--- title: Slope-Stability of Reflexive Twisted Sheaves url: https://www.emergentmind.com/topics/slope-stability-of-reflexive-twisted-sheaves type: topic --- # Slope-Stability of Reflexive Twisted Sheaves A reflexive twisted sheaf combines the structure of a reflexive coherent sheaf with an additional “twisting” by a fixed vector bundle and an endomorphism, central for moduli problems in complex and algebraic geometry. Slope-stability of reflexive twisted sheaves underlies the construction and stratification of moduli spaces of Higgs bundles and the study of the Hitchin morphism, with significant consequences for the geometry of moduli, especially under field extensions and morphisms of projective varieties. Techniques regarding slope-stability also generalize to key constructions such as the Lazarsfeld-Mukai sheaves via elementary transforms, ensuring large classes of stable, reflexive, and twisted sheaves. ## 1. Reflexive Twisted Sheaves and Slope-Stability Let $X$ be a normal projective variety of dimension $n$, $H$ an ample divisor, and $F$ a fixed vector bundle on $X$. An $F$-twisted sheaf is a pair $(\V,\theta)$ with $\V$ a coherent sheaf on $X$ and $\theta: \V\to\V\otimes F$ an $\O_X$-linear map. Such a pair is reflexive if $\V$ is reflexive as a sheaf. The (Mumford) slope of a torsion-free sheaf $\E$ of rank $r$ is $$ \mu_H(\E)=\frac{c_1(\E)\cdot H^{n-1}}{r}. $$ A reflexive $F$-twisted sheaf $(\V,\theta)$ is slope-stable (resp. semi-stable) if for every nonzero proper $F$-invariant reflexive subsheaf $0\subsetneq\W\subsetneq\V$, $$ \mu_H(\W)<\mu_H(\V)\qquad(\text{resp. }\le). $$ This definition extends the usual notion of stability for sheaves to the twisted, reflexive context. Such sheaves naturally arise in the classification of Higgs bundles and play a central role in the structure and geometry of moduli spaces [2601.08597]. ## 2. Behavior of Slope-Stability Under Covers The preservation of slope-stability under pullback is subtle. Let $\pi: Y\to X$ be a finite separable morphism of normal projective varieties. Slope-stability of a twisted bundle or reflexive twisted sheaf is preserved under pullback if and only if $\pi$ does not factor through a nontrivial quasi-étale cover—equivalently, if it is genuinely ramified in codimension $1$, i.e., $Y\times_XY$ is connected in codimension $1$ and $\pi_1^\et(Y_{\mathrm{sm}})\to \pi_1^\et(X_{\mathrm{sm}})$ is surjective. Under these conditions, - $(\V,\theta)$ is slope-stable if and only if $(\pi^*\V, \pi^*\theta)$ is slope-stable, - a reflexive $F$-twisted sheaf $(\V,\theta)$ is slope-stable if and only if its reflexive pullback $(\pi^{[*]}\V,\pi^{[*]}\theta)$ is slope-stable as an $F_{|Y}$-twisted reflexive sheaf. Hom-vanishing results show that, for genuinely ramified Galois covers, the homomorphism spaces between semistable twisted sheaves of the same slope are preserved, endomorphisms remain one-dimensional, and thus polystable pullbacks of stable sheaves remain stable [2601.08597]. ## 3. Detection of Stability via a Canonical Quasi-Étale Cover Slope-stability of a reflexive $F$-twisted sheaf of rank $r\ge2$ on $X$ with respect to all finite Galois covers is determined by its behavior on a single canonical quasi-étale Galois cover $\pi_r: X_r\to X$ of degree prime to $\chr(k)$. This "good cover" has the property that a reflexive $F$-twisted sheaf is stable on all such covers if and only if it is stable after pullback to $X_r$. The construction of $X_r$ proceeds by exploiting the fact that the prime-to-$p$ portion of the étale fundamental group of the smooth locus is topologically finitely generated, so only finitely many relevant covers of degree $\le r$ exist. A Galois cover dominating them all provides the desired $X_r$. A Jordan-type theorem (prime to $p$, in the sense of Brauer–Feit) bounds the size of decomposition groups and ensures that destabilizing factorizations occur already over covers of bounded degree. The universal detection property of the good cover underpins the stratification of the moduli of stable twisted sheaves [2601.08597]. ## 4. Moduli Stratification and the Hitchin Morphism Given a Hilbert polynomial $P$ of rank $r$, the moduli space $M_X^{P,F\text{-}s}$ of slope-stable $F$-twisted Higgs bundles decomposes into strata indexed by the decomposition type: \[ M_X^{P,F\text{-}s} = \bigsqcup_{m\mid r} Z^{P,F\text{-}s}(m) \] where $Z(m)$ is the locus of bundles whose decomposition type (the common rank $m$ of direct summands of the pullback) divides $r$. The decomposition type is stable under passage to dominating covers factoring through the good cover. The Hitchin morphism on the smallest closed stratum $Z^{P_0,F}(1)$, corresponding to completely split bundles, sends this locus onto a closed subset of the Hitchin base characterized by the property that its restriction to $X_r$ splits as a product of $r$ linear factors. In the case $X$ is hyperelliptic or abelian, with an étale Galois cover $Y\to X$ and $Y$ abelian, the image of the Dolbeault-Hitchin map is explicitly \[ (H^0(Y,\Omega^1_Y)^r\sslash S_r)^G \subset A_Y^{r,\Omega^1_Y}\cong (1)^{\oplus r} \] and for abelian $X$, exactly $H^0(X,\Omega^1_X)^r\sslash S_r$ [2601.08597]. ## 5. Construction and Stability of Reflexive Twisted Sheaves A central technique for constructing stable reflexive sheaves is the elementary transformation along divisors. For example, Lazarsfeld-Mukai reflexive sheaves are constructed as follows: if $X$ is smooth of dimension $N\ge2$ and $L$ is an ample, globally generated line bundle, then for a general smooth divisor $D\in|L|$ and general subspace $V \subset H^0(D,A)$ (with $A$ ample, globally generated on $D$), the kernel $E$ of the evaluation map $V\otimes \O_X \to F$ (for $F=i_*(A\otimes I_{Z(V)})$) gives a reflexive sheaf with rank $r=\dim V$, $\det E \cong L^\vee$, and $c_1(E) = -c_1(L)$. Slope-stability results for such sheaves include: - For $X$ with cyclic Picard group and $L=H$ ample, $E$ is $\mu_H$-stable. - For $X = \PP^N$, $d=1$, $E$ is $\mu_{\O(1)}$-stable for all $A$, $r\ge2$; further results describe (semi)stability for other choices of $A$ and $d$. - For arbitrary $X$, $E$ constructed via sufficiently general data is $\mu_L$-semistable if $2