Troubleshooting Determinant Import Issues in TREXIO with PySCF

[NastaMauger]

Hello,

I am trying to import my casci wavefunction into qp2. To do so I am using trexio-tools.
After importing my one- and two-electron integrals and my TREXIO data from PySCF using:

qp_import_trexio.py casci.hdf5
qp set_file casci.hdf5.ezfio/
qp set mo_basis mo_label Canonical
qp_run import_trexio_integrals casci.hdf5.ezfio
qp_run print_energy casci.hdf5.ezfio

I can verify that it returns the same energy components as PySCF.
However, I would also like to import my determinants. To do so, I ran:

qp_run import_trexio_determinants casci.hdf5.ezfio/

However, I am experiencing a few issues:

1. The files created in casci.hdf5.ezfio/determinants, named psi_det_qp_edit.gz and psi_det.gz, don’t seem to contain meaningful data, although the corresponding coefficients files appear to be correct.
2. While no error occurs when running the import_trexio_determinants command, the print_energy command now fails with this error:

Providing the one-electron integrals
* N_int                                                          1
 * mo_label          Canonical
* Number of determinants                                         5
* Dimension of the psi arrays                                    5
 Read psi_det
 Read psi_coef           5           1
.. >>>>> [ IO READ: mo_class ] <<<<< ..
  cpu  time :  4.399999999998849E-005 s
  wall time :  7.152557373046875E-006 s  ( x    6.15164586666506      )
 MO map initialized:                   1540
 ac_operator: iorb < 1
    -1822208          10
-1
Program exited with code 255.
Wall time: 0:00:01

My guess is that I might not be registering my determinants list correctly. However, if this were the case, how could the import_trexio_determinants command run without issues?

Here is how I handled this in my PySCF script:

# Determinants in TREXIO
from trexio_tools.group_tools import determinant as trexio_det
int64_num = int((mo_num-1)/64) + 1 

det_list = []
#occsa is my spin up list whereas occsb is my spin down
for occsa, occsb, _, _ in data_all_sorted:
    det_tmp = []
    det_tmp += trexio_det.to_determinant_list(occsa, int64_num)
    det_tmp += trexio_det.to_determinant_list(occsb, int64_num)
    det_list.append(det_tmp)
    print(det_tmp)

offset_file = 0 
trexio_file.set_state(1)
trexio.write_state_num(trexio_file, 1)
trexio.write_determinant_list(trexio_file, offset_file, num_determinants, det_list)

ci_coeff_list = []
for occsa, occsb, ci_coeff, _ in data_all_sorted:
    ci_coeff_list.append(ci_coeff)

if len(ci_coeff_list) == num_determinants:
    dset = np.array(ci_coeff_list)  
    trexio.write_determinant_coefficient(trexio_file, offset_file, num_determinants, dset)

I’ve attached my casci.hdf5 file, generated by Pyscf for reference.

If anyone has insights into why this error is occurring or suggestions on how to resolve it, I would really appreciate your help. Thank you in advance!

Best

[q-posev]

Hello @NastaMauger ,
When using @scemama's resultsFile package to convert Gaussian/GAMESS CI determinants into TREXIO, we shift the orbitals by 1 (see here). Are the orbital indices in occsa 0-based or 1-based?
Or perhaps there is a bug in import_trexio_determinants. This is @scemama's call.

[NastaMauger]

Hello @q-posev

I am a bit confused regarding the 0-based or 1-based question.
This is what it looks like with Pyscf:

Full list of occupied orbitals (with CI coefficients):
   det-alpha (occsa),         det-beta (occsb),
   [0 1]                              [0 1]         -0.99779049148461
   [0 2]                              [0 2]         0.033402646674729995
   [0 2]                              [1 2]         -0.03321881541135069
   [1 2]                              [0 2]         -0.03321881541135069
   [1 2]                              [1 2]         0.033036629721523374

Best

[q-posev]

So the PySCF orbital list is 0-based, meaning that the lowest orbital has index 0. I have a strong suspicion that the to_determinant_list function that you use expects the orbital list to be 1-based (namely the lowest orbital index starting from 1). Mainly because we compute modulo = int((orb-1)/64). 0-based vs 1-based is similar to Python versus Fortran difference for indexing arrays. To be confirmed by @scemama.

So you can use the same trick as we do for GAMESS/Gaussian:

orb_list_up  = [ orb+1 for orb in occsa ]
det_tmp += trexio_det.to_determinant_list(orb_list_up, int64_num)

and same for beta part.
Hope it helps!

[NastaMauger]

I have modified my Pyscf code this way

det_list = []
for occsa, occsb, _, _ in data_all_sorted:
  occsa_upshifted = [orb + 1 for orb in occsa]  
  occsb_upshifted = [orb + 1 for orb in occsb]
  det_tmp     = []
  det_tmp    += trexio_det.to_determinant_list(occsa_upshifted, int64_num)
  det_tmp    += trexio_det.to_determinant_list(occsb_upshifted, int64_num)                                                                  
  det_list.append(det_tmp)

And the export works fine now with qp2. qp_run print_ci_vectors casci.hdf5.ezfio/ correctly shows my CI vectors; however, qp_edit casci.hdf5.ezfio/ shows only one CI vector, even though n_det is 5 in my case. I’m not sure if this is normal or if it could be an issue for further calculations in QP2

Best

[NastaMauger]

Oh, I just realized you posted the same code solution as me, @q-posev. Sorry for the trouble and thank you so much! :)

[q-posev]

Good news! The rest is up to the QP2 gurus 😊

[scemama]

And the export works fine now with qp2. qp_run print_ci_vectors casci.hdf5.ezfio/ correctly shows my CI vectors; however, qp_edit casci.hdf5.ezfio/ shows only one CI vector, even though n_det is 5 in my case. I’m not sure if this is normal or if it could be an issue for further calculations in QP2

OK. There is a little thing missing in qp import_trexio_determinants to make qp_edit work. I will fix it. As long as qp_run print_ci_vectors is correct, everything is fine.

[scemama]

@NastaMauger Can you upload your trexio file so I can search for the qp_edit problem please?

[NastaMauger]

Hello,

Here is the casci.hdf5.ezfio.zip file. As you might notice, psi_det_qp_edit.gz and psi_det.gz are different. I believe the export code is missing the ezfio.set_qp_edit_determinant function call.

Best

[q-posev]

@NastaMauger small correction to your script: in TREXIO the ground state has id=0, so you should use trexio_file.set_state(0) to write the CI wave function for the ground state. Reference to the TREXIO doc:

[NastaMauger]

I have modified my script in #45.
Thank you for letting me know!