Some inverse problems in periodic homogenization of Hamilton-Jacobi equations

Homogenization in context of PDEs; PDEs in media with periodic structure (35B27) Hamilton-Jacobi equations (35F21) Inverse problems for PDEs (35R30) Viscosity solutions to Hamilton-Jacobi equations in optimal control and differential games (49L25)

Abstract: We look at the effective Hamiltonian

associated with the Hamiltonian

H (p, x) = H (p) + V (x)

in the periodic homogenization theory. Our central goal is to understand the relation between

V

and

. We formulate some inverse problems concerning this relation. Such type of inverse problems are in general very challenging. In the paper, we discuss several special cases in both convex and nonconvex settings.

Review text

Let \(\tilde H(x,p)=H(p)+V(x)\) for any \(x,p\in\mathbb R^n\) (\(n\geq 1\)), where \(H\) and \(V\) are continuous functions, \(V\) being \(\mathbb Z^n\)-periodic, and \(\lim_{|p|\to\infty}\tilde H(x,p)=\infty\) uniformly with respect to \(x\). Under the above assumptions, the viscosity solution \(u^{\varepsilon}\) to the Hamilton-Jacobi Cauchy problem \[ \begin{cases} u_t^{\varepsilon}(t,x)+\tilde H(\varepsilon^{-1}x,Du^{\varepsilon}(t,x))=0, &t\in (0,\infty),\;x\in\mathbb R^n,\\ u^{\varepsilon}(0,x)=g(x), &x\in\mathbb R^n, \end{cases}\tag{1} \] where \(g\in BUC(\mathbb R^n)\), converges locally uniformly to the solution to the Cauchy problem \[ \begin{cases} u_t+\overline H(Du)=0, &\text{in }(0,\infty)\times\mathbb R^n,\\ u(0,\cdot)=g, &\text{in }\mathbb R^n, \end{cases} \] where, for any \(p\in {\mathbb R}^n\), \(\overline H(p)\) is the only constant \(c\) such that the equation \(\tilde H(x,p+Dv)=c\) has a unique solution (in the viscosity sense) \((v,c)\in C({\mathbb T}^n)\times\mathbb R\), \({\mathbb T}^n\) being the \(n\)-dimensional torus. In this paper, the authors are interested in the inverse problem of recovering the potential energy \(V\) from the knowledge of \(H\) and the effective Hamiltonian \(\overline H\). To precisely state the main results of the paper, we denote by \(\tilde H_j=H+V_j\) (\(j=1,2\)) two Hamiltonians of problem (1) and by \(\overline H_j\) (\(j=1,2\)) the associated effective Hamiltonians. The first result of the paper shows that, if \(V_2\) identically vanishes, \(H\in C(\mathbb R^n)\) is differentiable at some point \(p_0\), which satisfies the condition \(DH(p_0)\cdot m\neq 0\) for any \(m\in\mathbb Z^n\setminus\{0\}\), \(\overline H_1(p_0)=\overline H_2(p_0)\) and \(\min_{\mathbb R^n}\overline H_1=\min_{\mathbb R^n}\overline H_2\), then \(V_1\equiv 0\). No convexity assumptions on \(H\) are assumed. In the case when neither \(V_1\) nor \(V_2\) is constant, the authors prove a weaker result. They assume \(V_1,V_2\in C^{\infty}(\mathbb T^n)\) and \(H\in C^2(\mathbb R^n)\) superlinear, with \(D^2H\) bounded. Under these conditions they prove that for any vector \(Q\in\mathbb R^n\), such that \(|Q\cdot k|\geq C|k|^{-\alpha}\) for some \(C,\alpha>0\) and any \(k\in\mathbb Z^n\setminus\{0\}\), it holds that \[ \int_{\mathbb T^n}V_i(x)dx=\lim_{\lambda\to\infty}(\overline H_i(\lambda P_{\lambda})-H(\lambda P_{\lambda})),\quad i=1,2, \] where \(P_{\lambda}\in\mathbb R^n\) is such that \(DH(\lambda P_{\lambda})=\lambda Q\). From the previous formula it follows that, if \(\overline H_1\equiv \overline H_2\), then \(V_1\) and \(V_2\) have the same average over \(\mathbb T^n\). This result is improved when \(H(p)=|p|^2/2\) for any \(p\in\mathbb R^n\). In particular, assuming a fast decay condition on the Fourier coefficients of \(V_1\) and \(V_2\), the authors show that if \(\overline H_1\equiv\overline H_2\) then \(V_1\) and \(V_2\) has the same \(L^2\)-norm over \(\mathbb T^n\). In the one-dimensional setting the analysis is easier and the authors prove, in particular, that, if \(V_1\) and \(V_2\) are continuous on the torus with the same distribution and \(H\) is quasi-convex then \(\overline H_1\) and \(\overline H_2\) coincide. The requirement on the distributions of \(V_1\) and \(V_2\) is almost necessary for \(\overline H_1\) and \(\overline H_2\) to coincide. Indeed, if \(H\) is strongly superlinear and \(\overline H_1\equiv \overline H_2\), then \(V_1\) and \(V_2\) have the same distribution, i.e., \[ \int_0^1f(V_1(x))dx=\int_0^1f(V_2(x))dx \] for any \(f\in C(\mathbb R)\). Some of the results proved in the \(n\)-dimensional setting are also extended to the viscous case.

Recommendations

Cites work

Cited in

(14)

This page was built for publication: Some inverse problems in periodic homogenization of Hamilton-Jacobi equations

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2630274)