KU Leuven

Function approximation, as a goal in itself or as an ingredient in scientific computing, typically relies on having a basis. However, in many cases of interest an obvious basis is not known or is not easily found. Even if it is, alternative representations may exist with much fewer degrees of freedom, perhaps by mimicking certain features of the solution into the “basis functions" such as known singularities or phases of oscillation. Unfortunately, such expert knowledge typically doesn’t match well with the mathematical properties of a basis: it leads instead to representations which are either incomplete or overcomplete. In turn, this makes a problem potentially unsolvable or ill-conditioned. We intend to show that overcomplete representations, in spite of inherent ill-conditioning, often work wonderfully well in numerical practice. We explore a theoretical foundation for this phenomenon, use it to devise ground rules for practitioners, and illustrate how the theory and its ramifications manifest themselves in a number of applications.

---

A link for this talk will be sent to our mailing list a day or two in advance.  If you are not on the list and wish to be sent a link, please contact @email.

I will discuss how  polynomials with a non-hermitian orthogonality on a contour in the complex plane arise in certain random tiling problems. In the case of periodic weightings the orthogonality is matrixvalued.

In work with Maurice Duits (KTH Stockholm) the Riemann-Hilbert problem for matrix valued orthogonal polynomials was used to obtain asymptotics for domino tilings of the two-periodic Aztec diamond. This model is remarkable since it gives rise to a gaseous phase, in addition to the more common solid and liquid phases.

In the structure theory of operator algebras, it has been a reliable theme that a classification of interesting classes of objects is usually followed by a classification of group actions on said objects. A good example for this is the complete classification of amenable group actions on injective factors, which complemented the famous work of Connes-Haagerup. On the C*-algebra side, progress in the Elliott classification program has likewise given impulse to the classification of C*-dynamics. Although C*-dynamical systems are not yet understood at a comparable level, there are some sophisticated tools in the literature that yield satisfactory partial results. In this introductory talk I will outline the (known) classification of finite group actions with the Rokhlin property, and in the process highlight some themes that are still relevant in today's state-of-the-art.

Highly oscillatory integrals arise in a range of wave-based problems. For example, they may occur when a basis for a boundary element has been enriched with oscillatory functions, or as part of a localised approximation to various short-wavelength phenomena. A range of contemporary methods exist for the efficient evaluation of such integrals. These methods have been shown to be very effective for model integrals, but may require expertise and manual intervention for
integrals with higher complexity, and can be unstable in practice.

The PathFinder toolbox aims to develop robust and fully automated numerical software for a large class of oscillatory integrals. In this talk I will introduce the method of numerical steepest descent (the technique upon which PathFinder is based) with a few simple examples, which are also intended to highlight potential causes for numerical instability or manual intervention. I will then explain the novel approaches that PathFinder uses to avoid these. Finally I will present some numerical examples, demonstrating how to use the toolbox, convergence results, and an application to the parabolic wave equation.

Functions are usually approximated numerically in a basis, a non-redundant and complete set of functions that span a certain space. In this talk we highlight a number of benefits of using overcomplete sets, in particular using the more general notion of a "frame". The main 

benefit is that frames are easily constructed even for functions of several variables on domains with irregular shapes. On the other hand, allowing for possible linear depencies naturally leads to ill-conditioning of approximation algorithms. The ill-conditioning is 

potentially severe. We give some useful examples of frames and we first address the numerical stability of best approximations in a frame. Next, we briefly describe special point sets in which interpolation turns out to be stable. Finally, we review so-called Fourier extensions and an efficient algorithm to approximate functions with spectral accuracy on domains without structure.

Rational Krylov subspaces have been proven to be useful for many applications, like the approximation of matrix functions or the solution of matrix equations. It will be shown that extended and rational Krylov subspaces —under some assumptions— can be retrieved without any explicit inversion or system solves involved. Instead we do the necessary computations of $A^{-1} v$ in an implicit way using the information from an enlarged standard Krylov subspace.

It is well-known that both for classical and extended Krylov spaces, direct unitary similarity transformations exist providing us the matrix of recurrences. In practice, however, for large dimensions computing time is saved by making use of iterative procedures to gradually gather the recurrences in a matrix. Unfortunately, for extended Krylov spaces one is required to frequently solve, in some way or another a system of equations. In this talk both techniques will be integrated. We start with an orthogonal basis of a standard Krylov subspace of dimension $m+m+p$. Then we will apply a unitary similarity built by rotations compressing thereby significantly the initial subspace and resulting in an orthogonal basis approximately spanning an extended or rational Krylov subspace.

Numerical experiments support our claims that this approximation is very good if the large Krylov subspace contains $A^{-(m+1)} v$, …, $A^{-1} v$ and thus can culminate in nonneglectable dimensionality reduction and as such also can lead to time savings when approximating, e.g., matrix functions.

This talk consists of three parts. As a motivation, we are first going to introduce von Neumann algebras associated with discrete groups and briefly describe their interplay with measurable group theory. Next, we are going to consider group von Neumann algebras of general locally compact groups and highlight crucial differences between the discrete and the non-discrete case. Finally, we present some recent results on group von Neumann algebras associated with certain locally compact HNN-extensions.

Lattice rules are equal-weight quadrature/cubature rules for the approximation of multivariate integrals which use lattice points as the cubature nodes. The quality of such cubature rules is directly related to the discrepancy between the uniform distribution and the discrete distribution of these points in the unit cube, and so, they are a kind of low-discrepancy sampling points. As low-discrepancy based cubature rules look like Monte Carlo rules, except that they use cleverly chosen deterministic points, they are sometimes called quasi-Monte Carlo rules.

\\

\\

The talk starts by motivating the usage of Monte Carlo and then quasi-Monte Carlo methods after which some more recent developments are discussed. Topics include: worst-case errors in reproducing kernel Hilbert spaces, weighted spaces and the construction of lattice rules and sequences.

\\

\\

In the minds of many, quasi-Monte Carlo methods seem to share the bad stanza of the Monte Carlo method: a brute force method of last resort with slow order of convergence, i.e., $O(N^{-1/2})$. This is not so.

While the standard rate of convergence for quasi-Monte Carlo is rather slow, being $O(N^{-1})$, the theory shows that these methods achieve the optimal rate of convergence in many interesting function spaces.

E.g., in function spaces with higher smoothness one can have $O(N^{-\alpha})$, $\alpha > 1$. This will be illustrated by numerical examples.

The evaluation of oscillatory integrals is often considered to be a computationally challenging problem. However, in many cases, the exact opposite is true. Oscillatory integrals are cheaper to evaluate than non-oscillatory ones, even more so in higher dimensions. The simplest strategy that illustrates the general approach is to truncate an asymptotic expansion, where available. We show that an optimal strategy leads to the construction of certain unconventional Gaussian quadrature rules, that converge at twice the rate of asymptotic expansions. We examine a range of one-dimensional and higher dimensional, singular and highly oscillatory integrals.