Colin Guillarmou (U Paris-Saclay and CNRS)
Thursday, Sept 24, 2026, 17:15-18:15, MSA 3.520
TBA
Registration is free but mandatory, here.Yilin Wang (ETH Zürich)
Thursday, Oct 29, 2026, 16-17, MNO 1.050
TBA
TBA
Registration is free but mandatory, here.Sophie Schbath (INRAE)
Thursday, Dec 3, 2026, 16-17, MSA 3.010
TBA
Registration is free but mandatory, here.Past talks
Richard Schwartz (Brown University)
Friday, June 27, 2025, 15-16, MSA 3.330
The optimal paper Moebius band
If you have a 1xL rectangular strip of paper, you can twist it in space and glue the length-1 ends together to make a Moebius band provided that L is large enough. If L is too small you can't do it. What is the cutoff? In this talk I will prove that you can do it if and only if L>sqrt(3). The proof is an elementary blend of geometry and topology, and anyone can understand it. This result resolves the conjecture about this, due to B. Halpern and C. Weaver, from 1977. I will also explain why an L-minimizing sequence of examples must converge to an equilateral triangle in space (up to isometries).
Olivier Benoist (CNRS / ENS - Université PSL)
Thursday, Oct 9, 2025, 17:15-18:15, MSA 3.350
Positivity and sums of squares
Hilbert's 17th problem, solved in 1927 by Emil Artin, states that a real polynomial that only takes nonnegative values is a sum of squares of rational functions. I will survey the history of this question, and present recent developments and related open problems.
Registration is free but mandatory, here.François Delarue (Univ. Cote d'Azur)
Thursday, Nov 13, 2025, 16-17, MSA 3.350
Control of large populations
The aim of this talk is to review the main ideas of large-population control. In general terms, we will consider dynamic particles, referred to as agents or individuals to emphasize that these particles are capable of controlling their own states. We will focus on the case where interactions between individuals are of mean-field type, that is, symmetric, with each agent having an asymptotically negligible influence on the rest of the population. Questions addressed by the community concern, on the one hand, how to formalize the control problem posed directly on an infinite population (a continuum), and on the other hand, how to explain the passage from a finite population to an infinite one. In this context, we will highlight the importance of the theory of partial differential equations posed on the space of probability measures.
Registration is free but mandatory, here.Thierry Levy (Sorbonne U)
Thursday, Apr 23, 2026, 16-17, MSA 3.370
Frobenius and Poincaré, a word in common
In 1895, in Paris, Henri Poincaré published his treatise /Analysis situs/, in which he did nothing less than founding the branch of mathematics known today as algebraic topology, almost exactly at the same time where in Berlin, Ferdinand Frobenius was writing his seminal paper /Über Gruppencharaktere/, of which the modern theory of group characters would quickly emerge. These two works seemed to have little else in common than the use of the notion of group, and for quite different reasons. However, a certain specific word in the elements of a group appeared in both of them: in the work of Poincaré as a relation defining the fundamental group of a surface, and in the work of Frobenius as an equation of which the notion of group characters allowed him to count the number of solutions. I will present the two mathematical threads that unexpectedly tie together around this word, and point to some modern consequences of this connection.
Registration is free but mandatory, here.Nicolas Verzelen (INRAE)
Thursday, June 18, 2026, 16-17, MSA 3.330
Statistical and Computational challenges in clustering
Clustering is one of the most iconic problems in unsupervised learning. To evaluate the performances of clustering procedures, it is standard to model the data as realizations of a Gaussian mixture model and to quantify the ability of these procedures to recover the unknown groups of this mixture. The best possible conditions under which perfect clustering is possible have been characterized for this model. Unfortunately, the corresponding clustering methods suffer from an exponential-time complexity, making them unable to deal with large-scale problems. Based on the replica heuristic from statistical physics, Lesieur et al. (2016) have conjectured the existence of computational barriers, that is regimes where clustering is statistically possible but no polynomial-time procedures is able to do so. Along with similar conjectures for a wide range of learning problems (sparse PCA, tensor models,...), this led to the emergence of a research field whose aim is to provide evidence for these computational barriers. In this presentation, I will describe recent advances in this research agenda with a strong focus on the low-degree polynomial paradigm.
Registration is free but mandatory, here.David Fisher (Rice U.)
Wednesday, July 1, 2026, 16-17, MSA 3.370
Totally geodesic submanifolds in negative curvature
The study of closed geodesics in negatively curved manifolds is an old topic. In this talk I will discuss the less investigated topic of closed totally geodesic submanifolds of dimension at least two. For generic metrics there are no such submanifolds, so the question becomes what are the consequences of assuming there are infinitely many. This turns out to be extremely restrictive for the geometry and topology of the manifold as has been discovered in a series of works in the last 7 years.
Registration is free but mandatory, here.