Customer call centers with hundreds of agents working in parallel are ubiquitous in many industries. These systems have a large amount of daily traffic that is stochastic in nature. It becomes more and more difficult to manage a call center because of its increasingly large scale and the stochastic variability in arrival and service processes. In call center operations, customer abandonment is a key factor and may significantly impact the system performance. It must be modeled explicitly in order for an operational model to be relevant for decision making. In this thesis, a large-scale call center is modeled as a queue with many parallel servers. To model the customer abandonment, each customer is assigned a patience time. When his waiting time for service exceeds his patience time, a customer abandons the system without service. We develop analytical and numerical tools for analyzing such a queue. We first study a sequence of G/G/n+GI queues, where the customer patience times are independent and identically distributed (iid) following a general distribution. The focus is the abandonment and the queue length processes. We prove that under certain conditions, a deterministic relationship holds asymptotically in diffusion scaling between these two stochastic processes, as the number of servers goes to infinity. Next, we restrict the service time distribution to be a phase-type distribution with d phases. Using the aforementioned asymptotic relationship, we prove limit theorems for G/Ph/n+GI queues in the quality- and efficiency-driven (QED) regime. In particular, the limit process for the customer number in each phase is a d-dimensional piecewise Ornstein-Uhlenbeck (OU) process. Motivated by the diffusion limit process, we propose two approximate models for a GI/Ph/n+GI queue. In each model, a d-dimensional diffusion process is used to approximate the dynamics of the queue. These two models differ in how the patience time distribution is built into them. The first diffusion model uses the patience time density at zero and the second one uses the entire patience time distribution. We also develop a numerical algorithm to analyze these diffusion models. The algorithm solves the stationary distribution of each model. The computed stationary distribution is used to estimate the queue's performance. A crucial part of this algorithm is to choose an appropriate reference density that controls the convergence of the algorithm. We develop a systematic approach to constructing a reference density. With the proposed reference density, the algorithm is shown to converge quickly in numerical experiments. These experiments also show that the diffusion models are good approximations of queues with a moderate to large number of servers.

Options are a special type of derivative securities because their values are derived from the value of some underlying security. Most options can be grouped into either of the two categories: European options which can be exercised only on the expiration date, and American options which can be exercised on or before the expiration date. American options are much harder to deal with than European ones. The reason being the optimal exercise policy of these options which led to free boundary problems. Ever since the seminal work of Black and Scholes [J. Pol. Econ. 81(3) (1973), 637-659], the differential equation approach in pricing options has attracted many researchers. Recently, numerical singular perturbation techniques have been used extensively for solving many differential equation models of sciences and engineering. In this thesis, we explore some of those methods which are based on spline approximations to solve the option pricing problems. We show a systematic construction and analysis of these methods to solve some European option problems and then extend the approach to solve problems of pricing American options as well as some exotic options. Proposed methods are analyzed for stability and convergence. Thorough numerical results are presented and compared with those seen in the literature.

The high costs and political tensions associated with Ballistic Missile Defense Systems (BMDS) has driven much of the testing and evaluation of BMDS to be performed through high fidelity Modeling and Simulation (M&S). In response, the M&S environments have become highly complex, extremely computationally intensive, and far too slow to be of use to systems engineers and high level decision makers. Regression models can be used to map the system characteristics to the metrics of interest, bringing about large quantities of data and allowing for real-time interaction with high-fidelity M&S environments, however the abundance of discontinuities and non-unique solutions makes the application of regression techniques hazardous. Due to these ambiguities, the transfer function from the characteristics to the metrics appears to have multiple solutions for a given set of inputs, which combined with the multiple inputs yielding the same set of outputs, causes troubles in creating a mapping. Due to the abundance of discontinuities, the existence of a neural network mapping from the system attributes to the performance metrics is not guaranteed, and if the mapping does exist, it requires a large amount of data to be for creating a regression model, making regression techniques less suitable to BMDS analysis. By employing Nested Neural Networks (NNNs), intermediate data can be associated with an ambiguous output which can allow for a regression model to be made. The addition of intermediate data incorporates more knowledge of the design space into the analysis. Nested neural networks divide the design space to form a piece-wise continuous function, which allows for the user to incorporate system knowledge into the surrogate modeling process while reducing the size of a data set required to form the regression model. This thesis defines nested neural networks along with methods and techniques for using NNNs to relieve the effects of discontinuities and non-unique solutions. To show the benefit of the approach, these techniques are applies them to a BMDS simulation. Case studies are performed to optimize the system configurations and assess robustness which could not be done without the regression models.

Agent-based systems usually operate in real-time, stochastic and dynamic environments. Many theoretical and applied techniques have been applied to the investigation of agent architecture with respect to communication, cooperation, and learning, in order to provide a framework for implementing artificial intelligence and computing techniques. Intelligent agents are required to be able to adapt and learn in uncertain environments via communication and collaboration (in both competitive and cooperative situations). The ability of reasoning and learning is one fundamental feature for intelligent agents. Due to the inherent complexity, however, it is difficult to verify the properties of the complex and dynamic environments a priori. Since analytic techniques are inadequate for solving these problems, reinforcement learning (RL) has appeared as a popular approach by mapping states to actions, so as to maximise the long-term rewards. Computer simulation is needed to replicate an experiment for testing and verifying the efficiency of simulation-based optimisation techniques. In doing so, a simulation testbed called robot soccer is used to test the learning algorithms in the specified scenarios. This research involves the investigation of simulation-based optimisation techniques in agent-based systems. Firstly, a hybrid agent teaming framework is presented for investigating agent team architecture, learning abilities, and other specific behaviors. Secondly, the novel reinforcement learning algorithms to verify goal-oriented agents; competitive and cooperative learning abilities for decision-making are developed. In addition, the function approximation technique known as tile coding (TC), is used to avoid the state space growing exponentially with the curse of dimensionality. Thirdly, the underlying mechanism of eligibility traces is analysed in terms of on-policy algorithm and off-policy algorithm, accumulating traces and replacing traces. Fourthly, the "design of experiment" techniques, such as Simulated Annealing method and Response Surface methodology, are integrated with reinforcement learning techniques to enhance the performance. Fifthly, a methodology is proposed to find the optimal parameter values to improve convergence and efficiency of the learning algorithms. Finally, this thesis provides a serious full-fledged numerical analysis on the efficiency of various RL techniques.

Philosophiae Doctor - PhD / Options are a special type of derivative securities because their values are derived from the value of some underlying security. Most options can be grouped into either of the two categories: European options which can be exercised only on the expiration date, and American options which can be exercised on or before the expiration date. American options are much harder to deal with than European ones. The reason being the optimal exercise policy of these options which led to free boundary problems. Ever since the seminal work of Black and Scholes [J. Pol. Econ. 81(3) (1973), 637-659], the differential equation approach in pricing options has attracted many researchers. Recently, numerical singular perturbation techniques have been used extensively for solving many differential equation models of sciences and engineering. In this thesis, we explore some of those methods which are based on spline approximations to solve the option pricing problems. We show a systematic construction and analysis of these methods to solve some European option problems and then extend the approach to solve problems of pricing American options as well as some exotic options. Proposed methods are analyzed for stability and convergence. Thorough numerical results are presented and compared with those seen in the literature. / South Africa

Let K be a compact subset of Cn . The polynomially convex hull of K is defined as The compact set K is said to be polynomially convex if = K. A closed subset is said to be locally polynomially convex at if there exists a closed ball centred at z such that is polynomially convex. The aim of this thesis is to derive easily checkable conditions to detect polynomial convexity in certain classes of sets in This thesis begins with the basic question: Let S1 and S2 be two smooth, totally real surfaces in C2 that contain the origin. If the union of their tangent planes is locally polynomially convex at the origin, then is locally polynomially convex at the origin? If then it is a folk result that the answer is, “Yes.” We discuss an obstruction to the presumed proof, and use a different approach to provide a proof. When dimR it turns out that the positioning of the complexification of controls the outcome in many situations. In general, however, local polynomial convexity of also depends on the degeneracy of the contact of T0Sj with We establish a result showing this. Next, we consider a generalization of Weinstock’s theorem for more than two totally real planes in C2 . Using a characterization, recently found by Florentino, for simultaneous triangularizability over R of real matrices, we present a sufficient condition for local polynomial convexity at of union of finitely many totally real planes is C2 . The next result is motivated by an approximation theorem of Axler and Shields, which says that the uniform algebra on the closed unit disc generated by z and h — where h is a nowhereholomorphic harmonic function on D that is continuous up to ∂D — equals . The abstract tools used by Axler and Shields make harmonicity of h an essential condition for their result. We use the concepts of plurisubharmonicity and polynomial convexity to show that, in fact, the same conclusion is reached if h is replaced by h+ R, where R is a nonharmonic perturbation whose Laplacian is “small” in a certain sense. Ideas developed for the latter result, especially the role of plurisubharmonicity, lead us to our final result: a characterization for compact patches of smooth, totallyreal graphs in to be polynomially convex.

Philosophiae Doctor - PhD / Options are a special type of derivative securities because their values are derived from the value of some underlying security. Most options can be grouped into either of the two categories: European options which can be exercised only on the expiration date, and American options which can be exercised on or before the expiration date. American options are much harder to deal with than European ones. The reason being the optimal exercise policy of these options which led to free boundary problems. Ever since the seminal work of Black and Scholes [J. Pol. Bean. 81(3) (1973), 637-659], the differential equation approach in pricing options has attracted many researchers. Recently, numerical singular perturbation techniques have been used extensively for solving many differential equation models of sciences and engineering. In this thesis, we explore some of those methods which are based on spline approximations to solve the option pricing problems. We show a systematic construction and analysis of these methods to solve some European option problems and then extend the approach to solve problems of pricing American options as well as some exotic options. Proposed methods are analyzed for stability and convergence. Thorough numerical results are presented and compared with those seen in the literature.

Data gathering is a constant in human history with ever increasing amounts in quantity and dimensionality. To get a feel for the data, make it interpretable, or find underlying laws it is necessary to fit a function to the finite and possibly noisy data. In this thesis we focus on a method achieving this, namely least squares approximation. Its discovery dates back to around 1800 and it has since then proven to be an indispensable tool which is efficient and has the capability to achieve optimal error when used right. Crucial for the least squares method are the ansatz functions and the sampling points. To discuss them, we gather tools from probability theory, frame subsampling, and $L_2$-Marcinkiewicz-Zygmund inequalities. With that we give results in the worst-case or minmax setting, when a set of points is sought for approximating a class of functions, which we model as a generic reproducing kernel Hilbert space. Further, we give error bounds in the statistical learning setting for approximating individual functions from possibly noisy samples. Here, we include the covariate-shift setting as a subfield of transfer learning. In a natural way a parameter choice question arises for balancing over- and underfitting effect. We tackle this by using the cross-validation score, for which we show a fast way of computing as well as prove the goodness thereof.:1 Introduction 2 Least squares approximation 3 Reproducing kernel Hilbert spaces (RKHS) 4 Concentration inequalities 5 Subsampling of finite frames 6 L2 -Marcinkiewicz-Zygmund (MZ) inequalities 7 Least squares in the worst-case setting 8 Least squares in statistical learning 9 Cross-validation 10 Outlook

Markov-type inequalities provide upper bounds on the norm of the (higher order) derivative of an algebraic polynomial in terms of the norm of the polynomial itself. The present thesis considers the cases in which the norms are of the Laguerre, Gegenbauer, or Hermite type, with respective weights chosen differently on both sides of the inequality. An answer is given to the question on the best constant so that such an inequality is valid for every polynomial of degree at most n. The demanded best constant turns out to be the operator norm of the differential operator. The latter conicides with the tractable spectral norm of its matrix representation in an appropriate set of orthonormal bases. The methods to determine these norms vary tremendously, depending on the difference of the parameters accompanying the weights. Up to a very small gap in the parameter range, asymptotics for the best constant in each of the aforementioned cases are given. / Markovungleichungen liefern obere Schranken an die Norm einer (höheren) Ableitung eines algebraischen Polynoms in Bezug auf die Norm des Polynoms selbst. Diese vorliegende Arbeit betrachtet den Fall, dass die Normen vom Laguerre-, Gegenbauer- oder Hermitetyp sind, wobei die entsprechenden Gewichte auf beiden Seiten unterschiedlich gewählt werden. Es wird die kleinste Konstante bestimmt, sodass diese Ungleichung für jedes Polynom vom Grad höchstens n erfüllt ist. Die gesuchte kleinste Konstante kann als die Operatornorm des Differentialoperators dargestellt werden. Diese fällt aber mit der Spektralnorm der Matrixdarstellung in einem Paar geeignet gewählter Orthonormalbasen zusammen und kann daher gut behandelt werden. Zur Abschätzung dieser Normen kommen verschiedene Methoden zum Einsatz, die durch die Differenz der in den Gewichten auftretenden Parameter bestimmt werden. Bis auch eine kleine Lücke im Parameterbereich wird das asymptotische Verhalten der kleinsten Konstanten in jedem der betrachteten Fälle ermittelt.

<div>The Subword Complexity of a character string refers to the number of distinct substrings of any length that occur as contiguous patterns in the string. The kth Subword Complexity in particular, refers to the number of distinct substrings of length k in a string of length n. In this work, we evaluate the expected value and the second factorial moment of the kth Subword Complexity for the binary strings over memory-less sources. We first take a combinatorial approach to derive a probability generating function for the number of occurrences of patterns in strings of finite length. This enables us to have an exact expression for the two moments in terms of patterns' auto-correlation and correlation polynomials. We then investigate the asymptotic behavior for values of k=a log n. In the proof, we compare the distribution of the kth Subword Complexity of binary strings to the distribution of distinct prefixes of independent strings stored in a trie. </div><div>The methodology that we use involves complex analysis, analytical poissonization and depoissonization, the Mellin transform, and saddle point analysis.</div>