deepmodeling
Error: The factorization could not be completed because the input is not positive-definite
Describe the bug
Hi, developers. Thank you very much for your previous responses and help to the questions I raised. I’m opening a new issue here because I feel this question is unrelated to the previous ones, which will make it easier for others to refer to and review later. After training using E3 model, when I tried to plot the band, I met the following error.
............
File "/dptb/lib/python3.10/site-packages/dptb/postprocess/elec_struc_cal.py", line 159, in get_eigs
data = self.eigv(data)
File "/dptb/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1190, in _call_impl
return forward_call(*input, **kwargs)
File "/dptb/lib/python3.10/site-packages/dptb/nn/energy.py", line 75, in forward
chklowt = torch.linalg.cholesky(data[self.s_out_field])
torch._C._LinAlgError: linalg.cholesky: (Batch element 0): The factorization could not be completed because the input
is not positive-definite (the leading minor of order 246 is not positive-definite).
Expected behavior
Could you help me check what the issue might be? The dataset was obtained from ABACUS and then parsing with dftio.
To Reproduce
dptb run band.json -i ./output/checkpoint/nnenv.best.pth -o ./ -stu ./STRU.vasp Sorry, I am unable to upload the file directly due to its large size. The Google Drive link below contains the files that might be necessary to reproduce the issue. https://drive.google.com/file/d/1AS3l2kiOhXFJy74Ep8DzbnxpYG50CIty/view?usp=sharing
Environment
No response
Additional Context
No response
Describe the bug
Hi, developers. Thank you very much for your previous responses and help to the questions I raised. I’m opening a new issue here because I feel this question is unrelated to the previous ones, which will make it easier for others to refer to and review later. After training using E3 model, when I tried to plot the band, I met the following error.
............ File "/dptb/lib/python3.10/site-packages/dptb/postprocess/elec_struc_cal.py", line 159, in get_eigs data = self.eigv(data) File "/dptb/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1190, in _call_impl return forward_call(*input, **kwargs) File "/dptb/lib/python3.10/site-packages/dptb/nn/energy.py", line 75, in forward chklowt = torch.linalg.cholesky(data[self.s_out_field]) torch._C._LinAlgError: linalg.cholesky: (Batch element 0): The factorization could not be completed because the input is not positive-definite (the leading minor of order 246 is not positive-definite).Expected behavior
Could you help me check what the issue might be? The dataset was obtained from ABACUS and then parsing with dftio.
To Reproduce
dptb run band.json -i ./output/checkpoint/nnenv.best.pth -o ./ -stu ./STRU.vasp Sorry, I am unable to upload the file directly due to its large size. The Google Drive link below contains the files that might be necessary to reproduce the issue. https://drive.google.com/file/d/1AS3l2kiOhXFJy74Ep8DzbnxpYG50CIty/view?usp=sharing
Environment
No response
Additional Context
No response
Hi, a brief guess would be that the basis used to construct the overlap matrix is near degenerated. This will require that the overlap fitting be extremely accurate. I will check on the cases shortly.
