What are the differences with regards to properties of $\mathbf{A}^T\mathbf{A}$ and $\mathbf{A}\mathbf{A}^T$? For example, the former has those mentioned here. Is the latter not positive (semi)definite as well? Will there always be eigenvalues for both of the matrix products?
-
4Identical objects do not have different properties ... – Hagen von Eitzen Sep 24 '20 at 19:37
-
Oh gosh! Typo fixed, thanks :) – That Guy Sep 24 '20 at 19:44
-
2In terms of the properties you mention, you should think of $B=A^T$, then $AA^T=B^T B$ and so they have the same properties. They also have joint properties (same eigen values, trace and determinant among others). – P. Quinton Sep 24 '20 at 19:45
-
@P.Quinton but isn't this a counter example: Let $\mathbf{A} = \left[\begin{array}{cc} 2&1 \end{array}\right]$. $\mathbf{A}\mathbf{A}^T$ does not have any eigenvalues(it is a scalar) but $\mathbf{A}^t\mathbf{A}$ is a matrix with eigenvalues 5 and 0? – That Guy Sep 24 '20 at 20:06
-
@P.Quinton Can you or anyone please help me with the question in the comment above?` – That Guy Sep 25 '20 at 00:34
-
A scalar is a Matrix, it's eigen value is it's value. – P. Quinton Sep 25 '20 at 06:04
1 Answers
Here are some properties:
Property 1. $AA^T$ and $A^TA$ are both symmetric.
Property 2. $\mathrm{rank}(A)=\mathrm{rank}(AA^T)=\mathrm{rank}(A^TA)$
If $Ax=0$ then $A^TAx=0$. If $A^TAx=0$, then $x^TA^TAx=0$, i.e. $||Ax||^2=0\Rightarrow Ax=0$. Thus $\mathrm{nullity}(A)=\mathrm{nullity}(A^TA)$. Interchanging $A$ and $A^T$, we have $\mathrm{nullity}(A)=\mathrm{nullity}(A^T)=\mathrm{nullity}(AA^T)$.
Property 3. $AA^T$ and $A^TA$ have the same non-zero eigenvalues.
If $\lambda$ is an eigenvalue of $A^TA$, then for some $v$, $A^TAv=\lambda v\Rightarrow AA^T(Av)=\lambda(Av)$. For the converse exchange $A^T$ and $A$.
We also see that $$\mathcal{C}(A^TA)\stackrel{(1)}{=}\mathcal{R}(A^TA)\stackrel{(2)}{=}\mathcal{R}(A)\ \text{and}\ \mathcal{R}(AA^T)\stackrel{(1')}{=}\mathcal{C}(AA^T)\stackrel{(2')}{=}\mathcal{C}(A),$$ where $(1)$ and $(1')$ follows from Property 1 (For a symmetric matrix $M$, $\mathcal{C}(M)=\mathcal{R}(M^T)=\mathcal{R}(M)$)
and $(2)$ and $(2')$ follows from the Property 2 (For any two matrix $M$ and $N$, $\mathcal{R}(MN)\subseteq\mathcal{R}(N)$ and $\mathcal{C}(MN)\subseteq\mathcal{C}(M)$. If $W$ is a subspace of $V$ and $\mathrm{dim}(W)=\mathrm{dim}(V)$, then $WW=V$).

- 12,644