1: \begin{abstract}

2: 	Recent works put much effort into \emph{tensor network structure search} (TN-SS), aiming to select suitable tensor network (TN) structures, involving the TN-ranks, formats, and so on, for the decomposition or learning tasks.

3: 	In this paper, we consider a practical variant of TN-SS, dubbed \emph{TN permutation search}~(TN-PS), 

4: 	in which we search for good mappings from tensor modes onto TN vertices (core tensors) for compact TN representations.

5: 	We conduct a theoretical investigation of TN-PS and propose a practically-efficient algorithm to resolve the problem.

6: 	Theoretically, we prove the counting and metric properties of search spaces of TN-PS, analyzing for the first time the impact of TN structures on these unique properties. 

7: 	Numerically, we propose a novel \emph{meta-heuristic} algorithm, in which the searching is done by randomly sampling in a neighborhood established in our theory, and then recurrently updating the neighborhood  until convergence.

8: 	Numerical results demonstrate that the new algorithm can reduce the required model size of TNs in extensive benchmarks, implying the improvement in the expressive power of TNs.

9: 	Furthermore, the computational cost for the new algorithm is significantly less than that in~\cite{li2020evolutionary}.	

10: \end{abstract}

11: