SpinPSO: An agent-based optimization workflow for identifying global noncollinear magnetic ground-states from first-principles

TL;DR

This paper introduces SpinPSO, a hybrid meta-heuristic algorithm combining particle swarm optimization and atomistic spin dynamics, designed to identify non-collinear magnetic ground-states from first-principles calculations.

Contribution

The paper presents a novel hybrid optimization workflow, SpinPSO, extending PSO to continuous spins and integrating it with DFT for accurate magnetic ground-state identification.

Findings

Successfully converged to experimental magnetic ground-states

Applied to diverse materials with exotic spin textures

Demonstrated effectiveness on high-performance computing platforms

Abstract

We propose and implement a novel hybrid meta-heuristic optimization algorithm for the identification of non-collinear global ground-states in magnetic systems. The inputs to this optimization scheme are directly from non-collinear density functional theory (DFT), and the workflow is implemented in the atomate code framework, making it suitable to run on high-performance computing architectures. The hybrid algorithm provides a seamless theoretical extension of particle swarm optimization (PSO) algorithms to continuous $\mathcal S^2$ spins, giving it the name SpinPSO. The hybrid nature of the algorithm stems from setting the dynamics of individual spins to be governed by physically motivated atomistic spin dynamics. Using this algorithm, we are able to achieve convergence to experimentally resolved magnetic ground-states for a set of diverse test case materials that exhibit exotic spin textures.