SuperVortexNet: Reconstructing Superfluid Vortex Filaments Using Deep Learning

Abstract

We introduce a novel approach to the three-dimensional reconstruction of superfluid vortex filaments using deep convolutional neural networks. Superfluid vortices, quantum mechanical phenomena of immense scientific interest, are challenging to image due to their small dimensions and intricate topology. Here, we propose a deep-learning methodology that serves as a proof-of-principle for fully reconstructing the topology of superfluid vortex filaments. We have trained a convolutional neural network on a large dataset of simulated superfluid density images obtained by solving the Gross--Pitaevskii equation at scale, enabling it to learn the complex patterns and features inherent to superfluid vortex filaments. The network ingests the integrated density along the axial, coronal, and sagittal directions and outputs the reconstructed superfluid vortex filaments in three dimensions. We demonstrate the success of this approach over a range of vortex densities of simulated isotropic quantum turbulence, enabling access to the characteristic scaling law of the decaying vortex line length.