I have two contradicting proofs for the property:
\begin{align}\liminf (a_nb_n) \geq (\liminf (a_n)) * (\liminf (b_n))\end{align} where ${a_n}$ and ${b_n}$ are two bounded non-negative sequences.

(Proof 1 is given in Book: Mathematical Analysis by SC Malik and Savita Arora Edition 5 Theorem 18; Proof 2 - I thought of it)

So, I was curious to know where the flaw is in the Proof 2

Proof 1: We note that if $\liminf a_n =0$ OR $\liminf b_n =0$, then the inequality follows immediately. Therefore we assume that $\liminf a_n =a>0$ and $\liminf b_n =b>0$.

Let $\epsilon>0$ be given. Then as it is known, (for limit inferiors) there exists integers $m_1$ and $m_2$ respectively, such that $$ a_n > a - (\frac{\epsilon}{2b}), \forall n\geq m_1 \\ b_n > b - (\frac{\epsilon}{2a}), \forall n\geq m_2 $$

Therefore, for all $n \geq \max(m_1, m_2)$, we have \begin{align} a_nb_n >(a -\frac{\epsilon}{2b})*(b-\frac{\epsilon}{2a})=ab-(\epsilon + \frac{\epsilon^2}{4ab})> ab - \epsilon \end{align}This implies that, lim inf($a_nb_n) \geq ab - \epsilon$ But $\epsilon > 0$ was arbitrary, hence the results follows:
\begin{align}\liminf (a_nb_n) \geq (\liminf (a_n)) * (\liminf (b_n))\end{align}

Proof 2: Since $a$ and $b$ are limit inferiors of $a_n$ and $b_n$ respectively the following holds, \begin{align} a_n < (a +\frac{\epsilon}{2b}) & \text{ for infinitely many values of n}\\ b_n < (b +\frac{\epsilon}{2a}) & \text{ for infinitely many values of n} \end{align}Considering the following property of limit inferior: \begin{align*} \text{If } \alpha \in \mathbb{R }\text{ is such that \{ } {\text{n: } a_n < \alpha} \text{ is infinite \}, then lim inf } a_n \leq \alpha. \end{align*}From (1) and (2) above, we have \begin{align*} a_nb_n < ab + \epsilon + \frac{\epsilon^2}{4ab} & \text{ for infinitely many values of n} \end{align*}Now since $\epsilon$ is taken arbitrarily, considering small $\epsilon$ and also applying above property, we see that: \begin{align}\liminf (a_nb_n) \leq (\liminf (a_n)) * (\liminf (b_n))\end{align}which is a contradiction to the proof mentioned in the first post.


The second proof states that \begin{align} a_n < (a +\frac{\epsilon}{2b}) & \text{ for infinitely many values of $n$}\\ b_n < (b +\frac{\epsilon}{2a}) & \text{ for infinitely many values of $n$} \end{align} but that are not necessarily the same values of $n$ in both sequences. Therefore the conclusion \begin{align*} a_nb_n < ab + \epsilon + \frac{\epsilon^2}{4ab} & \text{ for infinitely many values of $n$} \end{align*} is wrong. As an example consider the sequences $$ a_n = 1, 2, 1, 2, 1, 2, \ldots \\ b_n = 2, 1, 2, 1, 2, 1, \ldots $$ If $\epsilon $ is sufficiently small then \begin{align} a_n < (a +\frac{\epsilon}{2b}) & \text{ for odd $n$}\\ b_n < (b +\frac{\epsilon}{2a}) & \text{ for even $n$} \end{align} so that there is no common subsequence where both estimates hold.


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.