Hi In brag we use 2 sets of random numbers. The first that has been initialised diferently all the time is for the internal GEANT stuff = physics, while the other set is used for the input and vertex distributions. This means that in the files Djamel has generated he will in principle have the same input distribution for all XXX files but different "interaction" parameters. This is ofcourse pretty bad for the acceptance calculations, but when you average over a lot of settings and bins this effect will be reduced, so I agree with Flemming that this was/is not our biggest problem, but it is a problem that is easy to solve;) Cheers Peter P.s : I made some plots a while ago when I dicovered the problem : http://www.nbi.dk/~pchristi/testAcc.ps (with initialised) http://www.nbi.dk/~pchristi/testAccOld.ps (without initialised) The idea is to double the number of thrown particles each time : 1) 125K 2) 250K 3) 500K 4) 1M 5) 2M This is done in 4 files of 500000 thrown, so you should see the acceptance improve in the first 3 cases for the Old after which it does not improve. Pic 1 : 2d acepptances Pic 2 : 1d projection or slice (can't remember) Pic 3 : Number of accepted counts Pic 4 : Calculated errors Pic 5 : Estimated errors = deviation from best Hope this helps! On Wed, 18 Sep 2002, Flemming Videbaek wrote: > > Hi Djam > > You are certainly right that brag-1-1-12 used guxint 1.7 which had the PC > changes > overwritten. I will like Peter to comment specifically on the validity but I > would guess the errors on the acceptance map > is de facto larger since the additional files generated just adds to the > 'same' content after a while - > but this is probably NOT the dominating error in the analysis at this time > (statistics, cuts, effeciencies.) > > > > > This sounds not good at all. I generated the FS acceptance maps after you > > had improved the random generator intialization but I used brag-1-1-12 and > > the stuff you added was gone (as early as brag-1-1-8). Now, I wonder about > > the validity of my simulation... > > > > This clearly means that some people never update first their software > > before committing. This update is important since some conflicts are > > detected at this stage. Christian described the full procedure elsewhere > > (brat guide). This is not the best piece of literature ever written but > > please, read it at least once if you do care about the experiment. > > I agree whole hartedly- this is also why the versioning and tagging of code > + knowing what you used is paramount. > As a second comment - Hiro was not to blame for the change in guxint.F - but > someone else, in any case for all please recall the lesson. > > Flemming > > ------------------------------------------------------ > Flemming Videbaek > Physics Department > Brookhaven National Laboratory > > tlf: 631-344-4106 > fax 631-344-1334 > e-mail: videbaek@bnl.gov > > > -- :-) --------------------------- )-: Peter H L Christiansen @ NBI EMAIL : pchristi@nbi.dk OFFICE : Tb1@NBI (353 25269) HOME : Hjertensfrydsgade 3, st PHONE : 33330493(New)/ 40840492(mob.) :-D --------------------------- \-:
This archive was generated by hypermail 2b30 : Thu Sep 19 2002 - 03:48:25 EDT