2

In Diamond and Shurman, I am trying to understand the weak form of cubic reciprocity in section 4.11 - how it comes from the more standard version. In the following $A$ is the Eisenstein integers, and $\mu_3=e^{\frac{2\pi i}{3}}$.
The book says:

Let $d\in\mathbb{Z}$ be cubefree and $N=3\prod_{p|d}p$. Then there exists a character
$$\chi:A/NA)^*\to \{1,\mu_3,\mu_3^2\}$$ such that the multiplicative extension of $\chi$ to all of $A$ is trivial on $A^*$ and on primes $p \nmid N$, while on elements $\pi$ of $A$ such that $\pi\overline{\pi}$ is a prime $p\nmid N$ it is trivial if and only if $d$ is a cube modulo $p$

The standard version of cubic reciprocity also has a character the cubic residue character, however

1.The cubic residue character (the extended version) is defined for pairs $(\frac{\alpha}{\beta})$ where $\alpha,\beta\in A$ with $\beta$ coprime to $3$, whereas $\chi$ doesn't have pairs, it is defined on $(A/NA)^*$.
2.$\chi$ is supposed to be trivial on the units of $A$, and I believe the cubic reciprocity one is not.
3.$\chi$ is also (if I read correctly) trivial on rational primes $p\equiv 2\pmod{3}$, and I don't think the standard one is.

It seems the character he wants must be based on the standard one, but I don't see how.

2 Answers2

1

Let me start with the analogous question for quadratic reciprocity which may be more familiar.

The analogue of ''weak quadratic reciprocity'' is: Given a squarefree $d$, let $N=4|d|$. Then there exists a quadratic Dirichlet character $\chi$ on $(\Bbb Z/N\Bbb Z)^\times$ such that for any prime $p\nmid N$, $\chi(p)=1$ if and only if $N$ is a square mod $p$.

The character in the second part of your question is the Jacobi symbol: $\left(\frac db\right) = \prod_i\left(\frac d{p_i}\right)^{k_i}$ where $b = \prod_i p_i^{k_i}$ and $2\nmid b$.

Now for fixed $b$, $\left(\frac-b\right)$ is trivially a character mod $b$. The content of the weak quadratic reciprocity law above is that conversely for fixed $d$, $\left(\frac d-\right)$ is a Dirichlet character (*) (but in general mod $4|d|$, not mod $|d|$). This follows from the usual quadratic reciprocity law: For simplicity take $d=q$ to be an odd prime. Then for odd $n$, define $\chi(n) = (-1)^{(n-1)(p-1)/4}\left(\frac np\right)$. This is easily seen to define a Dirichlet character mod $N=4p$, and by quadratic reciprocity $\left(\frac pq\right)=\chi(q)$ for an odd prime $q\ne p$. Also note that $\chi$ is trivial on $\Bbb Z^\times$.

Now you are in the exact same situation, except we are dealing with cubic reciprocity. Again in the second part you have the cubic analogue of the Jacobi symbol (*): $\left(\frac \alpha\beta\right)_3 = \prod_i\left(\frac\alpha{\pi_i}\right)_3^{k_i}$. Now weak cubic reciprocity is saying that $\left(\frac{d}-\right)_3$ is a character mod $N$, and again this follows from the usual cubic reciprocity law.

(*) (there is a small problem in my answer: not every non-zero integer is of the form $\prod_i p_i^{k_i}$ with primes $p_i$, similarly not every non-zero element in $A$ is of the form $\prod_i \pi_i^{k_i}$ with primes $\pi_i$, in both cases the problem is units. The best way to fix this is to consider ideals instead. Then the character $\chi$ is really defined on the ray class group modulo $(N)$ which can be naturally identified with the quotient of $(A/NA)^\times$ by the image of $A^\times$. This also explains how it fits into class field theory: $\chi$ is the Artin symbol for the extension $K(\sqrt[3]d)/K$ with $K=\Bbb Q(\sqrt{-3}) =\Bbb Q(\omega)$)

leoli1
  • 8,908
  • Thanks for your answer. I get the gist! I’m still hoping to find an answer that only uses what is described e.g. in the Wikipedia article on cubic reciprocity. I have hope such exists because he calls it “Weak cubic reciprocity” and because at this point in the book he has assumed nothing advanced in number theory. In fact in his description of weak reciprocity the word prime only refers to rational primes; he doesn’t refer to Eisenstein integer primes at all, to the point where he says “element $\pi$ such that”, though it is an Eisenstein prime. Any ideas, anyone? – Jeff Margrave Nov 21 '25 at 20:15
0

I figured out an answer of the type I wanted. I'll call the character $\chi_N$ instead of $\chi$ to indicate its dependence on $N$. The requirements for $\chi_N$ are:
(a) It must be trivial on units
(b) It must be trivial on rational primes $p$ not dividing $N$
(c) If $\pi\in A$ is such that $\pi\overline{\pi}$ is a rational prime $p$ not dividing $N$, then $\chi(\pi)=1\iff d$ is a cube modulo $p$.

A basic fact from Wikipedia is that rational primes which are coprime to $3$ fall into two categories:

  • Those equivalent to $1$ modulo 3. These split into $2$ primes $\pi$ and $\overline{\pi}$ as described in (c) above.
  • Those equivalent to $2$ modulo $3$. These remain prime in $A$.

It is easy to see that the elements $p$ in (c) are the split rational primes, i.e. $\equiv 1\pmod{3}$, and so for proving requirement (b) we only need to consider $p\equiv 2\pmod{3}$.
In the Wikipedia article he defines a primary as an element coprime to $3$ and equivalent to $\pm 1\pmod{3}$. Under this definition, he then states that for any element $\lambda\in A$ coprime to $3$, exactly one of $\lambda,\mu_3\lambda\text{, and }\mu_3^2\lambda$ is primary. The definition has the weakness that the unique factorization that he states isn't quite correct because the negative of a primary is primary. Others define a primary as only being equivalent to $2\pmod{3}$, and state that any $\lambda$ coprime to $3$ has exactly one associate which is primary. Under this definition the unique factorization works. But I will use the Wikipedia definition. It is a minor difference, but I wanted to note it.

As suggested, use the character $\chi_N(r)=(\frac{N}{r})$, where $(\frac{N}{r})$ is the cubic character extended to non-primes. It is initially defined on all $R_N:=\{r\in A\ |\ r\text{ is coprime to } N\}$. Any such $r$ reduces an element of $A_N:= (A/NA))^*=R_N/NR_N$. But we need to show it is well-defined on $A_N$; I will show this below, but for now I will just use the cubic character itself.

First note that (a) is a general property of the cubic character. If the rational prime $p$ is $\equiv 2\pmod{3}$, then $3\nmid p-1$, so on $(\mathbb{Z}/p\mathbb{Z})^*$ cubing has trivial kernel, so is surjective. Therefore $d$ is a cube, say $N=N_1^3\pmod{p}$, and so $(\frac{N}{p})=(\frac{N_1}{p})^3=1$. So we only need to prove (c)

The next step is to break up $N$ into $3Part\times N_1$ where $N_1$ is coprime to $3$ and $3Part$ is $3$ or $9$. Correspondingly $\chi_N$ breaks into $\chi_{3Part}\chi_{N_1}$..
It is enough to show that if $r_1\equiv r_2\pmod{N}$, both coprime to $N$. then $\chi_{3Part}(r_1)=\chi_{3Part}(r_2)$ and $\chi_{N1}(r_1)=\chi_{N1}(r_2)$. By multiplying $r_1$ by a unit $u$, we can make it primary. So we can replace $r_1$ and $r_2$ by $r_1u$ and $r_2u$, giving the same value for $\chi_3$ and $\chi_1$. But now $r_1$ being primary and $r_1\equiv r_2\pmod{N}\implies r_1\equiv r_2\pmod{3}\implies r_2$ is primary.

The supplementary law for $3$ shows $\chi_{3Part}(r_1)=1=\chi_{3Part}(r_2)$. Also, since $r_1$ and $r_2$ are both primary, cubic reciprocity says that $\chi_{N_1}(r_1)=(\frac{r_1}{N_1})$ and $\chi_{N_1}(r_2)=(\frac{r_2}{N_1})$. But since $r_1\equiv r_2\pmod{N}$, $r_1\equiv r_2\pmod{N_1}$, and therefore by the equal property of the cubic character for numerators modulo the denominator, $(\frac{r_1}{N_1})=(\frac{r_2}{N_1})$, and by the same law again $\chi_{N_1}(r_1)=\chi_{N_1}(r_2)$. So $\chi$ is well-defined on $(A/NA)^*$.

Consider the requirement (c) by considering a rational prime $p\equiv 1 \pmod{3}$ with $N(\pi)=\pi\overline{\pi}=p$.
Lemma. $\pi\mid N^k-1\iff N\text{ is a cube }\pmod{\pi}\iff N\text{ is a cube }\pmod{p}$.
Proof. $A/pA$ has $p^2$ elements, and it breaks up into the direct sum of $A/\pi A$ and $A/\overline{\pi}A$, so these each have $p$ elements. Therefore the field inclusion $\mathbb{Z}/p\mathbb{Z}\to A/\pi A$ is an isomorphism, with cubes mapping to cubes. Q.E.D. This really works because $N$ is an ordinary integer.

The lemma shows that $$(\frac{N}{\pi})=1\iff N\text{ is a cube }\pmod{p}$$ whereas for part c) we need
$$(\frac{N}{\pi})=1\iff d\text{ is a cube }\pmod{p}$$

This is quite similar, but there are two differences (replacement of prime powers) between $d$ and $N$:
(1) Primes $q\ne 3$ dividing $d$ may occur as a square in $d$ but not in $N$.
(2) The power of $3$ in the definition of $N$ may be different.

Fix any prime $q$ dividing $N$ (so it is prime to $p$.).
Lemma. $q$ is a cube $\pmod{p}\iff q^2$ is.
$\implies$ Obvious. $\impliedby$ If $q^2\equiv x^3\pmod{p}$, then $$q^4\equiv x^3\pmod{p}\implies q\equiv (q^{-1}x)^3\pmod{p}$$ Since $A_N$ breaks up as a product over all primes dividing $N$ (including $3$), and since an element of $A_N$ is a cube if and only if each component is, we just need to consider cases 1) and 2)
For case 1), the lemma implies the replacement of $q^2$ in $d$ with $q$ in $N$ doesn't matter. For case 2), the possibilities are

  • $3$ doesn't divide $d$, in which case the 3Part acts trivially and so can be ignored
  • $3$ divides $d$ but $9$ doesn't, in which case $3Part$ is $3^2$, and the lemma suffices
  • $9$ divides $d$, in which case the 3 part of $N$ is $9$, the same as the previous item

I hope this is complete and typed in correctly..