Abstract:This paper provides parallel implementation of dynamic equations for solving complex eigenvalue problems, which are prevalent in engineering fields. The principle of the approach is that the eigenvalue problems is first transformed into the associated problems of solving large systems of nonlinear dynamic equations by optimization techniques and then waveform relaxation in circuit simulation is used to compute the resulting cubic dynamic equations.Because the stable state solutions of the nonlinear dynamic equations correspond to the solutions of the eigenvalue problems, often these dynamic equations were mapped into a neural network, then the transient response of the neural network is obtained by hardware or software implementation of the neural network. This paper introduces the waveform relaxation method to numerically compute the transient solutions of the dynamic equations in order to avoid the complicated neural network circuit design. The waveform relaxation method is a "divide and conquer" approach, that is, at a differential equation level it decouples a large dynamic system into some small dynamic subsystems in the time domain and these subsystems are then solved independently from each other in modern parallel processor environments. The proposed algorithm can efficiently determine all eigenvalues and their eigenvectors for a general complex matrix. It is the first time that the waveform relaxation method is used to compute the cubic dynamic equations. The algorithm is successfully implemented on the IBM RS/6000 SuperPOWER2 system, which is a message passing multi processor. The computer language is ANSI C. The parallel MPI software is applied to message passing between different processors. Authors run the parallel program on different numbers of processors, and acquire the speedup corresponding to the numbers of processors in use respectively. As illustrated by the experimental results on general complex matrices with dimensions 3000×3000, the algorithm of the paper is highly parallelized. The speedup of the parallel algorithm has a linear characteristic property. And the communication overhead of parallel processing is rather low. In view of the linear speedup and the low communication overhead, the parallel implementation reported here is effective for transient computations of large dynamic systems in practical applications.

The Waveform BiConjugate Gradient Method for Parallel Transient Simulation of Semiconductor Devices

The Hybrid Accelerated Waveform Technique for Parallel Device Transient Simulation

A Parallel Simulator for Massive Reservoir Models Utilizing Distributed-Memory Parallel Systems

An Improved Algorithm for Drift Diffusion Transport and Its Application on Large Scale Parallel Simulation of Resistive Random Access Memory Arrays

Two-Stage Parallel Waveform Relaxation Method for Large-Scale Power System Transient Stability Simulation

Analysis of a New Accelerated Waveform Relaxation Method Based on the Time-Parallel Algorithm

Parallel Computational Techniques for Simulating Dopant Diffusion in Silicon

On Parallel Implementation of Dynamic Equations for Solving Eigenvalue Problems of Complex Matrices

A Parallel Transient Stability Algorithm for Large-Scale Power System Based on GPU Platform

A Parareal Algorithm Based on Waveform Relaxation

Parallel Computation of Transient Stability Using Symplectic Gauss Method and GPU

Parallelized Computational Technique for Signal Propagation Analysis at Very High Speed Differential Transmission Lines

GPU_PBTE: an Efficient Solver for Three and Four Phonon Scattering Rates on Graphics Processing Units

A parareal waveform relaxation algorithm for semi-linear parabolic partial differential equations

Parallel computation of power system transient stability based on symplectic Gauss algorithm and preconditioned GMRES method

Parareal Schwarz Waveform Relaxation Methods.

Improving Convergence Performance of Relaxation-Based Transient Analysis by Matrix Splitting in Circuit Simulation

A New Parareal Waveform Relaxation Algorithm for Time-Periodic Problems

A robust hybridizable discontinuous Galerkin scheme with harmonic averaging technique for steady state of real-world semiconductor devices

The Parallel Waveform Relaxation Stochastic Runge–Kutta Method for Stochastic Differential Equations

A fast implicit solver for semiconductor models in one space dimension