Problem $\mathbf{4}$ If $x_{n+1}=\frac{x_n+y_n}2$, $y_{n+1}=\frac{2x_ny_n}{x_n+y_n}$ for $n=1,2,\ldots,n$, prove that both $\{x_n\}$ and $\{y_n\}$ are convergent, and $\lim\limits_{n\to\infty}x_n=\lim\limits_{n\to\infty}y_n$.

Solution: Note that $$4x_ny_n\le(x_n+y_n)^2$$ Hence, $$\frac{y_{n+1}}{x_{n+1}}=\frac{4x_ny_n}{(x_n+y_n)^2}\le 1$$ That is, $\color{crimson}{x_n\le y_n}$. Also, $$x_{n+1}-x_n=\frac{x_n+y_n}2-x_n=\frac{y_n-x_n}2\ge 0$$ implying that $\{x_n\}$ is monotonically increasing. Similarly, $$\frac{y_{n+1}}{y_n}=\frac{2x_n}{x_n+y_n}\le\frac{2x_n}{2x_n}=1$$ implying that $\{y_n\}$ is monotonically decreasing. Now we have $$x_1\le x_2\le\cdot\le x_n\le\cdot\le y_n\le\cdots\le y_2\le y_1$$ By the monotone convergence theorem, it follows immediately that both $\{x_n\}$ and $\{y_n\}$ are convergent. Let $$\lim_{n\to\infty}x_n=x,\lim_{n\to\infty}y_n=y$$ Taking limits on both sides of $x_{n+1}$ will yield that $$x=\frac{x+y}2$$ which implies that $\lim\limits_{n\to\infty}x_n=x=y=\lim\limits_{n\to\infty}y_n$

(Original image here.)

I have a question about how I can know that $x_n\le y_n$ (in red above).

I know that $y_{n+1} < x_{n+1}$, but how do I know $x_n\le y_n$?

Also, what is meant by ‘Taking limits on both sides of $x_{n+1}$’?

(Questioned parts highlighted in original image.)

  • $\begingroup$ What are $\;x_1,\,y_1\;$ ? Clearly the red part depends on this. Is there any more information you're withholding? $\endgroup$ – DonAntonio Oct 21 '16 at 15:13
  • $\begingroup$ This seems to be arithmetic-harmonic mean. $\endgroup$ – Martin Sleziak Oct 21 '16 at 15:24
  • $\begingroup$ @DonAntonio: We don’t need $x_1$ and $y_1$ for the theorem; the red inequality is backwards and needs to be restricted to $n\ge 2$, that’s all. $\endgroup$ – Brian M. Scott Oct 21 '16 at 15:24
  • $\begingroup$ @BrianM.Scott I'm not sure. What is shown there is $\;\cfrac{y_{n+1}}{x_{n+1}}\le 1\;$ , and from here they get (in red) $\; x_n\le y_n\;$ ...as if $\;x_n\;$ is negative. $\endgroup$ – DonAntonio Oct 21 '16 at 15:36
  • $\begingroup$ @DonAntonio: I think that that’s simply a mistake. However, you raise a good point anyway: the argument that I think was intended appears to rely on having $x_{n+1}>0$, which in turn does mean that they probably intended to require that $x_1\ge 0$ and $y_1>0$. $\endgroup$ – Brian M. Scott Oct 21 '16 at 15:45

The solution is simply wrong. The first step actually shows that $y_n\le x_n$ for $n\ge 2$. (Note that we do have to rule out $n=1$, since it’s possible that the initial values $x_1$ and $y_1$ don’t satisfy this inequality.)

Added: In fact to draw this conclusion we need to know that $x_{n+1}>0$, and at other points in the argument there appears to be an implicit assumption that $x_n$ and $y_n$ are positive. This will be the case if we add the assumption that $x_1\ge 0$ and $y_1>0$. A similar argument works if $x_n$ and $y_n$ are always both negative.

The second step should then be

$$x_{n+1}-x_n=\frac{x_n+y_n}2-x_n=\frac{y_n-x_n}2\le 0\;,$$

so that $\{x_n\}$ is monotonically decreasing. The next step then becomes


so that $\{y_n\}$ is monotonically increasing, and we have

$$y_1\le y_2\le\ldots y_n\le\ldots\le x_n\le\ldots x_2\le x_1\;.$$

It still follows from the monotone convergence theorem that both sequences are convergent, and we can let their limits be $x$ and $y$, as in the given solution. The final step is also correct, but I’ll expand on it to make clearer just what is being done. The point is that


so that


Thus, $2x=x+y$, and $x=y$.


This sequence is known under the name "arithmetic-harmonic mean". See (Proof the the Arithmetic-Harmonic Mean is expressible as the Geometric Mean) where it is proven that these sequences have a common limit which is the geometrical mean of the initial terms: $x=y=\sqrt{x_1y_1}$.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.