Probability and Mathematical Physics Seminar
This seminar covers a wide range of topics in pure and applied probability and in mathematical physics. Unless otherwise noted, the talks take place on Fridays, 11:10am–12pm, in room WWH 1302. See directions to the Courant Institute.
To be kept informed, you can subscribe to the Probability Seminar Mailing list.
Twice every semester, we will have a joint Columbia–Courant meeting (hosted once at each institution) as the Probability and the City seminar.
Seminar Organizer(s): The Probability group
Past Events

Friday, May 1, 2020CANCELLED
TBA
Nike Sun, MIT Mathematics Department 
Friday, April 17, 2020CANCELLED
ColumbiaCourant probability day at Columbia

Friday, April 10, 2020CANCELLED
TBA
Dima Ioffe, Technion 
Friday, April 3, 2020CANCELLED
TBA
Kyle Luh, Harvard University 
Friday, March 27, 2020CANCELLED
TBA.
Arjun Krishnan, University of Rochester 
Friday, March 20, 202011:10AM, Location TBA
No seminar
Springbreak 
Friday, March 13, 2020CANCELLED
Multitime distribution of TASEP
Zhipeng Liu, University of KansasSynopsis:
The Totally Asymmetric Simple Exclusion Process (TASEP) is the most studied model in the KardarParisiZhang (KPZ) university class. The one point limiting distributions are the wellknown TracyWidom distributions and their analogs, and the spatial processes are the socalled Airy processes. However, along the temporal direction much less is known until very recently. In this talk, we will first review some recent progresses along this direction, then discuss a novel formula of the finite time multipoint distribution formula of TASEP in the spacetime plane. This formula is in terms of multiple contour integrals of a Fredholm determinant, with initial conditions encoded in a symmetric polynomial, which is related to the dual Grothendieck polynomial and the inhomogeneous Schur polynomials. We are also able to find the limits of this multipoint distribution formula for both step and flat initial conditions when the times are different and go to infinity proportionally. These limits are believed to be universal, and hence are expected to be the limiting multitime distributions for all the models in the KPZ universality class (with step or flat initial conditions).

Friday, March 13, 2020CANCELLED
TBA
Dmitry Chelkak, ENS Paris 
Friday, March 6, 202012PM, Warren Weaver Hall 1302
A probabilistic construction of conformal blocks for Liouville CFT
Guillaume Rémy, Columbia UniversitySynopsis:
Liouville theory is a fundamental example of a conformal field theory (CFT) first introduced by A. Polyakov in the context of string theory. In recent years it has been rigorously studied using probabilistic techniques. In this talk we will present a probabilistic construction of the conformal blocks of Liouville CFT on the torus. These are the fundamental objects that allow to understand the integrable structure of CFT using the conformal bootstrap equation. We will also mention the connection with the AGT correspondence. Based on joint work with Promit Ghosal, Xin Sun, and Yi Sun.

Friday, March 6, 202011:10AM, Warren Weaver Hall 1302
Dynamics of Deep Neural Networks and Neural Tangent Hierarchy
HongTzer Yau, Harvard UniversitySynopsis:
The evolution of a deep neural network trained by the gradient descent can be described by its neural tangent kernel (NTK) as introduced by Jacot, Gabriel, Hongler, where it was proven that in the infinite width limit the NTK converges to an explicit limiting kernel and it stays constant during training. The NTK was also implicit in some other recent papers. In the overparametrization regime, a fullytrained deep neural network is indeed equivalent to the kernel regression predictor using the limiting NTK. And the gradient descent achieves zero training loss for a deep overparameterized neural network. However, it was observed in Arora&al that there is a performance gap between the kernel regression using the limiting NTK and the deep neural networks. This performance gap is likely to originate from the change of the NTK along training due to the finite width effect. The change of the NTK along the training is central to describe the generalization features of deep neural networks.
In the current paper, we study the dynamic of the NTK for finite width deep fullyconnected neural networks. We derive an infinite hierarchy of ordinary differential equations, the neural tangent hierarchy (NTH) which captures the gradient descent dynamic of the deep neural network. Moreover, under certain conditions on the neural network width and the data set dimension, we prove that the truncated hierarchy of NTH approximates the dynamic of the NTK up to arbitrary precision. This description makes it possible to directly study the change of the NTK for deep neural networks, and sheds light on the observation that deep neural networks outperform kernel regressions using the corresponding limiting NTK. 
Friday, February 28, 202011:10AM, Warren Weaver Hall 1302
Scaling limits of the two and threedimensional uniform spanning trees and the associated random walks
David Croydon, Kyoto UniversitySynopsis:
I will describe recent work regarding the structure of and random walks on the two and threedimensional uniform spanning trees. The results from the twodimensional case are from joint works with Martin Barlow (UBC) and Takashi Kumagai (Kyoto), and as well as a scaling limit, include demonstrating fluctuations in the ondiagonal part of the quenched heat kernel, and offdiagonal estimates for the averaged heat kernel. As for the threedimensional situation, this is being investigated in an ongoing work with Omer Angel (UBC), Sarai HernandezTorres (UBC) and Daisuke Shiraishi (Kyoto). In the latter work, we demonstrate the tightness of the graph and the random walk's annealed law under rescaling, and convergence along a particular subsequence. We also derive the random walk's walk dimension (with respect to both the intrinsic and Euclidean metric) and its spectral dimension, as well as heat kernel estimates for any diffusion that arises as a scaling limit.

Friday, February 21, 202011:10AM, Warren Weaver Hall 1302
The Landscape of the Planted Clique Problem: Dense subgraphs and the Overlap Gap Property
Ilias Zadik, NYU Center for Data ScienceSynopsis:
In this talk, we will focus on the planted clique model, denoted by G(n,1/2,k), which is originally introduced by Jerrum in 1992. An instance of G(n,1/2,k) is generated by planting a clique of size k uniformly at random in a vanilla ErdősRényi graph G(n,1/2). The inference goal is to recover the vertices of the planted clique from an instance of G(n,1/2,k). The study of the model have been the focus of a large line of work arising from the probability theory, computer science and statistics communities. While we know that as long as k > (2+\epsilon) log_2(n), for some \epsilon > 0, the clique is recoverable via bruteforce methods, no polynomialtime method is proven to succeed unless k > c \sqrt{n} for some c > 0. A convincing explanation for the failure of polynomialtime methods in the regime where k is much smaller than \sqrt{n} is currently missing in the literature of the model.
We are going to discuss some new results on the geometry of the dense subgraphs in an instance of G(n,1/2,k). Our results suggest that k = Θ (\sqrt{n}) admits an explanation as the phase transition point for the presence of a certain Overlap Gap Property (OGP) on the space of dense subgraphs. OGP is a disconnectivity notion which originates in spin glass theory and is known to suggest algorithmic harness when it appears. Finally, we show that OGP implies the failure of a large family of MCMC methods to recover the clique when k is much smaller than \sqrt{n}.
This is joint work with David Gamarnik.

Friday, February 14, 202011:10AM, Warren Weaver Hall 1302
Entanglement entropy in quantum spin chain models
Jani Virtanen, University of ReadingSynopsis:
I discuss entanglement entropy of bipartite systems using the von Neumann entropy as measure of entanglement. Some of the most widely studied systems include onedimensional quantum critical systems, such as quantum spin chains, which in their simplest setting consist of $N$ spins. Of particular interest is the XX spin chain model with zero magnetic field and the study of the von Neumann entropy of the subsystem $P$ of spins on lattice sites $\{1,2,\dots,m\}\cup\{2m+1,2m+2,\dots, 3m\}$, which can be analyzed using certain integral representations. For a single block subsystem, the integral representation involves Toeplitz determinants and the entropy can be calculated using the FisherHartwig asymptotic expansion of these determinants. In this talk, we consider a subsystem that consists of two blocks of spins separated by one spin and compute the mutual information between the two intervals explicitly and rigorously using the RiemannHilbert approach. Joint work with György Gehér (Reading University) and Alexander Its (Indiana UniversityPurdue University Indianapolis).

Friday, February 7, 202011:10AM, Warren Weaver Hall 1302
Stationary solutions for the stochastic Burgers' equation
Lenya Ryzhik, StanfordSynopsis:
We consider solutions to the viscous Burgers' equation with a random forcing in one dimension. We show that when the forcing is a spatial derivative of a smooth spacetime stationary random field, there exist spacetime stationary solutions. We also show that they describe the long time behavior for a reasonable class of initial conditions. These spacetime stationary solutions can, in turn, be used to construct shocklike timestationary solutions that connect such spacestationary states as their limits at infinity. This is a joint work with A. Dunlap and C. Graham. 
Friday, January 31, 202011:10AM, Warren Weaver Hall 1302
Zerofree regions and central limit theorems
Marcus Michelen, University of Illinois at ChicagoSynopsis:
Let X be a random variable taking values in {0,...,n} and f(z) be its probability generating function. Pemantle conjectured that if the variance of X is large and f has no roots close to 1 in the complex plane, then X must be approximately normal. We will discuss a complete resolution of this conjecture in a strong quantitative form, thereby giving the best possible version of a result of Lebowitz, Pittel, Ruelle and Speer. Additionally, if f has no roots with small argument, then X must be approximately normal, again in a sharp quantitative form. These results also imply a multivariate central limit theorem that answers a conjecture and completes a program of Ghosh, Liggett and Pemantle. This talk is based on joint work with Julian Sahasrabudhe.