errors in floating point arithmetic