Investigation of floating-point round-off errors within time-domain electromagnetic field computations

We present an investigation of floating-point round-off errors within time-domain electromagnetic field computations. Our eventual goal is to determine the minimal computational requirements for the successful implementation of algorithms for computational electromagnetics. Here, we specifically investigate these errors with respect to the Symmetric Condensed Node (SCN) Transmission Line Matrix (TLM) algorithm. Simulations are performed using the IEEE standard for single-precision floating-point arithmetic. An artificial truncation operation is added to an otherwise normal SCN-TLM code. The artificial truncation operation lowers the effective precision of the computation. We study the effect of this truncation operation in terms of the resultant stability and accuracy.