Le colloque a lieu (sauf exception) tous les jeudis à
16h15 en salle 17 de la Section
de mathématiques, rue du
Lièvre 2-4. Les dates indiquées en gris sont à
confirmer.
Le colloque est suivi d’un
apéro.
Le but de la géométrie
énumérative est de ``compter" le nombre des courbes holomorphes dans des
variétés algébriques. Une question naturelle est: quel est le comportement
asymptotique de ces nombres quand le dégrée ou le genre de la courbe devient
très grands? Dans ce colloque on va étudier ce
problème dans deux exemples très simples. Le premier exemple vient de la
théorie de l'intersection dans l'espace des moduli
des surfaces de Riemann, et le deuxième est encore plus élémentaire et vient de
la théorie des recouvrements de Hurwitz. Les réponses
(un théorème pour le premier exemple, une conjecture pour le deuxième) son
intimement lies a l'étude des phénomènes non-perturbatives
(instantons) en théorie de cordes.
Une formule simple permet
d'estimer la distribution spectrale du Laplacien au
voisinage de zéro à partir d'inégalités isoperimétriques
Le but de l'exposé est de
présenter quelques résultats (hélas encore trop partiels) qui vont dans ce
sens.
The matrix equations M^2 = 0 are quadratic, so to derive the linear equation Trace(M)=0 from them requires nonalgebraic operations. Are there corresponding "surprising" equations implied by the matrix equation XY=YX? This question was posed in the '60s, and still nobody knows. Even the (normalized) volume of this space {(X,Y) : XY=YX} is very difficult to compute for large matrices, and until recently was only known to start 1,3,31,1145. I'll talk about a bunch of related spaces of matrices, some of which are probably harder and some easier to understand than the commuting scheme {(X,Y) : XY=YX}, and the volumes of these spaces. Then I'll explain how physicists came up with the same set of numbers from a statistical mechanical model (making them much easier to compute), and why they are indeed the same. Some of this work is joint with Paul Zinn-Justin.
De nombreux systèmes
dynamiques chaotiques possèdent une unique mesure "physique",
décrivant les propriétés asymptotiques des orbites génériques pour la mesure de
Lebesgue. Si l'on considère une famille ft
de systèmes dynamiques dépendant différentiablement
d'un paramètre t, on peut se demander comment cette mesure mt dépend du
paramètre. D Ruelle a montré en 1997 que cette dépendance était différentiable
dans le cadre hyperbolique différentiable, et il a calculé la dérivée
("formule de la réponse linéaire"). Il a ensuite suggéré que, dans un
cadre assez vaste (plus forcément structurellement stable), cette
dépendance pourrait rester
différentiable (au sens de Whitney, si l'ensemble des bons paramètres n'est pas
un voisinage) et a proposé un candidat pour la derivée,
sous forme de série a priori divergente. Avec D Smania
nous avons montré que dans le cadre des applications dilatantes par morceaux en
dimension un, la mesure mt est dérivable en 0 si et seulement si la
famille ft est tangente en 0 à la classe
topologique de f0. La valeur
de la dérivée coïncide alors avec le candidat de Ruelle, resommé
de façon appropriée. Si le temps le permet, nous mentionnerons aussi des
conjectures et des résultats plus récents dans le cas non-uniformément
hyperbolique.
La construction principale des variétés à
clusters donne, à partir d'une matrice antisymétrisable
d'entiers, deux variétés algébriques de même dimension. L'une (de type A) est
munie d'une 2-forme fermée, l'autre (de type X) d'une structure de Poisson. Ces
variétés sont duales l'une de l'autre en un certain sens. Les deux sont munie d'une action d'un groupe discret et la variété de type
X admet une quantification canonique equivariante.
Parmi les variétés cluster on trouve les groupes de Lie, les espaces de Teichmueller, les espaces des paramètres de Stokes et
plusieurs autres variétés bien connues. On peut considérer la description
cluster comme une structure supplémentaire sur ces variétés unifiant leur propriétés géométriques, algébriques, combinatoires, Poisson
et arithmétiques.
Many applications nowadays involve high-dimensional data with p variables (or covariates), sample size n and the relation that p >> n. We focus on penalty-based estimation methods which are computationally feasible and have provable statistical and numerical properties. The Lasso (Tibshirani, 1996), an l_1-penalty method, became very popular in recent years for estimation in high-dimensional generalized linear models. Extensions to other models or data-types call for more flexible convex penalty functions, for example to handle categorical data or for improved control of smoothness in additive models. The Group-Lasso (Yuan and Lin, 2006) and a new sparsity-smoothness penalty are general and useful penalty functions for many high-dimensional models beyond GLM's. Fast coordinatewise descent algorithms can be used for solving the corresponding convex optimization problems which allow to easily deal with large dimensionality p (e.g. p ~ 10^6, n ~ 10^3). The talk includes: (i) a review of Lasso-type methods; (ii) new flexible penalty functions and fast algorithms (package "lasso"); and (iii) some illustrations for bio-molecular data.
Si X une variété Riemannienne, le Laplacien hypoelliptique est un
opérateur agissant sur l’espace total du fibré cotangent ou du fibré tangent de
X, dépendant d’un paramètre b > 0, qui interpole de manière adéquate entre
la théorie de Hodge usuelle de X, et le flot
géodésique. Elle est produite à l’aide d’une théorie de Hodge
exotique de l’espace total des fibrés tangents ou cotangents, par une
déformation convenable des opérateurs d + d*. Le Laplacien
hypoelliptique est essentiellement la somme de
l’oscillateur harmonique le long des fibres et du champ de vecteurs engendrant
le flot géodésique. Les motivations pour la construction de cette déformation
sont multiples. En théorie de de Rham,
on peut considérer le Laplacien hypoelliptique
comme une version semiclassique d’une déformation de
Witten du Laplacien sur l’espace des lacets, associé
à la fonctionnelle d’énergie. Dans l’exposé, on donnera les motivations
heuristiques pour la construction du Laplacien hypoelliptique, et on expliquera sa construction en théorie
de de Rham. On exposera les
résultats obtenus avec Gilles Lebeau sur l’analyse de cet opérateur. Enfin on
donnera diverses applications du Laplacien hypoelliptique.
I sketch what it is
supposed to mean to quantize gauge theory, and how this can be made more
concrete in perturbation theory and also by starting with a finite-dimensional
lattice approximation. Based on real experiments and computer simulations,
quantum gauge theory in four dimensions is believed to have a mass gap. This is
one of the most fundamental facts that makes the
Universe the way it is. arXiv:0812.4512
Enumerative geometry deals with counting geometric objects (lines, curves, planes) subject to certain restrictions (passage through a collection of points, tangency to a curve, etc).
I will discuss a wide class of such problems on a crossroads of a real enumerative geometry and smooth topology and explain its relation to the theory of finite type invariants. I will also present a simple approach to such problems using maps of configuration spaces and intersections. The talk does not assume any preliminary knowledge of the subject and should be easily accessible to students.
The classical Littlewood-Richardson coefficients are remarkable nonnegative integers which occupy a prominent place in combinatorics, representation theory and geometry. We review two versions of the original rule for their calculation then follow by a natural generalization of these coefficients called the Littlewood-Richardson polynomials and give a combinatorial rule for their calculation. Then we discuss two applications of this rule: we find the product of the Casimir elements for the general linear Lie algebra in the basis of the quantum immanants constructed by A. Okounkov and G. Olshanski. The same rule yields a positive and stable formula for the product of equivariant Schubert classes on the Grassmannian. The first positive formula for such a product was given by A. Knutson and T. Tao by using combinatorics of puzzles although the stability property was not apparent from their rule.
The equivariant quantum cohomology, K-theory, and elliptic cohomology of various hyperkahler quotients are identified with the quantum integrable systems, such as spin chains or many-body systems. In particular, the cotangent bundle to the Grassmanian is mapped to the Heisenberg magnet. The spectrum of the operators of quantum multiplication is identified with the solutions of the Bethe equations. The generalized Donaldson theory on the product of a two-sphere and a Riemann surface lead to the instanton corrected Bethe equations.
Ising's 1925 thesis on one-dimensional ferromagnetism has since spawned a wealth of mathematics and physics. This includes two geometrical processes known as the random-cluster (or FK) model and the random-current representation. These two systems provide jointly a powerful set of tools. This will be illustrated in this talk, both for the classical Ising model, and the quantum model of Lieb, Schultz, Mattis (1961).
Classical Hamiltonian Systems describe dynamical processes without friction. One can therefore expect that such systems are complicated, since small oscillations never decay. A good measure for the complexity of a dynamical system is its topological entropy. One can therefore expect that the topological entropy of classical Hamiltonian Systems is positive.
I will show how Floer Homology - a modern tool of symplectic geometry and Hamiltonian dynamics - can be used to confirm this expectation.
Nous donnerons un panorama
de résultats récents de rigidité d'actions de réseaux sur le cercle et
proposerons un point de vue unificateur reposant sur la cohomologie
bornée.
Vers la fin des années 50, Yu.
V. Linnik développe sa « méthode ergodique » pour
obtenir (ainsi que son élève Skubenko) plusieurs
résultats remarquables sur la répartition des représentations d'un entier par
une forme quadratique en trois variables (par exemple les représentations d'un
entier comme somme de trois carrés). Cette méthode avait plusieurs décades
d'avance sur son temps et ce n'est qu'à la fin des années 80 que les résultats
originaux de Linnik ont été significativement
améliorés, comme conséquence de progrès fondamentaux en théorie des formes
automorphes et en théorie analytique des nombres. Il faut dire aussi que cette
méthode est restée largement confidentielle, la raison principale étant que le
langage permettant de la formuler simplement n'existait pas à l'époque. Dans cette exposé, nous tâcherons d'exposer cette méthode en
termes aussi élémentaires que possible et d'en présenter plusieurs
applications. Nous montrerons qu'elle s'inscrit naturellement dans le cadre de
la dynamique des espaces homogènes de rang 1 et nous présenterons des résultats
récents, en rang supérieur, inspirés en partie par cette méthode. Il s'agit de
travaux en commun avec M. Einsiedler, E. Lindenstrauss et A. Venkatesh.
This is a joint work with A. Borisov and I. Kozakova. We prove that almost all 1-related groups with at least 3 generators are residually (finite p-) groups and are coherent (that is all finitely generated subgroups are finitely presented). The proof involves methods from geometric group theory, algebraic geometry and the classical theory of Brownian motion.
It is customary in the theory of diophantine approximation to concentrate on the asymptotic behaviour of the accuracy of rational approximations of an irrational number. The study of quasiperiodic dynamics, and in particular of the problem of stability of quasiperiodic orbits, allows in some cases to reinterpret classical diophantine conditions in dynamical terms. Moreover it also sheds new light on the approximation problem since not only the asymptotics but also the history can be relevant.
This talk is based on joint work with Pierre Moussa and Jean-Christophe Yoccoz.
Soit
une variété riemannienne qui est un revêtement
régulier d'une variété compacte. La vitesse de fuite l et l'entropie
de Kaimanovich h sont des invariants
géométriques définis par les propriétés asymptotiques du mouvement brownien sur
. Nous discuterons ces invariants et
leurs relations avec d'autres invariants. En particulier, un résultat général est que l2 ≤ h
This talk is about mesoscopic devices, which are driven slowly and periodically in time, and known as quantum pumps. In a first part I will present a discussion of transport properties, concerning charge, noise, and dissipation. Pump processes, which are "optimal" in this respect, may be characterized geometrically using the Hopf map. These considerations will be based on the scattering approach to pumping due to Buettiker. The system is viewed as consisting of a finite device connected to leads. It allows for scattering states at Fermi energy and is hence gapless. In a second part I will review an alternative approach due to Thouless, in which the system is idealized as being of infinite extent and gapped. The charge transported in a cycle is realized as a Chern number. It will be shown how to relate the seemingly disjoint approaches and in particular how they yield the same transported charge.
What is the maximum number of edges that a graph with n vertices can have if it can be drawn in the plane without 3 pairwise crossing edges? This innocent looking question and its relatives raised by Avital, Erdos, Hanani, Kupitz, Perles, and others can be considered as the starting point of a new discipline: Geometric Graph Theory. We describe some basic problems and results in this area. We also show a couple of recent results, joint with Jacob Fox, illustrating how some far-reaching generalizations of the Lipton-Tarjan separator theorem for planar graphs can be used to answer extremal questions in Geometric Graph Theory.
What is the link between Alternating Sign Matrices (ASM),
Totally Symmetric Self-Complementary Plane Partitions (TSSCPP) and the equivariant cohomology of the
variety of strictly upper triangular matrices with vanishing square?
Two-dimensional Integrable lattice models from statistical physics provide
the natural framework for this missing link. We show in particular how a physical
model, involving densely-packed loop configurations on an infinite surface, is
connected to all three subjects above. This model turns out to be integrable, and we'll use this fact to reformulate all of
the above in terms of polynomial solutions of the quantum Knizhnik-Zamolodchikov
equation. Results include a proof of the Razumov-Stroganov
sum rule, a new connection between ASM and TSSCPP, and the computation of the
(multi)degree of the variety M2=0.
We also present generalizations to the commuting variety and to Mk=0.
The mathematical theory of random graphs was initiated by Paul Erdös and Alfréd Rényi in 1959. Nowadays it is a fast growing and well-developed branch of discrete mathematics at the crossroads with graph theory and probability theory; it has numerous applications in computer science and engineering. In the talk I will discuss properties of discrete groups and high-dimensional spaces associated to random graphs. These are random right angled Artin groups and I will be interested in their Betti numbers, cohomological dimension and topological complexity. The latter is a numerical homotopy invariant reflecting the complexity of motion planning algorithms in robotics. Our main result (joint work with A. Costa) states that the topological complexity of a random right angled Artin group assumes, with probability tending to one, at most three values, when n tends to infinity.
Random graphs arise in numerous areas of mathematics and applications. A natural new framework for their study is provided by the notion of a graphed equivalence relation (originally used in ergodic theory and operator algebras). "Stochastic homogenization" of a certain family of infinite graphs consists in finding a probability measure invariant with respect to an equivalence relation whose classes are endowed with graph structures from this family. The role of such a measure is then similar to the role of an invariant measure for a usual dynamical system or of a stationary measure for a Markov chain.
It is easy to construct stochastically homogeneous random graphs by random perturbations of Cayley graphs of discrete groups. However, there are also graphs whose origin has nothing to do with groups. In this talk we shall discuss the homogenization problem for trees and certain tree-like graphs.
Already the notion of amenability for a group can be given in several equivalent but distinct ways.
- 1) Analytic: it is when we look at the decay, exponential or not, of the uniform norm of a convolution power of an appropriate measure.
- 2) Geometric: in terms of isoperimetric inequalities.
- 3) Algebraic: in terms of the Lie algebra in the case of Lie groups.
These are well-known facts. I propose a similar three-fold - Analytic-Geometric-Algebraic - classification different from the above, but very much in the same spirit, that I call the C and the NC (non-C) groups.
Just to illustrate the issue: an amenable group belongs to the NC class if the decay of the convolution powers if a measure is polynomial. If it is superpolynomial (it cannot be exponential!) we cay that it is a C-group.
This classification is done for Lie groups, where, just as for amenability, Lie algebras are considered first. The situation for discrete groups is not well understood, but it gives rise to a number of interesting problems. The main difficulty, as always for discrete groups, is that the algebra involved is far less tractable than a finite dimensional Lie algebra. The problems involved are truly fascinating, and I will make some propaganda for them.
Quelles propriétés topologiques
caractérisent un groupe de Lie (compact) ? Le fait qu'il admette une multiplication
conduit à définir les H-espaces et si l'on ajoute l'existence d'un inverse à
homotopie près on se restreint aux
espaces de lacets (finis). J'expliquerai quels problèmes on rencontre
lors d'une tentative de classification de ces objets et comment on est amené à
utiliser la complétion des espaces pour chaque nombre premier. A la fin de
l'exposé je dirai quelques mots sur le cas non compact.
Champs
magnétiques et inégalités de Morse pour la d''-cohomologie
Holomorphic Morse inequalities
I will present a survey of recent results about the long time asymptotic behaviour of random processes with long memory due to some rather natural local self-intaraction (self-repellence) of the trajectories. Typical examples are the so-called myopic (or "true") self-avoiding random walk and the self-repelling Brownian polymer models. The long time asymptotics of the displacement is expected to be robust (not depending on some microscopic details), but dimension dependent. It is expected that in 1d the motion is strongly superdiffusive, with time-to-the-two-thirds scaling; in 2d the motion is marginally superdiffusive with logarithmic multiplicative correction in the scaling; in three and more dimensions the displacement is diffusive. For some particular models some of these have been recently proved.
In this talk, I will survey my recent work on the subject of geometry of special holonomy. I will give a unified description of all these geometries in term of normed division algebras. Another closely related description is also explained by using vector cross product. We also applied these approaches to obtain new results in these geometries.
Gauss established a formula for the
number of ways to write a positive integer as a sum of three squares. We
will recall this result, reinterpret it and present a generalization in the
context of Eisenstein series on the group GL(3,Z).
This is a joint work with Omer Offen.
Trace formula is a powerful tool in the study of automorphic representations of reductive algebraic groups defined over number fields and the fields of functions of curves over finite fields. I will outline a conjectural framework of "geometric trace formulas" in the case that the curve is defined over the complex field, which exploits the categorical version of the geometric Langlands correspondence. This is joint work with Robert Langlands and Ngo Bao Chau (arXiv:1003.4578, arXiv:1004.5323).