Can not resolve "Error in routine cdiaghg (542): S matrix not positive definite" error.
Hello, I've been having difficulties running a system with noncollinear magnetism using a fully relativistic ECP and hybrid DFT functional. While I can achieve convergence for non-hybrid DFT, the code crashes for hybrids with the mentioned error in the title. I checked for some possible culprits:
- I checked that the geometry is OK since I can run hybrids in the same geometry using a scalar relativistic ECP in the ferromagnetic phase (FM).
- I checked for ghost states in my ECP and there appears to be non according to the OPIUM code. I also tried using another fully relativistic ECP and see the same error.
I then tried to modify my input parameters as following to resolve the error with no success:
- Increasing
mixing_ndim. - Using different
mixing_mode. - Decreasing
mixing_beta. - Changing the
smearingscheme. - Using larger pw-cutoff (400 Ry).
- Using a different k-mesh.
- Using
occupations = 'fixed'.
To give more info, I am trying to see the contribution of spin-orbit effects on the gap of this system. Particularly, I ran the FM phase in the scalar relativistic case, and therefore starting the magnetization in the noncollinear case from FM magnetization. I should note that nosym = .true. flag is needed for me to convert the orbitals for subsequent QMC calculations.
Any suggestions will be quite helpful, and the input/output files are attached below.
Version used: v.6.4.1
Machine: Andes/OLCF