On the number of compatibly Frobenius split subvarieties, prime \(F\)-ideals, and log canonical centers (Q609652)

Let \(X\) be a variety over a field of prime characteristic \(p>0\) and let \(F:X\to X\) denote the (absolute) Frobenius morphism, i.e., \(F\) is the identity map on the underlying topological space and \(\mathcal{O}_X\to F_*\mathcal{O}_X\) is the Frobenius map (or the \(p\)-th power map). \(X\) is called \(F\)-\textit{split} if there is an \(\mathcal{O}_X\)-module morphism \(\theta:F_*\mathcal{O}_X\to \mathcal{O}_X\) which splits the Frobenius map \(\mathcal{O}_X\to F_*\mathcal{O}_X\). And a subvarieties \(Z\subseteq X\) is called \textit{compatibly split with \(\theta\)} if the ideal sheaf \(\mathcal{I}_Z\) satisfies \(\theta(\mathcal{I}_Z)\subseteq \mathcal{I}_Z\). The study of \(F\)-split varieties (and compatibly split subvarieties) has played an important role in algebraic geometry, representation theory and combinatorics; and it is of great interest to bound the number of compatibly split subvarieties with a given \(F\)-splitting \(\theta\). The paper under review proves a \textit{sharp} bound when \(X\) is a projective variety: Theorem 1. Suppose that \(X\) is a projective variety over an \(F\)-finite field \(k\) of characteristic \(p>0\). Further, suppose that \(\mathcal{L}\) is an ample line bundle on \(X\) with associated section ring \(S=\bigoplus_{i\geq 0}H^0(X,\mathcal{L}^i)\). Let \(n\) be the embedding dimension of \(S\) at the irrelevant ideal \(S_+\). If \(\theta:F^e_*\mathcal{O}_X\to \mathcal{O}_X\) is a splitting of the \(e\)-th iterated Frobenius map, then there are at most \(\binom{n}{d+1}\) irreducible \(d\)-dimensional subschemes of \(X\) which are compatibly split with \(\theta\). The proof of Theorem 1 is based on the following result which is interesting in its own right: Theorem 2. Let \(\mathcal{D}\) be a collection of prime ideals in an excellent local ring \((R,\mathfrak{m})\) of embedding dimension \(n\). Suppose that the set of ideals \[ \mathcal{S}=\left\{\bigcap_{P\in \mathcal{D}'}P\;|\;\mathcal{D}' \text{ is a finite subset of }\mathcal{D}\right\} \] is closed under sums. Then the number of primes \(P\in \mathcal{D}\) such that \(R/P\) has dimension \(d\) is at most \(\binom{n}{d}\). Note that Theorem 2 is not restricted to positive characteristic and hence is applicable to other situations, for example, varieties over the complex number field. Indeed, the authors demonstrate the usefulness of Theorem 2 further by establishing a \textit{sharp} bound of the number of log canonical centers using Theorem 2: Theorem 3. Let \((X,\Delta)\) be a log canonical pair and \(x\in X\) be a point with embedding dimension \(n\). Then the number of \(d\)-dimensional log canonical centers of \((X,\Delta)\) which contain \(x\) is at most \(\binom{n}{d}\).