Same input -> different integrals output on different systems


I’m running psi4 on a Linux and a Mac. Given the exact same input (and same version of psi4), there are some massive differences in the integrals that are generated for a SCF calculation. The charge density matrices also differ widely.

However, the energies are (mostly) the same. So presumably, psi4 chooses different basis sets in the same Hilbert space, which leads to energy convergence, but different intermediate steps.

Is there some way to ensure this doesn’t happen? I am using the integrals generated by psi4 and running more calculations on that, so this discrepancy is leading to massive differences on Linux vs Mac. Interestingly, linux outputs are always the same. So is this somehow machine dependent?