Pith. sign in

REVIEW 1 major objections 6 minor 22 references

Convergence in law for quasi-linear SPDEs

T0 review · 1 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read The paper proves that the mild solutions of quasi-linear stochastic wave and heat equations driven by Gaussian noises whose spectral measures converge in a weighted vague sense converge in law to the solution driven by the limit noise.

desk verdict A strong, broadly applicable convergence-in-law theorem for quasi-linear SPDEs with rough noise; the wave and bounded-drift cases are solid, the general-drift heat case has a repairable gap in the truncation argument. read the letter →

arxiv 2505.22493 v2 pith:RATZKOWK submitted 2025-05-28 math.PR

Maria Jolis , Salvador Ortiz-Latorre , Lluís Quer-Sardanyons This is my paper
classification math.PR MSC 60H1560B1060G60
keywords stochasticwaveequationheatweakconvergencespectralmeasuremildsolutionspatiallyhomogeneousGaussiannoisefractionalRieszkernel
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper proves a stability theorem for stochastic PDEs: if a sequence of spatially homogeneous Gaussian noise fields has spectral measures $\mu_n$ that converge to a limit $\mu$ in the sense of weighted vague convergence, and the family satisfies a uniform tail bound (Hypothesis (H1)), then the mild solutions $u^n$ of the quasi-linear stochastic wave equation (dimensions $d\le 3$) and heat equation ($d\ge 1$) with additive Gaussian noise converge in law, in the space $C([0,T]\times\mathbb{R}^d)$, to the solution $u$ driven by the noise with spectral measure $\mu$. The result matters because it turns convergence of the noise's spatial correlation into convergence of the law of the full nonlinear solution, without requiring the Fourier transform of $\mu_n$ to be a function. The proof reduces the problem to the linear stochastic convolution $v^n$ and a continuous deterministic mapping induced by the drift. Applications include fractional noise in space (including Hurst parameters below $1/2$, where the Fourier transform is a genuine distribution), anisotropic fractional noise, and Riesz-kernel noise.

What carries the argument

The engine is the representation $u^n = F\circ T_{I_0}(v^n)$, where $v^n$ is the Gaussian stochastic convolution (the solution with zero drift and zero initial data) and $F$ maps a continuous path $\eta$ to the solution of the deterministic integral equation $z=\eta+G\star b(z)$. A mild solution is the random field solving this integral equation, with $G$ the fundamental solution of the wave or heat operator. $F$ and the translation $T_{I_0}$ are continuous on $C([0,T]\times\mathbb{R}^d)$, proved through two ad-hoc Gronwall lemmas (one adapted to the wave kernel, one to the bounded-drift heat kernel). Consequently the nonlinear convergence is inherited from the convergence in law of the Gaussian fields $v^n$, which is obtained by proving tightness via a multidimensional Kolmogorov-type criterion (stated and proved in an appendix) and identifying the limit via convergence of covariance functions. For the heat equation with general Lipschitz drift, the deterministic equation is not well posed, so the proof instead uses uniform Hölder-type estimates and a drift-truncation argument to identify finite-dimensional limits.

What would settle it

The decisive check is to search for a sequence $\mu_n$ satisfying (H1) and (H2) whose covariance functions converge but whose second-moment increment estimates (25)--(26) are not uniform in $n$; Proposition 3.3 asserts this cannot happen, so finding such a sequence would falsify the tightness step of Theorem 2.8, while proving the estimates in a new example would confirm the mechanism.

Watch

Extended reading notes

Core claim

The central claim is Theorem 2.8. Fix the fundamental solution $G$ of the wave operator ($d\in\{1,2,3\}$) or heat operator ($d\ge 1$), a globally Lipschitz drift $b$, and initial data satisfying the hypotheses in the statement. Under (H1) and (H2), the mild solution $u^n$ of $u^n = I_0 + G\star \dot W^n + G\star b(u^n)$ converges in law in $C([0,T]\times\mathbb{R}^d)$ to the unique mild solution $u$ of the analogous equation driven by a Gaussian spatially homogeneous noise with spectral measure $\mu$. Hypothesis (H1) is the uniform bound $\sup_n\int_{\mathbb{R}^d}\mu_n(d\xi)/(1+|\xi|^q)<\infty$ for some $q\in(0,2)$; Hypothesis (H2) is the convergence of $\int f\,d\mu_n$ to $\int f\,d\mu$ for continuous $f$ with $|f(\xi)|\le C/(1+|\xi|^2)$. The same theorem covers the linear case (zero drift and zero initial data), and the general statement is assembled from that linear case by path-by-path arguments, with the heat equation and unbounded Lipschitz drift handled by tightness plus convergence of finite-dimensional distributions.

Load-bearing premise

The load-bearing premise is that the noises' spatial spectral measures have a uniform tail bound: there is one exponent $q\in(0,2)$ such that $\sup_n\int_{\mathbb{R}^d}\mu_n(d\xi)/(1+|\xi|^q)<\infty$; if that fails, the family of stochastic convolutions may lose tightness and the whole convergence argument collapses.

Editorial extensions

If this is right

  • If the spectral measures converge in the stated sense and the uniform tail bound holds, the law of the solution depends continuously on the spatial covariance of the noise.
  • In space dimension one this recovers and generalizes the previously known continuity in law with respect to the fractional-noise parameter, now covering Hurst parameters in the full interval $(0,1)$.
  • In dimensions $d\ge 2$, the theorem applies to anisotropic fractional noise when the parameter sum stays above $d-1$, and to Riesz-kernel noise with $\alpha\in(0,2)$.
  • Because the convergence takes place in the continuous-path space $C([0,T]\times\mathbb{R}^d)$, continuous functionals such as suprema, level sets, and exit times of the solution converge in law as well.
  • The limit law depends only on $\mu$, not on the particular sequence $\mu_n$, so any approximating sequence satisfying (H1)--(H2) produces the same limit.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The continuous-mapping structure suggests the result should extend to other parametric families of spatially homogeneous noises: whenever the linear stochastic convolution converges in law and the deterministic fixed-point map is continuous, the same conclusion holds, so (H1)--(H2) are likely sufficient rather than necessary conditions.
  • The paper shows the isotropic fractional spectral measure fails the standard integrability condition unless $d=1$; an inference is that continuity in law for isotropic fractional noise would require either spatial smoothing in the equation or a different solution concept.
  • A natural testable extension is the multiplicative-noise analogue (Anderson-type models), which the authors explicitly leave for future work; the same confluence of tightness and covariance convergence may hold, but the path-by-path representation would have to be replaced.
  • The appendix's tightness criterion for random fields indexed by rectangles is reusable and may be of independent interest for proving convergence in law of other random fields without available proofs in the literature.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

1 major / 6 minor

Summary. This paper studies the mild solutions u^n of the quasi-linear stochastic wave equation in d <= 3 and the stochastic heat equation in d >= 1, driven by additive space-time homogeneous Gaussian noise whose spatial spectral measure mu_n is a tempered measure, possibly with a Fourier transform that is a genuine distribution. Under Hypothesis (H1), a uniform q-integrability bound sup_n integral mu_n(dxi)/(1+|xi|^q) < infinity for some q in (0,2), and Hypothesis (H2), a weighted vague convergence of mu_n to mu, together with explicit assumptions on the deterministic initial data, Theorem 2.8 asserts that u^n converges in law in C([0,T] x R^d) to the mild solution driven by a noise with spectral measure mu. The proof proceeds in three stages: the linear case (Section 3) establishes tightness from (H1) and identifies the Gaussian limit from (H2); the wave equation and the heat equation with bounded drift are handled by a pathwise deterministic representation u^n = F(u_0, v_0 + v^n) with a continuous functional F (Sections 4.1 and 4.2); the heat equation with general Lipschitz drift is handled by tightness (Proposition 5.2) and by a truncation of the drift for the finite-dimensional distribution identification (Proposition 5.3).

Significance. If the identified gap in Proposition 5.3 is closed, Theorem 2.8 is a genuinely useful unified convergence-in-law result: it covers rough spatial noises (fractional Hurst parameters below 1/2, whose spectral measures are not covered by Dalang's function-setting), treats all wave dimensions d <= 3 and heat dimensions d >= 1 under the same hypotheses, and includes the d = 1 fractional case of [11], the Riesz-kernel cases of [2,19], and the anisotropic fractional Brownian sheet in d >= 2 as worked examples. The paper is largely self-contained and unusually explicit about its infrastructure: Appendix A supplies a full proof of the multidimensional tightness criterion that is invoked repeatedly, Section 3 proves the linear case from scratch, and the path-space representation for the wave and bounded-drift heat cases is clean and yields convergence by the continuous mapping theorem. The spectral measure of the isotropic (Levy) fractional Brownian sheet is derived in Section 2.3.3 rather than taken from the literature.

major comments (1)
  1. [Proposition 5.3, Eq. (71); Section 5] The proof of the truncation estimate (71) is incomplete as written. The text states that the only auxiliary input needed is sup_n sup_{(t,x)} E|u^n(t,x)|^p < infinity and that this estimate "has been proved in Lemma 3.1"; however, Lemma 3.1 bounds the moments of the linear stochastic convolution v^n only, not those of the full solution u^n of the quasi-linear heat equation. A uniform-in-n moment bound for u^n requires a separate Gronwall argument applied to the mild equation (66), using the boundedness of I^d_0 and the uniform bound for v^n; moreover, in the drift comparison the term |b_m(u^n) - b(u^n)| vanishes in L^2 uniformly in n only if the moment bound holds for some p > 2. Since (71) is precisely what identifies the limit of the finite-dimensional distributions in case (c) of Theorem 2.8, this is a load-bearing gap rather than a cosmetic one. I consider it repairable within the scope of the paper, but the Gronwall step must be written out explicitly.
minor comments (6)
  1. [Lemma 3.1] In the displayed chain of inequalities, the comparison integral mu_n(dxi)/(1+|xi|^2) <= integral mu_n(dxi)/(1+|xi|^q) is not valid pointwise on {|xi| <= 1}: there 1/(1+|xi|^2) >= 1/(1+|xi|^q). The intended uniform bound still holds, for instance by estimating the unit-ball part separately via mu_n(B_1) <= 2 integral_{|xi|<=1} mu_n(dxi)/(1+|xi|^q), so the conclusion of the lemma is unaffected, but a factor-2 argument is needed.
  2. [Proof of Proposition 5.3] The sentence "An immediate consequence of (b) in Theorem 2.8" refers to a case of the very theorem being proved; since the bounded-drift heat case has been established earlier in Section 5 (via the continuous mapping argument), the reference should point to that already-proved statement or make clear that the bounded-drift sub-result is being used.
  3. [Section 4.1, proof of Theorem 4.2] There is a typo: "Sept 1" should read "Step 1".
  4. [Section 2.3.2] In the dominated-convergence argument for the anisotropic fractional noise, the phrase "replacing H^n_j by H^n_0" should read something like "replacing H^n_j by H^0_j"; the subscript and superscript are interchanged.
  5. [Section 4.2, Lemma 4.7] The proof of Lemma 4.7 is omitted with the justification that it follows the lines of Lemma 4.5; since the lemma is used to establish well-posedness of the deterministic equation (56) in the bounded-drift heat case, I recommend including the short proof or giving a precise pointer to the analogous argument.
  6. [Introduction and Section 2.2] There are a few typos: "the the main result" appears in the Introduction and again at the start of Section 2.2, and "Without loosing any generality" in the proof of Proposition 5.3 should read "losing".

Circularity Check

0 steps flagged · score 2.0 of 10

No constructional circularity: Theorem 2.8 follows from explicit sufficient hypotheses (H1)-(H2) with independent tightness and covariance arguments; only minor self-citations to [11] appear, plus a repairable proof gap that is not circularity.

full rationale

No significant circularity is present. Theorem 2.8 is not equivalent to its hypotheses by construction: (H1) and (H2) are stated sufficient conditions on the spectral measures, and the proof supplies independent content through the tightness estimates of Proposition 3.3, the covariance identification of Proposition 3.5, and the continuous-mapping argument for the wave and bounded-drift heat cases. The convergence v^n => v is not merely a restatement of (H2): the paper proves tightness and separately verifies that the covariance kernels I(ξ) and J(ξ) satisfy the admissible-function bound required by (H2). The paper cites [11], which shares two authors, for the one-dimensional fractional-noise case and for auxiliary Gronwall and truncation steps, but the central higher-dimensional and rough-noise claims do not reduce to [11]; they are established by the paper's own estimates. The proof of estimate (71) in Proposition 5.3 is delegated to [11, Sec. 4.3] and invokes Lemma 3.1, which as written bounds v^n rather than the full solution u^n; this is a correctness gap that would make the general-drift heat case incomplete as stated, since a uniform moment bound for u^n requires an additional Gronwall argument. This is a proof gap, not a circular identification of the theorem's conclusion with its assumptions. The reversed inequality in Lemma 3.1 is likewise an easily repaired technical error and does not make the argument circular. The overall score of 2 reflects only the presence of minor self-citations and no constructional circularity.

Assumptions & free parameters 0 free parameters · 4 assumptions · 0 invented entities

The paper introduces no new entities or fitted parameters. Its central claim rests on the standard SPDE integration theory, on a distribution-valued extension of the stochastic integral, and on the uniform integrability hypothesis (H1), which is an input condition rather than a fitted quantity.

assumptions (4)
  • domain assumption Extension of the stochastic integral to spectral measures whose Fourier transform is a genuine distribution, via [14, Lem. 3.2] without its non-negativity hypothesis.
    Invoked in Section 1 (around Eq. (2)) to cover fractional noise with H<1/2; the paper asserts the condition is unnecessary but does not supply the modified proof.
  • standard math Dalang's stochastic integration theory for spatially homogeneous Gaussian noise, including moment formulas used in Lemma 3.1.
    Used throughout Sections 3 and 4; placed on the cited results [5], [7], [11].
  • standard math Kolmogorov continuity criterion and Garsia-Rodemich-Rumsey lemma for random fields.
    Used in Theorem A.1 and in path-continuity proofs (Sections 3 and 4).
  • domain assumption Existence and uniqueness of mild solutions for each fixed n under Dalang's condition (Theorem 4.1), proved via Picard iteration with steps delegated to [11, Thm. 3.1].
    Used to define u^n and ensure L^p moment finiteness; the uniform-in-n version needed in Proposition 5.3 is not explicitly proved.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Convergence in law for quasi-linear SPDEs." pith.science (2026). https://pith.science/paper/RATZKOWK

@misc{pith2026250522493,
  author       = {Pith},
  title        = {Pith review of: Convergence in law for quasi-linear SPDEs},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RATZKOWK}},
  note         = {Machine review of arXiv:2505.22493}
}
abstract

We consider the quasi-linear stochastic wave and heat equations in $\mathbb{R}^d$ with $d\in \{1,2,3\}$ and $d\geq 1$, respectively, and perturbed by an additive Gaussian noise which is white in time and has a homogeneous spatial correlation with spectral measure $\mu_n$. We allow the Fourier transform of $\mu_n$ to be a genuine distribution. Let $u^n$ be the mild solution to these equations. We provide sufficient conditions on the measures $\mu_n$ and the initial data to ensure that $u^n$ converges in law, in the space of continuous functions, to the solution of our equations driven by a noise with spectral measure $\mu$, where $\mu_n\to\mu$ in some sense. We apply our main result to various types of noises, such as the anisotropic fractional noise. We also show that we cover existing results in the literature, such as the case of Riesz kernels and the fractional noise with $d=1$.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

22 extracted references · 22 canonical work pages

  1. [11]

    L. C. Evans. Partial Differential Equations. American Mathematical Society, Providence, RI, 1998

    work page 1998

  2. [1]

    (71) The proof of the above convergence follows exactly in the same way as in Step 2 of [11, Sec. 4.3]. The only needed auxiliary result is that, for all p ≥ 2, it holds: sup n≥1 sup (t,x)∈[0,T ]×Rd E [|un(t, x)|p] < ∞. This estimate has been proved in Lemma 3.1. Using the same arguments, one also shows that sup (t,x)∈[0,T ]×Rd E |um(t, x) − u(t, x)|2 − −...

    work page

  3. [2]

    The proof is similar to that of Step 3 in [11, Sec

    (72) 37 At this point, we have all the ingredients to show that the finite-dimensional distributions of un converge to those ofu. The proof is similar to that of Step 3 in [11, Sec. 4.3]. We will give it for the sake of completeness. Let (t1, x1), . . . ,(tk, xk) ∈ [0, T] × Rd and f : Rk → R be continuous and bounded. Then, we write E f un(t1, x1), . . . ...

    work page

  4. [3]

    , un(tk, xk)) − f (u(t1, x1),

    Hence, E [f (un(t1, x1), . . . , un(tk, xk)) − f (u(t1, x1), . . . , u(tk, xk))] < ε. Since ε can be taken arbitrary small, we can conclude the proof. A Tightness criterion In the paper, we have made use of the following tightness criterion several times. Although this result seems to be well-known, we have not been able to find a proof in the literature,...

    work page

  5. [4]

    Balan and X

    R.M. Balan and X. Liang. Continuity in law for solutions of SPDEs with space-time homogeneous Gaussian noise. Stochastics and Dynamics 23 (2023), no. 6

    work page 2023

  6. [5]

    P. Bezdek. On weak convergence of stochastic heat equation with colored noise. Stochastic Processes and their Applications 126 (2016) 2860–2875

    work page 2016

  7. [6]

    Billingsley

    P. Billingsley. Convergence of Probability Measures, 2nd ed. Wiley Series in Probability and Statis- tics: Probability and Statistics. New York: Wiley, 1999

    work page 1999

  8. [7]

    Conus and R.C

    D. Conus and R.C. Dalang. The non-linear stochastic wave equation in high dimensions. Electronic Journal of Probability 13 (2008)

    work page 2008

Show all 22 references
  1. [8]

    R.C. Dalang. Extending the martingale measure stochastic integral with applications to spatially homogeneous s.p.d.e’s. Electron. J. Probab. 4 (1999), 1-29

    1999

  2. [9]

    Dalang, D

    R.C. Dalang, D. Khoshnevisan and E. Nualart. Hitting probabilities for systems of non-linear stochastic heat equations with additive noise. ALEA, Lat. Am. J. Probab. Math. Stat. 3 (2007) 231- 271

    2007

  3. [10]

    Dalang and L

    R.C. Dalang and L. Quer-Sardanyons. Stochastic integrals for spde’s: A comparison. Expositiones Mathematicae 29 (2011) 67–109

    2011

  4. [12]

    G.B. Folland. Introduction to Partial Differential Equations, Princeton Univ. Press, 1976

    1976

  5. [13]

    Gel’fand and G.E

    I.M. Gel’fand and G.E. Shilov. Generalized Functions. V ol. 1. Properties and Operations. Academic Press, New York, 1964

    1964

  6. [14]

    Giordano, M

    L.M. Giordano, M. Jolis and L. Quer-Sardanyons. SPDEs with fractional noise in space: Continuity in law with respect to the Hurst index. Bernoulli 26 (2020), 352-386

    2020

  7. [15]

    Giordano, M

    L.M. Giordano, M. Jolis and L. Quer-Sardanyons. SPDEs with linear multi- plicative fractional noise: continuity in law with respect to the Hurst index. Stoch. Proc. Their Appl. 130 (2020), 7396- 7430

    2020

  8. [16]

    Millet and M

    A. Millet and M. Sanz-Solé. A stochastic wave equation in two space dimension: Smoothness of the law. Ann. Probab. 27 (1999), no. 2, 803-844

    1999

  9. [17]

    Nualart and L

    D. Nualart and L. Quer-Sardanyons. Existence and smoothness of the density for spatially homo- geneous spde’s. Potential Analysis 27 (2007), 281-299

    2007

  10. [18]

    Samorodnitski and M.S

    G. Samorodnitski and M.S. Taqqu. Stable non-Gaussian random processes, Chapman & Hall, Lon- don,1994. Paris, 1966

    1994

  11. [19]

    Sanz-Solé and M

    M. Sanz-Solé and M. Sarrà. Path properties of a class of Gaussian processes with applications to spde’s. Canadian Mathematical Society, Conference Proceedings 28 (2000) 303-316. 41

    2000

  12. [20]

    Sanz-Solé and M

    M. Sanz-Solé and M. Sarrà. Hölder Continuity for the Stochastic Heat Equation With Spatially Correlated Noise. In: Dalang, R.C., Dozzi, M., Russo, F. (eds) Seminar on Stochastic Analysis, Random Fields and Applications III. Progress in Probability, vol 52. Birkhäuser, Basel, 2002

    2002

  13. [21]

    Schwartz

    L. Schwartz. Théorie des Distributions, Hermann, Paris, 1966

    1966

  14. [22]

    W. Tao. On weak convergence of stochastic wave equation with colored noise onR. ArXiv Preprint arXiv:2408.10326 42

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.