Code for the paper Symplectic Recurrent Neural Networks, which appeared in ICLR 2020.
By unrolling a Hamiltonian system with a neural-network-parametrized Hamiltonian function using the leapfrog integrator, and together with initial state optimization, our model is able to learn the dynamics of complex, noisy and stiff physical systems.
Python 3.7
PyTorch 1.4
Matplotlib 3.1
Numpy 1.18
Scipy 1.4
Run script_chain_new for the spring-chain experiment
Run script_3body_new for the three-body experiment
This code is made available for research and replication purposes under the CC-BY-NC 4.0 license found in file LICENSE.