Online Spacepoints in EC fixed

1
Online Spacepoints in EC fixed

• After the fixes were applied, the IDScan
  inefficiency (with respect to EF) has decreased
  by 90, ie. out of an example set of 30
  previously failing electron tracks, 27 are now
  reconstructed.

• The fixes went into release 13. Additionally, for
  those who would like to use it in 12.0.6, they
  have been backported. Please check out
  TrigOnlineSpacePointTool-00-00-23-bf01.
• The fix will effect all slices at varying
  degrees and should help both IDScan and SiTrack.

2
Fixing the Spacepoints Some Details

• E. Özcan
• University College London

3
Symptoms

• Efficiency of IDScan with respect to EF is found
  to be less in EC than in barrel.
• Problem tracked to spacepoints.
• Comparison with offline SPs yielded two issues
• Smaller problem Online SCT clusters a slightly
  off compared to offline SCT clusters.
• Dominant problem Even if offline clusters are
  used, online SP formation creates different SPs
  than offline algorithm does.
• r and f coordinates are quite off.
• Occasionally no online SP corresponding to
  offline.

4
Before the Fix
Crosses Offline SPs Colored points SPs created
by the online algorithm from offline clusters To
see the effect of the fixes easily, follow, for
example the pink points.
5
Step 1 Fix f

• Lines along the strips do not intersect on the
  z-axis. (Strips are not really radial.)
• Result can be modeled by a small rotation around
  the center of the module (MC).
• Fix Create SP as usual. Then find the 2D local
  vector connecting MC to SP. Rotate that vector by
  the rotation angle. Revert back to global
  coordinates.
• double xsp rspcos(phisp), ysp
  rspsin(phisp)
  
  
  xsp - posPhi.x() ysp - posPhi.y()
• double tlsp atan2(ysp,xsp)
• double rlsp sqrt(xspxspyspysp)
• xsp rlspcos(tlsprotAng)
• ysp rlspsin(tlsprotAng)
• xsp posPhi.x() ysp posPhi.y()
• sp-gtphi(atan2(ysp,xsp))

posPhi Position vector of the center of the f
module. rotAng Angle between posPhi and
longitudinal local axis on the f module.
6
After the f Fix
Crosses Offline SPs Colored points SPs created
by the online algorithm from offline
clusters Note the orange points at the top,
which were off the figure before. All SPs have
roughly aligned at the correct f.
7
Step 2 Fix r

• r coordinate of SP calculated with
• Here a is the stereo angle, obtained from
• sina uv_element-gtsinStereo()
• alphaasin(sina)
• Unfortunately, when there is misalignment, this
  is incorrect! Fix by finding the angle between
  the transverse axes of phi and uv elements.
• const Hep3Vector dirT phi_element-gtphiAxis()
  
• const Hep3Vector dirT2 uv_element-gtphiAxis()
• double rotAng -asin((dirT.getX()posPhi.x()
• dirT.getY()posPhi.y())/posPhi.perp())
• alpha asin(dirT.getY()dirT2.getX()- dirT.ge
  tX()dirT2.getY())

10-20 effect in a causes r to be miscalculated
by centimeters!
8
After the f and r Fixes
Online SPs (colored points) in very good
agreement with offline SPs (crosses). Gray
crosses right in the middle (indicated in red
circle) gt One online SP is missing. Track (red
V) has changed. Previously no L2 track, now there
is one.
9
Step 3 Find missing SPs

• Disc 8 Inner ring is missing and middle ring
  is populated with short-middle modules.
• SiDetectorElementcenter() returns center of
  active area (red cross), not the point around
  which the stereo rotation is done (black/white
  circle, Crot).
• Major effect gt Essentially no SPs from disc 8
  middle ring.
• Fix Reduce rF (r of MC)

Crot
MCF
MCS
?correction?
rspRphiB if (maxLocLlt35 Rphigt400) //
short-mid strips double posDiff
sqrt(pow(posPhi.x()-posStereo.x(),2) pow(posPhi.
y()-posStereo.y(),2)) rsp(Rphi-posDiff/2./sin(
fabs(alpha)/2.))B
Crot
10
After Algorithm Fixes
Online SPs (colored points, created from offline
clusters) in very good agreement with offline SPs
(crosses). Gray crosses right in the middle gt
Now accompanied with cyan points!
11
Back to Clusters

• Problem in clustersgtSPs solved.
• So go back to investigate differences in
  clusters.
• Online and offline clusters have different
  coordinates even when they are made of exactly
  the same strips. This difference is usually on
  the order of 0.02-0.05 mm (less than 1
  pitch-length, surprising, but not significant).
• For strips on disk 8, difference much more
  significant Can be as high as 2.2mm or more!!!
• Source of problem Disk 8 middle ring is short
  module centers at higher radius compared to
  other rings. gt Different pitch length needed
  (instead of default one used for all middle-ring
  modules).
• Fix 2D array of pitch lengths filled at
  initialization

for (int disk0 disklts_ndisk disk) for (int
ieta0 ietalts_neta ieta) InDetDDSiDetecto
rElement element man-gtgetDetectorElement(2,
disk,0, ieta, 0) if (0element) continue
m_forward_pitchdiskietaelement-gtphiPitch()
12
Fixing the clusters
Left Disk-8 SPs created from online clusters
(squares) are significantly off from offline SPs
(crosses), before the cluster fix. Right All
online SPs, created from online clusters, are in
perfect agreement with offline SPs, after the
cluster fix.
13
Conclusion

• Tested on 30 events from Monika, where IDScan
  failed, but EF found a track.
• After fixes, 27 have tracks reconstructed.
• Likewise, studies by Mike Flowerdew with Z-gtee
  events immediately shows the improvement.
• The fix is in release 13, there also exists a
  backport that can be used with release 12.
• IDScan to be tuned to make best use of the fix.
  Preliminary studies show a 50 decrease in fake
  rates for the same efficiency.