Efficient computation of erfc(x) for large arguments

A new, infinite series representation for the error function is developed. It is especially suitable for computing erfc(x) for large x. For instance, for any x 4, the error function can be evaluated with a relative error less than 10 10 by using only eight terms. Similarly, the error function can be evaluated with a relative error less than 8 10 7 for any x 2 using just six terms. An analytical bound is derived to show that the total error due to series truncation and undersampling rapidly decreases as x increases. Comparisons with two other series are provided.