anthropocene
Computational Physics / Numerical Reality
◆ core20 mindraft

Conditioning versus stability

Two different questions that get confused constantly: is the problem sensitive, or is my algorithm sloppy? One you can fix, the other you cannot.

assumes you already have

Planned outline

  1. The condition number of a problem, and why it is a property of the question rather than the method.
  2. Backward stability: a stable algorithm gives the exact answer to a nearby question.
  3. Worked case: solving a nearly singular 2x2 system, where no algorithm can rescue you.
  4. Worked case: the quadratic formula, where the naive root is unstable and the rearranged one is not.
  5. Rule of thumb: relative error ≤ condition number x machine epsilon, for a backward-stable algorithm.

Interactives planned

Each numbered point above gets at least one thing you can manipulate, not a figure. See AGENTS.md for the widget vocabulary and the rule that every lesson ships at least one graded interaction and one recall card.