1: \begin{abstract}

2: We propose a novel rank-adaptive higher-order orthogonal iteration (HOOI) algorithm to compute the truncated Tucker decomposition of higher-order tensors with a given error tolerance, and prove that the method is locally optimal and monotonically convergent.

3: A series of numerical experiments related to both synthetic and real-world tensors are carried out to show that the proposed rank-adaptive HOOI algorithm is advantageous in terms of both accuracy and efficiency.

4: Some further analysis on the HOOI algorithm and the classical alternating least squares method

5: are presented to further understand why rank adaptivity can be introduce into the HOOI algorithm and how it works.

6: %reveals the reason why the rank-adaptive HOOI algorithm is superior to other methods.

7: 

8: %improves the previous algorithms based on truncated higher-order singular value decomposition ($t$-HOSVD) or sequentially truncated HOSVD ($st$-HOSVD). We prove that the rank-adaptive HOOI is local optimal and enjoys the monotonic convergence. Numerical experiments show that the proposed algorithm can obtain a more accurate truncation with only a small increase in time cost, especially a good initial guess is given.

9: \keywords{Truncated Tucker decomposition \and Low multilinear-rank approximation \and Higher-order singular value decomposition \and Rank-adaptive higher-order orthogonal iteration }

10: % \PACS{PACS code1 \and PACS code2 \and more}

11: % \subclass{MSC code1 \and MSC code2 \and more}

12: \subclass{15A69 \and 49M27 \and 65D15 \and 65F55}

13: \end{abstract}

14: