Distributed replica dynamics

Area of Science:

• Computational chemistry
• Statistical mechanics
• Materials science

Background:

• Rare-event molecular dynamics simulations are crucial for understanding complex systems.
• Existing methods like parallel replica dynamics (PRD) require synchronized computational resources.
• Distributed computing offers a flexible alternative but poses synchronization challenges.

Purpose of the Study:

• To introduce a novel distributed replica dynamics (DRD) method for rare-event molecular dynamics.
• To enable efficient simulations using asynchronous and heterogeneous distributed computing resources.
• To validate the accuracy of DRD in reproducing escape time distributions.

Main Methods:

• DRD simulates independent system replicas on distributed clients.
• Each replica runs molecular dynamics for a fixed time, reporting trajectory information to a server.
• A server-side clock tracks accumulated simulation time, initiating new calculations from observed transitions.

Main Results:

• DRD is proven to reproduce the correct probability distribution of escape times.
• Numerical simulations of Al(100) adatom diffusion using DRD and PRD yielded consistent results.
• The method accommodates asynchronous clients and dynamic addition/removal of computational resources.

Conclusions:

• DRD offers a robust and flexible approach for rare-event molecular dynamics on distributed systems.
• The method accurately captures system dynamics and escape time statistics.
• Guidelines for optimizing DRD parameters (number of replicas, trajectory length) are provided.