Abstract:In this paper, we introduce a new tensor decomposition for third order tensors, which decomposes a third order tensor to three third order low rank tensors in a balanced way. We call such a decomposition the triple decomposition, and the corresponding rank the triple rank. For a third order tensor, its CP decomposition can be regarded as a special case of its triple decomposition. The triple rank of a third order tensor is not greater than the middle value of the Tucker rank, and is strictly less than the middle value of the Tucker rank for an essential class of examples. These indicate that practical data can be approximated by low rank triple decomposition as long as it can be approximated by low rank CP or Tucker decomposition. This theoretical discovery is confirmed numerically. Numerical tests show that third order tensor data from practical applications such as internet traffic and video image are of low triple ranks. A tensor recovery method based on low rank triple decomposition is proposed. Its convergence and convergence rate are established. Numerical experiments confirm the efficiency of this method.

What problem does this paper attempt to address?

### What problems does this paper attempt to solve? This paper aims to introduce a new tensor decomposition method, called **triple decomposition**, specifically designed for handling third - order tensors. Specifically, it decomposes a third - order tensor into the product of three low - rank third - order tensors and defines the corresponding **triple rank**. The main objectives of the paper include: 1. **Propose a new tensor decomposition method**: Introduce triple decomposition and its related concepts and prove its theoretical properties. 2. **Compare the advantages and disadvantages of different decomposition methods**: Compare with the existing CP decomposition and Tucker decomposition, and show that triple decomposition has a lower rank in some cases, thus more effectively approximating the actual data. 3. **Apply to actual data**: Verify the effectiveness of triple decomposition in practical applications through numerical experiments, such as Internet traffic data and video image data. 4. **Develop a tensor recovery method**: Propose a new tensor recovery method based on low - rank triple decomposition and analyze its convergence and convergence rate. #### Specific problem description 1. **Definition and properties of triple decomposition**: - Defines triple decomposition, that is, decomposes a third - order tensor \( \mathcal{X} \in \mathbb{R}^{n_1\times n_2\times n_3} \) into three low - rank third - order tensors \( \mathcal{A}, \mathcal{B}, \mathcal{C} \) such that: \[ \mathcal{X}=\mathcal{A}\mathcal{B}\mathcal{C} \] where \( \mathcal{A} \in \mathbb{R}^{n_1\times r\times r}\), \( \mathcal{B} \in \mathbb{R}^{r\times n_2\times r}\), \( \mathcal{C} \in \mathbb{R}^{r\times r\times n_3}\). - Defines triple rank, that is, the minimum \( r \) value that satisfies the above decomposition. 2. **Relationship with other decomposition methods**: - Proves that triple decomposition can be regarded as a special case of CP decomposition. - Compares the relationship between triple rank and CP rank and Tucker rank, especially that triple rank usually does not exceed the intermediate value of Tucker rank and is strictly less than Tucker rank in some cases. 3. **Performance in practical applications**: - Verifies the effectiveness of triple decomposition in practical applications through numerical experiments, such as Internet traffic data and video image data. - Proposes a tensor recovery method based on low - rank triple decomposition and proves its convergence and convergence rate. #### Summary of mathematical formulas - **Triple decomposition**: \[ x_{ijt}=\sum_{p,q,s = 1}^r a_{iqs}b_{pjs}c_{pqt} \] - **Triple rank**: \[ \text{TriRank}(\mathcal{X})=r \] - **CP decomposition**: \[ \mathcal{X}=[[\mathbf{A},\mathbf{B},\mathbf{C}]] \] - **Tucker decomposition**: \[ \mathcal{X}=\mathcal{D}\times_1\mathbf{U}\times_2\mathbf{V}\times_3\mathbf{W} \] Through these methods, the paper shows the advantages of triple decomposition in handling high - dimensional data and provides new tools and methods for practical applications.

Triple Decomposition and Tensor Recovery of Third Order Tensors

Multi-Tubal Rank of Third Order Tensor and Related Low Rank Tensor Completion Problem

An Orthogonal Equivalence Theorem for Third Order Tensors

Optimal low rank tensor recovery

Canonical polyadic decomposition of third-order tensors: relaxed uniqueness conditions and algebraic algorithm

An Iterative Reweighted Method for Tucker Decomposition of Incomplete Multiway Tensors

Robust Tensor CUR Decompositions: Rapid Low-Tucker-Rank Tensor Recovery with Sparse Corruptions

Decomposition of linear tensor transformations

Tensor Star Decomposition

A Multi-resolution Low-rank Tensor Decomposition

Tensor N-tubal rank and its convex relaxation for low-rank tensor recovery

Singular Value Decomposition of Third Order Quaternion Tensors

Robust Tensor CUR Decompositions: Rapid Low-Tucker-Rank Tensor Recovery with Sparse Corruption

$p$-order Tensor Products with Invertible Linear Transforms

An Introduction to the Deviatoric Decomposition in Three-Dimensions Based on a Recursive Formula

Revisiting High-Order Tensor Singular Value Decomposition From Basic Element Perspective

Random Tensor Theory for Tensor Decomposition

Computational and Statistical Guarantees for Tensor-on-Tensor Regression with Tensor Train Decomposition

Irreducible Decompositions of Physical Tensors of High Orders

On uniqueness and computation of the decomposition of a tensor into multilinear rank-$(1, L_{r},L_{r})$ terms

Tensor CUR Decomposition under T-Product and Its Perturbation