Segfault in 1.4a1.dev60

#1

Hi!

I’m using the latest version of Psi4 (1.4a1.dev60) on Fedora 28/29 and encountered the problem with segmentation fault. Here is some input:

molecule {
0 3
   H        0.14748       -1.86719        0.82318
   N        0.22937       -0.87127        0.58052
   O        0.68847       -0.78564       -0.71211
   O       -0.64741       -0.08013        0.99612
   N       -0.23139        1.69102       -0.83138
   H       -1.25902        1.75249       -0.85594
   H        0.25985        1.10119       -0.80404
   H        0.02703        2.29628       -0.03930
}
set {
basis cc-pvdz
reference uhf
}
gradient('b3lyp')

It reproduced with every methods (DFT, MP2, OMP2, CCSD, SCF) and basis sets (cc-pvdz, sto-3g, 6-31g). I guess, the problem is analytic gradients. It is always stops at SCF GRAD module. With dertype='energy' the routine works clear.

#2

I cannot reproduce your error with the current psi4/label/dev from conda. It works fine. How did you install your version?

My conda version is 1.4a1.dev163, can you upgrade?

#3

I compiled Psi4 from source. Maybe that’s the point. GCC version 8.3.1.

Yes, I can upgrade to 163 and recompile the code. If the problem will appear again, I must revise the make scrit.

Thank you!

#4

After I recompiled Psi4 the problem is still here. Gradient, optimize and freq functions ends with segmentation fault.

Psi4 executable from conda repo is OK. I can calculate analytic gradients with every accesible methods.

So the problem is my compilation script. Question is closed, I guess.

#5

Any chance there’s an ubuntu or conda libint package around such that the build-time libint could be different from the runtime-loaded libint? If those two are built with different max AM, a runtime segfault is what happens. peripherally-relevant post

I can’t find a specific directly, but I’m confident psi has built and at least quick-tested with a GCC 8.

#6

For the record, I am building psi4 regularly with gcc 8.1 on my mac.

#7

Exactly! I tried to recompile Psi4 with different AM number. That’s my fault, I’m sorry :pensive: