CMS Tracking POG Planned Improvements for CTD 2015

New Developments for Run II

In order to face the new PU challengs for Run II, two new algorithms has been implemented: deterministic annealing filter for tracking reconstruction and cellular automata for tracking seeding. Preliminary results are shown below.

Deterministic Annealing Filter

The track parameters estimation by error minimization is made through a fit method. The Deterministic Annealing Filter is a soft assignment fit method, where several competing hits can contribute to the same track with different weights.
The implemented algorithm is basically an iterated Kalman Filter, where the wrong hit assignment to the track is avoided introducing an annealing factor which changes iteration by iteration.

Plot	Description
	Final distribution of weights for competing hits selected by the DAF algorithm. The distribution is shown for different samples: single muon (red), ttbar(green) and QCD jets(blue). Structures are visible around the values 0.3 and 0.5 due to the annealing scheduled chosen (30.0, 18.0, 14.0, 11.0, 6.0, 4.0, 2.0, 1.0).
	Final distribution of weights for competing hits selected by the DAF algorithm. The distribution is shown for different samples: single muon (red), ttbar(green) and QCD jets(blue). The properly tuned annealing schedule (15.0, 10.0, 6.0, 4.0, 2.0, 1.0, 1.0, 0.5, 0.1) shows a smooth distribution.
	The DAF algorithm uses the competing hits by including them in a single virtual hit, called MultiRecHit (MRH). This plot shows the distribution of the number of competing hits inside a MultiRecHit for different samples: single muon (red), ttbar(green) and QCD jets (blue).

Cellular Automaton Seeding

In a Cellular Automaton (CA), a network of cells evolves in discrete time steps from an initial state according to predefined rules, depending only on the values of the cells in the local neighborhood.
In the implementation of this algorithm in the tracking seeding context, the following consepts have been defined:
* cell: as a segment linking three hits in different layers (different configuration are chosen for the three pixel layer and the two innermost ones in the strip detector).
* neighborhood: two cells are considered neighborhood if they have in common pair of hits and similar eta.
* evolution rule: at each time step a cell increases its state if on its left it has a neighbor with the same state.
At the end of the network evolution, the neighbor fit triples are joint in a longer seed.

Plot	Description
	Full tracking efficiency as a function of the pseudorapidity for simulated ttbar + = 15 (with a bunch crossing of 25 ns) sample using different seeding in the first iteration: the standard one (blue) and the Cellular Automaton one (red).
	CPU time per event for the first iteration using the standard seeding (blue) and the Cellular Automaton one (red) for a simulated ttbar + = 15 (with a bunch crossing of 25 ns) sample. The x-axis bins show the time spent for the tracking sub-steps (seeding, track finding, fitting, final selection) as well as the total time for the iteration step. The additional time spent for a CA seeding is more than gained in the subsequent pattern recognition.

-- EricaBrondolin - 2015-02-03