Let $\left(u_{n}\right)$ and $\left(v_{n}\right)$ classically defined by $u_{0}=a>0$, $v_{0}=b>0$ and $\forall n \in \mathbb{N}, u_{n+1}=\frac{1}{2}\left(u_{n}+v_{n}\right)$, $v_{n+1}=\sqrt{u_{n} v_{n}}$.
Both $\left(u_{n}\right)$ and $\left(v_{n}\right)$ converge toward a common limit that we denote by $f(a,b)$.
Is the function $f$ differentiable over $\mathbb{R}_+^* \times \mathbb{R}_+^*$? Analytic?
P.S.: This is probably a very well-know result, but I couldn't find it here.
$f(a,b)$ has an expression in terms of a fairly simple integral, $$ \frac{\pi}{2 f(a,b)} = \int_0^{\pi/2} \frac{d\theta}{\sqrt{a^2\cos^2{\theta}+b^2\sin^2{\theta}}} , $$ (originally proven by Gauss). The integrand is $C^{\infty}$ on $(0,\infty) \times (0,\infty) \times (0,\pi/2)$, so differentiation under the integral sign implies that the integral is as well (and hence, by the chain rule, so is $f$). It should be analytic too, since any multidirectional derivative exists and is continuous.