Skip to content

NaN / Infinity gradients and general question concerning gradients #871

@steinmig

Description

@steinmig

Bug
With the attached input file and the current master built with gcc-9.2, I get a converged SCF solution, but if I request forces, I get NaN / infinity values (possibly coming from tblite? @awvwgk )

Additional question
Both the manual and the dftb_pin.hsd file suggest, that gradients in DFTB+ are calculated with finite differences, but requesting gradients via Analysis = {CalculateForces = Yes} has a close to identical runtime compared to without forces, suggesting that the gradients are analytical. Does tblite automatically support analytic gradients and the option in DFTB+ is a misnomen?

dftb_in.hsd.txt

Metadata

Metadata

Assignees

Labels

buglibrary: tbliteRelated to tblite external dependency (xTB Hamiltonian)

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions