Seminars from until

This is joint work with Marco Franciosi and Rita Pardini.

Godeaux surfaces, with $K^2=1$ and $p_g=q=0$, are the (complex projective) surfaces of general type with the smallest possible invariants. A complete classification, i.e. an understanding of their moduli space, has been an open problem for many decades.

The KSBA (after Kollár, Sheperd-Barron and Alexeev) compactification of the moduli includes so called stable surfaces. Franciosi, Pardini and Rollenske classified all such surfaces in the boundary which are Gorenstein (i.e., not too singular).

We prove that most of these surfaces corresponds to a point in the moduli which is nonsingular of the expected dimension 8. We expect that the methods used (which include classical and recent infinitesimal deformation theory, as well as algebraic stacks and the cotangent complex) can be applied to all cases, and to more general moduli as well.

The talk is aimed at a non specialist mathematical audience, and will focus on the less technical aspects of the paper.

The Gagliardo-Nirenberg-Sobolev (GNS) inequalities have played a major role in applied mathematics and mathematical physics in more than half a century. In this talk I will present several new results concerning the counterparts of the GNS inequalities on bounded domains. Of course concentration of the minimizers of the GNS inequalities is a main tool in the proof of existence of minimizers on bounded domains. Naturally concentration occurs in the interior for Dirichlet boundary conditions and on the boundary for Neumann boundary conditions. In the Neumann case this leaves a set of interesting open problems depending on the characteristic of the boundary. This is in part joint work with Cristobal Vallejos (Penn State U.) and Hanne Van Den Bosch (U. de Chile) and in part with Soledad Benguria (U. Wisconsin, Madison).

Thanks to neural networks (NNs), faster computation, and massive datasets, machine learning (ML) is under increasing pressure to provide automated solutions to even harder real-world tasks beyond human performance with ever faster response times due to potentially huge technological and societal benefits. Unsurprisingly, the NN learning formulations present a fundamental challenge to the back-end learning algorithms despite their scalability, in particular due to the existence of traps in the non-convex optimization landscape, such as saddle points, that can prevent algorithms from obtaining “good” solutions.

In this talk, we describe our recent research that has demonstrated that the non-convex optimization dogma is false by showing that scalable stochastic optimization algorithms can avoid traps and rapidly obtain locally optimal solutions. Coupled with the progress in representation learning, such as over-parameterized neural networks, such local solutions can be globally optimal.

Unfortunately, this talk will also demonstrate that the central min-max optimization problems in ML, such as generative adversarial networks (GANs), robust reinforcement learning (RL), and distributionally robust ML, contain spurious attractors that do not include any stationary points of the original learning formulation. Indeed, we will describe how algorithms are subject to a grander challenge, including unavoidable convergence failures, which could explain the stagnation in their progress despite the impressive earlier demonstrations.

Motivated by the Weak Gravity Conjecture, we uncover an intricate interplay between black holes, BPS particle counting, and Calabi-Yau geometry in five dimensions. In particular, we point out that extremal BPS black holes exist only in certain directions in the charge lattice, and we argue that these directions fill out a cone that is dual to the cone of effective divisors of the Calabi-Yau threefold. The tower and sublattice versions of the Weak Gravity Conjecture require an infinite tower of BPS particles in these directions, and therefore imply purely geometric conjectures requiring the existence of infinite towers of holomorphic curves in every direction within the dual of the cone of effective divisors. We verify these geometric conjectures in a number of examples by computing Gopakumar-Vafa invariants.

The Yamabe equation on a Riemannian manifold $(M,g)$ is relevant to the question of finding a constant scalar curvature metric on $M$ that is conformally equivalent to the given one.

An optimal $\ell$-partition for the Yamabe equation is a cover of $M$ by $\ell$ pairwise disjoint open subsets such that the Yamabe equation with Dirichlet boundary condition has a least energy solution on each one of these sets, and the sum of the energies of these solutions is minimal.

We will present some recent results obtained in collaboration with Angela Pistoia (La Sapienza Università di Roma) and Hugo Tavares (Universidade de Lisboa) that establish the existence and qualitative properties of such partitions.

If time allows, we will also present some results on symmetric optimal partitions obtained in collaboration with Angela Pistoia, and with Alberto Saldaña (UNAM) and Andrzej Szulkin (Stockholm University).

The Neural Tangent Kernel is a new way to understand the gradient descent in deep neural networks, connecting them with kernel methods. In this talk, I'll introduce this formalism and give a number of results on the Neural Tangent Kernel and explain how they give us insight into the dynamics of neural networks during training and into their generalization features.

Based off joint works with Arthur Jacot and Franck Gabriel.

We survey some known results and conjectures about framed and unframed refined Donaldson-Thomas invariants in this context, and compute attractor invariants explicitly for a variety of toric Calabi-Yau threefolds

Critical points having infinite Morse index and co-index are invisible to homotopy theory, since attaching an infinite dimensional cell does not produce any change in the topology of sublevel sets. Therefore, no classical Morse theory can possibly exist for strongly indefinite functionals (i.e. functionals whose all critical points have infinite Morse index and co-index). In this talk, we will briefly explain how to instead construct a Morse complex for certain classes of strongly indefinite functionals on a Hilbert manifold by looking at the intersection between stable and unstable manifolds of critical points whose difference of (suitably defined) relative indices is one. As a concrete example, we will consider the case of the Hamiltonian action functional defined by a smooth time-periodic Hamiltonian $H: S^1 \times T^*Q \to \mathbb R$, where $T^*Q$ is the cotangent bundle of a closed manifold $Q$. As one expects, in this case the resulting Morse homology is isomorphic to the Floer homology of $T^*Q$, however the Morse complex approach has several advantages over Floer homology which will be discussed if time permits. This is joint work with Alberto Abbondandolo and Maciej Starostka.

Linear zero-rest-mass fields generically develop logarithmic singularities at the critical sets where spatial infinity meets null infinity. Friedrich's representation of spatial infinity is ideally suited to study this phenomenon. These logarithmic singularities are an obstruction to the smoothness of the zero-rest-mass field at null infinity and, in particular, to peeling. In the case of the spin-2 field it has been shown that these logarithmic singularities can be precluded if the initial data for the field satisfies a certain regularity condition involving the vanishing, at spatial infinity, of a certain spinor (the linearised Cotton spinor) and its totally symmetrised derivatives. In this article we investigate the relation between this regularity condition and the staticity of the spin-2 field. It is shown that while any static spin-2 field satisfies the regularity condition, not every solution satisfying the regularity condition is static.

Together with Newtonian mechanics, Maxwell electromagnetism, Einstein relativity and quantum mechanics, Boltzmann-Gibbs (BG) statistical mechanics constitutes one of the pillars of contemporary theoretical physics, with uncountable applications in science and technology. This theory applies formidably well to a plethora of physical systems. Still, it fails in the realm of complex systems, characterized by generically strong space-time entanglement of their elements. On the basis of a nonadditive entropy (defined by an index $q$, which recovers, for $q=1$, the celebrated Boltzmann-Gibbs-von Neumann-Shannon entropy), it is possible to generalize the BG theory. We will briefly review the foundations and applications in natural, artificial and social systems.

A Bibliography is available at http://tsallis.cat.cbpf.br/biblio.htm

We study the Borel summation of the Gromov-Witten potential for the resolved conifold. The Stokes phenomena associated to this Borel summation are shown to encode the Donaldson-Thomas invariants of the resolved conifold, having a direct relation to the Riemann-Hilbert problem formulated by T. Bridgeland. There exist distinguished integration contours for which the Borel summation reproduces previous proposals for the non-perturbative topological string partition functions of the resolved conifold. These partition functions are shown to have another asymptotic expansion at strong topological string coupling.

The infamous sign problem leads to an exponential complexity in Monte Carlo simulations of generic many-body quantum systems. Nevertheless, many phases of matter are known to admit a sign-problem-free representative, allowing efficient simulations on classical computers. Motivated by long standing open problems in many-body physics, as well as fundamental questions in quantum complexity, the possibility of intrinsic sign problems, where a phase of matter admits no sign-problem-free representative, was recently raised but remains largely unexplored. I will describe results establishing the existence, and the geometric origin, of intrinsic sign problems in a broad class of topological phases in 2+1 dimensions. Within this class, these results exclude the possibility of 'stoquastic' Hamiltonians for bosons, and of sign-problem-free determinantal Monte Carlo algorithms for fermions. The talk is based on Phys. Rev. Research 2, 043032 and 033515.