Equilibrium-Informed Neural Networks Detect Critical Transitions

Swadesh Pal presents a Feb 14, 2026 arXiv preprint proposing equilibrium-informed neural networks (EINNs), a DNN approach that infers system parameters from candidate equilibrium states to locate critical thresholds. The method detects saddle-node bifurcations and multistability by mapping parameter landscapes and identifying abrupt infeasibility or discontinuities. EINNs offer a computationally efficient alternative to extensive forward simulations for early-warning analysis in high-dimensional nonlinear systems.