Mixing Time Guarantees for Unadjusted Hamiltonian Monte Carlo

Abstract: We provide quantitative upper bounds on the total variation mixing time of the Markov chain corresponding to the unadjusted Hamiltonian Monte Carlo (uHMC) algorithm. For two general classes of models and fixed time discretization step size h, the mixing time is shown to depend only logarithmically on the dimension. Moreover, we provide quantitative upper bounds on the total variation distance between the invariant measure of the uHMC chain and the true target measure. As a consequence, we show that an ε-accurate approximation of the target distribution μ in total variation distance can be achieved by uHMC for a broad class of models with O (d3/4ε−1/2 log(d/ε)) gradient evaluations, and for mean field models with weak interactions with O (d1/2ε−1/2 log(d/ε)) gradient evaluations. The proofs are based on the construction of successful couplings for uHMC that realize the upper bounds.

[1]  L. Rogers,et al.  Coupling of Multidimensional Diffusions by Reflection , 1986 .

[2]  Shmuel Friedland,et al.  Nuclear norm of higher-order tensors , 2014, Math. Comput..

[3]  Santosh S. Vempala,et al.  Optimal Convergence Rate of Hamiltonian Monte Carlo for Strongly Logconcave Distributions , 2019, APPROX-RANDOM.

[4]  A. Stuart,et al.  Spectral gaps for a Metropolis–Hastings algorithm in infinite dimensions , 2011, 1112.1392.

[5]  J. Marsden,et al.  Discrete mechanics and variational integrators , 2001, Acta Numerica.

[6]  Richard L. Tweedie,et al.  Markov Chains and Stochastic Stability , 1993, Communications and Control Engineering Series.

[7]  On explicit $L^2$-convergence rate estimate for piecewise deterministic Markov processes , 2020, 2007.14927.

[8]  Pierre E. Jacob,et al.  Estimating Convergence of Markov chains with L-Lag Couplings , 2019, NeurIPS.

[9]  Mateusz B. Majka,et al.  Quantitative contraction rates for Markov chains on general state spaces , 2018, Electronic Journal of Probability.

[10]  S. Armstrong,et al.  Variational methods for the kinetic Fokker-Planck equation , 2019, 1902.04037.

[11]  S. Rump.  Estimates of the determinant of a perturbed identity matrix , 2018, Linear Algebra and its Applications.

[12]  Alain Durmus,et al.  High-dimensional Bayesian inference via the unadjusted Langevin algorithm , 2016, Bernoulli.

[13]  Oren Mangoubi,et al.  Rapid Mixing of Hamiltonian Monte Carlo on Strongly Log-Concave Distributions , 2017, 1708.07114.

[14]  Radford M. Neal.  MCMC Using Hamiltonian Dynamics , 2011, 1206.1901.

[15]  A. Eberle.  Couplings, distances and contractivity for diffusion processes revisited , 2013 .

[16]  Andrew M. Stuart,et al.  Convergence of Numerical Time-Averaging and Stationary Measures via Poisson Equations , 2009, SIAM J. Numer. Anal..

[17]  Jonathan C. Mattingly,et al.  Asymptotic coupling and a general form of Harris’ theorem with applications to stochastic delay equations , 2009, 0902.4495.

[18]  Michael Ortiz,et al.  On the Γ-Convergence of Discrete Dynamics and Variational Integrators , 2004, J. Nonlinear Sci..

[19]  A. Eberle,et al.  Couplings and quantitative contraction rates for Langevin dynamics , 2017, The Annals of Probability.

[20]  Nawaf Bou-Rabee,et al.  Two-scale coupling for preconditioned Hamiltonian Monte Carlo in infinite dimensions , 2019 .

[21]  M. Kac.  Foundations of Kinetic Theory , 1956 .

[22]  Jonathan C. Mattingly,et al.  Ergodicity of the 2D Navier-Stokes equations with degenerate stochastic forcing , 2004, math/0406087.

[23]  C. Mouhot,et al.  HYPOCOERCIVITY FOR LINEAR KINETIC EQUATIONS CONSERVING MASS , 2010, 1005.1495.

[24]  Ravi Montenegro,et al.  Mathematical Aspects of Mixing Times in Markov Chains , 2006, Found. Trends Theor. Comput. Sci..

[25]  J. Heng,et al.  Unbiased Hamiltonian Monte Carlo with couplings , 2017, Biometrika.

[26]  A. Eberle,et al.  Coupling and convergence for Hamiltonian Monte Carlo , 2018, The Annals of Applied Probability.

[27]  Neal Madras,et al.  Quantitative bounds for Markov chain convergence: Wasserstein and total variation distances , 2010, 1102.5245.

[28]  S. Mischler,et al.  Kac’s program in kinetic theory , 2011, Inventiones mathematicae.

[29]  Martin J. Wainwright,et al.  Fast mixing of Metropolized Hamiltonian Monte Carlo: Benefits of multi-step gradients , 2019, J. Mach. Learn. Res..

[30]  E. Hairer,et al.  Simulating Hamiltonian dynamics , 2006, Math. Comput..

[31]  E. Hairer,et al.  Geometric Numerical Integration , 2022, Oberwolfach Reports.

[32]  M. Ledoux,et al.  Analysis and Geometry of Markov Diffusion Operators , 2013 .

[33]  Jesús María Sanz-Serna,et al.  Geometric integrators and the Hamiltonian Monte Carlo method , 2017, Acta Numerica.

[34]  Pierre Monmarch'e,et al.  High-dimensional MCMC with a standard splitting scheme for the underdamped Langevin diffusion. , 2020, Electronic Journal of Statistics.

[35]  S. Duane,et al.  Hybrid Monte Carlo , 1987 .