Similar to #710 we can reorder the axes so that c is the long axis. This should allow us to use a smaller qr_max on the inner loop and shorten the overall computation time.
This result is empirical. I noticed that accuracy improved for the octahedron model simply be reordering the axes but still using a fixed 76 point gaussian for both the inner and outer loop.
Add a three-way sort function to the library so the code isn't repeated across lots of different models.
Similar to #710 we can reorder the axes so that c is the long axis. This should allow us to use a smaller qr_max on the inner loop and shorten the overall computation time.
This result is empirical. I noticed that accuracy improved for the octahedron model simply be reordering the axes but still using a fixed 76 point gaussian for both the inner and outer loop.
Add a three-way sort function to the library so the code isn't repeated across lots of different models.