From chemistry-request&$at$&server.ccl.net Fri Jul 30 12:45:45 1999 Received: from postoffice.npac.syr.edu (postoffice.npac.syr.edu [128.230.7.230]) by server.ccl.net (8.8.7/8.8.7) with ESMTP id MAA20861 for ; Fri, 30 Jul 1999 12:45:45 -0400 Received: from snake.npac.syr.edu (snake.npac.syr.edu [128.230.162.60]) by postoffice.npac.syr.edu (8.9.3/8.9.3) with ESMTP id MAA19002; Fri, 30 Jul 1999 12:40:21 -0400 (EDT) Received: from localhost (bernhold /at\localhost) by snake.npac.syr.edu (8.9.3/8.9.3) with SMTP id MAA04693; Fri, 30 Jul 1999 12:40:08 -0400 (EDT) Message-Id: <199907301640.MAA04693(+ at +)snake.npac.syr.edu> X-Authentication-Warning: snake.npac.syr.edu: bernhold[ AT ]localhost didn't use HELO protocol Reply-to: bernhold: at :npac.syr.edu To: Jochen Kuepper cc: Iraj Daizadeh , chemistry(+ at +)ccl.net cc: bernhold %-% at %-% npac.syr.edu Subject: Re: CCL:Large matrix diagonalization. In-reply-to: Message from Jochen Kuepper of "Fri, 30 Jul 1999 09:29:56 +0200." <99073009311801.12040 $#at#$ bacchus.pc1.uni-duesseldorf.de> MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-ID: <4689.933352804.1 -A_T- snake.npac.syr.edu> Date: Fri, 30 Jul 1999 12:40:05 -0400 From: "David E. Bernholdt" The LAPACK library deals only with dense matrices, and is a de facto standard library for those operations on uniprocessors. Sparse linear algebra has a great deal of variation across fields and even across different applications, and is a lot less standardized at this point in time. Likewise, parallel dense linear algebra is also a lot less standardized, though that area seems to be consolidating of late. Mounting soapbox... My (humble?) opinion is that chemists (speaking broadly) aren't very smart about solving their linear algbra problems. I think it is getting better, along with the evolution of packages like LAPACK, but historically our track record is pretty poor. Chemists are all too willing to "roll their own" solutions to linear algebra problems, blithly ignoring the fact that there is an entire field of research that devotes itself to such problems. To the extent possible, we should be (well-informed) consumers of the high quality, often highly optimized mainstream linear algebra packages that one can easily obtain from NETLIB or other sites. To be sure, there will be problems that the standard linear algebra tools don't handle well for one reason or another. In that case, my recommendation is to make frieds with a good numerical analyst in the problem area. My experience is that they like new challenges -- it pushes their research & their field along just as chemical challenges do ours, andby virute of their training, they're likely to come up with a better solution than you can working in isolation. The more we chemists expose the numerical analysis community to our unique problems (relative to other fields), the more help we'll get from them, as chemically-derived problems become part of the standard test suites they use when testing new algorithms & libraries. Stepping off soapbox... -- David E. Bernholdt | Email: bernhold _-at-_)npac.syr.edu Northeast Parallel Architectures Center | Phone: +1 315 443 3857 111 College Place, Syracuse University | Fax: +1 315 443 1973 Syracuse, NY 13244-4100 | URL: http://www.npac.syr.edu