Re: Computers

["Donald E. Simanek" <dsimanek@eagle.lhup.edu>]

On Wed, 26 Feb 1997, Leigh Palmer wrote:

> Perhaps this will explain Macvangelism to Paul and others so bemused
> by it. We are true believers, I'm afraid.

Yes, it does seem perplexing to those of us not bitten by it. Some of us
are quite turned off by the interface, the pull down menus, the icons, and
*especially* the dependence on the mouse, features which initially attract
the novice, but stand in the way of productivity for the power user. And
these features now infect the DOS world as well, for the disease has taken
over the Windows environment nearly totally. Oh, one *can* still do
Windows 95 without the mouse most of the time, using keyboard equivalents
only, but there's still too much mucking around in what I consider a
user-hostile environment of both Mac and Windows platforms.

So, I'm not a religious advocate of either platform; I hate them both, and
whenever possible I stick to plain DOS. My entire Web page and all of my
HTML documents on it were created with a text editor. I've gotten to like
UNIX more and more as I get into it. A friend down the street has a
powerful DOS machine on which he has LINUX, DOS, OS2 and even Windows and
can switch readily between them. He is a programmer for a living, and
hates the Windows environment so much that he's gone into the Windows code
and modified it so he never sees the graphical interface and never has to
deal with it. He's reduced his system so he can use keyboard commands for
everything. And he can make that computer jump through hoops much faster
and more efficiently than anyone else I've ever watched. That's an
interesting form of extremism. I don't go that far.

But, I didn't post this just to fan the flames of the religious wars. A
colleague from chemistry brought to my attention an interesting difference
in the performance of Microsoft Excel in its Mac and DOS versions. Her
students were doing a chemistry problem requiring a data set of about 8
points on which a linear regression was to be done. Our students use both
Mac and DOS platforms, and our student labs have both. Excel is on both.
Students got the same answer for the slope of the line, a, but vastly
different results for the intercept, b. (y = ax + b). 

We independently checked with other programs. I even dusted off an old
program I wrote in Fortran back in 1965! That was the punched card era,
so I modified it to read and write to files, compiled it under Microsoft
Fortran and ran it. It, and several other spreadsheet packages, and
MathCad, all agreed with the answer given by MAC Excel, and we are
convinced that the DOS Excel has a bug in this routine. A colleague
who had the time to look into it says the two versions seem to be using
different algorithms for the calculation, and one is faulty.

And that's another pet peeve of mine. So much software of this type
doesn't provide the user with *any clue* what algorithms are being used
for curve fitting, etc., and which are appropriate for which purposes.
It's like using a black box you can't open. The only solution: write your
own software, then you know darn well what algorithms it uses. And none of
these products marketed for business applications are appropriate for
scientific work, but the scientific software costs *much* more. Exception:
My Mathcad manual not only documents the algorithms used, but gives
references to the literature for each.

I also tried a polynomial fitting routine, which gave the same results.

Try it yourself with your favorite software:

x      y

0.002  0.000
0.117  0.025
0.238  0.050
0.356  0.075
0.472  0.100
0.597  0.125

IBM Excel gives a = 4.751429,  b = 0.0050502
MAC Excel gives a = 4.7514  ,  b = -0.0002

The students obviously had different precision settings. Unfortunately
this software doesn't (so far as I know) readily compute the uncertainties
in a and b, which, in this case, will be quite large for b.

My hunch is that the difference is in the software and has nothing to do
with the relative merits of the hardware. One can, of course, simply wave
it away, saying that the difference is insignificant, and the intercept is
practically zero (as theory would predict in this case). And another pet
peeve. In a case such as this, where there's overpowering reason to say
that the fit must pass exactly through zero, there's no graceful way to
tell the commercial software this, forcing a fit through zero. Or maybe I
just don't know the trick to do it, since I never use Excel, Quattro, or
any of the other spreadsheet clones. 

	-- Donald

......................................................................
Dr. Donald E. Simanek                            Office: 717-893-2079
Prof. of Physics                    Internet: dsimanek@eagle.lhup.edu
Lock Haven University, Lock Haven, PA. 17745          CIS: 73147,2166 
Home page: http://www.lhup.edu/~dsimanek            FAX: 717-893-2047
......................................................................