Richard Clark wrote:
. . .
"Computer model"s are as useful as they are predictive. EZNEC
conforms to NEC which in turn conforms to field tests. The chain of
evidence is quite strong and I have yet to hear of any real-life
design that has defied NEC analysis.

I don't think you said quite what you meant.

There are many real-life designs which are beyond NEC's (and hence
EZNEC's) ability to analyze. One common example is a microstrip (patch)
antenna; NEC has no way to account for the dielectric. A user with even
minimal knowledge should realize the significance of this shortcoming,
and not expect to be able to get accurate results from a model with the
dielectric excluded.

However, and I think this is probably what you meant, the accuracy of
NEC is strikingly good whenever a model can be constructed that does a
good job of mimicking the real antenna. All this really proves are two
things: 1. The fundamental electromagnetic equations solved by NEC are
sound, and 2. Most of the bugs have been worked out of the code, so it
correctly solves the equations.

The equations solved by NEC can't be solved in closed form. That means
there is no formula into which you can plug numbers and calculate a
result. The method it uses can be done manually for very simple cases
and with very limited accuracy -- see Kraus' _Antennas_, 2nd Edition or
later, for a good example. A number of very simple antennas with simple
geometry can be analyzed using approximations which have been developed
over the years, but they're usually strictly limited to a narrow range
of conditions. For example, there have been many methods developed for
finding the input impedance of a simple, straight dipole of various
lengths and diameters. (I have a large collection of papers and
references on this topic, accumulated before MININEC became available.)
All are based on approximations, and some are better than others at
certain lengths and diameters. None are terribly good over a wide range,
and bending the elements, for example, invalidates any of the methods.
Now, why should we expect results from these methods to be better in
some way than results from NEC, "just" a computer model, when NEC can
solve the problem for any length, diameter, and shape, to an arbitrary
degree of accuracy, in a fraction of a second?

Roy Lewallen, W7EL