# Solution to Simultaneous Arithmetic/Geometric Mean Recursion Relations

I'm trying to solve the simultaneous convergent sequences of geometric/arithmetic means where $a_{n+1}=\frac{1}{2}(a_n+b_n)$ and $b_{n+1}=\sqrt{a_nb_n}$ and initial values are $a_0=1+x$ and $b_0=1-x$. I've tried an implementation using RSolve[] as follows but Mathematica just spits my input back at me without raising an error. Can someone point me in the right direction?

RSolve[{a[n + 1] == 1/2 (a[n] + b[n]), b[n + 1] == Sqrt[a[n] b[n]],
a == 1 + x, b == 1 - x}, {a, b}, n]


EDIT/UPDATE: I have taken Sasha's advice and implemented ArithmeticGeometricMean[]. The solution $M(x)$ matches the iterative solution obtained from the last element of the following recurrence table.

RecurrenceTable[{a[n + 1] == 1/2 (a[n] + b[n]),b[n + 1] == Sqrt[a[n] b[n]], a == 1 + x, b == 1 - x}, {a,b}, {n, 1, 10}]


That is, the coefficients of the series expansion for the approximate iterative solution match those for the series expansion of the ArithmeticGeometricMean[] function up to order 10 at least. Further, and for completeness, the series expansion of $\frac{1}{M(x)}$ also matches that of the following integral, which is actually what I was ultimately after. $$\frac{1}{\pi}\int_0^\pi\frac{d\phi}{\sqrt{1-x^2\cos^2(\phi)}}$$ Thanks everyone for the help! I'm closing the question now.

• I would have thought removing the indeterminate x from the initial conditions might coax it, but doing so still only returns the input on my system. – IPoiler Oct 2 '15 at 0:55
• It's strange, as soon as you remove the nonlinear relationships (Sqrt and the product of a[n] b[n]) from the geometric mean equation it evaluates, even with the x in the initial conditions. However, I wouldn't expect RSolve to be incapable of handling these relationships when the documentation gives examples of it solving linear fractional systems under "Scope->Systems of Difference Equations." – IPoiler Oct 2 '15 at 1:04

If you are looking for the $a(x) = \lim_{n \to \infty} a_n(x) = \lim_{n \to \infty} b_n\left(x\right)$, then Mathematica provides this as ArithmeticGeometricMean function.

Moreover, it can be expressed in terms of the complete elliptic integral of the first kind:

In:= FunctionExpand[ArithmeticGeometricMean[1 + x, 1 - x],
0 < x < 1]

Out= \[Pi]/(2 EllipticK[x^2])

• Wow. Mathematica has a function for everything. I ended up using a RecurrenceTable[] to get an iterative solution but it's nice to confirm that they match. Is there an obvious reason that $0<x<1$ and I'm just not seeing it? Does that also imply that $1<a_0<2$ and $0<b_0<1$? I just proved convergence of the sequences and I'm concerned that I never addressed this. – medley56 Oct 2 '15 at 14:10
Clear[f1, f2, f3]


The recurrence relationship can be implemented using FixedPoint

f1[x_?NumericQ] :=
FixedPoint[
{(#[] + #[])/2, Sqrt[#[]*#[]]} &,
{1 + x, 1 - x}][] /; -1 < x < 1;


Or with NestWhile

f2[x_?NumericQ] :=
NestWhile[
{(#[] + #[])/2, Sqrt[#[]*#[]]} &,
{1 + x, 1 - x}, Unequal @@ # &][] /; -1 < x < 1;


Or, as suggested by Sasha, with ArithmeticGeometricMean

f3[x_] = FunctionExpand[
ArithmeticGeometricMean[1 + x, 1 - x],
-1 < x < 1]

(*  Pi/(2 EllipticK[x^2])  *)

Plot[{f1[x], f2[x], f3[x]}, {x, -1, 1},
PlotStyle -> {
Directive[Lighter[Red], AbsoluteThickness],
Directive[Blue, AbsoluteDashing[{10, 10}]],
Directive[Green, AbsoluteDashing[{20, 15}]]},
PlotPoints -> 25,
PlotRange -> {0, 1.02},
PlotLegends -> {FixedPoint, NestWhile, AGM}] • I wish I could accept both answers! This was totally complete as well but I was specifically looking for that ArithmeticGeometricMean[] function. Thanks for the visualization too! – medley56 Oct 2 '15 at 14:12