SAPT0 on Ca, Mg complexes


#1

Hi,

I have been running some SAPT calculations on Ca imidazole complexes following Lao and Herbert’s paper JCTC 2015, 11, 2473, where they recommend to use def2TZVPP basis-sets for Ca and aug-cc-pvtz for C,H,O, N atoms. The HF calculations run properly but then it dies before running the SAPT0 calculations and find “SIGSEGV, segmentation fault” error in the error output file. On replacing Ca with Cl (and the corresponding charges) the calculations run without any errors. So, it has to do with the basis -set for Ca.

Here’s the input :

memory 40 Gb
molecule caimdz {
       0 1
C -2.131305   0.765215   -0.000153
C -0.803001   1.098737   0.000360
C -0.919454   -1.089722   0.000521
N -2.166820   -0.604754   -0.000461
H -3.005298   -1.172255   -0.000958
H -3.018593   1.371713   -0.000275
H -0.376669   2.087368   0.000629
H -0.689326   -2.140581   0.001022
N -0.035537   -0.067631   0.000218
 --
 2 1
Ca 2.281447   -0.004247   -0.000154
     units angstrom
     no_reorient
     symmetry c1
}

set globals {
   scf_type      df
   FREEZE_CORE true
}

basis {
   assign aug-cc-pVTZ
   assign Ca def2-TZVPP
}

df_basis_scf {
   assign aug-cc-pVTZ-jkfit
   assign Ca def2-TZVPP-jkfit
}

df_basis_sapt {
   assign aug-cc-pVTZ-RI
   assign Ca def2-TZVPP-ri
}

energy('sapt0')

Any help would be appreciated.
Thanks,
Arka


#2

Hi @arka,

Your error looks really quite similar to the one described in this post which was caused by a memory leak due to an integer overflow. This bug was fixed yesterday in this PR to the master branch of Psi4. Are you running a compiled Psi4, or one installed from a binary (either through conda or the installer)? If you’re running a compiled Psi4, try updating your clone to the current master and re-build.

If that doesn’t fix it, could you post the resulting error message & traceback (surrounded by triple backticks (```; this way, the plain text of the output file doesn’t get mangled by the forum’s auto formatting)?

Cheers & good luck,

~ Dom


#3

@dasirianni
Unless you’ve actually run the input files on an up-to-date version of Psi, I think it’s premature to say yesterday’s PR fixed the error described on that previous post. (The GH issue for the other bug is here. If it’s fixed now, fantastic!)

The issue in this topic isn’t about memory leaks. See discussion here. Investigation by myself and @tetrahydrofuran puts the blame at Psi not handling the frozen core correctly for Ca 2+, since Ca normally has two non-frozen electrons.


#4

FWIW, we have a PR that passes the tests and “fixes” this issue, here, but as discussed in the comments, I’d be in favour of a more thorough fix which is on my todo list for spring 2019.

In the meantime, the easiest way around the issue is to disable frozen core.