Jadwal Sholat

Memuat jadwal sholatโ€ฆ

Ilmu Komputer & AI editorial

Open AccessOA2024

Machine Learning of Reactive Potentials

This review examines the development and application of reactive machine learning potentials (RMLPs) for simulating chemical reactions, covering model architectures, data sampling strategies, and applications to reaction dynamics, free energy calculations, and rare events.
Yinuo Yang; Shu-Hao Zhang; Kavindri Ranasinghe; O. Isayev; A. Roitbergยท Annual review of physical chemistry (Print)ยท 2024ยท DOI 10.1146/annurev-physchem-062123-024417

The core problem

Machine learning potentials (MLPs) have driven significant developments in chemical, biological, and material sciences over the past two decades. The construction and training of MLPs enable fast and accurate simulations and analysis of thermodynamic and kinetic properties. This review focuses on the application of MLPs to reaction systems with consideration of bond breaking and formation. The authors review the development of MLP models, primarily with neural network and kernel-based algorithms, and recent applications of reactive MLPs (RMLPs) to systems at different scales. The review shows how RMLPs are constructed, how they speed up the calculation of reactive dynamics, and how they facilitate the study of reaction trajectories, reaction rates, free energy calculations, and many other calculations. Different data sampling strategies applied in building RMLPs are also discussed with a focus on how to collect structures for rare events and how to further improve their performance with active learning.

Innovation

The review covers the construction and training of reactive machine learning potentials (RMLPs). The primary algorithms discussed are neural network and kernel-based methods. The authors describe how RMLPs are built to handle bond breaking and formation, which requires training data that includes reactive configurations. Data sampling strategies are a key methodological component, particularly for capturing rare events such as transition states. The review discusses how to collect structures for rare events and how to improve RMLP performance with active learning. Active learning involves iteratively selecting new configurations to add to the training set based on model uncertainty or other criteria, thereby improving the potential's accuracy in regions relevant to the reaction. The review also covers how RMLPs are used to accelerate reactive dynamics simulations, enabling the study of reaction trajectories, reaction rates, and free energy calculations.
Introduction
Machine learning potentials (MLPs) have driven significant developments in chemical, biological, and material sciences over the past two decades. The construction and training of MLPs enable fast and accurate simulations and analysis of thermodynamic and kinetic properties. This review focuses on the application of MLPs to reaction systems with consideration of bond breaking and formation. The authors review the development of MLP models, primarily with neural network and kernel-based algorithms, and recent applications of reactive MLPs (RMLPs) to systems at different scales. The review shows how RMLPs are constructed, how they speed up the calculation of reactive dynamics, and how they facilitate the study of reaction trajectories, reaction rates, free energy calculations, and many other calculations. Different data sampling strategies applied in building RMLPs are also discussed with a focus on how to collect structures for rare events and how to further improve their performance with active learning.

Why it matters

The authors discuss the construction and application of reactive machine learning potentials (RMLPs) across different scales. They emphasize the importance of data sampling strategies, particularly for rare events, and the role of active learning in improving RMLP performance. The review highlights how RMLPs accelerate reactive dynamics calculations and enable detailed studies of reaction trajectories, rates, and free energy landscapes. The discussion also covers the development of MLP models using neural network and kernel-based algorithms, and how these models have driven significant developments in chemical, biological, and material sciences over the past two decades. The authors conclude that RMLPs provide a powerful framework for simulating and analyzing reactive systems with bond breaking and formation, and that active learning and targeted sampling are crucial for building accurate and efficient RMLPs.

Who should read this

Pembaca Research Digest dan praktisi rumpun ini.

Opening member contentโ€ฆ