1 Introduction
One of the central problems in geometry is decomposing varieties into basic building blocks to understand their geometric properties better. Building on foundational works [Reference MoriMor79, Reference Siu and YauSY80], structure theorems were established in the 1980s and 1990s for varieties with “semi-positive” curvature, including compact Kähler manifolds with semi-positive holomorphic bisectional curvature [Reference Howard, Smyth and WuHSW81, Reference MokMok88, Reference Mok and ZhongMZ86] and compact Kähler manifolds with nef tangent bundle [Reference Campana and PeternellCP91, Reference Demailly, Peternell and SchneiderDPS94], decomposing them into components with “positive” and “flat” curvature. More recently, these theorems have been extended to other semi-positive curvature conditions, such as nef anticanonical bundles [Reference CaoCao, Reference CaoCao19, Reference Cao and HöringCH19, Reference Druel and BiancoDB22, Reference Matsumura, Wang, Wu and ZhangMWWZ, Reference Naumann and WuNW25, Reference WuWu23, Reference ZhangZha05], pseudo-effective tangent bundles [Reference Hosono, Iwai and MatsumuraHIM22, Reference Iwai, Matsumura and ZhongIMZ, Reference MüllerMul25, Reference WuWu], and semi-positive holomorphic sectional curvature [Reference Heier and WongHW20, Reference MatsumuraMat22a, Reference YangYan18]. While significant progress has been made for projective varieties, a complete theory for compact Kähler manifolds is still lacking.
This paper focuses on the notion of semi-positive holomorphic sectional curvature and completes a comprehensive structure theorem for compact Kähler manifolds (see [Reference MatsumuraMat22a, Section 2] and references therein for the basic properties of holomorphic sectional curvature). The main result (see Theorem 1.1 below) resolves [Reference MatsumuraMat22b, Problem 3.7] and a question discussed in [Reference NiNi19], and extends the structure theorem, previously established for projective manifolds, to compact Kähler manifolds.
Theorem 1.1 (Main result).
Let X be a compact Kähler manifold admitting a Kähler metric g with semi-positive holomorphic sectional curvature. Then, the following statements hold:
-
(a) The $($
topological
$)$
fundamental group
$\pi _{1}(X)$
of X is a virtually abelian group
$($
i.e., there exists an abelian subgroup of
$\pi _{1}(X)$
of finite index
$)$
. -
(b) There exists a fibration $\phi \colon X \to Y$
(i.e., a proper surjective morphism with connected fibers) to a compact Kähler manifold Y with the following properties
$:$
-
(b1) The fibration $\phi \colon X \to Y$
is a locally constant fibration. In particular, the fibration
$\phi \colon X \to Y$
is a locally trivial fibration. -
(b2) The base Y is a finite étale quotient of a torus $($
i.e., there exists a finite étale cover
$T \to Y$
from a compact complex torus
$T)$
. -
(b3) The fiber F is a rationally connected projective manifold $($
i.e., any two points can be connected by a rational curve
$)$
. In particular, the fibration
$\phi \colon X \to Y$
is an MRC
$($
maximal rationally connected
$)$
fibration.
-
Here, a fibration
$\phi \colon X \to Y$
is said to be locally constant if it satisfies the following conditions:
-
• The fibration $\phi \colon X \to Y$
is an analytic fiber bundle with fiber F. -
• The fiber product $p_{X} \colon Y_{\mathrm {univ}} \times _Y X \to X$
is biholomorphic to the product $$ \begin{align*}Y_{\mathrm{univ}} \times_Y X \cong Y_{\mathrm{univ}} \times F,\end{align*} $$
where $p_{Y} \colon Y_{\mathrm {univ}} \to Y$
is the universal cover of Y. Furthermore, this identification is compatible with the natural maps, as in the following diagram: (1.1)
where $\mathrm {pr}_{i}$
denotes the projection. -
• There exists a representation $ \rho \colon \pi _1(Y)\to \mathrm {Aut}(F) $
of the fundamental group
$\pi _1(Y)$
into the automorphism group
$\mathrm {Aut}(F)$
such that X is isomorphic to the quotient $$ \begin{align*}X \cong (Y_{\mathrm{univ}} \times F)/\pi_1(Y) \quad \text{ over } Y. \end{align*} $$
Here, $\gamma \in \pi _1(Y)$
acts diagonally on
$Y_{\mathrm {univ}} \times F$
.
Remark 1.2. Under the same assumption as in Theorem 1.1, the compact Kähler manifold
$(X, g)$
also satisfies the following property:
-
(b4) There exist a Kähler metric $g_F$
on the fiber F and a Kähler metric
$g_Y$
on the base Y with the following properties
$:$
-
– The holomorphic sectional curvature of $g_F$
is semi-positive. -
– The curvature tensor of $g_Y$
is flat. -
– Consider the splitting of $Y_{\mathrm {univ}} \times _Y X$
of X obtained from
$\phi $
being a locally constant fibration
$:$
$$ \begin{align*}Y_{\mathrm{univ}} \times_Y X \cong \mathbb{C}^m \times F. \end{align*} $$
Then, this splitting is not only biholomorphic but also isometric with respect to the Kähler metrics $p_X^*g $
,
$p_{Y}^*g_{Y}$
, and
$g_F$
, where
$p_{Y}$
and
$p_X$
respectively denote the universal cover
$p_{Y} \colon \mathbb {C}^m \to Y$
of Y and the induced morphism
$p_X \colon Y_{\mathrm {univ}} \times _Y X \to X$
of X.
-
Note that property
$ (\mathrm {b_4}) $
is stated separately from Theorem 1.1 for convenience of citation in this paper, not due to any mathematical necessity.
We describe the contribution of this paper to the problem known as Yau’s conjecture [Reference YauYau82, Problem 72], one of the main motivations for studying holomorphic sectional curvature. Yau’s conjecture predicts that any compact Kähler manifold with positive holomorphic sectional curvature is rationally connected (equivalently, the base Y in Theorem 1.1 is a single point). This conjecture was resolved in the projective case in [Reference Heier and WongHW20] and was subsequently settled in full generality in [Reference YangYan18]. The conjecture was further extended to projective manifolds with semi-positive holomorphic sectional curvature (see [Reference MatsumuraMat22a, Corollary 1.3]) using the notion of truly flat tangent vectors.
As an application of Theorem 1.1, we extend [Reference MatsumuraMat22a, Corollary 1.3] to compact Kähler manifolds (see Corollary 1.3 below). The invariant
$n_{{\mathrm {tf}}}{(X, g)}$
in the corollary can be viewed as an analogue of the numerical Kodaira dimension (see [Reference Heier, Lu, Wong and ZhengHLWZ18] or the discussion after Corollary 1.3 for truly flat tangent vectors and the invariant
$n_{{\mathrm {tf}}}{(X, g)}$
). Therefore, the formulation of Corollary 1.3 can also be seen as an analogue of Hacon-McKernan’s question [Reference Hacon and McKernanHM07, Question 3.1], which was affirmatively resolved in [Reference Campana, Cao and MatsumuraCCM21, Reference Ejiri and GongyoEG19, Reference Ejiri, Iwai and MatsumuraEIM20]. This question remains open for compact Kähler manifolds, making Corollary 1.3 particularly interesting.
Corollary 1.3. Let
$(X, g)$
be a compact Kähler manifold equipped with a Kähler metric g with semi-positive holomorphic sectional curvature. Let
$\phi \colon X \dashrightarrow Y$
be an MRC fibration of X. Then, we have
In particular, the manifold X is rationally connected if
$n_{{\mathrm {tf}}}{(X, g)}=\dim X$
.
Recall that the invariant
$n_{{\mathrm {tf}}}{(X, g)}$
, which is an analogue of the numerical Kodaira dimension, is defined as follows. The holomorphic sectional curvature
$H_g$
is defined by
for any nonzero
$v\in T_{X,p}$
, where
$R_g$
is the curvature tensor defined by
for tangent vectors
$v, w, e, f \in T_X$
. A tangent vector
$v \in T_{X,p}$
is said to be truly flat if
Define
Then, the invariant
$n_{{\mathrm {tf}}}{(X, g)}$
is defined by
When the holomorphic sectional curvature
$H_{g}$
is quasi-positive (i.e., there exists a point
$p\in X$
such that
$H_g([v])>0$
for every nonzero tangent vector
$v\in T_{X,p}$
), Corollary 1.3 was already established in [Reference Zhang and ZhangZZ] by a different method. In this case, the condition
$n_{{\mathrm {tf}}}{(X, g)}=\dim X$
holds by definition.
We discuss the challenges in proving Theorem 1.1 compared to previous studies. Conclusion (a) directly follows from Conclusion (b), and Conclusion (b) was proved in the case where X is a projective manifold (see [Reference MatsumuraMat22a, Theorem 1.3]). In the proof of [Reference MatsumuraMat22a, Theorem 1.3], it was shown that an MRC fibration
$X \dashrightarrow Y$
can be chosen to satisfy the desired properties in Theorem 1.1. The smooth base Y of an MRC fibration is nonuniruled by [Reference Graber, Harris and StarrGHS03], and thus the canonical bundle
$K_{Y}$
is pseudo-effective by [Reference Boucksom, Demailly, Păun and PeternellBDPP13] when X (and consequently Y) is projective. The pseudo-effectivity of
$K_{Y}$
is a key point in the proof, but whether this property holds remains an open problem when X is a compact Kähler manifold, which presents a primary difficulty in proving Conclusion (b).
Conclusion (b) might be deduced from Conclusion (a). However, this approach is not straightforward, and proving Conclusion (a) directly is also challenging. Similar results have been established for compact Kähler manifolds with nef anticanonical bundles (see [Reference PăunPau97, Théorème 2], [Reference PăunPau17, Theorem 1.4], [Reference CampanaCam95, Theorem 2.2]), but adapting these methods to the context of holomorphic sectional curvature is highly nontrivial. Even for projective manifolds, the proof of Conclusion (a) relies on Conclusion (b).
We outline the proof of Theorem 1.1, keeping the aforementioned challenges in mind. Let
$(X, g)$
be a compact Kähler manifold equipped with a Kähler metric g with semi-positive holomorphic sectional curvature. The basic strategy is to consider the Albanese map
$\alpha \colon X \to Y := \mathrm {Alb}(X)$
instead of MRC fibrations. As mentioned earlier, since
$K_Y$
is a trivial line bundle (and thus pseudo-effective), the Albanese map satisfies all properties in Conclusion (b) except for property
$\mathrm {(b_{3})}$
. Therefore, our goal is to prove that the fiber F of the Albanese map
$\alpha \colon X \to Y$
is a rationally connected projective manifold.
If X is rationally connected, the augmented irregularity
$\hat {q}(X)$
is clearly zero. Conversely, if Theorem 1.1 holds, we can deduce that when the augmented irregularity
$\hat {q}(X)$
vanishes, the manifold X is a rationally connected projective manifold. On the basis of this observation, we divide the proof into two steps:
-
(1) Show that the fiber F has vanishing augmented irregularity: $\hat {q}(F) = 0$
. -
(2) Show that X with $\hat {q}(X) = 0$
is a rationally connected projective manifold.
The desired property
$\mathrm {(b_{3})}$
can be obtained by applying the result of Step (2) to the fiber F, which satisfies the conclusion
$\hat {q}(F) = 0$
of Step (1).
To achieve Step (1), we assume that X and F attain their augmented irregularities (i.e.,
$q(X) = \hat {q}(X)$
and
$q(F) = \hat {q}(F)$
) by replacing X with an appropriate finite étale cover. We then consider the relative Albanese map
$\beta \colon X \to Z := \mathrm {Alb}(X/Y)$
with the following diagram:

The fiber G of
$\gamma \colon Z \to Y$
is isomorphic to the Albanese torus
$\mathrm {Alb}(F)$
of F, and thus the canonical bundle
$K_G$
of G is trivial. Furthermore, since the canonical bundle
$K_Y$
of the base Y is pseudo-effective, the theory of variation of Bergman kernels shows that
$K_Z$
is pseudo-effective. Therefore, we conclude that Z is a finite étale quotient of a torus. If
$\dim Z> \dim Y$
, this contradicts the definition of the Albanese map of X. This shows that
$\dim Z = \dim Y$
, implying that the irregularity
$q(F)$
(and thus the augmented irregularity
$\hat {q}(F)$
) is zero.
To achieve Step (2), we focus on the notion of varieties of special type introduced by Campana. The structure theorem for projective manifolds shows that X is rationally connected if it is projective. If X is not projective, there exists a nontrivial holomorphic
$2$
-form
$\eta $
on X. Then, by [Reference Zhang and ZhangZZ, Theorem 1.4], the tangent bundle
$\mathrm {T}_{X}$
splits into
$V \oplus W$
such that the tangent vectors in V are truly flat. This result is crucial for our argument. This shows that V is a flat vector bundle, and thus, we can take the
$\mathrm {GL}$
-representation
$\rho \colon \pi _1(X) \to \mathrm {GL}(r \mathord {:} \,\mathbb {C})$
associated with the flat bundle V.
We show that
$\mathrm {Im} (\rho) $
is finite or, equivalently, that the flat vector bundle V is étale trivializable (i.e., there exists a finite étale cover
$\nu \colon X' \to X$
such that
$\nu ^* V$
is trivial). This implies that some finite étale cover of X admits a holomorphic
$1$
-form, contradicting our assumption
$\hat {q}(X) = 0$
. While it is generally difficult to directly prove that
$\pi _1(X)$
is finite, it suffices for our purposes to show that
$\mathrm {Im}(\rho) $
is finite. With this in mind, we prove that
$\mathrm {Im}(\rho) $
is virtually abelian by showing that X is a variety of special type (see Theorem 2.2). Since the abelianization of
$\pi _1(X)$
is finite due to
$\hat {q}(X) = 0$
, the universality of the abelianization guarantees that
$\mathrm {Im}(\rho) $
is finite.
We conclude with remarks on recent developments in holomorphic sectional curvature. The approach used in this paper is somewhat algebraic-geometric. It would be of interest to explore a more differential-geometric approach, such as one using holonomy groups or the de Rham decomposition, as suggested in [Reference Zhang and ZhangZZ, Theorem 1.4, Remark 1.6]. Although the structure theorem presented here is powerful, several open problems remain in the study of holomorphic sectional curvature. For instance, determining when a manifold admits (semi-)positive holomorphic sectional curvature remains challenging, even for rationally connected manifolds. For example, while the Hirzebruch surfaces are known to admit positive holomorphic sectional curvature [Reference HitchinHit75] (see [Reference Alvarez, Heier and ZhengAHZ18, Reference Chaturvedi and HeierHC20]), it is not known whether their blow-ups do (see [Reference MatsumuraMat20, Problem 3.8]). The study of holomorphic sectional curvature continues to develop in connection with other notions of curvature and positivity, though these aspects are not the focus of this paper. For further details, see [Reference Chu, Lee and ZhuCLZ25, Reference NiNi19, Reference NiNi21, Reference Ni and ZhengNZ22, Reference TangTan, Reference YangYan18b] and references therein.
Note that after we finished writing this paper, the result of [Reference Boucksom, Demailly, Păun and PeternellBDPP13] was generalized to Kähler manifolds by [Reference OuOu] (cf. [Reference Cao and PăunCP]). However, we believe that our methods, which focus on the notion of special varieties, provide a different perspective and are worth presenting.
Notation and conventions
We use the terms “invertible sheaves” and “line bundles” interchangeably and adopt the additive notation for tensor products (e.g.,
$L+M:=L\otimes M$
for line bundles L and M). Additionally, we use the terms “locally free sheaves” and “vector bundles” interchangeably. The term “fibrations” refers to a proper surjective morphism with connected fibers and the term “finite étale covers” refers to an unramified finite surjective morphism.
2 Varieties of special type
This section is devoted to the proof of Theorem 2.6. To this end, we first show that every compact Kähler manifold with semi-positive holomorphic sectional curvature belongs to the class of varieties of special type, as introduced by Campana. This result is proved by refining the argument in [Reference MatsumuraMat20, Reference MatsumuraMat22a], and we provide the relevant details here for the reader’s convenience.
In the proofs, we repeatedly use [Reference MatsumuraMat22a, Theorem 1.6], so we reformulate it here in our setting. Note that the terminology “locally constant fibrations” does not appear there, but this property is already established in the proof, specifically in [Reference MatsumuraMat22a, Lemma 2.5].
Theorem 2.1 [Reference MatsumuraMat22a, Theorem 1.6].
Let
$(X,g)$
be a compact Kähler manifold with semi-positive holomorphic sectional curvature, and let
$\phi \colon X \to Y$
be a fibration onto a compact Kähler manifold Y whose canonical bundle is pseudo-effective. Then,
$\phi \colon X \to Y$
satisfies all the properties listed in Theorem 1.1 and Remark 1.2, except for
$\mathrm {(a)}$
and
$\mathrm {(b_{3})}$
.
Furthermore, if
$\phi $
has a rationally connected fiber, then
$\phi \colon X \to Y$
satisfies all the properties, including
$\mathrm {(a)}$
and
$\mathrm {(b_{3})}$
.
Theorem 2.2. Let X be a compact Kähler manifold admitting a Kähler metric g with semi-positive holomorphic sectional curvature. Then, the manifold X is of special type in the sense of Campana.
Definition 2.3 [Reference CampanaCam04, Definitions 1.19, Proposition 1.25, Definitions 2.1, Theorem 2.22].
-
(1) A compact Kähler manifold X is said to be of special type in the sense of Campana if X does not admit any almost holomorphic dominant rational map $\phi \colon X \dashrightarrow Y$
of general type. -
(2) A dominant rational map $\phi \colon X \dashrightarrow Y$
is said to be of general type if some (equivalently, any) resolution
$\bar \phi \colon \Gamma \to Y$
of its indeterminacies satisfies $$ \begin{align*}\kappa \big(\mathcal{O}_{\Gamma}( {\bar \phi}^{*} K_{Y} )_{{\mathrm{sat}}} \big) \geq m:=\dim Y>0.\end{align*} $$
Here,
$$ \begin{align*}\mathcal{O}_{\Gamma}( {\bar \phi}^{*} K_{Y} )_{{\mathrm{sat}}} \subset \Omega_{\Gamma}^{m} \end{align*} $$
is the saturation of the invertible subsheaf defined by the pull-back of the canonical bundle $K_{Y}$
in the m-th exterior power
$\Omega _{\Gamma }^{m}$
of the cotangent bundle
$\Omega _{\Gamma }$
, where
$\kappa (\bullet )$
denotes the Kodaira dimension.
Proof. Let
$ \phi \colon X \dashrightarrow Y $
be an almost holomorphic dominant rational map, and let
$ \bar {\phi } \colon \Gamma \to Y $
be a resolution of its indeterminacy with a bimeromorphic morphism
$ \tau \colon \Gamma \to X $
, as shown in the following diagram:

Set
$ m := \dim Y> 0 $
. In [Reference MatsumuraMat22a, Theorem 3.1], we showed that if the canonical bundle
$ K_Y $
is pseudo-effective, then the invertible sheaf
defined by the reflexive hull (the double dual) is a Hermitian flat line bundle (equivalently,
$c_{1}(L)=0$
; see [Reference MatsumuraMat22a, page 767]). This is a crucial step and follows from [Reference MatsumuraMat22a, Claim 3.5], which is proved under the assumption that the holomorphic sectional curvature is semi-positive. This implies that the Kodaira dimension of
$ \mathcal {O}_\Gamma (\bar {\phi }^* K_Y) $
cannot be greater than or equal to one.
However, the definition of special type requires us to consider the saturation
of the pull-back
$\bar {\phi }^* K_Y$
in the m-th exterior power
$\Omega _\Gamma ^m$
. By essentially the same argument, if the above saturated sheaf is pseudo-effective, then the invertible sheaf
is also a Hermitian flat line bundle. Consequently, the Kodaira dimension of
$ \mathcal {O}_\Gamma (\bar {\phi }^* K_Y)_{{\mathrm {sat}}} $
cannot be greater than or equal to one. Therefore, the manifold X does not admit any almost holomorphic dominant rational map
$\phi \colon X \dashrightarrow Y$
of general type.
We briefly explain how to deduce the Hermitian flatness using the argument of [Reference MatsumuraMat22a]. In proving the above conclusion for L, we used only the properties that L is a pseudo-effective invertible subsheaf of
$ \Omega ^m_X $
, and that L coincides with the pull-back
$ \phi ^* K_Y $
on
$X_1$
, where
$ X_1 $
is the preimage of a Zariski open set
$ Y_1 \subset Y $
such that
is a smooth fibration. These properties were used in [Reference MatsumuraMat22a, Claim 3.5] via [Reference MatsumuraMat22a, (3.1)]. Although [Reference MatsumuraMat22a, (3.1)] is formulated on
$ X_0 $
(the locus where
$ \phi $
is holomorphic), it suffices for it to hold on
$ X_1 $
.
The invertible sheaf M satisfies the first property if the saturation
$ \mathcal {O}_\Gamma (\bar {\phi }^* K_Y)_{{\mathrm {sat}}} $
is pseudo-effective. Furthermore, we have
$M = \phi ^* K_Y$
on
$X_1$
, since
is a subline bundle over the smooth locus
$Y_1$
of
$\phi \colon X \to Y$
without taking saturation and
$\tau \colon \Gamma \to X$
is an isomorphism over
$Y_{1}$
. Therefore, the same argument applied to M shows that
$ M = (\tau _* ((\mathcal {O}_\Gamma (\bar {\phi }^* K_Y)_{{\mathrm {sat}}})))^{**} $
is a Hermitian flat line bundle, just like L.
We will apply the conclusion mentioned above to show that
$ X $
is of special type. By the definition of varieties of special type, we can take an almost holomorphic dominant rational map
$ \phi \colon X \dashrightarrow Y $
and a bimeromorphic morphism
$ \tau \colon \Gamma \to X $
admitting a fibration
$ \bar {\phi } \colon \Gamma \to Y $
of general type. According to the definition of fibrations of general type, the Kodaira dimension
$ \kappa (\mathcal {O}_{\Gamma }(\bar {\phi }^* K_{Y})_{{\mathrm {sat}}}) $
is equal to
$ \dim Y> 0 $
. In particular, the line bundle
$ M $
is pseudo-effective. The conclusion mentioned above shows that
$ M $
is a Hermitian flat line bundle. This contradicts the fact that
$ \kappa (\mathcal {O}_{\Gamma }(\bar {\phi }^* K_{Y})_{{\mathrm {sat}}}) $
is greater than or equal to one.
The fundamental group of a variety of special type is conjectured to be virtually abelian [Reference CampanaCam04, Conjecture 7.1]. This conjecture remains a challenging problem. Nevertheless, we can establish that the image of the
$\mathrm {GL}$
-representation of the fundamental group is virtually abelian. This weaker assertion, which is sufficient for our purpose, is proved in [Reference CampanaCam04] using the deep results in [Reference Campana, Claudon and EyssidieuxCCE15, Reference MokMok92, Reference ZuoZuo99]. Furthermore, by the following corollary, we confirm that the Albanese map
$ \phi \colon X \to Y := \mathrm {Alb}(X) $
satisfies the desired properties listed in Theorem 1.1, except for property
$\mathrm {(b_{3})}$
.
Corollary 2.4. Let
$(X, g)$
be a compact Kähler manifold admitting a Kähler metric g with semi-positive holomorphic sectional curvature. Then, the following statements hold
$:$
-
(1) Let $\rho \colon \pi _{1}(X) \to \mathrm {GL}(r \mathord {:} \,\mathbb {C})$
be a
$\mathrm {GL}$
-representation of the fundamental group
$\pi _{1}(X)$
. Then, the image
$\mathrm {Im}(\rho) $
is a virtually abelian group. -
(2) The Albanese map $\phi \colon X \to Y:=\mathrm {Alb}(X)$
satisfies the properties listed in Theorem 1.1 except for property
$\mathrm {(b_{3})}$
. Furthermore, the fiber F of
$\phi \colon X \to Y=\mathrm {Alb}(X)$
admits a Kähler metric
$g_{F}$
with semi-positive holomorphic sectional curvature.
Proof. Conclusion (1) is a direct consequence of [Reference CampanaCam04, Theorem 7.8] and Theorem 2.2.
Conclusion (2) has been implicitly proved in [Reference MatsumuraMat22a, Theorem 1.6]. For the reader’s convenience, we will specify the relevant parts of the proof. The Albanese map
$ X \to Y := \mathrm {Alb}(X) $
is a fibration by [Reference CampanaCam04, Theorem 7.4]. Thus, by [Reference MatsumuraMat22a, Theorem 1.6], there exists a holomorphic orthogonal splitting of the tangent bundle
$T_X$
:
The subbundle
$ \phi ^{*}T_Y \subset T_X $
is an integrable foliation (see [Reference MatsumuraMat22a, page 772]), indicating that
$ X \to \mathrm {Alb}(X) $
is a locally constant fibration (see, for example, [Reference MatsumuraMat22a, Lemma 2.5]). Moreover, any tangent vectors in
$ \phi ^{*}T_Y \subset T_X $
are truly flat (see [Reference MatsumuraMat22a, page 772]). Thus, by the holomorphic splitting and the Gauss-Codazzi-type formula (see [Reference MatsumuraMat22a, (2.2)]), the induced metric
$ g_{T_{X/Y}} $
on
$ T_{X/Y} $
satisfies
for any tangent vector
$ e \in T_{X/Y} \subset T_X $
. Here,
$\sqrt {-1}\Theta _h$
denotes the Chern curvature of a Hermitian vector bundle
$(E, h)$
:
and
$\langle \bullet , \bullet \rangle _h$
is the Hermitian inner product with respect to h. See [Reference MatsumuraMat22a, Section 2] for the precise notation and definition of truly flat tangent vectors.
The left-hand side equals the holomorphic sectional curvature
$H_g([e])$
evaluated at
$ e \in T_{X/Y} \subset T_X $
, which is semi-positive by assumption. This shows that the holomorphic sectional curvature of the induced metric
$ g_F := g_{T_{X/Y}}|_F $
on
$ T_F \cong T_{X/Y}|_F $
is semi-positive.
As explained in the proof of Theorem 2.2, the splitting (2.1) is derived from the existence of a pseudo-effective invertible subsheaf
$L \subset \Omega ^{m}_{X}$
with the geometric property that
$L = \phi ^{*} K_{Y}$
holds on
$X_{1}$
. In fact, as shown by an excellent result in [Reference Zhang and ZhangZZ], a splitting of the tangent bundle
$T_X$
can be obtained solely from the existence of a pseudo-effective invertible subsheaf
$L \subset \Omega ^{m}_{X}$
, without requiring the geometric property. This result extends a generalization of Yau’s conjecture to Kähler manifolds in the quasi-positive case (see [Reference Zhang and ZhangZZ, Theorem 1.5]) and plays a crucial role in this paper.
Theorem 2.5 [Reference Zhang and ZhangZZ, Theorem 1.4].
Let
$(X, g)$
be a compact Kähler manifold equipped with a Kähler metric g with semi-positive holomorphic sectional curvature. Suppose that there exists a pseudo-effective invertible subsheaf
$\mathcal {F} \subset \Omega _{X}^{m}$
. Then, the tangent bundle
$T_{X}$
admits a holomorphic orthogonal decomposition
such that V is a nonzero flat vector bundle
$($
i.e., there exists a
$\mathrm {GL}$
-representation
$\rho \colon \pi _{1}(X) \to \mathrm {GL}(r \mathord {:} \,\mathbb {C})$
of the fundamental group
$\pi _{1}(X)$
that determines
$V)$
.
Proof. By [Reference Zhang and ZhangZZ, Theorem 1.4], there exists a holomorphic orthogonal decomposition
$T_{X} \cong V \oplus W$
such that V is a nonzero vector bundle and any tangent vectors in
$V \subset T_{X}$
are truly flat. In the same way as in the proof of Corollary 2.4, due to the holomorphic splitting, the induced metric
$ g_{V} $
on
$ V $
satisfies
for any tangent vector
$ f \in V \subset T_{X} $
and any tangent vector
$v \in T_{X}$
. We can see that the left-hand side equals zero since
$f \in V \subset T_{X}$
is a truly flat tangent vector. Hence, the induced metric on V is Hermitian flat, and thus V is a flat vector bundle.
Based on the above preparations, we now proceed to prove the main theorem of this section.
Theorem 2.6. Let X be a compact Kähler manifold admitting a Kähler metric g with semi-positive holomorphic sectional curvature. Consider the augmented irregularity
where
is the irregularity of
$X'$
. Then, the following statements hold
$:$
-
(1) The inequality $\hat {q}(X) \leq \dim X$
holds. In particular, there exists a finite étale cover
$X' \to X$
such that
$q(X')=\hat {q}(X)$
holds. -
(2) If the augmented irregularity $\hat {q}(X)$
vanishes, then X is a rationally connected projective manifold.
Proof. For any finite étale cover
$\nu \colon X' \to X$
, the Kähler metric
$\nu ^{*} g$
defined by the pull-back has semi-positive holomorphic sectional curvature. Hence, the Albanese map
$\phi \colon X' \to Y' := \mathrm {Alb}(X')$
is a locally constant fibration (in particular, surjective) by Corollary 2.4. This shows that
This argument proves Conclusion (1).
We now consider Conclusion (2). Once we know that X is a projective manifold, its rational connectedness is easily verified. Indeed, since Theorem 1.1 has been proved for projective manifolds, there exists an MRC fibration
$X \to Y$
with the properties in Theorem 1.1. Then, we can easily see that
$\dim Y \leq \hat {q}(X)$
. Hence, by the assumption
$\hat {q}(X) = 0$
, we conclude that Y is a single point. This means that X itself is the fiber of the MRC fibration (i.e., it is rationally connected).
Suppose now that X is not projective. Then, there exists a nonzero holomorphic
$2$
-form
$\eta $
on X. Otherwise, the ample cone would coincide with the Kähler cone, which is nonempty, implying that X is projective (see, for example, [Reference HuybrechtsHuy05, Proposition 3.3.2 and Corollary 5.3.3]).
The strategy of the proof is to exploit the nonprojectivity of X to construct a foliation related to flatness. Similar ideas have appeared previously in the literature (see, for example, [Reference BrunellaBru06, Reference Höring and PeternellHP15, Reference Matsumura and QingMQ, Reference WuWu]). In our case, by applying Theorem 2.5, we obtain a splitting of the tangent sheaf, which yields the desired conclusion.
We consider the reflexive hull
of the sheaf
$\langle \eta \rangle _{\mathcal {O}_{X}}$
generated by
$\eta $
. This sheaf
$\mathcal {F} \subset \Omega _{X}^{2}$
is an effective invertible subsheaf. Thus, by Theorem 2.5, the tangent bundle
$T_{X}$
admits a holomorphic orthogonal decomposition
$T_{X} \cong V \oplus W$
such that V is a nonzero flat vector bundle. Let
be a
$\mathrm {GL}$
-representation that determines the flat vector bundle V, where
$r := \mathrm {rank} V> 0$
.
We will show that we may assume that
$\mathrm {Im}(\rho) $
is an abelian group. By Corollary 2.4, there exists an abelian normal subgroup
$A \subset \mathrm {Im}(\rho) $
of finite index. We consider the étale cover
$X' \to X$
corresponding to the kernel of the composition
The cover
$\nu \colon X' \to X$
is a finite étale cover since
$\mathrm {Im}(\rho) / A$
is a finite group. Note that
holds by assumption. We now consider the induced
$\mathrm {GL}$
-representation
By construction, the image
$\mathrm {Im}(\rho') $
is contained in A (in particular, it is an abelian group). Furthermore, we still have the holomorphic orthogonal splitting
and the
$\mathrm {GL}$
-representation
$\rho ' \colon \pi _{1}(X') \to \mathrm {GL}(r\mathord {:}\,\mathbb {C})$
determines the flat subbundle
Note that if
$X'$
is projective, so is X. Therefore, after replacing X with the finite étale cover
$X'$
, we may assume that the image
$\mathrm {Im}(\rho) $
is an abelian group.
The first homology group
$H_{1}(X, \mathbb {Z})$
can be regarded as the abelianization of
$\pi _{1}(X)$
. Since
$\dim H^{1}(X, \mathbb {C}) = 0$
holds by the assumption
$q(X) = 0$
and the Kählerness of X, the first homology group
$H_{1}(X, \mathbb {Z})$
is a torsion group (in particular, it is a finite group). Since
$\mathrm {Im}(\rho) $
is an abelian group, the universality of abelianizations shows that
$\rho \colon \pi _{1}(X) \to \mathrm {GL}(r\mathord {:}\,\mathbb {C})$
factors through
This implies that
$\mathrm {Im}(\rho) $
is a finite group, which means that V is étale trivializable (i.e., there exists a finite étale cover
$\nu \colon X' \to X$
such that the pull-back
$\nu ^{*}V $
is a trivial vector bundle on
$X'$
). Thus, by noting the splitting as in (2.2), we can see that the cover
$X'$
admits a nonzero
$1$
-form by
This contradicts the assumption
$\hat {q}(X) = 0$
.
3 Proof of the main result
This section is devoted to the proof of Theorem 1.1. To achieve this, we examine the Albanese map
$X \to Y := \mathrm {Alb}(X)$
in detail and prove that the fiber F of
$X \to Y = \mathrm {Alb}(X)$
has vanishing augmented irregularity
$\hat {q}(F) = 0$
to apply Theorem 2.6. We begin with an elementary lemma, which is well known to experts, but we provide a proof for the reader’s convenience.
Lemma 3.1. Let
$f \colon X \to Y$
be a smooth surjective morphism between smooth varieties and let
${X \xrightarrow {g} Z \xrightarrow {\sigma } Y}$
be its Stein factorization
$:$

Then, the variety Z is smooth, the morphism
$\sigma \colon Z \to Y$
is a finite étale cover, and the morphism
$g \colon X \to Z$
is a smooth fibration.
Proof. Note that Z is normal by Zariski’s main theorem. Let z be a smooth point of Z, and take a point
$x \in X$
such that
$z = g(x)$
. We consider the differential maps
$df$
,
$dg$
, and
$d\sigma $
defined on the tangent bundles around x and
$z = g(x)$
. Since
$f \colon X \to Y$
is a smooth morphism, the differential map
$df_{x} = d\sigma _{z} \circ dg_{x}$
is surjective. In particular, the differential map
$d\sigma _{z}$
is surjective, which shows that
$\sigma \colon Z \to Y$
is a finite étale cover on the smooth locus
$Z_{{\mathrm {reg}}}$
. The purity of the branched locus shows that
$\sigma \colon Z \to Y$
is étale by
$\mathrm {codim} Z_{{\mathrm {sing}}}\geq 2$
. Therefore, since Y is smooth, the variety Z is also smooth.
We now proceed to prove the main result of this paper.
Proof of Theorem 1.1.
We first check that it is sufficient to prove both Conclusions (a) and (b) after replacing X with a finite étale cover
$\nu \colon X' \to X$
. Indeed, this is obvious for Conclusion (a). Following the argument in [Reference Cao and HöringCH19], we check this for Conclusion (b). Suppose that there exists a finite étale cover
$\nu \colon X' \to X$
such that
$X'$
admits a locally constant fibration
$\phi ' \colon X' \to Y'$
with the properties listed in Theorem 1.1. Let
$\mathcal {F} \subset T_{X}$
be the (unique) saturated integrable subsheaf corresponding to an MRC fibration
$X \dashrightarrow Y$
of X, so that a general
$\mathcal {F}$
-leaf is a fiber of the MRC fibration
$X \dashrightarrow Y$
. For the reader’s convenience, we briefly explain that the subsheaf
$\mathcal {F} \subset T_{X}$
is independent of the choice of an MRC fibration
$X \dashrightarrow Y$
. Let
$\mathcal {F}$
and
$\mathcal {F}'$
be the foliations induced by MRC fibrations
$X \dashrightarrow Y$
and
$X \dashrightarrow Y'$
, respectively. After replacing
$Y'$
with a bimeromorphic model, we may assume that there exists a bimeromorphic morphism
$Y' \to Y$
, which yields a morphism
$\mathcal {F}' \to \mathcal {F}$
that is an isomorphism on a Zariski open subset of X. Consequently, the induced morphism
$T_X/\mathcal {F}' \to T_X/\mathcal {F}$
has a torsion kernel. Since
$T_X/\mathcal {F}'$
is torsion-free, this morphism is an isomorphism, and hence
$\mathcal {F}' \to \mathcal {F}$
is an isomorphism.
The pull-back
$\nu ^{*}\mathcal {F} \subset T_{X'}$
corresponds to the locally constant MRC fibration
$X' \to Y'$
of
$X'$
. This implies that
$\nu ^{*}\mathcal {F}$
is a locally free sheaf, and thus so is
$\mathcal {F}$
. Hence, by [Reference HöringHor07, Corollary 2.11], there exists a smooth RC (rationally connected) fibration
$\phi \colon X \to Y$
such that
$T_{X/Y} = \mathcal {F}$
. Once we know that
$K_{Y}$
is pseudo-effective, we can conclude that
$X \to Y$
is the desired fibration by Theorem 2.1.
To check that
$K_{Y}$
is pseudo-effective, we consider the Stein factorization
$X' \to Z \to Y$
of
${X' \to X \to Y}$
(see the diagram below). Note that Z is smooth and
$Z \to Y$
is a finite étale cover by Lemma 3.1.

We now check that
$g \colon X' \to Z$
is an RC fibration (i.e., a general fiber is rationally connected). Let
$z \in Z$
be a general point. The fiber
$X_{y} := \phi ^{-1}(y)$
at
$y := \sigma (z)$
is rationally connected, since
$y = \sigma (z)$
is also a general point in Y. Since
$\sigma \colon Z \to Y$
is étale, we obtain the decomposition into connected components:
where
$X^{\prime }_{w}:=g^{-1}(w)$
is the fiber of
$g \colon X' \to Z$
at w. This induces the étale cover
$\nu |_{X^{\prime }_{w}} \colon X^{\prime }_{w} \to X_{y}$
. The fiber
$X_{y}=\phi ^{-1}(y)$
is rationally connected and, in particular, simply connected. Therefore, the étale cover
$\nu |_{X^{\prime }_{w}} \colon X^{\prime }_{w} \to X_{y}$
is an isomorphism. In particular,
$X^{\prime }_{w}$
is rationally connected (i.e.,
$g \colon X' \to Z$
is an RC fibration).
By the universality of MRC fibrations (see [Reference CampanaCam92, Reference Kollár, Miyaoka and MoriKoMM92] for details), there exists a dominant rational map
$Z \dashrightarrow Y'$
. Since
$Y'$
is a finite étale quotient of a torus (in particular, it has no rational curves), this map
$Z \dashrightarrow Y'$
is actually an everywhere-defined fibration by [Reference FujikiFuj80, Proposition 1] (see also [Reference Hacon and McKernanHM07, Corollary 1.7]), and thus, it is a bimeromorphic morphism by
$\dim Y' = \dim Z$
. By comparing the canonical bundles, we can see that
$K_{Z}$
is pseudo-effective since
$Y'$
is smooth and
$K_{Y'}$
is Hermitian flat (in fact, torsion). This shows that
$K_{Y}$
is also a pseudo-effective line bundle since
$Z \to Y$
is an étale cover. Thus, the fibration
$X \to Y$
is the desired fibration by Theorem 2.1. This completes the proof of the reduction step showing that we may replace X with a finite étale cover.
We now prove all the properties of Conclusion (b) after replacing X with a suitable finite étale cover. Let
$\alpha \colon X \to Y := \mathrm {Alb}(X)$
be the Albanese map. By replacing X with a finite étale cover, we may assume that
by Theorem 2.6. Furthermore, as an additional reduction, we may assume that the fiber F of the Albanese map
$\alpha \colon X \to Y = \mathrm {Alb}(X)$
satisfies
by replacing X with a finite étale cover again. We will explain why this reduction is possible. Note that all the fibers are isomorphic to each other since
$\alpha \colon X \to Y = \mathrm {Alb}(X)$
is a locally constant fibration. The fiber F also admits a Kähler metric with semi-positive holomorphic sectional curvature by Corollary 2.4, and thus there exists a finite étale cover
$F' \to F$
such that
$q(F') = \hat {q}(F)$
by Theorem 2.6. It is sufficient for this reduction to find a finite étale cover
$F" \to F'$
(whose irregularity still attains the augmented irregularity of F) such that it is induced by some finite étale cover
$X" \to X$
of X.
By the definition of locally constant fibrations (see [Reference Matsumura and WangMW25a, Definition 2.3]), there exists a representation
where
$\gamma \in \pi _1(Y)$
acts diagonally by
We will construct the desired cover using the fundamental groups of this fibration. Since Y is a torus, we have
$\pi _2(Y)=0$
. Hence, the homotopy exact sequence gives
After fixing compatible base points in F and
$F'$
, let
$K\subset \pi _1(F)$
be a subgroup of finite index corresponding to
$F'\to F$
, and define the core H of
$K \subset \pi _1(X)$
by
Since
$\pi _1(F)$
is normal in
$\pi _1(X)$
, every
$gKg^{-1}$
is contained in
$\pi _1(F)$
and has the same index in
$\pi _1(F)$
. Moreover, a finitely generated group has only finitely many subgroups of a fixed finite index (see [Reference RoseRos78, Corollary 4.16]). Hence, we can deduce that
Since
$H\subset K \subset \pi _{1}(F)$
, we obtain the finite étale covers
This shows that
The converse inequality follows from the definition of the augmented irregularity, and hence we obtain
$q(F")=\hat q(F)$
.
We now construct a finite étale cover
$X" \to X$
of X inducing
$F"\to F$
on the fibers. The group
$\pi _1(F)/H$
is finite and satisfies
Since
$\pi _1(Y)$
is an abelian group, the commutator subgroup of
$\pi _1(X)/H$
is finite. Since
$\pi _1(X)/H$
is finitely generated, its center has finite index. Indeed, the centralizer of each element has finite index, and it is enough to intersect the centralizers of a finite set of generators. Thus, the center contains a torsion-free subgroup
of finite index in
$\pi _1(X)/H$
. Since
$\pi _1(F)/H$
is finite, we see that
Let
$G_X\subset \pi _1(X)$
be the inverse image of A under the quotient map
$\pi _1(X)\to \pi _1(X)/H$
, and let
$G_Y\subset \pi _1(Y)$
be its image. Then, we obtain the following commutative diagram of exact sequences:

All three vertical inclusions have finite index. Indeed, we have already proved that
$ [\pi _1(F):H]<\infty $
. Moreover, since
$G_X$
is the inverse image of A, we have
$ [\pi _1(X):G_X] = [\pi _1(X)/H:A] <\infty $
. Finally, since
$G_Y$
is the image of
$G_X$
under the surjection
$\pi _1(X)\to \pi _1(Y)$
, we have
$ [\pi _1(Y):G_Y] \leq [\pi _1(X):G_X] <\infty $
.
Let
be the finite étale covers corresponding to
$G_X$
and
$G_Y$
, respectively. Let
$p_F\colon {F}_{\mathrm {univ}}\to F$
be the universal covering map. The universal covering map of X factors as
Consequently, if
$g\in G_X$
maps to
$\gamma \in G_Y$
, then, by the lifting property, g acts as
where
$\widetilde {\tau }_g\in \mathrm {Aut}({F}_{\mathrm {univ}})$
satisfies
For every
$h\in H$
, the automorphism
$\widetilde {\tau }_g h\widetilde {\tau }_g^{-1}$
corresponds to
$ghg^{-1}\in H$
. Hence,
$\widetilde {\tau }_g$
normalizes H and descends to an automorphism of
$F"={F}_{\mathrm {univ}}/H$
. If two elements of
$G_X$
have the same image in
$G_Y$
, they differ by an element of H and hence induce the same automorphism of
$F"$
. Therefore, we obtain a representation
such that
where
$p\colon F"\to F$
is the covering map. Identifying
$G_Y$
with
$\pi _1(Y")$
, we obtain
$ X"\cong ({Y}_{\mathrm {univ}}\times F")/\pi _1(Y") $
, where
$\pi _1(Y")$
acts diagonally through
$\tau "$
. Consequently, the induced morphism
$ \alpha "\colon X"\longrightarrow Y" $
is a locally constant fibration with fiber
$F"$
.
After replacing X with the finite étale cover constructed above, we consider the Albanese map
$\alpha \colon X \to Y = \mathrm {Alb}(X)$
with fiber F. We will prove that the irregularity
$q(F)$
(and thus the augmented irregularity
$\hat {q}(F)$
) is zero, assuming that
To this end, we consider the relative Albanese map
associated with the Albanese map
$\alpha \colon X \to Y := \mathrm {Alb}(X)$
(see [Reference CampanaCam85, Reference FujikiFuj83] for relative Albanese maps).

The fibration
$\gamma \colon Z \to Y$
is locally trivial by the definition of relative Albanese maps, since the same holds for
$\alpha \colon X \to Y$
. For any point
$y \in Y$
, the fiber
$G_{y} := \gamma ^{-1}(y)$
is the Albanese torus
$\mathrm {Alb}(F_{y})$
, where
$F_{y} := \alpha ^{-1}(y)$
is the fiber of
$\alpha \colon X \to Y$
at y. In particular, the canonical bundle
$K_{G_{y}}$
is a trivial line bundle. Furthermore, since Y is a finite étale quotient of a torus, the canonical bundle
$K_{Y}$
is a Hermitian flat line bundle. Thus, the direct image sheaf
is a line bundle and admits the
$L^{2}$
-metric, which is a smooth Hermitian metric with semi-positive curvature by the positivity of direct images. Note that
$\gamma \colon Z \to Y$
is not necessarily projective but is a Kähler fibration. Nevertheless, we can apply the positivity of direct images by [Reference WangWan21, Theorem 2.6] (see also [Reference Berndtsson and PăunBP08, Reference CaoCao17, Reference Hacon, Popa and SchnellHPS18, Reference Păun and TakayamaPT18]). By the construction of the
$L^{2}$
-metric, the smoothness of the
$L^{2}$
-metric follows from the smoothness of
$\gamma \colon Z \to Y$
and the Hermitian metric on
$K_{Y}$
.
Moreover, since
$K_{G_{y}}$
is a trivial line bundle, the evaluation morphism
is surjective, and thus it is an isomorphism since
$\gamma ^{*}\gamma _{*}(\mathcal {O}_{Z}(K_{Z})) $
is a line bundle. In particular, the canonical bundle
$K_{Z}$
is pseudo-effective. Thus, by Theorem 2.1, there exists a finite étale cover
$T \to Z$
from a torus.
Consider the fiber product
$X' := X \times _Z T$
with the following diagram:

Note that
$X' \to X$
is a finite étale cover by construction. If
$\dim Z> \dim Y$
, then we obtain
which is a contradiction. Hence, we have
$\dim Y = \dim Z$
, and thus
$\gamma \colon Z \to Y$
is bimeromorphic. The bimeromorphic morphism
$\gamma \colon Z \to Y$
is an isomorphism since both
$K_{Y}$
and
$K_{Z}$
are Hermitian flat line bundles.
Consequently, we can obtain
from the definition of relative Albanese maps and the choice of F.
Theorem 2.6 shows that F is a rationally connected projective manifold since
$\hat {q}(F) = 0$
, which completes the proof of Conclusion (b). Since the rationally connected fiber F is simply connected, we can see that
by the homotopy exact sequence of the fibration
$X \to Y$
. Thus, we can see that
$\pi _{1}(X)$
is an abelian group, finishing the proof of Conclusion (a).
Finally, we prove Corollary 1.3 by applying Theorem 1.1 and Corollary 2.4.
Proof of Corollary 1.3.
By Theorem 1.1, there exists an MRC fibration
$\phi \colon X \to Y$
such that Y is a finite étale quotient of a torus. Then, by the proof of Corollary 2.4, the tangent bundle
$T_X$
splits as in (2.1), and any tangent vectors in
$\phi ^{*}T_Y \subset T_X$
are truly flat. This proves the desired inequality.
Acknowledgements
The author expresses his sincere gratitude to Professor Xi Zhang for sharing the excellent preprint [Reference Zhang and ZhangZZ], and to Dr. Shiyu Zhang for giving a talk on [Reference Zhang and ZhangZZ] at the 30th International Conference on Finite or Infinite Dimensional Complex Analysis and Applications, which inspired the author to revisit the topic of this paper. The author is also grateful to the two anonymous referees for pointing out an inaccuracy in the proof of Lemma 3.1, and for helpful comments that improved the readability of the paper. The author used generative AI tools solely for grammatical correction and to improve the fluency of the English; all mathematical content was written and verified by the author.
Funding statement
The author was partially supported by Grant-in-Aid for Scientific Research (B)
$\sharp $
21H00976 from JSPS, Fostering Joint International Research (A)
$\sharp $
19KK0342 from JSPS, and JST FOREST Program,
$\sharp $
PMJFR2368 from JST.


