8
$\begingroup$

If $$\lim_{x\to\infty}(f(x)+f'(x))=l$$ then prove that $$\lim_{x\to\infty}f(x)=l \text{ and } \lim_{x\to\infty}f'(x)=0 $$


I assume four cases $$\begin{array}{c|c|c|c|} & f(x) & f'(x) \\ \hline \text{1} & \infty & -\infty \\ \hline \text{2} & -\infty & \infty \\ \hline \text{3} & l & 0 \\ \hline \text{4} & 0 & l \\ \hline \end{array}$$ and elimination(1,2,4) of not possible cases can give the answer(3).


My work is not a correct/perfect or flaw proof.What would be a correct one or is this correct.

$\endgroup$
  • 1
    $\begingroup$ Why would you assume only those four cases and no others? $\endgroup$ – Michael Hardy Jun 20 '14 at 18:13
  • $\begingroup$ You are correct and also my note reminds everyone of the same $\endgroup$ – RE60K Jun 20 '14 at 18:14
  • 3
    $\begingroup$ Hint: Consider what happens if $\lim_{x\rightarrow\infty}f'(x) = C\neq 0$. What can you say about the asymptotic behavior of $f$? (Hint: Consider the mean-value theorem.) You also need to consider what happens if $f'$ oscillates - i.e. does not have a limit at all. $\endgroup$ – Cameron Williams Jun 20 '14 at 18:14
  • 3
    $\begingroup$ See the accepted answer to this. $\endgroup$ – David Mitra Jun 20 '14 at 18:17
  • $\begingroup$ @CameronWilliams do you mean that if f' doesn't approach zero the value of the limit wouldn't exist.thus f'x$\to$0 and fx$\to$l $\endgroup$ – RE60K Jun 20 '14 at 18:55
3
$\begingroup$

This is the solution coming straight from Hardy's Pure Mathematics and is truly beautiful. Before solving this problem Hardy solves a simple problem:

If $f(x) \to L$ and $f'(x) \to L'$ as $x \to \infty$ then $L' = 0$.

This is easy to do via mean value theorem as we have have $f(x) - f(x/2) = (x/2)f'(c)$ for some $c \in (x/2, x)$. If $L' \neq 0$ then we have contradiction as we take limits with $x \to \infty$ in the above relation.

[If $x \to \infty$ then $x/2 \to \infty$ and hence LHS $f(x) - f(x/2) \to L - L = 0$. Again since $x/2 < c < x$ therefore when $x \to \infty$ then $c$ also tends to $\infty$ and hence $f'(c) \to L'$. Then the RHS $(x/2)f'(c)$ tends to $(x/2)L'$ i.e. to $+\infty$ if $L' > 0$ and to $-\infty$ if $L' < 0$. Hence there is no choice for $L'$ except that $L' = 0$.]

Next we attack the current problem. We are given that $f(x) + f'(x) \to L$ when $x \to \infty$ and we need to show that $f(x) \to L$ and $f'(x) \to 0$ as $x \to \infty$. Hardy simplifies the problem by setting $f(x) = \phi(x) + L$ so that $f'(x) = \phi'(x)$ and $\phi(x) + \phi'(x) \to 0$ as $x \to \infty$.

[Hardy defines a new function $\phi(x) = f(x) - L$ and then $f(x) + f'(x) = \phi(x) + L + \phi'(x) \to L$ and cancel $L$ from both sides to get $\phi(x) + \phi'(x) \to 0$].

To solve the problem we need to show that both $\phi(x)$ and $\phi'(x)$ tend to $0$ as $x \to \infty$.

[If we show $\phi(x) \to 0$ then $f(x) = L + \phi(x) \to L$ and $f'(x) = \phi'(x) \to 0$ so that we get the original version of the problem.]

Now the argument by Hardy is beautiful and I quote him verbatim:

"If $\phi'(x)$ is of constant sign, say positive, for all sufficiently large values of $x$, then $\phi(x)$ steadily increases and must tend to a limit $A$ or to $\infty$.

[If the derivative $\phi'(x)$ is positive then original function $\phi(x)$ is strictly increasing for all large values of $x$. And there is a very standard theorem that if a function $\phi(x)$ is increasing then $\phi(x)$ tends to a limit or to $\infty$ as $x \to \infty$.]

If $\phi(x) \to \infty$ then $\phi'(x) \to -\infty$ [because $\phi(x) + \phi'(x) \to 0$] which contradicts our hypothesis [we have assumed that $\phi'(x) > 0$ for large values of $x$ and this is incompatible with $\phi'(x) \to -\infty$].

If $\phi(x) \to A$ then $\phi'(x) \to -A$ [because $\phi(x) + \phi'(x) \to 0$] and this is impossible unless $A = 0$ [from our previous result mentioned in beginning of answer].

Similarly we may dispose of the case in which $\phi'(x)$ is ultimately negative.

[If $\phi'(x)$ is negative then $\phi(x)$ is decreasing for large $x$. Now there is again a very standard theorem which says that if $\phi(x)$ is decreasing for all large $x$ then it either tends to a limit or to $-\infty$ as $x \to \infty$. If $\phi(x) \to -\infty$ then because of the relation $\phi(x) + \phi'(x) \to 0$ we get that $\phi'(x) \to \infty$ and this is incompatible with the fact that $\phi'(x)$ is negative for large $x$. If on the other hand we have $\phi(x) \to B$ then again $\phi'(x) \to -B$ and by the result mentioned in the beginning of answer this is possible only when $B = 0$.]

If $\phi'(x)$ changes sign for values of $x$ which surpass all limit, then these are the maxima and minima of $\phi(x)$.

[Note that a when a derivative changes sign it must also vanish somewhere in between and hence we will obtain points where $\phi'(x) = 0$ and before and after this point the derivative $\phi'(x)$ is of opposite signs. Such points are the maxima and minima of $\phi(x)$ and $\phi(x)$ takes local minimum and maximum values at these points. Since $\phi'(x)$ changes sign for values $x$ which go beyond any limit, this will lead to infinitely many points $x$ which are maxima or minima of $\phi(x)$.]

If $x$ has a large value corresponding to a maximum or a minimum of $\phi(x)$, then $\phi(x) + \phi'(x)$ is small [because it tends to zero] and $\phi'(x) = 0$ [because derivative vanishes at points of maxima/minima], so that $\phi(x)$ is small. A fortiori the other values of $\phi(x)$ are small when $x$ is large. [We see that the maximum and minimum values of $\phi(x)$ are small and hence all the other intermediate values of $\phi(x)$ are also small so that for large $x$ all values of $\phi(x)$ are small. And thus $\phi(x) \to 0$ in this case also.]"

Thus in all cases $\phi(x) \to 0$ and hence $\phi'(x) \to 0$.

Update: Based on comment from OP I add details inline enclosed in [].

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ It took me a long time figuring it out from your solution and am still a little confused; can you please explain a liitle bit more, for I am on the level of highschool mathematics. $\endgroup$ – RE60K Jun 21 '14 at 5:42
  • $\begingroup$ @Aditya: I have tried to explain the answer in detail by putting additional explanation in []. However you will need to know some standard results from theory of differential calculus like "increasing / decreasing nature of functions based on sign of derivative", "mean value theorem", "maxima/minima". $\endgroup$ – Paramanand Singh Jun 21 '14 at 6:38
4
$\begingroup$

Lemma Let $h:[a,+\infty)\to\mathbb{R}$ be a continuous function such that $\lim_{x\to\infty}h(x)=0$, Let $F$ be defined by $$F(x)=e^{-x}\int_a^xe^th(t)dt$$ Then $\lim_{x\to\infty}F(x)=0$.

Proof. Indeed, for $x>b>a$ we have $$\eqalign{ |F(x)|&=\left|e^{-x}\int_a^be^th(t)dt+e^{-x}\int_b^xe^th(t)dt\right|\cr &\leq e^{-x}\max_{x\in[a,b]}{(e^t|h(t)|)}+\left(\sup_{t\in[b,x]}|f(t)|\right)e^{-x}\int_b^xe^tdt\cr &\leq e^{-x}\max_{x\in[a,b]}{(e^t|h(t)|)}+\left(\sup_{t\in[b,x]}|f(t)|\right) }\tag{1} $$ Now, for $\epsilon>0$, there is $b>a$ such
that $$ \forall\,t>b,\qquad |f(t)|<\frac{\epsilon}{2}\tag{2} $$ With $b$ fixed as above, there is $x_0>b$ such that $$ \forall\,x>x_0,\qquad e^{-x}\max_{x\in[a,b]}{(e^t|h(t)|)}\leq\frac{\epsilon}{2}\tag{3} $$ combining $(1)$ and $(2)$ in $(1)$ we see that $|F(x)|<\epsilon$ for $x>x_0$. This proves the lemma since $\epsilon>0$ is arbitrary.$\qquad\square$

Now, let us apply the lemma to $h(x)=f(x)+f'(x)-\ell$. We have $$F(x)=e^{-x}\int_a^x((e^tf(t))'-\ell e^t)dt=f(x)-e^{-x+a}f(a)-\ell(1- e^{a-x})$$ By the Lemma $\lim_{x\to\infty}F(x)=0$ that is $\lim_{x\to\infty}f(x)=\ell$. and the desired conclusion follows.$\qquad\square$

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ You seem to use that $f'(x)$ is continuous. Is it possible to remove this limitation? $\endgroup$ – Paramanand Singh Jun 21 '14 at 17:27
  • $\begingroup$ The lemma can be proved for $h:[a,+\infty)\to\mathbb{R}$ measurable instead of continuous, you just use the Dominated Convergence Theorem to prove it. So, this version is just a simplification of the proof. $\endgroup$ – Omran Kouba Jun 21 '14 at 21:13
3
$\begingroup$

The problem is easy if one assumes that

$$\lim_{x \to \infty} f'(x) = k$$ exists.

Indeed, in this case, it is easy to prove $k=0$: By L'Hospital we have

$$\lim_{x \to \infty} \frac{f(x)}{x} = k \,.$$

Then, if $k= \pm \infty$ we get $\lim_{x \to \infty} f(x)$ is the same infinity, which contradicts the given condition; while if $k$ is finite and non-zero, we get that $\lim_{x \to \infty} f(x)$ is infinite, which contradicts again the given condition.

Now, since $$\lim_{x \to \infty} f'(x) = 0$$ we get $$\lim_{x \to \infty} f(x) =\left( \lim_{x \to \infty} f(x)+f'(x)\right) -\lim_{x \to \infty} f'(x)=l $$

To get around the existence of the limit $\lim_{x \to \infty} f'(x)$, I would try to look to $\limsup f'(x)$ and $\liminf f'(x)$.

| cite | improve this answer | |
$\endgroup$
2
$\begingroup$

If one were to assume that both of the limits $\lim_{x\to\infty}f(x) \text{ and } \lim_{x\to\infty}f'(x)$ exist, then the limit of $f'$ is some number $k$. If $k\ne0$, then for large enough $x$ we have $f'(x)$ being further away from $0$ than $k/2$. That would mean that as $x$ increases by $1$ then $f(x)$ changes by at least $k/2$, and always in the same direction. That follows from the mean value theorem. If $f(x)$ changes by more than $k/2$ and always in the same direction, then $f(x)$ is not approaching any limit, and there is a contradiction, so it must be that $k=0$.

The harder problem is showing that if the limit of the sum exists, then the two limits separately exist. Maybe I'll come back to that . . . . . .

| cite | improve this answer | |
$\endgroup$
1
$\begingroup$

Given: $$f'(x) = \lim_{h\to 0} \frac{f(x+h)-f(x)}{h}$$

$$l = \lim_{x\to \infty} f(x) + f'(x) = \lim_{x\to \infty}\left[ f(x) + \left[\lim_{h\to 0}\frac{f(x+h)-f(x)}{h}\right]\right]$$

$$l = \lim_{x\to\infty}\left[\lim_{h\to 0}\frac{f(x+h)+f(x)(h-1)}{h}\right]$$

Assuming $f(x)$ has a limit, we can write $|f(x+h) - f(x)| \leq \epsilon$ for large enough $x$. This means that we can bound the quantity inside the limits.

$$\frac{hf(x) - \epsilon}{h} \leq \frac{f(x+h)+f(x)(h-1)}{h} \leq \frac{hf(x) + \epsilon}{h}$$

$$f(x) - \frac{\epsilon}{h} \leq \frac{f(x+h)+f(x)(h-1)}{h} \leq f(x) + \frac{\epsilon}{h}$$

For any $|h| > 0$ we can find an $N$ so that for $x > N$, $\,|f(x+h) - f(x)| \leq h^2$. This is equivalent to setting $\epsilon = h^2$.

$$f(x) - h \leq \frac{f(x+h)+f(x)(h-1)}{h} \leq f(x) + h$$

Take the limit as $h\to 0$. Notice that doing this implies that $\epsilon \to 0$ which forces $x \to \infty$.

$$\lim_{x\to\infty} f(x) \leq \lim_{x\to\infty} f(x)+f'(x) \leq \lim_{x\to\infty} f(x) $$

The rest is clear from here.

I'm not going to pretend like this is formal as we essentially interchanged the limit operators but it should give you a good idea why this works.

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ @Aditya it should be better now but I will still take any constructive criticism. I don't see how it can be done without essentially interchanging limits. $\endgroup$ – Brad Jun 20 '14 at 19:02

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.