Bohrdt, Annabelle

Combining neural quantum states with physical insight to solve many-body problems

Annabelle Bohrdt - Ludwig Maximilian Universität, Munich, Germany

Strongly correlated quantum many-body systems are often both hard to understand theoretically and challenging to study numerically. In this talk, I will show how we can use neural networks to simulate interacting quantum many-body systems, in particular by taking advantage of available physical insight. After an introduction of neural quantum states as a variational Ansatz, I will discuss three ways in which we use physical knowledge to solve the many-body problem: (i) First, I will show how we can use data from quantum simulators to provide an advantageous starting point in an ensuing numerical ground state search. (ii) Next, we will consider the fermionic t-J model and show how we can combine our neural quantum state with a Gutzwiller projected Fermi sea initialization to find extremely efficient parametrizations of the ground state across the entire doping range. (iii) Finally, I will introduce an explicitly time-dependent neural quantum state, the t-NQS, which enables us to globally solve the time-dependent Schrödinger equation and thus to study non-equilibrium dynamics in large two-dimensional systems. Apart from global quenches, I will show how the t-NQS can also be used to study parameter ramps relevant for quantum simulation experiments.