Sums of quaternion squares and a theorem of Watson (Q2668131): Difference between revisions

In the paper under review, the authors prove an analogue of Waring's problem for sums of squares in some quaternion rings by applying a representability theorem of [\textit{G. L. Watson}, Integral quadratic forms. Cambridge: Cambridge University Press (1960; Zbl 0090.03103)].\\ More precisely, let \(LQ_{a,b}\) denote the quaternion ring \[\{\alpha_0+\alpha_1 \mathbf{i}+\alpha_2\mathbf{j}+\alpha_3\mathbf{k}~|~\alpha_n,a,b\in {\mathbb Z},\mathbf{i}^2=-a,\mathbf{j}^2=-b,\mathbf{i}\mathbf{j}=-\mathbf{j}\mathbf{i}=\mathbf{k}\},\] and let \(LQ_{a,b}^k\) denote the additive group generated by all \(k\)-th powers in \(LQ_{a,b}\). The \emph{Waring number} \(g_{a,b}(k)\) is the least positive integer such that every element in \(LQ_{a,b}^k\) can be written as the sum of at most \(g_{a,b}(k)\) \(k\)-th powers of elements of \(LQ_{a,b}\). The main theorem of the paper concerns sums of squares, i.e., the case when \(k=2\). Note that it was previously known that \(3\leq g_{a,b}(2)\leq 5\) for all \(a,b>0\) [\textit{P. Moree}, Math. Newsl., Ramanujan Math. Soc. 24, No. 4, 103--110 (2014; Zbl 1382.11008)].\\ Theorem 1.2. Given positive square-free integers \(a\) and \(b\), if \(\gcd(a,b)\leq 2\), then \(g_{a,b}(2)=3.\)\\ The ideas of proof are as follows. By equation (2) of [\textit{A. Cooke} et al., Involve 10, No. 4, 651--664 (2017; Zbl 1416.11054)] or Theorem 1.1(ii) of [\textit{P. Pollack}, Indag. Math., New Ser. 29, No. 5, 1259--1269 (2018; Zbl 1442.11138)], it is known that \[LQ_{a,b}^2=\{\alpha_0+2\alpha_1\mathbf{i}+2\alpha_2\mathbf{j}+2\alpha_3\mathbf{k}~|~\alpha_n\in {\mathbb Z}\}.\] Now for \(\alpha=\alpha_0+2\alpha_1\mathbf{i}+2\alpha_2\mathbf{j}+2\alpha_3\mathbf{k}\in LQ_{a,b}^2,\) the goal is to find \(x,y,z\in LQ_{a,b}\) such that \[\alpha=x^2+y^2+z^2.\] Letting \(z=1+\alpha_1\mathbf{i}+\alpha_2\mathbf{j}+\alpha_3\mathbf{k},\) one has \[z^2=1-a\alpha_1^2-b\alpha_2^2-ab\alpha_3^2+2\alpha_1\mathbf{i}+2\alpha_2\mathbf{j}+2\alpha_3\mathbf{k}.\] Since the pure part of \(z^2\) agrees with that of \(\alpha,\) one has \(\alpha-z^2\in {\mathbb Z}.\) Therefore, to prove Theorem 1.2, one shows first that any integer can be written as \(x^2+y^2\) with \(x,y\in LQ_{a,b}.\) To achieve this, let \(x=x_0\in {\mathbb Z}\) and \(y=y_1\mathbf{i}+y_2\mathbf{j}+y_3\mathbf{k}\) be a pure quaternion. Then the problem reduces to showing that the indefinite quadratic form \(x_0^2-ay_1^2-by_2^2-aby_3^2\) represents all integers. Under suitable congruence conditions, the theorem of Watson (Theorem 53 of [\textit{G. L. Watson}, Integral quadratic forms. Cambridge: Cambridge University Press (1960; Zbl 0090.03103)]) shows that such indefinite quadratic form does represent all integers. The authors then use Hensel's lemma and the Chinese remainder theorem to show that these conditions are satisfied with the given \(a,b\). Applying Watson's theorem, this shows that \(g_{a,b}(2)\leq 3.\) Combining with the bounds on \(g_{a,b}(2)\) mentioned above, one then has \(g_{a,b}(2)=3,\) as required.\\ As a corollary, the following density result is obtained with the help of Theorem 1 of [Zbl 1382.11008].\\ Corollary 1.3. Let \(k\) and \(g\) be positive integers, and let \[C(g,k)=\lim_{x\rightarrow \infty}\frac{|\{(a,b)\in {\mathbb N}\times{\mathbb N}~|~a,b\leq x,g_{a,b}(k)=g\}|}{x^2}.\] If \(C(3,2)\) exists, then \(C(3,2)\geq 0.322590.\)\\ Remark. In the statement of Lemma 3.4, it appears that the condition \(\gcd(m,n)=1\) is assumed but not stated. And in line \(3\) of the fourth paragraph of its proof, the statement ``Recall that \(\gamma_i=ms\alpha_i+nt\beta_i\)'' should be replaced by ``Recall that \(\gamma_i=nt\alpha_i+ms\beta_i\)''. \\