The spectral theorem can be explicitly expressed for an hermitian matrix by providing its eigen decomposition. In the more general case of a bounded self-adjoint operator with a continuous spectrum, the expression of the spectral theorem is $Ax=\int \lambda dF(\lambda) x$. This expression is rather abstract. Can someone provide a simple example of the continuous case, taking $x$ from a space of function, a differential operator A, and the detailed expressions of $F(\lambda)$ and $dF(\lambda)$ ?


Yes, let $X=L^{2}[0,1]$, and let $(Tx)(t)=tx(t)$, i.e., multiplication by $t$. This is the prototypical example in many regards. You don't have any eigenfunctions because $Tx=\lambda x$ would require $(t-\lambda)x(t)=0$ for a.e. $t\in[0,1]$ which means that $x=0$ a.e..

However, you have approximate eigenfunctions. For example, if $\lambda\in(0,1)$ and $\epsilon > 0$ is smaller than the distance of $\lambda$ to $\mathbb{R}\setminus[0,1]$, then $$ \varphi_{\epsilon} = \frac{1}{\sqrt{2\epsilon}}\chi_{[\lambda-\epsilon,\lambda+\epsilon]} $$ is a unit vector in $X$ (here $\chi_{S}$ means characteristic function of set $S$). And, $$ \|(T-\lambda I)\varphi_{\epsilon}\|^{2}=\frac{1}{2\epsilon}\int_{0}^{\epsilon}t^{2}\,dt = \frac{\epsilon^{2}}{6},\\ \|(T-\lambda I)\varphi_{\epsilon}\| = \frac{\epsilon}{\sqrt{6}}. $$ As $\epsilon\downarrow 0$, the function $\varphi_{\epsilon}$ looks more and more like an eigenvector with eigenvalue $\lambda$, but it does not converge to an actual eigenvector. Continuous spectrum looks like that.

In this case, the spectral measure $F$ is $$ F(S)x = \chi_{S}(t)x(t). $$ So, $F(\lambda)x=\chi_{[0,\lambda]}(t)x(t)$ and $F(\lambda+\delta)x-F(\lambda)x=\chi_{(\lambda,\lambda+\delta]}(t)x(t)$, which is related to the discussion of the previous paragraph.

  • $\begingroup$ Thank you, this is more detailed than the Wikipedia article, on the same example. $\endgroup$ – vkubicki Sep 20 '14 at 20:48
  • $\begingroup$ You're welcome. A building block for normal operators is multiplication by $z$ on some $L^{2}_{\mu}(S)$ where $S$ is a subset of $\mathbb{C}$ and $\mu$ is a finite Borel measure on $S$. You can glean most of what you need to know about the general case from looking at such an example. An atom $\lambda$ of $\mu$ is an eigenvalue with eigenvector $\chi_{\{\lambda\}}$. All other points $\lambda$ are in the continuous spectrum if $\lambda$ is in the support of $\mu$. Points outside the closure of $S$ are in the resolvent set, with $(M_{z}-\lambda I)^{-1}=M_{1/(z-\lambda)}$. $\endgroup$ – DisintegratingByParts Sep 21 '14 at 0:01
  • $\begingroup$ OK thank you ! I have a "soft" question connected to your answer: are Dirac distributions eigenvectors of the operator $T$, if you were to consider distributions instead of functions ? $\endgroup$ – vkubicki Sep 23 '14 at 5:38

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.