Core Concepts
Efficiently approximate products of H2-matrices with block-relative error control using specialized algorithms.
Abstract
H2-matrices exploit local low-rank structures in large matrices for efficient approximation. An algorithm is introduced to approximate the product of two H2-matrices with controllable accuracy. Specialized tree structures are used to represent the exact product, enabling rigorous error control strategies. The manuscript discusses the challenges in handling dense matrices arising from non-local operators and integral equations. Techniques like fast multipole methods and hierarchical matrices are employed to reduce complexity while maintaining accuracy. The article delves into the structure of H2-matrices, their properties, and efficient algorithms for matrix operations.
Stats
H2-matrices reduce storage requirements to O(nk) for n-dimensional matrices with local rank k.
Matrix-vector multiplication complexity is reduced to O(nk) operations.
Fast multipole method offers an efficient approach for integral equations with complexity O(nk log n).
Hierarchical matrices extend applications while maintaining optimal complexity of O(nk log n).
Algorithms for approximating products of hierarchical matrices have linear complexity O(nk^2).