Nested fibre bundles in Bott-Samelson varieties (Q2142477)

Let $G$ be a complex semisimple algebraic group with a fixed Borel subgroup $G$ and maximal torus $T\subseteq B$. For any sequence of simple reflections $s=(s_1,\ldots s_n)$ in the Weyl group $W$ corresponding to $B$, let $P_{s_i}=B\cup Bs_iB$ be the minimal parabolic subgroup for $s_i$. Then, $B^n$ acts on $P_{s_1}\times\cdots\times P_{s_n}$ as \[ (g_1,g_2,\ldots,g_n)\cdot (b_1,b_2,\ldots,b_n)=(g_1b_1,b_1^{-1}g_2b_2,\ldots, b_{r-1}^{-1}g_nb_n) \] and the Bott-Samelson variety is $\text{BS}(s)=P_{s_1}\times\cdots\times P_{S_n}/B^n$. It is known that $\text{BS}(s)$ is a smooth complex variety of dimension $n$ and the torus $T$ acts on the right on $\text{BS}(s)$ via the first factor. There is the $T$- equivariant projection $\pi(s) : \text{BS}(s)\rightarrow G/B$ that takes an orbit $(g_1,\dots,g_n)B^n$ to the coset $g_1\cdots g_nB$. When working with complex Lie groups $G$, the (original) definition of Bott-Samelson varieties considers the maximal compact subgroups $C\subseteq G$, $K\subseteq T$ and $C_{s_i}\subseteq P_{s_i}$ and the corresponding (compactly defined) Bott-Samelson variety is the quotient $\text{BS}_c(s)=C_{s_1}\times\cdots\times C_{S_n}/K^n$, for the similarly defined right action of the compact torus $K^n$ on the product. There is a similarly defined projection $\pi_c(s) : \text{BS}_c(s)\rightarrow C/K$. From the Iwasawa decomposition it follows that $\text{BS}(s)$ and $\text{BS}_c(s)$ are diffeomorphic, the corresponding projections $\pi(s)$ and $\pi_c(s)$ can be identified and the fibers $\pi(s)^{-1}(wB)$ and $\pi_c(s)^{-1}(wK)$ are homeomorphic for any $w\in W$. However, since the space $\text{BS}_c(s)$ can be defined for any sequence of reflections not necessarily simple ones, the main goal of the paper under review is to consider these compactly defined Bott-Samelson varieties for certain family of sequences of reflections. These sequences of reflections, or the corresponding element $w$ in the Weyl group appear as solutions to \textit{interval equations} for points $(c_1,\ldots,c_n)K^n\in \text{BS}_c(s)$ given as $c_1\cdots c_jK=wK$, where $i,j$ are integers that satisfy $1\leq i\leq j\leq r$. These equations do not depend on the choice of representative of the orbit $(c_1,\ldots,c_n)K$, and one notices that the similar equations for $\text{BS}(s)$ do not make sense unless $i=1$ or $w$ is the neutral element. Instances of these interval equations consider a subset $I\subseteq [n]=\{1,\ldots, n\}$ and an index function $v:I\rightarrow W$ where the corresponding equations are $c_i=v_iK$ for all $i\in I$. In Theorem 6, the author shows that there is a sequence of reflections $t=(t_1,\ldots,t_n)$ such that the set of solutions of this example is \[ \big\{(c_1,\ldots,c_n)K^n\in \text{BS}_c(s): c_i=v_iK\;\;\text{for all }i\in I\big\}\simeq \text{BS}_c(t). \] Although one may consider arbitrary interval equations, the author only considers the case when the intervals in the system of interval equations do not overlap. More precisely, the author considers pairs $R\subseteq \big\{(r_1,r_2)\in [n]\times [n]: r_1\leq r_2\big\}$ such that for all two distinct pairs $(r_1,r_2),(r_1',r_2')\in R$ the intervals $[r_1,r_2]\cap [r_1',r_2']=\emptyset$ in $[n]\times [n]$. For an indexing function $v:R\rightarrow W$, writing $r=(r_1,r_2)\in R$, consider the subspace of solutions of the corresponding interval equations \[ \text{BS}_c(s,v)=\big\{(c_1,\ldots,c_n)K^n\in\text{BS}_c(s): c_{r_1}c_{r_1+1}\cdots c_{r_2}K=v_rK\;\;\text{for all }r=(r_1,r_2)\in R\big\}. \] The first main result shows $\text{BS}_c(s,v)$ is the total space of a fiber bundle, a construction similar to the canonical line bundles over Bott-Samelson varieties. Iterating this process gives a decomposition of $\text{BS}_c(s,v)$ into a twisted product of elementary factors: smaller Bott-Samelson varieties $\text{BS}_c(t)$ or fibers $\pi_c(t)^{-1}(wK)$ of the canonical projections from $\text{BS}_c(t)$ to the flag variety $C/K$ over the $K$-fixed points $wK\in C/K$. This decomposition is too general to allow to determine $\text{BS}_c(s,v)$ uniquely, however when the space $\text{BS}_c(s,v)$ has an affine paving, the second main result gives a decomposition of the $K$-equivariant cohomology of $\text{BS}_c(s,v)$ into a tensor product of the $K$-equivariant cohomologies of the elementary factors of $\text{BS}_c(s,v)$ using a spectral sequence argument since the ordinary odd cohomology modules vanish.