}(u^N), and the fixed-point inequality for N_{k,beta}(u^N) would not close. The same pattern appears in the I_2 and I_4 displays in the uniqueness proof. If the 2k is a typo, it should be corrected throughout; if not, the moment bound in Proposition 2.1 requires a different argument.
- [3 (Proof of uniqueness)] The uniqueness proof transfers Propositions 2.1 and 2.4 from the truncated solutions u^N to arbitrary solutions u and v by saying that the bounds 'only use the linear growth constants' and therefore hold for u and v as well. This transfer is not automatic: those propositions are proved for u^N, and the proof uses properties specific to the truncated mild equation. The theorem should explicitly define the uniqueness class, for instance the class of random-field solutions satisfying sup_{t in [0,T]} sup_x E|u(t,x)|^k < infinity for all T>0 and k>=1, and the moment/tail bounds should either be proved directly for every solution in that class or the transfer should be justified. Without this, the uniqueness claim is ambiguous and the proof is circular.
- [3, after Eq. (3.13)] The passage from the estimate for a fixed moment order c to the claimed summability (3.1) for all k>=1 is stated very briefly. If c is chosen larger than a given k, the L^k norm is bounded by the L^c norm, so the argument is recoverable, but the manuscript should say this explicitly because c appears in (3.6) and (3.9) as a fixed constant and the relabeling step is not immediate.
minor comments (5)
- [Abstract] The first sentence says the equation is well posed 'solely' under local Lipschitz continuity and at most linear growth; this is stronger than Theorem 1.4, which requires the rate conditions in Assumption 1.3. The full-text abstract contains the qualifier 'with regularly behaved local Lipschitz constants,' but the metadata version omits it. The abstract should be worded so that the role of Assumption 1.3 is clear.
- [2, Remark 2.2] The claim that 'we can always choose, without incurring loss of generality, L_sigma large enough' is not a valid reduction, since L_sigma is fixed by the problem. The proposition's lower bound on k is acceptable as it stands; the remark should be deleted or rewritten.
- [3, Eq. (3.13)] In Eq. (3.13), the expression ||u^{N+1}(t,x)-u^{N}(t)||_k appears to have a typo; it should be ||u^{N+1}(t,x)-u^{N}(t,x)||_k.
- [3, convergence paragraph] There are two occurrences of 'y in R^d' in the convergence part of the existence proof; the spatial domain in this paper is R, not R^d.
- [3, after Eq. (3.5)] The constant in the displayed error term after (3.5) differs from a direct integration by a factor of 2 (one obtains 1/(256 L_sigma^4 k^2) rather than 1/(128 L_sigma^4 k^2)). Since all constants are non-explicit and the exponential rate is what matters, this is harmless, but the computation should be checked.
Circularity Check
No circular derivation: the theorem is proved from explicit hypotheses using standard SPDE estimates, and the only author-overlapping citation is a standard inequality that does not make the argument circular.
full rationale
The paper's derivation chain is not circular. Existence is proved by constructing truncated solutions u^N, proving explicit moment and tail estimates (Propositions 2.1, 2.3, 2.4), and then showing the series of differences in (3.1) converges via the N-dependent bounds in (3.11) and (3.13). The constants depend on the hypotheses L_b, L_sigma, L_{N,b}, and L_{N,sigma}; no parameter is fitted to the target conclusion, and Assumption 1.3 is a sufficient hypothesis rather than a renamed version of well-posedness. The uniqueness proof is only outlined and contains real gaps: it asserts without full detail that the moment and tail estimates extend to arbitrary solutions u and v, and the final inference that a constant summand must be zero is compressed. These are proof omissions, not circularity, because uniqueness does not assume the conclusion it establishes. The only citation involving a present author is [3] for the Burkholder-Davis-Gundy inequality, which is a standard published result and is used for its explicit constant; it is independent mathematical support and does not raise the circularity score. Remark 2.2's claim that one may assume L_sigma is large is mathematically suspect, but it is an erroneous technical reduction, not a circular step. Overall, the central claim has genuine independent content and the proof does not reduce to its inputs by construction.
Assumptions & free parameters
assumptions (6)
- standard math The Walsh stochastic integral against space-time white noise is well defined, and the heat kernel satisfies ||p_r||^2_{L2(R)} = 1/(2 sqrt(pi r)).
- standard math The Burkholder-Davis-Gundy inequality for Walsh stochastic integrals holds with the asymptotically optimal constant 4k, quoted from [3].
- standard math For globally Lipschitz truncations b^N and sigma^N, the SPDE (2.1) has a unique predictable mild solution with finite moments of all orders, by standard theory [2,11] and the forthcoming monograph [4, Theorems 4.2.1 and 4.2.8].
- domain assumption Assumptions 1.1 and 1.2: u0 is bounded and measurable; b and sigma are locally Lipschitz in the space variable uniformly in time, with at most linear growth.
- ad hoc to paper Assumption 1.3: L_{N,sigma}=o(N^{3/8}) and L_{N,b}/L^4_{N,sigma}=O(1) when L_sigma>0; or L_{N,sigma}=o(e^{N/2}) with the same ratio condition when sigma is bounded.
- ad hoc to paper Remark 2.2's claim that L_sigma can be chosen large enough to ensure L_b^{1/2} L_sigma^{-2} <= 2.
Cite this review
Pith. "Pith review of On the well-posedness of SPDEs with locally Lipschitz coefficients." pith.science (2026). https://pith.science/paper/C6EOYUJ7
@misc{pith2026241109381,
author = {Pith},
title = {Pith review of: On the well-posedness of SPDEs with locally Lipschitz coefficients},
year = {2026},
howpublished = {\url{https://pith.science/paper/C6EOYUJ7}},
note = {Machine review of arXiv:2411.09381}
}
abstract
We consider the stochastic partial differential equation, $\partial_t u = \tfrac12 \partial^2_x u + b(u) + \sigma(u) \dot{W},$ where $u=u(t\,,x)$ is defined for $(t\,,x)\in(0\,,\infty)\times\mathbb{R}$, and $\dot{W}$ denotes space-time white noise. We prove that this SPDE is well posed solely under the assumptions that the initial condition $u(0)$ is bounded and measurable, and $b$ and $\sigma$ are locally Lipschitz continuous functions and have at most linear growth. Our method is based on a truncation argument together with moment bounds and tail estimates of the truncated solution. The results naturally generalize to the case where $b$ and $\sigma$ are time dependent with uniform-in-time growth and oscillation properties. Additionally, our method can be extended to the stochastic wave equation.
Figures
Reference graph
Works this paper leans on
-
[1]
Le Chen, Mohammud Foondun, Jingyu Huang, and Michael Salins,Global solution for su- perlinear stochastic heat equation onRd under Osgood-type conditions, Preprint available at https://arxiv.org/abs/2310.02153(2023)
arXiv 2023
-
[2]
Robert C. Dalang,Extending the martingale measure stochastic integral with applications to spatially homogeneous s.p.d.e.’s, Electron. J. Probab.4(1999), no. 6, 29 pp. (electronic). MR1684157 (2000b:60132)
work page 1999
-
[3]
Robert C. Dalang, Davar Khoshnevisan, Carl Mueller, David Nualart, and Yimin Xiao,A Minicourse on Stochastic Partial Differential Equations(Khoshnevisan and Firas Rassoul- Agha, eds.), Lecture Notes in Mathematics, vol. 1962, Springer-Verlag, Berlin, 2009. MR1500166 (2009k:60009)
work page 1962
-
[4]
Robert C. Dalang and Marta Sanz-Solé,Stochastic Partial Differential Equations, Space- Time White Noise and Random Fields(forthcoming book), Current version available at https://arxiv.org/abs/2402.02119, 2024
arXiv 2024
-
[5]
István Gyöngy and Nicolai Krylov,Existence of strong solutions for Itô’s stochastic equations via approximations, Probability Theory and Related Fields105(1996), no. 2, 143–158
work page 1996
-
[6]
Thomas Kurtz,Weak and strong solutions of general stochastic models, Electronic Commu- nications in Probability19(2014), 1–16
work page 2014
-
[7]
Leonid Mytnik, Edwin Perkins, and Anja Sturm,On pathwise uniqueness for stochastic heat equations with non-Lipschitz coefficients, The Annals of Probability34(2006), no. 5, 1910– 1959
work page 2006
-
[8]
Daniel Revuz and Marc Yor,Continuous Martingales and Brownian Motion(third edition), Springer, Heidelberg, 1999. MR1725357 (2000h:60050)
work page 1999
Show all 11 references
-
[9]
Dyn.22(2022), no
Michael Salins,Existence and uniqueness of global solutions to the stochastic heat equation with superlinear drift on an unbounded spatial domain, Stoch. Dyn.22(2022), no. 5, Paper No. 2250014, 30. MR4486234
2022
-
[10]
2, 415–437
Tokuzo Shiga,Two contrasting properties of solutions for one-dimensional stochastic partial differential equations46(1994), no. 2, 415–437
1994
-
[11]
Walsh,An Introduction to Stochastic Partial Differential Equations, École d’été de Probabilités de Saint-Flour, XIV–1984, 1986, pp
John B. Walsh,An Introduction to Stochastic Partial Differential Equations, École d’été de Probabilités de Saint-Flour, XIV–1984, 1986, pp. 265–439. MR876085 (88a:60114) SPDES WITH LOCALLY LIPSCHITZ COEFFICIENTS 17 University of Strathclyde Email address:mohammud.foondun@strat...
1984
Reviewed August 12, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.