
ICM 2026 Invited Section Lecture - Benedetto Piccoli
transcript
show notes
Measure Evolution Equations for Multi-Agent Systems
Motivated by applications, several mathematical approaches have been proposed for the dynamics of large groups of intelligent agents. A typical example is vehicular traffic, where drivers make decisions and alter the energy of the system. A recent approach was developed by generalizing ordinary differential equations to measures, replacing vector fields with maps from probability measures on a manifold to probability measures on its tangent bundle. The new equations are called measure differential equations and allow for a multiscale representation of the physical system. A general theory, including existence and uniqueness results, generalization to multifunctions, and application to relaxed controls, is achieved using typical tools of optimal transport, such as the Wasserstein distance and its generalizations. While the existence of (weak) solutions is achieved under general assumptions, uniqueness is based on nonlinear functionals and on the concept of Dirac germ, which defines small-time evolution for finite sums of Dirac deltas. Circling back to applications, a general optimal control problem is formulated for the control of a large group via a small number of agents, also called Lagrangian controls. Mathematically, using the mean-field limit, one defines coupled systems of controlled ordinary differential equations and measure differential equations. The cost functionals include running costs defined in terms of (generalized) Wasserstein distance between the empirical measure of the controlled agents and the measure describing the uncontrolled ones, which elude classical tools. A list of open problems for future investigations is included.
Uploaded 14-09-26
Learn more about your ad choices. Visit megaphone.fm/adchoices





