3
$\begingroup$

In " Special functions: an introduction to classical functions of mathematical physics" by Nico M. Temme, at page 113 is reported this formula:

$$_2F_1(a,b;c;z)=\frac{\Gamma(c)\Gamma(b-a)}{\Gamma(b)\Gamma(c-a)}\,(-z)^{-a}\,_2F_1(a,1-c+a\ ;\,1-b+a\ ;\frac{1}{z})+ \\+\,\frac{\Gamma(c)\Gamma(a-b)}{\Gamma(a)\Gamma(c-b)}\,(-z)^{-b}\,_2F_1(b,1-c+b\ ;1-a+b\ ;\,\frac{1}{z})$$

Which, from what I understand, is the extension of the solution of hypergeometric differential equation for $\vert{z}\vert>1$, and therefore the convergent series representation of the $_2F_1(a,b;c;z)$ for $\vert{z}\vert>1$.

How is to possible to demonstrate this formula, for $z\in\mathbb{R}$, starting from this integral formulation of Gauss Hypergeoemtric series?

$$_2F_1(a,b;c;z)=\frac{\Gamma(c)}{\Gamma(b)\Gamma(c-b)}\int_0^1t^{b-1}\,(1-t)^{c-b-1}\,(1-tz)^{-a}dt $$

I have searched in various texts but unfortunately no one explicitly reports the demonstration. In particular "Handbook of Mathematical Functions with Formulas, Graphs, and Mathematical Tables" by M. Abramowitz and I. Stegun at page 559,suggest to get this result by Mellin-Barnes integral,but i would like to know if there is a (perhaps simpler) proof that does not include the calculation of the residuals.

$\endgroup$
3
$\begingroup$

Let's actualy assume $z\in\mathbb{C}$, ${\rm Im} z >0$, as that will allow ust to use a simple contour without worrying about the singularity at $z^{-1}$.

By choosing a countour made of the real line (avoiding singularities at $0$ an $1$ from above and closing it with the upper semicircle, it can be shown that (for $a,b,c$ from aprropriate range) $$ 0 = \Big(\int_{-\infty}^0 + \int_0^1 + \int_1^\infty \Big) t^{b-1}(1-t)^{c-b-1}(1-tz)^{-a} dt$$

We have \begin{align} & \int_{-\infty}^0 t^{b-1}(1-t)^{c-b-1}(1-tz)^{-a} dt = \\ & \qquad = - e^{i\pi b} \int_0^{\infty} s^{b-1}(1+s)^{c-b-1}(1+sz)^{-a} ds =^{s=\frac{u}{1-u}} \\ & \qquad= - e^{i\pi b} \int_0^1 u^{b-1}(1-u)^{a-c} (1-u(1-z))^{-a}du = \\ & \qquad= - e^{i\pi b} \frac{\Gamma(b)\Gamma(1+a-c)}{\Gamma(1+a+b-c)}\, _2F_1(a,b;1+a+b-c;1-z) \\ & \int_0^1 t^{b-1}(1-t)^{c-b-1}(1-tz)^{-a} dt = \\ &\qquad = \frac{\Gamma(b)\Gamma(c-b)}{\Gamma(c)}\, _2F_1(a,b;c;z) \\ & \int_1^\infty t^{b-1}(1-t)^{c-b-1}(1-tz)^{-a} dt = \\ & \qquad = -e^{i\pi(b-c)}\int_1^\infty t^{b-1}(t-1)^{c-b-1}(1-tz)^{-a} dt = ^{t=\frac{1}{u}} \\ &\qquad = -e^{i\pi(b-c)} \int_0^1 u^{a-c} (1-u)^{c-b-1}(u-z)^{-a}du = \\ &\qquad = -e^{i\pi(b-c)} (-z)^{-a} \int_0^1 u^{a-c} (1-u)^{c-b-1}(1-uz^{-1})^{-a}du = \\ &= -e^{i\pi(b-c)}\frac{\Gamma(1+a-c)\Gamma(c-b)}{\Gamma(1+a-b)} (-z)^{-a}\,_2F_1(a,1+a-c;1+a-b;z^{-1}) \end{align} In total we have $$ 0 = - e^{i\pi b} \frac{\Gamma(b)\Gamma(1+a-c)}{\Gamma(1+a+b-c)}\, _2F_1(a,b;1+a+b-c;1-z) + \frac{\Gamma(b)\Gamma(c-b)}{\Gamma(c)}\, _2F_1(a,b;c;z) -e^{i\pi(b-c)}\frac{\Gamma(1+a-c)\Gamma(c-b)}{\Gamma(1+a-b)} (-z)^{-a}\,_2F_1(a,1+a-c;1+a-b;z^{-1}) $$ By exchanging $a\leftrightarrow b$ and using a symmetry of a hypergeometric function, we get another identity $$ 0 = - e^{i\pi a} \frac{\Gamma(a)\Gamma(1+b-c)}{\Gamma(1+a+b-c)}\, _2F_1(a,b;1+a+b-c;1-z) + \frac{\Gamma(a)\Gamma(c-a)}{\Gamma(c)}\, _2F_1(a,b;c;z) -e^{i\pi(a-c)}\frac{\Gamma(1+b-c)\Gamma(c-a)}{\Gamma(1-a+b)} (-z)^{-b}\,_2F_1(b,1+b-c;1-a+b;z^{-1}) $$ By eliminating $_2F_1(a,b;1+a+b-c;1-z)$ from these equations, we get \begin{align}& \Big(e^{i\pi (c-b)}\frac{\Gamma(c-b)}{\Gamma(c)\Gamma(1+a-c)} - e^{i\pi (c-a)}\frac{\Gamma(c-a)}{\Gamma(c)\Gamma(1+b-c)}\Big) \,_2F_1(a,b;c;z) = \\ &\qquad = \frac{\Gamma(c-b)}{\Gamma(b)\Gamma(1+a-b)} (-z)^{-a}\,_2F_1(a,1+a-c;1+a-b;z^{-1}) + \\ &\qquad\quad - \frac{\Gamma(c-a)}{\Gamma(a)\Gamma(1-a+b)} (-z)^{-b}\,_2F_1(b,1+b-c;1-a+b;z^{-1}) \end{align} Now, by using the identity $\Gamma(x)\Gamma(1-x)=\frac{\pi}{\sin(\pi x)}$ it can be shown that $$ e^{i\pi (c-b)}\frac{\Gamma(c-b)}{\Gamma(c)\Gamma(1+a-c)} - e^{i\pi (c-a)}\frac{\Gamma(c-a)}{\Gamma(c)\Gamma(1+b-c)} = \frac{\sin(\pi(b-a))}{\pi}\frac{\Gamma(c-a)\Gamma(c-b)}{ \Gamma(c)} $$ $$ \frac{\Gamma(c-b)}{\Gamma(b)\Gamma(1+a-b)} = \frac{\sin(\pi(b-a))}{\pi} \frac{\Gamma(c-b)\Gamma(b-a)}{\Gamma(b)} $$ $$ \frac{\Gamma(c-a)}{\Gamma(a)\Gamma(1-a+b)} = -\frac{\sin(\pi(b-a))}{\pi} \frac{\Gamma(c-a)\Gamma(a-b)}{\Gamma(a)} $$ So after dividing the previous identity by a constant we get \begin{align}& \,_2F_1(a,b;c;z) = \\ &\qquad = \frac{\Gamma(c)\Gamma(b-a)}{\Gamma(b)\Gamma(c-a)} (-z)^{-a}\,_2F_1(a,1+a-c;1+a-b;z^{-1}) + \\ &\qquad\quad + \frac{\Gamma(c)\Gamma(a-b)}{\Gamma(a)\Gamma(c-b)} (-z)^{-b}\,_2F_1(b,1+b-c;1-a+b;z^{-1}) \end{align}

$\endgroup$
  • $\begingroup$ Thanks, a very comprehensive demonstration. The only statement that I don't understand is how to prove that the integral from -infinity to + infinitive of the integrand is zero? $\endgroup$ – C.C.12 May 1 at 18:18
  • 1
    $\begingroup$ The integrated function is holomorphic in the upper half of the complex plane (${\rm Im} t > 0$), so integral over any closed curve contained within this region is equal to $0$. In this case, the curve that we're using is an interval $[-R,R]$ closed with a semi-circle. For the right parameters $a,b,c$ the function vanishes quickly enough when $t\rightarrow\infty$, and the integral over the semicircle vanishes when we go with $R\rightarrow\infty$. That means that the integral over the real line is $0$. $\endgroup$ – Adam Latosiński May 1 at 20:14
  • 1
    $\begingroup$ One more issue that we need to be careful about is whether we can approach points $t=0$ and $t=1$ (which may be singularities) with no problems, but again, for $a,b,c$ in an appropriate range, it is possible. $\endgroup$ – Adam Latosiński May 1 at 20:14
  • $\begingroup$ Thanks a lot. Can you tell me a book where there is this demonstration? Because as I told you I didn't find even one that reported it and I didn't understand why, since it's not so trivial. $\endgroup$ – C.C.12 May 2 at 16:14
  • 1
    $\begingroup$ I've learned this method while proof-checking the latest publication for a fellow faculty member, prof. J. Dereziński. I don't know whether it has already been published, but it's available at his homepage, here's the link: fuw.edu.pl/~derezins/derezinski_CDDE.pdf The method itself appears on page 67. It uses a slightly diferent notation for hypergeometric functions, which is defined on page 58. $\endgroup$ – Adam Latosiński May 2 at 16:32

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.