From owner-chemistry@ccl.net Sat Jun 3 22:08:01 2006 From: "Tirath Ramdas tirath[#]tpg.com.au" To: CCL Subject: CCL: AO angular momentum statistical distribution Message-Id: <-31894-060603010628-24524-u41YKpoYMtQqyTTcLlWbPg[#]server.ccl.net> X-Original-From: Tirath Ramdas Content-Transfer-Encoding: 7bit Content-Type: text/plain; charset=US-ASCII; delsp=yes; format=flowed Date: Sat, 3 Jun 2006 14:14:12 +1000 Mime-Version: 1.0 (Apple Message framework v750) Sent to CCL by: Tirath Ramdas [tirath ~~ tpg.com.au] Hi all, I'm getting into ab initio computational quantum chemistry from a computer engineering background, I find it all terribly fascinating, but it gets a bit intractable sometimes! At the moment I'm keen on electron repulsion integrals. In the immediate future, I'm trying to get a grasp on typical workload characteristics. To this end, I have modified GAMESS a little to spit out a count of the number of roots required every time the Rys quadrature method is called (and I also changed the INTTYP command in my inp files to make sure RYSQUAD handles all the integrals). I have some data so far, which suggests no compelling trends at all regarding the distribution, but the data I have is essentially meaningless because I've only run a handful of jobs! But still, it was disappointing that I wasn't able to even identify very broad trends such as "low number of roots, corresponding to low angular momentum, appear to be the plurality"... I would like to identify the common case, and then optimise for it. Does anyone have some insight that they could share on this? Or is it, as I am beginning to suspect, too dependant on the nature of the workloads that the distribution changes wildly between application domains and between job types? If that is the case, would it make sense to assume a flat angular momentum distribution profile (up to some limit) for the purpose of proposing and validating computer architectural optimisations? Thanks! -tirath