Dear professors,
I am doing the simulation both with fluka and Geant4. As there are no cycles in Geant4, I wonder how to compare the two results using the same statistics. For example, if there are 100 primaries in G4, which is the equivalent way in fluka, 100 primaries in one cycle, or 20 primaries in five cycles?
I notice that in previous discussion, it is recommended to run at least five cycles in fluka to make the data more accurate and get the statistic error. However, this will repeat one initial primary in five times. Even with different random numbers, I wonder if the same primary in five times can actually be interpreted as five primaries.
Thank you for your help!
Dear @hanhechong,
I believe you are mixing two concepts: the statistical significance of the results of the simulations and the estimation of the uncertainty on the results.
If you run 1 cycle with 100 primaries, or 5 cycles with 20 primaries each, of 100 cycles with 1 primary, it makes no difference: the results of this 3 simulations will have the same statistical significance.
What would be instead very different is the estimation of the statistical uncertainty. If you would run only 1 cycle, Fluka would return a 100% statistical uncertainty. The recommendation to run at least five cycles is given because otherwise Fluka won’t be able to estimate the statistical uncertainty. Have a look at slide 28 of the Monte Carlo basics lecture.
1 Like
Dear @amario ,
Thank you for your reply! If I am only dedicated to the statistical significance of the results of the simulations, the three situations you mentioned are equivalent and I actually agree with you. However, as my input are reading sources from a text. Some people argue that 1 cycle with 100 primaries makes use of 100 particles in the sources while 5 cycles with 20 primaries actually run every single source five times. They believe that the first case is much better because it is more “independent”. How to reply to this?
Best regards,
Hechong Han
It depends on the design of the source routine you are using. The latter should read all your text at initialization and then sample randomly among all loaded particles, and not take them sequentially, such as not to always use the first ones and discard the last ones. With this sound design, the cycle “independence” is granted.
Thank you for your reply! Now I understand how to run certain amount of primaries.