School of Physics - Theses
Permanent URI for this collection
Search Results
1 - 10 of 21
-
ItemLearning invariant representations with applications to high-energy physics( 2020)In searches for new physics in high-energy physics, experimental analyses are primarily concerned with physical processes which are rare or hitherto unobserved. To claim a statistically significant discovery or exclusion of new physics when studying such decays, it is necessary to maintain an appropriate signal to noise ratio. This makes systems capable of efficient discrimination of signal from datasets overwhelmingly dominated by background events an important component of modern experimental analyses. However, na\"ive application of these methods is liable to raise poorly understood systematic effects which may ultimately degrade the significance of the final measurement. To understand the origin of these systematic effects, we note that there are certain protected variables in experimental analyses which should remain unbiased by the analysis procedure. Variables that the input parameters of models of new physics are strongly dependent upon and variables used to model background contributions to the total measured event yield fall into this category. Systems responsible for separating signal from background events achieve this by sampling events with signal-like characteristics from all candidate events. If this procedure introduces sampling bias into the distribution of protected variables, this introduces systematic effects into the analysis which are difficult to characterize. Thus it is desirable for these systems to distinguish between signal and background events without using information about certain protected variables. Beyond high-energy physics, building systems that make decisions independent of certain protected or sensitive information is an important theme in the real-world application of machine learning and statistics. We address this task as an optimization problem of finding a representation of the observed data that is invariant to the given protected quantities. This representation should satisfy two competing criteria. Firstly, it should contain all relevant information about the data so that it may be used as a proxy for arbitrary downstream tasks, such as inference of unobserved quantities or prediction of target variables. Secondly, it should not be informative of the given protected quantities, so that downstream tasks are not influenced by these variables. If the protected quantities to be censored from the intermediate representation contain information that can improve the performance of the downstream task, it is likely that removing this information will adversely affect this task. The challenge lies in balancing both objectives without significantly compromising either requirement. The contribution of this thesis is a new set of methods for addressing this problem. This thesis is divided into two parts, which are largely independent of one another. The first part of this thesis is about constraining the optimization procedure by which the representation is learnt to reduce the informativeness of the representation of the given protected quantities, such that the representation is invariant to changes in these quantities. The second part of this thesis approaches the problem from a latent variable model perspective, in which additional unobserved (latent) variables are introduced which explain the interaction between different attributes of the observed data. These latent variables can be interpreted as a more fundamental, compact lower-dimensional representation of the original high-dimensional unstructured data. By constraining the structure of this latent space, we demonstrate we can isolate the influence of the protected variables into a latent subspace. This allows downstream tasks to only access a relevant subset of the learned representation without being influenced by protected attributes of the original data. The feasibility of our proposed methods is demonstrated through application to a challenging experimental analysis in precision flavor physics at the Belle II experiment - the study of the $b \rightarrow s \gamma$ transition, a sensitive probe of potential new physics.
-
ItemMeasurements of the ATLAS tau trigger reconstruction and identification efficiencies using 2016 data from \(pp\) collisions at\(\sqrt s \) = 13 T\(e\)V( 2017)This thesis presents the performance of the tau trigger algorithm used by the ATLAS experiment to select hadronically decaying tau leptons in the LHC Run 2. Using the 33.3 \(f{b^{ - 1}}\) of \(pp\) collisions data recorded in 2016 at\(\sqrt s \) = 13 T\(e\)V, the performance of this algorithm is studied using a `tag-and-probe' based analysis in order to select Z boson decays to tau leptons, where one tau decays hadronically and the other leptonically. The reconstruction and identification efficiencies of the tau trigger algorithm are measured, and good performance is observed. The efficiency of the tau trigger in data is compared with that in simulation, and is parametrised as a function of the tau decay topology, its kinematics, and the average number of interactions per bunch crossing. The selection efficiency at each step of the high level trigger is measured, using dedicated intermediary triggers, and good agreement between data and simulation is observed. Using the comparison between reconstruction and identification efficiencies in data and simulation, correction factors for simulated events are measured, which are utilised by the entire ATLAS collaboration.
-
ItemB0→K0π0 and direct CP violation at Belle( 2017)Rare B-meson decays such as the B0 → Ksπ0 which proceed without a charm quark provide a probe for physics beyond the standard model. This decay proceeds mainly via the b → s penguin transition, with the b → u transition being colour suppressed, allowing CP-violating effects to be observable. The asymmetric e+e− KEKB collider and the Belle detector provide the large luminosity and data collection required to observe these rare B decays. Methods to reduce the large qq backgrounds are investigated. The use of optimised neural networks using TensorFlow shows a significant improvement compared to the commonly used NeuroBayes software. Techniques for reducing correlations between variables introduced by TensorFlow are also investigated, proving that the use of adversarial neural networks can provide an improved background suppression as compared to NeuroBayes, whilst minimising correlations introduced by the neural network. An improved method of measuring the direct CP violation is introduced. Using Monte Carlo data with sample sizes corresponding to the full Belle datatset of (771.581 ± 10.566) × 106 BB events, the statistical uncertainty in ACP using this method is reduced from the latest Belle result of 0.13 to 0.1035 ± 0.0032. This method would also provide an up to date measurement on B(B0 → K0π0).
-
ItemCharacterisation of sub-radiant modes of metasurfaces( 2016)Metallic nanostructures can be utilised as optical antennas: devices to assist the conversion between localised (near-field) and propagating (far-field) light. Optical antennas possess sub-radiant modes which cannot be easily coupled to propagating fields and have the advantages of longer life-times and higher Q resonances. This project has involved the computational and experimental investigation of the properties of sub-radiant modes of various nanoantennas that form the basis of a metasurface. These modes were found to be excited by off-normally incident light with a specific polarisation or vector beams with radial or azimuthal polarisation. It was also shown through studies of the optical far-field that the angular spectrum of the reflected fields was modified through interaction via a sub-radiant mode.
-
ItemBound states and structural properties of trap-imbalanced fermions( 2016)Ultracold quantum gases, which were only recently realised experimentally, have become one of the most active fields of modern research. This is due to the precision and power of the experiments, as well as the great variety of physical phenomena that they exhibit. In this thesis, the physics of few-body scattering in the strongly-interacting regime is studied. The study of few-body physics allows a better understanding of many-body systems, particularly with strong interactions, which make the usual many-body theoretical techniques untenable. The particular topic of this thesis is few-body scattering in heteronuclear systems, which contain two species of atom with different masses and/or harmonic trapping frequencies. These mass and trap imbalances lead to a variety of interesting physics that is not present in homonuclear systems. Deeply-bound Efimov states with unusual properties appear in systems containing two species of fermions when the ratio of the two species' masses becomes sufficiently large. Other types of deeply bound states also appear above a lower critical mass ratio. We use an implementation of a stochastic variational method to study states of this type under a trap imbalance i.e.\ with two species of fermion with different harmonic trapping frequencies. The stochastic variational method works by randomly generating trial functions, then using a competitive selection scheme to select the best contributions to the approximate variational solution. Using this method, it is shown that the introduction of a trap imbalance has no effect ont the physics of these bound states. Also using this variational method, the effect of trap imbalances on two- and three-body systems, with and without mass imbalances, is studied in detail. It is found that the trap imbalance has the immediate effect of lifting structural and energetic degeneracies between different total angular momentum states of the few-body system. Furthermore, trap imbalances significantly alter the usual physics of the three-fermion system, in which two atoms form a deeply-bound dimer while the third remains unbound. The trap imbalance changes this picture and causes all three atoms to overlap considerably in the ground state, forming a loosely-bound trimer state. Such alterations to the few-body collision properties can have significant effects on the many-body physics of an atomic gas. Thus these results indicate the possibility of additional methods of tuning and control for heteronuclear many-body systems. These results may also be of interest in explicitly few-body experiments, which remain largely unexplored at this time.
-
ItemTowards non-invasive quantum imaging of neuronal activity using optically-active spins in diamond( 2015)Resolving the biological neural network dynamics of the brain with subcellular spatial resolution remains a significant ongoing challenge. It has often been the case, however, that difficult problems in neuroscience have been illuminated by techniques developed at the intersection of various disciplines. This thesis focuses on progress towards such a technique, a nanoscale magnetic sensor for imaging neural networks that functions under physiological conditions. The nitrogen-vacancy (NV) centre defect in biocompatible diamond shows promise due to its nanoscale resolution, sensitivity, stable fluorescence, and room temperature operation. Although there is interest in using the NV centre as a non-invasive magnetic sensor of neural network activity, NV-based sensing techniques are ultimately limited by sources of magnetic noise that destroy the quantum phase coherence these techniques require for their operation. While there are several ways to improve the sensitivity of NV sensors, this thesis investigates the control settings needed to complement the ongoing improvements in diamond quality. We seek to reduce the control errors inherently present in NV sensing protocols: such errors are found to be reduced in the presence of sufficiently large magnetic and microwave fields, although it remains difficult to quantify these errors without accounting for hyperfine NV interactions. The long-term feasibility of NV sensors will be determined by their sensitivity to magnetic fields as well as the magnitude of magnetic fields that individual neurons generate. The external electric potentials generated by individual primary cortical neurons in mice are measured using multi-electrode arrays (MEAs), the prevailing non-invasive technology used to detect electric signals in neural networks. Signals detected by the MEAs are then converted into an equivalent magnetic field, which is found to be in the picotesla range. Although this estimate is lower than the best reported NV sensitivities to date, the actual fields are likely to be larger since MEAs can only detect the extracellular contribution to the magnetic field. This thesis complements existing progress towards realising the long-term goal of a wide-field NV neuron sensor with electrical co-recordings, and suggests that advances in control protocols and material quality may yet be needed to improve the overall sensitivity required to detect activity in neural networks.
-
ItemDevelopment of computational relativistic atomic theory with application to the oxygen spectra in astrophysical systems( 2015)Over many decades, atomic spectroscopy has repeatedly proven to be one of the most widely used applications across various different fields as it allows the detection, identification, and quantification of chemical species present within a particular sample. This is particularly true in the studies of astrophysics and cosmology. This thesis includes the development of new codes for the calculation of the electron self-energy – a quantum electrodynamic effect – as part of the relativistic atomic structure package grasp2k . The codes are shown to bring results towards better agreement with experimental as well as some of the most advanced atomic theories to date. Moreover, a review into the Breit interaction was conducted to address claims that the method of implementation can affect the overall results as well as the quality of gauge convergence. The theoretical development was used to study two oxygen forbidden lines that feature prominently in the aurora as observed on Earth, namely the green and ultraviolet lines with nominal wavelengths of 557.7nm and 297.2nm, respectively. Many have researched into these forbidden lines either through observation of the aurora, or experimentally measured in the laboratory, or theoretical calculations, dating back to the 1930s. Despite such tremendous effort spanning many decades, major discrepancies between observation, measurement, and theory remain. We also explore other optically allowed transitions of ionised oxygen, which showed great consistencies in our methodology. Fully relativistic atomic theories are employed for the first time in the calculation of these particular lines, with excellent overall results and convergence.
-
ItemMatrix product operator simulations of quantum algorithms( 2015)We develop simulation methods for matrix product operators, and perform simulations of the Quantum Fourier Transform, Shor’s algorithm and Grover’s algorithm using matrix product states and matrix product operators. By doing so, we provide numerical evidence that a constant number of QFTs can be efficiently classically simulated on any state whose Schmidt rank grows only polynomially with the number of qubits, and quantify the amount of entanglement present in Shor’s algorithm. The efficiency of the matrix product state and operator representation allows us to perform moderately large simulations of both Shor’s algorithm with Z errors and Grover’s algorithm with up to 15 X, Y and Z errors. While larger simulations have been performed, our results have been computed with little computational power and provide new methods to perform large-scale quantum algorithm simulations.
-
ItemSimulation of noise properties of the Murchison Wide-field Array( 2015)This thesis is focused on better understanding the origins of the noise in an image taken with the Murchison Wide-field Array (MWA). To investigate this, we process images of fields being used for Epoch of Reionization experiments, and use simulations to characterize the expected noise. For our simulation, we use the MIT Array Performance Simulator (MAPS), which was developed at the MIT Haystack Observatory in 2001. MAPS provides a flexible tool for a new generation of simulated data from low frequency radio telescopes. Here we use MAPS to model the performance of the prototype 32 Tiles instrument of the MWA (MWA−32T) and 128 Tiles (MWA−128T). To characterize the noise in the simulation, we need a good astronomical catalogue to provide input sources for the MAPS simulation. By matching source counterparts between the sources observed by the MWA, and the sources in different catalogues, we found that the best catalogue for simulating the model of the sky with respect to the frequency and the flux ranges of the observed sources is the PKSCAT90. In our initial simulation we assume that the sky is dominated by thermal noise. However, we also find noise in the image, which indicates that additional sources of noise are present. We use a counts model to estimate the numbers of unresolved sources in our simulation to better simulate the level of noise in the image. Using number counts of extragalactic radio sources, we find that these are the most dominant noise component. These extragalactic radio sources are usually unresolved at the angular resolution of the observations, and will cause a contamination for detecting the 21-cm line of the neutral hydrogen. In our simulations for the MWA−32T and MWA−128T, we use the EoR1 field, without the presence of resolved sources (FornaxA). Resolved sources can cause errors in the flux densities of the image, and decrease the accuracy of estimated noise in the image.
-
ItemThe effects of LO and NLO QCD corrections on the calculations of the CP violation observables S and C for the non-leptonic decays Bod → J/ΨKs and Bod→ ΦKs in the Standard Model and in the General Left-Right Symmetric Model( 2015)For full abstract please refer to abstract file.