i-PI 3.0: A flexible and efficient framework for advanced atomistic simulations.

Area of Science:

• Computational Physics
• Materials Science
• Quantum Chemistry

Background:

• Machine-learning interatomic potentials significantly advance atomic-scale simulations.
• These potentials bridge the accuracy of electronic structure calculations with large-scale modeling capabilities.
• The i-PI package integrates these potentials with advanced simulation techniques.

Purpose of the Study:

• To benchmark and optimize the i-PI package for enhanced performance in atomic-scale simulations.
• To introduce new features for advanced modeling and uncertainty quantification.
• To enable efficient simulation of complex quantum phenomena.

Main Methods:

• Benchmarking and optimization of the i-PI package in Python.
• Integration with popular machine-learning potentials (Behler-Parinello, DeePMD, MACE).
• Implementation of algorithms for bosonic/fermionic exchange, uncertainty quantification, and photon-nuclear dynamics.

Main Results:

• Computational overhead of i-PI is rendered negligible for large systems (tens of thousands of atoms).
• New features enable advanced simulations, including quantum exchange and coupled photon-nuclear dynamics.
• Optimized i-PI facilitates accurate and efficient atomic-scale modeling.

Conclusions:

• The optimized i-PI package significantly enhances the efficiency and scope of machine-learning-driven atomic-scale simulations.
• New functionalities expand its applicability to cutting-edge research in quantum dynamics and materials science.
• i-PI provides a robust platform for integrating advanced computational methods.