Machine learning transferable atomic forces for large systems from underconverged molecular fragments

Abstract

Machine learning potentials (MLP) enable atomistic simulations with first-principles accuracy at a small fraction of the costs of electronic structure calculations. Most modern MLPs rely on constructing the potential energy,...

Construction of High Accuracy Machine Learning Interatomic Potential for Surface/Interface of Nanomaterials—A Review

“…Employing a fragment-based dataset construction strategy rooted in chemical environment equivalence, they further undertook MLIP-based investigations on IRMOF-10 and IRMOF-16 systems. [204] More recently, several MLIP-based studies focusing on MOF-guest interactions have emerged. In 2022, Achar et al crafted a DP potential for UiO-66, applying it to probe neon and xenon diffusion within the crystal (Figure 13b).…”

Machine learning interatomic potentials for amorphous zeolitic imidazolate frameworks

“…This method was applied on different systems in three later works in different groups. 26–28 However, the molecular fragments approach suffers from a lack of universal method to choose the fragments working on all MOFs, and from the sensitivity of its predicted energies to the choice of fragments. To overcome its shortcomings, four other works relied on periodic DFT calculations on the primitive cell of the framework to train their NNP.…”

Quantum-level machine learning calculations to predict the PES of Selegiline

“…Recently, machine learning (ML), including deep learning (DL), techniques have entered the world of quantum chemical calculations, which has created a revolution in computing methods in terms of task completion time [35][36][37]. Machine learning methods have facilitated and guided a series of events and data-driven ndings in a wide range of sciences and specialized elds [38,39], and have the ability to approximate density functional theory (DFT) in a computationally e cient method which can signi cantly increase the effectiveness of computational methods in solving the problems of the atomic world and molecular systems [40,41] with quantum-level accuracy maintaining and much higher computational e ciency (∼1000×) than the ab initio [42][43][44]. Computational chemistry researchers have focused on arti cial intelligence and data management, and tools, software, and applications in this eld are rapidly developing and improving.…”