How to improve the TEP stability by adding the onsite term
Hi,
I tried to do somethings follow your suggestions but the issue cannot be solved.
-
I tried to add some configurations at higher temperatures(for example, if I want to run the MD at 300K, I used the 500K configurations from AIMD to fit the potential), to be honest, the stability can be improved a bit(the maximum running time is increased from 100ps to 300ps), but it still unstable after 300ps. And for this method, I think there is a problem that the RMSE will increase distinctly because of the configurations are extracted by the AIMD at higher temperatures.
-
For the regularization, I tried to fit the potential by using the ridge method (increase the alpha) and lasso, both these two methods have not distinct improvements to the stability of potential in my system. So I think the effectiveness of regularization is dependent with the research system.
If it is possible, could you tell me (or an example) how to add the onsite term to the TEP? (I need a longer time MD simulation because I want to get some phonon properties from the trajs of MD.)
Best wishes,