Past

The trapezoidal rule for numerical integration is remarkably accurate when

the integrand under consideration is smooth and periodic. In this

situation it is superior to more sophisticated methods like Simpson's rule

and even the Gauss-Legendre rule. In the first part of the talk we

discuss this phenomenon and give a few elementary examples. In the second

part of the talk we discuss the application of this idea to the numerical

evaluation of contour integrals in the complex plane.

Demonstrations involving numerical differentiation, the computation

of special functions, and the inversion of the Laplace transform will be

presented.

Marstrand's Theorem is a one of the classic results of Geometric Measure Theory, amongst other things it says that fractal measures do not have density. All methods of proof have used symmetry properties of Euclidean space in an essential way. We will present an elementary history of the subject and state a version of Marstrand's theorem which holds for spaces whose unit ball is a polytope.

We present a large deviation result for the behaviour of the

end-point of a diffusion under the influence of a strong drift. The rate

function can be explicitely determined for both attracting and repelling

drift. It transpires that this problem cannot be solved using

Freidlin-Wentzel theory alone. We present the main ideas of a proof which

is based on the Girsanov-Formula and Tauberian theorems of exponential type.

Let {X_n} be a sequence of bounded, real-valued random variables.

Assume that the partial-sums processes {S_n}, where S_n=X_1+...+X_n,

satisfies the large deviation principle with a convex rate-function, I().

Given an observation of the process {X_n}, how would you estimate I()? This

talk will introduce an estimator that was proposed to tackle a problem in

telecommunications and discuss it's properties. In particular, recent

results regarding the large deviations of estimating I() will be presented.

The significance of these results for the problem which originally motivated

the estimator, estimating the tails of queue-length distributions, will be

demonstrated. Open problems will be mentioned and a tenuous link to Oxford's

Mathematical Institute revealed.