New Run IIb L1Cal EM Algorithm

1
New Run IIb L1Cal EM Algorithm

• Greg Pawloski
• Rice University
• Run IIb L1Cal Meeting
• May 31, 2005

2
The Situation

• The original RunIIb Atlas EM algorithm found
  trigger objects by looking for a cluster of 2
  trigger towers (TTs) that was a local maximum (LM)

• Problem If two electrons are in the blue region
  then only the highest Et one will be found

3
Proposed Solutions

• To overcome this limitation a new style of
  algorithm was designed that uses an Object
  Inhibiting Mask instead of LM cuts
• 3 test algorithms were created
• Searches for a seed TT by requiring the TT to
  have the highest Et in a 3x3 TT area and then
  finds the highest Et nearest neighbor to form a
  cluster.
• Searches for a seed TT by requiring the TT to
  have a higher Et than its four nearest neighbors
  (a cross pattern) and then finds the highest Et
  nearest neighbor to form a cluster.
• Searches for entire 2TT clusters by looking for
  vertical and horizontal objects separately.
• Each test algorithm has a potential weakness (see
  next slides)

4
Potential Flaw in Algo 1

• The test algo that use a 3x3 search window to
  find seed TTs still has some cases where it will
  not find two neighboring electrons

Seed TTs must pass this condition
5
Potential Flaw in Algo 2

• The test algo that use a cross pattern search
  window to find seed TTs can form an object out of
  neighboring noise instead of the true signal

Seed TTs are found here
For the following configuration of TT Ets
The inhibiting mask prevents 2 neighboring
horizontal objects from forming
Only the yellow trigger objects are found
6
Potential Flaw in Algo 3

• The test algo that searches for entire trigger
  objects will occasionally form a trigger objects
  that contains the true hot TT, but not the
  correct nearest neighbor

Lots of Et now in isolation region Isolation cut
will now fail for this electron
For the following configuration of TT Ets
Only this trigger object is found
Algo finds a horizontal 2815 object and a
vertical 281 object. The mask removes
overlapping objects by choosing the vertical
object
7
Choosing a New EM algo

• Conditions for choosing a new EM algo
• The new algo must reproduce the old Atlas
  algorithms results when there are no neighboring
  electrons
• This will be tested with the CSG Z?ee Data Sample
• The new algo must out perform the Atlas algorithm
  when there are neighboring electrons
• This will be tested with the CSG J/Psi?ee Data
  Sample

8
Z?ee Data Sample
Et only cuts
The new algo w/ the 3x3 search window reproduces
the Atlas algorithm exactly The others are less
efficient
9
Z?ee Data Sample
w/ isolation and em fraction cuts
Again, the new algo w/ the 3x3 search window
reproduces the Atlas algorithm exactly
Trigger object w/ wrong NN kills isolation
efficiency
10
J/Psi?ee Data Sample
Et only cuts
For the same rate the new algos are more
efficient than the Atlas
For the trigger term CEM(2,3)
Atlas is 74 efficient The
new algos are 94 efficient
11
J/Psi?ee Data Sample
w/ isolation and em fraction cuts
For the same rate the new algos are more
efficient than the Atlas
12
Conclusion

• The test EM algo that uses a 3x3 search window
  would be the best choice for a new EM algorithm
• It reproduces the old Atlas when there are no
  neighboring electrons
• It out performs the old Atlas algorithm when
  there are neighboring electrons
• This algorithm is also known as Algo 4 on the
  DØ Run IIb Level 1 Calorimeter Trigger website
  http//www.nevis.columbia.edu/evans/l1cal/algos/e
  m_algo/working/em_comp.html