Restart q-point interpolation

[gbrunin]

Hi everyone,

I am using restart = .true. and restart_freq = 500 for a calculation. The file XXX.sigma_restart1 is well written, and I have the line

Code: Select all

Restart from:       3001/      3375

in my output file. My problem is that the interpolation seems to begin from the start again. It starts at iq = 1 and the calculation is now at iq = 395, thus higher than the 375 q-points remaining from the previous run. Here is my input file :

Code: Select all

EPW computation for Ca2N
&inputepw
  prefix      = 'Ca2N'
  dvscf_dir   = '../ph/save/'
  amass(1)    = 40.078
  amass(2)    = 14.007
  outdir      = '/scratch/ucl/naps/gbrunin/epw/'

  max_memlt   = 4

  !epbwrite    = .true.
  !epbread     = .false.
  epbwrite    = .false.
  epbread     = .true.

  epwwrite    = .true.
  epwread     = .false.
  kmaps       = .false
  elph        = .true.

  etf_mem     = .false.

  nbndsub     = 4
  nbndskip    = 1
  dis_froz_min   = -0.45
  dis_froz_max   = 3.92
  dis_win_min = -0.5
  dis_win_max = 3.92
  proj(1)     = 'N:px;py;pz'
  proj(2)     = 'f=0.5,0.5,0.5:s'

  !wannierize  = .true.
  wannierize  = .false.
  num_iter    = 1000
  iprint      = 2

  iverbosity  = 0

  elecselfen  = .true.
  phonselfen  = .true.
  a2f         = .true.

  parallel_k  = .true.
  parallel_q  = .false.

  fsthick     = 1  ! eV
  eptemp      = 300  ! K
  degaussw    = 0.025 ! eV
  degaussq    = 0.05 ! meV

  filukk      = './Ca2N.ukk'

  !rand_nk      = 20000
  !rand_nq      = 80000

  !rand_q      = .true.
  !rand_k      = .true.

  nkf1        = 70
  nkf2        = 70
  nkf3        = 70

  restart     = .true.
  restart_freq = 500

  nqf1        = 15
  nqf2        = 15
  nqf3        = 15

  nk1         = 24
  nk2         = 24
  nk3         = 24

  nq1         = 8
  nq2         = 8
  nq3         = 8
 /
      65 cartesian
...

Am I missing something ?
Thanks in advance

[sponce]

Hello Guillaume,

Indeed you are not doing a restart run. In addition to restart = .true. you also need:

Code: Select all

  epbwrite    = .false.
  epbread     = .false.

  epwwrite    = .false.
  epwread     = .true.
  kmaps       = .true.
  elph        = .true.

  etf_mem     = .false.

By the way, you should use the most recent version of the code.
In that case:

Code: Select all

 etf_mem     = 1 

Best,
Samuel

[gbrunin]

Hi Samuel,

Thanks for your answer. I did as you suggested but I still have the same problem. I have now

Code: Select all

------------------------------------------------------------------------
                   RESTART - RESTART - RESTART - RESTART
     Restart is done without reading PWSCF save file.
     Be aware that some consistency checks are therefore not done.
     ------------------------------------------------------------------------

at the beginning of my output file but the q-point interpolation starts at the first q-point, even if there is the message to say that it restart from another one.

Thanks for your suggestions,
Guillaume

[sponce]

Hello,

Its an information message. It's normal.

Are you sure that your XXX.sigma_restart1 is correct ?
You might have re-written on it since before you restarted incorrectly and start from the beginning. You said you were at 395 q which is less than 500 and therefore nothing
has been written to the restart file yet so if you restart it will restart from 1.

You can do a test by lowering restart_freq = 10 for example and then crash the code after 11 q-points. Then restart. It should restart from 10.

Best,
Samuel

[gbrunin]

Hi,

Yes, I thought I could have problems with that run because of that. I have two jobs that I want to restart, and I had not done anything on the second one yet. So I will explain with more details what I did for this one because it is cleaner.

First, I did a job for fine q-point and k-point grids of 20x20x20 (I know it's not enough) with the same input file as in my first post (except for the grids). The interpolation went up to the 7700 q-point out of 8000, and each 500 q-point a message was written that the restart point was created.
Then, I changed my input file as you said, and the restart calculation started at iq = 1 instead of 7501 as I expected.

I hope it is more clear now. I can do a test run but I think it will give me the same result.

Thanks,
Guillaume

[sponce]

Hello Guillaume,

Can you try the following input :

Code: Select all

EPW computation for Ca2N
&inputepw
  prefix      = 'Ca2N'
  dvscf_dir   = '../ph/save/'
  amass(1)    = 40.078
  amass(2)    = 14.007
  outdir      = './'

  epbwrite    = .false.
  epbread     = .false.

  epwwrite    = .false.
  epwread     = .true.
  kmaps       = .true.
  elph        = .true.

  etf_mem     = .false.

  nbndsub     = 4
  nbndskip    = 1
  dis_froz_min   = -0.45
  dis_froz_max   = 3.92
  dis_win_min = -0.5
  dis_win_max = 3.92
  proj(1)     = 'N:px;py;pz'
  proj(2)     = 'f=0.5,0.5,0.5:s'

  wannierize  = .false.

  elecselfen  = .true.
  phonselfen = .false.
  a2f           = .false.

  parallel_k  = .true.
  parallel_q  = .false.

  fsthick       = 3  ! eV
  eptemp      = 300  ! K
  degaussw   = 0.025 ! eV

  nkf1        = 24
  nkf2        = 24
  nkf3        = 24

  restart     = .true.
  restart_freq = 50

  nqf1        = 8
  nqf2        = 8
  nqf3        = 8

  nk1         = 24
  nk2         = 24
  nk3         = 24

  nq1         = 8
  nq2         = 8
  nq3         = 8
/
      65 cartesian
...

Manually crash the calculation just after the first 50 q-points have been written.

PS: Are you sure you need 24x24x24 k-grid for the fine grid ? This seems super dense. 16x16x16 should be already a lot.

Best,
Samuel

[gbrunin]

Hi Samuel,

Here is what it gives me.

Code: Select all

     Restart from:         11/      3375

     ===================================================================
     Electron (Imaginary) Self-Energy in the Migdal Approximation
     ===================================================================


     Fermi Surface thickness =   1.000000 eV

     Golden Rule strictly enforced with T =   0.025852 eV
     Gaussian Broadening:   0.025000 eV, ngauss=   1
     DOS =  0.522114 states/spin/eV/Unit Cell at Ef=  2.909832 eV

     Creation of a restart point
     Creation of a restart point

It does not print any information about the q-point interpolation in the output file, just when a restart point is created. I tried this input 3 times : the first time, I had only one "Creation of a restart point" message. The second time, it said that it started again from q-point 11 (I set restart_freq = 10), and created 2 restart points (see output above). Then the third time, it started again from q-point 21, but I guess it should have started from 31 if the restart point was working in the previous runs.

About the coarse k-point grid, will the computation be longer if it is denser ? Or "just" use more memory ?

Thanks for your help !
Guillaume

[sponce]

Hello,

I think it works as intended.

1) The printing information is hardcoded in EPW to be every 50 q-points (you can off course change that). This is the reason why you do not see the q-point information yet.

2) The reason why it restarted from 21 is because the first "creation of a restart point" is actually the re-writing to file of the first 10 q-points. So in practice you only go to 20 q-points.

3) yes it should be faster and less memory.

Best,
Samuel

[gbrunin]

Hi again,

I tried a little bit longer runs to be sure if it was working or not. I have the message (restart_freq = 50)

Code: Select all

Restart from:        151/      3375

and then

Code: Select all

Creation of a restart point
     Progression iq (fine) =        100/      3375
     Creation of a restart point
     Creation of a restart point

So I don't really understand if the job restarted properly or not.

Thanks,
Guillaume

[sponce]

Hello,

Ok that is indeed wrong.

Could you install the current development version of Quantum Espresso that includes EPW and try again:

https://github.com/QEF/q-e

You can do
git clone https://github.com/QEF/q-e.git

To get the latest version.

Sorry for the problems,
Best,
Samuel