VibroML: Automated Toolkit for Crystal Stability Using Machine Learning

Researchers have developed VibroML, an open-source Python toolkit that uses machine-learned interatomic potentials (MLIPs) to automatically resolve dynamical instabilities in computationally predicted crystalline materials. Unlike traditional methods that only identify instabilities, VibroML employs an energy-guided genetic algorithm to efficiently navigate the potential energy surface and discover diverse, dynamically stable polymorphs. The toolkit also includes an automated molecular dynamics workflow to evaluate finite-temperature structural retention, addressing the limitation that 0 K harmonic stability does not guarantee macroscopic viability. Additionally, VibroML integrates with ProtoCSP, a combinatorial structure prediction engine, to stabilize frustrated crystal topologies through targeted alloying. This work, published on arXiv (ID: 2604.27685), represents a shift from stability verification to automated structural remediation in materials science.