Past

Expander graphs are sparse finite graphs with strong connectivity properties, on account of which they are much sought after in the construction of networks and in coding theory. Surprisingly, the first examples of large expander graphs came not from combinatorics, but from the representation theory of semisimple Lie groups. In this introductory talk, I will outline some of the history of the emergence of such examples from group theory, and give several applications of expander graphs to group theoretic problems.

See http://www.essex.ac.uk/maths/people/fremlin/chap53.pdf (538S)

The reaction-diffusion master equation (RDME) is a popular model in systems biology. In the RDME, diffusion is modeled as discrete jumps between voxels in the computational domain. However, it has been demonstrated that a more fine-grained model is required to resolve all the dynamics of some highly diffusion-limited systems.

In Greenʼs Function Reaction Dynamics (GFRD), a method based on the Smoluchowski model, diffusion is modeled continuously in space.

This will be more accurate at fine scales, but also less efficient than a discrete-space model.

We have developed a hybrid method, combining the RDME and the GFRD method, making it possible to do the more expensive fine-grained simulations only for the species and in the parts of space where it is required in order to resolve all the dynamics, and more coarse-grained simulations where possible. We have applied this method to a model of a MAPK-pathway, and managed to reduce the number of molecules simulated with GFRD by orders of magnitude and without an appreciable loss of accuracy.

Derived moduli stacks extend moduli stacks to give families over simplicial or dg rings. Lurie's representability theorem gives criteria for a functor to be representable by a derived geometric stack, and I will introduce a variant of it. This establishes representability for problems such as moduli of sheaves and moduli of polarised schemes.

With the advent of powerful computers and the internet, our ability to collect and store large amounts of data has improved tremendously over the past decades. As a result, methods for extracting useful information from these large datasets have gained in importance. In many cases the data can be conveniently represented as a network, where the nodes are entities of interest and the edges encode the relationships between them. Community detection aims to identify sets of nodes that are more densely connected internally than to the rest of the network. Many popular methods for partitioning a network into communities rely on heuristically optimising a quality function. This approach can run into problems for large networks, as the quality function often becomes near degenerate with many near optimal partitions that can potentially be quite different from each other. In this talk I will show that this near degeneracy, rather than being a severe problem, can potentially allow us to extract additional information

based on joint work with Charles Fefferman (Princeton) and Kevin Luli (Yale).

Vinogradov's three prime theorem resolves the weak Goldbach conjecture for sufficiently large integers. We discuss some of the ideas behind the proof, and discuss some of the obstacles to completing a proof of the odd goldbach conjecture.

For a fixed background manifold $M$ and parameter-space $X$, the associated configuration space is the space of $n$-point subsets of $M$ with parameters drawn from $X$ attached to each point of the subset, topologised in a natural way so that points cannot collide. One can either remember or forget the ordering of the n points in the configuration, so there are ordered and unordered versions of each configuration space.

It is a classical result that the sequence of unordered configuration spaces, as $n$ increases, is homologically stable: for each $k$ the degree-$k$ homology is eventually independent of $n$. However, a simple counterexample shows that this result fails for ordered configuration spaces. So one could ask whether it's possible to remember part of the ordering information and still have homological stability.

The goal of this talk is to explain the ideas behind a positive answer to this question, using 'oriented configuration spaces', in which configurations are equipped with an ordering - up to even permutations - of their points. I will also explain how this case differs from the unordered case: for example the 'rate' at which the homology stabilises is strictly slower for oriented configurations.

If time permits, I will also say something about homological stability with twisted coefficients.

We find the order of growth of the typical number of components of zero sets of smooth random functions of several real variables. This might be thought as a statistical version of the (first half of) 16th Hilbert problem. The primary examples are various ensembles of Gaussian real-valued polynomials (algebraic or trigonometric) of large degree, and smooth Gaussian functions on the Euclidean space with translation-invariant distribution.

Joint work with Fedor Nazarov.

It is well known that electrical resistance arguments provide (usually) the best method for determining whether a graph is transient or recurrent. In this talk I will discuss a similar characterization of 'sub-diffusive behaviour' -- this occurs in spaces with many obstacles or traps.

The characterization is in terms of the energy of functions in annuli.

A simple formula is given for the $n$-field tree-level MHV gravitational

amplitude, based on soft limit factors. It expresses the full $S_n$ symmetry

naturally, as a determinant of elements of a symmetric ($n \times n$) matrix.

Shapiro's Conjecture says that two classical exponential polynomials over the complexes can have infinitely many common zeros only for algebraic reasons. I will explain the history of this, the connection to Schanuel's Conjecture, and sketch a proof for the complexes using Schanuel, as well as an unconditional proof for Zilber's fields.

In a variety of problems in engineering and applied science, the goal is to design or control a network of dynamical agents so as to achieve some desired asymptotic behaviour. Examples include consensus and rendez-vous problems in robotics, synchronization of generator angles in power grids or coordination of oscillations in bacterial populations. A pressing challenge in all of these problems is to derive appropriate analytical tools to prove convergence towards the target behaviour. Such tools are not only invaluable to guarantee the desired performance, but can also provide important guidelines for the design of decentralized control strategies to steer the collective behaviour of the network of interest in a desired manner. During this talk, a methodology for analysis and design of convergence in networks will be presented which is based on the use of a classical, yet not fully exploited, tool for convergence analysis: contraction theory. As opposed to classical methods for stability analysis, the idea is to look at convergence between trajectories of a system of interest rather that at their asymptotic convergence towards some solution of interest. After introducing the problem, a methodology will be derived based on the use of matrix measures induced by non-Euclidean norms that will be exploited to design strategies to control the collective behaviour of networks of dynamical agents. Representative examples will be used to illustrate the theoretical results.

We study exponential sums of Weyl type taken over roots of quadratic congruences. We are particularly interested in the situation where the range of summation is small compared to the discriminant of the polynomial. We are then able to give a number of arithmetic applications.

This is work which is joint with W. Duke and H. Iwaniec.

The advent of multicore processors represents a disruptive event in the history of computer science as conventional parallel programming paradigms are proving incapable of fully exploiting their potential for concurrent computations. The need for different or new programming models clearly arises from recent studies which identify fine-granularity and dynamic execution as the keys to achieve high efficiency on multicore systems. This talk shows how these models can be effectively applied to the multifrontal method for the QR factorization of sparse matrices providing a very high efficiency achieved through a fine-grained partitioning of data and a dynamic scheduling of computational tasks relying on a dataflow parallel programming model. Moreover, preliminary results will be discussed showing how the multifrontal QR factorization can be accelerated by using low-rank approximation techniques.

Complex structures on a closed surface of genus at least 2 are in

one-to-one correspondence with hyperbolic metrics, so that there is a

single space, Teichmüller space, parametrising all possible complex

and hyperbolic structures on a given surface (up to isotopy). We will

explore how complex and hyperbolic geometry interact in Teichmüller

space.

A construction by McCool gives rise to a finite presentation for the stabiliser of a finite set of conjugacy classes in a free group under the action of Aut(F_n) or Out(F_n). An important concept of my talk are rigid elements, which will allow to simplify these huge presentations. Finally I will sketch applications to centralisers in Aut(F_n).

In this talk, I will introduce stochastic models to describe the state of the chemical networks using continuous-time Markov chains.
First, I will talk about the multiscale approximation method developed by Ball, Kurtz, Popovic, and Rempala (2006). Extending their method, we construct a general multiscale approximation in chemical reaction networks. We embed a stochastic model for a chemical reaction network into a family of models parameterized by a large parameter N. If reaction rate constants and species numbers vary over a wide range, we scale these numbers by powers of the parameter N. We develop a systematic approach to choose an appropriate set of scaling exponents. When the scaling suggests subnetworks have di erent time-scales, the subnetwork in each time scale is approximated by a limiting model involving a subset of reactions and species.

After that, I will briefly introduce Gaussian approximation using a central limit theorem, which gives a model with more detailed uctuations which may be not captured by the limiting models in multiscale approximations.

Next, we consider modeling of a chemical network with both reaction and diffusion.
We discretize the spatial domain into several computational cells and model diffusion as a reaction where the molecule of species in one computational cell moves to the neighboring one. In this case, the important question is how many computational cells we need to use for discretization to get balance between e ective diffusion rates and reaction rates both of which depend on the computational cell size. We derive a condition under which concentration of species converges to its uniform solution exponentially. Replacing a system domain size in this condition by computational cell size in our stochastic model, we derive an upper bound
for the computational cell size.

Finally, I will talk about stochastic reaction-diffusion models of pattern formation. Spatially distributed signals called morphogens influence gene expression that determines phenotype identity of cells. Generally, different cell types are segregated by boundary between
them determined by a threshold value of some state variables. Our question is how sensitive the location of the boundary to variation in parameters. We suggest a stochastic model for boundary determination using signaling schemes for patterning and investigate their effects on the variability of the boundary determination between cells.