Let's consider an infinite dimensional space $X$ and a linear operator $T$. The resolvent operator of $T$ is $R_\lambda (T) = (T-\lambda I)^{-1}$. A regular value $\lambda$ of $T$ is a complex number such that:

(R1) $R_\lambda(T)$ exists

(R2) $R_\lambda(T)$ is bounded

(R3) $R_\lambda(T)$ defined on a set which is dense in $X$

The resolvent set $\rho(T)$ consists of all regular values $\lambda$ of $T$. The complement $\sigma(T)=C-\rho(T)$ is the spectrum of $T$ and we may distinguish parts of the spectrum:

point spectrum (eigenvalues) $\sigma_p(T)$: (R1) isn't satisfied

continuous spectrum $\sigma_c(T)$: (R2) isn't satisfied, but (R1) and (R3) are satisfied

residual spectrum $\sigma_r(T)$: (R3) isn't satisfied, (R1) is satisfied, (R2) - doesn't matter

Please, help me to clarify a couple of points:

Question 1: The point spectrum consists of eigenvalues and exists in finite dimensional case. So its meaning seems to be the same as in a finite dimensional case (scaling of eigenvectors that roughly represent orientation of the distortion by $T$). What is the meaning of the continuous and the residual spectrum?

Question 2: Why do we care about dense in the definitions? I have found a related question but didn't get the exact answer from it.


Question 1: If $\lambda$ is in the continuous spectrum, then $R_{\lambda}(T)$ is unbounded. That means there exists a sequence $\{ x_n \}$ such that $\|x_n\|=1$ and $\|R_{\lambda}(T)x_n\|\rightarrow\infty$. So, $y_n=\frac{1}{\|R_{\lambda}(T)x_n\|}R_{\lambda}(T)x_n$ is a sequence of unit vectors in $X$ such that $(T-\lambda I)y_n\rightarrow 0$ as $n\rightarrow\infty$. It is typical to refer to this sequence $\{ y_n \}$ of unit vectors as an "approximate eigenvector."

The reason one refers to this as "continuous spectrum" Historically had nothing to do with continuity; such spectrum was found to fill a continuum, rather than being discrete. For example, the multiplication operator $(Mf)(x)=xf(x)$ defined on $L^2[0,1]$ has no eigenvalues, but $[0,1]$ is in the continuous spectrum of $M$. You can see this because the step function $f_{\lambda,\delta}=\frac{1}{\sqrt{2\delta}}\chi_{[\lambda-\delta,\lambda+\delta]}$ is a unit vector for which $$ Mf_{\lambda,\delta} \approx \lambda f_{\lambda,\delta}, $$ in the sense that $$ \|(M-\lambda I)f_{\lambda,\delta}\|^2=\int_{\lambda-\delta}^{\lambda+\delta}(x-\lambda)^2f_{\lambda,\delta}^2dx \le 2\delta^2\|f_{\lambda,\delta}\|^2. $$ So $\|(M-\lambda I)f_{\lambda,\delta}\|\rightarrow 0$ as $\delta\rightarrow 0$ while $\|f_{\lambda,\delta}\|=1$.

The same type of spectrum occurs for the differentiation operator $\frac{1}{i}\frac{d}{dx}$ on $L^2(\mathbb{R})$ where you would like to identify $e^{isx}$ as an eigenvector with eigenvalue $s$, but it's not in the space. However, $s$ is in the continuous spectrum, a.k.a. approximate point spectrum. This is one of the earliest cases that motivated these definitions, where you could think of $$ F(x)=\int_{-\infty}^{\infty}c(s)e^{isx}ds $$ as a "continuous" (i.e. integral) sum of approximate eigenfunctions of the differentiation operator, with a coefficient function $c(s)$ defined on the continuum. And, when considered in $L^2$, the operator $\frac{1}{i}\frac{d}{dx}$ has continuous spectrum $\mathbb{R}$.

Question 2: For your second question, you don't care about a dense domain for $R_{\lambda}(T)$ if it is bounded because you can then extended it to the whole space. And, if $T$ is closable, then the extension of $R_{\lambda}(T)$ and the closure of $T$ remain inverses of each other. However, if $\lambda$ is an approximate eigenvalue and $T$ is closed, then $R_{\lambda}(T)$ must be unbounded, and you're basically stuck with a dense domain for $R_{\lambda}(T)$ because of the closed graph theorem.

  • $\begingroup$ Thank you for the reply! Can you please explain why the continuous spectrum is defined on a dense set in $X$? Could you tell a couple of words about the meaning of the residual spectrum? $\endgroup$ – Konstantin Jan 8 '17 at 19:09
  • 1
    $\begingroup$ @Konstantin : The continuous spectrum requires that you have an inverse that is unbounded. If $X$ is a complete space, then the inverse cannot be defined on the full space. It is standard to require the inverse to be defined on a dense subspace. If it is defined on a non-dense subspace, that falls into the miscellaneous category of residual spectrum. Residual should be regarded as the "left over" part of the spectrum. $\endgroup$ – DisintegratingByParts Jan 8 '17 at 21:06

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.