1. A Physics Analysis of a Particle Flow Algorithm
For Use with a Silicon Detector
Keith Fratus
Stanford Linear Accelerator Center
August 14th, 2008

2. The Large Hadron Collider:
Good for the time being
Collides protons at a center of mass energy of 14 TeV
Hope to see Higgs, along with other physics effects
Precision measurements limited by composite nature of hadrons
Less chaotic “events” are desirable
Photo Source: New York Times
The solution is...

3. The International Linear Collider
Proposed Linear Collider
Will collide opposing beams of electrons and positrons
500 GeV initially, with eventual upgrade to 1 TeV
Considerably less energy than LHC, but events are much “cleaner”
Must be linear, because of increased losses to synchrotron radiation due to light electron mass
Photo Source: lbl.gov

4. What type of detector? Several Choices
Linear Collider Detector (LCD) team at SLAC investigates this issue
SLAC, among other institutions, advocates a silicon based detector, referred to as “SiD”
Si/W based calorimeter, with a completely Si based tracker
Past experience suggests issues with other detector designs
Photo Source: linearcollider.org
Competing designs include ILD, with a gaseous tracker

5. How do we pick a detector?
Simulations!
Want to prove that certain benchmark physics properties are attainable to within some desired level
Design that produces best physics results with lowest cost will be chosen
Simulation issue becomes very non-trivial when the volume of data approaches that encountered at a particle accelerator
Photo Source: JAS3

6. How it works... 1.) Event Generation 2.) Simulated Detector Response
3.) Reconstruction
4.) Physics Analysis
Photo Source: Wikimedia Commons

7. My Work Verifying that step three provides results usable in step four
Lots of software, no hardware
Some software written on my own, other software already developed by others
Complexity of problems found dictates my ultimate involvement with them
Does not provide the sort of results that a “typical” research project would
Years of software development can ultimately lead to no actual physical realization
Photo Source: JAS3
Trade physical complications for software complications -> more annoying!

8. The Software To Be Tested
Three Packages for event generation and reconstruction
Using 500 GeV ZHH decays, 500 event samples
Only concerned with output of these packages
Most basic is “Fast Monte Carlo” simulation and reconstruction
Ron Cassell's PFA
Matt Charles' PFA
PFA is “Particle Flow Algorithm”
Photo Source: JAS3
JAS3

9. The First Step: LCFI Vertexing Package
“Linear Collider Flavor Identification” is a collaboration of researchers from several British universities, primarily Oxford
Identifies hadronic “jets” based on list of reconstructed particles
Hadronic jets help identify the quark structure of an event
Set up and monitor jobs on a remote Unix machine
Generates a data file with jet collections appended, for use with the next step
Photo Source: JAS3

10. The Next Step: The SiD ZHH Analysis Tool
Tool developed by mentor, Tim Barklow
Reads in jet information, and attempts to identify quark flavor by a “neural net analysis”
Focuses on events where the Z Boson decays to a quark/anti-quark pair, and both Higgs Bosons decay to bottom/anti-bottom pairs
Batch jobs submitted on a remote Unix machine
Generates data that can be used in a histogram to show quark content
Also reveals confidence in quark content
Photo Source: fnal.gov

11. The Results FastMC ran with no complications, as expected
Sample Histogram
Ron's PFA ran with no complications, and provided reasonably good physics information (not necessarily expected)‏
uds
Matt's PFA died! c
Issue discovered to be bad collection pointers
Issue was partly resolved with code fix, partly by ignoring certain unnecessary data collections b
Neural Net Result
With above fix, reasonable data could once again be obtained
Photo Source: PAW
When viewed in PAW, the data looks like...

12. FastMC reconstruction, with cheating Ron's PFA, with cheating
Matt's PFA, with cheating
FastMC reconstruction, without cheating
Ron's PFA, without cheating
Matt's PFA, with cheating

13. Some Sample Data... Still making extensive tabulations
Can give percentage of quarks definitively tagged as either 0 or 1 in b category, for case of cheating
Matt's PFA: 54.59%
Ron's PFA: 54.51%
FastMC: 55.92%
Verifies that PFA results are reasonable
Photo Source: EMACS text editor

14. For The Future...
Members of LCD are investigating why Matt's PFA has so many pointer issues
Other physics analyses must be deemed compatible with PFA
Further refining and development of PFA software

15. Further Reading:
Photo Source: art.com
And....

16. Thanks to Tim Barklow and everyone involved with the SULI program this summer!