Sharp error bounds for complex floating-point inversion (Q342874)

This paper deals with the accuracy of inversion of a nonzero complex number given by its real and imaginary parts as floating-point numbers. It is assumed that the underlying floating-point arithmetic has radix \(2\) and precision \(p\) greater than or equal \(2\). An unbounded exponent range is assumed so that the results apply to practical floating-point calculations according to IEEE 754 standard as long as underflow and overflow do not occur.