Fix an integer $k$. Let $X=G/P$ be a complex rational homogeneous variety. I assume here $G$ is a simply connected semi simple complex Lie group and $P=P_k$ is a maximal parabolic subgroup defined by dropping the $k$-th simple root (Assume we have chosen and fixed a Borel subgroup to avoid ambiguity, and we use Bourbaki convention about the order of simple roots).

I want to compute the cohomology of the homogeneous bundle $T_X\otimes L^{-\lambda_k}$ over $X$. Here, $T_X$ is the tangent bundle while $L^{-\lambda_k}$ is the line bundle corresponding to the 1-dimensional $P$-representation with character induced by $\lambda_k$. For example, if $X$ is a Grassmannian variety embedded into a projective space using Plücker embedding, then $L^{-\lambda_k}$ is just $\mathcal O(-1)$.

My strategy is to use Borel Weil Bott theorem. But $T_X$ is not irreducible in general (i.e. The $P$-representation $\mathfrak{g/p}$ is not irreducible), so is not $T_X\otimes L^{-\lambda_k}$. Hence, I have to find a $P$-representation filtration of $\mathfrak{g/p}$, say $0\subset s_1\subset s_2\subset\ldots\subset s_r=\mathfrak{g/p}$ with quotients $T_i$ irreducible $P$-representations. We shall use the same notation for the homogeneous vector bundles corresponding to $s_i, T_i$. My plan is: STEP I. compute the cohomology of $T_i\otimes L^{-\lambda_k}$ using Borel Weil Bott theorem for all $i$ since they are irreducible. STEP II. Using the filtration and step I to get the cohomology of $T_X\otimes L^{-\lambda_k}$.

STEP I is easily done by prudent computation, and STEP II can be done in most cases. But I meet some difficulties in step II for some special cases: I need to write the connection morphism down explicitly in these cases. I will use the following example to demonstrate my dilemma here.

From now on, let $G$ be the simply connected Lie group of type $B_l$ and $P=P_2$ be a maximal parabolic subgroup defined by dropping the second simple root. The the filtration of the tangent bundle of $X=G/P$ is $0\subset s_1\subset s_2=T_X$, and hence we have a short exact sequence $$ 0\to s_1\to T_X\to s_2/s_1\to 0$$ Tensoring $L^{-\lambda_2}$, we get another short exact sequence which we simply write as $$0\to s_1(-1)\to T_X(-1)\to s_2/s_1(-1)\to 0$$from which we have a long exact sequence. According to my computation using Borel Weil Bott theorem $$H^q(s_1(-1))=\mathbb C, q=1; 0, q\ne 1$$ $$H^q(s_2/s_1(-1))=\mathbb C, q=0; 0, q\ne 0$$

Hence, my long exact sequence looks like $$ 0\to H^0(T_X(-1))\to \mathbb C \stackrel{\delta}{\to} \mathbb C \to H^1(T_X)\to 0$$ Therefore, to determine what I want: $H^q(T_X(-1))$, I need to write down $\delta$ explicitly. Maybe there are other methods that can determine the cohomology of $T_X(-1)$ directly without $\delta$, so any idea is welcome and appreciated!


Let me show you a more geometric way to compute the desired cohomology for types $BCD$. I will illustrate it for $X=IGr(k, V)$, where $\dim V=2n$ and $3\leq k\leq n$.

Let $Y$ denote the Grassmannian $Gr(k,V)$, let $i:X\to Y$ be the natural closed embedding. Under $i$ the variety $X$ gets identified with the zero subscheme of a regular section $\omega\in H^0(Y, \Lambda^2U^*)\simeq \Lambda^2V^*$, where $U$ will denote the tautological both on $X$ and $Y$. In particular, the normal bundle $N_{X/Y}$ is isomorphic to $\Lambda^2U^*$.

Now, consider the short exact sequence $$ 0\to T_X(-1)\to i^*T_Y(-1)\to N_{X/Y}(-1)\to 0. $$ BBW on $X$ tells you that $H^\bullet(X, N_{X/Y}(-1))=0$. Thus, $$ H^\bullet(X, T_X(-1))\simeq H^\bullet(X, i^*T_Y(-1)) \simeq H^\bullet(Y, T_Y(-1)\otimes i_*\mathcal{O}) $$ I clame that the latter cohomology groups vanish.

Let us recall that $T_Y\simeq U^*\otimes V/U$, and use the Koszul resolution $$ 0\to \det \Lambda^2U\to\cdots\to\Lambda^2\Lambda^2U\to\Lambda^2U\to\mathcal{O} \to i_*\mathcal{O}\to 0. $$ Once you replace $i_*\mathcal{O}$ with its resolution, you get a spectral sequence with terms of the form $$ H^i(Y, U^*\otimes V/U(-1)\otimes \Lambda^t\Lambda^2U) \simeq \mathrm{Ext}^i(U^\perp, \Lambda^{k-1}U\otimes\Lambda^t\Lambda^2U).$$ It remains to show that for any irreducible summand $\Sigma^\alpha U\subset \Lambda^{k-1}U\otimes \Lambda^t\Lambda^2U$ one has $$ \mathrm{Ext}^\bullet(U^\perp, \Sigma^\alpha U)=0. $$ Here is the lazy way to show the latter: it is quite easy to see that every summand appearing in the decomposition of $\Lambda^t\Lambda^2 U$ is of highest weight $\gamma$ with $\gamma_1\leq k-1$ (actually, the decomposition is well known). Thus, from Pieri's formulas we get that $\alpha_1\leq k\leq 2n-k$. If one thinks of $\alpha$ as of a Young diagram, it is inscribed in the standard $k\times (2n-k)$ rectangle and has $|\alpha|=2t+k-1>1$ boxes.

Now, let $\beta$ be a Young diagram inscribed in the $(2n-k)\times k$ rectangle. It is a good exercise to check the following: $$ \mathrm{Ext}^i(\Sigma^\beta U^\perp, \Sigma^\alpha U)= \begin{cases} \mathsf{k}, & \text{if } \alpha=\beta^T \text{ and } i=|\alpha|,\\ 0 & \text{otherwise}. \end{cases} $$ (Actually, these are elements of dual exceptional collections in $D^b(Y)$ constructed by Kapranov.)

Finally, in our case $\beta$ consists of a single box. Thus, $|\alpha|>|\beta|$ and $\alpha\neq\beta^T$.

Teaser. Remark that when $k=2$, the variety $X$ is a hyperplane section of $Y$, and $N_{X/Y}(-1)\simeq \mathcal{O}$. What happens in the spectral sequence?


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.