1
$\begingroup$

I'm a student and I'm studying linear algebra. in Polar Decomposition we have:

for a linear operator $T$, there exist a linear isometry $S$ that: $$ T =S\sqrt{T^*T}$$

so if $S$ is a linear transformation then it must be $\dim(\operatorname{range}(T)) \leq \dim(\operatorname{range}(\sqrt{T^*T}))$.

But why?

edit: I think it must be equal, I mean:

$$\dim(\operatorname{range}(T)) = \dim(\operatorname{range}(\sqrt{T^*T}))$$

I know that $\dim(\operatorname{range}(T)) = \dim(\operatorname{range}(T^*))$ but because of the square root I cannot prove that $\dim(\operatorname{range}(T)) = \dim(\operatorname{range}(\sqrt{T^*T}))$.

$\endgroup$
  • $\begingroup$ If $f\colon U\to V$ and $g\colon V\to W$ are linear maps, then the rank of $g\circ f$ cannot be greater than the ranks of $g$ and $f$. $\endgroup$ – egreg Jan 6 at 17:24
1
$\begingroup$

As @egreg pointed out, for any linear maps $S,T$, the rank of $ST$ is always less than or equal to that of $T$. To see this, note that by dimension theorem $$\dim \text{ran} L =\dim \operatorname{dom} L-\dim\ker L\le\dim \operatorname{dom} L$$ for any linear map $L$. Now, the image of $ST$ can be seen as the image of $$ S\big|_{\text{ran} T} :\text{ran} T\to V, $$ we can see that $\dim \text{ran} (ST)\le \dim \text{ran} T$ as wanted. Moreover, equality holds if $\dim \ker S\big|_{\text{ran}T}=\dim [\ker S\cap\text{ran}T]=0$. (Here, $V$ denotes the vector space where $S,T$ are defined.)

To see that $\dim \text{ran}T =\dim \text{ran}(T^*T)$, note that $Tx = 0$ if and only if $T^*Tx =0$, which is saying that $\ker T = \ker (T^*T)$. Now, by dimension theorem, we have $$ \dim \text{ran} T = \dim V - \dim \ker T = \dim V - \dim \ker (T^*T) = \dim\text{ran}(T^*T). $$ (Or we can use the fact that $S$ is an isometry.)

$\endgroup$
1
$\begingroup$

If $T\colon U\to V$ and $S\colon V\to W$ are linear maps between finite dimensional vector spaces, then $$\DeclareMathOperator{\range}{range} \dim\range{ST}\le\dim\range(T) $$ This follows from the rank-nullity theorem: \begin{align} \dim U&=\dim\range(T)+\dim\ker(T) \\ \dim U&=\dim\range(ST)+\dim\ker(ST) \end{align} Therefore $$ \dim\range(ST)=\dim\range(T)+\dim\ker(T)-\dim\ker(ST)\le\dim\range(T) $$ because from $\ker(T)\subseteq\ker(ST)$ we have $\dim\ker(T)\le\dim\ker(ST)$.

In your case you can conclude that $$ \dim\range(T)\le\dim\range(\sqrt{T^*T}) $$ On the other hand, $S$ is an isometry, so it is invertible and $$ \sqrt{T^*T}=S^{-1}T $$ The same argument as before implies $$ \dim\range(\sqrt{T^*T})\le\dim\range(T) $$

$\endgroup$

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.