Let $H$ Hilbert and $A\colon D(A) \subset H \to H$ close (but unbounded) linear operator with dense domain. Assume moreover that it is maximally dissipative, i.e. $\langle Ax,x\rangle \leq 0$ for all $x \in D(A)$ and $Range(A-\lambda_0I)=H$ for some $\lambda_0 >0$.
Denote by $A^*$ the adjoint of $A$.

Then in [Fabbri, Gozzi, Swiech. "Stochastic optimal control in infinite dimension." Probability and Stochastic Modelling. Springer (2017).] p.175 it is claimed that the linear operator $$B=(I+AA^*)^{-1/2}$$ is well defined on all $H$ as a linear bounded operator.

I want to show it: if I prove that $I+AA^*$ is a strictly positive operator and moreover that it is also surjective, then I can take the inverse $(I+AA^*)^{-1}$ and define $B$ which is well defined at this point. Then I will work the details for the boundedness.

In order to prove $I+AA^*$ is a strictly positive operator it I should prove that $AA^*$ is a positive semidefinite operator (which I think should hold but don't know how to do it), how do you show it? How do you show $I+AA^*$ is surjective?


If $A : \mathcal{D}(A)\subseteq\mathcal{H}\rightarrow\mathcal{H}$ is closed and densely-defined on a Hilbert space $\mathcal{H}$, then the graph $\mathcal{G}(A)\subseteq \mathcal{H}\times\mathcal{H}$ is a closed linear subspace of $\mathcal{H}\times\mathcal{H}$, which gives $$ \mathcal{G}(A)\oplus\mathcal{G}(A)^{\perp}=\mathcal{H}\times\mathcal{H} \;\;\; (\dagger) $$ where $\mathcal{G}(A)^{\perp}$ is the orthogonal complement of $\mathcal{G}(A)$ in $\mathcal{H}\times\mathcal{H}$. Note that $(x,y)\in\mathcal{G}(A)^{\perp}$ iff $$ \langle z,x\rangle+\langle Az,y\rangle=0,\;\;\; z\in\mathcal{D}(A). $$ This condition is restated as $\langle Az,y\rangle = \langle z,-x\rangle$ holding for all $z\in\mathcal{D}(A)$, which is equivalent to having $y\in\mathcal{D}(A^*)$ with $A^*y=-x$. As a consequence, every $(a,b)\in\mathcal{H}\times\mathcal{H}$ may be uniquely written as the following some unique $x\in\mathcal{D}(A)$, $y\in\mathcal{D}(A^{*})$. $$ (x,Ax)+(-A^*y,y)=(a,b) $$ Setting $b=0$ gives $x-A^*y=a$ and $Ax+y=0$, or $x+A^*Ax=a$ for some $x\in\mathcal{D}(A^*A)$. That is enough to prove that $I+A^*A$ is surjective and symmetric. $I+A^*A$ is also injective because $$ \langle (I+A^*A)x,x\rangle=\|x\|^2+\|Ax\|^2,\;\;\; x\in\mathcal{D}(A^*A). $$ From this it follows that $I+A^*A$ self-adjoint, with a continuous inverse. So $(I+A^*A)^{-1/2}$ can be defined as the self-adjoint square root of the bounded self-adjoint operator $(I+A^*A)^{-1}$.

  • $\begingroup$ Almost all clear: but why the inverse of $I+AA^*$ is continuous? $\endgroup$
    – carlos85
    Oct 25 '21 at 14:37
  • $\begingroup$ because $ \lVert x \rVert \leq \lVert (I +AA^\star) x \rVert $ $\endgroup$ Oct 26 '21 at 13:59

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.