Introduction
The theory of foliations connects several areas of mathematics, including differential topology, algebraic topology, and dynamical systems. A classical example of this interaction is Tischler’s fibration theorem [39 ] , which relates three objects: (i) nowhere-vanishing closed real one-forms on smooth compact manifolds; (ii) fibrations over the circle; and (iii) smooth codimension-one foliations invariant under transverse real flows (see [35 , 21 ] ). This viewpoint is also useful in the study of certain Anosov flows [35 , 36 ] . The holomorphic setting has been considered in [36 ] .
In this work we investigate the noncompact holomorphic setting, principally under the assumption that the ambient space is Stein. Our first objective is to approximate closed holomorphic one-forms on noncompact Stein spaces by closed holomorphic one-forms with discrete periods. We therefore study closed holomorphic time forms normalized by a fixed complete vector field, especially the infinitesimal generator of a holomorphic C ∗ \mathbb C^* C ∗ -action. Our main results establish normalized approximation on normal Stein spaces under a local reflexive Poincaré hypothesis and identify the holomorphic fibrations associated with discrete periods. We also consider finite quotients, surface-specific existence and uniformization, and obstructions arising in singular and nonperiodic settings. The transverse-flow and suspension viewpoint builds on earlier work on invariant holomorphic foliations [36 , 21 ] .
Basic notions
Let M M M be a connected complex manifold, and let Ω M 1 \Omega_M^1 Ω M 1 denote its sheaf of holomorphic one-forms. For every open subset U ⊆ M U\subseteq M U ⊆ M , we have Ω M 1 ( U ) = H 0 ( U , T ∗ M ) , {\Omega_M^1(U)=H^0(U,T^*M),} Ω M 1 ( U ) = H 0 ( U , T ∗ M ) , where T ∗ M T^*M T ∗ M is the holomorphic cotangent bundle. Given a nowhere-vanishing holomorphic vector field X X X on M M M we put:
T X = { ω ∈ H 0 ( M , Ω M 1 ) : d ω = 0 , ω ( X ) = 1 } , F X = { β ∈ H 0 ( M , Ω M 1 ) : d β = 0 , β ( X ) = 0 } . \begin{equation}
\begin{aligned}
\mathcal T_X&=\{\omega\in H^0(M,\Omega_M^1):d\omega=0,\ \omega(X)=1\},\\
\mathcal F_X&=\{\beta\in H^0(M,\Omega_M^1):d\beta=0,\ \beta(X)=0\}.
\end{aligned} \tag*{(1.1)}
\end{equation} T X F X = { ω ∈ H 0 ( M , Ω M 1 ) : d ω = 0 , ω ( X ) = 1 } , = { β ∈ H 0 ( M , Ω M 1 ) : d β = 0 , β ( X ) = 0 } . (1.1)
We shall refer to T X \mathcal T_X T X as the space of closed time forms of X X X and to F X \mathcal F_X F X as the space of closed tangent forms of X X X . Whenever ω 0 ∈ T X ≠ ∅ \omega_0 \in \mathcal T_X\ne\varnothing ω 0 ∈ T X = ∅ we have the structure of an affine space T X = ω 0 + F X \mathcal T_X=\omega_0+\mathcal F_X T X = ω 0 + F X . Every ω ∈ T X \omega\in\mathcal T_X ω ∈ T X defines a nonsingular holomorphic codimension-one foliation F ω = ker ω \mathcal F_\omega=\ker\omega F ω = ker ω , invariant under the local flow of X X X ([21 ] , [36 ] ). Our approximation problem asks whether a closed time form ω \omega ω can be approximated on compact sets by ω j ∈ T X \omega_j\in\mathcal T_X ω j ∈ T X whose period images in C \mathbb C C are discrete. Throughout this work we shall adopt the following terminology. Given a holomorphic action Φ : C ∗ × M → M \Phi:\mathbb C^*\times M\to M Φ : C ∗ × M → M , it defines a complete holomorphic vector field, called infinitesimal generator, by
X ( p ) = d d t ∣ t = 0 Φ ( e t , p ) . X(p)=\left.\frac{d}{dt}\right|_{t=0}\Phi(e^t,p). X ( p ) = d t d t = 0 Φ ( e t , p ) .
Its flow is given by φ t ( p ) = Φ ( e t , p ) \varphi_t(p)=\Phi(e^t,p) φ t ( p ) = Φ ( e t , p ) , and φ 2 π i = id M \varphi_{2\pi i}=\operatorname{id}_M φ 2 π i = id M . The zeros of the infinitesimal generator are the fixed points of the action.
For a class ξ ∈ H 1 ( M , C ) \xi\in H^1(M,\mathbb C) ξ ∈ H 1 ( M , C ) , its period homomorphism is
Per ξ : H 1 ( M , Z ) ⟶ C , [ γ ] ⟼ ⟨ ξ , [ γ ] ⟩ . \operatorname{Per}_{\xi}:H_1(M,\mathbb Z)\longrightarrow\mathbb C,
\qquad [\gamma]\longmapsto\langle\xi,[\gamma]\rangle. Per ξ : H 1 ( M , Z ) ⟶ C , [ γ ] ⟼ ⟨ ξ , [ γ ]⟩ .
The discrete-period locus is defined as
D ( M ) = { ξ ∈ H 1 ( M , C ) : Per ξ ( H 1 ( M , Z ) ) is discrete in C } . \begin{equation}
\mathcal D(M)=\bigl\{\xi\in H^1(M,\mathbb C):
\operatorname{Per}_{\xi}(H_1(M,\mathbb Z))\text{ is discrete in }\mathbb C\bigr\}. \tag*{(1.2)}
\end{equation} D ( M ) = { ξ ∈ H 1 ( M , C ) : Per ξ ( H 1 ( M , Z )) is discrete in C } . (1.2)
Since every discrete additive subgroup of C \mathbb C C is free abelian of rank at most two, every period group associated with a class in D ( M ) \mathcal D(M) D ( M ) has rank at most two.
We turn to normal complex analytic spaces, using standard terminology as in [25 ] .
Definition 1.1 (Normalized reflexive holomorphic one-form). Let S S S be a normal complex analytic space, and let S r e g S_{\mathrm{reg}} S reg denote its regular locus. Write
j : S r e g ↪ S j:S_{\mathrm{reg}}\hookrightarrow S j : S reg ↪ S for the natural inclusion.
We denote by Ω S r e g 1 \Omega^1_{S_{\mathrm{reg}}} Ω S reg 1 the sheaf of holomorphic differential one-forms on the complex manifold S r e g S_{\mathrm{reg}} S reg , and define the sheaf of reflexive one-forms in S S S as
Ω S [ 1 ] : = j ∗ Ω S r e g 1 . \Omega_S^{[1]}:=j_*\Omega^1_{S_{\mathrm{reg}}}. Ω S [ 1 ] := j ∗ Ω S reg 1 . Let X X X be a holomorphic vector field on S S S . A reflexive holomorphic one-form
ω ∈ H 0 ( S , Ω S [ 1 ] ) \omega\in H^0(S,\Omega_S^{[1]}) ω ∈ H 0 ( S , Ω S [ 1 ] ) is said to be normalized with respect to X X X if
ω ( X ) = 1 on S r e g . \omega(X)=1 \quad\text{on } S_{\mathrm{reg}}. ω ( X ) = 1 on S reg . It is called a closed normalized reflexive holomorphic one-form if, in addition,
d ω = 0 on S r e g . d\omega=0 \quad\text{on } S_{\mathrm{reg}}. d ω = 0 on S reg .
Recall that, since S S S is normal, there is a canonical identification
Ω S [ 1 ] ≃ ( Ω S 1 ) ∗ ∗ , \Omega_S^{[1]}
\simeq (\Omega_S^1)^{**}, Ω S [ 1 ] ≃ ( Ω S 1 ) ∗∗ ,
where Ω S 1 \Omega_S^1 Ω S 1 denotes the sheaf of Kähler holomorphic differentials and ( ⋅ ) ∗ ∗ (\cdot)^{**} ( ⋅ ) ∗∗ denotes the double dual as an O S \mathcal O_S O S -module.
Thus, a reflexive holomorphic one-form is a holomorphic one-form on S r e g S_{\mathrm{reg}} S reg , regarded as a section of its reflexive extension to S S S . Such a form need not extend as a Kähler differential across the singular locus. When S S S is smooth, one has Ω S [ 1 ] = Ω S 1 \Omega_S^{[1]}=\Omega_S^1 Ω S [ 1 ] = Ω S 1 .
Definition 1.2 (Degree-one reflexive Poincaré property). Let S S S be a normal complex analytic space, and write j : S r e g ↪ S j:S_{\rm reg}\hookrightarrow S j : S reg ↪ S for the inclusion of its regular locus. Set
Z S 1 : = ker ( d : Ω S [ 1 ] ⟶ j ∗ Ω S r e g 2 ) . \mathcal Z_S^1:=\ker\bigl(d:\Omega_S^{[1]}\longrightarrow
j_*\Omega^2_{S_{\rm reg}}\bigr). Z S 1 := ker ( d : Ω S [ 1 ] ⟶ j ∗ Ω S reg 2 ) . The space S S S has the degree-one reflexive Poincaré property if the sequence of sheaves of abelian groups
0 ⟶ C ‾ S ⟶ O S → d Z S 1 ⟶ 0 \begin{equation}
0\longrightarrow\underline{\mathbb C}_S\longrightarrow\mathcal O_S
\xrightarrow{d}\mathcal Z_S^1\longrightarrow0 \tag*{(1.3)}
\end{equation} 0 ⟶ C S ⟶ O S d Z S 1 ⟶ 0 (1.3) is exact. Equivalently, every closed reflexive holomorphic one-form has a local holomorphic primitive, including at singular points.
The degree-one reflexive Poincaré property is not automatic for a normal Stein space. It holds for several natural classes of complex spaces. It is automatic for smooth complex manifolds by the classical holomorphic Poincaré lemma. It also holds for normal spaces with isolated complete-intersection singularities of complex dimension at least three, as well as their products with complex manifolds (Proposition 2.5 ).
Another class is provided by finite quotients of smooth complex manifolds. Indeed, closed reflexive one-forms pull back locally to closed holomorphic one-forms on the covering manifold; their local primitives can be averaged over the finite group and descended to the quotient. Thus the property holds, for instance, for the quadratic cone
{ x y = z 2 } ⊂ C 3 \{xy=z^2\}\subset\mathbb C^3 { x y = z 2 } ⊂ C 3
and, by Proposition 2.5 , for the three-dimensional ordinary double point
{ x y − z w = 0 } ⊂ C 4 . \{xy-zw=0\}\subset\mathbb C^4. { x y − z w = 0 } ⊂ C 4 .
Tischler approximation on normal Stein spaces
Our first main result extends normalized Tischler approximation to normal Stein spaces.
Theorem A (Tischler approximation on normal Stein spaces: analytic form). Let S S S be a connected normal Stein space equipped with a holomorphic C ∗ \mathbb C^* C ∗ -action with infinitesimal generator X X X . Assume that:
The first homology H 1 ( S , Z ) H_1(S,\mathbb Z) H 1 ( S , Z ) is finitely generated.
S S S has the degree-one reflexive Poincaré property.
Given a closed reflexive one-form ω \omega ω normalized by X X X , then ω \omega ω is a compact-open limit on S r e g S_{\rm reg} S reg of closed, normalized reflexive forms with discrete period groups. If b 1 ( S ) ≥ 2 b_1(S)\geq 2 b 1 ( S ) ≥ 2 , the approximating forms can be chosen so that their period groups are lattices of rank two in C \mathbb C C .
In complex dimension two, the existence of a closed normalized holomorphic one-form is automatic for fixed-point-free holomorphic C ∗ \mathbb C^* C ∗ -actions on normal Stein surfaces, as shown in Proposition 6.1 .
We can also state it in the geometric form below:
Theorem B (Tischler approximation on normal Stein spaces: geometric form). Let S S S be a connected normal Stein space with finitely generated H 1 ( S , Z ) H_1(S,\mathbb Z) H 1 ( S , Z ) , satisfying the degree-one reflexive Poincaré property. Let X X X be the infinitesimal generator of a holomorphic C ∗ \mathbb C^* C ∗ -action on S S S admitting a closed reflexive normalized holomorphic one-form ω \omega ω .
Then the holomorphic foliation defined by ω \omega ω on S r e g S_{\mathrm{reg}} S reg is approximated, in the compact-open topology, by fibrations F j : S ⟶ C / Λ j F_j:S\longrightarrow\mathbb C/\Lambda_j F j : S ⟶ C / Λ j where Λ j ⊂ C , j ∈ N \Lambda_j\subset\mathbb C, j \in \mathbb N Λ j ⊂ C , j ∈ N is a discrete subgroup.
A few remarks about the geometric version is:
∙ \bullet ∙ We have F j ∗ ( d t j ) ⟶ ω F_j^*(dt_j)\longrightarrow\omega F j ∗ ( d t j ) ⟶ ω uniformly on compact subsets of S r e g S_{\mathrm{reg}} S reg . Here d t j dt_j d t j denotes the invariant holomorphic one-form on C / Λ j \mathbb C/\Lambda_j C / Λ j induced by d z dz d z on C \mathbb C C .
∙ \bullet ∙ Each base C / Λ j \mathbb C/\Lambda_j C / Λ j is biholomorphic either to C ∗ \mathbb C^* C ∗ or to an elliptic curve.
∙ \bullet ∙ If b 1 ( S ) ≥ 2 b_1(S)\geq2 b 1 ( S ) ≥ 2 , the subgroups Λ j \Lambda_j Λ j may all be chosen to be lattices of rank two, so that every base C / Λ j \mathbb C/\Lambda_j C / Λ j is an elliptic curve.
Theorem C (Finite-quotient Tischler approximation). Let S = M / G S=M/G S = M / G be a finite quotient of a connected Stein manifold M M M with finitely generated first homology, and assume that M M M carries a holomorphic C ∗ \mathbb C^* C ∗ -action whose infinitesimal generator is nowhere zero and which commutes with G G G . Consequently S S S inherits a fixed-point-free holomorphic C ∗ \mathbb C^* C ∗ -action. Every G G G -invariant normalized closed holomorphic form on M M M admits G G G -invariant normalized approximants with discrete periods; the descended reflexive forms on S S S have discrete periods on S r e g S_{\rm reg} S reg .
Manifold case
When S S S is smooth, the reflexive Poincaré property follows from the classical holomorphic Poincaré lemma (see [40 ] ). The general approximation theorem specializes to the following smooth statement.
Corollary D (Tischler approximation: manifold case). Let M M M be a connected Stein manifold with finitely generated first homology group H 1 ( M , Z ) H_1(M,\mathbb Z) H 1 ( M , Z ) . Let X X X be the nowhere-vanishing infinitesimal generator of a holomorphic C ∗ \mathbb C^* C ∗ -action on M M M . Then the following three cases occur:
If b 1 ( M ) = 0 b_1(M)=0 b 1 ( M ) = 0 , then T X = ∅ \mathcal T_X=\varnothing T X = ∅ .
If b 1 ( M ) = 1 b_1(M)=1 b 1 ( M ) = 1 and T X ≠ ∅ \mathcal T_X\ne\varnothing T X = ∅ , every ω ∈ T X \omega\in\mathcal T_X ω ∈ T X has a nonzero discrete cyclic period group; its integral defines a holomorphic locally trivial fibration with connected fibers over C ∗ \mathbb C^* C ∗ , whose fibers are the leaves of ker ω \ker\omega ker ω .
If b 1 ( M ) ≥ 2 b_1(M)\ge2 b 1 ( M ) ≥ 2 and T X ≠ ∅ \mathcal T_X\ne\varnothing T X = ∅ , every ω ∈ T X \omega\in\mathcal T_X ω ∈ T X is a compact-open limit of closed normalized holomorphic forms with rank-two lattice period groups. Their leaves are connected fibers of holomorphic locally trivial fibrations over elliptic curves.
In each case where T X ≠ ∅ \mathcal T_X\ne\varnothing T X = ∅ , the cohomology classes of its members constitute the affine hyperplane { ξ ∈ H 1 ( M , C ) : ⟨ ξ , [ γ ] ⟩ = 2 π i } , \bigl\{\xi\in H^1(M,\mathbb C):\langle\xi,[\gamma]\rangle=2\pi i\bigr\}, { ξ ∈ H 1 ( M , C ) : ⟨ ξ , [ γ ]⟩ = 2 π i } , where [ γ ] [\gamma] [ γ ] is the common orbit-circle class.
On Stein surfaces, fixed-point-free multiplicative actions also yield an automatic existence result.
Theorem E (Uniformization theorem). Let M M M be a connected Stein surface carrying a fixed-point-free holomorphic C ∗ \mathbb C^* C ∗ -action with infinitesimal generator X X X . Then:
There is a closed holomorphic one-form ω \omega ω on M M M satisfying ω ( X ) = 1 \omega(X)=1 ω ( X ) = 1 . In particular H 1 ( M , C ) ≠ 0 H^1(M,\mathbb C)\ne0 H 1 ( M , C ) = 0 and b 1 ( M ) ≥ 1 b_1(M)\ge1 b 1 ( M ) ≥ 1 .
The universal cover M ~ \widetilde M M is biholomorphic to C 2 \mathbb C^2 C 2 or C × D \mathbb C\times\mathbb D C × D . In suitable global coordinates M ~ ≃ C × Σ \widetilde M\simeq\mathbb C\times\Sigma M ≃ C × Σ , Σ ∈ { C , D } \Sigma\in\{\mathbb C,\mathbb D\} Σ ∈ { C , D } , the lifted vector field is X ~ = ∂ / ∂ z \widetilde X=\partial/\partial z X = ∂ / ∂ z , with flow φ ~ t ( z , w ) = ( z + t , w ) \widetilde\varphi_t(z,w)=(z+t,w) φ t ( z , w ) = ( z + t , w ) .
If, moreover H 1 ( M , Z ) H_1(M,\mathbb Z) H 1 ( M , Z ) is finitely generated, then there is a holomorphic locally trivial fibration F : M → B F:M\to B F : M → B with connected Stein curve fibers transverse to X X X as follows:
B B B is a cylinder, B ≃ C ∗ B\simeq\mathbb C^* B ≃ C ∗ when b 1 ( M ) = 1 b_1(M)=1 b 1 ( M ) = 1 .
B B B is a torus, B ≃ C / Λ B\simeq\mathbb C/\Lambda B ≃ C /Λ for a rank-two lattice Λ ⊂ C \Lambda\subset\mathbb C Λ ⊂ C when b 1 ( M ) ≥ 2 b_1(M)\ge2 b 1 ( M ) ≥ 2 .
The fibers are the leaves of the kernel of a closed holomorphic one-form ω \omega ω normalized by X X X .
The universal-cover conclusion in Theorem E also holds for complete holomorphic flows carrying closed normalized time forms (Theorem 7.1 ). The approximation conclusion for multiplicative actions, however, differs from the situation for additive actions, as the counterexamples and cohomological discussion below illustrate.
The periodicity assumption in our approximation results is essential. Indeed, we construct Stein surfaces carrying free and proper holomorphic C \mathbb C C -actions which admit closed normalized time forms, but for which no such form has discrete periods. This construction is established in Theorem 11.3 .
We also consider holomorphic C ∗ \mathbb C^* C ∗ -actions with singularities. In this setting, we construct normalized logarithmic time forms associated with holomorphic semi-invariants and obtain discrete-period approximation on suitable invariant Stein open subsets (Theorem 12.2 ).
Foliations invariant under transverse flows
The hypotheses of the following theorem admit a natural interpretation in holomorphic foliation theory. Indeed, a nonsingular holomorphic foliation F \mathcal F F of codimension one on a complex manifold M M M , invariant under a complete holomorphic flow everywhere transverse to its leaves, is defined by a closed holomorphic one-form ω \omega ω normalized by the infinitesimal generator X X X of the flow, that is,
F = ker ω , d ω = 0 , ω ( X ) = 1. \mathcal F=\ker\omega,\qquad
d\omega=0,\qquad \omega(X)=1. F = ker ω , d ω = 0 , ω ( X ) = 1.
Conversely, any such pair ( ω , X ) (\omega,X) ( ω , X ) determines a nonsingular holomorphic foliation invariant under the complete transverse flow generated by X X X . This correspondence belongs to the classical framework of invariant holomorphic foliations developed in [36 ] ; see also [21 ] .
Theorem 1.3 (Closures of invariant leaves). Let M M M be a connected complex manifold equipped with a codimension one holomorphic foliation F \mathcal F F which is invariant under a transverse holomorphic flow Φ : C × M → M \Phi:\mathbb C\times M\to M Φ : C × M → M . Then we have the following possibilities for the leaves of F \mathcal F F :
All leaves are closed;
Their closures are proper real-analytic Levi-flat hypersurfaces;
All leaves are dense.
More precisely, let X X X be the infinitesimal generator of the flow, and let ω \omega ω be a closed holomorphic one-form on M M M defining F \mathcal F F and satisfying
ω ( X ) = 1. \omega(X)=1. ω ( X ) = 1.
Set H = Per ( ω ) ‾ ⊂ C H=\overline{\operatorname{Per}(\omega)}\subset\mathbb C H = Per ( ω ) ⊂ C , the closure of the period group of ω \omega ω . The three cases correspond, respectively, to H H H discrete, H H H proper and nondiscrete, and H = C H=\mathbb C H = C . Examples 10.4 and 10.5 realize the latter two cases on Stein surfaces.
Foliation-theoretic meaning. Theorem D approximates the nonsingular invariant foliation F ω = ker ω \mathcal F_\omega=\ker\omega F ω = ker ω , through foliations transverse to and invariant under the same complete flow, by holomorphic fiber-bundle foliations over C ∗ \mathbb C^* C ∗ or an elliptic curve. The approximation is uniform on compact subsets at the level of their defining one-forms and retains ω j ( X ) = 1 \omega_j(X)=1 ω j ( X ) = 1 exactly. For b 1 ( M ) ≥ 2 b_1(M)\ge2 b 1 ( M ) ≥ 2 the approximating bases can be taken to be elliptic curves. This is a holomorphic, noncompact counterpart to Tischler’s approximation of real closed nonsingular one-forms, with the additional requirement that a prescribed transverse flow remain fixed. The geometric reason that discrete periods yield closed fibers is made precise in Theorem 8.3 below.
Application to jointly-integrable holomorphic hyperbolic flows
We now turn to the relation between hyperbolic dynamics and the period group of a normalized closed holomorphic one-form, under a joint integrability hypothesis.
Theorem F (Period rigidity for jointly integrable hyperbolic flows). Let M M M be a connected complex manifold, and let X X X be a nowhere-vanishing complete holomorphic vector field with flow Φ : C × M → M \Phi:\mathbb C\times M\to M Φ : C × M → M . Assume an invariant holomorphic splitting T M = E s ⊕ C X ⊕ E u TM=E^s\oplus\mathbb CX\oplus E^u T M = E s ⊕ C X ⊕ E u with E s , E u E^s,E^u E s , E u of positive rank, with E s ⊕ E u E^s\oplus E^u E s ⊕ E u involutive, and a Hermitian metric h h h and constants C , λ > 0 C,\lambda>0 C , λ > 0 for which, for all real t ≥ 0 t\ge0 t ≥ 0 , ∥ D Φ t ( v s ) ∥ h ≤ C e − λ t ∥ v s ∥ h , ∥ D Φ − t ( v u ) ∥ h ≤ C e − λ t ∥ v u ∥ h . \|D\Phi_t(v_s)\|_h\le Ce^{-\lambda t}\|v_s\|_h,
\qquad
\|D\Phi_{-t}(v_u)\|_h\le Ce^{-\lambda t}\|v_u\|_h. ∥ D Φ t ( v s ) ∥ h ≤ C e − λ t ∥ v s ∥ h , ∥ D Φ − t ( v u ) ∥ h ≤ C e − λ t ∥ v u ∥ h . Let ω \omega ω be the unique normalized holomorphic one-form with ker ω = E s ⊕ E u \ker\omega=E^s\oplus E^u ker ω = E s ⊕ E u (necessarily closed). Suppose Φ T ( p ) = p \Phi_T(p)=p Φ T ( p ) = p for some p ∈ M p\in M p ∈ M and real T > 0 T>0 T > 0 . If the global dynamical stable and unstable manifolds of P = Φ T ∣ L p P=\Phi_T|_{L_p} P = Φ T ∣ L p on the joint leaf L p L_p L p intersect in exactly one point for every pair of base points, then Per ( ω ) = { t ∈ C : Φ t ( p ) = p } . \operatorname{Per}(\omega)=\{t\in\mathbb C:\Phi_t(p)=p\}. Per ( ω ) = { t ∈ C : Φ t ( p ) = p } . This group is nonzero and discrete, and ker ω \ker\omega ker ω defines a holomorphic fiber-bundle foliation over C / Per ( ω ) \mathbb C/\operatorname{Per}(\omega) C / Per ( ω ) . When M M M is Stein, this group is cyclic, so the base is C ∗ \mathbb C^* C ∗ .
The jointly integrable framework is developed in the body of the paper, where the dynamical product hypothesis yields the stated period rigidity.
In addition to [36 , 21 ] , related contributions to the existence of holomorphic submersions and closed holomorphic one-forms on Stein manifolds were obtained by Forstnerič [24 ] and Majcen [29 ] . Our approach builds on these ideas, with particular emphasis on the approximation of closed holomorphic one-forms normalized by a prescribed complete vector field.
We extend this approximation problem to normal Stein spaces using reflexive holomorphic differentials and investigate the holomorphic fibrations arising from discrete period groups. We also study the obstructions to normalization near singularities, drawing on the classical theory of holomorphic C ∗ \mathbb C^* C ∗ -actions developed by Orlik–Wagreich [33 ] and continued in [23 ] .
Reflexive holomorphic Tischler approximation on normal Stein spaces
We begin with the reflexive de Rham description of degree-one cohomology and then establish normalized approximation and its fibration consequence. Throughout this section S S S is a connected normal Stein complex space. We use the sheaves Ω S [ 1 ] \Omega_S^{[1]} Ω S [ 1 ] and Z S 1 \mathcal Z_S^1 Z S 1 and the degree-one reflexive Poincaré property from Definition 1.2 , in particular the exact sequence (1.3) .
Lemma 2.1 (Regular-locus representatives of loops). Let S S S be a connected normal complex space. For a base point in S r e g S_{\rm reg} S reg , the inclusion induces a surjection π 1 ( S r e g ) ⟶ π 1 ( S ) , \pi_1(S_{\rm reg})\longrightarrow\pi_1(S), π 1 ( S reg ) ⟶ π 1 ( S ) , and hence a surjection on first integral homology. In particular, the period character defined by local holomorphic primitives on S S S has the same image when computed on S r e g S_{\rm reg} S reg .
Proof. Complex spaces are locally contractible. By local conical structure and normality, each point has arbitrarily small contractible neighborhoods whose regular loci are connected and path connected. Subdivide a loop into finitely many arcs in such neighborhoods. Move the subdivision points into the regular locus in the overlaps, keeping the base point fixed, and join the resulting endpoints by paths in the regular part of each neighborhood. Contractibility shows that these replacements give a homotopic loop in S S S . This proves surjectivity on fundamental groups and on their abelianizations. The last assertion follows by restriction of the period character. Local contractibility also supplies the covering spaces used below. ◻
Proposition 2.2 (Reflexive de Rham). If a connected normal Stein space S S S satisfies (1.3) , then taking periods induces a natural isomorphism H 0 ( S , Z S 1 ) / d O ( S ) ≃ H 1 ( S , C ) . H^0(S,\mathcal Z_S^1)/d\mathcal O(S)\ \simeq\ H^1(S,\mathbb C). H 0 ( S , Z S 1 ) / d O ( S ) ≃ H 1 ( S , C ) . In particular, periods of global closed reflexive one-forms along loops of S r e g S_{\rm reg} S reg depend only on their images in H 1 ( S , Z ) H_1(S,\mathbb Z) H 1 ( S , Z ) .
Proof. The cohomology sequence of (1.3) yields
H 0 ( S , O S ) → d H 0 ( S , Z S 1 ) ⟶ H 1 ( S , C ‾ S ) ⟶ H 1 ( S , O S ) . H^0(S,\mathcal O_S)\xrightarrow{d}H^0(S,\mathcal Z_S^1)
\longrightarrow H^1(S,\underline{\mathbb C}_S)
\longrightarrow H^1(S,\mathcal O_S). H 0 ( S , O S ) d H 0 ( S , Z S 1 ) ⟶ H 1 ( S , C S ) ⟶ H 1 ( S , O S ) . The last term vanishes by Cartan’s Theorem B. The connecting map is the usual period class: local primitives differ by locally constant functions. This gives the isomorphism and the assertion about loops. ◻
The degree-one reflexive de Rham description makes it possible to approximate normalized classes without changing the prescribed vector field.
Theorem 2.3 (Reflexive Tischler approximation). Let S S S be a connected normal Stein space satisfying the degree-one reflexive Poincaré property, and assume H 1 ( S , Z ) H_1(S,\mathbb Z) H 1 ( S , Z ) is finitely generated. Let X X X generate a holomorphic C ∗ \mathbb C^* C ∗ -action, with complex-time flow φ 2 π i = i d \varphi_{2\pi i}=\mathrm{id} φ 2 π i = id . Suppose a global closed reflexive one-form ω ∈ H 0 ( S , Z S 1 ) \omega\in H^0(S,\mathcal Z_S^1) ω ∈ H 0 ( S , Z S 1 ) satisfies ω ( X ) = 1 \omega(X)=1 ω ( X ) = 1 on S r e g S_{\rm reg} S reg . Then there exist ω j ∈ H 0 ( S , Z S 1 ) \omega_j\in H^0(S,\mathcal Z_S^1) ω j ∈ H 0 ( S , Z S 1 ) such that ω j ( X ) = 1 , ω j ⟶ ω uniformly on compact subsets of S r e g , \omega_j(X)=1,\qquad
\omega_j\longrightarrow\omega\quad\text{uniformly on compact subsets of }S_{\rm reg}, ω j ( X ) = 1 , ω j ⟶ ω uniformly on compact subsets of S reg , and the period group of ω j ∣ S r e g \omega_j|_{S_{\rm reg}} ω j ∣ S reg is discrete in C \mathbb C C . When b 1 ( S ) ≥ 2 b_1(S)\ge2 b 1 ( S ) ≥ 2 , the approximants may be chosen with rank-two lattice periods. By Corollary 2.4 , they define holomorphic locally trivial fibrations of S S S with connected Stein analytic fibers, possibly singular.
Proof. Let γ \gamma γ be the homology class in H 1 ( S , Z ) H_1(S,\mathbb Z) H 1 ( S , Z ) of any circle orbit in S r e g S_{\rm reg} S reg . This locus is invariant under the action. Normalization implies ∫ γ ω = 2 π i \int_\gamma\omega=2\pi i ∫ γ ω = 2 π i , so γ \gamma γ is non-torsion. Since H 1 ( S , Z ) H_1(S,\mathbb Z) H 1 ( S , Z ) is finitely generated, the normalization condition selects the affine hyperplane
H γ = { ξ ∈ H 1 ( S , C ) : ⟨ ξ , γ ⟩ = 2 π i } . \mathcal H_\gamma=\{\xi\in H^1(S,\mathbb C):
\langle\xi,\gamma\rangle=2\pi i\}. H γ = { ξ ∈ H 1 ( S , C ) : ⟨ ξ , γ ⟩ = 2 π i } . This hyperplane contains a dense set of classes whose periods belong to a common discrete Gaussian rational lattice (which may depend on the class). Indeed, one chooses a rational basis for the free part of integral homology, solves the single rational affine constraint, and approximates the remaining real and imaginary coordinates by rationals. For b 1 ( S ) ≥ 2 b_1(S)\ge2 b 1 ( S ) ≥ 2 , one may ensure that the period image contains two real-linearly independent elements by a small choice of a nonreal free coordinate.
Choose classes ξ j → [ ω ] \xi_j\to[\omega] ξ j → [ ω ] in that rational locus. By Proposition 2.2 , represent a complex basis of H 1 ( S , C ) H^1(S,\mathbb C) H 1 ( S , C ) by global closed reflexive forms. Small linear combinations of these representatives give closed forms α j → ω \alpha_j\to\omega α j → ω on compact subsets of the smooth locus and with [ α j ] = ξ j [\alpha_j]=\xi_j [ α j ] = ξ j .
Average α j \alpha_j α j over the circle subgroup of the given C ∗ \mathbb C^* C ∗ -action. The resulting ω j \omega_j ω j are closed reflexive holomorphic forms, retain classes ξ j \xi_j ξ j because the circle action is homotopic to the identity, and converge to ω \omega ω : the latter is invariant by Cartan’s formula since d ω = 0 d\omega=0 d ω = 0 and ω ( X ) = 1 \omega(X)=1 ω ( X ) = 1 on the smooth locus. The averages are invariant, so Cartan’s formula gives d ( ω j ( X ) ) = 0 d(\omega_j(X))=0 d ( ω j ( X )) = 0 on S r e g S_{\rm reg} S reg ; this locus is connected by normality. The orbit integral then gives 2 π i ω j ( X ) = ⟨ ξ j , γ ⟩ = 2 π i 2\pi i\,\omega_j(X)=\langle\xi_j,\gamma\rangle=2\pi i 2 π i ω j ( X ) = ⟨ ξ j , γ ⟩ = 2 π i . Thus the normalization is exact. The selected cohomology classes have discrete period groups, and local exactness together with Lemma 2.1 identifies the periods on S r e g S_{\rm reg} S reg with those on S S S . ◻
Discrete periods have a direct geometric interpretation, which we now formulate for normal spaces.
Corollary 2.4 (Reflexive Tischler fibration). Let S S S be a connected normal Stein space satisfying the degree-one reflexive Poincaré property, and let X X X be the infinitesimal generator of a holomorphic C ∗ \mathbb C^* C ∗ -action on S S S . Suppose that ω ∈ H 0 ( S , Z S 1 ) \omega\in H^0(S,\mathcal Z_S^1) ω ∈ H 0 ( S , Z S 1 ) is a closed reflexive holomorphic one-form with ω ( X ) = 1 \omega(X)=1 ω ( X ) = 1 on S r e g S_{\rm reg} S reg and discrete period group Λ = Per ( ω ) ⊂ C \Lambda=\operatorname{Per}(\omega)\subset\mathbb C Λ = Per ( ω ) ⊂ C . Then integration induces a surjective holomorphic locally trivial fiber bundle, in the category of complex analytic spaces, F : S ⟶ C / Λ , F:S\longrightarrow\mathbb C/\Lambda, F : S ⟶ C /Λ , whose fibers are connected Stein analytic spaces, possibly singular. The identity F ∗ ( d t ) = ω F^*(dt)=\omega F ∗ ( d t ) = ω holds on S r e g S_{\rm reg} S reg , and the holomorphic flow satisfies F ( φ t ( p ) ) = F ( p ) + [ t ] ( t ∈ C ) . F(\varphi_t(p))=F(p)+[t]\quad(t\in\mathbb C). F ( φ t ( p )) = F ( p ) + [ t ] ( t ∈ C ) . Moreover, 2 π i ∈ Λ 2\pi i\in\Lambda 2 π i ∈ Λ , so C / Λ \mathbb C/\Lambda C /Λ is biholomorphic either to C ∗ \mathbb C^* C ∗ or to an elliptic curve. No finite-generation hypothesis on H 1 ( S , Z ) H_1(S,\mathbb Z) H 1 ( S , Z ) is required.
Proof. By the degree-one reflexive Poincaré property, the form ω \omega ω has local holomorphic primitives on S S S ; differences of primitives are locally constant. Their additive monodromy is the period homomorphism π 1 ( S ) → C \pi_1(S)\to\mathbb C π 1 ( S ) → C of ω \omega ω , whose image is Λ \Lambda Λ . The local primitive property ensures that the period homomorphism on the regular locus factors through π 1 ( S ) \pi_1(S) π 1 ( S ) : small loops around the singular set contribute no additional monodromy. Equivalently, the local primitives define a class in H 1 ( S , C ) H^1(S,\mathbb C) H 1 ( S , C ) whose restriction gives the periods on S r e g S_{\rm reg} S reg . Thus the local primitives glue modulo Λ \Lambda Λ into a holomorphic map F : S → C / Λ F:S\to\mathbb C/\Lambda F : S → C /Λ satisfying F ∗ ( d t ) = ω F^*(dt)=\omega F ∗ ( d t ) = ω on the regular locus. Here Lemma 2.1 identifies the period images on S S S and S r e g S_{\rm reg} S reg .
If f f f is a local holomorphic primitive of ω \omega ω , then X ( f ) = 1 X(f)=1 X ( f ) = 1 on the regular locus and hence throughout its domain. The local holomorphic flow therefore gives product charts transverse to the analytic level sets of f f f , including at singular points.
Let φ t \varphi_t φ t be the complete holomorphic flow. Since ω ( X ) = 1 \omega(X)=1 ω ( X ) = 1 on the regular locus and both sides extend holomorphically, the time derivative of F ( φ t ( p ) ) F(\varphi_t(p)) F ( φ t ( p )) is the invariant unit tangent vector on C / Λ \mathbb C/\Lambda C /Λ . Hence F ( φ t ( p ) ) = F ( p ) + [ t ] F(\varphi_t(p))=F(p)+[t] F ( φ t ( p )) = F ( p ) + [ t ] throughout S S S , which also gives surjectivity.
Form the pullback of the covering C → C / Λ \mathbb C\to\mathbb C/\Lambda C → C /Λ :
S ^ = S × C / Λ C , T : S ^ → C . \widehat S=S\times_{\mathbb C/\Lambda}\mathbb C,
\qquad T:\widehat S\to\mathbb C. S = S × C /Λ C , T : S → C . This covering is connected, since its monodromy surjects onto Λ \Lambda Λ by the definition of the period group. The lifted complete flow is given by φ ^ t ( p , z ) = ( φ t ( p ) , z + t ) \widehat\varphi_t(p,z)=(\varphi_t(p),z+t) φ t ( p , z ) = ( φ t ( p ) , z + t ) , and satisfies T ( φ ^ t ( q ) ) = T ( q ) + t T(\widehat\varphi_t(q))=T(q)+t T ( φ t ( q )) = T ( q ) + t . Setting L = T − 1 ( 0 ) L=T^{-1}(0) L = T − 1 ( 0 ) , we obtain mutually inverse holomorphic maps
L × C ⟶ S ^ , ( q , t ) ⟼ φ ^ t ( q ) , L\times\mathbb C\longrightarrow\widehat S,\qquad
(q,t)\longmapsto\widehat\varphi_t(q), L × C ⟶ S , ( q , t ) ⟼ φ t ( q ) , and
S ^ ⟶ L × C , q ⟼ ( φ ^ − T ( q ) ( q ) , T ( q ) ) . \widehat S\longrightarrow L\times\mathbb C,\qquad
q\longmapsto\bigl(\widehat\varphi_{-T(q)}(q),T(q)\bigr). S ⟶ L × C , q ⟼ ( φ − T ( q ) ( q ) , T ( q ) ) . Thus S ^ ≃ L × C \widehat S\simeq L\times\mathbb C S ≃ L × C and L L L is connected. Over a simply covered open disc in C / Λ \mathbb C/\Lambda C /Λ , this product structure descends to a holomorphic trivialization of F F F with fiber L L L . The fiber is a closed analytic subspace of the Stein space S S S and hence Stein. Finally, the orbit circle in the regular locus has period 2 π i 2\pi i 2 π i , so 2 π i ∈ Λ 2\pi i\in\Lambda 2 π i ∈ Λ . A discrete subgroup of C \mathbb C C containing 2 π i 2\pi i 2 π i has rank one or two, with quotient respectively C ∗ \mathbb C^* C ∗ or an elliptic curve. ◻
Proof of Theorem A . Under the hypotheses of Theorem A , Theorem 2.3 provides closed, normalized reflexive forms ω j → ω \omega_j\to\omega ω j → ω on compact subsets of S r e g S_{\rm reg} S reg , with discrete periods and, when b 1 ( S ) ≥ 2 b_1(S)\ge2 b 1 ( S ) ≥ 2 , with rank-two lattice periods. This proves every approximation assertion of the introductory theorem. For completeness, its geometric consequence follows by applying Corollary 2.4 to each ω j \omega_j ω j : we obtain a holomorphic locally trivial fibration S → C / Per ( ω j ) S\to\mathbb C/\operatorname{Per}(\omega_j) S → C / Per ( ω j ) with connected Stein analytic fibers. Since 2 π i ∈ Per ( ω j ) 2\pi i\in\operatorname{Per}(\omega_j) 2 π i ∈ Per ( ω j ) , the base is C ∗ \mathbb C^* C ∗ or an elliptic curve, according to the rank of the period group. ◻
For the geometric form we have:
Proof of Theorem B . By Theorem A, there exists a sequence of closed normalized reflexive holomorphic one-forms ω j \omega_j ω j with discrete period groups
Λ j = Per ( ω j ) \Lambda_j=\operatorname{Per}(\omega_j) Λ j = Per ( ω j ) such that ω j → ω \omega_j\to\omega ω j → ω uniformly on compact subsets of S r e g S_{\mathrm{reg}} S reg .
By Corollary 2.4 , integration of each ω j \omega_j ω j induces a holomorphic locally trivial fibration
F j : S ⟶ C / Λ j F_j:S\longrightarrow\mathbb C/\Lambda_j F j : S ⟶ C / Λ j with connected Stein analytic fibers, satisfying
F j ∗ ( d t j ) = ω j on S r e g . F_j^*(dt_j)=\omega_j
\quad\text{on }S_{\mathrm{reg}}. F j ∗ ( d t j ) = ω j on S reg . The required convergence follows immediately.
Since ω j ( X ) = 1 \omega_j(X)=1 ω j ( X ) = 1 , integration along an orbit circle gives
2 π i Z ⊆ Λ j . 2\pi i\mathbb Z\subseteq\Lambda_j. 2 π i Z ⊆ Λ j . Thus each discrete group Λ j \Lambda_j Λ j has rank one or two. If b 1 ( S ) ≥ 2 b_1(S)\geq2 b 1 ( S ) ≥ 2 , Theorem A allows the approximating forms to be chosen with rank-two lattice periods, yielding elliptic curves as bases.
Finally, the foliations induced by the fibrations are defined on S r e g S_{\mathrm{reg}} S reg by ker ω j \ker\omega_j ker ω j , and the convergence ω j → ω \omega_j\to\omega ω j → ω gives the asserted approximation of their normalized defining forms. ◻
The local primitive property is available beyond quotient singularities; the next proposition supplies a useful class of examples.
Proposition 2.5 (A non-quotient class with local primitives). A normal complex space with isolated complete-intersection singularities of complex dimension at least three satisfies the degree-one reflexive Poincaré property. The same is true for its product with a complex manifold.
Proof. The required local topological vanishing for isolated complete intersections is recorded in [28 Proposition 3.3] . We recall the elementary argument. At a smooth point this is the holomorphic Poincaré lemma. At an isolated complete-intersection singularity of dimension n ≥ 3 n\ge3 n ≥ 3 , a sufficiently small punctured representative is simply connected by Hamm’s connectivity theorem for isolated complete intersections [26 ] (for hypersurfaces, see Milnor[30 ] ). A closed holomorphic one-form there has a single-valued holomorphic primitive by path integration. Normality extends the primitive across the isolated point, and its differential agrees with the original reflexive form on the smooth locus. For a product with a complex manifold, integrate first on the punctured singular factor and on small coordinate polydiscs in the smooth factor; the resulting local primitive extends across the product singular locus by normality. Equivalently, use the local product decomposition of the punctured neighborhoods and the ordinary Poincaré lemma in the smooth directions. ◻
The following example illustrates the approximation theorem on a singular Stein space that is not locally a quotient of a smooth complex manifold by a finite group.
Example 2.6 (Non-quotient singular Stein approximation). Let Y = { x y − z w = 0 } ⊂ C 4 Y=\{xy-zw=0\}\subset\mathbb C^4 Y = { x y − z w = 0 } ⊂ C 4 , the three-dimensional ordinary double point, and let S = Y × ( C ∗ ) 2 S=Y\times(\mathbb C^*)^2 S = Y × ( C ∗ ) 2 . It is a normal affine Stein space with singularities and satisfies the degree-one reflexive Poincaré property by Proposition 2.5 . The local class group of the threefold node is infinite cyclic, so its singular germ is not a finite smooth quotient germ (which is Q \mathbb Q Q -factorial). With coordinates ( u , v ) (u,v) ( u , v ) on the torus factor, set
X = u ∂ u , ω λ = d u u + λ d v v . X=u\partial_u,\qquad
\omega_\lambda=\frac{du}{u}+\lambda\frac{dv}{v}. X = u ∂ u , ω λ = u d u + λ v d v . If λ ∈ R ∖ Q \lambda\in\mathbb R\setminus\mathbb Q λ ∈ R ∖ Q , then Per ( ω λ ) = 2 π i ( Z + λ Z ) \operatorname{Per}(\omega_\lambda)=2\pi i(\mathbb Z+\lambda\mathbb Z) Per ( ω λ ) = 2 π i ( Z + λ Z ) is nondiscrete. For λ j ∈ Q + i Q \lambda_j\in\mathbb Q+i\mathbb Q λ j ∈ Q + i Q with nonzero imaginary parts tending to λ \lambda λ , the ω λ j \omega_{\lambda_j} ω λ j are closed and normalized with rank-two lattice periods and converge to ω λ \omega_\lambda ω λ . This gives an explicit case beyond finite smooth quotients.
Equivariant approximation on finite Stein quotients
Finite quotients provide a complementary setting in which approximation is carried out equivariantly on a smooth Stein covering space before descent.
The preceding local obstructions concern the existence of normalized forms. There is nevertheless a substantial class of genuinely singular normal Stein spaces where the approximation procedure itself survives. We give a finite-equivariant version of the smooth theorem, and explicitly distinguish descent of closed forms and periods from the stronger assertion that their leaves form bundles on the singular quotient.
Proof. Since the quotient V / K V/K V / K is smooth near the image of the origin, the Chevalley–Shephard–Todd theorem implies that K K K is generated by complex pseudoreflections, that is, elements whose fixed-point subspaces are hyperplanes (see [37 ] ).
Let D D D denote the union of the reflecting hyperplanes of K K K . Away from the image of D D D , the quotient map π : V → V / K \pi:V\to V/K π : V → V / K is unramified, and every K K K -invariant holomorphic one-form descends uniquely to a holomorphic one-form on the corresponding open subset of V / K V/K V / K .
We now examine the extension across the branch divisor. At a generic point of a component of D D D , the stabilizer is cyclic, and suitable holomorphic coordinates may be chosen so that its action is given by
( z 1 , z ′ ) ⟼ ( ζ z 1 , z ′ ) , ζ m = 1. (z_1,z')\longmapsto(\zeta z_1,z'),
\qquad \zeta^m=1. ( z 1 , z ′ ) ⟼ ( ζ z 1 , z ′ ) , ζ m = 1. The corresponding quotient coordinates are
w 1 = z 1 m , w ′ = z ′ . w_1=z_1^m,\qquad w'=z'. w 1 = z 1 m , w ′ = z ′ . Let
α = a 1 ( z ) d z 1 + ∑ j > 1 a j ( z ) d z j \alpha=a_1(z)\,dz_1+
\sum_{j>1}a_j(z)\,dz_j α = a 1 ( z ) d z 1 + j > 1 ∑ a j ( z ) d z j be a K K K -invariant holomorphic one-form. Invariance under the cyclic stabilizer implies that its coefficients have the form
a 1 ( z ) = z 1 m − 1 b 1 ( z 1 m , z ′ ) , a j ( z ) = b j ( z 1 m , z ′ ) ( j > 1 ) , a_1(z)=z_1^{m-1}b_1(z_1^m,z'),
\qquad
a_j(z)=b_j(z_1^m,z')
\quad (j>1), a 1 ( z ) = z 1 m − 1 b 1 ( z 1 m , z ′ ) , a j ( z ) = b j ( z 1 m , z ′ ) ( j > 1 ) , where the functions b j b_j b j are holomorphic. Consequently,
α = 1 m b 1 ( w ) d w 1 + ∑ j > 1 b j ( w ) d w j , \alpha
=
\frac{1}{m}b_1(w)\,dw_1+
\sum_{j>1}b_j(w)\,dw_j, α = m 1 b 1 ( w ) d w 1 + j > 1 ∑ b j ( w ) d w j , showing that the descended form extends holomorphically across the generic points of every branch divisor.
The remaining exceptional subset of the smooth quotient germ has complex codimension at least two. Since the quotient is smooth, the coefficient functions of the descended form extend holomorphically across this subset by Hartogs’ extension theorem. Thus α \alpha α is the pullback of a holomorphic one-form on the quotient germ. Uniqueness follows from uniqueness of descent on the unramified locus, which is dense.
This conclusion is also consistent with Solomon’s description of invariant differential forms for finite complex reflection groups [38 ] .
Finally, suppose that α ν \alpha_\nu α ν is a sequence of K K K -invariant holomorphic one-forms converging uniformly on compact subsets upstairs. In the cyclic quotient coordinates above, the coefficients of the descended forms depend continuously on the invariant coefficients of α ν \alpha_\nu α ν , as follows from their power-series expansions and Cauchy estimates. Hence the descended forms converge uniformly on compact subsets away from the exceptional subset of codimension at least two.
Near that exceptional subset, choose small coordinate polydiscs in the smooth quotient whose distinguished boundaries avoid the exceptional set. The maximum modulus principle, applied to the holomorphically extended coefficient functions, gives uniform convergence on compact subsets of these polydiscs. A finite covering argument then proves compact-open convergence throughout the quotient germ. ◻
Theorem 3.2 (Tischler approximation on finite Stein quotients). Let M M M be a connected Stein manifold with finitely generated H 1 ( M , Z ) H_1(M,\mathbb Z) H 1 ( M , Z ) , and let G G G be a finite group acting holomorphically on M M M . Suppose that a holomorphic C ∗ \mathbb C^* C ∗ -action on M M M commutes with G G G and has nowhere-vanishing infinitesimal generator X X X . Given a G G G -invariant closed holomorphic one-form ω \omega ω such that ω ( X ) = 1 \omega(X)=1 ω ( X ) = 1 , there are G G G -invariant closed holomorphic one-forms ω j \omega_j ω j on M M M satisfying ω j ( X ) = 1 , ω j ⟶ ω uniformly on compact subsets of M , Per ( ω j ) ⊂ C discrete . \omega_j(X)=1,\qquad \omega_j\longrightarrow\omega
\quad\text{uniformly on compact subsets of }M,
\qquad \operatorname{Per}(\omega_j)\subset\mathbb C\ \text{discrete}. ω j ( X ) = 1 , ω j ⟶ ω uniformly on compact subsets of M , Per ( ω j ) ⊂ C discrete . Writing S = M / G S=M/G S = M / G and π : M → S \pi:M\to S π : M → S for the quotient map, the forms ω , ω j \omega,\omega_j ω , ω j descend to reflexive one-forms ω ˉ , ω ˉ j \bar\omega,\bar\omega_j ω ˉ , ω ˉ j on the normal Stein space S S S . Their restrictions to S r e g S_{\rm reg} S reg are closed and normalized by the descended vector field X ˉ \bar X X ˉ ; furthermore ω ˉ j → ω ˉ \bar\omega_j\to\bar\omega ω ˉ j → ω ˉ uniformly on compact subsets of S r e g S_{\rm reg} S reg and Per ( ω ˉ j ∣ S r e g ) ⊆ 1 ∣ G ∣ Per ( ω j ) , \operatorname{Per}(\bar\omega_j|_{S_{\rm reg}})
\subseteq\tfrac1{|G|}\operatorname{Per}(\omega_j), Per ( ω ˉ j ∣ S reg ) ⊆ ∣ G ∣ 1 Per ( ω j ) , so these descended period groups are discrete.
Proof. Let γ \gamma γ denote the common orbit-circle class in H 1 ( M , Z ) H_1(M,\mathbb Z) H 1 ( M , Z ) . Since ω ( X ) = 1 \omega(X)=1 ω ( X ) = 1 , its period is 2 π i 2\pi i 2 π i , so its image in H 1 ( M , Q ) H_1(M,\mathbb Q) H 1 ( M , Q ) is nonzero. Because G G G commutes with the C ∗ \mathbb C^* C ∗ -action, γ \gamma γ is G G G -fixed. Consider the rational vector space
V Q = H 1 ( M , Q ) G , V C = V Q ⊗ Q C = H 1 ( M , C ) G . V_\mathbb Q=H^1(M,\mathbb Q)^G,
\qquad V_\mathbb C=V_\mathbb Q\otimes_\mathbb Q\mathbb C=H^1(M,\mathbb C)^G. V Q = H 1 ( M , Q ) G , V C = V Q ⊗ Q C = H 1 ( M , C ) G . The equality follows from finite-group averaging. The affine equation ⟨ ξ , γ ⟩ = 2 π i \langle\xi,\gamma\rangle=2\pi i ⟨ ξ , γ ⟩ = 2 π i has a rational solution after division by 2 π i 2\pi i 2 π i : the functional ⟨ , γ ⟩ \langle\,,\gamma\rangle ⟨ , γ ⟩ is rational and nonzero on V Q V_\mathbb Q V Q . Thus the normalized affine hyperplane within V C V_\mathbb C V C has dense points with coordinates in 2 π i ( Q + i Q ) 2\pi i(\mathbb Q+i\mathbb Q) 2 π i ( Q + i Q ) relative to a rational basis, while preserving the orbit-period equation exactly. Choose such invariant classes ξ j → [ ω ] \xi_j\to[\omega] ξ j → [ ω ] . Because H 1 ( M , Z ) H_1(M,\mathbb Z) H 1 ( M , Z ) is finitely generated, the image of each ξ j \xi_j ξ j is contained in a Gaussian rational lattice N j − 1 2 π i Z [ i ] N_j^{-1}2\pi i\mathbb Z[i] N j − 1 2 π i Z [ i ] , for some positive integer N j N_j N j , and hence is discrete.
By holomorphic de Rham theory on the Stein manifold M M M , choose closed global holomorphic representatives of a finite basis of H 1 ( M , C ) H^1(M,\mathbb C) H 1 ( M , C ) . Perturbing ω \omega ω by small linear combinations gives closed forms α j → ω \alpha_j\to\omega α j → ω in the compact-open topology with [ α j ] = ξ j [\alpha_j]=\xi_j [ α j ] = ξ j . Average α j \alpha_j α j over G G G and then over the circle subgroup of C ∗ \mathbb C^* C ∗ . The averaged forms are closed and G G G -invariant, their cohomology classes remain ξ j \xi_j ξ j , and they converge to ω \omega ω (which is itself invariant under the circle by Cartan’s formula). By the orbit-period argument in Theorem 9.1 , every averaged form has contraction 1 1 1 with X X X . Denote these averaged forms by ω j \omega_j ω j .
For descent, replace G G G by its effective quotient G / ker ( G → Aut ( M ) ) G/\ker(G\to\operatorname{Aut}(M)) G / ker ( G → Aut ( M )) , if necessary. This does not change S = M / G S=M/G S = M / G or the invariant forms, and its order divides the order of the original group. With the resulting effective action, let M ∘ ⊂ M M^\circ\subset M M ∘ ⊂ M be the free-action locus and put S ∘ = M ∘ / G S^\circ=M^\circ/G S ∘ = M ∘ / G . The map π : M ∘ → S ∘ \pi:M^\circ\to S^\circ π : M ∘ → S ∘ is a finite unramified covering, so the invariant forms descend uniquely to S ∘ S^\circ S ∘ . For descent across smooth branch points, fix q ∈ S r e g q\in S_{\rm reg} q ∈ S reg and a lift p p p . Finite-group local linearization [22 ] models the quotient germ at q q q by the linear action of the stabilizer G p G_p G p . Since that germ is smooth, the Chevalley–Shephard–Todd criterion [37 ] applies. Lemma 3.1 therefore extends the invariant form uniquely to a holomorphic one-form near q q q . On the complement of S r e g S_{\rm reg} S reg , which has codimension at least two by normality, these forms determine a unique reflexive differential in j ∗ Ω S r e g 1 j_*\Omega^1_{S_{\rm reg}} j ∗ Ω S reg 1 . Closedness and normalization are equalities on the regular locus because they hold over the dense unramified set. Compact-open convergence of the descended forms on S r e g S_{\rm reg} S reg follows by applying the convergence assertion of Lemma 3.1 in finite local quotient charts.
To compare periods, represent a loop in S r e g S_{\rm reg} S reg by a loop avoiding the branch locus, which has real codimension two and can be avoided by a small perturbation; closedness preserves its integral. A lift to M ∘ M^\circ M ∘ ends at a point in the same G G G -orbit. If the monodromy element is g ∈ G g\in G g ∈ G , its order m m m divides ∣ G ∣ |G| ∣ G ∣ (Lagrange’s theorem). The m m m -fold concatenation lifts to a closed loop, giving m ∫ γ ω ˉ j ∈ Per ( ω j ) m\int_\gamma\bar\omega_j\in\operatorname{Per}(\omega_j) m ∫ γ ω ˉ j ∈ Per ( ω j ) . Therefore also ∣ G ∣ ∫ γ ω ˉ j ∈ Per ( ω j ) |G|\int_\gamma\bar\omega_j\in\operatorname{Per}(\omega_j) ∣ G ∣ ∫ γ ω ˉ j ∈ Per ( ω j ) , which proves the displayed containment and discreteness. This argument makes no assertion that integration defines a locally trivial fibration across S s i n g S_{\rm sing} S sing . ◻
Proof of Theorem C . The hypotheses of the introductory theorem are those of Theorem 3.2 . That theorem constructs G G G -invariant closed normalized approximants upstairs and proves both their descent to closed normalized reflexive forms on the quotient and discreteness of their periods on the regular locus. These are precisely the assertions of Theorem C . ◻
The following example illustrates the approximation theorem on a singular Stein space, with approximating forms whose period groups are lattices in C \mathbb C C .
Example 3.3 (Singular product approximation with elliptic periods). Let Y Y Y be any connected normal Stein space (in particular, a singular one), and put
S = Y × ( C ∗ ) 2 , X = z ∂ z , ω λ = d z z + λ d w w . S=Y\times(\mathbb C^*)^2,\qquad
X=z\partial_z,\qquad
\omega_\lambda=\frac{dz}{z}+\lambda\frac{dw}{w}. S = Y × ( C ∗ ) 2 , X = z ∂ z , ω λ = z d z + λ w d w . The forms are closed reflexive differentials on S S S , normalized by X X X , and their period groups are
Per ( ω λ ) = 2 π i ( Z + λ Z ) . \operatorname{Per}(\omega_\lambda)=2\pi i(\mathbb Z+\lambda\mathbb Z). Per ( ω λ ) = 2 π i ( Z + λ Z ) . Fix an irrational real λ \lambda λ and choose λ j ∈ Q + i Q \lambda_j\in\mathbb Q+i\mathbb Q λ j ∈ Q + i Q with Im λ j ≠ 0 \operatorname{Im}\lambda_j\ne0 Im λ j = 0 and λ j → λ \lambda_j\to\lambda λ j → λ . Then Per ( ω λ ) \operatorname{Per}(\omega_\lambda) Per ( ω λ ) is nondiscrete, whereas each Per ( ω λ j ) \operatorname{Per}(\omega_{\lambda_j}) Per ( ω λ j ) is a rank-two lattice, and ω λ j → ω λ \omega_{\lambda_j}\to\omega_\lambda ω λ j → ω λ uniformly on compact sets. For Y = { x y = z 0 2 } ⊂ C 3 Y=\{xy=z_0^2\}\subset\mathbb C^3 Y = { x y = z 0 2 } ⊂ C 3 this gives a genuinely singular normal Stein space exhibiting discrete-period approximation (the coordinate z 0 z_0 z 0 is distinct from the C ∗ \mathbb C^* C ∗ -coordinate z z z ). This example is elementary and does not assert a local fiber-bundle structure at singular points.
Normal Stein spaces: local period obstructions and flat connections
We now examine the local restrictions imposed by normalization, especially near singular points of normal Stein surfaces.
The preceding sections establish approximation results for normal Stein spaces, finite Stein quotients, and smooth Stein manifolds. We now examine local existence and obstruction phenomena on normal Stein spaces. Throughout this section, differentials on a normal space S S S are understood in the reflexive sense Ω S [ 1 ] = j ∗ Ω S r e g 1 \Omega_S^{[1]}=j_*\Omega_{S_{\mathrm{reg}}}^1 Ω S [ 1 ] = j ∗ Ω S reg 1 , where j : S r e g ↪ S j:S_{\mathrm{reg}}\hookrightarrow S j : S reg ↪ S . Closedness and contraction by vector fields are interpreted on the regular locus. We make no general claim that reflexive closed forms compute H 1 ( S , C ) H^1(S,\mathbb C) H 1 ( S , C ) .
Proposition 4.1 (Fixed singular points on normal surfaces). Let S S S be a normal complex analytic surface and let a connected complex Lie group act holomorphically on S S S . Every point of S i n g ( S ) \mathop{\mathrm{Sing}}(S) Sing ( S ) is fixed. In particular, a normal surface carrying a fixed-point-free holomorphic C ∗ \mathbb C^* C ∗ -action is smooth. This fixed-point observation is classical; compare [33 Section 1.3] .
Proof. Normality implies that S i n g ( S ) \mathop{\mathrm{Sing}}(S) Sing ( S ) has complex codimension at least two, hence is discrete. Every automorphism preserves S i n g ( S ) \mathop{\mathrm{Sing}}(S) Sing ( S ) . The orbit of a singular point is the connected image of a connected Lie group in a discrete space and is therefore a singleton. ◻
Two elementary consequences of normalization will be useful when analyzing singular orbits.
Lemma 4.2. Let S S S be a connected normal complex analytic space satisfying the degree-one reflexive Poincaré property. Let X X X generate a holomorphic C ∗ \mathbb C^* C ∗ -action on S S S , and suppose that there exists a closed reflexive holomorphic one-form ω \omega ω satisfying ω ( X ) = 1 on S r e g . \omega(X)=1
\quad\text{on }S_{\mathrm{reg}}. ω ( X ) = 1 on S reg . Then the action has no fixed points, and every orbit circle represents a nontorsion class in H 1 ( S , Z ) H_1(S,\mathbb Z) H 1 ( S , Z ) .
Proof. Near a hypothetical fixed point p p p , the degree-one reflexive Poincaré property gives ω = d f \omega=df ω = df for a holomorphic function f f f . Thus X ( f ) = 1 X(f)=1 X ( f ) = 1 on the regular locus and hence on the whole neighborhood by normality. At a fixed point, however, X ( f ) ( p ) = 0 X(f)(p)=0 X ( f ) ( p ) = 0 , a contradiction.
Let p ∈ S r e g p\in S_{\mathrm{reg}} p ∈ S reg , and consider the orbit circle
γ p ( θ ) = e i θ ⋅ p , 0 ≤ θ ≤ 2 π . \gamma_p(\theta)=e^{i\theta}\cdot p,
\qquad 0\leq\theta\leq2\pi. γ p ( θ ) = e i θ ⋅ p , 0 ≤ θ ≤ 2 π . Since γ ˙ p = i X ( γ p ) \dot\gamma_p=iX(\gamma_p) γ ˙ p = i X ( γ p ) , normalization gives
∫ γ p ω = 2 π i . \int_{\gamma_p}\omega=2\pi i. ∫ γ p ω = 2 π i . By the degree-one reflexive Poincaré property, the periods of ω \omega ω are defined on H 1 ( S , Z ) H_1(S,\mathbb Z) H 1 ( S , Z ) . If [ γ p ] [\gamma_p] [ γ p ] were torsion, its period would vanish, a contradiction. ◻
We first examine how invariant local forms can be obtained by averaging.
Lemma 4.3. Let S S S be a normal complex surface with a holomorphic C ∗ \mathbb C^* C ∗ -action and infinitesimal generator X X X . On any S 1 S^1 S 1 -invariant open set V ⊂ S V\subset S V ⊂ S , if a reflexive holomorphic one-form η \eta η satisfies η ( X ) = 1 \eta(X)=1 η ( X ) = 1 on V ∩ S r e g V\cap S_{\rm reg} V ∩ S reg , there exists a reflexive holomorphic one-form ω \omega ω on V V V with ω ( X ) = 1 , d ω = 0 on V ∩ S r e g . \omega(X)=1,\qquad d\omega=0\quad\text{on }V\cap S_{\rm reg}. ω ( X ) = 1 , d ω = 0 on V ∩ S reg .
Proof. Average the pullbacks of η \eta η over S 1 S^1 S 1 . The average defines a holomorphic differential on the regular locus, hence a reflexive differential on the normal surface. It remains normalized, since the action preserves X X X . Circle invariance implies L X ω = 0 \mathcal L_X\omega=0 L X ω = 0 by differentiation and holomorphicity. Cartan’s identity yields ι X d ω = 0 \iota_Xd\omega=0 ι X d ω = 0 . Since ω ( X ) = 1 \omega(X)=1 ω ( X ) = 1 , the field is nowhere zero on the regular locus; contraction by a nonzero vector is injective on holomorphic two-forms on a smooth complex surface. Therefore d ω = 0 d\omega=0 d ω = 0 . ◻
The next lemma identifies the role of local Stein geometry in normalization.
Lemma 4.4 (Local Stein normalization and Poincaré obstruction). Let p p p belong to a normal complex surface S S S with a holomorphic C ∗ \mathbb C^* C ∗ -action generated by X X X . If the stalk morphism ι X : Ω S , p [ 1 ] → O S , p \iota_X:\Omega_{S,p}^{[1]}\to\mathcal O_{S,p} ι X : Ω S , p [ 1 ] → O S , p is surjective, then there is a Stein neighborhood of p p p on which a normalized reflexive form exists. If p p p is fixed, one may choose a sufficiently small S 1 S^1 S 1 -invariant Stein neighborhood and obtain a closed normalized reflexive form there. Conversely, existence of a normalized form in a neighborhood implies surjectivity at p p p . If p p p is fixed and the degree-one reflexive Poincaré property holds at p p p , the contraction morphism is not surjective at p p p .
Proof. Surjectivity holds on some neighborhood of p p p , as the cokernel of the coherent contraction morphism has closed analytic support. Choose a smaller Stein neighborhood U U U of p p p . On U U U , the kernel of contraction is coherent, so Cartan’s Theorem B lifts the unit section to a normalized reflexive form. If p p p is fixed, average over S 1 S^1 S 1 after choosing U U U invariant under this compact group; an S 1 S^1 S 1 -invariant Stein neighborhood basis can be obtained by averaging a local strictly plurisubharmonic exhaustion and taking small invariant sublevel sets. Lemma 4.3 then makes the normalized form closed. The converse follows by contracting the normalized section. Finally, a closed normalized form near a fixed point is locally d f df df under the degree-one reflexive Poincaré property, so X ( f ) = 1 X(f)=1 X ( f ) = 1 on the regular locus and hence everywhere by normality. At a fixed point X ( f ) ( p ) = 0 X(f)(p)=0 X ( f ) ( p ) = 0 , a contradiction. ◻
These local considerations lead to a criterion for global reflexive normalization.
Theorem 4.5 (Reflexive normalization criterion). Let S S S be a normal Stein surface with a holomorphic C ∗ \mathbb C^* C ∗ -action, and let X X X be its infinitesimal generator. The following conditions are equivalent:
The coherent contraction morphism ι X : Ω S [ 1 ] ⟶ O S \iota_X:\Omega_S^{[1]}\longrightarrow\mathcal O_S ι X : Ω S [ 1 ] ⟶ O S is surjective.
There exists η ∈ H 0 ( S , Ω S [ 1 ] ) \eta\in H^0(S,\Omega_S^{[1]}) η ∈ H 0 ( S , Ω S [ 1 ] ) with η ( X ) = 1 \eta(X)=1 η ( X ) = 1 on S r e g S_{\rm reg} S reg .
There exists ω ∈ H 0 ( S , Ω S [ 1 ] ) \omega\in H^0(S,\Omega_S^{[1]}) ω ∈ H 0 ( S , Ω S [ 1 ] ) whose restriction to S r e g S_{\rm reg} S reg is closed and satisfies ω ( X ) = 1 \omega(X)=1 ω ( X ) = 1 .
Proof. If the contraction morphism is surjective, its kernel is coherent; Cartan’s Theorem B lifts the global unit section 1 ∈ O ( S ) 1\in\mathcal O(S) 1 ∈ O ( S ) to a global reflexive differential η \eta η . This proves (1)⇒ \Rightarrow ⇒ (2).
Conversely, for a form as in (2), the contraction η ( X ) \eta(X) η ( X ) is a holomorphic function on S S S because S S S is normal and the regular locus has codimension-two complement. Since it equals 1 1 1 on the regular locus, it equals 1 1 1 on S S S ; multiplication of η \eta η by local holomorphic functions therefore provides a sheaf-theoretic right inverse of contraction. Thus (2)⇒ \Rightarrow ⇒ (1).
The local averaging lemma, Lemma 4.3 , applied to V = S V=S V = S gives (3) from (2). Finally, (3)⇒ \Rightarrow ⇒ (2) is immediate. ◻
The period of a closed time form imposes a basic homological constraint on its orbits.
Proposition 4.6. Let U ∗ U^* U ∗ be a smooth complex space carrying a holomorphic vector field X X X . Suppose q ∈ U ∗ q\in U^* q ∈ U ∗ has a nonconstant periodic complex-time orbit γ ( t ) = φ t ( q ) \gamma(t)=\varphi_t(q) γ ( t ) = φ t ( q ) , t t t running from 0 0 0 to τ ∈ C ∖ { 0 } \tau\in\mathbb C\setminus\{0\} τ ∈ C ∖ { 0 } , and that the class of this loop vanishes in H 1 ( U ∗ , C ) H_1(U^*,\mathbb C) H 1 ( U ∗ , C ) . Then no closed holomorphic one-form ω \omega ω on U ∗ U^* U ∗ can satisfy ω ( X ) = 1 \omega(X)=1 ω ( X ) = 1 .
Proof. A normalized form would have ∫ γ ω = ∫ 0 τ ω ( X ) d t = τ ≠ 0 \int_\gamma\omega=\int_0^\tau\omega(X)\,dt=\tau\ne0 ∫ γ ω = ∫ 0 τ ω ( X ) d t = τ = 0 . Integration of a closed complex-valued one-form factors through H 1 ( U ∗ , C ) H_1(U^*,\mathbb C) H 1 ( U ∗ , C ) and therefore vanishes on [ γ ] [\gamma] [ γ ] , a contradiction. ◻
The preceding homological observation is particularly effective for rational-homology-sphere links.
Corollary 4.7 (Rational-homology-sphere links). Let ( S , p ) (S,p) ( S , p ) be an isolated normal surface singularity with a holomorphic C ∗ \mathbb C^* C ∗ -action fixing p p p . Choose a sufficiently small S 1 S^1 S 1 -invariant link L L L about p p p and a punctured neighborhood U ∗ U^* U ∗ containing it such that L ↪ U ∗ L\hookrightarrow U^* L ↪ U ∗ is a homotopy equivalence. Suppose H 1 ( L , Q ) = 0 H_1(L,\mathbb Q)=0 H 1 ( L , Q ) = 0 and the circle action has a nonconstant orbit in L L L . Then there is no closed holomorphic one-form on U ∗ U^* U ∗ normalized by the infinitesimal generator. No invariance of the bounded neighborhood under the full C ∗ \mathbb C^* C ∗ -action is assumed.
Proof. The nonconstant circle orbit is a closed complex-time orbit (with nonzero period), entirely contained in L ⊂ U ∗ L\subset U^* L ⊂ U ∗ . Its class is torsion in H 1 ( L , Z ) H_1(L,\mathbb Z) H 1 ( L , Z ) and hence zero in H 1 ( U ∗ , C ) H_1(U^*,\mathbb C) H 1 ( U ∗ , C ) . Apply Proposition 4.6 . ◻
Example 4.8 (The quadratic cone: obstruction even without closedness). Let S = { x y = z 2 } ⊂ C 3 S=\{xy=z^2\}\subset\mathbb C^3 S = { x y = z 2 } ⊂ C 3 , with the scalar action λ ⋅ ( x , y , z ) = ( λ 2 x , λ 2 y , λ 2 z ) \lambda\cdot(x,y,z)=(\lambda^2x,\lambda^2y,\lambda^2z) λ ⋅ ( x , y , z ) = ( λ 2 x , λ 2 y , λ 2 z ) . The origin is the unique singularity, and S ≃ C 2 / { ± I } S\simeq\mathbb C^2/\{\pm I\} S ≃ C 2 / { ± I } by ( u , v ) ↦ ( u 2 , v 2 , u v ) (u,v)\mapsto(u^2,v^2,uv) ( u , v ) ↦ ( u 2 , v 2 , uv ) . The regular locus has H 1 ( S r e g , Z ) ≃ Z / 2 Z H_1(S_{\rm reg},\mathbb Z)\simeq\mathbb Z/2\mathbb Z H 1 ( S reg , Z ) ≃ Z /2 Z ; thus a closed normalized form is excluded by Proposition 4.6 . More strongly, no reflexive normalized holomorphic form exists. Indeed, its pullback to C 2 ∖ { 0 } \mathbb C^2\setminus\{0\} C 2 ∖ { 0 } would be A ( u , v ) d u + B ( u , v ) d v A(u,v)\,du+B(u,v)\,dv A ( u , v ) d u + B ( u , v ) d v with u A + v B = 1 uA+vB=1 u A + v B = 1 . Hartogs extension extends A , B A,B A , B across the origin, making u A + v B = 1 uA+vB=1 u A + v B = 1 at the origin, an impossibility.
The homological obstruction extends far beyond rational homology sphere links. The Seifert structure and weighted-homogeneous classification are established results, not new assertions here: see Orlik–Wagreich [33 ] in the algebraic setting and Camacho–Movasati–Scárdua [23 ] for good analytic actions on normal Stein surfaces. Recall that the link of a weighted-homogeneous normal surface singularity with an isolated fixed point is naturally a Seifert three-manifold. We use the rational Seifert Euler number e o r b e_{\rm orb} e orb , with the sign convention immaterial for its nonvanishing. If the Seifert data are ( b ; ( α i , β i ) ) (b;(\alpha_i,\beta_i)) ( b ; ( α i , β i )) , one convention is e o r b = b + ∑ i β i / α i e_{\rm orb}=b+\sum_i\beta_i/\alpha_i e orb = b + ∑ i β i / α i .
Lemma 4.9 (Seifert fiber homology). Let L L L be a closed oriented Seifert three-manifold with oriented base orbifold, and let h ∈ H 1 ( L , Z ) h\in H_1(L,\mathbb Z) h ∈ H 1 ( L , Z ) be the class of a regular circle fiber. If e o r b ≠ 0 e_{\rm orb}\ne0 e orb = 0 , then h h h is torsion.
Proof. Let L L L denote the link of the singularity, endowed with its natural Seifert fibration. Write ( g ; b ; ( α 1 , β 1 ) , … , ( α r , β r ) ) (g;b;(\alpha_1,\beta_1),\ldots,
(\alpha_r,\beta_r)) ( g ; b ; ( α 1 , β 1 ) , … , ( α r , β r )) for its Seifert invariants, where α i ≥ 2 \alpha_i\geq 2 α i ≥ 2 is the multiplicity of the i i i -th exceptional fiber and gcd ( α i , β i ) = 1 \gcd(\alpha_i,\beta_i)=1 g cd( α i , β i ) = 1 .
Let h ∈ H 1 ( L , Z ) h\in H_1(L,\mathbb Z) h ∈ H 1 ( L , Z ) denote the homology class of an oriented regular fiber, and let q i ∈ H 1 ( L , Z ) q_i\in H_1(L,\mathbb Z) q i ∈ H 1 ( L , Z ) denote the classes arising from small loops around the exceptional points of the base orbifold, lifted using a section over their complement.
The standard presentation of the fundamental group of a Seifert fibered three-manifold, after abelianization, gives the relations
α i q i − β i h = 0 , i = 1 , … , r , \alpha_i q_i-\beta_i h=0,
\qquad i=1,\ldots,r, α i q i − β i h = 0 , i = 1 , … , r , and
q 1 + ⋯ + q r + b h = 0. q_1+\cdots+q_r+bh=0. q 1 + ⋯ + q r + bh = 0. Here the signs are fixed by our choice of orientations and Seifert invariants; see the classical presentation of Seifert fibered manifold groups.
Tensoring with Q \mathbb Q Q , the first relations give
q i = β i α i h . q_i=\frac{\beta_i}{\alpha_i}h. q i = α i β i h . Substitution into the remaining relation yields
( b + ∑ i = 1 r β i α i ) h = 0 in H 1 ( L , Q ) . \left(
b+\sum_{i=1}^r\frac{\beta_i}{\alpha_i}
\right)h=0
\quad\text{in }H_1(L,\mathbb Q). ( b + i = 1 ∑ r α i β i ) h = 0 in H 1 ( L , Q ) . Thus, if the rational Seifert Euler number
e o r b = b + ∑ i = 1 r β i α i e_{\mathrm{orb}}
=
b+\sum_{i=1}^r\frac{\beta_i}{\alpha_i} e orb = b + i = 1 ∑ r α i β i is nonzero, then h = 0 h=0 h = 0 in H 1 ( L , Q ) H_1(L,\mathbb Q) H 1 ( L , Q ) . Equivalently, the regular-fiber class is torsion in H 1 ( L , Z ) H_1(L,\mathbb Z) H 1 ( L , Z ) . After tensoring with Q \mathbb Q Q , elimination of the q i q_i q i gives ( b + ∑ i β i / α i ) h = 0 (b+\sum_i\beta_i/\alpha_i)h=0 ( b + ∑ i β i / α i ) h = 0 for the corresponding choice of signs. This coefficient is, up to the standard sign convention, the rational Euler number. Its nonvanishing implies h = 0 h=0 h = 0 in H 1 ( L , Q ) H_1(L,\mathbb Q) H 1 ( L , Q ) , i.e. h h h is torsion integrally. ◻
The Seifert description of an isolated surface singularity makes the orbit obstruction explicit.
Theorem 4.10 (Seifert Euler obstruction). Let ( S , p ) (S,p) ( S , p ) be an isolated normal surface singularity with a holomorphic C ∗ \mathbb C^* C ∗ -action fixing p p p . Assume there is a sufficiently small S 1 S^1 S 1 -invariant Seifert link L ⊂ S r e g L\subset S_{\rm reg} L ⊂ S reg and a punctured representative U ∗ U^* U ∗ containing L L L . Assume that the Seifert circles are the orbits of this S 1 S^1 S 1 -action, and that the induced Seifert fibration has nonzero rational Euler number e o r b ( L ) e_{\rm orb}(L) e orb ( L ) . Then no closed holomorphic one-form on U ∗ U^* U ∗ is normalized by the infinitesimal generator. In particular, the conclusion applies to the standard link fibrations of isolated normal surface singularities with good (weighted-homogeneous) C ∗ \mathbb C^* C ∗ -actions.
Proof. Choose a point of L L L on a regular (nonexceptional) Seifert fiber. Its parametrized circle orbit represents a nonzero integer multiple of the class h h h of a regular fiber and, by Lemma 4.9 , is torsion in H 1 ( L , Z ) H_1(L,\mathbb Z) H 1 ( L , Z ) . Hence it is torsion in H 1 ( U ∗ , Z ) H_1(U^*,\mathbb Z) H 1 ( U ∗ , Z ) . The inclusion L ↪ U ∗ L\hookrightarrow U^* L ↪ U ∗ maps torsion homology classes to torsion classes. The same orbit, parametrized by the complex-time holomorphic flow, has nonzero time period; Proposition 4.6 therefore excludes a closed normalized one-form. For an isolated good C ∗ \mathbb C^* C ∗ -action, an adapted small link carries the canonical Seifert fibration described by Orlik–Wagreich [33 Sections 1–2] ; the analytic setting and equivariant resolution are discussed in [23 ] . For the good-action link, the canonical star-shaped resolution plumbing has a negative-definite intersection matrix. Eliminating the negative-definite arm blocks gives a one-by-one Schur complement, which is nonzero because the full intersection matrix is negative definite. The standard Seifert plumbing formula identifies this Schur complement, up to sign, with the rational Euler number [32 ] . Hence e o r b ( L ) ≠ 0 e_{\rm orb}(L)\ne0 e orb ( L ) = 0 . For the weighted-action viewpoint see also [34 ] ; compare [32 , 31 ] . The link is chosen equivariantly, without demanding that a bounded punctured neighborhood be preserved by the full complex action. ◻
The same obstruction has a formulation in terms of the contraction morphism.
Corollary 4.11 (Non-surjectivity at a good singularity). Let S S S be a normal Stein surface with a good holomorphic C ∗ \mathbb C^* C ∗ -action having an isolated singular fixed point p p p . Then the contraction morphism ι X : Ω S [ 1 ] → O S \iota_X:\Omega_S^{[1]}\to\mathcal O_S ι X : Ω S [ 1 ] → O S is not surjective at p p p .
Proof. If it were surjective at p p p , coherence and the local definition of surjectivity would provide, on a neighborhood of p p p , a reflexive one-form η \eta η with η ( X ) = 1 \eta(X)=1 η ( X ) = 1 on the regular locus. After shrinking to a sufficiently small S 1 S^1 S 1 -invariant neighborhood, Lemma 4.3 produces a closed normalized reflexive form on the punctured neighborhood. This contradicts Theorem 4.10 . ◻
The following example illustrates the Seifert Euler obstruction for a classical isolated surface singularity whose projectivization is an elliptic curve.
Example 4.12 (An elliptic cubic cone). Let S = { x 3 + y 3 + z 3 = 0 } ⊂ C 3 S=\{x^3+y^3+z^3=0\}\subset\mathbb C^3 S = { x 3 + y 3 + z 3 = 0 } ⊂ C 3 with the scalar C ∗ \mathbb C^* C ∗ -action. Its projectivization is a smooth elliptic cubic E ⊂ P 2 E\subset\mathbb P^2 E ⊂ P 2 , and S ∗ S^* S ∗ is the complement of the zero section in O E ( − 1 ) \mathcal O_E(-1) O E ( − 1 ) , of degree − 3 -3 − 3 . The circle fibration has Euler number − 3 -3 − 3 , and its homology sequence yields H 1 ( S ∗ , Z ) ≃ Z 2 ⊕ Z / 3 Z H_1(S^*,\mathbb Z)\simeq\mathbb Z^2\oplus\mathbb Z/3\mathbb Z H 1 ( S ∗ , Z ) ≃ Z 2 ⊕ Z /3 Z . The scalar orbit is the torsion fiber class. Hence a closed normalized form cannot exist although H 1 ( S ∗ , Q ) ≃ Q 2 ≠ 0 H_1(S^*,\mathbb Q)\simeq\mathbb Q^2\ne0 H 1 ( S ∗ , Q ) ≃ Q 2 = 0 .
For comparison, principal bundles over smooth curves admit a direct description in terms of flat connections.
Proposition 4.13. Let π : P → B \pi:P\to B π : P → B be a holomorphic principal C ∗ \mathbb C^* C ∗ -bundle over a connected Riemann surface, with normalized infinitesimal generator X X X . The following conditions are equivalent:
P P P has a closed holomorphic one-form ω \omega ω with ω ( X ) = 1 \omega(X)=1 ω ( X ) = 1 ;
P P P has a holomorphic principal connection.
If B B B is compact, these are equivalent to deg L = 0 \deg L=0 deg L = 0 for the associated line bundle L L L . If B B B is noncompact, they always hold.
Proof. A closed normalized ω \omega ω is invariant by Cartan’s formula, and so is a connection form. Conversely, the curvature of a holomorphic principal connection is a basic holomorphic two-form, which vanishes on the one-dimensional base, hence the connection is closed. For compact B B B the degree-zero condition is the classical Atiyah–Weil criterion for holomorphic line bundles. For noncompact B B B , the Stein and Oka properties of open Riemann surfaces together with H 2 ( B , Z ) = 0 H^2(B,\mathbb Z)=0 H 2 ( B , Z ) = 0 imply that L L L is holomorphically trivial; then d q / q dq/q d q / q is the required form. ◻
The next example shows that, in complex dimension three, a singular normal Stein space may admit a fixed-point-free holomorphic C ∗ \mathbb C^* C ∗ -action together with a closed normalized reflexive holomorphic one-form.
Example 4.14 (A singular Stein threefold with normalized periods). Put Y = { x y = z 2 } ⊂ C 3 Y=\{xy=z^2\}\subset\mathbb C^3 Y = { x y = z 2 } ⊂ C 3 and S = Y × C ∗ S=Y\times\mathbb C^* S = Y × C ∗ . This is a normal Stein threefold with singular locus { 0 } × C ∗ \{0\}\times\mathbb C^* { 0 } × C ∗ . The action on the second factor is globally free, and X = q ∂ q X=q\partial_q X = q ∂ q has the global holomorphic reflexive form ω = d q / q \omega=dq/q ω = d q / q . The form is closed and normalized and has period group 2 π i Z 2\pi i\mathbb Z 2 π i Z . Thus the surface rigidity of Proposition 4.1 is dimension-sensitive.
Normalized time forms and invariant transverse foliations
The following elementary criteria relate normalized forms to invariant transverse holomorphic foliations.
Proposition 5.1 (Normalization is automatic; closedness is not). Every nowhere-vanishing holomorphic vector field on a Stein manifold admits a global holomorphic one-form α \alpha α with α ( X ) = 1 \alpha(X)=1 α ( X ) = 1 . Even for a complete vector field on a Stein surface, such a form need not be closed, and there need not exist any closed normalized form.
Proof. Contraction by X X X gives a surjective morphism of locally free coherent sheaves Ω M 1 → O M \Omega_M^1\to\mathcal O_M Ω M 1 → O M , whose kernel is coherent. The induced cohomology sequence and Cartan’s Theorem B give a global splitting of 1 ∈ O ( M ) 1\in\mathcal O(M) 1 ∈ O ( M ) . For the second assertion take M = C ∗ × C M=\mathbb C^*\times\mathbb C M = C ∗ × C , with coordinates ( z , w ) (z,w) ( z , w ) , and
X = e w z ∂ z , α = e − w d z z . X=e^w z\partial_z,\qquad \alpha=e^{-w}\,\frac{dz}{z}. X = e w z ∂ z , α = e − w z d z . The flow is ( z , w ) ↦ ( z e e w t , w ) (z,w)\mapsto(ze^{e^w t},w) ( z , w ) ↦ ( z e e w t , w ) for all t ∈ C t\in\mathbb C t ∈ C , so X X X is complete; α ( X ) = 1 \alpha(X)=1 α ( X ) = 1 , but d α = − e − w d w ∧ d z / z ≠ 0 d\alpha=-e^{-w}dw\wedge dz/z\ne0 d α = − e − w d w ∧ d z / z = 0 . For any other normalized one-form η \eta η , restriction to the circle γ w ( θ ) = ( e i θ , w ) \gamma_w(\theta)=(e^{i\theta},w) γ w ( θ ) = ( e i θ , w ) necessarily gives ∫ γ w η = 2 π i e − w \int_{\gamma_w}\eta=2\pi i e^{-w} ∫ γ w η = 2 π i e − w . The circles γ w \gamma_w γ w are homologous while the proposed periods vary with w w w . Therefore no closed normalized η \eta η exists. ◻
The following example illustrates how varying orbit periods may obstruct the existence of a closed normalized holomorphic one-form.
Example 5.2. On the Stein surface M = C ∗ × C M=\mathbb C^*\times\mathbb C M = C ∗ × C take X = e w z ∂ z X=e^wz\partial_z X = e w z ∂ z and α = e − w d z / z \alpha=e^{-w}dz/z α = e − w d z / z . Then X X X is nonsingular and complete and α ( X ) = 1 \alpha(X)=1 α ( X ) = 1 , yet d α ≠ 0 d\alpha\ne0 d α = 0 . The orbit through ( z , w ) (z,w) ( z , w ) has fundamental period 2 π i e − w 2\pi i e^{-w} 2 π i e − w , so the periods of a hypothetical closed normalized form around homologous orbit circles would depend on w w w . Therefore T X = ∅ \mathcal T_X=\varnothing T X = ∅ . This exhibits why existence of a normalized holomorphic form does not suffice for Tischler approximation.
On a complex surface, the invariance condition simplifies considerably.
Proposition 5.3 (Invariance criterion on a surface). If dim C M = 2 \dim_{\mathbb C}M=2 dim C M = 2 and α ( X ) = 1 \alpha(X)=1 α ( X ) = 1 , then d α = 0 ⟺ L X α = 0 ⟺ ker α is invariant under the local flow of X . d\alpha=0\quad\Longleftrightarrow\quad\mathcal L_X\alpha=0\quad\Longleftrightarrow\quad\ker\alpha\text{ is invariant under the local flow of }X. d α = 0 ⟺ L X α = 0 ⟺ ker α is invariant under the local flow of X . The last equivalence is for the normalized form α \alpha α .
Proof. Cartan’s formula gives L X α = ι X d α \mathcal L_X\alpha=\iota_Xd\alpha L X α = ι X d α ; since X ≠ 0 X\ne0 X = 0 , contraction with X X X is injective on holomorphic two-forms on a surface. Invariance of the kernel means L X α = f α \mathcal L_X\alpha=f\alpha L X α = f α locally. Evaluation on X X X gives f = ( L X α ) ( X ) = X ( α ( X ) ) = 0 f=(\mathcal L_X\alpha)(X)=X(\alpha(X))=0 f = ( L X α ) ( X ) = X ( α ( X )) = 0 . ◻
Surface obstructions and universal covers
In complex dimension two, invariance of a normalized form has an especially strong consequence for its closedness.
Proposition 6.1. Let N N N be a connected Stein surface carrying a holomorphic C ∗ \mathbb C^* C ∗ -action without fixed points, and let X X X be its infinitesimal generator, normalized by X ( x ) = d d t ∣ t = 0 ( e t ⋅ x ) X(x)=\left.\frac{d}{dt}\right|_{t=0}(e^t\cdot x) X ( x ) = d t d t = 0 ( e t ⋅ x ) . There exists a holomorphic, closed one-form η \eta η on N N N satisfying η ( X ) = 1 , ( s ⋅ ) ∗ η = η ( s ∈ C ∗ ) . \eta(X)=1,\qquad (s\cdot)^*\eta=\eta\quad (s\in\mathbb C^*). η ( X ) = 1 , ( s ⋅ ) ∗ η = η ( s ∈ C ∗ ) . Every orbit circle γ x ( θ ) = e i θ ⋅ x \gamma_x(\theta)=e^{i\theta}\cdot x γ x ( θ ) = e i θ ⋅ x , 0 ≤ θ ≤ 2 π 0\leq\theta\leq2\pi 0 ≤ θ ≤ 2 π , has period ∫ γ x η = 2 π i . \int_{\gamma_x}\eta=2\pi i. ∫ γ x η = 2 π i . Consequently its homology class has infinite order and H 1 ( N , C ) ≠ 0 H^1(N,\mathbb C)\ne0 H 1 ( N , C ) = 0 .
Proof. Since X X X has no zero, contraction is a surjective morphism of locally free coherent sheaves
ι X : Ω N 1 ⟶ O N . \iota_X:\Omega_N^1\longrightarrow\mathcal O_N. ι X : Ω N 1 ⟶ O N . Its kernel is coherent; Cartan’s Theorem B yields α ∈ H 0 ( N , Ω N 1 ) \alpha\in H^0(N,\Omega_N^1) α ∈ H 0 ( N , Ω N 1 ) with α ( X ) = 1 \alpha(X)=1 α ( X ) = 1 . Average over S 1 S^1 S 1 :
η = 1 2 π ∫ 0 2 π ( e i θ ⋅ ) ∗ α d θ . \eta=\frac1{2\pi}\int_0^{2\pi}(e^{i\theta}\cdot)^*\alpha\,d\theta. η = 2 π 1 ∫ 0 2 π ( e i θ ⋅ ) ∗ α d θ . Then η ( X ) = 1 \eta(X)=1 η ( X ) = 1 and η \eta η is S 1 S^1 S 1 -invariant. Holomorphicity in the group parameter implies invariance under C ∗ \mathbb C^* C ∗ , and hence L X η = 0 \mathcal L_X\eta=0 L X η = 0 . Cartan’s identity gives
ι X d η = L X η − d ( η ( X ) ) = 0. \iota_Xd\eta=\mathcal L_X\eta-d(\eta(X))=0. ι X d η = L X η − d ( η ( X )) = 0. On a complex surface, contraction of a holomorphic two-form by a nonvanishing vector field is injective at each point. Thus d η = 0 d\eta=0 d η = 0 . Finally, γ ˙ x = i X ( γ x ) \dot\gamma_x=iX(\gamma_x) γ ˙ x = i X ( γ x ) , so integration gives 2 π i 2\pi i 2 π i . A closed one-form has zero period over a torsion homology class, proving the assertion. ◻
A complete vector field with a closed normalized time form also constrains the geometry of the universal covering space.
Proof. Let q : N ~ → N q:\widetilde N\to N q : N → N be the universal covering map. Since q ∗ η q^*\eta q ∗ η is holomorphic and closed, simple connectivity gives a holomorphic primitive F : N ~ → C F:\widetilde N\to\mathbb C F : N → C with d F = q ∗ η dF=q^*\eta d F = q ∗ η . The complete holomorphic flow φ : C × N → N \varphi:\mathbb C\times N\to N φ : C × N → N generated by X X X lifts, with identity initial condition, to a holomorphic C \mathbb C C -action φ ~ \widetilde\varphi φ on N ~ \widetilde N N because C \mathbb C C is simply connected. Its infinitesimal generator X ~ \widetilde X X satisfies
F ( φ ~ t ( x ) ) = F ( x ) + t . F(\widetilde\varphi_t(x))=F(x)+t. F ( φ t ( x )) = F ( x ) + t . Thus Σ = F − 1 ( 0 ) \Sigma=F^{-1}(0) Σ = F − 1 ( 0 ) is a smooth complex curve, and
C × Σ ⟶ N ~ , ( t , y ) ⟼ φ ~ t ( y ) \mathbb C\times\Sigma\longrightarrow\widetilde N,\quad (t,y)\longmapsto\widetilde\varphi_t(y) C × Σ ⟶ N , ( t , y ) ⟼ φ t ( y ) is biholomorphic: its holomorphic inverse sends x x x to ( F ( x ) , φ ~ − F ( x ) ( x ) ) (F(x),\widetilde\varphi_{-F(x)}(x)) ( F ( x ) , φ − F ( x ) ( x )) . The product and the simple connectivity of N ~ \widetilde N N show that Σ \Sigma Σ is connected and simply connected. Hence uniformization gives Σ ≃ C , D \Sigma\simeq\mathbb C,\mathbb D Σ ≃ C , D , or P 1 \mathbb P^1 P 1 . The last case is impossible: composing the inclusion of the compact curve P 1 ↪ N ~ \mathbb P^1\hookrightarrow\widetilde N P 1 ↪ N with q q q would produce a nonconstant holomorphic map P 1 → N \mathbb P^1\to N P 1 → N , whereas holomorphic functions on a Stein manifold separate points and are constant on compact Riemann surfaces.
For any deck transformation γ \gamma γ , we have d ( F ∘ γ − F ) = 0 d(F\circ\gamma-F)=0 d ( F ∘ γ − F ) = 0 , hence F ∘ γ = F + c γ F\circ\gamma=F+c_\gamma F ∘ γ = F + c γ . It commutes with the lifted additive action: both lift the same action and agree at t = 0 t=0 t = 0 . The claimed product form follows. If the original flow has global period 2 π i 2\pi i 2 π i , then φ ~ 2 π i \widetilde\varphi_{2\pi i} φ 2 π i is a deck transformation, and its F F F -increment is 2 π i 2\pi i 2 π i , so it is nontrivial. It commutes with every deck transformation because it is part of the lifted flow. ◻
The preceding splitting has the following useful exact-time consequence.
Corollary 7.2 (Exact-time splitting). Let M M M be a connected complex manifold, X X X a complete holomorphic vector field, and T ∈ O ( M ) T\in\mathcal O(M) T ∈ O ( M ) satisfy X ( T ) = 1 X(T)=1 X ( T ) = 1 . Then the hypersurface L = T − 1 ( 0 ) L=T^{-1}(0) L = T − 1 ( 0 ) is connected and C × L ⟶ M , ( t , p ) ⟼ φ t ( p ) \mathbb C\times L\longrightarrow M,\qquad (t,p)\longmapsto\varphi_t(p) C × L ⟶ M , ( t , p ) ⟼ φ t ( p ) is a biholomorphism. In these coordinates X = ∂ t X=\partial_t X = ∂ t . In particular, if M M M is Stein, H 1 ( M , C ) = 0 H^1(M,\mathbb C)=0 H 1 ( M , C ) = 0 , and T X ≠ ∅ \mathcal T_X\ne\varnothing T X = ∅ , this splitting holds and every orbit is biholomorphic to C \mathbb C C .
Proof. The equation X ( T ) = 1 X(T)=1 X ( T ) = 1 and completeness yield T ( φ t ( p ) ) = T ( p ) + t T(\varphi_t(p))=T(p)+t T ( φ t ( p )) = T ( p ) + t . Thus the displayed map has holomorphic inverse q ↦ ( T ( q ) , φ − T ( q ) ( q ) ) q\mapsto\bigl(T(q),\varphi_{-T(q)}(q)\bigr) q ↦ ( T ( q ) , φ − T ( q ) ( q ) ) . Connectedness of M M M forces connectedness of L L L . In the final assertion, a normalized closed one-form has vanishing periods when H 1 ( M , C ) = 0 H^1(M,\mathbb C)=0 H 1 ( M , C ) = 0 , hence has a holomorphic primitive T T T . ◻
Sharpness of the hypotheses
The examples below indicate which assumptions in the uniformization statement cannot be omitted.
Example 7.3 (Steinness cannot be omitted). The scalar action on C 2 ∖ { 0 } \mathbb C^2\setminus\{0\} C 2 ∖ { 0 } is nonsingular, while H 1 ( C 2 ∖ { 0 } , C ) = 0 H^1(\mathbb C^2\setminus\{0\},\mathbb C)=0 H 1 ( C 2 ∖ { 0 } , C ) = 0 because this punctured space retracts to S 3 S^3 S 3 . The manifold is not Stein; thus it does not contradict Proposition 6.1 .
Example 7.4 (The surface restriction is essential). The free right action A ↦ A diag ( s , s − 1 ) A\mapsto A\operatorname{diag}(s,s^{-1}) A ↦ A diag ( s , s − 1 ) on the affine Stein threefold S L 2 ( C ) SL_2(\mathbb C) S L 2 ( C ) has no fixed points, but H 1 ( S L 2 ( C ) , C ) = 0 H^1(SL_2(\mathbb C),\mathbb C)=0 H 1 ( S L 2 ( C ) , C ) = 0 since S L 2 ( C ) SL_2(\mathbb C) S L 2 ( C ) retracts to S U ( 2 ) ≃ S 3 SU(2)\simeq S^3 S U ( 2 ) ≃ S 3 . Therefore the degree-one obstruction is special to the surface or codimension-one torus settings.
Discrete periods, suspensions, and fiber bundles
We next make explicit the suspension geometry associated with a closed normalized time form.
Theorem 8.1 (Suspension principle). Let M M M be connected and let X X X be a complete holomorphic vector field admitting η ∈ T X \eta\in\mathcal T_X η ∈ T X . Suppose Λ = Per ( η ) \Lambda=\operatorname{Per}(\eta) Λ = Per ( η ) is discrete. Then there exist a connected complex manifold L L L and a homomorphism ρ : Λ → Aut ( L ) \rho:\Lambda\to\operatorname{Aut}(L) ρ : Λ → Aut ( L ) such that M ≅ ( L × C ) / Λ , λ ⋅ ( p , t ) = ( ρ ( λ ) p , t + λ ) . M\cong (L\times\mathbb C)/\Lambda,
\qquad \lambda\cdot(p,t)=\big(\rho(\lambda)p,t+\lambda\big). M ≅ ( L × C ) /Λ , λ ⋅ ( p , t ) = ( ρ ( λ ) p , t + λ ) . The flow of X X X is induced by translation in t t t , and η \eta η is induced by d t dt d t . The period integral is a surjective, locally holomorphically trivial holomorphic map F : M ⟶ C / Λ F:M\longrightarrow\mathbb C/\Lambda F : M ⟶ C /Λ with connected fibers, precisely the leaves of ker η \ker\eta ker η .
Proof. Take the connected regular covering π : M ^ → M \pi:\widehat M\to M π : M → M corresponding to the kernel of the period homomorphism π 1 ( M ) → Λ \pi_1(M)\to\Lambda π 1 ( M ) → Λ . The pulled-back form is exact, π ∗ η = d T \pi^*\eta=dT π ∗ η = d T . The lifted vector field X ^ \widehat X X is complete and satisfies X ^ ( T ) = 1 \widehat X(T)=1 X ( T ) = 1 . Hence T ( Φ ^ t ( p ) ) = T ( p ) + t T(\widehat\Phi_t(p))=T(p)+t T ( Φ t ( p )) = T ( p ) + t . The fiber L = T − 1 ( 0 ) L=T^{-1}(0) L = T − 1 ( 0 ) is a complex submanifold and
L × C → M ^ , ( p , t ) ↦ Φ ^ t ( p ) L\times\mathbb C\to\widehat M,\quad (p,t)\mapsto\widehat\Phi_t(p) L × C → M , ( p , t ) ↦ Φ t ( p ) is a biholomorphism, with inverse q ↦ ( Φ ^ − T ( q ) q , T ( q ) ) q\mapsto(\widehat\Phi_{-T(q)}q,T(q)) q ↦ ( Φ − T ( q ) q , T ( q )) . In particular L L L is connected: the period covering M ^ \widehat M M is connected by construction and is biholomorphic to L × C L\times\mathbb C L × C . Thus connectedness of the fibers follows from the chosen period covering, not merely from connectedness of the original manifold. Every deck transformation commutes with Φ ^ \widehat\Phi Φ and changes T T T by λ ∈ Λ \lambda\in\Lambda λ ∈ Λ , so it acts as ( ρ ( λ ) p , t + λ ) \big(\rho(\lambda)p,t+\lambda\big) ( ρ ( λ ) p , t + λ ) , with ρ ( λ ) ∈ Aut ( L ) \rho(\lambda)\in\operatorname{Aut}(L) ρ ( λ ) ∈ Aut ( L ) . The quotient defines the asserted bundle, and fibers identify with L L L . ◻
We may now characterize closed leaves by the discreteness of the period group.
Theorem 8.3 (Global stability of invariant closed-form foliations). Let M M M be a connected complex manifold and X X X a complete holomorphic vector field. Let ω \omega ω be a closed holomorphic one-form with ω ( X ) = 1 \omega(X)=1 ω ( X ) = 1 , and put F = ker ω \mathcal F=\ker\omega F = ker ω and Λ = Per ( ω ) \Lambda=\operatorname{Per}(\omega) Λ = Per ( ω ) . The following assertions are equivalent:
F \mathcal F F has at least one closed leaf;
all leaves of F \mathcal F F are closed in M M M ;
Λ \Lambda Λ is a discrete subgroup of C \mathbb C C ;
integration of ω \omega ω induces a holomorphic fiber bundle F : M → C / Λ F:M\to\mathbb C/\Lambda F : M → C /Λ with connected fibers equal to the leaves of F \mathcal F F and F ∗ ( d z ) = ω F^*(dz)=\omega F ∗ ( d z ) = ω .
Every leaf of F \mathcal F F has trivial holonomy, independently of whether these conditions hold. When they do, the base is C \mathbb C C , C ∗ \mathbb C^* C ∗ , or an elliptic curve according as Λ \Lambda Λ has rank zero, one, or two. No Stein assumption is needed.
Proof. Since ω \omega ω is closed, its local holomorphic primitives give transverse coordinates for F \mathcal F F and differ on overlaps by additive constants. Holonomy acts in such coordinates by translations; as each holonomy germ fixes its basepoint, it is the identity.
The implications (iv)⇒ \Rightarrow ⇒ (ii)⇒ \Rightarrow ⇒ (i) are clear, and (iii)⇒ \Rightarrow ⇒ (iv) is the suspension principle, Theorem 8.1 . To prove the remaining implication, let L L L be a closed leaf. The period-cover construction, used in the proof of Theorem 8.1 , applies as an abstract cover even when Λ \Lambda Λ is not discrete as a subset of C \mathbb C C . It gives, for every p ∈ L p\in L p ∈ L ,
{ t ∈ C : φ t ( p ) ∈ L } = Λ . \{t\in\mathbb C:\varphi_t(p)\in L\}=\Lambda. { t ∈ C : φ t ( p ) ∈ L } = Λ. The left-hand side is closed because L L L is closed and the orbit map is continuous; thus Λ \Lambda Λ is closed in C \mathbb C C . As M M M is second-countable, its fundamental group is countable, so the image Λ \Lambda Λ of its period homomorphism is countable. A closed countable subgroup of C \mathbb C C is discrete: otherwise its identity component or its accumulation at the origin would, by the classification of closed subgroups of R 2 \mathbb R^2 R 2 , force an uncountable subgroup. This proves (i)⇒ \Rightarrow ⇒ (iii). ◻
Discrete periods do not replace completeness in the above result:
Example 8.5. Let M = D × C M=\mathbb D\times\mathbb C M = D × C , X = ∂ z X=\partial_z X = ∂ z for the disk coordinate z z z , and ω = d z \omega=dz ω = d z . Then d ω = 0 d\omega=0 d ω = 0 , ω ( X ) = 1 \omega(X)=1 ω ( X ) = 1 , and Per ( ω ) = { 0 } \operatorname{Per}(\omega)=\{0\} Per ( ω ) = { 0 } . The field is not complete on M M M , and integration gives the non-surjective submersion M → C M\to\mathbb C M → C , ( z , w ) ↦ z (z,w)\mapsto z ( z , w ) ↦ z . Hence the bundle conclusion of Theorem 8.1 requires completeness.
Normalized holomorphic Tischler approximation for C ∗ \mathbb C^* C ∗ actions
We return to periodic multiplicative actions and derive the corresponding approximation theorem in the smooth setting.
Let X X X generate a holomorphic C ∗ \mathbb C^* C ∗ -action ψ \psi ψ , with the complex-time flow Φ t = ψ e t \Phi_t=\psi_{e^t} Φ t = ψ e t . Let γ p \gamma_p γ p be the orbit circle θ ↦ ψ e i θ ( p ) \theta\mapsto\psi_{e^{i\theta}}(p) θ ↦ ψ e i θ ( p ) , 0 ≤ θ ≤ 2 π 0\le\theta\le2\pi 0 ≤ θ ≤ 2 π . Its homology class γ ∈ H 1 ( M , Z ) \gamma\in H_1(M,\mathbb Z) γ ∈ H 1 ( M , Z ) does not depend on p p p when M M M is connected. (A nontrivial finite isotropy group is permitted as long as X X X is nowhere zero.)
Theorem 9.1 (Affine hyperplane of normalized classes). Let M M M be a connected Stein manifold and let X X X be a nowhere-vanishing complete holomorphic vector field with φ 2 π i = id M \varphi_{2\pi i}=\operatorname{id}_M φ 2 π i = id M (equivalently, the infinitesimal generator of a fixed-point-free holomorphic C ∗ \mathbb C^* C ∗ -action). Let γ ∈ H 1 ( M , Z ) \gamma\in H_1(M,\mathbb Z) γ ∈ H 1 ( M , Z ) be the common homology class of the orbit circles θ ↦ e i θ ⋅ p \theta\mapsto e^{i\theta}\cdot p θ ↦ e i θ ⋅ p . The cohomology classes of closed holomorphic forms η \eta η with η ( X ) = 1 \eta(X)=1 η ( X ) = 1 are exactly { ξ ∈ H 1 ( M , C ) : ⟨ ξ , γ ⟩ = 2 π i } . \boxed{\;\{\xi\in H^1(M,\mathbb C):\ \langle\xi,\gamma\rangle=2\pi i\}\;.} { ξ ∈ H 1 ( M , C ) : ⟨ ξ , γ ⟩ = 2 π i } . Every class in the displayed set is realized by a closed holomorphic 1 1 1 -form η \eta η satisfying η ( X ) = 1 \eta(X)=1 η ( X ) = 1 . The set is empty if the image of γ \gamma γ in H 1 ( M , C ) H_1(M,\mathbb C) H 1 ( M , C ) is zero; otherwise it is an affine hyperplane.
Proof. Necessity follows from ∫ γ p η = ∫ 0 2 π i η ( X ) d θ = 2 π i \int_{\gamma_p}\eta=\int_0^{2\pi}i\,\eta(X)\,d\theta=2\pi i ∫ γ p η = ∫ 0 2 π i η ( X ) d θ = 2 π i . Conversely, represent any such class by a global closed holomorphic 1 1 1 -form α \alpha α (holomorphic de Rham theorem on Stein manifolds), and average over the compact circle:
α ˉ = 1 2 π ∫ 0 2 π ψ e i θ ∗ α d θ . \bar\alpha=\frac1{2\pi}\int_0^{2\pi}\psi_{e^{i\theta}}^*\alpha\,d\theta. α ˉ = 2 π 1 ∫ 0 2 π ψ e i θ ∗ α d θ . The average is holomorphic, closed, and preserves the cohomology class because each circle element acts by a map homotopic to the identity. The averaged form is S 1 S^1 S 1 -invariant, hence L X α ˉ = 0 \mathcal L_X\bar\alpha=0 L X α ˉ = 0 by holomorphic continuation. Cartan’s formula gives d ( α ˉ ( X ) ) = 0 d(\bar\alpha(X))=0 d ( α ˉ ( X )) = 0 , so α ˉ ( X ) = c \bar\alpha(X)=c α ˉ ( X ) = c is constant. Since 2 π i c = ∫ γ α ˉ = 2 π i 2\pi i c=\int_\gamma\bar\alpha=2\pi i 2 π i c = ∫ γ α ˉ = 2 π i , we have c = 1 c=1 c = 1 . ◻
The affine description of normalized classes leads to the smooth approximation theorem.
Theorem 9.3 (Normalized holomorphic Tischler approximation). Let M M M be a connected Stein manifold, let X X X be the nowhere-vanishing generator of a holomorphic C ∗ \mathbb C^* C ∗ -action, and assume that H 1 ( M , Z ) H_1(M,\mathbb Z) H 1 ( M , Z ) is finitely generated. Every ω ∈ T X \omega\in\mathcal T_X ω ∈ T X is a compact-open limit of closed holomorphic ω j ∈ T X \omega_j\in\mathcal T_X ω j ∈ T X with Per ( ω j ) \operatorname{Per}(\omega_j) Per ( ω j ) discrete. If b 1 ( M ) ≥ 2 b_1(M)\ge2 b 1 ( M ) ≥ 2 , the approximations can have rank-two lattice period groups; hence they define bundles over elliptic curves. All resulting foliations are invariant under the same complete holomorphic flow.
Proof. Fix ω ∈ T X \omega\in\mathcal T_X ω ∈ T X and put H 1 ( M , Z ) / t o r s ≃ Z r H_1(M,\mathbb Z)/\mathrm{tors}\simeq\mathbb Z^r H 1 ( M , Z ) / tors ≃ Z r . As ∫ γ ω = 2 π i \int_\gamma\omega=2\pi i ∫ γ ω = 2 π i , the orbit class γ \gamma γ has infinite order, so r ≥ 1 r\ge1 r ≥ 1 . By Smith normal form, choose an integral basis e 1 , … , e r e_1,\ldots,e_r e 1 , … , e r with [ γ ] = d e 1 [\gamma]=d e_1 [ γ ] = d e 1 modulo torsion, where d ≥ 1 d\ge1 d ≥ 1 . Under the period description of H 1 ( M , C ) H^1(M,\mathbb C) H 1 ( M , C ) , the normalization hyperplane of Theorem 9.1 is exactly
ξ ( e 1 ) = 2 π i / d , ξ ( e j ) ∈ C ( 2 ≤ j ≤ r ) . \xi(e_1)=2\pi i/d,\qquad \xi(e_j)\in\mathbb C\quad (2\le j\le r). ξ ( e 1 ) = 2 π i / d , ξ ( e j ) ∈ C ( 2 ≤ j ≤ r ) . Approximate each of the remaining r − 1 r-1 r − 1 coordinates of [ ω ] [\omega] [ ω ] by elements of 2 π i ( Q + i Q ) 2\pi i(\mathbb Q+i\mathbb Q) 2 π i ( Q + i Q ) , keeping ξ ( e 1 ) = 2 π i / d \xi(e_1)=2\pi i/d ξ ( e 1 ) = 2 π i / d fixed. Call the resulting characters ξ n \xi_n ξ n . The period values of any given ξ n \xi_n ξ n lie in 2 π i N n Z [ i ] \frac{2\pi i}{N_n}\mathbb Z[i] N n 2 π i Z [ i ] for a suitable integer N n ≥ 1 N_n\ge1 N n ≥ 1 ; hence their image is a discrete subgroup of C \mathbb C C . If r ≥ 2 r\ge2 r ≥ 2 , we can arrange, however close the approximation, that ξ n ( e 2 ) \xi_n(e_2) ξ n ( e 2 ) has nonzero real part. Together with the fixed nonzero purely imaginary value of ξ n ( e 1 ) \xi_n(e_1) ξ n ( e 1 ) , the image then has real span of dimension two and is a rank-two lattice. If r = 1 r=1 r = 1 , normalization forces the single free period value; it is already discrete.
Choose closed global holomorphic forms θ 1 , … , θ r \theta_1,\ldots,\theta_r θ 1 , … , θ r representing a cohomology basis, as permitted by holomorphic de Rham theory on Stein manifolds. Because ξ n − [ ω ] → 0 \xi_n-[\omega]\to0 ξ n − [ ω ] → 0 , suitable linear combinations of the θ j \theta_j θ j provide closed forms α n → ω \alpha_n\to\omega α n → ω uniformly on compact subsets with [ α n ] = ξ n [\alpha_n]=\xi_n [ α n ] = ξ n . Now average α n \alpha_n α n over S 1 S^1 S 1 :
ω n = 1 2 π ∫ 0 2 π ψ e i ϑ ∗ α n d ϑ . \omega_n=\frac1{2\pi}\int_0^{2\pi}\psi_{e^{i\vartheta}}^*\alpha_n\,d\vartheta. ω n = 2 π 1 ∫ 0 2 π ψ e i ϑ ∗ α n d ϑ . The proof of Theorem 9.1 shows that ω n ( X ) = 1 \omega_n(X)=1 ω n ( X ) = 1 exactly and that [ ω n ] = ξ n [\omega_n]=\xi_n [ ω n ] = ξ n , so its periods are discrete (and a rank-two lattice if r ≥ 2 r\ge2 r ≥ 2 ). Since ω \omega ω is invariant and the saturation of any compact set under S 1 S^1 S 1 is compact, the averaging preserves compact-open convergence to ω \omega ω . Theorem 8.1 gives the asserted holomorphic fibrations. ◻
The first Betti number sharpens the conclusion in three natural cases.
Corollary 9.4 (The first-Betti-number trichotomy). Let M M M be a connected Stein manifold endowed with a holomorphic C ∗ \mathbb C^* C ∗ -action whose infinitesimal generator X X X is nowhere zero, and suppose that H 1 ( M , Z ) H_1(M,\mathbb Z) H 1 ( M , Z ) is finitely generated. Then:
If b 1 ( M ) = 0 b_1(M)=0 b 1 ( M ) = 0 , then T X = ∅ \mathcal T_X=\varnothing T X = ∅ .
If b 1 ( M ) = 1 b_1(M)=1 b 1 ( M ) = 1 and T X ≠ ∅ \mathcal T_X\ne\varnothing T X = ∅ , every ω ∈ T X \omega\in\mathcal T_X ω ∈ T X already has a nonzero cyclic discrete period group. Integration of ω \omega ω gives a surjective, locally holomorphically trivial holomorphic map F ω : M → C ∗ F_\omega:M\to\mathbb C^* F ω : M → C ∗ with connected fibers equal to the leaves of ker ω \ker\omega ker ω ; no approximation is required.
If b 1 ( M ) ≥ 2 b_1(M)\ge2 b 1 ( M ) ≥ 2 and T X ≠ ∅ \mathcal T_X\ne\varnothing T X = ∅ , every ω ∈ T X \omega\in\mathcal T_X ω ∈ T X is a compact-open limit of normalized closed holomorphic forms whose periods are rank-two lattices; the associated leaf foliations are holomorphic fiber bundles over elliptic curves.
In particular, the first case cannot occur for a nowhere-vanishing C ∗ \mathbb C^* C ∗ -generator on a Stein surface, because such generators have T X ≠ ∅ \mathcal T_X\ne\varnothing T X = ∅ by Corollary 9.5 .
Proof. Fix p ∈ M p\in M p ∈ M and consider the orbit circle γ ( θ ) = e i θ ⋅ p \gamma(\theta)=e^{i\theta}\cdot p γ ( θ ) = e i θ ⋅ p . If ω ∈ T X \omega\in\mathcal T_X ω ∈ T X then ∫ γ ω = 2 π i \int_\gamma\omega=2\pi i ∫ γ ω = 2 π i , whence [ γ ] [\gamma] [ γ ] is nontorsion and b 1 ( M ) ≥ 1 b_1(M)\ge1 b 1 ( M ) ≥ 1 . This proves (i).
Suppose b 1 ( M ) = 1 b_1(M)=1 b 1 ( M ) = 1 . Because the first integral homology is finitely generated, its free part is Z e \mathbb Ze Z e . Write [ γ ] = d e [\gamma]=d e [ γ ] = d e modulo torsion, with a nonzero integer d d d (choose the sign of e e e so that d > 0 d>0 d > 0 ). Normalization forces
∫ e ω = 2 π i d , Per ( ω ) = 2 π i d Z . \int_e\omega=\frac{2\pi i}{d},\qquad
\operatorname{Per}(\omega)=\frac{2\pi i}{d}\mathbb Z. ∫ e ω = d 2 π i , Per ( ω ) = d 2 π i Z . The complete-flow suspension Theorem 8.1 therefore gives a holomorphic bundle over C / ( ( 2 π i / d ) Z ) ≃ C ∗ \mathbb C/((2\pi i/d)\mathbb Z)\simeq\mathbb C^* C / (( 2 π i / d ) Z ) ≃ C ∗ with connected fibers. This proves (ii). Assertion (iii) is Theorem 9.3 . The final assertion follows from Corollary 9.5 . ◻
Proof of Corollary D . The three cases and their period assertions follow from Corollary 9.4 . When b 1 ( M ) = 1 b_1(M)=1 b 1 ( M ) = 1 , the associated discrete cyclic periods yield a bundle over C ∗ \mathbb C^* C ∗ by Theorem 8.1 ; when b 1 ( M ) ≥ 2 b_1(M)\ge2 b 1 ( M ) ≥ 2 , Theorem 9.3 provides rank-two lattice approximants whose foliations are bundles over elliptic curves, again by the suspension theorem. Finally, Theorem 9.1 identifies the cohomology classes of normalized forms with the stated affine hyperplane. ◻
In complex dimension two, a fixed-point-free multiplicative action also supplies the required closed time form.
Corollary 9.5 (Automatic normalization on Stein surfaces). If M M M is a Stein surface and X X X is a nowhere-vanishing generator of a holomorphic C ∗ \mathbb C^* C ∗ -action, there exists ω ∈ T X \omega\in\mathcal T_X ω ∈ T X . If H 1 ( M , Z ) H_1(M,\mathbb Z) H 1 ( M , Z ) is finitely generated, then M M M admits a holomorphic fiber bundle over C ∗ \mathbb C^* C ∗ or an elliptic curve whose fibers are connected, Stein, and transverse to X X X ; when b 1 ( M ) ≥ 2 b_1(M)\ge2 b 1 ( M ) ≥ 2 , the base can be elliptic.
Proof. Contraction with X X X defines a surjection of locally free sheaves Ω M 1 → O M \Omega_M^1\to\mathcal O_M Ω M 1 → O M . Cartan B lifts 1 1 1 to a global holomorphic form α \alpha α with α ( X ) = 1 \alpha(X)=1 α ( X ) = 1 . Average α \alpha α over S 1 S^1 S 1 to obtain invariant α ˉ \bar\alpha α ˉ with α ˉ ( X ) = 1 \bar\alpha(X)=1 α ˉ ( X ) = 1 . Then ι X d α ˉ = L X α ˉ − d ( 1 ) = 0 \iota_Xd\bar\alpha=\mathcal L_X\bar\alpha-d(1)=0 ι X d α ˉ = L X α ˉ − d ( 1 ) = 0 . On a complex surface, contraction of a holomorphic 2 2 2 -form by a nowhere-zero vector field is injective, so d α ˉ = 0 d\bar\alpha=0 d α ˉ = 0 . Apply Theorems 9.3 and 8.1 . A closed complex submanifold of Stein M M M is Stein. ◻
Proof of Theorem E . By Proposition 6.1 , the fixed-point-free holomorphic C ∗ \mathbb C^* C ∗ -action on the Stein surface admits a closed holomorphic one-form ω \omega ω normalized by X X X , and its orbit circle has nonzero period. Thus H 1 ( M , C ) ≠ 0 H^1(M,\mathbb C)\ne0 H 1 ( M , C ) = 0 and b 1 ( M ) ≥ 1 b_1(M)\ge1 b 1 ( M ) ≥ 1 . Theorem 7.1 proves the claimed universal-cover splitting and translation form of the lifted flow. If H 1 ( M , Z ) H_1(M,\mathbb Z) H 1 ( M , Z ) is finitely generated, apply Corollary 9.4 : for b 1 ( M ) = 1 b_1(M)=1 b 1 ( M ) = 1 the normalized form already has cyclic discrete periods, while for b 1 ( M ) ≥ 2 b_1(M)\ge2 b 1 ( M ) ≥ 2 we choose one of the rank-two lattice approximants. Corollary 2.4 (or, equivalently, Theorem 8.1 in the smooth case) gives holomorphic locally trivial bundles with connected Stein fibers and the respective bases C ∗ \mathbb C^* C ∗ and an elliptic curve. As M M M is a surface, each fiber has complex dimension one. ◻
Example 9.7 (Betti-number b 1 = 0 b_1=0 b 1 = 0 : a free action with no closed time form). Let M = S L 2 ( C ) M=SL_2(\mathbb C) M = S L 2 ( C ) , the affine Stein threefold { A ∈ Mat 2 × 2 ( C ) : det A = 1 } \{A\in\operatorname{Mat}_{2\times2}(\mathbb C):\det A=1\} { A ∈ Mat 2 × 2 ( C ) : det A = 1 } , and let
s ⋅ A = A ( s 0 0 s − 1 ) , s ∈ C ∗ . s\cdot A=A\begin{pmatrix}s&0\\0&s^{-1}\end{pmatrix},
\qquad s\in\mathbb C^*. s ⋅ A = A ( s 0 0 s − 1 ) , s ∈ C ∗ . This action is free, and its nowhere-zero generator is X ( A ) = A H X(A)=AH X ( A ) = A H , where H = diag ( 1 , − 1 ) H=\operatorname{diag}(1,-1) H = diag ( 1 , − 1 ) . The manifold retracts onto S U ( 2 ) ≃ S 3 SU(2)\simeq S^3 S U ( 2 ) ≃ S 3 , hence b 1 ( M ) = 0 b_1(M)=0 b 1 ( M ) = 0 . Consequently T X = ∅ \mathcal T_X=\varnothing T X = ∅ by Corollary 9.4 : in fact, if a closed normalized form η \eta η existed, its integral along the orbit circle would be 2 π i 2\pi i 2 π i even though H 1 ( M , C ) = 0 H^1(M,\mathbb C)=0 H 1 ( M , C ) = 0 .
The obstruction is specifically to closed normalization: letting ϑ = A − 1 d A \vartheta=A^{-1}dA ϑ = A − 1 d A be the holomorphic Maurer–Cartan form,
α = 1 2 tr ( H ϑ ) \alpha=\tfrac12\operatorname{tr}(H\vartheta) α = 2 1 tr ( H ϑ ) is a global holomorphic form with α ( X ) = 1 \alpha(X)=1 α ( X ) = 1 . It is not closed. Indeed, if E = ( 0 1 0 0 ) E=\left(\begin{smallmatrix}0&1\\0&0\end{smallmatrix}\right) E = ( 0 0 1 0 ) and F = ( 0 0 1 0 ) F=\left(\begin{smallmatrix}0&0\\1&0\end{smallmatrix}\right) F = ( 0 1 0 0 ) , then [ E , F ] = H [E,F]=H [ E , F ] = H and the Maurer–Cartan formula at the identity gives d α ( E , F ) = − α ( H ) = − 1 d\alpha(E,F)=-\alpha(H)=-1 d α ( E , F ) = − α ( H ) = − 1 . Thus the b 1 = 0 b_1=0 b 1 = 0 case is genuinely realized by a nonsingular C ∗ \mathbb C^* C ∗ -action on a Stein manifold, but not by a Stein surface.
Example 9.8 (First Betti number one: an existing fibration). Take M = C ∗ × C M=\mathbb C^*\times\mathbb C M = C ∗ × C with coordinates ( z , w ) (z,w) ( z , w ) and the free action s ⋅ ( z , w ) = ( s z , w ) s\cdot(z,w)=(sz,w) s ⋅ ( z , w ) = ( sz , w ) . Then b 1 ( M ) = 1 b_1(M)=1 b 1 ( M ) = 1 and
X = z ∂ z , ω = d z z , Per ( ω ) = 2 π i Z . X=z\partial_z,\qquad \omega=\frac{dz}{z},\qquad
\operatorname{Per}(\omega)=2\pi i\mathbb Z. X = z ∂ z , ω = z d z , Per ( ω ) = 2 π i Z . The form is holomorphic, closed, and normalized by X X X . The associated holomorphic fibration is already the projection
F : M ⟶ C ∗ , F ( z , w ) = z , F:M\longrightarrow\mathbb C^*,\qquad F(z,w)=z, F : M ⟶ C ∗ , F ( z , w ) = z , whose connected fibers are the copies { z } × C \{z\}\times\mathbb C { z } × C , precisely the leaves of ker ω \ker\omega ker ω . No perturbation is necessary.
Example 9.9 (First Betti number two: elliptic approximation of a nonclosed foliation). On M = ( C ∗ ) 2 M=(\mathbb C^*)^2 M = ( C ∗ ) 2 , where b 1 ( M ) = 2 b_1(M)=2 b 1 ( M ) = 2 , take the free action s ⋅ ( z , w ) = ( s z , w ) s\cdot(z,w)=(sz,w) s ⋅ ( z , w ) = ( sz , w ) , whose generator is X = z ∂ z X=z\partial_z X = z ∂ z , and put
ω = d z z + 2 d w w , ω n = d z z + ( 2 + i n ) d w w ( n ≥ 1 ) . \omega=\frac{dz}{z}+\sqrt2\frac{dw}{w},\qquad
\omega_n=\frac{dz}{z}+\left(\sqrt2+\frac{i}{n}\right)\frac{dw}{w}\quad(n\ge1). ω = z d z + 2 w d w , ω n = z d z + ( 2 + n i ) w d w ( n ≥ 1 ) . All these forms are closed and normalized by X X X , and ω n → ω \omega_n\to\omega ω n → ω compact-open. The initial period group 2 π i ( Z + 2 Z ) 2\pi i(\mathbb Z+\sqrt2\mathbb Z) 2 π i ( Z + 2 Z ) is dense in i R i\mathbb R i R , whereas the periods of ω n \omega_n ω n form the rank-two lattice
2 π i Z + 2 π i ( 2 + i n ) Z . 2\pi i\mathbb Z+2\pi i\left(\sqrt2+\frac{i}{n}\right)\mathbb Z. 2 π i Z + 2 π i ( 2 + n i ) Z . Thus every ω n \omega_n ω n defines a holomorphic bundle over an elliptic curve with fibers transverse to the same complete field X X X , while the original invariant foliation has nonclosed leaves.
The closures of invariant holomorphic leaves
Period groups also govern the topology of leaf closures for invariant holomorphic foliations.
Theorem 10.1 (Leaf-closure classification). Let M M M be connected, let X X X be complete, and let ω ∈ T X \omega\in\mathcal T_X ω ∈ T X . Write Λ = Per ( ω ) \Lambda=\operatorname{Per}(\omega) Λ = Per ( ω ) and H = Λ ‾ ⊂ C H=\overline\Lambda\subset\mathbb C H = Λ ⊂ C . There is a continuous surjection Q : M → C / H Q:M\to\mathbb C/H Q : M → C / H such that, for every leaf L L L of ker ω \ker\omega ker ω , L ‾ = Q − 1 ( Q ( L ) ) . \boxed{\overline L=Q^{-1}(Q(L)).} L = Q − 1 ( Q ( L )) . Exactly one of these alternatives holds:
H H H is discrete: every leaf is closed, and there is a holomorphic fiber bundle M → C / H M\to\mathbb C/H M → C / H whose connected fibers are leaves.
H H H is proper and nondiscrete: every leaf is dense in a proper, closed, real-analytic Levi-flat hypersurface. The hypersurfaces are fibers of a real-analytic submersion onto R \mathbb R R or S 1 S^1 S 1 .
H = C H=\mathbb C H = C : every leaf is dense in M M M .
Proof. On the period covering, use the splitting M ^ ≃ L 0 × C \widehat M\simeq L_0\times\mathbb C M ≃ L 0 × C from the proof of Theorem 8.1 , which does not require the period image to be discrete as a subset of C \mathbb C C : the deck group is an abstract quotient of π 1 ( M ) \pi_1(M) π 1 ( M ) acting properly on M ^ \widehat M M , and acts by translation by λ ∈ Λ \lambda\in\Lambda λ ∈ Λ on the time coordinate. Thus Q ( [ p , t ] ) = t ( m o d H ) Q([p,t])=t\pmod H Q ([ p , t ]) = t ( mod H ) is well defined, continuous, and surjective.
For a fixed leaf L L L of the foliation, λ ∈ Λ \lambda\in\Lambda λ ∈ Λ precisely when Φ λ ( L ) = L \Phi_\lambda(L)=L Φ λ ( L ) = L (use the product cover). If h ∈ H h\in H h ∈ H , take λ n ∈ Λ \lambda_n\in\Lambda λ n ∈ Λ with λ n → h \lambda_n\to h λ n → h ; then Φ λ n ( p ) ∈ L \Phi_{\lambda_n}(p)\in L Φ λ n ( p ) ∈ L tends to Φ h ( p ) \Phi_h(p) Φ h ( p ) for each p ∈ L p\in L p ∈ L . The flow saturates M M M by translates of L L L , and therefore the entire fiber Q − 1 ( Q ( L ) ) Q^{-1}(Q(L)) Q − 1 ( Q ( L )) lies in L ‾ \overline L L . Conversely Q Q Q is constant on L L L and has closed point fibers since H H H is closed, giving the reverse inclusion.
A proper nondiscrete closed subgroup of the real plane C \mathbb C C is either a real line or a real line plus a transverse infinite cyclic subgroup. Its quotient is respectively R \mathbb R R or S 1 S^1 S 1 . Locally Q Q Q is the real part of a complex-linear multiple of a holomorphic time coordinate, modulo a real period, hence is a real-analytic submersion with Levi-flat level hypersurfaces. ◻
Proof of Theorem 1.3 . The transverse splitting T M = T F ⊕ C X TM=T\mathcal F\oplus\mathbb CX T M = T F ⊕ C X defines a unique holomorphic one-form ω \omega ω with ker ω = T F \ker\omega=T\mathcal F ker ω = T F and ω ( X ) = 1 \omega(X)=1 ω ( X ) = 1 . Invariance gives L X ω = 0 \mathcal L_X\omega=0 L X ω = 0 , hence ι X d ω = 0 \iota_Xd\omega=0 ι X d ω = 0 by Cartan’s formula. Integrability gives d ω ( Y , Z ) = − ω ( [ Y , Z ] ) = 0 d\omega(Y,Z)=-\omega([Y,Z])=0 d ω ( Y , Z ) = − ω ([ Y , Z ]) = 0 for sections Y , Z Y,Z Y , Z of T F T\mathcal F T F . These identities imply d ω = 0 d\omega=0 d ω = 0 . Thus the hypotheses of Theorem 10.1 hold. Apply the period-cover construction and the classification of closed subgroups of C \mathbb C C in that theorem. Discrete closure of the period group gives closed leaves; a proper nondiscrete closure gives real-analytic Levi-flat leaf closures; and H = C H=\mathbb C H = C gives dense leaves. None of these steps requires Steinness. ◻
The same period-cover method extends the leaf-closure description to normal complex spaces.
Corollary 10.2 (Leaf closures on normal complex spaces). Let S S S be a connected normal complex analytic space satisfying the degree-one reflexive Poincaré property. Let X X X be a complete holomorphic vector field on S S S , with holomorphic flow Φ : C × S → S \Phi:\mathbb C\times S\to S Φ : C × S → S , and let ω ∈ H 0 ( S , Z S 1 ) \omega\in H^0(S,\mathcal Z_S^1) ω ∈ H 0 ( S , Z S 1 ) be a closed reflexive holomorphic one-form satisfying ω ( X ) = 1 \omega(X)=1 ω ( X ) = 1 on S r e g S_{\rm reg} S reg . Write F r e g = ker ( ω ∣ S r e g ) \mathcal F_{\rm reg}=\ker(\omega|_{S_{\rm reg}}) F reg = ker ( ω ∣ S reg ) , let Λ = Per S ( ω ) \Lambda=\operatorname{Per}_S(\omega) Λ = Per S ( ω ) be the image of the period character defined by local primitives on S S S , and set H = Λ ‾ ⊂ C H=\overline\Lambda\subset\mathbb C H = Λ ⊂ C . Then there is a continuous surjection Q : S ⟶ C / H Q:S\longrightarrow\mathbb C/H Q : S ⟶ C / H such that, for every leaf L L L of F r e g \mathcal F_{\rm reg} F reg , L ‾ S = Q − 1 ( Q ( L ) ) . \boxed{\overline L^{\,S}=Q^{-1}(Q(L)).} L S = Q − 1 ( Q ( L )) . More precisely:
If H H H is discrete, Q Q Q is a holomorphic locally trivial fiber bundle with connected normal analytic fibers. Each fiber is the closure in S S S of one regular leaf, and its regular locus is that leaf.
If H H H is proper and nondiscrete, C / H \mathbb C/H C / H is isomorphic as a real-analytic manifold to R \mathbb R R or S 1 S^1 S 1 . Locally in S S S , the map Q Q Q is induced by a real-linear functional of a holomorphic time coordinate. Its fibers are closed real-analytic subsets of real codimension one, whose intersections with S r e g S_{\rm reg} S reg are Levi-flat real-analytic hypersurfaces; the fibers may contain singular points of S S S .
If H = C H=\mathbb C H = C , every leaf of F r e g \mathcal F_{\rm reg} F reg is dense in S S S .
Neither Steinness nor finite generation of H 1 ( S , Z ) H_1(S,\mathbb Z) H 1 ( S , Z ) is assumed.
Proof. The local primitive property gives a period homomorphism π 1 ( S ) → C \pi_1(S)\to\mathbb C π 1 ( S ) → C with image Λ \Lambda Λ . On the connected covering S ^ → S \widehat S\to S S → S corresponding to its kernel, the pullback of ω \omega ω has a global holomorphic primitive T : S ^ → C T:\widehat S\to\mathbb C T : S → C . Normalization and completeness lift to T ( Φ ^ t ( q ) ) = T ( q ) + t T(\widehat\Phi_t(q))=T(q)+t T ( Φ t ( q )) = T ( q ) + t . As in the suspension argument,
S ^ ≃ N × C , N = T − 1 ( 0 ) , \widehat S\simeq N\times\mathbb C,\qquad N=T^{-1}(0), S ≃ N × C , N = T − 1 ( 0 ) , by ( n , t ) ↦ Φ ^ t ( n ) (n,t)\mapsto\widehat\Phi_t(n) ( n , t ) ↦ Φ t ( n ) . The space N N N is connected and normal: the covering is connected and normal, and the product isomorphism implies both assertions for N N N . In particular, N r e g N_{\rm reg} N reg is connected and dense in N N N .
A deck transformation indexed by λ ∈ Λ \lambda\in\Lambda λ ∈ Λ acts in these coordinates as
( n , t ) ⟼ ( ρ λ ( n ) , t + λ ) , (n,t)\longmapsto(\rho_\lambda(n),t+\lambda), ( n , t ) ⟼ ( ρ λ ( n ) , t + λ ) , where ρ λ \rho_\lambda ρ λ is a biholomorphism of N N N . Thus Q ( [ n , t ] ) = t + H Q([n,t])=t+H Q ([ n , t ]) = t + H is well defined, continuous and surjective. The regular leaves are the images of slices N r e g × { t } N_{\rm reg}\times\{t\} N reg × { t } : they are connected because N r e g N_{\rm reg} N reg is connected, and the product decomposition identifies the regular locus of S ^ \widehat S S with N r e g × C N_{\rm reg}\times\mathbb C N reg × C .
Fix such a leaf L L L and a point [ n , t + h ] ∈ Q − 1 ( Q ( L ) ) [n,t+h]\in Q^{-1}(Q(L)) [ n , t + h ] ∈ Q − 1 ( Q ( L )) , where h ∈ H h\in H h ∈ H and t t t is the slice parameter of L L L . Choose λ j ∈ Λ \lambda_j\in\Lambda λ j ∈ Λ with λ j → h \lambda_j\to h λ j → h . Since the slice N r e g × { t + λ j } N_{\rm reg}\times\{t+\lambda_j\} N reg × { t + λ j } projects onto the same leaf L L L , and N r e g N_{\rm reg} N reg is dense in N N N , one can choose n j ∈ N r e g n_j\in N_{\rm reg} n j ∈ N reg tending to n n n . The projected points [ n j , t + λ j ] ∈ L [n_j,t+\lambda_j]\in L [ n j , t + λ j ] ∈ L converge to [ n , t + h ] [n,t+h] [ n , t + h ] . Hence Q − 1 ( Q ( L ) ) ⊆ L ‾ S Q^{-1}(Q(L))\subseteq\overline L^{\,S} Q − 1 ( Q ( L )) ⊆ L S . The reverse inclusion follows because Q Q Q is continuous, H H H is closed and Q Q Q is constant on L L L .
If H H H is discrete, the local primitives of ω \omega ω glue modulo H = Λ H=\Lambda H = Λ to a holomorphic map Q : S → C / Λ Q:S\to\mathbb C/\Lambda Q : S → C /Λ . The product decomposition and local sections of C → C / Λ \mathbb C\to\mathbb C/\Lambda C → C /Λ make it a locally trivial holomorphic bundle. The fibers identify with N N N , so are connected and normal, and their regular loci are precisely the regular leaves.
If H H H is proper and nondiscrete, classification of closed subgroups of C ≃ R 2 \mathbb C\simeq\mathbb R^2 C ≃ R 2 gives C / H ≃ R \mathbb C/H\simeq\mathbb R C / H ≃ R or S 1 S^1 S 1 . In a local holomorphic time coordinate f f f provided by a primitive of ω \omega ω , the quotient map is locally represented by a nonzero real-linear functional of f f f (possibly modulo a real period). Thus its level sets are locally products of a normal analytic space with a real line in the time disc. On the regular locus they are real-analytic Levi-flat hypersurfaces, while in S S S they are real-analytic subsets of real codimension one. The case H = C H=\mathbb C H = C follows immediately from the closure identity. ◻
As a consequence, the first Betti number restricts the possible leaf-closure behavior.
Corollary 10.3. Under the hypotheses of Theorem 10.1 , if H 1 ( M , Z ) H_1(M,\mathbb Z) H 1 ( M , Z ) is finitely generated, then b 1 ( M ) ≤ 1 b_1(M)\le1 b 1 ( M ) ≤ 1 forces closed leaves; b 1 ( M ) = 2 b_1(M)=2 b 1 ( M ) = 2 permits closed or intermediate Levi-flat closures but not full density. Full density requires b 1 ( M ) ≥ 3 b_1(M)\ge3 b 1 ( M ) ≥ 3 . On a Stein manifold of complex dimension at least two, closed positive-dimensional holomorphic leaves are noncompact.
Proof. Λ \Lambda Λ is generated by at most b 1 ( M ) b_1(M) b 1 ( M ) complex numbers. An abelian subgroup of C \mathbb C C generated by at most two elements is either discrete or lies in a real line, where its closure may be nondiscrete. Density in C \mathbb C C needs at least three generators. A compact positive-dimensional complex submanifold cannot sit inside a Stein manifold. ◻
Example 10.5 (Every transverse leaf is dense). On M = C ∗ × ( C ∖ { 0 , 1 } ) M=\mathbb C^*\times(\mathbb C\setminus\{0,1\}) M = C ∗ × ( C ∖ { 0 , 1 }) , let
X = z ∂ z , ω = d z z + i d w w + ( 2 + i 3 ) d w w − 1 . X=z\partial_z,\quad
\omega=\frac{dz}{z}+i\frac{dw}{w}+(\sqrt2+i\sqrt3)\frac{dw}{w-1}. X = z ∂ z , ω = z d z + i w d w + ( 2 + i 3 ) w − 1 d w . Its period image is 2 π i [ Z + i Z + ( 2 + i 3 ) Z ] 2\pi i[\mathbb Z+i\mathbb Z+(\sqrt2+i\sqrt3)\mathbb Z] 2 π i [ Z + i Z + ( 2 + i 3 ) Z ] , dense in C \mathbb C C by Kronecker’s theorem; hence all leaves are dense in M M M .
Additive actions and generic counterexamples
Additive actions illustrate why the existence of a normalized time form need not lead to discrete-period approximation.
Theorem 11.1 (Stein orbit quotient: positive approximation). Suppose a complete holomorphic X X X generates a free proper C \mathbb C C -action on Stein M M M , and the quotient Riemann surface B = M / C B=M/\mathbb C B = M / C is Stein. If H 1 ( M , Z ) H_1(M,\mathbb Z) H 1 ( M , Z ) is finitely generated, every ω ∈ T X \omega\in\mathcal T_X ω ∈ T X admits normalized discrete-period approximation.
Proof. The projection π : M → B \pi:M\to B π : M → B is a principal C \mathbb C C -bundle. Its class in H 1 ( B , O B ) H^1(B,\mathcal O_B) H 1 ( B , O B ) vanishes because B B B is Stein, so M ≅ B × C M\cong B\times\mathbb C M ≅ B × C and X = ∂ t X=\partial_t X = ∂ t . Every normalized closed form is d t + π ∗ α dt+\pi^*\alpha d t + π ∗ α , with d α = 0 d\alpha=0 d α = 0 . Approximate [ α ] ∈ H 1 ( B , C ) [\alpha]\in H^1(B,\mathbb C) [ α ] ∈ H 1 ( B , C ) by discrete-period classes in the finite-dimensional cohomology space and realize small corrections by global closed holomorphic forms on Stein B B B . The forms d t + π ∗ α j dt+\pi^*\alpha_j d t + π ∗ α j are normalized and converge compact-open. ◻
We identify the relevant period constraints on a suspension through the cohomology of its base.
Proposition 11.2 (Identification with quotient Dolbeault cohomology). Let π : M → B \pi:M\to B π : M → B be a holomorphic principal C \mathbb C C -bundle over a connected Riemann surface, and assume that the total space M M M is Stein. Let X X X be its infinitesimal generator and let O X \mathcal O_X O X denote the sheaf of local holomorphic first integrals of X X X . Then there are canonical isomorphisms O X ≃ π − 1 O B , Q X : = O ( M ) / X O ( M ) ≃ H 1 ( M , O X ) ≃ H 1 ( B , O B ) . \mathcal O_X\simeq\pi^{-1}\mathcal O_B,\qquad
\mathcal Q_X:=\mathcal O(M)/X\mathcal O(M)
\simeq H^1(M,\mathcal O_X)\simeq H^1(B,\mathcal O_B). O X ≃ π − 1 O B , Q X := O ( M ) / X O ( M ) ≃ H 1 ( M , O X ) ≃ H 1 ( B , O B ) . If B B B is compact of genus g g g , Dolbeault theory identifies the last group with H 0 , 1 ( B ) H^{0,1}(B) H 0 , 1 ( B ) , of complex dimension g g g .
Proof. We first justify the identification of sheaves and the passage from total space to base, rather than relying only on the contractibility of the fiber. On every holomorphic bundle chart π − 1 ( U ) ≃ U × C \pi^{-1}(U)\simeq U\times\mathbb C π − 1 ( U ) ≃ U × C , write the fiber coordinate as t t t . Then X = ∂ t X=\partial_t X = ∂ t , and a holomorphic function is killed by X X X precisely when it is locally the pullback of a holomorphic function of the base coordinate. This gives the asserted sheaf isomorphism O X ≃ π − 1 O B \mathcal O_X\simeq\pi^{-1}\mathcal O_B O X ≃ π − 1 O B . Moreover the local flow-box primitive gives an exact sequence of sheaves of complex vector spaces
0 ⟶ O X ⟶ O M → X O M ⟶ 0. \begin{equation}
0\longrightarrow\mathcal O_X\longrightarrow\mathcal O_M
\xrightarrow{\,X\,}\mathcal O_M\longrightarrow0. \tag*{(11.1)}
\end{equation} 0 ⟶ O X ⟶ O M X O M ⟶ 0. (11.1) Since M M M is Stein, the long exact cohomology sequence and Cartan’s Theorem B applied to O M \mathcal O_M O M imply
H 1 ( M , O X ) ≃ O ( M ) / X O ( M ) = Q X . H^1(M,\mathcal O_X)\simeq\mathcal O(M)/X\mathcal O(M)=\mathcal Q_X. H 1 ( M , O X ) ≃ O ( M ) / X O ( M ) = Q X . For the base-cohomology identification, take a sufficiently small Stein coordinate disc U ⊂ B U\subset B U ⊂ B over which the bundle is holomorphically trivial. The total space π − 1 ( U ) ≃ U × C \pi^{-1}(U)\simeq U\times\mathbb C π − 1 ( U ) ≃ U × C is Stein. For an arbitrary h ∈ O ( U × C ) h\in\mathcal O(U\times\mathbb C) h ∈ O ( U × C ) , the holomorphic function
H ( z , t ) = ∫ 0 t h ( z , s ) d s H(z,t)=\int_0^t h(z,s)\,ds H ( z , t ) = ∫ 0 t h ( z , s ) d s is globally well defined, and ∂ t H = h \partial_t H=h ∂ t H = h . Applying cohomology to (11.1) on U × C U\times\mathbb C U × C , using this surjectivity and Cartan’s Theorem B, gives
H q ( π − 1 ( U ) , O X ) = 0 ( q ≥ 1 ) . H^q\bigl(\pi^{-1}(U),\mathcal O_X\bigr)=0\qquad(q\ge1). H q ( π − 1 ( U ) , O X ) = 0 ( q ≥ 1 ) . Also, the X X X -invariant holomorphic functions on the entire fibers over U U U are exactly pullbacks of O B ( U ) \mathcal O_B(U) O B ( U ) , so π ∗ O X ≃ O B \pi_*\mathcal O_X\simeq\mathcal O_B π ∗ O X ≃ O B . The displayed local vanishing proves R q π ∗ O X = 0 R^q\pi_*\mathcal O_X=0 R q π ∗ O X = 0 for q ≥ 1 q\ge1 q ≥ 1 . The Leray spectral sequence therefore degenerates to yield
H 1 ( M , O X ) ≃ H 1 ( B , O B ) . H^1(M,\mathcal O_X)\simeq H^1(B,\mathcal O_B). H 1 ( M , O X ) ≃ H 1 ( B , O B ) . If B B B is compact, the Dolbeault isomorphism H 1 ( B , O B ) ≃ H 0 , 1 ( B ) H^1(B,\mathcal O_B)\simeq H^{0,1}(B) H 1 ( B , O B ) ≃ H 0 , 1 ( B ) completes the proof. ◻
The next result shows that this constraint can exclude discrete periods generically.
Theorem 11.3 (Generic failure of normalized discrete periods). Let B B B be any compact Riemann surface of genus g ≥ 3 g\ge3 g ≥ 3 , with universal covering D → B \mathbb D\to B D → B and deck group Γ \Gamma Γ . There exists a residual subset G B ⊂ Hom ( Γ , R ) ≃ H 1 ( B , R ) , \mathscr G_B\subset\operatorname{Hom}(\Gamma,\mathbb R)
\simeq H^1(B,\mathbb R), G B ⊂ Hom ( Γ , R ) ≃ H 1 ( B , R ) , whose complement has Lebesgue measure zero, such that for every ρ ∈ G B \rho\in\mathscr G_B ρ ∈ G B the complex surface M ρ = ( D × C ) / Γ , γ ⋅ ( z , t ) = ( γ z , t + ρ ( γ ) ) , M_\rho=(\mathbb D\times\mathbb C)/\Gamma,
\qquad \gamma\cdot(z,t)=(\gamma z,t+\rho(\gamma)), M ρ = ( D × C ) /Γ , γ ⋅ ( z , t ) = ( γ z , t + ρ ( γ )) , is connected and Stein, and its translation vector field X X X is complete and generates a free, proper holomorphic C \mathbb C C -action. Moreover, H 1 ( M ρ , Z ) ≃ Z 2 g , T X ≠ ∅ , H_1(M_\rho,\mathbb Z)\simeq\mathbb Z^{2g},\qquad \mathcal T_X\ne\varnothing, H 1 ( M ρ , Z ) ≃ Z 2 g , T X = ∅ , but no η ∈ T X \eta\in\mathcal T_X η ∈ T X has a discrete period group: { η ∈ T X : Per ( η ) is discrete } = ∅ . \boxed{\;\{\eta\in\mathcal T_X:\operatorname{Per}(\eta)
\text{ is discrete}\}=\varnothing.\;} { η ∈ T X : Per ( η ) is discrete } = ∅ . In particular, normalized discrete-period approximation fails at every η ∈ T X \eta\in\mathcal T_X η ∈ T X . For g = 3 g=3 g = 3 this gives the original genus-three example.
Proof. Fix the compact surface B B B . Since g ≥ 3 g\ge3 g ≥ 3 , its universal covering is D \mathbb D D and Γ ≃ π 1 ( B ) \Gamma\simeq\pi_1(B) Γ ≃ π 1 ( B ) . A real character ρ : Γ → R \rho:\Gamma\to\mathbb R ρ : Γ → R defines the indicated holomorphic principal C \mathbb C C -bundle π : M ρ → B \pi:M_\rho\to B π : M ρ → B . Translation in t t t commutes with the Γ \Gamma Γ -action, so it descends to a complete, free holomorphic C \mathbb C C -action; local triviality over the Hausdorff base makes the action proper. As the fibers are contractible, the bundle projection is a homotopy equivalence, and H 1 ( M ρ , Z ) ≃ H 1 ( B , Z ) ≃ Z 2 g H_1(M_\rho,\mathbb Z)\simeq H_1(B,\mathbb Z)\simeq\mathbb Z^{2g} H 1 ( M ρ , Z ) ≃ H 1 ( B , Z ) ≃ Z 2 g .
The coordinate differential d t dt d t is Γ \Gamma Γ -invariant and descends to a closed holomorphic form ω ρ \omega_\rho ω ρ with ω ρ ( X ) = 1 \omega_\rho(X)=1 ω ρ ( X ) = 1 . Under the canonical identification H 1 ( M ρ , C ) ≃ H 1 ( B , C ) H^1(M_\rho,\mathbb C)\simeq H^1(B,\mathbb C) H 1 ( M ρ , C ) ≃ H 1 ( B , C ) , its cohomology class corresponds to the class of the additive character ρ \rho ρ (up to the chosen deck convention). The character ρ \rho ρ is nontrivial on the generic set under consideration; the suspension is then exactly the universal globalization of the compact-hyperbolic-base model of Iannuzzi–Trapani. Thus M ρ M_\rho M ρ is Stein by [27 Proposition 4.1(i)] ; see also [27 Remark 4.3] .
We now determine all normalized closed holomorphic forms. If η ∈ T X \eta\in\mathcal T_X η ∈ T X , then β = η − ω ρ \beta=\eta-\omega_\rho β = η − ω ρ satisfies d β = 0 d\beta=0 d β = 0 and β ( X ) = 0 \beta(X)=0 β ( X ) = 0 . Cartan’s formula gives L X β = 0 \mathcal L_X\beta=0 L X β = 0 . Thus β \beta β is basic for the principal bundle, and descends to a holomorphic one-form α \alpha α on B B B . Conversely any such form may be pulled back. Hence
T X = ω ρ + π ∗ H 0 ( B , Ω B 1 ) . \begin{equation}
\boxed{\;\mathcal T_X=\omega_\rho+\pi^*H^0(B,\Omega_B^1).\;} \tag*{(11.2)}
\end{equation} T X = ω ρ + π ∗ H 0 ( B , Ω B 1 ) . (11.2) Consider the Hodge decomposition
H 1 ( B , C ) = H 1 , 0 ( B ) ⊕ H 0 , 1 ( B ) , dim C H 0 , 1 ( B ) = g , H^1(B,\mathbb C)=H^{1,0}(B)\oplus H^{0,1}(B),
\qquad\dim_\mathbb CH^{0,1}(B)=g, H 1 ( B , C ) = H 1 , 0 ( B ) ⊕ H 0 , 1 ( B ) , dim C H 0 , 1 ( B ) = g , and let q : H 1 ( B , C ) → H 0 , 1 ( B ) q:H^1(B,\mathbb C)\to H^{0,1}(B) q : H 1 ( B , C ) → H 0 , 1 ( B ) be projection. Every η ∈ T X \eta\in\mathcal T_X η ∈ T X has the same antiholomorphic component:
q ( [ η ] ) = q ( [ ρ ] ) , \begin{equation}
q([\eta])=q([\rho]), \tag*{(11.3)}
\end{equation} q ([ η ]) = q ([ ρ ]) , (11.3) where the classes on M ρ M_\rho M ρ have been identified with classes on B B B .
If Per ( η ) ⊂ C \operatorname{Per}(\eta)\subset\mathbb C Per ( η ) ⊂ C is discrete, it is a free abelian group of rank at most two. Choose generators a , b ∈ C a,b\in\mathbb C a , b ∈ C (with one or both zero if the rank is smaller). The integral coordinate characters of the period homomorphism provide u , v ∈ H 1 ( B , Z ) u,v\in H^1(B,\mathbb Z) u , v ∈ H 1 ( B , Z ) with
[ η ] = a u + b v , q ( [ ρ ] ) = a q ( u ) + b q ( v ) . [\eta]=a\,u+b\,v,
\qquad q([\rho])=a\,q(u)+b\,q(v). [ η ] = a u + b v , q ([ ρ ]) = a q ( u ) + b q ( v ) . Thus q ( [ ρ ] ) q([\rho]) q ([ ρ ]) belongs to the complex vector space spanned by q ( u ) q(u) q ( u ) and q ( v ) q(v) q ( v ) .
Define the exceptional set
E B = ⋃ u , v ∈ H 1 ( B , Z ) span C { q ( u ) , q ( v ) } ⊂ H 0 , 1 ( B ) . \mathscr E_B=
\bigcup_{u,v\in H^1(B,\mathbb Z)}
\operatorname{span}_\mathbb C\{q(u),q(v)\}
\ \subset H^{0,1}(B). E B = u , v ∈ H 1 ( B , Z ) ⋃ span C { q ( u ) , q ( v )} ⊂ H 0 , 1 ( B ) . Because g ≥ 3 g\ge3 g ≥ 3 , each constituent has complex dimension at most two and is therefore a proper closed linear subspace of H 0 , 1 ( B ) H^{0,1}(B) H 0 , 1 ( B ) . There are countably many pairs ( u , v ) (u,v) ( u , v ) , hence E B \mathscr E_B E B is meagre and has Lebesgue measure zero. The restriction
q : H 1 ( B , R ) ⟶ H 0 , 1 ( B ) q:H^1(B,\mathbb R)\longrightarrow H^{0,1}(B) q : H 1 ( B , R ) ⟶ H 0 , 1 ( B ) is an isomorphism of real vector spaces: an element of H 1 , 0 ( B ) H^{1,0}(B) H 1 , 0 ( B ) which is also real must vanish, and both real dimensions are 2 g 2g 2 g . Consequently
G B = H 1 ( B , R ) ∖ q − 1 ( E B ) \mathscr G_B
=H^1(B,\mathbb R)\setminus q^{-1}(\mathscr E_B) G B = H 1 ( B , R ) ∖ q − 1 ( E B ) is residual and of full measure. Since 0 ∈ E B 0\in\mathscr E_B 0 ∈ E B , every ρ ∈ G B \rho\in\mathscr G_B ρ ∈ G B is nonzero, so Steinness applies. Equation (11.3) contradicts discreteness of the periods for any η ∈ T X \eta\in\mathcal T_X η ∈ T X . This completes the proof. ◻
Example 11.4 (Genus-three Stein principal bundle). Let B = D / Γ B=\mathbb D/\Gamma B = D /Γ be compact of genus three and let ρ : Γ → R \rho:\Gamma\to\mathbb R ρ : Γ → R belong to the residual full-measure set G B \mathscr G_B G B constructed in Theorem 11.3 . Then M ρ = ( D × C ) / Γ M_\rho=(\mathbb D\times\mathbb C)/\Gamma M ρ = ( D × C ) /Γ , with γ ( z , t ) = ( γ z , t + ρ ( γ ) ) \gamma(z,t)=(\gamma z,t+\rho(\gamma)) γ ( z , t ) = ( γ z , t + ρ ( γ )) , is Stein, X = ∂ t X=\partial_t X = ∂ t induces a free proper C \mathbb C C -action, and the descending d t dt d t is closed and normalized. No normalized closed form on M ρ M_\rho M ρ has discrete periods, by the Hodge-theoretic argument in that theorem.
Discrete-period approximation may also occur for nonperiodic additive flows:
Example 11.5 (Nonperiodic additive flow with discrete-period approximation). On M = ( C ∗ ) 2 M=(\mathbb C^*)^2 M = ( C ∗ ) 2 , let
X = z ∂ z + 2 w ∂ w , ω = 1 2 d z z + 1 2 2 d w w . X=z\partial_z+\sqrt2\,w\partial_w,\qquad
\omega=\frac12\frac{dz}{z}+\frac1{2\sqrt2}\frac{dw}{w}. X = z ∂ z + 2 w ∂ w , ω = 2 1 z d z + 2 2 1 w d w . The complete flow Φ t ( z , w ) = ( e t z , e 2 t w ) \Phi_t(z,w)=(e^tz,e^{\sqrt2 t}w) Φ t ( z , w ) = ( e t z , e 2 t w ) defines a faithful additive C \mathbb C C -action (no nonzero common time period). For real ε ≠ 0 \varepsilon\ne0 ε = 0 set
ω ε = ( 1 2 + i ε ) d z z + 1 2 − i ε 2 d w w . \omega_\varepsilon=\Big(\tfrac12+i\varepsilon\Big)\frac{dz}{z}
+\frac{\tfrac12-i\varepsilon}{\sqrt2}\frac{dw}{w}. ω ε = ( 2 1 + i ε ) z d z + 2 2 1 − i ε w d w . Then d ω ε = 0 d\omega_\varepsilon=0 d ω ε = 0 , ω ε ( X ) = 1 \omega_\varepsilon(X)=1 ω ε ( X ) = 1 , ω ε → ω \omega_\varepsilon\to\omega ω ε → ω , and the two period generators are R \mathbb R R -linearly independent for ε ≠ 0 \varepsilon\ne0 ε = 0 . Thus period groups are lattices. This illustrates why the genus-three obstruction is genuinely geometric rather than a consequence of the absence of a circle subgroup.
Singular multiplicative actions and logarithmic time
When the infinitesimal generator has zeros, logarithmic forms provide normalized time coordinates away from invariant divisors.
In this subsection X X X is allowed to vanish, so T X \mathcal T_X T X need not exist on the whole manifold. A meromorphic normalization η ( X ) = 1 \eta(X)=1 η ( X ) = 1 is interpreted as an identity in the field of meromorphic functions, not as a pointwise equality at the fixed points.
Theorem 12.1 (A logarithmic normalized time form for a singular action). Let M M M be a connected Stein manifold and let X ≢ 0 X\not\equiv0 X ≡ 0 be the holomorphic infinitesimal generator of a holomorphic C ∗ \mathbb C^* C ∗ -action, possibly with fixed points. Then there are k ∈ Z ∖ { 0 } k\in\mathbb Z\setminus\{0\} k ∈ Z ∖ { 0 } and a nonzero holomorphic semi-invariant f ∈ O ( M ) f\in\mathcal O(M) f ∈ O ( M ) satisfying X ( f ) = k f X(f)=k f X ( f ) = k f . Its zero divisor D = ( f = 0 ) D=(f=0) D = ( f = 0 ) is invariant and contains Sing ( X ) \operatorname{Sing}(X) Sing ( X ) . The meromorphic form η = 1 k d f f \eta=\frac1k\frac{df}{f} η = k 1 f df is closed and satisfies η ( X ) = 1 \eta(X)=1 η ( X ) = 1 meromorphically. The complement U = M ∖ D U=M\setminus D U = M ∖ D is Stein; X ∣ U X|_U X ∣ U is nowhere zero and complete; the restriction of η \eta η is normalized, holomorphic and closed with Per U ( η ) ⊆ 2 π i k Z . \operatorname{Per}_U(\eta)\subseteq \frac{2\pi i}{k}\mathbb Z. Per U ( η ) ⊆ k 2 π i Z . Moreover f : U → C ∗ f:U\to\mathbb C^* f : U → C ∗ is a surjective holomorphic submersion. If H 1 ( U , Z ) H_1(U,\mathbb Z) H 1 ( U , Z ) is finitely generated, each closed holomorphic ω \omega ω on U U U with ω ( X ) = 1 \omega(X)=1 ω ( X ) = 1 is a compact-open limit on U U U of normalized closed forms with discrete period groups, and each such approximant defines a holomorphic suspension over C / Per U ( ω j ) \mathbb C/\operatorname{Per}_U(\omega_j) C / Per U ( ω j ) .
Proof. Choose p ∈ M p\in M p ∈ M with X ( p ) ≠ 0 X(p)\ne0 X ( p ) = 0 . By holomorphic separability of tangent vectors on a Stein manifold choose h ∈ O ( M ) h\in\mathcal O(M) h ∈ O ( M ) such that X ( h ) ( p ) ≠ 0 X(h)(p)\ne0 X ( h ) ( p ) = 0 . Its circle Fourier coefficients
h k ( q ) = 1 2 π ∫ 0 2 π e − i k θ h ( e i θ ⋅ q ) d θ , k ∈ Z , h_k(q)=\frac1{2\pi}\int_0^{2\pi}e^{-ik\theta}h(e^{i\theta}\cdot q)\,d\theta,\qquad k\in\mathbb Z, h k ( q ) = 2 π 1 ∫ 0 2 π e − ik θ h ( e i θ ⋅ q ) d θ , k ∈ Z , are holomorphic and satisfy h k ( s ⋅ q ) = s k h k ( q ) h_k(s\cdot q)=s^k h_k(q) h k ( s ⋅ q ) = s k h k ( q ) for all s ∈ C ∗ s\in\mathbb C^* s ∈ C ∗ , by holomorphic continuation from S 1 S^1 S 1 . Since X ( h ) ( p ) ≠ 0 X(h)(p)\ne0 X ( h ) ( p ) = 0 , at least one coefficient of nonzero weight is nonzero. Take f = h k f=h_k f = h k for such k k k . The semi-invariance yields X ( f ) = k f X(f)=kf X ( f ) = k f and implies that its zero divisor is invariant and contains every zero of X X X . On the meromorphic level, d ( d f / f ) = 0 d(df/f)=0 d ( df / f ) = 0 and ( d f / f ) ( X ) = X ( f ) / f = k (df/f)(X)=X(f)/f=k ( df / f ) ( X ) = X ( f ) / f = k . The complement U U U is biholomorphic to the closed submanifold { ( p , v ) ∈ M × C : v f ( p ) = 1 } \{(p,v)\in M\times\mathbb C:vf(p)=1\} {( p , v ) ∈ M × C : v f ( p ) = 1 } of the Stein manifold M × C M\times\mathbb C M × C , hence is Stein. Completeness and preservation of U U U follow from the original C ∗ \mathbb C^* C ∗ -action. Since X ( f ) = k f X(f)=kf X ( f ) = k f and k ≠ 0 k\ne0 k = 0 , f f f is a submersion on U U U ; the nonempty image is a C ∗ \mathbb C^* C ∗ -orbit under f ( s ⋅ p ) = s k f ( p ) f(s\cdot p)=s^k f(p) f ( s ⋅ p ) = s k f ( p ) , hence is all of C ∗ \mathbb C^* C ∗ . For any closed path γ \gamma γ in U U U , the winding number of f ∘ γ f\circ\gamma f ∘ γ is integral and ∫ γ η = ( 2 π i / k ) wind ( f ∘ γ , 0 ) \int_\gamma\eta=(2\pi i/k)\operatorname{wind}(f\circ\gamma,0) ∫ γ η = ( 2 π i / k ) wind ( f ∘ γ , 0 ) . Thus the periods lie in the displayed discrete group. Finally apply the nonsingular normalized approximation Theorem 9.3 to U U U under the stated finite-generation hypothesis, followed by the suspension Theorem 8.1 . The fibers of f f f itself need not be connected; the suspension associated with the full period group supplies connected fibers. ◻
The logarithmic construction produces a corresponding statement for singular multiplicative actions.
Theorem 12.2 (Singular multiplicative actions). Let M M M be a connected Stein manifold and X ≢ 0 X\not\equiv0 X ≡ 0 the infinitesimal generator of a holomorphic C ∗ \mathbb C^* C ∗ -action, possibly with zeros. There exist k ∈ Z ∖ { 0 } k\in\mathbb Z\setminus\{0\} k ∈ Z ∖ { 0 } and a nonzero holomorphic semi-invariant f f f with X ( f ) = k f X(f)=kf X ( f ) = k f . The closed meromorphic form k − 1 d f / f k^{-1}df/f k − 1 df / f satisfies η ( X ) = 1 \eta(X)=1 η ( X ) = 1 meromorphically. Its polar divisor is invariant and contains Sing ( X ) \operatorname{Sing}(X) Sing ( X ) ; on its Stein complement U U U the form is holomorphic, normalized and has discrete periods contained in ( 2 π i / k ) Z (2\pi i/k)\mathbb Z ( 2 π i / k ) Z . Under finite generation of H 1 ( U , Z ) H_1(U,\mathbb Z) H 1 ( U , Z ) , all normalized closed forms on U U U admit discrete-period approximation.
Proof of Theorem 12.2 . Theorem 12.1 constructs a nonzero semi-invariant holomorphic function f f f of nonzero integral weight k k k and establishes the normalized logarithmic form k − 1 d f / f k^{-1}df/f k − 1 df / f , its invariant polar divisor, Stein complement, and discrete period containment. Its final assertion follows by applying normalized approximation on that complement, provided its first integral homology is finitely generated. These conclusions give Theorem 12.2 . ◻
Example 12.3 below displays the logarithmic pole at a fixed point.
Example 12.3 (The logarithmic pole at a fixed point). For the action s ⋅ z = s z s\cdot z=sz s ⋅ z = sz on M = C M=\mathbb C M = C , the generator X = z ∂ z X=z\partial_z X = z ∂ z vanishes at the origin. The semi-invariant f = z f=z f = z has weight 1 1 1 , and η = d z / z \eta=dz/z η = d z / z is a closed meromorphic form with η ( X ) = 1 \eta(X)=1 η ( X ) = 1 . On U = C ∗ U=\mathbb C^* U = C ∗ , its period group is 2 π i Z 2\pi i\mathbb Z 2 π i Z , even though H 1 ( C , C ) = 0 H^1(\mathbb C,\mathbb C)=0 H 1 ( C , C ) = 0 . Thus the relevant cohomology is H 1 ( U , C ) H^1(U,\mathbb C) H 1 ( U , C ) and logarithmic residues matter.
Hyperbolicity and Period Rigidity in Holomorphic Flows
Finally, transverse hyperbolicity together with global dynamical product structure supplies a dynamical mechanism for discreteness of periods distinct from approximation.
The connection between jointly integrable Anosov dynamics and invariant closed one-forms goes back to Plante’s treatment of smooth Anosov flows [35 ] and, in the holomorphic setting, to [36 §4] . The latter develops the compact complex theory, whereas the arguments below concern complete complex-time flows on noncompact manifolds. We use holomorphic hyperbolic splitting in a specified sense, without identifying this definition with every notion of Anosov flow in the compact holomorphic literature.
Definition 13.1 (Jointly integrable holomorphic hyperbolic flow). Let M M M be a connected complex manifold and let X X X be a nowhere-vanishing complete holomorphic vector field, with complex-time flow Φ : C × M → M \Phi:\mathbb C\times M\to M Φ : C × M → M . We say that Φ \Phi Φ is a jointly integrable holomorphic hyperbolic flow if there are holomorphic subbundles E s , E u ⊂ T M E^s,E^u\subset TM E s , E u ⊂ T M , both of positive complex rank and invariant under D Φ t D\Phi_t D Φ t for every t ∈ C t\in\mathbb C t ∈ C , such that
T M = E s ⊕ C X ⊕ E u , E s u : = E s ⊕ E u \begin{equation}
TM=E^s\oplus\mathbb CX\oplus E^u,\qquad E^{su}:=E^s\oplus E^u \tag*{(13.1)}
\end{equation} T M = E s ⊕ C X ⊕ E u , E s u := E s ⊕ E u (13.1) is an involutive holomorphic distribution, and if there exist a Hermitian metric h h h on M M M and constants C , λ > 0 C,\lambda>0 C , λ > 0 for which, for all real t ≥ 0 t\ge0 t ≥ 0 ,
∥ D Φ t ( v s ) ∥ h ≤ C e − λ t ∥ v s ∥ h , ∥ D Φ − t ( v u ) ∥ h ≤ C e − λ t ∥ v u ∥ h \begin{equation}
\|D\Phi_t(v_s)\|_h\le Ce^{-\lambda t}\|v_s\|_h,
\qquad
\|D\Phi_{-t}(v_u)\|_h\le Ce^{-\lambda t}\|v_u\|_h \tag*{(13.2)}
\end{equation} ∥ D Φ t ( v s ) ∥ h ≤ C e − λ t ∥ v s ∥ h , ∥ D Φ − t ( v u ) ∥ h ≤ C e − λ t ∥ v u ∥ h (13.2) for v s ∈ E s v_s\in E^s v s ∈ E s and v u ∈ E u v_u\in E^u v u ∈ E u . We call E s E^s E s and E u E^u E u the stable and unstable distributions, and the integral foliation of E s u E^{su} E s u the joint stable–unstable foliation.
These are uniform real-time estimates for a flow complete in complex time. In the noncompact setting the metric is part of the data: the estimates are not invariant under an arbitrary change of Hermitian metric. We do not require completeness of h h h in the definition, although both positive examples below admit complete metrics of the indicated kind. The term “hyperbolic” is used in this specified sense, without asserting equivalence with all definitions of holomorphic Anosov flow on compact complex manifolds.
Lemma 13.2 (The normalized joint time form). Under the invariant integrable splitting (13.1) there is a unique holomorphic one-form ω \omega ω satisfying ω ( X ) = 1 \omega(X)=1 ω ( X ) = 1 and ker ω = E s u \ker\omega=E^{su} ker ω = E s u . It is closed and invariant under every Φ t \Phi_t Φ t , t ∈ C t\in\mathbb C t ∈ C .
Proof. The decomposition of the tangent bundle defines a holomorphic projection to the line generated by X X X , hence the indicated form. Since the splitting is invariant, Φ t ∗ ω = ω \Phi_t^*\omega=\omega Φ t ∗ ω = ω and L X ω = 0 \mathcal L_X\omega=0 L X ω = 0 . Cartan’s formula and the normalization give ι X d ω = 0 \iota_Xd\omega=0 ι X d ω = 0 . If Y , Z Y,Z Y , Z are local sections of E s u E^{su} E s u , involutivity gives d ω ( Y , Z ) = − ω ( [ Y , Z ] ) = 0 d\omega(Y,Z)=-\omega([Y,Z])=0 d ω ( Y , Z ) = − ω ([ Y , Z ]) = 0 . These identities exhaust the components of d ω d\omega d ω under the direct-sum decomposition of T M TM T M . ◻
For p ∈ M p\in M p ∈ M let L p L_p L p denote the leaf of ker ω \ker\omega ker ω through p p p . If τ ∈ C ∖ { 0 } \tau\in\mathbb C\setminus\{0\} τ ∈ C ∖ { 0 } satisfies Φ τ ( p ) = p \Phi_\tau(p)=p Φ τ ( p ) = p , then the orbit loop has ω \omega ω -integral τ \tau τ , so that τ ∈ Λ : = Per ( ω ) \tau\in\Lambda:=\operatorname{Per}(\omega) τ ∈ Λ := Per ( ω ) . Therefore Φ τ \Phi_\tau Φ τ preserves L p L_p L p , and its restriction P τ = Φ τ ∣ L p P_\tau=\Phi_\tau|_{L_p} P τ = Φ τ ∣ L p is a holomorphic return automorphism. The relation between a period of the time form and a return time of a point is central to the next proposition.
Proposition 13.3 (Isolated-return criterion). Let M M M be a connected complex manifold, X X X a complete holomorphic vector field and ω ∈ T X \omega\in\mathcal T_X ω ∈ T X . Suppose Φ τ ( p ) = p \Phi_\tau(p)=p Φ τ ( p ) = p for some p ∈ M p\in M p ∈ M and τ ≠ 0 \tau\ne0 τ = 0 . Suppose there is a neighborhood U U U of p p p in M M M such that U ∩ L p ∩ Fix ( Φ τ ) = { p } . U\cap L_p\cap\operatorname{Fix}(\Phi_\tau)=\{p\}. U ∩ L p ∩ Fix ( Φ τ ) = { p } . Then Per ( ω ) \operatorname{Per}(\omega) Per ( ω ) is discrete. Accordingly the foliation ker ω \ker\omega ker ω is the foliation by connected fibers of a holomorphic fiber bundle onto C / Per ( ω ) \mathbb C/\operatorname{Per}(\omega) C / Per ( ω ) .
Proof. Put Λ = Per ( ω ) \Lambda=\operatorname{Per}(\omega) Λ = Per ( ω ) . For every s ∈ Λ s\in\Lambda s ∈ Λ , the time-s s s map preserves L p L_p L p . Commutativity of the flow implies
P τ ( Φ s ( p ) ) = Φ s ( Φ τ ( p ) ) = Φ s ( p ) , P_\tau(\Phi_s(p))=\Phi_s(\Phi_\tau(p))=\Phi_s(p), P τ ( Φ s ( p )) = Φ s ( Φ τ ( p )) = Φ s ( p ) , so Φ s ( p ) \Phi_s(p) Φ s ( p ) is a fixed point of P τ P_\tau P τ . If Λ \Lambda Λ were not discrete, it would contain nonzero elements s j → 0 s_j\to0 s j → 0 . Then Φ s j ( p ) → p \Phi_{s_j}(p)\to p Φ s j ( p ) → p in M M M ; the ambient isolation hypothesis forces Φ s j ( p ) = p \Phi_{s_j}(p)=p Φ s j ( p ) = p for large j j j . But the isotropy subgroup { t : Φ t ( p ) = p } \{t:\Phi_t(p)=p\} { t : Φ t ( p ) = p } is discrete because X ( p ) ≠ 0 X(p)\ne0 X ( p ) = 0 , a contradiction. The bundle conclusion is Theorem 8.1 . ◻
The return map at a real hyperbolic periodic orbit gives a complementary period-discreteness criterion under the same ambient isolation hypothesis.
Theorem 13.4 (Discreteness from a real hyperbolic periodic orbit with ambient isolation). Let Φ \Phi Φ be a jointly integrable holomorphic hyperbolic flow in the sense of Definition 13.1 , and let ω \omega ω be the form from Lemma 13.2 . If there exists p ∈ M p\in M p ∈ M with Φ T ( p ) = p \Phi_T(p)=p Φ T ( p ) = p for a real T > 0 T>0 T > 0 and an ambient neighborhood U U U of p p p satisfying U ∩ L p ∩ Fix ( Φ T ) = { p } , U\cap L_p\cap\operatorname{Fix}(\Phi_T)=\{p\}, U ∩ L p ∩ Fix ( Φ T ) = { p } , then Per ( ω ) \operatorname{Per}(\omega) Per ( ω ) is discrete and ker ω \ker\omega ker ω is a holomorphic fiber-bundle foliation over C / Per ( ω ) \mathbb C/\operatorname{Per}(\omega) C / Per ( ω ) . If M M M is Stein, the base is C ∗ \mathbb C^* C ∗ or an elliptic curve.
Proof. Write P = Φ T ∣ L p P=\Phi_T|_{L_p} P = Φ T ∣ L p and A = D Φ T ( p ) ∣ T p L p A=D\Phi_T(p)|_{T_pL_p} A = D Φ T ( p ) ∣ T p L p . The invariant splitting of T p L p T_pL_p T p L p and the estimates for D Φ n T ( p ) = A n D\Phi_{nT}(p)=A^n D Φ n T ( p ) = A n show that the spectrum of A A A on E p s E^s_p E p s has modulus less than one. Applying the same argument to A − n A^{-n} A − n on E p u E^u_p E p u shows that the spectrum on E p u E^u_p E p u has modulus greater than one. In particular, A − I A-I A − I is invertible. The holomorphic inverse function theorem applied to P − i d P-\mathrm{id} P − id in a chart of the leaf at p p p makes p p p an isolated fixed point of P P P in the intrinsic leaf topology. This alone does not imply ambient isolation. The additional ambient hypothesis permits application of Proposition 13.3 , giving discreteness and the bundle structure. Finally T ∈ Per ( ω ) T\in\operatorname{Per}(\omega) T ∈ Per ( ω ) rules out the zero subgroup, leaving rank one or two. ◻
The hyperbolicity estimates concern real time. The ambient-isolation criterion itself applies to every nonzero complex return time. A stronger conclusion is available when the global stable and unstable manifolds of the return map have a product property on the actual joint leaf, rather than only on a covering space. This global product hypothesis does not require the ambient isolation assumption of Theorem 13.4 .
Theorem 13.5 (Cyclicity under global dynamical product structure). Let Φ \Phi Φ satisfy Definition 13.1 , let ω \omega ω be its normalized joint time form, and suppose that Φ T ( p ) = p \Phi_T(p)=p Φ T ( p ) = p for some real T > 0 T>0 T > 0 . Put P = Φ T ∣ L p P=\Phi_T|_{L_p} P = Φ T ∣ L p . Suppose that for every a , b ∈ L p a,b\in L_p a , b ∈ L p the global dynamical stable manifold W P s ( a ) W^s_P(a) W P s ( a ) and global dynamical unstable manifold W P u ( b ) W^u_P(b) W P u ( b ) intersect in exactly one point. Then Per ( ω ) = { t ∈ C : Φ t ( p ) = p } . \operatorname{Per}(\omega)=\{t\in\mathbb C:\Phi_t(p)=p\}. Per ( ω ) = { t ∈ C : Φ t ( p ) = p } . If M M M is Stein, this common group is a nonzero discrete cyclic subgroup of C \mathbb C C , and the joint foliation defines a holomorphic fiber bundle over C ∗ \mathbb C^* C ∗ .
Proof. Suppose q ∈ L p q\in L_p q ∈ L p is fixed by P = Φ T ∣ L p P=\Phi_T|_{L_p} P = Φ T ∣ L p . The point r ∈ W P s ( p ) ∩ W P u ( q ) r\in W^s_P(p)\cap W^u_P(q) r ∈ W P s ( p ) ∩ W P u ( q ) is unique. Since P P P preserves these two sets, uniqueness implies P ( r ) = r P(r)=r P ( r ) = r . Its positive iterates converge to p p p , and its negative iterates converge to q q q . As r r r is fixed, r = p = q r=p=q r = p = q . Therefore Fix ( P ) = { p } \operatorname{Fix}(P)=\{p\} Fix ( P ) = { p } . For every s ∈ Per ( ω ) s\in\operatorname{Per}(\omega) s ∈ Per ( ω ) , the preceding commuting-return identity makes Φ s ( p ) \Phi_s(p) Φ s ( p ) a fixed point of P P P . It follows that Φ s ( p ) = p \Phi_s(p)=p Φ s ( p ) = p . Conversely, integration along an orbit loop shows that every element of the isotropy group of p p p belongs to Per ( ω ) \operatorname{Per}(\omega) Per ( ω ) . The two groups coincide.
The common group is discrete because X ( p ) ≠ 0 X(p)\ne0 X ( p ) = 0 , so its isotropy has no accumulation at the origin; it is nonzero because it contains T T T . A discrete rank-two isotropy lattice would make the orbit through p p p a compact elliptic curve C / Λ \mathbb C/\Lambda C /Λ , embedded as a complex orbit in M M M . This is impossible when M M M is Stein: every holomorphic map from a compact connected complex curve to a Stein manifold is constant. Thus the isotropy, and hence the period group, is cyclic. The suspension principle completes the proof. ◻
Proof of Theorem F
Now we are ready to prove the last of our main results.
Proof of Theorem F . The invariant splitting, involutivity, and hyperbolic estimates are precisely the jointly integrable hypotheses of Definition 13.1 . Lemma 13.2 constructs the unique normalized closed holomorphic time form with kernel E s ⊕ E u E^s\oplus E^u E s ⊕ E u . The real-period and global stable–unstable product hypotheses allow Theorem 13.5 to be applied. It identifies the period group with the nonzero discrete isotropy of p p p , and Theorem 8.1 gives the holomorphic fiber bundle. When M M M is Stein the group is cyclic, so its quotient base is C ∗ \mathbb C^* C ∗ , completing the proof. ◻
Here W P s ( a ) W^s_P(a) W P s ( a ) and W P u ( b ) W^u_P(b) W P u ( b ) denote the global dynamical stable and unstable sets of P P P , respectively, defined by
W P s ( a ) = { z ∈ L p : d ( P n ( z ) , P n ( a ) ) → 0 as n → + ∞ } , W^s_P(a)=\{z\in L_p:d(P^n(z),P^n(a))\to0
\text{ as }n\to+\infty\}, W P s ( a ) = { z ∈ L p : d ( P n ( z ) , P n ( a )) → 0 as n → + ∞ } ,
and
W P u ( b ) = { z ∈ L p : d ( P − n ( z ) , P − n ( b ) ) → 0 as n → + ∞ } , W^u_P(b)=\{z\in L_p:d(P^{-n}(z),P^{-n}(b))\to0
\text{ as }n\to+\infty\}, W P u ( b ) = { z ∈ L p : d ( P − n ( z ) , P − n ( b )) → 0 as n → + ∞ } ,
where d d d is a compatible metric.
Examples
Next we give a holomorphic flow with hyperbolic return dynamics and a cyclic period group.
Example 13.6. Take M = C ∗ × C 2 M=\mathbb C^*\times\mathbb C^2 M = C ∗ × C 2 with coordinates ( q , x , y ) (q,x,y) ( q , x , y ) , choose a , T > 0 a,T>0 a , T > 0 , and let
X = 2 π i T q ∂ q − a x ∂ x + a y ∂ y , ω = T 2 π i d q q . X=\frac{2\pi i}{T}q\partial_q-ax\partial_x+ay\partial_y,
\qquad \omega=\frac{T}{2\pi i}\frac{dq}{q}. X = T 2 π i q ∂ q − a x ∂ x + a y ∂ y , ω = 2 π i T q d q . The complete flow is
Φ t ( q , x , y ) = ( e 2 π i t / T q , e − a t x , e a t y ) , \Phi_t(q,x,y)=(e^{2\pi it/T}q,e^{-at}x,e^{at}y), Φ t ( q , x , y ) = ( e 2 π i t / T q , e − a t x , e a t y ) , so d ω = 0 d\omega=0 d ω = 0 , ω ( X ) = 1 \omega(X)=1 ω ( X ) = 1 , and Per ( ω ) = T Z \operatorname{Per}(\omega)=T\mathbb Z Per ( ω ) = T Z . The joint foliation consists of the fibers { q } × C 2 \{q\}\times\mathbb C^2 { q } × C 2 . The point ( 1 , 0 , 0 ) (1,0,0) ( 1 , 0 , 0 ) has real period T T T , and its return on the fiber is P T ( x , y ) = ( e − a T x , e a T y ) P_T(x,y)=(e^{-aT}x,e^{aT}y) P T ( x , y ) = ( e − a T x , e a T y ) , with unique fixed point. The invariant bundles E s = C ∂ x E^s=\mathbb C\partial_x E s = C ∂ x and E u = C ∂ y E^u=\mathbb C\partial_y E u = C ∂ y satisfy (13.2) for the complete Hermitian metric
h = ∣ d x + a T 2 π i x d q q ∣ 2 + ∣ d y − a T 2 π i y d q q ∣ 2 + ∣ d q q ∣ 2 . h=\left|dx+\frac{aT}{2\pi i}x\frac{dq}{q}\right|^2
+\left|dy-\frac{aT}{2\pi i}y\frac{dq}{q}\right|^2
+\left|\frac{dq}{q}\right|^2. h = d x + 2 π i a T x q d q 2 + d y − 2 π i a T y q d q 2 + q d q 2 . Indeed the three displayed coframe elements are respectively multiplied by e − a t , e a t e^{-at},e^{at} e − a t , e a t , and 1 1 1 under pullback by Φ t \Phi_t Φ t . For completeness, lift a rectifiable finite-length path to a logarithmic coordinate u u u with q = e u q=e^u q = e u . The ∣ d u ∣ 2 |du|^2 ∣ d u ∣ 2 term gives a finite variation of u u u , so its image is bounded. Along the path, the other coframe terms are d x + ( a T / ( 2 π i ) ) x d u dx+(aT/(2\pi i))x\,du d x + ( a T / ( 2 π i )) x d u and d y − ( a T / ( 2 π i ) ) y d u dy-(aT/(2\pi i))y\,du d y − ( a T / ( 2 π i )) y d u . The corresponding inhomogeneous linear integral inequalities, with bounded u u u and finite integrals of the coframe norms, show via Gronwall’s inequality that x x x and y y y are bounded and Cauchy toward a finite-length endpoint. Thus no divergent curve has finite length, which proves completeness. This furnishes a noncompact Stein model with both stable and unstable directions and a genuine real periodic orbit. In particular, it is a jointly integrable holomorphic hyperbolic flow in the sense of Definition 13.1 .
Example 13.7 (Nonperiodic jointly integrable hyperbolic flow). Let M = C 3 M=\mathbb C^3 M = C 3 with coordinates ( x , y , z ) (x,y,z) ( x , y , z ) , choose a > 0 a>0 a > 0 , and set
X = ∂ z − a x ∂ x + a y ∂ y , ω = d z . X=\partial_z-ax\partial_x+ay\partial_y,\qquad \omega=dz. X = ∂ z − a x ∂ x + a y ∂ y , ω = d z . The complete complex-time flow is
Φ t ( x , y , z ) = ( e − a t x , e a t y , z + t ) . \Phi_t(x,y,z)=(e^{-at}x,e^{at}y,z+t). Φ t ( x , y , z ) = ( e − a t x , e a t y , z + t ) . The translation in z z z shows that every orbit is nonperiodic and biholomorphic to C \mathbb C C . The invariant holomorphic bundles E s = C ∂ x E^s=\mathbb C\partial_x E s = C ∂ x and E u = C ∂ y E^u=\mathbb C\partial_y E u = C ∂ y have positive rank and E s u = ker d z E^{su}=\ker dz E s u = ker d z is integrable. Furthermore d ω = 0 d\omega=0 d ω = 0 , ω ( X ) = 1 \omega(X)=1 ω ( X ) = 1 , and Per ( ω ) = { 0 } \operatorname{Per}(\omega)=\{0\} Per ( ω ) = { 0 } . Put
θ s = d x + a x d z , θ u = d y − a y d z , θ 0 = d z . \theta_s=dx+ax\,dz,\qquad
\theta_u=dy-ay\,dz,\qquad \theta_0=dz. θ s = d x + a x d z , θ u = d y − a y d z , θ 0 = d z . These forms constitute a global holomorphic coframe, with Φ t ∗ θ s = e − a t θ s \Phi_t^*\theta_s=e^{-at}\theta_s Φ t ∗ θ s = e − a t θ s , Φ t ∗ θ u = e a t θ u \Phi_t^*\theta_u=e^{at}\theta_u Φ t ∗ θ u = e a t θ u , and Φ t ∗ θ 0 = θ 0 \Phi_t^*\theta_0=\theta_0 Φ t ∗ θ 0 = θ 0 . Thus the metric h = ∣ θ s ∣ 2 + ∣ θ u ∣ 2 + ∣ θ 0 ∣ 2 h=|\theta_s|^2+|\theta_u|^2+|\theta_0|^2 h = ∣ θ s ∣ 2 + ∣ θ u ∣ 2 + ∣ θ 0 ∣ 2 satisfies (13.2) with C = 1 C=1 C = 1 and λ = a \lambda=a λ = a . It is complete: in the global coordinates u = e a z x u=e^{az}x u = e a z x , v = e − a z y v=e^{-az}y v = e − a z y , and z z z , it reads
h = e − 2 a Re z ∣ d u ∣ 2 + e 2 a Re z ∣ d v ∣ 2 + ∣ d z ∣ 2 . h=e^{-2a\operatorname{Re}z}|du|^2
+e^{2a\operatorname{Re}z}|dv|^2+|dz|^2. h = e − 2 a Re z ∣ d u ∣ 2 + e 2 a Re z ∣ d v ∣ 2 + ∣ d z ∣ 2 . A finite-length path has bounded z z z -variation, and on bounded z z z -regions the metric is uniformly comparable with the Euclidean metric in ( u , v , z ) (u,v,z) ( u , v , z ) . Hence no divergent path can have finite length. This is a jointly integrable holomorphic hyperbolic flow in the sense of Definition 13.1 , with an exact normalized time form and no periodic complex-time orbit.
Example 13.8 (Elliptic periods without hyperbolicity). Let M = ( C ∗ ) 2 × C 2 M=(\mathbb C^*)^2\times\mathbb C^2 M = ( C ∗ ) 2 × C 2 , with coordinates ( z , w , x , y ) (z,w,x,y) ( z , w , x , y ) , and set
X = 2 π i z ∂ z − x ∂ x + y ∂ y , ω = 1 2 π i d z z + 1 2 π d w w . X=2\pi i z\partial_z-x\partial_x+y\partial_y,
\qquad
\omega=\frac{1}{2\pi i}\frac{dz}{z}
+\frac{1}{2\pi}\frac{dw}{w}. X = 2 π i z ∂ z − x ∂ x + y ∂ y , ω = 2 π i 1 z d z + 2 π 1 w d w . Then Φ t ( z , w , x , y ) = ( e 2 π i t z , w , e − t x , e t y ) \Phi_t(z,w,x,y)=(e^{2\pi it}z,w,e^{-t}x,e^ty) Φ t ( z , w , x , y ) = ( e 2 π i t z , w , e − t x , e t y ) , and ω ( X ) = 1 \omega(X)=1 ω ( X ) = 1 . Direct integration over the two standard generators of H 1 ( ( C ∗ ) 2 , Z ) H_1((\mathbb C^*)^2,\mathbb Z) H 1 (( C ∗ ) 2 , Z ) gives
Per ( ω ) = Z + i Z . \operatorname{Per}(\omega)=\mathbb Z+i\mathbb Z. Per ( ω ) = Z + i Z . The point ( 1 , 1 , 0 , 0 ) (1,1,0,0) ( 1 , 1 , 0 , 0 ) has real period 1 1 1 , but w w w is constant under the flow. The tangent direction w ∂ w w\partial_w w ∂ w contributes a neutral direction to ker ω \ker\omega ker ω after subtraction of a multiple of X X X , so the splitting (13.1) cannot be uniformly hyperbolic on the entire joint distribution. This example shows that Steinness, a closed normalized time form, and a real periodic orbit alone do not exclude elliptic periods. The next example realizes elliptic periods even with the full hyperbolicity of Definition 13.1 .
Example 13.9 (Hyperbolic returns with nondiscrete or elliptic periods). Let a > 0 a>0 a > 0 , α ∈ C \alpha\in\mathbb C α ∈ C , and put
V = { ( x , y ) ∈ C 2 : x y ≠ 1 } , M = C ∗ × V , f = x y − 1. V=\{(x,y)\in\mathbb C^2:xy\ne1\},\qquad
M=\mathbb C^*\times V,\qquad f=xy-1. V = {( x , y ) ∈ C 2 : x y = 1 } , M = C ∗ × V , f = x y − 1. This is a connected smooth affine Stein threefold. Indeed it is biholomorphic to the closed affine submanifold
{ ( q , r , x , y , v ) ∈ C 5 : q r = 1 , v ( x y − 1 ) = 1 } . \{(q,r,x,y,v)\in\mathbb C^5:qr=1,\ v(xy-1)=1\}. {( q , r , x , y , v ) ∈ C 5 : q r = 1 , v ( x y − 1 ) = 1 } . Set
X = 2 π i q ∂ q − a x ∂ x + a y ∂ y , ω = 1 2 π i ( d q q + α d f f ) . X=2\pi i q\partial_q-ax\partial_x+ay\partial_y,
\qquad
\omega=\frac1{2\pi i}\left(\frac{dq}{q}
+\alpha\frac{df}{f}\right). X = 2 π i q ∂ q − a x ∂ x + a y ∂ y , ω = 2 π i 1 ( q d q + α f df ) . The complete complex-time flow is
Φ t ( q , x , y ) = ( e 2 π i t q , e − a t x , e a t y ) . \Phi_t(q,x,y)=(e^{2\pi it}q,e^{-at}x,e^{at}y). Φ t ( q , x , y ) = ( e 2 π i t q , e − a t x , e a t y ) . Since f f f is invariant, d ω = 0 d\omega=0 d ω = 0 and ω ( X ) = 1 \omega(X)=1 ω ( X ) = 1 . The loop around the q q q -factor has period 1 1 1 , while the loop q = 1 q=1 q = 1 , x = 1 x=1 x = 1 , y = 1 + ε e i θ y=1+\varepsilon e^{i\theta} y = 1 + ε e i θ , for 0 ≤ θ ≤ 2 π 0\le\theta\le2\pi 0 ≤ θ ≤ 2 π and ε > 0 \varepsilon>0 ε > 0 , has period α \alpha α . Every period is an integral combination of these values, by the winding-number formula. Consequently
Per ( ω ) = Z + α Z . \operatorname{Per}(\omega)=\mathbb Z+\alpha\mathbb Z. Per ( ω ) = Z + α Z . Define global holomorphic vector fields
V s = ∂ x − α q y f ∂ q , V u = ∂ y − α q x f ∂ q . V_s=\partial_x-\alpha\frac{qy}{f}\partial_q,
\qquad
V_u=\partial_y-\alpha\frac{qx}{f}\partial_q. V s = ∂ x − α f q y ∂ q , V u = ∂ y − α f q x ∂ q . They span line bundles E s = C V s E^s=\mathbb CV_s E s = C V s , E u = C V u E^u=\mathbb CV_u E u = C V u , with
T M = E s ⊕ C X ⊕ E u , E s ⊕ E u = ker ω . TM=E^s\oplus\mathbb CX\oplus E^u,
\qquad E^s\oplus E^u=\ker\omega. T M = E s ⊕ C X ⊕ E u , E s ⊕ E u = ker ω . The joint distribution is involutive because ω \omega ω is closed, and direct differentiation gives
D Φ t ( V s ∣ p ) = e − a t V s ∣ Φ t ( p ) , D Φ t ( V u ∣ p ) = e a t V u ∣ Φ t ( p ) ( t ∈ C ) . D\Phi_t(V_s|_p)=e^{-at}V_s|_{\Phi_t(p)},\qquad
D\Phi_t(V_u|_p)=e^{at}V_u|_{\Phi_t(p)}
\quad(t\in\mathbb C). D Φ t ( V s ∣ p ) = e − a t V s ∣ Φ t ( p ) , D Φ t ( V u ∣ p ) = e a t V u ∣ Φ t ( p ) ( t ∈ C ) . The holomorphic coframe
θ s = d x + a x ω , θ u = d y − a y ω , ω \theta_s=dx+ax\omega,\qquad
\theta_u=dy-ay\omega,\qquad \omega θ s = d x + a x ω , θ u = d y − a y ω , ω is dual to ( V s , V u , X ) (V_s,V_u,X) ( V s , V u , X ) and satisfies
Φ t ∗ θ s = e − a t θ s , Φ t ∗ θ u = e a t θ u , Φ t ∗ ω = ω . \Phi_t^*\theta_s=e^{-at}\theta_s,\qquad
\Phi_t^*\theta_u=e^{at}\theta_u,\qquad
\Phi_t^*\omega=\omega. Φ t ∗ θ s = e − a t θ s , Φ t ∗ θ u = e a t θ u , Φ t ∗ ω = ω . Thus h = ∣ θ s ∣ 2 + ∣ θ u ∣ 2 + ∣ ω ∣ 2 h=|\theta_s|^2+|\theta_u|^2+|\omega|^2 h = ∣ θ s ∣ 2 + ∣ θ u ∣ 2 + ∣ ω ∣ 2 is a Hermitian metric satisfying (13.2) with C = 1 C=1 C = 1 and λ = a \lambda=a λ = a . This verifies all the hypotheses of Definition 13.1 , which does not require completeness of h h h .
The point p = ( 1 , 0 , 0 ) p=(1,0,0) p = ( 1 , 0 , 0 ) has real period 1 1 1 . On the joint leaf through p p p , the derivative of P = Φ 1 ∣ L p P=\Phi_1|_{L_p} P = Φ 1 ∣ L p at p p p has eigenvalues e − a e^{-a} e − a and e a e^a e a . In particular, p p p is an isolated fixed point in the intrinsic leaf topology. Nevertheless, if α ∈ R ∖ Q \alpha\in\mathbb R\setminus\mathbb Q α ∈ R ∖ Q , then Per ( ω ) \operatorname{Per}(\omega) Per ( ω ) is nondiscrete. For nonzero s j ∈ Z + α Z s_j\in\mathbb Z+\alpha\mathbb Z s j ∈ Z + α Z tending to zero, the distinct points Φ s j ( p ) = ( e 2 π i s j , 0 , 0 ) \Phi_{s_j}(p)=(e^{2\pi i s_j},0,0) Φ s j ( p ) = ( e 2 π i s j , 0 , 0 ) lie on L p L_p L p , are fixed by P P P , and converge to p p p in M M M . They do not converge in the intrinsic leaf topology: a local plaque through p p p is given by branches of log q + α log f = c o n s t a n t \log q+\alpha\log f=\mathrm{constant} log q + α log f = constant , and meets { x = y = 0 } \{x=y=0\} { x = y = 0 } only at p p p .
If Im α ≠ 0 \operatorname{Im}\alpha\ne0 Im α = 0 , then Per ( ω ) \operatorname{Per}(\omega) Per ( ω ) is a rank-two lattice, so the joint foliation is a holomorphic fiber bundle over an elliptic curve. For α = i \alpha=i α = i , the points Φ n i ( p ) = ( e − 2 π n , 0 , 0 ) \Phi_{ni}(p)=(e^{-2\pi n},0,0) Φ ni ( p ) = ( e − 2 π n , 0 , 0 ) , n ∈ Z n\in\mathbb Z n ∈ Z , are distinct isolated fixed points of P P P on the same joint leaf. Thus elliptic periods can occur under full hyperbolicity as defined here.
The following example exhibits a holomorphic flow with hyperbolic stable and unstable directions whose joint distribution E s ⊕ E u E^s\oplus E^u E s ⊕ E u is contact and hence nonintegrable.
Example 13.10 (A complete hyperbolic Reeb flow on an affine Stein group). Let G = S L ( 2 , C ) G=\mathrm{SL}(2,\mathbb C) G = SL ( 2 , C ) , the smooth affine Stein hypersurface a d − b c = 1 ad-bc=1 a d − b c = 1 in C 4 \mathbb C^4 C 4 . In its Lie algebra choose
H = 1 2 ( 1 0 0 − 1 ) , E = ( 0 1 0 0 ) , F = ( 0 0 1 0 ) . H=\frac12\begin{pmatrix}1&0\\0&-1\end{pmatrix},\quad
E=\begin{pmatrix}0&1\\0&0\end{pmatrix},\quad
F=\begin{pmatrix}0&0\\1&0\end{pmatrix}. H = 2 1 ( 1 0 0 − 1 ) , E = ( 0 0 1 0 ) , F = ( 0 1 0 0 ) . Then [ H , E ] = E [H,E]=E [ H , E ] = E , [ H , F ] = − F [H,F]=-F [ H , F ] = − F , and [ E , F ] = 2 H [E,F]=2H [ E , F ] = 2 H . Define
Φ t ( g ) = g exp ( t H ) , t ∈ C . \Phi_t(g)=g\exp(tH),\qquad t\in\mathbb C. Φ t ( g ) = g exp ( t H ) , t ∈ C . This is a complete holomorphic flow. Write H L , E L , F L H^L,E^L,F^L H L , E L , F L for the left-invariant holomorphic vector fields (A g L = d ( L g ) e A = g A A^L_g=d(L_g)_eA=gA A g L = d ( L g ) e A = g A ). The infinitesimal generator is X = H L X=H^L X = H L . Right multiplication acts in the left trivialization by Ad ( exp ( − t H ) ) \operatorname{Ad}(\exp(-tH)) Ad ( exp ( − t H )) , hence
D Φ t ( E g L ) = e − t E Φ t ( g ) L , D Φ t ( F g L ) = e t F Φ t ( g ) L . D\Phi_t(E^L_g)=e^{-t}E^L_{\Phi_t(g)},\qquad
D\Phi_t(F^L_g)=e^tF^L_{\Phi_t(g)}. D Φ t ( E g L ) = e − t E Φ t ( g ) L , D Φ t ( F g L ) = e t F Φ t ( g ) L . For the left-invariant Hermitian metric h h h making H L , E L , F L H^L,E^L,F^L H L , E L , F L orthonormal, these are exact real-time hyperbolicity estimates with C = λ = 1 C=\lambda=1 C = λ = 1 . The underlying left-invariant Riemannian metric is complete, as is every left-invariant Riemannian metric on a connected Lie group. The line fields E s = C E L E^s=\mathbb CE^L E s = C E L and E u = C F L E^u=\mathbb CF^L E u = C F L are invariant and holomorphic, but their sum is not integrable, because [ E L , F L ] = 2 H L [E^L,F^L]=2H^L [ E L , F L ] = 2 H L .
Let α \alpha α be the left-invariant holomorphic one-form defined by α ( H L ) = 1 \alpha(H^L)=1 α ( H L ) = 1 , α ( E L ) = α ( F L ) = 0 \alpha(E^L)=\alpha(F^L)=0 α ( E L ) = α ( F L ) = 0 . The Maurer–Cartan formula yields
d α ( E L , F L ) = − α ( [ E L , F L ] ) = − 2 , α ∧ d α ≠ 0. d\alpha(E^L,F^L)=-\alpha([E^L,F^L])=-2,
\qquad \alpha\wedge d\alpha\ne0. d α ( E L , F L ) = − α ([ E L , F L ]) = − 2 , α ∧ d α = 0. Also α ( X ) = 1 \alpha(X)=1 α ( X ) = 1 and ι X d α = 0 \iota_Xd\alpha=0 ι X d α = 0 , so X X X is the holomorphic Reeb vector field of the contact form α \alpha α . In contrast to the jointly integrable case, no closed holomorphic one-form η \eta η on G G G can satisfy η ( X ) = 1 \eta(X)=1 η ( X ) = 1 . Indeed, exp ( 4 π i H ) = I \exp(4\pi iH)=I exp ( 4 π i H ) = I , and the associated complex-time orbit loop would have η \eta η -integral 4 π i 4\pi i 4 π i ; this contradicts the fact that G G G is simply connected, hence that every closed holomorphic one-form is exact. Thus T X = ∅ \mathcal T_X=\varnothing T X = ∅ . The real subgroup S L ( 2 , R ) \mathrm{SL}(2,\mathbb R) SL ( 2 , R ) carries the corresponding homogeneous real-time geodesic-flow model, up to its standard central quotient. This familiar homogeneous construction is used here as a geometric contrast, not asserted to be original.
Thus, uniform real-time hyperbolicity alone does not imply the joint integrability of E s ⊕ E u E^s\oplus E^u E s ⊕ E u required in Definition 13.1 .
Statements and Declarations
Funding. This research was supported by Fundação Getúlio Vargas (FGV-EMAp, Rio de Janeiro).
Competing interests. There are no conflicts of interest. Data availability. No datasets were generated or analysed in this theoretical mathematical study.
Use of generative AI tools. The author used ChatGPT (OpenAI) to assist with language editing, LaTeX compilation and text grammar review. All mathematical statements, proofs, and references are due to and were reviewed by the author, who takes full responsibility for the accuracy, originality, and final content of the manuscript.