5
$\begingroup$

Let $c_n$ be an unbounded sequence: $c_n\to\infty$ when $n\to\infty$

Find the value of the limit: $$\lim_{n\to\infty}\int_1^n\frac{\ln x}{c_n+x\ln x}\,dx$$

What I managed to find was the fact that $$\lim_{n\to\infty}\frac{\ln x}{c_n+x\ln x}=0$$ I was thinking that the limit is going to be $0$, but I have to prove it still. Maybe it is not and I am wrong I do not know for sure.

I denoted $$I(n)=\int_1^n\frac{\ln x}{c_n+x\ln x}\,dx$$

I was thinking of finding a recurrence formula for this integral, Though, I could not be able to make any significant progress.

What should I do?

$\endgroup$
7
  • 1
    $\begingroup$ Do you want the specific case $c_n=n\log n$ or the more general $c_n \to \infty$? $\endgroup$
    – Henry
    Apr 30 at 20:10
  • $\begingroup$ I would want the general case, but maybe a particular case might help with the general case. If you have a solution for that particular case I would be grateful if you post it so as to have an idea how the general case might be solved. $\endgroup$
    – shangq_tou
    Apr 30 at 20:18
  • 1
    $\begingroup$ If ${c_n\over n\log n}\to \infty$ the limit is equal $0.$ So the case $c_n=n\log n$ is interesting. $\endgroup$ Apr 30 at 20:38
  • $\begingroup$ Why is the limit equal with zero for $\frac{c_n}{n\log _n}\rightarrow \infty $? $\endgroup$
    – shangq_tou
    Apr 30 at 22:10
  • 1
    $\begingroup$ Following the comments of Henry and Ryszard: from $$\frac{\log x}{c_n + x\log x}\mathbb{1}_{(1,n]}(x)\leq \frac{\log n}{c_n}\mathbb{1}_{(1,n]}(x)$$ we have $$\int^n_1\frac{\log x}{c_n + x\log x}\,dx\leq \frac{n\log n}{c_n}$$ Thus, of $\lim_n\frac{c_n}{n\log n}=\infty$, the case discussed in the comments has $0$ limit. $\endgroup$ Apr 30 at 22:36

3 Answers 3

5
$\begingroup$

As @Ryszard Szwarc mentioned in the comment, if $\frac{c_n}{n\ln n}\to\infty,\,\,\lim_{n→∞} \int _1^n\frac{\ln\:x}{c_n\:+\:x\:\ln x}dx=0$, and the most interesting is the case $c_n=n\ln n$.

So, we want to find $\lim_{n→∞} I(n)=\lim_{n→∞}\int _1^n\frac{\ln\:x}{n\ln n\:+\:x\:\ln x}dx$. Making the substitution $x=e^t$ $$I(n)=\int _1^n\frac{\ln\:x}{n\ln n\:+\:x\:\ln x}dx=\int_0^{\ln n}\frac{te^t}{n\ln n+e^t t}dt$$ Making another substitution $t=s\ln n$ $$I(n)=\frac{\ln n}{n}\int_0^1\frac{sn^s}{1+\frac{s}{n^{1-s}}}ds$$ As $\frac{s}{n^{1-s}}\leqslant1$, we can decompose the denominator into the series: $$I(n)=\frac{\ln n}{n}\int_0^1n^ss\big(1-sn^{s-1}+s^2n^{2s-2}-s^3n^{3s-3}+-...\big)ds$$ Integrating by part, we get the terms with the different rate of convergence, for example: $$\int_0^1sn^sds=\frac{1}{\ln^2n}\int_0^{\ln n}se^sds=\frac{1}{\ln^2n}\big(n\ln n-n\big)=\frac{n}{\ln n}-\frac{n}{\ln^2 n}$$ The same story happens to all the terms; only the first ones will give a non-zero limit at $n\to\infty$. Therefore, keeping only these main terms $$I(n)=\frac{\ln n}{n}\Big(\frac{1}{\ln^2n}\int_0^{\ln n}e^xxdx-\frac{1}{n\ln^3n}\frac{1}{2^3}\int_0^{2\ln n}e^xx^2dx+\frac{1}{n^2\ln^4n}\frac{1}{3^4}\int_0^{3\ln n}e^xx^3dx-+...\Big)$$ The pattern is clear. Integrating by part and keeping only first terms of every integral $$I(n)=\frac{\ln n}{n}\Big(\frac{1}{\ln^2n}n\ln n-\frac{1}{n\ln^3n}\frac{1}{2^3}2^2n^2\ln^2n+\frac{1}{n^2\ln^4n}\frac{1}{3^4}3^3n^3\ln^3n-+...\Big)+O\Big(\frac{1}{\ln n}\Big)$$ $$\Rightarrow\,\,\boxed{\,\,\lim_{n\to\infty} I(n)=1-\frac{1}{2}+\frac{1}{3}-\frac{1}{4}+-...=\ln 2=0.693147...\,\,}$$ The numeric evaluation of the limit confirms this conclusion: $$\int _1^n\frac{\ln\:x}{n\ln n\:+\:x\:\ln x}dx\,\Big|_{n=10^{300}}=0.692143...$$ though the rate of convergence, of course, is monstrously low :)

It is not difficult to evaluate in the same way the second and third terms of the asymptotics. I got $$\boxed{\,\,I(n)=\ln2-\frac{\ln2}{\ln n}+\Big(\ln2-\frac{\pi^2}{12}\Big)\frac{1}{\ln^2n}+O\Big(\frac{1}{\ln^3n}\Big)\,\,}$$

$\mathbf{PS}$

I checked numerically the revised asymptotics. It seems it works! $$\int _1^n\frac{\ln\:x}{n\ln n\:+\:x\:\ln x}dx\,\Big|_{n=1000}=0.590148...$$ Asymptotics gives $$I(n)\sim\ln2-\bigg(\frac{\ln2}{\ln n}+\Big(\ln2-\frac{\pi^2}{12}\Big)\frac{1}{\ln^2n}\bigg)\,\bigg|_{n=1000}=0.590093...$$

$\endgroup$
4
  • 1
    $\begingroup$ You may want to stick with either $c_n$ or $C_n$ at the beginning. $\endgroup$
    – Gary
    May 1 at 5:56
  • 1
    $\begingroup$ @Gary , thank you. Correcting $\endgroup$
    – Svyatoslav
    May 1 at 5:57
  • 2
    $\begingroup$ Also in the middle of the post you give the expansion for $I(n)$ but then you say it equals to its limit as $n \to \infty$. Since you keep only the first term of each integral, the expansion is not equal to $I(n)$. $\endgroup$
    – Gary
    May 1 at 5:59
  • 1
    $\begingroup$ @Gary, thank you for catching this. Also correcting $\endgroup$
    – Svyatoslav
    May 1 at 6:01
4
$\begingroup$

It turns out that we have the following inequalities:

$$ \log\biggl(1 + \varliminf_{n\to\infty} \frac{n\log n}{c_n}\biggr) \leq \varliminf_{n\to\infty} I(n) \leq \varlimsup_{n\to\infty} I(n) \leq \log\biggl(1 + \varlimsup_{n\to\infty} \frac{n\log n}{c_n}\biggr) $$


Let $g : [0, \infty) \to [1, \infty)$ be the inverse of $x \mapsto x \log x$ for $x \geq 1$, that is, $g$ solves the functional equation

$$ g(t) \log g(t) = t $$

(In fact, we can write $g(t) = t/W(t)$ where $W(t)$ is the Lambert W-function. However, this observation is not needed here.) Then by substituting $t = x \log x$, or equivalently, $x = g(t)$, and noting that

$$ \frac{\mathrm{d}x}{x} = \frac{g'(t) \, \mathrm{d}t}{g(t)} = \frac{\mathrm{d}t}{g(t)(\log g(t) + 1)} = \frac{\mathrm{d}t}{t + g(t)} $$

we have

\begin{align*} I(n) = \int_{1}^{n} \frac{x \log x}{c_n + x \log x} \, \frac{\mathrm{d}x}{x} = \int_{0}^{n \log n} \frac{t}{c_n + t} \, \frac{\mathrm{d}t}{t + g(t)} \end{align*}

Writing $\rho_n = (n \log n) / c_n$ and substituting $t = c_n u$,

\begin{align*} I(n) &= \int_{0}^{\rho_n} \frac{u}{1 + u} \cdot \frac{1}{u + g(c_n u)/c_n} \, \mathrm{d}u \end{align*}

Now, let $ \displaystyle \alpha = \varliminf_{n\to\infty} \rho_n$ and $ \displaystyle \beta = \varlimsup_{n\to\infty} \rho_n$. Then

$$ \varlimsup_{n\to\infty} I(n) \leq \varlimsup_{n\to\infty} \int_{0}^{\rho_n} \frac{u}{1 + u} \cdot \frac{1}{u} \, \mathrm{d}u = \varlimsup_{n\to\infty} \log(1+\rho_n) = \log(1+\beta) \tag{1} $$

On the other hand, together with the fact that $g(t) \ll t$ and the Fatou's lemma,

\begin{align*} \varliminf_{n\to\infty} I(n) &\geq \int_{0}^{\infty} \varliminf_{n\to\infty} \frac{u}{1 + u} \cdot \frac{1}{u + g(c_n u)/c_n} \mathbf{1}_{[0, \rho_n]}(u) \, \mathrm{d}u \\ &= \int_{0}^{\infty} \frac{u}{u+1} \cdot \frac{1}{u} \mathbf{1}_{[0, \alpha)}(u) \, \mathrm{d}u \\ &= \log(1+\alpha) \tag{2} \end{align*}

$\endgroup$
1
  • 1
    $\begingroup$ +1 This fully addresses the question $\endgroup$
    – FShrike
    May 1 at 10:15
3
$\begingroup$

This is not an entire solution, but a long comment.

Following the comments of Henry and Ryszard: from $$\frac{\log x}{c_n + x\log x}\mathbb{1}_{(1,n]}(x)\leq \frac{\log n}{c_n}\mathbb{1}_{(1,n]}(x)$$ we have $$\int^n_1\frac{\log x}{c_n + x\log x}\,dx\leq \frac{n\log n}{c_n}$$ Thus, of $\lim_n\frac{c_n}{n\log n}=\infty$, the case discussed in the comments has $0$ limit.

For the general case, the limit if exists, may not be $0$ $$\frac{\log(m-1)}{c_n+m\log m}\leq \int^m_{m-1}\frac{\log x}{c_n + x\log x}\,dx \leq\frac{\log m}{c_n+ (m-1)\log (m-1)}\leq \frac{\log m}{c_n}$$ adding all terms for $2\leq m\leq n$ yields $$\frac{\log((n-1)!)}{c_n+n\log n}\leq\sum^n_{m=2}\frac{\log(m-1)}{c_n+m\log m}\leq \int^n_1\frac{\log x}{c_n+x\log x}\,dx\leq \frac{1}{c_n}\log(n!)$$

$\endgroup$
1

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.