- categories: Linear algebra, Observation
An matrix of rank has free parameters. This count arises from the properties of the Singular Value Decomposition (SVD).
Singular Value Decomposition (SVD)
An matrix of rank can be expressed as:
where:
- is an orthogonal matrix containing the left singular vectors.
- is an orthogonal matrix containing the right singular vectors.
- is a diagonal matrix with non-zero singular values .
The non-zero singular values in determine the rank of .
Free Parameters
-
Singular Value Matrix :
- is (diagonal), and its entries are the non-zero singular values.
- The singular values are free parameters: free parameters.
-
Matrix (Left Singular Vectors):
- is and orthogonal. However, only the first columns of (denoted as ) contribute to .
- spans an -dimensional subspace of . To parameterize this subspace:
- requires parameters to define column vectors.
- However, must be orthonormal, introducing constraints to ensure orthogonality.
- Net free parameters for :
-
Matrix (Right Singular Vectors):
- Similarly, only the first columns of (denoted as ) contribute to .
- requires parameters to define column vectors.
- Like , orthonormality introduces constraints.
- Net free parameters for :
Total Free Parameters
Summing the contributions:
- : parameters.
- : parameters.
- : parameters.
Total:
Simplify:
Conclusion
The number of free parameters in an matrix of rank is:
This count is a balance between the degrees of freedom in , , and , and the orthonormal constraints.