Example: Higher order Ti model
One way or another I think it is necessary to be able to produce a hiphive
model which reproduces the Ti meam potential energy surface very well. This is not really an issue per say but I think it is important to know that we can reproduce a higher order energy landscape.
Since we have access to any amounts of data we want for the Ti meam so the problem should be with atleast one of the below points
- Picking good training data
- Picking a good cluster space (cutoffs, orders)
- Correctness of the code in regards to 4th order and beyond
- Fitting method
Solving this issue will yield great insight for good practices when constructing higher order models.