Unconditional Primality Certificates for the Hexagonal 3-smooth Family p = 3m(m+1) + 1: Deterministic Pocklington Witnesses and Arithmetic Filters

Since m+1=2^{a}3^{b}, p-1=3m(m+1)=3m\cdot 2^{a}3^{b}=2^{a}3^{b+1}m. As p=3m^2+3m+1<3(m+1)^2, we get \sqrt p<\sqrt3\,(m+1)<3(m+1)=2^{a}3^{b+1}=F.

m=2^{a}3^{b}-1 is odd, so m(m+1) is even and p=3m(m+1)+1 is odd; and 3\mid 3m(m+1) gives p\equiv1\pmod3. By CRT, p\equiv1\pmod6.

q\mid p is equivalent to 3m^2+3m+1\equiv0\pmod q; completing the square, (6m+3)^2\equiv-3\pmod q, so -3 must be a quadratic residue mod q. By quadratic reciprocity [IR], \left(\frac{-3}{q}\right)=+1 iff q\equiv1\pmod3. For q\equiv2\pmod3 there is no solution, so q\nmid p.

Since \mathrm{ord}_7(2)=3 and \mathrm{ord}_7(3)=6, t:=2^{a}3^{b}\bmod7 depends only on (a\bmod3,b\bmod6), and m\equiv t-1, p\equiv3t^2-3t+1\pmod7. Multiplying 7\mid p by 5\equiv3^{-1} gives t^2-t-2\equiv0, i.e. (t-2)(t+1)\equiv0, so t\in\{2,6\}. Enumerating the 18 values of t (Table ?) yields exactly \mathcal F_7.

By (?), w_2=5 is valid iff \left(\frac{5}{p}\right)=-1. As 5\equiv1\pmod4, quadratic reciprocity gives \left(\frac{5}{p}\right)=\left(\frac{p}{5}\right), and \left(\frac{p}{5}\right)=-1 iff p\equiv2,3\pmod5. Modulo 5, p=3m(m+1)+1 takes the values 1,2,4,2,1 for m\equiv0,1,2,3,4; hence p\equiv2,3\pmod5 (the value 3 never occurs) iff m\equiv1,3\pmod5. Finally m+1=2^{a}3^{b} and 3\equiv2^{3}\pmod5 give m+1\equiv2^{\,a+3b}\pmod5 with \mathrm{ord}_5(2)=4, so m\equiv1\iff a+3b\equiv1 and m\equiv3\iff a+3b\equiv2\pmod4. Since a+3b\equiv a-b\pmod4, the claim follows.

3m^2+3m+1=(m+1)^3-m^3 (centred hexagonal identity). With X^3-Y^3=(X-Y)(X-\omega Y)(X-\omega^2Y), X=m+1, Y=m: p=\bigl((1+m)-\omega m\bigr)\bigl((1+m)-\omega^2 m\bigr). A direct computation gives N(\pi)=(1+m)^2-(1+m)(-m)+m^2=1+3m+3m^2=p; since p\equiv1\pmod3 is prime it splits into two conjugate primes of norm p, and \pi is one of them.

Factor 7=\lambda\overline\lambda with \lambda=3+\omega, \overline\lambda=2-\omega (N(\lambda)=7). Their primary associates (those \equiv2\bmod3) are \lambda^{*}=2+3\omega=-\omega^2\lambda and \overline\lambda^{*}=-1-3\omega=-\omega\,\overline\lambda, equivalently \lambda=-\omega\,\lambda^{*} and \overline\lambda=-\omega^2\,\overline\lambda^{*}. Since m\equiv2\pmod3, the primary associate of \pi=(1+m)-m\omega is

The cubic symbol depends only on the prime ideal in its lower entry, so \left(\frac{7}{\pi}\right)_{\!3}=\left(\frac{7}{\pi^{*}}\right)_{\!3}. By multiplicativity and \left(\frac{-1}{\cdot}\right)_{\!3}=1,

and \left(\frac{\omega}{\pi^{*}}\right)_{\!3}\left(\frac{\omega^2}{\pi^{*}}\right)_{\!3}=\left(\frac{\omega^3}{\pi^{*}}\right)_{\!3}=1: the units cancel. Eisenstein’s law of cubic reciprocity [IR, Ch. 9], applied to the primary primes \lambda^{*},\overline\lambda^{*},\pi^{*}, gives \left(\frac{\lambda^{*}}{\pi^{*}}\right)_{\!3}=\left(\frac{\pi^{*}}{\lambda^{*}}\right)_{\!3} and similarly for \overline\lambda^{*}, so

We evaluate in \mathbb{Z}[\omega]/(\lambda^{*})\cong\mathbb{F}_7 and \mathbb{Z}[\omega]/(\overline\lambda^{*})\cong\mathbb{F}_7. From \lambda^{*}=2+3\omega\equiv0 we get \omega\equiv4\pmod7 (and \omega^2\equiv2); from \overline\lambda^{*}=-1-3\omega\equiv0, \omega\equiv2 (and \omega^2\equiv4). Hence

Since \left(\frac{\alpha}{\theta}\right)_{\!3}\equiv\alpha^{(N(\theta)-1)/3}=\alpha^{2}\pmod\theta for N(\theta)=7, we have \left(\frac{\pi^{*}}{\lambda^{*}}\right)_{\!3}\equiv(5-m)^2 and \left(\frac{\pi^{*}}{\overline\lambda^{*}}\right)_{\!3}\equiv(3+4m)^2 in \mathbb{F}_7, each lying in the cube-roots subgroup \{1,2,4\}=\{1,\omega,\omega^2\} under the identifications above (\omega\mapsto4 resp. \omega\mapsto2). Multiplying in \{1,\omega,\omega^2\}, the product equals 1 exactly for m\equiv2,6\pmod7; this is the finite verification of Table ?. (The omitted classes m\equiv1,5\pmod7 give 5-m\equiv0 or 3+4m\equiv0, i.e. 7\mid p.)

If p=7 then m=1 and \pi is associate to \lambda, so w_3=7 does not apply; discard this case. For p>7: since b\ge1, 3\mid3^b, so m+1\equiv0\pmod3, i.e. m\equiv2\pmod3; and m+1=2^{a}3^{b} is coprime to 7, so m\not\equiv6\pmod7; finally m\equiv1,5\pmod7 give 7\mid p (Theorem 3.4), excluded. Thus m\bmod7\in\{0,2,3,4\} and Proposition 4.4 applies: 7 is a cubic residue iff m\equiv2 or 6\pmod7, hence (as m\not\equiv6) iff m\equiv2\pmod7. By (?), w_3=7 is valid iff 7 is a cubic non-residue, i.e. m\not\equiv2\pmod7. The translation to (a\bmod3,b\bmod6) follows from m\equiv2\pmod7\iff m+1\equiv3\pmod7, whose solutions in Table ? are \{(0,1),(1,5),(2,3)\}.

As in Proposition 4.4, the primary associate of \pi=(1+m)-m\omega is \pi^{*}=u\pi, where u=-1,-\omega,-\omega^2 according as m\equiv0,1,2\pmod3 — the unique unit making the leading coefficient \equiv2\pmod3. In every case the unit factors cancel in \left(\frac{7}{\pi}\right)_{\!3}=\left(\frac{\pi^{*}}{\lambda^{*}}\right)_{\!3}\left(\frac{\pi^{*}}{\overline\lambda^{*}}\right)_{\!3} (the argument of Proposition 4.4 uses only \left(\frac{-1}{\cdot}\right)_{\!3}=1 and \left(\frac{\omega}{\cdot}\right)_{\!3}\left(\frac{\omega^2}{\cdot}\right)_{\!3}=1, which hold for any associate). Reducing \pi^{*} modulo \lambda^{*} and \overline\lambda^{*} gives the linear forms of Table ?; squaring (the exponent (7-1)/3=2) and combining in \{1,\omega,\omega^2\} yields \left(\frac{7}{\pi}\right)_{\!3}=1 exactly on the listed classes. A direct check confirms 0 exceptions over all 1797 such primes with m<9000 (Appendix).