Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Following the result of Ivan Niven, in his paper

That is whenever $d\equiv 3 \pmod 4$, every algebraic integer of the form $a + 2b\sqrt{-d}$ can be expressed as the sum of three squares. Which means that $-1$ can be expressed as $s^2 + t^2 + u^2$ in $\mathcal{O}_{\mathbb{Q}(\sqrt{-7})}$. Can someone show me the representation?

Following Jyrki Lahtonen's answer, I have edited this question (adding the condition $d = 4k + 3$).

share|cite|improve this question
hmmmm... you cannot have two of the $s,t,u$ real as $4\cdot n^2\cdot 7-1\equiv 3\mod 4$ for all $n\in\mathbb Z$ hence cannot be written as a sum of $2$ squares. – Alex Becker Feb 28 '12 at 12:37
Wait, note that since $7\equiv 3\mod 4$ the integers of $\mathbb Q[\sqrt{-7}]$ are of the form $a/2+b\sqrt{-7}/2$. – Alex Becker Feb 28 '12 at 12:44
up vote 7 down vote accepted

I don't think that this can be done. $-7\equiv 1\pmod8$, so there is a square root of $-7$ in the 2-adic field $\mathbf{Q}_2$. Therefore the ring of integers of $\mathbf{Q}[\sqrt{-7}]$ can be viewed as a subring of the 2-adic integers $\mathbf{Z}_2$. But the equation $$ s^2+t^2+u^2=-1 $$ has no solutions with $s,u,t\in\mathbf{Z}_2$, because reducing such an equation modulo 8 contradicts the fact that $s^2,u^2,t^2$ are all congruent to either $0,1$ or $4$ modulo $8\mathbf{Z}_2$.

I don't have the time to check Niven's argument to see, whether he might have made a mistake somewhere. A priori it is more likely that I missed something than he would have done so.

share|cite|improve this answer
Mr Lahtonen, I think Niven gives the condition for $d$, that is $d \equiv 3 mod 4$, which is satisfied for $d=7$. I have edited my question. – Ajat Adriansyah Feb 28 '12 at 13:08
@Ajat: My argument makes Niven's claim suspect for the rings of integers of all the fields $\mathbb{Q}[\sqrt{-d}]$ for $d=7,23,31,47,\ldots$. Have you had any luck with those? – Jyrki Lahtonen Feb 28 '12 at 14:10
Indeed $(1+\sqrt{-7})/2 \mapsto 3 \in \mathbb{Z}/8\mathbb{Z}$ extends to a homomorpshism on the ring of integers of $\mathbb{Q}(\sqrt{-7})$. – WimC Feb 28 '12 at 15:48
@JyrkiLahtonen: I'd say all $d \equiv -1$ (mod 8) are suspect since $X^2-X+(d+1)/4$ has a root in $\mathbb{Z}/8\mathbb{Z}$ by Hensel lifting. – WimC Feb 28 '12 at 17:00
@WimC: For the benefit of those reader who are not familiar with the 2-adics you could consider writing the details of the use of a homomorphism to $\mathbf{Z}/8\mathbf{Z}$ as an answer! Most of us see it from the comment, but I'm sure many would prefer to see this settled that way! – Jyrki Lahtonen Feb 28 '12 at 19:35

According to the review by E G Straus,

Eljoseph, Nathan, On the representation of a number as a sum of squares, Riveon Lematematika 7, (1954). 38–43, MR0058627 (15,401d)

"points out an error in the proofs of Theorems 4 and 10 of I. Niven's paper, Trans. Amer. Math. Soc. 48, 405-417 (1940) [MR0003000 (2,147b)].

"He points out that instead of proving that every number $a+2b\sqrt{-m}$ ($a,b,m$ rational integers, $m\ge2$, square free) is the sum of squares of three integers in ${\bf Q}(\sqrt{-m})$, all that Niven's argument for Theorem 4 proves is:

"Either $a+2b\sqrt{−m}=\alpha^2+\beta^2+\gamma^2$ or $a+2b\sqrt{−m}=\alpha^2-\beta^2-\gamma^2$ where $\alpha,\beta,\gamma$ are integers in ${\bf Q}(\sqrt{-m})$.

"The author gives a valid counterexample to Theorem 10 and an invalid counter-example to Theorem 4. However Theorem 4 was disproved by C. L. Siegel [Ann. of Math. (2) 46, 313--339 (1945); MR0012630 (7,49b)], who gives an infinite number of imaginary quadratic fields, including ${\bf Q}(\sqrt{-7})$, in which 7 is not the sum of three squares."

share|cite|improve this answer
thank you very much for the reference :) – Ajat Adriansyah Feb 29 '12 at 1:49
+1: Nice digging, Gerry. Wish I knew how to do that :-( – Jyrki Lahtonen Feb 29 '12 at 11:05
@Jyrki, I think I just went to the review of Niven's paper at Math Reviews and then clicked on the link to other papers/reviews that referenced Niven's paper. – Gerry Myerson Feb 29 '12 at 21:43

This answer explains an earlier comment in more detail, as requested by Jyrki Lahtonen. I'll write $\mathbf{Z}_8$ for $\mathbb{Z}/8\mathbb{Z}$. First, as Jyrki remarked, the equation $a^2+b^2+c^2=-1$ has no solution in $\mathbf{Z}_8$. This follows by inspection from the fact that $x^2 \in \{0,1,4\}$ for all $x\in \mathbf{Z}_8$. Now let $d$ be a square free positive integer such that $d \equiv 7 \ (\bmod 8)$. Then $d \equiv 3\ (\bmod 4)$ and I'll just state the fact that in this case the ring of integers $\mathcal{O}_d$ in $\mathbb{Q}(\sqrt{-d})$ is a lattice:

$$ \mathcal{O}_d = \mathbb{Z} + \mathbb{Z} \alpha, \textrm{ where } \alpha =\frac{1+\sqrt{-d}}{2}. $$

Define $k := (d+1)/4 \in 2\mathbb{Z}$. Then the minimum polynomial of $\alpha$ is $X^2-X+k$. The next step is to show that this polynomial has two roots in $\mathbf{Z}_{2^m}$ for all $m \geq 1$. This follows by induction.

First note that $0,1$ are both roots in $\mathbf{Z}_2$. Now suppose that $x \in \mathbf{Z}_{2^m}$ is a root, so $x^2-x+k=y2^m$ for some integer $y$. Then

$$ (x+y2^m)^2-(x+y2^m)+k = xy2^{m+1} + y^22^{2m} \equiv 0\ (\bmod 2^{m+1}) $$

and in particular the minimum polynomial has a root $x' \in \mathbf{Z}_{2^{m+1}}$ such that $x' \equiv x\ (\bmod 2^m)$. In fact closer inspection of this procedure shows that there are exactly two roots in $\mathbf{Z}_{2^m}$.

Fix a root $x \in \mathbf{Z}_8$ and define a map $\varphi: \mathcal{O}_d \rightarrow \mathbf{Z}_8$ by

$$ \varphi: a + b \frac{1+\sqrt{-d}}{2} \mapsto a + bx. $$

It is easy to check explicitly that $\varphi$ is in fact a homomorphism. Multiplication works out because both $x \in \mathbf{Z}_8$ and $\alpha \in \mathcal{O}_d$ satisfy an identical quadratic relation. Now we can conclude that for any $a,b,c \in \mathcal{O}_d$

$$ \varphi(a^2+b^2+c^2) = \varphi(a)^2+\varphi(b)^2+\varphi(c)^2 \neq -1 $$

and therefore $a^2+b^2+c^2 \neq -1$.

share|cite|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.