Speaker
Description
Evaluation of the impact parameter in a single event is crucial for correct and efficient data processing in collision-based nuclear and particle physics experiments. Real-time estimates of the impact parameter allows experimentalists to preselect the most informative events at the data acquisition stage, before any processing. Here we consider a number of model setups to check whether a neural network can evaluate the impact parameter from the spacial and time-of-flight data collected in real time by a set of inexpensive microchannel plate ring detectors.
We evaluate several detector geometries, including the geometries considered for SPD detector [1] at NICA, and several neural network architectures.
We have shown that even low spacial resolution detectors in realistic geometry would make it possible to separate low $-$ less than 6 fm $-$ impact parameter events from other collisions with 84% probability.
The analysis of the full $4\pi$ geometry would rise the probability of the low impact parameter collision identification to 97%. Appropriate usage of the time of flight information is crucial to obtain these results. Without time information the quality of identification of low impact parameter events does not exceed 64%, with especially high contamination from high $-$ greater than 12 fm $-$ impact parameters.
The presented computational experiments prove application of neural network techniques for direct impact parameter evaluation useful for future experimental setups.
This work is partially supported by Russian Foundation for Basic Research grants 18-02-40104 mega and 18-02-40097 mega.
- V.M. Abazov et al., arXiv:2102.00442v2 (2021).
