Arkavo Hait, Santanu Mahapatra · Journal of Open Source Software, 10(109), 7589 · 2025

Atomistic spin texture simulations are crucial for understanding magnetic materials at the nanoscale, but efficient parallelisation of Metropolis Monte Carlo is challenging because the Markov property makes evolving states interdependent. Prior parallel attempts are restricted to the simpler Ising model and do not transfer to 2D materials, which have finite magnetocrystalline anisotropy, complex crystal structures, and long-range interactions — a cost that scales as N², becoming prohibitive beyond N = 100. We present CUDA-METRO, a GPU-based open-source code built on a new algorithm that simultaneously updates multiple spins irrespective of their mutual correlation, harnessing the SIMD potential of the GPU to reach the ground state of ultra-large supercells. We evaluate it by simulating complex topological spin textures and temperature-dependent magnetic phase transitions across diverse 2D crystal structures with long-range magnetic interactions, reaching the ground state of a 750×750 supercell from a random initial configuration in 9 hours on an A100-SXM4 GPU.

Arkavo Hait, Santanu Mahapatra · IEEE International Conference on High Performance Computing, Data and Analytics Workshop (HiPCW) · 2025

Traditional Markov Chain Metropolis Monte Carlo (MCMC) simulations face severe scalability bottlenecks when applied to complex, large-scale magnetic systems. To overcome this, we introduce an “unbound” parallelization algorithm for the Heisenberg model, explicitly designed for modern GPU architectures using CUDA. By introducing a tunable Parallelization Ratio (P), our method decouples state evaluations, allowing the simulation to bypass the critical slowing down inherent to serial MCMC. This approach not only delivers massive computational acceleration but also enables the system to escape deep local energy minima that trap standard algorithms. We demonstrate the efficacy of this method by simulating micron-scale megastructures and complex spin textures, such as skyrmions, paving the way for rapid, fully in-silico material discovery.

Accepted manuscript. © 2025 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works.