This code implements the Control Volume Physics-informed Neural Networks (CVPINNs) method in [1]. cvpinns.py implements the PDE class. Objects of this class are constructed with mesh, quadrature rule, and PDE specifications, and provide the getRES function for computing CVPINNs residuals. euler.ipynb is an example script applying CVPINNs to recover the equation of state from the analytical solution to a Sod shock problem.
Python >= 3.5
numpy
scipy
matplotlib
toolz
tensorflow >= 2.2
SAND No: SAND2021-2386 O
[1] R. G. Patel, I. Manickam, N. A. Trask, M. A. Wood, M. Lee, I. Tomas, E. C. Cyr. Thermodynamically consistent physics-informed neural networks for hyperbolic systems. arXiv preprint arXiv:2012.05343, 2020.