Past Stochastic Analysis Seminar

The notion quantization originates from information theory, where it refers to the approximation of a continuous signal on a discrete set. Our research on quantization is mainly motivated by applications in quadrature problems. In that context, one aims at finding for a given probability measure $\mu$ on a metric space a discrete approximation that is supported on a finite number of points, say $N$, and is close to $\mu$ in a Wasserstein metric.

In general it is a hard problem to find close to optimal quantizations, if  $N$ is large and/or  $\mu$ is given implicitly, e.g. being the marginal distribution of a stochastic differential equation. In this talk we analyse the efficiency of empirical measures in the constructive quantization problem. That means the random approximating measure is the uniform distribution on $N$ independent $\mu$-distributed elements.

We show that this approach is order order optimal in many cases. Further, we give fine asymptotic estimates for the quantization error that involve moments of the density of the absolutely continuous part of $\mu$, so called high resolution formulas. The talk ends with an outlook on possible applications and open problems.

The talk is based on joint work with Michael Scheutzow (TU Berlin) and Reik Schottstedt (U Marburg).

We present some recent results on the metastability of continuous time Markov chains on finite sets using potential theory. This approach is applied to the case of supercritical zero range processes.

I will report joint work with Wolfhard Hansen, Tomasz Jakubowski, Sebastian Sydor and Karol Szczypkowski on perturbations of semigroups and integral kernels, ones which produce comparable semigroups and integral kernels.

Several stochastic simulation algorithms (SSAs) have been recently

proposed for modelling reaction-diffusion processes in cellular and molecular biology. In this talk, two commonly used SSAs will  be studied. The first SSA is an on-lattice model described by the  reaction-diffusion master equation. The second SSA is an off-lattice model based on the simulation of Brownian motion of individual  molecules and their reactive collisions. The connections between SSAs  and the deterministic models (based on reaction- diffusion PDEs) will  be presented. I will consider chemical reactions both at a surface  and in the bulk. I will show how the "microscopic" parameters should  be chosen to achieve the correct "macroscopic" reaction rate. This  choice is found to depend on which SSA is used. I will also present  multiscale algorithms which use models with a different level of  detail in different parts of the computational domain

We treat the first n terms of general orthogonal series evolving with n as the partial sum process, and proved that under Menshov-Rademacher condition, the partial sum process can be enhanced into a geometric 2-rough process. For Fourier series, the condition can be improved, with an equivalent condition on limit function identified.

We consider the forest fire process on Z: on each site, seeds and matches fall at random, according to some independent Poisson processes. When a seed falls on a vacant site, a tree immediately grows. When a match falls on an occupied site, a fire destroys immediately the corresponding occupied connected component. We are interested in the asymptotics of rare fires. We prove that, under space/time re-scaling, the process converges (as matches become rarer and rarer) to a limit forest fire process.
Next, we consider the more general case where seeds and matches fall according to some independent stationary renewal processes (not necessarily Poisson). According to the tail distribution of the law of the delay between two seeds (on a given site), there are 4 possible scaling limits.
We finally introduce some related coagulation-fragmentation equations, of which the stationary distribution can be more or less explicitely computed and of which we study the scaling limit.

We will consider a branching Brownian motion where particles have a drift $-\rho$, binary branch at rate $\beta$ and are killed if they hit the origin. This process is supercritical  if $\beta>\rho^2/2$ and we will discuss the survival probability in the regime as criticality is approached. (Joint work with Elie Aidekon)

We describe a construction of the Brownian measure on Jordan curves with respect to the Weil-Petersson metric. The step from Brownian motion on the diffeomorphism group of the circle to Brownian motion on Jordan curves in the complex plane requires probabilistic arguments well beyond the classical theory of conformal welding, due to the lacking quasi-symmetry of canonical Brownian motion on Diff(S1). A new key step in our construction is the systematic use of a Kählerian diffusion on the space of Jordan curves for which the welding functional gives rise to conformal martingales.

To identify the Martin boundary for a transient Markov chain with Green's function G(x,y), one has to identify all possible limits Lim G(x,y_n)/G(0,y_n) with y_n "tending to infinity". For homogeneous random walks, these limits are usually obtained from the exact asymptotics of Green's function G(x,y_n). For non-homogeneous random walks, the exact asymptotics af Green's function is an extremely difficult problem. We discuss several examples where Martin boundary can beidentified by using large deviation technique. The minimal Martin boundary is in general not homeomorphic to the "radial"  compactification obtained by Ney and Spitzer for homogeneous random walks in Z^d : convergence of a sequence of points y_n toa point on the Martin boundary does not imply convergence of the sequence y_n/|y_n| on the unit sphere. Such a phenomenon is a consequence of non-linear optimal large deviation trajectories.

 In the recent series of papers Kleban, Simmons, and Ziff gave a non-rigorous computation  (base on Conformal Field Theory) of probabilities of several connectivity events for critical percolation. In particular they showed that the probability that there is a percolation cluster connecting two points on the boundary and a point inside the domain can be factorized in therms of pairwise connection probabilities. We are going to use SLE techniques to rigorously compute probabilities of several connectivity events and prove the factorization formula.

Planar maps are graphs embedded in the plane, considered up to continuous deformation. They have been studied extensively in combinatorics, and they have also significant geometrical applications. Particular cases of planar maps are p-angulations, where each face (meaning each component of the complement of edges) has exactly p adjacent edges. Random planar maps have been used in theoretical physics, where they serve as models of random geometry.Our goal is to discuss the convergence in distribution of rescaled random planar maps viewed as random metric spaces.More precisely, we consider a random planar map M(n) which is uniformly distributed over the set of all p-angulations with n vertices. We equip the set of vertices of M(n) with the graph distance rescaled by the factor n to the power -1/4. Both in the case p=3 and when p>3 is even, we prove that the resulting random metric spaces converge as n tends to infinity to a universal object called the Brownian map. This convergence holds in the sense of the Gromov-Hausdorff distance between compact metric spaces. In the particular case of triangulations (p=3), this solves an open problem stated by Oded Schramm in his 2006 ICM paper. As a key tool, we use bijections between planar maps and various classes of labeled trees

We consider nonlinear stochastic wave equations in dimension d\le 3.

Using Malliavin Calculus, we give upper bounds for the small eigenvalues of the inverse of two point densities.These provide a rate of degeneracy when points go close to each other.  Then, we analyze the consequences of this result on lower estimates for hitting probabilities. 

We define a notion of discrete Ricci curvature for a metric measure space by looking at whether "small balls are closer than their centers are". In a Riemannian manifolds this gives back usual Ricci curvature up to scaling. This definition is very easy to apply in a series of examples such as graphs (eg the discrete cube has positive curvature). We are able to generalize several Riemannian theorems in positive curvature, such as concentration of measure and the log-Sobolev inequality. This definition also allows to prove new theorems both in the Riemannian and discrete case: for example improved bounds on spectral gap of the Laplace-Beltrami operator, and fast convergence results for some Markov Chain Monte Carlo methods

We establish a large deviations principle for the block sizes of a uniformly random non-crossing partition. As an application we obtain a variational formula for the maximum of the support of a compactly supported probability measure in terms of its free cumulants, provided these are all non-negative. This is useful in free probability theory, where sometimes the R-transform is known but cannot be inverted explicitly to yield the density.

The vacant set is the set of vertices not visited by a random walk on a graph G before a given time T. In the talk, I will discuss properties of this random subset of the graph, the phase transition conjectured in its connectivity properties (in the `thermodynamic limit'

when the graph grows), and the relation of the problem to the random interlacement percolation.  I will then concentrate on the case when G is a large-girth expander or a random regular graph, where the conjectured phase transition (and much more) can be proved.

Duality methods can be very powerful tools for the analysis of stochastic

processes. However, there seems to be no general theory available

yet. In this talk, I will discuss and aim to clarify various notions

of duality, give some recent rather striking examples (applied to

stochastic PDEs, interacting particle systems and combinatorial stochastic

processes)

and try to give some systematic insight into the type of questions

that can in principle be tackled. Finally, I will try to provide you

with some intuition for this fascinating technique.

The talk will explain how a geometric principle gave rise to a new variational description of information-theoretic entropy and how this led to the solution of a problem dating back to the 50's: whether the the central limit theorem is driven by an analogue of the second law of thermodynamics.

This problem is classically addressed by the so-called Skorohod Embedding problem. We instead develop a stochastic control approach. Unlike the previous literature, our formulation seeks the optimal no arbitrage bounds given the knowledge of the distribution at some (or various) point in time. This problem is converted into a classical stochastic control problem by means of convex duality. We obtain a general characterization, and provide explicit optimal bounds in some examples beyond the known classical ones. In particular, we solve completely the case of finitely many given marginals.

In 1986 Kardar, Parisi, and Zhang proposed a stochastic PDE for the motion of driven interfaces,
in particular for growth processes with local updating rules. The solution to the 1D KPZ equation
can be approximated through the weakly asymmetric simple exclusion process. Based on work of 
Tracy and Widom on the PASEP, we obtain an exact formula for the one-point generating function of the KPZ
equation in case of sharp wedge initial data. Our result is valid for all times, but of particular interest is
the long time behavior, related to random matrices, and the finite time corrections. This is joint work with 
Tomohiro Sasamoto.

: Backward error analysis is a technique that has been extremely successful in understanding the behaviour of numerical methods for ordinary differential equations.  It is possible to fit an ODE (the so called modified equation) to a numerical method to very high accuracy. Backward error analysis has been of particular importance in the numerical study of Hamiltonian problems, since it allows to approximate symplectic numerical methods by a perturbed Hamiltonian system, giving an approximate statistical mechanics for symplectic methods. 

Such a systematic theory in the case of numerical methods for stochastic differential equations (SDEs) is currently lacking. In this talk we will describe a general framework for deriving modified equations for SDEs with respect to weak convergence. We will start by quickly recapping of how to derive modified equations in the case of ODEs and describe how these ideas can be generalized in the case of SDEs. Results will be presented for first order methods such as the Euler-Maruyama and the Milstein method. In the case of linear SDEs, using the Gaussianity of the underlying solutions, we will derive a SDE that the numerical method solves exactly in the weak sense. Applications of modified equations in the numerical study of Langevin equations and in the calculation of effective diffusivities will also be discussed, as well as the use of modified equations  as a tool for constructing higher order methods for stiff stochastic differential equations.

This is joint work with A. Abdulle (EPFL). D. Cohen (Basel), G. Vilmart (EPFL).

Recently, Liang, Lyons and Qian developed a new methodology for the study of backward stochastic differential equations (BSDEs) on general filtered probability spaces. Their approach is based on the analysis of a particular class of functional differential equations, where the driver of the equation does not depend only on the present, but also on the terminal value of the solution.

The purpose of this work is to study fully coupled systems of forward functional differential equations, which are related to a broad class of fully coupled forward-backward stochastic dynamics with respect to general filtrations. In particular, these systems of functional differential equations have a more homogeneous structure with respect to the underlying forward-backward problems, allowing to partly avoid the conflicting nature between the forward and backward components.

Another advantage of the approach is that its generality allows to consider many other types of forward-backward equations not treated in the classical literature: this is shown with the help of several examples, which have interesting applications to mathematical finance and are related to parabolic integro-partial differential equations. In the second part of the talk, we introduce a numerical scheme for the approximation of decoupled systems, based on a time discretization combined with a local iteration approach.

I will introduce the notion of a nonlinear Levy process, discuss basic well-posednes, SDE links and the connection with interacting particles. The talk is aimed to be an introduction to the topic of my recent CUP monograph 'Nonllinear Markov processes and kinetic equations'.

The paper analyses structural models for the evaluation of risky debt following H.E. LELAND [2], with an approach of optimal stopping problem (for instance cf. N. EL KAROUI [1]) and within a more general context: a dividend is paid to equity holders, moreover a different tax schedule is introduced, depending on the firm current value. Actually, an endogenous default boundary is introduced and a nonlinear convex tax schedule allowing for a possible switching in tax benefits. The aim is to find optimal capital structure such that the failure is delayed, meaning how to decrease the failure level VB, anyway preserving D debtholders and E equity holders’interests: for the firm VB is needed as low as possible, for the equity holder, an optimal equity is requested, finally an optimal coupon C is asked  for the total value.

Keywords: corporate debt, optimal capital structure, default,

Numerical Approximations of Non-linear Stochastic Systems. Abstract:  The explicit solution of stochastic differential equations (SDEs can be found only in a few cases. Therefore, there is a need fo accurate numerical approximations that could, for example, enabl  Monte Carlo Simulations. Convergence and stability of these methods are well understood for SDEs with Lipschit  continuous coefficients. Our research focuses on those situations wher  the coefficients of the underlying SDEs are non-Lipschitzian  It was demonstrated in the literature,  that in this case using the classical methods we may fail t  obtain numerically computed paths that are accurate for small step-sizes, or to obtain qualitative information about the behaviour of numerical methods over long time intervals. Our work addresses both of these issues, giving a customized analysis of the most widely used numerical methods.

We consider parabolic scalar conservation laws perturbed by a (conservative) noise. Large deviations are investigated in the singular limit of jointly vanishing viscosity and noise. The model is supposed to feature the same behavior of "asymmetric" particles systems (e.g. TASEP) under Euler scaling.

A first large deviations principle is obtained in a space of Young measures. A "second order" large deviations principle is then discussed, including connections with the Jensen and Varadhan functional. As time allows, more recent "long correlation" models will be treated.

Azema martingales arise naturally in the study of the chaotic representation property; they also provide classical interpretations of quantum stochastic calculus. The talk will not insist on these aspects, but only define these processes and address the problem of their classification. This raises algebraic questions concerning tensors. Everyone knows that matrices can be diagonalized in some common orthonormal basis if and only if they are symmetric and commute with each other; we shall see an analogous statement for tensors with more
than two indices. This, and other theorems in the same vein, make it possible to associate to any multidimensional Azema martingale an orthogonal decomposition of the state space into one- and two-dimensional subspaces; the behaviour of the process becomes simpler when split into its components in these sub-spaces.

"Rough paths of inhomogeneous degree of smoothness (Pi-rough paths) can be treated as p-rough paths (of homogeneous degree of
smoothness) for a sufficiently large p. The theory of integration with respect to p-rough paths can be applied to prove existence and uniqueness of solutions of differential equations driven by Pi-rough paths. However the required conditions on the one-form determining the differential equation are too strong and can be weakened. The talk proves the existence and uniqueness under weaker conditions and explores some applications of Pi-rough paths

The behaviour of the tail of the distribution of the first passage time over a fixed level has been known for many years, but until recently little was known about the behaviour of the probability mass function or density function. In this talk we describe recent results of Vatutin and Wachtel, Doney, and Doney and Rivero which give such information whenever the random walk or Levy process is asymptotically stable.

Let $X$ be a Poisson point process and $K$ a d-dimensional convex set.
For a point $x \in X$ denote by $v_X(x)$ the Voronoi cell with respect to $X$, and set $$ v_X (K) := \bigcup_{x \in X \cap K } v_X(x) $$ which is the union of all Voronoi cells with center in $K$. We call $v_X(K)$ the Poisson-Voronoi approximation of $K$.
For $K$ a compact convex set the volume difference $V_d(v_X(K))-V_d(K) $ and the symmetric difference $V_d(v_X(K) \triangle K)$ are investigated.
Estimates for the variance and limit theorems are obtained using the chaotic decomposition of these functions in multiple Wiener-Ito integrals

We consider a process $X_t\in\R^d$, $t\ge0$, introduced by Durrett and Rogers in 1992 in order to model the shape of a growing polymer; it undergoes a drift which depends on its past trajectory, and a Brownian increment. Our work concerns two conjectures by these authors (1992), concerning repulsive interaction functions $f$ in dimension $1$ ($\forall x\in\R$, $xf(x)\ge0$).

We showed the first one with T. Mountford (AIHP, 2008, AIHP Prize 2009), for certain functions $f$ with heavy tails, leading to transience to $+\infty$ or $-\infty$ with probability $1/2$. We partially proved the second one with B. T\'oth and B. Valk\'o (to appear in Ann. Prob. 2011), for rapidly decreasing functions $f$, through a study of the local time environment viewed from the

particule: we explicitly display an associated invariant measure, which enables us to prove under certain initial conditions that $X_t/t\to_{t\to\infty}0$ a.s., that the process is at least diffusive asymptotically and superdiffusive under certain assumptions.

Thanks to a Last Passage Percolation model, 3 colored sources are in competition to fill all the positive quadrant N2. There is coexistence when the 3 souces have infected an infinite number of sites.
A coupling between the percolation model and a particle system -namely, the TASEP- allows us to compute the coexistence probability.

A very general model of evolving graphs was introduced by Cooper and Frieze in 2003, and further analysed by Cooper. At each stage of the process, either a new edge is added
between existing vertices, or a new vertex is added and joined to some number of existing vertices. Each vertex gaining a new neighbour may be chosen either uniformly, or by preferential attachment, i.e., with probability proportional to the current degree.
It is known that the degrees of vertices in any such model follow a ``power law''. Here we study in detail the degree sequence of a graph obtained from such a procedure, looking at the vertices of large degree as well as the numbers of vertices of each fixed degree.
This is joint work with Graham Brightwell.

The talk will explain how a geometric principle gave rise to a new variational description of information-theoretic entropy and how this led to the solution of a problem dating back to the 50's: whether the the central limit theorem is driven by an analogue of the second law of thermodynamics.

Abstract: A random polytope $K_n$ is, by definition, the convex hull of $n$ random independent, uniform points from a convex body $K subset R^d$. The investigation of random polytopes started with Sylvester in 1864. Hundred years later R\'enyi and Sulanke began studying the expectation of various functionals of $K_n$, for instance number of vertices, volume, surface area, etc. Since then many papers have been devoted to deriving precise asymptotic formulae for the expectation of the volume of $K \setminus K_n$, for instance. But with few notable exceptions, very little has been known about the distribution of this functional. In the last couple of years, however, two breakthrough results have been proved: Van Vu has given tail estimates for the random variables in question, and M. Reitzner has obtained a central limit theorem in the case when $K$ is a smooth convex body. In this talk I will explain these new results and some of the subsequent development: upper and lower bounds for the variance, central limit theorems when $K$ is a polytope. Time permitting, I will indicate some connections lattice polytopes.

Abstract: We consider the inverse problem of finding the diffusion coefficient of a linear uniformly elliptic partial differential equation in divergence form, from noisy measurements of the forward solution in the interior. We adopt a Bayesian approach to the problem. We consider the prior measure on the diffusion coefficient to be either a Besov or Gaussian measure. We show that if the functions drawn from the prior are regular enough, the posterior measure is well-defined and Lipschitz continuous with respect to the data in the Hellinger metric. We also quantify the errors incurred by approximating the posterior measure in a finite dimensional space. This is joint work with Stephen Harris and Andrew Stuart.

The signature of the path is an essential object in rough path theory which takes value in tensor algebra and it is anticipated that the expected signature of Brownian motion might characterize the rough path measure of Brownian path itself. In this presentation we study the expected signature of a Brownian path in a Bananch space E stopped at the first exit time of an arbitrary regular domain, although we will focus on the case E=R^{2}. We prove that such expected signature of Brownian motion should satisfy one particular PDE and using the PDE for the expected signature and the boundary condition we can derive each term of expected signature recursively. We expect our method to be generalized to higher dimensional case in R^{d}, where d is an integer and d >= 2.

Abstract: We construct solutions to Burgers type equations perturbed by a multiplicative

space-time white noise in one space dimension. Due to the roughness of the driving noise, solutions are not regular enough to be amenable to classical methods. We use the theory of controlled rough paths to give a meaning to the spatial integrals involved in the definition of a weak solution. Subject to the choice of the correct reference rough path, we prove unique solvability for the equation. We show that our solutions are stable under smooth approximations of the driving noise. A more general class of approximations will also be discussed. This is joint work with Martin Hairer and Jan Maas.

We analyse stability properties of stochastic Lagrangian Navier stokes flows on compact Riemannian manifolds.

Abstract: In this talk, we first introduce the notion of ergodic BSDE which arises naturally in the study of ergodic control. The ergodic BSDE is a class of infinite-horizon BSDEs:
Y_{t}^{x}=Y_{T}^{x}+∫_{t}^{T}[ψ(X^{x}_{σ},Z^{x}_{σ})-λ]dσ-∫_{t}^{T}Z_{σ}^{x}dB_{σ}, P-<K1.1/>, ∀0≤t≤T<∞,
<K1.1 ilk="TEXTOBJECT" > <screen-nom>hbox</screen-nom> <LaTeX>\hbox{a.s.}</LaTeX></K1.1> where X^{x} is a diffusion process. We underline that the unknowns in the above equation is the triple (Y,Z,λ), where Y,Z are adapted processes and λ is a real number. We review the existence and uniqueness result for ergodic BSDE under strict dissipative assumptions.
Then we study ergodic BSDEs under weak dissipative assumptions. On the one hand, we show the existence of solution to the ergodic BSDE by use of coupling estimates for perturbed forward stochastic differential equations. On the other hand, we show the uniqueness of solution to the associated Hamilton-Jacobi-Bellman equation by use of the recurrence for perturbed forward stochastic differential equations.
Finally, applications are given to the optimal ergodic control of stochastic differential equations to illustrate our results. We give also the connections with ergodic PDEs.

Probability measures in infinite dimensional spaces especially that induced by stochastic processes are the main objects of the talk. We discuss the role played by measures on analysis on path spaces, Sobolev inequalities, weak formulations and local versions of such inequalities related to Brownian bridge measures.

We relate the partition function associated with a certain Brownian directed polymer model to a diffusion process which is closely related to a quantum integrable system known as the quantum Toda lattice. This result is based on a `tropical' variant of a combinatorial bijection known as the Robinson-Schensted-Knuth (RSK) correspondence and is completely analogous to the relationship between the length of the longest increasing subsequence in a random permutation and the Plancherel measure on the dual of the symmetric group.

We study a simple heat-bath type dynamic for a simple model of
polymer interacting with an interface. The polymer is a nearest neighbor path in
Z, and the interaction is modelised by energy penalties/bonuses given when the
path touches 0. This dynamic has been studied by D. Wilson for the case without
interaction, then by Caputo et al. for the more general case. When the interface
is repulsive, the dynamic slows down due to the appearance of a bottleneck in the
state space, moreover, the systems exhibits a metastable behavior, and, after time
rescaling, behaves like a two-state Markov chain.