5
$\begingroup$

I stumbled upon the following integral:

$$\int e^{-\left(\frac{\gamma}{2} + ip \right)^2} H_n \left( q-\frac{\gamma}{2} \right) H_n \left( q+\frac{\gamma}{2} \right) \, d\gamma \, ,$$

and I really need a closed form of it. I'm pretty sure it exists, and probably in terms of $\Gamma$ functions, since

$$ \int e^\frac{-\gamma^2}{2} H_n(\gamma) d\gamma = 4^n \sqrt{2} \,\Gamma \left( n + \frac{1}{2} \right) \, .$$

The problem is that Mathematica can't compute this in closed form, although it can do it term by term and hint that a closed expression probably exists:

Table[Integrate[
 HermiteH[i, q + 1/2 \[Gamma]] HermiteH[i, 
   q - 1/2 \[Gamma]] Exp[-(\[Gamma]/2 + I p)^2], {\[Gamma], -Infinity,
   Infinity}],{i,0,2}]

gives

$$\left\{2 \sqrt{\pi },4 \sqrt{\pi } \left(2 p^2+2 q^2-1\right),16 \sqrt{\pi } \left(2 \left(p^4+2 p^2 \left(q^2-1\right)+q^4\right)-4 q^2+1\right)\right\} \, .$$

Can someone give me a tip?

$\endgroup$
  • $\begingroup$ Use a generating function for Hermit polynomials $\exp(2xt-t^2)=\sum_n H_n(x) t^n/n!$ $\endgroup$ – yarchik Sep 5 '16 at 21:50
  • $\begingroup$ How would that help, since I don't have a summation? $\endgroup$ – QuantumBrick Sep 5 '16 at 21:57
  • $\begingroup$ But you know how to extract each term using series. $\endgroup$ – yarchik Sep 5 '16 at 21:58
  • $\begingroup$ My suggestion is not to use Mathematica, but to consult a table of integrals like Abramowitz and Stegun. $\endgroup$ – QuantumDot Sep 5 '16 at 22:00
  • $\begingroup$ Can we make any assumptions about the parameters - Real, Integer, Positive...? $\endgroup$ – mikado Sep 5 '16 at 22:04
5
$\begingroup$

This is an integral I happen to have worked with before. The answer is a Laguerre polynomial, as can be verified like this for the first three indices as in the question:

ll = 
  2 Table[LaguerreL[i, 2 (p^2 + q^2)] (-2)^i Sqrt[Pi] i!, {i, 0, 2}];

l1 = Table[
   Integrate[
    HermiteH[i, q + 1/2 γ] HermiteH[i, q - 1/2 γ] Exp[-(γ/2 + I p)^2], 
   {γ, -Infinity, Infinity}], {i, 0, 2}];

Simplify[ll/l1]

(* ==> {1, 1, 1} *)

Here, ll is my closed form, and l1 is the list copied from the question. I verify that they give the same answers by forming the ratios.

The formula originally comes from Gradshtein and Ryzhik, equation 7.377, and I used it in my very first paper, Conductance quantization and backscattering below equation (13) to expand a shifted harmonic oscillator function in terms of the unshifted one. That's why I remembered where to look this up...

$\endgroup$
  • $\begingroup$ I can't thank you enough. Indeed, I just remembered the seminal work by Groenewald, "On the principles of quantum mechanics", uses thais formula to calculate the Wigner function for a harmonic oscillator. Thank you again! $\endgroup$ – QuantumBrick Sep 6 '16 at 11:13
4
$\begingroup$

It is interesting to solve this integral without relying on Gradshteyn and Ryzhik. As I suggested in the comments, one can use the generating function approach. The solution consists of 7 steps. Human input is needed on step 5. All other steps are pretty much automatic.

  1. Substitute generating functions $\exp(2xt-t^2)=\sum_n\frac{1}{n!}H_n(x)t^n$ and integrate

    g[1] = Integrate[Exp[-(Y/2 + I p)^2] Exp[2(q-Y/2)t-t^2] Exp[2(q+Y/2)s-s^2], 
                   {Y, -Infinity, Infinity}]
    

    We have for g[1]=$2 \sqrt{\pi } e^{-2 i p (s-t)+2 q (s+t)-2 s t}=\sum_{mn}\frac{1}{m!n!}I_{m,n}(p,q)t^ms^n$, where we are interested in $I_{n,n}(p,q)$.

  2. Next we need to extract from this generating function coefficients in front of $t^ns^n$, i.e., diagonal elements of the expansion. Let us introduce the substitution {t -> x/y, s -> y} and search for coefficients in front of $x^n y^0$.

    g[2] = g[1] /. {t -> x/y, s -> y};
    
  3. It is the exponential function with the $2\sqrt\pi$ prefactor and power:

    pow = Cases[g[2], Exp[x_] -> x] // First;
    

    i.e., g[2]=2 Sqrt[Pi] Exp[pow]

  4. We simplify the argument of the exponent little bit:

    g[3] = Collect[pow, {y, x}]
    

    $2(q+ip)\,x/y+2(q-ip)\,y -2 x$

  5. Now, we can easily extract the coefficient in front of $y^0$ in the series expansion of g[2] (this is the only nontrivial step in this derivation). For the summation, we rely on the amazing capabilities of Mathematica:

    g[4] = 2 Sqrt[Pi] E^(-2 x) Sum[1/(2 n)! Binomial[2n,n] (2 I p + 2 q)^n x^n 
            (-2 I p + 2 q)^n,{n, 0, Infinity}]
    (*2 E^(-2 x) Sqrt[Pi] Hypergeometric0F1[1, 4 (p^2 + q^2) x]*)
    
  6. At this point we have the generating function, i.e., $2\sqrt{\pi} e^{-2x}{\,}_0F_1\left[;1,(p^2 + q^2) x\right]=\sum_n \frac{1}{(n!)^2}I_{n,n}x^n$. Finally, we extract corresponding series coefficients.

    g[5] = (n!)^2 SeriesCoefficient[g[4], {x, 0, n}]; 
    
  7. The solution in terms of the generalized hypergeometric function can be simplified as follows:

    FullSimplify[FunctionExpand[g[5]], Assumptions -> n >= 0 && n \[Element] Integers]
    

And we are done:

g[6]=(-1)^n 2^(1 + n) n! Sqrt[Pi] LaguerreL[n, 2 (p^2 + q^2)]
$\endgroup$
  • $\begingroup$ Actually you're still using Gradshteyn & Ryzhik because the generating function from which you start is equation GR 8.957... $\endgroup$ – Jens Sep 9 '16 at 4:05
  • $\begingroup$ @Jens GR is such a comprehensive book that I am not surprised that the generating function of Hermite polynomials can be found there. I this case, I vaguely remembered that the function has an exponential form which is useful here. To be sure I just check it on wiki. But thank you to putting a reference to this equation in GR, this makes the answer more complete. $\endgroup$ – yarchik Sep 9 '16 at 6:39

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.