From semiconductors to superconductors: A data-driven pipeline for exploring complex energy

The rational design of advanced functional materials—from semiconductors to superconductors—demands efficient exploration of complex, high-dimensional energy landscapes, a task that remains intractable for conventional first-principles methods alone. In this work, we present a unified data-driven pipeline that integrates entropy maximization and Wyckoff-splitting structure generation, MLIPs, and ML-accelerated property screening. This workflow systematically compresses the structural search space and prioritizes physically meaningful configurations, enabling efficient exploration of materials stability and functional properties.We demonstrate the effectiveness of this framework in three representative applications. For semiconductor CdSe, a NequIP-trained machine-learning potential combined with PALLAS pathway sampling reveals detailed phase-transition pathways, including a previously unreported zinc-blende intermediate in agreement with experimental trends. For Boron-Carbon clathrate superconductors, Wyckoff-splitting enumeration generates high-symmetry structural candidates, from which high-fidelity DFT data are used to train an accurate interatomic potential, while an EOSNet-based model predicts superconducting critical temperatures. A hierarchical screening strategy combining ML phonon-stability filtering with selective first principles validation significantly reduces computational cost while maintaining high predictive accuracy. Finally, for representative systems including carbon, silicon, and NaCl under high pressure, the framework efficiently generates compact yet diverse datasets by integrating local entropy maximization, a global entropy filter, and fine-tuned machine-learning interatomic potentials.Taken together, these results establish that integrating crystallographic principles with modern machine learning can accelerate both the discovery and mechanistic understanding of complex materials by orders of magnitude relative to purely first-principles approaches. The proposed pipeline provides a generalizable, modular framework applicable to diverse materials classes, laying the groundwork for the rational design of next-generation semiconductors, superconductors, and functional materials under extreme conditions.