1
$\begingroup$

Suppose I have a matrix $A_{p \times q}$ and let $I_p$ and $I_q$ be the identity matrices of dimension $p \times p$ and $q\times q$, respectively. I would like to show that $\det(I_p-AA^T) = \det(I_q - A^TA)$. I have thought about taking transposes, but that doesn't work. Is there something I'm missing here?

$\endgroup$
3
$\begingroup$

Let the SVD decomposition of $A=UDV^T$, where $U$ and $V$ are orthogonal.

$$AA^T=UDD^TU^T$$ $$I-AA^T=U(I_p-DD^T)U^T$$

Hence $$\det(I-AA^T)=\det(I_p-DD^T)$$

Similarly, we have

$$A^TA=VD^TDV^T$$

$$I-A^TA=V(I-D^TD)V^T$$

Hence $$\det(I-A^TA)=\det(I_q-D^TD)$$

We can see that $$\det(I_p-DD^T)=\det(I_q-D^TD)$$

Edit to explain the equality above:

For the case where $p \leq q$, we can write $D= \begin{bmatrix} \hat{D} & 0\end{bmatrix}$, where $\hat{D} \in \mathbb{R}^{p \times p}.$

$$DD^T=\begin{bmatrix} \hat{D} & 0\end{bmatrix}\begin{bmatrix} \hat{D} \\ 0\end{bmatrix}=\hat{D}^2$$

$$I_p-DD^T=I_p-\hat{D}^2$$

$$D^TD=\begin{bmatrix} \hat{D} \\ 0\end{bmatrix}\begin{bmatrix} \hat{D} & 0\end{bmatrix}=\begin{bmatrix} \hat{D}^2 & 0 \\ 0 & 0\end{bmatrix}$$

$$I_q-D^TD=\begin{bmatrix} I_p-\hat{D}^2 & 0 \\ 0 & I_{q-p}\end{bmatrix}$$

$$\det(I_q-D^TD)=\det(I_{q-p})\det(I_p-\hat{D}^2)=\det(I_p-\hat{D}^2)=\det(I-DD^T)$$

$\endgroup$
  • $\begingroup$ Thanks, can I ask why $\det(I_p-DD^T)=\det(I_q-D^TD)$? I am guessing there is something related to the singular values, but why? $\endgroup$ – user136503 Dec 19 '16 at 4:52
  • $\begingroup$ Is it because since $D$ is NOT a square matrix, yet is still diagonal, the only way this happens is if $DD^T$ or $D^TD$ has an extra row of $0$'s, and so $I_p-DD^T$ and $I_q-D^TD$ differ by one having an extra row with $1$ at the end, but that doesn't matter since determinant of a diagonal matrix is the product and so we multiply by $1$? Is there a formal way to show this? $\endgroup$ – user136503 Dec 19 '16 at 5:02
  • $\begingroup$ I added some explaination for the case where $p\leq q$, the case of $p>q$ is similar. $I_p-DD^T$ and $I_q-D^TD$ differs by a block of identity matrix. along the diagonal. $\endgroup$ – Siong Thye Goh Dec 19 '16 at 7:13
4
$\begingroup$

Consider $\mathbf{M} = \begin{bmatrix} \mathbf{I}_p & \mathbf{A} \\ \mathbf{A}^\top & \mathbf{I}_q \end{bmatrix}$. Applying the Schur determinant formula on $\mathbf{M}$ for the top-left and bottom-right blocks results in:

$\det(\mathbf{M}) = \det(\mathbf{I}_p - \mathbf{A}\mathbf{A}^\top) = \det(\mathbf{I}_q - \mathbf{A}^\top\mathbf{A})$.

$\endgroup$

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.