Skip to content

fix: Ensure proper tensor conversion for numpy solver in Eigenvalues … #287

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 3 commits into from
Nov 30, 2025
Merged

fix: Ensure proper tensor conversion for numpy solver in Eigenvalues … #287

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
95bd47c
fix: Ensure proper tensor conversion for numpy solver in Eigenvalues …
AsymmetryChou Nov 28, 2025
c6f87e6
use local variable to ensure dtype correctness in data.dict
AsymmetryChou Nov 28, 2025
52d5928
fallback for h_transformed_np when overlap is False
Lonya0 Nov 29, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
17 changes: 11 additions & 6 deletions dptb/nn/energy.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -83,23 +83,28 @@ def forward(self,
for i in range(int(np.ceil(num_k / nk))):
data[AtomicDataDict.KPOINT_KEY] = kpoints[i*nk:(i+1)*nk]
data = self.h2k(data)
h_transformed_np = None
if self.overlap:
data = self.s2k(data)
if eig_solver == 'torch':
chklowt = torch.linalg.cholesky(data[self.s_out_field])
chklowtinv = torch.linalg.inv(chklowt)
data[self.h_out_field] = (chklowtinv @ data[self.h_out_field] @ torch.transpose(chklowtinv,dim0=1,dim1=2).conj())
elif eig_solver == 'numpy':
chklowt = np.linalg.cholesky(data[self.s_out_field].detach().numpy())
s_np = data[self.s_out_field].detach().cpu().numpy()
h_np = data[self.h_out_field].detach().cpu().numpy()
chklowt = np.linalg.cholesky(s_np)
chklowtinv = np.linalg.inv(chklowt)
data[self.h_out_field] = (chklowtinv @ data[self.h_out_field].detach().numpy() @ np.transpose(chklowtinv,(0,2,1)).conj())
else:
data[self.h_out_field] = data[self.h_out_field]

h_transformed_np = chklowtinv @ h_np @ np.transpose(chklowtinv,(0,2,1)).conj()

if eig_solver == 'torch':
eigvals.append(torch.linalg.eigvalsh(data[self.h_out_field]))
elif eig_solver == 'numpy':
eigvals.append(torch.from_numpy(np.linalg.eigvalsh(a=data[self.h_out_field])))
if h_transformed_np is None:
h_transformed_np = data[self.h_out_field].detach().cpu().numpy()
eigvals_np = np.linalg.eigvalsh(a=h_transformed_np)
# Preserve dtype by converting to the Hamiltonian's original dtype
eigvals.append(torch.from_numpy(eigvals_np).to(dtype=self.h2k.dtype, device=self.h2k.device))

data[self.out_field] = torch.nested.as_nested_tensor([torch.cat(eigvals, dim=0)])
if nested:
Expand Down