Consider the Schroedinger operator

$$H = -\Delta + \frac{c}{\vert x \vert^2}$$ in two dimensions with $c >0$

This operator has a self-adjoint realization, since it is a positive symmetric operator on $C_c^{\infty}( \mathbb R^2 \setminus \{0\}).$ I would like to understand three things:

0.) What are possible domains $D(H)$ on which $H$ is self-adjoint?

1.) Does there exist a domain $D(H)$ of $H$ on which $-\Delta$ and $H$ are both self-adjoint.

2.) On these domains $D(H)$, can we expect $$\Vert \frac{f}{\vert x \vert^2}\Vert_{L^2} \le \alpha_1 \Vert f \Vert_{L^2} + \alpha_2 \Vert \Delta f \Vert_{L^2} \text{ for some }\alpha_i \ge 0?$$

I would appreciate insights on any of these questions. My main interest is in small $c>0.$

One way to think about it may be to follow Example 4 in Reed-Simon II on page 160.

There the authors prove, going via a 1D reduction, that if

$c \ge 1,$ then $H$ is essentially self-adjoint (there is a unique self-adjoint extension) on $C_c^{\infty}(\mathbb R^2 \setminus \{0\})$ and if

$c \in [ \frac{1}{4},1),$ then $H$ is not essentially self-adjoint on $C_c^{\infty}(\mathbb R^2 \setminus \{0\}).$

The case $c \in (0,\frac{1}{4})$ seems to be missing, but I think one quickly sees by analysing the argument that it is the same case as the last one above (not essentially self-adjoint).

In a separate question Iosif Pinelis has shown that the quadratic form generated by the potential is not relatively form bounded with respect to the Dirichlet form.

The reason I asked this question was that I suspected that $H$ was self-adjoint on the Dirichlet domain of $-\Delta$, i.e. $H_0^1(\mathbb R^2\setminus \{0\}) \cap H^2(\mathbb R^2\setminus \{0\})$, but I am now not so sure about my conjecture anymore.

  • $\begingroup$ This physics literature has a number of papers about finding the correct self-adjoint realization of this operator, e.g. starting from here: pubs.aip.org/aapt/ajp/article/42/11/960/1045509/… $\endgroup$
    – Buzz
    Feb 26 at 4:49
  • $\begingroup$ See also the more recent papers by Derezinski. Note that this operator is usually called the "Bessel operator" which might help when doing a literature search. The radial part is equivalent to the generator of a Bessel process which might also help with intuition. $\endgroup$ Feb 26 at 9:40

1 Answer 1


Taking advantage of the spherical symmetry to decompose this into a sum of one-dimensional problems sounds like the right approach. I will probably just be redoing what Reed-Simon had in mind here.

The Laplacian is given in polar coordinates by $$ \Delta=\frac{1}{r}\frac{\partial}{\partial r} r\frac{\partial}{\partial r} + \frac{1}{r^2}\frac{\partial^2}{\partial\theta^2} . $$ We identify $L^2(\mathbb R^2)=L^2((0,\infty); rdr)\otimes L^2(S^1)$ and decompose $L^2(S^1)=\bigoplus L(e^{in\theta})$ and also get rid of the weight in the radial component with the help of the unitary map $U:L^2(r\, dr)\to L^2(0,\infty)$, $(Uf)(r)=r^{1/2}f(r)$, then $H$ becomes (is unitarily equivalent to, to be precise) an orthogonal sum $H=\bigoplus H_n$, with each $H_n$ acting in $L^2(0,\infty)$ and $$ H_n f = -f''+\frac{n^2-1/4+c}{r^2}f . $$ This operator is essentially self-adjoint if and only if $n^2-1/4+c\ge 3/4$ or $c\ge 1-n^2$. See for example the corresponding discussion in my answer here.

So $H$ on $C_0^{\infty}(\mathbb R^2\setminus\{ 0\})$ itself is essentially self-adjoint if and only if this happens for all $n\in\mathbb Z$, which is indeed equivalent to $c\ge 1$. For $0<c<1$, $H_n$ is still essentially self-adjoint on $C_0^{\infty}(0,\infty)$ for $|n|\ge 1$, while the closure of $H_0$ on this domain is symmetric with deficiency index $(1,1)$. So in this case there is a one-parameter family of boundary conditions, to be imposed on the elements of $D(H_0^*)$, and this gives all self-adjoint realizations.

As for (1), we can just set $c=0$ in this analysis and everything is still correct, and we are now discussing the self-adjoint extensions of $-\Delta$ on $C_0^{\infty}(\mathbb R^2\setminus \{ 0\})$. Since there is nothing to choose in the higher harmonics, the question is if $-d^2/dr^2+k/r^2$ can ever be given the same domain as $-d^2/dr^2-1/(4r^2)$ for $-1/4<k<3/4$. The discussion here strongly suggests that the answer is no, though it's not completely conclusive since we fixed specific boundary conditions there.

Added later (with one detail corrected later still): This seems quite clear actually: When $c=0$, the possible domains $D(H_0)$ can be obtained by putting an arbitrary solution of $-f''-f/(4r^2)=0$ into $D(H_0)$. This is an Euler equation with solution basis $f_1=r^{1/2}$, $f_2=r^{1/2}\log r$. Note also that when we undo the unitary transformation from above, these correspond to $U^*f_j=1, \log r$. On the other hand, $f$ can never be in $D(H_0)$ when $c>0$ because then $-f''+(c-1/4)f/r^2= cf/r^2\notin L^2$. So we never have $D(H)=D(\Delta)$.

For the record ("what is the correction?"), in an earlier version I put $f_1=r^{1/2}$ into $D(H_0)$ and claimed that this corresponds to $D_0=H_0^1\cap H^2$, which you suggested as a possible domain. This was nonsense of course (I mixed up my signs in the exponent when applying $U^*$), since in that case we put $g=1$ (near $r=0$) into $D(\Delta)$ and we simply obtain the plain Laplacian on $H^2(\mathbb R^2)$. Any other choice, namely put $g=b+\log r$ into $D(\Delta)$, is certainly closer to what you had in mind, but it's not really the domain $D_0$. In fact, $H_0^1$ on $\mathbb R^2\setminus \{0\}$ is simply $H^1$ since the function $h(r)=\log a^2/\log r^2$ gets us from $h(0)=0$ to $h(a)=1$ and $\|h\|_{H^1(r<a)}\to 0$ as $a\to 0$.

Finally, (2) is now moot, but it would be a consequence of (1) since if $D(S)\supseteq D(T)$ for closed operators, then $S$ is $T$-bounded (= Theorem 5.9 in Weidmann, Linear operators in Hilbert spaces).

  • $\begingroup$ So you are saying the Dirichlet domain is indeed an admissible domain? If I read your answer correctly, then you are saying that we just have to know the self-adjoint extensions of the Laplacian and we will have a complete description for $c<1$? $\endgroup$ Feb 25 at 21:12
  • $\begingroup$ Thanks Christian, I think I still stumble over the same point: You say "we can just set 𝑐=0 in this analysis and everything is still correct". Are you suggesting that the conditions are in some sense independent of setting $c=0$ as long as $c<1$? This is not obvious to me. $\endgroup$ Feb 25 at 21:45
  • $\begingroup$ @AntónioBorgesSantos: No, I'm just saying that we can decompose $-\Delta=\bigoplus H_n$ and then extend this, and since the $H_n$, $n\not= 0$ are already essentially self-adjoint, it's only about finding a domain for $H_0$. This works for any $c\in\mathbb R$, including $c=0$. $\endgroup$ Feb 25 at 21:49
  • $\begingroup$ ah, but you think the domains will be different for $c \neq 0$ from the ones you can obtain for $c=0.$ So both (1) and (2) are wrong. $\endgroup$ Feb 25 at 21:50

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.