An 'invariant compiler' uses LLMs to translate physics requirements into Neural ODE architectures that satisfy conservation laws by construction.
March 26, 2026
Original Paper
An Invariant Compiler for Neural ODEs in AI-Accelerated Scientific Simulation
arXiv · 2603.23861
The Takeaway
Instead of using 'soft' loss penalties that can still drift, this framework enforces domain invariants (like energy conservation) through the architecture itself. It provides a systematic design pattern for creating physically grounded neural surrogates for scientific simulation.
From the abstract
Neural ODEs are increasingly used as continuous-time models for scientific and sensor data, but unconstrained neural ODEs can drift and violate domain invariants (e.g., conservation laws), yielding physically implausible solutions. In turn, this can compound error in long-horizon prediction and surrogate simulation. Existing solutions typically aim to enforce invariance by soft penalties or other forms of regularization, which can reduce overall error but do not guarantee that trajectories will