Michael, I suppose you are running the example without changing the fdf file. I have just run th example using the serial version compiled with gfortran, and had no such errors. If the fdf is unmodified, which compiler are you using, intel? Version? In any case, avoid using -O3, it can give you wrong numerical results. Stick to -O2, it's on the safe side and sometimes it works faster than -O3.
Marcos On Mon, Mar 7, 2011 at 5:18 PM, Michael Shin <[email protected]> wrote: > Dear All Thanks for your help. > > How to solve the following WARNING > Siesta Version: siesta-3.0-rc2 > Architecture : x86_64-unknown-linux-gnu--unknown > Compiler flags: mpif90 -g -O3 > PARALLEL version > > * Running on 4 nodes in parallel > >> Start of run: 2-MAR-2011 10:58:02 > > *********************** > * WELCOME TO SIESTA * > *********************** > > reinit: Reading from standard input > ************************** Dump of input data file > **************************** > # $Id: Fe.fdf,v 1.1 1999/04/20 12:52:43 emilio Exp $ > # > ----------------------------------------------------------------------------- > # FDF for bcc iron > # > # GGA, Ferromagnetic. > # Scalar-relativistic pseudopotential with non-linear partial-core > correction > # > # E. Artacho, April 1999 > # > ----------------------------------------------------------------------------- > SystemName bcc Fe ferro GGA # Descriptive name of the system > SystemLabel Fe # Short name for naming files > # Output options > WriteCoorStep > WriteMullikenPop 1 > WriteXML F > # Species and atoms > NumberOfSpecies 1 > NumberOfAtoms 1 > %block ChemicalSpeciesLabel > 1 26 Fe > %endblock ChemicalSpeciesLabel > # Basis > PAO.EnergyShift 50 meV > PAO.BasisSize DZP > %block PAO.Basis > Fe 2 > 0 2 P > 6. 0. > 2 2 > 0. 0. > %endblock PAO.Basis > LatticeConstant 2.87 Ang > %block LatticeVectors > 0.50000 0.500000 0.500000 > 0.50000 -0.500000 0.500000 > 0.50000 0.500000 -0.500000 > %endblock LatticeVectors > KgridCutoff 15. Ang > %block BandLines > 1 0.00000 0.000000 0.000000 \Gamma > 40 2.00000 0.000000 0.000000 H > 28 1.00000 1.000000 0.000000 N > 28 0.00000 0.000000 0.000000 \Gamma > 34 1.00000 1.000000 1.000000 P > %endblock BandLines > xc.functional GGA # Exchange-correlation functional > xc.authors PBE # Exchange-correlation version > SpinPolarized true # Logical parameters are: yes or no > MeshCutoff 150. Ry # Mesh cutoff. real space mesh > # SCF options > MaxSCFIterations 40 # Maximum number of SCF iter > DM.MixingWeight 0.1 # New DM amount for next SCF cycle > DM.Tolerance 1.d-3 # Tolerance in maximum difference > # between input and output DM > DM.UseSaveDM true # to use continuation files > DM.NumberPulay 3 > SolutionMethod diagon # OrderN or Diagon > ElectronicTemperature 25 meV # Temp. for Fermi smearing > # MD options > MD.TypeOfRun cg # Type of dynamics: > MD.NumCGsteps 0 # Number of CG steps for > # coordinate optimization > MD.MaxCGDispl 0.1 Ang # Maximum atomic displacement > # in one CG step (Bohr) > MD.MaxForceTol 0.04 eV/Ang # Tolerance in the maximum > # atomic force (Ry/Bohr) > # Atomic coordinates > AtomicCoordinatesFormat Fractional > %block AtomicCoordinatesAndAtomicSpecies > 0.000000000000 0.000000000000 0.000000000000 1 > %endblock AtomicCoordinatesAndAtomicSpecies > ************************** End of input data file > ***************************** > > reinit: > ----------------------------------------------------------------------- > reinit: System Name: bcc Fe ferro GGA > reinit: > ----------------------------------------------------------------------- > reinit: System Label: > Fe > reinit: > ----------------------------------------------------------------------- > > initatom: Reading input for the pseudopotentials and atomic orbitals > ---------- > Species number: 1 Label: Fe Atomic number: 26 > Ground state valence configuration: 4s02 3d06 > Reading pseudopotential information in formatted form from Fe.psf > > Pseudopotential generated from a relativistic atomic calculation > There are spin-orbit pseudopotentials available > Spin-orbit interaction is not included in this calculation > > Valence configuration for pseudopotential generation: > 4s( 2.00) rc: 2.00 > 4p( 0.00) rc: 2.00 > 3d( 6.00) rc: 2.00 > 4f( 0.00) rc: 2.00 > For Fe, standard SIESTA heuristics set lmxkb to 3 > (one more than the basis l, including polarization orbitals). > Use PS.lmax or PS.KBprojectors blocks to override. > Warning: Empty PAO shell. l = 1 > Will have a KB projector anyway... > > <basis_specs> > > =============================================================================== > Fe Z= 26 Mass= 55.850 Charge= 0.17977+309 > Lmxo=2 Lmxkb=3 BasisType=split Semic=F > L=0 Nsemic=0 Cnfigmx=4 > n=1 nzeta=2 polorb=1 > splnorm: 0.15000 > vcte: 0.0000 > rinn: 0.0000 > rcs: 6.0000 0.0000 > lambdas: 1.0000 1.0000 > L=1 Nsemic=0 Cnfigmx=4 > L=2 Nsemic=0 Cnfigmx=3 > n=1 nzeta=2 polorb=0 > splnorm: 0.15000 > vcte: 0.0000 > rinn: 0.0000 > rcs: 0.0000 0.0000 > lambdas: 1.0000 1.0000 > > ------------------------------------------------------------------------------- > L=0 Nkbl=1 erefs: 0.17977+309 > L=1 Nkbl=1 erefs: 0.17977+309 > L=2 Nkbl=1 erefs: 0.17977+309 > L=3 Nkbl=1 erefs: 0.17977+309 > > =============================================================================== > </basis_specs> > > atom: Called for Fe (Z = 26) > > read_vps: Pseudopotential generation method: > read_vps: ATM3 Troullier-Martins > Total valence charge: 8.00000 > > read_vps: Pseudopotential includes a core correction: > read_vps: Pseudo-core for xc-correction > > xc_check: Exchange-correlation functional: > xc_check: GGA Perdew, Burke & Ernzerhof 1996 > V l=0 = -2*Zval/r beyond r= 2.7645 > V l=1 = -2*Zval/r beyond r= 2.7645 > V l=2 = -2*Zval/r beyond r= 2.7645 > V l=3 = -2*Zval/r beyond r= 2.7645 > All V_l potentials equal beyond r= 1.9726 > This should be close to max(r_c) in ps generation > All pots = -2*Zval/r beyond r= 2.7645 > Using large-core scheme for Vlocal > > atom: Estimated core radius 2.76453 > atom: Maximum radius for 4*pi*r*r*local-pseudopot. charge 3.05528 > atom: Maximum radius for r*vlocal+2*Zval: 2.79930 > GHOST: No ghost state for L = 0 > GHOST: No ghost state for L = 1 > GHOST: No ghost state for L = 2 > GHOST: No ghost state for L = 3 > > KBgen: Kleinman-Bylander projectors: > l= 0 rc= 2.047986 el= -0.388305 Ekb= 4.259322 kbcos= 0.262992 > l= 1 rc= 2.047986 el= -0.097543 Ekb= 2.850785 kbcos= 0.194191 > l= 2 rc= 2.022544 el= -0.553241 Ekb=-12.567334 kbcos= -0.683368 > l= 3 rc= 2.047986 el= 0.003178 Ekb= -1.649997 kbcos= -0.006611 > > KBgen: Total number of Kleinman-Bylander projectors: 16 > atom: > ------------------------------------------------------------------------- > > atom: SANKEY-TYPE ORBITALS: > atom: Selected multiple-zeta basis: split > > SPLIT: Orbitals with angular momentum L= 0 > > SPLIT: Basis orbitals for state 4s > > izeta = 1 > lambda = 1.000000 > rc = 6.000769 > energy = -0.359899 > kinetic = 0.368794 > potential(screened) = -0.728693 > potential(ionic) = -6.200046 > *WARNING: Minimum split_norm parameter: 0.52689. Will not be able to > generate orbital with split_norm = 0.15000 > See manual for new split options > ERROR STOP from Node: 0 > * > --- On *Tue, 3/1/11, Marcos Veríssimo Alves < > [email protected]>* wrote: > > > From: Marcos Veríssimo Alves <[email protected]> > > Subject: Re: [SIESTA-L] Parallel_version > To: [email protected] > Date: Tuesday, March 1, 2011, 3:50 PM > > > Michael, > > I'm pretty sure this issue has already been discussed before on the list, > but here goes the short version of the story. > > Siesta needs infiniband or myrinet (in general, a low latency internode > connection) to work properly in parallel, otherwise you should try to run it > in parallel only on one node with many cores, using mpi. If you have to use > many nodes, Gigabit interconnection won't do the job. Even in the case of > mpi on only one node with many cores, if your system is too small, > parallelism will not work efficiently. > > No modifications in the fdf file are needed to run siesta in parallel, > except for the Parallel Over Orbitals, and Parallel Over K modes. Check the > manual and the list archives for that. > > Marcos > > On Tue, Mar 1, 2011 at 6:32 PM, Michael Shin > <[email protected]<http://us.mc1619.mail.yahoo.com/mc/[email protected]> > > wrote: > > Hello out Huan, > It works and I successfully compiled parallel SIESTA. However, When I run > the example of Fe, I found the computational time increases with increasing > the number of nodes. For example with 1 node it takes about 3 mints, and > with 2 nodes it takes about 4 mints. > But with 4 nodes, it took about 30 mints. > See the attached arch,make file, and two output files. > Do I have to make any special changes in FDF file when I want to run > parallel calculations. > Any help from any one is welcome. > > > > --- On *Sat, 2/26/11, Huan Tran > <[email protected]<http://us.mc1619.mail.yahoo.com/mc/[email protected]> > >* wrote: > > > From: Huan Tran > <[email protected]<http://us.mc1619.mail.yahoo.com/mc/[email protected]> > > > Subject: Re: [SIESTA-L] Parallel_version > To: > [email protected]<http://us.mc1619.mail.yahoo.com/mc/[email protected]> > Date: Saturday, February 26, 2011, 11:10 AM > > > For those who need to install parallel siesta, here is something I found on > the web. For me, it works. > > Installation of SIESTA-3.0-b(parallel version): > > 1. Installation of MPICH2: > > a) to get mpich2 > > http://www-unix.mcs.anl.gov/mpi/mpich/ > > > b) installation instruction: > > http://hydra.nac.uci.edu/~sev/docs/mpich2-readme.txt > > > > > > $ tar xfz mpich2.tar.gz > > $ cd mpich2-1.0.1 > $ ./configure # default installation directory (/urs/local/bin/) > $ make > $ make install ( need to be a root) > > > $ which mpd # (print /usr/local/bin ) > $ which mpiexec #(print /usr/local/bin/) > > $ which mpirun #(print /usr/local/bin/) > > $ cd #(go to home directory) > $ vi .mpd.conf #(type secretword = what ever you want and save) > > > > > $ chmod 600 .mpd.conf > > $ mpd $ > $ mpdtrace (print the localhost name) > > $ mpdallexit > > $ vi hostfile (type localhost X no, X = no of core of that machine ) > > > 2. BLAS installation: > > > > > to get BLAS > > http://www.netlib.org/blas/blas.tgz > > > for installation > > http://wiki.ifca.es/e-ciencia/index.php/BLAS > > > > > or > $ tar -zxvf blas.tgz > $ cd BLAS > $ vi make.inc (need to edit ---see below) > > $ make > > edit > > FORTRAN = /usr/local/bin/mpif90 > LOADER = /usr/lcoal/bin/mpif90 > > > > > > $ make clean (for cleaning) > > 3. lapack installation: > > to get > http://www.netlib.org/blas/lapack.tgz > > > > > > > for installation > go to > http://wiki.ifca.es/e-ciencia/index.php/LAPACK > or > $ tar -zxvf lapack.tgz > $ cd lapack-3.2/ > > > > > $ cp make.inc.example make.inc > $ vi make.inc (need to edit ...see below) > $ make all > > edit > > FORTRAN = /usr/local/bin/mpif90 > LOADER = /usr/local/bin/mpif90 > TIMER = EXT_ETIME > > > > > BLASLIB = /home/../../BLAS/blas$(PLAT).a (give the full path) > > > for cleaning > > $ rm *.a #(make clean is not going to work) > > > 4. BLACS installation: > > to get > > > > > > > http://www.netlib.org/blacs/mpiblacs.tgz > http://www.netlib.org/blacs/mpiblacs-patch03.tgz > > > > > > for installation > > http://wiki.ifca.es/e-ciencia/index.php/BLACS > or > > $ tar -zxvf mpiblacs.tgz > > > > > $ tar -zxvf mpiblacs-patch03.tgz > > $ cd BLACS/ > $ cp BMAKES/Bmake.MPI-LINUX ./Bmake.inc > $ vi Bmake.inci (need to edit .....see below) > $ make mpi > > edit > > BTOPdir = $(HOME)/Software/BLACS # directory we're compiling BLACS at > > > > > MPIdir = /usr/local/ # Open MPI directory > MPILIBdir = > MPIINCdir = /usr/local/include/ > MPILIB = /usr/local/lib/libmpich.a > SYSINC = > INTFACE = -DAdd_ > TRANSCOMM = -DUseMpich > > > > > F77 = /usr/local/bin/mpif90 # MPI wrapper for Fortran compiler > CC = /usr/local/bin/mpicc # MPI wrapper for C compiler > CCFLAGS = -O3 > > > for cleaning > > $ find ./* -name '*.o' | xargs rm > > > > > > > 5. Scalapack installation: > > to get > > http://www.netlib.org/scalapack/scalapack.tgz > > for installation > > > > > > http://wiki.ifca.es/e-ciencia/index.php/ScaLAPACK > or > > $ tar -zxvf TARs/scalapack.tgz > $ cd scalapack-1.8.0/ > > > > > $ cp SLmake.inc.example SLmake.inc > > $ vi SLmake.inc (need to edit ...see below) > $ make > > for cleaning > > $ make clean > $ rm libscalapack.a > > > > edit > > > home = $(HOME)/Software/scalapack-1.8.0 > > > > > BLACSdir = $(HOME)/Software/BLACS/LIB > > SMPLIB = /usr/local/bin/libmpich.a > BLACSFINIT = $(BLACSdir)/blacsF77init_MPI-$(PLAT)-$(BLACSDBGLVL).a > BLACSCINIT = $(BLACSdir)/blacsCinit_MPI-$(PLAT)-$(BLACSDBGLVL).a > > > > > BLACSLIB = $(BLACSdir)/blacs_MPI-$(PLAT)-$(BLACSDBGLVL).a > > F77 = /usr/local/bin/mpif90 > CC = /usr/local/bin/mpicc > F77FLAGS = -O2 $(NOOPT) > > > > > CCFLAGS = > > CDEFS = -DAdd_ -DNO_IEEE $(USEMPI) > BLASLIB = $(HOME)/Software/BLAS/blas_LINUX.a > LAPACKLIB = $(HOME)/Software/lapack-3.2/lapack_LINUX.a > > > > > > > 6. Siesta installation: > > > go to > http://wiki.ifca.es/e-ciencia/index.php/SIESTA#Serial_version > > > > > > 1. $ tar -zxvf siesta-3.0-b.tgz > 2. $ cd siesta-3.0-b/Obj/ > > 3. $ sh ../Src/obj_setup.sh > 4. $ ../Src/configure > 5. $ vi arch.make (need to edit this file .... see the web page...as > > > > > it depends on where did you install the other libraries.... ... ) > 6. $ make ( this is for siesta, for transiesta see below ) > > 7 $ make clean (to clean the *.o file and recompile again > > > > > to compile transiesta type make transiesta instead of make) > > > edit > FC= /usr/local/bin/mpif9 > FFLAGS=-g -O2 > FPPFLAGS= -DMPI -DFC_HAVE_FLUSH -DFC_HAVE_ABORT > > BLAS_LIBS=$(HOME)/../BLAS/blas_LINUX.a > > > > > LAPACK_LIBS=$(HOME)/../lapack-3.2/lapack_LINUX.a > BLACS_LIBS=$(HOME)/../BLACS/LIB/blacsF77init_MPI-LINUX-0.a \ > $(HOME)/../BLACS/LIB/blacs_MPI-LINUX-0.a \ > > $(HOME)/../BLACS/LIB/blacsCinit_MPI-LINUX-0.a > > > > > SCALAPACK_LIBS=$(HOME)/../scalapack-1.8.0/libscalapack.a > > COMP_LIBS=dc_lapack.a > MPI_INTERFACE=libmpi_f90.a > MPI_INCLUDE=. > > 7. To run siesta from any directory > > > $ cd yourdirectory > $ ln -s ../path to siesta/Obj/siesta . ( to link the siesta ) > $ mpdboot -n 1 -f hostfile > $ mpiexec -n <np> ./siesta < xyz.fdf | tee xyz.out & > > > > > **** make sure to kill the mpd after every run **** > $ mpdallexit > > > On Sat, Feb 26, 2011 at 9:47 AM, Michael Shin > <[email protected]<http://us.mc1619.mail.yahoo.com/mc/[email protected]> > > wrote: > > Dear All > Can any one guide me and send me some related steps for installation of > parallel version of SIESTA-3.0-rc2. > I will be grateful if some has successfully installed the parallel version > of SIESTA-3.0-rc2 and can send me arch.make file. > Also tell me the FORTRAN compiler name which can support parallel version > of SIESTA-3.0-rc2 > Regrads > > > > > >
