5

Per the title of this question, how does one go about calculating $$\lim_{n\rightarrow \infty} \frac{1}{n^2}\left(\ln\left(\frac{2^n}{3^n}\right)+\ln\left(\frac{5^n}{4^n}\right)+\cdots+\ln\left(\frac{(3n-1)^n}{(n+2)^n}\right)\right)\ ?$$

Thanks!

josh
  • 259
  • 1
  • 4

1 Answers1

1

For a mechanical* way to do this:

Hint:

$$\log \frac{(3k-1)}{k+2} = \log 3 + \log (1- \frac{1}{3k}) - \log (1 + \frac{2}{k})$$

and

$$\log (1-x) = -x - \frac{x^2}{2} - \frac{x^3}{3} - \dots $$

for $|x| \lt 1$.

More details:

Using the above hint, the $k^{th}$ term is $$ n \log \frac{(3k-1)}{k+2} = n \log 3 + \frac{5n}{3k} + \mathcal{O}(\frac{n}{k^2})$$

and so your sum is

$$ \frac{1}{n^2} \sum_{k=1}^{n} (n \log 3 + \frac{5n}{3k} + \mathcal{O}(\frac{n}{k^2}))$$

$$ = \frac{1}{n^2}(n^2\log 3 + \frac{5n\log n}{3} + \mathcal{O}(n))$$

Here we used the estimate $\displaystyle \sum_{k=1}^{n} \frac{1}{k} = \log n + \mathcal{O}(1)$ and $\displaystyle \sum_{k=1}^{n} \frac{1}{k^2} = \mathcal{O}(1)$

Thus the sum is $$ = \log 3 + \frac{5\log n}{3n} + \mathcal{O}(\frac{1}{n})$$

and so your limit is $$\log 3$$

Note that we don't really need the estimate $\displaystyle \sum_{k=1}^{n} \frac{1}{k} = \log n + \mathcal{O}(1)$

All we need to show is that $\displaystyle \sum_{k=1}^{n} \frac{1}{k} = o(n)$ and this easily follows from the following classic theorem:

If $\displaystyle a_n \to 0$, then $\displaystyle \frac{1}{n} \sum_{k=1}^{n} a_k \to 0$

*As Didier points out (see comments below), this last theorem can be used to skip all the mechanical calculations done above by applying it directly to the terms of your sequence.

Aryabhata
  • 82,206
  • If I consider the sequence as the integer, can I work it out? – 89085731 Feb 26 '12 at 02:09
  • @Gingerjin: I don't understand your question. Can you please elaborate? – Aryabhata Feb 26 '12 at 02:11
  • I mean I want to create an integral that is the answer of the sequence, Can I work it out? SOrry, I haven't explained it correctly. – 89085731 Feb 26 '12 at 02:14
  • @Gingerjin: Yes it is possible, I think. After cancelling one $n$, it looks like a Riemann Sum. – Aryabhata Feb 26 '12 at 02:19
  • Could you please show me that? I couldn't work it out.3ks – 89085731 Feb 26 '12 at 02:21
  • @Gingerjin: Maybe I was wrong. We need to get $\frac{k}{n}$ in there somehow. – Aryabhata Feb 26 '12 at 02:34
  • Thanks. Is it true that Taylor expansion is commonly used in this kind of problems? – 89085731 Feb 26 '12 at 02:37
  • @Gingerjin: Well, it is a useful and powerful tool. If you are talking about in exams or something like that, the answer is, it depends on the teacher I guess. – Aryabhata Feb 26 '12 at 02:41
  • @Aryabhata: Could you add in more detail? I can't quite see the entire solution yet (in case this is a concern - this isn't homework, just a question from a practice exam ;). Thanks! – josh Feb 26 '12 at 03:14
  • @josh: Perhaps if you edit the question with your attempt at completing this answer, I can help you better. Knowing where exactly you are having problems will help us focus on the right things. Otherwise I will have to try and cover all possibilities... btw, if you hover your mouse over the block at the end of the answer, you will see more detail. – Aryabhata Feb 26 '12 at 03:16
  • @Aryabhata: For the most part I don't understand the second step in the 'complete answer' section, where you go on to claim the complete sum without $n^2$ in the denominator is (such and such). Why does the $log(n)$ appear there, and why do we get the $O(n)$? (I'm used to remainder in the form of $R_4(n)$ moreso than big-O or small-O) – josh Feb 26 '12 at 03:30
  • @josh: I am using $\sum_{k=1}^{n} \frac{1}{k} = \log n + \gamma + O(1/n)$ and the fact that $\sum_{k=1}^{\infty} \frac{1}{k^2}$ converges and so $\sum_{k=1}^{n} \frac{1}{k^2} = O(1)$. – Aryabhata Feb 26 '12 at 03:37
  • Ah, okay. Thank you. We have not learned the 1+1/2+...->ln(n) approximation in this course. Or well, we have, since we learned about Taylor polynomials, but I guess I just don't understand the big-O notation quite well. – josh Feb 26 '12 at 03:43
  • @josh: You don't really need to know that approximation to solve this problem. All you need to prove is that $\sum_{k=1}^{n} \frac{1}{k} = o(n)$ (note: small Oh, not big Oh). For that: This is a well-known theorem: if $a_n \to 0$, then $\frac{1}{n}\sum_{k=1}^{n} a_k \to 0$. – Aryabhata Feb 26 '12 at 03:46
  • @Aryabhata Your $k$ in the answer should be an $n$. When you sum from $n,\log 3$, the index of summation is $n$, and the sum $n(n-1)\log3/2$. The limit is $\log3/2$, as can also be seen using Stolz's criterium. – Julián Aguirre Feb 26 '12 at 10:22
  • @JuliánAguirre: If you see the first version, that is what I had. But I realized that the question is actually $\lim_{n \to \infty} \frac{1}{n^2} \sum_{k=1}^{n} n \log (\frac{3k-1}{k+2})$ and not $\lim_{n \to \infty} \frac{1}{n^2} \sum_{k=1}^{n} k \log (\frac{3k-1}{k+2})$ which gives $\frac{\log 3}{2}$ – Aryabhata Feb 26 '12 at 10:26
  • @Aryabhata I see. – Julián Aguirre Feb 26 '12 at 10:39
  • 2
    @Aryabatha: the argument in the very last line of your post allows to skip everything before since the sum one is interested in is $(\log3)+\frac1n\sum\limits_{k=1}^na_k$ with $a_n=\log\left(\frac{3n-1}{3n+3}\right)\to0$. – Did Feb 26 '12 at 11:24
  • @DidierPiau: Yes agreed. – Aryabhata Feb 26 '12 at 11:33