First-Principles AI for Quantum Matter

Strongly interacting electrons exhibit a rich variety of striking phenomena in quantum materials, from the fractional quantum Hall effect to chiral superconductivity. Yet our understanding remains limited by the difficulty of solving the many-electron Schrödinger equation in a vast Hilbert space. In this talk, a “first-principles AI” framework is introduced in which neural networks as universal and systematically improvable variational wavefunctions for many-electron quantum states. Crucially, these neural wavefunctions are optimized entirely by energy minimization, without any external training data or input physics knowledge. These architectures are provably universal approximators of fermionic wavefunctions while enforcing antisymmetry, providing a rigorous foundation for tackling the many-electron problem with AI. Within this framework, several new results are presented. First result presented is how first-principles AI discovers an electron quasicrystal phase in a single semiconductor quantum well, where interacting electrons spontaneously form a twisted-bilayer-like structure with quasiperiodic charge order. Second, will demonstrate the crystallization of a fractional quantum Hall liquid at strong Landau-level mixing, a regime that has remained inaccessible to previous methods. All results are obtained from a single unified neural architecture, which captures electron correlation through self-attention.

Speaker: Liang Fu, Massachussets Institute of Technology