How to remove the spurious resonances from ring polymer molecular dynamics

Mariana Rossi, Michele Ceriotti, David E. ManolopoulosThe Journal of Chemical Physics 2014 · DOI · arXiv

Abstract: Two of the most successful methods that are presently available for simulating the quantum dynamics of condensed phase systems are centroid molecular dynamics (CMD) and ring polymer molecular dynamics (RPMD). Despite their conceptual differences, practical implementations of these methods differ in just two respects: the choice of the Parrinello-Rahman mass matrix and whether or not a thermostat is applied to the internal modes of the ring polymer during the dynamics. Here, we explore a method which is halfway between the two approximations: we keep the path integral bead masses equal to the physical particle masses but attach a Langevin thermostat to the internal modes of the ring polymer during the dynamics. We justify this by showing analytically that the inclusion of an internal mode thermostat does not affect any of the established features of RPMD. In particular, the resulting method is still optimum in the short-time limit, and the transition state approximation to its reaction rate theory remains closely related to the semiclassical instanton approximation. We find that a broad range of frictions give similar results, and that these results are immune to both the resonance problem of RPMD and the curvature problem of CMD.

Introduction

Simulating the quantum dynamics of atoms in liquids or molecules is notoriously difficult. Two popular approximations—Ring Polymer Molecular Dynamics (RPMD) and Centroid Molecular Dynamics (CMD)—use the "path integral" formalism, where a single quantum particle is mapped onto a "necklace" of classical beads (a ring polymer). While these methods have revolutionized our ability to study quantum effects in the condensed phase, both suffer from well-known dynamical artifacts that limit their accuracy in vibrational spectroscopy.

The "Ghosts in the Machine"

Standard path integral methods are plagued by two primary issues:

  1. RPMD's Resonance Problem: The internal vibrations of the ring polymer beads can accidentally match the frequency of real physical vibrations (like an O-H bond stretch). This causes spurious, non-physical splits and peaks in the calculated spectra.
  2. CMD's Curvature Problem: To avoid resonances, CMD averages over the ring polymer's shape. At low temperatures, the polymer "spreads out" around corners in the potential energy surface, leading to artificial broadening and "red-shifting" (shifting to lower frequencies) of spectral peaks.
Figure 1: Comparison of vibrational spectra for an anharmonic OH model. Panels (a) and (c) show the failure of CMD (red-shift) and RPMD (spurious peaks) at low temperatures, while (b) and (d) show TRPMD's consistent performance.

Thermostatted Ring Polymer Molecular Dynamics (TRPMD)

The authors introduce TRPMD as a "middle way." The core idea is to keep the bead masses equal to the physical masses (the RPMD approach) but attach a Langevin thermostat specifically to the internal modes of the ring polymer. This modification ensures the "centroid" (the average position, which carries the main dynamical information) moves undisturbed by the friction, while the noisy internal modes are dampened.

Analytically, the authors prove that TRPMD preserves the same short-time accuracy as standard RPMD. By utilizing the PILE (Path Integral Langevin Equation) thermostat at near-critical damping, the method effectively "washes out" the coherent internal mode vibrations that cause resonances without over-dampening the physical signal.

Quantitative Performance and Results

The method was evaluated on several benchmark systems, including model OH molecules, the Zundel cation ($H_5O_2^+$), and bulk liquid water. In every case, TRPMD demonstrated superior spectral purity.

One of the most striking improvements is seen in the temperature dependence of the peak position. As shown in the quantitative analysis below, CMD's predicted frequency drops sharply as temperature decreases—a purely artificial effect of the curvature problem. TRPMD, by contrast, maintains a stable peak position that aligns with physical expectations.

Figure 4: Temperature dependence of the FWHM (top) and peak wavenumber (bottom) for the OH stretch. CMD (yellow squares) shows a dramatic unphysical red-shift at low temperatures, whereas TRPMD (red circles) remains stable.

In complex systems like liquid water, TRPMD successfully removed the resonance "shoulders" seen in RPMD. It produced a peak that was slightly blue-shifted compared to CMD, which is likely more physically accurate given CMD's known curvature issues. Furthermore, the authors confirmed that thermostatting the internal modes does not negatively impact the calculation of bulk transport properties like diffusion coefficients.

Conclusion

TRPMD solves a long-standing trade-off in quantum simulations. It is easier to implement and computationally cheaper than CMD, yet it avoids the spurious resonances of RPMD. By simply dephasing the internal modes of the ring polymer, TRPMD provides a robust new standard for calculating vibrational spectra in condensed phase systems.

Summary updated

Abstract

Two of the most successful methods that are presently available for simulating the quantum dynamics of condensed phase systems are centroid molecular dynamics (CMD) and ring polymer molecular dynamics (RPMD). Despite their conceptual differences, practical implementations of these methods differ in just two respects: the choice of the Parrinello-Rahman mass matrix and whether or not a thermostat is applied to the internal modes of the ring polymer during the dynamics. Here, we explore a method which is halfway between the two approximations: we keep the path integral bead masses equal to the physical particle masses but attach a Langevin thermostat to the internal modes of the ring polymer during the dynamics. We justify this by showing analytically that the inclusion of an internal mode thermostat does not affect any of the established features of RPMD: thermostatted RPMD is equally valid with respect to everything that has actually been proven about the method as RPMD itself. In particular, because of the choice of bead masses, the resulting method is still optimum in the short-time limit, and the transition state approximation to its reaction rate theory remains closely related to the semiclassical instanton approximation in the deep quantum tunneling regime. In effect, there is a continuous family of methods with these properties, parameterised by the strength of the Langevin friction. Here, we explore numerically how the approximation to quantum dynamics depends on this friction, with a particular emphasis on vibrational spectroscopy. We find that a broad range of frictions approaching optimal damping give similar results, and that these results are immune to both the resonance problem of RPMD and the curvature problem of CMD.
BibTeX
@article{rossi2014how,
  author = {Mariana Rossi and Michele Ceriotti and David E. Manolopoulos},
  title = {{How to remove the spurious resonances from ring polymer molecular dynamics}},
  journal = {The Journal of Chemical Physics},
  year = {2014},
  doi = {10.1063/1.4883861},
}
Figure