0
$\begingroup$

In this ArXiv paper by Wilk and Wlodarczyk (published in Physical Review Letters), equation 16 has essentially the following definition of a function: $$\text{f(x)=}\frac{c}{2Dx^2}\exp[\int^x_0 \frac{\mu}{t^2}+\frac{1-2\alpha}{t} dt]$$

They claim that with the normalization condition

$$\int^\infty_0 f(x)\,dx = 1$$

it becomes (with some variable changes)

$$\text{f(x)=}\frac{1}{\Gamma\text{(}\alpha\text{)}}\mu\text{(}\frac{\mu}{x}\text{)}^{\text{(}\alpha-1\text{)}}\exp\text{(-}\frac{\mu}{x}\text{)}.$$


I can't understand the integral in the exponential function, because it appears to be infinite. Can you help me by telling how can I solve this type of integral?

$\endgroup$
10
  • 1
    $\begingroup$ This question was a repeat of mathoverflow.net/questions/55714. Do not post the same question multiple times - if you think your question should be reopened you can edit it and flag it for moderator attention, or start a thread on tea.mathoverflow.net. $\endgroup$ Feb 17, 2011 at 16:26
  • 1
    $\begingroup$ SMH was unable to edit the previous question because two user accounts were created. @SMH: you should register, so your work is less volatile. $\endgroup$
    – S. Carnahan
    Feb 17, 2011 at 16:35
  • 2
    $\begingroup$ Can you just email the authors asking what exactly they mean by the function $f$, which---as noted above---is not defined in a traditional sense, at any (!) point? $\endgroup$ Feb 17, 2011 at 17:33
  • 1
    $\begingroup$ Why do you expect mathematics published in physics journals to make sense to mathematicians? $\endgroup$ Feb 17, 2011 at 18:22
  • 2
    $\begingroup$ @Gerald: Indeed. On the other hand, Talagrand... $\endgroup$
    – Did
    Feb 17, 2011 at 18:37

1 Answer 1

8
$\begingroup$

The trouble (as was already explained to you) lies in the starting point $t=0$ of the integral in the exponential. Fortunately, W+W are only interested in steady solutions of equation (15) of their arXiv preprint and these can be written as the function in their equation (16) provided one replaces the starting point $t=0$ of the integral by any starting point $t=x_0$ with $x_0>0$. Changing $x_0$ only modifies $f$ by a multiplicative factor so the normalisation condition saves the day.

Assuming for example that $x_0=1$, one gets $$f(x)=\frac{c}{x^2}\exp\left(\int^x_1 \left(\frac{\mu}{t^2}+\frac{1-2\alpha}{t}\right) \mathrm{d}t\right)=\frac{c}{x^2}\exp\left(\mu-\frac{\mu}{x}+(1-2\alpha)\log(x)\right), $$ hence $$ f(x)=c\mathrm{e}^{\mu}x^{-1-2\alpha}\mathrm{e}^{-\mu/x}. $$ This is not W+W's formula (18) so either I made a mistake in this post or there is a misprint in W+W's preprint. Note that the function $f_{0}$ written in (18) of W+W's preprint and in your question here is not integrable if $\alpha\le2$ because $f_{0}(x)$ behaves like a multiple of $x^{1-\alpha}$ when $x\to\infty$, hence for such values of $\alpha$, $f_{0}$ cannot be normalized to get a probability density function (even assuming that one got rid of the problem of the starting point $t=0$ as I explained). The function I obtained above is integrable for every positive $\alpha$ and $\mu$.

If $f$ is the density of the distribution of a random variable $X$, the distribution of $Y=\mu/X$ has density $$ g(y)=c\mathrm{e}^{\mu}\mu^{-2\alpha}y^{2\alpha-1}\mathrm{e}^{-y}, $$ hence $Y$ is a standard Gamma random variable of exponent $2\alpha$ and $$ c=\mathrm{e}^{-\mu}\mu^{2\alpha}/\Gamma(2\alpha). $$ (As I said before, this question belongs to math.SE.)

$\endgroup$
7
  • $\begingroup$ All this might mean that (20) in W+W's preprint should be corrected to $q=1+2\tau D$. Or maybe not. $\endgroup$
    – Did
    Feb 17, 2011 at 17:55
  • $\begingroup$ I want to thank Didier Piau for the answer. It is really helpful. but I think there is a mistake in g(y). g(y) is proportional to $\y^{2\alpha+1}$. I believe that in W+W's article $\2\alpha$ must be just $\alpha$ in obtaining the coefficients of the Fokker-Planck eqn (17). so with the point of evaluationg the integrals that you said, I obtain the distribution like this :$$\text{f(x)=}\frac{1}{\Gamma\text{(}\alpha\text{)}}\mu\text{(}\frac{\mu}{x}\text{)}^{\text{(}\alpha+1\text{)}}\exp\text{(-}\frac{\mu}{x}\text{)}.$$ I have $\alpha+1$ not $\alpha-1$. $\endgroup$
    – SMH
    Feb 17, 2011 at 19:17
  • $\begingroup$ @SMH: No, $y^{2\alpha-1}$ is correct. $\endgroup$
    – Did
    Feb 17, 2011 at 19:22
  • $\begingroup$ @Didier: sorry! Can you explain it a little more please? $\endgroup$
    – SMH
    Feb 17, 2011 at 19:45
  • 2
    $\begingroup$ @SMH: The empiricist in me finds rather limited the size of your sample, to reach any valuable conclusion about ISI papers in general. The cynic in me finds rather naive your reaction about the mathematical accuracy of physics papers. Best wishes for your future studies. $\endgroup$
    – Did
    Feb 18, 2011 at 11:33

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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