Reverse approximation of gradient flows as Minimizing Movements: a conjecture by De Giorgi (1711.07256v1)

Abstract: We consider the Cauchy problem for the gradient flow \begin{equation} \label{eq:81} \tag{$\star$} u'(t)=-\nabla\phi(u(t)),\quad t\ge 0;\quad u(0)=u_0, \end{equation} generated by a continuously differentiable function $\phi:\mathbb H \to \mathbb R$ in a Hilbert space $\mathbb H$ and study the reverse approximation of solutions to ($\star$) by the De Giorgi Minimizing Movement approach. We prove that if $\mathbb H$ has finite dimension and $\phi$ is quadratically bounded from below (in particular if $\phi$ is Lipschitz) then for every solution $u$ to ($\star$) (which may have an infinite number of solutions) there exist perturbations $\phi_\tau:\mathbb H \to \mathbb R \ (\tau>0)$ converging to $\phi$ in the Lipschitz norm such that $u$ can be approximated by the Minimizing Movement scheme generated by the recursive minimization of $\Phi(\tau,U,V):=\frac 1{2\tau}|V-U|^2+ \phi_\tau(V)$: \begin{equation} \label{eq:abstract} \tag{$\star\star$} U_\tau^n\in \operatorname{argmin}{V\in \mathbb H} \Phi(\tau,U\tau^{n-1},V)\quad n\in\mathbb N, \quad U_\tau^0:=u_0. \end{equation} We show that the piecewise constant interpolations with time step $\tau > 0$ of all possible selections of solutions $(U_\tau^n)_{n\in\mathbb N}$ to ($\star\star$) will converge to $u$ as $\tau\downarrow 0$. This result solves a question raised by Ennio De Giorgi. We also show that even if $\mathbb H$ has infinite dimension the above approximation holds for the distinguished class of minimal solutions to ($\star$), that generate all the other solutions to ($\star$) by time reparametrization.