From owner-chemistry@ccl.net Tue Oct 2 15:13:01 2007 From: "Uwe Huniar uwe.huniar%x%cosmologic.de" To: CCL Subject: CCL: Gaussian 03 MP2 performance Message-Id: <-35290-071002093834-8353-G8zJQe1fF5fM8it41CrSKg,,server.ccl.net> X-Original-From: Uwe Huniar Content-Transfer-Encoding: 8bit Content-Type: text/plain; charset=ISO-8859-1; format=flowed Date: Tue, 02 Oct 2007 14:41:52 +0200 MIME-Version: 1.0 Sent to CCL by: Uwe Huniar [uwe.huniar~~cosmologic.de] the list of programs with RI-MP2 lacks of the one that has RI-MP2 energy and gradients since about 10 years: Turbomole RI-MP2 gradients have originally been published in 1997 (Theor. Chem. Acc. 97, 1997, 331) by Weigend and Häser, and implemented in Turbomole. Meanwhile RI-MP2 calculations can also be done in parallel (based on MPI), here an example (taken from Köhn, Hättig, Chem. Phys. Lett. 358, 2002, 350, and from a talk a bit later that used faster CPUs): Calicheamicin fragment, 39 atoms, cc-pVTZ basis, 924 basis functions, C1, MP2 energy, on (ancient) Athlon 1800 MP CPUs, 100MBit Ethernet: CPU Total wall time in minutes 1 60 7 18 13 11 A short note about the scaling: the N**5 step that is dominant for larger systems on 1, 7, 13 CPUs takes 34, 5.5, 3 minutes, resp. The MP2 energy was -4.26 Hartree, the error from RI 0.0005 Hartree when using the auxiliary basis sets of Weigend and Häser (included in Turbomole and also available from the Turbomole ftp server). On a Core 2 machine an RI-MP2 single point calculation should take just a few minutes (and yes: that is shorter than the ground state Hartree-Fock calculation, but one can use RI-JK as pre-step for RI-MP2 with Turbomole). Regards, Uwe > MP2 does get expensive as the number of basis functions increases > (scaling as n**5). If it's only single-point calculations that you need, > you may want to look at "local" or "RI" implementations (you find them > in Q-Chem, NWChem, MOLPRO, ORCA, and probably several other programs > nowadays; some of them even have gradients, which used to be the problem > until lately). -- ------------------------------------------------------------------------------ Dr. Uwe Huniar COSMOlogic GmbH&Co.KG Burscheider Str. 515 D-51381 Leverkusen Germany Tel. +49 - 2171 - 36 36 65 Fax +49 - 2171 - 73 16 89 EMail: uwe.huniar,+,cosmologic.de turbomole,+,cosmologic.de info,+,turbomole.com