\(O(d \log N)\)-quantics approximation of \(N\)-\(d\) tensors in high-dimensional numerical modeling (Q717125)

This paper deals with new prospects for tensor-structured data formats in high-dimensional applications based on the idea of quantics representations. The described quantics-based tensor approximation methods are applied to discrete functions over \(N\)-\(d\) tensors, and to the related matrices representing the discrete analogs of differential-integral operators. The results obtained make an important step towards the almost grid-independent representation of high order tensors. In particular, it is proved that the ``exponential, trigonometric, and polynomial'' vectors have fixed QTT-rank uniformly in the vector size \(N\). Numerical illustrations demonstrate the value of the QTT method presented in this paper in tensor decomposition of large multidimensional data sets arising in numerical solution of partial differential equations and in numerical quantum chemistry.