Aubry-Mather theory and Hamilton-Jacobi equations (Q1400916)

Let \(L(q,\dot q,t)\) be a Lagrangian subject to the periodicity conditions \[ L(q+ 1,\dot q,t)= L(q,\dot q,t)= L(q,\dot q,t+1) \] and to some further regularity and convexity assumptions. In order to study the trajectories of an associated Hamilton-Jacobi flow the author considers functions \(h^\infty_c(x,t,\xi)\) given by: \[ h^\infty_c(x,t,\xi)= \liminf_{n\to\infty} \min\Biggl\{\int^n_t (L(q(s),\dot q(s),s)- c\dot q(s)+ \alpha(c))\Biggr\} ds,\tag{1} \] where \(q(s)\) ranges over functions absolutely continuous on \([t,n]\) with values on the unit circle \(S_1\) and such that \(q(t)= x\), \(q(n)=\xi\). Here \(c\) is an auxiliary parameter while \(\alpha(c)\) is defined in terms of \(c\) and \(L(q,\dot q,t)\) in a somewhat involved way. The family \(\{h^\infty_c\}\) in (1), introduced by J. N. Mather, has the following properties. Let \(H(q,p,t)\) be the Hamiltonian associated with \(L(q,\dot q,t)\) via Legendre transform and consider the Hamilton-Jacobi equation: \[ \partial_t v- H(x,c- \partial_x v,t)+\alpha(c)= 0,\quad v(x+1,t)= v(x,t)= v(x,t+1).\tag{2} \] Then as shown by P. Lions and C. Mantegazza, C. A. Menucci, \(h^\infty_c(x,t,\xi)\) is a viscosity solution of (2). On the other hand, Jansen, Kreiss, Moser considered the equation \[ \textstyle{{1\over 2}}\nu^2 \partial_{xx}u+ \partial_tu- H(x,c-\partial_x u,t)+ \beta(\nu)= 0,\;u(x+ 1,t)= u(x,t)= u(x,t+1),\tag{3} \] which for suitably chosen \(\beta(\nu)\) admits a unique globally defined solution \(u_\nu(x,t)\). They showed that if \(\nu_k\searrow 0\) then the associated sequence \(\{u_{\nu_k}\}\) of solutions to (3) admits a subsequence which converges uniformy toward a viscosity solution \(v^*\) of (2) which however may be different from \(h^\infty_c\). The question arises: can one achieve that for a suitably chosen sequence \(\{u_{\nu_k}\}\) and an appropriate \(\xi\) the above limit \(v^*\) equals \(h^\infty_c(x,t,\xi)\)? This question, considered earlier in the literature, is answered affirmatively by the author, provided that the Lagrangian satisfies a somewhat involved condition (A). His Theorem 1 asserts that if (A) holds and some further assumption is satisfied, then one finds \(\xi_0\), related to these assumptions, such that \[ \lim\partial_x u_\nu(x,t)= \partial_x h^\infty_c(x,t,\xi_0)\quad\text{a.e. as }\nu\searrow 0. \] There are further results, too technical to be stated here. An interesting feature of the proof is that it makes use (among others) of results on stochastic differential equations obtained by W. H. Fleming.