A new preprint on arXiv argues that neural emulation need not require learning a substitute for known physics. The authors investigate PolyNets, which couple a continuous-time neural representation of polynomial dynamics with a simple recurrent network for discrete-time.

The central idea is that the physics itself can serve as the network structure, rather than being approximated by a learned model. Because the approach is described as training-free, it may offer a faster and more interpretable route to neural emulation than conventional learned surrogates.

The abstract is truncated, so the announcement does not include full methodological details or quantitative results. As a preprint, the work has not yet undergone peer review.