Author Topic: ** On entry to DSTEDC, parameter number 8 had an illegal value  (Read 3496 times)

0 Members and 1 Guest are viewing this topic.

Offline Alireza

  • Regular QuantumATK user
  • **
  • Posts: 13
  • Country: de
  • Reputation: 0
    • View Profile
Hello qatk,

I am using qatk/2022.12 and performing bandstructure calculations for twisted BL MoS2 in which the total number of atoms is at least 6160. I also have much bigger systems. In the beginning, I faced with out of memory error and was able to solve this issue by setting 
Quote
optimize_for_speed_over_memory=True,
in DiagonalizationSolver. I also set
Quote
store_basis_on_grid=True,
and
Quote
store_energy_density_matrix=True,
in AlgorithmParameters too. With these new settings, the calculation uses almost 20T memory with 124 cores to run. However, in the bandstructure analysis, I faced with an error stating
Quote
** On entry to DSTEDC, parameter number  8 had an illegal value
.

I was thinking since this is an expensive calculation with a large number of atoms, I should restrict the number of conduction bands to a certain value. Hence, I set
Quote
bands_above_fermi_level=5,
both in DiagonalizationSolver and Bandstructure analysis. But was not helpful and I got the same error. I learned that if
Quote
processes_per_kpoint=1
the diagonalization wants to use LAPACK algorithm, otherwise ELPA. And I think this error coming from ELPA.

The calculation is at dftb level using SLAKOs parameter.

In the attachment, you can find input, log and error files.

Offline Tue Gunst

  • Heavy QuantumATK user
  • ***
  • Posts: 46
  • Country: dk
  • Reputation: 3
    • View Profile
Re: ** On entry to DSTEDC, parameter number 8 had an illegal value
« Reply #1 on: February 15, 2023, 16:00 »
Hi Alireza,
Thanks for the detailed comments.
I see that the flexibility in controlling which solver to be used is not present in Bandstructure currently.
We are currently implementing/exposing iterative solvers in Bandstructure - but you will first be able to use that in the next release.
In the next release, it should be possible to perform bandstructure simulations on millions of atoms with iterative solvers (avoiding calculating all the valance bands).
For the present release, the only solution is to optimize for memory and reduce the number of k-points.
Tue

Offline Alireza

  • Regular QuantumATK user
  • **
  • Posts: 13
  • Country: de
  • Reputation: 0
    • View Profile
Re: ** On entry to DSTEDC, parameter number 8 had an illegal value
« Reply #2 on: February 16, 2023, 09:36 »
Hi Tue,

Thank you for the reply.
Is there any chance to use the preliminary version of iterative solver in bandstructure calculations now, as we urgently need it?

Alireza