StringNET is the code accompanying the paper:
StringNET: Neural Network based Variational Method for Transition Pathways
Jiayue Han, Zhiyou Wu, Shuting Gu, Xiang Zhou
https://arxiv.org/abs/2408.12621
Key Words: Neural Network, Variational Method, Maximum Flux Path (MFP), Minimum Energy Path (MEP), Minimum Action Path (MAP), High-Dimensional Applications, Ginzburg–Landau Functional
StringNET is a neural network–based method for computing transition pathways between metastable states. Our approach is built on a variational formulation over curves in the path space and can be used to compute the Maximum Flux Path (MFP), and pre-train for Minimum Energy Path (MEP), and Minimum Action Path (MAP).
Our code consist of a set of numerical examples that clearly demonstrate the method on various test cases, including:
- Double-Well and Three-Well Potentials
- Alanine Dipeptide Conformational Transitions
- High-Dimensional Muller Potential
- 2D and 4D Ginzburg–Landau Functionals
For more details, please refer to our preprint in arxiv. StringNET: Neural Network based Variational Method for Transition Pathways.