Pith. sign in

REVIEW 3 major objections 6 minor 39 references

Strong convergence in the infinite horizon of numerical methods for stochastic delay differential equations

T0 review · 3 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

Pith's one-line read A general four-condition patching argument gives time-uniform strong error bounds for SDDE numerical methods, applied to backward and truncated Euler-Maruyama methods and to invariant measures.

desk verdict Useful abstraction, but the key contraction step in Appendices A and B is algebraically wrong; the proof as written does not establish the time-uniform bound. read the letter →

arxiv 2505.14262 v1 pith:VWZYG2PO submitted 2025-05-20 math.NA cs.NA

classification math.NAcs.NA
keywords numericalmethodsconvergencedelaydifferentialequationshorizoninfinite
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

Stochastic delay differential equations describe systems whose future depends on their past (for example, a cell's response to hormone levels from minutes ago) plus random noise. Simulating them accurately over long time horizons is hard because standard error estimates grow exponentially with total time, forcing ever smaller time steps. This paper proposes a way to get error bounds that stay flat forever, provided the equation is contracting (solutions from different starts converge to each other exponentially) and the numerical method is stable and accurate on finite intervals.

The proof cuts time into fixed-length windows. On each window, it compares the numerical path with a fresh true solution started from the numerical value. Because the true system pulls trajectories together, errors from earlier windows decay, while each window's new error is controlled by a standard finite-time convergence result. The sum stays bounded, giving a time-independent bound proportional to a power of the step size. The authors apply the recipe to two schemes: the backward Euler-Maruyama method and the truncated Euler-Maruyama method, and they show the numerical segment process also approximates the invariant distribution of the true system, which is useful for computing long-run statistics.

The idea is simple and potentially general, but the written proof contains an algebraic slip in the central contraction step. The numerical experiments are qualitative, with no code or error bars.

Extended reading notes

Core claim

Theorem 3.3: Under Condition 3.1, sup_{t≥-τ} E|x(t;0,ξ)-X(t;0,ξ)|^p ≤ C Δ^q with C independent of T, so the strong error of the numerical solution is bounded uniformly in the infinite horizon. Theorem 3.6 extends the same bound to segment processes, and Theorem 3.7 uses it to show d_L(P_{t_k}, π) → 0 as k→∞, Δ→0. If the paper is correct, backward and truncated Euler-Maruyama methods inherit these uniform-in-time strong error bounds under the stated dissipativity and growth assumptions.

Load-bearing premise

Condition 3.1(iii) and 3.4(iii): the true SDDE solution is exponentially contractive in p-th moment, E|x(t;0,ξ)-x(t;0,η)|^p ≤ M2 sup E|ξ-η|^p e^{-M3 t}. The whole interval-patching argument depends on this decay to make the accumulated error over successive windows summable; if the underlying SDDE is not exponentially attracting, no time-uniform bound of this form can be expected. This assumption is verified in the applications by dissipativity conditions (4.3) and (5.2), but it is a genuine restriction on the dynamics.

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

3 major / 6 minor

Summary. The paper proposes a general technique for proving strong convergence over the infinite horizon for numerical methods for stochastic delay differential equations (SDDEs). The idea is to partition time into windows of fixed length and to compare the numerical solution on each window with a newly restarted true solution, using exponential contraction of the true SDDE to make the accumulated errors summable. Theorem 3.3 claims a uniform-in-time bound sup_{t\ge -\tau} E|x(t;0,\xi)-X(t;0,\xi)|^p \le C\Delta^q under Condition 3.1. Theorem 3.6 claims the analogous uniform bound for segment processes under Condition 3.4, and Theorem 3.7 uses this to show that the law of the numerical segment process converges to the invariant measure of the true segment process as k\to\infty and \Delta\to0. The technique is illustrated on the truncated Euler-Maruyama method and on the backward Euler-Maruyama method, with numerical experiments included.

Significance. If the proof gap identified below is repaired, the paper would make a useful contribution: it provides a framework for obtaining strong error bounds whose constants do not grow with the time horizon, and it gives a route to numerical approximation of invariant measures that avoids proving existence of an invariant measure for the numerical scheme itself. The paper correctly identifies exponential contraction of the true dynamics as the key structural assumption, and it applies the framework to two practically relevant methods. The reliance on finite-time convergence results and on explicit moment/contraction assumptions is clearly stated, so the overall strategy is transparent. However, the central window-patching argument in the appendices is currently not valid as written, and the omitted proof of Theorem 3.6 makes that part of the paper impossible to verify.

major comments (3)
  1. [Appendix A, Eq. (A.6)] The inequality 2^p M2 e^{-M3 T} \le e^{-M3 T}/2 is equivalent to 2^p M2 \le 1/2, which is not implied by Condition 3.1. The proof defines T := 2\tau + (2\log(2M1))/M2, using the numerical moment bound M1 rather than the contraction constants M2 and M3; when M2 > 2^{-p}, no value of T satisfies the displayed inequality. Also, T can be nonpositive when M1 < 1/2, since M1 is only assumed positive. This step is load-bearing because the contraction coefficient is what makes the successive error terms summable in (A.9)-(A.11), so Theorem 3.3 is not established as written. A repair appears plausible by choosing T so that 2^p M2 e^{-M3 T} \le 1/2 and then repeating the iteration, but the proof must be corrected explicitly.
  2. [Appendix B, Eq. (A.4)] The same algebraic defect occurs in the proof of Lemma 3.5. With T = 4\tau + (4\log(2K1))/K2, the displayed identity 2K1 e^{-K2 T} = e^{-3K2 T/4} is false unless \tau = 0; the term 4\tau introduces an extra factor e^{-K2\tau}. Moreover, when K1 < 1/2, the quantity \log(2K1) is negative and T may fail to be positive. Consequently the asserted bound 2(K1 e^{-K2 T} + C_T^2 \Delta^q) \le e^{-K2 T/2} is not justified. Since Lemma 3.5 is needed for Theorem 3.6, this is a second load-bearing gap. The repair is to choose T from K1 and K2 so that 2K1 e^{-K2 T} \le 1/2 and to handle the \Delta^q term separately for sufficiently small step size.
  3. [Section 3, Condition 3.4, Lemma 3.5, and Theorem 3.6] Condition 3.4 is stated for a generic moment order p in items (ii)-(iv), but Lemma 3.5 proves a second-moment bound and its proof invokes Condition 3.4 with power 2; Theorem 3.6 then concludes sup_k E\|x_{t_k} - X_{t_k}\|^2 \le C\Delta^q. As written, these results only follow when Condition 3.4 is specialized to p=2, or when the conclusion is stated for the same p as in the condition. In addition, the proof of Theorem 3.6 is omitted as 'similar' to Theorem 3.3; given that the patching argument in Lemma 3.5 needs repair, the paper should provide the details rather than rely on analogy.
minor comments (6)
  1. [Section 3, Condition 3.1] The condition list contains two items labelled (iii); the second occurrence should be labelled (iv).
  2. [Section 4, Assumption 4.2] The constants b1, b2, b3 appear twice without any distinction, although the two sets are clearly meant to be different; one set should be barred or otherwise renamed.
  3. [Appendix A, Eq. (A.1)] The first displayed inequality in Appendix A writes E(x(t;0,\xi)-X(t;0,\xi)) without the absolute value and without the p-th power; it should be E|x(t;0,\xi)-X(t;0,\xi)|^p to match the rest of the proof.
  4. [Lemma 3.5] The statement contains the typo 'satisyes' instead of 'satisfies'.
  5. [Condition 3.4, item (iv)] The text says that C_{T2-T1} is a constant dependent on T2 - T2; this should be T2 - T1.
  6. [Section 4, Lemma 4.9] The notation z(t; T1, z_{T1}) is used without a prior definition; since z is a numerical scheme starting from time 0, the flow-property definition for restarting at T1 should be made explicit.
Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

No free parameters are fitted to data. The only hand-chosen quantities are the window lengths T in the proofs, whose stated choices are inconsistent with the needed contraction inequalities. The theorems rest on standard stochastic calculus, on structural assumptions guaranteeing existence and Markov/flow properties, and on exponential attraction of the true solution.

free parameters (2)
  • Window length T in Theorem 3.3 proof = 2τ + 2 log(2M1)/M2 (as written)
    Chosen by hand to make the patching sum converge, but the stated value does not imply the contraction inequality (A.6); the proof needs T to be a function of the attraction constant M2 and rate M3 instead.
  • Window length T in Lemma 3.5 proof = 4τ + 4 log(2K1)/K2 (as written)
    Same issue: the claimed decay estimate uses a limit that does not match this T.
assumptions (4)
  • standard math Standard stochastic calculus tools (Itô formula, Burkholder-Davis-Gundy, Gronwall, Young, Hölder) are applied as background facts.
    Used throughout Sections 4, 5 and appendices; standard for this literature.
  • domain assumption For Assumptions 4.1-4.3 (TEM) and 5.1-5.2 (BEM), SDDE (2.1) has a unique global solution and satisfies the flow property x(t;0,ξ)=x(t;s,x_s) at grid points.
    Invoked in Lemma 4.9 and Appendix A; cited to Mao and Yuan [28, p.78].
  • domain assumption The numerical segment processes {Z_{t_k}} and {X_{t_k}} are time-homogeneous Markov chains.
    Proved in Lemma 4.5 and 5.5 under structural assumptions (5.4); the Markov property is what lets the patching argument restart at each window.
  • domain assumption Exponential attraction of true solutions and segment processes (Condition 3.1(iii), 3.4(iii)).
    Verified for the two examples via dissipativity (Assumptions 4.3 and 5.2); it is the key contraction input for the infinite-horizon bound.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Strong convergence in the infinite horizon of numerical methods for stochastic delay differential equations." pith.science (2026). https://pith.science/paper/VWZYG2PO

@misc{pith2026250514262,
  author       = {Pith},
  title        = {Pith review of: Strong convergence in the infinite horizon of numerical methods for stochastic delay differential equations},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VWZYG2PO}},
  note         = {Machine review of arXiv:2505.14262}
}
read the original abstract

In this work, we present a general technique for establishing the strong convergence of numerical methods for stochastic delay differential equations (SDDEs) in the infinite horizon. This technique can also be extended to analyze certain continuous function-valued segment processes associated with the numerical methods, facilitating the numerical approximation of invariant measures of SDDEs. To illustrate the application of these results, we specifically investigate the backward and truncated Euler-Maruyama methods. Several numerical experiments are provided to demonstrate the theoretical results.

Figures

Figures reproduced from arXiv: 2505.14262 by the authors.

Figure 1
Figure 1. Left: The BEM method; Right:The TEM method; [PITH_FULL_IMAGE:figures/full_fig_p025_1.png] view at source ↗
Figure 2
Figure 2. Left:Difference between empirical distribution [PITH_FULL_IMAGE:figures/full_fig_p025_2.png] view at source ↗
Figure 3
Figure 3. Sample means of E(|f(z(k))|) (k ∈ N) with different initial data ξ and  1 + |x| 2 7 h2x, f(x, y)i + 15|g(x, y)| 2  ≤ 2 7 (4 − 12|x| 18 + |y| 18), h2x, f(x, y)i + 15|g(x, y)| 2 ≤ 30 − 3x 2 + y 2 2hx − x, f(x, y) − f(x, y)i + 2|g(x, y) − g(x, y)| 2 ≤ −3|x − x| 2 + |y − y| 2 . Moreover, for any s, t ∈ [−1, 0], E|cos(W(t)) − cos(W(s))| p ≤ E|W(t) − W(s)| p ≤ Kp(t − s) p 2 , where Kp is a positive constant dependent … view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

39 extracted references · 39 canonical work pages

  1. [1]

    Uniform in time convergence of numerical schemes for stochastic differential equations via Strong Exponential stability: Euler methods, Split-Step and Tamed Schemes

    L. Angeli, D. Crisan, and M. Ottobre. Uniform in time conv ergence of numerical schemes for stochastic differen- tial equations via strong exponential stability: Euler met hods, split-step and tamed schemes. arXiv:2303.15463, 2023

  2. [2]

    Arriojas, Y

    M. Arriojas, Y . Hu, S.-E. Mohammed, and G. Pap. A delayed b lack and scholes formula. Stochastic Anal. Appl., 25(2):471 – 492, 2007

  3. [3]

    J. Bao, J. Shao, and C. Y uan. Invariant probability measu res for path-dependent random diffusions. Nonlinear Anal., 228, 2023

  4. [4]

    E. Buckwar. Introduction to the numerical analysis of st ochastic delay differential equations. J. Comput. Appl. Math., 125(1-2):297 – 307, 2000

  5. [5]

    Cacace, A

    F. Cacace, A. Germani, C. Manes, and M. Papi. Predictor-b ased control of stochastic systems with nonlinear diffusions and input delay. Automatica, 107:43 – 51, 2019

  6. [6]

    Y . Cai, Y . Kang, M. Banerjee, and W . Wang. A stochastic sir s epidemic model with infectious force under intervention strategies. J. Differential Equations, 259(12):7463 – 7502, 2015

  7. [7]

    Crisan, P

    D. Crisan, P . Dobson, and M. Ottobre. Uniform in time esti mates for the weak error of the Euler method for SDEs and a pathwise approach to derivative estimates for diffusion semigroups. Trans. Amer . Math. Soc., 374(5):3289– 3330, 2021

  8. [8]

    J. P . D´ ecamps, F. Gensbittel, and T. Mariotti. Investment timing and technological breakthroughs, 2021

Show all 39 references
  1. [9]

    N.H. Du, D.H. Nguyen, and G.G. Yin. Conditions for perman ence and ergodicity of certain stochastic predator prey models. J. Appl. Probab., 53(1):187 – 202, 2016

  2. [10]

    S. Gao, X. Li, and Z. Liu. Stationary distribution of the milstein scheme for stochastic differential delay equations with first-order convergence. Appl. Math. Comput. , 458, 2023. 29 A PREPRINT - S EPTEMBER 10, 2025

  3. [11]

    Grenadier and A

    S. Grenadier and A. Weiss. Investment in technological innovations: An option pricing approach. Journal of Financial Economics, 44(3):397–416, 1997

  4. [12]

    Q. Guo, X. Mao, and R. Y ue. The truncated euler–maruyama method for stochastic differential delay equations. Numer . Algorithms., 78(2):599 – 624, 2018

  5. [13]

    D. J. Higham, X. Mao, and A. M. Stuart. Exponential mean- square stability of numerical solutions to stochastic differential equations. LMS J. Comput. Math. , 6:297 – 313, 2003

  6. [14]

    Semi-implicit euler-maruyama scheme for stiff stochastic equations

    Y aozhong Hu. Semi-implicit euler-maruyama scheme for stiff stochastic equations. In Stochastic analysis and related topics, V (Silivri, 1994) , volume 38 of Progr . Probab., pages 183–202. Birkh¨ auser Boston, Boston, MA, 1996

  7. [15]

    Ikeda and S

    N. Ikeda and S. Watanabe. Stochastic differential equations and diffusion processes. North-Holland, Amsterdam, 1989

  8. [16]

    Ji and C

    Y . Ji and C. Y uan. Tamed em scheme of neutral stochastic d ifferential delay equations. J. Comput. Appl. Math. , 326:337 – 357, 2017

  9. [17]

    Lei and M

    J. Lei and M. C. Mackey. Stochastic differential delay e quation, moment stability, and application to hematopoi- etic stem cell regulation system. SIAM J. Appl. Math. , 67(2):387 – 407, 2006

  10. [18]

    X. Li, Q. Ma, H. Y ang, and C. Y uan. The numerical invarian t measure of stochastic differential equations with markovian switching. SIAM J. Numer . Anal., 56(3):1435 – 1455, 2018

  11. [19]

    Li and X

    X. Li and X. Mao. Population dynamical behavior of non-a utonomous lotka-volterra competitive system with random perturbation. Discrete Contin. Dynam. Systems , 24(2):523 – 545, 2009

  12. [20]

    X. Li, X. Mao, and G. Song. Explicit approximation of inv ariant measure for stochastic delay differential equa- tions with the nonlinear diffusion term. J. Theoret. Probab., 2023

  13. [21]

    X. Li, X. Mao, and G. Yin. Explicit numerical approximat ions for stochastic differential equations in finite and infinite horizons: Truncation methods, convergence in p th moment and stability. IMA J. Numer . Anal. , 39(2):847–892, 2019

  14. [22]

    M. Liu, W . Cao, and Z. Fan. Convergence and stability of t he semi-implicit euler method for a linear stochastic differential delay equation. J. Comput. Appl. Math. , 171(1-2):255 – 268, 2004

  15. [23]

    X. Mao. Exponential stability of equidistant euler-ma ruyama approximations of stochastic differential delay equations. J. Comput. Appl. Math. , 200(1):297 – 316, 2007

  16. [24]

    X. Mao. Stochastic Differential Equations and Applications . Horwood, Chichester, UK, 2 edition, 2007

  17. [25]

    X. Mao. Almost sure exponential stability in the numeri cal simulation of stochastic different equations. SIAM J. Numer . Anal., 53(1):370 – 389, 2015

  18. [26]

    Mao and M

    X. Mao and M. J. Rassias. Khasminskii-type theorems for stochastic differential delay equations. Stochastic Anal. Appl., 23(5):1045 – 1069, 2005

  19. [27]

    Mao and S

    X. Mao and S. Sabanis. Numerical solutions of stochasti c differential delay equations under local lipschitz condition. J. Comput. Appl. Math. , 151(1):215 – 227, 2003

  20. [28]

    Mao and C

    X. Mao and C. Y uan. Stochastic Differential Equations with Markovian Switchi ng. Imperial College Press, London, 2 edition, 2006

  21. [29]

    E. J. McShane. Stochastic Calculus and Stochastic Models. Academic . Academic Press, 1974

  22. [30]

    Milstein, E

    G.N. Milstein, E. Platen, and H. Schurz. Balanced impli cit methods for stiff stochastic systems. SIAM J. Numer . Anal., 35(3):1010 – 1019, 1998

  23. [31]

    Mohammed

    S.E.A. Mohammed. Stochastic Functional Differential Equations. Longman, New Y ork, 1986

  24. [32]

    M. Reiß, M. Riedle, and O. van Gaans. Delay differential equations driven by l´ evy processes: stationarity and feller properties. Stochastic Process. Appl., 116(10):1409 –1432, 2006

  25. [33]

    Ru´ e, J

    P . Ru´ e, J. Vill` a-Freixa, and K. Burrage. Simulation methods with extended stability for stiff biochemical kinetics. BMC Syst. Biol. 4 (110) (2010) 1–13. , 4(110):1 – 13, 2010

  26. [34]

    G. Song, J. Hu, S. Gao, and X. Li. The strong convergence a nd stability of explicit approximations for nonlinear stochastic delay differential equations. Numer . Algorithms, 89(2):855 – 883, 2022

  27. [35]

    Y . Wang, F. Wu, and X. Mao. Stability in distribution of s tochastic functional differential equations. Systems Control Lett., 132, 2019

  28. [36]

    Y uan and X

    C. Y uan and X. Mao. Asymptotic stability in distributio n of stochastic differential equations with markovian switching. Stochastic Process. Appl., 103(2):277 – 291, 2003. 30 A PREPRINT - S EPTEMBER 10, 2025

  29. [37]

    Y uan and X

    C. Y uan and X. Mao. Stability in distribution of numerical solutions for stochastic differential equations. Stochas- tic Anal. Appl. , 22(5):1133 – 1150, 2004

  30. [38]

    Y uan, J

    C. Y uan, J. Zou, and X. Mao. Stability in distribution of stochastic differential delay equations with markovian switching. Systems Control Lett., 50(3):195 – 207, 2003

  31. [39]

    S. Zhou. Strong convergence and stability of backward e uler–maruyama scheme for highly nonlinear hybrid stochastic differential delay equation. Calcolo, 52(4):445 – 473, 2015. 31

Pith tools

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