Code is at:

/star/u/rjreed/PPV2009/v3/

Issue:  Converges for:

   if (dT2<dmax2)      chi2+=dT2/t->ery2;    else      chi2+=dmax2/t->ery2;
    if (dZ2<dmax2)      chi2+=dZ2/t->erz2;   else      chi2+=dmax2/t->erz2;
 

Does not converge for:

      if ((dT2<dmax2)&&(dZ2<dmax2))          chi2+=(dT2/t->ery2+dZ2/t->erz2);
      else          chi2+=(dmax2/t->ery2+dmax2/t->erz2);
  

But, this statement is supposed to be for pile-up rejection.  One would naively think that a pile-up track is a pile-up track and if it fails either if statement, it should be thrown out.  My tentative hyposthesis is that by connecting the XY plane directly to the Z, the likelihood space becomes just bumpy enough that minimization is difficult.  I have also seen the same problem that Jan did with attempting to fit for all four parameters.  (i.e. it stops in strange places....)  Some MC sets do allow the second fit to work and converge to a sensible answer, but not all.

Beamline parameters where Vx,Vy and Ux,UY were fit seperately.  I used the first (convergent) set of if statements as well.

Errors need to be determined by looking at the various likelihood cross-sections.  This is in progress.

Figure 1:  Vx,Vy at Z = 0 for the beamline both truncated likelihood cases.  Each point represents one fill.  Blue pionts are much closers together and tighter packed.  Minuit converged for every blue point.  It did not converge for any red points.  The average difference between the points found with the two methods is 0.02.

In any case, if Minuit doesn't converge the answer is not sensible.  Obviously there must be something else going on with the likelihood space that when we connect the XY plane to the Z, it becomes a problem.  Solution should be to get a working robust potential, which will take care of the outliers without the if statement.  However, given time constraints, perhaps we should use these beam-line parameters while working in parallel.