GSIM Focus Group

Purpose: to focus on the problem of generating the first publication-quality acceptance calculations with the GEANT-based code GSIM.

 

Core Group Members (33% of time or more)

David Rowntree (MIT) 33%	Jianguo Zhao (MIT) 33%
Kyungseon Joo (UVA) 33%	Franz Klein (JLAB) 33%
Laurent Farhi (Saclay) 40%	Will Brooks (JLAB) 33%
Mike Vineyard (UR) 33% (during summer, starting May 1998)

Contributing Members:

Cole Smith (UVA) ?%	Maurik Holtrop (UNH) 35%
Larry Dennis (FSU) 20% (until fall)	Stepan Stepanyan (YPI/JLAB) ?%
Dave Tedeschi (USC) 25%	Allena Opper (OU) 20%
Hall Crannell (CUA) 20%(summer), 10% during the year	Dan Sober (CUA) 20%(summer), 10% during the year
Burin Asavapibhop (UMASS) ?%	Junho Yun (ODU) ?%
Dennis Weygand (JLAB) ?%	Volker Burkert (JLAB) 10%
John Price (LTU)?%	Bryan Carnahan (CUA) 40%
1 UR student in summer ?%	Alex Vlassov (ITEP) ?%
Andi Klein (ODU) ?%	Rob Feuerbach (CMU) ?%
Hendrik Ayvazian (CUA) 40%Summer	Robert Pollard (ODU) ?%
Alex Skabelin (MIT)	Si McAleer (FSU)
Costy Loukachine (VPI), Summer 98	Steve Dytman (UPitt)
2 MIT students, Summer 98	Alan Coleman (W&M)

Regular `virtual' meetings, using speakerphone and Web-based whiteboard/slideshow program; minutes distributed to the offline and gsim mailing lists.

Web: Hall B home page -> GSIM home page -> GSIM Focus Group home page

 

Recent Accomplishments

• Got time-based tracking to work on GSIM events and implemented a scheme for applying drift time resolution and dead channels to DC simulated data (Burin Asavapibhop, Kyungseon Joo, Dennis Weygand)
• Determined the reasons for slowness of code and devised a multi-track plan to speed it up (David Rowntree, Jianguo Zhao, Will Brooks, Larry Dennis, John Price, Dave Tedeschi, Kyungseon Joo, Franz Klein, Stepan Stepanyan, Volker Burkert)
• Added the photon cryotarget to GSIM (Hall Crannell)
• Determined GSIM cut parameters for optimal compromise between speed and accuracy (David Rowntree, Jianguo Zhao, Will Brooks)
• Completed the first round of multi-million event runs for acceptance calculations (Cole Smith, David Rowntree, Junho Yun, Franz Klein, Andi Klein)
• Designed a web-viewable databasing system for bookkeeping of GSIM runs (Greg Riccardi, Will Brooks, Franz Klein)
• Made a number of improvements to the CELEG event generator, e.g., particle filter, PART bank with selection options, improved kaon simulation, etc. (Dan Sober, Bryan Carnahan, John Price)
• Identified and fixed a number of bugs in GSIM, improved performance and accuracy via coding, e.g., (Maurik Holtrop, Cole Smith, Dan Sober, Alex Vlassov)
• Made a plan for a faster, simpler simulation program (Larry Dennis)
• Performed first acceptance calculations (Kyungseon Joo, Hovanes Egiyan, Latifa Elouadrhiri, Will Brooks)

 

Near-term Projects

• Implement the quality control/bookkeeping system for distributed computations (Greg Riccardi, Franz Klein, Sam Ewell)
• Implement and test the electromagnetic shower parameterization (Dave Tedeschi, Maurik Holtrop, Mike Vineyard, Will Brooks, Serguei Boiarinov)
• Investigate regions of `zero electron acceptance', implement event generator filter if such regions are found (David Rowntree, Will Brooks)
• Investigate optimizing volume organization in GSIM, implement improvements (John Price, Jerry Gilfoyle (?), others?)
• Further investigate the step size parameters in GSIM to see if we can speed the code up without sacrificing accuracy (David Rowntree)
• Fix CELEG elastic scattering cross section
• Add radiative effects to CELEG (Ralph Minehart, Dan Sober)
• Random number generation in GSIM (Maurik Holtrop)
• Add the electron cryotarget to GSIM (Hall Crannell)
• Study the existing high-statistics simulated data to find further problems and to understand known problems
• Prepare for extended computational runs (Cole Smith, Junho Yun, David Rowntree, Arne Freyberger, Larry Dennis(?))
• Produce a prototype trigger simulator (Allena Opper)
• Develop a working version of FSIM (Larry Dennis)
• Add extended target capablility to the event generators

 

How fast is GSIM, and how fast do we need it to be?

 

 

In ideal case:

W bins of 20 MeV, 0.9 GeV -> beam energy

Q-squared bins 0.2 -> 1.1*beam energy, 10 bins logarithmically spaced

Hadronic theta bins - 10 degree bins from 8 to 140 degrees

Hadronic phi bins 20 degree bins from -180 to 180

 

At 1.6 GeV, need 35 (W) x 10 (Q2) x 13 (theta) x 18 (phi) = 81,900 bins

82 kbin x 1 kevent/bin(3% statistics) x 3(electron fraction) = 246 million events

246 million events @ 3.1 second/event = 8800 cpu days

 

8800 cpu days @ 20 dedicated Linux cpu's = 14.7 months

(per energy, magnetic field setting, and event type)

 

Can probably tolerate 0.75 month, => need factor of 20 in:

(number of cpus) x (cpu speed increase) x (increase in code speed)

 

e.g., get 60 Linux cpu's and 10 Alpha's, and speed up code by a factor of 5

=> factor of 20

 

If aim for 10% statistics instead of 3%, above number becomes 6 weeks.

(But need more statistics at the higher energies)

 

=> there probably exists a feasible solution, given more work and cpu's <=

 

 

Method of Calculating Acceptance

• `Acceptance' includes all efficiencies, any well-known correlation effects, etc. Not just geometry.
• Start with the event generator: CELEG or radiated AO (`aao')
• Definitely need events which include all final states in realistic ratios with realistic distributions, particularly for small cross sections
• May also need events which are purely one final state, especially for small cross sections
• Run events through GSIM, then GSIMKO, then TSIM, producing a BOS format file.
• Reconstruct using your favorite reconstruction program.
• Choose cuts and bins, and apply them identically to: a) reconstructed data, b) reconstructed simulated data, and c) thrown data.
• Must bin in small bins in any variable where simulated data distributions do not match real data distributions well (integrated quantities)
• Must bin in small enough bins so that distribution change across bins is approximately linear
• Resolution from simulation must match data well or get divide-by-zeros
• Multiply each data bin contents by the thrown bin contents divided by the accepted simulated data bin contents.

 

Comments on Other Methods for Calculating Acceptances

- Throwing multiparticle final states uniformly

- Throwing single particles, using single particle acceptances with a separate event generator

• Particle correlations affect acceptance calculations
• Effect of correlations on acceptance is minimized when:
• correlations are not strong
• acceptance is large
• backgrounds are minimized
• Using kinematic correlations in the event generator enforces kinematic boundaries via statistical errors from the acceptance calculation. The resulting error estimates are straightforward to interpret.
• Not using kinematic correlations in the event generator produces finite acceptance for unphysical events which may accentuate background events. Cutting these out may prove difficult, especially for complicated final states.
• Any scheme not using kinematic correlations to generate events for GSIM `wastes' cpu cycles by generating events in unphysical regions, and by not generating fewer events in physical regions where there are few real counts.
• Any scheme using single particle acceptances will miss second-order effects such as multiple SC hits, complicated EC hit, two-adjacenttrack events or crossover tracks in DC, etc. May be adequate accuracy for some experiments, but not for all. (Can recover approximate single-particle acceptances from results of proposed scheme).
• Need many bins for single-particle acceptances (e.g., target z and r).
• Schemes using single particle acceptances will save somewhat on overall cpu time, but not dramatically. A better solution for moderate quality acceptance calculations would be to use FSIM.

==> The alternative schemes mentioned are workable, however, the proposed approach is somewhat technically superior.

 

How to Get Acceptance Calculations for Your Experiment

• Let the GSIM Focus Group know of your requirements as soon as possible

 

• Come to one of our meetings to discuss your requirements (can `attend' by phone and Web if you're not local)

 

 

Some Issues for Collaboration Input

• More cpu's needed, at JLAB or elsewhere.
• Still many important GSIM tasks without volunteers.
• Is the present proposed method for generating acceptances optimal and suitable for the collaboration as a whole?
• Will a radiated version of CELEG be an adequate event generator for multiple final states for photon and electron beam experiments? Needs to be developed and improved by the collaboration.
• Next long runs will attempt to accumulate some statistics for electron beam simulation at 1.6, 2.4, 4.0, 40% and 60%B-field; then start on photon beam simulations, plus fill in special simulations as needed.