Multi-determinant support for PySCF wavefunctions

[huangxxxxxxxx]

The newest version of QMCPACK is not support the multi-derterminant input file gernerated by Pyscf now. I would like to ask if this feature can be added in the next update

[prckent]

Please can you be specific about which version of PySCF you are using and attach a reproducer (molecule, basis set, specific methodology). Since there are several steps in getting from PySCF to QMCPACK, which step fails? I assume that there has been a recent update on the PySCF side. If you know that either older PySCF or older QMCPACK works, that will let the problem be identified faster.

[huangxxxxxxxx]

The PySCF is 3.8 version, and the QMCPACK is 3.11.0 version. In QMCPACK manual, if i want to use multideterminant, the introduction about it is "Multiple schemes to generate a multideterminant wavefunction are possible, from CASSF to full CI or selected CI.The QMCPACK converter can convert MCSCF multideterminant wavefunctions from GAMESS [SBB+93] and CIPSI [EG13] wavefunctions from Quantum Package [Sce17] (QP). Full details of how to run a CIPSI calculation and convert the wavefunction for QMCPACK are given in CIPSI wavefunction interface." There is not include PySCF software.

[prckent]

Please try with the latest QMCPACK release v3.17.1. See also https://qmcpack.readthedocs.io/en/develop/additional_tools.html#supported-codes for a PySCF example.

[anbenali]

Dear Huang

Am I correct in understanding that what you want is running a multideterminant calculation in PYSCF and then using that multideterminant wavefunction directly in QMCPACK? If that is the case, that functionality never existed. We tend to use other programs for our MSD expansions.

There are no technical reasons for this not to be implemented, except for a "priority" reason and developers being oversubscribed.

Let us know exactly what is the input file from pyscf you have and we will try to see if time allows if we can make that converter work.

Best

On Wed, Sep 20, 2023, 8:59 AM Paul R. C. Kent @.***> wrote:

Please try with the latest release 3.17.1. See also https://qmcpack.readthedocs.io/en/develop/additional_tools.html#supported-codes for a PySCF example.

— Reply to this email directly, view it on GitHub https://github.com/QMCPACK/qmcpack/issues/4740#issuecomment-1727793124, or unsubscribe https://github.com/notifications/unsubscribe-auth/AF4XVQ7QSL6F4DZVCLPOXT3X3LZE5ANCNFSM6AAAAAA475DGXI . You are receiving this because you are subscribed to this thread.Message ID: @.***>

[huangxxxxxxxx]

o2.py.zip This is the input file from PySCF, the multi-determinant will obtain when we run o2.py, but it can't convert to a format of QMCPACK

[prckent]

Did you find an alternative solution? (GAMESS, QP) or do you still need PySCF?

[huangxxxxxxxx]

GAMESS has been uesd before, but it can't offer a kohn-sham orbital of DFT. The QP hasn't been used now.

[huangxxxxxxxx]

we have a test that the orbital information is offered by .H5 which gennerated by PySCF software and the information about multideterminant is added to the .wfj.xml file with GAMESS format by munual. when run it by QMCPACK 3.9.0 version, the process is successful, but the variance of energy is too large. However, when run it by QMCPACK 3.17.0 version, the process is stop because some errors. part of input of *.wfj.xml follows: <?xml version="1.0"?>

[huangxxxxxxxx]