Student researchers from Alperton Community School, in collaboration with Orbyts Fellows Cathy Darling and Srinjoy Ganguly from UCL Physics and Astronomy, conducted a study to test a Generative Quantum Eigensolver (GQE) for predicting the ground state energy of molecules. This research, which extends the application of GQE to a larger molecule than previously attempted, aims to demonstrate the potential of classical generative AI models to solve complex quantum chemistry problems.

The team trained a GPT model on quantum circuits known to minimise ground-state energy, then used the model to generate predictions for the ground state energy of two molecules: the simple hydrogen ion (H3+​) and the more complex water molecule (H2​O). They chose these molecules because their true ground state energies were already known, providing a benchmark for the model's performance.

The GQE model was highly successful for the hydrogen ion, as its predicted energies converged accurately with the known ground state value. This success was reflected in the low training losses, which indicated effective learning on the part of the model. This result is particularly compelling because it demonstrates GQE's capability on a molecule with more atoms than previously tested. However, the model struggled significantly with the more complex water molecule, yielding predicted energies that did not converge and showing higher training losses.

The study successfully demonstrated that GQE can predict the ground state energy of the hydrogen ion (H3+​). This confirms the potential of GQE, but highlights the increasing difficulty in training the model as molecular complexity, such as the greater number of atoms and the more intricate overlap between orbitals, increases. Future work would involve running the model for more iterations (>10,000) or optimising other parameters to achieve better predictions for more complex molecules like H2​O!