The biggest problem when solving multidimensional linear systems in a tensor form is their exponential growth. This master thesis focuses on tensors which can be approximated by a low rank tensor. As an example we give parametric linear systems. A proof for their good approximation is provided.
We present hierarchical Tucker decomposition of a tensor. It is based on higher order singular value decomposition and has space complexity linear in the order of tensor. We discuss basic operations on tensors in hierarchical Tucker decomposition, used when truncating tensors and finding iterative solution of a problem. Some examples of application of the decomposition are given, some of which are compared to known methods for such problems.
|