Tag Archives: performance_profile

Let and be distinct floating point numbers. How small can the relative difference between and be? For IEEE double precision arithmetic the answer is , which is called the unit roundoff. What if we now let and be vectors and … Continue reading →