Correlation effects in parallel tempering and the role of the swapping frequency

Area of Science:

• Computational Physics
• Statistical Mechanics
• Molecular Dynamics

Background:

• Parallel tempering (PT), also known as replica exchange molecular dynamics (REMD), is a powerful simulation technique.
• It enhances sampling efficiency by running multiple simulations at different temperatures and exchanging configurations.
• The effectiveness of REMD relies on the frequency of these attempted exchanges.

Purpose of the Study:

• To investigate the impact of swapping frequency in parallel tempering on sampled equilibrium distributions.
• To identify potential biases introduced by high-frequency swap attempts in REMD simulations.

Main Methods:

• Utilized a simple deterministic Ising model, originally proposed by M. Creutz.
• Analyzed the sampled equilibrium distributions at various temperatures under different swapping frequencies.
• Investigated the correlation effects between configurations at close temperatures influencing swap probabilities.

Main Results:

• Demonstrated that excessively high swap frequencies can lead to systematic bias in REMD sampled distributions.
• Identified correlation effects among configurations at nearby temperatures as the cause of this simulation bias.
• Observed that this bias can affect generic Hamiltonians, including those used in atomistic simulations.

Conclusions:

• High swapping frequencies in parallel tempering (REMD) are not always beneficial and can introduce significant bias.
• Careful consideration of swap frequency is crucial to avoid erroneous results in REMD simulations.
• The findings provide a cautionary note for researchers using REMD, especially in complex systems like atomistic simulations.