[Paper Review] Coercive functions from a topological viewpoint and properties of minimizing sets of convex functions appearing in image restoration

René Ciak|arXiv (Cornell University)|Jun 23, 2015

Sparse and Compressive Sensing Techniques12 references4 citations

TL;DR

This paper investigates coercive functions and minimizing sets of convex functions in the context of image restoration, using topological and variational analysis. It establishes that equality in a key optimality condition holds only when the candidate point is not a minimizer, demonstrating the necessity of this condition via a counterexample with Ψ(x) = x².

ABSTRACT

Many tasks in image processing can be tackled by modeling an appropriate data fidelity term $Φ: \mathbb{R}^n ightarrow \mathbb{R} \cup \{+\infty\}$ and then solve one of the regularized minimization problems \begin{align*} &{}(P_{1,τ}) \qquad \mathop{ m argmin}_{x \in \mathbb R^n} \big\{ Φ(x) \;{ m s.t.}\; Ψ(x) \leq τ\big\} \\ &{}(P_{2,λ}) \qquad \mathop{ m argmin}_{x \in \mathbb R^n} \{ Φ(x) + λΨ(x) \}, \; λ> 0 \end{align*} with some function $Ψ: \mathbb{R}^n ightarrow \mathbb{R} \cup \{+\infty\}$ and a good choice of the parameter(s). Two tasks arise naturally here: \begin{align*} {}& ext{1. Study the solver sets ${ m SOL}(P_{1,τ})$ and ${ m SOL}(P_{2,λ})$ of the minimization problems.} \\ {}& ext{2. Ensure that the minimization problems have solutions.} \end{align*} This thesis provides contributions to both tasks: Regarding the first task for a more special setting we prove that there are intervals $(0,c)$ and $(0,d)$ such that the setvalued curves \begin{align*} τ\mapsto {}& { m SOL}(P_{1,τ}), \; τ\in (0,c) \\ {} λ\mapsto {}& { m SOL}(P_{2,λ}), \; λ\in (0,d) \end{align*} are the same, besides an order reversing parameter change $g: (0,c) ightarrow (0,d)$. Moreover we show that the solver sets are changing all the time while $τ$ runs from $0$ to $c$ and $λ$ runs from $d$ to $0$. In the presence of lower semicontinuity the second task is done if we have additionally coercivity. We regard lower semicontinuity and coercivity from a topological point of view and develop a new technique for proving lower semicontinuity plus coercivity. Dropping any lower semicontinuity assumption we also prove a theorem on the coercivity of a sum of functions.

Motivation & Objective

• To understand the role of coercivity and topological properties in characterizing minimizing sets of convex functions.
• To clarify conditions under which optimality conditions achieve equality, particularly focusing on the necessity of non-minimality.
• To investigate the structure of sublevel sets and their relationship to subdifferentials in variational problems.
• To provide a theoretical foundation for image restoration models relying on convex minimization.

Proposed method

• Analyzes the function Ψ(x) = x² as a canonical example to test optimality conditions.
• Examines the sublevel set S = lev_Ψ(0)Ψ = {0} at the minimizer x* = 0 ∈ int(dom Ψ).
• Compares the positive ray spanned by the subdifferential R_+ ∂Ψ(x*) = {0} with the subdifferential of the indicator function ∂ι_S(x*).
• Uses topological arguments to show that equality in the optimality condition fails when x* is a minimizer.
• Applies variational analysis tools, including subdifferential calculus and indicator functions, to study minimizing sets.
• Establishes the necessity of the non-minimizer condition via contradiction using the given example.

Experimental results

Research questions

• RQ1 Under what conditions does equality hold in the optimality condition (LABEL:wichtig) for convex functions?
• RQ2 Why is it essential that x* is not a minimizer for equality to be achieved in the given condition?
• RQ3 How do the subdifferentials of Ψ and the indicator function ι_S relate at the minimizer x* = 0?
• RQ4 What topological and variational properties characterize minimizing sets in convex optimization?
• RQ5 How does the example Ψ(x) = x² illustrate the necessity of the non-minimizer condition?

Key findings

• The equality in the optimality condition (LABEL:wichtig) fails when x* is a minimizer, as demonstrated by the counterexample Ψ(x) = x².
• At x* = 0, the subdifferential R_+ ∂Ψ(x*) = {0} is strictly contained in the full subdifferential ∂ι_S(x*) = R, showing a strict inclusion.
• The set S = lev_Ψ(0)Ψ = {0} is a singleton, indicating that the minimizer is isolated and the sublevel set is trivial.
• The analysis confirms that the condition that x* is not a minimizer is essential for equality in the stated optimality condition.
• The example illustrates that coercivity and topological structure of the domain influence the behavior of subdifferentials at minimizers.
• The result highlights a subtle but critical distinction in subdifferential calculus when applied to minimizing points in convex analysis.