Hi all,
My cluster system has a limited time (less than 4 days) to run a single job. So, when i submitted a TDDFT-time-evolution job, it usually broke off in the middle of given time scale.
In this case, how to restart a TDDFT time evolution ?
Junjie He
BCCMS, Universität Bremen
Last edit: Junjie He 2019-12-09
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
From what I know, TDDFT jobs are not restartable.
You could try reducing paramaters that define quality of the calculation.
For dielectric responce, those are lmaxo and gmaxrf.
Reducing number of points in wplot might help as well.
Finally, use more nodes if possible, although do some tests - more nodes does not always mean more speed.
Finally, worth trying to ask admins of the cluster to make an exception for you - for instance, a separate queue with no time limit.
Best regards.
Andrew
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
The charge density and magnetization and many other time-dependent quantities are discontinuous after the restart.
You can try it with any simple example.
Regards,
A.K.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I try to restart TDDFT in the calculation of laser induced current for GaAs in examples by using 461, and the time-dependent current density is discontinuous after the restart. Can this problem be solved maybe?
Best regards.
Lu
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi all,
My cluster system has a limited time (less than 4 days) to run a single job. So, when i submitted a TDDFT-time-evolution job, it usually broke off in the middle of given time scale.
In this case, how to restart a TDDFT time evolution ?
Junjie He
BCCMS, Universität Bremen
Last edit: Junjie He 2019-12-09
Dear Junjie,
From what I know, TDDFT jobs are not restartable.
You could try reducing paramaters that define quality of the calculation.
For dielectric responce, those are lmaxo and gmaxrf.
Reducing number of points in wplot might help as well.
Finally, use more nodes if possible, although do some tests - more nodes does not always mean more speed.
Finally, worth trying to ask admins of the cluster to make an exception for you - for instance, a separate queue with no time limit.
Best regards.
Andrew
You can restart TD-DFT by using
tasks
461
Dear Andrew and Sharma:
thank you for your answer
With regards,
Junjie
In fact, Elk has an additional backup system for TDDFT runs because they are so expensive and time-consuming. Just set (for example)
and Elk will write the files EVECSV_TD_BACKUP.OUT and TIMESTEP_BACKUP.OUT every 100 time steps.
If the code is terminated while writing EVECSV_TD.OUT and TIMESTEP.OUT then rename the corresponding backup files to these and run task 461.
Kay.
Hi,
The charge density and magnetization and many other time-dependent quantities are discontinuous after the restart.
You can try it with any simple example.
Regards,
A.K.
Hi,
I try to restart TDDFT in the calculation of laser induced current for GaAs in examples by using 461, and the time-dependent current density is discontinuous after the restart. Can this problem be solved maybe?
Best regards.
Lu