Inconsistent GW quasiparticle energies in v4.0.1

[fabio.caruso]

Dear Yambo developers,

I would like to report an issue with the latest release (v4.0.1).
To test the new version, I have run a small calculation to compare the parallel and serial performance of Yambo for the QP energies of Silicon.
The calculations are not supposed to be converged, thus I have only used 8 bands (4 occ, 4 empty), 6x6x6 k-point mesh (centered at Gamma), and small cutoffs
In particular, I have set:
EXXRLvcs= 1 RL
NGsBlkXp= 1 RL

All input/output files are attached.
The calculations with 4 CPUs (input file: yambo.4cpu.in; output files: o-4CPU.qp, r-4CPU_em1d_ppa_HF_and_locXC_gw0) and only 1 CPU (input file: yambo.1cpu.in; output files: o-1CPU.qp, r-1CPU_em1d_ppa_HF_and_locXC_gw0) give different quasiparticle energies. Both exchange and correlation energy seem to differ.
The calculations have been run with commands:

mpirun -np 4 yambo -I ../ -F yambo.4cpu.in -J 4CPU
yambo -I ../ -F yambo.1cpu.in -J 1CPU

Is there anything wrong with my input files?

Thank you in advance for your help.

Best,
Fabio Caruso

[Daniele Varsano]

Dear Fabio,
that's sound really strange, and something looks to fail in your parallel run.
Did you compile the code using OpenMPI, if so we experienced some problem that are solved by using the flag:

Code: Select all

--enable-openmpi 

when configuring the code.
If this is not the case, can you also post your config.log file?

Best,
Daniele

[fabio.caruso]

Hi Daniele,
thanks for the quick reply. No, I am not using OpenMPI. I configured using the following command:

./configure \
F77=ifort \
FC=ifort \
FCFLAGS="-O3 -xW -assume bscc -nofor_main" \
--with-fft-path="/usr/lib" \
--with-iotk-path="/home/caruso/QE/espresso-5.0.2/iotk" \
--with-blas-libs="-L/opt/intel/composer_xe_2013_sp1.1.106/mkl/lib/intel64/ -lmkl_core -lmkl_intel_lp64 -lmkl_sequential" \
--with-lapack-libs="-L/opt/intel/composer_xe_2013_sp1.1.106/mkl/lib/intel64/ -lmkl_core -lmkl_intel_lp64 -lmkl_sequential" \
--with-blacs-libs="-L/opt/intel/composer_xe_2013_sp1.1.106/mkl/lib/intel64/ -lmkl_core -lmkl_intel_lp64 -lmkl_sequential" \
--with-scalapack-libs="-L/opt/intel/composer_xe_2013_sp1.1.106/mkl/lib/intel64/ -lmkl_core -lmkl_intel_lp64 -lmkl_sequential"

This is the message that I got at the end of the ./configure.


# [VER] 4.0.1 r.88
#
# - GENERAL CONFIGURATIONS -
#
# [SYS] linux@x86_64
# [SRC] /home/caruso/QE/espresso-5.0.2/yambo-4.0.1-rev.89
# [BIN] /home/caruso/QE/espresso-5.0.2/yambo-4.0.1-rev.89/bin
# [-] Double precision
# [X] Redundant compilation
# [-] Run-Time timing profile
#
# - PARALLEL SUPPORT -
#
# [X] MPI (not open-mpi kind)
# [-] OpenMP
# [-] Blue-Gene specific procedures
#
# - LIBRARIES (E=external library; I=internal library; -=not used;) -
#
# I/O
# [ E ] IOTK : /home/caruso/QE/espresso-5.0.2/iotk/src/libiotk.a (QE 5.0)
# [ - ] ETSF_IO:
# [ I ] NETCDF : -lnetcdff -lnetcdf (No large files support)
# [ - ] HDF5 :
#
# MATH
# [ I ] FFT : Internal Goedecker FFT with 0 cache
# [ E ] BLAS : -L/opt/intel/composer_xe_2013_sp1.1.106/mkl/lib/intel64/ -lmkl_core -lmkl_intel_lp64 -lmkl_sequential
# [ E ] LAPACK : -L/opt/intel/composer_xe_2013_sp1.1.106/mkl/lib/intel64/ -lmkl_core -lmkl_intel_lp64 -lmkl_sequential
# [ - ] SCALAPACK: -L/opt/intel/composer_xe_2013_sp1.1.106/mkl/lib/intel64/ -lmkl_core -lmkl_intel_lp64 -lmkl_sequential
#
# OTHER
# [ I ] LibXC : -lxc
# [ - ] MPI library:
#
# - COMPILERS, MAKE and EDITOR -
#
# [ CPP ] gcc -E -P -D_NETCDF_IO -D_MPI -D_FFTSG
# [ C ] gcc -g -O2 -D_C_US -D_FORTRAN_US
# [MPICC] mpicc -g -O2 -D_C_US -D_FORTRAN_US
# [ F90 ] ifort -O3 -xW -assume bscc -nofor_main
# [MPIF ] mpif90 -O3 -xW -assume bscc -nofor_main
# [ F77 ] ifort -O3 -xW -assume bscc -nofor_main
# [Cmain] -nofor_main
# [NoOpt] -assume bscc -g -O0
#
# [ MAKE ] make
# [EDITOR] vim

I also attached the full config.log, which might contain some relevant information that I might have overlooked.

Thanks!

Best,
Fabio

[fabio.caruso]

Dear Daniele,

a quick update. I have reconfigured and recompiled yambo with the flag --enable-openmpi. Now the calculations with 1 and 4 CPU are in good agreement (discrepancies are of the order of ~1meV or smaller). Now, at the end of the configure summary I get:

(...)
# [X] MPI (open-mpi kind)
(...)

(Before I was getting # [X] MPI (not open-mpi kind)).
This seems to solve the issue for now.
Many thanks for your help.

Fabio