[Amps] The Philosophy of Science

[Kim Elmore]

Hi Eric,

I understand your gripe, and agree with your sentiments. yet, I feel the 
need to weigh in. I have a fair bit of experience with numerical models, 
though not those involved in EE. Yet, the basic ideas behind numerical 
models are all quite similar. In particular, you say:

>One statement by you and others ( in some of those OTHER armchairs) regards
>the term 'computer modelling'. There is somewhat of a semantic problem here,
>as follows. The computer models which we use are EXACT, precise physical
>devices whose electronic equations we can write precisely.

Can we really?  I know that you're bright and that you do credible work. 
I'm not an EE, but, for example, are there *any* assumptions made in any of 
the models?  I'll bet there are, because without assumptions of some type, 
it's nearly impossible to start the process. For example, unless each 
component invariably behaves in a way that can be represented by a 
closed-form solution of some type, an approximation is involved.  The 
approximation could be linear (leading to errors that are O(2)), as in 
I*R=E. If there are assumptions of any kind, then the model is not exact. 
In this, I agree with Dave Kirby and Conrad (G0RUZ).

You proposed a straw-man: Where might the simple model I*R=E fail?  A 
simple error occurs if the temperature coefficient of the device is 
ignored.  Yet, the temperature coefficient is unlikely to be truly linear, 
or quadratic, or cubic, or quartic or to fit any order of polynomial.  It's 
likely to have non-linear components, even though those may be small. 
What's worse, the nature of these will vary from one device to another with 
identical ratings. If any of those components are ignored, then the answer 
isn't exact: it's approximate.  This isn't to say that the answer is *bad*, 
or useless: the answer we get may be supremely useful even if it isn't exact.

Non-linear partial differential equations (which must appear abundantly in 
any EE model) almost never have a closed-form solution.  So, some numerical 
methods must be derived to solve them.  As you know, there are lots of 
them.  A common, robust one is Runge-Kutta.  But there are many different 
flavors of Runge-Kutta.  For example, the order of the method (2nd, 3rd, 
etc.) helps define the errors in the converged solution.  If we watch each 
iteration of the solution, we'll see the error term rapidly shrink, but, in 
the end, no matter how many iterations we make, we'll never see it simply 
cease to change: it will bounce around a small neighborhood.  We stop 
iterating when the error term is "small," or "close enough," defined by our 
application.

When viewed in this light, *none* of our numerical models of real-world 
devices or processes are exact.  In fact, because we don't really know 
"everything" about any device, we wouldn't recognize an exact answer of it 
were given to us. Some might call knowing everything "divine knowledge" 
while a quantum physicist would simply say the concept of an exact answer 
has no meaning. Regardless, the answers we get are very, very good. 
Ideally, they are so good that we can't measure whatever errors they 
contain. Even short of that ideal, they allow us to design and build things 
undreamed of 25 years ago.

There.  Done.  Said my peace.  Back to RF!

73,

Kim Elmore
                           Kim Elmore, Ph.D.
                        University of Oklahoma
         Cooperative Institute for Mesoscale Meteorological Studies
"All of weather is divided into three parts: Yes, No, and Maybe. The
greatest of these is Maybe" The original Latin appears to be garbled.