Tensor numerical methods in scientific computing (Q263816)

This is an up to date monograph on tensor numerical methods, a field, which has been essentially shaped by the author, who holds since 1999 the position of a senior researcher at the Max-Planck Institute for Mathematics in the Sciences in Leipzig (Germany). Among the 373 references cited in the book his own name appears in 89 of them as author or coauthor. \vskip 0.8em The newly developed tensor numerical methods provide a powerful tool for scientific computing when dealing with high-dimensional problems, e.g., multivariate convolution, Hartree-Fock spectral problems, dynamical Fokker-Planck equations, chemical master equations, stochastic and parametric PDEs, many particle modeling, elliptic equations with highly oscillating coefficients, homogenization problems, all of them studied, including numerical experiments, in detail in Chapter 5. \vskip 0.8em The book is full of a large variety of different topics which cannot all be mentioned here. To convey at least an impression of the contents, the headings of the five chapters are given here: {\parindent=0.7cm\begin{itemize}\item[--] Chapter 1: Introduction (8 pages) \item[--] Chapter 2: Theory on separable approximation of multivariate functions (61 pages) \item[--] Chapter 3: Multilinear algebra and nonlinear tensor approximation (83 pages) \item[--] Chapter 4: Superfast computations via quantized tensor approximation (73 pages) \item[--] Chapter 5: Tensor approach to multidimensional integrodifferential equations (132 pages) \end{itemize}} The titles of some sections in Chapters 3 and 4 may provide an additional idea of what can be found in the course of the text: canonical and Tucker tensor formats; direct methods of low rank approximation; alternating least squares iteration; additive and multiplicative dimension splitting, tensor train formats; the quantized tensor train approximation QTT (which was invented by the author in 2009 reducing the asymptotic storage costs to the logarithmic scale); QTT-FFT, convolution in logarithmic time.