Abstract:Since the popularization of the Stiefel manifold for numerical applications in 1998 in a seminal paper from Edelman et al., it has been exhibited to be a key to solve many problems from optimization, statistics and machine learning. In 2021, Hüper et al. proposed a one-parameter family of Riemannian metrics on the Stiefel manifold, subsuming the well-known Euclidean and canonical metrics. Since then, several methods have been proposed to obtain a candidate for the Riemannian logarithm given any metric from the family. Most of these methods are based on the shooting method or rely on optimization approaches. For the canonical metric, Zimmermann proposed in 2017 a particularly efficient method based on a pure matrix-algebraic approach. In this paper, we derive a generalization of this algorithm that works for the one-parameter family of Riemannian metrics. The algorithm is proposed in two versions, termed backward and forward, for which we prove that it conserves the local linear convergence previously exhibited in Zimmermann's algorithm for the canonical metric.

What problem does this paper attempt to address?

### What problem does this paper attempt to solve? This paper aims to solve the problem of computing the Riemannian logarithm on the Stiefel manifold, especially for a family of Riemannian metrics with one parameter proposed by Hüper et al. in 2021. Specifically, the goals of the paper are: 1. **Generalize existing algorithms**: The paper generalizes the efficient matrix - algebraic method proposed by Zimmermann in 2017 for the classical metric to be applicable to this family of Riemannian metrics with parameters. This generalization includes two versions of the algorithm: forward and backward iterations. 2. **Ensure convergence**: The author proves that the new algorithm maintains the characteristics of Zimmermann's original algorithm in terms of local linear convergence and provides an explicit expression for the convergence rate of the backward iteration. 3. **Improve computational efficiency**: To reduce the computational cost, the paper introduces three types of forward iterations, which are more efficient than the backward iteration while maintaining local linear convergence. ### Background information The Stiefel manifold is an important class of manifolds and has wide applications in optimization, statistics, and machine learning. In particular, when dealing with orthogonal frames, subspaces, fixed - rank matrices, etc., the Stiefel manifold is a key tool. However, computing the shortest geodesic (i.e., the Riemannian distance) between two points on the Stiefel manifold is a challenging task, known as the geodesic endpoint problem. This problem is crucial for numerical applications because many operations depend on the distance calculation between two points. ### Main contributions of the paper - **Generalize existing algorithms**: Generalize Zimmermann's classical metric algorithm to a family of Riemannian metrics with parameters. - **Convergence analysis**: Prove the local linear convergence of the new algorithm and provide a theoretical analysis of the convergence rate. - **Improve efficiency**: Propose more efficient forward iteration methods to reduce the computational cost. Through these improvements, the paper provides new tools and methods for more complex and efficient numerical calculations on the Stiefel manifold.

An efficient algorithm for the Riemannian logarithm on the Stiefel manifold for a family of Riemannian metrics

Riemannian optimization on the symplectic Stiefel manifold using second-order information

Symplectic Stiefel manifold: tractable metrics, second-order geometry and Newton's methods

Riemannian optimization with a preconditioning scheme on the generalized Stiefel manifold

A Riemannian optimization method on the indefinite Stiefel manifold

Convergence of Inexact Steepest Descent Algorithm for Multiobjective Optimizations on Riemannian Manifolds Without Curvature Constraints

New vector transport operators extending a Riemannian CG algorithm to generalized Stiefel manifold with low-rank applications

Vector Transport Free Riemannian LBFGS for Optimization on Symmetric Positive Definite Matrix Manifolds

Riemannian Stochastic Proximal Gradient Methods for Nonsmooth Optimization over the Stiefel Manifold

Decentralized Riemannian Conjugate Gradient Method on the Stiefel Manifold

Riemannian Accelerated Zeroth-order Algorithm: Improved Robustness and Lower Query Complexity

Riemannian Levenberg-Marquardt Method with Global and Local Convergence Properties

On the Injectivity Radius of the Stiefel Manifold: Numerical investigations and an explicit construction of a cut point at short distance

Learning-Rate-Free Stochastic Optimization over Riemannian Manifolds

An Inexact Riemannian Proximal Gradient Method

Optimal Riemannian metric for Poincar{é} inequalities and how to ideally precondition Langevin dymanics

A Riemannian Alternating Descent Ascent Algorithmic Framework for Nonconvex-Linear Minimax Problems on Riemannian Manifolds

Riemannian Optimization for Non-convex Euclidean Distance Geometry with Global Recovery Guarantees

Optimization without Retraction on the Random Generalized Stiefel Manifold

Learning to Optimize on Riemannian Manifolds