I am using EZNEC+ 4.0 and I am trying to model the effect of
increasing the spacing between the inside ends of a simple split dipole.
I can’t figure out how to place the source so it feeds the first
segment of each of the inside ends of the two wires that constitute
the dipole. I thought this was what a split source was, but after
trying that and reading the help file, I see I had that all wrong.
Is there a way to do this?

1. Put a source on each segment - which puts it always in the middle of the segment
2. Use a split source - which does the same for you except that it assumes that the two legs are connected.
3. Put a short wire between the two dipole wires and put the source on that wire. The source will be in the middle of this added wire.
4. Make the number of segments high enough that the sources are very close to the actual wire end. You DO have to make sure that you don’t make them TOO short. Some recommendations suggest that segments should be at least 4 times the radius of the wire used.

NONE of the options does precisely what you want, but you should be able to get some feeling for the impact. The biggest problem to me (beyond NEC2’s placement of the source) is that the individual segments are likely to be large enough to mask the impact of separating the two legs of the dipole.

Remember, the whole thing is an approximation to the solution of the
integral equations governing the creation of the radiation field. The
assumptions are good in general for most large scale antenna problems, but the effect of changing the gap might very well differ from the calculation results because the scale of the change is too small.

One other possibility is to model the antenna fed by a balanced feed line from a source at a distance from the antenna. Even that’s not guaranteed to give you good answers, but they should be comparable.

What you might try is use several of these options and compare the results. If you try option 4, try increasing the number of segments and running the model at smaller and smaller segment sizes and see what happens to the solution. Does it converge? At some point does making the segments smaller lead to a divergence in the results?