Abstract:We introduce and study the class of totally dissipative multivalued probability vector fields (MPVF) $\boldsymbol{\mathrm F}$ on the Wasserstein space $(\mathcal{P}_2(\mathsf{X}),W_2)$ of Euclidean or Hilbertian probability measures. We show that such class of MPVFs is in one to one correspondence with law-invariant dissipative operators in a Hilbert space $L^2(\Omega,\mathcal{B},\mathbb{P};\mathsf{X})$ of random variables, preserving a natural maximality property. This allows us to import in the Wasserstein framework many of the powerful tools from the theory of maximal dissipative operators in Hilbert spaces, deriving existence, uniqueness, stability, and approximation results for the flow generated by a maximal totally dissipative MPVF and the equivalence of its Eulerian and Lagrangian characterizations. We will show that demicontinuous single-valued probability vector fields satisfying a metric dissipativity condition are in fact totally dissipative. Starting from a sufficiently rich set of discrete measures, we will also show how to recover a unique maximal totally dissipative version of a MPVF, proving that its flow provides a general mean field characterization of the asymptotic limits of the corresponding family of discrete particle systems.Such an approach also reveals new interesting structural properties for gradient flows of displacement convex functionals with a core of discrete measures dense in energy.
What problem does this paper attempt to address?
The main problem that this paper attempts to solve is the study of the evolution semigroup generated by the λ - dissipative multi - valued probability vector fields (MPVF) in the Wasserstein space. Specifically, the authors focus on the following aspects:
1. **Definition and Properties**: A class of completely dissipative multi - valued probability vector fields (MPVF) on the Wasserstein space \((P_2(X), W_2)\) is introduced and studied, where \(X\) is a Euclidean or Hilbert probability measure space. There is a one - to - one correspondence between this type of MPVF and the conservative - dissipative operators in the Hilbert space \(L^2(\Omega, \mathcal{B}, P; X)\), and the natural maximality property is retained.
2. **Tools and Methods**: Through this correspondence, the authors introduce powerful tools from the theory of maximal dissipative operators in Hilbert space into the Wasserstein framework, thereby deriving the existence, uniqueness, stability and approximation results of the flow generated by the maximal completely dissipative MPVF, as well as the equivalence of its Euler and Lagrangian characterizations.
3. **Specific Applications**:
- **Discrete Measures**: It is proved that a single - valued probability vector field satisfying specific metric dissipation conditions is actually completely dissipative. Starting from a rich set of discrete measures, it is shown how to recover a unique maximal completely dissipative MPVF version, whose flow provides a general mean - field characterization of the asymptotic limit of the corresponding discrete particle system.
- **Gradient Flows**: This method also reveals new and interesting structural properties of the gradient flows of displacement convex functionals, whose cores are composed of energy - intensive discrete measures.
4. **Extensions and Generalizations**: The paper further explores the method of constructing the maximal completely dissipative MPVF from the discrete core, that is, deriving the mean - field description from a family of compatible discrete particle systems. This not only guarantees the unique global extension of the local - time evolution but also induces a semigroup, which can be interpreted as the unique mean - field limit driven by the particle system.
5. **Geometrically Convex Functionals**: The behavior of geometrically convex functionals in \(P_2(X)\) is studied, especially when the appropriate domain of these functionals contains an energy - intensive discrete core, they are completely convex.
In summary, this paper aims to expand and deepen the understanding of the theory of probability measure evolution in the Wasserstein space by introducing the concept of completely dissipative MPVF, especially in the applications of mean - field dynamics and the asymptotic behavior of discrete particle systems.