Geometry of normed bilinear maps and the 16-square problem (Q1105630)

Results on the existence and nonexistence of sums of square formulas of type [r,s,n], i.e., identities of the form \[ (*)\quad (x_ 1^ 2+...+x^ 2_ r)(y^ 2_ 1+...+y^ 2_ s)=f_ 1(x,y)^ 2+...+f_ n(x,y)^ 2 \] in the polynomial ring \({\mathbb{R}}[x_ 1,...,x_ r,y_ 1,...,y_ s]\), have been of great interest, especially since the last century when it was confirmed that [2,2,2], [4,4,4], [8,8,8] are possible types while [16,16,16] is impossible. The smallest integer n, denoted by \(16*_{{\mathbb{R}}} 16\), for which there exists a formula of type [16,16,n] remains unknown. Various methods from the 1960's narrow down the range of n to 23\(\leq n\leq 32\). The values 23, 24 were subsequently ruled out. In this paper, the authors present a deeper study of the geometry behind the formula (*), and use sophisticated methods in algebraic topology and homotopy theory to establish \(16*_{{\mathbb{R}}} 16\geq 29\).