A library for solving differential equations using neural networks based on PyTorch, used by multiple research groups around the world, including at Harvard IACS.
This would be great! Would it be useful to simple time-evolution of the Schrödinger equation in 2D? With H = p²/2m +V(r), linear case.
Currently I solve this with Chebyshev expansion of the time-evolution operator and to avoid large matrices I use FFT to represent the momentum operators. But even after optimizations, it is still a costly numerical task.
Do you expect that NNs would reduce the computational cost here as well?
This would be great! Would it be useful to simple time-evolution of the Schrödinger equation in 2D? With H = p²/2m +V(r), linear case.
Currently I solve this with Chebyshev expansion of the time-evolution operator and to avoid large matrices I use FFT to represent the momentum operators. But even after optimizations, it is still a costly numerical task.
Do you expect that NNs would reduce the computational cost here as well?
Any plans to develop this IBVP2D case soon?