# Prove that the limit exists and then find its limit

We are given a function $h(x)$ which strictly positive. The function $h$ is defined on $\mathbb{R}$ and that satisfies the following:\

$$\lim_{x\to 0}(h(x)+\frac{1}{h(x)})=2$$

The question is to prove that the limit of $h$ exists at$0$ and then find its limit as $x\to 0$

If we assume that the limit exists (say it's equal to $\lambda$, then we have $\lambda+\frac{1}{\lambda}=2$ and by solving this equation, we get $\lambda=1$. However, I am completely clueless on how to prove the existence of the limit of $h(x)$ when $x\to 0$.

Let $g(x)=\sqrt{h(x)}$. We have $$h(x)-\frac{1}{h(x)}-2=\left(g(x)-\frac{1}{g(x)}\right)^2.$$

This has limit $0$, so $g(x)-\frac{1}{g(x)}$ has limit $0$. We show that $g(x)$ has limit $1$. Note that $|g(x)|$ is bounded above, so $g^2(x)-1$ has limit $0$. Since $g(x)+1\gt 1$, it follows that $g(x)-1$ has limit $0$.

We have $h(x) +\dfrac 1{h(x)} \to 2$, so $$f(x) :=h(x) +\dfrac 1{h(x)}-2=\frac{h(x)^2 -2h(x) + 1}{h(x)}=\frac{(h(x) - 1)^2}{h(x)}\to 0$$as $x \to 0$.

Show that for $x$ sufficiently small $h(x)$, we have $\frac12<h(x) < 2$ (any upper bound will work). So $$f(x)>\frac12\big(h(x)-1\big)^2$$

Now can you use the squeeze theorem to derive the result?

• Since $(h - 1)^{2}/h \to 0$ and $h$ is bounded we can say that $(h - 1)^{2} \to 0$ and then $h \to 1$. – Paramanand Singh Sep 9 '14 at 6:32

Suppose there is a sequence $(x_n)$ such that $x_n \to 0$ but $h(x_n) \to 1$ is not true.

Then there is a $\delta > 0$ such that either,

a) $h(x_n) > 1 + \delta$ for infinitely many values of $n$,or,

b) $0 < h(x_n) < \dfrac{1}{1+\delta}$ for infinitely many values of n.

If b) holds assume (by choosing a subsequence if necessary) that $0 < h(x_n) < \dfrac{1}{1+\delta}$ for all $n$.

The boundedness of $h(x_n)$ implies we can find a subsequence $(y_n)$ of $(x_n)$ such that $\lim h(y_n) = L$ exists and $0 \leq L \leq \dfrac{1}{1+\delta} < 1$.

$L \neq 0$ as a 0 limit will contradict $h(y_n) + \dfrac{1}{h(y_n)} \to 2$ a consequence of the given condition.

But this implies $\dfrac{1}{h(y_n)}$ also converges to $\dfrac{1}{L}$ with $L + \dfrac{1}{L} = 2$ and $L = 1$, a contradiction to $L < 1$.

If $h(x_n) > 1 + \delta$ infinitely often, then $0 < \dfrac{1}{h(x_n)} < 1/(1+\delta)$ infinitely often, and the argument proceeds similarly to obtain a contradiction.

Hence, for any $x_n \to 0$ $h(x_n) \to 1$, hence $\lim h(x) = 1$ as $x \to 0.$

Let $g(x) = x + x^{-1}$. Then your limit is

$$\lim_{x \to 0} g(h(x)) = 2$$

$g$ isn't quite an invertible function: for every point $a \in (2, \infty)$ there are two solutions to $g(x) = a$. But in some sense it's a continuous two-valued 'function': if I let $f^+$ and $f_-$ be the larger and smaller of the two solutions, then they are both continuous functions:

$$\lim_{x \to 0} f_+(g(h(x))) = f_+(2) = 1$$ $$\lim_{x \to 0} f_-(g(h(x))) = f_-(2) = 1$$

So in some sense, we whould be able to say

$$\lim_{x \to 0} g^{-1}(g(h(x)) = g^{-1}(2) = \{1\}$$ $$\lim_{x \to 0} \{ h(x), h(x)^{-1} \} = \{1\}$$

and since we just get one limit point, the individual limits should exist (i.e. $h(x) \to 1$). All that's left is finding a way to make a precisely statement that captures this idea, and then prove the theorem saying we can find limits like this. This is a fair bit of work, but a good idea in the long run since it gives us a new tool useful for dealing with more problems!

• I'm not sure I understand. We haven't assumed anything about the continuity of $f$ here – Mathmo123 Sep 9 '14 at 12:24

Note that if either $h(x)$ or $\frac{1}{h(x)}$ is not bounded in $(-\varepsilon, \varepsilon)$ then the original limit cannot exist. We'll use that property further, refering it as $(*)$

$$\lim_{x\to 0}\left(h(x) + \frac{1}{h(x)}\right) = 2 \Leftrightarrow \lim_{x\to 0}\left(h(x) -2 + \frac{1}{h(x)}\right) = 0 \Leftrightarrow \lim_{x\to 0}\left(\sqrt{h(x)} - \frac{1}{\sqrt{h(x)}}\right)^2 = 0 \Leftrightarrow \lim_{x\to 0}\left(\frac{h(x) - 1}{\sqrt{h(x)}}\right)^2 = 0 \Leftrightarrow \lim_{x\to 0}\frac{(h(x) - 1)^2}{h(x)} = 0 \stackrel{(*)}{\Rightarrow} \lim_{x\to 0}(h(x)-1)^2 = 0$$

Wich means that $$\lim_{x\to 0} h(x) = 1$$

• The last $\iff$ requires that you know that both the numerator and denominator have limits. But that the denominator has limit is what you're trying to prove! – Kim Jong Un Sep 9 '14 at 2:09
• You are saying that $\lim\frac fg = \frac{\lim f}{\lim g}$. Why doesn't this assume the convergence of $f$ and $g$? – Mathmo123 Sep 9 '14 at 2:10
• If $h(x) \to \pm\infty$ as $x \to 0$ then $\lim \left( h(x) + \frac{1}{h(x)} \right)$ does not exist. This says nothing about other kinds of divergence. – user164587 Sep 9 '14 at 2:22
• @user164587 But that very limit exists by hypothesis of the problem. – Darth Geek Sep 9 '14 at 2:23
• That's my point! $h(x)$ is bounded. – user164587 Sep 9 '14 at 2:37