We introduce a new integrable hierarchy of nonlinear differential-difference equations which we call constrained Toda hierarchy (C-Toda). It can be regarded as a certain subhierarchy of the 2D Toda lattice obtained by imposing a constraint on the Lax operators of the latter. We prove the existence of the tau-function of the C-Toda hierarchy and show that it is the square root of the 2D Toda lattice tau-function. In this and some other respects the C-Toda is a Toda analogue of the CKP hierarchy. It is also shown that zeros of the tau-function of elliptic solutions satisfy the dynamical equations of the Ruijsenaars-Schneider model restricted to turning points in the phase space. The spectral curve has holomorphic involution which interchange the marked points in which the Baker-Akhiezer function has essential singularities.
This is a joint work with I. Krichever
Classical approximation theory establishes power-law bounds on the approximation rates of parametric models that are linear or have a continuous parameter assignment. However, neural networks are neither linear nor necessarily continuous w.r.t. the weight assignment. It turns out that by using very deep and narrow networks one can exceed the classical rates. Moreover, by exploiting special (“superexpressive”) activation functions, a single network with a fixed finite number of neurons can be used to approximate any continuous function with arbitrary accuracy. This result is related to the Kolmogorov(-Arnold) Superposition Theorem. While the functions involved in the KST are highly complex and non-elementary, superexpressive neural networks can be constructed using elementary activations, e.g. {sin, arcsin}. At the same time, Khovansky’s theory of Pfaffian functions implies that commonly used activations (such as ReLU, sigmoid, Heaviside, Gaussian, etc.) are not superexpressive
This is based on joint works with Jan de Gier, Sam Watson and Istvàn Prause. The five vertex model is a special case of the six-vertex model, and generalizes the lozenge tiling model.
We find the free energy of the model, and explicit parameterizations of its limit shapes such as the “boxed-plane partition” limit shape generalizing that for the lozenge tiling case
I will discuss similarities and differences between the Teichmuller spaces and the moduli of bundles and the symplectic structure on the affine bundles of flat and central connections over the moduli spaces of stable vector bundles over a compact Riemann surface of genus g>1
Dynamics of a quantum system can be described by coupled Heisenberg equations. In a generic many-body system these equations involve an exponentially large hierarchy of operators that is intractable without approximations. In contrast, in an integrable system one may hope to find a relatively small subset of relatively simple operators closed with respect to commutation with the Hamiltonian and analytically solve the corresponding system of linear differential equations. We have successfully applied this idea to models where the Hamiltonian is an element of the Onsager algebra (such as the transverse-field Ising spin-(1/2) chain and the superintegrable chiral n-state Potts models), and to the Kitaev model on the honeycomb-like Bethe lattice. For these models, we present analytical results for the quench dynamics inaccessible by other state-of-the art methods. Further prospects of this approach and a number of related open problems will be outlined and discussed.
Based on
O.Lychkovskiy. Closed hierarchy of Heisenberg equations in integrable models with Onsager algebra // SciPost Phys. 10, 124 (2021)
O.Gamayun, O. Lychkovskiy. Out-of-equilibrium dynamics of the Kitaev model on the Bethe lattice via a set of Heisenberg equations // arXiv 2110.13123
Folding transformation of the Painlev\’e equations is an algebraic (of degree greater than 1) transformation between solutions of different equations. In 2005 Tsuda, Okamoto and Sakai classified folding transformations of differential Painleve equations, they are in correspondence with automorphisms of affine Dynkin diagrams.
We give a complete classification of folding transformations of the q-difference Painleve equations, they are in correspondence with certain subdiagrams of the affine Dynkin diagrams (possibly with automorphism). The method is based on Sakai’s approach to Painleve equations through rational surfaces.
Based on joint work with M. Bershtein to appear on coming Tuesday on Arxiv
We consider the triangular lattice version of the two-dimensional ice model with suitable boundary conditions, leading to an integrable 20 Vertex model. Configurations give rise to generalizations of Alternating Sign Matrices, which we call Alternating Phase Matrices (APM). After reviewing a few facts on the square lattice version and the role of integrability, we compute the number of APM of any given size in the form of a determinant, which turns out to match the number of quarter-turn symmetric domino tilings of a quasi-Aztec square with a central cross-shaped hole. We also present results/conjectures for triangular Ice with other types of boundary conditions, and results on the limit shape of large typical configurations, obtained by applying the so-called “Tangent Method” of Colomo and Sportiello.
(based on work with E. Guitter and B. Debin)
Certain often encountered mathematical objects such as a function on a ball/sphere, a set of points in R^n, or a full weighted graph, at first sight, possessing no nontrivial topological information, have nevertheless a nontrivial invariant, called bar-code, that describes the dependence of topological structure on the scale and which consists in each dimension of a set of real numbers partitioned into pairs “birth”-“death”. I’ll describe the construction of this invariant and outline some of its applications, notably in the domain of topological data analysis to the following two subjects. First, how to distinguish natural and generated texts with help of a transformer deep neural network. Second, the topological obstructions in deep neural networks’ learning.
The talk is based on the following works of the speaker, and in collaboration:
1)”Framed Morse complex and its invariants” Advances in Soviet Mathematics,Vol 21, 93-116 (1994)
2)”Canonical forms=Persistence diagrams”(EuroCG’2021)
3)arXiv:2012.15834 (NeurIPS’2020 workshop)
4)arXiv:2109.04825 (EMNLP’2021)
Fock module of quantum toroidal gl_1 appears in different areas of mathematics and mathematical physics. The algebra has SL(2,Z)-symmetry, this enables us to consider the Fock module twisted by an element of SL(2,Z). We give an explicit construction of a slope n’/n twisted Fock module via vertex operators of quantum affine gl_n. The main technical tool is semi-infinite construction via double affine Hecke algebra.
The results are consistent with Gorsky-Negut conjecture (Kononov-Smirnov theorem) on stable envelopes for Hilbert schemes of points in the plane and can be viewed as a manifestation of (gl_1, gl_n)-duality
Smooth Gaussian fields appear in many areas of science (oceanography, astrophysics, engineering, …) but they are also extremely interesting and rich mathematical objects. My initial interest was motivated by conjectures of Berry and Bogomolny-Schmit that connected Laplace eigenfunctions, random plane waves and critical percolation. My main interest is in the geometric properties of level lines of these fields and their connection to percolation theory.
In this talk, I will explain what kind of fields I am interested in, give an overview of some recent results, explain what kind of ideas and methods are currently used in this area and present some open problems and conjectures
arXiv
| Spring 2022 | Fall 2021 | Spring 2021 | Fall 2020 | Spring 2020 | Fall 2019 | Spring 2019 | Fall 2018 | Spring 2018 | Fall 2017 | Spring 2017 | Fall 2016 |