Take the vector space of all polynomial functions from $\Bbb R$ to $\Bbb R$ with an inner product $$\langle f,g \rangle = \int_{-\infty}^{\infty} f(x)g(x)\sigma(x)dx$$ where $\sigma$ is a positive, smooth function such that for all natural numbers $n$ $$\lim_{|x| \to \infty} \sigma(x)x^n=0.$$ (For example, $\sigma (x)$ could be $e^{-x^2}$.)

Producing a metric from this inner product in the usual way turns the space of polynomial functions into a metric space. I think I can show that this metric space is not complete. For instance, I believe the exponential function is a limit that does not exist in the space. What I am not certain of is how to go about finding the completion of this space.

So my question is: what is the completion of this space?

I have a feeling that it is something like the space of all functions which are square-integrable, do not go to infinity faster than $\sigma$ goes to zero, and are equal to their Taylor series on all of $\Bbb R$... but I don't know how to show that.


  • 1
    $\begingroup$ Are there any other conditions on $\sigma$, such as $\sigma > 0$? Otherwise, you could take $\sigma = 0$ and not even get an inner product. $\endgroup$ – Paul Sinclair Oct 19 '15 at 22:39
  • $\begingroup$ Ah, thanks. You are correct, $\sigma$ must be everywhere-positive. $\endgroup$ – Matt Dickau Oct 19 '15 at 22:41
  • $\begingroup$ It's pretty clear that any Taylor series with infinite radius of convergence is in the closure, But I think the closure is likely to be much bigger. It seems likely to me, though I haven't proved it, that the closure is all of $L^2(\Bbb R)$. $\endgroup$ – Paul Sinclair Oct 20 '15 at 2:17
  • $\begingroup$ @PaulSinclair : If $\sigma(x)=e^{-|x|}$, then $e^{2x}$ is not in closure, even though $e^{2x}$ has an infinite radius of convergence. $\endgroup$ – DisintegratingByParts Oct 20 '15 at 20:14
  • $\begingroup$ @TrialAndError - true. I misspoke, and the $L^2(\Bbb R)$ in my previous comment should be $L^2(\Bbb R, \sigma)$ as well. $\endgroup$ – Paul Sinclair Oct 20 '15 at 20:35

In the case of $e^{-x^{2}}$, you have an answer, but a general answer is non-trivial. In the case of $\sigma(x)=e^{-x^{2}}$, Hermite polynomials give you the answer: $$ H_n(x) = (-1)^{n}C_ne^{x^{2}}\frac{d^{n}}{dx^{n}}e^{-x^{2}} $$ You have $$ \int_{-\infty}^{\infty}H_{n}(x)H_{m}(x)e^{-x^{2}}dx =0,\;\;\; n \ne m. $$ By choosing positive constants $C_n$ appropriately, $\{ H_n \}_{n=0}^{\infty}$ becomes an orthonormal set with respect to the weighted inner product. For any $f \in L^{2}(\mathbb{R})$, it is known that $\{ h_n=e^{-x^{2}/2}H_{n}\}$ is an orthonormal basis of $L^{2}(\mathbb{R})$. Therefore, if $f \in L^{2}(\mathbb{R})$, then $$ \int_{-\infty}^{\infty}|f-\sum_{n=0}^{N}(f,h_n)h_n|^{2}dx =\int_{-\infty}^{\infty}|fe^{x^{2}/2}-\sum_{n}^{N}(f,h_n)H_n|^{2}e^{-x^{2}}dx\rightarrow 0 $$ as $N\rightarrow \infty$. So the completion of the space with weight $\sigma(x)=e^{-x^{2}}$ is $e^{x^{2}/2}L^{2}(\mathbb{R})$, which is as nice as you can reasonably expect. The completion $X$ of your space is isometrically isomorphic to $L^{2}$ under the map $U : L^{2}(\mathbb{R})\rightarrow X$ defined by $Uf=e^{x^{2}/2}f$. Proving the completeness of the Hermite functions $\{ h_n \}$ is not a simple matter, and, for an arbitrary weight $\sigma$, I don't think things turn out so nice. I wish I had a reference for you, but I don't. You might try searching "orthogonal polynomials" for references.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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