EZGA: An evolutionary structure exploration framework
Description
Evolutionary algorithms provide a powerful route to exploring the structural complexity of molecules and materials, enabling efficient exploration of the system's energy landscape to identify the minima related to stable and metastable configurations. Yet, existing frameworks often trade flexibility for chemical fidelity or scalability, leading to unphysical structures or overly restricted search spaces. The here presented framework {is designed to address several of these limitations} through a modular evolutionary architecture that preserves chemical consistency while enabling scalable exploration. combines an interchangeable evolutionary pipeline with chemically grounded molecular and crystalline encodings, {ensuring physically meaningful sampling across different atomistic problem classes}. A hierarchical Supercell Escalation (HiSE) protocol propagates low-energy motifs from minimal cells to larger supercells, improving sampling efficiency in extended systems. The execution model integrates synchronous parallelism for physical evaluations with an asynchronous multi-agent island strategy, while a hybrid SQL–HDF5 archive ensures efficient, reproducible, and fault-tolerant data management. Across benchmarks spanning peptide conformations, Lennard–Jones nanoclusters, binary-oxide convex-hull reconstruction, and CuO/Cu2O surface phase diagrams, {recovers the targeted low-energy basins and thermodynamic envelopes across the tested benchmarks}, resolves complex phase boundaries, and d{shows low orchestration overhead and efficient strong-scaling behavior for the tested benchmark configuration}. High-level YAML workflows enable autonomous discovery campaigns {with reduced manual intervention in the tested workflows, positioning as a flexible framework for scalable, chemically consistent atomistic exploration.}