Handling massive spatial volumes in time domain simulations using Triggered Cells Method

TL;DR

This paper introduces the Triggered Cells Method (TCM), a technique that significantly reduces memory and computation requirements for time domain wave simulations, especially effective for pulse or burst propagation scenarios.

Contribution

The paper presents the TCM, a novel approach that enhances memory capacity and reduces computation time in wave simulations, applicable to various differential equations and grid types.

Findings

TCM increases available memory by about two orders of magnitude.

TCM reduces calculation time proportionally, enabling efficient single CPU simulations.

The method effectively resolves boundary reflection issues without complex phase matching layers.

Abstract

Simulating wave propagation on discrete grid in time domain requires the description of both the parameters of the media and the values of the wave field at two different time steps within the simulated domain. For most practical cases this operation is very demanding on the system's memory and runtime. In some cases this demands renders the time domain simulation useless. By using the Triggered Cells Method (TCM) one may effectively increase the available memory of a given system by approximately two orders of magnitude in chosen scenarios, and reduce the calculation time by the same factor. The method is most applicable for pulse or bursts propagation, and can achieve outstanding results using single CPU, although it is easily parallelizable for multiple CPU's. This paper will introduce the method using the simple case of the Finite Difference Time Domain implementation of the scalar wave equation, although applicable for most differential equations under any grid type. A hybrid computational engine will be presented, that utilizes the Heterogenic Bulking Method. It will be demonstrated that in most relevant cases, the boundary reflection problem will be resolved using the TCM, without the need to implement sophisticated Phase Matching Layers.