Lately, I've been reading a couple of papers from different one-dimensional PDE contexts on which operators like $\mathcal{L}:=-\partial_x^2+c_*+\Phi$ repeatedly appear. Usually, on these contexts $\Phi$ is a smooth exponentially decaying function and $c_*\in\mathbb{R}$ is a positive constant.

I am very surprised that in all of these papers the authors, just by knowing these facts, they immediately conclude that the continuous spectrum of $\mathcal{L}$ is exactly $[c_*,+\infty)$ and the rest of the spectrum consists on a finite number of eigenvalues. My question is, how do they know that the continuous spectrum starts exactly at $c_*$? I've seen this at least five times for different values of $c_*$ and different functions $\Phi$ (all of them smooth with exponential decay). Does anyone have an explanation for this? Or any reference?

A second question (I know that the difficulty of the question can exponentially grow now so I am actually very happy just by understanding the previous part!): What if I consider a non-smooth but still exponentially decaying $\Phi$? For instance $\Phi=e^{-\vert x\vert}$? The previous "result" still holds?

  • 2
    $\begingroup$ The 3D case is treated here around page 88. I can't remember whether they did the 1D case, or how it was proved. $\endgroup$ Commented Mar 25, 2020 at 3:00

1 Answer 1


Let $A$ be a self-adjoint operator with domain $D(A)\subset\mathcal H$ ($\mathcal H$ is some Hilbert space). An operator $C$ with $D(A)\subset D(C)$ is called relatively compact with respect to $C$ if $C(A-zI)^{-1}$ is compact for some (hence all) $z\notin\sigma(A)$. Paraphrasing Corollary 2, page 113 Section XIII.4, in [1], we have

If $C$ is relatively compact with respect to $A$, then $B:=A+C$ is closed with domain $D(B)=D(A)$, and $$\sigma_{ess}(B)=\sigma_{ess}(A).$$

In fact, this is elementary once one realises that $C(A-zI)^{-1}:\mathcal H\to \mathcal H$ is compact if and only if $C:D(A)\to\mathcal H$ is compact.

In your case, setting $A = -\partial_x^2 + c_*$ and $C$ the multiplication operator by $\Phi$, is it not hard to prove that $C:H^2(\mathbb R)\to L^2(\mathbb R)$ is compact. We then obtain that $$ \sigma_{\mathrm{ess}}(\mathcal{L}) = \sigma_{\mathrm{ess}}(-\partial_x^2 + c_*). $$ Since it is well know that $\sigma_{\mathrm{ess}}(-\partial_x^2 + c_*) = [c_*,+\infty)$, the result follows.

For your second question, this answer is yes the result hold for $\Phi(x) = e^{-|x|}$. In fact it will hold in dimension $d\leq 3$ for any $\Phi\in L^2(\mathbb R^d)$. As explained earlier, it is enough to show that multiplication by $\Phi$ is (well-defined and) compact from $H^2(\mathbb R^d)\to L^2(\mathbb R^d)$; let us prove this result.

Let $(f_n)_{n\geq0}$ be a bounded sequence in $H^2(\mathbb R^d)$, $d\leq 3$. The core of the argument is the following fact.

$(f_n)_{n\geq0}$ is bounded in $L^\infty$ and, up to extraction, converges $L^\infty$-locally to some function $f$.

Note that $f$ then has to be bounded as well. Once this is established, we see that $$ \begin{align*} \limsup |\Phi f_n-\Phi f|_{L^2}^2 & = \limsup \int|\Phi|^2\cdot|f_n-f|^2 \\\\ & \leq \limsup |f_n-f|_{L^\infty([-k,k]^d)}^2\cdot\int|\Phi|^2{\mathbf 1}_{[-k,k]^d} \\\\ & \quad + \limsup (|f_n|_{L^\infty}+|f|_{L^\infty})^2\cdot\int|\Phi|^2{\mathbf 1}_{([-k,k]^d)^\complement} \\\\ & \leq |\Phi|_{L^2}^2\cdot0 + 4M^2\cdot\limsup \int |\Phi|^2{\mathbf 1}_{([-k,k]^d)^\complement} \end{align*} $$ for $M$ a bound on the norms $|f_n|_{L^\infty}$, and $\limsup$ the limit superior along a convergent subsequence. Because $\Phi$ is in $L^2$, Lebesgue's dominated convergence theorem guarantees that the limit is zero, and $\Phi f_n$ converges to $\Phi f$ as expected.

Let us turn to the proof of the fact. The boundedness of $(f_n)_{n\geq0}$ on compact sets follows from the continuous embeddings from $H^2(\ell+[0,1]^d)$ to $\mathcal C^0(\ell+[0,1]^d)$ for all $\ell\in\mathbb Z^d$ (because $d\leq3$). Since the norm of these injections does not depend on $\ell$, $(f_n)_{n\geq0}$ is in fact uniformly bounded over $\mathbb R^d$

As for the convergence up to extraction, according to the usual Sobolev embeddings/inequalities, the sequence of restrictions $\big((f_n)_{|[-k,k]^d}\big)_{n\geq0}$ is relatively compact in $\mathcal C^0([-k,k]^d)$ for all $k\geq1$ (we use again that $d\leq3$). We can then extract diagonally a subsequence $(f_{\sigma(m)})_{m\geq0}$ that converges to some continuous function $f$ uniformly over the compact sets, concluding the proof of the fact.

[1] Reed, M., Simon, B. (1978). Methods of Modern Mathematical Physics. IV Analysis of Operators. New York: Academic Press.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.