Let $H = -\partial^2_x + V(x) : L^2(\mathbb{R}) \to L^2(\mathbb{R})$ be a one dimensional Schrödinger operator, where the potential $V$ is real-valued, belongs to $L^\infty(\mathbb{R})$, and, as $|x| \to \infty$, decays as $|x|^{-2 - \delta}$ for some $\delta > 0$. Since $V \in L^\infty$, by the Kato-Rellich theorem, the operator $H$ is self-adjoint with respect to the domain the Sobolev space $H^2(\mathbb{R})$.

I would like to show that $0$ cannot be an eigenvalue of $H$, that the spectrum of $H$ on $(-\infty, 0)$ consists only of eigenvalues, and finally that there exists $\mu > 0$ sufficiently small so that $(-\mu, 0) \subseteq \rho(H)$.

These properties of $H$ are asserted but not proved in the article by Jensen and Nenciu, Rev. Math. Phys. 13(6) (2001). It seems that the above properties are quite classical, because I have consulted many of the prominent modern references (the books of Reed--Simon and Yafaev, articles by Jensen, Kato, Gesztesy, etc.), but have not found any concrete discussion about them. Hint, solutions, or pointers to relevant references are greatly appreciated. I would like to establish each of these properties without using too much abstract machinery, if possible.

Letting $H_0 = -\partial^2_x$, I think I have worked out that, due to the decay of $V$, $V(H_0 - \lambda^2)^{-1}: L^2(\mathbb{R}) \to L^2(\mathbb{R})$ is compact in $\text{Im} \, \lambda> 0$. We should also have $(H - \lambda^2)^{-1} = (H_0 - \lambda^2)^{-1}(I + V(H_0 - \lambda^2)^{-1})$ exists for $\text{Im} \, \lambda \gg 1$ and (by the analytic Fredholm theorem) is meromorphic in $\text{Im} \, \lambda> 0$. But this is as far as I have got toward proving any of the above properties.


1 Answer 1


As a general rule of thumb, it's usually most convenient in one-dimensional problems to work with solutions of the ODE $-y''+Vy=Ey$ rather than operator theoretic methods.

Here, everything follows from the behavior of the $E=0$ solutions. To analyze the spectrum below $E=0$, we can replace the actual potential by a smaller potential (by min-max). The borderline case for finite spectrum below $E=0$ is $V_0=-(1/4)|x|^{-2}$. This we can see by solving $-y''-cx^{-2}y=0$ explicitly (for $x\ge 1$, say), which is possible since this is an Euler equation. For $c<1/4$ the solutions have only finitely many zeros, and then the claim follows from oscillation theory.

Finally, $E=0$ cannot be an eigenvalue because a square integrable solution $y$ would also solve $$ y(x)=1+\int_x^{\infty} (t-x)V(t)y(t)\, dt , $$ but then $y\to 1$ as $x\to\infty$.

  • $\begingroup$ How you get $1$ (or any non-zero number) in the displayed formula? It looks like a circular argument.. but probably I missed something. $\endgroup$ Commented Dec 27, 2022 at 11:38
  • $\begingroup$ Thank you for your answer. I am also struggling to see how $1$ or any nonzero constant appears in the display. Here is my approach. If $y'' = -Vy$ is in $H^2$, then $y'$ and $\int_x^\infty Vydt$ have the same distributional derivative, thus $y' = c + \int_x^\infty Vydt$ for some $c$. Sending $x \to \pm \infty$ gives $c = 0 = -\int_{-\infty}^{\infty} Vydt$. Performing a similar "antidifferentiation" calculation gives $y = - \int_x^\infty (t -x)Vydt$ and I found $\int_{-\infty}^\infty tVydt = 0$. $\endgroup$
    – JZS
    Commented Dec 27, 2022 at 13:14
  • 1
    $\begingroup$ @GiorgioMetafune: You are both right, I was a bit sloppy. I didn't really think about this but just adapted the usual approach to discuss Jost solutions to $E=0$ (the $1$ on the RHS is what one needs to obtain the solution with $y=1$ for large $x$ for compactly supported $V$). But I think essentially the argument is fine: View the equation I wrote as an integral equation for $y$. This has a solution (by iteration) $y$, and $y\to 1$. Moreover, $y$ also solves $-y''+Vy=0$. A similar argument with the modified equation $y=x+ \ldots$ gives a second solution $y\simeq x$ (cont'd) $\endgroup$ Commented Dec 27, 2022 at 18:09
  • $\begingroup$ (cont'd) Obviously, no linear combination of these is square integrable, so $-y''+Vy=0$ doesn't have $L^2$ solutions. $\endgroup$ Commented Dec 27, 2022 at 18:09
  • 1
    $\begingroup$ @JZS: Yes, right again, nothing is very obvious here. The statement (in a more general version) is Lemma 3.1.4 in Marchenko, and it has a lengthy proof. The integral equation to consider is $y=x+\int_0^x \ldots$, this avoids the convergence issues you pointed out. $\endgroup$ Commented Dec 27, 2022 at 22:19

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.