McKean-Vlasov SDEs under measure dependent Lyapunov conditions (Q2041835)

An equation \[ x_t=x_0+\int_0^tb(s,x_s,\mathscr L(x_s))\,ds+\int_0^t\sigma(s,x_s,\mathscr L(x_s))\,dw_s \] is considered on an interval \(I\) that is either \([0,T]\) or \([0,\infty)\). Here, \(w\) is a \(d^\prime\)-dimensional Wiener process, the solution \(x\) is assumed to take values in an open domain \(D\subseteq\mathbb R^d\), \(\mathscr L(x_s)\) denotes the law of \(x_s\) and \(b:I\times D\times\mathcal P(D)\to\mathbb R^d\), \(\sigma:I\times D\times\mathcal P(D)\to\mathbb R^d\otimes\mathbb R^{d^\prime}\) are measurable mappings, where \(\mathcal P(D)\) is the space of Borel probability measures on \(D\) equipped with the topology induced by the weak convergence of probability measures. The authors define three types of Lyapunov functions \(v:I\times D\times\mathcal P(D)\to\mathbb R\), say (a), (b) and (c), and two types of global Lyapunov functions \(v:\mathbb R^d\to\mathbb R\), say (d) and (e). For any such \(v\), they further define \(v\)-initial distributions (a subclass of initial distribution of \(x_0\)), \(v\)-continuity, a local \(v\)-boundedness and an integrated \(v\)-growth. Let a Lyapunov function \(v\) be fixed. It is proved that if the initial distribution \(\mu\) is a \(v\)-initial distribution, \(b\) and \(\sigma\) are \(v\)-continuous and locally \(v\)-bounded and either \(v\) is type (a), or \(v\) is type (b) or (c) and \(b\) and \(\sigma\) satisfy the integrated \(v\)-growth property then there exists a weak solution (in the probabilistic sense) of the above equation with continuous paths in \(D\), with the initial distribution \(\mu\) and such that \[ \sup_{t\in I}\mathbb E\,v(t,x_t,\mathscr L(x_t))<\infty. \] Moreover, if \(I=[0,\infty)\), \(b\) and \(\sigma\) are time-homogeneous, the Markov semigroup \((\mathscr P_t)\) associated with the above equation is Feller and \(v\) can be minorized in the pointwise or in an integral sense by a suitable function \(V:I\times D\to[0,\infty)\) such that \(V(t,x)\ge|x|^2\) near the boundary of \(D\) in the one-point compactification of \(\mathbb R^d\) then there exists an invariant probability measure for \((\mathscr P_t)\). Finally, if a global Lyapunov function \(\bar v\) exists and satisfies either (d), or it satisfies (e) and \(v\) is of type (a), (b) or (c) and \(b\) and \(\sigma\) are locally \(v\)-bounded then \[\mathbb E\,\bar v(x^1_t-x^2_t)\le C(t)\mathbb E\,\bar v(x^1_0-x^2_0),\qquad t\in I \] holds for every two solutions \(x^1\) and \(x^2\) and for a deterministic function \(C\) that depends only on the global Lyapunov function \(\bar v\). In particular, pathwise uniqueness holds provided that \([\bar v=0]=\{0\}\).