Show that this polynomial (involving n-th roots of unity) has rational coefficients

287 Views Asked by At

I am tasked with showing that the polynomial $\Phi(x) = \prod_{i=0}^{\phi(n)} (x - \omega_i)$ contains only rational coefficients.

To give you an example, if $n = 10$, the polynomial is just $\Phi(x) = (x - \omega^1)(x - \omega^3)(x - \omega^7)(x - \omega^9)$.

Upon working this out, I do get that the coefficients are rational numbers.

How would I go about proving the general case?

Some things I know are that the n-th cyclotomic extension (the splitting field for $x^n - 1$) has $\phi(n)$ distinct generators when viewed as a multiplicative group. I'm not sure how that helps though.

1

There are 1 best solutions below

0
On

Suppose we seek to show that

$$\Phi_n(x) = \prod_{k=1, \; \gcd(k,n)=1}^n (x - \exp(2\pi i k/n))$$

has integer coefficients.

First observe that for $n\ge 2$ $$\sum_{k=1, \; \gcd(k,n)=1}^n k = \frac{1}{2} \varphi(n) n.$$

This implies that

$$[x^0] \Phi_n(x) = \Phi_n(0) = (-1)^{\varphi(n)} \exp(2\pi i \times \varphi(n)n/2/n) \\ = (-1)^{\varphi(n)} \exp(\pi i \times \varphi(n)) = (-1)^{2\varphi(n)} = 1.$$

Observe that

$$x^n - 1 = \prod_{k=1}^n (x - \exp(2\pi i k/n)) = \prod_{d|n} \prod_{k=1, \; (k,n)=d}^n (x - \exp(2\pi i k/n)) \\ = \prod_{d|n} \prod_{q=1, \; (qd,n)=d}^{n/d} (x - \exp(2\pi i qd/n)) = \prod_{d|n} \prod_{q=1, \; (q,n/d)=1}^{n/d} (x - \exp(2\pi i q/(n/d))) \\ = \prod_{d|n} \Phi_{n/d}(x) = \prod_{d|n} \Phi_d(x).$$

We now prove the claim by induction, it holds when $n=1$ where we have $\Phi_1(x) = x - 1.$ Write

$$x^n - 1 = \Phi_n(x) \prod_{d|n, \; d \lt n} \Phi_d(x) = \Phi_n(x) g_n(x).$$

Note that the constant coefficient of $g_n(x)$ is

$$[x^0] g_n(x) = \prod_{d|n, \; d \lt n} \Phi_d(0) = -1 \times \prod_{d|n, \; 1\lt d \lt n} \Phi_d(0) = -1.$$

Let $\Phi_n(x) = \sum_{q=0}^{\varphi(n)} c_q x^q$ and $g_n(x) = \sum_{q=0}^{n-\varphi(n)} a_q x^q$

and suppose to the contrary that there is a non-integer coefficient $c_k = [x^k] \Phi_n(x)$ in $\Phi_n(x)$ with $k$ being the smallest to produce such a coefficient. Here $k\ge 1$ since we know the constant coefficient is an integer.

This would imply

$$[x^k] (x^n-1) = [x^k] \Phi_n(x) g_n(x) = \sum_{q=0}^{\min(k,\varphi(n))} c_q a_{k-q}.$$

With $k\le \varphi(n)$ by definition we thus have

$$[x^k] (x^n-1) = \sum_{q=0}^{k} c_q a_{k-q}.$$

The first $k$ terms in this sum are integers by construction (the $c_q$ precede $c_k$ and the $a_{k-q}$ are integers by induction). The last term is $c_k [x^0] g_n(x) = - c_k,$ which is not an integer. The sum of a series of integers and a single non-integer fraction is not an integer. We now have a contradiction because the coefficient $[x^k] (x^n-1)$ is always an integer, QED. (Here some of the $a_{k-q}$ could potentially be zero for small $q$ but this does not affect the argument.)

Source. This introductory lecture / textbook style proof was found at the following Internet link.