From P.Sherwood@dl.ac.uk Sun Aug 11 15:16:51 1996 Received: from mserv1.dl.ac.uk for P.Sherwood@dl.ac.uk by www.ccl.net (8.7.5/950822.1) id OAA10656; Sun, 11 Aug 1996 14:46:38 -0400 (EDT) Received: from tcsi.dl_sun_server by mserv1.dl.ac.uk with SMTP id TAA16765 (8.6.13/5.3[ref postmaster@dl.ac.uk] for dl.ac.uk from P.Sherwood@dl.ac.uk); Sun, 11 Aug 1996 19:46:33 +0100 Precedence: first-class Received: by tcsi.dl_sun_server (4.1/SMI-4.1) id AA15302; Sun, 11 Aug 96 19:47:29 BST Date: Sun, 11 Aug 96 19:47:29 BST From: P.Sherwood@dl.ac.uk (P. Sherwood) Message-Id: <9608111847.AA15302@tcsi.dl_sun_server> To: CHEMISTRY@www.ccl.net Subject: quantum chemistry databases Dear CCL Does anyone know of any databases of quantum chemical results, other than the "Quantum Chemistry Literature Data Base" from JAICI (Japan). I have a recollection of a Gaussian results archive being maintained - but haven't heard anything recently. Many thanks for any information Paul Sherwood ====================================================================== Dr Paul Sherwood, CLRC Daresbury Lab, email: p.sherwood@daresbury.ac.uk Warrington, tel: +44-1925-603553 WA4 4AD, UK fax: +44-1925-603634 ====================================================================== From D.van.der.Spoel@chem.rug.nl Sun Aug 11 16:16:53 1996 Received: from mailhost.rug.nl for D.van.der.Spoel@chem.rug.nl by www.ccl.net (8.7.5/950822.1) id QAA10823; Sun, 11 Aug 1996 16:02:11 -0400 (EDT) Received: from rugch4.chem.rug.nl by mailhost.rug.nl with SMTP (PP); Sun, 11 Aug 1996 22:02:03 +0200 Received: from rugmd0.chem.rug.nl by rugch4.chem.rug.nl (WAA04179); Sun, 11 Aug 1996 22:02:00 +0200 Received: from rugmd17.chem.rug.nl by rugmd0.chem.rug.nl (5.x/SMI-SVR4) id AA24797; Sun, 11 Aug 1996 22:01:58 +0200 Received: by rugmd17.chem.rug.nl (5.x/SMI-SVR4) id AA21085; Sun, 11 Aug 1996 22:02:00 +0200 From: D.van.der.Spoel@chem.rug.nl (David van der Spoel) Message-Id: <9608112002.AA21085@rugmd17.chem.rug.nl> Subject: Bug in Sun CC To: chemistry@www.ccl.net Date: Sun, 11 Aug 1996 22:01:59 +0200 (MET DST) Reply-To: D.van.der.Spoel@chem.rug.nl X-Mailer: ELM [version 2.4 PL23] Mime-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Hi. I am running my MD code (Ansi-C plus a little Fortran) on various platforms. Recently I found out that on my Sun SparcStation 5 the code produces erroneous results. Moreover, with the same input, results may be different from time to time. Note that I don't use any random generators or whatsoever. On a Sun SS-10 there is no problem (using the same executable), and neither is there on a Silicon Graphics. The errors do only show up when optimization is set to -xO4, at -xO2 there is no problem either. I have had similar problems with the GNU C compiler. Did anyone ever find something similar ? David van der Spoel --------------------------------------------------------- EMAIL: spoel@chem.rug.nl WWW: http://rugmd0.chem.rug.nl/~spoel PHONE: 31-50-3634327 FAX: 31-50-3634800 MAIL: Nijenborgh 4, 9747 AG Groningen, The Netherlands. --------------------------------------------------------- From kneth@lord.Colorado.EDU Sun Aug 11 18:16:53 1996 Received: from lord.Colorado.EDU for kneth@lord.Colorado.EDU by www.ccl.net (8.7.5/950822.1) id SAA11060; Sun, 11 Aug 1996 18:12:12 -0400 (EDT) Received: from localhost by lord.Colorado.EDU via SMTP (920330.SGI/911001.SGI) for chemistry@www.ccl.net id AA29282; Sun, 11 Aug 96 16:11:39 -0600 Date: Sun, 11 Aug 1996 16:11:39 -0600 (MDT) From: Kenneth Geisshirt To: chemistry@www.ccl.net Subject: Re: Bug in Sun CC In-Reply-To: <9608112002.AA21085@rugmd17.chem.rug.nl> Message-Id: Mime-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII On Sun, 11 Aug 1996, David van der Spoel wrote: > The errors do only show up when optimization is set to -xO4, at -xO2 > there is no problem either. How large is the error? Some compiler may violate the IEEE standard for floating-point operations in order to get faster code. And then, there might be problems. Check the man-page for that. I guess it is a well-known problem that optimized code and non-optimized code produce different results. If you have code like foo = x*y - x*z it will be rearranged to foo = x*(y-z) since multiplication is very slow. Due to round-off error, this might get you a different result. I don't know how your MD program works, but from your name I guess that it is a parallel MD program (gromacs?). I have some experience with parallel MD programs, so let me ask a few 'silly' questions: (i) Do you get the same result with 1 processor and 2 processors at low optimization? (ii) Can you see the error after one time step? I.e. run only one time step and see what happens. (iii) Is the error independent of integrator? If you can answer NO to (i), you have a bug in the parallization. If you can say YES to (ii) and (iii) it might be the force calculation, otherwise it is the integrator. You have tried Sun and SGI, any other platforms? Best regards Kenneth +-----------------------------------------------------------------+ | Kenneth Geisshirt Ph.D.-student | | Department of Life Sciences and Chemistry, Roskilde University | | Currently at: University of Colorado at Boulder | +-----------------------------------------------------------------+ | E-mail: kneth@fatou.ruc.dk | | Homepage: http://virgil.ruc.dk/~KNETH | +-----------------------------------------------------------------+ | Old chemists don't die - they just reach equilibrium | +-----------------------------------------------------------------+