On the strong convergence of the trajectories of a Tikhonov regularized second order dynamical system with asymptotically vanishing damping (Q6042664)
From MaRDI portal
scientific article; zbMATH DE number 7681528
Language | Label | Description | Also known as |
---|---|---|---|
English | On the strong convergence of the trajectories of a Tikhonov regularized second order dynamical system with asymptotically vanishing damping |
scientific article; zbMATH DE number 7681528 |
Statements
On the strong convergence of the trajectories of a Tikhonov regularized second order dynamical system with asymptotically vanishing damping (English)
0 references
3 May 2023
0 references
Let \(\mathcal H\) be a Hilbert space, \(g:\mathcal{H}\to \mathbb{R}\) a convex continuously Fréchet differentiable function where the solution set \(\mathrm{argmin}\, g\) of all points \(x^*\in\mathcal{H}\) where \(g(x)\) attains its minimum is nonempty, with \(\nabla g\) Lipschitz continuous on bounded sets. The authors consider the asymptotics of the second order dynamical system \begin{align*} x''(t) + \frac{\alpha}{ t^q} x'(t) + \nabla g(x(t)) + \frac{a}{t^p} x(t) &= 0 \quad \ (t > 0) \\ x(t_0) = u_0 , \quad x'(t_0) &= v_0 \quad (t_0 > 0) \tag{1} \end{align*} (all constants positive) as \(t\to+\infty\). The subject is the convergence (weak or strong) of trajectories of (1) to an element \( x^*\) of \(\mathrm{argmin}\, g,\) estimates of the form \[ g(x(t)) - g(x^*) = O\Big(\frac{1}{t^\gamma}\Big) \text{ or } g(x(t)) - g(x^*) = o\Big(\frac{1}{t^\gamma} \Big)\tag{2} \] and similar estimates for the derivative \[ \|x'(t)\| = O\Big(\frac{1}{t^\gamma} \Big),\text{ or } \|x'(t)\| = o\Big(\frac{1}{t^\gamma}\Big)\tag{3} \] for \(\gamma > 0\) depending on the constants in (1). Under suitable conditions on the constants the author obtains convergence results (both weak and strong) for \(x(t)\) and estimates of the forms (2) and (3). There are also auxiliary integral estimates, \[ \int_{t_0}^\infty t^q \|x'(t)\|^2 dt < \infty, \qquad \int_{t_0}^\infty t^q \big( g(x(t)) - g(x^*) \big) dt < \infty, \] the first instrumental in the convergence proof. The author compares in detail his conclusions with theorems in the existing literature. The results are computationally checked in several examples.
0 references
convex optimization
0 references
continuous second-order dynamical system
0 references
asymptotics as \(t \to \infty
0 references
\) Tikhonov regularization
0 references
strong convergence
0 references
convergence rate
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