Mathematik
Refine
Year of publication
Document Type
- Doctoral Thesis (76) (remove)
Has Fulltext
- yes (76)
Is part of the Bibliography
- no (76)
Keywords
Institute
- Mathematik (76)
- Informatik (3)
This thesis covers the analysis of radix sort, radix select and the path length of digital trees under a stochastic input assumption known as the Markov model.
The main results are asymptotic expansions of mean and variance as well as a central limit theorem for the complexity of radix sort and the path length of tries, PATRICIA tries and digital search trees.
Concerning radix select, a variety of different models for ranks are discussed including a law of large numbers for the worst case behavior, a limit theorem for the grand averages model and the first order asymptotic of the average complexity in the quantile model.
Some of the results are achieved by moment transfer techniques, the limit laws are based on a novel use of the contraction method suited for systems of stochastic recurrences.
This work is concerned with two topics at the intersection of convex algebraic geometry and optimization.
We develop a new method for the optimization of polynomials over polytopes. From the point of view of convex algebraic geometry the most common method for the approximation of polynomial optimization problems is to solve semidefinite programming relaxations coming from the application of Positivstellensätze. In optimization, non-linear programming problems are often solved using branch and bound methods. We propose a fused method that uses Positivstellensatz-relaxations as lower bounding methods in a branch and bound scheme. By deriving a new error bound for Handelman's Positivstellensatz, we show convergence of the resulting branch and bound method. Through the application of Positivstellensätze, semidefinite programming has gained importance in polynomial optimization in recent years. While it arises to be a powerful tool, the underlying geometry of the feasibility regions (spectrahedra) is not yet well understood. In this work, we study polyhedral and spectrahedral containment problems, in particular we classify their complexity and introduce sufficient criteria to certify the containment of one spectrahedron in another one.
The cones of nonnegative polynomials and sums of squares arise as central objects in convex algebraic geometry and have their origin in the seminal work of Hilbert ([Hil88]). Depending on the number of variables n and the degree d of the polynomials, Hilbert famously characterizes all cases of equality between the cone of nonnegative polynomials and the cone of sums of squares. This equality precisely holds for bivariate forms, quadratic forms and ternary quartics ([Hil88]). Since then, a lot of work has been done in understanding the difference between these two cones, which has major consequences for many practical applications such as for polynomial optimization problems. Roughly speaking, minimizing polynomial functions (constrained as well as unconstrained) can be done efficiently whenever certain nonnegative polynomials can be written as sums of squares (see Section 2.3 for the precise relationship). The underlying reason is the fundamental difference that checking nonnegativity of polynomials is an NP-hard problem whenever the degree is greater or equal than four ([BCSS98]), whereas checking whether a polynomial can be written as a sum of squares is a semidefinite feasibility problem (see Section 2.2). Although the complexity status of the semidefinite feasibility problem is still an open problem, it is polynomial for fixed number of variables. Hence, understanding the difference between nonnegative polynomials and sums of squares is highly desirable both from a theoretical and a practical viewpoint.
We consider a class of nonautonomous nonlinear competitive parabolic systems on bounded radial domains under Neumann or Dirichlet boundary conditions. We show that, if the initial profiles satisfy a reflection inequality with respect to a hyperplane, then bounded positive solutions are asymptotically (in time) foliated Schwarz symmetric with respect to antipodal points. Additionally, a related result for (positive and sign changing solutions) of scalar equations with Neumann or Dirichlet boundary conditions is given. The asymptotic shape of solutions to cooperative systems is also discussed.
A multiple filter test for the detection of rate changes in renewal processes with varying variance
(2014)
The thesis provides novel procedures in the statistical field of change point detection in time series.
Motivated by a variety of neuronal spike train patterns, a broad stochastic point process model is introduced. This model features points in time (change points), where the associated event rate changes. For purposes of change point detection, filtered derivative processes (MOSUM) are studied. Functional limit theorems for the filtered derivative processes are derived. These results are used to support novel procedures for change point detection; in particular, multiple filters (bandwidths) are applied simultaneously in oder to detect change points in different time scales.
The work presented in this thesis is devoted to two classes of mathematical population genetics models, namely the Kingman-coalescent and the Beta-coalescents. Chapters 2, 3 and 4 of the thesis include results concerned with the first model, whereas Chapter 5 presents contributions to the second class of models.
We study the price-setting problem of market makers under perfect competition in continuous time. Thereby we follow the classic Glosten-Milgrom model that defines bid and ask prices as the expectation of a true value of the asset given the market makers partial information that includes the customers trading decisions. The true value is modeled as a Markov process that can be observed by the customers with some noise at Poisson times.
We analyze the price-setting problem by solving a non-standard filtering problem with an endogenous filtration that depends on the bid and ask price process quoted by the market maker. Under some conditions we show existence and uniqueness of the price processes. In a different setting we construct a counterexample to uniqueness. Further, we discuss the behavior of the spread by a convergence result and simulations.
In this thesis, the asymptotic behaviour of Pólya urn models is analyzed, using an approach based on the contraction method. For this, a combinatorial discrete time embedding of the evolution of the composition of the urn into random rooted trees is used. The recursive structure of the trees is used to study the asymptotic behavior using ideas from the contraction method.
The approach is applied to a couple of concrete Pólya urns that lead to limit laws with normal distributions, with non-normal limit distributions, or with asymptotic periodic distributional behavior.
Finally, an approach more in the spirit of earlier applications of the contraction method is discussed for one of the examples. A general transfer theorem of the contraction method is extended to cover this example, leading to conditions on the coefficients of the recursion that are not only weaker but also in general easier to check.
A stochastic model for the joint evaluation of burstiness and regularity in oscillatory spike trains
(2013)
The thesis provides a stochastic model to quantify and classify neuronal firing patterns of oscillatory spike trains. A spike train is a finite sequence of time points at which a neuron has an electric discharge (spike) which is recorded over a finite time interval. In this work, these spike times are analyzed regarding special firing patterns like the presence or absence of oscillatory activity and clusters (so called bursts). These bursts do not have a clear and unique definition in the literature. They are often fired in response to behaviorally relevant stimuli, e.g., an unexpected reward or a novel stimulus, but may also appear spontaneously. Oscillatory activity has been found to be related to complex information processing such as feature binding or figure ground segregation in the visual cortex. Thus, in the context of neurophysiology, it is important to quantify and classify these firing patterns and their change under certain experimental conditions like pharmacological treatment or genetical manipulation. In neuroscientific practice, the classification is often done by visual inspection criteria without giving reproducible results. Furthermore, descriptive methods are used for the quantification of spike trains without relating the extracted measures to properties of the underlying processes.
For that reason, a doubly stochastic point process model is proposed and termed 'Gaussian Locking to a free Oscillator' - GLO. The model has been developed on the basis of empirical observations in dopaminergic neurons and in cooperation with neurophysiologists. The GLO model uses as a first stage an unobservable oscillatory background rhythm which is represented by a stationary random walk whose increments are normally distributed. Two different model types are used to describe single spike firing or clusters of spikes. For both model types, the distribution of the random number of spikes per beat has different probability distributions (Bernoulli in the single spike case or Poisson in the cluster case). In the second stage, the random spike times are placed around their birth beat according to a normal distribution. These spike times represent the observed point process which has five easily interpretable parameters to describe the regularity and the burstiness of the firing patterns.
It turns out that the point process is stationary, simple and ergodic. It can be characterized as a cluster process and for the bursty firing mode as a Cox process. Furthermore, the distribution of the waiting times between spikes can be derived for some parameter combination. The conditional intensity function of the point process is derived which is also called autocorrelation function (ACF) in the neuroscience literature. This function arises by conditioning on a spike at time zero and measures the intensity of spikes x time units later. The autocorrelation histogram (ACH) is an estimate for the ACF. The parameters of the GLO are estimated by fitting the ACF to the ACH with a nonlinear least squares algorithm. This is a common procedure in neuroscientific practice and has the advantage that the GLO ACF can be computed for all parameter combinations and that its properties are closely related to the burstiness and regularity of the process. The precision of estimation is investigated for different scenarios using Monte-Carlo simulations and bootstrap methods.
The GLO provides the neuroscientist with objective and reproducible classification rules for the firing patterns on the basis of the model ACF. These rules are inspired by visual inspection criteria often used in neuroscientific practice and thus support and complement usual analysis of empirical spike trains. When applied to a sample data set, the model is able to detect significant changes in the regularity and burst behavior of the cells and provides confidence intervals for the parameter estimates.
We investigate multivariate Laurent polynomials f \in \C[\mathbf{z}^{\pm 1}] = \C[z_1^{\pm 1},\ldots,z_n^{\pm 1}] with varieties \mathcal{V}(f) restricted to the algebraic torus (\C^*)^n = (\C \setminus \{0\})^n. For such Laurent polynomials f one defines the amoeba \mathcal{A}(f) of f as the image of the variety \mathcal{V}(f) under the \Log-map \Log : (\C^*)^n \to \R^n, (z_1,\ldots,z_n) \mapsto (\log|z_1|, \ldots, \log|z_n|). I.e., the amoeba \mathcal{A}(f) is the projection of the variety \mathcal{V}(f) on its (componentwise logarithmized) absolute values. Amoebas were first defined in 1994 by Gelfand, Kapranov and Zelevinksy. Amoeba theory has been strongly developed since the beginning of the new century. It is related to various mathematical subjects, e.g., complex analysis or real algebraic curves. In particular, amoeba theory can be understood as a natural connection between algebraic and tropical geometry.
In this thesis we investigate the geometry, topology and methods for the approximation of amoebas.
Let \C^A denote the space of all Laurent polynomials with a given, finite support set A \subset \Z^n and coefficients in \C^*. It is well known that, in general, the existence of specific complement components of the amoebas \mathcal{A}(f) for f \in \C^A depends on the choice of coefficients of f. One prominent key problem is to provide bounds on the coefficients in order to guarantee the existence of certain complement components. A second key problem is the question whether the set U_\alpha^A \subseteq \C^A of all polynomials whose amoeba has a complement component of order \alpha \in \conv(A) \cap \Z^n is always connected.
We prove such (upper and lower) bounds for multivariate Laurent polynomials supported on a circuit. If the support set A \subset \Z^n satisfies some additional barycentric condition, we can even give an exact description of the particular sets U_\alpha^A and, especially, prove that they are path-connected.
For the univariate case of polynomials supported on a circuit, i.e., trinomials f = z^{s+t} + p z^t + q (with p,q \in \C^*), we show that a couple of classical questions from the late 19th / early 20th century regarding the connection between the coefficients and the roots of trinomials can be traced back to questions in amoeba theory. This yields nice geometrical and topological counterparts for classical algebraic results. We show for example that a trinomial has a root of a certain, given modulus if and only if the coefficient p is located on a particular hypotrochoid curve. Furthermore, there exist two roots with the same modulus if and only if the coefficient p is located on a particular 1-fan. This local description of the configuration space \C^A yields in particular that all sets U_\alpha^A for \alpha \in \{0,1,\ldots,s+t\} \setminus \{t\} are connected but not simply connected.
We show that for a given lattice polytope P the set of all configuration spaces \C^A of amoebas with \conv(A) = P is a boolean lattice with respect to some order relation \sqsubseteq induced by the set theoretic order relation \subseteq. This boolean lattice turns out to have some nice structural properties and gives in particular an independent motivation for Passare's and Rullgard's conjecture about solidness of amoebas of maximally sparse polynomials. We prove this conjecture for special instances of support sets.
A further key problem in the theory of amoebas is the description of their boundaries. Obviously, every boundary point \mathbf{w} \in \partial \mathcal{A}(f) is the image of a critical point under the \Log-map (where \mathcal{V}(f) is supposed to be non-singular here). Mikhalkin showed that this is equivalent to the fact that there exists a point in the intersection of the variety \mathcal{V}(f) and the fiber \F_{\mathbf{w}} of \mathbf{w} (w.r.t. the \Log-map), which has a (projective) real image under the logarithmic Gauss map. We strengthen this result by showing that a point \mathbf{w} may only be contained in the boundary of \mathcal{A}(f), if every point in the intersection of \mathcal{V}(f) and \F_{\mathbf{w}} has a (projective) real image under the logarithmic Gauss map.
With respect to the approximation of amoebas one is in particular interested in deciding membership, i.e., whether a given point \mathbf{w} \in \R^n is contained in a given amoeba \mathcal{A}(f). We show that this problem can be traced back to a semidefinite optimization problem (SDP), basically via usage of the Real Nullstellensatz. This SDP can be implemented and solved with standard software (we use SOSTools and SeDuMi here). As main theoretic result we show that, from the complexity point of view, our approach is at least as good as Purbhoo's approximation process (which is state of the art).