Vertex Reconstructing Neural Network at the ZEUS Central Tracking Detector

Gideon Dror,Erez Etzion

DOI: https://doi.org/10.1063/1.1405272

2000-11-19

Abstract:An unconventional solution for finding the location of event creation is presented. It is based on two feed-forward neural networks with fixed architecture, whose parameters are chosen so as to reach a high accuracy. The interaction point location is a parameter that can be used to select events of interest from the very high rate of events created at the current experiments in High Energy Physics. The system suggested here is tested on simulated data sets of the ZEUS Central Tracking Detector, and is shown to perform better than conventional algorithms.

High Energy Physics - Experiment

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to determine the vertex position of particle collision events (i.e., the location where the event occurs) efficiently and accurately in high - energy physics experiments. Specifically, the paper focuses on the following issues:

1. **High - precision vertex positioning**: In high - energy physics experiments, especially in colliders like HERA (Hadron Electron Ring Accelerator), electrons and protons collide once every 96 nanoseconds. These collisions generate a large amount of event data, of which only a very small part are the signal events of interest. Therefore, a method is required to be able to quickly and accurately determine the vertex position of each event for further analysis or trigger recording.

2. **Improving the efficiency of the trigger system**: Due to the extremely high event rate produced in the experiment, traditional trigger systems (such as hardware - or software - based filtering algorithms) may not be able to meet the requirements of real - time processing. Therefore, a more efficient algorithm is needed to improve the performance of the trigger system and ensure that valuable events can be screened out in a timely manner.

3. **Overcoming the limitations of existing methods**: Traditional vertex reconstruction methods (such as histogram - based methods) have certain limitations in precision. The paper proposes a new method based on neural networks, aiming to overcome these limitations and provide higher resolution and accuracy.

### Specific solutions

The paper proposes a vertex reconstruction method based on two feed - forward neural networks (FFNNs). These two networks are used to process information on two - dimensional planes (xy coordinates) and one - dimensional lines (z coordinates) respectively:

- **The first neural network**: It is responsible for identifying local line segments and arc segments from the input hit - point data and finally constructing a global arc segment. This process is similar to the way of processing image information in the visual system.

- **The second neural network**: Using the results of the first network, it calculates the z - coordinates of all trajectory intersection points, thereby determining the vertex position.

Through this method, the paper shows the test results on the simulated data set, indicating that this neural network method is superior to the traditional histogram method in vertex positioning accuracy. Specifically, the resolution of the neural network method reaches approximately 4.5 centimeters, while the resolution of the traditional method is 8.4 centimeters.

### Summary

The main objective of the paper is to develop an efficient, neural - network - based vertex reconstruction algorithm to improve the precision and speed of vertex positioning in high - energy physics experiments, thereby better supporting real - time triggering and data analysis.