UniMatSim: A High-Throughput Materials Simulation Automation Framework Based on Universal Machine Learning Potentials

Abstract

Universal machine learning interatomic potentials (UMLIPs) offer accuracy close to first-principles calculations at a fraction of the cost, showing significant potential for large-scale material simulations. However, the fragmented UMLIPs ecosystem lacks unified interface standards and integration frameworks, hindering their automated deployment in high-throughput workflows. To address this, we present UniMatSim, a modular Python framework. It systematically integrates various UMLIPs (e.g., CHGNet, M3GNet, MACE) and automates workflows from structural optimization to stability verification. The framework enables seamless model switching via abstracted interfaces, incorporates task orchestration, and provides standardized modules for key properties (elasticity, phonons, molecular dynamics), including automated handling for low-dimensional materials. As a test case, using the 2D Lieb lattice system, we constructed a multi-stage high-throughput screening workflow covering structural optimization, elastic stability, and phonon spectrum calculations. Starting from 1,176 candidate compositions, a four-model consensus pipeline yields 393 stable structures. These are refined by magnetic-state screening and DFT band-structure calculations to 59 Lieb-lattice candidates with staggered-magnetic-band characteristics. Results show UniMatSim significantly improves computational efficiency and reproducibility, providing a reliable infrastructure for data-driven materials discovery and design.

Figures (76)

• Figure 1: Architectural overview of the UniMatSim framework. The schematic illustrates the system's core workflow. It accepts diverse structural inputs (e.g., bulk crystals, 2D monolayers) and processes them through the UniMatSim Automated Core. This core features a Unified Potential Interface that supports various models (CHGNet, M3GNet, MACE, MatterSim and others) and a Smart Workflow Engine (DAG Scheduler) for executing tasks such as geometry optimization, elastic tensor calculation, phonon spectra analysis, and thermodynamic stability checks. The framework integrates dimensionality awareness for low-dimensional handling and outputs high-throughput screening results, including convex hull energy diagrams, phonon dispersion band structures, and mechanical modulus bar charts.
• Figure 2: High-throughput stability-screening workflow for Lieb lattices implemented in UniMatSim. The diagram illustrates the parallel screening process starting from 1176 initial candidate structures. Four independent models (Models A--D) are employed to run parallel screening pipelines, each performing structural optimization (convergence criteria $f_\text{max} \leq 0.05\ \text{eV/Å}$), elastic stability verification, and phonon spectrum calculations. The individual models identified 449, 438, 438, and 428 stable candidates, respectively. These sets enter the Multi-Judge Consensus System, where a strict intersection check confirms 393 final consistent candidates for subsequent analysis.
• Figure 3: Workflow for the identification of altermagnetic Lieb lattices. The process begins with the relaxation of 393 candidates in a ferromagnetic (FM) state. Antiferromagnetic (AFM) configurations are generated by flipping the spin signs (MAGMOM) of specific atoms. Candidates are retained only if the AFM state is energetically more favorable than the FM state ($E_{\text{AFM}} < E_{\text{FM}}$). The resulting 61 AFM-stable structures undergo band structure analysis to verify altermagnetism, yielding 59 final candidates.
• Figure 4: Geometry and electronic structure of the representative Lieb-lattice material TaCo$_2$Se$_4$. (a) Schematic of the Lieb lattice: light blue and dark blue spheres denote two distinct transition-metal sites, and green spheres denote chalcogen or pnictogen sites. (b) Atomic structure of TaCo$_2$Se$_4$ showing the Lieb-lattice framework and coordination environment. Co, Ta, and Se atoms are colored dark blue, light blue, and green, respectively. (c) DFT-calculated spin-polarized band structure, revealing the characteristic spin splitting and staggered-magnetic-band character.
• Figure S1: UpSet plot showing the overlap in phonon-stable materials identified by four independently fine-tuned MatterSim models. A total of 393 materials were predicted as stable by all four models.