Euclidean rings of S-integers in complex quadratic fields

The procedure described in the previous section leads to the proof of Theorem 1. Implementing this procedure on a computer gives the result immediately, but this could be done by hand as the number of intervals necessary is not large. Below we give the value of X necessary and the number of intervals produced.

The proof is a simple application of Corollary 1. Let y\in[0,1]\cap \mathbb{Q}. Set X=\sqrt{D}/\sqrt{3}. Suppose that S contains all primes \leq \lceil X\rceil so that T contains all positive integers \leq \lceil X \rceil. We will follow the proof of Dirichlet’s approximation theorem to show that there exists a c \in T such that \{cy\} < 1/X or \{cy\}>1-1/X. Subdivide the unit interval as

where k=\lceil X\rceil is the number of sub-intervals. Each of the k+1 numbers \{0y\}, \{y\}, \ldots \{ky\} must land in one of the k subintervals. By the Pigeonhole Principle, there must exist an s,t \in \mathbb{Z} satisfying 0 \leq s < t \leq \lceil X \rceil such that \{sy\},\{ty\} differ by less than X. Upon setting c=t-s we obtain the desired result.

Let d>0 be squarefree and S=\{p\} with p odd and (-d/p)\neq 1. Set \xi_0=(1+w)/2. Let \alpha\in\mathcal{O}_{S} be arbitrary and write \alpha=(a+bw)/p^n where a,b,n\in\mathbb{Z}, n\geq 0, with a and b not both divisible by p. Then we have

where we have rewritten 2a-p^n=p^m A and 2b-p^n=p^mB with m\geq 0 maximal. Note that A,B\neq 0 and both of A,B cannot be divisible by p.

First consider the case where -d\equiv 2,3\pmod{4}. We have

Suppose (-d/p)=-1. We claim that A^2+dB^2 is not divisible by p. Indeed if, p divides A^2+dB^2, then we must have A,B\not\equiv 0\pmod {p}, lest both A and B are divisible by p; in this case, one can solve A^2+dB^2\equiv 0\pmod{p} to show that -d is a square mod p, a contradiction. Therefore N_S(\alpha-\xi_0)\geq (1+d)/4\geq 1 when d\geq 3. We turn to the case where (-d/p)=0. This is the same except that A^2+dB^2 is possibly divisible by one copy of p. If A+dB^2 is not divisible by p, the result follows as before, so we may assume p divides A^2+dB^2. In this case, A must be divisible by p as well, which implies that A^2+dB^2\geq p^2+d and hence N_S(\alpha-\xi_0)\geq (p^2+d)/(4p)\geq 1 when d\geq 3. Consequently, K is not S-norm-Euclidean when -d\equiv 2,3\pmod{4} and d\geq 3.

Next we turn to the case -d\equiv 1\pmod{4}. We have

If (-d/p)=-1, then (2A+B)^2+dB^2 is not divisible by p, and it follows that then N_S(\alpha-\xi_0)\geq (1+d)/16\geq 1 when d\geq 15. If (-d/p)=0 then (2A+B)^2+dB^2 is divisible by at most one copy of p. As before, we may assume p divides (2A+B)^2+dB^2. In this case, we find that 2A+B is divisible by p and thus N_S(\alpha-\xi)\geq (p^2+d)/(16p). This gives N_S(\alpha-\xi_0)\geq 1 when d\geq 15 except when (d,p) equals one of the pairs (15,3), (15,5), (35,5), (35,7). We deal with the remaining cases in Section ?.

Let d>0 be squarefree and S=\{2\}. Set \xi_0=(1+w)/3. Let \alpha\in\mathcal{O}_{S} be arbitrary and write \alpha=(a+bw)/2^n where a,b,n\in\mathbb{Z}, n\geq 0. We aim to show that N_S(\alpha-\xi_0)\geq 1. Observe that

We rewrite 3a-2^n=2^m A and 3b-2^n=2^m B where A,B\in\mathbb{Z} and m\geq 0 is maximal; note that A,B\neq 0 and that A and B cannot both be even.

First consider the case where -d\equiv 2,3\pmod{4}. We have

We claim that A^2+dB^2 is divisible by at most one power of 2; indeed, the assumption A^2+dB^2\equiv 0\pmod{4} together with -d\equiv 2,3\pmod{4} leads to A,B\equiv 0\pmod{2}, a contradiction. Whence

which implies N_S(\alpha-\xi_0)\geq 1 when d\geq 17. Since \alpha\in\mathcal{O}_S was arbitrary, this proves that K is not S-norm-Euclidean when -d\equiv 2,3\pmod{4} and d\geq 17.

This leaves open the cases of d=10,13,14. However, if d is even then the assumption that 2 divides A^2+dB^2 leads to 2 divides A and therefore N_S(\alpha-\xi_0)\geq (2^2+d)/18\geq 1 when d\geq 14. We will return to the particular cases of d=10,13 later.

Next, consider the case where -d\equiv 1\pmod {4}. We assume, in addition, that -d\not\equiv 1\pmod{8} so that (1+d)/4 is an odd integer. We have

We claim that A^2+AB+((1+d)/4)B^2 is not divisible by 2; indeed, if this were the case, then A^2+AB+B^2\equiv 0\pmod{2} which leads to A,B\equiv 0\pmod{2}, a contradiction. Therefore

Hence N_S(\alpha-\xi_0)\geq 1 when d\geq 35.

Now we return to the case of d=13. We choose \xi_0=w/3. In a similar manner as before, we write

where 2^mA=3a, 2^mB=3b-2^n, A,B not both even, and A^2+13B^2 is divisible by at most one power of 2. Notice that B\neq 0 but that A could equal 0. If 2 does not divide A^2+13B^2, then N_S(\alpha-\xi_0)\geq 13/9\geq 1. Hence we may assume 2 divides A^2+13B^2. In this case, it follows that A and B are both odd, and since A is a multiple of 3, we have N_S(\alpha-\xi_0)\geq (3^3+13)/18\geq 1.

It turns out the case of (d,p)=(10,2) will require a little more care. We return to this case in the next section.

Observe

and cz\xi-cz\alpha\in\mathcal{O} which implies

Let \xi\in K be arbitrary. We must show that there exists \alpha\in\mathcal{O}_{S} such that N_S(\xi-\alpha)<1. Without loss of generality, we may assume that \xi=(x+yw)/z where z is not divisible by any primes in S, by multiplying \xi by an element of T if necessary; indeed, if u\in T, then N_S(u\xi-\alpha)=N_S(u)N_S(\xi-u^{-1}\alpha) and N_S(u)=1. Moreover, we may assume \xi\in\mathcal{F} by subtracting an appropriate element of \mathcal{O}.

Noting that N(\xi - \alpha) = |\xi - \alpha|^2, Lemma 1 gives the result.

Suppose [0,1]\cap\mathbb{Q}\subseteq \bigcup_{j,k} I_{j}^k. Taking \pi^{-1} of both sides allows use to observe that

In light of the preceding discussion, we conclude that K is S-norm-Euclidean.

For the first part of the proof, we will show that

To establish this, we show that the condition in Corollary 1 holds for all y\in[0,1] except when y\in\{1/3,2/3\}. First, observe that y\in\{1/3,2/3\} implies the fractional part \{2^ny\} belongs to \{1/3,2/3\} for all n\geq 0, and hence the condition clearly fails for these two values of y. Note that \sqrt{3/40}\approx 0.274<1/3.

Let y\in[0,1] be arbitrary. If there exists n\geq 0 such that \{2^ny\}\in[0,1/4]\cup[3/4,1], then we are done, since 1/4<\sqrt{3/40}. By way of contradiction, suppose that \{2^n y\}\in[1/4,1/2] for all n\geq 0. We have y\in J_0:=[1/4,1/2] or y\in J'_0:=[1/2, 3/4]. For sake of concreteness, suppose y\in J_0, but the case of y\in J'_0 is treated in a similar manner. Since y\in [1/4,1/2], we have 2y\in[1/2,1] but by hypothesis it must be that 2y\in J_1:=[1/2,3/4]. Similarly, 4y\in J_2:=[5/4,3/2]. We inductively define a sequence of intervals J_0,J_1,J_2,\dots in this manner, all of length 1/4, having the property that 2^ny\in J_n. Therefore y\in\bigcap_n 2^{-n}J_n, and since this intersection contains at most one element, we get a contradiction, except for one possible value of y. Similarly, if y\in J'_0, we can inductively define J'_0,J'_1,J'_2,\dots and obtain a contradiction, except for one possible value of y. Hence we have proven that the condition in Corollary 1 holds with at most two possible exceptions. But since we already know 1/3 and 2/3 are exceptions, this proves (Proof).

It suffices to show that given \xi=r/s+w/3\in\mathcal{F} with s odd, there exists \alpha\in\mathcal{O}_S such that N_S(\xi-\alpha)<1. (If \xi is of the form r/s+2w/3, then multiply by 2 and translate back into \mathcal{F} using an element of \mathcal{O}.) Notice that 0\leq r/s\leq 1 since \xi\in\mathcal{F}. First set \alpha=w/2. We find

Therefore N_S(\alpha-\xi_0)<2(r/s)^2+5/9<1 when r/s<\sqrt{2}/3\approx 0.47. If \alpha=(2+w)/2, then we find N_S(\alpha-\xi_0)<2(1-r/s)^2+5/9<1 when r/s>1-\sqrt{2}/3\approx 0.53.

If \alpha=(2+w)/4, then N_S(\alpha-\xi_0)\leq 4(1/2-r/s)^2+5/9<1 when 1/6\leq r/s\leq 5/6. This completes the proof.

Suppose S=\{p\} with p=3 or p=5. We proceed in a similar manner as in Proposition 1. First we claim that \bigcup_{j,k}I_j^k \supseteq[0,1]\setminus\left\{1/2\right\}. For sake of concreteness, we prove this when p=3. Let y\in[0,1] be arbitrary. Suppose \{3^n y\}\in[1/3,2/3] for all n\geq 0. (If this condition does not hold, then we are done since 1/3<\sqrt{3/15}.) This means that y\in J_0:=[1/3,2/3], 3y\in J_1:=[4/3,5/3], 9y\in J_2:=[13/3,14/3], and so on. Inductively, we find

Therefore y\in \bigcap_n 3^{-n}J_n =\{1/2\}. This proves the claim when p=3. When p=5, the proof is similar. Noting that 2/5<\sqrt{3/15}, we can assume that \{5^n y\}\in[2/5,3/5] for all n\geq 0. This leads to J_n:=[5^n/2-1/10, 5^n/2+1/10] and the result follows as before.

Let \xi=r/s+w/2\in\mathcal{F} with (p,s)=1. We may assume 0\leq r/s<1. If \alpha=w, then N_S(\xi-\alpha)\leq(r/s-1/4)^2+15/16<1 when 0<r/s<1/2. Similarly, if \alpha=1, then N_S(\xi-\alpha)\leq (r/s-3/4)^2+15/16<1 when 1/2<r/s<1. This leaves only the points \xi=w/2 and \xi=(1+w)/2. When \xi=w/2, we compute N(\xi)=15/16 so that N_S(\xi)<1 in either case. Consider \xi=(1+w)/2. We have N(\xi)=3/2 so that N_S(\xi)=1/2<1 in the case p=3. For the case p=5, we compute N_S(\xi-2)=5/2 so that N_S(\xi-2)=1/2<1. This completes the proof.

We have

\xi-\alpha=(A+Bw)/(cz) where A=cx-az and B=cy-bz so that

As in the proof of Lemma 1 we have

If in addition, the conditions (4) hold, then 2A+B\equiv 0\pmod{p} which implies the numerator is divisible by p, and therefore N_S(\xi-\alpha)\leq p^{-1}c^2 N(\xi-\alpha).

Set S=\{p\}. We first consider p=5. Making use of Lemmas 2 and 4 we obtain a list of 14 circles of various radii that suffice to cover \mathcal{F}. See Table Proof for the centers of the chosen circles (as elements \alpha\in\mathcal{O}_S) and the radius afforded by our lemmas.

In principle, one could verify by hand that the 14 provided circles cover \mathcal{F}. We believe the picture in Figure Proof that displays the covering is fairly convincing. If one is not convinced by the picture, one could partition the fundamental domain into 5^6 parallelograms of equal area in the obvious way, and use a computer to check that each parallelogram lies entirely inside one of the given circles. When p=7, the argument is similar. We give a list of 20 circles (see Table Proof) that carry out the covering (also depicted in Figure Proof).