Magnetic Molecular Dynamics

Molecular dynamics has been a cornerstone of computational materials science, providing groundbreaking solutions to challenges such as defect dynamics, phase transitions. However, the focus of these successes has largely been on structural evolution and mechanical properties. The simulation of magnetic properties, a long-standing goal in the field, presents unique challenges. Notable contributions include Dudarev et al.'s SPILADY^[1] and Julien's Spin-Lattice dynamics^[2].<br/>In this talk, we will highlight four essential aspects crucial to advancing molecular dynamics simulations of magnetic materials. First, high-quality data^[3] is paramount; it must capture the excitation states with precision, particularly the magnetic states, encompassing both longitudinal and transverse variations with full degrees of freedom. Second, effective feature correlation^[4] is needed to accurately represent the interplay between the lattice and magnetic degrees of freedom. Additionally, efficient exploration schemes are required to navigate the complexities of spin interactions and to describe phenomena across a broad range of energy scales. Machine learning has emerged as a promising tool to address these challenges, offering new approaches to feature extraction and data-driven models.<br/>However, current energy minimization algorithms or time-evolution algorithms, which are used to find ground states and simulate magnetic phase structures at specific temperatures and pressures, typically treat spin and lattice as independent degrees of freedom, evolving separately or iteratively. This approach not only slows down computations - especially when utilizing machine-learning-based potential functions - but also introduces significant issues with energy conservation, limiting the applicability of these methods for magnetic potential energy functions.<br/>Therefore, the third and fourth key aspects of this talk will focus on the development of co-optimization^[5] and co-evolutionary algorithms that integrate both degrees of freedom, ensuring energy conservation while providing efficient structure optimization and accurate time-dependent simulations. By addressing these challenges, we can unlock the full potential of molecular dynamics for simulating magnetic materials.<br/><br/><b>References:</b><br/>[1] Ma P W, Dudarev S L, Woo C H. SPILADY: A parallel CPU and GPU code for spin–lattice magnetic molecular dynamics simulations[J]. Computer Physics Communications, 2016, 207: 350-361.<br/>[2] Tranchida J, Plimpton S J, Thibaudeau P, et al. Massively parallel symplectic algorithm for coupled magnetic spin dynamics and molecular dynamics[J]. Journal of Computational Physics, 2018, 372: 406-425.<br/>[3] Cai Z, Wang K, Xu Y, et al. A self-adaptive first-principles approach for magnetic excited states[J]. Quantum Frontiers, 2023, 2(1): 21.<br/>[4] Yang T, Cai Z, Huang Z, et al. Deep learning illuminates spin and lattice interaction in magnetic materials[J]. Physical Review B, 2024, 110(6): 064427.<br/>[5] Xu B, Huang Z, Yang T, et al. Synergistic Optimization of Lattice and Spin in Atomic Scale Simulation of Magnetic Materials[J]. 2024.