1. CBM RICH ring recognition G.Ososkov, B.Kostenko, A.Solovjev Laboratory of Information Technologies Joint Institute for Nuclear Research Abstract Method of robust fitting with rough center guidance is developed for the RICH1 Cherenkov ring recognition.

2. CBM RICH ring recognition The first attempt to simulate CBM RICH1 data has been made without taking into account the RICH geometry, signal amplitude information and ring center guidance like it has been done for RICH real data of the CERES/NA-45 experiment at CERN. CERES RICH1 event (see NIM A394 (1997) pp 225-231) Two rings from this event extracted by center guidance

3. CBM RICH1 data were simulated in G4CBM framework (IHEP Protvino), without information about signal amplitudes and center guidance. As one can clearly see, the latter is needed especially because of the high density of rings in the central area. CBM RICH1 design 2D plot of one event (axis are in cm)

4. Therefore we continue the development of ring recognition algorithms on the local data models simulating signal amplitudes and combined data contamination by neighboring rings and randomly distributed noise signals. One example of 3D ring image MC radius is 10 (in pad size), amplitude range is (a min =3,300) and two examples of 2D ring images Measurements are in pad size

5. Circle fitting by pads On the first stage we neglected the pad amplitudes and started considering pad centers as points: (x i,y i ), i=1,2,...n to be fit to a circle (R,a,b). Minimized functional L(R,a,b) = Σw i e i 2 with and where c T =5. and then decreased slowly on the next iterations. (See: H.Agakishiev, G.Ososkov at al, “New Robust Fitting Algoritm For Vertex Reconstruction in the CERES Experiment”, Nucl. Instr. and Methods A394 (1997) 225-231) Due to nonlinearity of L(R,a,b) we use, first, the MINUIT program with our robust addition as a benchmark for the next comparison with our faster circle fitting algorithms. As initial parameter values we used either result of the zero iteration with w i Ξ1 or centers of gravity (CoG) of all points, as (a (0),b (0) ), and average distance of them to CoG, as R (0). Some preliminary accuracy results are presented on the next slide (in pad size units) for Δa=a MC -a Cal ; Δ b=b MC -b Cal ; Δ R=R MC -R Cal, where MC and Cal denote Monte-Carlo values and their corresponding estimations calculated on 1000 events.

6. First MINUIT accuracy results (in pad size units) after 1000 events Results of 20 fit iterations (non-robust) Results of 20 robust fit iteraions ΔR histogram with with w i Ξ 1 (right) and robust weights (left)

7. Concluding remarks Robust algorithms of fast ring finding have been elaborated. With the present local model we are going to complete our algorithm of the fast robust ring fit and to calculate its efficiency versus contamination ratio. For further development of the algorithms (in particular, for more accurate consideration of rings' intersections) data of G4CBM simulations including signal amplitudes and ring center guidance are necessary. A problem of recognition and fit of splintered rings corresponding to cases when incoming particle passes between two RICH1 mirrors has been considered. That will require more sophisticated algorithm including a possibility to fit a non-complete ring by its partial arch.