pith. sign in

arxiv: 2605.08916 · v3 · pith:HG6E6ZSFnew · submitted 2026-05-09 · 💻 cs.CE · cs.NA· math.NA· math.PR

Diffusion Restore: Real-Time Markov Chain Monte Carlo Light Transport

Sascha Holl , Gurprit Singh , Hans-Peter Seidel This is my paper

Pith reviewed 2026-05-21 08:56 UTC · model grok-4.3

classification 💻 cs.CE cs.NAmath.NAmath.PR
keywords MCMC light transportdiffusion dynamicsnonreversible Markov chainsRestore frameworkreal-time renderingpath samplingGPU ray tracing
0
0 comments X

The pith

Diffusion Restore integrates nonreversible diffusion dynamics into the Restore framework to enable faster MCMC sampling for light transport without Metropolis adjustments.

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

The paper establishes that diffusion-based local dynamics can be selected inside the Restore framework for MCMC light transport such that they stay valid without any Metropolis adjustment. By modeling the dynamics as nonreversible and adding momentum to the drift, the approach produces more directed local exploration that avoids excessive backtracking while still balancing global discovery. A sympathetic reader would care because MCMC sampling is often the only practical way to handle complex, high-dimensional light path distributions when direct sampling fails, and improved local dynamics directly raise the quality and speed of the resulting integral approximations. The authors supply theoretical support for the dynamics and show through experiments that the method beats prior MCMC techniques across many scenes while running in real time on GPU hardware, even surpassing conventional path tracing for interactive use.

Core claim

We present Diffusion Restore, a real-time framework for diffusion-based MCMC light transport. We show how to choose diffusion-based local dynamics within the Restore framework while completely avoiding Metropolis adjustment. Furthermore, we model these dynamics as nonreversible, introducing momentum in the drift and thereby enabling more directed exploration of the target distribution compared to reversible, random-walk-like dynamics. We provide a theoretical justification for the validity of our choice of local dynamics.

What carries the argument

Nonreversible diffusion-based local dynamics with momentum in the drift, placed inside the Restore framework to handle local exploration while the framework manages global discovery.

If this is right

  • The method outperforms all existing MCMC light transport methods across diverse scenes.
  • It establishes a new state of the art in MCMC sampling efficiency for light transport.
  • A GPU implementation in ray tracing and compute shaders delivers real-time frame rates.
  • It outperforms traditional Path Tracing in real-time rendering settings such as interactive applications and games.

Where Pith is reading between the lines

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

  • Similar nonreversible diffusion dynamics could be tested in other MCMC settings that involve high-dimensional integrals, such as global illumination variants or related Monte Carlo problems.
  • The real-time capability opens the door to using MCMC sampling inside dynamic or interactive pipelines that previously relied only on faster but biased methods.
  • One could examine whether the momentum term helps or hinders sampling when the target distribution changes over time, as in animated scenes.

Load-bearing premise

The chosen diffusion-based local dynamics stay valid for the light transport distribution without Metropolis adjustment and gain better exploration from nonreversibility and momentum.

What would settle it

Run the sampler on a scene whose path space contains isolated modes and measure whether the variance of the lighting estimate drops slower than with reversible dynamics or whether the GPU implementation fails to sustain interactive frame rates.

Figures

Figures reproduced from arXiv: 2605.08916 by Gurprit Singh, Hans-Peter Seidel, Sascha Holl.

Figure 1
Figure 1. Figure 1: We implement a novel diffusion-based real-time Markov chain Monte Carlo (MCMC) framework on the GPU. The [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Exploration of a mode of a target density [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: MSE as a function of rendering time (up to 360 seconds, averaged over 100 realizations) for the scenes shown in [PITH_FULL_IMAGE:figures/full_fig_p015_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Equal-rendering-time comparison (1/30 seconds of computation) of PT (left), MALA Restore (middle), and Diffusion Restore (right) for the VEACH, AJAR scene provided by Bitterli (2016). The inset error maps show the corresponding mean absolute percentage error (MAPE) for each rendering. Path Tracing MALA Restore Diffusion Restore Path Tracing MALA Restore Diffusion Restore [PITH_FULL_IMAGE:figures/full_fig_… view at source ↗
Figure 5
Figure 5. Figure 5: Equal-rendering-time comparison (1/30 seconds of computation) of PT (left), MALA Restore (middle), and Diffusion Restore (right) for the TORUS scene provided by Li et al. (2015). The inset error maps show the corresponding MAPE for each rendering. 16 [PITH_FULL_IMAGE:figures/full_fig_p016_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Equal-rendering-time comparison (1/30 seconds of computation) of PT (left), MALA Restore (middle), and Diffusion Restore (right) for the GLASS OF WATER scene provided by Bitterli (2016). The inset error maps show the corresponding MAPE for each rendering. Path Tracing MALA Restore Diffusion Restore Path Tracing MALA Restore Diffusion Restore [PITH_FULL_IMAGE:figures/full_fig_p017_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Equal-rendering-time comparison (1/30 seconds of computation) of PT (left), MALA Restore (middle), and Diffusion Restore (right) for the SALLE DE BAIN scene provided by Bitterli (2016). The inset error maps show the corresponding MAPE for each rendering. Path Tracing MALA Restore Diffusion Restore Path Tracing MALA Restore Diffusion Restore [PITH_FULL_IMAGE:figures/full_fig_p017_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Equal-rendering-time comparison (1/30 seconds of computation) of PT (left), MALA Restore (middle), and Diffusion Restore (right) for the SWIMMING POOL scene provided by Rioux-Lavoie et al. (2020). The inset error maps show the corresponding MAPE for each rendering. 17 [PITH_FULL_IMAGE:figures/full_fig_p017_8.png] view at source ↗
read the original abstract

We present Diffusion Restore, a real-time framework for diffusion-based MCMC light transport. MCMC methods are highly suitable for sampling from complex high-dimensional distributions and for approximating integrals over them. In practice, they are often the only viable solution when direct sampling is not possible and alternative methods are either inefficient or cannot be applied due to the structure of the target distribution. However, controlling the exploration of the target distribution in MCMC methods remains challenging. Efficient exploration requires a balance between local exploration and global discovery, and local dynamics must rapidly explore individual modes without getting stuck or exhibiting excessive backtracking. The problem of global discovery has recently been addressed by the introduction of the Restore framework. In this work, we build on this framework and focus on improving local exploration. We show how to choose diffusion-based local dynamics within the Restore framework while completely avoiding Metropolis-adjustment, which is known to slow down convergence. Furthermore, we model these dynamics as nonreversible, introducing momentum in the drift and thereby enabling more directed exploration of the target distribution compared to reversible, random-walk-like dynamics. We provide a theoretical justification for the validity of our choice of local dynamics. Empirically, we demonstrate across diverse scenes that Diffusion Restore outperforms all existing MCMC light transport methods and establishes a new state of the art. In addition, we present a GPU implementation in ray tracing and compute shaders and achieve real-time frame rates. This demonstrates that Diffusion Restore is not only superior in offline rendering, but also outperforms traditional Path Tracing methods in real-time rendering settings, such as interactive applications and games.

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

2 major / 2 minor

Summary. The paper introduces Diffusion Restore, a real-time MCMC framework for light transport that extends the Restore approach with diffusion-based local dynamics. These dynamics are chosen to be nonreversible (incorporating momentum in the drift) and are used without any Metropolis-Hastings correction. The authors provide a theoretical justification for why the chosen generator leaves the target path-space measure invariant, demonstrate empirical superiority over prior MCMC light-transport methods across diverse scenes, and report a GPU implementation (ray tracing plus compute shaders) that achieves real-time frame rates, outperforming traditional path tracing in interactive settings.

Significance. If the nonreversible diffusion dynamics are shown to preserve the correct stationary distribution and the empirical gains hold under rigorous error analysis, the work would represent a meaningful advance in MCMC rendering: it directly addresses the local-exploration bottleneck that has limited MCMC methods in production, while simultaneously moving the technique into the real-time regime. The combination of a parameter-light nonreversible sampler with a practical GPU realization is a concrete strength that could influence both offline and interactive rendering pipelines.

major comments (2)
  1. [Theoretical justification] Theoretical justification section (around the derivation of the nonreversible generator): the claim that the chosen drift and diffusion coefficients leave the target path measure invariant without Metropolis adjustment is load-bearing for the entire method. The manuscript should supply an explicit verification—either a Fokker-Planck derivation showing that the stationary density equals the desired path-space measure, or a direct numerical check (e.g., long-run histogram convergence on a known low-dimensional integral). The current abstract-level statement is insufficient to confirm correctness of the sampler.
  2. [§4] §4 (or wherever the local dynamics are defined): the momentum term and diffusion coefficients are stated to be free parameters. The paper must clarify whether these parameters are tuned per scene or held fixed across the reported experiments; if they are scene-dependent, the “parameter-free” or “theoretically justified” framing of the dynamics needs qualification.
minor comments (2)
  1. [Results figures] Figure captions and axis labels in the real-time performance plots should explicitly state the hardware platform, resolution, and whether the reported frame rates include denoising or only raw sampling.
  2. [Comparison tables] The comparison tables would benefit from an additional column reporting effective sample size per unit time (or a similar normalized efficiency metric) rather than raw error alone, to separate exploration quality from raw speed.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive feedback and for recognizing the potential of Diffusion Restore to advance MCMC methods in rendering. We address each major comment below with clarifications and revisions to strengthen the manuscript.

read point-by-point responses

  1. Referee: [Theoretical justification] Theoretical justification section (around the derivation of the nonreversible generator): the claim that the chosen drift and diffusion coefficients leave the target path measure invariant without Metropolis adjustment is load-bearing for the entire method. The manuscript should supply an explicit verification—either a Fokker-Planck derivation showing that the stationary density equals the desired path-space measure, or a direct numerical check (e.g., long-run histogram convergence on a known low-dimensional integral). The current abstract-level statement is insufficient to confirm correctness of the sampler.

    Authors: We agree that an explicit verification strengthens the theoretical foundation. The original manuscript presented a high-level argument based on the Restore framework's invariance properties combined with the specific form of the nonreversible generator. In the revision we will add a dedicated subsection containing the full Fokker-Planck derivation that shows the stationary density is exactly the target path-space measure. We will also include a brief numerical sanity check on a low-dimensional integral to illustrate convergence to the correct distribution. revision: yes

  2. Referee: [§4] §4 (or wherever the local dynamics are defined): the momentum term and diffusion coefficients are stated to be free parameters. The paper must clarify whether these parameters are tuned per scene or held fixed across the reported experiments; if they are scene-dependent, the “parameter-free” or “theoretically justified” framing of the dynamics needs qualification.

    Authors: The momentum and diffusion coefficients are held fixed for all scenes and experiments; they were selected once on the basis of the invariance condition and kept constant thereafter. We will revise Section 4 to state this explicitly and to qualify the “parameter-free” description by noting that the functional form is theoretically justified while the specific numerical values are chosen to ensure both invariance and practical mixing, remaining unchanged across the reported results. revision: yes

Circularity Check

0 steps flagged

Builds on Restore framework with independent theoretical justification for new nonreversible dynamics

full rationale

The paper extends the Restore framework by introducing diffusion-based local dynamics without Metropolis adjustment and with nonreversible momentum terms. It explicitly claims to provide a theoretical justification for the validity of these choices as preserving the target path measure. No equations or steps in the provided abstract reduce a prediction or result to a fitted parameter or self-definition by construction. The central sampler validity claim has independent content beyond the cited framework, and empirical superiority is presented separately. This yields only minor self-citation load that is not load-bearing for the new contributions.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on the validity of the Restore framework for global discovery, the theoretical justification for the new local dynamics, and empirical results across scenes; specific diffusion and momentum parameters are likely present but not quantified in the abstract.

free parameters (1)
  • diffusion and momentum parameters
    Parameters controlling the diffusion process and momentum in the nonreversible dynamics are introduced or tuned to achieve the reported exploration behavior.
axioms (1)
  • domain assumption The Restore framework correctly addresses global discovery in MCMC sampling.
    The paper explicitly builds on this framework to focus on local exploration improvements.

pith-pipeline@v0.9.0 · 5819 in / 1226 out tokens · 55522 ms · 2026-05-21T08:56:10.111163+00:00 · methodology

Review history (3 revisions) →

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

96 extracted references · 96 canonical work pages

  1. [1]

    The Annals of Mathematical Statistics , volume =

    Blackwell, David , title =. The Annals of Mathematical Statistics , volume =

    work page

  2. [2]

    Radhakrishna , title =

    Rao, C. Radhakrishna , title =. Bulletin of the Calcutta Mathematical Society , volume =

    work page

  3. [3]

    ACM Trans

    Bitterli, Benedikt and Wyman, Chris and Pharr, Matt and Shirley, Peter and Lefohn, Aaron and Jarosz, Wojciech , title =. ACM Trans. Graph. , month = aug, articleno =. 2020 , issue_date =. doi:10.1145/3386569.3392481 , abstract =

    work page doi:10.1145/3386569.3392481 2020

  4. [4]

    Bitterli, Benedikt , Year =

    work page

  5. [5]

    2025 , howpublished =

    Pharr, Matt , title =. 2025 , howpublished =

    work page 2025

  6. [6]

    PAMM , year =

    Daniel Rudolf and Björn Sprungk , title =. PAMM , year =

    work page

  7. [7]

    Çinlar, Erhan , title =

    work page

  8. [8]

    Randal Douc and Eric Moulines and Pierre Priouret and Philippe Soulier , title =

    work page

  9. [9]

    Engel, Klaus-Jochen and Nagel, Rainer , title =

    work page

  10. [10]

    Ethier and Thomas G

    Stewart N. Ethier and Thomas G. Kurtz , title =. 2009 , publisher=

    work page 2009

  11. [11]

    , title =

    Kajiya, James T. , title =. 1986 , issue_date =. doi:10.1145/15886.15902 , journal =

    work page doi:10.1145/15886.15902 1986

  12. [12]

    2021 , url =

    Pharr, Matt and Jakob, Wenzel and Humphreys, Greg , title =. 2021 , url =

    work page 2021

  13. [13]

    and Willems, Yves D

    Lafortune, Eric P. and Willems, Yves D. Rendering Participating Media with Bidirectional Path Tracing. Rendering Techniques '96. 1996

    work page 1996

  14. [14]

    Bidirectional Estimators for Light Transport

    Veach, Eric and Guibas, Leonidas. Bidirectional Estimators for Light Transport. Photorealistic Rendering Techniques. 1995

    work page 1995

  15. [15]

    Foundations of Modern Probability , author =

    work page

  16. [16]

    Probability Theory: A Comprehensive Course , author =

    work page

  17. [17]

    Markov Chains and Stochastic Stability , author =

    work page

  18. [18]

    Durmus, Alain Oliviero and Eberle, Andreas , title =

    work page

  19. [19]

    and Roberts, Gareth O

    Robert, Christian P. and Roberts, Gareth O. , year =. Rao-. 2101.01011 , archivePrefix =

    work page arXiv

  20. [20]

    2009 , eprint =

    Does waste-recycling really improve Metropolis-Hastings Monte Carlo algorithm? , author =. 2009 , eprint =

    work page 2009

  21. [21]

    Monte Carlo simulation of a many-fermion study , author =. Phys. Rev. B , volume =. 1977 , month =. doi:10.1103/PhysRevB.16.3081 , url =

    work page doi:10.1103/physrevb.16.3081 1977

  22. [22]

    , volume =

    Douc, Randal and Robert, Christian P. , volume =. A vanilla. The Annals of Statistics , number =. 2011 , month =

    work page 2011

  23. [23]

    The Journal of Chemical Physics , volume =

    Equation of state calculations by fast computing machines , author =. The Journal of Chemical Physics , volume =. 1953 , month =

    work page 1953

  24. [24]

    Hastings, W. K. , journal =. Monte. 1970 , month =

    work page 1970

  25. [25]

    1998 , journal =

    Tierney, Luke , title =. 1998 , journal =

    work page 1998

  26. [26]

    Robust Monte Carlo Methods for Light Transport Simulation , author =

    work page

  27. [27]

    A simple and robust mutation strategy for the

    Kelemen, Csaba and Szirmay-Kalos, László and Antal, György and Csonka, Ferenc , journal =. A simple and robust mutation strategy for the. 2002 , volume =

    work page 2002

  28. [28]

    Bernoulli , year =

    Exponential convergence of Langevin distributions and their discrete approximations , author =. Bernoulli , year =

    work page

  29. [29]

    Physics Letters B , year =

    Hybrid Monte Carlo , author =. Physics Letters B , year =

    work page

  30. [30]

    Langevin

    Luan, Fujun and Zhao, Shuang and Bala, Kavita and Gkioulekas, Ioannis , journal =. Langevin. 2020 , volume =

    work page 2020

  31. [31]

    Anisotropic gaussian mutations for

    Li, Tzu-Mao and Lehtinen, Jaakko and Ramamoorthi, Ravi and Jakob, Wenzel and Durand, Fr\'. Anisotropic gaussian mutations for. ACM Transactions on Graphics , year =

    work page

  32. [32]

    Europhysics Letters (EPL) , publisher =

    Simulated Tempering: A New Monte Carlo Scheme , author =. Europhysics Letters (EPL) , publisher =

    work page

  33. [33]

    Journal of Statistics , year =

    Mira, Antonietta , title =. Journal of Statistics , year =

    work page

  34. [34]

    Liu, J. S. and Liang, F. and Wong, W. H. , title =. Journal of the American Statistical Association , year =

    work page

  35. [35]

    , title =

    Green, P.J. , title =. Biometrika , year =

    work page

  36. [36]

    ACM Transactions on Graphics , year =

    Pantaleoni, Jacopo , title =. ACM Transactions on Graphics , year =

    work page

  37. [37]

    ACM Transactions on Graphics , year =

    Rioux-Lavoie, Damien and Litalien, Joey and Gruson, Adrien and Hachisuka, Toshiya and Nowrouzezahrai, Derek , title =. ACM Transactions on Graphics , year =

    work page

  38. [38]

    Coherent

    Segovia, Benjamin and Iehl, Jean Claude and P. Coherent. 2007 , url =

    work page 2007

  39. [39]

    Reversible jump

    Bitterli, Benedikt and Jakob, Wenzel and Nov\'. Reversible jump. ACM Transactions on Graphics , year =

    work page

  40. [40]

    ACM Transactions on Graphics (TOG) , volume=

    Gradient-domain metropolis light transport , author=. ACM Transactions on Graphics (TOG) , volume=. 2013 , publisher=

    work page 2013

  41. [41]

    ACM Transactions on Graphics (TOG) , volume=

    Ensemble Metropolis Light Transport , author=. ACM Transactions on Graphics (TOG) , volume=. 2021 , publisher=

    work page 2021

  42. [42]

    and Hanika, Johannes and Dachsbacher, Carsten and Hachisuka, Toshiya , title =

    Otsu, Hisanari and Kaplanyan, Anton S. and Hanika, Johannes and Dachsbacher, Carsten and Hachisuka, Toshiya , title =. ACM Transactions on Graphics , year =

    work page

  43. [43]

    ACM Transactions on Graphics , year =

    Otsu, Hisanari and Hanika, Johannes and Hachisuka, Toshiya and Dachsbacher, Carsten , title =. ACM Transactions on Graphics , year =

    work page

  44. [44]

    Computer Graphics Forum , year =

    Hanika, Johannes and Kaplanyan, Anton and Dachsbacher, Carsten , title =. Computer Graphics Forum , year =

    work page

  45. [45]

    ACM Transactions on Graphics , year =

    Jakob, Wenzel and Marschner, Steve , title =. ACM Transactions on Graphics , year =

    work page

  46. [46]

    and Dachsbacher, Carsten , journal =

    Hachisuka, Toshiya and Kaplanyan, Anton S. and Dachsbacher, Carsten , journal =. Multiplexed. 2014 , volume =

    work page 2014

  47. [47]

    Stratified

    Gruson, Adrien and West, Rex and Hachisuka, Toshiya , journal =. Stratified

    work page

  48. [48]

    , school =

    Wang, Andi Q. , school =. Theory of killing and regeneration in continuous-time

    work page

  49. [49]

    and Pollock, Murray and Roberts, Gareth O

    Wang, Andi Q. and Pollock, Murray and Roberts, Gareth O. and Steinsaltz, David , title =. The Annals of Applied Probability , year =

    work page

  50. [50]

    2024 , eprint =

    Sampling using Adaptive Regenerative Processes , author =. 2024 , eprint =

    work page 2024

  51. [51]

    2024 , eprint=

    Concatenation of Markov processes , author=. 2024 , eprint=

    work page 2024

  52. [52]

    2025 , issue_date =

    Holl, Sascha and Singh, Gurprit and Seidel, Hans-Peter , title =. 2025 , issue_date =

    work page 2025

  53. [53]

    SIGGRAPH Conference Papers '26: Special Interest Group on Computer Graphics and Interactive Techniques Conference Papers , year =

    Holl, Sascha and Singh, Gurprit and Seidel, Hans-Peter , title =. SIGGRAPH Conference Papers '26: Special Interest Group on Computer Graphics and Interactive Techniques Conference Papers , year =. doi:10.1145/3799902.3811041 , isbn =

    work page doi:10.1145/3799902.3811041

  54. [54]

    dpt , url =

    Li, Tzu-Mao and Lehtinen, Jaakko and Ramamoorthi, Ravi and Jakob, Wenzel and Durand, Fr\'. dpt , url =

    work page

  55. [55]

    Pharr, Matt and Jakob, Wenzel and Humphreys, Greg , title =

    work page

  56. [56]

    Luan, Fujun and Zhao, Shuang and Bala, Kavita and Gkioulekas, Ioannis , title =

    work page

  57. [57]

    2026 , eprint=

    Forget Superresolution, Sample Adaptively (when Path Tracing) , author=. 2026 , eprint=

    work page 2026

  58. [58]

    Denoising with Kernel Prediction and Asymmetric Loss Functions , journal =

    Vogels, Thijs and Rousselle, Fabrice and McWilliams, Brian and R. Denoising with Kernel Prediction and Asymmetric Loss Functions , journal =. 2018 , month = jul, pages =

    work page 2018

  59. [59]

    Computer Graphics Forum , volume =

    Kuznetsov, Alexandr and Khademi Kalantari, Nima and Ramamoorthi, Ravi , title =. Computer Graphics Forum , volume =. 2018 , doi =

    work page 2018

  60. [60]

    , title =

    Mitchell, Don P. , title =. SIGGRAPH Computer Graphics , volume =. 1987 , month = aug, doi =

    work page 1987

  61. [61]

    Lee, M. E. and Redner, R. A. , title =. IEEE Computer Graphics and Applications , volume =. 1990 , doi =

    work page 1990

  62. [62]

    and Ward, Gregory J

    Rushmeier, Holly E. and Ward, Gregory J. , title =. Proceedings of SIGGRAPH '94 , pages =. 1994 , publisher =

    work page 1994

  63. [63]

    and Pattanaik, S

    Xu, R. and Pattanaik, S. N. , title =. IEEE Computer Graphics and Applications , volume =. 2005 , doi =

    work page 2005

  64. [64]

    and Donner, Craig and Ramamoorthi, Ravi , title =

    Overbeck, Ryan S. and Donner, Craig and Ramamoorthi, Ravi , title =. ACM Transactions on Graphics (TOG) , volume =. 2009 , month = dec, doi =

    work page 2009

  65. [65]

    Proceedings of the ACM on Computer Graphics and Interactive Techniques , volume =

    Kazmierczyk, Pawel and Kim, Sungye and Uss, Wojciech and Kalinski, Wojciech and Galaj, Tomasz and Maciejewski, Mateusz and Harihara, Rama , title =. Proceedings of the ACM on Computer Graphics and Interactive Techniques , volume =. 2025 , month = may, pages =

    work page 2025

  66. [66]

    , title =

    Thomas, Manu Mathew and Liktor, Gabor and Peters, Christoph and Kim, Sungye and Vaidyanathan, Karthik and Forbes, Angus G. , title =. Proceedings of the ACM on Computer Graphics and Interactive Techniques , volume =. 2022 , month = jul, pages =

    work page 2022

  67. [67]

    Non-reversible Metropolis-Hastings , volume =

    Bierkens, Joris , year =. Non-reversible Metropolis-Hastings , volume =. Statistics and Computing , publisher =. doi:10.1007/s11222-015-9598-x , number =

    work page doi:10.1007/s11222-015-9598-x

  68. [68]

    Andrieu, Christophe and Livingstone, Samuel , title =

    work page

  69. [69]

    Eberle, Andreas and Lörler, Francis , title =

    work page

  70. [70]

    2017 , eprint =

    Nonreversible Langevin Samplers: Splitting Schemes, Analysis and Implementation , author =. 2017 , eprint =

    work page 2017

  71. [71]

    2004 , eprint=

    Improving Asymptotic Variance of MCMC Estimators: Non-reversible Chains are Better , author=. 2004 , eprint=

    work page 2004

  72. [72]

    Accelerating reversible Markov chains , journal =

    Ting-Li Chen and Chii-Ruey Hwang , keywords =. Accelerating reversible Markov chains , journal =. 2013 , issn =. doi:https://doi.org/10.1016/j.spl.2013.05.002 , url =

    work page doi:10.1016/j.spl.2013.05.002 2013

  73. [73]

    Convergence of non-reversible Markov processes via lifting and flow Poincar

    Eberle, Andreas and Guillin, Arnaud and Hahn, Leo and Lörler, Francis and Michel, Manon , year =. Convergence of non-reversible Markov processes via lifting and flow Poincar. 2503.04238 , archivePrefix =

    work page arXiv

  74. [74]

    Stochastic Analysis and Applications , volume =

    Expansion of the global error for numerical schemes solving stochastic differential equations , author =. Stochastic Analysis and Applications , volume =

    work page

  75. [75]

    2020 , eprint =

    Convergence in total variation of the Euler-Maruyama scheme applied to diffusion processes with measurable drift coefficient and additive noise , author =. 2020 , eprint =

    work page 2020

  76. [76]

    1992 , publisher =

    Numerical Solution of Stochastic Differential Equations , author =. 1992 , publisher =

    work page 1992

  77. [77]

    Numerical approximations of stochastic differential equations with non-globally Lipschitz continuous coefficients , volume =

    Hutzenthaler, Martin and Jentzen, Arnulf , year =. Numerical approximations of stochastic differential equations with non-globally Lipschitz continuous coefficients , volume =. Memoirs of the American Mathematical Society , doi =

    work page

  78. [78]

    2004 , doi =

    An Introduction to Partial Differential Equations , author =. 2004 , doi =

    work page 2004

  79. [79]

    2011 , edition =

    PDE and Martingale Methods in Option Pricing , author =. 2011 , edition =. doi:10.1007/978-88-470-1781-8 , isbn =

    work page doi:10.1007/978-88-470-1781-8 2011

  80. [80]

    and Roberts, Gareth O

    Pollock, Murray and Johansen, Adam M. and Roberts, Gareth O. , title =. Bernoulli , number =. 2016 , doi =

    work page 2016

Showing first 80 references.