2

Let $f_S(m)$ be defined as in my previous question:

Let $S = \{1/n^2 : n \in \mathbb{N} \}$. Let $f_S(m)$ be the sum of the products of all $m$-tuples chosen from $S$. That is $$f_S(m) = \sum_{X \in {S \choose m}} \prod_{x \in X} x $$

I have used inclusion-exclusion to derive a formula for $f_S(m)$.

For example: Let $a,b,c$ be the $3$ multiplicands when expanding $$\zeta(2)^3 = \left( 1 + \frac 1 4 + \frac 1 9 + \cdots \right)^3$$

Then $f_S(3)$ can be solved by excluding all products that satisfy at least one of these rules: $a=b, \ b=c, \ a=c$.

  • The values that satisfy $a=b$ let $c$ be any value, so the sum of their products is $\left(1 + \left(\frac 1 4\right)^2 + \left(\frac 1 9\right)^2 + \cdots \right) \left( 1 + \frac 1 4 + \frac 1 9 + \cdots \right) =\zeta(4)\zeta(2)$. The same applies for $b=c$ and $a=c$.

  • The values that satisfy $a=b \wedge b=c$ have $a=b=c$, so the sum of the products is $\zeta(6)$. The same applies for $a=b \wedge a=c$ and $a=c \wedge b=c$.

  • For $a=b \wedge b=c \wedge a=c$ the result is again $\zeta(6)$.

By inclusion-exclusion and symmetry ordering $a,b,c$ $$f_S(3) = \frac{1}{3!}(\zeta(2)^3 - 3\zeta(4)\zeta(2) + 3\zeta(6) - \zeta(6)) = \frac 1 6 (\zeta(2)^3 - 3\zeta(4)\zeta(2) + 2\zeta(6)) $$

In fact, for larger values of $m$ satisfying rules like $a=b$ and $c=d$ turns out to be the disjoint union problem. If the rules we satisfy are $a=b$ and $c=d$, then we have the disjoint sets $(a,b), (c,d)$ and the result will be $\zeta(4)^2$. If the rules are like $a=b$ and $b=c$ then the disjoint sets are $(a,b,c), (d)$ and the result is $\zeta(6)\zeta(2)$.

I wrote a Mathematica program to calculate $f_S(m)$. Unfortunately it's quite slow, from evaluating all the inclusion-exclusion subsets.

altTotal[x_] := Total@x[[;; ;; 2]] - Total@x[[2 ;; ;; 2]];

inex[n_, zeta_] :=
 Module[{productPairs, summands},
  productPairs = Subsets[Range[n], {2}];
  summands = Table[
    conComp = ConnectedComponents /@ 
      Map[Join[#, Table[{i, i}, {i, 1, n}]] &, 
       Subsets[productPairs, {depth}]];
    Total[Times @@@ Map[zeta[2 Length[#]] &, conComp, {2}]],
    {depth, 1, Length[productPairs]}];
  Return[(zeta[2]^n - altTotal[summands])/n!]
  ]

I have determined the following:

\begin{align} f_S(3) &= \frac 1 6 (\zeta(2)^3 - 3\zeta(4)\zeta(2) + 2\zeta(6)) \\ f_S(4) &= \frac{1}{24} \left(\zeta(2)^4-6 \zeta(4) \zeta(2)^2+8 \zeta(6) \zeta(2)+3 \zeta(4)^2-6 \zeta(8)\right) \\ f_S(5) &= \frac{1}{120} \left(\zeta (2)^5-10 \zeta (4) \zeta (2)^3+20 \zeta (6) \zeta (2)^2+15 \zeta (4)^2 \zeta (2)-30 \zeta (8) \zeta (2)-20 \zeta (4) \zeta (6)+24 \zeta (10)\right) \\ f_S(6) &= \frac{1}{720} \left(\zeta (2)^6-15 \zeta (4) \zeta (2)^4+40 \zeta (6) \zeta (2)^3+45 \zeta (4)^2 \zeta (2)^2-90 \zeta (8) \zeta (2)^2-120 \zeta (4) \zeta (6) \zeta (2)+144 \zeta (10) \zeta (2)-15 \zeta (4)^3+40 \zeta (6)^2+90 \zeta (4) \zeta (8)-120 \zeta (12)\right) \end{align}

My question is how can I derive these expressions? (my linked answer has the closed-form already) The coefficients match A181897. Wikiversity: Permutations by cycle type looks insightful; unfortunately I don't know any group theory.

What is the connection between the cycle index of the symmetric group and this inclusion-exclusion process for $f$?

(Program that calculates much more efficiently)

f[m_, zeta_] := ((-1)^m) CycleIndexPolynomial[SymmetricGroup[m], -zeta /@ Range[2, 2 m, 2]]
qwr
  • 11,362

2 Answers2

1

These may be computed with the unlabeled set operator

$$\def\textsc#1{\dosc#1\csod} \def\dosc#1#2\csod{{\rm #1{\small #2}}} \textsc{SET}$$

applied to the Riemann Zeta function

$$\zeta(s) = \sum_{n\ge 1} \frac{1}{n^s}.$$

We use the recurrence by Lovasz for the cycle index $Z(P_n)$ of the set operator $\textsc{SET}_{=n}$ on $n$ slots, which is

$$Z(P_n) = \frac{1}{n} \sum_{l=1}^n (-1)^{l-1} a_l Z(P_{n-l}) \quad\text{where}\quad Z(P_0) = 1.$$

This recurrence lets us calculate the cycle index $Z(P_n)$ very easily. We are interested in

$$f(s, m) = Z(P_m; \zeta(s))$$

and hence the recurrence becomes

$$f(s,m) = \frac{1}{m} \sum_{l=1}^m (-1)^{l-1} \zeta(ls) f(s, m-l) \quad\text{where}\quad f(s,0) = 1.$$

This yields e.g. for $f(2,4)$

$$1/24\, \left( \zeta \left( 2 \right) \right) ^{4}-1/4 \,\zeta \left( 4 \right) \left( \zeta \left( 2 \right) \right) ^{2}+1/3\,\zeta \left( 6 \right) \zeta \left( 2 \right) +1/8\, \left( \zeta \left( 4 \right) \right) ^{2}-1/4\,\zeta \left( 8 \right)$$

and for $f(2,5)$

$${\frac { \left( \zeta \left( 2 \right) \right) ^{5}}{ 120}}-1/12\,\zeta \left( 4 \right) \left( \zeta \left( 2 \right) \right) ^{3}+1/6\,\zeta \left( 6 \right) \left( \zeta \left( 2 \right) \right) ^{2}+1 /8\,\zeta \left( 2 \right) \left( \zeta \left( 4 \right) \right) ^{2}\\-1/4\,\zeta \left( 8 \right) \zeta \left( 2 \right) -1/6\,\zeta \left( 4 \right) \zeta \left( 6 \right) +1/5\,\zeta \left( 10 \right)$$

We also have for $f(3,7)$

$$-1/12\, \left( \zeta \left( 3 \right) \right) ^{2} \zeta \left( 6 \right) \zeta \left( 9 \right) +1/8\, \zeta \left( 3 \right) \zeta \left( 6 \right) \zeta \left( 12 \right) -1/6\,\zeta \left( 18 \right) \zeta \left( 3 \right)\\ -{\frac {\zeta \left( 6 \right) \left( \zeta \left( 3 \right) \right) ^{5}}{240}}+{ \frac {\zeta \left( 9 \right) \left( \zeta \left( 3 \right) \right) ^{4}}{72}}+1/48\, \left( \zeta \left( 3 \right) \right) ^{3} \left( \zeta \left( 6 \right) \right) ^{2}-1/24\,\zeta \left( 12 \right) \left( \zeta \left( 3 \right) \right) ^{3}\\+1/10\, \zeta \left( 15 \right) \left( \zeta \left( 3 \right) \right) ^{2}-1/48\,\zeta \left( 3 \right) \left( \zeta \left( 6 \right) \right) ^{3}\\+1/18\,\zeta \left( 3 \right) \left( \zeta \left( 9 \right) \right) ^{2}+1/24\, \left( \zeta \left( 6 \right) \right) ^{2}\zeta \left( 9 \right) -1/10\,\zeta \left( 6 \right) \zeta \left( 15 \right) \\-1/12\,\zeta \left( 9 \right) \zeta \left( 12 \right) +1/7\,\zeta \left( 21 \right) +{\frac { \left( \zeta \left( 3 \right) \right) ^{7}}{5040}}.$$

(Sorted and formatted by Maple.)

Marko Riedel
  • 64,728
  • Whew, this answer went way over my head. Do you know of any introductory books or articles where I can learn about this? (You mentioned Lovasz, I'm not sure what exactly that refers to) – qwr Aug 22 '18 at 17:41
  • It refers to the book Combinatorial problems and exercises, being one reference for the cycle index. The computation presented here is actually an application of the Polya Enumeration Theorem (PET) and there are many more examples using Dirichlet series at the following link at MSE Meta, section is "number theory and algebra." – Marko Riedel Aug 22 '18 at 17:56
  • I still don't understand why picking $m$-tuples is related to the symmetric group and the cycle index. Is my application of "rules" like $a=b$ the reason why the formula is related to the symmetric group? – qwr Aug 28 '18 at 18:31
0

I think I finally have some intuition on the relationship between my inclusion-exlcusion "rules" and cycles in permutations.

Suppose I were at the stage in my inclusion-exclusion calculations where I needed to consider $(a = b) \wedge (b=c) \wedge (d=e)$. This corresponds to $a=b=c$ and $d=e$. If we consider $a=b$ as $1 \mapsto 2$, $b=c$ as $2 \mapsto 3$, and $d=e$ as $4 \mapsto 5$, there is a bijection between these rules and a permutation consisting of a 3-cycle and 2-cycle: $(1 \ 2 \ 3)(4 \ 5)$. (Here $a=1, b=2,\dots$)

Similarly if my rules were $(a=c) \wedge (b=c)$ this corresponds to $1 \mapsto 3 \mapsto 2$ or the cycle $(1 \ 3 \ 2)$. Since every cycle can be represented uniquely by having its least element written first, I believe every combination of rules corresponds exactly to every permutation in $S_n$, organized by which cycles it contains.

qwr
  • 11,362