Fine tuning classical and quantum molecular dynamics using a generalized Langevin equation

Mariana Rossi, Venkat Kapil, Michele CeriottiThe Journal of Chemical Physics 2017 · DOI · arXiv

Abstract

Generalized Langevin Equation (GLE) thermostats have been used very effectively as a tool to manipulate and optimize the sampling of thermodynamic ensembles and the associated static properties. Here we show that a similar, exquisite level of control can be achieved for the dynamical properties computed from thermostatted trajectories. We develop quantitative measures of the disturbance induced by the GLE to the Hamiltonian dynamics of a harmonic oscillator, and show that these analytical results accurately predict the behavior of strongly anharmonic systems. We also show that it is possible to correct, to a significant extent, the effects of the GLE term onto the corresponding microcanonical dynamics, which puts on more solid grounds the use of non-equilibrium Langevin dynamics to approximate quantum nuclear effects and could help improve the prediction of dynamical quantities from techniques that use a Langevin term to stabilize dynamics. Finally we address the use of thermostats in the context of approximate path-integral-based models of quantum nuclear dynamics. We demonstrate that a custom-tailored GLE can alleviate some of the artifacts associated with these techniques, improving the quality of results for the modeling of vibrational dynamics of molecules, liquids, and solids.

Introduction: The Challenge of Thermostatted Dynamics

In molecular dynamics (MD) simulations, thermostats are essential for maintaining a constant temperature and ensuring proper sampling of the thermodynamic ensemble. However, thermostats act as a source of friction and random noise, which inevitably disturbs the natural "Hamiltonian" dynamics of the system. This makes it difficult to calculate dynamical properties—like vibrational spectra or diffusion coefficients—from thermostatted trajectories.

The problem is even more acute in Thermostatted Ring Polymer Molecular Dynamics (TRPMD), a method used to include nuclear quantum effects (NQEs). Standard TRPMD uses white-noise thermostats that cause "spurious broadening," blurring out high-frequency spectral features (like OH-stretch peaks) and making them look significantly wider than they actually are.

Methods: A Quantitative Framework for GLE Control

The researchers leverage the Generalized Langevin Equation (GLE), which uses a memory kernel to describe the interaction between the system and the heat bath. By mapping the non-Markovian GLE into a Markovian process using auxiliary degrees of freedom, they enable highly tunable "colored noise" thermostats.

To manage the dynamical disturbance, the authors defined three analytical measures—shift ($w_{shift}$), width ($w_{width}$), and shape ($w_{shape}$)—based on how the GLE affects a simple harmonic oscillator. They discovered that these metrics accurately predict the behavior of strongly anharmonic systems like liquid water. Furthermore, they utilized the Iterative Image Space Reconstruction Algorithm (ISRA) to perform mathematical deconvolution. If the GLE kernel is known, it is possible to take a "distorted" spectrum from a thermostatted run and reconstruct the "clean" microcanonical (NVE) spectrum.

Key Results: Recovering Unperturbed Dynamics

Spectral Reconstruction and Deconvolution

For liquid water, the authors demonstrated that they could take a spectrum distorted by a very strong thermostat and use the ISRA algorithm to reconstruct the NVE density of states with high fidelity. This proves that researchers do not have to choose between stable temperature control and accurate dynamics; by knowing the "fingerprint" of the thermostat, the damage to the spectrum can be analytically reversed.

Reconstruction of the vibrational density of states for liquid water. The GLE->NVE (green) line successfully recovers the NVE (dashed grey) spectrum from the distorted GLE (red/blue) trajectories across different friction regimes.

Resolving Quantum Nuclear Effects

In systems like isolated water molecules and the Zundel cation, custom-tailored GLE-TRPMD successfully resolved the splitting between symmetric and anti-symmetric OH-stretch peaks. These features are typically blurred into a single feature by standard white-noise TRPMD due to spurious broadening. The GLE approach allows for the recovery of these high-frequency details while still accounting for quantum nuclear effects.

Vibrational spectra of water comparing NVE, TRPMD, and the corrected GLE approach (QT->NVE). The custom GLE resolves high-frequency features that are blurred in standard TRPMD.

Limitations and Future Outlook

While the custom GLEs significantly improved high-frequency peaks, they still introduced a slight unphysical "blue shift" in low-frequency rotational and librational bands. The authors suggest this is due to the coupling between the thermostat and the curvilinear motion of the molecule's centroid. Future work will focus on designing GLEs specifically to target and mitigate these rotational artifacts while maintaining the sampling efficiency required for complex molecular systems.

Summary updated

Abstract

Generalized Langevin Equation (GLE) thermostats have been used very effectively as a tool to manipulate and optimize the sampling of thermodynamic ensembles and the associated static properties. Here we show that a similar, exquisite level of control can be achieved for the dynamical properties computed from thermostatted trajectories. We develop quantitative measures of the disturbance induced by the GLE to the Hamiltonian dynamics of a harmonic oscillator, and show that these analytical results accurately predict the behavior of strongly anharmonic systems. We also show that it is possible to correct, to a significant extent, the effects of the GLE term onto the corresponding microcanonical dynamics, which puts on more solid grounds the use of non-equilibrium Langevin dynamics to approximate quantum nuclear effects and could help improve the prediction of dynamical quantities from techniques that use a Langevin term to stabilize dynamics. Finally we address the use of thermostats in the context of approximate path-integral-based models of quantum nuclear dynamics. We demonstrate that a custom-tailored GLE can alleviate some of the artifacts associated with these techniques, improving the quality of results for the modeling of vibrational dynamics of molecules, liquids, and solids.
BibTeX
@article{rossi2017fine,
  author = {Mariana Rossi and Venkat Kapil and Michele Ceriotti},
  title = {{Fine tuning classical and quantum molecular dynamics using a generalized Langevin equation}},
  journal = {The Journal of Chemical Physics},
  year = {2017},
  doi = {10.1063/1.4990536},
}
Figure