Neutron transport simulations are essential for nuclear reactor design and shielding, but they come with high computational costs, particularly when modeling high-dimensional features. A new arXiv preprint proposes a machine learning method to address this bottleneck. The algorithm, named pcNN-SN, couples a physics-enhanced predictor-corrector neural network with the discrete ordinates (SN) method.

The abstract describes the work as a super-resolution machine learning approach, a class of methods that reconstruct high-resolution details from lower-resolution inputs. The authors suggest that such techniques are promising for reducing the computational expense of traditional transport simulations. However, the abstract is truncated, so the paper's specific results, benchmarks, and implementation details are not available from the source.

The name pcNN-SN indicates that the neural network is designed to work alongside the discrete ordinates method, but the abstract does not elaborate on the coupling mechanism. As an arXiv preprint, the work has not yet been formally peer-reviewed.