Mobius function and Euler Product Triangle

287 Views Asked by At

Gevorg Hmayakyan proposed last year (17 September 2017) on StackExchange the following relationship concerning the mobius function. Gevorg stated that if $$∏^{n−1}_{i=1}(1−x^i)=∑_{k=0}^{n(n−1)/2}a_{n,k}x^{k} $$ then $$μ(n)=a_{n,1}+a_{n,n+1}+a_{n,2n+1}+a_{n,3n+1}+...$$

Does anyone know how prove this?

During related computer computation on this function, I also noted that $$\phi (n)=a_{n,0}+a_{n,n}+a_{n,2n}+a_{n,3n}+...$$

but why this is the case I do not know? Any insights?

PS. I had noticed that the above generating function is equivalent to a row of Euler Truncated Product Triangle (see the paper by Alex Mennen at http://www.alex.mennen.org/mahoniantri.pdf). Thus Using Mennens notation then they become $$μ(n)=∑_{k=0}^{n(n−1)/2}P(n−2,kn+1)$$ $$\phi(n)=∑_{k=0}^{n(n−1)/2}P(n−2,kn)$$

2

There are 2 best solutions below

1
On

For $\phi(n)$, series multisection means that $$\sum_{k}a_{n,kn}x^{kn}=\frac1n \sum_{j=0}^n f_n(\zeta^j x)$$ where $f_n(x)=\prod_{i=1}^{n-1}(1-x^i)$ and $\zeta=\exp(2\pi i/n)$. Therefore $$\sum_{k}a_{n,kn}=\frac1n \sum_{j=0}^n f_n(\zeta^j).$$ If $\gcd(j,n)>1$ then $f_n(\zeta^j)=0$. For $\gcd(j,n)=1$ then $f_n(\zeta^j)=f_n(\zeta)=\prod_{j=1}^{n-1}(1-\zeta_j)$ and so $$\sum_{k}a_{n,kn}=\frac{\phi(n)}n f_n(\zeta).$$ But $$f_n(\zeta)=\lim_{x\to1}\frac{(x-1)(x-\zeta)(x-\zeta^2)\cdots(x-\zeta^{n-1})}{x-1}=\lim_{x\to1}\frac{x^n-1}{x-1}=n.$$ We conclude $$\sum_{k}a_{n,kn}=\phi(n).$$

ADDED IN EDIT

The $\mu(n)$ case follows similar lines. In this case $$\sum_{k}a_{n,kn+1}=\frac1n \sum_{j=0}^n\zeta^{-j} f_n(\zeta^j).$$ This reduces to $\sum_{j:gcd(n,j)=1}\zeta^{-j}$. This is the sum of all primitive $n$-th roots of unity, which is $\mu(n)$.

0
On

Thank you "Lord Shark the Unknown". What an elegant answer. Since then I have done some further research based on your answer in relation to the general case. I deduced from Wikipedia that these are related to Ramujanim Sums. $$\sum_{k=0}^{n(n-1)/2}a_{n,kn+j} = c_n(j) =\mu (\frac{n}{(n,j)})\frac{\phi (n)}{\phi (\frac{n}{n,j})}$$