Convergence of the alternating least squares algorithm for CP tensor decompositions

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Hu, Nicholas, Iwen, Mark A., Needell, Deanna, Wang, Rongrong
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909617078927360
author Hu, Nicholas
Iwen, Mark A.
Needell, Deanna
Wang, Rongrong
author_facet Hu, Nicholas
Iwen, Mark A.
Needell, Deanna
Wang, Rongrong
contents The alternating least squares (ALS/AltLS) method is a widely used algorithm for computing the CP decomposition of a tensor. However, its convergence theory is still incompletely understood. In this paper, we prove explicit quantitative local convergence theorems for CP-AltLS applied to orthogonally decomposable and incoherently decomposable tensors. Specifically, we show that CP-AltLS converges polynomially with order $N-1$ for $N$th-order orthogonally decomposable tensors and linearly for incoherently decomposable tensors, with convergence being measured in terms of the angles between the factors of the exact tensor and those of the approximate tensor. Unlike existing results, our analysis is both quantitative and constructive, applying to standard CP-AltLS and accommodating factor matrices with small but nonzero mutual coherence, while remaining applicable to tensors of arbitrary rank. We also confirm these rates of convergence numerically and investigate accelerating the convergence of CP-AltLS using an SVD-based coherence reduction scheme.
format Preprint
id arxiv_https___arxiv_org_abs_2505_14037
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Convergence of the alternating least squares algorithm for CP tensor decompositions
Hu, Nicholas
Iwen, Mark A.
Needell, Deanna
Wang, Rongrong
Numerical Analysis
15A69, 65F99, 65F45
The alternating least squares (ALS/AltLS) method is a widely used algorithm for computing the CP decomposition of a tensor. However, its convergence theory is still incompletely understood. In this paper, we prove explicit quantitative local convergence theorems for CP-AltLS applied to orthogonally decomposable and incoherently decomposable tensors. Specifically, we show that CP-AltLS converges polynomially with order $N-1$ for $N$th-order orthogonally decomposable tensors and linearly for incoherently decomposable tensors, with convergence being measured in terms of the angles between the factors of the exact tensor and those of the approximate tensor. Unlike existing results, our analysis is both quantitative and constructive, applying to standard CP-AltLS and accommodating factor matrices with small but nonzero mutual coherence, while remaining applicable to tensors of arbitrary rank. We also confirm these rates of convergence numerically and investigate accelerating the convergence of CP-AltLS using an SVD-based coherence reduction scheme.
title Convergence of the alternating least squares algorithm for CP tensor decompositions
topic Numerical Analysis
15A69, 65F99, 65F45
url https://arxiv.org/abs/2505.14037