Finite Convergence of Circumcentered-Reflection Method on Closed Polyhedral Cones in Euclidean Spaces

Hongzhi Liao

2024-11-30

Abstract:The Circumcentered Reflection Method (CRM) is a recently developed projection method for solving convex feasibility problems. It offers preferable convergence properties compared to classic methods such as the Douglas-Rachford and the alternating projections method. In this study, our first main theorem establishes that CRM can identify a feasible point in the intersection of two closed convex cones in \(\mathbb{R}^2\) from any starting point in the Euclidean plane. We then apply this theorem to intersections of two polyhedral sets in \(\mathbb{R}^2\) and two wedge-like sets in \(\mathbb{R}^n\), proving that CRM converges to a point in the intersection from any initial position finitely. Additionally, we introduce a modified technique based on CRM, called the Sphere-Centered Reflection Method. With the help of this technique, we demonstrate that CRM can locate a feasible point in finitely many iterations in the intersection of two proper polyhedral cones in \(\mathbb{R}^3\) when the initial point lies in a subset of the complement of the intersection's polar cone. Lastly, we provide an example illustrating that finite convergence may fail for the intersection of two proper polyhedral cones in \(\mathbb{R}^3\) if the initial guess is outside the designated set.

Optimization and Control

What problem does this paper attempt to address?

The problem that this paper attempts to solve is about the finite convergence of the Circumcentered Reflection Method (CRM) under specific geometric structures. Specifically, the author focuses on the performance of CRM in the intersection of closed polyhedral cones and proves that under certain conditions, CRM can find a feasible point within a finite number of steps.

### Main problems and goals:

1. **Finite convergence**:
- The author aims to prove that CRM can find a feasible point in the intersection of two closed convex cones within a finite number of steps.
- Specifically, the paper proves that in two - dimensional Euclidean space, CRM can find a feasible point in the intersection of two closed convex cones within at most three steps (see Theorem 6).

2. **Extension to polyhedral sets**:
- The author extends the above results to polyhedral sets in two - dimensional space and wedge - like sets in higher - dimensional spaces and proves that CRM also has finite convergence in these cases (see Theorem 9 and Corollary 11).

3. **Polyhedral cones in three - dimensional space**:
- In three - dimensional space, the author introduces an improved technique - Sphere - Centered Reflection Method (SRM) and proves that for two appropriate polyhedral cones, if the initial point is located within a certain region, CRM can find a feasible point within a finite number of steps (see Theorem 18).
- At the same time, the author also provides a counter - example to show that when the initial point is not in this specific region, the finite convergence may not hold (see Example 19).

### Key conclusions:

- **Closed convex cones in two - dimensional space**: CRM can find a feasible point in the intersection of two closed convex cones within at most three steps.
- **Polyhedral sets**: CRM also shows finite convergence in two - dimensional polyhedral sets and wedge - like sets in higher - dimensional spaces.
- **Polyhedral cones in three - dimensional space**: By introducing SRM, CRM can find a feasible point in the intersection of two polyhedral cones in three - dimensional space under certain conditions, but the position of the initial point has an important impact on the convergence.

### Mathematical formula representation:

- **Reflection operator**:
\[
R_C(x)=2P_C(x)-x
\]
where \(P_C(x)\) represents the projection of point \(x\) onto set \(C\).

- **Circumcentered Reflection Operator**:
\[
C_T(x): = C(x,R_A(x),R_B(R_A(x)))
\]
where \(C(x,y,z)\) represents the circumcenter of points \(x,y,z\).

- **Distance definition**:
\[
d_C(x):=\inf_{y\in C}\|x - y\|
\]

Through these formulas and theoretical analysis, the author shows the finite convergence performance of CRM under different geometric structures and provides new ideas and methods for future research.