Prof. Dr. Stephan Paul

Lehrveranstaltungen und Termine

Ausgeschriebene Angebote für Abschlussarbeiten

Implementation of a Bayesian Particle Filter for Event Reconstruction on GPUs

The Multi-purpose Active-target Particle Telescope (MAPT) is a radiation detector for space applications currently under development at our institute. The detector shall be used to monitor the radiation environment on spacecraft and satellites. It is most sensitive to low-energy protons and ions and can distinguish radiation particles by their interactions with the material of the detector.

To calculate the flux of particles from measurement data, the direction, energy, and species of the detected particles must be reconstructed. We use an algorithm called particle filter, which is the optimal mathematical solution of the given task but requires a large computational effort. To use the algorithm efficiently, we need an implementation of it that can be run on a graphics processing unit (GPU), allowing a massive parallelization of the underlying ray-tracing problem.     

In this thesis, a GPU-based implementation of the particle filter shall be developed and implemented. The algorithm shall first be tested with simulated detector data and later verified with experimental data from an experiment conducted at a high-energy particle accelerator facility.

 

Tasks

• Acquire the necessary theoretical understanding of interactions of particles with matter and the particle-filter algorithm.
• Implement the particle-filter algorithm on a graphics processing unit and verify the algorithm’s ability to reconstruct particle tracks.
• Measure the performance of the developed reconstruction algorithm.
•  Integrate the reconstruction algorithm into the MAPT analysis framework.

 

Prerequisites

Experience in C/C++ programming and CUDA is helpful, but not required.

Contact

Thomas Pöschl, Room PH1 3257, Thomas.poeschl@ph.tum.de

geeignet als

• Masterarbeit Kern-, Teilchen- und Astrophysik
• Masterarbeit Applied and Engineering Physics

Themensteller(in): Stephan Paul

Search for New Particles at the COMPASS Experiment

In nature, quarks and gluons cannot exist as free particles and are always confined into hadrons. Unfortunately, the equations of the strong interaction that governs the behavior of quarks and gluons cannot be directly solved at energy scales where hadrons form. Hence theoretical predictions of hadron properties like masses and decay modes are very difficult. This is in particular true for hadrons that are made of the three lightest quarks: „up“, „down“ and „strange“. Also on the experimental side much confusion exists on what concerns masses, decay widths, and the assignment of quantum numbers of some observed states, let alone the interpretation of their internal structure.

Our group participates in the COMPASS experiment at CERN, where we study the production of mesons (hadrons with integer spin) in the scattering of a 190 GeV pion beam off a stationary proton target. The produced highly excited mesons have extremely short life times of the order of 10^-24 seconds and can hence be measured only via their decay products. Our group employs and develops mathematically involved analysis tools in order to identify the produced mesons with high sensitivity and to measure their properties with high precision. We have recently found a new meson with surprising properties and also did groundbreaking work on precision measurements of the properties of the pion. This science is directly connected to the understanding of the strong force at large distances and low energies and the search for new forms of matter and therefore addresses one of the last open questions of the Standard Model.

This science project, involves the use of several state-of-the-art technologies:

• Building of analysis models in close collaboration with theorists
• Handling of very large data sets (several Petabytes)
• Elaborate data fitting with more than 1000 parameters
• Use of large computing clusters (200 cores at E18 + 2000 cores at LRZ/Excellence cluster)
• Software development to exploit new CPU/GPU technologies

In addition, we operate and maintain a large-scale scientific apparatus, which involves tasks like:

• Calibration of particle detectors
• Adaptation of large simulation codes

We offer thesis topics at various levels of difficulty, which cover a wide range of subjects in strong-interaction physics, statistics and/or computer science. They include for example

• Data handling and event selection
• Simulation of high-energy scattering processes and detector response
• Model building and model selection
• Estimation of model parameters from high-dimensional data
• Parallelization of analysis software

You have the opportunity to perform your own science analysis, e.g. searching for new particles and determining their properties. The analysis work can start at different levels within the analysis chain: from the selection of a reaction process itself in order to study the feasibility of a full-fletched analysis up to the final fitting process leading to scientific publications. The topic can be chosen according to personal preference and interest. Thesis projects usually involve travels to CERN.

Experience in programming (C++, Python) is helpful, but not required.

Contact: Boris Grube bgrube@tum.de, room PH1 3574, Tel. 089 289 12588