scalar objective function

One of my soap-box issues (inherited from Roweis) is that whenever you make a measurement with data, you better be optimizing a justified scalar objective function. If you aren't, then you haven't made the best measurement you can, in any sense. Parenthetically, this is one of my objections to the HST drizzle algorithm for making combined images; it has no such scalar.

Marshall and I spent Friday night, past midnight, working on the correct scalar objective function for the automated lens finder. It is not an easy problem; we don't have a final solution, but we learned a lot. Some desiderata: It should be close to zero when lens models don't work well on the source, and it should be close to unity when the lens model under consideration explains most of the image intensity. It should not depend strongly on whether the images have had their foreground and sky subtraction done correctly on large scales, or other details of subtraction of the foreground galaxy light.