Aug-cc-pvtz failed to converge as part of cbs

Hi.
We are implementing the MP2.X protocol and one of the steps

cbs_basis = 'aug-cc-pv[dt]z'
cbs(name='mp2', scf_wfn='scf', corl_wfn='mp2', corl_basis=cbs_basis, corl_scheme=corl_xtpl_helgaker_2)

) is failing to converge. Following a related post, I tried changing the guess to both default and READ but am still seeing a problem. I am looking for further guidance to help ensure convergence.
Much appreciated!
-Ben

Psi4 0.4.156 Driver

Begin Error

//>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>//
// CBS Computation: MP2 / AUG-CC-PVTZ //
//<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<//
.
.
.
@DF-RHF iter 100: -136164741186.71992492675781 2.13623e-04 1.57695e-02 DIIS

==> Post-Iterations <==

Failed to converged.
NOTE: MO Coefficients will not be saved to Checkpoint.
Iterations did not converge.
An error has occurred. Traceback:
<type ‘exceptions.RuntimeError’>:
Fatal Error: Iterations did not converge.
Error occurred in file: /process.cc on line: 202
The most recent 5 function calls were:

psi::PsiException::PsiException(std::string, char const*, int)
psi::die_if_not_converged()
psi::scf::HF::finalize_E()
psi::scf::HF::compute_energy()
psi::scf::scf(boost::shared_ptrpsi::Wavefunction, psi::Options&, _object*, _object*)
: File “”, line 50, in

File “/psi4//python/wrappers.py”, line 2435, in complete_basis_set
mc[‘f_energy’] = call_function_in_1st_argument(func, molecule=molecule, **kwargs)

End error

BEGIN input file example

molecule {
0 1
C 30.11940 -21.09390 -0.00000
H 30.67060 -20.15740 -0.00000
C 28.72490 -21.08260 0.00000
H 28.18310 -20.14070 0.00010
C 28.04530 -22.30360 0.00010
H 26.96040 -22.34880 0.00010
C 28.80450 -23.47970 0.00000
H 28.31290 -24.45090 0.00010
N 30.14530 -23.51210 -0.00010
C 30.79300 -22.32940 -0.00010
N 32.96400 -23.51780 0.00060
C 34.30200 -23.25620 0.00080
H 35.00780 -24.07930 0.00120
C 34.65530 -21.92070 0.00050
H 35.64470 -21.48280 0.00050
S 33.24870 -20.95260 -0.00010
C 32.26780 -22.39340 0.00010
}

set freeze_core true

set basis 6-31+g*
energy(‘mp3’)
e_scf = psi4.get_variable(‘SCF TOTAL ENERGY’)
ce_mp2 = psi4.get_variable(‘MP2 CORRELATION ENERGY’)
ce_mp3 = psi4.get_variable(‘MP3 CORRELATION ENERGY’)
mp3_basis = psi4.get_global_option(“BASIS”)
clean()

cbs_basis = ‘aug-cc-pv[dt]z’

cbs(name=‘mp2’, scf_wfn=‘scf’, corl_wfn=‘mp2’, corl_basis=cbs_basis, corl_scheme=corl_xtpl_helgaker_2)
#END

Edited: removed full paths

I had to modify your syntax a bit for changes made this year, but otherwise the input ran. See if the output below and attached is what you were expecting.

forum356.txt (107.0 KB)

   ==> Components <==

  ---------------------------------------------------------------------------------------------------------
                          Method / Basis                      Rqd      Energy [Eh]   Variable
  ---------------------------------------------------------------------------------------------------------
                             scf / aug-cc-pvtz                  *    -812.99870983   SCF TOTAL ENERGY
                              hf / aug-cc-pvdz                       -812.89482881   HF TOTAL ENERGY
                             mp2 / aug-cc-pvdz                  *    -814.45103060   MP2 TOTAL ENERGY
                              hf / aug-cc-pvtz                       -812.99870990   HF TOTAL ENERGY
                             mp2 / aug-cc-pvtz                  *    -814.85340570   MP2 TOTAL ENERGY
                             scf / aug-cc-pvdz                  *    -812.89482874   SCF TOTAL ENERGY
  ---------------------------------------------------------------------------------------------------------

   ==> Stages <==

  ---------------------------------------------------------------------------------------------------------
      Stage               Method / Basis                       Wt      Energy [Eh]   Scheme
  ---------------------------------------------------------------------------------------------------------
        scf                  scf / aug-cc-pvtz                  1    -812.99870983   xtpl_highest_1
       corl                  mp2 / aug-cc-pv[dt]z               1    -815.02282680   corl_xtpl_helgaker_2
       corl                  scf / aug-cc-pv[dt]z              -1    -813.04244923   corl_xtpl_helgaker_2
  ---------------------------------------------------------------------------------------------------------

   ==> CBS <==

  ---------------------------------------------------------------------------------------------------------
      Stage               Method / Basis                               Energy [Eh]   Scheme
  ---------------------------------------------------------------------------------------------------------
        scf                  scf / aug-cc-pvtz                       -812.99870983   xtpl_highest_1
       corl                  mp2 / aug-cc-pv[dt]z                      -1.98037757   corl_xtpl_helgaker_2
      total                  CBS                                     -814.97908740   
  ---------------------------------------------------------------------------------------------------------

*** Psi4 exiting successfully. Buy a developer a beer!

Thanks very much for your time in looking at this and running it. This looks like what I was expecting.

Is the conclusion that the latest version of Psi4 has some changes to help with convergence? Or is there a workaround I could implement with my currently installed version?

I’d guess your Psi4 is late 2015 or early 2016. There have been renovations of the SAD guess implementation and flow of scf guess in general in early-mid to mid 2016 by Daniel Smith. More drastically, late-mid to fall 2016 saw build, python binding, directory layout, and git history rewriting improvements, so the usual channels of updating code are not available. When you’re ready to migrate, you’ll want to do a git diff of the changes you’ve made since cloning. If you have an existing fork of psi4 on github, you’ll want to back it up locally (if there’s anything you want to preserve), delete the fork and refork it. Then reclone, build, test. Then apply by hand the changes you made to psi4. Assuming you were working in the driver, note that it’s moved from /lib/python to /psi4/driver.

I’m sorry this isn’t a couple lines to update, but we did a lot of infrastructure changes this year. And at least we did them all at once, so there’s nothing on the horizon that would induce us to reshuffle files and break history again.

Thanks very much for your clear response. Very helpful.