Weighted energy-dissipation principle for gradient flows in metric spaces (Q2317071)

From MaRDI portal
scientific article
Language Label Description Also known as
English
Weighted energy-dissipation principle for gradient flows in metric spaces
scientific article

    Statements

    Weighted energy-dissipation principle for gradient flows in metric spaces (English)
    0 references
    0 references
    0 references
    0 references
    0 references
    8 August 2019
    0 references
    The very interesting paper under review focuses on gradient flows in metric spaces. The authors extend to the metric framework the analysis carried out within a Hilbert space \(H\) in [\textit{A. Mielke} and \textit{U. Stefanelli}, ESAIM, Control Optim. Calc. Var. 17, No. 1, 52--85 (2011; Zbl 1218.35007)], where the gradient flow \[ u'(t)+\partial \phi(u(t))\ni 0\quad \text{in}\ H\qquad \text{for a.a.}\ t\in (0,T) \] was studied by considering the functional of trajectories \(u\in H^1(0,T;H)\to (-\infty,\infty]\) defined by \[ \mathcal{I}_{\varepsilon,T}[u]=\int_0^T \mathrm{e}^{-t/\varepsilon}\left(\dfrac{1}{2}|u'(t)|^2+\dfrac{1}{\varepsilon}\phi(u(t))\right)\mathrm{d}t. \] This functional features both the energy and the (quadratic) dissipation terms, with an exponential weight, and is referred to as WED (Weighted Energy-Dissipation) functional. The authors deal with gradient flows set in a complete metric space \((X,\mathbf{d})\) by considering the functional \[ \mathcal{I}_{\varepsilon}[u]=\int_0^\infty \mathrm{e}^{-t/\varepsilon}\left(\dfrac{1}{2}|u'|^2(t)+\dfrac{1}{\varepsilon}\phi(u(t))\right)\mathrm{d}t. \] It is proved that, for a reasonably wide class of driving energy functionals \(\phi,\) the WED minimizers \(u_\varepsilon\) of \(\mathcal{I}_{\varepsilon}[u]\) among the natural class of absolutely continuous curves \(u:\ [0,\infty)\to X\) satisfying a given initial condition \(u(0) = x,\) converge up to subsequences to curves of maximal slope for \(\phi\) which are characterized by \[ -(\phi\circ u)'(t)= \dfrac{1}{2}|u'|^2(t)+\dfrac{1}{2}|\partial \phi|^2(u(t))=|u'|^2(t)=|\partial \phi|^2(u(t))\quad \text{for a.a.}\ t\in (0,\infty). \] Since the WED approach does not involve directly the distance \(\mathbf{d}\) but it relies on the notion of length and quadratic action of a curve, the metric strategy developed in the paper could be extended to more general cases of length structures where the length of a curve may be strongly affected by the geometry of the sublevels of the functional \(\phi.\)
    0 references
    gradient flow
    0 references
    metric space
    0 references
    curve of maximal slope
    0 references
    weighted energy-dissipation functionals
    0 references
    variational principle
    0 references
    Hamilton-Jacobi equation
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references

    Identifiers

    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references