Large deviations, free energy functional and quasi-potential for a mean field model of interacting diffusions (Q3833402)

From MaRDI portal

!

This is the item page for this Wikibase entity, intended for internal use and editing purposes. Please use the normal view instead:

scientific article; zbMATH DE number 4109813
Language Label Description Also known as
default for all languages
No label defined
    English
    Large deviations, free energy functional and quasi-potential for a mean field model of interacting diffusions
    scientific article; zbMATH DE number 4109813

      Statements

      Large deviations, free energy functional and quasi-potential for a mean field model of interacting diffusions (English)
      0 references
      0 references
      0 references
      1989
      0 references
      mean field model of N interacting diffusions
      0 references
      invariant distribution of the empirical process
      0 references
      Gibbs free energy
      0 references
      large deviation principle
      0 references
      domain of attraction
      0 references
      McKean-Vlasov dynamics
      0 references
      minimizing path
      0 references
      The authors investigate the mean field model of N interacting diffusions given by the Itô equations \[ (1)\quad dx_ k=-[\nabla U(x_ k)+(\theta /N)\sum^{N}_{j=1}(x_ k-x_ j)]dt+\sigma dw_ k\quad (k=1,...,N), \] describing the motion of N particles in \({\mathbb{R}}^ d\) under the influence of a potential U, a mean field interaction with strength \(\theta\), and internal white noise fluctuations \(w_ 1,...,w_ N\) (independent Wiener processes) with intensity \(\sigma^ 2\). The main result of the paper establishes a close relationship beetween the ``quasipotential'' as a dynamical notion of the model and the ``free energy functional'' of the invariant distribution of the empirical process. More precisely, let \(\pi_ N\) denote the invariant distribution of the empirical process \(\{X_ N(t)\}\) associated with (1), and let F denote the free energy functional on M, the space of all probability measures on \({\mathbb{R}}^ d\). Set \(f=\min \{F(\nu):\) \(\nu\in M\}\), the Gibbs free energy. It is proved that \(\{\pi_ N\}\) satisfies a large deviation principle, i.e. \[ (2)\quad \pi_ N(A)=\exp \{-N\inf_{\nu \in A}I(\nu)\quad (1+o(1))\} \] as \(N\to \infty\) for suitable subsets \(A\subset M\), where the large deviation functional I is given by \(I=\beta (F-f)\), \(\beta\) being the inverse temperature. On the other hand, large deviations of the empirical measure process from the nonequilibrium dynamics can be characterized via a certain ``action functional'' \(S_{s,t}\) on M, which is associated with the nonlinear McKean-Vlasov equation describing the limiting deterministic dynamics of \(\{X_ N(t)\}\) [cf. the authors, Stochastics 20, 247-308 (1987; Zbl 0613.60021)]. Now, if \(\rho\) is a stationary state of the latter dynamics, define the quasipotential \(Q_{\rho}\) with respect to \(\rho\) by \[ (3)\quad Q_{\rho}(\nu)=\inf \{S_{s,t}(\mu (\cdot)):\quad -\infty <s<t<\infty,\quad \mu (s)=\rho,\quad \mu (t)=\nu \}. \] The aforementioned relation between \(Q_{\rho}\) of (3) and the free energy functional F appearing in (2) is then established as follows: \[ (4)\quad Q_{\rho}(\nu)\geq \beta (F(\nu)-F(\rho))\quad for\quad all\quad \nu,\quad and \] \[ (5)\quad Q_{\rho}(\nu)=\beta (F(\nu)-F(\rho)), \] if \(\nu\) belongs to the domain of attraction of \(\rho\). If in addition, \(F(\nu)<\infty\), the minimum in the definition of \(Q_{\rho}(\nu)\) is attained by the time-reversed trajectory of the McKean-Vlasov dynamics connecting \(\rho\) with \(\nu\). Up to a time shift this minimizing path is unique.
      0 references

      Identifiers

      0 references
      0 references
      0 references
      0 references
      0 references
      0 references