2
$\begingroup$

Let $H$ be a Hilbert space. For a bounded linear operator $T:H\to H$ we write $T\geq 0$ to mean that $T$ is self-adjoint and that $\langle Tx, x\rangle \geq 0$ for all $x\in H$. For two bounded linear operators $S$ and $T$ we will write $S\geq T$ to mean $S-T\geq 0$.

Suppose $T_1, T_2, T_3, \ldots$ is a sequence of self-adjoint operators on $H$ such that

1) $T_{n+1}\geq T_n$ for all $n$.

2) There is a self adjoint operator $T'$ such that $T'\geq T_n$ for all $n$.

Problem. Show that there is a self adjoint operator $T$ such that $T_nx\to Tx$ (convergence in norm) for all $x\in H$.

It is clear that $\langle T_nx, x\rangle$ is a bounded and increasing sequence of real numbers and hence has a limit. Therefore so does $\langle T_n(x+y), x+y\rangle$. From here we can conclude that $\langle T_nx, y\rangle$ is a Cauchy sequence for each $x$ and $y$ in $H$.

But what I really need to show is that $\{T_nx\}$ is a Cauchy sequence and I am unable to do that.

$\endgroup$

1 Answer 1

2
$\begingroup$

The sequence $(T_{n}x,x)$ is convergent: $(T'x,x)\geq(T_{n+1}x,x)\geq(T_{n}x,x)$ for each $n=1,2,...$

Then sequence $(T_{n}x,y)$ is also convergent by means of polarization.

We can define $(Tx,y)=\lim_{n}(T_{n}x,y)$. It is easy to see that $T$ is self-adjoint because of those $T_{n}$.

On the other hand, by Uniform Boundedness Principle, one can show that $\|T_{n}\|\leq C$ and hence $T$ is also bounded.

Indeed, consider the sesquilinear form $U(x,y)=(T_{n}x,y)$, then we have \begin{align*} \|T_{n}x\|^{2}&=U(x,T_{n}x)\\ &\leq U(x,x)^{1/2}U(T_{n}x,T_{n}x)^{1/2}\\ &=(T_{n}x,x)^{1/2}(T_{n}^{2}x,T_{n}x)^{1/2}\\ &\leq(T_{n}x,x)^{1/2}(T'(T_{n}x),T_{n}x)^{1/2}\\ &\leq(T_{n}x,x)^{1/2}\|T'\|^{1/2}\|T_{n}x\|, \end{align*} so $\|T_{n}x\|\leq(T_{n}x,x)^{1/2}\|T'\|^{1/2}\leq(T'x,x)^{1/2}\|T'\|^{1/2}$, so for each $x$, $\|T_{n}x\|$ is bounded, now we apply Uniform Boundedness Principle.

Now consider the sesquilinear form $S(x,y)=((T-T_{n})(x),y)$.

We have \begin{align*} \|(T-T_{n})x\|^{2}&=S(x,(T-T_{n})x)\\ &\leq S(x,x)^{1/2}S((T-T_{n})x,(T-T_{n})(x))^{1/2}\\ &=((T-T_{n})(x),x)^{1/2}((T-T_{n}x)^{2},(T-T_{n})(x))\\ &\leq((T-T_{n})(x),x)^{1/2}\|(T-T_{n})(x)\|^{1/4}\|(T-T_{n})^{2}(x)\|^{1/4}\\ &\leq C'((T-T_{n})(x),x)^{1/2}\|x\|^{1/2}\\ &\rightarrow 0. \end{align*}

Edit:

The boundedness of $\|T_{n}\|$ can be proved without appealing to $T'$.

Indeed, the sequence $(T_{n}x,y)$ is bounded by applying polarization to the boundedness of $(T_{n}x,x)$.

Now we consider $S_{n}(y)=(y,T_{n}x)$, for each fixed $x$, Uniform Boundedness Theorem gives $\|S_{n}\|\leq M_{x}$, then $|(T_{n}x,y)|\leq M_{x}$ for any $y$ with $\|y\|\leq 1$. We set $y=T_{n}x/\|T_{n}x\|$ to get $\|T_{n}x\|\leq M_{x}$, once again Uniform Boundedness Theorem gives $\sup_{n}\|T_{n}\|<\infty$.

$\endgroup$
4
  • $\begingroup$ I am unable to see the first inequality in the last part. If we write $L$ in place of $T-T_n$, then we have $S(x, y) = (Lx, y)$. So the inequality becomes $\|Lx\|^2 \leq \sqrt{(Lx, x)}\sqrt{(L^2x, Lx)}$. I am unable to see why this should be the case. Can you please help. Thanks. $\endgroup$ Nov 2, 2019 at 16:32
  • 1
    $\begingroup$ The point is to use the Cauchy-Shwarz to the sesquilinear form: $S(x,y)\leq S(x,x)^{1/2}S(y,y)^{1/2}$. $\endgroup$
    – user284331
    Nov 2, 2019 at 16:39
  • $\begingroup$ When using the Cauchy-Schwarz for $U$, do we need to assume that $T_n$ is positive semi-definite, or just the self-adjointness of $T_n$ suffices? I ask this because $\langle T_nx, x\rangle$ might be negative and therefore its square root will cause problems. $\endgroup$ Nov 5, 2019 at 10:47
  • $\begingroup$ Positivity is needed, I almost forgot, but luckily in this case they are. $\endgroup$
    – user284331
    Nov 5, 2019 at 14:29

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.