From landman@hal.physics.wayne.edu Sun Oct 4 08:00:27 1992 Date: Sun, 4 Oct 92 12:00:27 EDT From: (Joseph I. Landman) To: chemistry@ccl.net Subject: PC numerical instabilities I have been following the discussion, and I would like to add my 2 cents worth. About 1 year ago I was developing the innards of a nearly complete (and easy to read/use/change) version of a Herman-Skillman type program. I had been using MS fortran v5.0 to generate code for testing on an 8086 PC and a 80286 PC. The 286 code worked nicely, converging quickly to the desired solution. The 8086 ran EXACTLY the same code (copied from a floppy) in exactly the same environment, and had a great deal of convergence trouble/rounding problems. I was using an 8087 on the 8086, and a 80287 on the 80286. I investigated furthur, and found, much to my chagrin, that the 8087 control word was set up improperly. That is, it was not told how to do its rounding properly, thus allowing LARGE errors to accumulate quickly. I reported this in comp.lang.fortran and several other groups. Within one week, Micro soft had a diskette with the correct libraries out to me. They seemed to work, but I have been wary of them/their products ever since. I had spoken at length to one of their reps who told me that "fortran was not a very important product to them" (paraphrased), and that they were not going to follow up with bug fix releases. I have since given my PC fortran business to watcom. There product is IMHO much better/more mature than Microsoft's. (they even have an internet based tech support address... not common to PC world software). I suspect the problem you see is the same one I saw. I will be happy to ftp the patch as supplied to me by Msoft to anyone who wants it. I strongly recommmend switching fortran compilers though. Microsoft is not committed to its fortran product, and thus I am not committed to Msoft. epilog: my version of the SCF code is almost done (I am squashing the last little critters this week) due largely to my switch from Msofts product, to Watcom's. I have no financial or other connection to Watcom, I am an extremely satisfied customer. Joe #,:::::::;, :: Joe Landman Graduate Student (PhD), # :: :: Computational Solid State Physics. #;: :: :: Physics Dept., Wayne State Univ. #`::::; oe ;:::; andman 666 Hancock Avenue, Detroit MI 48202 #Internet: landman@hal.physics.wayne.edu landman@rhic2.physics.wayne.edu # kil@pandora.cs.wayne.edu jland@ss0.eng.wayne.edu #ATTnet: (313) 577-2720