4
$\begingroup$

I'm having some trouble solving this limit:

$$\lim_{x\to\infty} x\left[\left(\cosh x\right)^ \frac1x - \left(1+\frac1x\right)^x\right]$$

It's part of a set of limits wich should be solved using taylor. I tried this road:

$$\lim_{x\to\infty} x\left[e^{\frac1x\ln(\cosh x)}-e^{\ln\left(1+\frac1x\right)x}\right]$$

I then tried with algebraic manipulation using the definition of hyperbolic cosine $\frac12(e^x+e^{-x})$ and then I also played a bit with l'Hopital but it turns into something suspiciously compicated... I'm taking real analysis 1 and my toolset is:

-algebraic manipulation

-talyor series

-l'Hopital (if and only if all else fails)

I can't use more advanced techniques since they are not part of the course. I'm sure there's something obvious I'm missing. Any idea on how to proceed?

$\endgroup$
  • 1
    $\begingroup$ One thing to try is that your given limit equals $\lim_{y\rightarrow 0^+}\frac{\cosh(1/y)^y-(1+y)^{1/y}}{y}$. L'Hopital is appropriate for this fraction (by checking that both terms in the numerator have the same limit). $\endgroup$ – Michael Burr Jun 4 '16 at 13:21
  • $\begingroup$ @MichaelBurr thank you very much for your advice, however, being a beginner in analysis, I prefer to avoid l'Hopital whenever I can, since while it may give you results really fast at the same time it keeps you from noticing where your weaknesses are. $\endgroup$ – edo Jun 4 '16 at 18:51
2
$\begingroup$

I have tried several approaches, and the fastest seems the following [it does not require L'Hopital]:

$$ \lim_{x\to\infty} x[e^{\ln(\cosh x)\frac1x}-e^{\ln(1+\frac1x)x}]=\lim_{x\to\infty} e^{\ln x}[e^{\ln(\cosh x)\frac1x}-e^{\ln(1+\frac1x)x}] $$ $$ \lim_{x\to\infty} e^{\phi_1(x)}-e^{\phi_2(x)}\ . $$ Let us analyze the series expansions of the two functions in the exponent up to second next-to-leading order $$ \phi_1(x)=\ln x+\frac{1}{x}\ln(\cosh(x))=\ln x+\frac{1}{x}\ln\left(\frac{1}{2}(e^x+e^{-x})\right)=\ln x-\frac{\ln 2}{x}+1+\frac{1}{x}\ln(1+e^{-2x}) $$ $$ \sim \ln x+1-\frac{\ln 2}{x} $$ $$ \phi_2(x)=\ln x+x \ln(1+1/x)\sim\ln x +1-\frac{1}{2x} $$ Therefore, Taylor-expanding the exponentials [of the type $e^{-c/x}$] around $x=\infty$ $$ e^{\phi_1(x)}\sim e x e^{-\ln 2/x}\sim e x -e \ln (2)+\frac{e \ln ^2(2)}{2 x}-\frac{e \ln ^3(2)}{6 x^2}+\ldots $$ $$ e^{\phi_2(x)}\sim e x-\frac{e}{2}+\frac{e}{8 x}-\frac{e}{48 x^2}+\ldots $$ we get the final result by subtracting one from the other $$ \boxed{\lim_{x\to\infty} x[e^{\ln(\cosh x)\frac1x}-e^{\ln(1+\frac1x)x}]=\frac{e}{2}-e\ln 2\approx -0.525028...} $$

$\endgroup$
  • $\begingroup$ thanx a lot, it took me a minute to understand why you stated this: $$\ln x-\frac{\ln 2}{x}+1+\frac{1}{x}\ln(1+e^{-2x})\sim \ln x+1-\frac{\ln 2}{x}$$ and that was the root of my problem: $\\$we can ignore $~~$ $\frac{1}{x}\ln\left(1+e^{-2x}\right)$ $~~$ because it's approaching zero "way faster" than$~~$ $-\frac{\ln 2}{x}$. $\\$ I guess I'm still a bit shaky on when to use asymptotic equalities. Thanx for the editing as well, I'm new to MathJax. All the best! $\endgroup$ – edo Jun 4 '16 at 19:54
  • $\begingroup$ You're welcome. Very challenging limit indeed...and remember that limits and asymptotic expansions can only be learnt by the combination of three tools: i) PRACTICE ii) PRACTICE, and I let you figure out the third... $\endgroup$ – Pierpaolo Vivo Jun 4 '16 at 20:01

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.