From owner-chemistry@ccl.net Wed Dec 6 05:41:12 1995 Received: from bedrock.ccl.net for owner-chemistry@ccl.net by www.ccl.net (8.6.10/950822.1) id FAA19470; Wed, 6 Dec 1995 05:38:22 -0500 Received: from gaia.msicam.co.uk for rnobes@msicam.co.uk by bedrock.ccl.net (8.7.1/950822.1) id FAA02009; Wed, 6 Dec 1995 05:38:18 -0500 (EST) Received: by gaia.msicam.co.uk (940816.SGI.8.6.9/SMI-4.1 (MSI-Cambridge)) id KAA27163; Wed, 6 Dec 1995 10:33:06 GMT From: rnobes@msicam.co.uk (Ross Nobes) Message-Id: <199512061033.KAA27163@gaia.msicam.co.uk> Subject: Re: CCL:G:G94: B3LYP energy and Charge To: frisch@lorentzian.com (Mike Frisch) Date: Wed, 6 Dec 1995 10:33:05 +0000 (GMT) Cc: chemistry@ccl.net Reply-To: rnobes@msi.com In-Reply-To: <9512051730.AA14738@mjf> from "Mike Frisch" at Dec 5, 95 12:30:57 pm X-Mailer: ELM [version 2.4 PL23] Content-Type: text Mike Frisch writes: > > > The answer is that both are right! The Gaussian definition of B3LYP > > uses the VWN local correlation functional; most other programs use > > what Gaussian terms VWN5. > > > > No, B3LYP was defined to use VWN3. Using the same name for two different > functionals is a disservice to everyone and makes reproducing other peoples > results impossible. This used to be a major problem in the DFT area, and > things are finally improving with regard to terminology. To take a name for > something that is specified in the literature and use it for something > different is a major step backwards. Published and in-press tests of B3LYP > all use VWN3; an assumption that the same conclusions apply to a different > functional just because it has the same name are not valid. > Point taken. All that I was implying in my original answer was that there are no 'numerical' problems here - both programs are giving correct results within their own definitions of B3LYP. Most programs (such as Mulliken, CADPAC, MOLPRO, Turbomole) use the VWN5 functional uniformly as the local correlation part of functionals such as SVWN, B3P86 and B3LYP. It is unfortunate that Gaussian is alone in using an alternative VWN local correlation functional and (as Mike says) that the terminology is not consistently defined. The issue is further confused by a bug in the Gaussian 92/DFT open-shell VWN5 code which leads to small discrepancies with the other programs even when extremely large quadrature grids are used. -- +----------------------------------------------------------------+ | Dr Ross Nobes | | Senior Scientist, Quantum Mechanics ! | | | BIOSYM/Molecular Simulations | | 240/250 The Quorum Phone: + 44 1223 413300 | | Barnwell Road Fax: + 44 1223 413301 | | Cambridge CB5 8RE, UK E-mail: rnobes@msicam.co.uk | +----------------------------------------------------------------+