On the best rank-1 approximation to higher-order symmetric tensors (Q2472705): Difference between revisions

Higher-order tensors can be encountered in many fields of theoretical disciplines. This tool is particularly useful in their linear or locally linear approaches as it enables very transparent and effective formulations (solid state or fluid mechanics, physics of field interactions, signal processing, higher-order statistics, etc.). An order-\(m\) tensor is an \(m\)-way array whose elements are accessed via \(m\) indices. The tensor is called symmetric if its elements are invariant under any permutation of their indices. A very important problem attracting many authors represents the existence of an optimal approximation of these tensors using simpler structures independent from any particular frame and techniques of their determination. Many steps and elements used in this study remember simpler analogies developed in classical matrix algebra. The respective generalizations presented by the authors (eigenvalues, eigenvectors, bi-orthogonal systems, etc.) are transparent and provide an inspiration for physical applications as well as for further research in mathematics. The most important result presented in the paper is a proof of the theorem affirming that the best rank-1 (vector) approximation of a symmetric tensor with the even order \(m\) can be determined by \(m/2\) unit vectors. The authors introduce and recall all definitions needed as well as previously published lemmas and theorems. Many references are given including those related with physical applications. After the general theorem a special case of the symmetric tensor \((m=4, n=2)\) is discussed including all details necessary to understand the internal structure of the whole approximation process. Some results remember properties of the elasticity tensor in the 2D theory of linear elasticity, its invariants and auxiliary orthogonal matrix. Nevertheless some more attention should be paid to solution uniqueness, which can be unclear especially for higher \(m\).