GEWUM: General Exploration Workflow for the Utopia of Materials: A Unified Platform for Automated Structure Generation, Selection, and Validation

TL;DR

GEWUM is an open-source, unified platform that automates materials discovery by integrating structure generation, machine learning potentials, and stability assessments, optimized for HPC environments.

Contribution

It introduces a modular, scalable workflow combining structure search, stability validation, and property calculations within a single platform for materials discovery.

Findings

Successfully predicted low-energy polymorphs in Al-Sc-N system.

Identified a new P-62c phase of U3Si5.

Predicted high-pressure structure of ThH10 at 150 GPa.

Abstract

The discovery of materials with tailored properties is increasingly reliant on computational methods. However, the fragmented landscape of existing software often hinders the seamless integration of large-scale structure prediction with rigorous stability validation, particularly in high-performance computing (HPC) environments. To address this gap, we present GEWUM (General Exploration Workflow for the Utopia of Materials), a unified, open-source platform designed to automate and accelerate materials discovery. GEWUM integrates the Selective Random Structure Search (SRSS) strategy with universal Machine Learning Interatomic Potentials (uMLIPs), enabling efficient exploration of vast chemical spaces. Its core architecture features a modular design with native support for SLURM-based HPC clusters. The platform unifies the entire workflow, from random structure generation and diversity-preserving selection to thermodynamic and dynamic stability assessments, as well as advanced property calculations (e.g., elastic constants, thermal conductivity, and quasi-harmonic approximations). We demonstrate GEWUM's capabilities through three distinct case studies: (1) the prediction of low-energy polymorphs in the complex Al-Sc-N nitride system; (2) the identification of a P-62c phase of U3Si5, distinct from the known AlB2 type; and (3) the high-pressure structure prediction of ThH10 at 150 GPa. Furthermore, benchmark tests show reasonable agreement in predicting thermophysical properties. By bridging the gap between uMLIPs and automated high-throughput workflows, GEWUM serves as a valuable framework to facilitate efficient and scalable materials exploration.