For modelling non-linear dynamical systems on a mesh, you can "lift" the states using some function such that you have a linear operation in a higher dimensional space. Koopman operators are the linear operators in Hilbert spaces, and recently neural networks have been applied to learn system dynamics in this space. This paper proposes a pytorch module for Koopman Neural Operators.

#MachineLearning #DynamicalSystem #NeuralOperators
https://arxiv.org/abs/2301.01104