Newton polytopes of constants of locally nilpotent derivations (Q4500816)

Let \(B=k[x_1, \dots, x_m]\) be the ring of polynomials in the free generators \(x_1, \dots, x_m\) over an arbitrary field \(k\) of characteristic zero. Any polynomial in this ring can be written as a finite sum \(\alpha= \sum_v \alpha_vz^v\) (\(v \in \mathbb Z_+^m\)). The convex hull of all \(v\)'s such that \(\alpha_v\neq 0\) is called the Newton polytope of \(\alpha\). A derivation \(\partial\) of \(B\) is called locally nilpotent if for any element \(\beta\in B\) there exists \(n\) such that \(\partial^n(\beta)=0\). The subring \(A=\ker\partial\) of \(B\) is called the ring of constants of the derivations \(\partial\). Any polytope having vertices with nonnegative integer coordinates is the Newton polytope of some polynomial. However, most of such polynomials are not Newton polytopes of constants of any locally nilpotent derivation. In the paper, it is shown that the Newton polytopes of constants of locally nilpotent derivations are quite restricted. Their vertices must lie on the coordinate hyperplanes, and their edges are either contained in the coordinate hyperplanes or possess one coordinate that divides the other coordinates. In the two variable case, Rentschler's classification of locally nilpotent derivations of \(k[x,y]\) is obtained as a corollary.