Okounkov bodies and toric degenerations (Q2376890)

If \(D\) is a big Cartier divisor on an irreducible projective variety \(X\) of dimension \(d\), then the choice of a chain (or flag) \(Y_\bullet\) of nested irreducible subvarieties \(Y_r\subset X\), all nonsingular at \(Y_d\), determines a semigroup \(\Gamma_{Y_\bullet}\subset {\mathbb N}^{d+1}\) and the corresponding Okounkov body \(\Delta_{Y_\bullet}(D)\subset {\mathbb R}^d\). This is the intersection of \(\{1\}\times{\mathbb R}^d\) with the closure of the convex hull of \(\Gamma_{Y_\bullet}\) and in general it can be rather arbitrary. However, if \(\Gamma_{Y_\bullet}\) happens to be finitely generated, then \(\Delta_{Y_{\bullet}}(D)\) is a rational polytope, which immediately suggests a connection with toric varieties. It is that connection that is examined here. If \(X\) is already a smooth toric variety and \(D\) is ample, then \(\Gamma_{Y_\bullet}\) is finitely generated and the Okounkov body is the polytope associated with \(X\). What is shown here is, very roughly, that the converse holds up to deformations. More precisely, if \(V\) is a not necessarily complete linear system then one can define \(\Gamma_{Y_\bullet}(V)\), and then then there is a flat family over \({\mathbb A}^1\) whose general fibre is the image of \(X\) under \(V\) and whose fibre \(X_0\) at \(0\) is \(\text{Proj}(k[\Gamma_{Y_{\bullet}}(V)])\). The normalisation of \(X_0\) is the toric variety associated with \(\Delta_{Y\bullet}(V)\). Moreover this construction is \(T\)-equivariant when it should be, and behaves well under restriction to linear subsystems and under restriction to \(Y_r\). It is in general dfficult to decide whether \(\Gamma_{Y_{\bullet}}(V)\) is finitely generated, but the author gives some conditions under which this does hold. An effective divisor in a linear system \(V\) is said to be maximal with respect to a prime divisor \(Y\) if it is supported on \(Y\), base divisors, and divisors in the boundary of the pseudoeffective cone. The existence of such a maximal divisor for \(Y_1\) gives finite generation for part of the semigroup, and so a suitable recursive condition (with compatibilities) provides a sufficient and in principle verifiable condition for \(\Gamma_{Y_{\bullet}}(V)\) to be finitely generated.