From chemistry-request@server.ccl.net Tue Apr 23 11:57:03 2002 Received: from hotmail.com ([64.4.9.46]) by server.ccl.net (8.11.6/8.11.0) with ESMTP id g3NFv3A07453 for ; Tue, 23 Apr 2002 11:57:03 -0400 Received: from mail pickup service by hotmail.com with Microsoft SMTPSVC; Tue, 23 Apr 2002 08:56:52 -0700 Received: from 63.184.71.13 by lw9fd.law9.hotmail.msn.com with HTTP; Tue, 23 Apr 2002 15:56:51 GMT X-Originating-IP: [63.184.71.13] From: "Fernando De La Vega" To: CHEMISTRY@ccl.net Subject: Summary: Gaussian98 Warning Message Date: Tue, 23 Apr 2002 08:56:51 -0700 Mime-Version: 1.0 Content-Type: text/plain; format=flowed Message-ID: X-OriginalArrivalTime: 23 Apr 2002 15:56:52.0535 (UTC) FILETIME=[7C321C70:01C1EADF] Dear All, Last week I asked a question with regard to the warning received in Gaussian98 that reads: "Warning!!: The largest alpha MO coefficient is #########", where the #'s represent a given (usually large) number. I want to thank Douglas Fox (from Gaussian Tech Support), Antonio Marquez and Christoph van Wüllen for their response to my inquiry. Below I reproduced their answers. ---------------------------------------------------- Douglas J. Fox: Fernando, The warning is most relevant for post-HF calculations like Moller Plesset perturbation theory or coupled cluster, where the accuracy of the result is related to the accuracy of the MO integrals. When you have near linear dependencies in the basis set you will get large MO coefficients related to maintaining orthogonality. The side effect of large coefficients is loss of precision on any existing machine. Depending on the order that you add C1*I1+C2*I2-C3*I3+C4*I4 where C1 and C3 are large and of the same sign, if the integrals are all about the same size and C1*I1-C3*I3 is near zero the contribution from C2 or C4 can be lost if they are added before C3 is subtracted. For HF and DFT calculations the code which checks this is run but seldom is it an issue, it is actually just after the SCF completes. For any post-HF method look to see that the correlation corrections are a moderate fraction of the total energy. It may not be an error but it is worth comparing with a slightly smaller basis or a different correlation method, CCSD is less sensitive than MP4, as a check. --------------------------------------------------------------- Antonio Marquez (marquez@us.es): Dear Fernando, I guess that you mean that ####### is a LARGE number. This means that you have a nearly linearly dependent AO basis set. The closer you are to have a really linearly dependent basis set the biggest will be the number. For HF calculations this is not usually a problem as the large number(s) is(are) in one (or more) of the virtual MO that are just the orthogonal complement to your occupied MO. Problems may arise if these orbitals are used for a subsequent correlated ab initio calculation. During the 2e-integrals transformation to the MO basis the AO integrals will be mutiplied by the all the MO coefficients. If one or more of your coefficients if very high the numerical precision on your transformed MO integrals will be compromised and you have to be carefull with the result. The ways to circumvent this problem are either reduce the linear dependency in your AO basis set or use an algorithm to compute the 2e repulsion integrals that mantains a high degree of numerical accuracy. I hope this clarifies a little bit your doubts. ---------------------------------------------------------------- Christoph van Wüllen (Christoph.vanWullen@TU-Berlin.De): This means that you have a near-linear-dependency in your basis sets, e.g. if you have diffuse functions on two atoms which are quite close. Since very similar wavefunctions can result from very different coefficients, numerical problems might arise in force constant or correlation calculations. If you are sure that there is no input error, proceed (with care). ----------------------------------------------------------- _________________________________________________________________ MSN Photos is the easiest way to share and print your photos: http://photos.msn.com/support/worldwide.aspx