This thread is locked.Only browsing is available.
Top Page > Browsing
Problem with jx in ver 3.9.1
Date: 2020/01/16 06:25
Name: Zsolt   <gercsiz@tcd.ie>

Dear Developers,

Thank you for the great work you provide to this community.
I very much appreciate it.

I've came across a problem related to jx routine in ver. 3.9.1.
The compilation was with the latest intel compiler - 2020.0.166 .
(Openmx3.9.1 passed the runtest. )

The bulk example for jx (Fe_Bulk_jx.dat) should give
Previous eigenvalue solver = Band
atomnum = 2
ChemP = -0.205912787451 (Hartree)
E_Temp = 300.000000000000 (K)

Jij calculation for a periodic structure
Number of k-grids: 27 27 27
flag_periodic_sum = 0: coupling between site i at cell 0 and site j at cell R
Number of poles of Fermi-Dirac continued fraction (PRB.75.035123): 60

i j c1 c2 c3 J [meV] J [mRy] time_eig [s] ...
---------------------------------------------------------------------------------- ...
1 1 -2 -2 -2 -0.845809571401 -0.062165857439 0.51534 ...
1 1 -2 -2 -1 0.274300677331 0.020160728111 0.00000 ...
1 1 -2 -2 0 0.036006012552 0.002646393135 0.00000 ...
1 1 -2 -2 1 0.274300705154 0.020160730156 0.00000 ...
1 1 -2 -2 2 -0.845809596417 -0.062165859278 0.00000 ...
...
in parallel mode.

However I find that the result depends on the number of cpu used and it is always wrong.
For 6 cpus:

Previous eigenvalue solver = Band
atomnum = 2
ChemP = -0.205912787453 (Hartree)
E_Temp = 300.000000000000 (K)

Jij calculation for a periodic structure
Number of k-grids: 27 27 27
flag_periodic_sum = 0: coupling between site i at cell 0 and site j at cell R
Number of poles of Fermi-Dirac continued fraction (PRB.75.035123): 60

i j c1 c2 c3 J [meV] J [mRy] time_eig [s] time_Jij [s]
-------------------------------------------------------------------------------------------------
1 1 -2 -2 -2 8.678245081649 0.637839254616 3.23103 4.10107
1 1 -2 -2 -1 223.039091480043 16.393071009323 0.00000 4.10567
1 1 -2 -2 0 34753.684204497680 2554.348698335151 0.00000 4.09168
1 1 -2 -2 1 97.403607123658 7.159033143233 0.00000 4.01299
1 1 -2 -2 2 -5.418626443363 -0.398261701434 0.00000 4.06748
1 1 -2 -1 -2 14.103431481762 1.036583103982 0.00000 3.99666
1 1 -2 -1 -1 229.219325498758 16.847309835578 3.02332 3.85880
1 1 -2 -1 0 162208.235362253676 11922.085509518232 0.00000 3.91607
1 1 -2 -1 1 93.499173136160 6.872062535599 0.00000 3.91077
1 1 -2 -1 2 -1.363511443916 -0.100216243591 0.00000 3.92502
1 1 -2 0 -2 13.486008722914 0.991203367804 0.00000 3.89397
1 1 -2 0 -1 226.815840335952 16.670656932797 0.00000 3.88508


For 2 cpus:

Previous eigenvalue solver = Band
atomnum = 2
ChemP = -0.205912787453 (Hartree)
E_Temp = 300.000000000000 (K)

Jij calculation for a periodic structure
Number of k-grids: 27 27 27
flag_periodic_sum = 0: coupling between site i at cell 0 and site j at cell R
Number of poles of Fermi-Dirac continued fraction (PRB.75.035123): 60

i j c1 c2 c3 J [meV] J [mRy] time_eig [s] time_Jij [s]
-------------------------------------------------------------------------------------------------
1 1 -2 -2 -2 471.474735520948 34.652754219939 8.74824 10.80863
1 1 -2 -2 -1 2155.174348775489 158.402394407656 0.00000 11.40773
1 1 -2 -2 0 307385.293306481093 22592.402555789042 0.00000 11.30063
1 1 -2 -2 1 1696.624931437430 124.699633560557 0.00000 11.26372
1 1 -2 -2 2 273.478936648984 20.100331284043 0.00000 11.13096
1 1 -2 -1 -2 477.766147094407 35.115164445825 0.00000 11.06395
1 1 -2 -1 -1 2170.059564944067 159.496437626813 8.49868 10.60332
1 1 -2 -1 0 1447521.639098873828 106390.879104715481 0.00000 10.88354
1 1 -2 -1 1 1684.750838336693 123.826903806862 0.00000 10.93297
1 1 -2 -1 2 274.531764676297 20.177712717485 0.00000 10.94530
...

and so on. It also gives different and wrong results for different k-points used.

I have experienced the same problem on another magnetic system , too.
I wonder if anyone here come across the same issue?

With best wishes,
Zsolt
メンテ
Page: [1]

Re: Problem with jx in ver 3.9.1 ( No.1 )
Date: 2020/01/16 12:54
Name: Po-Hao  <chang.pohao@gmail.com>

Hi,

Is it the compiler issue?
I compiled with compiler/intel/16 and tested on CrBr3. I got the following



1 cpu:
i j c1 c2 c3 J [meV] J [mRy] time_eig [s] time_Jij [s]
-------------------------------------------------------------------------------------------------
19 24 0 0 0 0.017657171539 0.001297778183 152.78443 8.99369
19 22 0 0 0 3.172469987400 0.233172245427 0.00000 9.15662
19 19 1 0 0 0.270400919300 0.019874101180 0.00000 9.17260
19 24 1 0 0 -0.000440496537 -0.000032375899 0.00000 9.16182
19 22 1 0 0 -0.023497741705 -0.001727052176 0.00000 9.10821

6 cpu
i j c1 c2 c3 J [meV] J [mRy] time_eig [s] time_Jij [s]
-------------------------------------------------------------------------------------------------
19 24 0 0 0 0.017657171539 0.001297778183 28.13279 1.82858
19 22 0 0 0 3.172469987400 0.233172245427 0.00000 1.74988
19 19 1 0 0 0.270400919300 0.019874101180 0.00000 1.77849
19 24 1 0 0 -0.000440496537 -0.000032375899 0.00000 1.81862
19 22 1 0 0 -0.023497741705 -0.001727052176 0.00000 1.75223


the speed up is nearly perfect and the results are identical


Po-Hao


メンテ
Re: Problem with jx in ver 3.9.1 ( No.2 )
Date: 2020/01/16 17:38
Name: Asako Terasawa

Hello Zsolt,

And thank Po-Hao for information!

The mpi-dependent problem can also be quite dependent on the computer environment, so I would appreciate if you write down the information about the compiler and libraries you are using.

Thanks,
Asako Terasawa
メンテ
Re: Problem with jx in ver 3.9.1 ( No.3 )
Date: 2020/01/16 21:47
Name: Zsolt  <gercsiz@tcd.ie>


Thanks for both of you for the fast reply!

I use Ubuntu 18.04.3 LTS with gnome 3.28.2 in 64-bit on i7-5820K CPU @ 3.30GHz × 12 .
The compiled jx file was linked with these libraries below.


linux-vdso.so.1 (0x00007ffec4330000)
libmkl_blacs_intelmpi_lp64.so => /media/joy/c6a81621-d733-4e5f-9999-58b0a235b634/opt/intel/compilers_and_libraries_2020.0.166/linux/mkl/lib/intel64_lin/libmkl_blacs_intelmpi_lp64.so (0x00007f89831ae000)
libmkl_scalapack_lp64.so => /media/joy/c6a81621-d733-4e5f-9999-58b0a235b634/opt/intel/compilers_and_libraries_2020.0.166/linux/mkl/lib/intel64_lin/libmkl_scalapack_lp64.so (0x00007f8982890000)
libmkl_intel_lp64.so => /media/joy/c6a81621-d733-4e5f-9999-58b0a235b634/opt/intel/compilers_and_libraries_2020.0.166/linux/mkl/lib/intel64_lin/libmkl_intel_lp64.so (0x00007f8981d24000)
libmkl_core.so => /media/joy/c6a81621-d733-4e5f-9999-58b0a235b634/opt/intel/compilers_and_libraries_2020.0.166/linux/mkl/lib/intel64_lin/libmkl_core.so (0x00007f897da04000)
libmkl_intel_thread.so => /media/joy/c6a81621-d733-4e5f-9999-58b0a235b634/opt/intel/compilers_and_libraries_2020.0.166/linux/mkl/lib/intel64_lin/libmkl_intel_thread.so (0x00007f897b498000)
libpthread.so.0 => /lib/x86_64-linux-gnu/libpthread.so.0 (0x00007f897b279000)
libifcore.so.5 => /media/joy/c6a81621-d733-4e5f-9999-58b0a235b634/opt/intel/compilers_and_libraries_2020.0.166/linux/compiler/lib/intel64_lin/libifcore.so.5 (0x00007f897af15000)
libiomp5.so => /media/joy/c6a81621-d733-4e5f-9999-58b0a235b634/opt/intel/compilers_and_libraries_2020.0.166/linux/compiler/lib/intel64_lin/libiomp5.so (0x00007f897ab25000)
libm.so.6 => /lib/x86_64-linux-gnu/libm.so.6 (0x00007f897a787000)
libmpifort.so.12 => /media/joy/c6a81621-d733-4e5f-9999-58b0a235b634/opt/intel/compilers_and_libraries_2020.0.166/linux/mpi/intel64/lib/libmpifort.so.12 (0x00007f897a3c9000)
libmpi.so.12 => /media/joy/c6a81621-d733-4e5f-9999-58b0a235b634/opt/intel/compilers_and_libraries_2020.0.166/linux/mpi/intel64/lib/release/libmpi.so.12 (0x00007f897932f000)
libdl.so.2 => /lib/x86_64-linux-gnu/libdl.so.2 (0x00007f897912b000)
librt.so.1 => /lib/x86_64-linux-gnu/librt.so.1 (0x00007f8978f23000)
libgcc_s.so.1 => /lib/x86_64-linux-gnu/libgcc_s.so.1 (0x00007f8978d0b000)
libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (0x00007f897891a000)
/lib64/ld-linux-x86-64.so.2 (0x00007f89833f0000)
libimf.so => /media/joy/c6a81621-d733-4e5f-9999-58b0a235b634/opt/intel/compilers_and_libraries_2020.0.166/linux/compiler/lib/intel64_lin/libimf.so (0x00007f897837c000)
libsvml.so => /media/joy/c6a81621-d733-4e5f-9999-58b0a235b634/opt/intel/compilers_and_libraries_2020.0.166/linux/compiler/lib/intel64_lin/libsvml.so (0x00007f89769f5000)
libintlc.so.5 => /media/joy/c6a81621-d733-4e5f-9999-58b0a235b634/opt/intel/compilers_and_libraries_2020.0.166/linux/compiler/lib/intel64_lin/libintlc.so.5 (0x00007f897677e000)
libfabric.so.1 => /media/joy/c6a81621-d733-4e5f-9999-58b0a235b634/opt/intel/compilers_and_libraries_2020.0.166/linux/mpi/intel64/libfabric/lib/libfabric.so.1 (0x00007f8976539000)

Let me know, if you have any suggestion to try.
Best wishes,
Zsolt


メンテ
Re: Problem with jx in ver 3.9.1 ( No.4 )
Date: 2020/01/16 22:02
Name: Zsolt  <gercsiz@tcd.ie>


OK, I think I found the issue.
The optimization had to be reduced to -O1 (for jx only!) and the problem goes away.
(-O2 doesn't work either.)

Thanks again,
Zsolt
メンテ
Re: Problem with jx in ver 3.9.1 ( No.5 )
Date: 2020/01/17 11:29
Name: Asako Terasawa

Thanks, Zsolt!

The optimization option of compilation makes a problem sometimes but it is indeed environment dependent and the reason is not always clear.
I'm still working on further development of jx, and I will also check the issue you mentioned sooner or later.
Thanks a lot for your feedback!

AT
メンテ

Page: [1]