Higher-order derivatives not working -- Libint2 limitation?

[chertianser]

I'm attempting to reproduce the example after following the installation instructions provided in the README. Quax runs fine for the first and second order derivative calls, but orders above two will fail:

import quax 
import psi4

molecule = psi4.geometry('''
                         0 1
                         H 0.0 0.0 -0.80000000000
                         H 0.0 0.0  0.80000000000
                         units bohr
                         ''')

dz1_dz1_dz2_dz2 = quax.core.geom_deriv(molecule, '6-31g', 'ccsd(t)', deriv_order=4, partial=(2,2,5,5))

python3: /h/292/ctser/functor/libint/BUILD/libint-2.9.0/PREFIX/include/libint2/./engine.impl.h:483: const target_ptr_vec& libint2::Engine::compute1(const libint2::Shell&, const libint2::Shell&): Assertion `deriv_order_ <= 2 && "feature not implemented"' failed.
/var/spool/slurmd/job216925/slurm_script: line 15: 1964948 Aborted                 (core dumped) python3 run_quax.py

The check for derivative order stated at the end of issue #1 have been since removed from quax/methods/ints.py. Could this be reinstated somehow? Thanks!

[EricaCMitchell]

Seems that the Libint wiki needs to be fixed, "libint currently supports computation of arbitrary-order geometric derivative integrals; this interface has supported arbitrary-order derivatives since version 2.7.0 (versions 2.6.0 and older only exposed up to 2nd-order derivatives of 2-body integrals)."

I can bring back the tmp_potential function, but it should be noted that integrals computed using JAX are extremely slow, which is why they were removed in the current version.

I've moved groups so Quax does not have a primary maintainer currently, but I do work in the group maintaining Libint now. So, I can talk to the maintainers of Libint about this issue with the potential integrals, but integration of arbitrary-order derivatives for the potential integrals is not a high priority for them.