3

it is known that the generalized hypergeometric function is defined by: \begin{equation} \label{e:pFq} {}_{p}F_q(\textbf{a};\textbf{b};z)=\sum_{n=0}^{\infty}\frac{(a_1)_n\cdots (a_p)_n}{(b_1)_n\cdots(b_q)_n}\frac{z^n}{n!}, \end{equation} where \begin{equation} \label{relposchgamma} (z)_n = \frac{\Gamma(z+n)}{\Gamma(z)} \end{equation} is the Pochammer symbol, $\textbf{a}\in\mathbb{C}^p$, $\textbf{b}\in\mathbb{C}^q$ and $z\in\mathbb{C}$. I am studying the case $p=q+1$.

On the one hand, I know by comparison limit test that it is absolutely convergent if $\alpha=\Re\left(\sum_{j=0}^{q}b_j-\sum_{j=0}^{q+1}a_j\right)>0$.

On the other hand, I read in the literature that if $\alpha\leq -1$ it is divergent; and, if $-1<\alpha\leq 0$ and $z\neq 1$, the series converges, but I don't know how to prove it, so I would like you to give me some help to try it out.

Thanks a lot, Ivan.

ivan
  • 97

1 Answers1

1

Define $$A_n:=\frac{(a_1)_n\cdot\ldots\cdot(a_{q+1})_n}{(b_1)_n\cdot\ldots\cdot(b_q)_n}\frac{1}{n!}$$ Notice that \begin{align} \Big(\frac{|A_{n+1}|}{|A_n|}\Big)^n&=\Big(\frac{\prod^{q+1}_{k=1}|a_k+n|}{(n+1)\prod^p_{j=1}|b_j+n|}\Big)^n\\ &=\Big|1+\frac{a_{q+1}-1}{n+1}\Big|^n\prod^q_{k=1}\Big|1+\frac{a_k-b_k}{n+b_k}\Big|^n\xrightarrow{n\rightarrow\infty}e^{\mathfrak{R}(a_{q+1}-1)}\prod^q_{k=1}e^{\mathfrak{R}(a_k-b_k)}\\ &=\exp\Big(\mathfrak{R}\big(\sum^{q+1}_{k=1}a_j-\sum^q_{k=1}b_k -1\Big)\Big) \end{align} Consequently, $\sum_n|A_n|$ converges if $\mathfrak{R}\big(\sum^{q+1}_{k=1}a_j-\sum^q_{k=1}b_k\big)<0$; $\sum_n|A_n|$ diverges if $\mathfrak{R}\big(\sum^{q+1}_{k=1}a_j-\sum^q_{k=1}b_k\big)>0$; the test is inconclusive if $\mathfrak{R}\big(\sum^{q+1}_{k=1}a_j-\sum^q_{k=1}b_k\big)=0$. Thus is based on the the following not well known convergence test:

Theorem: Suppose $a_n>0$ for all $n$.

  1. If $\limsup_n\Big(\frac{a_{n+1}}{a_n}\Big)^n<\frac1e$, then the series $\sum_na_n$ converges.
  2. If there is $N$ such that for all $n\geq N$, $\Big(\frac{a_{n+1}}{a_n}\Big)^n\geq\frac1e$, then the series $\sum_na_n$ diverges.

An alternative, a proof similar to the one used for $p=2$here works. Notice that \begin{align} \frac{A_{n+1}}{A_n}&=\frac{\prod^{p+1}_{k=1}(a_k+n)}{(n+1)\prod^p_{j=1}(b_j+n)}=1+\frac{\prod^{p+1}_{k=1}(a_k+n)-(n+1)\prod^p_{j=1}(b_j+n)}{\prod^{p+1}_{k=1}(a_k+n)}\\ &=1+\frac{\big(\sum^{p+1}_ja_k - \sum^p_kb_k-1)\big)\big)n^p+P(n)}{Q(n)}\tag{1}\label{one} \end{align} where $P$ is a polynomial in $n$ of degree at most $p-1$ and $Q$ is a monomial in $n$ of degree $p+1$. It follows immediately that the series $f(z)=\sum_nA_nz^n$ has radius of convergence $1$, hence $f$ is analytic in $B(0;1)$.

Now for $|z|=1$, a further inspection of \eqref{one} yields $$\frac{A_{n+1}}{A_n}=1+\frac{\xi}{n z_n}+r_n$$ where $\xi=\sum^{p+1}_ja_k - \sum^p_kb_k-1$, $z_n\xrightarrow{n\rightarrow\infty}1$, and $nr_n\xrightarrow{n\rightarrow\infty}0$. Then \begin{align} n\Big(1-\Big|\frac{A_{n+1}}{A_n}\Big|\Big)&=n\Big(1-\big|1+\frac{\xi}{n z_n}+r_n\big|\Big)\\ &=n\left(\frac{1- \big|1+\frac{\xi}{n z_n}+r_n\big|^2}{1+ \big|1+\frac{\xi}{n z_n}+r_n\big|}\right)\\ &=n\left(\frac{-2\mathfrak{R}(\frac{\xi}{n z_n}+r_n)-\tfrac{1}{n^2}\mathfrak{R}^2(\frac{\xi}{z_n}+nr_n) +\frac{1}{n^2}\mathfrak{I}^2(\frac{\xi}{z_n}+nr_n)}{1+ \big|1+\frac{\xi}{n z_n}+r_n\big|} \right) \end{align} Hence $$ n\Big(1-\Big|\frac{A_{n+1}}{A_n}\Big|\Big)\xrightarrow{n\rightarrow\infty}-\mathfrak{R}(\xi)=-\mathfrak{R}\Big(\sum^{p+1}_ja_k - \sum^p_kb_k-1\Big)=1+\mathfrak{R}\big(\sum^p_kb_k-\sum^{p+1}_ja_k\big) $$ The conclusion then follows by the Raabe's theorem: if $-\mathfrak{R}(\xi)>1$, $\sum_nA_ne^{in \theta}$ converges absolutely; when $-\mathfrak{R}(\xi)<1$, $\sum_n|A_n|$ diverges; when $\mathfrak{R}(\xi)=1$ the test is inconclusive.

Mittens
  • 46,352
  • Thank you very much for your evidence. Now it only remains to be seen, why the original series is divergent. – ivan Apr 05 '23 at 06:38
  • @Ivan: I don't know which literature you are referring to, but must likely, they referred to the absolute convergence of the series. If you care to furnish the literature in your posting, then maybe someone would have an answer. – Mittens Apr 05 '23 at 15:38