In this project we study the use of neural networks as a tool for particle track pattern recognition with the possibility of its implementation in the Trigger system at the ATLAS experiment [1]. By using a method named Hough transform we created a Convolutional Neural Network (CNN) that is able to train on the transformed images of muons merged with minimum bias. We give an overview of how the CNN works and compare the results from the CNN with the old cut based method. We believe to have managed to find an alternative to the previously used algorithm, that is faster and more efficient.

Uncertainties in radiative effects of the quarks in -background in the form of final state radiation (FSR) are significant when it comes to reducing all forms of systematics that can arise from measuring the jets energy. Analysis on FSR is in general conducted on different simulated samples where one has included the radiative effect using algorithms such as PYTHIA[29]. The hypothesis is that through the re-weighting of the -background nominal sample one could add a better representation of the FSR effect. Finding a simple way to include a better description of FSR would not only save time in the simulation process but it would also be a way to reduce the systematic errors originating from limited MC statistics. Due to statistical effects coming from the simulations one cannot use the basic approach to define the effect of FSR as simply the difference between nominal and FSR. Two methods are tested to estimate the FSR effects; the first method uses a set of efficiency factors to represent the signal regions, the second method is to add a weight to the events of the nominal sample. The first method show positive results, especially in SR2, compared to a basic analysis, with an uncertainty of the FSR effect of: SR1:±29% SR2: ±51% SR3: ±37%. While a basic analysis gave an uncertainty of ±42%, ±122% and 36%. The second method shows positive signs where the re-weighted sample moves closer to the behaviour of the FSR sample. However, both methods are based on insufficient amount of statistics to draw any absolute conclusions.

The IceCube Neutrino Observatory at the South Pole has instrumented one cu-bic kilometer of ice by deploying digital optical modules (DOMs) in 86 drillholes, each containing a string of DOMs. So far IceCube has used the locationof the drill tower for the positions of DOMs in the transverse directions (x andy) to the vertical direction (z). Attempts to improve on this have so far failed.This thesis presents a new method for calibrating the positions of the DOMs.For a large selection of muon tracks, a maximum likelihood-based approach isused to determine the positions of DOMs. As a proof of concept, four centralstrings are studied to keep systematics uncertainties as low as possible. Themethod can find x and y (z) positions to 0.2m (0.5m) as found using simulationcorresponding to four days of data. In four days of real data we find that thereconstructed z position has a systematic offset of around 5m, which is not aphysical shift as the positions are known to 0.2m. For the x and y positions theresults are consistent with nominal positions except for string 36. As string 36is a string at the center of the IceCube array, it should be the most symmet-ric and an accurate estimate could indicate that the method has found a realdeviation.

There is by now compelling evidence that most of the matter in the universe is in the form of dark matter, a form of matter quite different from the matter we experience in every day life. The gravitational effects of this dark matter have been observed in many different ways but its true nature is still unknown. In most models dark matter particles can annihilate with each other into standard model particles. The direct or indirect observation of such annihilation products could give important clues for the dark matter puzzle. For signals from dark matter annihilations to be detectable, typically high dark matter densities are required. Massive objects, such as stars, can increase the local dark matter density both via scattering off nucleons and by pulling in dark matter gravitationally as the star forms. Dark matter annihilations outside the star would give rise to gamma rays and this is discussed in the first paper. Furthermore dark matter annihilations inside the star would deposit energy inside the star which, if abundant enough, could alter the stellar evolution. Aspects of this are investigated in the second paper. Finally, local dark matter overdensities formed in the early universe could still be around today; prospects of detecting gamma rays from such clumps are discussed in the third paper. / Introduktionsdelen till en sammanläggningsavhandling

QC 20100420

This thesis examines how the kinetic properties of pair-produced Higgs bosons depend on whether the process is mediated by particles in the Standard Model or a simplified supersymmetric model. The analysis is based on simulated data made using MadGraph. The examination within this thesis is split into two parts, one performed on the output of MadGraph (parton level) and the second performed on the output of a simplified simulation of the ATLAS detector response (reconstruction level). The first part also contains a section showing that the supersymmetric model results are consistent with the Standard Model results in cases where no supersymmetric particles are present in the di-Higgs production and decay, and a section detailing the minor differences in kinematics of the Higgs bosons depending on the mass of the supersymmetric partner of the top quark. The second part of the thesis covers the decay of the Higgs boson into bottom and anti-bottom quarks, and includes a section showing that there is no difference if the decay is done using the MadSpin module or the Pythia module for MadGraph, before exploring the kinematics of the jets produced. Both when studying Higgs bosons at the parton level and when studying b-jets at the reconstruction level, it is found that there are clear differences in the kinematics between the Standard Model and the supersymmetric model. At both levels, a neural network has been designed, trained and tested. For the parton level neural network 59 % of all events are classified correctly, while at the reconstruction level 51 % of all events are classified correctly. These neural network results show that it is possible to train a neural network to learn on data like this, and that with enough di-Higgs events detected, their kinematic properties could be used to indirectly infer the presence of physics beyond the Standard Model.

The Large Hadron Collider (LHC) located at the European Organization for Nuclear Research (CERN) is the most powerful particle accelerator in the world in terms of collision energy and luminosity. This thesis presents a search for supersymmetric particles in proton-proton collision data recorded by the ATLAS experiment. A search for direct production of chargino and slepton pairs in a final state characterised by the presence of two leptons and missing transverse momentum is presented. This analysis is done using L = 20 fb-1 proton-proton collisions at √s = 8 TeV collected in 2012. No significant excess over background is observed. Exclusion limits at 95% confidence level on chargino, neutralino and slepton production are set. In 2011-12 the LHC was providing collisions every 50 ns. This puts very strong requirements on the energy measurement in presence of energy deposits from different collisions in the same read-out window and in the same calorimeter channel (pile-up). A quality factor computed offline for each collision and for each channel in the ATLAS Tile Calorimeter (TileCal) is studied. It is shown that the quality factor can be used to select channels that need a special treatment to account for large energy deposition from pile-up. Efficient criteria to detect pile-up in TileCal channels are proposed.

Characteristics of the electromagnetic calorimeter of PANDA (anti-Proton ANnihilations at DArmstadt), one of the key experiments at FAIR, have been measured at the tagged photon beam facility at MAX-lab for 61 energies in the range 12-63 MeV. The tested detector array consisted of 5x5 PbWO4 (lead tungstate) crystals designed for the forward end-cap. The array was cooled to -25 ºC and read out with conventional photomultiplier tubes. The measured energy resolution, σ/E, (for example 6 % at 20 MeV) shows that the current design meets the criterion of PANDA. The array is now ready to be equipped with vacuum phototriods designed for the forward end-cap.         As a part of the experiment, characteristics of the tagging spectrometer at MAX-lab were measured. This revealed a lowered performance in terms of resolution for low energies. A discrepancy between the nominal and the measured tagged energies was also discovered. Possible explanations to these observations are presented.

A study of photon-jets and their identification variables computed in the electromagnetic calorimeter of an in-house, ATLAS-like, detector model developed at Uppsala University for full-simulations is presented. The signal, which is based on the Composite Higgs Model, considers the pair production of vector-like top quarks, denoted t', with the successive decays t' to S and a Standard Model top quark and S to two photons, where S is a new pseudo-scalar boson. For clustering purposes, Topological Clustering was implemented and compared against the already implemented Sliding Window clustering algorithm. A modification to the Topological Clustering algorithm used in the ATLAS detector was made and compared against the original version. The performance test showed a significant mass reconstruction improvement with Topological Clustering. Furthermore, the performance test showed some improvement with the modifications in the Topological Clustering for photons in a collimated scenario. The signal is compared to Standard Model background consisting of one photon plus zero, one or two jets, and isolated photons from the Higgs boson decaying into two photons. Even though some discrimination could be found, the analysis showed that standard cut and count strategy won't work and hence, further analysis with machine learning should be tested.

Measuring lifetimes of excited nuclear states can give important information on the internal structure of the nucleus. This thesis is based on two experiments performed in Italy and in the USA in order to deduce the lifetimes of excited states in 165Lu and 107Cd. The lifetimes were measured using the Recoil Distance Method and the reduced transition probabilities between states have been calculated from the lifetimes in a model independent way. In the analysis of the data from the experiment on 165Lu, the shape of the nucleus is investigated using a collective rotational model to describe the observed excited states. A possibility of a triaxial shape of this nucleus is discussed. In the second experiment the obtained reduced transition probabilities for the nucleus 107Cd are compared to theoretical predictions for vibrational and rotational excitation modes / QC 20101108