From chemistry-request@server.ccl.net Wed Jan 23 10:14:39 2002 Received: from mail-b.bcc.ac.uk ([144.82.100.22]) by server.ccl.net (8.11.6/8.11.0) with ESMTP id g0NFEcc20010 for ; Wed, 23 Jan 2002 10:14:38 -0500 Received: from socrates-a.ucl.ac.uk by mail-b.bcc.ac.uk with SMTP (Mailer) with ESMTP; Wed, 23 Jan 2002 15:13:47 +0000 Received: (from uccatvm@localhost) by socrates-a.ucl.ac.uk (8.10.2+Sun/8.9.3) id g0NFDlN00482; Wed, 23 Jan 2002 15:13:47 GMT From: uccatvm Message-Id: <200201231513.g0NFDlN00482@socrates-a.ucl.ac.uk> Subject: Re: CCL:linearly dependent basis set or the mystery of the large alpha MO To: chemistry@ccl.net Date: Wed, 23 Jan 2002 15:13:46 +0000 (GMT) Cc: T.vanMourik@ucl.ac.uk (Tanja van Mourik) In-Reply-To: from "Elena Jakubikova" at Jan 22, 2002 08:22:35 PM X-Mailer: ELM [version 2.5 PL3] MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit Dear Elena, > I would like to learn more about the problem of nearly linearly > dependent basis set and very large MO coefficients that arise thanks > to this problem. I have read in the Gaussian manual that these large > coefficients may lead to subtantial numerical errors and that in case > any MO is larger than 1000, the calculation is aborted. Does any of > you have an idea how big these numerical errors might be if my largest > MO is of the order 100? When do I really need to start worry about > numerical accuracy of the calculation, at which value of the largest > MO coefficient? And finally, how does Gaussian program deal with the > problem of very large coefficients, is it trying to correct the > problem in any way (for example by excluding certain basis functions > > from further calculation) or it just continues calculation > irrespectively of what the error might be on the end? Gaussian deletes eigenvectors of the overlap matrix belonging to small eigenvalues (threshold value 10E-5). You may want to read David Feller's messages to CCL on this subject: see CCL archive of Jan 21, 2000. In my experience, you have to really careful if you compare energies obtained with different programs, as these may have different threshold values for deleting the eigenvectors. Feller reports an energy difference of more than 4 millihartree for Ar computed with G98 and other programs that do not delete almost linearly dependent functions. > > I also tried to learn some more about this problem from the > literature, but I was not very successful. The only paper touching > this problem I found was 'A. Wallis, D.L.S. McElwain, H.O. Pritchard, > "Variation method and the algebraic eigenvalue problem.", > Int.J.Quantum Chem. 3(5), 711-722 (1969).' Does anybody know where can > I find some more information about this? I encountered problems with almost linearly dependent functions with basis sets containing large sets of bondfunctions: see T. van Mourik, R.J. Vos, J.H. van Lenthe and F.B. van Duijneveldt, Int. J. Quant. Chem. 63, 805 (1997). It is a bit of a special case, but it may be of interest if you want to learn more about this subject. Hope this of any use to you, Tanja -- ===================================================================== Tanja van Mourik Royal Society University Research Fellow Chemistry Department University College London phone: +44 (0)20-7679-4663 20 Gordon Street e-mail: work: T.vanMourik@ucl.ac.uk London WC1H 0AJ, UK home: tanja@netcomuk.co.uk http://www.chem.ucl.ac.uk/people/vanmourik/index.html =====================================================================