Probabilistic Numerical Methods
Spécialité Mathématiques de la Modélisation du M2 Mathématiques et Applications de Sorbonne Université.
Course objectives
The aim of this course is to present a number of numerical methods for solving deterministic problems—such as numerical integration, function optimization, and the solution of partial differential equations—using stochastic algorithms, that is, algorithms based on random sampling. The course therefore covers both theoretical topics, including random variables and limit theorems, Markov chains, stochastic processes, and stochastic differential equations, and numerical methods, including Monte Carlo techniques, the Metropolis–Hastings algorithm, and the discretization of stochastic differential equations. Applications to a variety of problems and models arising in physics and statistics are also presented.
Prerequisites
Students are expected to be familiar with the fundamentals of measure and integration theory, as well as the basic concepts of probability theory, including random variables, the law of large numbers, and the central limit theorem.
Course content
The course is divided into two parts, each consisting of four sessions. The first part is common to all students. It begins with methods for simulating random variables, followed by the Monte Carlo method for numerical integration and several associated variance reduction techniques. It then introduces the theory of Markov chains required to present the Metropolis–Hastings algorithm, which plays a central role in computational physics and statistics.
For the second part of the course, students may choose between a computational project track and a theoretical advanced track.
The computational project track consists of implementing the methods introduced in the first part of the course on a real-world application. Assessment is based on an oral project presentation. This option is recommended for students who wish to gain practical experience with Monte Carlo methods while remaining within a relatively elementary theoretical framework.
The theoretical advanced track provides an introduction to the tools of stochastic calculus, including Brownian motion, Itô calculus, stochastic differential equations, and their connection with partial differential equations. These tools form the mathematical foundation of most modern sampling algorithms used in molecular dynamics and statistical machine learning. This option is recommended for students intending to pursue doctoral studies in these areas and is assessed by a final written examination.
The updated lecture notes will be posted on the course's Moodle page every week after the lectures.
Syllabus
Part 1
References