Georgia Institute of Technology College of Sciences

I read Benoist's paper Convexes Divisibles IV (2006, Invent. Math.), and will talk about it. The main result is a striking structural theorem for triangles in the boundaries of 3-dimensional properly convex divisible domains O that are not strictly convex (which exist). These bound "flats" in O. Benoist shows that every Z^2 subgroup of the group G preserving O preserves a unique such triangle. Conversely, all such triangles are disjoint and any such triangle descends to either a torus or Klein bottle in the quotient M = O/G (and so must have many symmetries!). Furthermore, this "geometrizes" the JSJ decomposition of M, in the sense that cutting along these tori and Klein bottles gives an atoroidal decomposition of M.

We develop a method to construct solutions of some (retarded or advanced) equations. A prime example could be the motion of point charges interacting via the fully relativistic Lienard-Wiechert potentials (as suggested by J.A. Wheeler and R.P. Feynman in the 1940's). These are retarded equations, but the delay depends implicitly on the trajectory. We assume that the equations considered are formally close to an ODE and that the ODE admits hyperbolic solutions (that is, perturbations transversal to trajectory grow exponentially either in the past or in the future) and we show that there are solutions of the functional equation close to the hyperbolic solutions of the ODE. The method of proof does not require to formulate the delayed problem as an evolution for a class of initial data. The main result is formulated in an "a-posteriori" format and allows to show that solutions obtained by non-rigorous approximations are close (in some precise sense) to true solutions. In the electrodynamics (or gravitational) case, this allows to compare the hyperbolic solutions of several post-newtonian approximations or numerical approximations with the solutions of the Lienard-Weichert interaction. This is a joint work with R. de la Llave and J. Yang.

A knot K in S^3 is (smoothly) slice if K is the boundary of a properly embedded disk D in B^4, and K is ribbon if this disk can be realized without any local maxima with respect to the radial Morse function on B^4. In dimension three, a knot K with nice topology – that is, a fibered knot – bounds a unique fiber surface up to isotopy. Thus, it is natural to wonder whether this sort of simplicity could extend to the set of ribbon disks for K, arguably the simplest class of surfaces bounded by a knot in B^4. Surprisingly, we demonstrate that the square knot, one of the two non-trivial ribbon knots with the lowest crossing number, bounds infinitely many distinct ribbon disks up to isotopy. This is joint work with Jeffrey Meier.

Recent advancements in machine learning have paved the way for groundbreaking opportunities in the realm of molecular discovery. At the forefront of this evolution are improved computational tools with proper inductive biases and efficient optimization. In this talk, I will delve into our efforts around these themes from a geometry, sampling and optimization perspective. I will first introduce how to encode symmetries in the design of neural networks and the balance of expressiveness and computational efficiency. Next, I will discuss how generative models enable a wide range of design and optimization tasks in molecular discovery. In the third part, I will talk about how the advancements in stochastic optimal control, sampling and optimal transport can be applied to find transition states in chemical reactions.