Skip to content

Change-of-Basis Matrix

Artifact — instantiates Independent Generating Set Design

The concrete invertible matrix that rewrites any object's coordinates from one basis into another, letting two teams using different bases exchange the same object exactly.

Change-of-Basis Matrix is the reusable artifact — a stored matrix P and its inverse — that mechanically re-expresses any object's coordinates from one basis of a space into another basis of the same space. It is not a discovery method and not a test; it is the plumbing. Once built for a given pair of bases, it converts every object between them exactly and identically, which is what lets two parties holding the same space under different coordinate conventions trade vectors without ambiguity. This is the numeric counterpart to Dual-Basis Transform: where the dual transform is a conceptual move to a complementary representation for insight, the change-of-basis matrix is the concrete, general-purpose converter for any two bases.

Example

In color science, a pixel measured as sRGB (r, g, b) and the same color written in CIE XYZ are one physical color in two coordinate systems. The change-of-basis matrix is the 3×3 built from the display's primaries: XYZ = M · RGB, and its inverse M⁻¹ carries XYZ back to sRGB. A rendering pipeline stores M and M⁻¹ once and applies them to millions of pixels; because M is invertible, no color information is lost in either direction, and a value that round-trips through both matrices returns to itself. Two systems — a camera speaking sRGB and a print engine speaking XYZ — exchange exact colors through this single artifact.

How it works

Each vector of the new basis is written in the old coordinates and stacked as a column to form P; then a new coordinate vector is recovered as P⁻¹ times the old one (up to the row/column convention in force). Invertibility — full rank — is the contract that guarantees the conversion loses nothing and can be undone. For operators rather than vectors, the same P conjugates them as P⁻¹AP, a similarity transformation, so the matrix serves double duty: converting coordinates and re-expressing the transformations that act on them.

Tuning parameters

  • Convention — active vs. passive, which direction P maps, row- vs. column-vector layout. This must be pinned down explicitly; a wrong convention silently transposes or inverts everything downstream.
  • Precomputed inverse vs. solve-on-demand — store P⁻¹ for speed (risky if P is near-singular) or solve the linear system each time for numerical safety.
  • Numeric form — exact/rational vs. floating point, and whether to orthogonalize P for stability at some cost to exactness.
  • Target-basis normalization — scaling the new basis rescales the matrix's rows or columns and hence the magnitude of the resulting coordinates.

When it helps, and when it misleads

Its strength is interoperability: one artifact makes two representations of a space exchangeable, exactly and repeatably, so teams, tools, and file formats can each keep their preferred basis and still agree on every object.

Its failure modes are numerical and conventional. If the two bases are nearly aligned, P is ill-conditioned and P⁻¹ amplifies error — a matrix that is invertible in theory but corrosive in practice.[1] And because a coordinate list carries no label saying which basis or convention produced it, a mismatched convention silently corrupts everything that touches it. The classic misuse is applying a near-singular change of basis to make quantities look separable or decorrelated when they are not. The discipline is to store the convention and both bases with the matrix, and to run a conditioning check before trusting the inverse.

How it implements the components

Change-of-Basis Matrix realizes the conversion side of the archetype's machinery — the components that govern how coordinates move and are encoded:

  • basis_change_contract — it is the artifact that encodes the contract for how any object's coordinates transform between the two bases.
  • coordinate_encoding_contract — applying P (or P⁻¹) is exactly the operation of encoding an object's coordinates in the target basis.

It does not judge whether the target basis is numerically stable (that is Basis Conditioning and Perturbation Audit), verify that conversions round-trip faithfully on real cases (Coordinate Round-Trip Test), or reinterpret coordinates into a complementary meaning (Dual-Basis Transform).

Notes

The change-of-basis matrix between a basis and its dual is precisely the artifact a Dual-Basis Transform executes — the two mechanisms meet there, one supplying the concept, the other the numbers. Keeping the artifact separate from the transform is what lets you reuse the same matrix for purely mechanical conversions that carry no reinterpretation at all.

References

[1] Re-expressing an operator under a new basis as P⁻¹AP is a similarity transformation; it preserves eigenvalues and rank but not numerical conditioning, which is why an invertible change-of-basis can still be a dangerous one. ↩