Let $\gamma_n: \mathbb{R}^n\to\mathbb{R}$ be the Gaussian distribution function defined by $$ \gamma_n(x):=(2 \pi)^{-\frac{n}{2}} e^{-\frac{|x|^2}{2}}. $$ Let $d\gamma_n$ denote the following measure (weight) $\gamma_n(x) dx$ and consequently $L^2(\mathbb{R}^n,d\gamma_n)$ and $H^1(\mathbb{R}^n,d\gamma_n)$ be the weighted versions of the Sobolev spaces with respect to the measure $d\gamma_n$.

Does the following weighted version of the Rellich-Kondrachov Theorem hold?

$$ H^1(\mathbb{R}^n,d\gamma_n) \text{ is } \textbf{compactly} \text{ embedded in } L^2(\mathbb{R}^n,d\gamma_n). $$


1 Answer 1


I'll start with several known facts.

Proposition 1. Suppose that $E$ is a real Hilbert space with norm $\Vert -\Vert$ and $K: E\to E$ is a a compact, selfadjoint positive operator. Denote by $R(K)$ the range of $K$. Equip the range with the norm

$$ \Vert x\Vert_K:=\Vert K^{-1} x\Vert,\;\;x\in R(K). $$

Then the inclusion $(R(K), \Vert-\Vert_K)\to (E,\Vert-\Vert)$ is compact.


Proposition 2. Suppose that $E$ is a real Hilbert space and $K:E\to E$ is a positive selfadjoint operator such that for a sufficiently large positive integer $p$ the operator $K^p$ is trace class. Then the operator $K$ is compact.

Suppose now that $\newcommand{\bR}{\mathbb{R}}$ $\newcommand{\bZ}{\mathbb{Z}}$ that $E=L^2(\bR^n,\gamma_n)$. Then $E$ admits a natural orthogonal basis consisting of the Hermite polynomials

$$ H_\alpha(x_1,\dotsc, x_n)=H_{\alpha_1}(x_1)\dotsc H_{\alpha_n}(x_n), $$

$$\alpha=(\alpha_1,\dotsc,\alpha_n)\in\bZ^n_{\geq 0}. $$

Above, $H_k(x)$ denotes the degree $k$ Hermite polynomial in one variable $x$ defined by

$$ H_k(x) =\delta^k 1, $$

where $\delta :C^\infty(\bR)\to C^\infty(\bR)$ is the creation operator $$ \delta f(x)=-f'(x)+x f(x),\;\;\forall f\in C^\infty(\bR). $$ We have

$$\Vert H_\alpha\Vert^2 =\alpha!:=(\alpha_1!)\cdots (\alpha_n!), $$

where $\Vert-\Vert$ denotes the norm in $E=L^2(\bR^n,\gamma_n)$.

Denote by $E_m$ the subspace of $E$ spanned by the polynomials $H_\alpha$such that $|\alpha|=m$, where $|\alpha|=\alpha_1+\cdots +\alpha_n$. $\DeclareMathOperator{\Proj}{Proj}$ Note that

$$\dim E_m =\binom{m+n-1}{n-1} = O(m^{n-1})\;\;\mbox{as $m\to \infty$}. $$

Let $\Proj_m$ denote the orthogonal projection onto $E_m$. The norm $\Vert-\Vert_*$ on $H^1(\bR^n,\gamma_n)$ can be described in terms of the projectors $\Proj_m$. More precisely

$$\Vert f\Vert_*^2=\Vert f\Vert^2+\sum_{j=1}^n \Vert \partial_{x_j}\; f\Vert^2=\sum_{m\geq 0} (1+m)\Vert \Proj_m f\Vert^2 . $$

Now consider the operator

$$K : E\to E,\;\;K f= \sum_{m\geq 0}(1+m)^{-1/2} \Proj_m f. $$

This shows that $H^1(\bR^n,\gamma_n)$ can be identified with the range of $K$ equipped with the norm $\Vert -\Vert_K$ defined as in Proposition 1.

Clearly, for $p$ sufficiently large

$$\sum_{m\geq 0} (m+1)^{-p/2} \dim E_m <\infty. $$

This shows that $K^p$ is trace class for large $p$ and thus, by Proposition 2, $K$ is compact. Now conclude using Proposition 1.

  • $\begingroup$ Interesting! And all the more so because $p$-Schatten-class embeddings are actually much stronger than compact embeddings. $\endgroup$ Dec 15, 2015 at 23:01
  • $\begingroup$ I think the definition of the norm $\Vert - \Vert_*$ is missing. Maybe after "Then the norm on $H^1(\bR^n,\gamma_n)$ is equivalent with the norm..."? $\endgroup$ Dec 16, 2015 at 4:11
  • $\begingroup$ @LorenzoCavallina Thanks for pointing this out. $\endgroup$ Dec 16, 2015 at 10:17

Your Answer

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

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