QR Decomposition¶
An exact matrix factorization into an orthogonal or unitary coordinate factor and an upper-triangular or trapezoidal coefficient factor.
Core Idea¶
QR decomposition writes a matrix as \(A=QR\), separating an orthogonal or unitary coordinate frame \(Q\) from an upper-triangular coefficient factor \(R\). For a real tall matrix, a full form uses square orthogonal \(Q\) and a zero-padded triangular factor; an economy form keeps only the needed orthonormal columns. Multiplication reconstructs \(A\) exactly in algebra.[^lapack]
The identity is not one algorithm. Gram–Schmidt and Householder methods can construct QR with different numerical properties. A rank-deficient matrix can still have QR, but the familiar nonsingular backsolve and positive-diagonal uniqueness claims require explicit full-rank assumptions.[ref-3e8165d848c5][mit]
Scope of Application¶
In full-column-rank least squares, \(Q\) preserves residual norm and \(R\) yields a triangular solve. In eigenvalue computation, QR-based iterations on Hessenberg matrices are steps toward Schur form. These are distinct uses of the factorization, not competing definitions of it. Complex matrices use unitary rather than real orthogonal geometry.[lapack][ref-985161512489]
Clarity¶
QR makes the difference between a column-space frame and coordinates within that frame explicit. It also separates “the matrix has a QR representation” from “a particular least-squares solution is unique.” The latter needs rank, and a whole QR eigenvalue algorithm needs iteration beyond one product identity.[lapack][ref-985161512489]
Manages Complexity¶
The dense matrix becomes two structured factors: norm-preserving \(Q\) and ordered triangular \(R\). This turns a qualifying least-squares problem into coordinate separation plus back-substitution. Yet numerical construction still matters: Householder QR is more reliable than classical Gram–Schmidt in the cited Cornell comparison, and the normal-equation alternative squares the condition number of \(A\).[lapack][ref-3e8165d848c5]
Abstract Reasoning¶
Because orthogonal \(Q\) preserves the 2-norm, minimizing \(\lVert Ax-b\rVert_2\) can be recast in \(Q\) coordinates. When \(A\) has full column rank, the fitted component gives an invertible triangular equation in \(R\); the complementary coordinates describe the residual. If rank fails, do not infer that same unique backsolve merely from the presence of QR.[^lapack]
Knowledge Transfer¶
The same orthogonal–triangular roles transfer literally from least-squares matrices to matrices encountered in eigenvalue iteration, while the downstream objective changes. More broadly, QR instantiates live prime Decomposition's recoverable-whole pattern. Live Factorization is a neighbor, but its same-type and irreducible-factor commitments do not strictly describe every rectangular QR.
[^lapack]: E. Anderson et al., LAPACK Users' Guide, 3rd ed. (1999), QR Factorization.
[^ref-985161512489]: E. Anderson et al., LAPACK Users' Guide, Eigenvalue Problems.
[^ref-3e8165d848c5]: Cornell University CS 4220/5223 numerical methods review, §5.3, PDF p. 26.
[^mit]: MIT 18.700 Lesson Plan 19, Theorem 10.
[^ref-9cb942c6d756]: Encyclopedia of Abstractions, live prime_abstractions/v2/decomposition.md, inspected 2026-10-01.
[^ref-9cb942c6d756-2]: Encyclopedia of Abstractions, live prime_abstractions/v2/factorization.md, inspected 2026-10-01.
Relationships to Other Abstractions¶
Current abstraction QR Decomposition Domain-specific
Parents (1) — more general patterns this builds on
-
QR Decomposition is a kind of Decomposition Prime
QR decomposes a matrix into exactly recombinable orthogonal/unitary and triangular factors.
Hierarchy path (1) — routes to 1 parentless root
- QR Decomposition → Decomposition
Neighborhood in Abstraction Space¶
QR Decomposition sits in a sparse region of the domain-specific corpus (60th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.
Family — Codes, Matrices & Combinatorial Problems (30 abstractions)
Nearest neighbors
- Matrix — 0.87
- Schur decomposition — 0.85
- LU Decomposition — 0.85
- Tensor — 0.85
- Hermite normal form — 0.84
Computed from structural-signature embeddings · 2026-10-08