I encountered this integral in my work, and it would be really convenient if it had a closed form in terms of any known special functions (which Mathematica could handle):

$$J(\alpha,\beta)=\int_0^\infty \frac{e^{-x^2} I_0 \left(\beta x \right) d x}{\sqrt{ \alpha^2+x^2}}$$

It's easy to see that:

$$J(\alpha,0)=\frac12 e^{\alpha^2/2} K_0 \left( \frac{\alpha^2}{2} \right)$$

Using series expansion of the Bessel function and a little help from Mathematica, I was able to get a series expression:

$$J(\alpha,\beta)=\frac12 e^{\alpha^2/2} K_0 \left( \frac{\alpha^2}{2} \right)+\frac{\sqrt{\pi}}{16} \beta^2 \sum_{k=0}^\infty \frac{(2k+1)!}{k!^3 (k+1)^2} U \left(\frac{1}{2},-k,\alpha^2 \right) \frac{\beta^{2k}}{4^{2k}}$$

Where $U$ is the confluent hypergeometric function, which in this case (according to Mathematica) is a linear combination of Bessel functions $K_0$ and $K_1$ though I wasn't able to get the general expression for the coefficients yet.

This series is good for small values of $\beta$, but for large $\beta$ I need too many terms, and the computation time is slower than I'd like.

I know that it's simple to get a quadrature formula for the integral, but still, a closed form would be better.

The integral can be thought as a Hankel transform with imaginary argument, but I wasn't able to find it in the tables of Hankel transforms either.

Clarification re: R. Burton's comment:

The final expression has the form:

$$\alpha e^{-\beta^2/2} J(\alpha,\beta)$$

Which should make the expression finite for every choice of variables.

Now that I think about it, I should probably use asymptotics for the Bessel function for large $\beta$:

$$I_0 \left(\beta x \right) \asymp \frac{1}{\sqrt{2 \pi \beta x}}e^{\beta x}$$

Then I get:

$$J(\alpha,\beta) \asymp \frac{1}{\sqrt{2 \pi \beta}} \int_0^\infty \frac{e^{-x^2+\beta x} d x}{\sqrt{x}\sqrt{ \alpha^2+x^2}}, \quad \beta \to \infty$$

Which I still don't know the closed form for.

A more simple integral which does have a closed form:

$$\int_0^\infty e^{-x^2} I_0 \left(\beta x \right) d x= \frac{\sqrt{\pi}}{2} e^{\beta^2/8} I_0 \left(\frac{\beta^2}{8} \right)$$

  • $\begingroup$ I don't know about a closed form, but you can use the definition of $$I_0(x)=\sum_{n=0}^\infty \frac{(\beta x)^{2n}}{4^nn!^2}$$ to get $$J(\alpha,\beta,x)=\int_0^\infty \frac{e^{-x^2}}{\left\Vert(\alpha,x)\right\Vert}\sum_{n=0}^\infty \frac{(\beta x)^{2n}}{4^nn!^2}\ dx$$ If you truncate the sum it should at least reduce the computation time. $\endgroup$
    – R. Burton
    Dec 26, 2018 at 16:39
  • $\begingroup$ Also, I tried running this through Scilab, and for very large $\beta$ (ex 130+), I ended up with things like $x\times10^{103}$, %inf, and error messages. Suffice to say that for sufficiently large $\beta$, the output of $J$ becomes impractically large. $\endgroup$
    – R. Burton
    Dec 26, 2018 at 16:46
  • $\begingroup$ @R.Burton, re: your first comment: this is what I did to get the series $\endgroup$
    – Yuriy S
    Dec 26, 2018 at 17:15
  • $\begingroup$ @R.Burton, I posted an edit to answer your second observation. The point is that for large $\beta$ there's an exponential factor before the integral which should make it small. I guess instead of the series we could derive some kind of asymptotic expression for that case $\endgroup$
    – Yuriy S
    Dec 26, 2018 at 18:07
  • $\begingroup$ Is this somehow related to probability distributions or wavefunctions? $\endgroup$
    – R. Burton
    Dec 27, 2018 at 22:57

1 Answer 1


Changing $x=\beta t$, one can express the asymptotic expression as \begin{align} J(\alpha,\beta) &\asymp \frac{1}{\sqrt{2 \pi \beta}} \int_0^\infty \frac{e^{-x^2+\beta x} d x}{\sqrt{x}\sqrt{ \alpha^2+x^2}}, \quad \beta \to \infty\\ &\asymp \frac{1}{\sqrt{2 \pi }} \int_0^\infty\frac{e^{-\beta^2\left( t^2-t \right)}}{\sqrt{\alpha^2+\beta^2t^2}}\frac{dt}{\sqrt{t}} \end{align} The exponential term reaches its maximum when $t=1/2$, while the remaining part of the integrand behaves smoothly. One may use the Laplace method: \begin{equation} \int_a^b h(t)e^{Mg(t)}\,dt\asymp\sqrt{\frac{2\pi}{M\left|g''(t_0)\right|}}h(t_0)e^{Mg(t_0)}, \quad M \to \infty \end{equation} here $M=\beta^2,g(t)=-t^2+t,h(t)=t^{-1/2}\left( \alpha^2+\beta^2t^2 \right)^{-1/2},t_0=1/2,g(t_0)=1/4,g''(t_0)=-2$ \begin{equation} J(\alpha,\beta) \simeq \frac{2e^{\frac{\beta^2}{4}}}{\beta\sqrt{4\alpha^2+\beta^2}} \end{equation} For $\alpha=1.2345,\beta=10$, we find numerically $\alpha e^{-\beta^2/2}J(\alpha,\beta)=3.4603.10^{-13}$, while the approximation gives $3.3289^{-13}$. Additional terms, both in the $I_0$ expansion and (more laboriously) in the Laplace method, may be added in this way.


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.