How can we prove that $$\lim_{x\rightarrow 0}\cfrac{e^x-1-x}{x^2}=\cfrac{1}{2}$$ Without using L'hopital rule, and Taylor expansions?


  • 4
    $\begingroup$ What is your definition of $e^x$? What can you use about this function? $\endgroup$ – ajotatxe Aug 30 '16 at 14:40
  • 12
    $\begingroup$ There is no pre-calculus definition of $e^{x}$ $\endgroup$ – preferred_anon Aug 30 '16 at 14:47
  • 8
    $\begingroup$ @DanielLittlewood I'd also say that there is no pre-calculus definition of limit... $\endgroup$ – ajotatxe Aug 30 '16 at 14:51
  • 3
    $\begingroup$ It can be worked out if you use $e^x = \lim_{n \to \infty} \left (1 + \frac xn \right)^n$ presuming the limit exists in the first place. $\endgroup$ – Umberto P. Aug 30 '16 at 14:55
  • 1
    $\begingroup$ @ajotatxe we can only use the definition stated in Umberto P. comment $\endgroup$ – An old man in the sea. Aug 30 '16 at 14:56

Use the limit laws and the binomial theorem: you have $$\frac{e^x - (1+x)}{x^2} = \lim_{n \to \infty} \left( \frac{ (1+ \frac xn)^n - (1+x)}{x^2} \right) = \lim_{n \to \infty} \sum_{k=2}^n \binom nk \frac{x^{k-2}}{n^k} \\ = \frac 12 + x \left( \lim_{n \to \infty} \sum_{k=3}^n \binom nk \frac{x^{k-3}}{n^k} \right)$$ provided that the limit $ \displaystyle e^x = \lim_{n \to \infty} \left(1 + \frac xn \right)^n$ is assumed to exist.

As a by-product of this computation you get that $$\lim_{n \to \infty} \sum_{k=3}^n \binom nk \frac{x^{k-3}}{n^k}$$ exists too. With $x=1$ this implies $$\sup_n \sum_{k=3}^n \binom nk \frac{1}{n^k} < \infty$$ and consequently if $|x| \le 1$ then $$\sup_{n} \left| \sum_{k=3}^n \binom nk \frac{x^{k-3}}{n^k} \right| \le \sup_{n} \sum_{k=3}^n \binom nk \frac{|x^{k-3}|}{n^k} \le \sup_n \sum_{k=3}^n \binom nk \frac{1}{n^k} < \infty.$$

So, if $|x| \le 1$ then $$\left| \frac{e^x - (1+x)}{x^2} - \frac 12 \right| \le |x| \sup_n \sum_{k=3}^n \binom nk \frac{1}{n^k}.$$

Now let $x \to 0$.

  • $\begingroup$ Nothing wrong with that but for the main result we do not need to know that the sequence $(\sum_{k=3}^nx^{k-3}n^{-k}\binom {n}{k})_{n\geq 3} $converges: For $n\geq k$ we have $0<n^{-k}\binom {n}{k}\leq 1.$ So for $n\geq 3$ and $|x|\leq 1/2$ we have $ |\sum_{k=3}^nn^{-k}x^{n-3}\binom {n}{k}|\leq$ $ \sum_{k=3}^n|x|^{k-3}\leq$ $ \sum_{k=3}^n2^{-(k-3)}<2.$ So for $|x|\leq 1/2$ we have $|(e^x-(1+x)/x^2-\frac {1}{2}|\leq 2|x|.$ $\endgroup$ – DanielWainfleet Aug 31 '16 at 7:20

For every natural $n\ge 2$ we have $$\lim_{x\to 0}\frac{\left(1+\frac xn\right)^n-x-1}{x^2}=\lim_{x\to0}\frac{1+n\frac xn+\binom n2\frac{x^2}{n^2}+x^3P(x)-x-1}{x^2}=\frac{n-1}{2n}$$ where $P(x)$ is a polynomial.

This alone does not imply that your limit is $1/2$. We need to assume that the function $$e^x=\lim_{n\to\infty}\left(1+\frac xn\right)^n$$ is defined in some interval $I$ around $0$ (that is, that the limit exists for every $x\in I$) and continuous.

  • $\begingroup$ I don't see how this proves the result. You've shown $\lim_{x\to 0}\frac{(1+x/n)^n - x - 1}{x^2} = \frac{1}{2}$ for each $n.$ But why is $$\lim_{x\to 0}\lim_{n\to \infty}\frac{(1+x/n)^n - x - 1}{x^2}= \lim_{n\to \infty}\lim_{x\to 0}\frac{(1+x/n)^n - x - 1}{x^2}?$$ $\endgroup$ – zhw. Sep 3 '16 at 19:46

Are derivatives allowed? Assuming the existence of $$L = \lim_{x \to 0} \frac{e^x - 1 - x}{x^2} = \lim_{x \to 0} \frac{\left(\frac{e^x - 1}x\right) - 1}x$$ it is equal to the derivative of the (continuous) function $$f(x) = \begin{cases} \frac{e^x - 1}x & x \neq 0 \\ 1 & x = 0 \end{cases}$$ at $x = 0$. Thus, we have (formally) $$L = \lim_{t \to 0}\left.\frac{d}{dx}\left(\frac{e^x - 1}x\right)\right|_{x = t} = \lim_{t \to 0} \left(\frac{e^t}t - \frac{e^t - 1}{t^2}\right) = \lim_{t \to 0} \left(\frac{e^t - 1}t - \frac{e^t - 1 - t}{t^2}\right) = 1 - L$$

  • $\begingroup$ Nice! ${{{{}}}}$ $\endgroup$ – user00000 Aug 30 '16 at 19:32
  • $\begingroup$ Hi, in you 1st equality on the last line, you're assuming that (e^x-1)/x is differentiable for any t in a neighbourhood of zero? and implicitly assuming that (e^x-1)/x is diff. also at zero? this is exactly what we want to prove... $\endgroup$ – An old man in the sea. Aug 31 '16 at 19:13
  • $\begingroup$ @oldman (that sounds derogatory, but I don't know what else to call you!), I believe I am assuming that $(e^x-1)/x$ is differentiable at zero, since that is equivalent to assuming that the limit exists. It is certainly differentiable in a deleted neighborhood of 0, as it is elementary and continuous off 0, but I did not explain why the derivative extends to 0, and why the derivative extends continuously to 0. Then again, such matters may be beyond the grasp of truly beginner students. $\endgroup$ – ALGebraist Sep 1 '16 at 3:29
  • $\begingroup$ ALGebraist, there's no problem, I'm an old man(or becoming one =D). When you try to tag someone with @ don't cut their usernames, otherwise they will not be notified. For e.g. you should have written @anoldmaninthesea (you can use the tab key to speed it up) $\endgroup$ – An old man in the sea. Sep 12 '16 at 10:38

We have that $f(x)=e^x$ is an increasing convex function, hence for any $\varepsilon>0$ there is a neighbourhood of the origin $U$ where the inequality $$ (1-\varepsilon)x\leq e^x-1 \leq (1+\varepsilon)x \tag{1}$$ holds. If we consider some $z\in U$ and integrate every term of $(1)$ over $(0,z)$ we get $$ (1-\varepsilon)\frac{z^2}{2}\leq e^z-1-z \leq (1+\varepsilon)\frac{z^2}{2}\tag{2} $$ hence $$ \liminf_{z\to 0}\frac{e^z-1-z}{z^2}\geq\frac{1-\varepsilon}{2},\qquad \limsup_{z\to 0}\frac{e^z-1-z}{z^2}\leq \frac{1+\varepsilon}{2} \tag{3}$$ and since $\varepsilon$ is arbitrary, the claim follows.


I guess we're allowed limits here, so take $e^x = \lim_{k\rightarrow \infty}(1+x/k)^k$. So for large $k$, your limit should be about the same as $$\lim_{x\rightarrow 0} \frac{(1+x/k)^k - 1 - x}{x^2} = \lim_{x\rightarrow 0} \frac{1 + {k\choose 1}\frac{x}{k} + {k \choose 2}\frac{x^2}{k^2} + \mbox{HOT} - 1 - x}{x^2} =\lim_{x\rightarrow 0} \frac{{k \choose 2}\frac{x^2}{k^2} + \mbox{HOT}}{x^2} =\frac{1}{2}.$$

If we can figure out a way to swap the two limits, we'll have a rigorous proof.

  • $\begingroup$ Last equality is not true. It is $(k-1)/(2k)$. And this tends to $1/2$. $\endgroup$ – ajotatxe Aug 30 '16 at 15:22
  • $\begingroup$ Like I said, swap the limits. $\endgroup$ – B. Goddard Aug 30 '16 at 15:23
  • $\begingroup$ How do you swap limits without calculus? $\endgroup$ – Bernard Aug 30 '16 at 16:01
  • $\begingroup$ How do you have limits at all without calculus? "Calculus is algebra plus limits." -- Famous Wise Person. $\endgroup$ – B. Goddard Aug 30 '16 at 18:38
  • 1
    $\begingroup$ What's $\text{HOT}$ a short for? $\endgroup$ – user00000 Aug 30 '16 at 19:31

If you allow a few basic calculus results into the game, then note

$$e^x-x - 1 = \int_0^x (e^t-1)\, dt = \int_0^x [(e^t-1)/t]t\, dt.$$

Now $(e^t-1)/t \to 1$ as $t\to 0,$ simply because the derivative of $e^x$ is $1$ at $0.$ That implies the last integral above $\approx \int_0^x t\, dt = x^2/2$ and that will lead to the result. So a few details missing, but it's not too hard to fill them in.


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.