Simulation to Determine Optimum Bin Size for Level 1


The basic concept of STAR Level 1 trigger is to make a decision during the drift time of the TPC (100 microseconds) using a limited set of information from the fast detectors (MWC and CTB). The fast detectors digitize their data at every RHIC bunch crossing, and that information is used at Level 0 to make a decision to open the gate on the slow detectors (TPC and SVT) to allow the charge created by ionizing tracks to drift.

The two baseline trigger detectors of interest, the CTB and MWC, cover a limited region in psuedorapidity, from about -2 to 2. The information content from these two detectors is also limited to charged particle multiplicities, but the information is available for each bunch crossing. Therefore the baseline level 0 trigger will act on global charged particle multiplicity distributions while the level 1 algorithms are sensitive to event by event fluctuations in d2N/deta dphi.


The baseline trigger for level 1 uses charged particle multiplicity information that is passed from level 0 in a two dimensional course pixel array. The size and structure of this array for optimum use in selecting anomalous events is the subject of the simulation which will be described below.

The "standard" pixel array was assumed to be an 8 X 4 (eta X phi) array where the central 4 X 4 pixels originate from the CTB and the 2 outer 2 X 4 contain MWC information. Since the CTB is constrained mechanically to 2 slats per tray and 2 trays in psuedorapidity, the central 4 X 4 array remained fixed in the different binning models used for this study. The only time this was changed was when TOF geometry was assumed in place of the CTB in which case the pixel array was an 18 X 1 array, or basically 18 one dimensional pixels in psuedorapidity. The MWC has fewer physical constraints, since the approximately 8000 wires can be segmented into as many pixels, however there are also bandwith limitations with the increasing number of bins being passed from level 0 to level 1. The "standard" bin model used two 2 X 4 arrays, but this was compared to cases with the MWC arrays being 4 X 2 and 8 X 1. The binning models used for this study are represented in the first figure. The total number of bins used was 32, except in the case of using the TOF, where the total number of bins was 34. So there were four different binning models used, three with the CTB and the three different MWC bin arrangements, and one case with the TOF and the MWC arranged in two 8 X 1 arrays.

The events used to test these different models were generated by Ron Longacre and based on the HIJET event generator. A set of 100 standard events were generated (RHIC Events) as well as three sets of 100 events with some random degree of fluctuation introduced, Smoke Events, Landau Events and Chiral Events. There were also twenty sample plasma events generated for testing level 1 algorithms. These twenty events consisted of ten standard HIJET events and ten anomalous events with one or two plasma bubbles of various sizes in them. These events were processed through the STAR GEANT Monte Carlo and files with hits were stored for analysis.

These files were then processed through the trigger detector TAS simulation packages, and then through the level 0 and level 1 packages. The output from the CTS simulation package was ten bit ADC values and from the MWS package the output was eight bit ADC values. These ouputs then went into the rl0 package (level 0) which in turn output eight bit values into the rl1 package (level 1).

The level 1 algorithm used for this particular simulation was the Kolmogorov-Smirnov (K-S) Test, which compares a given distribution with a standard cumulative distribution. For these simulations, the standard distribution was taken as an average of five central Au Au FRITIOF events. The next figure shows the result of the test on the twenty sample plasma events as plotted with the level 1 result against charged particle multiplicity as derived from the level 0 analysis. The points lie in two regions, where the group in the upper region (above .045) contains the anomalous events and the lower region (less than .03) contains the standard events. For the following analysis, three different level 1 cutoff values were used to determine event selectivity, .03, .04 and .05.

The next figure shows a sample result, in this case using 20 one dimensional bins and looking at the Chiral Events. The circles represent the twenty sample events, used here as a guide to compare to the Chiral Events plotted as triangles. Using a cutoff value of .03 one is able to pass 99% of the Chiral events, 98% with a .04 value and 86% with a value of .05. This procedure was followed for all the events and bin models, and the results summarized in a table.


For the unmodified Rhic Events, few are passed regardless of the number of bins used. For the Chiral Events, almost all are passed using the finer segmentation of the MWC (20 bins vs. 12 bins). However one loses infornmation when segmenting the CTB to a full TOF. However one improves seletivity of events with narrow fluctuations when using the TOF and fine pixelation in the central region. One again loses a little with the events with wide longitudinal fluctuations (Landau) when using the full TOF as compared to using the CTB. The loss however is fairly small (68% vs. 60%) and based on this alone would not be an argument against the TOF. In fact, these results point to the advantage of having a TOF since one could use its fine segmentation for certain events and then sum certain bins in the software to get course binning in the central eta region. For the MWC, it seems that for the cases tested, it is advantageous to have the finer pixel sizes. Further tests are needed to determine at what point one reaches in the MWC information where further segmenting yields no new information.