Neural Wavefunctions for Interacting Bosons 

We present a permutation-symmetric neural-wavefunction architecture for interacting bosons trained with variational Monte Carlo. We study the strategies to mitigate the amplitude-collapse instability of bosonic heads and compare dense and attention-based models under matched compute. As a testbed, we study bosons in a 2D disk trap, both non-interacting and interacting. Beyond ground-state energies, we report radial density profiles and pair-correlation functions and examine the condensate's rotational response to a Coriolis term in the Hamiltonian. The framework offers a flexible route to accurate bosonic neural wavefunctions in confined geometries and sets the stage for extensions to vortex physics and more complex interactions.