4
$\begingroup$

Consider the Confluent hypergeometric function $U(a,b,z)$, which is a solution of the Kummer's Equation : $$zw''+(b-z)w'-aw=0$$

it has the following integral representation when $- \pi/2 < \arg z< \pi/2$ $$U(a,z)= \frac{1}{\Gamma(a)} \int^{\infty}_0 t^{a-1}(1+t)^{b-a-1} e^{-tz} dt$$

I know that there is the following asymptotic series for $U(a,1,z)$ :

$$\mathop{U}\nolimits\!\left(a,1,z\right)=-\frac{1}{\mathop{\Gamma}\nolimits% \!\left(a\right)}\left(\mathop{\ln}\nolimits z+\mathop{\psi}\nolimits\!% \left(a\right)+2\gamma\right)+\mathop{O}\nolimits\!\left(z\mathop{\ln}% \nolimits z\right) \ \ \ \ (*) \ \ \ \ z \to 0$$

http://dlmf.nist.gov/13.2.E19

Where $\gamma$ is the Euler constant and $\psi(z) = \displaystyle \frac{\Gamma'(z)}{\Gamma(z)}$ is the digamma function.

I would be very thankful if one can explain how should I start deriving asymptotic series $(*)$

$\endgroup$
3
$\begingroup$

It might be easier to start with one of the other integral representations, but here's one way. We'll begin with some manipulations to strip away the lower order terms:

$$ \begin{align} &\int_0^\infty t^{a-1} (1+t)^{-a} e^{-zt}\,dt \\ &\qquad = \left(\int_0^1 + \int_1^\infty\right) t^{a-1} (1+t)^{-a} e^{-zt}\,dt \\ &\qquad = \int_0^1 t^{a-1} (1+t)^{-a}\,dt + O(z) + \int_1^\infty t^{a-1} (1+t)^{-a} e^{-zt}\,dt \\ &\qquad = \int_0^1 t^{a-1} (1+t)^{-a}\,dt + O(z) + \int_1^\infty \left[t^{a-1} (1+t)^{-a} - t^{-1} + at^{-2}\right]e^{-zt}\,dt \\ &\qquad \qquad - a\int_1^\infty t^{-2}e^{-zt}\,dt + \int_1^\infty t^{-1} e^{-zt}\,dt. \tag{1} \end{align} $$

The last integral in the last line diverges logarithmically as $z \to 0^+$ while the other integrals converge. Thus the leading-order behavior comes from the last integral. To extract this behavior let's first make the change of variables $zt=s$, then integrate by parts:

$$ \begin{align} \int_1^\infty t^{-1} e^{-zt}\,dt &= \int_z^\infty s^{-1} e^{-s}\,ds \\ &= e^{-s}\log s \Bigr|_z^\infty + \int_z^\infty e^{-s}\log s\,ds \\ &= -e^{-z}\log z + \int_0^\infty e^{-s}\log s\,ds - \int_0^z e^{-s}\log s\,ds. \end{align} $$

To get the last line we split the integral like $\int_z^\infty = \int_0^\infty - \int_0^z$. Note that we could repeatedly integrate the last integral by parts to obtain further terms of the asymptotic expansion if we desire. We only need the terms up to $O(z\log z)$, so it's enough to notice that

  • $e^{-z}\log z = \log z + O(z\log z)$,

  • $\int_0^\infty e^{-s}\log s\,ds = -\gamma$, and

  • $\int_0^z e^{-s}\log s\,ds = O(z\log z)$ by L'Hopital's rule.

Thus

$$ \int_1^\infty t^{-1} e^{-zt}\,dt = -\log z - \gamma + O(z\log z). \tag{2} $$

It remains to estimate the remaining integrals in $(1)$, namely

$$ \int_1^\infty \left[t^{a-1} (1+t)^{-a} - t^{-1} + at^{-2}\right]e^{-zt}\,dt \qquad \text{and} \qquad \int_1^\infty t^{-2}e^{-zt}\,dt. $$

The integrand in the first of these is $O(t^{-3})e^{-zt}$, so we can apply the mean value theorem (i.e. $e^{-zt} = 1 -zte^{-zc(t)}$ with $0 < c(t) < t$) to conclude that

$$ \int_1^\infty \left[t^{a-1} (1+t)^{-a} - t^{-1} + at^{-2}\right]e^{-zt}\,dt = \int_1^\infty \left[t^{a-1} (1+t)^{-a} - t^{-1} + at^{-2}\right]dt + O(z).$$ $$ \tag{3} $$

For the second integral we proceed as we did with the singular integral in $(2)$:

$$ \begin{align} \int_1^\infty t^{-2}e^{-zt}\,dt &= \int_1^\infty t^{-2}\,dt + \int_1^\infty t^{-2}(e^{-zt}-1)\,dt \\ &= \int_1^\infty t^{-2}\,dt + z\int_z^\infty s^{-2}(e^{-s}-1)\,ds \\ &= \int_1^\infty t^{-2}\,dt + z\int_1^\infty s^{-2}(e^{-s}-1)\,ds - z\int_1^z s^{-2} (e^{-s}-1)\,ds. \end{align} $$

Now

$$ \int_1^z s^{-2} (e^{-s}-1)\,ds = O(\log z) $$

by L'Hopital's rule, so

$$ \int_1^\infty t^{-2}e^{-zt}\,dt = \int_1^\infty t^{-2}\,dt + O(z\log z). \tag{4} $$

Combining $(3)$ and $(4)$ we get

$$ \begin{align} &\int_1^\infty \left[t^{a-1} (1+t)^{-a} - t^{-1} + at^{-2}\right]e^{-zt}\,dt - a\int_1^\infty t^{-2}e^{-zt}\,dt \\ &\qquad = \int_1^\infty \left[t^{a-1} (1+t)^{-a} - t^{-1}\right]dt + O(z\log z). \end{align} $$

Substituting this and $(2)$ into the first equation we got, we conclude that

$$ \int_0^\infty t^{a-1} (1+t)^{-a} e^{-zt}\,dt = -\log z - \gamma + f(a) + O(z\log z), $$ where $$ f(a) = \int_0^1 t^{a-1} (1+t)^{-a}\,dt + \int_1^\infty \left[t^{a-1} (1+t)^{-a} - t^{-1}\right]dt. $$

It shouldn't be hard to show that

$$ f(a) = \int_0^1 \frac{t^{a-1}-1}{1-t}\,dt = -\psi(a)-\gamma, $$

from which the result would follow.

$\endgroup$
  • $\begingroup$ Thank you so much for your answer ! I will try to follow every single step, I hope I can get it all $\endgroup$ – the8thone Apr 14 '15 at 2:16
  • $\begingroup$ Sure thing. I hope it's helpful. This is the usual method I use for these types of problems, though I suspect there is a more elegant approach out there... $\endgroup$ – Antonio Vargas Apr 14 '15 at 6:41

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.