8
$\begingroup$

In Sept. 2013, I investigated the determinant $$D_n=\det[\gcd(i-j,n)]_{1\le i,j\le n}$$ and computed the values $D_1,\ldots,D_{100}$ (cf. http://oeis.org/A228884). To my surprise, they are all positive!

Question. Does $D_n>0$ hold for all $n=1,2,3,\ldots$?

I believe that $D_n$ is always positive. How to prove this?

It is easy to see that $D_n$ is divisible by $\sum_{k=1}^n\gcd(k,n)=\sum_{d\mid n}\varphi(d)\frac nd$. It seems that $\varphi(n)^{\varphi(n)}\sum_{k=1}^n\gcd(k,n)$ divides $D_n$. Maybe there is a simple explanation for this.

Your comments are welcome!

$\endgroup$

1 Answer 1

13
$\begingroup$

Denote $f(k)={\rm gcd}(k,n)$. Clearly $f$ is an $n$-periodic function. $D_n$ is the circulant $D_n=\det(f(i-j):0\leqslant i,j\leqslant n-1)$ which equals $\prod_{k=0}^{n-1}h(\omega^k)$ where $\omega=e^{2\pi i/n}$ and $h(t)=f(0)+f(1)t+\ldots+f(n-1)t^{n-1}$. We have $$ h(t)=\sum_{d|n} \varphi(d)(1+t^d+t^{2d}+\ldots+t^{(n/d-1)d}). $$ If $t=\omega^k$, the sum $1+t^d+t^{2d}+\ldots+t^{(n/d-1)d}$ either equals $n/d$ (if $n$ divides $kd$) or equals $(1-\omega^{kn})/(1-\omega^{kd})=0$ otherwise. In any case it is a non-negative integer, and for $d=n$ it is strictly positive.

$\endgroup$

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.