Past

I will introduce a class of mean-field games under forward performance and for general risk preferences. Players interact through competition in fund management, driven by relative performance concerns in an asset diversification setting. This results in a common-noise mean field game. I will present the value and the optimal policies of such games, as well as some concrete examples. I will also discuss the partial information case, i.e.. when the risk premium is not directly observed. 

Via Poincaré duality, fundamental groups of aspherical manifolds have (appropriately shifted) isomorphisms between their homology and cohomology. In a 1973 Inventiones paper, Bieri and Eckmann defined a broader notion of a Duality Group, where the isomorphism between homology and cohomology can be twisted by what they called a Dualizing Module. Examples of these groups in geometric group theory (after passing to a finite-index subgroup) include $GL(n,\mathbb{Z})$, mapping class groups, and automorphism groups of free groups.

In work-in-progress with Thomas Wasserman we are looking into the following puzzle: the examples of duality groups that we know of that do not come from manifolds all have classifying spaces that satisfy a weaker local condition called the Cohen-Macaulay property. These spaces also satisfy weaker (twisted) versions of Poincaé duality via their local homology sheaves (or local cohomology cosheaves), and we are attempting to understand more about the links between these geometric versions of duality and the algebraic notion of a duality group. The goal of the talk is to explain more about the words used in the above paragraphs and say where we have got to so far.


 

A loop of Hamiltonian diffeomorphisms of a symplectic manifold $X$ defines, by clutching, a symplectic fibration over the two-sphere with fibre $X$.  We prove that the integral cohomology of the total space splits additively, answering a question of McDuff, and extending the rational cohomology analogue proved by Lalonde-McDuff-Polterovich in the late 1990’s. The proof uses a virtual fundamental class of moduli spaces of sections of the fibration in Morava K-theory. This talk reports on joint work with Mohammed Abouzaid and Mark McLean.

Non-smooth optimization is a core ingredient of many imaging or machine learning pipelines. Non-smoothness encodes structural constraints on the solutions, such as sparsity, group sparsity, low-rank and sharp edges. It is also the basis for the definition of robust loss functions such as the square-root lasso.  Standard approaches to deal with non-smoothness leverage either proximal splitting or coordinate descent. The effectiveness of their usage typically depend on proper parameter tuning, preconditioning or some sort of support pruning. In this work, we advocate and study a different route. By over-parameterization and marginalising on certain variables (Variable Projection), we show how many popular non-smooth structured problems can be written as smooth optimization problems. The result is that one can then take advantage of quasi-Newton solvers such as L-BFGS and this, in practice, can lead to substantial performance gains. Another interesting aspect of our proposed solver is its efficiency when handling imaging problems that arise from fine discretizations (unlike proximal methods such as ISTA whose convergence is known to have exponential dependency on dimension). On a theoretical level, one can connect gradient descent on our over-parameterized formulation with mirror descent with a varying Hessian metric. This observation can then be used to derive dimension free convergence bounds and explains the efficiency of our method in the fine-grids regime.

The line operators in a 2+1D TQFT form an algebraic structure called a modular tensor category (MTC). There is a natural action of a Galois group on MTCs which maps a given TQFT to other 'Galois conjugate' TQFTs. I will describe this Galois action and give several examples of Galois conjugate TQFTs. Galois action on a unitary TQFT can result in a non-unitary TQFT. I will derive a sufficient condition under which unitarity is preserved. Finally, I will describe the invariance of 0-form and 1-form symmetries of TQFTs under Galois action.    

*Note the different room location (L2) to usual Fridays@4 sessions*

This week is Mental Health Awareness Week. To mark this, Rebecca Reed from Siendo will deliver a session on mental health and wellbeing. The session will cover the following things: 

- The importance of finding a balance with achievement and managing stress and pressure.
- Coping mechanisms work with stresses at work in a positive way (not seeing all stress as bad).
- The difficulties faced in the HE environment, such as the uncertainty felt within jobs and research, combined with the high expectations and workload. 

I'll review 2104.13932, where we analyze the supersymmetric index of N=4 SU(N) Super Yang-Mills using the Bethe Ansatz approach, expressing it as a sum and concentrating on some family of contributions to the sum. We show that in the large N limit each term in this family corresponds to the contribution of a different euclidean black hole to the partition function of the dual gravitational theory. By taking into account non-perturbative contributions (wrapped D3-branes), we further show a one to one match between the contributions of the gravitational saddles and this family of contributions to the index, both at the perturbative and non-perturbative levels. I'll end with some new results regarding the Bethe Ansatz expansion and the information one could extract from it.

The statistical analysis of data which lies in a non-Euclidean space has become increasingly common over the last decade, starting from the point of view of shape analysis, but also being driven by a number of novel application areas. However, while there are a number of interesting avenues this analysis has taken, particularly around positive definite matrix data and data which lies in function spaces, it has increasingly raised more questions than answers. In this talk, I'll introduce some non-Euclidean data from applications in brain imaging and in linguistics, but spend considerable time asking questions, where I hope the interaction of statistics and topological data analysis (understood broadly) could potentially start to bring understanding into the applications themselves.

We can learn a lot about an integral domain by studying the Galois group of its fraction field. These groups are generally quite complicated and hard to understand, but their representations, so-called Galois representations, contain more easily accessible information. These also play the lead in many important theorems and conjectures of modern maths, such as the Modularity theorem and the Langlands programme. In this talk we give a quick introduction to Galois representations, motivated by lots of examples aimed at a general algebraist audience, and talk about some open problems.

Understanding the molecular mechanisms that control the biology of health and disease requires development of models that traverse multiple scales of organisation in order to encapsulate the relationships between genes and linking to observable phenotypes. Measuring, parameterising and simulating the entire system that determines these phenotypes in exhaustive detail is typically impossible due to the underlying biological complexity, our limited knowledge and the paucity of available data. For example, approximately one third of human genes are poorly characterised and most genes perform multiple functions, which manifest according to the surrounding biochemical context. Indeed, new functions continue to emerge even for deeply studied genes. Therefore, simplifying abstractions in concert with empirical analysis of matched genome-scale and descriptive data are valuable strategies to fill knowledge gaps relevant to a focused biomedical question or hypothesis.

Epithelial plasticity is a key driver of cancer progression and is associated with the most life-threatening phenotypes; specifically, metastasis and drug resistance. Computational methods developed in my group enable modelling the molecular control of important cancer phenotypes. We applied a machine learning approach for genome-wide context-specific biochemical interaction network inference (CoSNI) to map gene function for the Epithelial to Mesenchymal Transition cell programme (EMT_MAP), predicting new mechanisms in control of cancer invasion. Analysis of patient data with EMT_MAP and our NetNC algorithm [Cancers 2020;12:2823; https://github.com/overton-group/NetNC] enabled discovery of candidate renal cancer prognostic markers with clear advantages over standard statistical approaches. NetNC recovers the network-defined signal in noisy data, for example distinguishing functional EMT Transcription Factor targets from ‘neutral’ binding sites and defining biologically coherent modules in renal cancer drug response time course data. These and other approaches, including SynLeGG (Nucleic Acids Research 2021;49:W613-8, www.overton-lab.uk/synlegg) and an information-theoretic approach to causality (GABI) offer mechanistic insights and opportunity to predict candidate cancer Achilles’ heels for drug discovery. Computational results were validated in follow-up experiments, towards new clinical tools for precision oncology.

The discrete Fourier transform is fundamental in modern communication systems.  It is used to generate and process (i.e. modulate and demodulate) the signals transmitted in 4G, 5G, and wifi systems, and is always implemented by one of the fast Fourier transforms (FFT) algorithms.  It is possible to generalize the FFT to work correctly on input vectors with periodic missing values.   I will consider whether this has applications, such as more general transmitted signal waveforms, or further applications such as spectral density estimation for time series with missing data.  More speculatively, can we generalize to "recursive" missing values, where the non-missing blocks have gaps?   If so, how do we optimally recognize such a pattern in a given time series?

Last summer, there was a lot of activity regarding an old conjecture of van der Waerden, culminating in its solution by Bhargava, and including joint work by Sam Chow and myself on which I want to report in this talk: We showed that the number of irreducible monic integer polynomials of degree n, with coefficients in absolute value bounded by H, which have Galois group different from S_n and A_n, is of order of magnitude O(H^{n-1.017}), providing that n is at least 3 and is different from 7,8,10. Apart from the alternating group and excluding degrees 7,8,10, this establishes the aforementioned conjecture to the effect that irreducible non-S_n polynomials are significantly less frequent than reducible polynomials.

The Langlands program is a far-reaching collection of conjectures that relate different areas of mathematics including number theory and representation theory. A fundamental problem on the representation theory side of the Langlands program is the construction of all (irreducible, smooth, complex or mod-$\ell$) representations of p-adic groups. I will provide an overview of our understanding of the representations of p-adic groups, with an emphasis on recent progress including joint work with Kaletha and Spice that introduces a twist to the story, and outline some applications.

That the linear systems arising upon the discretization of elliptic PDEs can be solved efficiently is well-known, and iterative solvers that often attain linear complexity (multigrid, Krylov methods, etc) have proven very successful. Interestingly, it has recently been demonstrated that it is often possible to directly compute an approximate inverse to the coefficient matrix in linear (or close to linear) time. The talk will argue that such direct solvers have several compelling qualities, including improved stability and robustness, the ability to solve certain problems that have remained intractable to iterative methods, and dramatic improvements in speed in certain environments.

After a general introduction to the field, particular attention will be paid to a set of recently developed randomized algorithms that construct data sparse representations of large dense matrices that arise in scientific computations. These algorithms are entirely black box, and interact with the linear operator to be compressed only via the matrix-vector multiplication.

We derive quantitatively the weak and strong Harnack inequality for kinetic Fokker--Planck type equations with a non-local diffusion operator for the full range of the non-locality exponents in (0,1).  This implies Hölder continuity.  We give novel proofs on the boundedness of the bilinear form associated to the non-local operator and on the construction of a geometric covering accounting for the non-locality to obtain the Harnack inequalities.  Our results apply to the inhomogeneous Boltzmann equation in the non-cutoff case.

Macroscopic Transport in Heterogeneous Porous Materials

Lucy Auton

Solute transport in porous materials is a key physical process in a wide variety of situations, including contaminant transport, filtration, lithium-ion batteries, hydrogeological systems, biofilms, bones and soils. Despite the prevalence of solute transport in porous materials, the effect of microstructure on flow and transport remains poorly understood and improving our understanding of this remains a major challenge.  In this presentation, I consider a two-dimensional microstructure comprising an array of obstacles of smooth but arbitrary shape, the size and spacing of which can vary along the length of the porous medium, allowing for anisotropy.  I use a nontrivial extension to classical homogenisation theory via the method of multiple scales to rigorously upscale the novel problem involving cells of varying area. This results in simple effective continuum equations for macroscale flow and transport where the effect of the microscale geometry on the macroscopic transport and removal is encoded within these simple macroscale equations via effective parameters such as an effective local anisotropic diffusivity and an effective local adsorption rate.  For a simple example geometry I exploit the two degrees of microstructural freedom in this problem, obstacle size and obstacle spacing, to investigate scenarios of uniform porosity but heterogenous microstructure, noting the impact this heterogeneity has on filter efficiency. 

This model constitutes the development of the core framework required to consider other crucial problems such as solute transport within soft porous materials for which there does not currently exist a simple macroscale model where the effective diffusivity and removal depend on the microstructure. Further, via this methodology I will  derive a  bespoke model for fluoride and arsenic removal filters. With this model I will be able to optimise the design of fluoride-removal filters which are being deployed across rural India. The design optimisation will both increase filter lifespan and reduce filter cost, enabling more people to access safe drinking water

Averaged interface conditions: evaporation fronts in porous media

Ellen Luckins

Homogenisation methods are powerful tools for deriving effective PDE models for processes incorporating multiple length-scales. For physical systems in which interface processes are crucial to the overall system, we might ask how the microstructure impacts the effective interface conditions, in addition to the PDEs in the bulk. In this talk we derive an effective model for the motion of an evaporation front through porous media, combining homogenisation and boundary layer analysis to derive averaged interface conditions at the evaporation front. Our analysis results in a new effective parameter in the boundary conditions, which encodes how the shape and speed of the porescale evaporating interfaces impact the overall drying process.

We will give a fairly self contained introduction to acylindrically hyperbolic groups, using mapping class groups as a motivating example. This will be a mainly expository talk, the aim is to make my topology seminar talk in week 5 more accessible to people who are less familiar with these topics.

The first edition of Thompson’s famous book On Growth and Form appeared in 1917. It has subsequently been regarded as a foundational work in mathematical biology and a revolutionary contribution to the field of morphology. Most existing literature credits Thompson as a lone genius who produced the 793 pages of the 1917 edition and 1116 pages of the 1942 edition. Thompson’s correspondence presents a very different picture of this tome as one arising from extensive and ongoing – perhaps sometimes unwitting? – collaboration.

G2 structure manifolds are a key ingredient in supersymmetric compactifications on seven-manifolds. We will discuss the fact that G2 structure manifolds admit refinements in the form of almost contact (3-) structures.  In fact, there are infinite dimensional spaces of these structures. We will discuss topological and differential geometric aspects of (the space of) these refinements. We will then explore applications in physics, including supersymmetry enhancement. This is based on 2101.12605.

The string 2-group is a fundamental object in string geometry, which is a refinement of spin geometry required to describe the spinning string. While many models for the string 2-group exist, the construction of a representation for it is new. In this talk, I will recall the notion of strict 2-group, and then give two examples: the automorphism 2-group of a von Neumann algebra, and the string 2-group. I will then describe the representation of the string 2-group on the hyperfinite III_1 factor, which is a functor from the string 2-group to the automorphism 2-group of the hyperfinite III_1 factor.

In this talk, I will introduce the BPS cohomology of the moduli space of Higgs bundles on a smooth projective curve of rank r and degree d using cohomological Donaldson-Thomas theory. The BPS cohomology and the intersection cohomology coincide when r and d are coprime, but they are different in general. We will see that the BPS cohomology does not depend on d. This is a generalization of the Hausel-Thaddeus conjecture to non-coprime case. I will also explain that Toda's χ-independence conjecture (and hence the strong rationality conjecture) for local curves can be proved in the same manner. This talk is based on a joint work with Naoki Koseki and another joint work with Naruki Masuda.

The sinoatrial node (SAN) is the pacemaker region of the heart.
Recently calcium signals, believed to be crucially important in heart
rhythm generation, have been imaged in intact SAN and shown to be
heterogeneous in various regions of the SAN. However, calcium imaging
is noisy, and the calcium signal heterogeneity has not been
mathematically analyzed to distinguish meaningful signals from
randomness or to identify signalling regions in an objective way. In
this work we apply methods of random matrix theory (RMT) developed for
financial data and used for analysis of various biological data sets
including β-cell collectives and EEG data. We find eigenvalues of the
correlation matrix that deviate from RMT predictions, and thus are not
explained by randomness but carry additional meaning. We use
localization properties of the eigenvectors corresponding to high
eigenvalues to locate particular signalling modules. We find that the
top eigenvector captures a common response of the SAN to action
potential. In some cases, the eigenvector corresponding to the second
highest eigenvalue appears to yield a possible pacemaker region as its
calcium signals predate the action potential. Next we study the
relationship between covariance coefficients and distance and find
that there are long range correlations, indicating intercellular
interactions in most cases. Lastly, we perform an analysis of nearest
neighbor eigenvalue distances and find that it coincides with the
universal Wigner surmise. On the other hand, the number variance,
which captures eigenvalue correlations, is a parameter that is
sensitive to experimental conditions. Thus RMT application to SAN
allows to remove noise and the global effects of the action potential
and thereby isolate the correlations in calcium signalling which are
local. This talk is based on joint work with Chloe Norris with a
preprint found here:
https://www.biorxiv.org/content/10.1101/2022.02.25.482007v1.

For graphs $G$ and $H$, we say $G \stackrel{r}{\to} H$ if every $r$-colouring of the edges of $G$ contains a monochromatic copy of $H$. Let $H[t]$ denote the $t$-blowup of $H$. The blowup Ramsey number $B(G \stackrel{r}{\to} H;t)$ is the minimum $n$ such that $G[n] \stackrel{r}{\to} H[t]$. Fox, Luo and Wigderson refined an upper bound of Souza, showing that, given $G$, $H$ and $r$ such that $G \stackrel{r}{\to} H$, there exist constants $a=a(G,H,r)$ and $b=b(H,r)$ such that for all $t \in \mathbb{N}$, $B(G \stackrel{r}{\to} H;t) \leq ab^t$. They conjectured that there exist some graphs $H$ for which the constant $a$ depending on $G$ is necessary. We prove this conjecture by showing that the statement is true when $H$ is a $3$-chromatically connected, which includes all cliques on $3$ or more vertices. We are also able to show perhaps surprisingly that for any forest $F$ there is $f(F,t)$ such that  for any $G \stackrel{r}{\to} H$, $B(G \stackrel{r}{\to} H;t)\leq f(F,t)$ i.e. the function does not depend on the ground graph $G$. This is joint work with Robert Hancock.