This repository contains code to run the inexact iterative numerical algebra schemes for spectral estimation and forecasting from https://arxiv.org/abs/2303.12534.
python=3.8|3.9jaxoptaxfor optimizersnumpyscipynumbato for numerical acceleration
See the Jax repository for more detailed installation instructions on GPUs.
The Jupyter notebook VPM_Muller_Brown.ipynb contains template code to simulate the Müller-Brown potential (as described in the paper) and solve the eigenproblem with subspace iteration and forecast problems (MFPT and committor). The neural-network training will be most efficient with a GPU.