Implementation of a methodology for crystal structure prediction using genetic algorithms integrated into the Python ASE library

This work is dedicated to the development and implementation of a methodology for crystal structure prediction using genetic algorithms integrated into the Python ASE library. Crystal structure prediction plays a critical role in materials science, chemistry, and nanotechnology, enabling the discovery of novel compounds with tailored properties. By combining the flexibility of ASE with the speed of classical relaxers and the accuracy of DFTbased methods, our approach significantly reduces computational costs while maintaining predictive reliability. The methodology was validated on polymorphs of silica (SiO₂), where our system successfully recovered both global and local minima of the energy landscape. We also explore the integration of neural network relaxers such as MACE and AIMNet2 to further accelerate the search process. This study lays the groundwork for efficient, scalable, and accurate predictive modeling of crystalline materials.