Let $(a_n)_{n\geq 0}$ be a positive sequence of real numbers. I have two questions about sequences that are related to log-convex and log-concave. I'll just copy the necessary part in an article (Lemma 4.1) I'm reading

Question 1:

Suppose that $(a_{n+1}/a_n)_{n\geq 0}$ is eventually increasing to $\lambda\in (1,\infty]$. Then the proof says: for any $\lambda_0\in (1,\lambda)$, we have $a_{n+1}\geq \lambda_0 a_n$ for $n$ sufficiently large. Why?

Question 2:

Suppose that $(a_{n+1}/a_n)_{n\geq 0}$ is eventually decreasing to $\lambda\in [0,\infty)$. Then, the proof says: if $\lambda\in[0,1)$, then $\sum a_n<\infty$. Why?


The first statement says that $a_{n+1}/a_n \le \lambda$ but $$ \lim_{n \to \infty} \frac{a_{n+1}}{a_n} = \lambda \in (1,\infty). $$ So if $1 < \lambda_0 < \lambda$, eventually (for $n$ large enough), we will have $$ \frac{a_{n+1}}{a_n} > \lambda_0 \iff a_{n+1} > \lambda_0 a_n. $$

Here, $$ \lim_{n \to \infty} \frac{a_{n+1}}{a_n} = \lambda \in [0,1) $$ so analogously letting $0 \le \lambda \le \mu < 1$, for $n$ large enough you will get $a_{n+1} < \mu a_n$, and from that $n$ onwards, we can bound $\sum a_n$ by a geometric series.


Let $(b_n)_{n = 0}^\infty$ be a sequence with $b_n \le L \in (1,\infty)$ and assume $\lim_n b_n \to L$. Let $\ell \in (1,L)$. Prove that for $n$ large enough we have $b_n > \ell$.


Since $\lim_n b_n \to L$, there must exist $N \in \mathbb{N}$ such that $\forall \epsilon >0$ we have $|b_n - L| < \epsilon$ whenever $n > N$. In particular, pick $\epsilon = (L- \ell)/2$. Hence, there must exists $N \in \mathbb{N}$ such that for all $n > N$ we have $|b_n - L| < \epsilon$. Since $b_n < L$, we have $|b_n - L| = L-b_n$ and our inequality becomes $$L -b_n < \frac{L - \ell}{2}.$$ Eliminating $L/2$ and solving for $b_n$ on the LHS we get $$ b_n > L - \frac{L - \ell}{2} = \frac{L + \ell}{2} > \frac{\ell + \ell}{2} = \ell. $$

| cite | improve this answer | |
  • $\begingroup$ Thanks for your answer! You wrote on your first part that $\frac{a_{n+1}}{a_n}>\lambda_0$ eventually. Why is this the case? That was my question. $\endgroup$ – James Apr 17 at 14:02
  • $\begingroup$ @James see the update for a formal argument. The idea is, since you are increasing to a limit, you can get as close to $\lambda$ as you like for $n$ large enough. So you can get within half the distance between $\lambda$ and $\lambda_0$... $\endgroup$ – gt6989b Apr 17 at 14:53
  • $\begingroup$ Thanks for the update. It's been very helpful. I've one last question on your last part. You let $\mu\in [\lambda,1)$. Then in a similar proof (as you have written), one gets $a_{n+1}<\mu a_n$ for all large $n$. Then you wrote about a geometric series, I do not understand it. Did you mean that $\sum_{n}\mu^n a_n<\infty$ since $\mu\in [0,1)$? $\endgroup$ – James Apr 17 at 15:50
  • $\begingroup$ @James if $a_{n+1} < \mu a_n$ for all $n \ge N$, then $a_{n+2} < \mu a_{n+1} < \mu^2 a_n$ and analogously $a_{n+k} < \mu^k a_n$ and you get $$\sum_{k=n+1}^\infty a_{n+k} < \sum_{k=n+1}^\infty \mu^k a_n = a_n \sum_{k=n+1}^\infty \mu^k$$ where the right hand side is now a geometric series $\endgroup$ – gt6989b Apr 17 at 18:31
  • $\begingroup$ Thank you for your help. By the way, this is a direct consequence of the ratio test! :) $\endgroup$ – James Apr 17 at 23:15

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.