Fix UKS gradient threading

[loriab]

I just noticed that the UKS gradients are off when nthreads>1. Failing tests are: dft_grad_lr2, dft_grad_lr3, dft_grad_meta, dft-grad2. An example case is:

func_reference = psi4.Matrix.from_list([                         #TEST
 [ 0.,          0.,         -5.23600907e-02],                    #TEST
 [ 0.,          0.,          5.23600907e-02]])                   #TEST
cam_b3lyp_reference = psi4.Matrix.from_list([                    #TEST
 [ 0.,          0.,         -1.09138077e-01],                    #TEST
 [ 0.,          0.,          1.09137576e-01]])                   #TEST
func_uks_reference = psi4.Matrix.from_list([                     #TEST
 [ 0.,          0.,          1.77777928e-03],                    #TEST
 [ 0.,          0.,         -1.77777928e-03]])                   #TEST
cam_b3lyp_uks_reference = psi4.Matrix.from_list([                #TEST
 [ 0.,          0.,         -3.46813335e-02],                    #TEST
 [ 0.,          0.,          3.46809033e-02]])                   #TEST

molecule {
    H  0.000  0.000  0.000
    F  0.000  0.000  1.100
}

set {
    scf_type              df
    basis                 cc-pvdz
    dft_radial_points     120
    dft_spherical_points  434
    e_convergence         12
    d_convergence         10
    points 5
}

molecule {
    1 2
    H  0.000  0.000  0.000
    F  0.000  0.000  1.100
}

set reference uks

anl_grad = gradient('scf', dft_functional="CAM-B3LYP", dertype=1)
fd_grad = gradient('scf', dft_functional="CAM-B3LYP", dertype=0)
compare_matrices(cam_b3lyp_uks_reference, anl_grad, 6, "Analytic vs Reference CAM Gradients")    #TEST
compare_matrices(anl_grad, fd_grad, 4, "Analytic vs FD CAM Gradients")    #TEST

[JonathonMisiewicz]

The bug was introduced in #2839, which parallelized UKS gradients. I assumed the feature was ready to go, somebody had just forgotten to enable it, but apparently not.

I'll fix this before the release.

[susilehtola]

The bug was introduced in #2839, which parallelized UKS gradients. I assumed the feature was ready to go, somebody had just forgotten to enable it, but apparently not.

The question is why was this not caught by the test suite..?

[JonathonMisiewicz]

I recall that the test suite runs in serial. We'd love a parallel version, but @loriab has had other priorities.

Do correct me if I'm wrong.

[loriab]

The parallel test suite is a matter of adding -n2 in 2-3 places. I really should make it an option. CI (that won't tell us how many threads are even available) hasn't been the best place for running a parallel suite.

[q-posev]

GitHub runners with Ubuntu have 2 cores, the MacOS ones have 3. Source