Mathematical Sciences Research Institute

Home » Workshop » Schedules » Level set volume preserving diffusions

Level set volume preserving diffusions

Fluid Mechanics, Hamiltonian Dynamics, and Numerical Aspects of Optimal Transportation October 14, 2013 - October 18, 2013

October 14, 2013 (03:30 PM PDT - 04:30 PM PDT)
Speaker(s): Yann Brenier (École Polytechnique)
Location: MSRI: Simons Auditorium
Primary Mathematics Subject Classification No Primary AMS MSC
Secondary Mathematics Subject Classification No Secondary AMS MSC


Abstract We discuss diffusion equations that are constrained to preserve the volume of each level set during the time evolution (which excludes the standard heat equation). We consider, in particular, the gradient flow of the Dirichlet integral under suitable volume-preserving transportation metrics. The resulting equations are non-linear and very degenerate, admitting as stationary solutions all scalar functions which are functions of their own Laplacian. (In particular, in 2D, all stationary solutions of the Euler equations for incompressible fluids.) We relate them to both combinatorial optimization and linear algebra, through the quadratic assignmemt problem (a NP combinatorial optimization problem including the travelling salesman problem) and the Brockett-Wegner diagonalizing flow for linear operators. For these equations, we provide a concept of "dissipative solutions" that exist globally in time and are unique as long as they stay smooth, following some works of P.-L. Lions (for the Euler equations) and Ambrosio-Gigli-Savare (for the heat equation in metric spaces).
18890?type=thumb Brenier 994 KB application/pdf Download
Video/Audio Files


H.264 Video v1180.m4v 308 MB video/mp4 rtsp://videos.msri.org/data/000/018/433/original/v1180.m4v Download
Quicktime v1180.mov 431 MB video/quicktime rtsp://videos.msri.org/data/000/018/434/original/v1180.mov Download
Troubles with video?

Please report video problems to itsupport@msri.org.

See more of our Streaming videos on our main VMath Videos page.