Fubini-Griffiths-Harris rigidity and Lie algebra cohomology (Q1939221)

This article studies questions of extrinsic rigidity of certain homogeneous projective algebraic varieties, in particular of the adjoint variety of a simple Lie algebra and of the Segre varieties. Let us denote by \(X\) such a variety and by \(Y\) some smooth subvariety of the same projective space such that \(\dim(Y)=\dim(X)\). Then one looks at the Fubini forms up to some order \(k\) in a general point \(y\in Y\) and in some point \(x\in X\) up to conjugation by linear isomorphisms of the tangent and normal spaces in the two points. Then \(X\) is called rigid at order \(k\) if coincidence of these Fubini forms up to order \(k\) implies that there is a projective transformation of the ambient projective space mapping \(Y\) to \(X\). In contrast, \(X\) is called flexible at order \(k\) if the space of varieties \(Y\) having the same Fubini forms up to order \(k\) is infinite dimensional. (The intermediate concept of quasi--rigidity does not occur in any of the cases studied in the article.) The article contains rigidity results for the adjoint varieties (at order three), Veronese varieties, quadrics in the Veronese embedding, and Segre varieties. Moreover, some flexibility results are proved, which imply sharpness of the rigidity results. In particular, it is shown that the adjoint variety of \(\mathfrak{sl}(3,\mathbb C)\) is flexible at order two, with some of the varieties \(Y\) showing up even being non-flat as path geometries (so even the intrinsic geometries inherited from the Fubini forms are not locally isomorphic). The main tool used in the article are exterior differential systems (EDS). The crucial idea is to formulate the rigidity questions in such a way that in the EDS problems, one can use the information on Lie algebra cohomology groups provided by Kostant's theorem instead of Spencer cohomology groups, which are notoriously difficult to compute. In order to do this, the authors develop a general concept of filtered EDS and their prolongations, which should be interesting in its own right.