From owner-nwchem-users@emsl.pnl.gov Wed Dec 29 22:33:45 2004 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 iBU6Xjxk003934 for ; Wed, 29 Dec 2004 22:33:45 -0800 (PST) Received: (from majordom@localhost) by odyssey.emsl.pnl.gov (8.12.10/8.12.10/Submit) id iBU6Xj4N003933 for nwchem-users-outgoing; Wed, 29 Dec 2004 22:33:45 -0800 (PST) Date: Wed, 29 Dec 2004 22:30:01 -0800 From: David Pullman Subject: ccsd problem for >1 node To: nwchem-users@emsl.pnl.gov Message-id: <5.2.0.9.0.20041229213831.02400430@sciences.sdsu.edu> MIME-version: 1.0 X-Mailer: QUALCOMM Windows Eudora Version 5.2.0.9 Content-type: text/plain; charset=us-ascii; format=flowed X-COS-MailScanner: Found to be clean X-MailScanner-SpamCheck: not spam, SpamAssassin (score=0.029, required 8, AWL 0.03) X-MailScanner-From: dpullman@sciences.sdsu.edu Sender: owner-nwchem-users@emsl.pnl.gov Precedence: bulk Dear all, We're trying the coupled cluster module of NWChem 4.6 to calculate energies but generally can't seem to reproduce results from run to run. A simple example input file for Ne dimer is given below. Specifically, during different runs the ccsd iterations give different "delta energies" and "rms errors." Interestingly, if only one node in our computer cluster is used, the results are exactly reproducible; however, if two or more nodes are used, the results become irreproducible, and sometimes the maximum number of allowed iterations (20) is reached with no convergence-- when one node is used, 6 iterations are required for convergence. Has anyone seen/solved this problem? We're running on a 16 node Athlon XP cluster using Rocks 3.2.0. NWChem was compiled with Intel Fortran 8.0 and gcc 3.2.3; the ATLAS 3.6.0 library was used. Here's the Ne dimer input file: scratch_dir /export permanent_dir /home/dpullman/nwchem/ne_dimer geometry noprint units angstroms Ne1 0.0 0.0 0.0 Ne2 0.0 0.0 3.1 end basis spherical Ne1 library Ne cc-pvdz Ne2 library Ne cc-pvdz end task ccsd(t) energy