Closed janosh closed 1 year ago
Sorry about the noise here. I should stop opening unfinished PRs when I'm tired.
The problem was I was reconstructing pymatgen
Structures from the TrajectoryObserver
's atom positions which are Cartesian. pymatgen by default interprets coords as fractional. Simple fix.
Now the relaxation behavior is very good. CHGNet finds its way to sensible atom configurations even if you heavily scramble the structure.
This PR adds the 2nd example notebook that showcases how to visualize atom positions and energy of a structure relaxation over the course of a CHGNet relaxation in a Plotly Dash app built with Crystal Toolkit.
Here's a screen recording of what the Dash app looks like in action.
https://user-images.githubusercontent.com/30958850/226071804-c9015f22-d0dd-46a7-85d4-ae4033635723.mp4
Something seems to be going wrong with the relaxation though. @BowenD-UCB Maybe you can take a look what's happening?