Pith. sign in

REVIEW 2 major objections 4 minor 48 references

This paper claims that minimizing a single second-order stochastic residual — one scalar space–time network trained on one-step Brownian residuals plus terminal penalties — provably controls the full jet (value, gradient, Hessian) of fully

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 ↗

T0 review · deepseek-v4-flash

2026-08-01 20:10 UTC pith:WXLCC57H

load-bearing objection This is the first population-level full-jet convergence theory for a deep Hessian-dependent parabolic solver, and the main theorems check out; the advertised O(h^1/2) rate, however, rests on an O(h) best-approximation premise that the paper proves only qualitatively. the 2 major comments →

arxiv 2607.16730 v1 pith:WXLCC57H submitted 2026-07-18 math.NA cs.NAmath.OCstat.ML

A Deep Second-Order Stochastic Residual Method for Fully Nonlinear Parabolic PDEs

Zhenhua Zhao , Jihao Long This is my paper
classification math.NA cs.NAmath.OCstat.ML MSC 35K5568T0760H3565M75
keywords fully nonlinear parabolic PDEdeep learning solversecond-order Brownian residualBrownian occupation normHessian-dependent nonlinearitya posteriori error estimatepopulation convergencehigh-dimensional PDE
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 paper introduces D2SRM, a deep-learning solver for high-dimensional fully nonlinear parabolic PDEs — equations where the nonlinearity depends on the Hessian of the unknown solution. It claims that a single scalar space–time network, differentiated to supply value, gradient, and Hessian, can be trained by jointly minimizing second-order Brownian one-step residuals together with terminal value and gradient penalties. The central result is an a posteriori guarantee: for any admissible candidate, the squared full-jet error in the Brownian occupation norm is bounded by the time step plus the candidate's own population objective, and for approximate population minimizers the bound splits explicitly into time discretization, neural representation error, and optimizer suboptimality. When the latter two are O(h), the full-jet occupation error is O(h^{1/2}). A sympathetic reader would care because this is a population-level convergence theory for a Hessian-dependent equation — a setting where, the authors state, this chain of estimates has not previously been established for a deep solver. The analysis is deliberately population-level: Section 9 states that finite-sample generalization, optimizer dynamics, and quantitative network rates are outside its scope.

Core claim

The paper's central claim: the D2SRM population objective is both reliable and attainable. Reliability (Theorem 3.8): under globally Lipschitz data, identity diffusion, and small-gain condition 2√2L₂<1, every admissible candidate satisfies PE²_h ≤ C(h + L_h(θ)) — a small objective forces the whole jet (value, gradient, Hessian) to track the true solution in the Brownian occupation law. Attainability (Theorem 3.9): approximate population minimizers obey PE²_h ≤ C(h + inf_θ A_h(U_θ) + ε_opt,h), separating time discretization, representation error, and optimizer suboptimality; with the latter two O(h), the full-jet occupation norm is O(h^{1/2}). The chain runs through a source-Picard contractio

What carries the argument

The load-bearing objects are the Brownian occupation space H_β = L²(ν_β) with weight r_β(t)p_t(x) dt dx — the norm in which Hessian recovery is proved stable — and the source-update map Φ(F) = f(t,x,u_F,∇u_F,D²u_F) with Lipschitz constant λ(β) = L₀/β + L₁/√β + 2√2L₂. The 2√2 factor comes from a Gaussian affine-complement coercivity: on the Gaussian law at time t, the projected generator A^μ_t Q^aff_t is bounded by √2 times the Hessian norm, so taking β large makes Φ a contraction and yields a unique occupation mild solution. On the grid the argument shifts to an implicit reference scheme whose conditional Gaussian projections P⁰, P¹, P² extract value, first-chaos, and second-chaos channels;

Load-bearing premise

The load-bearing premise is the small-gain condition 2√2 L₂ < 1 on the driver's Hessian-Lipschitz constant (Assumption 3.1); it drives the source-Picard contraction, the stability of Theorem 6.4, and hence both main theorems. For the natural diagonal-Hessian driver f = αΣ|γ_kk| it reads α < 1/√(8d) (Remark 6.5) — dimension-hostile and narrower than the monotone-scheme regime α < 1/2. If it fails, the paper gives no bound; Section 9 also places finite-sample generalization and

What would settle it

Within the assumptions (2√2 L₂ < 1, identity diffusion, global Lipschitz driver), find one admissible Markov candidate with small population objective L_h(θ) but large full-jet occupation error PE²_h; the theorem says none exists. A concrete program: take the d = 1, f = g = 0 case of Remark 6.6 and rebuild the counterexample with σ(X_i)-measurable state-local variables instead of the path-dependent second-chaos variable Q₀ — Theorem 6.4 asserts this cannot be done, so any Markov triple with bounded J_h and divergent first-node Hessian error as h→0 would refute it. Alternatively, in the proved

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

If this is right

  • A posteriori certification at the population level: the training objective doubles as an error certificate — small L_h(θ) provably means small value, gradient, and Hessian error in the Brownian occupation norm, with no extra regularity required of the candidate.
  • Error separability: for approximate minimizers the bound is O(h) + best-representation error + optimizer gap, so the three error sources can be studied and reduced independently; O(h) in the last two yields O(h^{1/2}) full-jet accuracy.
  • Well-posedness by contraction: the exact source-Picard iteration converges geometrically in the occupation space under weak Hessian coupling, identifying the solution as the fixed point the network is implicitly trying to match.
  • The Markov/local comparison class is structurally necessary: the counterexample of Remark 6.6 (d = 1, f = g = 0) shows no h-independent bound can hold for arbitrary adapted path-dependent triples, delimiting any L²-residual-based analysis of second-order solvers.
  • The proved regime is narrower than monotone-scheme regimes: for f = αΣ|γ_kk| the small-gain requirement α < 1/√(8d) is dimension-hostile, and the paper's own experiments show residual informativeness degrading as α grows toward the boundary.

Where Pith is reading between the lines

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

  • If the reliability bound is right, the empirical loss could serve as a practical, conservative monitor of full-jet accuracy during training — with the caveat that the theorem is population-level and says nothing about the SGD gap or finite-sample noise.
  • The small-gain boundary points at the fixed Brownian occupation law as the bottleneck rather than the network class; a testable extension is to train against a small mixture of occupation laws and check whether jet error improves in the strong-coupling regime the paper leaves open.
  • The additive, linear-in-h discretization term suggests an adaptive time-stepping heuristic: local residual contributions could flag where to refine, since discretization error enters the final bound separately from representation and optimization terms.
  • Because the attainability proof relies on C³ activations (centered Softplus, tanh) and qualitative density in a third-order derivative norm, a natural empirical question is whether common smooth-but-non-C³ activations still reach the proved O(h^{1/2}) jet rate — the theory would not cover them.

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

2 major / 4 minor

Summary. The paper introduces D2SRM, a deep learning method for fully nonlinear parabolic PDEs in which a single scalar space-time network supplies value, gradient, and Hessian through derivative-consistent jets and is trained by minimizing one-step second-order Brownian residuals plus terminal penalties. The theoretical core is a new Brownian occupation-space framework. Under global Lipschitz assumptions on the driver and a small-gain condition 2√2 L2 < 1, Theorem 3.7 proves existence and uniqueness of a nonlinear occupation mild solution via a contracting source Picard map, with the explicit contraction constant λ(β)=L0/β+L1/√β+2√2 L2. Theorem 3.8 gives the a posteriori reliability bound PE_h^2 ≤ C(h+L_h(θ)) for admissible candidates. Theorem 3.9 gives the population convergence bound PE_h^2 ≤ C(h+inf_θ A_h(U_θ)+ε_opt,h), and, when the approximation and optimization terms are O(h), the full-jet occupation norm is O(h^{1/2}). Experiments on a 100-dimensional manufactured equation compare terminal treatments and Hessian couplings and report decreasing errors with decreasing step size, with code provided.

Significance. If the conditional claims are accepted, this is a substantial step: it is one of the first analyses that converts a small L2 learning residual into control of the full jet (value, gradient, Hessian) for Hessian-dependent fully nonlinear parabolic PDEs, and it separates time-discretization, representation, and optimization errors. The paper is unusually honest: the small-gain boundary is stated explicitly, the population-level scope is acknowledged, and Section E.4 explicitly says that the best-approximation term is only shown to vanish qualitatively. The explicit constants in Lemma 4.1, Theorem 4.2, and Theorem 3.7 are useful and appear internally consistent. The main weakness is that the advertised O(h^{1/2}) rate is conditional on an O(h) approximation premise that no concrete network family is shown to satisfy, and the numerical time-discretization study is run mostly outside the proved small-gain regime. These are limitations rather than contradictions of the formal statements, but they materially affect the paper's headline claims.

major comments (2)
  1. [§3.3, Eq. (3.6); §E.4; abstract] The advertised O(h^{1/2}) rate is not realized by any demonstrated network sequence. Theorem 3.9 gives PE_h^2≤C(h+inf_{θ∈Θ_h}A_h(U_θ)+ε_opt,h), so the rate requires inf A_h(U_θ)=O(h). Proposition E.3 proves only qualitative vanishing, inf A_h→0, and the text explicitly states that a coefficient-dependent O(h) rate requires additional estimates. Thus the unconditional consequence of Theorem 3.9 is convergence with no rate, and the O(h^{1/2}) statement is a conditional corollary with an unverified premise. The paper acknowledges this, but because the abstract's rate claim is a central selling point, the abstract and introduction should state clearly that the rate is not known to be attained by any concrete neural family, or a positive approximation-rate result should be supplied.
  2. [Assumption 3.1; Remark 6.5; §8.4] The small-gain condition 2√2 L2<1 is load-bearing for Theorems 3.7, 6.4, 3.8, and 3.9. For the benchmark driver f_α=α∑|γ_kk|, it reads α<1/√(8d), which for d=100 is about 0.035. Experiment 3, the time-discretization study whose fitted exponents are compared with the theoretical O(h^{1/2}) rate, uses α=0.25, which is outside the proved regime. Consequently the numerical evidence for the headline rate is not covered by the theorem. The paper does say there is 'no guarantee' outside the regime, but the presentation should make explicit that Experiment 3 is a heuristic probe, not a verification of Theorem 3.9. Either add a time-convergence run inside the proved range or qualify the comparison in the text and figure captions.
minor comments (4)
  1. [Assumptions 3.2 and 3.3] Assumption 3.2 begins 'Theorem 3.1 holds' and Assumption 3.3 begins 'Theorem 3.2 holds'; these should refer to Assumptions 3.1 and 3.2, respectively.
  2. [Remarks 6.5, 6.8; §8.2] Several cross-references are wrong: 'Theorem 6.5' and 'Theorem 6.8' should be Remark 6.5 and Remark 6.8, and 'σcsp defined in Theorem 7.2' should refer to Proposition E.3 or §E.4.
  3. [Definition 2.1 and Theorem 3.9] The parameter sets Θ_h are allowed to depend on h, but this is not made explicit in the definitions. For the infimum in Theorem 3.9 to vanish, the network family must generally grow with h; the text should state that no fixed finite architecture is claimed to achieve the approximation bound.
  4. [Eq. (8.3)] The denominator '+ε_rel^2 B N' is ambiguous: it is unclear whether the regularizer is ε_rel^2/(B N) or ε_rel^2 multiplied by B N. Please clarify the displayed formula.

Circularity Check

0 steps flagged

No significant circularity: the bounds are conditional a posteriori stability results, no fitted constants feed the derivation, and the self-citations are not load-bearing.

full rationale

The derivation chain is self-contained and not circular. The paper defines the population objective L_h and the path error PE2_h, then proves in Theorem 3.8 that PE2_h(Y^theta,Z^theta,Gamma^theta) <= C(h+L_h(theta)) by inserting the auxiliary implicit reference scheme and using exact-grid consistency (Proposition 5.8), mesh-uniform residual stability (Theorem 6.4), and convergence of the reference scheme (Corollary 6.7). Proposition 7.1 bounds L_h by C(h+A_h), so Theorem 3.9 follows as an upper bound, not by construction. Nothing is fitted to the numerical results: the small-gain condition 2*sqrt(2)*L2 < 1 and the rate constants are derived before Section 8, and the experimental log-log slopes are presented as empirical checks rather than as inputs. The conditional O(h^{1/2}) assertion is the direct square-root of inequality (3.6) under the stated O(h) hypotheses; Section E.4 explicitly proves only qualitative vanishing of the best-approximation term and says a quantitative O(h) rate requires additional estimates, which is an acknowledged unverified premise rather than a circular reduction. The self-citations ([6], [17], [19]) appear in related-work and forward-looking remarks and are not load-bearing for the main theorems; no uniqueness theorem by the same authors is invoked to force the choice of method or objective. I therefore cannot exhibit any specific equation or fitted parameter that makes a supposed prediction equal to its own input.

Axiom & Free-Parameter Ledger

3 free parameters · 6 axioms · 1 invented entities

The theory contains no fitted constants. Its parameters are the norm weight beta (chosen, not tuned), the benchmark knob alpha, and the training configuration; none enter the theorem constants via fitting. The main pulls from the hat are the regularity assumptions (3.1)-(3.3) and the Markov-class restriction, both explicitly flagged in the text (Assumption 3.1, Remark 6.6). The single most restrictive premise is the small-gain condition 2*sqrt(2)*L2 < 1: sufficient and dimension-free in statement, but dimension-hostile for f = alpha*sum|gamma_kk|.

free parameters (3)
  • beta (occupation weight) = not fixed numerically; any sufficiently large value with lambda(beta) < 1 (no data fitting)
    Weight of the Brownian occupation norm H_beta. A theory parameter, not fitted to data; constants C_beta depend on it, but all theorems hold for any admissible beta.
  • alpha (benchmark Hessian coupling) = 0.025, 1/sqrt(800) ~ 0.035, 0.25, 0.45
    Experimental knob scanning the proved boundary 1/sqrt(8d) for d=100. Not used to fit any theory constant.
  • network/training hyperparameters (width 64x3, batch 256, LR schedules) = standard choices from the deep-PDE-solver literature
    Training configuration for the experiments; the theory is agnostic to them and they do not enter the theorem constants.
axioms (6)
  • domain assumption Global Lipschitz continuity of f in (y,z,gamma) with constants L0, L1, L2 and the small-gain condition 2*sqrt(2)*L2 < 1
    Assumption 3.1; drives the contraction lambda(beta) < 1 (Theorem 3.7) and the mesh-uniform residual stability (Theorem 6.4). For the natural driver family f = alpha*sum|gamma_kk| this reads alpha < 1/sqrt(8d), which is dimension-hostile (Remark 6.5).
  • domain assumption g in H^2(mu_T), spatial Lipschitz continuity of f, time 1/2-Holder and Brownian-state regularity (Lt, LB)
    Assumptions 3.2-3.3; used in Proposition 5.1 (weak derivative of the source), Lemma 5.6 (source freezing), and the canonical trace machinery (Lemma 5.3, Lemma C.3).
  • domain assumption Identity diffusion X_t = x0 + W_t with a fixed deterministic start
    Section 2.1 asserts the fixed-start setup is a presentational choice, but the general-diffusion extension (Appendix F) is only conditional and explicitly excluded from the main convergence chain.
  • standard math Standard Gaussian analytic toolkit: Poincare inequality, Gaussian Bochner/OU identity, Hermite-Mehler expansions, Ito formula, Schur test and Hardy-Copson weighted inequalities
    Used in Lemma 4.1/B.1, Theorem 4.2/B.2, and Appendices A, C, D; cited to [26, 37, 36, 30, 29, 16, 28].
  • domain assumption Markov comparison class C_h: the sampled jet must be sigma(X_i)-measurable at each node
    Section 2.2. Remark 6.6 proves J_h alone fails for arbitrary adapted path-dependent triples (a counterexample with J_h = T a^2/2 but first-node Hessian error T N a^2), so the state-local restriction is structural, not merely technical.
  • standard math Hornik-type C3 derivative density of the centered-Softplus ridge class with respect to the finite measure nu_h
    Proposition E.3/(E.9): used only for qualitative attainability (inf A_h -> 0), never for rates. The O(h) antecedent of Theorem 3.9 is left unproved (Remark 7.2).
invented entities (1)
  • Nonlinear occupation mild solution (source fixed point F* in H_beta) independent evidence
    purpose: Gives L2 well-posedness and the source Picard contraction that matches the residual objective
    A mathematical construction, not a physical entity. It is internally consistent with classical solutions (Remark 3.6) and with the classical 2BSDE representation (Lemma 5.3/(5.5)); no new physics or unobserved entities are introduced.

pith-pipeline@v1.3.0-alltime-deepseek · 46389 in / 43835 out tokens · 389255 ms · 2026-08-01T20:10:30.202075+00:00 · methodology

0 comments
read the original abstract

We introduce the Deep Second-Order Stochastic Residual Method (D2SRM) for high-dimensional, Hessian-dependent fully nonlinear parabolic PDEs. A single scalar space--time network generates derivative-consistent approximations of the solution, gradient, and Hessian, which are trained jointly through second-order Brownian one-step residuals and terminal value and gradient penalties. For globally Lipschitz equations with identity diffusion and sufficiently weak Hessian coupling, we establish well-posedness in a Brownian occupation space and develop a population-level convergence theory. Under additional regularity, an a posteriori estimate bounds the squared full-jet occupation error of any admissible candidate by the time step and its population objective. For approximate population minimizers, the error bound separates time discretization, neural approximation, and population suboptimality; when the latter two terms are $O(h)$, the full-jet occupation norm is $O(h^{1/2})$. Experiments on a 100-dimensional manufactured benchmark compare terminal treatments, probe Hessian couplings inside and outside the proved small-gain range, and show decreasing errors as the time step decreases. The code is available at https://github.com/ZZHPKU/D2SRM.

discussion (0)

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

Reference graph

Works this paper leans on

48 extracted references

  1. [1]

    and Long, J

    Han, J. and Long, J. , title =. Probab. Uncertain. Quant. Risk , volume =. 2020 , eid =

    2020

  2. [2]

    and Han, J

    E, W. and Han, J. and Jentzen, A. , title =. Commun. Math. Stat. , volume =. 2017 , pages =

    2017

  3. [3]

    Deep Backward Schemes for High-Dimensional Nonlinear

    Hur. Deep Backward Schemes for High-Dimensional Nonlinear. Math. Comp. , volume =. 2020 , pages =

    2020

  4. [4]

    and Touzi, N

    Fahim, A. and Touzi, N. and Warin, X. , title =. Ann. Appl. Probab. , volume =. 2011 , pages =

    2011

  5. [5]

    and Zhang, J

    Guo, W. and Zhang, J. and Zhuo, J. , title =. Ann. Appl. Probab. , volume =. 2015 , pages =

    2015

  6. [6]

    Adapted Solution of a Backward Stochastic Differential Equation , journal =

    Pardoux,. Adapted Solution of a Backward Stochastic Differential Equation , journal =. 1990 , pages =

    1990

  7. [7]

    , title =

    Janson, S. , title =

  8. [8]

    , title =

    Nualart, D. , title =

  9. [9]

    and Peccati, G

    Nourdin, I. and Peccati, G. , title =

  10. [10]

    and Persson, L.-E

    Kufner, A. and Persson, L.-E. and Samko, N. , title =. 2017 , doi =

    2017

  11. [11]

    Adams, R. A. and Fournier, J. J. F. , title =

  12. [12]

    , title =

    Hornik, K. , title =. Neural Networks , volume =. 1991 , pages =

    1991

  13. [13]

    Error Bounds for Approximations with Deep

    G. Error Bounds for Approximations with Deep. Anal. Appl. (Singap.) , volume =. 2020 , pages =

    2020

  14. [14]

    2021 , pages =

    On the Approximation of Functions by Tanh Neural Networks , journal =. 2021 , pages =

    2021

  15. [15]

    and Schmocker, P

    Neufeld, A. and Schmocker, P. , title =. Anal. Appl. (Singap.) , volume =. 2026 , pages =

    2026

  16. [16]

    and Yang, H

    Hon, S. and Yang, H. , title =. Neural Networks , volume =. 2022 , pages =

    2022

  17. [17]

    Aronson, D. G. , title =. Bull. Amer. Math. Soc. , volume =. 1967 , pages =

    1967

  18. [18]

    , title =

    Sheu, S.-J. , title =. Ann. Probab. , volume =. 1991 , pages =

    1991

  19. [19]

    , title =

    Ledoux, M. , title =. Ann. Fac. Sci. Toulouse Math. (6) , volume =. 2000 , pages =

    2000

  20. [20]

    Stroock, D. W. and Turetsky, J. , title =. Comm. Anal. Geom. , volume =. 1998 , pages =

    1998

  21. [21]

    and Long, J

    Chen, H. and Long, J. and Wu, L. , title =. Ann. Statist. , volume =. 2025 , pages =

    2025

  22. [22]

    Halmos, P. R. , title =. A Hilbert Space Problem Book , edition =. 1982 , pages =

    1982

  23. [23]

    and Shreve, S

    Karatzas, I. and Shreve, S. E. , title =. 1991 , doi =

    1991

  24. [24]

    and Soner, H

    Cheridito, P. and Soner, H. M. and Touzi, N. and Victoir, N. , title =. Comm. Pure Appl. Math. , volume =. 2007 , pages =

    2007

  25. [25]

    and Hornung, F

    Grohs, P. and Hornung, F. and Jentzen, A. and von Wurstemberger, P. , title =. Mem. Amer. Math. Soc. , volume =. 2023 , eid =

    2023

  26. [26]

    and Jentzen, A

    Hutzenthaler, M. and Jentzen, A. and Kruse, T. and Nguyen, T. A. , title =. Partial Differ. Equ. Appl. , volume =. 2020 , eid =

    2020

  27. [27]

    and Shukla, K

    Hu, Z. and Shukla, K. and Karniadakis, G. E. and Kawaguchi, K. , title =. Neural Networks , volume =. 2024 , eid =

    2024

  28. [28]

    and Shi, Z

    Hu, Z. and Shi, Z. and Karniadakis, G. E. and Kawaguchi, K. , title =. Comput. Methods Appl. Mech. Engrg. , volume =. 2024 , eid =

    2024

  29. [29]

    and Guillin, A

    Cattiaux, P. and Guillin, A. , title =. S. 2014 , pages =

    2014

  30. [30]

    and Li, X.-M

    Chen, X. and Li, X.-M. and Wu, B. , title =. Ann. Probab. , volume =. 2023 , pages =

    2023

  31. [31]

    Time-Uniform

    Monmarch. Time-Uniform. Electron. J. Probab. , volume =. 2024 , eid =

    2024

  32. [32]

    and Jentzen, A

    Han, J. and Jentzen, A. and E, W. , title =. Proc. Natl. Acad. Sci. USA , volume =. 2018 , pages =

    2018

  33. [33]

    and Han, J

    E, W. and Han, J. and Jentzen, A. , title =. Nonlinearity , volume =. 2022 , pages =

    2022

  34. [34]

    and Souganidis, P

    Barles, G. and Souganidis, P. E. , title =. Asymptot. Anal. , volume =. 1991 , pages =

    1991

  35. [35]

    Soner, H. M. and Touzi, N. and Zhang, J. , title =. Probab. Theory Related Fields , volume =. 2012 , pages =

    2012

  36. [36]

    and E, W

    Beck, C. and E, W. and Jentzen, A. , title =. J. Nonlinear Sci. , volume =. 2019 , pages =

    2019

  37. [37]

    and Spiliopoulos, K

    Sirignano, J. and Spiliopoulos, K. , title =. J. Comput. Phys. , volume =. 2018 , pages =

    2018

  38. [38]

    and Perdikaris, P

    Raissi, M. and Perdikaris, P. and Karniadakis, G. E. , title =. J. Comput. Phys. , volume =. 2019 , pages =

    2019

  39. [39]

    and Hu, W

    Han, J. and Hu, W. and Long, J. and Zhao, Y. , title =. SIAM J. Sci. Comput. , volume =. 2026 , pages =

    2026

  40. [40]

    and Ouyang, D

    Jia, Y. and Ouyang, D. and Pham, H. and Zhou, X. Y. , title =. 2026 , eprint =

    2026

  41. [41]

    and Warin, X

    Pham, H. and Warin, X. and Germain, M. , title =. Partial Differ. Equ. Appl. , volume =. 2021 , eid =

    2021

  42. [42]

    , title =

    Tan, X. , title =. Electron. J. Probab. , volume =. 2013 , eid =

    2013

  43. [43]

    and Molinaro, R

    Mishra, S. and Molinaro, R. , title =. IMA J. Numer. Anal. , volume =. 2023 , pages =

    2023

  44. [44]

    and Darbon, J

    Shin, Y. and Darbon, J. and Karniadakis, G. E. , title =. Commun. Comput. Phys. , volume =. 2020 , pages =

    2020

  45. [45]

    Nguwi, J. Y. and Penent, G. and Privault, N. , title =. J. Comput. Phys. , volume =. 2024 , eid =

    2024

  46. [46]

    and Loeper, G

    Lefebvre, W. and Loeper, G. and Pham, H. , title =. Digital Finance , volume =. 2023 , pages =

    2023

  47. [47]

    and Han, J

    Zhou, M. and Han, J. and Lu, J. , title =. SIAM J. Sci. Comput. , volume =. 2021 , number =

    2021

  48. [48]

    and Lu, J

    Zhou, M. and Lu, J. , title =. SIAM J. Control Optim. , volume =. 2025 , number =

    2025