Algebraic \(K\)-theory and sums-of-squares formulas (Q2572490)

The authors use algebraic geometry to extend to every field \(F\) of odd characteristic a classical result on sum of squares. More precisely, a \textit{sum-of-squares formula of type \([r,s,n]\) over a field \(F\)} is a polynomial identity in the ring \(F[x_1,\dots,x_r,y_1,\dots,y_s]\) of the form: \[ (x_1^2 + \dots + x_r^2)(y_1^2 + \cdots + y_s^2) = z_1^2 + \cdots + z_n^2, \] where the \(z_k\) are bilinear expressions in the \(x_i\) and the \(y_j\). The main result of the paper is the following theorem (Th. 1.1): Assume that the characteristic of \(F\) is different of 2. Then if a sum-of-square formula of type \([r,s,n]\) exists over \(F\), the 2-valuation of the binomial coefficient \((^n_i)\)is at least \([\frac{s-1}{2}]-i+1\), for \(i=n-r,\dots,[\frac{s-1}{2}]\). For instance this result shows that a formula of type \([13,13,16]\) cannot exist over such a field. In case \(F\) is the real field \(\mathbb R\), the above theorem was essentially proven by \textit{M. F. Atiyah} [Topology 1, 125--132 (1962; Zbl 0109.41101)] as pointed out by \textit{S. Yuzvinsky} [Mich. Math. J. 28, 131--145 (1981; Zbl 0469.15013)] and it can be extended to any field with characteristic 0 by a purely algebraic argument due to K.Y. and T. Y. Lam, as explained by \textit{D. B. Shapiro} [Expo. Math. 2, 235--261 (1984; Zbl 0541.10025)]. The proof of the authors relies on a modification of Atiyah's original argument: by computations in algebraic \(K\)-theory one can determine restrictions on the geometric dimension of certain algebraic vectors bundles given by such a square formula. And this yields the theorem above.