Subject: The Bugs Crawl In, The Bugs Crawl Out
From: Bob Nolty (nolty@hep211.cithep.caltech.edu)
Date: Wed Aug 02 2000 - 15:44:34 EDT
Hi all --
I have discovered some bizarre behavior which I presume is a bug in
the reconstructions (in DREAM) of the width of strip hits, affecting
all muon analyses. Please refer to
http://www.hep.caltech.edu/macro/protected/notes/weirdStripWidths.ps
(usual account and password). The figure shows scatter plots of strip
hits, with the D-coordinate of the strip hit on the vertical axis and
the width of the strip hit (as reported in the coordinate bank of
DREAM) on the horizontal axis. The first figure, which includes all
the hits in horizontal streamer tube planes 3 through 10, shows the
nominal situation. Most hits are either one strip (about 3 cm) or two
strips (about 6 cm) wide.
The second figure, which includes all the hits in planes 1 and 2,
shows the most serious problem. In these planes, in supermodule 1,
the reported width of the hit depends on which strip is hit. There is
a clear locus of points corresponding to a hit one strip wide in SM1
M1, another locus of points for one strip wide in SM1 M2, another
locus of points for two strips wide in SM1 M1, etc. Hits on strips
near the edge of a module (i.e. short strips) are shown to be less
wide than strips near the center of a module (i.e. long strips). I
guess it is possible there is some real variation that I didn't know
about in the width of strips as installed, but I don't think that is
what this shows -- because for a given location, the reported width
for a 2-strip-wide hit is not twice the width of a 1-strip-wide hit;
it is only about 3 cm greater than a 1-strip-wide hit. So I expect it
is a computational bug, not a database error.
There is a similar effect in the attico horizontal planes (the third
figure), although the amplitude of the effect is much less. Again,
the reported width of hits on strips near the edge of a SM-pair (short
strips) are smaller than strips in the middle (long strips).
In GMACRO simulated data reconstructed in DREAM, the SM1 planes 1-2
effect does not appear. However, the attico planes effect *does*
appear.
I suspect the error is introduced during reconstructions in DREAM, and
is not inherent in the raw event record. As I understand it, the STAS
equipment essentially reports which channels were hit, but does not
speculate on the location or width of the channels. Because the
error is introduced in DREAM, it infects MUADST data and FARFALLA
data (I have verified this).
The behavior is persistent. I have seen it in each of four different
runs I have checked, ranging from run 8534 to run 17668.
This is a serious-looking bug, but I think the effect on published
analyses will be quite minimal. The width of some strip hits is
overestimated, so the relative contribution of those hits to
determining tracking parameters is undervalued.
I have not tried to find the bug in DREAM code, I have merely
investigated the effect.
Bob
This archive was generated by hypermail 2a24 : Wed Aug 02 2000 - 15:44:43 EDT