Infinite/Recursive Cesàro Summation of $\zeta(1)$

155 Views Asked by At

Is anything known about this kind of `infinite' Cesàro summation (or any related types of summation)?

If we have a function we wish to sum $f(n)$, but $$ S^0[f] = \sum_{n=1}^\infty f(n) $$ diverges, so we take the average of partial terms (Cesàro summation) $$ s_1=\frac{1}{n_1}\sum_{n_0=1}^{n_1} f(n_0)\\ S^1[f] = \lim_{n_1 \to \infty}s_1=\lim_{n_1 \to \infty}\frac{1}{n_1}\sum_{n_0=1}^{n_1} f(n_0) $$ but say this also diverges, so we iterate $$ s_2 = \frac{1}{n_2}\sum_{n_1=1}^{n_2} \frac{1}{n_1}\sum_{n_0=1}^{n_1} f(n_0)\\ S^2[f] = \lim_{n_2 \to \infty} s_2 = \lim_{n_2 \to \infty}\frac{1}{n_2}\sum_{n_1=1}^{n_2} \frac{1}{n_1}\sum_{n_0=1}^{n_1} f(n_0) $$ then what can be said about $S^{\infty}$, if this is performed through a limit on the partial $s_k$'s?

Some examples:

For, $f(n)=n$ the intermediate terms look like $$ s_k=\frac{(2^k-1)n_k + n_k^2}{2^k n_k}\\ S^{\infty}[f] = \lim_{k \to \infty} \frac{(2^k-1)n_k + n_k^2}{2^k n_k} = 1 $$

For, $f(n)=n^2$ the intermediate terms appear to look like $$ s_k=\frac{(2^{k+1}-3^{k+1}+6^k)n_k + 3(3^k-2^k)n_k^2 + 2^k n_k^3}{6^k n_k}\\ S^{\infty}[f] = \lim_{k \to \infty} \frac{(2^{k+1}-3^{k+1}+6^k)n_k + 3(3^k-2^k)n_k^2 + 2^k n_k^3}{6^k n_k} = 1 $$

  • For $f(n)=n+1$, $S^{\infty}[f]=2$
  • For $f(n)=n-1$, $S^{\infty}[f]=0$
  • It seems that for $f(n)=n+k$, $S^{\infty}[f]=1+k$

The most interesting case seems to be $f(n)=\frac{1}{n}$ which then relates to $\zeta(1)$. We have

\begin{equation} s_1 = \frac{H(n)}{n} \\ s_2 = \frac{H(n)^2+H^{(2)}(n)}{2n} \\ s_3 = \frac{H(n)^3+3H(n)H^{(2)}(n)+2H^{(3)}(n)}{6n} \\ s_4 = \frac{H(n)^4+6H(n)^2H^{(2)}(n)+3H^{(2)}(n)^2 + 8 H(n)H^{(3)}(n)+6H^{(4)}(n)}{24n} \\ \end{equation} where the sum in $s_j$ seems to be over products of harmonic numbers $H^{(k_i)}(n)$ such that $\sum_i k_i = j$. It is not clear how to take the limit for $s_\infty$. The coefficients seem related to A102189 (and A036039), which makes sense. The paper referenced in A102189 does mention similar coefficients in a related expansion for multiple zeta values but in terms of powers of $\zeta(k)$ rather than powers of $H^{(k)}(n)$. In some sense the infinite limit $s_{\infty}$ is the related to $\zeta(1,1,1,\cdots)$ for an infinite number of $1$'s.

This seems to relate to this Wikipedia article. Which would imply $$ s_k(n) = \frac{1}{n}\sum_{m_1 + 2m_2 + \cdots + km_k = k \atop m_1\ge 0, \ldots, m_k\ge 0} \prod_{i=1}^k \frac{H^{(i)}(n)^{m_i}}{m_i ! i^{m_i}} $$ but it is not clear how an infinite limit $k \to \infty$ would be taken.

Edit:

Based on an identity here, it would seem that $$ \text{Li}_k(z) = \sum_{n \ge 1}\frac{(-1)^{n-1}n s_k(n)z^n}{(1-z)^{n+1}} $$

1

There are 1 best solutions below

0
On

If we put the terms of the partial sums into a column-vector $S_0$ then with left-multiplication of a matrix $$S_1 = C \cdot S_0 $$ we get the partial sums $S_1$ and so on.

The matrix $C$ for the getting of $S_1$ looks like:

       1    .    .    .    .    .    .    .
     1/2  1/2    .    .    .    .    .    .
     1/3  1/3  1/3    .    .    .    .    .
     1/4  1/4  1/4  1/4    .    .    .    .
     1/5  1/5  1/5  1/5  1/5    .    .    .
     1/6  1/6  1/6  1/6  1/6  1/6    .    .
     1/7  1/7  1/7  1/7  1/7  1/7  1/7    .
     1/8  1/8  1/8  1/8  1/8  1/8  1/8  1/8

The matrix for getting $S_2$ is its second power $C^2$ by $S_2 = C^2 \cdot S_0$ :

            1         .         .         .         .        .       .     .
          3/4       1/4         .         .         .        .       .     .
        11/18      5/18       1/9         .         .        .       .     .
        25/48     13/48      7/48      1/16         .        .       .     .
      137/300    77/300    47/300     9/100      1/25        .       .     .
       49/120    29/120    19/120    37/360    11/180     1/36       .     .
      363/980   223/980   153/980  319/2940  107/1470   13/294    1/49     .
     761/2240  481/2240  341/2240  743/6720  533/6720  73/1344  15/448  1/64

and for the limit to $n \to \infty$ this goes towards $C^\infty$

     1  .  .  .  .  .  .  .
     1  .  .  .  .  .  .  .
     1  .  .  .  .  .  .  .
     1  .  .  .  .  .  .  .
     1  .  .  .  .  .  .  .
     1  .  .  .  .  .  .  .
     1  .  .  .  .  .  .  .
     1  .  .  .  .  .  .  .

To make still sense of using the limit $n \to \infty$ of $C^n$ one could normalize the columns of the powers of $C_n=C^n$ such that its diagonalentries are always $1$; then we get in the limit $C_\infty^* = P$ where $P$ is the Pascal-/Binomialmatrix.

It is nice to see in the case of finite dimension, how that $P$ matrix provides also a diagonalization of $C$: $C^k=P \cdot Z^k \cdot P^{-1}$ where $Z=diagonal([1,1/2,1/3,...])$ contains the eigenvalues $[1,1/2,1/3,...])$ of $C$.

      P                            diag(Z)   P^{-1}
    ------------------------------------------------------------------------------
     1  .   .   .   .   .  .  .  |    1  |   1   .    .    .    .   .   .  .  |
     1  1   .   .   .   .  .  .  |  1/2  |  -1   1    .    .    .   .   .  .  |
     1  2   1   .   .   .  .  .  |  1/3  |   1  -2    1    .    .   .   .  .  |
     1  3   3   1   .   .  .  .  |  1/4  |  -1   3   -3    1    .   .   .  .  |
     1  4   6   4   1   .  .  .  |  1/5  |   1  -4    6   -4    1   .   .  .  |
     1  5  10  10   5   1  .  .  |  1/6  |  -1   5  -10   10   -5   1   .  .  |
     1  6  15  20  15   6  1  .  |  1/7  |   1  -6   15  -20   15  -6   1  .  |
     1  7  21  35  35  21  7  1  |  1/8  |  -1   7  -21   35  -35  21  -7  1  |
  .  -  -   -   -   -   -  -  -  +    -  +   -   -    -    -    -   -   -  -  +

and we see, again, that for $Z^\infty$ this scheme gives the $C_\infty$ matrix only.

Curiosity-note: by this diagonalization we can even define fractional powers of $C$ !