-1

I want to prove that there exists a number such that $$\lim_{h\to 0} \frac{x^h-1}{h}=1$$

So to prove it you can say let $$f(x) = \lim_{h\to 0} \frac{x^h-1}{h}$$ Then by interchange of limit and differentiation operator $f'(x)= 1/x$.

My first question is :How to somehow show you can interchange limit and differentiation operator in this case?I know that who can not exchange them always .

Then you can say $$f(x)= \int_1^x\frac1x\, dx $$ Now you need to show this integral increases without bound. It is easy to see it is continuous on the interval $(0,\infty]$ since $1/x$ is continuous on that interval. And so, if it increase without bound, it must take value $1$.

How to show it increases without bound? And is my above proof correct?

Vivaan Daga
  • 5,531
  • First of all, can you prove that the limit defining $f(x)$ always exists? – Angina Seng Jun 15 '20 at 08:26
  • @AnginaSeng it is continuous – Vivaan Daga Jun 15 '20 at 08:30
  • If you know $f$ is continuous, all you need is that there are $x_0$ and $x_1$ with $f(x_0)<1<f(x_1)$. Obviously, you can take $x_0=0$. – Angina Seng Jun 15 '20 at 08:39
  • @AnginaSeng how can you show a upper bound? – Vivaan Daga Jun 15 '20 at 08:46
  • First show that $f$ is defined in $(0,\infty) $. This involves a definition of $x^h, x>0$ which is independent of logarithm. Next show that $f(xy) =f(x) +f(y) $ and then finally $f(x) \leq x-1$. From these properties you can conclude that $f$ is strictly increasing and a bijection from $(0,\infty)$ to $\mathbb {R} $. These properties characterize logarithm uniquely. – Paramanand Singh Jun 15 '20 at 08:55
  • See https://math.stackexchange.com/a/1735035/72031 – Paramanand Singh Jun 15 '20 at 09:01
  • @VivaanDaga I'd try using Bernoulli's inequality, – Angina Seng Jun 15 '20 at 09:28
  • @ParamanandSingh Ok but how do you explain interchange of limit and differentiation operator? – Vivaan Daga Jun 16 '20 at 11:28
  • @VivaanDaga: one does not show the interchange of limit and differentiation but rather one establishes $f(x) \leq x - 1$ via inequalities and then deduces $f'(x) =1/x$. – Paramanand Singh Jun 16 '20 at 11:36
  • 1
    For $f(x) \leq x - 1$ see inequality $(8)$ in this answer. – Paramanand Singh Jun 16 '20 at 11:43
  • @ParamanandSingh how did you deduce f’ =$1/x$ – Vivaan Daga Jun 16 '20 at 12:01
  • 1
    Use definition of derivative and you will find that you need to prove $\lim_{x\to 1}\dfrac{f(x)}{x-1}=1$ in order to show $f'(x) =1/x$. We have $f(x) \leq x - 1$ and replacing $x$ by $1/x$ we get $f(x) \geq \dfrac{x-1}{x}$. Use squeeze theorem on $$\frac{1}{x}\leq\frac{f(x)}{x-1}\leq 1$$ – Paramanand Singh Jun 16 '20 at 12:09
  • @ParamanandSingh as another way could you somehow use Monotone convergence thm for exchanging limit. Also can you please expand on how using def of derivative and getting the limit equal one is enough to show f’(x)=1/x – Vivaan Daga Jun 16 '20 at 14:45
  • It is best that you try to write the definition of derivative and apply it on $f$. Its a trivial part of the overall proof. – Paramanand Singh Jun 16 '20 at 15:00
  • @ParamanandSingh ok but is there any advanced theorem eg.MCT or something that can do this faster? – Vivaan Daga Jun 16 '20 at 15:19
  • I think this is already simpler and fast. I don't know much about the general theorems like mct. – Paramanand Singh Jun 16 '20 at 16:12
  • @ParamanandSingh ok is this a better defintion than just defining log to be the integral because it is more motivated or not? – Vivaan Daga Jun 16 '20 at 16:33
  • It is best to all multiple approaches to the theory of logarithmic and exponential functions. Each has its pros and cons. But in general the more intuitive approaches are formally the difficult ones. See typical approaches as described in my blog posts starting with this one. – Paramanand Singh Jun 16 '20 at 16:49
  • @ParamanandSingh but don’t you still need to prove it exists before saying its derivative is 1/x what about that? – Vivaan Daga Jun 17 '20 at 04:58
  • @ParamanandSingh don’t you need to prove the limit exists to take derivative or am I missing something – Vivaan Daga Jun 17 '20 at 09:05
  • If you are talking the limit in question then you are right. You need to prove that for every $x>0$ the limit in question exists and hence defines a function $f(x) $. This is the first step and then you proceed further. Existence of $f(x) $ can be proved by showing that $(x^h-1)/x$ decreases as $h$ decreases and is also bounded. See the linked answer in my previous comments which deals with inequalities. – Paramanand Singh Jun 17 '20 at 10:04

4 Answers4

2

A non-constructive proof:

Let $$f(x):=\lim_{h\to0}\dfrac{x^h-1}h.$$ We will prove that $f$ is continuous and $f(x)$ straddles $1$.


By the generalized binomial development, $$f(x+\delta)-f(x)=\lim_{h\to0}\dfrac{(x+\delta)^h-x^h}h \\=\lim_{h\to0}x^h\dfrac{1+h\delta+h(h-1)\dfrac{\delta^2}2+h(h-1)(h-2)\dfrac{\delta^3}{3!}+\cdots-1}h \\=\lim_{h\to0}\left(\delta+(h-1)\frac{\delta^2}2+(h-1)(h-2)\frac{\delta^3}{3!}+\cdots\right)<\delta $$

for $\delta$ is sufficiently small (the sum converges when $|\delta|<1$).

Hence for any $\epsilon$, we can find $\delta$ such that

$$|f(x+\delta)-f(x)|<\epsilon$$ and this proves the continuity of $f$.


Next, by the binomial theorem, $$2n^n<(n+1)^n=n^n+n^n+(n-1)\frac{n^{n-1}}2+(n-1)(n-2)\frac{n^{n-2}}{3!}+\cdots<n^n\left(1+1+\frac12+\frac1{3!}+\cdots\right)<3n^n$$ (last step because $n!\ge 2^n$).

Hence,

$$2<\left(1+\frac1n\right)^n<3$$

and

$$n(2^{1/n}-1)<1,\\n(3^{1/n}-1)>1.$$

Hence by the IVT, there must be an $x\in(2,3)$ such that $f(x)=1.$


Final note:

If we want, can refine the value of $x$ such that $f(x)=1$ by a tighter bracketing of

$$\left(\frac{n+1}n\right)^n.$$

An upper bound is

$$1+1+\frac12+\frac1{3!}+\cdots$$

Remains to check if it is tight.

  • Why can’t you just prove it is continuous because it is diffrentiable you can switch limit and differentiation operator in this case (uniform convergence) – Vivaan Daga Jun 15 '20 at 10:56
  • @VivaanDaga: because I don't want to take differentiability for granted. If you accept that, why not use that the limit is well-know to be $\log(x)$ and $\log(e)=1$ ? –  Jun 15 '20 at 10:59
  • log(x) is not defined without e I don’t understand – Vivaan Daga Jun 15 '20 at 11:00
  • Hence,

    $$2<\left(1+\frac1n\right)^n<3$$

    and

    $$n(2^{1/n}-1)<1,\n(3^{1/n}-1)>1.$$

    – Vivaan Daga Jun 15 '20 at 11:18
  • Should’nt you have a bound for f(x) how do you make that connection form the above? – Vivaan Daga Jun 15 '20 at 11:20
  • @VivaanDaga: I don't understand your comments. –  Jun 15 '20 at 11:36
  • How do the above bounds imply that you can use IVT for f(x) – Vivaan Daga Jun 15 '20 at 11:41
  • @VivaanDaga: rewrite with $h=1/n$ and take the limit. –  Jun 15 '20 at 11:42
  • Ok ,i understand the argument .As another argument to demonstrate the same fact could you convert the integral in my question into a riemann sum and then show that is unbounded with f(1)=0 to show its unbounded compare it to harmonic series . Can you please tell me would such an argument work. – Vivaan Daga Jun 15 '20 at 13:17
  • @VivaanDaga: sorry, no. –  Jun 15 '20 at 13:20
  • https://math.stackexchange.com/questions/2489317/proof-that-the-number-e-exists-based-on-the-following-definition but it says so here – Vivaan Daga Jun 15 '20 at 13:32
  • is there any way to prove an integral increases without bound somehow? without evaluating it – Vivaan Daga Jun 15 '20 at 13:36
1

Let $$ \alpha_n=\left(1+\frac1n\right)^n\quad\text{and}\quad\beta_n=\left(1+\frac1n\right)^{n+1}\tag1 $$ Then, for non-zero $|h|\le\frac1n$, Bernoulli's Inequality says $$ \begin{align} \frac{\alpha_n^h-1}h &=\frac{\left(1+\frac1n\right)^{nh}-1}h\\[3pt] &\le\frac{(1+h)-1}h\\[6pt] &=1\tag2 \end{align} $$ Note that for $h\lt0$, the sense of the inequality in the numerator changes, but then the negative denominator restores the sense. Furthermore, $$ \begin{align} \frac{\alpha_n^h-1}h &=\frac{\left(1-\frac1{n+1}\right)^{-nh}-1}h\\[3pt] &\ge\frac{\left(1+h\frac{n}{n+1}\right)-1}h\\[6pt] &=\frac{n}{n+1}\tag3 \end{align} $$ Thus, $$ \bbox[5px,border:2px solid #C0A000]{\frac{n}{n+1}\le\frac{\alpha_n^h-1}h\le1}\tag4 $$ Similarly, for non-zero $|h|\le\frac1{n+1}$ $$ \begin{align} \frac{\beta_n^h-1}h &=\frac{\left(1+\frac1n\right)^{(n+1)h}-1}h\\[3pt] &\le\frac{\left(1+h\frac{n+1}n\right)-1}h\\[6pt] &=\frac{n+1}n\tag5 \end{align} $$ and $$ \begin{align} \frac{\beta_n^h-1}h &=\frac{\left(1-\frac1{n+1}\right)^{-(n+1)h}-1}h\\[3pt] &\ge\frac{(1+h)-1}h\\[6pt] &=1\tag6 \end{align} $$ Thus, $$ \bbox[5px,border:2px solid #C0A000]{1\le\frac{\beta_n^h-1}h\le\frac{n+1}n}\tag7 $$ Since $\frac{x^h-1}h$ is monotonically increasing in $x$ for non-zero $|h|\lt\frac1{n+1}$, $(4)$ and $(7)$ say that for all $x\in[\alpha_n,\beta_n]$ and non-zero $|h|\lt\frac1{n+1}$, $$ \frac{n}{n+1}\le\frac{x^h-1}h\le\frac{n+1}n\tag8 $$ As is shown in this answer, $\alpha_n$ is increasing and $\beta_n$ is decreasing. Since $[\alpha_n,\beta_n]$ is a decreasing nested sequence of non-empty compact sets, Cantor's Intersection Theorem says that $$ \bigcap_{n=1}^\infty\,[\alpha_n,\beta_n]\ne\emptyset\tag9 $$ In fact, since $\beta_n-\alpha_n=\frac1{n+1}\beta_n$ decreases to $0$, the intersection in $(9)$ consists of one point, called $e$.

Thus, we have that for non-zero $|h|\lt\frac1{n+1}$ $$ \frac{n}{n+1}\le\frac{e^h-1}h\le\frac{n+1}n\tag{10} $$ Inequality $(10)$ and The Squeeze Theorem imply $$ \lim_{h\to0}\frac{e^h-1}{h}=1\tag{11} $$

robjohn
  • 345,667
  • Another way to prove the existence can be prove $lim_{h\to 0} x^h-1/h$ exists for all $x>0$ this may be prove by the fact $a^x$ is monotonic and continuous and that a monotonic continuous function has to be differentiable almost everywhere and by $a^{(x+y)}$ = $a^xa^y$ it means that if it is differentiable at point it is differentiable everywhere the limit function then will come form the definition of the derivative.Then prove the limit function is continuous and unbounded hence by IVT it is equal to $1$ for a number $e$ Then d/dx $e^x$ =$e^x$ . – Vivaan Daga Aug 06 '20 at 09:56
  • 1
    This has nothing to do with this question. You might want to start another question for this. – robjohn Aug 13 '20 at 13:21
  • Using this limit entire theory of log and exp can be developed from that can it be somehow proven that Bernoulli inequality will be strict ? – Vivaan Daga Aug 13 '20 at 13:27
  • 1
    Please start another question. Please do not start extensive unrelated comment threads. – robjohn Aug 13 '20 at 15:18
  • it can be proved using the fact that the tangent of a monotone function only touches the curve at one point the equation of the tangent at $x=0$ is equal to $1+nx$ and there is equality at $x=0$ For any n since at rationals x >0 and fixed n > 1 the function is greater then for reals also has to be greater than. – Vivaan Daga Aug 13 '20 at 16:01
  • "the tangent of a monotone function only touches the curve at one point" - false. It is true for a strictly convex or strictly concave function. – robjohn Aug 13 '20 at 18:21
  • it does not matter . $f’(x)$ is strictly increasing . – Vivaan Daga Aug 14 '20 at 05:56
  • Umm... that says $f$ is strictly convex. – robjohn Aug 14 '20 at 06:29
  • How does $$1-1/n <lim sup x^h-1/h <1 $$ imply that the middle term will be one ? – Vivaan Daga Aug 17 '20 at 05:35
  • I have reworked the proof to remove the use of limsup and liminf so that it is simpler to follow. – robjohn Aug 17 '20 at 22:21
  • +1 surely...why is there the last boxed formula is open? I have seen that this happen often. – Sebastiano Aug 17 '20 at 22:34
  • @Sebastiano: are you talking about this formula? – robjohn Aug 17 '20 at 22:40
  • How does $\alpha_{n}-\beta_{n}\to 0 $ imply that there will only be one point? – Vivaan Daga Aug 18 '20 at 14:12
  • Yes of course. But I now see the formula correctly. Why? Thank you very much for your reply. – Sebastiano Aug 18 '20 at 14:18
  • 1
    @Sebastiano: That is what I see. I don't know why it is different on your system. – robjohn Aug 18 '20 at 16:27
  • @VivaanDaga: If $\alpha_n$ is increasing and $\beta_n$ is decreasing and $\beta_n-\alpha_n\to0$, how many points could be in $\bigcap\limits_{n=1}^\infty[\alpha_n,\beta_n]$? If there were two points, how far apart could they be? – robjohn Aug 18 '20 at 16:30
  • I know this is a shot in the dark, but if there is something wrong with what I did, would the downvoter, or anyone, please point it out? – robjohn Apr 04 '22 at 21:55
0

For $x>0$ fix, let $g(t):=x^t= e^{t \ln x}.$

Then $g'(t)=x^t \cdot \ln x.$ Thus $g'(0)= \ln x.$

On the other hand, $ \lim_{h\to 0} \frac{x^h-1}{h}= \lim_{h \to 0}\frac{g(h)-g(0)}{h-0}=g'(0).$

Hence

$$ \lim_{h\to 0} \frac{x^h-1}{h}=1 \iff \ln x=1 \iff x=e.$$

Consequence: $g(t)=e^t.$

Fred
  • 77,394
0

If $\frac {x^h - 1}{h} = 1$ then $x = (1+h)^\frac {1}{h}$

Now taking the limit as $h$ approaches $0,$ will give us the value we seek for $x$

$x = \lim_\limits{h\to 0} (1+h)^\frac 1h = \lim_\limits{n\to \infty} (1+\frac {1}{n})^n = e$

by definition.

Actually, if you only need to prove existence,

$f(x)$ is continuous.

$f(1) = 0$

$f(4) = \lim_\limits{n\to \infty} \frac {4^\frac 1n - 1}{\frac 1n}$

Multiply numerator and denominator by $(4^{1-\frac {1}{n}} + 4^{1-\frac {2}{n}} + \cdots + 4^{1-\frac {n-1}{n}} + 1)$

$f(4) = \lim_\limits{n\to \infty} \frac {4 - 1}{\frac 1n(4^{1-\frac {1}{n}} + 4^{1-\frac {2}{n}} + \cdots + 4^{1-\frac {n-1}{n}} + 1)}$

$4^\frac 12 = 2$

If ${1-\frac {k}{n}} \le \frac 12$ then $4^{1-\frac {k}{n}} < 2$

and if $\frac 12 < {1-\frac {k}{n}} < 1$ then $4^{1-\frac {k}{n}} < 4$

$(4^{1-\frac {1}{n}} + 4^{1-\frac {2}{n}} + \cdots + 4^{1-\frac {n-1}{n}} + 1) < \frac n2\cdot 4 + \frac n2 \cdot 2 = 3n$

$f(4) > \lim_\limits{n\to\infty}\frac {3}{(\frac 1n) 3n} = 1$

By the IVT, there exists $1<x<4$ such that $f(x) = 1$

Doug M
  • 57,877