forum@abinit.org
Subject: The ABINIT Users Mailing List ( CLOSED )
List archive
- From: Fabien Bruneval <fabien.bruneval@polytechnique.fr>
- To: forum@abinit.org
- Subject: Re: [abinit-forum] Parallel GW Problem
- Date: Fri, 01 Jun 2007 19:01:58 +0200
Dear Maurice,
The first step of the GW calculation is to output the Kohn-Sham energies and wavefunctions in a so-called KSS file. If you choose kssform=1 (that is the default), the code is going to diagonalize the KS hamiltonian using a direct diagonalization with lapack. In this part, there is only parallelization over k-points and nothing else. As you are using only the gamma point, there is no way to parallelize this part.
You should run the first step in serial.
I have several comment on your input file.
1) try to run a smaller calculation first. GW calculations are extremely heavy. Personnaly, I would be scared by a cell containing 288 atoms and 800 electrons.
2) nbandkss is set to 50, even though your system contains around 400 occupied bands. You should be aware that the GW calculations requires a huge number of empty states.
nbandkss 800 or 1200 would be more reasonable...
Riad Shaltaf and Matteo Giantomasi are currently working on the parallelization over bands. I advise you to use the next version of abinit as soon as it is available...
Regards,
Fabien
Maurice de Koning wrote:
Hi,
I'm trying to carry out a GW calculation for hexagonal ice I, running abinip 5.3.4 on a 16-CPU SGI Altix system. I'm using a large proton-disordered cell containing 96 water molecules, Gamma point only, PBE GGA, Troullier-Martins Pseudopotentials and a large ecut=850 eV. Based on the geometry-optimized structure I am trying to carry out the 3 steps required in the GW calculation. The problem already occurs for the first step, when trying to calculate the Kohn-Sham electronic structure by the diagonalization of the Kohn-Sham Hamiltonian based on the converged density.
The SCF part of the calculation, parallelized over npbands*npftt=8*2=16,
seems OK, but the problems seem to
start when abinip starts writing the _DEN file, after which all hell breaks loose:
*************************************************************************************************
At SCF step 20, etot is converged :
for the second time, diff in etot= 1.632E-06 < toldfe= 1.000E-05
Cartesian components of stress tensor (hartree/bohr^3)
sigma(1 1)= 1.48708394E-04 sigma(3 2)= 6.91251884E-08
sigma(2 2)= 1.50207276E-04 sigma(3 1)= -1.23849931E-07
sigma(3 3)= 1.50516609E-04 sigma(2 1)= 3.56082287E-08
ioarr: writing density data
ioarr: file name is IhGWo_DS1_DEN
*** glibc detected *** malloc(): memory corruption: 0x6000000004f71ab0 ***
MPI: On host opala, Program
/home/esmeralda/dekoning/dekoning/abinit/abinit-5.3.4/src/main/abinip, Rank
3, Process 21986 received signal SIGSEGV(11)
MPI: --------stack traceback-------
*** glibc detected *** double free or corruption (out): 0x6000000004f67730 ***
*** glibc detected *** free(): invalid next size (normal): 0x6000000004f20ec0
***
*** glibc detected *** free(): invalid next size (normal): 0x6000000004f1dc90
***
*** glibc detected *** free(): invalid next size (normal): 0x6000000004f6a3e0
***
*** glibc detected *** double free or corruption (!prev): 0x6000000004f6f740
***
*** glibc detected *** free(): invalid next size (normal): 0x6000000004fb6b30
***
*** glibc detected *** double free or corruption (out): 0x6000000005004830 ***
*** glibc detected *** double free or corruption (!prev): 0x6000000004fbdfc0
***
*** glibc detected *** double free or corruption (!prev): 0x600000000500c840
***
ioarr: data written to disk file IhGWo_DS1_DEN
bonds_lgth_angles : about to open file IhGWo_DS1_GEO
======================================================================
Calculating and writing out Kohn-Sham electronic Structure file
Using diagonalized wavefunctions and energies (kssform=1)
Additional amount of memory required by "outkss" routine= 1615.00 Mbytes.
Since the number of g's to be written on file
was 0 or too large, it has been set to the max. value.,
computed from the union of the sets of G vectors for the different k-points.
Number of g-vectors written on file is: 10068
Since the number of bands to be computed was (-1) or
too large, it has been set to the max. value. allowed for each k,
thus, the minimum of the number of plane waves for each k point.
Number of bands written on file is: 10068
Since the number of bands to be computed
is equal to the nb of G-vectors found for each k-pt,
the program will perform complete diagonalizations.
Amount of disk space required by _STA file= 1547.12 Mbytes.
Sorting g-vecs for an output of states on an unique "big" PW basis.
The inversion was not found in the symmetries list.
- outkss - COMMENT :
GW program uses the original set of symmetries
Shells found:
number of shell number of G vectors cut-off energy
1 1 0.024
2 2 0.024
3 3 0.051
4 4 0.051
5 5 0.056
6 6 0.056
7 7 0.056
8 8 0.056
9 9 0.083
10 10 0.083
11 11 0.083
12 12 0.083
13 13 0.132
14 14 0.132
15 15 0.152
16 16 0.152
.......
**** 10048 31.177
**** 10049 31.180
**** 10050 31.180
**** 10051 31.180
**** 10052 31.180
**** 10053 31.182
**** 10054 31.182
**** 10055 31.182
**** 10056 31.182
**** 10057 31.183
**** 10058 31.183
**** 10059 31.183
**** 10060 31.183
**** 10061 31.208
**** 10062 31.208
**** 10063 31.208
**** 10064 31.208
**** 10065 31.210
**** 10066 31.210
**** 10067 31.210
**** 10068 31.210
Opening file for KS structure output: IhGWo_DS1_KSS
number of Gamma centered plane waves 10068
number of Gamma centered shells 10068
number of bands 10068
maximum angular momentum components 4
number of symmetry operations 1 (without inversion)
* g( 1)= 0 1 0
* g( 2)= 0 -1 0
* g( 3)= 0 1 1
* g( 4)= 0 -1 1
* g( 5)= 1 1 0
* g( 6)= -1 1 0
* g( 7)= 1 -1 0
* g( 8)= -1 -1 0
* g(****)=-25 21 5
* g(****)= 25-21 5
* g(****)=-25-21 5
* g(****)= 16 31 1
* g(****)=-16 31 1
* g(****)= 16-31 1
* g(****)=-16-31 1
-P-0000
-P-0000 k-point 1
*** glibc detected *** double free or corruption (!prev): 0x6000000004f5cf90
***
*** glibc detected *** free(): invalid pointer: 0x6000000004f64fa0 ***
*** glibc detected *** double free or corruption (!prev): 0x6000000004f582b0
***
*** glibc detected *** free(): invalid pointer: 0x6000000004f602c0 ***
*** glibc detected *** malloc(): memory corruption: 0x6000000004f582b0 ***
*** glibc detected *** double free or corruption (!prev): 0x6000000004f67730
***
*** glibc detected *** free(): invalid pointer: 0x6000000004f6f740 ***
*** glibc detected *** double free or corruption (!prev): 0x6000000004f623d0
***
*** glibc detected *** free(): invalid pointer: 0x6000000004f6a3e0 ***
-P-0000 Calculating <G|H|G'> elements
*** glibc detected *** double free or corruption (!prev): 0x6000000004fa28f0
***
*** glibc detected *** free(): invalid pointer: 0x6000000004faa900 ***
*** glibc detected *** double free or corruption (!prev): 0x6000000004fa9080
***
*** glibc detected *** free(): invalid pointer: 0x6000000004fb1090 ***
*** glibc detected *** double free or corruption (out): 0x6000000004f682d0 ***
*** glibc detected *** double free or corruption (out): 0x6000000004f6cfb0 ***
*** glibc detected *** double free or corruption (!prev): 0x6000000004fecf30
***
*** glibc detected *** free(): invalid pointer: 0x6000000004ff4f40 ***
*** glibc detected *** double free or corruption (!prev): 0x6000000004ff5170
***
*** glibc detected *** free(): invalid pointer: 0x6000000004ffd180 ***
*** glibc detected *** double free or corruption (out): 0x6000000004fb2910 ***
*** glibc detected *** double free or corruption (!prev): 0x6000000005037570
***
*** glibc detected *** free(): invalid pointer: 0x600000000503f580 ***
******************************************************************************************************************
Here is the input file:
----------------------------------------------------------------------------------------------------------------------------------------------------------------------
#-----------------------------------------------------------------
# Ice PBE GGA
#-----------------------------------------------------------------
#Definition of the unit cell
acell 13.1493536801862057 15.1976493759670106 14.3247149284289197 angstrom
#Definition of the planewave basis set
ecut 850 eV # Maximal kinetic energy cut-off, eV
#Definition of k-points
nkpt 1
ndtset 1
# Definition of parameters for the calculation of the KSS file
nbandkss 50 # Number of bands in KSS file (-1 means the maximum
possible)
symmorphi 0 # No symmorphic symmetry operations
nsym 0
# Definition of the SCF procedure
nstep 30 # Maximal number of SCF cycles
toldfe 1.0d-5 # Will stop when this tolerance is achieved on total energy
diemac 12.0
#Parallelization specifications
npband 8 npfft 2
timopt -1
fftalg 401
nloalg=4
wfoptalg 4
fft_opt_lob 2
iprcch 0
intxc 0
istwfk 1 #Do not take advantage of time-reversal symmetry
#Exchange-correlation functional
ixc=11 #PBE
#Electron bands
nband 400
#Output options
enunit 2 #Print energies in Hartrees and eV
prtden 1
prtwf 1
prtgeo 1
chkexit 1 #Check for soft exit
#Definition of the atoms
natom 288
ntypat 2
typat 96*2 192*1
znucl 1 8
xred 8.4312914696E-02 8.3660261848E-02 9.3809915685E-02
2.5074724775E-01 1.6659125325E-01 1.5681072076E-01
8.3324114394E-02 8.3419501461E-02 4.0579067959E-01
2.5013948449E-01 1.6724399173E-01 3.4402885286E-01
2.5040058344E-01 3.3278509043E-01 9.4311450028E-02
8.3103792575E-02 4.1649451464E-01 1.5571120920E-01
......
---------------------------------------------------------------------------------------------------------------------------------
Any idea what may be the problem?
Thanks,
Maurice
- Parallel GW Problem, Maurice de Koning, 06/01/2007
- Re: [abinit-forum] Parallel GW Problem, Fabien Bruneval, 06/01/2007
- Parallelization over bands/fft, Maurice de Koning, 06/01/2007
- Re: [abinit-forum] Parallelization over bands/fft, gilles zerah, 06/02/2007
- Re: [abinit-forum] Parallelization over bands/fft, Maurice de Koning, 06/04/2007
- Re: [abinit-forum] Parallelization over bands/fft, Gilles Zerah, 06/04/2007
- Re: [abinit-forum] Parallelization over bands/fft, Xavier Gonze, 06/04/2007
- Re: [abinit-forum] Parallelization over bands/fft, gilles zerah, 06/04/2007
- Re: [abinit-forum] Parallelization over bands/fft, Gilles Zerah, 06/04/2007
- Re: [abinit-forum] Parallelization over bands/fft, Maurice de Koning, 06/04/2007
- Re: [abinit-forum] Parallelization over bands/fft, gilles zerah, 06/02/2007
- Parallelization over bands/fft, Maurice de Koning, 06/01/2007
- Re: [abinit-forum] Parallel GW Problem, Deyu Lu, 06/01/2007
- Re: [abinit-forum] Parallel GW Problem, Fabien Bruneval, 06/01/2007
Archive powered by MHonArc 2.6.16.