This post follows on from the one below. The correct lengths to use for chi-squared calc would be the blue ones below:But it would be annoying to have to manipulate the error as it is bound up in cross terms and the like. So we can use the red ones as thier ratio is the same as the blue. We know that the red residual is = blue/sin(theta) and blue is the projection of the 3D poca on the plane perpendicular to the 2D poca, ie blue = sqrt(3DPOCA**2-2DPOCA**2) Implementing this gives a much better looking step.
I should prob say there were a couple of iterations between this and the initial chi-squared, but I'm sure this is the right calculation. Of course there is the further question of what the FORTRAN is doing..... and is it deliberate?