Yonghong Yao,Lateef O. Jolaoso,Yekini Shehu
Abstract:Many recently proposed gradient projection algorithms with inertial extrapolation step for solving quasi-variational inequalities in Hilbert spaces are proven to be strongly convergent with no linear rate given when the cost operator is strongly monotone and Lipschitz continuous. In this paper, our aim is to design an inertial type gradient projection algorithm for quasi-variational inequalities and obtain its linear rate of convergence. Therefore, our results fill in the gap for linear convergence results for inertial type gradient projection algorithms for quasi variational inequalities in Hilbert spaces. We perform numerical implementations of our proposed algorithm and give numerical comparisons with other related inertial type gradient projection algorithms for quasi variational inequalities in the literature.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is that when solving Quasi - Variational Inequalities (QVIs) in Hilbert space, although the existing gradient projection algorithms with inertial extrapolation steps can converge strongly, there is a lack of results on the linear convergence rate. Specifically, when the cost operator is strongly monotone and Lipschitz continuous, these algorithms have been proven to converge strongly, but no linear convergence rate has been given. The goal of this paper is to design a gradient projection algorithm with inertial extrapolation steps and prove its linear convergence rate, thus filling this research gap.
### Main Contributions
1. **Linear Convergence Result**: This paper proposes a gradient projection algorithm with inertial extrapolation steps and proves that under strongly monotone and Lipschitz continuous conditions, the algorithm has a linear convergence rate.
2. **Extension of Inertial Factor**: Different from the inertial factor in existing algorithms which is restricted to [0, 1], the algorithm proposed in this paper allows the inertial factor to take negative and non - negative values, which increases the flexibility of the algorithm.
3. **Numerical Experiments**: Through numerical experiments on standard test problems, this paper shows the efficiency of the proposed algorithm and compares it with other related algorithms. The results show that the proposed algorithm is superior to other algorithms in many cases.
### Mathematical Background
- **Definition 2.1**:
- An operator \(A: H\rightarrow H\) is \(L\)-Lipschitz continuous if for all \(x, y\in H\):
\[
\|A(x) - A(y)\|\leq L\|x - y\|
\]
- An operator \(A: H\rightarrow H\) is \(\mu\)-strongly monotone if for all \(x, y\in H\):
\[
\langle A(x) - A(y), x - y\rangle\geq\mu\|x - y\|^{2}
\]
- **Lemma 2.2**:
- Suppose \(A: H\rightarrow H\) is \(L\)-Lipschitz continuous and \(\mu\)-strongly monotone, and the set - valued mapping \(K(\cdot)\) has non - empty, closed and convex values, and there exists \(\lambda\geq0\) such that:
\[
\|P_{K}(x)(z)-P_{K}(y)(z)\|\leq\lambda\|x - y\|
\]
and satisfies:
\[
\lambda+\frac{r}{1-\frac{\mu^{2}}{L^{2}}}< 1
\]
then the QVI has a unique solution.
### Main Results
- **Algorithm 1**: This paper proposes a gradient projection algorithm with inertial extrapolation steps and proves its linear convergence.
- **Theorem 3.3**: Suppose \(A\) is \(\mu\)-strongly monotone and \(L\)-Lipschitz continuous, and there exists \(\lambda\geq0\) satisfying condition (13), then the sequences \(\{x_{k}\}\) and \(\{z_{k}\}\) generated by Algorithm 1 converge linearly to the unique solution \(x^{*}\in K(x^{*})\) of QVI (1).
### Numerical Experiments
- **Experimental Setup**: Use the problems in the QVILIB test library for numerical experiments and compare the performance of the proposed algorithm with that of existing algorithms.
- **Performance Indicators**: Including the number of iterations, execution time, the number of feasible solutions, etc.
- **Experimental Results**: The proposed algorithm shows better performance in most cases, especially in terms of the number of iterations and execution time.
### Conclusion
This paper proposes a gradient projection algorithm with inertial extrapolation steps, proves its effectiveness in terms of linear convergence rate, and verifies its superiority through numerical experiments. This provides a new and effective tool for solving quasi - variational inequalities.