The Hitchin morphism for K K -trivial varieties Aryaman Patel and Dario Weissmann Abstract . We study the Hitchin morphism for higher dimensional varieties and show that, for a certain class of varieties which we call r r -small , the set-theoretic image of the Hitchin morphism from the Dolbeault moduli space coincides with the spectral base. In other words, a stronger version of the conjecture of Chen and Ngô holds for this class of varieties, which includes K K -trivial varieties. As part of the proof, we slightly modify the construction of spectral covers to obtain normal spectral covers. 1. Introduction We work over an algebraically closed field \mathop{{k}}\nolimits of characteristic zero. For a smooth proper scheme X over \mathop{{k}}\nolimits , the Hitchin morphism is a map from the moduli stack {M}^{r}_{X} of Higgs bundles of rank r to the Hitchin base {A}^{r}_{X}:=\bigoplus_{i=1}^{r}H^{0}(X,\mathrm{Sym}^{i}\Omega^{1}_{X}) , which sends a Higgs bundle (\mathcal{E},\theta) to the characteristic polynomial of the Higgs field \theta . We denote it by h^{r}_{X}:{M}^{r}_{X}\to{A}^{r}_{X} . When X is a curve, the Hitchin morphism is known to be surjective. Indeed, in this case \Omega^{1}_{X} is a line bundle, so the integrability condition \theta\wedge\theta=0 is automatically satisfied by any vector bundle with a \Omega^{1}_{X} -twisted endomorphism. However, for higher dimensional X , the Hitchin morphism is in general far from being surjective. A natural question then is: can we describe the image of h^{r}_{X} when \mathrm{dim}X>1 ? Recent progress has been made in this direction by Chen and Ngô in [ 2 ] , where they show that h^{r}_{X} factors through a closed subscheme {B}^{r}_{X} of {A}^{r}_{X} , called the spectral base , which is in general a non-linear subspace of smaller dimension. The morphism {M}^{r}_{X}\to{B}^{r}_{X} is called the spectral data morphism and is conjectured to be surjective. Conjecture 1.1 ( [ 2 , Conjecture 5.2] ) . For every point s\in{B}^{r}_{X} , the fiber (h^{r}_{X})^{-1}(s) is non-empty. Although the conjecture has been formulated more generally for G -Higgs bundles where G is a reductive group, we only treat the case G=GL_{r} in this article. Some recent progress has been made towards proving Conjecture 1.1 . For example, Chen and Ngô proved it for ruled and elliptic surfaces in [ 2 ] , and Song-Sun proved it more generally for all surfaces in [ 11 ] . The case of hyperelliptic varieties was treated by the authors in [ 9 ] . In [ 6 ] He and Liu proved a stronger version of the conjecture for rank 2 Higgs bundles, namely that the Hitchin morphism from the moduli stack of polystable rank 2 Higgs bundles surjects onto the spectral base {B}^{2}_{X} . In [ 7 ] He, Liu, and Mok show that {B}^{r}_{X}=\{0\} for all r when X is a smooth projective quotient of a bounded symmetric domain. In particular, Conjecture 1.1 holds for such varieties. The motivating question for the present article is: for which projective varieties X is the Hitchin morphism from the moduli space M^{P,G-ss}_{X} of Gieseker semistable Higgs bundles with fixed Hilbert polynomial P surjective onto the spectral base? A natural choice for P is the Hilbert polynomial P_{0} of the trivial bundle of rank r , in which case M^{P_{0},G-ss}_{X}=:M^{r}_{X,\mathrm{Dol}} is the Dolbeault moduli space . The following result partially answers this question. Theorem 1.2 ( Theorem 4.11 ) . Let X be a smooth projective variety whose canonical divisor K_{X} is numerically trivial. Then for all r\in\mathbb{N} the restricted Hitchin morphism h^{r}_{X,\mathrm{Dol}}:M^{r}_{X,\mathrm{Dol}}\to{B}^{r}_{X} is surjective. The surjectivity of h^{r}_{X,\mathrm{Dol}} actually holds for a larger class of varieties that satisfy a geometric condition on the vanishing loci of symmetric differentials. Theorem 1.2 is thus a special case of the following more general statement. Theorem 1.3 ( Proposition 4.8 ) . Let X be a smooth projective scheme and let r\in\mathbb{N} be fixed. Suppose that \mathrm{Sym}^{k(k-1)}\Omega^{1}_{X} does not have a non-zero global section whose vanishing locus has codimension 1 in X , for all 1\leq k\leq r . Then the restricted Hitchin morphism h^{k}_{X,\mathrm{Dol}}:M^{k}_{X,\mathrm{Dol}}\to{B}^{k}_{X} is surjective for all 1\leq k\leq r . In fact, a stronger statement holds – every fiber of h^{k}_{X,\mathrm{Dol}} contains a Higgs bundle whose underlying vector bundle is semistable and has vanishing Chern classes (see Proposition 4.8 and Remark 4.9 ). The proof relies essentially on a modification of the spectral cover construction and purity of branch locus, which is recalled in the next section. Although many varieties satisfy the hypothesis of Theorem 1.3 , the condition that all non-zero global sections of \mathrm{Sym}^{r(r-1)}\Omega^{1}_{X} have vanishing locus of codimension \geq 2 in X is hard to check. This motivates the notion of r -small varieties . Definition 1.4 . Let X be a smooth projective variety and let r\in\mathbb{N} be fixed. Then X is said to be r -small if \Omega^{1}_{X} is slope-semistable and \mathrm{Sym}^{k(k-1)}\Omega^{1}_{X} satisfies the slope inequality \mu(\mathrm{Sym}^{k(k-1)}\Omega^{1}_{X})<1 for all 1\leq k\leq r . Remark 1.5 . Note that in characteristic 0 , a vector bundle \mathcal{V} being semistable implies that the bundle \mathrm{Sym}^{j}\mathcal{V} is semistable for all j\geq 0 . A standard argument shows that any r -small variety satisfies the hypothesis of Theorem 1.3 . It follows that the Hitchin morphism h^{r}_{X,\mathrm{Dol}}:M^{r}_{X,\mathrm{Dol}}\to{B}^{r}_{X} is surjective for any r -small variety X . Since \Omega^{1}_{X} is known to be slope-semistable of slope 0 for any projective variety X with numerically trivial canonical divisor K_{X} (e.g. by [ 4 , Theorem A] ), it is immediate that any such variety is r -small. In general, the restricted Hitchin morphism h^{r}_{X,\mathrm{Dol}} from the Dolbeault moduli space is not surjective (see [ 6 , Example 3.4 (5)] ). It would then be interesting to determine necessary and sufficient conditions that X must satisfy so that h^{r}_{X,\mathrm{Dol}} is surjective. We believe that this question can be answered once we have more information on the geometric and topological properties of the spectral base. In [ 11 ] , Song-Sun prove that the spectral base is invariant under smooth birational modifications. We use this along with the standard argument of resolving the singularities of a torsion-free sheaf to reformulate 1.1 as follows. Conjecture 1.6 (equivalent to Conjecture 1.1 ) . The set-theoretic image \mathrm{im}(h^{r}_{X}) of the Hitchin morphism is invariant under smooth birational modifications. Notation We work over an algebraically closed base field \mathop{{k}}\nolimits of characteristic 0 . By variety we mean an integral separated \mathop{{k}}\nolimits -scheme of finite type. Given an affine group scheme G over \mathop{{k}}\nolimits acting on an affine scheme A , we denote the quotient stack by [A/G] which over a scheme S parametrizes G -torsors over S together with a G -invariant morphism to A . We denote by \mathrm{B}G the quotient stack [\mathop{\mathrm{Spec}}\nolimits(\mathop{{k}}\nolimits)/G] , where we equip the point with the trivial action. For a projective variety X we implicitly fix a polarization \mathcal{O}_{X}(1) and define the notion of Gieseker (semi)stability, slope-(semi)stability, Hilbert polynomial, and degree with respect to \mathcal{O}_{X}(1) . 2. Spectral covers In this section, we recall the definitions of the Hitchin base, the spectral base, and the construction of spectral covers from [ 2 , Section 6] . We also define a stratification of the spectral base in terms of multiplication of characteristic polynomials, similar to the stratification on the Chow scheme defined in [ 11 ] , and analogous to the stratification of the Hitchin base defined in [ 9 ] . The spectral covers arising from the open locus of irreducible spectral data are close to being integral, see Lemma 2.4 for the precise formulation. Further, we recall the discriminant and show that we can factor every spectral datum into a product of irreducible spectral data with non-zero discriminant, see Lemma 2.5 . 2.1. The spectral base In this subsection, we recall the definitions of the Hitchin base, the spectral base, and the construction of spectral covers from [ 2 , Section 6] . We refer the reader to loc. cit. for a complete and rigorous treatment. Let X be a smooth variety of dimension d , and let r\in\mathbb{N} be fixed. Define A^{r} as the affine space \mathbb{A}^{d}\times\mathrm{Sym}^{2}(\mathbb{A}^{d})\times\dots\times\mathrm{Sym}^{r}(\mathbb{A}^{d}) and B^{r} as the Chow scheme \mathrm{Chow}^{r}(\mathbb{A}^{d}):=(\mathbb{A}^{d})^{r}\sslash S_{r} , which classifies 0 -dimensional cycles of length r in \mathbb{A}^{d} . Here, S_{r} denotes the symmetric group in r elements and acts via permutation. A closed point \omega\in B^{r} can be represented as an unordered tuple of r closed points \omega=[\omega_{1},...,\omega_{r}] , where \omega_{i}\in\mathbb{A}^{d} for 1\leq i\leq r . We denote the i -th elementary symmetric polynomial in \omega_{1},\dots,\omega_{r} by \sigma_{i}(\omega)=\sigma_{i}(\omega_{1},\dots,\omega_{r})\in\mathrm{Sym}^{i}(\mathbb{A}^{d}) . For example, \sigma_{1}(\omega)=\omega_{1}+\dots+\omega_{r}\in\mathbb{A}^{d} , \sigma_{2}(\omega)=\sum_{i1 at x , i.e., if and only if \omega_{i}(x)=\omega_{j}(x) for some i\neq j . We observe that for any nonzero s\in{B}^{r}_{X} , the polynomial P(s) can be decomposed as a product of irreducible polynomials, such that each irreducible factor has nonzero discriminant. Lemma 2.5 . Let 0\neq s\in{B}^{r}_{X} be a \mathop{{k}}\nolimits -point. Then s can be written as a product s=\prod_{k=1}^{n}t_{k}^{m_{k}} of irreducible factors t_{k}\in{B}^{l_{k}}_{X}(\mathop{{k}}\nolimits) , m_{k}\in\mathbb{N} , each of which has nonzero discriminant. Proof. Let s=\prod_{k=1}^{n}t_{k}^{m_{k}} be a decomposition of s into irreducible factors. Then t_{k}\in(l_{k})\subset{B}^{l_{k}}_{X} for some l_{k}\in\mathbb{N} and m_{k}\in\mathbb{N} denotes the multiplicity of t_{k} in s for 1\leq k\leq n . We know from Lemma 2.4 that the spectral cover \pi_{t_{k}}:X_{t_{k}}\to X corresponding to t_{k} is generically étale of degree l_{k} . Thus, there is a dense open subset X^{\circ}_{k}\subset X over which \pi_{t_{k}} is étale. Recall that the fibers of \pi_{t_{k}} consist of roots of P(t_{k}) , which are all pairwise distinct over X^{\circ}_{k} . This means that D(t_{k}) is nonzero over X^{\circ}_{k} and is thus a nonzero section of \mathrm{Sym}^{l_{k}(l_{k}-1)}\Omega^{1}_{X} for 1\leq k\leq n . ∎ Remark 2.6 . We can draw the following conclusions from Lemma 2.5 . • If s\in{B}^{r}_{X} is an irreducible \mathop{{k}}\nolimits -point (i.e. if s\in(r) ), then D(s)\neq 0 . • In order to prove Conjecture 1.1 , it suffices to prove that for every point s\in(r)\subset{B}^{r}_{X} , the fiber (h^{r}_{X})^{-1}(s) is non-empty. 3. Normalized spectral covers In this section we modify the construction of a spectral cover of an irreducible spectral datum to obtain a normal spectral cover. We also observe that the Hitchin morphism from the stack of reflexive Higgs sheaves factors via the spectral base and that this morphism is onto. Let X be a smooth proper variety and let r\in\mathbb{N} be fixed. Let s\in{B}^{r}_{X} be a closed point. In view of Remark 2.6 , we assume that s\in(r) . Henceforth, we slightly abuse notation by using s and P(s) interchangeably. Let \widetilde{X}_{s} denote the reduced subscheme underlying the unique irreducible component of X_{s} that is mapped onto X via \pi_{s} . Note that \widetilde{X}_{s} is isomorphic to X_{s} over the dense open subset X^{\circ}\subset X where \pi_{s} is étale, and therefore \widetilde{X}_{s} is integral and generically étale over X by Lemma 2.4 . Recall from the last paragraph of Section 2 that, in general, \widetilde{X}_{s} may not be normal or Cohen-Macaulay. Let \nu:X^{\prime}_{s}\to\widetilde{X}_{s} denote the normalization map. We call X^{\prime}_{s} the normalized spectral cover corresponding to s . We have the following commutative triangle {\lx@inpgf@ignorespaces X^{\prime}_{s}} {\lx@inpgf@ignorespaces X} {\lx@inpgf@ignorespaces\widetilde{X}_{s}} \scriptstyle{\lx@inpgf@ignorespaces\pi^{\prime}_{s}} \scriptstyle{\lx@inpgf@ignorespaces\nu} \scriptstyle{\lx@inpgf@ignorespaces\pi_{s}} where \pi^{\prime}_{s}=\pi_{s}\circ\nu is finite and surjective. Let Z\subset X and Z^{\prime}\subset X denote the loci where the morphisms \pi_{s} and \pi^{\prime}_{s} are branched, respectively, and note that Z=X\setminus X^{\circ} . It follows from the construction of the spectral cover that the morphism \pi_{s} is ramified precisely over those points x\in X where P(s) has roots with multiplicity >1 at x ( [ 2 , Proposition 6.1] ). In other words, we have the following description. \displaystyle Z=\{x\in X\;|\;D(s)(x)=0\}\subset X. Next, we observe that passing from \widetilde{X}_{s} to its normalization X^{\prime}_{s} does not add any additional branching. Lemma 3.1 . With notation as above, we have Z^{\prime}\subset Z . Proof. Let U and U^{\prime} denote the loci in X where the maps \pi_{s}:\widetilde{X}_{s}\to X and \pi^{\prime}_{s}:X^{\prime}_{s}\to X respectively, are étale. Then it is clear that U\subset U^{\prime} . Since Z and Z^{\prime} are complements of U and U^{\prime} respectively, the claim follows. ∎ Since X^{\prime}_{s} is normal and X is smooth, it follows from the purity of branch locus ( [ 12 , Tag 0BMB ] ) that either Z^{\prime} has pure codimension one in X , or Z^{\prime} is empty. We want to determine sufficient conditions such that, for a general \mathop{{k}}\nolimits -point s in (r)\subset{B}^{r}_{X} , the normalized spectral cover \pi^{\prime}_{s}:X^{\prime}_{s}\to X is étale. We begin with the following construction. Since \widetilde{X}_{s} is a closed subscheme of \Omega^{1}_{X} , the vector bundle \pi_{s}^{*}\Omega^{1}_{X} on \widetilde{X}_{s} has a tautological global section \lambda\in H^{0}(\widetilde{X}_{s},\pi_{s}^{*}\Omega^{1}_{X}) . Multiplying by \lambda gives a natural \pi_{s}^{*}\Omega^{1}_{X} -twisted bundle on \widetilde{X}_{s} \displaystyle\mathcal{O}_{\widetilde{X}_{s}}\xrightarrow{\times\lambda}\mathcal{O}_{\widetilde{X}_{s}}\otimes\pi_{s}^{*}\Omega^{1}_{X}. Pushing forward along \pi_{s} , we obtain a Higgs sheaf on X by [ 10 , Lemma 6.8] \displaystyle\theta:\pi_{s*}\mathcal{O}_{\widetilde{X}_{s}}\to\pi_{s*}(\mathcal{O}_{\widetilde{X}_{s}}\otimes\pi_{s}^{*}\Omega^{1}_{X})\cong\pi_{s*}\mathcal{O}_{\widetilde{X}_{s}}\otimes\Omega^{1}_{X}. By the generalization of the BNR correspondence ( [ 1 , Proposition 3.6] ) to higher dimensional varieties ( [ 2 , Proposition 6.3] ), the characteristic polynomial of this Higgs sheaf is precisely s . We now look at the vector bundle \nu^{*}\pi_{s}^{*}\Omega^{1}_{X}={\pi^{\prime}_{s}}^{*}\Omega^{1}_{X} on X^{\prime}_{s} , and the section \nu^{*}\lambda\in H^{0}(X^{\prime}_{s},{\pi^{\prime}_{s}}^{*}\Omega^{1}_{X}) . As before, multiplying by \nu^{*}\lambda gives a natural {\pi^{\prime}_{s}}^{*}\Omega^{1}_{X} -twisted bundle on X^{\prime}_{s} \displaystyle\mathcal{O}_{X^{\prime}_{s}}\xrightarrow{\times\nu^{*}\lambda}\mathcal{O}_{X^{\prime}_{s}}\otimes{\pi^{\prime}_{s}}^{*}\Omega^{1}_{X}. Pushing forward along \pi^{\prime}_{s} , we obtain an \Omega^{1}_{X} -twisted sheaf on X \displaystyle\theta^{\prime}:\pi^{\prime}_{s*}\mathcal{O}_{X^{\prime}_{s}}\to\pi^{\prime}_{s*}(\mathcal{O}_{X^{\prime}_{s}}\otimes{\pi^{\prime}_{s}}^{*}\Omega^{1}_{X})\cong\pi^{\prime}_{s*}\mathcal{O}_{X^{\prime}_{s}}\otimes\Omega^{1}_{X}. Lemma 3.2 . The pair (\pi^{\prime}_{s*}\mathcal{O}_{X^{\prime}_{s}},\theta^{\prime}) is a Higgs sheaf on X with spectral datum s . Proof. Recall that we have assumed s\in(r) and that \widetilde{X}_{s} is reduced. The morphisms \pi_{s} and \pi^{\prime}_{s} are both étale over the dense open subset U=X\setminus Z of X , and clearly \pi_{s}^{-1}(U)\cong{\pi_{s}^{\prime}}^{-1}(U) . In particular, we have (\pi^{\prime}_{s*}\mathcal{O}_{X^{\prime}_{s}}|_{U},\theta^{\prime}|_{U})\cong(\pi_{s*}\mathcal{O}_{\widetilde{X}_{s}}|_{U},\theta|_{U}) as \Omega^{1}_{X}|_{U} -twisted bundles. Since (\pi_{s*}\mathcal{O}_{\widetilde{X}_{s}}|_{U},\theta|_{U}) is a Higgs bundle over U , we have the vanishing (\theta\wedge\theta)|_{U}=\theta|_{U}\wedge\theta|_{U}=0 , from which it follows that (\theta^{\prime}\wedge\theta^{\prime})|_{U}=0 . Since \theta^{\prime}\wedge\theta^{\prime} vanishes over the dense open subset U , we conclude that \theta^{\prime}\wedge\theta^{\prime}=0 on all of X . Let s^{\prime}=(s^{\prime}_{1},...,s^{\prime}_{r}) be the spectral datum associated with (\pi^{\prime}_{s*}\mathcal{O}_{X^{\prime}_{s}},\theta^{\prime}) . Since (\pi_{s*}\mathcal{O}_{\widetilde{X}_{s}}|_{U},\theta|_{U})\cong(\pi^{\prime}_{s*}\mathcal{O}_{X^{\prime}_{s}}|_{U},\theta^{\prime}|_{U}) as Higgs bundles on U , the characteristic polynomials of \theta|_{U} and \theta^{\prime}|_{U} agree. Therefore, we have s_{i}|_{U}=s^{\prime}_{i}|_{U} for all 1\leq i\leq r . Again, U\subset X being a dense open subset, s_{i}|_{U}=s^{\prime}_{i}|_{U} implies that s_{i}=s^{\prime}_{i} for all 1\leq i\leq r , i.e., s^{\prime}=s . ∎ Remark 3.3 . Since X^{\prime}_{s} and X are both normal and \pi^{\prime}_{s} is a finite surjective morphism, \pi^{\prime}_{s*}\mathcal{O}_{X^{\prime}_{s}} is in fact a reflexive Higgs sheaf on X ( [ 5 , Corollary 1.7] ). This recovers a result of He, Liu, and Mok (see [ 7 , Proposition 3.5] ), which says that for every \mathop{{k}}\nolimits -point s\in{B}^{r}_{X} , there is a reflexive Higgs sheaf whose associated spectral datum is precisely s . Let {M}^{r,\mathrm{refl}}_{X} denote the stack of reflexive Higgs sheaves of rank r over X which is an open substack of the stack {M}^{r,\mathrm{tf}}_{X} of torsion-free Higgs sheaves of rank r over X . We denote the Hitchin morphisms by h^{r,\mathrm{tf}}_{X}:{M}^{r,\mathrm{tf}}_{X}\to{A}^{r}_{X} and h^{r,\mathrm{refl}}_{X}:{M}^{r,\mathrm{refl}}_{X}\to{A}^{r}_{X} . Lemma 3.4 . The morphism h^{r,\mathrm{tf}}_{X} factors through the spectral base {B}^{r}_{X} . Proof. Let S be a \mathop{{k}}\nolimits -scheme of finite type and consider an S -point (\mathcal{E},\varphi)\in{M}^{r,\mathrm{tf}}_{X}(S) , that is, (\mathcal{E},\varphi) is a family of reflexive Higgs sheaves on X\times S flat over S . Fix a closed point s\in S and consider the torsion-free sheaf \mathcal{E}_{s} as the base change of \mathcal{E} along X\xrightarrow{\mathop{\mathrm{id}}\nolimits\times s}X\times S . Then \mathcal{E}_{s} is a vector bundle over a big open subset U(s)\subset X . Since being a family of vector bundles is an open condition we find an open neighborhood V(s)\subset S of s such that \mathcal{E}_{\mid U(s)\times V(s)} is a family of vector bundles flat over V(s) . Since S is quasi-compact, there exist finitely many closed points s_{1},\dots,s_{n} such that V(s_{1}),\dots,V(s_{n}) cover S . In particular, setting U:=\bigcap_{i=1}^{n}U(s_{i}) yields a big open subset of X such that \mathcal{E}_{\mid U\times S} is a family of vector bundles on U flat over S . Then h_{U}^{r}(\mathcal{E}_{\mid U\times S}) lies in {B}^{r}_{U}(S) . Since {B}^{r}_{X}\to{B}^{r}_{U} is an isomorphism by [ 11 , Lemma 5.2] we conclude. ∎ Combining Remark 3.3 and Lemmas 3.2 and 3.4 , we arrive at the following weaker version of the statement of Conjecture 1.1 Proposition 3.5 . The morphism h^{r,\mathrm{refl}}_{X} is surjective onto the spectral base {B}^{r}_{X} . Remark 3.6 . This allows us to formulate Conjecture 1.1 without using the spectral base as: the set-theoretic images of h^{r}_{X} and h^{r,\mathrm{refl}}_{X} agree. In particular, restricting to the case \mathrm{dim}(X)=2 , and observing that a reflexive sheaf on a smooth scheme is locally free outside a subset of codimension three, we see that Chen-Ngô’s conjecture is true for surfaces (see also [ 11 , Theorem 1.2] ). Alternatively, one can also use the fact that a normal surface is Cohen-Macaulay ( [ 12 , Tag 033P ] ), and hence by miracle flatness ( [ 12 , Tag 00R4 ] ) the map \pi^{\prime}_{s} is flat. This recovers the main result of Song and Sun. Corollary 3.7 ( [ 11 , Theorem 1.2] and [ 7 , Remark 2.16] ) . Let X be a smooth, proper surface. Then the spectral base {B}^{r}_{X} is equal to the set-theoretic image of h^{r}_{X} for all r . 3.1. Birational morphisms Let f:Y\to X be a birational morphism between smooth proper schemes. The natural morphism f^{*}\Omega^{1}_{X}\to\Omega^{1}_{Y} induces morphisms \mathrm{Sym}^{k}f^{*}\Omega^{1}_{X}\to\mathrm{Sym}^{k}\Omega^{1}_{Y} for all k\geq 1 . Taking global sections and composing with H^{0}(X,\mathrm{Sym}^{k}\Omega^{1}_{X})\to H^{0}(X,\mathrm{Sym}^{k}f^{*}\Omega^{1}_{X}) for 1\leq k\leq r , gives a map {A}^{r}_{X}\to{A}^{r}_{Y} between Hitchin bases, which we denote by f^{*} . Restricting to {B}^{r}_{X} , we obtain a morphism f^{*}:{B}^{r}_{X}\to{B}^{r}_{Y} between spectral bases, which is actually an isomorphism by [ 7 , Lemma 2.6] or [ 11 , Theorem 1.4] . Lemma 3.8 . Let X be a smooth proper variety of dimension d and fix r\in\mathbb{N} . Consider a \mathop{{k}}\nolimits -point s\in{B}^{r}_{X} . Then there exists a birational morphism \pi:Y\to X of smooth proper varieties such that \pi^{*}s lies in the image of h^{r}_{Y} . Proof. Via decomposing s into irreducible factors and applying Proposition 3.5 we obtain a reflexive Higgs sheaf \mathcal{F} on X with characteristic polynomial s . By [ 12 , Tag 0ESN ] we can find a birational morphism \pi^{\prime}:Y^{\prime}\to X such that the strict transform of \pi^{\prime\ast}\mathcal{F} is locally free. Since we are in characteristic 0 we can further resolve the singularities of Y^{\prime} to find a birational morphism \pi:Y\to X with smooth proper Y such that the strict transform \pi^{[\ast]}\mathcal{F} of \pi^{\ast}\mathcal{F} is locally free. We also can pullback the Higgs field \mathcal{F}\to\mathcal{F}\otimes\Omega^{1}_{X} to Y . Taking strict transform and composing with the natural morphism \pi^{\ast}\Omega^{1}_{X}\to\Omega^{1}_{Y} yields a Higgs field on the strict transform \pi^{[\ast]}\mathcal{F} on Y . Via restricting to the open locus where \pi is an isomorphism we conclude that the characteristic polynomial of \pi^{[\ast]}\mathcal{F} coincides with \pi^{\ast}s . ∎ In other words, Conjecture 1.1 is equivalent to the conjecture that the set-theoretic image of the Hitchin morphism is a birational invariant. 4. A stronger version of the conjecture In this section we prove our main theorem: the Hitchin morphism from the Dolbeault moduli space surjects onto the spectral base for K -trivial varieties, i.e., a stronger version of the Chen-Ngô conjecture holds in this case. Our approach is to study the vanishing locus of the symmetric differential arising from the discriminant. Let X be a smooth proper variety. Let r\in\mathbb{N} be fixed and let s\in{B}^{r}_{X} be a \mathop{{k}}\nolimits -point. By Lemma 3.1 , we have the inclusion Z^{\prime}\subset Z of branch loci of \pi^{\prime}_{s}:X^{\prime}_{x}\to X and \pi_{s}:X_{s}\to X (normalized) spectral cover constructed in Section 3 , and Z is given by the vanishing locus of the discriminant D(s) . If \mathrm{codim}_{X}Z=1 , then in general the (normalized) spectral cover may not be flat. However, if \mathrm{codim}_{X}Z\geq 2 , then it follows that Z^{\prime}=\emptyset and \pi_{s}^{\prime}:X^{\prime}_{s}\to X is étale by purity of branch locus. We want to determine sufficient conditions for which the latter case occurs. Consider the vector space H^{0}(X,\mathrm{Sym}^{k}\Omega^{1}_{X}) for each k\geq 1 , and define the subsets W^{i} as follows. \displaystyle W^{i}:=\{\omega\in H^{0}(X,\mathrm{Sym}^{k}\Omega^{1}_{X})\;|\;\mathrm{codim}_{X}V(\omega)\leq i\}, where V(\omega)\subset X denotes the vanishing locus of \omega . The W^{i} are closed algebraic subsets of H^{0}(X,\mathrm{Sym}^{k}\Omega^{1}_{X}) , but they may not be linear unless i=1 and k=1 ( [ 3 , Theorem 1.11] ). By definition, we have the inclusion W^{i}\subset W^{i+1} for all i . This defines a stratification of H^{0}(X,\mathrm{Sym}^{k}\Omega^{1}_{X}) given by \displaystyle W^{1}\subset W^{2}\subset\dots\subset W^{d}\subset H^{0}(X,\mathrm{Sym}^{k}\Omega^{1}_{X}). Consider a spectral datum s\in{B}^{r}_{X}(\mathop{{k}}\nolimits) such that its image D(s) via the discriminant map lies in H^{0}(X,\mathrm{Sym}^{r(r-1)}\Omega^{1}_{X})\setminus W^{1} . Then, by the purity of branch locus, the normalized spectral cover X^{\prime}_{s}\to X is étale and X^{\prime}_{s} is smooth. For a general point s\in{B}^{r}_{X} , it seems a difficult but interesting problem to determine conditions such that D(s)\in H^{0}(X,\mathrm{Sym}^{r(r-1)}\Omega^{1}_{X})\setminus W^{1} . This is related to the vanishing loci of symmetric differentials, which (to our knowledge) have not been widely studied. Lemma 4.1 . Let X be a smooth proper variety such that {B}^{r}_{X} is irreducible. Suppose there is a point s\in{B}^{r}_{X} such that D(s)\in H^{0}(X,\mathrm{Sym}^{r(r-1)}\Omega^{1}_{X})\setminus W^{1} . Then the preimage D^{-1}(\mathrm{im}(D)\setminus W^{1})\subset{B}^{r}_{X} is a dense open subset. Proof. The hypothesis implies that the image of D is not contained entirely in W^{1} , therefore \mathrm{im}(D)\cap W^{1} is a closed subset of the image. The complement of its preimage is a non-empty open, thus dense in {B}^{r}_{X} by irreducibility. ∎ Remark 4.2 . We can weaken the irreducibility condition in Lemma 4.1 to the condition that each irreducible component of {B}^{r}_{X} contains a point s such that D(s)\in H^{0}(X,\mathrm{Sym}^{r(r-1)}\Omega^{1}_{X})\setminus W^{1} to arrive at the same conclusion. A natural follow-up question is: Question 4.3 . (When) is {B}^{r}_{X} irreducible? There are several examples of varieties X such that the spectral base {B}^{r}_{X} is irreducible for all r\in\mathbb{N} . Any variety that has a ”sufficiently negative” cotangent bundle has {A}^{r}_{X}=\{0\} and thus {B}^{r}_{X}=\{0\} . Interestingly, quotients of bounded symmetric domains of sufficiently large dimension have {B}^{r}_{X}=\{0\} for all r , but the dimension of {A}^{r}_{X} is large for r sufficiently large (see [ 7 , Theorem 1.3] ). It follows from [ 2 , Proposition 8.1] that ruled surfaces and non-isotrivial elliptic surfaces with reduced fibers satisfy {B}^{r}_{X}={A}^{r}_{X} for all r\in\mathbb{N} , and thus have irreducible spectral base. The spectral base for an Abelian variety X is given by {B}^{r}_{X}\cong\mathrm{Chow}^{r}(H^{0}(X,\Omega^{1}_{X})) (see [ 2 , Example 5.1] ), and is thus irreducible. It follows from [ 9 , Theorem 1.4] that hyperelliptic varieties also have irreducible spectral base. A nice consequence of the normalized spectral cover associated to s being étale is that the vector bundle \pi^{\prime}_{s*}\mathcal{O}_{X^{\prime}_{s}} corresponds to a representation of the étale fundamental group \pi_{1}^{\text{\'{e}t}}(X) . More precisely, the following is true. Lemma 4.4 . Let f:Y\to X be a finite étale cover between normal proper varieties. Then f_{*}\mathcal{O}_{Y} is an étale trivializable vector bundle on X , that is, there exists a finite étale cover X^{\prime}\to X such that the pullback of f_{*}\mathcal{O}_{Y} is isomorphic to the trivial bundle on X^{\prime} . In particular, if X and Y are projective, then f_{*}\mathcal{O}_{Y} is a polystable vector bundle on X and all its Chern classes vanish. Proof. Let g:Z\to Y\to X be the Galois closure of f . Applying Chow’s lemma and normalization allow us to find a normal projective variety \tilde{X} together with a birational morphism \pi:\tilde{X}\to X . Set \tilde{Y}:=\tilde{X}\times_{X}Y and \tilde{Z}:=\tilde{X}\times_{X}Z . Then \tilde{g}:\tilde{Z}\to\tilde{Y}\xrightarrow{\tilde{f}}\tilde{X} are étale covers of normal projective varieties and \tilde{Z}\to\tilde{X} is Galois. Further, the morphisms \pi_{Z}:\tilde{Z}\to Z and \pi_{Y}:\tilde{Y}\to Y are birational. Then we have \pi_{Z,\ast}\mathcal{O}_{\tilde{Z}}=\mathcal{O}_{Z} and by the projection formula it suffices to show that \pi^{\ast}_{Z}g^{\ast}f_{\ast}\mathcal{O}_{Y} is the trivial bundle. Since f is affine, we have \pi^{\ast}_{Z}g^{\ast}f_{\ast}\mathcal{O}_{Y}=\tilde{g}^{\ast}\pi^{\ast}f_{\ast}\mathcal{O}_{Y}=\tilde{g}^{\ast}\tilde{f}_{\ast}\mathcal{O}_{\tilde{Y}}, see [ 12 , Lemma 02KG ] . Thus, we are reduced to showing the lemma in the case where X and Y are normal projective varieties. We assume this from now on. Note that g is an étale Galois cover satisfying Z\times_{X}Z\cong\bigsqcup_{\sigma\in\mathrm{Gal}(Z/X)}Z . Then we conclude via affine base change that g^{\ast}g_{\ast}\mathcal{O}_{Z}\cong\bigoplus_{\sigma\in\mathrm{Gal}{Z/X}}\mathcal{O}_{Z} . By adjunction we have f_{\ast}\mathcal{O}_{Y}\subseteq g_{\ast}\mathcal{O}_{Z} and conclude by flatness of g that g^{\ast}f_{\ast}\mathcal{O}_{Y}\subseteq g^{\ast}g_{\ast}\mathcal{O}_{Z} . Thus, by the previous discussion we find that g^{\ast}f_{\ast}\mathcal{O}_{Y} is a subsheaf of the trivial bundle. Fix a polarization \mathcal{O}_{X}(1) of X and let \mathcal{O}_{Y}(1) and \mathcal{O}_{Z}(1) the polarizations obtained via pullback to Y and Z respectively. We consider slope-(semi)stability with respect to these polarizations. Recall that slope-semistability is preserved under pushforward and pullback along an étale cover, by [ 8 , Lemma 3.2.1] . Recall that the degree of a bundle does not change under pushforward along an étale cover, and the degree under pullback is multiplied by the degree of the cover. Thus, g^{\ast}f_{\ast}\mathcal{O}_{Y} slope-semistable of slope 0 and contained in the trivial bundle by the above discussion. We conclude that g^{\ast}f_{\ast}\mathcal{O}_{Y} is trivial. Since we are in characteristic 0 this implies that f_{\ast}\mathcal{O}_{Y} is polystable by [ 8 , Lemma 3.2.3] . Note that pullback along g induces an injection between cohomology groups of X and Z . The naturality property of Chern classes then implies that all Chern classes of f_{\ast}\mathcal{O}_{Y} vanish. ∎ Recall that M^{P,G-ss}_{X} denotes the good moduli space of Gieseker semistable Higgs bundles of rank r on X with Hilbert polynomial P , and P_{0} denotes the Hilbert polynomial of the trivial bundle of rank r . The Dolbeault moduli space is defined as the moduli space M^{r}_{X,\mathrm{Dol}}:=M^{G-ss,P_{0}}_{X} . The \mathop{{k}}\nolimits -points of M^{r}_{X,\mathrm{Dol}} correspond to Gieseker polystable rank r Higgs bundles with vanishing Chern classes. We return to the situation where s\in{B}^{r}_{X}(\mathop{{k}}\nolimits) is a spectral datum such that its discriminant D(s) lies in H^{0}(X,\mathrm{Sym}^{r(r-1)}\Omega^{1}_{X})\setminus W^{1} . Recall from Lemma 3.2 that \pi^{\prime}_{s*}\mathcal{O}_{X^{\prime}_{s}} is equipped with a Higgs field \theta^{\prime} whose associated spectral datum is s . It follows from Proposition 4.4 that the Higgs bundle (\pi^{\prime}_{s*}\mathcal{O}_{X^{\prime}_{s}},\theta^{\prime}) corresponds to a point in the Dolbeault moduli space M^{r}_{X,\mathrm{Dol}} . Thus, one might wonder when a stronger version of Chen-Ngô’s conjecture holds. Question 4.5 . When is the restricted Hitchin morphism h^{r}_{X,\mathrm{Dol}}:M^{r}_{X,\mathrm{Dol}}\to{B}^{r}_{X} surjective? Remark 4.6 . Note that for an irreducible spectral datum, that is, a \mathop{{k}}\nolimits -point s of (r)\subset{B}^{r}_{X} , any torsion-free Higgs sheaf with spectral datum s is in fact slope-stable: it does not even have a non-trivial saturated Higgs-subsheaf since the characteristic polynomial is multiplicative in short exact sequences. However, direct sums of slope-stable sheaves need not be slope-polystable since the slope may vary and the spectral cover construction does not lend itself to controlling the slope (as far as we can see). In general, h^{r}_{X,\mathrm{Dol}} is not surjective; see [ 6 , Example 3.4 (5)] for a counterexample. However, using the properness of h^{r}_{X,\mathrm{Dol}} and assuming that {B}^{r}_{X} is irreducible, we make the following observation, which partially answers the above question. Proposition 4.7 . Let X be a smooth projective scheme and let r\in\mathbb{N} be fixed. Suppose that {B}^{r}_{X} is irreducible, and that there is a point s\in{B}^{r}_{X} such that D(s) lies in H^{0}(X,\mathrm{Sym}^{r(r-1)}\Omega^{1}_{X})\setminus W^{1} . Then the map h^{r}_{X,\mathrm{Dol}}:M^{r}_{X,\mathrm{Dol}}\to{B}^{r}_{X} is surjective. Proof. It follows from the hypothesis and from Lemma 4.1 that the subset U of spectral data s such that D(s) lands outside W^{1} is a dense open subset of {B}^{r}_{X} . For every point s\in U , the normalized spectral cover \pi^{\prime}_{s}:X^{\prime}_{s}\to X is étale, and thus the associated Higgs bundle (\pi^{\prime}_{s*}\mathcal{O}_{X^{\prime}_{s}},\theta^{\prime}) is étale trivializable by Proposition 4.4 and corresponds to a point in M^{r}_{X,\mathrm{Dol}} . In other words, U is contained in \mathrm{im}(h^{r}_{X,\mathrm{Dol}}) . By [ 10 , Theorem 6.11] , we know that h^{r}_{X,\mathrm{Dol}} is proper. In particular, \mathrm{im}(h^{r}_{X,\mathrm{Dol}}) is closed in {B}^{r}_{X} . Since U\subset\mathrm{im}(h^{r}_{X\mathrm{Dol}}) and U is dense in {B}^{r}_{X} , we conclude that \mathrm{im}(h^{r}_{X,\mathrm{Dol}}) coincides with {B}^{r}_{X} set-theoretically. ∎ Similarly as for Lemma 4.1 , one can replace the irreducibility of {B}^{r}_{X} by the condition that every irreducible component of {B}^{r}_{X} contains a point s such that D(s) is not contained in W^{1} , to arrive at the same conclusion. Although Proposition 4.7 provides a sufficient condition for the restricted Hitchin morphism M^{r}_{X,\mathrm{Dol}}\to{B}^{r}_{X} to be surjective, the irreducibility of {B}^{r}_{X} is not a reasonable condition to check. We can replace this with the condition that the subsets W^{1} be zero and still achieve the surjectivity of M^{r}_{X,\mathrm{Dol}}\to{B}^{r}_{X} . Proposition 4.8 . Let X be a smooth proper scheme and let r\in\mathbb{N} be fixed. Suppose that W^{1}=\{0\}\subset H^{0}(X,\mathrm{Sym}^{k(k-1)}\Omega^{1}_{X}) for all 1\leq k\leq r . Then for every s\in{B}^{k}_{X} the fiber (h^{k}_{X})^{-1}(s) contains an étale trivializable Higgs bundle for all 1\leq k\leq r . In particular, if X is also projective, the morphism h^{k}_{X}:M^{k}_{X,\mathrm{Dol}}\to{B}^{k}_{X} is surjective for all 1\leq k\leq r . Proof. For s=0\in{B}^{k}_{X} it is clear that the trivial bundle of rank k is contained in the fiber (h^{k}_{X})^{-1}(0) . Suppose 0\neq s\in{B}^{k}_{X} is a \mathop{{k}}\nolimits -point such that D(s)\neq 0 . Then the normalized spectral cover \pi^{\prime}_{s}:X^{\prime}_{s}\to X is étale and (\pi^{\prime}_{s*}\mathcal{O}_{X^{\prime}_{s}},\theta^{\prime}) is an étale trivializable Higgs bundle for all 1\leq k\leq r by Proposition 4.4 . If X is projective, this corresponds to a point in M^{k}_{X,\mathrm{Dol}} for all 1\leq k\leq r . Now, let 0\neq s\in{B}^{k}_{X} be a point such that D(s)=0 . Then by Lemma 2.5 the characteristic polynomial associated with s splits as a product s=\prod_{j}t_{j} , where t_{j}\in(m_{j})\subset{B}^{m_{j}}_{X} for some m_{j} such that \sum_{j}m_{j}=k and non-vanishing discriminant D(t_{j})\in H^{0}(X,\mathrm{Sym}^{m_{j}(m_{j}-1)}\Omega^{1}_{X})\setminus\{0\} . By hypothesis, the normalized spectral cover \pi^{\prime}_{t_{j}}:X^{\prime}_{t_{j}}\to X is étale and the Higgs bundle (\pi^{\prime}_{t_{j}*}\mathcal{O}_{X^{\prime}_{t_{j}}},\theta^{\prime}_{j}) is étale trivializable for each j , again by Proposition 4.4 . Then the direct sum \bigoplus_{j}(\pi^{\prime}_{t_{j}*}\mathcal{O}_{X^{\prime}_{t_{j}}},\theta^{\prime}_{j}) is an étale trivializable Higgs bundle of rank k , and thus corresponds to a point in M^{k}_{X,\mathrm{Dol}} if X is projective for all 1\leq k\leq r . ∎ Remark 4.9 . From the proof of Proposition 4.8 , we see that in the case where X is projective, we have a stronger statement. That is, for each point s\in{B}^{r}_{X} , the fiber (h^{r}_{X,\mathrm{Dol}})^{-1}(s) contains a Higgs bundle whose underlying vector bundle is étale trivializable. The hypothesis in Proposition 4.8 is still somewhat intractable. However, they are satisfied by r -small varieties. Theorem 4.10 . Let X be a r -small variety, as in Definition 1.4 and let r\in\mathbb{N} be fixed. Then h^{k}_{X,\mathrm{Dol}}:M^{k}_{X,\mathrm{Dol}}\to{B}^{k}_{X} is surjective for all 1\leq k\leq r . Proof. It suffices to show that any r -small variety satisfies the hypothesis of Proposition 4.8 . By definition, \Omega^{1}_{X} is slope-semistable, which implies that \mathrm{Sym}^{j}\Omega^{1}_{X} is slope-semistable for all j\geq 1 since we are working in characteristic 0 . We claim that W^{1}=\{0\}\subset H^{0}(X,\mathrm{Sym}^{k(k-1)}\Omega^{1}_{X}) for all 1\leq k\leq r . Indeed, suppose that there was a non-zero section \sigma\in H^{0}(X,\mathrm{Sym}^{k(k-1)}\Omega^{1}_{X}) such that \sigma vanishes along a divisor E in X . Consider the injective morphism \mathcal{O}_{X}\hookrightarrow\mathrm{Sym}^{k(k-1)}\Omega^{1}_{X} induced by \sigma . Denote the quotient by Q:=\mathop{\mathrm{Sym}}\nolimits^{k(k-1)}\Omega^{1}_{X}/\mathcal{O}_{X} . Let \mathcal{L} be the saturation of \mathcal{O}_{X} in \mathop{\mathrm{Sym}}\nolimits^{k(k-1)}\Omega^{1}_{X} . Then the torsion subsheaf T of Q fits into a short exact sequence 0\to\mathcal{O}_{X}\to\mathcal{L}\to T\to 0 . Since \sigma vanishes along E by assumption, we have that T has support \mathop{\mathrm{Supp}}\nolimits(T)=E and we conclude that \mathcal{L} has strictly positive degree, that is, \deg(\mathcal{L})\geq 1 . This contradicts the slope-semistability of \mathop{\mathrm{Sym}}\nolimits^{k(k-1)}\Omega^{1}_{X} which is assumed to be slope-semistable of slope <1 . We conclude that W^{1}=\{0\} as claimed and that h^{k}_{X,\mathrm{Dol}} is surjective for all 1\leq k\leq r by Proposition 4.8 . ∎ It is known that the cotangent bundle of a smooth projective variety with numerically trivial canonical class is slope-semistable in characteristic zero (see, e.g. [ 4 , Theorem A] ). Thus, every K_{X} -numerically trivial variety is r -small, and we arrive at the following special case. Theorem 4.11 . Let X be a smooth projective scheme with K_{X} -numerically trivial. Then h^{r}_{X,\mathrm{Dol}}:M^{r}_{X,\mathrm{Dol}}\to{B}^{r}_{X} is surjective for all r\in\mathbb{N} . Next, we make the following observation about the structure of the spectral base for varieties which satisfy the hypothesis in Proposition 4.8 . Proposition 4.12 . Let r\in\mathbb{N} be fixed and let X be as in Proposition 4.8 . Then there are finitely many étale-trivializable vector bundles \mathcal{V}_{1},\dots,\mathcal{V}_{m} of rank r on X for some m\in\mathbb{N} such that for each point s\in{B}^{r}_{X} , the fiber (h^{r}_{X})^{-1}(s) contains some (\mathcal{V}_{i},\theta_{i})\in M^{r}_{X,\mathrm{Dol}} . Proof. By induction on r , it suffices to prove the assertion for each point in the stratum (r)\subset{B}^{r}_{X} . By assumption, the discriminant of each point s\in(r) has a vanishing locus of codimension \geq 2 . The corresponding normalized spectral cover \pi^{\prime}_{s}:X^{\prime}_{s}\to X is thus étale of degree r and by Proposition 4.4 , the vector bundle \pi^{\prime}_{s*}\mathcal{O}_{X^{\prime}_{s}} underlying the Higgs bundle (\pi^{\prime}_{s*}\mathcal{O}_{X^{\prime}_{s}},\theta^{\prime}) is étale trivializable. Since there are finitely many étale covers of X of degree r up to isomorphism, there are finitely many such étale trivializable vector bundles \pi^{\prime}_{s*}\mathcal{O}_{X^{\prime}_{s}} of rank r up to isomorphism underlying the Higgs bundles (\pi^{\prime}_{s*}\mathcal{O}_{X^{\prime}_{s}},\theta^{\prime}) , and we are done. ∎ Remark 4.13 . With notation as in Proposition 4.12 , let V_{i} denote the vector space of all Higgs fields on the vector bundle \mathcal{V}_{i} , for 1\leq i\leq m . The Hitchin morphism induces maps V_{i}\to{B}^{r}_{X} , \theta\mapsto h^{r}_{X}(\mathcal{V}_{i},\theta) for each i . Proposition 4.12 then implies that we can express the spectral base as a finite union {B}^{r}_{X}=\bigcup_{i=1}^{m}h^{r}_{X}(V_{i}) . However, it is unclear whether this description would imply irreducibility of the spectral base in this case. Rather, it may provide us with an example of the spectral base of a K -trivial variety where the spectral base is not irreducible.