Siesta

TSHS takes too much memory in phonon calculation

Asked by Leonardo Fonseca

Dear all,

I am running a phonon calculation using a rather small system containing 28 atoms. During the calculation transiesta generates 6N+1 TSHS files, each one occupying considerable memory space. The total memory taken by all TSHS files combined is enormous. How to manage this problem? I plan to increase the number of unit cells in each direction to test for convergence but right now I am unable to calculate even for the smallest 28-atom case.

Leo

Question information

Language:
English Edit question
Status:
Solved
For:
Siesta Edit question
Assignee:
No assignee Edit question
Solved by:
Nick Papior
Solved:
Last query:
Last reply:
Revision history for this message
Nick Papior (nickpapior) said : 		#1

This is a known problem for phonon calculations using Inelastica.

Currently there is no way around it. Note that the TSHS files will only be generated for the displaced atoms.

Since regular phonon calculations using vibra does not require the TSHS files, you could do a regular FC run without writing the TSHS files.

Lastly, if this _is_ related to Inelastica, you should probably open an issue here: https://github.com/tfrederiksen/inelastica

Revision history for this message
Leonardo Fonseca (fonseca65) said : 		#2

Hi Nick,

Thanks for your quick reply. How do I instruct transiesta not to write the TSHS files?

My unit cell is a crystal containing 28 atoms, therefore I need all 28 atoms displaced to obtain the complete phonon spectrum. This should not be a big calculation challenge. In the examples you considered a Si supercell with 250 atoms. So I guess the problem is really to avoid writing the TSHS files, right?

Revision history for this message
Best Nick Papior (nickpapior) said : 		#3

TS.HS.Save false ;)

As for your calculation. If you have symmetry you don't need to displace all atoms, only the unique atoms. The rest is handled by symmetry.

Revision history for this message
Leonardo Fonseca (fonseca65) said : 		#4

Thanks Nick Papior, that solved my question.