Exercise 2: Resolving the Apparent Chapter 4 vs. Chapter 6 Contradiction — Possible Solution ==================================================================== WHY IT LOOKS LIKE A CONTRADICTION AT FIRST ------------------------------ Chapter 4 showed that rewriting the unstable branch of the quadratic formula (avoiding a direct subtraction of two nearly-equal quantities) produced a dramatically more accurate result. This chapter shows that the subtraction operation a - b has its own fixed condition number, (|a|+|b|)/|a-b|, which is large whenever a and b are close - and conditioning, unlike stability, was explicitly described as something no algorithm can fix. Put those two claims side by side and it looks like Chapter 4 is claiming to have fixed something Chapter 6 says can't be fixed. WHY THEY DON'T ACTUALLY CONTRADICT ------------------------------ The key is that "the problem" being solved and "a specific sub-operation inside one way of solving it" are not the same thing. The quadratic formula's small root is the actual problem being solved - and that root has its own well-defined value regardless of how it's computed. The naive formula's version of solving that problem happens to route the calculation through a specific subtraction of two nearly-equal numbers (-b and sqrt(disc)) - and THAT SPECIFIC SUBTRACTION, as its own isolated operation, genuinely is ill-conditioned, exactly as this chapter describes. Chapter 4's fix did not make that particular subtraction any less sensitive - it restructured the overall calculation (using q = -0.5*(b+sign(b)*sq), then dividing to get the small root) so that the final answer is reached WITHOUT ever performing that ill-conditioned subtraction at all. THE GENERAL PRINCIPLE THAT RESOLVES IT ------------------------------ A single mathematical problem (like "find this root") can typically be computed via more than one sequence of intermediate operations, and different sequences can have completely different sub-operations with completely different condition numbers, even though they all arrive at the same final answer. Choosing a good algorithm means choosing a sequence of operations that avoids routing the computation through any individual ill-conditioned sub-operation, when a well-conditioned path to the same final answer exists. This is different from this chapter's near-singular linear-system example, where the ILL-CONDITIONING BELONGS TO THE FINAL ANSWER ITSELF (how sensitive the true intersection point is to the input coefficients), not to one arbitrary intermediate step that a smarter algorithm could route around. WHY THIS WORKS AS AN ANSWER ------------------------------ The explanation identifies the precise distinction that resolves the apparent contradiction (an ill-conditioned intermediate operation that a restructured algorithm can avoid, versus ill-conditioning that belongs to the final answer itself and cannot be routed around), and applies it correctly to both this chapter's own linear-system example and Chapter 4's quadratic-formula fix rather than treating the two as unrelated cases.