onsagernet

OnsagerNet package

This package implements the basic routines for building and training (with or without model reduction/closure modelling) OnsagerNet and variants.

Main Modules

References

  1. Chen, X. et al. Constructing custom thermodynamics using deep learning. Nature Computational Science 4, 66-85 (2024).
  2. Novoselov, K. S. & Li, Q. Learning physical laws from observations of complex dynamics. Nature Computational Science 1-2 (2024).
  3. Yu, H., Tian, X., E, W. & Li, Q. OnsagerNet: Learning stable and interpretable dynamics using a generalized Onsager principle. Phys. Rev. Fluids 6, 114402 (2021).
 1"""
 2# OnsagerNet package
 3
 4This package implements the basic routines for building and training
 5(with or without model reduction/closure modelling)
 6OnsagerNet and variants.
 7
 8## Main Modules
 9- `onsagernet.dynamics`: Dynamic models, including SDEs, OnsagerNet, and variants
10- `onsagernet.transformations`: Transformations for dimensionality reduction and reconstruction
11- `onsagernet.trainers`: Training routines for SDEs and those with dimensionality transformations
12- `onsagernet.models`: Basic model definitions (mostly fully connected neural networks) that are used in references [1-3]
13
14## References
151. Chen, X. et al. Constructing custom thermodynamics using deep learning. Nature Computational Science 4, 66-85 (2024).
162. Novoselov, K. S. & Li, Q. Learning physical laws from observations of complex dynamics. Nature Computational Science 1-2 (2024).
173. Yu, H., Tian, X., E, W. & Li, Q. OnsagerNet: Learning stable and interpretable dynamics using a generalized Onsager principle. Phys. Rev. Fluids 6, 114402 (2021).
18"""