Restarting Eliashberg calculations for further temperatures

[tveskaeg]

Hey everybody,
I am using EPW v.5.2.0, and I have completed an anisotropic Eliashberg run at certain temperatures (using nstemp, tempsmin, tempsmax), but now I am struggling to launch a restart for different temperatures: I am using the same number of nodes and pools on the same machine, I have all .ephmat, .freq, .egnv, and .ikmap files, but the calculation crashes with

Finish reading .egnv file


%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Error in routine invmat (1):
error in DGETRF
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

stopping ...

Before the crash, the program runs normally with the steps
- recognizing a restart ("RESTART ... Restart is done without reading PWSCF save file)
- starting the Eliashberg part ("Solve anisotropic Eliashberg equations")
- reading some of the files ("Finish reading .freq file", "Finish reading .egnv file")

In the first run I used these flags:
ep_coupling = .true.
elph = .true.
epbread = .false.
epbwrite = .true.
epwread = .false.
epwwrite = .true.
kmaps = .false.
wannierize = .true.

eliashberg = .true.
ephwrite = .true.
laniso = .true.
limag = .true.
lpade = .true.
lifc = .true.
mp_mesh_k = .true.

For the restart I tried these flags (giving the additional temperatures with temp(1) = ..., temp(2) = ...):
ep_coupling = .false.
elph = .false.
epbread = .false.
epbwrite = .false.
epwread = .true.
epwwrite = .false.
kmaps = .true.
wannierize = .false.

eliashberg = .true.
ephwrite = .false.
laniso = .true.
limag = .true.
lpade = .true.
lifc = .true.
mp_mesh_k = .true.

Do you have any idea what could cause this crash?
Thanks and all the best

[hlee]

Dear tveskaeg:

Please read my previous answer at viewtopic.php?p=3854#p3854 .

Or please rerun calculations with the recent version of EPW.
In general. I would suggest users to always use the recent version of code since several bugs or issues in previous versions were fixed.

Sincerely,

H. Lee

[hpaudya1]

Hi tveskaeg:

In addition to H. Lee's answer, I would like to add the following.

As far as I remember, while mapping k and k+q, the code uses symmetry (in the version you are using, and not required in the latest versions), which is missing in your restart.

To make it work, you can do the following;
1. run your scf.in and nscf.in (like what you did in the very first step to create prefix.save directory)
2. run your restart epw.in (see the input flags, I have modified your restart input)

Code: Select all

ep_coupling = .false.
elph = .false.

epbread = .false.
epbwrite = .false.

epwread = .false.
epwwrite = .true.

kmaps = .true.
wannierize = .false.

eliashberg = .true.
ephwrite = .false.

laniso = .true.
limag = .true.
lpade = .true.
lifc = .true.
mp_mesh_k = .true.

Please note that, you need to use the same number of cores that you have used to write prefix.ephmatXX files (#of cores = #of prefix.ephmatXX files) in your restart.

Lest me know what happens,

Best,
Hari Paudyal

[tveskaeg]

Dear Hyungjun, dear Hari,

thank you for your answers and suggestions! In future calculations I will stick to the latest EPW release, but in that special case I had to work on an existing run with version 5.2. Unfortunately I couldn't try another version for the restart, as the changes in 5.1, 5.2 and 5.3 affected essentially the way the Eliashberg quantities are written to file.

However, I tried Hari's suggestion (Eliashberg restart with epwwrite = .true.) and this works perfectly in EPW 5.2!
I have one more question, out of curiosity: I noticed that the epwwrite quantities (epwdata.fmt and XX.epmatwpX) are actually not rewritten in that setting and the EPW std-output claims regular mode instead of "RESTART", so I wonder which other flag(s) determine this behaviour? Or is it really that special set of flags you suggested, Hari?

All the best
Roman / tveskaeg

[hpaudya1]

Hi Roman,

I have one more question, out of curiosity: I noticed that the epwwrite quantities (epwdata.fmt and XX.epmatwpX) are actually not rewritten in that setting and the EPW std-output claims regular mode instead of "RESTART", so I wonder which other flag(s) determine this behaviour? Or is it really that special set of flags you suggested, Hari?

I actually did not understand your question clearly, can you please explain a little bit in detail. (what do you want to do).

Happy EPWing,
Hari Paudyal