Prove that

$$\lim_{k \to \infty} \int_{-\infty}^{\infty} \frac{ke^{-k^2x^2}}{\pi}h(x) \, dx= h(0) \tag{A}$$


From the left side of (A)

$$ \frac{k}{\pi}\lim_{k \to \infty} \int_{-\infty}^{\infty} e^{-k^2x^2}h(x)\, dx$$

Integration by parts?

$$ \frac{k}{\pi}\lim_{k \to \infty} \int_{-\infty}^{\infty} e^{-k^2x^2}h(x)\, dx = \frac{k}{\pi}\lim_{k \to \infty} \left[ \left( \int e^{-k^2x^2} \right)h(x)\bigg|_{-\infty}^{\infty} + \int_{-\infty}^{\infty} \left( \int e^{-k^2x^2} \right)h'(x)\, dx \right]$$

But here there's no antiderivative for the Gaussian-type function, unless I can apply those limits of integration to it, but I don't see how. The definite integral of that piece would be $\sqrt{\pi}/{k}. $

Then we'd have

$$ \frac{k}{\pi}\lim_{k \to \infty} \int_{-\infty}^{\infty} e^{-k^2x^2}h(x)\, dx = \frac{k}{\pi}\lim_{k \to \infty} \left[ \frac{\sqrt{\pi}}{{k}}h(x)\bigg|_{-\infty}^{\infty} + \int_{-\infty}^{\infty} \left( \int e^{-k^2x^2} \right)h'(x)\, dx \right]$$

Don't know where to go from here. The actual question is to prove that the kernel in the integrand in (A) is Dirac Delta sequence.

  • $\begingroup$ What are the assumptions on $h$? Just integrable? Or continuous as well? $\endgroup$ – M10687 Mar 8 '18 at 18:51
  • 2
    $\begingroup$ You cannot factor the dummy parameter $k$ from the limit. $\endgroup$ – Mark Viola Mar 8 '18 at 18:53
  • $\begingroup$ Oh yep, mistake. $\endgroup$ – Zduff Mar 8 '18 at 18:54
  • $\begingroup$ This is not correct as written. Take $h(x)= c$ then the limit is $\frac{c}{\sqrt{\pi}} = \frac{h(0)}{\sqrt{\pi}}$. Change the denominator to $\sqrt{\pi}$ and you are good $\endgroup$ – Winther Mar 8 '18 at 19:03

If $h\in C^{0}(-\infty,\infty)\cap L^{\infty}(-\infty,\infty)$, then \begin{align*} \dfrac{1}{\sqrt{\pi}}\int_{-\infty}^{\infty}e^{-u^{2}}(h(u/k)-h(0))du\rightarrow 0,~~~~k\rightarrow\infty \end{align*} because $h(u/k)\rightarrow h(0)$ for every $u$ as $k\rightarrow\infty$ and $|h(u/k)-h(0)|\leq 2\|u\|_{L^{\infty}(-\infty,\infty)}$ and $e^{-u^{2}}\in L^{1}(-\infty,\infty)$, so Lebesgue Dominated Convergence Theorem.

Now use the fact that $\displaystyle\int_{-\infty}^{\infty}e^{-u^{2}}du=\sqrt{\pi}$, we have \begin{align*} \dfrac{1}{\sqrt{\pi}}\int_{-\infty}^{\infty}e^{-u^{2}}h(u/k)du\rightarrow h(0),~~~~k\rightarrow\infty. \end{align*} For the question, let $u=kx$ for the substitution. If the denominator in question is $\pi$, then it may not go through.

  • $\begingroup$ You beat me to it. (+1) Of course, the tacit assumption is that $h$ is continuous. $\endgroup$ – Mark Viola Mar 8 '18 at 18:54

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.