I wrote:
ASCII art fails me, Jack. You can model the coax as a wire to
ground, but that wire has to have all the geometric twists and
turns that your actual coax takes getting to the transmitter.

Jack Twilley:
That'll be a royal pain in the butt, to put it kindly.

That's *real* antenna analysis. Think about doing that
for all the wires in a missile silo to assess EMP
vulnerability and you get an idea about how the big
boys use NEC.

Actually, you can fudge it a bit. Any bump or kink
less that 1/10 wavelength in extent can be approximated
with a chord between its endpoints.

In reality, there is no such thing as "ground" with RF.

This appears to be one of those subtleties that
has slipped past me.

Yeah, the best one can do is equipotential surfaces,
with the hope that one of those surfaces is parallel
to that big hunk of lossy dielectric beneath your
feet.

Interesting. I do actually have two current-mode baluns -- the
feedpoint has eight or nine turns of coax that were wrapped around a
coffee can before being tied down and the coffee can removed, and the
transceiver has three turns of coax through three large rare earth
magnets. How does one model those with NEC2?

As fixed inductances in series with the "coax shield" wire.
I make your first balun to be about 10 uH, which is roughly
200 ohms at 3 MHz. Your guess is as good as mine on the
one with the rare-earth magnets. What you want to look for
is reducing the current in the "coax shield" wire to zero.

73,
Jim, K7JEB