Exploring Quantum Phase Transitions with an Extended FermiNet Architecture
FermiNet is extended to detect quantum phase transitions
Artificial neural networks (ANNs), which can analyze large amounts of data quickly and accurately, have proven to be very useful in research settings. Google’s British AI division DeepMind, which is a subsidiary of Google, used a new ANN called the Fermionic Neural Network (FermiNet), in 2020 to solve a chemistry problem – the Schrodinger equation – for electrons within molecules.
The Schroedinger Equation is a partial difference equation that uses the well-established theory on energy conservation to solve problems relating to the properties and behavior of matter. DeepMind was able to solve the equation using FermiNet – a simple conceptual method – in the context chemistry. The results were very accurate and comparable with those of highly sophisticated quantum chemistry methods.
Researchers from Imperial College London (ICL), DeepMind (Lancaster University), and University of Oxford have recently adapted FermiNet to solve a quantum-physics problem. They used FermiNet in their paper published in Physical Review Letters to study the homogeneous electron gas, a simplified quantum mechanics model of electrons interfacing in solids.