Towards a transferable fermionic neural wavefunction for molecules

In the following, we briefly outline our approach and how it extends existing work (A multi-compound wavefunction ansatz). We show the fundamental properties of our ansatz such as extensivity and equivariance with respect to the sign of reference orbitals. We demonstrate the transferability of the ansatz when pre-training on small molecules and re-using it on larger, chemically similar molecules. We also compare its performance against GLOBE, a concurrent work²⁶. Lastly, we present a first wavefunction base model pre-trained on a large diverse dataset of 360 geometries and evaluate its…