Suppose that is a linear map between inner product spaces. The adjoint $T^{}TT^{}:W\to V$ satisfying
To see that $T^{}w\phi_w:v\mapsto \langle w, Tv\rangle\phi_w\Hom(V,F)v’\in V\phi_w(v)=\langle v’,v\ranglev’=T^{}w$ by definition. The adjoint has the properties:
and
.
The identity is its own adjoint.
The spectral theorem (over ) applies to self-adjoint operators where is an inner product space. The singular value decomposition is a generalization to operators . So let and be real inner product spaces and a linear map.
Let . Then is a map from :
is self-adjoint.
for all . A self adjoint operator with this property is called positive.
The null space of is the same as the null space of .
The range (column space) of is the same as the column space of .
The ranks of , , and all coincide.
Proof:
The key point is that . So is greater than or equal to zero; and if it equals zero then and conversely. This proves that .
In general:
null space of is orthogonal to the range of .
range of is orthogonal to the null space of .
null space of is orthogonal to the range of .
range of is orthogonal to the nullspace of .
So since the null space of and the null space of coincide, and is self-adjoint, we have:
Also
Definition: If is a linear operator, then is diagonalizable. Let be diagonal the matrix of in an orthonomormal basis given by the spectral theorem, with eigenvalues listed in decreasing order. The singular values of are the entries in – the square roots of the eigenvalues of .
Proposition: (Singular value decomposition) Let be a linear map of inner product spaces with singular values . Then there are orthonormal basis and for and such that for all . In matrix terms this basically means that is “diagonal” in the bases given by the ’s and ’s (although needn’t be square).
This is usually written like this for matrices.
Theorem: (SVD) Let be an matrix over . Then there are orthogonal matrices and such that where is , is , and is . The “diagonal” entries of are the singular values of .
This is the matrix version of the previous statement.
Proof:
There’s an orthonormal basis of so that where the are the eigenvalues of the self-adjoint operator . For the nonzero eigenvectors for , the vectors are orthonormal. We can complete the set of (if needed) to a basis of . Any can be written The sought-after formula comes from applying to this.