Efficient stochastic thermostatting of path integral molecular dynamics

Michele Ceriotti, Michele Parrinello, Thomas E. Markland, David E. ManolopoulosThe Journal of Chemical Physics 2010 · DOI

TL;DR

This paper introduces two novel stochastic thermostats, PILE and GLE, for Path Integral Molecular Dynamics (PIMD) simulations. These new methods offer comparable or superior sampling efficiency to the established Nosé-Hoover chain (NHC) thermostat but are computationally simpler and more efficient, making them valuable tools for future quantum simulations.

What Problem Does This Paper Address?

Simulating quantum mechanical properties of condensed phase systems, especially those involving light atoms like hydrogen, is crucial for understanding their behavior. Path Integral Molecular Dynamics (PIMD) is a powerful technique for this, but it introduces "ring polymer" replicas of the system, leading to a wide range of vibrational frequencies that are challenging to sample efficiently. Traditional methods like the Nosé-Hoover chain (NHC) thermostat, while effective, are computationally complex due to their deterministic nature and the need for auxiliary variables. This paper aims to develop simpler, more efficient stochastic thermostats that can overcome these sampling challenges.

Methods

The authors propose two new stochastic thermostats:

  • Path Integral Langevin Equation (PILE) thermostat: This method leverages analytical knowledge of the free ring polymer's normal mode frequencies to efficiently sample the system. It's a direct adaptation of the Andersen thermostat to the PIMD framework.
  • Generalized Langevin Equation (GLE) thermostat: This approach uses a colored-noise thermostat, which automatically adapts to a wide range of frequencies without requiring prior knowledge of the system's specific vibrational modes. This is achieved through a matrix-based implementation of the Langevin equation.

Both new thermostats are compared against the established Nosé-Hoover chain (NHC) thermostat, as well as a simple white-noise Langevin equation (WNLE) thermostat. The comparison focuses on sampling efficiency, measured by correlation times of various physical observables.

Datasets

The paper does not introduce new datasets. Instead, it uses two well-established benchmark systems for evaluation:

  • Liquid water: A flexible four-site water model (q-TIP4P/F) at room temperature (298 K) and standard density (0.997 gcm⁻³), with 216 molecules and 32 beads per ring polymer.
  • Hydrogen in palladium: A system with a single hydrogen atom interstitial in a palladium lattice, modeled using an embedded atom potential. This system uses a 10-bead ring polymer for hydrogen and classical treatment for the palladium atoms.

Key Insights and Ideas

  • Stochastic thermostats can rival deterministic ones: The PILE and GLE thermostats demonstrate that simpler stochastic approaches can achieve sampling efficiencies comparable to, and sometimes better than, the more complex deterministic NHC thermostat.
  • Targeting normal modes is key: Both the PILE and NHC thermostats, by specifically targeting the ring polymer's normal modes, effectively overcome the ergodicity issues inherent in PIMD.
  • Global vs. Local Thermostatting: For systems with slow, collective motions (like liquid water), global thermostats (applied to the entire system or its centroid) are significantly more efficient than local thermostats (applied to individual beads or modes). However, for systems with distinct, localized vibrations (like H in Pd), local thermostats can be more effective.
  • GLE's robustness: The GLE thermostat's ability to adapt to a wide frequency range without prior knowledge makes it a robust and versatile option, performing well across various conditions.
  • Computational efficiency: The PILE thermostat, in particular, offers a significant computational advantage due to its simpler implementation and lack of auxiliary variables compared to NHC.

Key Results

  • Water simulations: The PILE-G and NHC-G (global versions) thermostats were found to be the most efficient for sampling properties of liquid water, including kinetic and potential energy, and the squared dipole moment. The GLE also performed well, significantly outperforming the simple WNLE.
  • Hydrogen in Palladium simulations: For properties sensitive to individual H atom vibrations (radius of gyration, kinetic energy), PILE and NHC (both local and global) showed superior performance. However, for properties sensitive to slower motions (potential energy, diffusion), global thermostats were less effective, highlighting the importance of local thermostatting for such systems.
  • PILE's efficiency: The PILE thermostat was found to perform "just as well as the NHC thermostat while allowing for a computationally more efficient implementation" in nearly all cases.
  • GLE's robustness: The GLE thermostat consistently delivered near-optimum sampling efficiency across a wide range of thermostat relaxation times.

Limitations and Future Work

  • Anisotropic potentials: The paper notes that for anisotropic potentials, the performance of NHC thermostats can be sensitive to the alignment of eigenvectors, potentially degrading sampling efficiency.
  • Global thermostatting caveats: While global thermostats are beneficial for collective motions, the authors caution that they can mask poor sampling of individual degrees of freedom. Therefore, monitoring local properties is still recommended.
  • Computational cost: While stochastic methods are generally simpler, the GLE thermostat, due to its matrix operations, incurs a higher computational overhead than PILE. However, this is still often negligible compared to the cost of evaluating forces in ab initio simulations.
  • Future applications: The authors express confidence that these stochastic thermostats will find widespread use in future PIMD simulations due to their efficiency and simplicity. They suggest that further exploration of global vs. local thermostatting strategies for various systems could be beneficial.

Summary updated

Abstract

The path integral molecular dynamics (PIMD) method provides a convenient way to compute the quantum mechanical structural and thermodynamic properties of condensed phase systems at the expense of introducing an additional set of high frequency normal modes on top of the physical vibrations of the system. Efficiently sampling such a wide range of frequencies provides a considerable thermostatting challenge. Here we introduce a simple stochastic path integral Langevin equation (PILE) thermostat which exploits an analytic knowledge of the free path integral normal mode frequencies. We also apply a recently developed colored noise thermostat based on a generalized Langevin equation (GLE), which automatically achieves a similar, frequency-optimized sampling. The sampling efficiencies of these thermostats are compared with that of the more conventional Nosé-Hoover chain (NHC) thermostat for a number of physically relevant properties of the liquid water and hydrogen-in-palladium systems. In nearly every case, the new PILE thermostat is found to perform just as well as the NHC thermostat while allowing for a computationally more efficient implementation. The GLE thermostat also proves to be very robust delivering a near-optimum sampling efficiency in all of the cases considered. We suspect that these simple stochastic thermostats will therefore find useful application in many future PIMD simulations.
BibTeX
@article{ceriotti2010efficient,
  author = {Michele Ceriotti and Michele Parrinello and Thomas E. Markland and David E. Manolopoulos},
  title = {{Efficient stochastic thermostatting of path integral molecular dynamics}},
  journal = {The Journal of Chemical Physics},
  year = {2010},
  doi = {10.1063/1.3489925},
}
Figure