For a block diagonal matrix, we have an identity for its cholesky decomposition i.e.

$chol(Z) = chol(blockdiag(A,B,...)) = blockdiag(chol(A),chol(B),...)$

(Here, $Z = blockdiag(A,B,...)$)

I want to know whether there any such similar identities for SVD of a block diagonal matrix.



Yep. If $A = U_A D_A V_A$ and $B = U_B D_B V_B$ with the $U$ and $V$ matrices unitary and the $D$ matrices diagonal, then it is easy to check that $$\operatorname{blockdiag}(A, B) = U \operatorname{blockdiag}(D_A,D_B) V$$ with $U = \operatorname{blockdiag}(U_A,U_B)$ and $V = \operatorname{blockdiag}(V_A,V_B)$ unitary.

Notice that the block SVD does not generally have its singular values in descending order, as is commonly the convention.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.