scientific article; zbMATH DE number 1025926

From MaRDI portal
Publication:4342633





The author considers a geometric method for constructing solutions of the Hamilton Jacobi equation \[ {{\partial u}\over{\partial t}}(t,x)=H(t,x,u,D_xu),\quad u(t_0,x)=u_0(x),\leqno{(HJ)} \] where \(t_0,t\in\mathbb{R}\), \(x\in N\), a compact manifold. In the first place, the classical ``method of characteristics for solving \((HJ)\) is recalled, to serve as natural introduction to ``geometric solutions, that is Lagrangian submanifolds of \(T^*(\mathbb{R}\times N)\). Next, it is shown that, upon knowing a generating \(C^\infty\)-function \(S\colon(\mathbb{R}\times N)\times\mathbb{R}^k\to\mathbb{R}\), with \(S(t,x,\cdot)\) a nondegenerate quadratic form \(Q(\cdot)\), which describes a geometric solution \(L\subset T^*(\mathbb{R}\times N)\) to \((HJ)\) (i.e. \(L=\{\bigl(t,x,\partial_t S(t,x,\xi),\partial_x S(t,x,\xi)\bigr);\;\partial_\xi S(t,x,\xi)=0\})\), one can get a solution \(u\) to \((HJ)\) given by a ``minmax of \(S\). Namely, with \(\mathbb{R}^k=E=E_+\oplus E_-\) (\(E_\pm\) the positive/negative space associated with \(Q\)), \(u\) is given by \[ u(t,x)=\inf\{a\in\mathbb{R};\;\alpha\not=0\text{ in }H^*(E^a,E_-^b)\}, \] where \(\alpha(\xi)=\rho(\xi_-)dv_-\), with \(\xi_-\) the \(E_-\)-projection of \(\xi, dv_-\) the volume form on \(E_-\), \(\rho\) a smooth function with non-zero integral on \(E_-\), and \(b\in\mathbb{R}\) is such that the cohomology class of \(\alpha\) in \(H^*(E,E_-^b)\) is non-zero. Finally, the properties of such a solution \(u\) are gathered in the main theorem: denoting by \(J_{t_0,t}(H,u_0)\) the operator which associates the problem \((HJ)\) with its ``variational solution \(u(t,x)\), one has the following Theorem. (1) \(J\) extends to a continuous map \(C^{Lip}(T^*(\mathbb{R}\times N))\times C^0(N)\to C^0(\mathbb{R}\times N)\). (2) For small \(t\), \(J\) coincides with the solution given by the method of characteristics. (3) If \(p\mapsto H(\cdot,\cdot,\cdot,p)\) is convex, \(J_{t_0,t}(H,u_0)\) coincides with the viscosity solution. (4) In the general case, \(J_{t_0,t}(u_0)\) does not coincide with the viscosity solution. In particular, one generally has \(J_{t,s}\circ J_{0,t}\not=J_{0,t+s}\).




Cited in
(27)








This page was built for publication:

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4342633)