Neural quantum states

Quantum chemistry relies on accurate descriptions of electronic states. While we have good approximation methods, getting experimental accuracy is very difficult because the number of possible electronic configurations becomes too large to treat exactly. While some methods such as density-matrix renormalisation group (DMRG) can reduce the computational cost, they struggle with the large numbers of near-degenerate states that are important for lanthanide spectroscopy. This project will explore the use of machine learning to overcome this challenge by developing neural-network representations of multiconfigurational wavefunctions. The student will use established quantum chemistry calculations as a starting point and investigate whether neural-network wavefunctions can recover the effects of dynamic electron correlation and improve predictions of lanthanide electronic spectra. Depending on the length and success of the project, the student may assess the accuracy of the approach for excited-state energies and spectroscopic properties of representative lanthanide compounds.