Sparse matrix computations with application to solve system of nonlinear equations

Abstract

Numerical linear algebra is an essential ingredient in algorithms for solving problems in optimization, nonlinear equations, and differential equations. Spanning diverse application areas, from economic planning to complex network analysis, modeling and solving problems arising in those areas share a common theme: numerical calculations on matrices that are sparse or structured or both. Linear algebraic calculations involving sparse matrices of order 10 9 are now routine. In this article, we give an overview of scientific calculations where effective utilization of properties such as sparsity, problem structure, etc. play a vital role and where the linear algebraic calculations are much more complex than their dense counterpart. This is partly because operation and storage involving known zeros must be avoided, and partly because the fact that modern computing hardware may not be amenable to the specialized techniques needed for sparse problems. We focus on sparse calculations arising in nonlinear equation solving using the Newton method.

Fast and computationally efficient generative adversarial network algorithm for unmanned aerial vehicle–based network coverage optimization

“…To tackle this challenge, we employed a correction mechanism to address this issue, which is designed as follows. The proposed correction algorithm takes the sparse matrix Y , and to accelerate the computations, it transforms it into a scalable coordinate storage scheme represented by the set Y , 25 solely encompassing the space coordinates [ x , y ] where UAVs are present. Note that the transformation f of the sparse matrix to the compressed set f : Y → Y allows for faster computations than that with regular sparse matrices.…”

Integer partition manifolds and phonon damping in one dimension

“…] and find the evolution of the system's state of the system using the modern methods for sparse systems of differential equations [30].…”