Explore "approximate" gradients

[WardLT]

We cannot get gradients with MP2 in NWChemEx (yet!) and they are additional expense anyway.

What happens if we train without gradients or using numerical approximations of 1 or more directions?

[WardLT]

Update from SPEC (excited states):

• CC has 95% complete gradients (missing just one term). Are they "close enough" to be useful?
  • They have it for "singles and doubles," which is generally accepted to be better than MP2
  • Not yet sure what the timeline for it is (check in with them in mid/late-Jan about it)