Past

Relative entropy weighted optimization is convex optimization problem over the space of probability measures. Many convex optimization problems can be rephrased as such a problem. This is particularly useful since this problem type admits a quasi-explicit solution (i.e. as the expectation over a random variable), which immediately provides a Monte-Carlo method for numerically computing the solution of the optimization problem.

In this talk we discuss the background and application of this approach to stochastic optimal control problems, which may be considered as relative entropy weighted problems with Wiener space as probability space, and its connection with the theory of large deviations for Brownian functionals. As a particular application we discuss the minimization of the local time in a given point of Brownian motion with drift.

Recent years have seen rapid expansion of the capabilities 
to recreate in vivo conditions using in vitro microfluidic assays.  
A wide range of single cell and cell population behaviors can now 
be replicated, controlled and imaged for detailed studies to gain 
new insights.  Such experiments also provide useful fodder for 
computational models, both in terms of estimating model parameters 
and for testing model-generated hypotheses.  Our experiments have 
focused in several different areas.  
1) Single cell migration experiments in 3D collagen gels have 
revealed that interstitial flow can lead to biased cell migration 
in the upstream direction, with important implications to cancer 
invasion.  We show this phenomenon to be a consequence of 
integrin-mediated mechanotransduction.  
2) Endothelial cells seeded in fibrin gels form perfusable 
vascular networks within 2-3 days through a process termed 
“vasculogenesis”.  The process by which cells sense their 
neighbours, extend projections and form anastomoses, and 
generate interconnected lumens can be observed through time-lapse 
microscopy.  
3) These vascular networks, once formed, can be perfused with 
medium containing cancer cells that become lodged in the 
smaller vessels and proceed to transmigrate across the endothelial 
barrier and invade into the surrounding matrix.  High resolution 
imaging of this process reveals a fascinating sequence of events 
involving interactions between a tumour cell, endothelial cells, 
and underlying matrix.  These three examples will be presented 
with a view toward gaining new insights through computational 
modelling of the associated phenomena.

We will present in this lecture the global existence of weak solutions and the local existence and uniqueness of strong-in-time solutions for the fully-coupled Navier-Stokes/Q-tensor system on a bounded domain $\O\subset\mathbb{R}^d$ ($d=2,3$) with inhomogenerous Dirichlet and Neumann or mixed boundary conditions. Our result is valid for any physical parameter $\xi$ and we consider the Navier-Stokes equations with a general (but smooth) viscosity coefficient.

The good use of condiments is one of the secrets of a tasty quiche. If you want to delight your guests, add a pinch of ground pepper or cinnamon to the yellow liquid formed by the mix of the eggs and the crème fraiche. Here, is a surprise : even if the liquid is at rest, the pinch of milled pepper spreads by itself at the surface of the mixture. It expands in a circular way, and within a few seconds, it covers an area equal to several times its initial one. Why does it spread like that ? What factors influence this dispersion ? I will present some experiments and mathematical models of this process.

*****     PLEASE NOTE THIS SEMINAR TAKES PLACE ON TUESDAY     *****

Reduced ODE models describing coarsening dynamics of droplets in nanometric polymer film interacting on solid substrate in the presence of large slippage at the liquid/solid interface are derived from one-dimensional lubrication equations. In the limiting case of the infinite slip length corresponding to the free suspended films a collision/absorption model then arises and is solved explicitly. The exact collision law is derived. Existence of a threshold at which the collision rates switch from algebraic to exponential ones is shown.

*****     PLEASE NOTE THIS SEMINAR TAKES PLACE ON TUESDAY     *****

*****     PLEASE NOTE THIS SEMINAR TAKES PLACE ON MONDAY     *****

Ultralow interfacial tension mixtures have interfacial tensions that are 1,000 times, or more, lower than typical oil-water systems. Despite the recent utility of ultralow interfacial tension mixtures in industry and research, quantifying the interfacial tension remains challenging. Here I describe a technique that measures ultralow interfacial tensions by magnetically deflecting paramagnetic spheres in a co-flow microfluidic device. This method involves the tuning of the distance between the co-flowing interface and the magnetic field source, and observing the behavior of the magnetic particles as they approach the liquid-liquid interface--the particles either pass through or are trapped. I demonstrate the effectiveness of this technique for measuring very low interfacial tensions by testing solutions of different surfactant concentrations, and show that the results are comparable with measurements made using a spinning drop tensiometer.

*****     PLEASE NOTE THIS SEMINAR TAKES PLACE ON MONDAY     *****

*****     PLEASE NOTE THIS SEMINAR WILL TAKE PLACE ON MONDAY 24TH JUNE 2013     *****

Energy equations describing magnetic and inertial confinement functions (ICF) are strongly coupled, time dependent non-linear PDEs. The huge disparity of the coefficients in the coupled non-linear equations brings tremendous numerical difficulties to get high resolution solutions. It results in highly ill-conditioned linear systems in each non-linear iteration. Solving the resulted non-linear systems is time-consuming which takes up to 90% in the total simulation time. Many customized numerical techniques have to be employed to get a robust and accurate solution.This talk will present an inexact Newton-Krylov-Schwarz framework to solve the problem, demonstrating how to integrate preconditioning, partial Jacobian matrix forming techniques, parallel computing techniques with the Newton-Krylov solvers to solve the challenging problem. The numerical results will be shown and other numerical problems will be mentioned.

*****     If anyone is planning to take the 11.36 train after the seminar to the NA conference in Glasgow a taxi from the Gibson building is being arranged. Please contact Jude, @email, to book a place in the taxi.     *****

The Bayesian approach to inverse problems is of paramount importance in quantifying uncertainty about the input to and the state of a system of interest given noisy observations. Herein we consider the forward problem of the forced 2D Navier Stokes equation. The inverse problem is inference of the forcing, and possibly the initial condition, given noisy observations of the velocity field. We place a prior on the forcing which is in the form of a spatially correlated temporally white Gaussian process, and formulate the inverse problem for the posterior distribution. Given appropriate spatial regularity conditions, we show that the solution is a continuous function of the forcing. Hence, for appropriately chosen spatial regularity in the prior, the posterior distribution on the forcing is absolutely continuous with respect to the prior and is hence well-defined. Furthermore, the posterior distribution is a continuous function of the data.

\\ \\

This is a joint work with Andrew Stuart and Kody Law (Warwick)

*****     PLEASE NOTE THIS SEMINAR WILL COMMENCE AT 12.00     *****

I will present experimental work on collective dynamics in two different systems: (i) a collection of self propelled droplets and (ii) coupled mechanical oscillators.  

In the first part, I will talk about microswimmers made from water-in-oil emulsion droplets. Following a brief description of the swimming mechanism, I will discuss some of the collective effects that emerge in quasi 1 and 2 dimensional confinements of swimming droplets. Specifically, I dwell on hydrodynamic and volume exclusion interactions, only through which these droplets can couple their motions. 

In the second part, I will present recent results about an intriguing dynamic known as a chimera state. In the world of coupled oscillators, a chimera state is the co-existence of synchrony and asynchrony in a population of identical oscillators, which are coupled nonlocally. Following nearly 10 years of intense theoretical research, it has been an imminent question whether these chimera states exist in real systems. Recently, we built an experiment with of springs, swings and metronomes and realised, for the first time, these symmetry breaking states in a purely physical system.

*****     PLEASE NOTE THIS SEMINAR WILL COMMENCE AT 12.00     *****

*****     PLEASE NOTE THIS SEMINAR WILL TAKE PLACE ON MONDAY 17TH JUNE 2013     *****

Computing is an exercise of discretization of the real world into space, time, and value. While discretization in time and space is well understood in the sciences, discretization of value is a scientific domain full of opportunity. Maxeler's Multiscale Dataflow Computing allows the programmer to finely trade off discretization of value with real performance measured in wallclock time.

In this talk I will show the connection between discretization of value and Kolmogorov Complexity on one hand and approximation theory on the other. Utilizing the above concepts together with building general purpose computing systems based on dataflow concepts, has enabled us to deliver production systems for Oil & Gas imaging (modelling, multiple elimination, RTM, Geomechanics), Finance Risk (derivatives modelling and scenario analysis), as well as many scientific application such as computing weather models, Astrochemistry, and brain simulations. Algorithms range from 3D Finite Difference, Finite Elements (sparse matrix solvers), pattern matching, conjugate gradient optimization, to communication protocols and bitcoin calculations. Published results of users of our machines show a 20-50x total advantage in computations per unit space (1U) and computations per Watt.

*****     PLEASE NOTE THIS SEMINAR WILL TAKE PLACE ON MONDAY 17TH JUNE 2013     *****

Advanced models such as Lévy models require advanced numerical methods for developing efficient pricing algorithms. Here we focus on PIDE based methods. There is a large arsenal of numerical methods for solving parabolic equations that arise in this context. Especially Galerkin and Galerkin inspired methods have an impressive potential. In order to apply these methods, what is required is a formulation of the equation in the weak sense.

We therefore classify Lévy processes according to the solution spaces of the associated parabolic PIDEs. We define the Sobolev index of a Lévy process by a certain growth condition on the symbol. It follows that for Lévy processes with a certain Sobolev index b the corresponding evolution problem has a unique weak solution in the Sobolev-Slobodeckii space with index b/2. We show that this classification applies to a wide range of processes. Examples are the Brownian motion with or without drift, generalised hyperbolic (GH), CGMY and (semi) stable Lévy processes.

A comparison of the Sobolev index with the Blumenthal-Getoor index sheds light on the structural implication of the classification. More precisely, we discuss the Sobolev index as an indicator of the smoothness of the distribution and of the variation of the paths of the process.

An application to financial models requires in particular to admit pure jump processes as well as unbounded domains of the equation. In order to deal at the same time with the typical payoffs which can arise, the weak formulation of the equation has to be based on exponentially weighted Sobolev-Slobodeckii spaces. We provide a number of examples of models that are covered by this general framework. Examples of options for which such an analysis is required are calls, puts, digital and power options as well as basket options.

The talk is based on joint work with Ernst Eberlein.

Colloidal suspensions do not freeze uniformly; rather, the frozen phase (e.g. ice) becomes segregated, trapping bulk regions of the colloid within, which leads to a fascinating variety of patterns that impact both nature and technology. Yet, despite the central importance of ice segregation in several applications, the physics are poorly understood in concentrated systems and continuum models are available only in restricted cases. I will discuss a particular set of steady-state ice segregation patterns that were obtained during a series of directional solidification experiments on concentrated suspensions. As a case study, I will focus of one of these patterns, which is very reminiscent of ice lenses observed in freezing soils and rocks; a form of ice segregation which underlies frost heave and frost weathering. I will compare these observations against an extended version of a 'rigid-ice' model used in previous frost heave studies. The comparison between theory and experiment is qualitatively correct, but fails to quantitatively predict the ice-lensing pattern. This leaves open questions about the validity of the assumptions in 'rigid-ice' models. Moreover, 'rigid-ice' models are inapplicable to the study of other ice segregation patterns. I conclude this talk with some possibilities for a more general model of freezing colloidal suspensions.