Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

Explicit modified Euler methods keep superlinear SDE laws within O(τ|ln τ|) of the invariant measure, uniformly in time.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

Explicit modified Euler methods for superlinear multiplicative-noise SDEs are shown to converge in W1 distance to the invariant measure with rate τ|lnτ| under contractivity at infinity.

T0 review reviewed 2026-08-04 challenge →

load-bearing objection This paper has a real new result—non-asymptotic W1 bound for explicit modified Euler schemes for superlinear multiplicative-noise SODEs under contractivity at infinity—but the proof of Lemma 3.5(2) relies on an unverified transfer via [17, Theorem 4.5] that needs checking. the 3 major comments →

arxiv 2509.08410 v1 pith:XEGEHWX2 submitted 2025-09-10 math.NA cs.NA

Non-asymptotic Error Analysis of Explicit Modified Euler Methods for Superlinear and Non-contractive SODEs

Zhihui Liu , Xiaojie Wang , Xiaoming Wu , Xiaoyan Zhang This is my paper
classification math.NA cs.NA MSC 60H3537M2565C30
keywords explicit modified Euler methodnumerical Lyapunov structureuniform-in-time weak error estimateWasserstein distancesuperlinear SDEmultiplicative noiseinvariant measurenon-asymptotic error analysis
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The reading

This paper constructs a family of explicit time-stepping schemes—modified Euler methods that tame or project the coefficients—for stochastic differential equations whose drift and diffusion grow faster than linearly. It proves that, under a contractivity-at-infinity condition, the law of the numerical solution stays within Wasserstein-1 distance C* e^{-λτ n} + C* τ|ln τ| of the true invariant measure for every step size τ and every time step n. The exponential term decays with time, so the uniform-in-time weak error is O(τ|ln τ|), independent of how long the simulation runs. This gives the same O(τ|ln τ|) rate between the numerical and exact invariant measures, answering a question left open in earlier work on tamed schemes.

Core claim

The central claim is Theorem 2.3: for the explicit modified Euler family (2.8), under Assumptions 2.1–2.5, W1(L(Y_n^{x0}), π) ≤ C* e^{-λτ n} + C* τ|ln τ| for every n and every τ ∈ (0,1). The first term is the known exponential contraction of the continuous dynamics toward its invariant measure; the second is a uniform-in-time weak error of the scheme, proved for Lipschitz test functions only. Corollary 2.4 then gives |π(φ) − π_τ(φ)| ≤ C τ|ln τ| for all 1-Lipschitz φ, i.e., the numerical invariant measure converges to the exact one at the same O(τ|ln τ|) rate, resolving the question left open in [11].

What carries the argument

The machinery is the explicit modified Euler method (2.8): a state Y_n is first mapped by P (identity for the tamed scheme, radial truncation for the projected scheme), then advanced with modified coefficients b_τ, σ_{j,τ}. Assumptions 2.4–2.5 encode that this family preserves the SDE's Lyapunov structure, yielding uniform 2p-th moment bounds for the numerical chain (Theorem 3.2). For the weak error, the paper uses the Kolmogorov equation and the Bismut–Elworthy–Li formula to control derivatives of u(t,x)=E φ(X_t^x) with only Lipschitz φ; Lemma 3.5 gives t^{-1/2} singular bounds for t≤1 and exponential decay for t>1. Telescoping one-step weak errors over N steps then yields the τ|ln τ| bound

Load-bearing premise

The load-bearing premise is Assumption 2.3—the diffusion stays uniformly elliptic and the drift becomes contractive only at infinity; without it, the exponential contraction toward the invariant measure and the time-independent derivative decay used in the proof are not available, and the uniform W1 bound collapses to a time-dependent weak error.

What would settle it

Simulate a scalar SDE with superlinear drift and a diffusion that vanishes at infinity, e.g., dx = (−x^3 + x) dt + (1+|x|)^{-1} dW, which satisfies dissipativity but violates the uniform-ellipticity part of Assumption 2.3. With the tamed Euler scheme at several step sizes τ, estimate the W1 distance between the empirical law at large n and a high-accuracy reference for the invariant measure; if it is not O(τ|ln τ|) uniformly in n, the theorem's assumption is doing essential work.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • A fixed small step size can be used for arbitrarily long simulations of superlinear multiplicative-noise SDEs without the one-step errors accumulating beyond O(τ|ln τ|).
  • The numerical invariant measure of the scheme converges to the exact one at rate O(τ|ln τ|) in W1, making explicit tamed or projected schemes viable long-time samplers in this regime.
  • The error bound is stated for Lipschitz test functions, which is exactly the Kantorovich dual of W1, so the result is a genuine law-level guarantee, not a smooth-test-function artifact.
  • Both the tamed Euler and projected Euler schemes are covered by one set of assumptions, so results transfer between them without re-proving the moment and regularity estimates.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The uniform ellipticity in Assumption 2.3 looks necessary for the e^{-λτ n} term; a degenerate-diffusion version of the same theorem would need a different contraction mechanism and is a natural stress test.
  • The logarithmic factor τ|ln τ| arises from the last O(1) steps near the terminal time where the t^{-1} derivative singularity is integrated; refining the test-function regularity might remove the log and give O(τ).
  • The same framework should extend to higher-order explicit schemes, such as modified Milstein methods, as long as the Lyapunov-structure and Kolmogorov-regularity assumptions hold, giving uniform-in-time weak error rates for those methods too.
  • A practical by-product: the constants in Theorem 4.4 depend on moments of the initial condition, so a user can choose τ by balancing the initial-condition term and the target tolerance.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper constructs a family of explicit modified Euler methods (MEMs) for superlinear SODEs with multiplicative noise, including tamed and projected schemes as special cases. The main result, Theorem 2.3, asserts that under Assumptions 2.1–2.5 the law of the MEM at step n satisfies W1(L(Y_n^{x0}), π) ≤ C_* e^{-λτ n} + C_* τ|ln τ| for all τ∈(0,1) and n∈N, where π is the invariant measure of the SDE. Corollary 2.4 derives an O(τ|ln τ|) rate for the invariant-measure approximation. The proof combines Lyapunov moment estimates (Theorems 3.1–3.2), regularity estimates for the Kolmogorov equation obtained via the Bismut–Elworthy–Li formula (Lemmas 3.3–3.5), and a telescoping weak-error decomposition in Section 4 (Lemmas 4.1–4.3 and Theorem 4.4).

Significance. If the main result is correct, it is a meaningful advance: it extends uniform-in-time non-asymptotic W1 error analysis to non-contractive, superlinear, multiplicative-noise settings, going beyond the global contractivity condition used in much of the prior literature. The general framework of Assumptions 2.4–2.5 is useful, and the paper makes a clear effort to reduce the regularity required of test functions via the Bismut–Elworthy–Li formula. The moment bounds and the telescoping structure are substantial and are presented in detail. However, the proof as written contains a load-bearing gap in Lemma 3.5(2), where the exponential derivative decay is imported from [17, Theorem 4.5] without verifying its hypotheses, and several key estimates in Lemmas 4.2–4.3 are omitted. The central claim is therefore not fully verifiable in the current version.

major comments (3)
  1. [Lemma 3.5, Eqs. (3.20)–(3.22)] The time-independent exponential factor e^{-λ(t-1)} in these estimates is what makes the weak-error sums in Lemmas 4.1–4.3 independent of N, and hence is load-bearing for Theorem 2.3. The proof of part (2) defines Φ_t(x)=u(t-1,x)-∫φ dπ, checks only the pointwise bound |Φ_t(x)|≤C_* e^{-λ(t-1)}(1+|x|), and then states that the estimates of DV(t,x) and D^2V(t,x) at t=1 in [17, Theorem 4.5] give (3.20)–(3.21). The manuscript never states the hypotheses of [17, Theorem 4.5] nor verifies that Φ_t satisfies them. If that theorem requires the test function to be Lipschitz, C^k, or to satisfy a different growth/regularity condition, the conclusion may fail or may hold only for fixed t without the exponential decay. Without (3.20)–(3.22), the proof reduces to a finite-time weak error estimate and Theorem 2.3 does not follow. This gap must be repaired by either stating and verifying the hypotheses
  2. [Lemmas 4.2–4.3 and Theorem 4.4] Several estimates are asserted to follow by 'similar' arguments and are omitted: T2, T3, T5, T6 in Lemma 4.2, and R5, R6, R72, R73, R842 in Lemma 4.3. These are not merely cosmetic. The final O(τ|ln τ|) rate in Theorem 4.4 emerges from the near-T singular terms such as R83 and R84, and the precise admissible L^p exponents in (4.4) depend on the treatment of the omitted terms. As written, the proof of Theorem 4.4 is incomplete. The authors should provide complete estimates for the omitted terms, or at minimum give the details of the singular near-T cases and state the resulting moment exponents.
  3. [Theorem 2.3, Assumption 2.5, Theorem 3.2] Theorem 2.3 states the bound for all τ∈(0,1), but Theorem 3.2 only establishes the Lyapunov moment estimate (3.3) for τ<τ_max := min{1/K4,1}. Since Assumption 2.5 does not impose K4≥1, the interval used in Theorem 2.3 can be strictly larger than the interval in which the proof operates. The statement of Theorem 2.3 (and similarly Theorem 4.4, which relies on Theorem 3.2) must either restrict τ to (0,τ_max) or provide a separate argument for τ in (τ_max,1).
minor comments (5)
  1. [Lemma 4.1 and Lemma 4.2 proofs] The proofs cite 'Theorem 3.3', but the paper contains no Theorem 3.3; the intended reference appears to be Theorem 3.2. Please correct the cross-reference.
  2. [Lemma 3.5 statement] The lemma states u(t,·)∈C^3_b(R^d), but the estimates (3.17), (3.20) allow |Du| to grow linearly in |x|. Either C^3_b is used with a nonstandard norm, or the statement should say that the derivatives have polynomial growth. Please clarify the notation.
  3. [Assumption 2.2, equation display (2.3)] The notation 1_{γ≤k} and 1_{γ>k} is terse. Since γ is a real parameter and k is an integer, this is understandable, but a brief explanation of the convention would improve readability, especially because the exponents in later lemmas depend on these cases.
  4. [Theorem 4.4 and Corollary 2.4] The paper claims to 'answer a question left in [11]', where [11] is an arXiv preprint by two of the authors. The relation to [11] should be stated precisely: whether [11] conjectured the O(τ|ln τ|) rate, and whether the present verification for the tamed scheme is independent of [11]'s Assumptions 2.4–2.5 verification for that scheme.
  5. [Overall presentation] The paper has no numerical experiments. This is not required for the theoretical claim, but a small numerical illustration of the W1 rate would help readers assess the practical sharpness of the O(τ|ln τ|) bound.

Circularity Check

0 steps flagged

No significant circularity: the W1 bound follows from stated assumptions, in-paper weak-error estimates, and external ergodicity/regularity imports; self-citations are not used to define the result.

full rationale

Theorem 2.3 is proved by the triangle inequality W1(L(Y_N),π) ≤ W1(L(X_{Nτ}),π) + W1(L(Y_N),L(X_{Nτ})). The first term is controlled by exact-process exponential ergodicity (2.7), imported from the external source [19] under Assumption 2.3, not by the numerical scheme. The second term is the weak error estimate of Theorem 4.4, which is derived in the paper from a telescoping decomposition, the Itô formula, and Lemmas 4.1–4.3; the latter are proved from Assumptions 2.1–2.5 with explicit inequalities. Lemma 3.5(1) (finite-time derivative bounds) is proved in-paper via the Bismut–Elworthy–Li formula. Lemma 3.5(2) imports the exponential derivative decay for t>1 from [17, Theorem 4.5]; while [17] shares an author and the hypotheses are not restated or checked, this is an incomplete justification / rigor gap rather than a circular reduction: the paper does not define any quantity in terms of the target bound, fit a parameter and relabel it as a prediction, or invoke a uniqueness theorem. The only other self-citation, [11], is used to verify that the tamed Euler example satisfies Assumptions 2.4–2.5; that verification is not needed for the general MEM theorem and is not equivalent to Theorem 2.3. No ansatz is smuggled in via citation and no known result is merely renamed. Hence the derivation chain is essentially self-contained apart from standard imported estimates, and no pattern of circularity is present.

Axiom & Free-Parameter Ledger

0 free parameters · 7 axioms · 0 invented entities

No free parameters are fitted to data; all constants are existential from assumptions. The paper introduces no physical entities. Its numerical scheme is a construction, not a postulate. The central claim rests on the stated assumptions and on [19] for exponential contraction.

axioms (7)
  • domain assumption Assumptions 2.1-2.2: dissipativity, polynomial growth, and third-order smoothness of b and σ
    Imposed on the SDE coefficients to guarantee moment bounds and Kolmogorov regularity; not derived in the paper.
  • domain assumption Assumption 2.3: contractivity at infinity with uniform ellipticity of σσ^T
    Key condition imported from [19] to obtain W1 exponential ergodicity; used in Remark 2.1, Lemma 3.5, and Theorem 2.3.
  • ad hoc to paper Assumptions 2.4-2.5: conditions on the modification P and modified coefficients bτ, στ
    The MEM framework is constructed in this paper; these assumptions are tailored sufficient conditions for the Lyapunov structure of the numerical scheme.
  • domain assumption W1 exponential ergodicity of the exact semigroup (2.7) from [19]
    Imported as a black box; supplies the C* e^{-λt} term in the final bound.
  • standard math BEL formula and Kolmogorov equation (3.7) for Lipschitz test functions
    Used in Lemma 3.4 to convert derivative estimates on X into estimates on u without requiring smooth test functions.
  • domain assumption Existence and uniqueness of invariant measures π and πτ
    Implied by Assumptions 2.1 and 2.3 plus the numerical Lyapunov structure, following standard ergodic theory.
  • standard math Itô formula, Gronwall inequality, Hölder and Young inequalities, Fatou lemma, Markov property
    Basic tools used throughout Sections 3 and 4.

reviewed 2026-08-04 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Non-asymptotic Error Analysis of Explicit Modified Euler Methods for Superlinear and Non-contractive SODEs." pith.science (2026). https://pith.science/paper/XEGEHWX2

@misc{pith2026250908410,
  author       = {Pith},
  title        = {Pith review of: Non-asymptotic Error Analysis of Explicit Modified Euler Methods for Superlinear and Non-contractive SODEs},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XEGEHWX2}},
  note         = {Machine review of arXiv:2509.08410}
}
Share X Bluesky LinkedIn Reddit HN
abstract

A family of explicit modified Euler methods (MEMs) is constructed for long-time approximations of super-linear SODEs driven by multiplicative noise. The proposed schemes can preserve the same Lyapunov structure as the continuous problems. Under a non-contractive condition, we establish a non-asymptotic error bound between the law of the numerical approximation and the target distribution in Wasserstein-1 ($\mathcal{W}_1$) distance through a time-independent weak convergence rate for the proposed schemes. As a by-product of this weak error estimate, we obtain an $\mathcal{O}(\tau|\ln \tau|)$ convergence rate between the exact and numerical invariant measures.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Discretization, Uniform-in-Time Estimations and Approximation of Invariant Measures for Nonlinear Stochastic Differential Equations with Non-Uniform Dissipativity

    math.NA 2025-11 conditional novelty 6.0

    The numerical invariant measure of an explicit truncated Euler–Maruyama scheme converges to the true invariant measure at order h^{1/2} in the L1-Wasserstein distance, uniformly in time, for drifts that are dissipativ...

Reference graph

Works this paper leans on

20 extracted references · 15 canonical work pages · cited by 1 Pith paper

  1. [1]

    W.-J. Beyn, E. Isaak, and R. Kruse. Stochastic C-stability and B-consistency of explicit and implicit Euler-type schemes.J. Sci. Comput., 67(3):955–987, 2016

    work page 2016

  2. [2]

    Br´ ehier

    C.-E. Br´ ehier. Approximation of the invariant measure with an Euler scheme for stochastic PDEs driven by space-time white noise.Potential Anal., 40(1):1–40, 2014

    work page 2014

  3. [3]

    C.-E. Brehier. Approximation of the invariant distribution for a class of ergodic SDEs with one-sided Lipschitz continuous drift coefficient using an explicit tamed Euler scheme.ESAIM Probab. Stat., 27:841–866, 2023

    work page 2023

  4. [4]

    Cerrai.Second order PDE’s in finite and infinite dimension, volume 1762 ofLecture Notes in Mathematics

    S. Cerrai.Second order PDE’s in finite and infinite dimension, volume 1762 ofLecture Notes in Mathematics. Springer-Verlag, Berlin, 2001. A probabilistic approach. 26 ZHIHUI LIU, XIAOJIE W ANG, XIAOMING WU, AND XIAOYAN ZHANG

    work page 2001

  5. [5]

    K. D. Elworthy and X.-M. Li. Formulae for the derivatives of heat semigroups.J. Funct. Anal., 125(1):252–286, 1994

    1994

  6. [6]

    Fang and M

    W. Fang and M. B. Giles. Adaptive Euler-Maruyama method for SDEs with nonglobally Lipschitz drift.Ann. Appl. Probab., 30(2):526–560, 2020

    work page 2020

  7. [7]

    Hutzenthaler, A

    M. Hutzenthaler, A. Jentzen, and P. E. Kloeden. Strong and weak divergence in finite time of Euler’s method for stochastic differential equations with non-globally Lipschitz continuous coefficients.Proc. R. Soc. Lond. Ser. A Math. Phys. Eng. Sci., 467(2130):1563–1576, 2011

    2011

  8. [8]

    L. Li, M. Wang, and Y. Wang. Error estimates of the Euler’s method for stochastic differential equations with multiplicative noise via relative entropy.arXiv:2409.04991, 2024

    Pith/arXiv arXiv 2024

  9. [9]

    X. Li, F. Wang, and L. Xu. Unadjusted Langevin algorithms for SDEs with H¨ older drift.Sci. China Math., 2025

    work page 2025

  10. [10]

    W. Liu, X. Mao, and Y. Wu. The backward Euler-Maruyama method for invariant measures of stochastic differential equations with super-linear coefficients.Appl. Numer. Math., 184:137– 150, 2023

    work page 2023

  11. [11]

    Liu and X

    Z. Liu and X. Wu. Geometric ergodicity and strong error estimates for tamed schemes of super-linear SODEs.arXiv:2411.06049, 2024

    Pith/arXiv arXiv 2024

  12. [12]

    Lytras and S

    I. Lytras and S. Sabanis. Taming under isoperimetry.Stochastic Process. Appl., 188:Paper No. 104684, 2025

    work page 2025

  13. [13]

    Neufeld, M

    A. Neufeld, M. Ng, and Y. Zhang. Non-asymptotic convergence bounds for modified tamed unadjusted Langevin algorithm in non-convex setting.J. Math. Anal. Appl., 543(1):Paper No. 128892, 53, 2025

    work page 2025

  14. [14]

    Pag` es and F

    G. Pag` es and F. Panloup. Unadjusted Langevin algorithm with multiplicative noise: total variation and Wasserstein bounds.Ann. Appl. Probab., 33(1):726–779, 2023

    work page 2023

  15. [15]

    Pang and X

    C. Pang and X. Wang. Antithetic multilevel Monte Carlo method for approximations of SDEs with non-globally Lipschitz continuous coefficients.Stochastic Process. Appl., 178:Paper No. 104467, 30, 2024

    work page 2024

  16. [16]

    C. Pang, X. Wang, and Y. Wu. Linear implicit approximations of invariant measures of semi- linear SDEs with non-globally Lipschitz coefficients.J. Complexity, 83:Paper No. 101842, 45, 2024

    work page 2024

  17. [17]

    C. Pang, X. Wang, and Y. Wu. Projected Langevin Monte Carlo algorithms in non-convex and super-linear setting.J. Comput. Phys., 526:Paper No. 113754, 33, 2025

    work page 2025

  18. [18]

    Santambrogio.Optimal transport for applied mathematicians, volume 87 ofProgress in Nonlinear Differential Equations and their Applications

    F. Santambrogio.Optimal transport for applied mathematicians, volume 87 ofProgress in Nonlinear Differential Equations and their Applications. Birkh¨ auser/Springer, Cham, 2015. Calculus of variations, PDEs, and modeling

    work page 2015

  19. [19]

    F. Wang. Exponential contraction in Wasserstein distances for diffusion semigroups with negative curvature.Potential Anal., 53(3):1123–1144, 2020

    work page 2020

  20. [20]

    Wu and X

    X. Wu and X. Wang. Strong convergence rates for long-time approximations of SDEs with non-globally Lipschitz continuous coefficients.arXiv:2406.10582, 2024. Department of Mathematics & National Center for Applied Mathematics Shenzhen (NCAMS) & Shenzhen International Center for Mathematics, Southern University of Science and Technology, Shenzhen 518055, Ch...

    Pith/arXiv arXiv 2024

This paper was first reviewed by deepseek-v4-flash on August 4, 2026.