Convergence testing

The fact of having a small relative erro in a score does not always guarantee that it has converged to a good result. This may be specially true when there are contributions of very different weights to the scorer, and the high weight scores have not been properly sampling with your statistics. If you suspect of a wrong behaviour you can activate the convergence test with the parameter (as these tests consume some CPU time and memorey, bu default they are deactivated):

/gamos/setParam SCORER_NAME:ConvergenceTester CONVERGENCE_NAME

where SCORER_NAME is the name of the scorer and CONVERGENCE_NAME is the name you want to give to the convergence tester, which will be used in the report. If you are using a point detector scorer, you should substitute SCORER_NAME by GmPDS.

The convergence tests are taken from the Geant4 code, and are inspired in the MCNP tests. The tests are based in the analysis of the sum of scores at the end of each event. The following variables are printed about the sum of scores :

To get a feeling of how big are the fluctuations the same variables, i.e. mean, variance, R, shift and FOM, are printed again but adding to the values a new one equal to the largest value (so that this value is counted twice). And also the ratio of thies affected to the original ones.

Then the results of eight convergence tests are shown:

Finalyy it prints the evolution of several variables, i.e. the cahnge of these variables when more events are added. The variables are printed each N/16 events, where N is the total number of events. The variables printed are the following: