Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.2 9/18/84; site oddjob.UUCP
Path: utzoo!watmath!clyde!burl!ulysses!mhuxr!mhuxn!ihnp4!oddjob!sra
From: sra@oddjob.UUCP (Scott R. Anderson)
Newsgroups: net.physics
Subject: Re: IBM supercomputer to test QCD
Message-ID: <952@oddjob.UUCP>
Date: Wed, 28-Aug-85 12:31:42 EDT
Article-I.D.: oddjob.952
Posted: Wed Aug 28 12:31:42 1985
Date-Received: Fri, 30-Aug-85 09:01:45 EDT
References: <278@ecsvax.UUCP> <939@oddjob.UUCP> <313@ecsvax.UUCP>
Reply-To: sra@oddjob.UUCP (Scott R. Anderson)
Distribution: net
Organization: University of Chicago, Department of Physics
Lines: 63
Summary: 

Somewhere in net.physics, D Gary Grady writes:
>> >Now, does anyone remember the name of the guy who wrote in Physics Today
>> >a few years ago that all physicists should go and buy Commodore 64s so
>> >they wouldn't have to use big computers any more?
>> 
>I was only kidding (sorry about the missing :=) ), but it's worth noting
>that Bak's article was a little overly enthusiastic, if I'm remembering
>it correctly.

I have to agree that it was somewhat "overly enthusiastic"; I took exception
(over-enthusiastically? (:-) because I use Sun Workstations in much the
same way that Per Bak used his Commodore, although not at home (I wish!).
For most of my work, they are very adequate.

>The main problem was software, including compilers and subroutine
>libraries.  Many BASIC interpreters (and I believe Bak specifically gave
>examples of programs in BASIC) have poorly written scientific functions,
>for example, which give little if any accuracy in parts of their domains.

Using BASIC was his biggest mistake!  I don't believe that any part of
his program used scientific functions, though.  Can I trust a Commodore
64's floating point software?

>Unfortunately, many scientists are self-taught computerists

Like me (:-).

>and either don't know enough to worry about such problems or are
>inclined to think it "won't happen to them" in the fine tradition of
>seat belt usage.

It isn't just scientists, but also engineers, business people, etc.,...
I guess this is a case of shoddy American worksmanship and "caveat emptor" (:-).

>If you think I'm overly pessimistic, think about how
>many of your colleagues use IBM's SSP (including, ghu help us, RANDU).

The people around here stay as far away from IBM computers as possible
(unless they are given to them. (:-)

>Have YOU ever checked out the claimed accuracy of your library
>functions?

Not the math functions.  We run 4.2BSD; can anybody tell me which of these
routines have limitations and what they are?  I don't recall even *seeing*
any claims of accuracy.

The random number generators are another story.  I quickly discovered that
rand() is useless.  However, did you know that even random() generates an
incorrect second moment?  (Details on request.)  I am currently using the
generator referenced in Bak's article, and have found it to be better than
both of these, and to be of intermediate speed.

>An article in Scientific American compared the accuracy of
>various compilers and interpreters on micros and mainframes, with most
>of them making an unimpressive showing compared with a pocket
>calculator.

Can someone give me a reference for this article?  Also, can anyone give
me any references for writing mathematical function subroutines?

				Scott Anderson
				ihnp4!oddjob!kaos!sra