Solving Baer Wave Equation Reduced to Three-Parameter Eigenvalue Problem by Dynamic Thread-Based Computing

Abstract

Real-time computation of eigenvalues is valuable in science and engineering. This is possible via memory-efficient, scalable, robust, and high-performance algorithms when we take advantage of supercomputing. Baer wave equation arises from applying the separation of variables to the Helmholtz equation. When the Baer wave equation is discretized, a three-parameter eigenvalue problem is obtained. In this study, we consider the computationally challenging problem of finding eigenvalue tuples in a three-parameter eigenvalue problem reduced from the Baer wave equation. We solve this problem using a fused parameter optimization algorithm by implementing a dynamic thread-based computation in C and MATLAB. We achieved scaled speed-up for the dense coefficient matrices of the problem from the Baer wave equation to run up to 64 threads in our C implementation. To the best of our knowledge, this is the first study to solve the three-parameter eigenvalue problem using parallel thread-based computing.