The quadratic subfield of $\mathbb{Q}(\zeta_p)$ is given by $\mathbb{Q}(\sqrt{p^*})$, where $p^*$ is the choice of $\pm p$ which is $1$ mod $4$. By some elementary Galois theory, the cyclotomic polynomial $\Phi_p = \frac{x^p-1}{x-1}$ factors into two irreducible polynomials of degree $\frac{p-1}{2}$ over this quadratic subfield, which one can describe as $$ P_{QR}(x) = \prod_{k\ \text{QR}} (x-\zeta_p^k) $$ $$ P_{QNR}(x) = \prod_{k\ \text{QNR}} (x-\zeta_p^k) $$ where $k$ ranges over the (nonzero) quadratic residues and quadratic nonresidues modulo $p$ respectively.

Is there an explicit description for the coefficients of these polynomials in terms of $\sqrt{p^*}$? The first coefficient, the one in front of $x^{\frac{p-3}{2}}$, is $\frac{1\mp\sqrt{p^*}}{2}$ by the statement of Gauss sums.


Some trivial observations. We have $$P_{QR}(1/x) x^{(p-1)/2} = \prod_{QR} (1 - x \zeta^k),$$ $$P_{QNR}(1/x) x^{(p-1)/2} = \prod_{QNR} (1 - x \zeta^k),$$ which are easier to work with. On the other hand, if $p \ge 5$, the product of $\zeta^k_p$ over quadratic residues is one, and the product over non-residues is also one. Hence we can write $$\begin{aligned} P_{QR}(1/x) x^{(p-1)/2} = & \ \prod_{QR} ( 1 - x \zeta^k) = \prod_{QR} ( \zeta^{-k} - x) \\ = & \ (-1)^{(p-1)/2} \prod_{QR} (x - \zeta^{-k}) \\ = & \ \begin{cases} P_{QR}(x), & p \equiv 1 \mod 4 \\ - P_{QNR}(x), & p \equiv 3 \mod 4 \end{cases} \end{aligned}$$ because $(-1/p) = (-1)^{(p-1)/2}$. The same swapping occurs for $P_{QNR}$. Let's consider the highest powers of $P_{QR}(x)$ and $P_{QNR}(x)$, or equivalently the lowest powers of $P_{QR}(1/x) x^{(p-1)/2}$ and $P_{QNR}(1/x) x^{(p-1)/2}$. We have $$\frac{P_{QR}(1/x) x^{(p-1)/2}}{P_{QNR}(1/x) x^{(p-1)/2}} = \prod ( 1 - x \zeta^k)^{ \left( \frac{k}{p}\right)}.$$

As mentioned there is the Gauss sum: $$\sum \left( \frac{k}{p}\right) \zeta^{k} = \sqrt{p^*},$$ and similarly $$\sum \left( \frac{k}{p}\right) \zeta^{nk} = \left( \frac{n}{p} \right) \sqrt{p^*},$$ applying $[n] \in (\mathbb{Z}/p\mathbb{Z})^{\times} = \mathrm{Gal}(\mathbb{Q}(\zeta)/\mathbb{Q})$ to both sides; when $p|n$ we interpret the RHS as being zero, and this is still correct, although we won't actually care about terms this deep into the power series above. We deduce that $$- \log \left( \frac{P_{QR}(1/x) x^{(p-1)/2}}{P_{QNR}(1/x) x^{(p-1)/2}} \right) = \sum_{n=1}^{\infty} \frac{x^n}{n} \sum \left( \frac{k}{p}\right) \zeta^{nk} = \sqrt{p^*} \sum_{n=1}^{\infty} \left( \frac{n}{p}\right) \frac{x^n}{n}.$$ Naturally $$- \log \left( P_{QR}(1/x) x^{(p-1)/2} P_{QNR}(1/x) x^{(p-1)/2} \right) = \log \left( \frac{1 - x^p}{1 - x} \right) = \log(1-x) + O(x^p) = - \sum \frac{x^n}{n},$$ and so (for example) $$\log(P_{QR}(1/x) x^{(p-1)/2}) = \frac{1}{2} \sum \frac{x^n}{n} \left(1 - \left( \frac{n}{p}\right) \sqrt{p^*}\right) + O(x^p),$$ You can now formally expand this out to get the first few terms. For example, the first non-zero term is $$\frac{1 - \sqrt{p^*}}{2},$$ and the second is $$ \frac{3 + p^* - 2 \sqrt{p^*}\left(1 + \left( \frac{2}{p}\right) \right) }{8}$$ For example, if $p \equiv 3,5 \mod 8$ so $(2/p) = -1$, this is $$\frac{3 + p^*}{8}.$$ Note the conditions on $p$ ensure that this is an algebraic integer, as it has to be. As you keep going, you get more and more terms involving the quadratic residues $(n/p)$ for small $n$, and it becomes messier and messier, and dependent on $p$ modulo higher integers. The third term, for example, is $$\frac{15 - 9 \sqrt{p^*} + 3 p^* - p^* \sqrt{p^*} + 6(p^* - \sqrt{p^*})(2/p) - 8 \sqrt{p^*} (3/p)}{48}.$$

| cite | improve this answer | |
  • $\begingroup$ Also, one can internet-search on "Aurefeuillian factorization". $\endgroup$ – paul garrett Apr 2 at 21:59
  • $\begingroup$ Thanks, this is great!! $\endgroup$ – Achim Krause Apr 2 at 22:50

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.