I was wondering if an unbounded sequence may have a convergent cesaro mean ($\frac{1}{n}\sum_{k=1}^n a_n$). I was maybe thinking of $$a_n = (-n)^n$$ as a sequence having a convergent mean, but I might be wrong. Anyways, how would you proceed to prove such an intuition?

  • $\begingroup$ @JackD'Aurizio I don't think this example works, the Cesaro means of this sequence have $\pm 1/2$ as accumulation points. $\endgroup$ – Wojowu Oct 19 '16 at 20:19
  • 3
    $\begingroup$ @Wojowu: oh, you're right. Well, minor fix: $a_n=(-1)^n \sqrt{n}$ works (actually, $a_n=(-1)^n n^{1-\varepsilon}$ for some $\varepsilon>0$ works as well). $\endgroup$ – Jack D'Aurizio Oct 19 '16 at 20:23
  • $\begingroup$ Oops, don't mind me missing the unbounded requirement. $\endgroup$ – Simply Beautiful Art Oct 19 '16 at 20:27
  • 1
    $\begingroup$ In general, $a_n$ must be alternating in some sense for convergence to be possible. $\endgroup$ – Simply Beautiful Art Oct 19 '16 at 20:27
  • 1
    $\begingroup$ Let's try $a_n = (-n)^n.$ If $n$ is even, then $$a_1 + \cdots + a_n >n^n - \sum_{k=1}^{n-1}k^k > n^n - \sum_{k=1}^{n-1}(n-1)^{n-1} = n^n - (n-1)\cdot (n-1)^{n-1} = n^n(1 - [(n-1)/n]^n).$$ Now $[(n-1)/n]^n \to 1/e.$ Thus for large $n$ we have $$a_1 + \cdots + a_n > n^n(1 - 1/2).$$ Dividing that by $n$ gives rapid divergence to $\infty.$ $\endgroup$ – zhw. Oct 20 '16 at 0:17

Here's an example where all $a_n$ are nonnegative. If $n = 2^m, m = 1,2, \dots ,$ define $a_n = m.$ For all other $n$ define $a_n =0.$ Then $(a_n)$ is unbounded. But if $2^m\le n < 2^{m+1},$ then

$$\frac{a_1 + \cdots + a_n}{n} \le \frac{1 + 2 + \cdots + m}{2^m} = \frac{m(m+1)/2}{2^m}.$$

The fraction on the right $\to 0$ as $m\to \infty,$ showing the Cesaro means $\to 0.$

  • $\begingroup$ I think this answer deserves to be chosen as the best answer over mine! $\endgroup$ – String Oct 20 '16 at 13:10
  • $\begingroup$ @AlexQ: Why not chose this answer as the correct one? $\endgroup$ – String Oct 20 '16 at 13:11
  • $\begingroup$ Clear and precise $\endgroup$ – Alex Q Oct 20 '16 at 14:53

Here is another example following the same basic idea that zhw gave, but with non-zero limit:

For a given $n$, determine the highest power of $2$ that divides $n$, say $2^m$. Define $a_n:=m$. So we are talking about the sequence: $$ \begin{array}{|c|c|} \hline n&1&2&3&4&5&6&7&8&9&10&11&12&13&14&15&16&...\\ \hline a_n&0&1&0&2&0&1&0&3&0&1&0&2&0&1&0&4&...\\ \hline \end{array} $$ The idea is that $1/2$ of the sequence is at least $1$, $1/4$ is at least $2$, $1/8$ is at least $3$ etc. so on average we have $1/2+1/4+1/8+...=1$ as the Cesaro mean in the limit, as I will show in detail below.

Then it turns out that $$ \sum_{k=1}^n a_k=\sum_{m=1}^{\lfloor\log_2 n\rfloor}\lfloor n/2^m\rfloor \approx n $$ Each term in the last sum will differ less than $1$ from the content of the floor function $$ \lfloor n/2^m\rfloor\in(n/2^m-1,n/2^m] $$ and so $$ \sum_{m=1}^{\lfloor\log_2 n\rfloor}\lfloor n/2^m\rfloor >\sum_{m=1}^{\lfloor\log_2 n\rfloor} (n/2^m-1) =\sum_{m=1}^{\lfloor\log_2 n\rfloor} n/2^m-\lfloor\log_2 n\rfloor $$ and since $$ \sum_{m=1}^t n/2^m=n-\frac{n}{2^t} $$ we have $$ \sum_{m=1}^{\lfloor\log_2 n\rfloor} n/2^m-\lfloor\log_2 n\rfloor=n-\frac{n}{2^{\lfloor\log_2 n\rfloor}}-\lfloor\log_2 n\rfloor $$ On the other hand $\sum_{m=1}^\infty n/2^m=n$ is an upper bound on this sum from which it follows that $$ n-\frac{n}{2^{\lfloor\log_2 n\rfloor}}-\lfloor\log_2 n\rfloor\leq\sum_{k=1}^n a_k\leq n $$ Hence it should be evident that $\frac1n\sum_{k=1}^n a_k\to 1$. Finally, note that $a_{2^m}=m$ so $\{a_n\}$ is unbounded.

Multiplying the sequence from above by any $c\in\mathbb R$ we can construct a non-bounded sequence with Cesaro means converging to that $c$.

  • $\begingroup$ you could also let $a_n$ be what I had, and set $b_n=1$ for all $n.$ Then $c_n = a_n+b_n$ is unbounded and the C-means of $c_n$ tend to $1.$ $\endgroup$ – zhw. Oct 19 '16 at 23:56
  • $\begingroup$ @zhw: Yes, you are absolutely right of course :o) My aim was to make the non-zero elements "denser" to see how that could lead to non-zero limits. Your solution is very nice BTW! $\endgroup$ – String Oct 20 '16 at 0:00

Take the following example








$(u_n)$ is unbounded.

$(v_n)$ goes to $0$.

  • $\begingroup$ That is far more intuitive than my sequence, how would you prove that (vn) converges to 0 (which seems logical)? $\endgroup$ – Alex Q Oct 19 '16 at 20:32
  • $\begingroup$ the two susequences $v_{2n}$ and $v_{2n+1}$ tend to $0$, so ... $\endgroup$ – hamam_Abdallah Oct 19 '16 at 20:40

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.