0

The question is:

Evaluate the following limit without L'Hôpital, Taylor, differentiation (or integration): $$\lim_{x \to 0} \frac{\ln\left[\dfrac{(1-3x)(1+x)^3}{(1+3x)(1-x)^3}\right]}{x^3}$$

This came up while I was trying to answer to this question. My own answere is there too.

I managed to cancel out most of the terms without L'Hôpital/Taylor/differentiation until I reached the above limit. Then I used a differentiation (not L'Hôpital) and finished with the definition of a derivative. The OP to that question claimed that even differentiation is not allowed. He believes it could be solvable only using such methods that can be derived (algebraically) from something like $\lim_{x\to0} \frac{\sin x}{x}$ or $\lim_{n\to\infty} \left(1+\frac1n\right)^n$, etc. I think that a sandiwch theorem will be fine too.

Using Taylor seriese or L'Hôpital's theorem, the above limit is trivial to solve. And the below is how I solved it using differentiation (which was not allowed either). I am wondering if there's a way to evaluate it even without differentiation, so I am posting here. Thanks.

My method: \begin{align} \lim_{x \to 0} \frac{\ln\left[\dfrac{(1-3x)(1+x)^3}{(1+3x)(1-x)^3}\right]}{x^3} &=\lim_{x \to 0} \frac{\ln\left((1-3x)(1+x)^3\right)-\ln\left((1+3x)(1-x)^3\right)}{x^3}\\ &=\lim_{x \to 0} \frac{\ln(1-6x^2-8x^3-3x^4)-\ln(1-6x^2+8x^3-3x^4)}{x^3}\\ &=\lim_{x \to 0} \frac{\ln(1-6x^2-8x^3)-\ln(1-6x^2+8x^3)}{x^3}\\ &=\lim_{h \to 0} \frac{\ln(1-6h^{2/3}-8h)-\ln(1-6h^{2/3}+8h)}{h}\\ &=\left.\left\{\ln(1-6x^{2/3}-8x)-\ln(1-6x^{2/3}+8x)\right\}'\right|_{x=0}\\ &=\left.\left(\frac{-\frac4{\sqrt[3]x}-8}{1-6x^{2/3}-8x}-\frac{-\frac4{\sqrt[3]x}+8}{1-6x^{2/3}+8x}\right)\right|_{x=0}\\ &=\left.\frac{-16+32x^{2/3}}{(1-6x^{2/3}-8x)(1-6x^{2/3}+8x)}\right|_{x=0}\\ &=-16\\ \end{align}


Here's the alternative answer that I think may be acceptable. It only uses: $$\frac1{1+x}=1-x+x^2-x^3+\cdots$$ and $$\lim_{x\to0}\frac{\ln(1+x)}x=1$$ that comes from $$\lim_{n\to\infty}\left(1+\frac1n\right)^n=e$$ So the method-2 is: \begin{align} &\dfrac{(1-3x)(1+x)^3}{(1+3x)(1-x)^3}\\ =&\dfrac{(1-3x)(1+3x+3x^2+x^3)}{(1+3x)\left(1-x(3+3x-x^2)\right)}\\ =&(1-3x)(1+3x+3x^2+x^3)\\ \times&(1-3x+9x^2-27x^3+\cdots)\\ \times&\left(1+x(3+3x-x^2)+x^2(3+3x-x^2)^2+x^3(3+3x-x^2)^3+\cdots\right)\\ =&(1-3x)(1+3x+3x^2+x^3)(1-3x+9x^2-27x^3+\cdots)(1+3x+6x^2+10x^3+\cdots)\\ =&1-16x^3+O(x^4) \end{align} Therefore, \begin{align} \lim_{x \to 0} \frac{\ln\left[\dfrac{(1-3x)(1+x)^3}{(1+3x)(1-x)^3}\right]}{x^3}&=\lim_{x \to 0} \frac{\ln\left(1-16x^3+O(x^4)\right)}{x^3}\\ &=-16\lim_{x \to 0} \frac{\ln\left(1-16x^3\right)}{-16x^3}\\ &=-16\lim_{x\to0}\frac{\ln(1+x)}x\\ &=-16 \end{align}


Using the Essaidi's answer below, here's the solution that will (I believe) satisfy the requirement of "using only knowledge learned before Calculus" (this is the updated compilation of the requirement to the original problem that I could understand from the OOP's comment to other person below the original question).

\begin{align} &\lim_{x \to 0} \frac{\ln\left[\dfrac{(1-3x)(1+x)^3}{(1+3x)(1-x)^3}\right]}{x^3}\\ &=\lim_{x \to 0} \frac{\ln\left[1+\dfrac{(1-3x)(1+x)^3}{(1+3x)(1-x)^3}-1\right]}{\dfrac{(1-3x)(1+x)^3}{(1+3x)(1-x)^3}-1}\frac{\dfrac{(1-3x)(1+x)^3}{(1+3x)(1-x)^3}-1}{x^3}\\ &=\lim_{y \to 0}\frac{\ln(1+y)}{y}\cdot\lim_{x \to 0}\frac{\dfrac{(1-3x)(1+x)^3}{(1+3x)(1-x)^3}-1}{x^3}\\ &=1\cdot\lim_{x \to 0}\frac{(1-3x)(1+x)^3-(1+3x)(1-x)^3}{x^3(1+3x)(1-x)^3}\\ &=\lim_{x \to 0}\frac{-16x^3}{x^3}\\ &=-16 \end{align}

Kay K.
  • 9,931
  • It is kind of weird to forbid Taylor/L'Hôpital and not differentiation as they are basically the same. – nejimban Jan 08 '22 at 16:53
  • @nejimban Differentiation is forbidden too. – Kay K. Jan 08 '22 at 16:54
  • Well your method uses differentiation, right?... – Adam Rubinson Jan 08 '22 at 17:01
  • So one essentially needs that $\ln(1+x)\sim x$ which amounts to the differentiability of $x\mapsto\ln(1+x)$ at $0$. You can recover it from the squeezing theorem and the “Taylor” inequalities (say for $x>0$) $$\ln(1+x)=\int_1^{1+x}\frac{\mathrm dt}t\le\int_1^{1+x}\frac{\mathrm dt}{1}=x$$ and $$x-\ln(1+x)=\int_1^{1+x}\frac{t-1}t,\mathrm dt\le\int_1^{1+x}\frac{t-1}1,\mathrm dt=\left[\frac{(t-1)^2}2\right]_1^{1+x}=\frac{x^2}2,$$ but wouldn't this also qualify as a “differentiation” argument? – nejimban Jan 08 '22 at 17:04
  • @AdamRubinson Yes, so it was not good, although the OP generously accepted it. – Kay K. Jan 08 '22 at 17:05
  • @nejimban The approximation should be precise at least up to $x^3$ term, because all $x^2$ terms should be cancelled out and the denominator is $x^3$. – Kay K. Jan 08 '22 at 17:08
  • Oops you're right. One can surely integrate again to get $\ln(1+x)\le x-\frac{x^2}2+\frac{x^3}3$. – nejimban Jan 08 '22 at 17:10
  • @nejimban Okay, the OOP (orignal poster to the opriginal post) claimed no integration either, but, could you post you answer? Thanks. – Kay K. Jan 08 '22 at 17:12
  • @Kay K. See e.g., this post. You can also use the Binomial theorem to show that $\mathrm e^x-1-x\sim\frac{x^2}2$, which by changing variables gives $y-\ln(1+y)\sim\frac{y^2}2$. – nejimban Jan 08 '22 at 17:17
  • @nejimban Actually I had learned a lot from the RE60K's answer on that post. And I used the method up to the point I got stuck with this $\ln$. Thanks. I'll think about the binomial theorem. (However, I think that it will still be an alternative way to get the Taylor series up to $x^3$, which I think may be the only way.) – Kay K. Jan 08 '22 at 17:23
  • 1
    Isn't Bionomial expansion of $(1+x)^{n}$ a form of Taylor Series? (where $x \in (-1,1)$ and $n \in \mathbb {Q}$) – Justlookingforthemoment Jan 08 '22 at 18:47
  • @Justlookingforthemoment No, I don't think so. Taylor series is an infinite series you can't directly get without getting the coefficients by differentiation. – Kay K. Jan 08 '22 at 19:04
  • 1
    Oops, I am sorry, I meant where $x \in (-1,1)$ and $n \notin \mathbb{N}$ and $n \neq 0$ – Justlookingforthemoment Jan 08 '22 at 19:08
  • you can't even properly define logarithms without concepts of differentiation or integration or power series, so how would you expect to compute limits involving it? – peek-a-boo Jan 09 '22 at 00:09
  • @Justlookingforthemoment I don't think the generalized binomial expansion is a Taylor series, because Sir. Issac Newton invented it even before he invented Calculus. However, it may be as higher level as the Taylor series, so I don't think the solution you teacher had in his/her mind involves it (if he had the solution). – Kay K. Jan 09 '22 at 00:56
  • @peek-a-boo I am sorry that the restriction is not clear indeed. However, the OOP (Justlookingforthemoment in this conversation is the OOP) said that $lim_{x\to0} \frac{\ln(1+x)}x = 1$ is allowed to use as it comes from $\lim_{n\to\infty} (1+1/n)^n=e$. – Kay K. Jan 09 '22 at 00:58
  • @Justlookingforthemoment I added the method-2. Please see if it may be accepted by your teacher. BTW, can you please ask you teacher what the solution that he/she had in mind? Thanks. – Kay K. Jan 09 '22 at 01:19
  • 1
    Perfect! You exactly understood how I wanted my problem to be solved, and you did that! Congrats! – Justlookingforthemoment Jan 09 '22 at 05:51
  • 1
    Also, I see you updated your original answer to my question. Did a great thing. – Justlookingforthemoment Jan 09 '22 at 05:52

3 Answers3

2

You have a limit of type : $$\lim_{x \to 0} \dfrac{\ln f(x)}{g(x)}$$ with : $$\lim_{x \to 0} f(x) = 1$$ Write : $$\lim_{x \to 0} \dfrac{\ln f(x)}{f(x) - 1} \dfrac{f(x) - 1}{g(x)}$$ and you have : $$\lim_{x \to 0} \dfrac{\ln f(x)}{f(x) - 1} = 1$$ The other part : $$\lim_{x \to 0} \dfrac{f(x) - 1}{g(x)}$$ is a simple limit of a fraction.

Essaidi
  • 1,733
  • Thanks. That's basically what I did in my method-2. – Kay K. Jan 09 '22 at 02:55
  • But I think it adds clarity. So let me +1. – Kay K. Jan 09 '22 at 03:22
  • Thinking about this more, now I see this is not same to my method-2 (maybe I wanted to think so). If I had known this level of clear view, I would have not needed the messy expansion of the $\frac1{1+x}$. As you said, this indeed makes it just a matter of the limit of a fraction. I Let me accept this one. Thanks. – Kay K. Jan 09 '22 at 03:30
1

Blockquote $\displaystyle\lim_{x \to 0} \frac{\ln\left[\dfrac{(1-3x)(1+x)^3}{(1+3x)(1-x)^3}\right]}{x^{3}}$
$=\displaystyle\lim_{x \to 0}\dfrac{\ln(1-3x)-\ln(1+3x)+3\ln(1+x)-3\ln(1-x)}{x^{3}}$
we use Taylor series
$\displaystyle\lim_{x \to 0}\dfrac{(-3x-\dfrac{(3x)^{2}}{2}-\dfrac{(3x)^{3}}{3}-\dfrac{(3x)^{4}}{4}-\dfrac{(3x)^{5}}{5}-\cdots)-(3x-\dfrac{(3x)^{2}}{2}+\dfrac{(3x)^{3}}{3}-\dfrac{(3x)^{4}}{4}+\dfrac{(3x)^{5}}{5}-\cdots)+3(x-\dfrac{(x)^{2}}{2}+\dfrac{(x)^{3}}{3}-\dfrac{(x)^{4}}{4}+\dfrac{(x)^{5}}{5}-\cdots)-3(-x-\dfrac{(x)^{2}}{2}-\dfrac{(x)^{3}}{3}-\dfrac{(x)^{4}}{4}-\dfrac{(x)^{5}}{5}-\cdots)}{x^{3}}$
$\displaystyle\lim_{x \to 0}\dfrac{-3x-\dfrac{(3x)^{2}}{2}-\dfrac{(3x)^{3}}{3}-\dfrac{(3x)^{4}}{4}-\dfrac{(3x)^{5}}{5}-\cdots-3x+\dfrac{(3x)^{2}}{2}-\dfrac{(3x)^{3}}{3}+\dfrac{(3x)^{4}}{4}-\dfrac{(3x)^{5}}{5}+\cdots+3x-3\dfrac{(x)^{2}}{2}+3\dfrac{x)^{3}}{3}-3\dfrac{(x)^{4}}{4}+3\dfrac{(x)^{5}}{5}-\cdots+3x+3\dfrac{(x)^{2}}{2}+3\dfrac{(x)^{3}}{3}+3\dfrac{(x)^{4}}{4}+3\dfrac{(x)^{5}}{5}+\cdots}{x^{3}}$
$=\displaystyle\lim_{x \to 0}\dfrac{-2(\dfrac{(3x)^{3}}{3}+\dfrac{(3x)^{5}}{5}+\dfrac{(3x)^{7}}{7}+\cdots +6(\dfrac{(x)^{3}}{3}+\dfrac{(x)^{5}}{5}+\dfrac{(x)^{7}}{7}+\cdots}{x^{3}}$
$=\displaystyle\lim_{x \to 0}-2(\dfrac{(3x)^{3}}{3x^{3}}+\dfrac{(3x)^{5}}{5x^{3}}+\dfrac{(3x)^{7}}{7x^{3}}+\cdots +6(\dfrac{(x)^{3}}{3x^{3}}+\dfrac{(x)^{5}}{5x^{3}}+\dfrac{(x)^{7}}{7x^{3}}+\cdots$
$=\displaystyle\lim_{x \to 0}-2(\dfrac{27(x)^{3}}{3x^{3}}+\dfrac{((3)^{5}(x)^{5}}{5x^{3}}+\dfrac{(3)^{7}(x)^{7}}{7x^{3}}+\cdots +6(\dfrac{(x)^{3}}{3x^{3}}+\dfrac{(x)^{5}}{5x^{3}}+\dfrac{(x)^{7}}{7x^{3}}+\cdots$
$=\displaystyle\lim_{x \to 0}-2(9+\dfrac{((3)^{5}(x)^{2}}{5}+\dfrac{(3)^{7}(x)^{4}}{7}+\cdots +6(\dfrac{1}{3}+\dfrac{(x)^{2}}{5}+\dfrac{(x)^{4}}{7}+\cdots$
$=-2(9)+6(\dfrac{1}{3})$
$=-18+2$ $=-16$

HAD HAN
  • 21
1

The use of an identity is meaningful for the problem.

$arctanh(z)=\frac{ln(1+z)-ln(1-z)}{2}$

With this Your problem writes as

$lim_{x\rightarrow 0}\frac{6 arctanh(x)-2 arctanh(3 x)}{x^3}$

This works straight forward using

$6 arctanh(x)-2arctanh(3x)=-3 ln(1-x)+3 ln(1+x)-ln(1+3x)-ln(1-3x)=ln(\frac{(1+x)^3(1-3x)}{(1-x)^3(1+3x)})$

using $ln(a b)=ln(a)+ln(b)$ and $ln(\frac{a}{b})=ln(a)-ln(b)$.

Now $arctanh$ has a series representation that is not necessarily a Taylor series. It is a transcendental function. This series representation can be truncated with little error for small values of $x$.

$arctanh(x)=\sum_{k=1}^{\infty}\frac{x^{2k-1}}{(2k-1)!}\approx x- \frac{x^3}{3} + $ error of order $5$

$\frac{6(x-\frac{x^3}{3})-2 (3 x - \frac{(3 x)^3}{3})}{x^3}$

$\frac{(2-18)x^3}{x^3}=-16$

Because of the series representation, everything is like required: no Taylor expansion, no l'Hopital, no differentiation or integration.

Indeed used is an identity for logarithms and the series for Arcustangenshyperbolicus. Because we have two terms for the numerator both cancel in linear order as x gets smaller and the third-order term remains as is suggested already by the given fraction. As x is taken closer to zero the next order is five with coefficient $-96$ but with denominator, this reduces to second order. This cancels down to the constant term and that is $-16$ as shown in my calculation. A little disambiguation remains since the Taylor series around 0 coincides with the series representation of Arcustangenshyperbolicus. For limits usually, only the Taylor expansion up to noncritical orders is used by definition. Mind this is per se only valid for $x>0$. For $x<0$ a similar calculation is possible and the results are the same.

  • Thanks. I agree that we should use any kind of series representation of the logarithm. Your finding of the $arctanh$ is very interesting. However, I wonder how to convert the $arctanh$ into the series without differentiation. Anyway thank a lot for your answer. So I give your answer +1. Encouraged by your answer, I also added my solution-2. – Kay K. Jan 09 '22 at 01:26