Abstract:We consider the optimization problem with a generally quadratic matrix constraint of the form $X^TAX = J$, where $A$ is a given nonsingular, symmetric $n\times n$ matrix and $J$ is a given $k\times k$ matrix, with $k\leq n$, satisfying $J^2 = I_k$. Since the feasible set constitutes a differentiable manifold, called the indefinite Stiefel manifold, we approach this problem within the framework of Riemannian optimization. Namely, we first equip the manifold with a Riemannian metric and construct the associated geometric structures, then propose a retraction based on the Cayley transform, and finally suggest a Riemannian gradient descent method using the attained materials, whose global convergence is guaranteed. Our results not only cover the known cases, the orthogonal and generalized Stiefel manifolds, but also provide a Riemannian optimization solution for other constrained problems which has not been investigated. Numerical experiments are presented to justify the theoretical results.
What problem does this paper attempt to address?
The main problem that this paper attempts to solve is the optimization problem with a specific quadratic matrix constraint. Specifically, it considers the optimization problem under the general quadratic matrix constraint in the form of \( X^TAX = J \), where \( A \) is a given non - singular symmetric \( n\times n \) matrix, \( J \) is a given \( k\times k \) matrix (satisfying \( J^2 = I_k \), that is, \( J \) is an involutory matrix), and \( k\leq n \).
Since the feasible set forms a differentiable manifold, called the indefinite Stiefel manifold, the author adopts the Riemannian optimization framework to deal with this problem. The specific steps are as follows:
1. **Define the Riemannian metric**: First, equip the manifold with a Riemannian metric and construct the relevant geometric structure.
2. **Construct the retraction mapping**: Propose a retraction method based on the Cayley transform.
3. **Propose the gradient - descent algorithm**: Using the above materials, propose a Riemannian gradient - descent method and ensure its global convergence.
In addition, the results of the paper not only cover the known cases (such as the orthogonal Stiefel manifold and the generalized Stiefel manifold), but also provide Riemannian optimization solutions for other constraint optimization problems that have not been studied yet. The effectiveness of the theoretical results is verified through numerical experiments.
In summary, this paper aims to solve the optimization problem with a general quadratic matrix constraint on the indefinite Stiefel manifold through the Riemannian optimization method and provides a new and effective solution approach.
### Formula Summary
- **Constraint condition**:
\[
X^TAX = J
\]
where \( A\in\mathbb{R}^{n\times n} \) is a symmetric non - singular matrix, and \( J\in\mathbb{R}^{k\times k} \) satisfies \( J^2 = I_k \).
- **Indefinite Stiefel manifold**:
\[
iSt_{A,J}(k,n)=\{X\in\mathbb{R}^{n\times k}:X^TAX = J\}
\]
- **Tangent space**:
\[
T_X iSt_{A,J}(k,n)=\{Z\in\mathbb{R}^{n\times k}:Z^TAX + X^TAZ = 0\}
\]
- **Riemannian metric**:
\[
g_{M_X}(Z_1,Z_2)=\text{tr}(Z_1^T M_X Z_2)
\]
- **Cayley retraction**:
\[
R_{\text{cay}}^X(Z)=\left(I_n-\frac{1}{2}S_{X,Z}A\right)^{-1}\left(I_n+\frac{1}{2}S_{X,Z}A\right)X
\]
where \( S_{X,Z}=XJZ^TAXJX^T - XJZ^T+ZJX^T \).
These formulas and methods together form the core content of the paper and solve the optimization problem with special quadratic matrix constraints.