Abstract
This paper introduces Physics-Informed Deep Equilibrium Models (PIDEQs) forsolving initial value problems (IVPs) of ordinary differential equations(ODEs). Leveraging recent advancements in deep equilibrium models (DEQs) andphysics-informed neural networks (PINNs), PIDEQs combine the implicit outputrepresentation of DEQs with physics-informed training techniques. We validatePIDEQs using the Van der Pol oscillator as a benchmark problem, demonstratingtheir efficiency and effectiveness in solving IVPs. Our analysis includes keyhyperparameter considerations for optimizing PIDEQ performance. By bridgingdeep learning and physics-based modeling, this work advances computationaltechniques for solving IVPs, with implications for scientific computing andengineering applications.