Hi Urs,
yesterday I spent some time scanning the n00 and o00 root files and I
found that:
- event selection is basically identical
- mxhad differs if an event has a kaon in one case and not in the
other one
- mxhad is identical if the number of identified charged kaons is
the same
- the number of events with a changed kaon is similar
- data are identical (only MC changed)
It looks like the kaonID in MC was different and I agree with you that it
looks like the random seed was different. I also tried running a few
events over and over again with different seeds and the order of magnitude
of the changes is the one observed between n00 and o00.
I support your interpretation of the changes as MC statistics
fluctuations, probably due to this anaQA.cc incomprehentions. I am glad
that I managed to recover your result to the last digit with VubAnalysis
ursl-032803. We should probably rename it PRL-tag ...
(are your scripts already in VubAnalysis? otherwise I would add them
before making the very final tag)
ciao
ric
On Mon, 31 Mar 2003, Urs Langenegger wrote:
>
> Hoi,
>
> I looked a bit into this.
>
> 0. The n00 jobs were run Jan 28, the last three crashed jobs ended
> very early on Jan 29 (that day was bad for computing). The next
> batch of jobs n01 (with modified recoilNtp.cc) did not start
> running before Jan 29 afternoon. I still think that I was careful
> not to mess up the executables.
>
> 1. I started with VubAnalysis ale-012303 and RecoilAnalysis
> ursl-011403 as the closest tags before running the n00 jobs and ran
> on csx-bpcock-2002. I could not reproduce the results of n00.
> (results is p.d. the number of events with IERR < 0).
>
> 2. Just to check, I put back in the K-lepton charge correlation test.
> No reproduction of n00. Again, I think that I was careful back in
> January, despite the usual high-pressure zone approaching from the
> east ....
>
> 3. I looked at CVS. I see that on Thu Jan 23 07:29:03 2003 someone
> committed a version of anaQA.cc that re-introduced
>
> gRandom->SetSeed(processID);
>
> instead of
>
> gRandom->SetSeed(randomSeed);
>
> into the executable. This is something that had sneaked in
> previously and had been fixed before (read the commit messages of
> Nov 28 01:28:41 2002, and also Wed Dec 11 01:28:25 2002) ...
>
> I fixed this once again in my commit of Thu Jan 30 00:10:16 2003,
> but I do not know whether that affected n00. I guess not.
>
> Therefore I currently think that n00 was running with a "random"
> random seed (and the printout in the logfile is not relevant, as you
> may see when perusing CVS web or the source of the correct tag). I
> don't know how to find this out without a timestamp of the executable.
> I think I'll add that as a default dump like the env dump we have in
> IbuApp.
>
> Archaeology is a bit difficult under time-constraints (not from the
> east, this time), so I may be wrong. But I don't think so. But if you
> manage to invest your own time and find something else, please post.
>
> I still see no reason why n00 should be better than o00 or p00.
>
>
> Cheers,
> --U.
>
>
|