-1
$\begingroup$

Let $(\mathscr{H}, \langle \cdot, \cdot \rangle)$ be a separable complex Hilbert space, and let $\mathscr{D}$ be a dense subset of $\mathscr{H}$. Let $P: \mathscr{D} \to \mathscr{H}$ and $N: \mathscr{D} \to \mathscr{H}$ be positive and negative (unbounded) operators, respectively, meaning that $$ \langle Pv,v \rangle \geq 0 \quad \text{and} \quad \langle Nv,v \rangle \leq 0 $$ for all $v \in \mathscr{D}$. Moreover, assume that $P$ and $N$ commute, and $P \neq -N$.

Consider the operator sum $X := P + N$. If $\mathscr{D}$ decomposes as a finite sum of eigenspaces for $X$, is it true that $\mathscr{D}$ also decomposes into eigenspaces for $P$ and $N$, respectively?

$\endgroup$
3
  • 1
    $\begingroup$ What if $N = -P$? $\endgroup$
    – David Gao
    Commented Oct 19 at 19:11
  • $\begingroup$ You're absolutely right about that. However, I would like to exclude that case. $\endgroup$
    – d'Alembert
    Commented Oct 19 at 19:19
  • 4
    $\begingroup$ As indicated by Christian Remling's answer, there are clearly an infinite number of cases of similar flavour to the extreme example that @DavidGao has pointed out. I recommend that you go back to the example(s) that you are interested in attacking and work out what extra conditions they satisfy (and then perhaps edit this question or ask a new one) $\endgroup$
    – Yemon Choi
    Commented Oct 19 at 20:05

1 Answer 1

3
$\begingroup$

No, this isn't working at all.

Since $N,P$ commute (you would have to be more specific what exactly you mean by this since the operators are unbounded, but let's just assume we have the right version), they are functions of a third operator. So passing to a spectral representation, we may assume that $N$ is multiplication by $f(x)\le 0$ and $P$ is multiplication by $g(x)\ge 0$ in $L^2(\mathbb R, \rho)$ (really in a sum of such spaces, but let's ignore multiplicity).

Now the assumption is that $f(x)+g(x)$ takes only finitely many values, and we would like to deduce that $f,g$ are also of this type, but clearly this is completely hopeless.

$\endgroup$

You must log in to answer this question.

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