5

I have an expression which involves Meijer G-functions, to be precise $$ G_{1,4}^{2,1}\left(x \left\vert \begin{array}{c} 1 \\ 1,1,0,\frac{1}{2} \\ \end{array} \right)\right. $$ After messing around a bit with the Mathematica software I found that the following appears to hold: $$ G_{1,4}^{2,1}\left(x \left\vert \begin{array}{c} 1 \\ 1,1,0,\frac{1}{2} \\ \end{array} \right)\right. -\frac{1}{\sqrt{\pi}} = - G_{0,3}^{2,0}\left(x \left \vert \begin{array}{c} 0,1,\frac{1}{2} \\ \end{array} \right)\right. $$ With "appears to hold" I mean the plots overlay perfectly. But I would like to prove it. I know that you can lower the order of the G function if the same parameters appear in the $a$ and $b$ vectors. But this only applies if the $1$ in $b$ are in its second half. So in my example it does not cancel. Any suggestions on how to proceed are highly appreciated.

PS: Input for Mathematica

Plot[MeijerG[{{1}, {}}, {{0, 1}, {1/2, 1}}, x] - 1/Sqrt[Pi] + MeijerG[{{1}, {}}, {{1, 1}, {0, 1/2}}, x], {x, 0, 4}, PlotRange -> All]

Which just gives machine precision noise.

user170231
  • 19,334
Mike Jordan
  • 122
  • 6

1 Answers1

5

From the definition, \begin{align} G_{1,4}^{2,1}\left(x \left\vert \begin{array}{c} 1 \\ 1,1,0,\frac{1}{2} \\ \end{array} \right)\right.&=\frac{1}{2i\pi}\int_{\mathcal L}\frac{[\Gamma(1+s)]^2\Gamma(-s)}{\Gamma(1-s)\Gamma(1/2-s)}x^{-s}\,ds \end{align} Here, the parameters of the function corresponds to the case (i) of the of the definition of the Meijer function given above. The contour $\mathcal L$ can be a straight line $(\gamma-i\infty,\gamma+i\infty)$ with $-1<\gamma<0$.

Using the functional relation of the Gamma function, simplifications of the integrand leads to the alternative expression \begin{equation} G_{1,4}^{2,1}\left(x \left\vert \begin{array}{c} 1 \\ 1,1,0,\frac{1}{2} \\ \end{array} \right)\right. =-\frac{1}{2i\pi}\int_{\mathcal L}\frac{\Gamma(1+s)\Gamma(s)}{\Gamma(1/2-s)}x^{-s}\,ds \end{equation} Under the simplified form the integrand could correspond to another Meijer function, but the contour must be checked carefully. Indeed, \begin{equation} G_{0,3}^{2,0}\left(x \left \vert \begin{array}{c} 0,1,\frac{1}{2} \\ \end{array} \right)\right.=\frac{1}{2i\pi}\int_{\mathcal L'}\frac{\Gamma(1+s)\Gamma(s)}{\Gamma(1/2-s)}x^{-s}\,ds \end{equation} where, $\mathcal L'$ can be the straight line $(\gamma'-i\infty,\gamma'+i\infty)$, but here $0<\gamma'<1$. From the residue theorem, we can shift the contour to make it correspond to $\mathcal L$ by including the residue at $s=0$: \begin{equation} G_{0,3}^{2,0}\left(x \left \vert \begin{array}{c} 0,1,\frac{1}{2} \\ \end{array} \right)\right.=\frac{1}{2i\pi}\int_{\mathcal L}\frac{\Gamma(1+s)\Gamma(s)}{\Gamma(1/2-s)}x^{-s}\,ds+\frac{1}{\sqrt{\pi}} \end{equation} Then, \begin{equation} G_{1,4}^{2,1}\left(x \left\vert \begin{array}{c} 1 \\ 1,1,0,\frac{1}{2} \\ \end{array} \right)\right.= -G_{0,3}^{2,0}\left(x \left \vert \begin{array}{c} 0,1,\frac{1}{2} \\ \end{array} \right)\right.+\frac{1}{\sqrt{\pi}} \end{equation} as expected.

Paul Enta
  • 14,113