The coherent cohomology ring of an algebraic group (Q385539)

Let \(k\) be a field and let \(G\) be a \(k\)-group scheme with multiplication map \(\mu : G \times G \to G\), neutral element \(e_G \in G(k)\) and inverse map \(\iota :G \to G\). Then \(H^*(G):=\bigoplus_{i \geq 0} H^i(G,\mathcal O_G)\) has the structure of a graded commutative Hopf algebra with comultiplication \(\mu^*\), counit \(e_G^*\) and antipode \(\iota^*\). When \(G\) is affine, the Hopf algebra \(H^*(G)\) is just the algebra \(\mathcal O(G)\) of global sections of \(\mathcal O_G\), which uniquely determines the group scheme \(G\) (indeed, \(G=\text{Spec}\,\mathcal O(G)\)). But the latter is not the case if \(G\) is not affine. For example, if \(G\) is an abelian variety, then the structure of \(H^*(G)\) only depends of \(g := \text{dim}(G)\). Indeed, by a result of Serre, \(H^*(G)\) is the exterior algebra \(\Lambda^*(H^1(G))\), where \(H^1(G)\) has dimension \(g\). Further, \(H^1(G)=P^*(G)\), where \(P^*(G)\) denotes the graded subspace of \(H^*(G)\) consisting of primitive elements (recall that \(\gamma \in H^*(G)\) is primitive if \(\mu^*(\gamma) = \gamma \otimes 1 + 1 \otimes \gamma\)). In this paper, the author obtains the following generalization of Serre's result. Theorem. Assume that \(G\) is of finite type over \(k\). Then the graded algebra \(H^*(G)\) is the exterior algebra of the \(\mathcal O(G)\)-module \(H^1(G)\), which is free of finite rank. If \(G\) is connected, then there exists an isomorphism of graded Hopf algebras \[ H^*(G) \cong \mathcal O(G) \otimes \Lambda^*(P^1(G)). \] Moreover, \(P^i(G) = 0\) for all \(i \geq 2\). As a consequence of the above, the graded Lie algebra \(P^*(G)\) equals \(P^0(G) \oplus P^1(G)\), and hence is abelian; also, the vector space \(P^1(G)\) is finite-dimensional. Note that \(P^0(G)\) consists of the homomorphisms of group schemes \(G \to \mathbb G_a\); this vector space is finite-dimensional in characteristic \(0\), but not in prime characteristics (already for \(G = \mathbb G_a\)). When \(G\) is an abelian variety and \(k\) is perfect, the structure of \(H^*(G)\) follows without difficulty from that of connected graded-commutative Hopf algebras and from the isomorphism of \(H^1(G)\) with the Lie algebra of the dual abelian variety. But for an arbitrary group scheme \(G\), the above theorem is not a direct consequence of general structure results on Hopf algebras such as those of Cartier-Gabriel-Kostant and Milnor-Moore, since \(H^*(G)\) is neither connected nor cocommutative. The proof of the theorem is based on the affinization theorem. This theorem asserts that \(G\) has a smallest normal subgroup scheme \(H\) such that the quotient \(G/H\) is affine; then \(\mathcal O(G/H) \cong \mathcal O(G)\) via the quotient morphism \(G \to G/H\), which is therefore identified with the canonical morphism \(G \to \text{Spec}\, \mathcal O(G)\). In particular, the \(k\)-algebra \(\mathcal O(G)\) is finitely generated. Moreover, \(H\) is smooth, connected and contained in the center of the neutral component \(G^o\); in particular, \(H\) is commutative. Also, one has \(\mathcal O(H) = k\), i.e., \(H\) is ``anti-affine''. In fact, \(H\) is the largest anti-affine subgroup scheme of \(G\), which is denoted \(G_{\text{ant}}\) below. By analyzing the quotient morphism \(G \to G/G_{\text{{ant}}}\), one obtains an isomorphism of \(\mathcal O(G)\)-modules \(\psi : H^*(G) \overset{\cong}{\longrightarrow} \mathcal O(G) \otimes H^*(G_{\text{ant}})\) which identifies the pull-back \(H^*(G) \to H^*(G_{\text{ant}})\) to \(e_G^* \otimes \text{id}\). On the other hand, using the structure of anti-affine groups and additional arguments, it is shown that the Hopf algebra \(H^*(G_{\text{ant}})\) is the exterior algebra of \(H^1(G_{\text{ant}})\), which is a finite-dimensional vector space. This implies the first assertion of the theorem. When \(G\) is connected, it is shown that the above map \(\psi\) is an isomorphism of graded Hopf algebras. Moreover, \(P^1(G) \cong H^1(G_{\text{ant}})\) via pull-back. This yields a description of the primitive elements which takes very different forms in characteristic \(0\) and in positive characteristics. A rather unexpected consequence follows: In positive characteristics, the group schemes \(G\) such that \(H^*(G) = k\) are trivial; in characteristic \(0\), they are exactly the fibered products \(S \times_A E\), where \(S\) is an anti-affine extension of an abelian variety \(A\) by a torus, and \(E\) is the universal vector extension of \(A\).