Writing binomial coefficients as sums of three squares (Q607708)

According to the classical representation theorem by \textit{A. M. Legendre} [Essay on number theory. Cambridge Library Collection - Mathematics. Cambridge: Cambridge University Press (2009; Zbl 1205.11005)] and by \textit{C. F. Gauß} [Disquisitiones Arithmeticae. (Reprint) Remagen: Verlag Kessel (2009; Zbl 1167.11001)], a natural number \(n\) can be represented as a sum of three squares if and only if it cannot be written as \(4^{t} (8u+7)\), \(t,u\in \mathbb N\). By using elementary number theory, some methods from stochastic matrices and graph theory, \textit{A. Granville} and \textit{Y. Zhu} [Am. Math. Mon. 97, No. 6, 486--493 (1990; Zbl 0717.11016)] proved the following two conjectures: a) For every positive integer \(n\), except 1,2,3,4,5,9,14,17,18,20,21,35,41 there exists an integer \(k\), \(0\leq k\leq n\), divisible by 4, for which \(\binom{n}{k}\) cannot be represented as a sum of three squares; b) The set of integers \(n\), for which \(\binom{2n}{n}\) cannot be represented as a sum of three squares, has asymptotic density 1/8 in the set of all positive integers. Although they provided a fundamental improvement in writing binomial coefficients as sums of three squares, nevertheless Granville and Zhu left some open questions, especially in the conjecture (a), like the eventual existence of a superior limit for \(k\) and the possibility to drop the condition \(4|k\). Aim of the present paper is to complete such previous work in a generalized theorem-proving frame and to refine it through relatively straightforward new strategies corroborated by computer checking. Let \({\mathcal{S}}\) denote the set consisting of those integers which can be written as sums of three squares: the authors prove that if \(0 \leq k \leq n\) and \(\binom{n}{0}, \binom{n}{1},\dots, \binom{n}{k} \in \mathcal{S}\), then \(k \leq 73\). Then they study how many consecutive binomial coefficients may belong to \({\mathcal{S}}\), and prove that for any given \(k\) we can find infinitely many values of \(n\) such that at least \(k\) consecutive coefficients \(\binom{n}{j}, \binom{n}{j+1},\dots,\binom{n}{j+k-1}\) belong to \({\mathcal{S}}\). The authors also prove the existence of infinitely many quadruples of consecutive binomial coefficients that cannot be written as sums of three squares and that from five consecutive binomial coefficients at least one is a sum of three squares.