Deformation of Bott-Samelson varieties and variations of isotropy structures (Q2301465)

Let \(G\) be a semisimple algebraic group and \(B \subset G\) be a Borel subgroup. For \(P_1,P_2,\dots,P_r\) parabolic subgroups of \(G\) containing \(B\) one can define a Bott-Samelson variety as \[ (P_1\times ^B P_2\times ^B \dots \times ^B P_r)/B. \] These varieties are also described as desingularizations of Schubert varieties in the complete flag variety \(G/B\) and can be constructed recursively as towers of \({\mathbb P}^1\)-bundles (for details, see Section 3 of the paper under review). In this last interpretation, Bott-Samelson varieties have been useful to characterize some varieties whose all elementary contractions are \({\mathbb P}^1\)-bundles, as complete flag varieties (see [\textit{G. Occhetta} et al., Annali della Scuola normale superiore di Pisa, Classe di scienze XVII(2): 573--607 (2017; Zbl 1390.14128)]). The idea is to use the intersection matrix of the relative anticanonical bundles of the contractions dot their different fibers and show it is the Cartan matrix of a particular semisimple Lie algebra. Then prove, via a recursive construction of Bott-Samelson varieties starting form a point, that the variety is isomorphic to the complete flag variety determined by this Lie algebra. This process needs to control the construction of the \({\mathbb P}^1\)-bundle structure at any step which has to be compatible with that of the corresponding flag bundle. For groups \(G\) whose Dynkin diagram is simply laced, this has been done in the reference above, because the corresponding \(H^1\)'s are one dimensional, so that the construction is unique; and also for those whose Dynkin diagram is not simply laced, except \(F_4\) and \(G_2\), by means of the choice of what the authors call a \textit{good word}, avoiding possible choices of \(H^1\) groups of dimension bigger than one. This paper completes the picture dealing with the cases \(F_4\) and \(G_2\), previously considered by different techniques. In this cases a good word cannot be chosen. In the paper under review the authors interpretate the excess of parameters for these two groups as the existence of a geometrical structure (orthogonal or skew-symmetric) on a partial flag for which the complete flag is defined using the notion of isotropy with respect to this structure. Moreover, they show that, also for \(F_4\) and \(G_2\), one can choose a \textit{flag-compatible} word, which satisfies that two successors can be different as \({\mathbb P}^1\)-bundles but isomorphic as varieties, unifying the characterization of complete flag varieties.