Abstract:In this paper, we present new equivalent conditions for the growth of proper lower semicontinuous functionals at strict local minimizers. The main conditions are a variant of the so-called tilt stability property of local minimizers and an analog of the classic Polyak-Łojasiewicz condition, where the gradient is replaced by linear perturbations. We derive the following tilting principle: stability of minimizers under linear perturbations can infer their stability under nonlinear ones. We show how growth conditions can be used to give convergence rates for the proximal point algorithm. Finally, we give an application to elliptic tracking problems, establishing a novel equivalence between second-order conditions and the sensitivity of solutions with respect to uncertainty in data.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is about the growth conditions of non - convex functions at strictly local minimum points. Specifically, the author aims to establish new equivalent conditions that describe the growth behavior of appropriate lower - semicontinuous functions at strictly local minimum points. Main attention is focused on two aspects:
1. **Tilt Stability**: This is a property similar to that introduced by Rockafellar and Poliquin, which can be interpreted as the metric subregularity of the Moreau - Rockafellar subdifferential in a certain neighborhood.
2. **Polyak - Łojasiewicz - like Conditions**: This is a variant of the classical Polyak - Łojasiewicz condition, in which the gradient is replaced by a linear perturbation.
Through these conditions, the author establishes the following principle: if the behavior of the minimum points under linear perturbations is known, then their behavior under certain nonlinear perturbations can be inferred. This principle is of great significance in the design and analysis of optimization algorithms, especially for the convergence rate analysis of the proximal point algorithm.
In addition, the paper also provides an application in the elliptic tracking problem, proving the equivalence between the second - order conditions and the Lipschitz stability of the solution with respect to data uncertainty. This result provides key insights for ensuring the robustness of the optimal solution under data uncertainty.
### Main Contributions
- **New Equivalent Conditions**: Established new equivalent conditions for the growth conditions of non - convex functions at strictly local minimum points.
- **Tilt Stability and Subregularity**: Explored the relationship between tilt stability and the metric subregularity of the subdifferential mapping.
- **Convergence Rate of the Proximal Point Algorithm**: Using these conditions, provided the convergence rate of the proximal point algorithm.
- **Application in the Elliptic Tracking Problem**: Proved the equivalence between the second - order conditions and the Lipschitz stability of the solution with respect to data uncertainty.
### Mathematical Formulas
- **Growth Condition**:
\[
f(x)\geq f(\bar{x})+\gamma\|x - \bar{x}\|^{p}\quad\text{for all }x\in B(\bar{x},\delta)
\]
- **Monotonicity Condition**:
\[
\langle\xi,x - \bar{x}\rangle\geq\sigma\|x - \bar{x}\|^{p}\quad\text{for all }x\in\arg\min_{y\in B(\bar{x},\delta)}\{f(y)-\langle\xi,y\rangle\}
\]
- **Tilt Sub - stability**:
\[
\|x - \bar{x}\|\leq\kappa\|\xi\|^{q / p}\quad\text{for all }x\in\arg\min_{y\in B(\bar{x},\delta)}\{f(y)-\langle\xi,y\rangle\}
\]
- **Polyak - Łojasiewicz - like Condition**:
\[
f(x)-f(\bar{x})\leq\mu\|\xi\|^{q}\quad\text{for all }x\in\arg\min_{y\in B(\bar{x},\delta)}\{f(y)-\langle\xi,y\rangle\}
\]
These conditions and results provide important tools and methods for the theoretical analysis and algorithm design of non - convex optimization problems.