Penalized Nonreversible Langevin for Constrained Sampling

Abstract

We propose penalized nonreversible Langevin algorithms for sampling from π(x)∝ e^−f(x)1_𝒞(x), where 𝒞⊂ℝ^d is a compact convex set. The algorithms combine a squared distance penalty with constant or compatible state dependent skew symmetric perturbations that preserve the penalized Gibbs distribution. For smooth, possibly nonconvex f, we derive nonasymptotic total variation bounds for the full gradient algorithm under a log Sobolev inequality. When unbiased stochastic gradients are available, we establish $2$-Wasserstein bounds under global contraction and Lipschitz conditions on the full drift in an adapted quadratic metric. For a fixed penalty parameter, the error relative to the penalized Gibbs distribution decays exponentially to an 𝒪(√η) neighborhood, where η is the stepsize. We also bound the discrepancy between the penalized Gibbs distribution and the constrained target. In a two dimensional quadratic model, we establish nonreversible acceleration by tuning the skew perturbation to the curvature imbalance induced by penalization. With the target accuracy and smaller curvature fixed and initial Wasserstein distances uniformly bounded, tuning the skew perturbation improves the sufficient Euler iteration bound from linear to logarithmic in the curvature ratio. Numerical experiments evaluate the algorithms on constrained Bayesian regression, classification, neural networks, and truncated sampling, and examine the acceleration mechanism in a stochastic quadratic model.

Two-Loop Stochastic Mirror Langevin Algorithms for Constrained Sampling
Ruiting Tong, Antik Chakraborty, Raghu Pasupathy
arXiv7 Oct 2026PreprintCC BY 4.0Full textStatisticsarXiv
Nonlinear Assimilation via Score-Based Sequential Langevin Sampling
Zhao Ding, Chenguang Duan, Yuling Jiao +3
IEEE Transactions on Information Theory21 Jul 2026Peer-reviewedLicence not statedFull text14 citedStatisticsOpenAlex
Neural Actor--Critic Learning with Minimax Critics and Reflected Langevin Sampling
Kunqi Huang, Lican Kang, Hongrui Shu +1
arXiv2 Nov 2026PreprintarXiv licenceFull textStatisticsarXiv
Wasserstein Bounds for Euler and Randomized Midpoint Langevin Sampling with Noisy Gradients
Yifei Xiong, Vinayak Rao
arXiv27 Oct 2026PreprintCC BY 4.0Full textNumerical analysis & scientific computing StatisticsarXiv
A CONSTRAINED WASSERSTEIN GRADIENT FLOW FOR MEAN-FIELD LANGEVIN DYNAMICS
Mohamed Alfaki Aboubacrine Assadek
HAL2 Oct 2026PreprintCC BY 4.0Full textAnalysis & PDEs Mathematical physicsHALarXiv

Cite

BibTeX RIS

BibTeX

@misc{ali2026penalized,
  title = {{Penalized Nonreversible Langevin for Constrained Sampling}},
  author = {Pervez Ali and Weihao Dong and Xiaoyu Wang},
  howpublished = {arXiv},
  year = {2026},
  month = oct,
  doi = {10.48550/arxiv.2609.25381},
  eprint = {2609.25381},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2609.25381},
}

RIS

TY  - UNPB
TI  - Penalized Nonreversible Langevin for Constrained Sampling
AU  - Pervez Ali
AU  - Weihao Dong
AU  - Xiaoyu Wang
PY  - 2026
DA  - 2026/10/02
PB  - arXiv
DO  - 10.48550/arxiv.2609.25381
UR  - https://arxiv.org/abs/2609.25381
AB  - We propose penalized nonreversible Langevin algorithms for sampling from π(x)∝ e^−f(x)1_𝒞(x), where 𝒞⊂ℝ^d is a compact convex set. The algorithms combine a squared distance penalty with constant or compatible state dependent skew symmetric perturbations that preserve the penalized Gibbs distribution. For smooth, possibly nonconvex f, we derive nonasymptotic total variation bounds for the full gradient algorithm under a log Sobolev inequality. When unbiased stochastic gradients are available, we establish $2$-Wasserstein bounds under global contraction and Lipschitz conditions on the full drift in an adapted quadratic metric. For a fixed penalty parameter, the error relative to the penalized Gibbs distribution decays exponentially to an 𝒪(√η) neighborhood, where η is the stepsize. We also bound the discrepancy between the penalized Gibbs distribution and the constrained target. In a two dimensional quadratic model, we establish nonreversible acceleration by tuning the skew perturbation to the curvature imbalance induced by penalization. With the target accuracy and smaller curvature fixed and initial Wasserstein distances uniformly bounded, tuning the skew perturbation improves the sufficient Euler iteration bound from linear to logarithmic in the curvature ratio. Numerical experiments evaluate the algorithms on constrained Bayesian regression, classification, neural networks, and truncated sampling, and examine the acceleration mechanism in a stochastic quadratic model.
ER  -

Abstract

Related

Cite