Adaptive multi-stage integration schemes for Hamiltonian Monte Carlo. (English) Zbl 07833809
Summary: Hamiltonian Monte Carlo (HMC) is a powerful tool for Bayesian statistical inference due to its potential to rapidly explore high dimensional state space, avoiding the random walk behavior typical of many Markov Chain Monte Carlo samplers. The proper choice of the integrator of the Hamiltonian dynamics is key to the efficiency of HMC. It is becoming increasingly clear that multi-stage splitting integrators are a good alternative to the Verlet method, traditionally used in HMC. Here we propose a principled way of finding optimal, problem-specific integration schemes (in terms of the best conservation of energy for harmonic forces/Gaussian targets) within the families of 2- and 3-stage splitting integrators. The method, which we call Adaptive Integration Approach for statistics, or s-AIA, uses a multivariate Gaussian model and simulation data obtained at the HMC burn-in stage to identify a system-specific dimensional stability interval and assigns the most appropriate 2-/3-stage integrator for any user-chosen simulation step size within that interval. s-AIA has been implemented in the in-house software package HaiCS without introducing computational overheads in the simulations. The efficiency of the s-AIA integrators and their impact on the HMC accuracy, sampling performance and convergence are discussed in comparison with known fixed-parameter multi-stage splitting integrators (including Verlet). Numerical experiments on well-known statistical models show that the adaptive schemes reach the best possible performance within the family of 2-, 3-stage splitting schemes.
MSC:
| 65Cxx | Probabilistic methods, stochastic differential equations |
| 65Lxx | Numerical methods for ordinary differential equations |
| 60Jxx | Markov processes |
Keywords:
Hamiltonian Monte Carlo; multi-stage integrators; adaptive integration; Bayesian inference; stability limit; velocity VerletReferences:
| [1] | Duane, S.; Kennedy, A.; Pendleton, B. J.; Roweth, D., Hybrid Monte Carlo, Phys. Lett. B, 195, 2, 216-222, (1987) |
| [2] | Neal, R. M., MCMC using Hamiltonian dynamics, (Handbook of Markov Chain Monte Carlo, vol. 2(11), (2011)), 2 · Zbl 1229.65018 |
| [3] | Sanz-Serna, J.; Calvo, M., Numerical Hamiltonian Problems, (1994), Chapman and Hall: Chapman and Hall London · Zbl 0816.65042 |
| [4] | Bou-Rabee, N.; Sanz-Serna, J. M., Geometric integrators and the Hamiltonian Monte Carlo method, Acta Numer., 27, 113-206, (2018) · Zbl 1431.65004 |
| [5] | Radivojević, T.; Akhmatskaya, E., Modified Hamiltonian Monte Carlo for Bayesian inference, Stat. Comput., 30, 2, 377-404, (2020) · Zbl 1436.62098 |
| [6] | Verlet, L., Computer “experiments” on classical fluids. I. Thermodynamical properties of Lennard-Jones molecules, Phys. Rev., 159, 98-103, (1967) |
| [7] | Swope, W. C.; Andersen, H. C.; Berens, P. H.; Wilson, K. R., A computer simulation method for the calculation of equilibrium constants for the formation of physical clusters of molecules: application to small water clusters, J. Chem. Phys., 76, 1, 637-649, (1982) |
| [8] | Blanes, S.; Casas, F.; Sanz-Serna, J. M., Numerical integrators for the hybrid Monte Carlo method, SIAM J. Sci. Comput., 36, 4, A1556-A1580, (2014) · Zbl 1312.65209 |
| [9] | Campos, C. M.; Sanz-Serna, J., Palindromic 3-stage splitting integrators, a roadmap, J. Comput. Phys., 346, 340-355, (2017) · Zbl 1378.65154 |
| [10] | Calvo, M.; Sanz-Alonso, D.; Sanz-Serna, J., HMC: reducing the number of rejections by not using leapfrog and some results on the acceptance rate, J. Comput. Phys., 437, Article 110333 pp., (2021) · Zbl 07505916 |
| [11] | Fernández-Pendás, M.; Akhmatskaya, E.; Sanz-Serna, J. M., Adaptive multi-stage integrators for optimal energy conservation in molecular simulations, J. Comput. Phys., 327, 434-449, (2016) · Zbl 1422.65447 |
| [12] | Akhmatskaya, E.; Fernández-Pendás, M.; Radivojević, T.; Sanz-Serna, J. M., Adaptive splitting integrators for enhancing sampling efficiency of modified Hamiltonian Monte Carlo methods in molecular simulation, Langmuir, 33, 42, 11530-11542, (2017), pMID: 28689416 |
| [13] | Bonilla, M. R.; García Daza, F. A.; Fernández-Pendás, M.; Carrasco, J.; Akhmatskaya, E., Multiscale modelling and simulation of advanced battery materials, (Cruz, M.; Parés, C.; Quintela, P., Progress in Industrial Mathematics: Success Stories, (2021), Springer International Publishing: Springer International Publishing Cham), 69-113 |
| [14] | Bonilla, M. R.; García Daza, F. A.; Ranque, P.; Aguesse, F.; Carrasco, J.; Akhmatskaya, E., Unveiling interfacial Li-Ion dynamics in Li7La3Zr2O12/PEO (LiTFSI) composite polymer-ceramic solid electrolytes for all-solid-state lithium batteries, ACS Appl. Mater. Interfaces, 13, 26, 30653-30667, (2021) |
| [15] | Bonilla, M. R.; Daza, F. A.G.; Cortés, H. A.; Carrasco, J.; Akhmatskaya, E., On the interfacial lithium dynamics in Li7La3Zr2O12: poly (ethylene oxide) (LiTFSI) composite polymer-ceramic solid electrolytes under strong polymer phase confinement, J. Colloid Interface Sci., 623, 870-882, (2022) |
| [16] | Escribano, B.; Lozano, A.; Radivojević, T.; Fernández-Pendás, M.; Carrasco, J.; Akhmatskaya, E., Enhancing sampling in atomistic simulations of solid-state materials for batteries: a focus on olivine NaFePO 4, Theor. Chem. Acc., 136, 1-15, (2017) |
| [17] | Radivojević, T., Enhancing Sampling in Computational Statistics Using Modified Hamiltonians, (2016), UPV/EHU: UPV/EHU Bilbao (Spain), Ph.D. thesis |
| [18] | Tuckerman, M.; Berne, B. J.; Martyna, G. J., Reversible multiple time scale molecular dynamics, J. Chem. Phys., 97, 3, 1990-2001, (1992) |
| [19] | Blanes, S.; Casas, F.; Murua, A., Splitting and composition methods in the numerical integration of differential equations, (2008) · Zbl 1242.65276 |
| [20] | McLachlan, R. I., On the numerical integration of ordinary differential equations by symmetric composition methods, SIAM J. Sci. Comput., 16, 1, 151-168, (1995) · Zbl 0821.65048 |
| [21] | Takaishi, T.; de Forcrand, P., Testing and tuning symplectic integrators for the hybrid Monte Carlo algorithm in lattice QCD, Phys. Rev. E, 73, Article 036706 pp., (2006) |
| [22] | McLachlan, R. I.; Atela, P., The accuracy of symplectic integrators, Nonlinearity, 5, 2, 541-562, (1992) · Zbl 0747.58032 |
| [23] | Radivojević, T.; Fernández-Pendás, M.; Sanz-Serna, J. M.; Akhmatskaya, E., Multi-stage splitting integrators for sampling with modified Hamiltonian Monte Carlo methods, J. Comput. Phys., 373, 900-916, (2018) · Zbl 1416.65517 |
| [24] | Mazur, A. K., Common molecular dynamics algorithms revisited: accuracy and optimal time steps of Störmer-Leapfrog integrators, J. Comput. Phys., 136, 2, 354-365, (1997) · Zbl 0886.65123 |
| [25] | Mazur, A. K., Hierarchy of fast motions in protein dynamics, J. Phys. Chem. B, 102, 2, 473-479, (1998) |
| [26] | Predescu, C.; Lippert, R. A.; Eastwood, M. P.; Ierardi, D.; Xu, H.; Jensen, M.Ø.; Bowers, K. J.; Gullingsrud, J.; Rendleman, C. A.; Dror, R. O.; Shaw, D. E., Computationally efficient molecular dynamics integrators with improved sampling accuracy, Mol. Phys., 110, 9-10, 967-983, (2012) |
| [27] | Akhmatskaya, E.; Reich, S., GSHMC: an efficient method for molecular simulation, J. Comput. Phys., 227, 10, 4934-4954, (2008) · Zbl 1148.82316 |
| [28] | Escribano, B.; Akhmatskaya, E.; Reich, S.; Azpiroz, J. M., Multiple-time-stepping generalized hybrid Monte Carlo methods, J. Comput. Phys., 280, 1-20, (2015) · Zbl 1349.82077 |
| [29] | Akhmatskaya, E.; Reich, S., New Hybrid Monte Carlo methods for efficient sampling: from physics to biology and statistics, Prog. Nucl. Sci. Technol., 2, 447-462, (2011) |
| [30] | Akhmatskaya, E.; Reich, S., Meso-GSHMC: a stochastic algorithm for meso-scale constant temperature simulations, Proc. Comput. Sci., 4, 1353-1362, (2011) |
| [31] | Nagar, L.; Fernández-Pendás, M.; Sanz-Serna, J. M.; Akhmatskaya, E., Finding the optimal integration coefficient for a palindromic multi-stage splitting integrator in HMC applications to Bayesian inference, (2023) |
| [32] | Schlick, T.; Mandziuk, M.; Skeel, R. D.; Srinivas, K., Nonlinear resonance artifacts in molecular dynamics simulations, J. Comput. Phys., 140, 1, 1-29, (1998) · Zbl 0905.65084 |
| [33] | LeCun, Y.; Simard, P.; Pearlmutter, B., Automatic learning rate maximization by on-line estimation of the Hessian’s eigenvectors, (Hanson, S.; Cowan, J.; Giles, C., Advances in Neural Information Processing Systems, vol. 5, (1992), Morgan-Kaufmann), 156-163 |
| [34] | Beskos, A.; Pillai, N.; Roberts, G.; Sanz-Serna, J. M.; Stuart, A., Optimal tuning of the hybrid Monte Carlo algorithm, Bernoulli, 19, 5A, 1501-1534, (2013) · Zbl 1287.60090 |
| [35] | Inouzhe, H.; Rodríguez-Álvarez, M. X.; Nagar, L.; Akhmatskaya, E., Dynamic SIR/SEIR-like models comprising a time-dependent transmission rate: Hamiltonian Monte Carlo approach with applications to COVID-19, (2023) |
| [36] | Hoffman, M. D.; Gelman, A., The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo, J. Mach. Learn. Res., 15, 1, 1593-1623, (2014) · Zbl 1319.60150 |
| [37] | Liu, J. S., Monte Carlo Strategies in Scientific Computing, vol. 10, (2001), Springer: Springer New York · Zbl 0991.65001 |
| [38] | Lichman, M., UCI machine learning repository, (2013) |
| [39] | Plummer, M.; Best, N.; Cowles, K.; Vines, K., CODA: convergence diagnosis and output analysis for MCMC, R News, 6, 1, 7-11, (2006) |
| [40] | Kruschke, J. K., Doing Bayesian Data Analysis, Second Edition: A Tutorial with R, JAGS, and Stan, (2015), Academic Press: Academic Press Boston |
| [41] | Vehtari, A.; Gelman, A.; Simpson, D.; Carpenter, B.; Bürkner, P.-C., Rank-normalization, folding, and localization: an improved R̂ for assessing convergence of MCMC (with discussion), Bayesian Anal., 16, 2, 667-718, (2021) · Zbl 07637221 |
| [42] | Gelman, A.; Rubin, D. B., Inference from iterative simulation using multiple sequences, Stat. Sci., 7, 4, 457-472, (1992) · Zbl 1386.65060 |
| [43] | Brooks, S. P.; Gelman, A., General methods for monitoring convergence of iterative simulations, J. Comput. Graph. Stat., 7, 4, 434-455, (1998) |
| [44] | Geyer, C. J., Practical Markov Chain Monte Carlo, Stat. Sci., 7, 4, 473-483, (1992) |
| [45] | Stan reference manual, (2023), Version 2.33 |
| [46] | Kermack, W. O.; McKendrick, A. G., A contribution to the mathematical theory of epidemics, Proc. R. Soc. Lond., 115, 772, 700-721, (1927) · JFM 53.0517.01 |
| [47] | Hindmarsh, A. C.; Serban, R.; Balos, C. J.; Gardner, D. J.; Reynolds, D. R.; Woodward, C. S., User documentation for CVODE v5. 7.0 (sundials v5. 7.0), (2021) |
This reference list is based on information provided by the publisher or from digital mathematics libraries. Its items are heuristically matched to zbMATH identifiers and may contain data conversion errors. In some cases that data have been complemented/enhanced by data from zbMATH Open. This attempts to reflect the references listed in the original paper as accurately as possible without claiming completeness or a perfect matching.
