Simulating Quantum Field Theories with Neural Quantum States

by Prof. Di Luo (MIT)

Monday 19 Dec 2022, 14:00 → 16:00 US/Eastern

Quantum field theories are known to be fundamental while challenging to be simulated due to high dimensionality and sign problems. In this talk, I will discuss several advancements of simulating quantum field theories with neural quantum states. First, I will present the development of gauge equivariant neural network wave functions. The approach has been applied to discrete group gauge theories with abelian and non-abelian symmetries, and recently extended to 2+1D U(1) gauge theory. Next, I will talk about a gauge invariant approach with a neural flow wavefunction, Gauge-Fermion FlowNet. The new algorithm is able to represent the U(1) freedom without cutoff, obey the Gauss's law with dynamical fermions, sample efficiently without autocorrelation time and simulate theories with sign problems. We use the new method to study 2+1D quantum electrodynamics at finite density, investigating phenomena such as string breaking, charge crystal to vacuum phase transition and magnetic phase transition. If time permits, I will conclude with our new results on neural network simulation of continuum quantum field theories.