Probabilistic Construction and Analysis of Seminormalized Hadamard Matrices

TL;DR

This paper introduces probabilistic algorithms for constructing seminormalized Hadamard matrices, analyzes their probabilistic properties, and demonstrates their effectiveness in low-order cases, offering a resource-efficient alternative to exhaustive enumeration.

Contribution

It develops and analyzes probabilistic methods like RVS and OSA for constructing SH-matrices, providing insights into their success probabilities and distribution discrepancies.

Findings

Probabilistic algorithms effectively construct SH-matrices.

Orthogonal probability p between SH-vectors is analyzed.

Distribution discrepancy between known and expected SH-matrices is addressed.

Abstract

Let o be a 4k-length column vector whose all entries are 1s, with k a positive integer. Let V={v_i} be a set of semi-normalized Hadamard (SH)-vectors, which are 4k-length vectors whose 2k entries are -1s and the remaining 2k are 1s. We define a 4k-order QSH (Quasi SH)-matrix, Q, as a 4kx4k matrix where the first column is o and the remaining ones are distinct v_i in V. When Q is orthogonal, it becomes an SH-matrix H. Therefore, 4k-order SH-matrices can be built by enumerate all possible Q from every combination of v_i, then evaluate the orthogonality of each one of them. Since such exhaustive method requires a large amount of computing resource, we can employ probabilistic algorithms to construct H, such as, by Random Vector Selection (RVS) or the Orthogonalization by Simulated Annealing (OSA) algorithms. We demonstrate the constructions of low-order SH-matrices by using these methods. We also analyze some probabilistic aspects of the constructions, including orthogonal probability p between a pair of randomly selected SH-vectors, the existence probability p_H|Q that a randomly generated Q is in fact an SH-matrix H, and address the discrepancy of the distribution between the known number of SH-matrix with expected number derived from the probabilistic analysis.