Skip directly to content

Coming dissertations at TekNat

  • Deep probabilistic models for sequential and hierarchical data

    Author: Carl Andersson
    Publication date: 2022-05-02 14:39

    Consider the problem where we want a computer program capable of recognizing a pedestrian on the road. This could be employed in a car to automatically apply the brakes to avoid an accident. Writing such a program is immensely difficult but what if we could instead use examples and let the program learn what characterizes a pedestrian from the examples. Machine learning can be described as the process of teaching a model (computer program) to predict something (the presence of a pedestrian) with help of data (examples) instead of through explicit programming.

    This thesis focuses on a specific method in machine learning, called deep learning. This method can arguably be seen as sole responsible for the recent upswing of machine learning in academia as well as in society at large. However, deep learning requires, in human standards, a huge amount of data to perform well which can be a limiting factor.  In this thesis we describe different approaches to reduce the amount of data that is needed by encoding some of our prior knowledge about the problem into the model. To this end we focus on sequential and hierarchical data, such as speech and written language.


  • New Roads for an Ancient Enzyme : Whole-Cell Studies and New Cofactors for [FeFe] Hydrogenases

    Author: Marco Lorenzi
    Publication date: 2022-05-02 11:21

    [FeFe] hydrogenases rare Nature’s best H2-processing catalysts, and one of the best candidates to satisfy societal need for cheap and efficient catalyst for H2-evolution. These enzymes owe their remarkable catalytic activities to their organometallic active site, called “H-cluster”. The H-cluster can be described as a canonical [4Fe4S] cluster linked via a bridging cysteine residue to a [2Fe] subsite, which is in turn coordinated by three CO, two CN− and one bidentate azadithiolate ligand. This unique cofactor allows these enzymes to function with virtually no overpotential requirements and TOFs up to 20 000 s-1. We have now reached a good understanding of the catalytic cycle by which these enzymes operate, yet many questions remain open especially regarding the physiological relevance of some the proposed intermediates. 

    In this thesis, we have used FTIR and EPR spectroscopies on whole-cell samples of [FeFe] hydrogenases to study the influence of the intracellular environment on the catalytic cycle of these enzymes. Moreover, we have investigated how the bacterial cytoplasm influences the stability of the H-cluster and favours the formation of sulfide-inhibited states,...

  • Computational fluid dynamics for dispersion calculation in urban surroundings

    Author: Jan Burman
    Publication date: 2022-05-02 09:09

    Increased knowledge on dispersion processes in urban environment will enhance the ability in the society to handle events where releases of toxic substances can occur. Also, the ability to increase preparedness at locations where such events potentially can emerge. 

    Can Computational Fluid Dynamics (CFD) models contribute to increased knowledge and what type of models are most suitable considering dispersion in urban environment? 

    CFD-models can simulate almost any scenario but urban scales are still computationally de-manding. Simplifications of the basic equations are needed. Mainly two methods to do this is feasible, namely Reynolds Averaged Navier-Stokes models (RANS) and Large Eddy Simula-tion models (LES). These methods are commonly used for hydrodynamic flow studies. In this thesis the eddy viscosity hypothesis is implemented and used in all turbulence models. 

    The scenarios studied includes flow and dispersion past objects at the side of a road, flow over buildings, dispersion in urban environments and on synthetic stochastic boundary condi-tions. The basic flow around objects assume that the turbulence is realistically modelled. In RANS the flow is...

  • First-principles theory of electrically-induced spin and orbital magnetization

    Author: Leandro Salemi
    Publication date: 2022-04-29 10:46

    Spin-orbit torques (SOTs) have emerged recently as practical tools to control the magnetization in spintronic devices, but it is debated what the underlying fundamental processes are that enable fast and energy-efficient magnetization switching.

    In this thesis, we investigate theoretically possible means of controlling magnetization in magnetic materials and heterostructures. To this end we employ relativistic density functional theory and linear-response theory to compute electrically induced spin currents and spin polarizations. We focus initially on two effects, the spin Hall effect (SHE) and the spin Rashba-Edelstein effect (SREE) that are due to spin-orbit coupling (SOC).

    First, we investigate the electric-field induced local magnetization in two antiferromagnets, CuMnAs and Mn2Au. Our explicit calculations show that there is not only an SREE-induced local spin polarization, but also a surprisingly large orbital polarization, due to what we call an orbital Rashba-Edelstein effect (OREE). We show that the induced orbital polarization does not require SOC and that it exhibits a staggered, Rashba symmetry in contrast to the induced spin polarization that can have...

  • The effects of internally expressed Contact-Dependent growth Inhibition (CDI) toxins in bacteria

    Author: Magnus Stårsta
    Publication date: 2022-04-29 08:38

    Bacteria, both pathogenic and non-pathogenic, have developed multiple forms of competition mechanisms to combat each other including, but not limited to, Contact-Dependent growth Inhibition (CDI) systems, Type VI Secretion Systems and the associated Rearrangement hotspot (Rhs) toxin system. These systems usually confers a great fitness advantage as they allow for precise delivery of toxic molecules into competing bacteria whilst sister cells are protected from auto-inhibition by producing a cognate immunity protein. Delivery between sister cells may serve as a form of “self-recognition” whilst maintaining selection pressure for these genes within the population. How these genes are maintained in conditions where delivery does not occur has until now not been fully understood.

    This thesis describes secondary functions, maintained selection pressure and regulation of Rhs and CDI systems in three parts. In Paper I, we made a novel discovery that rhs toxin and immunity genes from Salmonella enterica serovar Typhimurium are transcribed from internal transcriptional start sites independent of the full length delivery gene. This results in functional cytosolic...

  • Localization of N=2 Field Theories: a Twisted Path Across Four Manifolds

    Author: Lorenzo Ruggeri
    Publication date: 2022-04-27 13:40

    Supersymmetric quantum field theories provide a framework where certain physical observables can be computed exactly. In those cases, one not only has control over perturbative contributions but also over non-perturbative contributions. In this thesis the main focus are N=2 supersymmetric quantum field theories on compact manifolds with U(1)xU(1) isometry and a Killing vector with isolated fixed points.

    In Part I, focusing on pure gauge theories, it is explained how equivariant Donaldson-Witten theory and a certain class of non-topological theories, related to the well-known result of Pestun on the four-dimensional sphere, can be described as two instances of an underlying framework. Employing this formalism, a general formula for the partition functions has been proposed which is valid both for equivariant Donaldson-Witten and Pestun-like theories. On top of perturbative contributions, the partition functions get contributions from instantons and fluxes.

    In Part II, the results appearing in the papers attached to this thesis are presented. First, a formal treatment of the perturbative part is discussed. Then, the dependence on flux of the partition function is...

  • Heterotic Compactifications in the Era of Data Science

    Author: Robin Schneider
    Publication date: 2022-04-27 13:38

    The goal of this thesis is to review and investigate recent applications of machine learning to problems in string theory. String theory, the leading candidate for a unification of gravity and the standard model of particle physics, requires the introduction of additional space-time dimensions. To match experimental observations of our universe, these additional dimensions need to be curled up on a compact space. The most common choice to describe this compact space are manifolds of Calabi-Yau type. These manifolds come with favourable mathematical and phenomenological properties.

    In the first half of this thesis Calabi-Yau manifolds, which are complex Kähler manifolds admitting a Ricci-flat metric, are introduced. The popular construction as complete intersections in products of complex projective space is explained and the necessary mathematical machinery to compute their topological quantities presented. This part is followed by a review of machine learning applications to study their Hodge numbers and the cohomologies of line bundles. In a next step the new Python library cymetric is presented for modeling numerical approximations of the unknown Ricci-flat metric. The...

  • Numerical and laboratory studies of seismic properties and elements of rock fabric from the microscale to the macroscale

    Author: Mohsen Bazargan
    Publication date: 2022-04-27 11:22

    Physical properties of rocks studied in the laboratory are useful to provide constraints on the dynamics of Earth’s interior. This may include direct constraints on in-situ seismic properties, such as elastic wave velocity measurements that can be compared to seismological data, or petrofabric indicators such as anisotropy of magnetic susceptibility (AMS). Another approach that provides predictive insight into the physical properties of Earth’s interior are computer models. Numerical modelling, in particular, can be used to investigate the dynamic propagation of elastic waves or the flow of a material to generate a fabric or texture (i.e., petrofabric in rocks). This thesis focuses on an integrative approach, utilizing both laboratory measurements and numerical modelling, to understand physical properties and petrofabric development in rocks originating in Earth’s crust. The physical properties of rocks are affected by both intrinsic sources (e.g., inherent to crystals) and extrinsic sources (e.g., layering, microcracks, shape preferred orientation of crystals, grain size, presence of geological fluids). A versatile numerical elastic wave propagation model is...

  • Studies of second coordination sphere effects and metal variations on [FeFe]-hydrogenase mimics

    Author: Holly J. Redman
    Publication date: 2022-04-26 08:57

    Mitigation of climate change motivates researchers to explore hydrogen as a potential energy carrier. Unfortunately, widespread use of hydrogen as an energy carrier is limited by numerous challenges in its production, including high energy consumption; high economic cost; current reliance on rare metals such as platinum. Diiron hydrogenases could provide a solution to the above-mentioned challenges because they are able to turnover hydrogen at very high frequencies, and utilise earth abundant iron as the redox active centre. However, diiron hydrogenases are not currently a scalable technology, and more research is needed to fully understand their reaction mechanism, and to allow engineering of optimal proton reduction catalysts. The H-cluster is an hexanuclear iron cluster in the active site, which consists of a [Fe4S4]-cluster and a [Fe2S2] cofactor. The [Fe4S4]-cluster behaves as a redox active ligand for the [Fe2S2] cofactor. The [Fe2S2] cofactor is a diiron complex in which the irons share a bridging azadithiolato ligand, and one bridging carbonyl ligand. Each iron has one terminal carbonyl ligand and one cyanide ligand. The [Fe4S4]-cluster and [Fe2S2] cofactor are coupled...

  • Stratified algebras and classification of tilting modules

    Author: Elin Persson Westin
    Publication date: 2022-04-25 14:30

    This thesis contains three papers in representation theory of algebras. It mainly studies two types of algebras; quasi-hereditary algebras and standardly stratified algebras.

    Paper I provides a classification of generalized tilting modules and full exceptional sequences for a family of quasi-hereditary algebras and for another related family of algebras. These algebras are referred to as leaf quotients of type A zig-zag algebras. We also give a characterization of the first family of algebras as quasi-hereditary algebras with a simple preserving duality, where exactly one indecomposable projective module is not injective.

    Paper II proves uniqueness of the essential order for standardly stratified algebras having a simple preserving duality. We use this result to classify, up to equivalence, regular blocks of S-subcategories in the BGG category O. We also establish some derived equivalences between blocks in type A. Additionally, the paper provides explicit formulas for the projective dimension of certain structural modules in S-subcategories of O and for the finitistic dimension of these subcategories. 

    Paper III provides a classification of generalized...