On extensions of algebraic groups with finite quotient (Q903158)

Let \(k\) be a field, and suppose \[ 1\rightarrow N\rightarrow G\rightarrow Q\rightarrow 1 \] is a short exact sequence of \(k\)-group schemes of finite type. In general, such a sequence is not split. However, we can ask whether \(G=N\cdot H\) for some subgroup \(H\) of \(G\) with \(N\cap H\) finite, where \(N\cdot H=(N\rtimes H)/(N\cap H)\) and \((N\cap H\)) is viewed as a subgroup of \((N\rtimes H)\) via \(x\mapsto\left( x,x^{-1}\right)\). If such an \(H\) exists, then the extension is called quasisplit. It is known that if \(Q\) is smooth and \(N\) is an abelian variety then the sequence is quasisplit. The main result of this paper is that if \(Q\) is finite then the extension is quasisplit. Furthermore, the \(H\) above can be chosen to be finite, and if \(Q\) is étale and \(k\) is perfect then \(H\) can be chosen to be étale (and finite). The proof is accomplished by reducing to the case where \(Q\) is étale and \(N\) is either a smooth connected unipotent group, a torus, or an abelian variety. Applications to the above theorem are given. Chevalley's Structure Theorem is extended:\ here it is shown that any algebraic group \(G\) has a smallest affine normal subgroup \(N\) such that \(G/N\) is proper, and that this \(N\) is connected. Also, there is an extension to the Rosenlicht decomposition of smooth connected groups: in the new version, for \(G\) an algebraic group there exists an affine subgroup \(H\) of \(G\) such that \(G=G_{\text{ant}}\cdot H\) where \(G_{\text{ant}}\) is the largest antiaffine subgroup of \(G\); furthermore if \(G\) is smooth and \(k\) is perfect then we can choose \(H\) to be smooth. Finally, is is shown that any extension with \(Q\) reductive and char \(k=0\) is quasisplit.