4
$\begingroup$

Let $f:[0,\infty)\rightarrow R $ be a continuously differentiable function. Show that if

$ \lim_{x \rightarrow \infty} f'(x)<1 $

then

$ f(x_0)<x_0 $ for some $x_0$ large enough. (An example of a function that satisfies these assumption is $f(x) = \sqrt x $). I am struggling with the proof. I've tried with the mean value theorem:

Choose $0<x<y$. Then by the MVT there esists a number $c \in (x,y)$ such that

$ \dfrac{f(y)-f(x)}{y-x} = f'(c) $

But at this point I got stuck... The other info I have is that there exists a number $M>0$ such that if $x>M$ then $f'(x)<1$ (the limit condition stated in the hp). Is there a way to combine these two facts in order to prove what I want?

$\endgroup$
  • 2
    $\begingroup$ I added the "real-analysis" tag to your post. Cheers! $\endgroup$ – Robert Lewis Oct 4 '18 at 16:48
2
$\begingroup$

Take $r\in\left(\lim_{x\to\infty}f'(x),1\right)$. Since $\lim_{x\to\infty}f'(x)<r$, there is a $M>0$ such that $x\geqslant M\implies f'(x)<r$. Then, when $x>M$, by the mean value theorem$$\frac{f(x)-f(M)}{x-M}=f'(c)<r,$$for some $c$ between $M$ and $x$. So,$$x>M\implies f(x)<f(M)+r(x-M).$$But, if $x$ is large enough, we have $f(M)+r(x-M)<x$, since $0<r<1$, and therefore, if $x$ is large enough, we have $f(x)<x$.

$\endgroup$
  • $\begingroup$ The first two sentences of your answer make no sense. $\lim f > r$ no given in the problem, also how you find the $M$ ? $\endgroup$ – Mikey Spivak Oct 4 '18 at 16:51
  • 1
    $\begingroup$ @Neymar THere was a typo in the second sentence: I wrote $f$ intead of $f'$. What's the problem with the first sentence? Since $\lim_{x\to\infty}f'(x)<1$, I can consider the interval $\left(\lim_{x\to\infty}f'(x),1\right)$. $\endgroup$ – José Carlos Santos Oct 4 '18 at 16:57
  • $\begingroup$ i dont understand really its too terse the way you write $\endgroup$ – Mikey Spivak Oct 4 '18 at 17:43
  • $\begingroup$ Thanks but I don't understand the last statement: "if x is large enough, we have $f(M)+r(x−M)<x$ since $0<r<1$" $\endgroup$ – Alessandro Oct 4 '18 at 17:50
  • 2
    $\begingroup$ Note that$$f(M)+r(x-M)<x\iff f(M)-rM<(1-r)x,$$which is true if $x$ is large enough. $\endgroup$ – José Carlos Santos Oct 4 '18 at 17:54
2
$\begingroup$

Let $g(x) =f(x) - x$ so that $g'(x) \to L<0$ as $x\to\infty$ and therefore $g'(x) $ is negative as $x\to\infty $. It thus follows that $g(x)$ eventually decreases and hence either tends to a limit $M$ or to $-\infty$. If it tends to $-\infty$ then $g$ is eventually negative and we are done. But if $g(x) \to M$ then by mean value theorem $g'(\xi) = g(x+1)-g(x)\to 0$ and this contradicts that $g'(x) \to L<0$.

Note: There is no need to assume that $f$ is continuously differentiable and we just need the hypotheses that $\lim_{x\to\infty} f'(x) $ exists and this limit is less than $1$.

$\endgroup$
  • $\begingroup$ Paramand.Very nice and clear, as usual +. $\endgroup$ – Peter Szilas Oct 4 '18 at 17:53
1
$\begingroup$

So I think it is safe to say that the hypothesis

$\displaystyle \lim_{x \to \infty} f'(x) < 1 \tag 1$

gives us two important facts concerning the function $f(x)$:

first, that $\lim_{x \to \infty} f'(x)$ actually exists, that is, there is some $L \in \Bbb R$ such that, given any $\epsilon > 0$, there also exists $M_\epsilon \in \Bbb R$ such that

$x \ge M_\epsilon \Longrightarrow \vert f'(x) - L \vert < \epsilon; \tag 2$

second, that in fact

$L < 1; \tag 3$

then

$1 - L > 0, \tag 4$

so we can choose, say

$\epsilon \le \dfrac{1 - L}{2} = \dfrac{1}{2}(1 - L); \tag 5$

then we may write (2) in the form

$x \ge M_\epsilon \Longrightarrow L - \epsilon < f'(x) < L + \epsilon \le L + \dfrac{1}{2}(1 - L) = \dfrac{1}{2}(L + 1) < 1; \tag 6$

the key here is that by taking $x$ sufficiently large, we may ensure $f'(x)$ is bounded by a number, in this case $(1/2)(L + 1)$, which is strictly less than $1$ which, intuitively, shows us that $f(x)$ grows more slowly than $x$ for $x$ large enough.

Now consider the function $x - f(x)$; for $\xi \ge M_\epsilon$ we have

$x - f(x) - (\xi - f(\xi)) = \displaystyle \int_\xi^x (s - f(s))' \; ds$ $=\displaystyle \int_\xi^x (1 - f'(s)) \; ds \ge \int_\xi^x \dfrac{1}{2}(1 - L) \; ds = \dfrac{1}{2}(1 - L)(x - \xi), \tag 7$

or

$x - f(x) \ge \dfrac{1}{2}(1 - L)(x - \xi) +(\xi - f(\xi)) \to \infty \; \text{as} \ x \to \infty \tag 8$

since

$\dfrac{1}{2}(1 - L) > 0, \tag 9$

that is, the right-hand side of (8) describes a line of positive slope; we further see from (8) that this line intersects the $x$-axis at that unique $x$ such that

$\dfrac{1}{2}(1 - L)(x - \xi) +(\xi - f(\xi)) = 0, \tag 9$

i.e., where

$x = -\dfrac{2(\xi - f(\xi))}{1 - L} + \xi; \tag{10}$

therefore

$x > f(x) \tag{11}$

for

$x > -\dfrac{2(\xi - f(\xi))}{1 - L} + \xi. \tag{12}$

Nota Bene: It is not necessary to assume $f(x)$ actually has a limit as $x \to \infty$; scrutiny of the above argument reveals that the desired result, $f(x) < x$ for sufficiently large $x$, binds under the weaker assumption that

$f'(x) \le L < 1, \; \forall x \; \text{"big enough"}. \tag{13}$

End of Note.

$\endgroup$
  • $\begingroup$ +1 for the note at the end. $\endgroup$ – Paramanand Singh Oct 5 '18 at 0:55

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.