I am trying to do the band structure calculation considering the Spin Orbit Coupling but getting the following error:

Computing kpt #:    92
 total cpu time spent up to now is     3908.6 secs

 Computing kpt #:    93
 total cpu time spent up to now is     3955.6 secs

 Computing kpt #:    94

     Error in routine cdiaghg (463):    
     S matrix not positive definite     

I have tried doing this calculation with both PAW and Ultrasoft fully relativistic pseudopotential and getting the same error in both the cases. Before it I have already done the scf calculation without any error with both the pseudopotential. I have tried increasing the plane wave cutoff too, but still getting the same error.

The link for the scf, nscf input out file is here.

I am wondering why this is happening. I have already checked for the overlap of atoms in xcrysden but I don't think that is the problem cause then scf calculation would not have been done successfully. Please suggest me how can I remove this errors as I am dealing with it for last three days.
Thank you!

  • 1
    $\begingroup$ Did you try with diago_david_ndim=2? $\endgroup$
    – 147875
    Oct 9 '21 at 11:42
  • 1
    $\begingroup$ @147875 Nope, I do not know about it. Will it resolve this issue? $\endgroup$
    – UJM
    Oct 9 '21 at 12:26
  • 1
    $\begingroup$ May be, you should tweak the parameters. Read these articles QE-Developer Emails, QE bug, Research Gate. If still problem persists, try changing the diagonalization = 'cg' $\endgroup$
    – 147875
    Oct 9 '21 at 12:37
  • 1
    $\begingroup$ @147875 Thank you for providing links. I have already read the thread from research gate link and also tried with diagonalization = 'cg' but it didn't work for me. Let me try with diago_david_ndim=2 and revert back. $\endgroup$
    – UJM
    Oct 9 '21 at 12:48
  • 1
    $\begingroup$ We have a QE chatroom in case you need to chat a lot more! $\endgroup$ Oct 9 '21 at 16:34

This error has been resolved now. Though I am not an expert but here are few thoughts. There may be several reasons for this error:

  • This error might appear due to numerical instability from overlapping atoms.
  • As mentioned by @Phil Hasnip, if the S-matrix eigenvalues are really small.
  • Some pseudopotential may not fit with calculation, USPP giving non-positive charge density, leading to a violation of positiveness of the S matrix appearing in the USPP formalism or maybe the presence of the ghost states (I am not aware of these states)

These steps may help in resolving the issue:

  • Changing the pseudopotential file, it may be corrupted.
  • Set the higher ecutwfc and ecutrho values.
  • Changing the parameter diago_david_ndim to 2(or maybe increase to 8). Currently, the default value is 4.
  • Run the calculation with the diagonalization = 'cg' instead of default value davidson. cg diagonalization is more robust.

In my case, changing the value of diago_david_ndim to '2' has resolved the issue for 'davidson' diagonalization.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.