pith. sign in

arxiv: 2605.21388 · v1 · pith:KKU6MJMHnew · submitted 2026-05-20 · 💻 cs.LG · cs.AI· cs.NA· math.NA· stat.ML

On the Regularity and Generalization of One-Step Wasserstein-guided Generative Models for PDE-Induced Measures

Likun Lin , Zhongjian Wang , Jack Xin , Zhiwen Zhang This is my paper

Pith reviewed 2026-05-21 05:20 UTC · model grok-4.3

classification 💻 cs.LG cs.AIcs.NAmath.NAstat.ML
keywords generative modelsoptimal transportPDE-induced measuresdoubling conditionsHölder continuityWasserstein distancegeneralization boundsone-step models
0
0 comments X

The pith

Optimal transport maps from uniform sources to PDE-induced measures are Hölder continuous under standard assumptions.

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

The paper establishes that normalized target densities from linear elliptic and parabolic PDEs on bounded domains, plus diffusion and Fokker-Planck equations on the torus, yield measures that satisfy doubling conditions when standard structural assumptions hold. Pairing this property with existing regularity results for optimal transport between doubling measures shows that the map transporting a uniform source distribution to the target is Hölder continuous. This continuity supplies an approximation-theoretic reason why one-step generative models can learn such PDE-induced distributions through a single learned pushforward. A reader would care because the result moves beyond pessimistic statistical bounds toward concrete justification for simple transport-based generators in scientific computing settings.

Core claim

Under standard structural assumptions, the target measures associated with linear elliptic and parabolic equations on bounded domains, as well as diffusion and Fokker-Planck equations on the torus, satisfy doubling conditions. Combining this fact with regularity theory for optimal transport between doubling measures shows that the optimal transport map from a uniform source measure to the target measure is Hölder continuous. This regularity yields an approximation-theoretic justification for one-step generative models that learn PDE-induced distributions via a single pushforward map. As a concrete case, excess-risk bounds are derived for DeepParticle that measure the gap to the population OT

What carries the argument

Doubling conditions on the PDE-induced target measures, which trigger Hölder regularity of the optimal transport map from a uniform source via established optimal transport theory.

If this is right

  • Excess-risk bounds quantify how well a learned one-step map approximates the true population OT map for models such as DeepParticle.
  • Robustness estimates quantify stability of the learned map under shifts in the target PDE-induced measure.
  • One-step pushforward models receive approximation-theoretic support for representing distributions arising from elliptic, parabolic, and Fokker-Planck equations.
  • Generalization rates follow from the Hölder exponent once the map is approximated by a neural network or similar function class.

Where Pith is reading between the lines

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

  • The same doubling-plus-regularity route may apply to other linear or semilinear PDEs once their densities are checked for the required structural properties.
  • Architectural choices in generative networks could be guided by the expected Hölder exponent to reduce approximation error.
  • The framework suggests that transport-based generators may scale better than iterative sampling methods when the target measure is known to be doubling.
  • Numerical PDE solvers could be recast as distribution-learning tasks whose accuracy is controlled by the same OT regularity.

Load-bearing premise

The normalized target densities from the listed PDEs meet the structural conditions that produce the doubling property for the associated measures.

What would settle it

Numerical approximation of the optimal transport map for the heat equation on a unit square, followed by direct verification of whether the map satisfies a Hölder bound with positive exponent strictly less than one.

Figures

Figures reproduced from arXiv: 2605.21388 by Jack Xin, Likun Lin, Zhiwen Zhang, Zhongjian Wang.

Figure 1
Figure 1. Figure 1: Validation Wasserstein–2 error versus sample size for the one- and two￾dimensional model problems. In each panel, the blue curve shows the observed mean validation error and the red line is the least-squares fit in log-log coordinates. One-dimensional experiment We use the source and target measures from Example 5.1 and the exact optimal transport map given there. The network architecture is [1, 256, 256, … view at source ↗
read the original abstract

Despite the remarkable empirical success of generative models, the available theory on their statistical accuracy in scientific computing remains largely pessimistic. This paper develops a theoretical framework for understanding the regularity of transport maps and the generalization properties of one-step Wasserstein-guided generative models for PDE-induced probability measures. We consider normalized target densities associated with linear elliptic and parabolic equations on bounded domains, as well as diffusion and Fokker--Planck equations on the torus. Under standard structural assumptions, we prove that these target measures satisfy doubling conditions. By combining this fact with regularity theory for optimal transport between doubling measures, we show that the optimal transport map from a uniform source measure to the target measure is H\"older continuous. This regularity yields an approximation-theoretic justification for one-step generative models that learn PDE-induced distributions via a single pushforward map. As a representative instance, we study DeepParticle and derive excess-risk bounds characterizing the discrepancy between the learned map and the population-optimal map. We also establish a robustness estimate under target shift and illustrate the theory with experiments which support the derived rates.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 1 minor

Summary. The paper develops a theoretical framework for the regularity of transport maps and generalization properties of one-step Wasserstein-guided generative models targeting PDE-induced probability measures. It considers normalized target densities from linear elliptic and parabolic equations on bounded domains as well as diffusion and Fokker-Planck equations on the torus. Under standard structural assumptions, the target measures are shown to satisfy doubling conditions; combined with existing OT regularity theory for doubling measures, this yields Hölder continuity of the optimal transport map from a uniform source to the target. The regularity is used to justify one-step generative models, with excess-risk bounds derived for the representative case of DeepParticle, plus a robustness estimate under target shift and supporting experiments.

Significance. If the central claims on doubling conditions and the resulting Hölder regularity hold with explicit verification, the work would provide a valuable approximation-theoretic justification for one-step generative models in scientific computing applications involving PDE-induced distributions. It connects PDE structure to OT map regularity in a way that could inform generalization bounds and model design for physical measures, and the excess-risk analysis for DeepParticle plus robustness result add concrete quantitative content.

major comments (1)
  1. [Abstract] Abstract: The central step asserting that the normalized target densities satisfy doubling conditions 'under standard structural assumptions' is load-bearing for the subsequent application of OT regularity theory and the Hölder continuity conclusion, yet the assumptions (e.g., uniform positivity/boundedness of the density, boundary behavior for domain cases, periodicity for torus cases) are invoked without explicit statement or verification that they hold uniformly across the four PDE families listed.
minor comments (1)
  1. The abstract references experiments that 'support the derived rates' but provides no quantitative details on the experimental setup, error bars, or specific rates shown; adding a brief summary or pointer to the relevant figure/table would improve clarity.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their careful reading of our manuscript and for the constructive comments. We address the major comment below.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central step asserting that the normalized target densities satisfy doubling conditions 'under standard structural assumptions' is load-bearing for the subsequent application of OT regularity theory and the Hölder continuity conclusion, yet the assumptions (e.g., uniform positivity/boundedness of the density, boundary behavior for domain cases, periodicity for torus cases) are invoked without explicit statement or verification that they hold uniformly across the four PDE families listed.

    Authors: We thank the referee for highlighting this point. In the manuscript, the structural assumptions are introduced and applied separately for each PDE family in the dedicated sections (linear elliptic and parabolic PDEs on bounded domains in Sections 3 and 4; diffusion and Fokker-Planck equations on the torus in Sections 5 and 6). These include uniform ellipticity and boundedness of coefficients, smooth or Lipschitz boundary conditions ensuring positive densities bounded away from zero and infinity, and standard periodicity for the torus cases. Under these conditions, the normalized densities are comparable to the Lebesgue measure on compact sets, which directly yields the doubling property via standard measure-theoretic arguments. We agree, however, that a unified and explicit enumeration would improve clarity and transparency. In the revision we will add a short preliminary subsection (new Section 2.3) that lists the precise assumptions for all four families side-by-side and recalls the elementary verification that each implies the doubling condition. This addition will be referenced from the abstract and introduction, without changing any theorems or proofs. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation applies external OT regularity to independently established doubling conditions

full rationale

The paper's chain begins with normalized target densities from linear elliptic/parabolic PDEs on domains and diffusion/Fokker-Planck equations on the torus. Under explicitly invoked standard structural assumptions, it proves these measures satisfy doubling conditions. It then combines this with existing regularity theory for optimal transport maps between doubling measures to obtain Hölder continuity of the map from uniform source to target. This supplies an approximation-theoretic justification for one-step models and excess-risk bounds for DeepParticle. No quoted step reduces by construction to a fitted input, self-definition, or load-bearing self-citation whose content is itself unverified; the OT regularity is treated as an external result, and the doubling proof is presented as a direct consequence of the structural assumptions rather than a renaming or ansatz smuggling. The derivation therefore remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claims rest on standard structural assumptions from PDE theory that are used to prove the doubling property; no free parameters or new entities are introduced in the abstract.

axioms (1)
  • domain assumption Standard structural assumptions on normalized target densities for linear elliptic, parabolic, diffusion and Fokker-Planck equations.
    Invoked to establish that the target measures satisfy doubling conditions.

pith-pipeline@v0.9.0 · 5734 in / 1354 out tokens · 44190 ms · 2026-05-21T05:20:50.248239+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

48 extracted references · 48 canonical work pages · 1 internal anchor

  1. [1]

    Wasserstein gener- ative adversarial networks

    Martin Arjovsky, Soumith Chintala, and L´ eon Bottou. Wasserstein gener- ative adversarial networks. In Doina Precup and Yee Whye Teh, editors, Proceedings of the 34th International Conference on Machine Learning, volume 70 ofProceedings of Machine Learning Research, pages 214–223. PMLR, 06–11 Aug 2017

    work page 2017

  2. [2]

    Some theoretical insights into wasserstein GANs.Journal of Machine Learning Research, 22(119):1–45, 2021

    G´ erard Biau, Maxime Sangnier, and Ugo Tanielian. Some theoretical insights into wasserstein GANs.Journal of Machine Learning Research, 22(119):1–45, 2021

    work page 2021

  3. [3]

    American Mathematical Society, Providence, Rhode Island, 1 edition, 2019

    Sergey Bobkov and Michel Ledoux.One-dimensional empirical measures, order statistics, and Kantorovich transport distances, volume 261 ofMemoirs of the American Mathematical Society. American Mathematical Society, Providence, Rhode Island, 1 edition, 2019

    work page 2019

  4. [4]

    Caffarelli

    Luis A. Caffarelli. The regularity of mappings with a convex potential. Journal of the American Mathematical Society, 5(1):99–104, 1992

    work page 1992

  5. [5]

    Bartlett

    Saptarshi Chakraborty and Peter L. Bartlett. On the statistical properties of generative adversarial models for low intrinsic data dimension.Journal of Machine Learning Research, 26:1–80, 2025

    work page 2025

  6. [6]

    Distribution approximation and statistical estimation guarantees of generative adversarial networks, 2022

    Minshuo Chen, Wenjing Liao, Hongyuan Zha, and Tuo Zhao. Distribution approximation and statistical estimation guarantees of generative adversarial networks, 2022

    work page 2022

  7. [7]

    Xiaoli Chen, Phoebus Rosakis, Zhizhang Wu, and Zhiwen Zhang. Solving nonconvex energy minimization problems in martensitic phase transitions with a mesh-free deep learning approach.Computer Methods in Applied Mechanics and Engineering, 416:116384, 2023

    work page 2023

  8. [8]

    Podno: Proper orthogonal decomposition neural operators, 2025

    Zilan Cheng, Zhongjian Wang, Li-Lian Wang, and Mejdi Azaiez. Podno: Proper orthogonal decomposition neural operators, 2025

    work page 2025

  9. [9]

    A variational neural network approach for glacier modelling with nonlinear rheology.Communi- cations in Computational Physics, 34(4):934–954, 2023

    Tiangang Cui, Zhongjian Wang, and Zhiwen Zhang. A variational neural network approach for glacier modelling with nonlinear rheology.Communi- cations in Computational Physics, 34(4):934–954, 2023

    work page 2023

  10. [10]

    Convergence of denoising diffusion models under the manifold hypoth- esis.arXiv preprint arXiv:2208.05314,

    Valentin De Bortoli. Convergence of denoising diffusion models under the manifold hypothesis.arXiv preprint arXiv:2208.05314, 2022

    work page arXiv 2022

  11. [11]

    The deep Ritz method: A deep learning-based numerical algorithm for solving variational problems.Communications in Mathematics and Statistics, 6(1):1–12, 2018

    Weinan E and Bing Yu. The deep Ritz method: A deep learning-based numerical algorithm for solving variational problems.Communications in Mathematics and Statistics, 6(1):1–12, 2018. 24

    work page 2018

  12. [12]

    Evans.Partial differential equations

    Lawrence C. Evans.Partial differential equations. American Mathematical Society, Providence, R.I., 2010

    work page 2010

  13. [13]

    On the rate of convergence in wasser- stein distance of the empirical measure.Probability Theory and Related Fields, 162:707–738, 2015

    Nicolas Fournier and Arnaud Guillin. On the rate of convergence in wasser- stein distance of the empirical measure.Probability Theory and Related Fields, 162:707–738, 2015

    work page 2015

  14. [14]

    Trudinger.Elliptic partial differential equations of second order

    David Gilbarg and Neil S. Trudinger.Elliptic partial differential equations of second order. Grundlehren der mathematischen Wissenschaften ; 224. Springer-Verlag, Berlin, 2nd ed. edition, 1983

    work page 1983

  15. [15]

    Goodfellow, J

    I. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. Courville, and Y. Bengio. Generative adversarial nets.Advances in Neural Information Processing Systems, 27, 2014

    work page 2014

  16. [16]

    Jain, and Pieter Abbeel

    Jonathan Ho, Ajay N. Jain, and Pieter Abbeel. Denoising diffusion prob- abilistic models. InAdvances in Neural Information Processing Systems, volume 33. Curran Associates, Inc., 2020

    work page 2020

  17. [17]

    Neural network Poisson-Boltzmann electrostatics for biomolecular interactions

    Zunding Huang, Bo Li, Zhongming Wang, and Zhiwen Zhang. Neural network Poisson-Boltzmann electrostatics for biomolecular interactions. Journal of Computational Physics, page 114446, 2025

    work page 2025

  18. [18]

    On the regularity of optimal transports between degenerate densities.Archive for Rational Mechanics and Analysis, 245(2):819–861, June 2022

    Yash Jhaveri and Ovidiu Savin. On the regularity of optimal transports between degenerate densities.Archive for Rational Mechanics and Analysis, 245(2):819–861, June 2022

    work page 2022

  19. [19]

    A generalization of caffarelli’s con- traction theorem via (reverse) heat flow.Mathematische annalen, 354(3):827– 862, 2012

    Young-Heon Kim and Emanuel Milman. A generalization of caffarelli’s con- traction theorem via (reverse) heat flow.Mathematische annalen, 354(3):827– 862, 2012

    work page 2012

  20. [20]

    Kobyzev, S

    I. Kobyzev, S. Prince, and M. Brubaker. Normalizing flows: An introduction and review of current methods.IEEE Transactions on Pattern Analysis and Machine Intelligence, 2020

    work page 2020

  21. [21]

    Approximating smooth functions by deep neural net- works with sigmoid activation function.Journal of Multivariate Analysis, 182:104696, 2021

    Sophie Langer. Approximating smooth functions by deep neural net- works with sigmoid activation function.Journal of Multivariate Analysis, 182:104696, 2021

    work page 2021

  22. [22]

    Fourier neural operator for parametric partial differential equations

    Zongyi Li, Nikola Kovachki, Kamyar Azizzadenesheli, Burigede Liu, Kaushik Bhattacharya, Andrew Stuart, and Anima Anandkumar. Fourier neural operator for parametric partial differential equations. InInternational Conference on Learning Representations, 2021

    work page 2021

  23. [23]

    Physics- informed neural operator for learning partial differential equations.ACM / IMS J

    Zongyi Li, Hongkai Zheng, Nikola Kovachki, David Jin, Haoxuan Chen, Burigede Liu, Kamyar Azizzadenesheli, and Anima Anandkumar. Physics- informed neural operator for learning partial differential equations.ACM / IMS J. Data Sci., 1(3), May 2024. 25

    work page 2024

  24. [24]

    How well generative adversarial networks learn distribu- tions.Journal of Machine Learning Research, 22(228):1–41, 2021

    Tengyuan Liang. How well generative adversarial networks learn distribu- tions.Journal of Machine Learning Research, 22(228):1–41, 2021

    work page 2021

  25. [25]

    G. M. Lieberman.Second Order Parabolic Differential Equations. World Scientific, 1996

    work page 1996

  26. [26]

    Cell-induced densification and tether formation in fibrous extracellular matrices with biomimetic physics-informed neural networks

    Anci Lin, Xiaohong Liu, Zhiwen Zhang, Weidong Zhao, and Wenju Zhao. Biomimetic PINNs for cell-induced phase transitions: UQ-R3 sampling with causal gating.arXiv preprint arXiv:2603.29184, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  27. [27]

    Learning nonlinear operators via DeepONet based on the universal approximation theorem of operators.Nature Machine Intelligence, 3:218–229, 2021

    Lu Lu, Pengzhan Jin, Guofei Pang, Zhongqiang Zhang, and George Em Karniadakis. Learning nonlinear operators via DeepONet based on the universal approximation theorem of operators.Nature Machine Intelligence, 3:218–229, 2021

    work page 2021

  28. [28]

    A convergent interacting particle method and computation of KPP front speeds in chaotic flows.SIAM Journal on Numerical Analysis, 60(3):1136–1167, 2022

    Junlong Lyu, Zhongjian Wang, Jack Xin, and Zhiwen Zhang. A convergent interacting particle method and computation of KPP front speeds in chaotic flows.SIAM Journal on Numerical Analysis, 60(3):1136–1167, 2022

    work page 2022

  29. [29]

    Opti- mal transport mapping via input convex neural networks

    Ashok Makkuva, Amirhossein Taghvaei, Sewoong Oh, and Jason Lee. Opti- mal transport mapping via input convex neural networks. In Hal Daum´ e III and Aarti Singh, editors,Proceedings of the 37th International Confer- ence on Machine Learning, volume 119 ofProceedings of Machine Learning Research, pages 6672–6681. PMLR, 13–18 Jul 2020

    work page 2020

  30. [30]

    Pathway to o( √ d) complexity bound under wasserstein metric of flow-based models, 2025

    Xiangjun Meng and Zhongjian Wang. Pathway to o( √ d) complexity bound under wasserstein metric of flow-based models, 2025

    work page 2025

  31. [31]

    Spectral normalization for generative adversarial networks, 2018

    Takeru Miyato, Toshiki Kataoka, Masanori Koyama, and Yuichi Yoshida. Spectral normalization for generative adversarial networks, 2018

    work page 2018

  32. [32]

    Adaptive computation and machine learning series

    Mehryar Mohri, Afshin Rostamizadeh, and Ameet Talwalkar.Foundations of machine learning. Adaptive computation and machine learning series. MIT Press, Cambridge, MA, 2012

    work page 2012

  33. [33]

    Lawrence E. Payne. Maximum principles in differential equations (murray h. protter and hans f. weinberger).SIAM Review, 10(3):386–387, July 1968

    work page 1968

  34. [34]

    Raissi, P

    M. Raissi, P. Perdikaris, and G. Karniadakis. Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations.Journal of Computational Physics, 378:686–707, 2019

    work page 2019

  35. [35]

    Z. Shen, Z. Wang, J. Xin, and Z. Zhang. Two-step diffusion: Fast sampling and reliable prediction for 3D Keller-Segel and KPP equations in fluid flows, 2026

    work page 2026

  36. [36]

    Deep network approximation characterized by number of neurons.Communications in Computational Physics, 28(5):1768–1811, November 2020

    Zuowei Shen, Haizhao Yang, and Shijun Zhang. Deep network approximation characterized by number of neurons.Communications in Computational Physics, 28(5):1768–1811, November 2020. 26

    work page 2020

  37. [37]

    Dgm: A deep learning algorithm for solving partial differential equations.Journal of Computational Physics, 375:1339–1364, 2018

    Justin Sirignano and Konstantinos Spiliopoulos. Dgm: A deep learning algorithm for solving partial differential equations.Journal of Computational Physics, 375:1339–1364, 2018

    work page 2018

  38. [38]

    Wasserstein generative adversarial networks are minimax optimal distribution estimators

    Arthur St´ ephanovitch, Eddie Aamari, and Cl´ ement Levrard. Wasserstein generative adversarial networks are minimax optimal distribution estimators. The Annals of Statistics, 52(5):2167 – 2193, 2024

    work page 2024

  39. [39]

    Wasserstein bounds for generative diffusion models with gaussian tail targets.arXiv preprint arXiv:2412.11251, 2024

    Xixian Wang and Zhongjian Wang. Wasserstein bounds for generative diffusion models with gaussian tail targets.arXiv preprint arXiv:2412.11251, 2024

    work page arXiv 2024

  40. [40]

    Z. Wang, J. Xin, and Z. Zhang. DeepParticle: Learning invariant measure by a deep neural network minimizing Wasserstein distance on data generated from an interacting particle method.Journal of Computational Physics, 464:111309, 2022

    work page 2022

  41. [41]

    Z. Wang, J. Xin, and Z. Zhang. A DeepParticle method for learning and gen- erating aggregation patterns in multi-dimensional Keller-Segel chemotaxis systems.Physica D, 460:134082, 2024

    work page 2024

  42. [42]

    Computing effective dif- fusivity of chaotic and stochastic flows using structure-preserving schemes

    Zhongjian Wang, Jack Xin, and Zhiwen Zhang. Computing effective dif- fusivity of chaotic and stochastic flows using structure-preserving schemes. SIAM Journal on Numerical Analysis, 56(4):2322–2344, 2018

    work page 2018

  43. [43]

    Sharp error estimates on a stochastic structure-preserving scheme in computing effective diffusivity of 3D chaotic flows.Multiscale Modeling & Simulation, 19(3):1167–1189, 2021

    Zhongjian Wang, Jack Xin, and Zhiwen Zhang. Sharp error estimates on a stochastic structure-preserving scheme in computing effective diffusivity of 3D chaotic flows.Multiscale Modeling & Simulation, 19(3):1167–1189, 2021

    work page 2021

  44. [44]

    A mesh-free method for interface problems using the deep learning approach.Journal of Computational Physics, 400:108963, 2020

    Zhongjian Wang and Zhiwen Zhang. A mesh-free method for interface problems using the deep learning approach.Journal of Computational Physics, 400:108963, 2020

    work page 2020

  45. [45]

    Zhizhang Wu, Renaud Raqu´ epas, Jack Xin, and Zhiwen Zhang. Computing large deviation rate functions of entropy production for diffusion processes by an interacting particle method.SIAM Journal on Scientific Computing, 47(6):A3330–A3355, 2025

    work page 2025

  46. [46]

    On the rates of conver- gence for learning with convolutional neural networks.SIAM Journal on Mathematics of Data Science, 7(4):1755–1772, 2025

    Yunfei Yang, Han Feng, and Ding-Xuan Zhou. On the rates of conver- gence for learning with convolutional neural networks.SIAM Journal on Mathematics of Data Science, 7(4):1755–1772, 2025

    work page 2025

  47. [47]

    Zhang, Z

    T. Zhang, Z. Wang, J. Xin, and Z. Zhang. A bidirectional DeepParticle method for efficiently solving low-dimensional transport map problems. Journal of Computational Physics, 2026 to be appear

    work page 2026

  48. [48]

    A convergent interacting particle method for computing KPP front speeds in random flows.SIAM/ASA Journal on Uncertainty Quantification, 13(2):639–678, 2025

    Tan Zhang, Zhongjian Wang, Jack Xin, and Zhiwen Zhang. A convergent interacting particle method for computing KPP front speeds in random flows.SIAM/ASA Journal on Uncertainty Quantification, 13(2):639–678, 2025. 27

    work page 2025