Let $K$ be a number field, with ring of integers $\mathcal{O}_k$. For $x\in \mathcal{O}_K$, let $f(x) = |N_{K/\mathbb{Q}}(x)|$, the (usual) absolute value of the norm of $x$ over $\mathbb{Q}$. Dirichlet's unit theorem tells us, in a precise sense, "how many" units (i.e., elements with $f(x)=1$) there are in $\mathcal{O}_K$.

My question is, are there results out there about the size (or structure) of the set of elements $\alpha\in \mathcal{O}_K$ with $f(\alpha)=n$, for some given positive integer $n$? (To make things more concrete, we could just focus on $n=2$, if it helps.)

Of course, if $f(\alpha)=n$ and $u$ is a unit, $f(u\alpha)=n$ too, so we should mod out by the action of $\mathcal{O}_K^*$ on $\mathcal{O}_K$ by multiplication, i.e., look at equivalence classes of associated elements of $\mathcal{O}_K$. If we let $S =$ this set of equivalence classes, $f$ is well-defined on $S$. For a given $n\in\mathbb{N}$, is $f^{-1}(n)$ finitely generated or (dare I hope it) finite?


Okay here is part of an answer to your question:

Let $x \in O_K$. Then it can be shown that the norm of $x$ is the norm of the ideal $(x)$ (I am presuming you know what the norm of an ideal is. If not, take a look at Pierre Samuel's book "Algebraic Theory of Numbers"). But, since $O_K$ is a Dedekind domain, it can be shown (using an argument of unique factorization of ideals into products of primes) that there are only finitely many integral ideals in $O_K$ of a given norm. Thus, up to units, there will be only finitely many elements in $O_K$ of a given norm.

Here is the reason why there are only finitely many ideals of norm $n \in \mathbb{N}$:

Let $I \subset O_K$ be an ideal of norm $n$. Then by definition of the norm of an ideal, $|O_K/I| = n$. Thus, $n \in I$. This means that $(n)O_K \subset I$. Factorize $(n)O_K$ into a product of prime ideals. Then one can see that there are only finitely many choices for $I$.

  • $\begingroup$ I should mention that this is a standard proof, which you can read in most ANT textbooks. I feel bad, because I have left out details in the proof. If any part of my argument is unclear to you, then read up some relevant facts about factorization of ideals in Dedekind domains. $\endgroup$ – Rankeya Dec 18 '11 at 4:12
  • 1
    $\begingroup$ Don't feel bad! This answers my basic question of whether it's finite. $\endgroup$ – John Conecker Dec 18 '11 at 4:25
  • 7
    $\begingroup$ You don't need to know anything about prime ideals to prove finiteness. Just note that $n\mathcal{O}_K\subseteq I\subseteq \mathcal{O}_K$, so it's enough to prove finiteness of $\mathcal{O}_K/n\mathcal{O}_K$. But $\mathcal{O}_K$ is a free $\mathbb{Z}$-module of rank $d=$ degree of $K$, so that quotient is isomorphic to $(\mathbb{Z}/n\mathbb{Z})^d$ as an abelian group, and the ideal $I$ corresponds to a subgroup of that. The multiplicative structure of $I$ only enters in a very weak way: all you need to know is that $I$ is a subgroup of $\mathcal{O}_K$ of full $\mathbb{Z}$-rank. $\endgroup$ – Alex B. Dec 18 '11 at 4:38
  • 1
    $\begingroup$ Well, that is true, but if you are studying ANT, it is good to know about factorization of ideals in Dedekind domains. $\endgroup$ – Rankeya Dec 18 '11 at 4:40

Counting elements of a given norm modulo units is the same as counting principal ideals of a given norm. It's a quite difficult problem. For example, how many elements of norm $N$ are there in $\mathbf{Z}[i]$? In other words, in how many ways can we write $N$ as the sum of two squares? A theorem of Jacobi says that this is equal to four times the excess of the number of divisors of $N$ which are $\equiv 1 \mod 4$ over the number of divisors of $N$ which are $\equiv 3 \mod 4$, generalizing the theorem of Euler-Fermat about representations of primes as sums of two squares. Thus we see that for even the simplest number field $\mathbf{Q}(i)$, the situation is quite complicated (but in this case, Jacobi's theorem provides a simple formula).

The number $a_N$ of ideals of norm $N$ in $\mathcal{O}_K$ is related to the residue of the Dedekind zeta function $\zeta(s)$ at $s=1$, which is in turn related to all important invariants of $K$.

We can write

$$\zeta_K(s)=\sum_{n \geq 1}\frac{a_n}{n^s},$$

where the sum converges absolutely for $\Re s>1$.

According to the Wiener–Ikehara Tauberian theorem, since $\zeta_K$ has a simple pole at $s=1$,

$$\sum_{n\leq N}a_n \sim C_K N$$

where $C_K$ is the residue of $\zeta_K$ at $s=1$, given explicitly by the class number formula.

This gives an upper bound for the number of principal ideals of norm at most $N$, and an asymptotic expression in case $\mathcal{O}_K$ is a P.I.D.

Addendum: As pointed out by Matt, ideals are uniformly distributed among the $h$ ideal classes of $\mathcal{O}_K$. This is the heart of the class number formula.

By definition of the ideal class group of $\mathcal{O}_K$, the set of principal ideals of $\mathcal{O}_K$ is the identity element of the ideal class group. By the uniformity of the distribution, if we denote $L$ the average over $n$ of the number of principal ideals of norm $n$ as $n \to \infty$, then

$$\frac{1}{h}\sum_{n\leq N}a_n \sim \frac{1}{h}C_KN \sim LN$$

hence $$L=\frac{1}{h}C_K = \frac{2^{r_1}(2\pi)^{r_2}\text{Reg}_K}{\omega_K \sqrt{|D_K|}}$$

where $r_1$ and $r_2$ are respectively the number of real and complex infinite primes of $K$, $\text{Reg}$ is the regulator (the covolume of the unit group $\mathcal{O}_K^*$ in $K^*$), $D_K$ is the discriminant (more or less the square of the covolume of $\mathcal{O}_K$ in $K$), and $\omega_K$ is the number of roots of unity in $K^*$.

Addendum #2: I just want to point out what I think is the neatest proof of of Jacobi's theorem, using $L$-functions.

We let $K=\mathbb{Q}(i)$. By definition of the $\zeta_K(s)$, and using the fact that there are two primes of norm $p$ above $p \equiv 1 \mod 4$, one prime of norm $p^2$ above $p \equiv 3 \mod 4$, and one prime of norm $2$ above $p=2$,

$$\zeta_K(s)= (1-2^{-s})^{-1}\prod_{p \equiv 3 \mod 4} (1-p^{-2s})^{-1} \prod_{p \equiv 1 \mod 4} (1-p^{-s})^{-2}$$

which, be rearranging, can be written as $\zeta(s)L(s, \chi)$, where $\chi$ is the Dirichlet character $$p \mapsto \begin{cases}\left(\frac{-1}{p}\right) = (-1)^{(p-1)/2} && p \text{ odd}; \\ 0 && p=2\end{cases}.$$

Hence, from this identity, the coefficient of $N^{-s}$ in $\zeta_K$ is

$$a_N = \sum_{d \mid N}\chi(d),$$

which is precisely what Jacobi's theorem says, since for a general odd integer $n$ we have $\chi(n)=(-1)^{(n-1)/2} = \pm 1$ according as to $n\equiv 1$ or $n \equiv 3$ mod $4$.


  • 3
    $\begingroup$ Dear Bruno, By considering the $L$-functions $L(\chi,s)$ as $\chi$ ranges over the various characters of the ideal class group, can't we deduce that for each ideal class $c$, the sum $\sum_{\mathfrak a \in c} N(\mathfrak a) \simeq \dfrac{1}{h} C_K N$? Regards, $\endgroup$ – Matt E Dec 18 '11 at 5:07
  • $\begingroup$ Dear @MattE, you are right! I will add it immediately. Thank you! $\endgroup$ – Bruno Joyal Dec 18 '11 at 5:19
  • 1
    $\begingroup$ Dear Bruno, You're welcome. Regards, $\endgroup$ – Matt E Dec 18 '11 at 7:03
  • $\begingroup$ Rankeya answered my basic question of finiteness, but thank you for the additional information! Lovely stuff. $\endgroup$ – John Conecker Dec 19 '11 at 17:32
  • 3
    $\begingroup$ The post has been un-wikified. $\endgroup$ – Zev Chonoles Dec 28 '11 at 15:01

This is an addendum to Addendum 2 in Bruno's great answer.

Let $d$ be a square free integer, let $K$ be the quadratic field $\mathbb Q(\sqrt d)$, let $\mathbb Z_K$ be its ring of integers, and let $D$ be the integer which equal to $d$ if $d\equiv1\bmod4$, and to $4d$ otherwise.


Borevich-Shafarevich, Number Theory, p. 237,

Borevitch-Chafarevitch, Théorie des Nombres, p. 263,

define $\chi:\mathbb Z\to\mathbb Z$ as follows. If $x$ is not prime to $D$, put $\chi(x)=0$. If $x$ is prime to $D$, let



$$d=2d'\implies\chi(x)=(-1)^{[(x^2-1)/8]+[(x-1)/2]+[(d'-1)/2]}\left(\frac{x}{|d'|}\right),$$ where $(\frac{a}b)$ is the Jacobi symbol.

Then $\chi$ induces a multiplicative map from $\mathbb Z/D\mathbb Z$ to $\mathbb Z$. Moreover, if $p$ is prime, then

$\bullet$ $\chi(p)=1$ if $p\mathbb Z_K$ is the product of two prime ideals of norm $p$,

$\bullet$ $\chi(p)=-1$ if $p\mathbb Z_K$ is a prime ideal of norm $p^2$,

$\bullet$ $\chi(p)=0$ if $p\mathbb Z_K$ is the square of a prime ideal of norm $p$.

This implies $$ \zeta_K(s)=\zeta(s)\ L(s,\chi), $$ that is $$ \text{(Dedekind)}=\text{(Riemann)}\cdot\text{(Dirichlet)}, $$ and the number of ideals of $\mathbb Z_K$ of norm $n$ is $$ \sum_{m|n}\ \chi(m). $$ Now assume $d < 0$.

$(*)$ The number $u$ of units of $\mathbb Z_K$ is $4$ if $d=-1$, it is $6$ if $d=-3$, and it is $2$ otherwise.

See for instance this entry of PlanetMath.

Suppose in addition that $\mathbb Z_K$ is principal, i.e. that $d$ is one of the numbers

$$-1, -2, -3, -7, -11, -19, -43, -67, -163.$$

See for instance this sub-entry of Wikipedia.

The number of elements of norm $n$ is then $$ u\ \sum_{m|n}\ \chi(m), $$ where $u$ is the number of units of $\mathbb Z_K$, see $(*)$.


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.