I am working on Aharonov-Bohm operators for my thesis, and one of the articles given to me as a reference (here it is) tells me (intequality (2.1) on p. 6 in the link) that the magnetic Sobolev space defined with the AB operator is continuously embedded into $H_0^1$, that is:

$$\|u\|_{H_0^1(\Omega)}\leq C\|(i\nabla+A_a)u\|_{L^2(\Omega)}.$$

(Notes: $A_a$ is $\alpha(-\frac{x_2-a_2}{|x-a|^2},\frac{x_1-a_1}{|x-a|^2})$, $\Omega$ is a bounded simply connected domain containing $a$, $u$ is a generic function in the magnetic Sobolev space, and the RHS of the above inequality is taken as the norm w.r.t. which we complete $\mathcal C_0^\infty(\Omega\smallsetminus\{a\})$ to get the magnetic Sobolev space, denoted $\mathcal D^{1,2}_{A_a}(\Omega)$.)

It seems relatively easy to prove that there is a continuous embedding into $L^2$, seen as, being $\Omega$ bounded, so is $|x-a|$, so that $\frac{1}{|x-a|}$ is bounded away from zero, and we have inequality (2.2) (which holds also for $D=\Omega$, provided $u$ has zero trace, and holds for the $D$'s stated in the article for general $u$) which allows us to complete the following inequality chain:

$$\|u\|_{L^2}^2\leq\sup|x-a|^2\left\|\frac{u}{|x-a|}\right\|_{L^2}^2\leq C^2\sup|x-a|^2\|u\|_{\mathcal D^{1,2}_{A_a}(\Omega)}^2.$$

However, the Sobolev norm is either:

$$\|u\|_{H_0^1}=\sqrt{\|u\|_{L^2}^2+\|\nabla u\|_{L^2}^2},$$

or, equivalently (since the trace is zero and by Poincaré inequality):

$$\|u\|_{H_0^1}=\|\nabla u\|_{L^2}.$$

So I have the gradient to deal with. What I need to prove is therefore that for some $K$ we have:

$$\|\nabla u\|_{L^2(\Omega)}\leq K\|u\|_{\mathcal D^{1,2}_{A_a}(\Omega)}.$$

I tried working on the RHS without a constant and making lower estimates, but I only managed to show non-negativity, which is pretty useless. Here is what I did:

\begin{align*} \|(i\nabla+A_a)u\|_{L^2}^2={}&\int_\Omega|i\nabla u+A_au|^2dx=\int_\Omega(|\nabla u|^2+|A_au|^2+i(\nabla u;A_a)\overline u-i(A_a;\nabla u)u)dx={} \\ {}={}&\int_\Omega(|\nabla u|^2+|A_au|^2+2\operatorname{Im}[(A_a;\nabla u)u])dx, \end{align*}


$$(v;w)=v^T\overline w,$$

the Euclidean inner product (with conjugation on the second entry rather than the first one). So I would have to estimate that imaginary part, hopefully to eliminate the $|A_au|^2$ and only partially cancel out the $|\nabla u|^2$. The only estimate I was able to come up with for this is Cauchy-Schwarz plus Young, giving me:

$$\left|2\operatorname{Im}[(A_a;\nabla u)u]\right|\leq2|A_a||\nabla u||u|\leq|\nabla u|^2+|A_a|^2|u|^2=|\nabla u|^2+|A_au|^2,$$

but to use that I would have to say twice that imaginary part is bigger than the opposite of that right-hand side, which happens to cancel out everything and leave me with:


which I already knew to be true. So what can I do to prove my desired inequality? Is there a useful estimate for that imaginary part? Did I make any mistake in the above?


Thinking about it, I realized that, if $u$ is real-valued (as I guess the article assumes), the imaginary part in question is zero, so I'm already done. So is there a way to generalize this estimate to complex-valued $u$s? Is it even true in that generality?


1 Answer 1


The following argument is basically given in the proof of Lemma 6.2. It does not directly estimate that Imaginary part, but it shows the continuous embedding.

\begin{align*} \|u\|_{H_0^1(\Omega)}={}&\|\nabla u\|_{L^2(\Omega)}=\|i\nabla u\|_{L^2(\Omega)}\leq\|(i\nabla+A_a)u\|_{L^2(\Omega)}+\|A_au\|_{L^2(\Omega)}={} \\ {}={}&\|u\|_{\mathcal D^{1,2}_{A_a}(\Omega)}+\||A_a||u|\|_{L^2(\Omega)}=\|u\|_{\mathcal D^{1,2}_{A_a}(\Omega)}+\left\|\frac{u}{|x-a|}\right\|_{L^2(\Omega)}\leq{} \\ {}\leq{}&\left(1+C\right)\|u\|_{\mathcal D^{1,2}_{A_a}(\Omega)}, \end{align*}

where the $C$ is the constant of an inequality proved here. The passage from $A_a$ to $\frac{1}{|x-a|}$ is because:

\begin{align*} |A_a|={}&\sqrt{\frac{(x_1-a_1)^2}{((x_2-a_2)^2+(x_2-a_2)^2)^2}+\frac{(x_1-a_1)^2}{((x_1-a_1)^2+(x_2-a_2)^2)^2}}={} \\ {}={}&\sqrt{\frac{1}{(x_1-a_1)^2+(x_2-a_2)^2}}=\frac{1}{|x-a|}. \end{align*}

So here I go: the embedding of the magnetic Sobolev space into the regular Sobolev space is provedly continuous.


You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .