From owner-nwchem-users@emsl.pnl.gov Mon Oct 24 13:07:25 2005 Received: from odyssey.emsl.pnl.gov (localhost [127.0.0.1]) by odyssey.emsl.pnl.gov (8.12.10/8.12.10) with ESMTP id j9OK7PSN000864 for ; Mon, 24 Oct 2005 13:07:25 -0700 (PDT) Received: (from majordom@localhost) by odyssey.emsl.pnl.gov (8.12.10/8.12.10/Submit) id j9OK7P7N000863 for nwchem-users-outgoing-0915; Mon, 24 Oct 2005 13:07:25 -0700 (PDT) Date: Mon, 24 Oct 2005 14:00:16 -0600 From: Daniel Knapp Subject: [NWCHEM] Band calculation problem To: NWChem Users Message-id: MIME-version: 1.0 (Apple Message framework v734) X-Mailer: Apple Mail (2.734) Content-type: text/plain; charset=US-ASCII; delsp=yes; format=flowed Content-transfer-encoding: 7bit X-UCalgary-MailScanner-Information: Please contact IT Help Desk at (403) 220-5555 for more information X-UCalgary-MailScanner: Found to be clean X-UCalgary-MailScanner-From: dknapp@ucalgary.ca Sender: owner-nwchem-users@emsl.pnl.gov Precedence: bulk Hello, I am trying to run a NWPW band calculation on a parallel computer and have been finding some strange behaviour. When the calculation is run on four or fewer processors the change in density from iteration to iteration rapidly falls to zero. When the very same calculation is run on more than four processors the change in density converges to a positive constant ~ 10. It seems as though the energy and the change in energy converge to the same values in either case, so I'm wondering if there is some kind of weird bug in how the density change is calculated, or if there is some problem with the setup (see below for details)? Thank you for your help, Daniel The computer I am running on has the following specs: ---------- 128 nodes of dual AMD 2.4 GHz Opteron processors Voltaire Infiniband switched-fabric interconnect Hewlett Packard's XC software stack RedHat Enterprise Linux Operating System Parallel communications run through MPI, code compiled with g77 (version 3.2.3). Environment variables: NWCHEM_TARGET linux64 USE_MPI yes USE_MPIF yes LIBMPI -l mpich -l hpmpi The input file I am using is: ---------- TITLE "Cu 3x2x3 Planar Surface BAND" START SCRATCH_DIR /scratch/dan/test2 PERMANENT_DIR /scratch/dan/test2 PRINT high #will write more info to output files ECHO #writes input file to output #***** Enter geometry using fractional coordinates ***** GEOMETRY units angstroms center noautosym noautoz print SYSTEM molecule #can be polymer, surface, crystal, molecule (no transl. symm.) END #system Cu 0.63900 4.24230 0.00000 Cu 3.19500 4.24230 0.00000 Cu 5.75100 4.24230 0.00000 Cu 7.02900 2.02879 0.00000 Cu 4.47300 2.02879 0.00000 Cu 1.91700 2.02879 0.00000 Cu 1.91700 3.50439 2.08700 Cu 4.47300 3.50439 2.08700 Cu 7.02900 3.50439 2.08700 Cu 5.75100 1.29089 2.08700 Cu 3.19500 1.29089 2.08700 Cu 0.63900 1.29089 2.08700 Cu 0.63900 2.76650 4.17400 Cu 3.19500 2.76650 4.17400 Cu 5.75100 2.76650 4.17400 Cu 7.02900 0.55300 4.17400 Cu 4.47300 0.55300 4.17400 Cu 1.91700 0.55300 4.17400 END #geometry #***** Setup nwpw gamma point code ***** NWPW SIMULATION_CELL LATTICE lat_a 14.4904d0 # 7.668A lat_b 8.3658d0 # 4.427A lat_c 26.7850d0 # 14.174A END #lattice ngrid 20 20 20 #in same ratios as lattice vectors END #simulation_cell # Wavefunction_cutoff 12.5 Ewald_ncut 8 #number of unit cells in Ewald real space summation Unrestricted #occupied and unoccupied states in a given orbital not required to have the same energy mult 2 xc PBE96 #***** k-points sampled in first BZ ***** monkhorst-pack 1 1 1 END #nwpw SET nwpw:minimizer 2 SET nwpw:psi_nolattice .true. # turns off unit cell checking for wavefunctions SET nwpw:stress_numerical .true. # numerical stress ONLY works with band DRIVER clear maxiter 100 END TASK band energy -- Daniel Knapp PDF in Fuel Cells Ziegler Research Group University of Calgary dan@danielk.ca http://www.danielk.ca