Iterative Methods for Computing the Moore-Penrose Pseudoinverse of Quaternion Matrices, with Applications (2508.16979v1)

Abstract: We develop iterative methods to compute the Moore-Penrose pseudoinverse of quaternion matrices directly in $\mathbb{H}$. We introduce a damped Newton-Schulz (NS) iteration, proving its convergence under spectral scaling $X_0=\alpha A^H$ with $\alpha\in(0,2/|A|_2^2)$. We derive higher-order hyperpower NS schemes with optimized factorizations. Beyond NS, we propose a randomized sketch-and-project method (RSP-Q) and a conjugate gradient on normal equations (CGNE-Q). Numerically, spectrally-scaled NS achieves superior accuracy and runtime. In applications (image completion, filtering, deblurring), the NS family delivers state-of-the-art performance with the lowest wall time, providing efficient solvers for large-scale quaternion inverse problems.