I installed PSI 4 rev. 1.1 and tried to run a ccsd calculation on Ne atom with a cc-pV7Z basis set. Of course I got the dreaded “Libint max angular momentum error”.
I did try
conda install am7 -c psi4
and verified this installs the newly downloaded libint library in the library folder, but I still get the error. Then I tried
conda install am8 -c psi4
and still no luck. I tried stripping the K functions and noticed the calculation still crashes — but when I revert to the original libint that calculation ran to completion.
Does this mean updating libint requires recompilation of everything else? It’s not a “dropin” replacement?
many thanks in advance,
Jan Martin
PS: I did try building from source, but the antediluvian toolchain on our faculty HPC cluster (which runs Red Hat Enterprise Linux 6.6, with Intel compilers and MKL) apparently doesn’t agree with the cmake script.
Yes, I was just trying this out today and verify what you describe. It is intended as a dropin replacement not requiring recompilation. My current hypothesis is that the smaller MAX_AM_ERI was compiled into the AM-checking error messages that you received in libmints/eri_base.cc. So the libraries are willing to run your calc, but the AM-checking is preventing it. Only flaw in this reasoning is that I had tested dropin replacement a couple months ago, so how did that work?
The conclusion is that because the datastructure that libmints (psi4) acquires from the libint (not-psi) header is AM-dependent, libraries cannot be drop-in replacements. you can drop a new AM in and have a light compile, but that won’t work if psi is conda, not source.
Practical conclusion is that since above, we’ve switched the primary libint conda package to be AM=8, so a bit heftier but accommodates all computations.
The upcoming switch to Libint2 has more knobs for AM, so a smaller conda package may be in reach again.