Polynomials vanishing on Cartesian products: the Elekes-Szabó theorem revisited (Q504018)

This 50-pages paper has a 6-page introduction, which succinctly summarizes the motivation and the results of earlier papers. I recommend reading this introduction to anyone, who wants to know more about this paper. G. Elekes and his collaborators, working on hard Erdős problems in discrete geometry, identified an algebraic problem behind the geometric problems: how many zeros a trivariate polynomial \(F(x,y,z)\) can have on a finite Cartesian product set \(A\times B\times C\)? It turned out that the answer depends on whether the polynomial is generic or belongs to a very special family, at least for the instance \(F(x,y,z)=f(x,y)-z\). The paper under review generalizes the existing results, and reproduces the best earlier bounds, which were available for special cases. The main theorem states: Let \(F\in \mathbb{C}[x,y,z]\) be an irreducible polynomial of degree \(d\), and assume that none of the derivatives \(\partial F/\partial x\), \(\partial F/\partial y\), \(\partial F/\partial z\) is identically zero. Let \(Z(F) \) denote the zero set of \(F\). Then one of the following two statements hold: (i) For all \(A,B,C \subset \mathbb{C}\) with \(|A|=|B|=|C|=n\), we have \(|Z(F)\cap(A\times B\times C)|=O(d^{13/2}n^{11/6})\), (ii) There exists a 1-dimensional subvariety \(Z_0\subset Z(F)\) such that for all \(v\in Z(F)\setminus Z_0\) there exist open sets \(D_1,D_2,D_3\subset \mathbb{C}\) and one-to-one analytic functions \(\phi_i:D_i\rightarrow \mathbb{C}\) with analytic inverses, for \(i=1,2,3\), such that \(v\in D_1\times D_2\times D_3\) and for all \((x,y,z)\in D_1\times D_2\times D_3\), \((x,y,z)\in Z(F)\) if and only if \(\phi_1(x)+\phi_2(y)+\phi_3(z)=0\). The theorem can be extended (a) to reducible polynomials, (b) for cases, where \(A,B,C\) are not of the same size, (c) for polynomials of real variables instead of complex variables. The bounds in part (i) of the theorem are unlikely to be tight. One application of the main theorem is the following: Any \(n\) points on a constant-degree irreducible algebraic curve in \(\mathbb{C}^2\) determines \(O(n^{11/6})\) collinear triples of 3 distinct points, unless the curve is a line or a cubic.