6

Let $A:D(a)\to L^2(\mathbb{R}^n)$ be an elliptic partial differential operator $$ A(f)=\sum_{i,j=1}^{\infty}\partial_{x_j}(a_{ij}(x)\partial_{x_i}f) $$ where $a_{ij}\in C^{\infty}_b(\mathbb{R}^n)$, this means they are bounded continuously differentiable functions with bounded derivative of all orders. Assume that there is a $c>0$ such that for every $y=(y_1,\ldots,y_n)$ $$ \sum_{i,j=1}^{n}a_{ij}(x)y_iy_j>c|y|^2,\ \text{for every $x$.} $$ Assume also that $a_{ij}=a_{ji}$. The question is how to prove that, on the domain $D(A)=H^2(\mathbb{R}^n)$, the operator $A$ is selfadjoint.

user80839
  • 101
  • Integration by parts. – Shuhao Cao Jun 03 '13 at 23:01
  • 2
    If you content yourself with proving that the operator is symmetric, then you just need to use integration by parts as Shuhao Cao suggests. If you need to prove that the operator is self-adjoint you may try using the basic criterion of self-adjointness, which you may find here, Proposition 3. – Giuseppe Negro Jun 03 '13 at 23:24
  • I know that symmetry follows by integration by parts as Shuhao Cao suggests but the problem is to show that it is self-adjoint. I will try to see if I can show the basic criterium for selfadjointness thank you very much. – user80839 Jun 04 '13 at 14:40

1 Answers1

5

We recall definitions first. Let $H$ be a Hilbert space. A linear operator $A\colon D(A)\subset H\to H$ is said to be symmetric if \begin{equation} (Af, g)=(f, Ag),\qquad \forall f, g\in D(A), \end{equation} and it is said to be self-adjoint if \begin{equation} D(A)=D(A^\star)\quad \text{and}\quad (Af, g)=(f, Ag),\quad \forall f,g\in D(A). \end{equation} Here $D(A^\star)$ is the biggest domain on which the adjoint operator $A^\star$ can be defined, that is \begin{equation} D(A^\star)=\left\{g\in H\ |\ \text{the map }f\in D(A)\mapsto (Af, g)\ \text{is a continuous linear functional}\right\}\end{equation} A self-adjoint operator is necessarily closed, meaning that its graph $\{(f, Af)\in D(A)\times H\}$ is a closed subset of $H\times H$. This is equivalent to the fact that $D(A)$ is a Banach space when equipped with the graph norm $$\lVert f\rVert_{G(A)}=\lVert f\rVert+\lVert Af\rVert,\qquad f\in D(A).$$

To prove that a symmetric operator is self-adjoint one can use the following criterion, named basic criterion of self-adjointess on Reed & Simon's Methods of Modern Mathematical Physics, volume I, Theorem VIII.3.

Theorem. Let $A$ be symmetric. Then the following three statements are equivalent:

  1. $A$ is self-adjoint;
  2. $A$ is closed and $\ker(A^\star\pm i)=\{0\}$;
  3. $A\pm i$ are both surjective.

We will show that the operator $A$ given in the question is self-adjoint by proving that it satisfies property 2. First of all we note that $A$ is closed, because from elliptic regularity theory (see for example Evans's book on PDE, 2nd edition, §6.3.5) we have for the graph norm $$\lVert f\rVert_{G(A)}=\lVert f\rVert_{L^2}+\lVert Af\rVert_{L^2}\ge C\lVert f\rVert_{H^2},\qquad \forall f\in D(A),$$ so a $\lVert\cdot \rVert_{G(A)}$-Cauchy sequence is a $\lVert \cdot \rVert_{H^2}$-Cauchy sequence also and $D(A)$ is a Banach space with the graph norm. We now turn to the equation \begin{equation}\tag{1} (A^\star + i)f=0. \end{equation} We claim that any $f\in D(A^\star)$ satisfying \eqref{1} is necessarily null. Indeed, again by regularity theory we can observe that $f\in C^\infty$. So \begin{equation} 0=(A^\star+i)f=(A+i)f, \end{equation} and because of symmetry we have $\lVert (A+i)f\rVert_{L^2}^2=\lVert Af\rVert_{L^2}^2+\lVert f\rVert_{L^2}^2$, from which we conclude that $f\equiv 0$. We have thus proved that $A$ satisfies all properties of the basic criterion of self-adjointness.

  • Dear Giuseppe Negro, thank you very much for your solution, however there is something I don't understand yet. Evans's estimates work on bounded domains, however the operator is on $H^2(\mathbb{R}^n)$. Are those estimates valid in that context? – user80839 Jun 04 '13 at 22:00
  • Yes, the estimates you see on Evans's book are valid on the whole space too. See here for some references. The proof should be much easier because you do not have to worry about the boundary, but I haven't performed the computation yet. – Giuseppe Negro Jun 05 '13 at 00:12
  • @GiuseppeNegro are (uniformly) parabolic Differential operators closed also? – ABIM Feb 08 '18 at 15:03
  • @CSA: You should specify a domain, to ask about closedness. Anyway, I don't know, but I would be surprised if they were not closed on their "natural" domains (whatever they might be). – Giuseppe Negro Feb 08 '18 at 15:13
  • Do you know of a good reference, I was looking in the Evans one your provided for the elliptic PDEs case. – ABIM Feb 08 '18 at 15:20
  • https://www.cambridge.org/core/books/spectral-theory-and-differential-operators/93D1D33A1395B4BA34C81CF615E21EF6 – Giuseppe Negro Feb 08 '18 at 15:22