Abstract
Quantum materials, characterized by novel quantum effects, serve as a critical physical foundation for disruptive next-generation technologies such as information technology, quantum computing, and clean energy. However, their complex physical mechanisms pose significant challenges to traditional research methods, rendering material discovery and design time-consuming and costly. To address these challenges, we have developed a suite of AI-empowered methods for quantum materials research. We independently developed the Property Analysis and Simulation Package for materials (PASP), which integrates diverse techniques including the four-state method, effective Hamiltonian methods, Monte Carlo simulations, molecular dynamics (including spin-lattice dynamics), SpinGNN (a neural potential function for magnetic systems), and DREAM (a neural network potential for dielectric systems). Notably, we pioneered a universal foundation model (Uni-HamGNN) for predicting material electronic structures. This model incorporates E(3) equivariant symmetry and was trained on massive first-principles data, achieving the first accurate prediction of electronic Hamiltonians for complex systems across the entire periodic table. Recently, through effective Hamiltonian decomposition, this universal model has also achieved precise modeling of spin-orbit coupling effects. Building on this foundation, we established the world's first large-scale, AI-based electronic structure database and open online prediction platform (sci-ai.cn), paving the way for the rapid discovery of new materials.
Leveraging these theoretical methods and software tools, we have made significant advances in frontier quantum materials research. In two-dimensional magnetic materials, we revealed the existence of novel strong Kitaev interactions and high-order magnetic interactions. In ferroelectrics, we proposed the concept of fractional quantum ferroelectricity (FQFE), extending ferroelectricity to non-polar crystals—a regime previously considered impossible—thereby breaking through traditional ferroelectric theory. Regarding multiferroics, we established a unified model for spin-order-induced ferroelectricity, clarifying long-standing controversies regarding its physical origin. Looking ahead, AI methods that deeply integrate data-driven approaches with physical principles are poised to become the core engine of quantum materials research, dramatically accelerating the discovery of new materials.
Anyone interested is welcome to attend.