31
$\begingroup$

I am stuck with the following problem.

Prove that $$\limsup_{n \to \infty} (a_n+b_n) \le \limsup_{n \to \infty} a_n + \limsup_{n \to \infty} b_n$$

I was thinking of using the triangle inequality saying $$|a_n + b_n| \le |a_n| + |b_n|$$ but the problem is not about absolute values of the sequence.

Intuitively it's clear that this is true because $a_n$ and $b_n$ can "reduce each others magnitude" if they have opposite signs, but I cannot express that algebraically...

Can someone help me out ?

$\endgroup$
  • $\begingroup$ I'm sorry, my book (Rudin) doesn't use the notation $\sup_{k \ge n}a_k$ and I am not sure what it means. $\endgroup$ – hyg17 Jun 1 '13 at 2:11
37
$\begingroup$

Define for all natural numbers $k$: $A_k = \sup\{ a_n: n \ge k \}$, $B_k = \sup\{ b_n: n \ge k \}$ (where $A_k, B_k \in \mathbb{R} \cup \{+\infty\}$, (They are decreasing because for larger $k$ we take the $\sup$ of fewer terms), so that by definition $\limsup_{n \to \infty} a_n = \lim_{k \to \infty} A_k$ and similarly for $B_k$ and $\limsup_{n \to \infty} b_n$. Also we consider the $C_k = \sup \{ (a_n + b_n) : n \ge k \}$, so that $\lim_{k \to \infty} C_k = \limsup_{n \to \infty} (a_n+b_n)$.

Now, fix an index $k$, then for all $n \ge k$ we have $a_n + b_n \le A_k + B_k$, because we estimate $a_n$ by the supremum of all terms of $(a_n)$ with $n \ge k$ and likewise for the $b_n$. As (for fixed $k$) the right hand side is fixed:

$$C_k = \sup \{ (a_n + b_n : n \ge k \} \le A_k + B_k\mbox{.}$$

This holds for all $k$, so we take the $\inf$ or $\lim$ on both sides as $k$ tends to infinity, and this preserves the inequality and we are done.

$\endgroup$
  • $\begingroup$ It's looking a lot clearer, but I'm still puzzled about the part where you said, "$a_k+b_k \le A_k +B_k \Rightarrow a_k+b_k \le A_n+B_n$". If $A_n+B_n$ is smaller than $A_k+B_k$, how is the above inequality guaranteed ? $\endgroup$ – hyg17 Jun 1 '13 at 17:56
  • $\begingroup$ @hyg17 you are right, I fixed the argument (hopefully). $\endgroup$ – Henno Brandsma Jun 1 '13 at 18:19
  • $\begingroup$ Okay, now it's 100% lucid. Thanks ! It was a lot more helpful than you can imagine. $\endgroup$ – hyg17 Jun 1 '13 at 18:29
  • $\begingroup$ @hyg17 Glad to be able to help! $\endgroup$ – Henno Brandsma Jun 1 '13 at 18:31
  • 1
    $\begingroup$ @AshokVardhan try two sequences that alternate between two values $\endgroup$ – Henno Brandsma Aug 28 '15 at 15:34
8
$\begingroup$

Hint:

Given two sequences, $\;\displaystyle \{a_n\}_{n \in \Bbb N},\;$ $\,\{b_n\}_{n \in \Bbb N},\;$

and given the definition of the supremum of a sequence, we can see that for every $k\geq n$, $$(a_k + b_k) \;\; \leq \;\;\sup_{k\geq n} a_k + \sup_{k\geq n} b_k\,.$$

Now how does this imply that $$\lim_{n\to \infty} \sup(a_n + b_n) \;\; \leq \;\; \lim _{n \to \infty} \sup a_n + \lim_{n\to \infty} \sup b_n\quad ?$$

Added: see Definition 3.16, Theorem 3.17, 3.19: perhaps you'd prefer to use the notation used there.

$\endgroup$
  • $\begingroup$ I am puzzled. From my understanding $\sup a_k$ and $\sup b_k$ is a fixed number, so if $a_k + b_k$ happens to be increasing, the supremum of $a_k+b_k$ can only equal to the sum of the suprema... right ? $\endgroup$ – hyg17 Jun 1 '13 at 2:10
  • $\begingroup$ Recall that we are taking the limit of the suprema as $n\to \infty$. The supremum say of $a_n$ is fixed for a fixed n, or for a fixed k. We are simply stating that lim sup of the sum of the terms (an + bn) is necessarily less than or equal to the lim sup of an + lim sup of bn, when $n \to \infty$. $\endgroup$ – Namaste Jun 1 '13 at 2:20
  • $\begingroup$ Examine Thm. 3.19 $\endgroup$ – Jake Color Jun 1 '13 at 2:20
  • $\begingroup$ @amWhy I don't understand. The supremum of a sequence is not always larger than the terms of a sequence. As a matter of fact, {$1 \over n$} is a sequence where all of it's temrs are larger than the supremum. So how can $a_k \le \sup a_k$ at all ? $\endgroup$ – hyg17 Jun 1 '13 at 2:31
  • $\begingroup$ @hyg17 for your example $(a_n) = (\frac{1}{n})$, $s_k = \sup \{a_n: n \ge k \} = \frac{1}{k}$ for all $k$ and its limit is $0$, which is thus the $\limsup a_n$. But the supremum of the tail of the sequence $s_k$ is at least as large as all the terms it is supremum of.. $\endgroup$ – Henno Brandsma Jun 1 '13 at 5:04
3
$\begingroup$

I happened to making solutions for PMA, so I would like to share my solution here:

If either $\limsup_{n \to \infty} a_n = +\infty$ or $\limsup_{n \to \infty} b_n = +\infty$, there is nothing to prove. So we may assume $\limsup_{n \to \infty} a_n = A, \limsup_{n \to \infty} b_n = B$, where $A < +\infty, B < +\infty$ (but each of them can possibly take $-\infty$).

Given $\varepsilon > 0$, there exist $N_1, N_2 \in \mathbb{N}$, such that $a_n < A + \varepsilon/2$ for all $n \geq N_1$ and $b_n < B + \varepsilon/2$ for all $n \geq N_2$. Take $N = \max(N_1, N_2)$, it follows that for all $n \geq N$, $$a_n + b_n < A + \varepsilon/2 + B + \varepsilon/ 2 = A + B + \varepsilon.$$ Let $n \to \infty$ in the above equation, we conclude that $\limsup_{n \to \infty} (a_n + b_n) \leq A + B + \varepsilon$. Since $\varepsilon$ is arbitrary, it follows that $\limsup_{n \to \infty}(a_n + b_n) \leq A + B$, proving the result.

$\endgroup$
2
$\begingroup$

This question occurs in Rudin's PMA, page 78, Exercise 5. Below is my trial of proof. Is it right? Any comments are welcome!

5. $\quad $For any two real sequences $\{a_n\}, \{b_n\},$ prove that \begin{gather*} \limsup_{n\to\infty}(a_n+b_n)\leq \limsup_{n\to\infty} a_n+\limsup_{n\to\infty} b_n, \end{gather*} provided the sum on the right is not of the form $\infty-\infty.$

proof: $\quad$ Put $a^*=\limsup\limits_{n\to\infty} a_n, b^*=\limsup\limits_{n\to\infty} b_n$ and $c^*=\limsup\limits_{n\to\infty} (a_n+b_n).$ We shall show $c^*\leq a^*+b^*.$

Since it is excluded that the right is of the form $\infty-\infty,$ if one of $a^*$ and $ b^*$ is $-\infty,$ then the assertion holds. So we assume that $a^*, b^*>-\infty.$ And if one of $a^*$ or $b^*$ is $+\infty,$ then $c^*=a^*+b^*,$ so we we need to consider the case $a^*, b^*\in\mathbb{R}.$

By Theorem 3.17 (see Page 56 of Rudin's PMA), for every $\epsilon>0$ there exists $N_1, N_2$ such that \begin{gather*} a_n<a^*+\frac{\epsilon}{2},\qquad \forall n>N_1,\\ b_n<b^*+\frac{\epsilon}{2},\qquad \forall n>N_2. \end{gather*} Hence we have \begin{gather*} a_n+b_n<a^*+b^*+\epsilon,\qquad \forall n>\max\{N_1, N_2\}. \end{gather*} Then, for every convergent subsequence $a_{\sigma(n)}+b_{\sigma(n)}$ of sequence $(a_n+b_n)_{n\in\mathbb{N}},$ we have \begin{gather*} \lim_{n\to\infty} (a_{\sigma(n)}+b_{\sigma(n)})\leq a^*+b^*+\epsilon. \end{gather*} Thus we see that $a^*+b^*+\epsilon$ is an upper bound of the set $C$ of subsequential limits of $(a_n+b_n)_{n\in\mathbb{N}}.$ Hence $c^*=\sup C\leq a^*+b^*+\epsilon.$ By the arbitrariness of $\epsilon>0,$ it follows that $c^*\leq a^*+b^*.$ $\Box$

$\endgroup$

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.