Interpreting electron paramagnetic resonance (EPR) spectra is a recurring task in fields that use magnetic resonance to probe unpaired electrons. It usually involves running forward simulations many times while adjusting spin Hamiltonian parameters until the modeled spectrum matches the experimental one.
TorchSpin, described in a recent arXiv preprint, is a differentiable EPR simulation module built in PyTorch. Because the simulation is differentiable, it can in principle be optimized with gradient-based methods, which may reduce the number of forward passes needed during fitting.
The abstract notes that established software already provides mature simulation and fitting methods, but the sentence is cut off in the source. The paper's contribution appears to be a new implementation that leverages PyTorch's automatic differentiation, though the full details and benchmarks are not available from the abstract alone.