GammaLib

Here a ctobssim analysis that illustrates the problem. The default run gave

2017-08-24T19:18:02:  MC source photons .........: 3991 [Crab]
2017-08-24T19:18:02:  MC source events ..........: 1387 [Crab]
2017-08-24T19:18:02:  MC identifier 1 ...........: Crab
2017-08-24T19:18:02:  MC events .................: 1387 (all models)

while doubling the normalisation gave

2017-08-24T19:18:48:  MC source photons .........: 15639 [Crab]
2017-08-24T19:18:48:  MC source events ..........: 5518 [Crab]
2017-08-24T19:18:48:  MC identifier 1 ...........: Crab
2017-08-24T19:18:48:  MC events .................: 5518 (all models)

This is about four times the number of events compared to the default run.

After removing the multiplication with the normalization factor in the GModelTemporalLightCurve::mc method the results look better.

The default run now gives

2017-08-24T19:29:43:  MC source photons .........: 5269 [Crab]
2017-08-24T19:29:43:  MC source events ..........: 1868 [Crab]
2017-08-24T19:29:43:  MC identifier 1 ...........: Crab
2017-08-24T19:29:43:  MC events .................: 1868 (all models)

while doubling the normalization factor gives

2017-08-24T19:29:59:  MC source photons .........: 10548 [Crab]
2017-08-24T19:29:59:  MC source events ..........: 3689 [Crab]
2017-08-24T19:29:59:  MC identifier 1 ...........: Crab
2017-08-24T19:29:59:  MC events .................: 3689 (all models)

which is a factor of 1.97, hence within the statistical fluctuations close to 2. Note that the default run did not produce the same number of events since the time interval was a bit modified.

Here the code:

        // Compute mean number of times by multiplying the rate with the
        // effective duration. Note that the light curve normalization factor
        // is already included in the effective rate, hence we should not
        // multiply it here again (see #2181).
        double lambda = rate * m_mc_eff_duration;