RESCU+ Cell Relaxation : Page 2 - Nanoacademic forum

RESCU+ Cell Relaxation

rafaelfreire

Dear @vincentm ,
I have been testing the capabilities of RESCU+ software, but I had some issues with the parallelization scheme presented in this post.
When I use
inp["solver"]["mpidist"] = {"kptprc": 5}
the calculation failed in the end and I got the following message:

Note: The following floating-point exceptions are signalling: IEEE_INVALID_FLAG
ERROR STOP dispMat_mtrx_ift_dns: mat is null.

Please, could you give some hints on that problem?
Thanks in advance.

vincentm

rafaelfreire I have been testing the capabilities of RESCU+ software, but I had some issues with the parallelization scheme presented in this post.

To be sure, are you executing RESCU+ with a number of processes which is a multiple of 5 (or kptprc in general)?
Also, could you try other divisors of the number of processes you are using (e.g. 2, 4, 5 for 40 proc) and let us know what happens?

I have to say, this works on our side, but we want to make it work for you. We are working on a containerized solution at the moment, which will avoid this kind of issue in the future.

rafaelfreire

vincentm To be sure, are you executing RESCU+ with a number of processes which is a multiple of 5 (or kptprc in general)?

Yes. I am. I think I also tried multiples of 4. Anyway, I will do the tests you asked for and give you feedback later.
Thanks a lot for your fast answer! By the way, I am using the license provided by Ilum/CNPEM/Brazil.

rafaelfreire

I have done some tests using a different number of processors, namely 32, 40, and 64, and using "nkpprc" as a multiple of those numbers (2,4,5,8) whenever possible. However, I still get the same error message.
The error appears right after the end of the self-consistent loop. I will write down the resculog.out here, because I couldn't attach it.

>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>><<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<
------------------------------------------------------------------------
                       rescuplus_scf (1.1.0-rc.1)
------------------------------------------------------------------------
>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>><<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<
------------------------------------------------------------------------
                       enter rescuplus_scf program
------------------------------------------------------------------------
............................................................
\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\//////////////////////////////
This calculation is using the library Libxc 4.2.3.

Please cite Libxc as follows

Susi Lehtola, Conrad Steigemann, Micael J. T. Oliveira,
and Miguel A. L. Marques, Recent developments in Libxc -
A comprehensive library of functionals for density
functional theory, Software X 7, 1 (2018)
//////////////////////////////\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\
............................................................
------------------------------------
        mpi distribution info
------------------------------------
      proc num: 32
....................................
              grid dist
....................................
        levels: x y z k
           blk: 5 5 5 1
           prc: 2 4 2 2
           siz: 137 137 118 1
....................................
              orb dist
....................................
        levels: mu nu k
           blk: 32 32 1
           prc: 4 4 2
           siz: 624 624 1
------------------------------------
             atoms info
------------------------------------
      # species 1
        # atoms 48
          # aob 624
  ionic charge:    192.000
       valence:    192.000
    net charge:      0.000
....................................
           species #1 info
....................................
      atom num: 48
         label: C
        charge:   4.000
      psp path: /mnt/beegfs/home/rafael.freire/work_vasp/rescu/graphene_bilayer/flakes/test/fixedplane/2base-ring/R30/prc-tests/prc-2/C_TM_PBE_DZP.mat
         vnl L: 0 1 2 3
       vnl num: 16
      aob path: /mnt/beegfs/home/rafael.freire/work_vasp/rescu/graphene_bilayer/flakes/test/fixedplane/2base-ring/R30/prc-tests/prc-2/C_TM_PBE_DZP.mat
         aob L: 0 0 1 1 2
       aob num: 13
------------------------------------
              cell info
------------------------------------
lattice vectors (Bohr)
_______________________________________________
    51.74070126      0.00000000      0.00000000
     0.00000000     51.73885745      0.00000000
     0.00000000      0.00000000     44.59753652
_______________________________________________
      boundary: 0 0 0 0 0 0
       bravais: TET
          grid: 137 137 118
           blk: 5
           prc: 2 4 2
------------------------------------
            k-point info
------------------------------------
reciprocal lattice vectors (Bohr)
_______________________________________________
     0.12143603      0.00000000      0.00000000
     0.00000000      0.12144036      0.00000000
     0.00000000      0.00000000      0.14088638
_______________________________________________
          type: full
          grid: 1 1 1
         shift: 0.00 0.00 0.00
------------------------------------
           population info
------------------------------------
          type: ga
    sigma (ha): 0.367E-02
valence charge:   192.000
------------------------------------
               xc info
------------------------------------
    functional: ( 101) gga_x_pbe
    functional: ( 130) gga_c_pbe
------------------------------------------------------------------------
                          enter pre-processing
------------------------------------------------------------------------
computing ao-plans
computing vnl(R)
computing s(R), t(R)
computing fields
computing short-range energy
------------------------------------------------------------------------
                           exit pre-processing                               3.726 (s)
------------------------------------------------------------------------
------------------------------------------------------------------------
                       enter self-consistent loop
------------------------------------------------------------------------
    Iter       Etot (Ha)      dEtot (Ha)     dRho (/ e-)        Time (s)
       1   -0.273842E+03    0.273842E+03    0.857323E-01           4.751
       2   -0.276549E+03    0.270687E+01    0.414806E-01           4.668
       3   -0.267461E+03    0.908777E+01    0.318496E-01           4.575
       4   -0.271331E+03    0.386927E+01    0.120543E-01           4.554
       5   -0.271584E+03    0.253645E+00    0.587174E-02           4.596
       6   -0.271192E+03    0.391929E+00    0.187296E-02           4.585
       7   -0.271224E+03    0.319896E-01    0.502443E-03           4.638
       8   -0.271263E+03    0.385274E-01    0.131565E-03           4.580
       9   -0.271265E+03    0.212958E-02    0.525151E-04           4.578
      10   -0.271264E+03    0.842435E-03    0.173318E-04           4.644
      11   -0.271264E+03    0.564562E-03    0.518036E-05           4.605
      12   -0.271264E+03    0.546857E-04    0.190703E-05           4.631
      13   -0.271264E+03    0.380528E-05    0.119331E-05           4.650
      14   -0.271264E+03    0.358319E-04    0.305923E-06           4.709
      15   -0.271264E+03    0.219100E-05    0.132870E-06           4.674
      16   -0.271264E+03    0.332000E-05    0.717888E-07           4.701
      17   -0.271264E+03    0.893639E-06    0.320133E-07           4.681
      18   -0.271264E+03    0.416836E-06    0.102409E-07           4.755
      19   -0.271264E+03    0.253078E-06    0.266792E-08           4.728
 ```
The log file stops here...

Kan

vincentm Hello. The new release rescu+ works on niagara for scf calculation now, but when we need to relax the cell, there would be some errors if we execute from rescupy.ase.calculators.rescuplus import Rescuplus. It seems like the rescupy problem.

Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/home/h/hongguo/kan/.local/miniconda3/envs/rescuplus/lib/python3.8/site-packages/rescupy/ase/calculators/rescuplus.py", line 37, in <module>
register_io_formats()
File "/home/h/hongguo/kan/.local/miniconda3/envs/rescuplus/lib/python3.8/site-packages/rescupy/ase/calculators/rescuplus.py", line 22, in register_io_formats
F("rescuplus-in", "rescuplus input file", "1F", module="rescuplus", ext=extin),
File "/home/h/hongguo/kan/.local/miniconda3/envs/rescuplus/lib/python3.8/site-packages/ase/io/formats.py", line 290, in define_io_format
raise ValueError('extension "{}" already registered'.format(ext))
ValueError: extension "rsi" already registered

vincentm

Is it possible that this ASE is from an old installation? In this case, the old ASE Rescuplus class might conflict with the new. I would try to clear ASE and install a new one (from Pypi and without Rescuplus).

Kan

vincentm The rescuplus.py will create a soft link in the ase folder. So this may need to be executed once in vincentm account.

File "/gpfs/fs0/project/h/hongguo/vincentm/rescuf08/88c33acc/venv/lib/python3.8/site-packages/rescupy/ase/calculators/rescuplus.py", line 34,in register_ase_io_rescuplus
os.symlink(rsciopath / "rescuplus.py", aseiopath / "rescuplus.py")
PermissionError: [Errno 13] Permission denied: '/gpfs/fs0/project/h/hongguo/vincentm/rescuf08/88c33acc/venv/lib/python3.8/site-packages/rescupy/ase/io/rescuplus.py' -> '/gpfs/fs0/project/h/hongguo/vincentm/rescuf08/88c33acc/venv/lib/python3.8/site-packages/ase/io/rescuplus.py'

vincentm

Correct. I added write rights to python3.8/site-packages/ase/io. If this doesn't work, the simplest thing might be to create your own venv. Could you give it a shot?

Kan

vincentm Hi! It works. Thanks a lot.

Kan

vincentm
Hi,

I found a warning like this.

To use Cannons algorithm, np_cols must be a multiple of np_rows. Switching to elpa Hermitian and scalapack

So how to set the np_cols and np_rows in the rescuplus input file?

Kan

siaodani

Kan HI, sorry for the late reply.
Could you try setting
solver.eig.reduAlgo = 2

I suppose that this warning shows up when
solver.eig.algo = 'elpa'
solver.eig.reduAlgo= 0 (or 3)

Kan

siaodani Hi, thanks for your reply.
I've test the reduAlgo values. The calc.solver.eig.set_algo("elpa") and calc.solver.eig.set_reduAlgo(3) requires the number of process rows divides the number of process columns, while the number of process rows and columns are related to the grids and can't be set separately.
But we could change the number of process by calc.solver.set_mpi_command("mpiexec -n 121 --oversubscribe"), namely the the square of 11, or some other numbers like 324, 400, and the number of process could be set a little more than the physical cores applied. The calc.solver.eig.set_reduAlgo(3) has fastest computational speed for the large system.

« Previous Page