Skip to content

svd_intro: add the Courant-Fischer min-max characterisation as a route to PCA #1017

Description

@mmcky

svd_intro.md currently derives principal components three ways — as an application of the SVD (Application: Principal Components Analysis (PCA)), through the explicit Relationship of PCA to SVD, and again in PCA with Eigenvalues and Eigenvectors. What none of them gives is the variational characterisation: that the principal components are the successive maximisers of the Rayleigh quotient, which is what the Courant–Fischer min-max theorem states.

That framing answers a question the current derivations leave open — not just what the principal components are, but why they are optimal, and in what precise sense the k-th component is the best remaining direction.

This was requested in QuantEcon/meta#28, Tom Sargent's brief on eigenvalues and eigenfunctions, which asked to "use principal components analysis as an application of the min-max theorem". That issue has been closed against the material that did ship; this is one of the residuals it did not cover, spun out so it is not lost.

Suggested shape: a section after PCA with Eigenvalues and Eigenvectors stating the Rayleigh quotient, giving the Courant–Fischer characterisation for a symmetric matrix, and connecting it back to the components already computed earlier in the lecture — ideally reusing the same worked example so the reader sees the two routes agree.

The original brief also links the Wikipedia min-max theorem article as a starting reference.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions