Life as a Physicist

I finished giving my SLAC seminar a few hours ago. A bunch of good questions including a suggestion on how to make the Matrix Element calculation more efficient. But one good question that I never felt like I was able to really answer well was how can you prove that the fact you model the single top discriminate at low values means you understand it at high values. In other words can you really say that the fact you model your background well in one regime means you model it in another as well?

That is a tough one. Peskin, who asked this question, pointed out that when looking at something simple like a jet pT you know enough about the matrix element that you can extrapolate it out to very high pT because you understand the matrix element. You can’t do that for one of our decision tree outputs or our matrix element outputs.

We are left with cross checks. This is all we can do. We don’t have a direct calculation between the matrix element that drives a particular process and a particular variable that we are looking at — and so it is impossible to gain an intuitive understanding from the matrix element as to the discriminates behavior.

It isn’t that the cross checks don’t satisfy. It is just that there isn’t an argument you can draw on from first principles. I fear we are in for a lot of this — analyses are moving towards using more and more multivariate techniques and I see this getting worse before it gets any better.