Particle Physics Summer Projects

A number of Summer projects targeted primarily at undergraduate students entering their third or fourth year of studies are available through the University of Liverpool particle physics groups. Two separate programs are run, where projects are provided within the group, and in collaboration with host institutes.

Note: The deadline for projects for Summer 2021 at Liverpool has now passed. This page will be updated with details for the 2022 program at a later date

University of Liverpool Particle Physics Summer Projects 2021

The particle physics group at Liverpool is offering six virtual four-week summer projects which will take place between July and September 2021. The projects are listed below and to apply, send a copy of your C.V. and a one page cover letter to Dr. Stephen Farry before the deadline of April 21st 2021. In your letter, please indicate your first and second choice projects. Please also include the names of two referees we may contact.

1. Charged lepton flavour violation at the ATLAS detector

Supervisors: Dr Carl Gwilliam, Dr Matt Sullivan
Number of students: 1
Schedule: 4 weeks between July and September 2021
It is known that flavour is not conserved in the Standard Model (SM), and has been observed in both the neutrino- and quark-sectors, via mixing. It is thus natural to expect charged lepton flavour violation (cLFV) but, despite hints of flavour anomalies at g-2 experiment and B factories, it has never been seen in nature. Stringent limits have been set in the electron and muon channels, but the constraints in the tau case are much weaker. This project will study the extremely rare decay of a tau lepton into three muons, which if observed would be unambiguous evidence of new physics. The student will study tau leptons produced in the decay of a W boson using a combination of data collected at the ATLAS detector in Run-2, and simulation. Physics observables related to the decay of the tau lepton to three muons will be studied in order to separate these rare decays from the huge background present at the LHC. To improve separation, machine learning techniques such as the use of deep neural networks will be utilised.


Supervisors: Dr. Eva Vilella, Sam Powell
Number of students: 1
Schedule: 4 weeks between July and September 2021
UKRI-MPW0 is a High Voltage-Monolithic Active Pixel Sensors (HV-MAPS) prototype, recently designed and submitted for manufacture by the Liverpool HV-CMOS R&D Group.
The student's task will consist of analysing measurement data from the UKRI-MPW0 Data AcQuisition (DAQ) system. This will require writing analysis scripts, preferably in Python or other languages such as ROOT, to extract S-curve data, plot the results and calculate values for the noise and gain of the sensor. Once the sensor has been characterised the results will be compared against design simulations. Previous knowledge of programming would be a plus.
Notes: Previous knowledge of programming would be a plus.

3. Unfolding neutrino cross section with the OmniFold algorithm

Supervisor(s): Dr Ka Ming Tsui
Number of students: 1
Schedule: 4 weeks between July and September 2021
Neutrino cross sections are one of the most important inputs to neutrino oscillation experiments. In cross-section measurements, data must be corrected for detector effects ("unfolded") to physical observables at "truth level" before it can be compared to theoretical calculations and results from other experiments. The traditional methods of unfolding often require the measured observables to be binned in histograms which does not allow multi-differential cross-section measurements beyond two or three dimensions. In this project, we will use the OmniFold algorithm which utilizes machine learning to capitalize on all available information from arbitrarily high-dimensional data and enables simultaneous measurement of all observables. In particular we will apply the method to recent T2K cross-section measurement(s) and compare its performance to the traditional approach. Student is expected to use a lot of python programming in the process, and learn how to train a neural network and make useful physics plots.

4. Quantum Machine Learning

Supervisor(s): Dr Nikolaos Rompotis
Number of students: 1
Schedule: 4 weeks between July and September 2021
It has been claimed in a few particle physics studies that quantum computing algorithms can lead to better performance of multivariate discriminants when the number of available samples for training and testing is limited in comparison to classical techniques (see e.g. [1,2]). It is most probable, however, that this is not generically true [3] and it is certainly not known whether it can lead to any quantum advantage. This projects aims to create a setup in which such claims can be tested in simple scenarios of LHC new physics searches using simulated proton-proton collision data. The development will use python packages like sklearn and the quantum computing implementation will use IBM's quantum simulator python package qiskit.
Note: The student is already expected to be fluent in python programming, including numpy arrays and, possibly, some experience in sklearn. As a test to see whether you are up to scratch consider this dataset here and try to plot a histogram of the variables meeHigh and meeLow using numpy histogram class and matplotlib (unweighted histograms would suffice).

5. Tagging Jets at LHCb using Machine Learning

Supervisor(s): Dr Stephen Farry, Dr Eduardo Rodrigues, in collaboration with LHCb colleagues at Padova
Number of students: 2
Schedule: 4 weeks between July and September 2021
The hadronisation of quarks and gluons at the Large Hadron Collider results in high energy "sprays" of particles known as jets. The identification of these jets is a crucial part of the physics programme of the LHC experiments, and allows events of interest to be separated from large backgrounds. Since this is a classification task, it is also ideal for the exploitation of machine learning techniques, which have become more widely used in recent years. Classical ML algorithms such as Deep Neural Networks and Tree Tensor Networks have proved to be far more efficient than standard methods since they rely on the whole jet substructure.  In the proposed exercise the aim is to identify jets produced by b and b-bar quarks, which is fundamental to performing important physics searches at LHCb such as for example the measurement of the b - b-bar charge asymmetry, which could be sensitive to New Physics.

Two Projects are offered within the LHCb Group on jet tagging using Machine Learning:

Project 1: A new approach is to develop Quantum ML algorithms and measure their performances on the above-mentioned b vs b-bar task, using official LHCb simulated (open) data. Up to now the Variational Quantum Classifier (VQC) approach has been considered, using as input 16 variables coming from the jet substructure. Several circuit geometries have already been studied, and up to now everything runs on local simulators. Libraries from Qiskit and Pennylane have been used.

Our task can be framed in the category of Supervised Learning which consists in a training phase during which a fraction of the full dataset is used to optimize the variational parameters of the VQC by means of a cost function. Once the VQC has been optimized,
the performances of the model are evaluated on the remaining fraction of the dataset. Preliminary results show that QML algorithms perform almost as good as classical ML algorithms, despite using less events due to time constraints. There are several possible new studies starting from the code already developed, going from simple to more difficult:
- Re-do the same exercise by using different methods and compare with the one available;
- Perform the same exercise on QC hardware in order to make a comparison between hardware and simulators, both regarding classification and timing performances, if enough resources can be identified;
- Use the same framework to classify b-quark against c-quark initiated jets.

Project 2: Current b-jet taggers at LHCb proceed first by requiring a reconstructed secondary vertex (or SV) in the event, and the classifiers are trained on the properties of this SV and the jet as a whole. However, the SV reconstruction only has a reconstruction efficiency of approximately 65-70%. For low statistics studies, the loss of efficiency at this stage reduces the statistical power of the measurement. Additionally, the use of both jet and SV information in the tagger biases some measurements studying jet hadronisation. The project aims to produce three different taggers to suit different tasks. First you will aim to reproduce the current LHCb jet performance using the standard technique, then you will develop a classifier that bypasses the secondary vertex requirement, and finally you will train a tagger that only uses the secondary vertex information. You will compare and contrast the performance of the three for the final results.


Projects with International Partners for Summer 2021

Projects are also available for this summer for University of Liverpool students which will be undertaken with collaborators at institutes and laboratories. To apply, please send your CV and half a page of application letter to Prof. Uta Klein by 1.5.2021 .

1. Calculation of the Synchrotron Light Parameters in the LHeC Interaction Region

Host supervisor(s): Dr Bernhard Holzer (CERN accelerator division), Kevin Daniel Joel Andre
Host partner: CERN
Number of Students: 2
The LHeC project studies the design of a new accelerator to provide intense, high energy electrons for collisions with the protons of the LHC storage ring. The responsibility of the two trainees in this international task force will be the calculation of the characteristics of the emitted synchrotron light. The photons are emitted during beam separation in the direct vicinity of the LHeC Interaction Region (IR). As part of a  detailed study of synchrotron light aspects of the lepton-proton collider, the critical energy and power of the emitted light will be calculated and the geometry of the emitted light cone has to be determined.
The work will continue and complete the calculations that have been done by the two recent Liverpool trainees. In collaboration with the experts of the machine detector interface at CERN, shielding concepts like absorbers and collimators will be studied. The task gives you deep insights into general accelerator-detector integration aspects and opens the way for further master or PhD thesis work in the wider field of accelerator, detector and particle physics.

2. Dark Matter searches with ATLAS

Host supervisor(s): Prof. Nuno Castro, Dr. Rute Pedro, Universidade do Minho, Braga, Portugal and  (the “Portuguese CERN”)
Host partner: LIP - Laboratory of Instrumentation and Experimental Particle Physics, Lisbon
Number of students: 1

The current project aims at searching for new physics, namely the hint for dark matter production, by analyzing ATLAS data. Topologies with a boosted top quark and missing transverse energy will be considered and the use of advanced machine learning techniques will be explored.

The expected learning outcomes include: the understanding of an analysis workflow in collider data, as well as the understanding of the motivation for dark matter searches and the technical challenges in the analysis of fully hadronic topologies; the mastering of advanced machine learning tools in the context of high energy physics; and competences in the statistical interpretation of the obtained results.

Previous experience in python and/or C++ is desirable.

3. Deep Neural Networks applications in experimental physics data analysis

Host supervisor(s): Prof. Helena Santos (LIP researcher and professor at the University of Lisbon)
Host partner: LIP - Laboratory of Instrumentation and Experimental Particle Physics, Lisbon
Number of students: 1
Schedule: 5 weeks between June, 7 - July, 18

One of the focus of the Heavy Ion program of the LHC in Run 3 is the study of heavy flavour jets (collimated sprays of particles originating in the hadronization of charm and bottom quarks). Deep Neural Networks to tag these jets in the demanding Pb+Pb collisions environment are under development. The student will collaborate to these efforts by implementing an hyper parameter scanner with the goal of optimising the DNN. The work will be developed online and the student will integrate the ATLAS LIP Group. Working Plan Weeks 1 and 2- Introduction to the problem; familiarisation with Jupyter notebooks and online platforms for data analysis provided by CERN. Week 3 , 4, and 5 - Implementation of the algorithm. Analysis of the DNN performance.
Requirements - some experiencing in programming is desirable. The codes are in Python.

4. Evaluating new methods for identifying Higgs boson decays into a pair of b-quarks at high transverse momentum

Host supervisor(s): Inês Ochoa (LIP), Patricia Conde Muíño (LIP, IST)
Host partner: LIP - Laboratory of Instrumentation and Experimental Particle Physics, Lisbon
Number of students: 1 or 2
Schedule:7.06.2021-18-07.2020 (Flexible).

Since the discovery of the Higgs boson, the focus of the ATLAS and CMS experiments at the LHC has been on precisely measuring its properties. Measuring the rate of production of Higgs bosons at high momentum is particularly interesting, since it is a regime that can be sensitive to physics beyond the Standard Model. This project targets the decay of the Higgs boson to a pair of b-quarks and in the WH production channel. Simulations of WH(bb) production and its main background processes will be used and recent developments in Higgs to bb tagging will be explored, with the goal of pinpointing main sources of background and identifying future avenues for improvement.

Approximate weekly benchmarks
Week 1: Familiarisation with ATLAS, High Energy Physics and ROOT. First analysis of the WH-> lnu bb search with ATLAS Open Data.
Week 2: Benchmarking state-of-the-art algorithms in signal efficiency and background rejection (considering top-pair production and V+jets processes as the main backgrounds).
Week 3: Identification of main sources of background in terms of flavour composition and jet substructure. Comparison of expectations in simulation with real data distributions.
Week 4: Finalisation of the results and preparation of the final presentation.

5. Analysis example for workshop on CMS Open Data for theorists

Host supervisor(s): Dr. Achim Geiser
Host partner: Desy, Hamburg
Number of students: 1
An online workshop on the use of CMS Open Data for theorists is scheduled for this summer. An analysis example based on the latest developments of data formats for the CMS Open Data is to be worked out, e.g. some dilepton mass peak, possibly combined with jets. If successful, this example may then be actually used as a hands-on exercise for the workshop participants.
Requirements: basic familiarity with linux and the Root analysis package, at the level of a basic Root tutorial on a linux platform.

6. Gas-flow and pressure simulations of the GTS-LHC ion source using Molfow+

Host supervisor(s): Dr Detlef Küchler & Toke Koevner (Accelerators & Beam
Physics Group, CERN)
Number of students: 1
The Grenoble Test Source (GTS-LHC) is the ion source that provides the CERN accelerator chain with heavy (usually lead) ions as required for the study of lead-lead collisions at the LHC experiments. The source uses a magnetic field to confine the plasma in a large chamber. As the plasma chamber does not provide a lot of access for diagnostics, like pressure gauges, simulations help to better understand the pressure profile of the ion source and thus to improve the efficiency of producing ions for the experiments. A model of the GTS-LHC ion source to determine the gas pressure inside the plasma chamber already exists. This model needs refinement at the position of the gas injection system. This will require the student to model the gas injection system in a CAD program and import the geometry to the Molflow+ model, where Molfow+ is the tool for gas flow simulations in the molecular flow regime that was developed at CERN. The student will then study the consequences of the refinement on the pressure distribution. Additionally, the model can be used to study the distribution of the lead vapor coming from ovens inside of the ion source. It should be determined how the lead is deposited onto the plasma chamber walls and if it can be expected that lead vapor is also entering the beamline of the accelerator linac3. Tools that will be used: A CAD program, Molflow+ and a data analysis tool of choice.

7. Magnetic field simulations of the GTS-LHC ion source using Radia

Host supervisor(s): Dr Detlef Küchler & Toke Koevner (Accelerators & Beam
Physics Group, CERN)
Number of students: 1
The GTS-LHC ion source provides the CERN accelerator chain with heavy (usually lead) ions. The magnetic field of the ion source influences the properties of the extracted ion beam. Magnetic field simulations as an input for beam simulations can be done, using Radia, a Mathematica package developed by the ESRF in France. A model of the ion source in Radia exists but the export of 3D field maps takes a long time still. A supposedly faster export function using interpolations of the field is in development but needs to be benchmarked and improved. The student will compare the 3D field maps created by the two existing export functions to determine if the output is consistent. Additionally, the student will assess options to improve the existing function or develop a better one. Furthermore, the student will use the Radia model to assess the forces between the different magnetic components, as they could be a possible explanation for previously observed source malfunctions. The outcome of the project will be crucial in the further improvement of the ion beam delivered to the experiments. Tools that will be used: Mathematica, including Radia.