memory error in EPW calculations

[simba2828]

Dear Experts,

I am getting this large memory issue, which I am unable to resolve. I tried looking for a solution into the forum but did not find a proper solution.

Code: Select all

Size of required memory per pool: ~=  171.5761 Gb

 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
     Error in routine mem_size_eliashberg (1):
     Size of required memory exceeds max_memlt
 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

I am using

Code: Select all

iverbosity  = 2
max_memlt =  18.0d0
nsiter   = 800
  degaussq     = 0.5 ! meV
  degaussw    = 0.04 ! eV ~  1/4 of fsthick
  wscut = 0.2   ! eV 10 times of  Upper limit over frequency integration/summation in the Elisashberg eq(1 cm-1 ~ 1/8000 eV)
  fsthick     = 0.2  ! eV ~ 4 times the maximum phonon frequency
  muc     = 0.1

mp_mesh_k = .true.
  nkf1 = 40
  nkf2 = 40
  nkf3 = 40

  nqf1 = 20
  nqf2 = 20
  nqf3 = 20

Can somebody help with the issue?

Thank you,
Shubham

[hpaudya1]

Hi Shubham,

You currently have "max_memlt = 18.0d0" specified in the input, meaning this is the maximum amount of memory can be allocated per pool. Is this value set to the maximum available memory per core in the cluster you are using? The code crashed because the memory requirement is approximately 172 GB per pool, which exceeds the allocation. I recommend verifying the usable memory on your system and adjusting the "max_memlt" accordingly. Additionally, you may consider reducing some parameters, such as k/q meshes, fsthick, and wscut, to optimize memory usage.

Happy EPWing,
Hari

[simba2828]

Hi Hari,

Yes, the maximum memory available per core in my system is 18GB. I can't use more than that. I will check the calculations reducing the parameters you suggested. However, is there any other parameter which can reduce the memory usage, such as, `etf_mem`?

Please let me know.

Thank you,
Shubham

[Shashi]

Hi Subham,

You can try removing the max_memlt tag from the input. It will run kernel vertex calculation on the fly instead of reading every time from the file. Give it a try if it solves your issue.

Second, etf_mem=1 is default which uses less memory and is used for interpolation not for solving Eliashberg equations. So it will not be of help in your case.

Regards,
Shashi

[simba2828]

Hi Shashi,
When I remove max_memlt tag from the input, I get the following memory related error:

Code: Select all

stopping ...

 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
     Error in routine mem_size_eliashberg (1):
     Size of required memory exceeds max_memlt
 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

     stopping ...
 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

I reduced k points `nkf` to reduce the memory required per pool but it still requires 35GB (per pool). And as I mentioned earlier, my system has maximum 18GB memory per core. Anything else you would suggest?

Thank you,
Shubham

[Shashi]

Hi Subham,

Could you confirm whether you are using the latest version of EPW v5.9? If not, I recommend using the latest version and try again.

Best regards,
Shashi

[simba2828]

Hi SHashi,

I am using EPW 5.6, I will try with the latest version and get back to you.

Thank you