I'm following the proof on Yosida's Functional Analysis of the spectral theorem for (unbounded) self-adjoint operators, and on the first implication (that, given a spectral resolution $(E_\lambda)_{\lambda \in \mathbb{R}}$ on a Hilbert space $\mathcal{H}$, the operator $A = \int_{-\infty}^{\infty} f(\lambda) dE_\lambda$ defined on

$$\mathcal{D} = \{ \psi \in \mathcal{H} \ |\ \int_{-\infty}^{\infty} |f(\lambda)|^2 d\langle \psi, E_\lambda \psi \rangle < \infty \} \ $$

is self-adjoint) he argues that the operator $A$ defined above is densely defined on $\mathcal{D}$ because this subspace contains every $\varphi \in \mathcal{H}$ such that $\varphi = E((-N, N])\varphi$ for some $0 < N < \infty$, where $E((-N, N]) = E_N - E_{-N^-}$, but I fail to see first why this is true and second, why that would imply that $A$ is densely defined because of it.


Because the approach seems to be a classical approach using he Riemann-Stieltjes integral, then it would follow that the author is assuming $f$ is bounded on finite intervals. And, if $M$ is a bound for $f$ on $[-N,N]$, $$ \int_{-\infty}^{\infty} |f(\lambda)|^2 d\|E_{\lambda}E(-N,N]x\|^2 \\ \le M^2\int_{-N+0}^{N}d\|E_{\lambda}x\|^2 \le M^2\|x\|^2 $$ The set $\{E(-N,N]x : N > 0, \; x\in\mathcal{H} \}$ is a dense subspace of $\mathcal{H}$ because $\lim_{N\rightarrow\infty}E(-N,N]x=x$.

  • $\begingroup$ I forgot to mention that $f$ is a continuous function. Well, its clear why now. I'm just not sure about the integration limits on the second inequality, because I think $E(-N, N] = E_N - E_{-N^-}$, as resolutions of identity are defined as strong left-continuous... $\endgroup$ – B. Chinaski Aug 19 '18 at 4:35
  • $\begingroup$ If you're not sure about the integration limits, err on the side of the largest quantity, and everything is still okay. $\endgroup$ – DisintegratingByParts Aug 19 '18 at 4:41

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.