Hi all!
I had some trouble implementing the grid calculation for phonon
calculation, which splits the calculation over q and representations.
I have spend last few days try to do the splitting following the
run_example script in GRID_example folder.
I am running the QE V4.3 on HPC machine
(IBM Blade Center Linux Cluster, henry2, 981 dual Xeon compute nodes with
Intel Xeon Processors (mix of single-, dual-, quad-, and six- core), 2-4GB
per core distributed memory, dual gigabit Ethernet interconnects)
8-16 cpus are assigned to each splitted jobs
After some calculations, I found that the splitting is actually not
feasible on HPC!
Before you can split the job over q points and representations, you have to
copy PREFIX.save (which is generated by scf calculation) to each sub
directories where you want to do the ph calculation separately. If the
system is slightly large (more 20 atoms in unit cell), the PREFIX.save will
be several GBs, then let's say if you want to split over 3 q points and 48
representations, you have to copy the PREFIX.save files 3x48=144 times!
(144x 3GB + 144 X 5GB > 1TB)
I had a script doing this job, but it quickly used up the disk space (it
used more 1TB )! which made this approach unusable on HPC machine!
Is there anyway to do the splitting phonon calculation without copying the
PREFIX.save generated by scf to each sub directories?
Can we let all the splitted phonon calculations look for the same input
directory for PREFIX.save so that we do not need to copy files around?
Any comment on how to do the splitting job without using up the disk space
will be greatly appreciated!
Thank you,
Rui Mao
Part of the script that copy the PREFIX.save files generated by scf to each
sub directories:
mkdir $TMP_DIR/$q.$irr
cp -r $TMP_DIR/$PREFIX.save $TMP_DIR/$q.$irr
mkdir -p $TMP_DIR/$q.$irr/_ph0/$PREFIX.phsave
cp -r $TMP_DIR/_ph0/$PREFIX.phsave/* $TMP_DIR/$q.$irr/_ph0/$PREFIX.phsave
Scf putfile:
&control
calculation = 'scf',
restart_mode='from_scratch',
wf_collect=.true.,
prefix='graphite',
tstress = .true.,
tprnfor = .true.,
forc_conv_thr = 1.0D-3,
etot_conv_thr = 1.0D-4,
pseudo_dir = '/home/rmao/Pseudo/',
outdir='/share2/Ray/grid_graphite/scf',
/
&system
ibrav=4,
celldm(1) =9.2961,
celldm(3)=1.3579,
nat=16,
ntyp= 1,
ecutwfc =70,
ecutrho=700,
occupations='smearing', smearing='mp', degauss=0.003,
/
&electrons
diagonalization='cg',
mixing_mode = 'plain',
mixing_beta = 0.5,
conv_thr = 1.0d-8,
/
ATOMIC_SPECIES
C 12.01070 C.pz-rrkjus.UPF
ATOMIC_POSITIONS {angstrom}
C 1.183416184 0.709967199 -3.339999736
C 1.183397032 2.130056591 -3.339999831
C 2.413378375 2.839955863 -3.339999742
C -0.046321206 2.840033775 -3.339999742
C 2.413196667 4.260044397 -3.339999829
C -0.046512842 4.260108061 -3.339999831
C 1.183461168 4.970022576 -3.339999741
C 3.643012948 2.129829924 -3.339999823
C 0.000008073 -0.000036198 -0.000000172
C -0.000009822 1.420051269 -0.000000260
C 1.229970973 2.129951526 -0.000000171
C -1.229729485 2.130029208 -0.000000172
C 1.229791025 3.550039979 -0.000000260
C -1.229918926 3.550102386 -0.000000260
C 0.000053606 4.260017465 -0.000000171
C 2.459606230 1.419825978 -0.000000260
K_POINTS {automatic}
8 8 16 0 0 0
Ph inputfile
phonons of Graphite_2x2
&inputph
tr2_ph=1.0d-16,
alpha_mix(1)=0.5,
prefix='graphite',
ldisp=.true.,
epsil=.false.,
recover=.false.,
reduce_io=.false.,
amass(1)=12.01070,
start_irr=0,
last_irr=0,
nq1=1, nq2=1, nq3=4,
outdir='/share2/Ray/grid_graphite/scf',
fildyn='graphite.dyn'
/
--
Rui Mao
============================================================
Department of Electrical and Computer Engineering (ECE)
North Carolina State University (NCSU)
Raleigh, NC, 27606
Email: rmao at ncsu.edu
Email: ruimao20 at gmail.com
============================================================
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
http://www.democritos.it/pipermail/pw_forum/attachments/20120515/d44aa498/attachment.htm
