Optics Meeting Nov 19 2024 0100PM ET

From Moller Wiki
Revision as of 22:56, 19 November 2024 by Vdoomra (Talk | contribs) (Agenda)

Jump to: navigation, search

Back to Main Page >> Optics Meetings

>> following meeting

>> previous meeting

Agenda

  1. Tracking Software (David)
  2. C++ Fitting code (Tyler)
  3. Kinematic Factor (Vassu)[1]

Attendance

David, Vassu, Sakib, Tyler, Paul S, Kate

Minutes

  1. (David) Reported on a meeting with a small working group to discuss event-mode (Tracking) software. The meeting is described in some detail here: [2]. Framework will be Podd, and much of the GEM code (decoding, clustering, track finding/fitting) will be based on SBS code; Chandan Ghosh has done the work so far. Has been tested with cosmic data and MOLLER GEM prototypes. For this group's purpose need to know how it will interface to remoll simulation. Idea is that remoll will produce root trees that will have "ideal" track hits (r,phi,z) at the GEM planes (no resolution smearing, no inefficiency or misalignments). A "post-processor" code will than apply resolution smearing, inefficiency, noise (electronic and low-energy gamma, detector misalignment) and produce a new set of hits (r,phi, z) as input to the tracking code at a place where the subsequent track fitting will be identical for simulated and real data. Then this code will produce output tracks with (r,r',phi,phi') as input to the optics fitting code.
  2. (Tyler) Update on status of C++ Optics fitting code. Is available on github. Probably no work on it will happen over the next few months, as Tyler gets established in his new position at Mainz. Code fits data from one foil at a time, assumes that the HoleID branch already filled (so the original sieve hole for each track has already been established). Applies a simple cut on the number of sigma in r away from the mean GEM r to suppress radiated events. Uses the Root MultiDimFit class to apply a chi-squared fit, which easily adjustable fit function. Splits data into a training set and a testing set. However, problem with fit results at the moment (earlier good results may have been overfitting because a bug meant that had used essentially all the data in training set), does not give stable fit results. Discussion - only using < 10K events in present tests, may need higher statistics to stabilize. Maybe need to add (arbitrary) error bars to data points; maybe tweak some of the knobs in the (ultimately Minuit-based) chi-squared minimization.

Meeting link information

See email invitation, or contact David Armstrong, Kate Evans, or Ciprian Gal for Zoom link