What is usually missing from these articles is even a simplified error analysis. We can all count the floating-point operations, and today caches sometimes inf

What is usually missing from these articles is even a simplified error analysis. We can all count the floating-point operations, and today caches sometimes influence matrix algorithms more than raw flop counts. The problem gets more interesting when the matrix becomes ill conditioned and different algorithms will produce very different results.