MÉTHODOS · the way · numerical methods: how a machine computes real answers to continuous problems, one controlled approximation at a time · kept by HERON (Hero of Alexandria, the ancient numerical algorithmist)

THE LU DECOMPOSITION ◧ 2D · ◍ 3D · ◆ 4D · ◐ shadow · 👶 TAP

[[the-gaussian-elimination|Gaussian elimination]] solves one system — but if you have to solve with the SAME matrix again and again (new right-hand sides), why redo the work? LU factors the matrix ONCE into a Lower-triangular times an Upper-triangular piece; then each new solve is two cheap triangular sweeps. Factor once, solve forever. It’s how real solvers handle many loads on one structure. Slide to factor and read off L and U.

◆ LIT▲ AMBER
◧ THE MEASURE · 2D
◍ FACTOR ONCE, SOLVE MANY · 3D · the reusable form of a matrix
◆ THE FOURTH · 4D · a tesseract turns
◐ THE SHADOW · one dimension down
👶 THE TODDLER CORNER — one fat tap
L·U = A?
new solve cost
O(n²)
factor cost
O(n³)
L·U
= A

◆ LIT — exact / checkable

LU decomposition factors a matrix A=LU into a unit-LOWER-triangular L (the elimination multipliers) and an UPPER-triangular U (the reduced matrix) — the bookkeeping of [[the-gaussian-elimination|Gaussian elimination]] made reusable. Factoring costs O(n³) ONCE; then each solve Ax=b is a forward sweep Ly=b and a back sweep Ux=y, only O(n²). So many right-hand sides on the same A (and the determinant ∏uᵢᵢ, and the inverse) come cheap. Partial pivoting gives PA=LU for stability. It is the standard dense direct solver (LAPACK). A fail-loud self-check throws unless L·U reproduces A. ◆ real numerical methods, node-verified.

▲ AMBER — the figure

Needs pivoting (PA=LU) for stability, and dense O(n³) factoring is costly for very large sparse systems (sparse/iterative methods win). The L·U=A factorization and reuse economics are exact.

MÉTHODOS: the continuous is never solved, only APPROACHED — the art is bounding the error.  — HERON
David Lee Wise / ROOT0, with AVAN · MÉTHODOS — kept by HERON, the reckoner