8
$\begingroup$

Let $H$ be a Hilbert space, and $T:D(T)\subset H\rightarrow H$ and $S:D(S)\subset H\rightarrow H$ be unbounded self-adjoint operators.

Is $T+S:D(T)\cap D(S)\rightarrow H$ self-adjoint?

$\endgroup$
1
  • 9
    $\begingroup$ Usually, $T+S$ is not even densely defined. $\endgroup$ Commented Feb 26, 2013 at 14:48

4 Answers 4

8
$\begingroup$

The answer is no. $\newcommand{\bR}{\mathbb{R}}$ Consider the Hilbert space of $L^2$-functions

$$ u:[0,1]\to \bR^2,\;\;u(t)=(x(t), y(t)) . $$

Define

$$ D(T)=\Bigl\lbrace u\in H;\;\int_0^1 \bigl\vert u'(t)\bigr\vert^2 dt <\infty,\;;\;x(0)=x(1)=0\;\Bigr\rbrace, $$

$$ D(S)=\Bigl\lbrace u\in H;\;\int_0^1 \bigl|u'(t)\bigr|^2 dt <\infty,\;;\;y(0)=y(1)=0\;\Bigr\rbrace. $$

Denote by $J:\bR^2\to\bR^2$ the linear operator $(x,y)\mapsto (-y,x)$. For $u\in D(T)$ defines

$$ Tu=J\frac{du}{dt}, $$

while for $ u\in D(S)$ define

$$ Su=J\frac{du}{dt}. $$

Both operators $S$ and $T$ are selfadjoint. Note that

$$ D(T)\cap D(S)= \Bigl\lbrace u\in H;\;\int_0^1 \bigl|u'(t)\bigr|^2 dt <\infty,\;\;u(0)=u(1)=0\,\Bigr\rbrace. $$

The operator $A:=T+S:D(A)= D(T)\cap D(S)\to H$ is symmetric, but not selfadjoint. Indeed, $ v\in D(A^*)$ if and only if, there exists $C>0$ such that

$$ \bigl\vert\;(Au, v)_{H} \;\bigr\vert \leq C\Vert u\Vert_H,\;\;\forall u\in D(A). $$

The constant function $t\mapsto v(t)= (1,1)$ satisfies this condition because for any $u\in D(A)$ we have

$$ (Au, v)_H=\int_0^1 \bigl(\; 2Ju'(t), v(t)\;\bigr) dt = 2\int_0^1\frac{d}{dt} \bigl(\; Ju(t), v(t)\;\bigr) =0. $$

We have produced a function $v\in D(A^*)\setminus D(A)$.

$\endgroup$
6
  • $\begingroup$ you may want to fix the boundary conditions in the domains and possibly also the operators, which are currently the same. $\endgroup$ Commented Feb 26, 2013 at 18:49
  • $\begingroup$ The boundary conditions are different. The philosophy of this example goes bak about 10 years to work of H. Weyl. Thr selfadjoint extensions of a symmetric operator are classified by the lagrangian subspaces of a certain symplectic vector space associated to this operator. The symmetric extensions correspond to the isotropic subspaces of this symplectic space. I chose the boundary conditions following this principle. The operators I've just described play an important role in symplectic Floer theory. $\endgroup$ Commented Feb 27, 2013 at 10:30
  • $\begingroup$ 100 years perhaps? $\endgroup$ Commented Feb 27, 2013 at 11:31
  • $\begingroup$ Yep!I meant 100 years. It was really early when I wrote that. L $\endgroup$ Commented Feb 27, 2013 at 12:18
  • $\begingroup$ I agree with your idea. I was just saying that there was a typo in your formulae - and there still is. What does it mean $x(0)=x(1)=0$, since you are introducing a class of functions $u$? $\endgroup$ Commented Feb 27, 2013 at 17:55
6
$\begingroup$

Let $T:D(T)\to H$ be a self-adjoint unbounded operator.

Take $S=-T$ with domain $D(S)=D(T)$.

Then $S$ is a self-adjoint unbounded operator but the sum $T+S$ is the null operator $0:D(T)\to H$, which is not self-adjoint if $D(T)\neq H$.

$\endgroup$
1
  • $\begingroup$ Really clear simple counterexample. $\endgroup$ Commented Apr 17, 2017 at 11:35
5
$\begingroup$

This is one of the central problems of modern mathematical physics. One could fill libraries with the mathematics generated by the most important special case---the Schrödinger operator. Mathematically, this is the question of when the sum of the Laplace operator and a multiplication operator on $L^2$ is self-adjoint (more precisely, essentially self-adjoint). The writings of Barry Simon, many of which are available online, are a good place to start.

$\endgroup$
1
$\begingroup$

In general if T,S, and T+S are densely defined linear operator on a Hilbert space then

$$ T^* + S^* \subset (T+S)^* $$

If S is bounded then

$$ T^* + S^* = (T + \bar{S})^* $$

So if S is bounded then $ T + \bar{S} $ is self adjoint. Of course there could be more weaker conditions for them to be self adjoint.

$\endgroup$

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .