Generalized solutions of nonlinear parabolic equations and diffusion processes (Q996725)

The aim of the paper is to construct a weak solution to a nonlinear parabolic equation in \(\mathbb{R}^d\), \[ \frac{\partial f}{\partial t} = L(f) f \quad (0<t\leq T),\qquad f(0,x) = f_0(x). \tag{1} \] Here \(L(g)\) stands for a linear differential operator of the second order, \[ L(g)f(t,x) = a_i(x,g(t,x))\nabla_i f(t,x) + \frac12 \, F_{ij}(x,g(t,x))\nabla_i \nabla_j f(t,x), \] with \(F_{ij} = A_{ik}A_{jk}\). A probabilistic way to prove existence of a classical \(C^2\) solution based on the auxiliary problem, \[ d\xi_s = a(\xi_s,f(t-s,\xi_s))\,ds + A(\xi_s,f(t-s,\xi_s))\,dw_s, \quad \xi_0=x, \quad f(t,x) = E_{0,x}f_0(\xi_t), \tag{2} \] has been developed earlier by the first author jointly with \textit{Yu.~L.~Daletskij} [Izv. Vyssh. Uchebn. Zaved., Mat. 1978, No. 12(199), 6--17 (1978; Zbl 0405.35042)]. Here \(w_s, \, s\geq 0\), is a \(d\)-dimensional Wiener process. In this paper weak solutions of Sobolev class \(H^1\) are constructed, and only one continuous derivative with respect to both variables \(x, f\), is assumed on the coefficients \(a\), \(A\), and, in addition to that, on the ``Stratonovich term'' \(m_j(x) = \frac 12 A_{ik}(x)\nabla_i A_{jk}(x)\). Also it is assumed that the coefficients \(a\) and \(A\) satisfy Lipschitz conditions and linear growth in a special norm. Non-degeneracy of \(F\) is not used. The main result states existence and uniqueness of solutions of the system (2) under \(f_0\in H^1\), and existence and uniqueness in \(H^1\) of the initial equation (1), along with the representation \(f(t,x) = E_{0,x}f_0(\xi_t)\) on a sufficiently small finite time interval \([0,T]\). The point of the construction is successive approximations for the system (2) for \(k = 0, 1, \dots \) \[ f^0(t,x) = f_0(x), \quad d\xi^{k}_s = a(\xi^{k}_s,f^{k}(t-s,\xi^{k}_s))\,ds + A(\xi^{k}_s,f^{k}(t-s,\xi^{k}_s))\,dw_s, \quad \xi^k_0=x, \] \[ f^{k+1}(t,x) := E_{0,x}f^{k}_0(\xi_t). \] In fact, after the introduction the authors reverse time in equation (1), which is more convenient for the presentation. Lipschitz assumptions eventually allow to use the Gronwall lemma on some finite interval of time, both in the uniqueness part and in inequalities leading to convergence of successive approximations. Because the initial data \(x\) shows up in system (3), it is also worthwhile to say that the authors benefit from stochastic flow theory due to Kunita.