Hello. I’m experiencing problems when trying to use basis sets with large maximum angular momentum.
The output is the following:
ERROR: ERI - libint cannot handle angular momentum this high (7).
In a fresh object directory, reconfigure libint for higher angular momentum, then recompile.
LibInt was compiled for angular momentum of: 7
An error has occurred python-side. Traceback (most recent call last):
File “”, line 51, in
File “/home/hernan/miniconda2/share/psi4/python/driver.py”, line 442, in energy
wfn = procedures[‘energy’][lowername](lowername, molecule=molecule, **kwargs)
File “/home/hernan/miniconda2/share/psi4/python/procedures/proc.py”, line 124, in select_mp2
return func(name, **kwargs)
File “/home/hernan/miniconda2/share/psi4/python/procedures/proc.py”, line 2610, in run_dfmp2
RuntimeError: value for ERI - libint cannot handle angular momentum this high.
In a fresh object directory, reconfigure libint for higher angular momentum, then recompile. exceeded.
allowed: 7 actual: 7
I am using a fresh binary install (1.0rc3) following the instructions given in:
How can I solve it?
Also, I noticed that there is not Dunning’s 6Z JKFIT basis sets. How does Psi4 handle this? Are results reliables?
Thank you very much for your time.
When running the setup script you can add
--max-am-eri n where n is the highest angular momenta you wish compiled.
It should be noted that most DF basis sets are 1-2 angular momenta higher than their orbital counterparts. As 7 is quite high already, it might be better if you describe what you are doing.
Thank you for reply. If I understand correctly, I must discard the binary install and compile Psi4. Right?
It is just single point RO-MP2 energies with Dunning’s 6Z basis set.
Ah yes, in this case you would need to compile from scratch. An alternative would be to use a conventional algorithm by setting:
The normal binaries are compiled with max AM 6, which is sufficient for most Sherrill group production calculations. Occasionally I build a AM 7 binary on request, so if you
conda remove psi4 and then
conda install psi4=0.5.92 you can get an AM 7 binary from 2 months ago. We’ve a plan in the works that will allow very high AM to be distributed with every nightly build binary without incurring the prohibitive integrals build every night, but that’s not quite ready.
When fitting basis sets aren’t available, psi4 defaults to def2-qzvpp. Your skepticism does you credit– the df a6z results aren’t reliable when a quad-zeta fitting basis is used. There’s a ticket to raise the alarm when such calcs are initiated (https://github.com/psi4/psi4/issues/144). For now, if you’re wanting sextuple-zeta quality results, you’re better off
set scf_type pk and
set mp2_type conv to avoid the missing auxiliary sets problem and avoid df error as well.
Thank you very much. Is it possible to use
scf_type direct ? I think that it will be impossible to me afford a 6z pk calculation.
direct or any non-
df will be fine.
pk is just my default non-
Thank your for the help. I will try as soon as the actual calculations finish.
On the latest clone I tried:
cmake -H. -DLIBINT_OPT_AM=11 …
Manually-specified variables were not used by the project:
I also looked for the “setup” file so that I could do it the other way, but couldn’t find the setup file.
What should I do?
You’re building from scratch? It’s
-DMAX_AM_ERI=11 and that’ll trigger deriv of 12. That’s extreme. https://github.com/evaleev/libint/blob/v1/README_CMake.md
You can grab up to AM=8 from conda (http://psicode.org/psi4manual/master/build_planning.html#how-to-get-high-angular-momentum-integrals-from-conda), and I think Susi’s got int=10/deriv=6 for Fedora.