Studies on the neural z-Vertex-Trigger for the Belle II Particle Detector

Sumitted to PubDB: 2018-11-19

Category: Master Thesis, Visibility: Public

Tags: -

Authors Fernando Abudinen, Christian Kiesling
Non-Belle II authors Jochen Schieck
Date Jan. 1, 2014
Belle II Number BELLE2-MTHESIS-2018-007
Abstract The electron-positron particle collider KEKB together with its Belle detector in Tsukuba, Japan are currently being upgraded. So far, KEKB held the world record luminosity of 2.11x10^{34}cm^{-2}s^{-1}. The new collider, SuperKEKB, will reach a 40 times larger luminosity than its predecessor. Additionally, a new high-resolution Pixel-Vertex-Detector will be built into the new Belle II detector. In consequence, a higher data production and a better vertex resolution will be achieved. This will constrain the unitarity triangle and the search of new physics. Because of the new beam parameters, the load due to undesired scattering process, e.g. Touschek scattering, will be larger by a factor $30$. The scattered particles produce background signals inside the detector. Since the enormous data production of the detector exceeds any data transference capacity, the background has to be filtered out in advance. From the Belle experiment, it is known that these background events do not come from the collision point. Therefore, a good possibility to reduce the background could be to identify the tracks of charged particles and to reject events which do not stem from the collision point. For a sufficient background suppression, a resolution of about 2 cm is needed.Since the decision has to me made in real time, there is not enough time for usual track reconstruction. The most promising method up to now has been a neural z-vertex trigger composed by several multi layer perceptrons (MLP). Previous experiments were based on idealized simulated events within narrow detector regions using single particle tracks, but without background. In this work, the MLP has been optimized for specific detector regions. The networks were trained and tested with more realistic data samples including background. On the basis of the achieved results, the total amount of required MLPs to cover the whole detector could be approached to about 2x10^6. The lowest and the highest resolution reached, using simulated events under the presence of background, is sigma_{max} = 1.70 +- 0.02 cm sigma_{min} = 0.85 +- 0.04 cm. Thus it could be shown that, also under the presence of background, the neural z-vertex trigger is capable to reach the required resolution of less than 2 cm.
Conference Munich

Files