0
$\begingroup$

How to find $C$ such that $CC^T = AA^T + BB^T$, $A$ and $B$ are known.

$A = \left(\begin{matrix}X\\Y\end{matrix}\right)$, $B = \left(\begin{matrix}0\\cY\end{matrix}\right)$, $c$ is a constant.

To clarify. Cholesky decomposition can give the solution. I was wondering is there any closed form solution?

$\endgroup$
  • 4
    $\begingroup$ Are you familiar with the Cholesky decomposition? $\endgroup$ – Patrick Sanan Oct 5 '15 at 19:57
  • $\begingroup$ @MarkLStone's answer below is deleted, but is the correct answer anyway. $\endgroup$ – Wolfgang Bangerth Oct 6 '15 at 12:00
5
$\begingroup$

If A and B are known, then the right hand side is known, and must of necessity be symmetric positive semi-definite. Therefore, there is a solution C to the stated equation, and this is the (lower triangular) Cholesky factor of $ AA^T + BB^T$ . Cholesky factorization is a standard routine in any linear algeabra library. MATLAB's chol computes the upper triangular Cholesky factor, so using that, C would be chol(A * A' + B * B')' .

I'll leave it to others to determine whether there is a "closed form" solution in terms of X, Y, and c, but nothing jumps out at me.

Edit: Sorry, did not mean for this to be deleted.

$\endgroup$
  • 1
    $\begingroup$ Forming that matrix squares the condition number. qr([[A B]) should be significantly more stable for ill-conditioned matrices. $\endgroup$ – Federico Poloni Sep 27 '17 at 19:00
3
$\begingroup$

Probably the Cholesky factor is what you are looking for, as noted by the other answers. However the rectangular matrix $C=[A,B]$ does satisfy $$CC^T=AA^T+BB^T$$ so is technically a closed form solution.

$\endgroup$
1
$\begingroup$

Substituting A and B into the equation gives you the following symmetric matrix:

$$CC^T=K=\left[\begin{matrix}XX^T & XY^T\\YX^T & (1+c^2)YY^T\end{matrix}\right]$$

To find $C$, you could use the following MATLAB commands:

C = chol(K,'lower')

Positive definiteness of $K$ should of course be ensured.

$\endgroup$
  • 1
    $\begingroup$ Forming that matrix squares the condition number. qr([[A B]) should be significantly more stable for ill-conditioned matrices. $\endgroup$ – Federico Poloni Sep 27 '17 at 18:59

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.