Message-passing neural quantum states for the homogeneous electron gas
Gabriel Pescia, Jannes Nys, Jane Kim, Alessandro Lovato, Giuseppe Carleo
Abstract
The immense parameter complexity of current neural-network quantum states (NQS) in continuous space renders their application to realistic physical systems impossible. Here, the authors rigorously motivate a many-body coordinate transformation as basis for a variational neural-network Ansatz, implement it via message-passing neural networks, and show a resulting reduction of two orders of magnitude in parameter complexity and convergence speed. With this, the authors open the door for future NQS studies of extended systems in the thermodynamic limit.
Topics & Concepts
HomogeneousFermi gasQuantumElectronComputer sciencePhysicsQuantum mechanicsStatistical physicsQuantum, superfluid, helium dynamicsSpectroscopy and Quantum Chemical StudiesAdvanced Chemical Physics Studies