Since PySCF-2.3, B3LYP (and B3P86) are changed...

[Konjkov]

Use your best judgment to provide a useful level of information. Depending on the nature of the issue, consider including, e.g.

• Which version of the software you're running
• Any logging or error output you see

Hello QMCTorch developers.
I'am running tests of your program of latest (0.3.2) and
platform linux -- Python 3.10.12, pytest-6.2.5, py-1.10.0, pluggy-0.13.0

on simple test pytest scf/test_gto2sto_fit.py
I get a lot of warnings:

../../../.local/lib/python3.10/site-packages/pyscf/dft/libxc.py:771
  /home/vladimir/.local/lib/python3.10/site-packages/pyscf/dft/libxc.py:771: UserWarning: Since PySCF-2.3, B3LYP (and B3P86) are changed to the VWN-RPA variant, corresponding to the original definition by Stephens et al. (issue 1480) and the same as the B3LYP functional in Gaussian. To restore the VWN5 definition, you can put the setting "B3LYP_WITH_VWN5 = True" in pyscf_conf.py
    warnings.warn('Since PySCF-2.3, B3LYP (and B3P86) are changed to the VWN-RPA variant, '

tests/scf/test_gto2sto_fit.py::TestGTO2STOFit::test_forward
  /home/vladimir/.local/lib/python3.10/site-packages/torch/__init__.py:1144: UserWarning: torch.set_default_tensor_type() is deprecated as of PyTorch 2.1, please use torch.set_default_dtype() and torch.set_default_device() as alternatives. (Triggered internally at ../torch/csrc/tensor/python_tensor.cpp:432.)
    _C._set_default_tensor_type(t)

and error

FAILED scf/test_gto2sto_fit.py::TestGTO2STOFit::test_forward - ValueError: array must not contain infs or NaNs

maybe the error is a consequence of using an outdated method torch.set_default_tensor_type, but it seems to me that the tests are a bit outdated.

Best, Vladimir.

[NicoRenaud]

Thanks for the report @Konjkov !
I'm going to update the code as a few issues are popping up.
I'll keep everyone updated when I'm done !

Thanks

[NicoRenaud]

Hey @Konjkov,

I've fixed the torch warning that you flagged in the branch issue170. The test are passing on GH Action (here). Could you tell me two things:

• which OS are you using ? I've only tested the code on Ubuntu so the code might misbehave on OSX
• do you remember how you installed qmctorch ? Did you use the pip release or did you clone the repo and pip installed it ?

Thanks again for your help !

[Konjkov]

Hello Nico.

I just wanted to get acquainted with the project. I'am usinh Ubuntu 22.04, clone the repo and pip installed, torch==2.5.1.

I put .pyscf_conf.py with

B3LYP_WITH_VWN5=True

this fixes the 2nd warning but I can't understand if this will affect the execution of the rest of the tests.

I also constantly get an error

Error : Nan detected in local energy

perhaps it is discussed elsewhere #170

Vladimir

[NicoRenaud]

Hi Vladimir,

Thanks for the info !
I've found the issue, some obscure problem with the new implementation of the scipy.special.f2 function.
I've fixed the issue in the branch called issue170 and I will merge that with main as soon as the test are done.
That should fixed the NaN issue.

I think that the pyscf warning you ran into is something different and I'm not sure I can do something about it on my side. I'll look into it though :)

Thanks !

[NicoRenaud]

I've merged the fix in master, the NaN error should not happen anymore