Abstract:We show that finding rank-$R$ decompositions of a 3D tensor, for $R\le 4$, over a fixed finite field can be done in polynomial time. However, if some cells in the tensor are allowed to have arbitrary values, then rank-2 is NP-hard over the integers modulo 2. We also explore rank-1 decomposition of a 3D tensor and of a matrix where some cells are allowed to have arbitrary values.
What problem does this paper attempt to address?
The main problem that this paper attempts to solve is the computational complexity of low - rank tensor decomposition over a finite field. Specifically, the paper explores how to find a decomposition of rank \(R\) (where \(R\leq4\)) for a given three - dimensional tensor \(T\) in a fixed finite field, and proves that under certain conditions this problem can be solved in polynomial time. In addition, the paper also studies the computational complexity of low - rank tensor decomposition when some elements in the tensor can take arbitrary values (i.e., "wildcards").
### Main Problems
1. **Polynomial - Time Algorithm for Low - Rank Tensor Decomposition**:
- The paper shows that for a three - dimensional tensor of a fixed rank \(R\leq4\), the rank - \(R\) decomposition over a finite field can be completed in polynomial time.
- Specifically, for an \(n\times n\times n\) tensor \(T\), finding a rank - \(R\) decomposition or determining its non - existence over a finite field \(F\) can be completed in \(O(n^3 + f(|F|, R)n^2)\) time, where \(f\) is some function related to the finite field and the rank.
2. **Computational Complexity of Low - Rank Tensor Decomposition with Wildcards**:
- When some elements in the tensor are allowed to take arbitrary values, the paper proves that the rank - 2 decomposition is NP - hard in the case of integers modulo 2.
- This is proved by reducing the Not - All - Equal 3SAT (NAE - 3SAT) problem to the rank - 2 decomposition problem with wildcards.
### Solution Overview
- **Polynomial - Time Algorithm**: The paper proposes a polynomial - time algorithm to solve the three - dimensional tensor decomposition problem of rank \(R\leq4\). This algorithm simplifies the calculation by extracting the basis of tensor slices and converting it into a matrix problem.
- **NP - Hardness Result**: For the case with wildcards, the paper proves by the reduction method that the rank - 2 decomposition is NP - hard in the case of integers modulo 2. This shows that in this case, the complexity of the problem increases significantly.
### Key Formulas
- The basic formula for tensor decomposition is:
\[
T_{i,j,k}=\sum_{r = 0}^{R-1}A_{r,i}B_{r,j}C_{r,k}\quad\forall i,j,k
\]
or written as:
\[
T=\sum_{r = 0}^{R-1}A_r\times B_r\times C_r
\]
where \(\times\) represents the outer product.
### Summary
This paper mainly solves the computational complexity problem of low - rank tensor decomposition over a finite field, especially for the case of a fixed rank \(R\leq4\), provides a polynomial - time algorithm, and explores the complexity when there are wildcards. These results are of great significance for understanding the computational properties of tensor decomposition, especially in application fields such as fast matrix multiplication.