I'm trying to understand the asymptotic behavior of the hypergeometric function $$ _2F_1(a, b; c; z) $$ at fixed argument $0 < z < 1$ when some of the parameters $a$, $b$ and $c$ are large.
There are two specific cases I'm interested in, but I only know the answer in one of them. Let me start with the solved example to set the stage:
1) Consider $$ _2F_1(a + n, b + n; c + 2n; z) $$ in the limit $n \to \infty$. In this case one can use the saddle-point method in the integral representation of the hypergeometric function to show that $$ _2F_1(a + n, b + n; c + 2n; z) \approx (1-z)^{(c-a-b-1/2)/2} \left( 2 \frac{1 - \sqrt{1-z}}{z} \right)^{2n + c - 1} $$ This is the kind of result I'm after: it shows that the hypergeometric function grows at most like $2^{2n}$ when $|z|<1$.
2) Now consider the other case of interest $$ _2F_1(a + n, b - n; c; z) $$ It is similar to the first case in the sense that the parameters are "balanced" in $n$. But now one of these parameters goes negative at large $n$, and I can observe numerically that the function oscillates in $n$. I assume that there should be an asymptotic expression for the amplitude of these oscillations, but I am not able to get one.
I cannot make progress neither with the saddle-point method nor with a variety of hypergeometric identities. Does anyone have a solution, or at least a suggestion on how to attack the problem?