EPW was killed when allocating arrays and open files

[snake_skin]

Dear all,

I have recently been using EPW to study polarons in transition metal oxides. When I increase nkf and nqf to 13 or above, the EPW program encounters the following error:
===================================================================
Use the band from 1 to 24 total 24
Including bands: 1 2 3 4 5 6 7 8 9 10
Including bands: 11 12 13 14 15 16 17 18 19 20
Including bands: 21 22 23 24
Use the band from 1 to 24 total 24 in saving g
Gathering eigenvalues of 24 bands and 2197 k points
The band extremes are at band 24
Fermi Energy is 10.4835306 (eV) located at kpoint 1 0.000 0.000 0.000 band 24
Allocating arrays and open files.

===================================================================================
= BAD TERMINATION OF ONE OF YOUR APPLICATION PROCESSES
= RANK 0 PID 24374 RUNNING AT cu74
= KILLED BY SIGNAL: 9 (Killed)
===================================================================================

should I use etf_mem = 2? But I don't know how to use this parameter, like how to extract formation energy from result(I tried this when I was learning to use EPW in the examples, but the epw2.out just has some basic information, I don't know the workflow).

And should I use efermi_read? I don't know what impact this parameter has on the results. When should I enable it?

Thanks a lot!

Regards
Snake

Attached is the input file.
epw2.13.in
&inputepw
prefix = 'm-HfO2'
outdir = './'

elph = .true.
epwwrite = .false.
epwread = .true.
lpolar = .true.
etf_mem = 1

nbndsub = 24
dvscf_dir = './save/'

wannierize = .false.
num_iter = 200
iprint = 2
proj(1) = 'O:p'
bands_skipped = 'exclude_bands = 1-24'
dis_win_max = 10.5
dis_froz_max = 10.5
dis_froz_min = 5
wannier_plot = .true.
wannier_plot_supercell = 3 3 3

! efermi_read = .true.
! fermi_energy = 10.4835

! lifc = .true.
! asr_typ = 'simple'

plrn = .true.
restart_plrn = .false.
type_plrn = 1
init_plrn = 1
init_sigma_plrn = 2.0
niter_plrn = 500
conv_thr_plrn = 1E-4
ethrdg_plrn = 1E-4
! adapt_ethrdg_plrn = .true.

nk1 = 6
nk2 = 6
nk3 = 6
nq1 = 3
nq2 = 3
nq3 = 3

nkf1 = 13
nkf2 = 13
nkf3 = 13
nqf1 = 13
nqf2 = 13
nqf3 = 13
/

[stiwari]

Hi,

Can you explain a bit more regarding the calculation? How many cores you were using etc.?
Also, will be helpful if you can attach the traceback.

Best regards,
Sabya.

[snake_skin]

Hi Sabya,

I used 64 cores to compute with 515472 memory in default display in one NODE. And then I tried to use 32 cores with 2063084 memory in default display in another NODE, I can increase nkf and nqf to 15.

I used 6×6×6 coarse grid in scf, 6×6×6 grid in nscf, 3×3×3 q-grid.

Unfortunately, I have not save the traceback. But I remember that our SBTACH logs also showed that the program had been killed, just like above.
To further study this material, I have increased the q-grid because of bad decay in decay.epmatp. I think this problem will happen again, and I can provide the traceback when this problem happen.

By the way, could I use atom labels such as "O1", "O2" to identify non-equivalent atoms in the unit cell? Thanks a lot!

Best regards,
Snake.

[stiwari]

Hi Snake,

If you face memory problem again, please use the latest develop version of EPW from here: https://gitlab.com/epw/q-e.git, This version of EPW will be released soon as EPW v6.1 alongside QE v7.6, and can be image parallelized even for the coarse grid calculations.
To use image parallelization use

Code: Select all

mpirun -np <cores> epw.x -nk <npool>  -ni <nimage> -in epw1.in 

Code: Select all

cores = npool * nimage

Using images will allow you to utilize more cores and will reduce memory requirement per core. Also, do report the error in this thread if it re-appears.

Best,
Sabya.