Closed kcpearce closed 5 years ago
Cfour would be a good choice. Not going to be accuracy-affecting approximations, easy to get a Cartesian Hessian out of. Known for higher-order derivatives. Let me know if you want a reference run.
Cool! Thank you for your continued work on this.
@loriab I'm currently having issues getting my version of CFOUR to work. If you get a chance, can you give me a reference run with the following following input/options.
Full disclosure: there could be problems with my input file. I only want the analytic hessian, but can't find the appropriate keyword to get that by itself, so I went with a vib freq calc instead. I also can't find out the highest print level allowed, but I'd like to get as much info in the output as possible.
O
H 1 R
H 1 R 2 A
R=1.1
A=104
*CFOUR(CALC=MP2,
BASIS=STO-3G,
VIB=ANALYTIC,
SCF_CONV=12,
PRINT=10)
CALC_level=mp2
vibration=exact
FD_PROJECT=off
@kcpearce This seems close, what else is needed before it is ready to go?
@dgasmith Just need to debug my t2 derivative code. Then it should be straightforward to get the hessians matching. And definitely a little cleaning up before review. Working on it now.
Awesome! I think this will make a really nice addition to the Psi4Numpy tutorials.
@kcpearce Looks like these are pretty close to ready to go?
@dgasmith Doing some further cleaning up of my code, but it's about ready to go! I'll get it pushed for review by the end of today.
@dgasmith Quick question: I'm using CFOUR's hessians to test mine, and I know we can use psi4 to interface to cfour for simple things (energies, gradients, etc.), but we can NOT do this for hessians, correct?
You can run hessians through the interface but the results aren’t parsed.
@loriab Okay, so I take it that I should just keep them hard coded in for my test at the end of the code?
Yes, hardcoding is the way to go here. Even in an analytic bs findif case, best to have a hard coded reference.
I ran pytest locally, and am failing test_SAPT0_ROHF
on the final psi4.compare_values
call. Any idea as to why?
Sorry no idea off the top of my head, how wrong is the answer?
Ind200,r: computed value (-0.000251) does not match (0.000000) to 5 digits
Ah pretty wrong... @loriab This looks like Psi4 changed the names of its variables?
Note: that test failure isn't caught by travis since it's marked as a long test and not run by travis
Can you make an issue about this then and we can go ahead and merge your PR when ready.
This is an elemental bug that I have just issues a patch for and not your fault. If this is ready to go we can review and merge.
Ready for review and merge on my end
@dgasmith Good to go on my end!
Description
Adds MP2 Hessians
What are your new additions? Please provide a brief list.
Any questions for the community?
Status