3
$\begingroup$

Consider two densely defined, strictly positive, self-adjoint operators $A$ and $B$ with the following property $$\|A^k x\| \simeq \|B^k x\|, \quad\forall x \in D(A^k)= D(B^k),$$ for $k=1,2,\cdots, M$, where $M$ can be $\infty$.

Do we have for fixed $t>0$, $$\|T(t) x \| \simeq \|S(t) x\|, $$ where $T(t),S(t)$ are the semigroups generated by $A,B$?

Does it hold only when $M = \infty$? Or does it hold under some extra assumptions?

Notation: $a \simeq b$ means there exist a positive constant $C$ such that $\frac{1}{C}a \leq b \leq C a.$

$\endgroup$
1
$\begingroup$

The condition $\|A^k x\| \simeq \|B^k x\|$ can be rewritten, by squaring both sides, as $\langle x, A^{2k} x \rangle \simeq \langle x, B^{2k} x \rangle$, i.e. $A^{2k} \leqslant C B^{2k}$ and $B^{2k} \leqslant C A^{2k}$. Recall now that the function $s \mapsto s^{\alpha}$ is operator monotone for $\alpha \in (0,1]$. It means that if you fix $t$, then you should take any $k \geqslant t$ and consider the function $s \mapsto s^{\frac{t}{k}}$, to conclude that $A^{2t} \leqslant C^{\frac{t}{k}} B^{2t}$ and $ B^{2t} \leqslant C^{\frac{t}{k}} A^{2t}$, from which it follows that $\|T(t)x\|\simeq \|S(t) x\|$. In this answer I assumed that in this setting by semigroups $T(t)$ and $S(t)$ you meant $A^{t}$ and $B^{t}$.

EDIT: It also holds for $T(t)=\exp(tA)$ and $S(t)=\exp(tB)$. First of all, note that by our norm condition the sets of analytic vectors for $A$ and $B$ are the same, so we can work on this common analytic domain and disregard all the issues coming from unboudedness. If $x$ is analytic then $\|T(t)x\|^2 = \sum_{k=0}^{\infty} \frac{(2t)^{k}}{k!} \langle x, A^{k} x\rangle$ and $\|S(t)x\|^2 = \sum_{k=0}^{\infty} \frac{(2t)^{k}}{k!} \langle x, B^{k} x\rangle$, so the same computation as in the first version gives $\|S(t)x\| \simeq \|T(t)x\|$.

EDIT #2: It turns out that if there exists a constant $C$ such that $\frac{1}{C} \|A^{k} x\| \leqslant \|B^{k} x\| \leqslant C \|A^{k}x\|$ and $D(A^k)=D(B^k)$ for any $k$ then $A=B$. As remarked above, both operators have the same analytics vectors and they form a core, so it suffices to work with them. We can therefore write $\frac{1}{C} \langle x, A^{2k} x\rangle \leqslant \langle x, B^{2k} x\rangle \leqslant C \langle x, A^{2k} \rangle$, i.e. $\frac{1}{C} A^{2k} \leqslant B^{2k} \leqslant C A^{2k}$. Since the function $s \mapsto s^{\frac{1}{2k}}$ is operator monotone, we get $\frac{1}{C^{\frac{1}{2k}}} A \leqslant B \leqslant C^{\frac{1}{2k}} A$, i.e. $\frac{1}{C^{\frac{1}{2k}}}\langle x, A x\rangle \leqslant \langle x, Bx\rangle \leqslant C^{\frac{1}{2k}}\langle x, Ax\rangle$ for any analytic vector $x$. Note that the proof of operator monotonicity of $s \mapsto s^{\frac{1}{2k}}$ follows from an appropriate integral representation of this function, so our analyticity assumption shows that we can perform it in our situation. Letting $k\to \infty$ we get $\langle x, Ax\rangle= \langle x, Bx\rangle$ for any analytic vector $x$. By polarisation and density of analytic vectors we may conclude that $Ax=Bx$ for any analytic vector and therefore $A=B$, since the analytic domain forms a core.

$\endgroup$
  • 2
    $\begingroup$ I think that the semigroups are rather meant to be $\exp(tA)$, i.e.\ $A$ and $B$ are the infinitesimal generators. $\endgroup$ – Jochen Wengenroth Mar 6 '17 at 13:18
  • 1
    $\begingroup$ @JochenWengenroth: It may well be the case, so I also added the proof in this case. First I assumed the other version because of the positivity assumption on $A$ and $B$. $\endgroup$ – Mateusz Wasilewski Mar 6 '17 at 16:02
  • $\begingroup$ Thanks for your answer.In the exponential case, do we require $M = \infty$? Since we are taking infinite sum. $\endgroup$ – newbie Mar 6 '17 at 22:31
  • $\begingroup$ @newbie: Yes, I think so. $\endgroup$ – Mateusz Wasilewski Mar 7 '17 at 5:57
  • $\begingroup$ What if there are only few analytic vectors? $\endgroup$ – András Bátkai Mar 7 '17 at 13:08

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.