0
$\begingroup$

I would like to understand the asymptotic behaviour of the following integrals with fixed $x_0>0$: $$J_m=\int^{+\infty}_{x_0}|H_m(x)|^2 e^{-x^2}dx,$$ where $H_m(x)$ is the $m-$th Hermite polynomial. For instance, I would like to understand if $$\frac{J_m}{\sqrt{2^m m!}}=O(m^p)$$ with $p\in\mathbb{R}^−$. Thank you in advance!

$\endgroup$
1
  • $\begingroup$ Do you want here $\sqrt{2^m m!}$ or $2^m m!$? With $\sqrt{2^m m!}$, your conjecture is certainly false. $\endgroup$ Commented Nov 14 at 13:48

2 Answers 2

1
$\begingroup$

Your conjecture is not true. Moreover, $$J_m(y):=\int_y^\infty dx\, e^{-x^2}H_m(x)^2\sim c_m:=\frac12\,\pi^{1/2}\,2^m m! \tag{1}\label{1}$$ whenever $$0\le y=o(m^{1/2}).$$

Indeed, recalling that $H_m^2$ is an even function, we have $$J_m(0)=c_m. \tag{2}\label{2}$$ Also, by Theorem 8.22.9(a) in Szegő's book Orthogonal Polynomials, for any real $h\in(0,1)$, $$e^{-x^2/2}H_m(x)=O(c_m^{1/2}m^{-1/4})$$ (as $m\to\infty$) uniformly in $x\in[0,(2m)^{1/2}h]$. So, $$J_m(0)-J_m(y)=\int_0^y dx\, (e^{-x^2/2}H_m(x))^2=O((c_m^{1/2}m^{-1/4})^2y)=o(c_m).$$ So, in view of \eqref{2}, we do have \eqref{1}. $\quad\Box$

$\endgroup$
3
  • $\begingroup$ I think the factor $\pi^{1/2}$ in your $c_m$ should be $\tfrac{1}{2}\pi^{1/2}$, at least that is what I find; one way to check this, is to set $y=0$, when $J_m=\tfrac{1}{2}\pi^{1/2} 2^m m!$ exactly. $\endgroup$ Commented Nov 14 at 21:04
  • $\begingroup$ @CarloBeenakker : Thank you for your comment. This is now fixed. $\endgroup$ Commented Nov 14 at 22:18
  • $\begingroup$ I actually wanted to ask if $\frac{J_m}{{2^m m!}}=O(m^p)$ but you already answered implicitly to my question. Thank you very much ! $\endgroup$
    – Darius
    Commented Nov 15 at 16:12
1
$\begingroup$

Let me first consider the case $x_0=0$, when the integral has a closed form expression: $$\int^{\infty}_{0}|H_m(x)|^2 e^{-x^2}\,dx=\sqrt{\pi } \,2^{m-1} m!\;\;(m\in\mathbb{N}).$$ See, for example, this MSE post for a derivation.

Considering next a nonzero $x_0$, to evaluate the integral from $0$ to $x_0$ (and subtract it from the answer given above) I substitute the large-$m$ asymptotics of the Legendre polynomials, $$e^{-\frac{x^2}{2}}H_m(x) \rightarrow \frac{2^m}{\sqrt \pi}\Gamma\left(\tfrac{1}{2}+m/2\right) \cos \left(x \sqrt{2 m}- \tfrac{1}{2}m\pi \right),$$ to obtain the asymptotics $$\int^{x_0}_{0}|H_m(x)|^2 e^{-x^2}\,dx\rightarrow \left[\frac{2^m}{\sqrt \pi}\Gamma(\tfrac{1}{2}+m/2)\right]^2\int^{x_0}_0 \cos^2 \left(x \sqrt{2 m}- \tfrac{1}{2}m\pi \right)\,dx\rightarrow\frac{x_0}{2}\left[\frac{2^m}{\sqrt \pi}\Gamma(\tfrac{1}{2}+m/2)\right]^2,$$ which is smaller than the integral from $0$ to $\infty$ by a factor $\sqrt m$.

Hence I conclude that $$\lim_{m\rightarrow\infty}\frac{1}{2^m m!}\int_{x_0}^{\infty}|H_m(x)|^2 e^{-x^2}\,dx=\tfrac{1}{2}\sqrt\pi.$$

Note that the OP uses the convention that $H_m(x)$ is the socalled physicists Hermite polynomial.
$\endgroup$
4
  • $\begingroup$ (I) The limit for $m\to\infty$ cannot depend on $m$. (ii) how do you get this asymptotic? $\endgroup$ Commented Nov 14 at 15:52
  • $\begingroup$ (i) Again, the limit for $m\to\infty$ cannot depend on $m$, because $m$ is then a dummy variable. (ii) In the linked Wikipedia article, the asymptotic formula you are apparently trying to use here is given without any specifications on $x$, and it cannot possibly make sense when the value of the cosine term is $0$. (iii) Even if the asymptotic formula were true pointwise, the convergence of the corresponding integrals must be proved/justified. $\endgroup$ Commented Nov 14 at 18:28
  • 1
    $\begingroup$ @IosifPinelis The only problem here is that $\rightarrow$ should be replaced with $\sim$. The stated asymptotic formula for Hermite polynomials holds pointwise and in fact locally uniformly on $\mathbb R$. This is enough to deduce the stated asymptotics of the integrals. $\endgroup$ Commented Nov 15 at 11:17
  • $\begingroup$ I actually wanted to ask if $\frac{J_m}{{2^m m!}}=O(m^p)$ but you already answered implicitly to my question. Thank you very much ! $\endgroup$
    – Darius
    Commented Nov 15 at 16:12

You must log in to answer this question.

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