Introduction
The present paper provides a generalization of Simpson’s Lefschetz theorem from connected compact Kähler manifolds to connected smooth complex quasi-projective varieties, in the setting relevant to real characters of the fundamental group, and uses it to study quasi-projective groups, i.e., groups that can be realized as the fundamental group of a connected smooth complex quasi-projective variety. Simpson’s original theorem applies to an arbitrary nonzero holomorphic 1-form, as well as to its real part, which is a real harmonic 1-form on a compact Kähler manifold. In contrast, here we start with a nonzero real character
χ : G = π 1 ( X ) → R \chi:G=\pi_1(X)\to\mathbb{R} χ : G = π 1 ( X ) → R
and consider the real part of a suitably chosen logarithmic holomorphic 1-form representing χ \chi χ . We say a nonzero χ \chi χ is rational if its image has rank 1 over Z \mathbb{Z} Z ; otherwise, we say it is irrational .
We begin by recalling the following fundamental result of Simpson; see [38 Theorem 1] or [12 Theorem 2.1] .
Theorem 1.1 (Simpson). — Suppose X X X is a connected compact Kähler manifold, and let ω \omega ω be a nonzero (closed) holomorphic 1-form. Let p : X ^ → X p:\widehat{X}\to X p : X → X be a regular covering such that Re p ∗ ω \operatorname{Re}p^*\omega Re p ∗ ω is exact. Denote by
g : X ^ → R g:\widehat{X}\to\mathbb{R} g : X → R a primitive function of Re p ∗ ω \operatorname{Re}p^*\omega Re p ∗ ω . Then one of the following holds:
(i) There exists a Riemann surface C C C , a surjective holomorphic morphism
with connected fibers, and a closed holomorphic 1-form η \eta η on C C C such that
ω = f ∗ η . \omega=f^*\eta. ω = f ∗ η . (ii) For every t ∈ R t\in\mathbb{R} t ∈ R , g − 1 ( t ) g^{-1}(t) g − 1 ( t ) is connected and the inclusion induces a surjection
π 1 ( g − 1 ( t ) ) ↠ π 1 ( X ^ ) . \pi_1(g^{-1}(t))\twoheadrightarrow\pi_1(\widehat{X}). π 1 ( g − 1 ( t )) ↠ π 1 ( X ) .
This theorem, originally stated for smooth projective varieties, along with other results, can be generalized to the compact Kähler case; see [12 Remark 2.1.1] . We refer to this result as Simpson’s Lefschetz alternative . It has played a central role in the study of Kähler groups, namely, groups that arise as fundamental groups of connected compact Kähler manifolds. For Kähler groups, the reader may refer to [2 , 3 , 1 , 32 , 33 , 12 , 8 , 34 ] and so on. Remarkably, even its restriction to rational characters has significant geometric consequences. A closely related result was established independently by Napier and Ramachandran through a quite different approach, as part of their work on proper holomorphic fibrations from complete noncompact Kähler manifolds to Riemann surfaces; see [29 , 35 , 30 NR01, 33 ] .
Theorem 1.2. — (Napier–Ramachandran, [NR01, Theorem 4.3]) Suppose X X X is a connected compact Kähler manifold. Assume that there exists a surjective homomorphism
χ : π 1 ( X ) → Z \chi:\pi_1(X)\to\mathbb{Z} χ : π 1 ( X ) → Z such that its kernel is not finitely generated. Then X X X admits a surjective holomorphic mapping onto a Riemann surface C C C , and the homomorphism factors through π 1 ( C ) \pi_1(C) π 1 ( C ) .
This result follows from Simpson’s Lefschetz alternative. Indeed, for a rational character χ \chi χ , choose a holomorphic 1-form ω \omega ω whose real part α = Re ω \alpha=\operatorname{Re}\omega α = Re ω represents χ \chi χ . Then the primitive function of α \alpha α on the corresponding Z \mathbb{Z} Z -covering is proper, so every regular level set is compact and smooth. If the second alternative of Theorem 1.1 held, the fundamental group of such a level set would be finitely generated and would surject onto ker χ \ker\chi ker χ , forcing ker χ \ker\chi ker χ to be finitely generated. Thus the first alternative must occur. By contrast, the essential geometric argument underlying Theorem 1.2 accounts for much of the technical difficulty in proving the alternative Theorem 1.1 itself, as is evident from § 3 and § 4 . This naturally raises the following Question: To what extent does Simpson’s Lefschetz theorem [38 ] and the theorem of Napier–Ramachandran extend to smooth complex quasi-projective varieties?
The first question has been studied from different perspectives. Zhou [Zho22] proved a Lefschetz-type theorem for weakly 1-complete complex manifolds under an additional properness assumption, while Corlette and Simpson [9 Theorem 5.11] treated the quasi-projective case when the relevant holomorphic 1-form extends holomorphically across a smooth projective compactification.
Here we start with a nonzero real character χ : G → R \chi:G\to\mathbb{R} χ : G → R and consider a specific logarithmic holomorphic 1-form whose real part represents χ \chi χ . In this setting, we show that Simpson’s Lefschetz alternative continues to hold for smooth complex quasi-projective varieties. In full generality, however, one should perhaps not expect the same statement to hold for an arbitrary holomorphic 1-form or real harmonic 1-form. For example, the holomorphic 1-form sin ( 1 / z ) d z \sin(1/z)\,dz sin ( 1/ z ) d z on C ∗ \mathbb{C}^{*} C ∗ has zeros accumulating at the boundary; see Theorem 3.10.
Theorem A (Lefschetz alternative, see Theorem 4.1 for this general version and Theorem 3.1 for the rational version) Let X X X be a smooth connected complex quasi-projective variety and let χ : π 1 ( X ) → R \chi:\pi_1(X)\to\mathbb{R} χ : π 1 ( X ) → R be a nonzero group homomorphism. Let p : X ^ → X p:\widehat{X}\to X p : X → X be the corresponding χ \chi χ -covering, with π 1 ( X ^ ) ≅ ker χ \pi_1(\widehat{X})\cong\ker\chi π 1 ( X ) ≅ ker χ . Then one of the following holds:
(i) There exists a surjective algebraic morphism with connected general fiber f : X → C f:X\to C f : X → C onto a smooth quasi-projective curve such that, after equipping C C C with the orbifold structure determined by the gcd multiplicities m \mathbf{m} m of the fibers of f f f , there exists a homomorphism
χ C o r b : π 1 o r b ( C , m ) → R \chi_C^{\mathrm{orb}}:\pi_1^{\mathrm{orb}}(C,\mathbf{m})\to\mathbb{R} χ C orb : π 1 orb ( C , m ) → R satisfying
χ = χ C o r b ∘ f ∗ o r b . \chi=\chi_C^{\mathrm{orb}}\circ f_*^{\mathrm{orb}}. χ = χ C orb ∘ f ∗ orb . (ii) There exists a smooth closed real 1-form α \alpha α on X X X representing χ \chi χ such that, for any primitive function
u : X ^ → R , d u = p ∗ α , u:\widehat{X}\to\mathbb{R},\qquad du=p^*\alpha, u : X → R , d u = p ∗ α , every level set L t = u − 1 ( t ) L_t=u^{-1}(t) L t = u − 1 ( t ) is connected and the inclusion induces a surjection
π 1 ( L t ) ↠ π 1 ( X ^ ) . \pi_1(L_t)\twoheadrightarrow\pi_1(\widehat{X}). π 1 ( L t ) ↠ π 1 ( X ) . Equivalently, the pair ( X ^ , L t ) (\widehat{X},L_t) ( X , L t ) is 1-connected for every t ∈ R t\in\mathbb{R} t ∈ R .
We call it a Lefschetz alternative since it is not a dichotomy; see Theorem 3.2 and Figure 3 for the rational case. Here in (ii), the smooth closed 1-form α \alpha α is the real part of a logarithmic holomorphic 1-form on X X X with nonzero purely imaginary residues at infinity, constructed in sections 3.2 and 4.1 for rational and general case respectively; more specifically, see eqs. (5) , (7) and (22) .
Figure 3. logical relations for rational Lefschetz alternative: here (i) refers to the specific α \alpha α constructed in section 3.2.
Correspondingly, we have the following analogue of Theorem 1.2 in the smooth quasi-projective setting:
Theorem B (Theorem 3.3 ). — Let X X X be a smooth connected complex quasi-projective variety and let χ : π 1 ( X ) → Z \chi:\pi_1(X)\to\mathbb{Z} χ : π 1 ( X ) → Z be an epimorphism. If ker χ \ker\chi ker χ is not finitely generated, then there exists a surjective algebraic morphism with connected general fiber f : X → C f:X\to C f : X → C onto a smooth quasi-projective curve such that, after equipping C C C with the orbifold structure determined by the gcd multiplicities m \mathbf{m} m of the fibers of f f f , ( C , m ) (C,\mathbf{m}) ( C , m ) is hyperbolic and there exists an epimorphism
χ C o r b : π 1 o r b ( C , m ) → Z \chi_C^{\mathrm{orb}}:\pi_1^{\mathrm{orb}}(C,\mathbf{m})\to\mathbb{Z} χ C orb : π 1 orb ( C , m ) → Z such that
χ = χ C o r b ∘ f ∗ o r b . \chi=\chi_C^{\mathrm{orb}}\circ f_*^{\mathrm{orb}}. χ = χ C orb ∘ f ∗ orb .
As in the compact case, the geometric argument underlying this factorization accounts for much of the work required for the quasi-projective extension of Theorem 1.1 , say Theorem A. The proof of rational case combines ideas of Simpson [38 ] , especially the local picture developed there, with ideas of Zhou [Zho22], in particular the construction of a flow used to push a path in a prescribed direction. The main point is to ensure that the deformed path remains in the same relative homotopy class as the original one, while being pushed in the direction determined by the primitive function u u u on the regular Z \mathbb{Z} Z -covering associated with χ \chi χ . We carry out this argument in section 3 and then adapt it to the general χ \chi χ -cover in section 4. An outline of the proof strategy appears at the beginning of section 3.
A cautious reader may notice that Theorem B involves a hyperbolic holomorphic fibration, whereas Theorem A refers only to a holomorphic fibration. This distinction is one of the main reason why the two alternatives in Theorem A need not be mutually exclusive, see Figure 3 . We explain this phenomenon in terms of the BNS invariant in section 5.3.
Given the central role of Simpson’s Lefschetz theorem in the study of Kähler groups, one may expect its extension to the smooth quasi-projective setting to have a number of consequences for quasi-projective groups. Here we mention only those consequences that are most closely related to the previous work of the author and his collaborator [25 , 23 ] . First, we describe the complement of the BNS invariant for quasi-projective groups, in analogy with Delzant’s result for Kähler groups [12 ] .
Theorem C (Theorem 5.3). — Let X X X be a connected smooth complex quasi-projective variety with G = π 1 ( X ) G = \pi_1(X) G = π 1 ( X ) . Then
Σ 1 ( G ) c = S ( ⋃ f im ( f ∗ : H 1 ( C , m ; R ) → H 1 ( X ; R ) ) ) . \Sigma^1(G)^{\mathrm{c}} = \mathsf{S}\left(\bigcup_f \operatorname{im}\left(f^*: H^1(C,\mathbf{m};\mathbb{R}) \to H^1(X;\mathbb{R})\right)\right). Σ 1 ( G ) c = S f ⋃ im ( f ∗ : H 1 ( C , m ; R ) → H 1 ( X ; R ) ) . where the union runs over all hyperbolic orbifold fibrations f : X → ( C , m ) f : X \to(C,\mathbf{m}) f : X → ( C , m ) , which is a finite union. Here S \mathsf{S} S means the projection onto the unit sphere in H 1 ( X ; R ) H^1(X;\mathbb{R}) H 1 ( X ; R ) .
For restrictions on quasi-projective groups, we have
Proposition 1.3. — (a) (Theorem 5.5) Any group admitting a properly ascending (or descending) HNN decomposition with finitely generated base group is not a quasi-projective group.
(b) (Theorem 5.6) Any Thompson–Stein group F ( ℓ , Z [ 1 n 1 ⋯ n k ] , ⟨ n 1 , … , n k ⟩ ) F\left(\ell,\mathbb{Z}\left[\frac{1}{n_1\cdots n_k}\right],\langle n_1,\ldots,n_k\rangle\right) F ( ℓ , Z [ n 1 ⋯ n k 1 ] , ⟨ n 1 , … , n k ⟩ ) is not quasi-projective.
We also remark that (a) and (b) with k = 1 k = 1 k = 1 require only the rational Lefschetz alternative.
Remark 1.4. — A Lefschetz-type theorem for smooth quasi-projective varieties was also claimed by Rogov in [36 Appendix A] . The main issue we see in the proof occurs in the local trivialization of [Rog24, Lemma A.13]. There, the strict monotonicity of Re g \operatorname{Re} g Re g along a complete vector field is used to conclude that every trajectory meets the inverse image of a prescribed real line in C \mathbb{C} C exactly once. In the noncompact setting, however, completeness and strict monotonicity alone do not imply such a global crossing property: a complete trajectory may escape to infinity while Re g \operatorname{Re} g Re g remains bounded. Thus an additional uniform estimate forcing the relevant trajectories to cross the prescribed hypersurface would be needed, but such an estimate is not established. Consequently, the local trivialization in [36 Lemma A.13] is not justified by the stated argument.
Our proof differs substantially from his, both conceptually and technically. In particular, we replace the patching argument and global construction in [38 ] with a modification of the geometric flow constructed in [Zho22].
We also emphasize in this paper the relation between the Lefschetz-type theorem and the theorem of Napier–Ramachandran, which is one of the reasons for treating the rational case first. As the proof shows, although the lifting picture in the rational case is easier to describe, it already contains most of the essential difficulties of the argument.
There is also a second motivation coming from cohomology jump loci. As shown by Delzant in [11 ] , the rational part of the complement of the BNS invariant alone is sufficient to obtain a geometric description of the rank-one cohomology jump loci over any algebraically closed field of Kähler groups, while this rational part is closely related to the rational case of Simpson’s Lefschetz theorem. These considerations motivate our emphasis on the rational case.
In an upcoming paper with Yongqiang Liu [26 ] , we use the rational Lefschetz alternative, Theorem 3.1, to study twisted rank-one cohomology jump loci (see [27 ] ) over any algebraically closed field in the smooth quasi-projective setting. Our approach there follows a path different from that of [12 ] : rather than using the BNS invariant, we exploit the tropical geometry of cohomology jump loci, continuing the approach developed in [25 ] .
The paper is organized as follows. In section 2, we introduce the basic notation concerning hyperbolic Riemann orbifolds and holomorphic fibrations. In section 3, we prove the rational version of the Lefschetz alternative in the smooth quasi-projective setting; in section 4, we prove Theorem A for arbitrary nonzero real characters and explain the essential modifications needed to pass from the rational to the general case. Finally, in section 5, we discuss applications of Theorem A to quasi-projective groups and, in section 5.3, compare Theorem A with Theorem B from the viewpoint of the BNS invariant.
Preliminaries on Riemann orbifolds and fibration
In this section, we recall the notion of a Riemann orbifold, namely a Riemann surface equipped with an orbifold structure allowing finitely many points with nontrivial finite isotropy groups. For proofs of the results presented below, we refer the reader to [11 §4] , [33 §5] and the references therein.
Let X X X be a connected complex manifold, and let C C C be a Riemann surface. Let h : X → C h:X \to C h : X → C be a surjective holomorphic map with connected general fibers, which we call a holomorphic fibration . For a subset A ⊂ C A \subset C A ⊂ C , denote X A : = h − 1 ( A ) X_A := h^{-1}(A) X A := h − 1 ( A ) ; in particular, for a point z ∈ C z \in C z ∈ C , denote X z = h − 1 ( z ) X_z = h^{-1}(z) X z = h − 1 ( z ) . Assume that the set of critical values of h h h is locally finite. For each critical value p p p , let m p m_p m p be the greatest common divisor of the multiplicities of the irreducible components of the divisor h − 1 ( p ) h^{-1}(p) h − 1 ( p ) . Let
P : = { p ∈ C : m p ≥ 2 } . P := \{p \in C : m_p \ge2\}. P := { p ∈ C : m p ≥ 2 } .
For each p ∈ P p \in P p ∈ P , let γ p \gamma_p γ p be a positively oriented loop tracing the boundary circle of a coordinate disk D p D_p D p in C C C centered at p p p , chosen so that D p ‾ ∩ P = { p } \overline{D_p} \cap P = \{p\} D p ∩ P = { p } . Then the orbifold fundamental group is given by
π 1 o r b ( C , m ) : = π 1 ( C ∖ P ) / N \pi_1^{\mathrm{orb}}(C,\mathbf{m}) := \pi_1(C \setminus P)/N π 1 orb ( C , m ) := π 1 ( C ∖ P ) / N
where N N N is the normal subgroup of π 1 ( C ∖ P ) \pi_1(C \setminus P) π 1 ( C ∖ P ) generated by the loops { γ p m p } \{\gamma_p^{m_p}\} { γ p m p } with m p ≥ 2 m_p \ge2 m p ≥ 2 , and m ⊂ P N \mathbf{m} \subset P^{\mathbb{N}} m ⊂ P N is the data of critical points with multiplicities. For a regular value z ∈ C ∖ P z \in C \setminus P z ∈ C ∖ P of h h h , we have
π 1 ( X z ) ⟶ π 1 ( X ) → h ∗ : = h ∗ o r b π 1 o r b ( C ) ⟶ 1 (1) \pi_1(X_z) \longrightarrow\pi_1(X) \xrightarrow{h_*:=h_*^{\mathrm{orb}}} \pi_1^{\mathrm{orb}}(C) \longrightarrow1 \tag*{(1)} π 1 ( X z ) ⟶ π 1 ( X ) h ∗ := h ∗ orb π 1 orb ( C ) ⟶ 1 (1)
When h h h is proper or algebraic, we have the kernel of π \pi π is finitely generated (see, for example, [37 ] ).
For a compact orbifold ( C , m ) (C,\mathbf{m}) ( C , m ) , defines its orbifold Euler characteristic as:
χ o r b ( C , m ) = 2 − 2 g ( C ) − ∑ p ∈ P ( 1 − 1 m p ) \chi^{\mathrm{orb}}(C,\mathbf{m}) = 2 - 2g(C) - \sum_{p \in P}\left(1 - \frac{1}{m_p}\right) χ orb ( C , m ) = 2 − 2 g ( C ) − p ∈ P ∑ ( 1 − m p 1 )
where g ( C ) g(C) g ( C ) is the genus of its smooth compactification. If χ o r b ( C , m ) < 0 \chi^{\mathrm{orb}}(C,\mathbf{m}) < 0 χ orb ( C , m ) < 0 , we say it is hyperbolic .
When X X X is compact, so is C C C , then P P P is a finite set of points. The following finiteness of holomorphic fibrations for X X X fixed is well known, for example, see [11 Theorem 2] . Note that a holomorphic map h : X → ( C , m ) h:X \to(C,\mathbf{m}) h : X → ( C , m ) means a holomorphic map such that, for each orbifold point p i p_i p i , the multiplicity of every irreducible component of h − 1 ( p i ) h^{-1}(p_i) h − 1 ( p i ) is divisible by m i m_i m i . Two holomorphic fibrations h i : X → Σ i h_i:X \to\Sigma_i h i : X → Σ i , i = 1 , 2 i = 1,2 i = 1 , 2 are considered equivalent if there exists an orbifold isomorphism ψ : Σ 1 → Σ 2 \psi:\Sigma_1 \to\Sigma_2 ψ : Σ 1 → Σ 2 such that h 2 = ψ ∘ h 1 h_2 = \psi\circ h_1 h 2 = ψ ∘ h 1 .
Theorem 2.1. — Let X X X be a compact complex manifold, then there exists, up to equivalence, a finite set of pairs ( C i , h i ) (C_i,h_i) ( C i , h i ) , where C i = ( C i , m i ) C_i = (C_i,\mathbf{m}_i) C i = ( C i , m i ) is a hyperbolic Riemann orbifold, and h i : X → ( C i , m ) h_i:X \to(C_i,\mathbf{m}) h i : X → ( C i , m ) is a holomorphic map with connected fibers.
In the compact case, by applying the Stein factorization theorem and replacing the target by the corresponding intermediate Riemann surface if necessary, we may assume that every fiber of a holomorphic fibration is connected. Similarly, in the noncompact case, if h : X → C h : X \to C h : X → C is proper, the same argument shows that, after replacing C C C by the corresponding intermediate Riemann surface, we may assume that every fiber is connected. A related factorization result for smooth fiber bundle without the properness assumption, under the additional assumptions that h h h is a holomorphic submersion and that every fiber has finitely many connected components, is given in [Zho22, Theorem 2]; Zhou refers to this as a noncompact Stein factorization.
However, a holomorphic surjection from a noncompact quasi-projective manifold to an open Riemann surface need not be proper, and the hypotheses of [Zho22, Theorem 2] need not be satisfied. Thus, in the general quasi-projective setting, one can in general require only that the general fiber be connected. From now on, by a holomorphic fibration from a compact complex manifold we mean a surjective holomorphic map with connected fibers, whereas in the noncompact case we require only that the general fiber be connected.
For the case when X X X and C C C are noncompact, and the Riemann surface C C C is a compact Riemann surface with finitely many points removed, say
( C ∖ { p 1 , ⋯ , p r } , m ) (C \setminus\{p_1,\cdots,p_r\},\mathbf{m}) ( C ∖ { p 1 , ⋯ , p r } , m )
with m \mathbf{m} m the data of orbifold points in C ∖ { p 1 , … , p r } C \setminus\{p_1,\ldots,p_r\} C ∖ { p 1 , … , p r } ; we denote this data as a triple ( C , r , m ) (C,r,\mathbf{m}) ( C , r , m ) . The Euler characteristic of such an open Riemann surface is
χ o r b ( C , r , m ) = 2 − 2 g ( C ) − r − ∑ p ∈ P ( 1 − 1 m p ) . (2) \chi^{\mathrm{orb}}(C,r,\mathbf{m})=2-2g(C)-r-\sum_{p\in P}\left(1-\frac{1}{m_p}\right). \tag*{(2)} χ orb ( C , r , m ) = 2 − 2 g ( C ) − r − p ∈ P ∑ ( 1 − m p 1 ) . (2)
If χ o r b ( C , r , m ) < 0 \chi^{\mathrm{orb}}(C,r,\mathbf{m})<0 χ orb ( C , r , m ) < 0 , we say it is hyperbolic .
For our applications, the real part of a holomorphic 1 1 1 -form on C C C must represent a nonzero class in H 1 ( C ; R ) H^1(C;\mathbb{R}) H 1 ( C ; R ) . We therefore assume that its dimensions b 1 ( C ) ≥ 1 b_1(C)\geq1 b 1 ( C ) ≥ 1 . If C C C is compact, this is equivalent to g : = g ( C ) ≥ 1 g:=g(C)\geq1 g := g ( C ) ≥ 1 ; if C C C is noncompact, it is equivalent to 2 g + r − 1 ≥ 1 2g+r-1\geq1 2 g + r − 1 ≥ 1 . We also need to pay attention to the case when χ o r b = 0 \chi^{\mathrm{orb}}=0 χ orb = 0 when 2 g + r ≥ 2 2g+r\geq2 2 g + r ≥ 2 , namely
2 g + r = 2 , (3) 2g+r=2, \tag*{(3)} 2 g + r = 2 , (3)
and we have no multiply fibers.
We have the following two cases:
g = 1 g=1 g = 1 , r = 0 r=0 r = 0 : then C C C is an elliptic curve, and π 1 o r b ( C ) = π 1 ( C ) ≅ Z 2 \pi_1^{\mathrm{orb}}(C)=\pi_1(C)\cong\mathbb{Z}^2 π 1 orb ( C ) = π 1 ( C ) ≅ Z 2 ;
g = 0 g=0 g = 0 , r = 2 r=2 r = 2 : then C ≅ C ∗ C\cong\mathbb{C}^* C ≅ C ∗ , and π 1 o r b ( C ) = π 1 ( C ) ≅ Z \pi_1^{\mathrm{orb}}(C)=\pi_1(C)\cong\mathbb{Z} π 1 orb ( C ) = π 1 ( C ) ≅ Z .
In either case, the (orbifold) fundamental group is free abelian.
Under the condition 2 g + r ≥ 2 2g+r\geq2 2 g + r ≥ 2 , the finiteness of hyperbolic fibrations in the noncompact case follows from [4 Theorem 1.6] ; see also [14 §4.1] . We state the following results which is needed in Theorem 5.3 .
Theorem 2.2. — Let X X X be a smooth connected complex quasi-projective variety. Then, up to equivalence, there exist only finitely many pairs ( C i , h i ) (C_i,h_i) ( C i , h i ) , where C i o r b : = ( C i , r i , m i ) C_i^{\mathrm{orb}}:=(C_i,r_i,\mathbf{m}_i) C i orb := ( C i , r i , m i ) is a hyperbolic Riemann orbifold with 2 g ( C i ) + r i ≥ 2 2g(C_i)+r_i\geq2 2 g ( C i ) + r i ≥ 2 , and h i : X → C i h_i:X\to C_i h i : X → C i is a surjective algebraic morphism with general fiber connected.
We finish this section by a result on the BNS-invariant of hyperbolic Riemann orbifolds (See § 5.1 for essential definitions).
Proposition 2.3. — Let ( C , m ) (C,\mathbf{m}) ( C , m ) be a hyperbolic Riemann orbifold, open or closed, then we have Σ 1 ( C , m ) = ∅ \Sigma^1(C,\mathbf{m})=\varnothing Σ 1 ( C , m ) = ∅ .
Proof. — This result is well known. For the compact case, for example, we can use [25 Proposition 5.5] ; for the noncompact case, a similar calculation on rank one jump loci is carried out in [24 ] , and the same tropical upper bound of Σ 1 ( G ) \Sigma^1(G) Σ 1 ( G ) (see [25 Theorem 1.1] and Theorem 5.3 ) gives the same result.
The emptiness of Σ 1 ( G ) \Sigma^1(G) Σ 1 ( G ) for ( C , m ) (C,\mathbf{m}) ( C , m ) noncompact can also be seen directly as follows. For ( C , r , m ) (C,r,\mathbf{m}) ( C , r , m ) a noncompact Riemann orbifold, it is well-known that
π 1 o r b ( C , r , m ) ≅ F 2 g + r − 1 ∗ Z / m 1 Z ∗ ⋯ ∗ Z / m s Z , \pi_1^{\mathrm{orb}}(C,r,\mathbf{m})\cong F_{2g+r-1}*\mathbb{Z}/m_1\mathbb{Z}*\cdots*\mathbb{Z}/m_s\mathbb{Z}, π 1 orb ( C , r , m ) ≅ F 2 g + r − 1 ∗ Z / m 1 Z ∗ ⋯ ∗ Z / m s Z ,
We only need to consider the case in which the character sphere is nonempty, namely 2 g + r − 1 ≥ 1 2g+r-1\ge1 2 g + r − 1 ≥ 1 . If 2 g + r ≥ 3 2g+r\ge3 2 g + r ≥ 3 , then the nonabelian free group F 2 g + r − 1 F_{2g+r-1} F 2 g + r − 1 has empty BNS invariant, and so does π 1 o r b ( C , r , m ) \pi_1^{\mathrm{orb}}(C,r,\mathbf{m}) π 1 orb ( C , r , m ) by Theorem 5.2(a); if 2 g + r = 2 2g+r=2 2 g + r = 2 , then there exists multiply fibers, and π 1 o r b ( C , r , m ) ≅ Z ∗ Z / m 1 Z ∗ ⋯ ∗ Z / m s Z \pi_1^{\mathrm{orb}}(C,r,\mathbf{m})\cong\mathbb{Z}*\mathbb{Z}/m_1\mathbb{Z}*\cdots*\mathbb{Z}/m_s\mathbb{Z} π 1 orb ( C , r , m ) ≅ Z ∗ Z / m 1 Z ∗ ⋯ ∗ Z / m s Z , whose BNS-invariant is empty by direct computation.
Lefschetz alternative of smooth quasi-projective manifolds in the rational case
In this section, we prove a version of the Lefschetz alternative for connected smooth complex quasi-projective varieties under the assumption that the real character is rational . We begin with this special case because of its relative clarity, and because its proof illustrates several of the main ideas that will also enter the proof of the general case.
Let X X X be a smooth quasi-projective manifold, and let G = π 1 ( X ) G=\pi_1(X) G = π 1 ( X ) . Up to rescaling, let χ : G ↠ Z \chi:G\twoheadrightarrow\mathbb{Z} χ : G ↠ Z be a nonzero rational character and set N = ker χ N=\ker\chi N = ker χ . Let p : X ^ → X p:\widehat{X}\to X p : X → X be the corresponding regular Z \mathbb{Z} Z -cover, so that π 1 ( X ^ ) = N \pi_1(\widehat{X})=N π 1 ( X ) = N and the deck transformation group is naturally identified with Z \mathbb{Z} Z . Denote by T T T the deck transformation corresponding to 1 ∈ Z 1\in\mathbb{Z} 1 ∈ Z . Choose a smooth closed 1 1 1 -form θ \theta θ on X X X representing χ \chi χ , in the sense that
∫ γ θ = χ ( γ ) \int_\gamma\theta=\chi(\gamma) ∫ γ θ = χ ( γ )
for every γ ∈ π 1 ( X ) \gamma\in\pi_1(X) γ ∈ π 1 ( X ) . Since p ∗ θ p^*\theta p ∗ θ has trivial periods on X ^ \widehat{X} X , it is exact. Hence there exists a smooth function u : X ^ → R u:\widehat{X}\to\mathbb{R} u : X → R such that d u = p ∗ θ du=p^*\theta d u = p ∗ θ . For a fixed choice of θ \theta θ , the function u u u is unique up to an additive constant. Moreover,
u ( T x ) = u ( x ) + 1 u(Tx)=u(x)+1 u ( T x ) = u ( x ) + 1
for every x ∈ X ^ x\in\widehat{X} x ∈ X . We call u u u a primitive function associated with χ \chi χ on the Z \mathbb{Z} Z -cover.
Theorem 3.1. — Let X X X be a smooth connected complex quasi-projective variety and let χ : π 1 ( X ) ↠ Z \chi:\pi_1(X)\twoheadrightarrow\mathbb{Z} χ : π 1 ( X ) ↠ Z be an epimorphism. Let p : X ^ → X p:\widehat{X}\to X p : X → X be the corresponding Z \mathbb{Z} Z -covering. Then one of the following holds:
(i) There exists a surjective algebraic morphism with connected general fiber f : X → C f:X\to C f : X → C onto a smooth quasi-projective curve such that, after equipping C C C with the orbifold structure determined by the gcd multiplicities of the fibers of f f f , there exists a homomorphism
χ C o r b : π 1 o r b ( C , m ) → Z \chi_C^{\mathrm{orb}}:\pi_1^{\mathrm{orb}}(C,\mathbf{m})\to\mathbb{Z} χ C orb : π 1 orb ( C , m ) → Z satisfying
χ = χ C o r b ∘ f ∗ o r b . \chi=\chi_C^{\mathrm{orb}}\circ f_*^{\mathrm{orb}}. χ = χ C orb ∘ f ∗ orb . (ii) There exists a smooth closed 1 1 1 -form α \alpha α on X X X representing χ \chi χ such that, for a primitive function
u : X ^ → R , d u = p ∗ α , u:\widehat{X}\to\mathbb{R},\qquad du=p^*\alpha, u : X → R , d u = p ∗ α , every level set L t = u − 1 ( t ) L_t=u^{-1}(t) L t = u − 1 ( t ) is connected and the inclusion induces a surjection
π 1 ( L t ) ↠ π 1 ( X ^ ) . \pi_1(L_t)\twoheadrightarrow\pi_1(\widehat{X}). π 1 ( L t ) ↠ π 1 ( X ) . Equivalently, the pair ( X ^ , L t ) (\widehat{X},L_t) ( X , L t ) is 1 1 1 -connected for every t ∈ R t\in\mathbb{R} t ∈ R .
Remark 3.2. — The two alternatives are not mutually exclusive. For example, take X = C ∗ X=\mathbb{C}^* X = C ∗ and let χ : π 1 ( X ) ≅ Z → Z \chi:\pi_1(X)\cong\mathbb{Z}\to\mathbb{Z} χ : π 1 ( X ) ≅ Z → Z be the identity. Then both (i) and (ii) hold. For an example of compact case, take X X X to be an elliptic curve, with χ : π 1 ( X ) → Z \chi:\pi_1(X)\to\mathbb{Z} χ : π 1 ( X ) → Z any nonzero character.
Theorem 3.3. — Let X X X be a smooth connected complex quasi-projective variety and χ : π 1 ( X ) ↠ Z \chi:\pi_1(X)\twoheadrightarrow\mathbb{Z} χ : π 1 ( X ) ↠ Z an epimorphism. If ker ( χ ) \ker(\chi) ker ( χ ) is not finitely generated, there is a surjective algebraic morphism with connected general fiber f : X → C f:X\to C f : X → C onto a smooth quasi-projective
curve such that, after equipping C C C with the gcd multiplicities m \mathbf{m} m of the fibers of f f f , ( C , m ) (C,\mathbf{m}) ( C , m ) is hyperbolic and
χ = χ C ∘ f o r b , χ C : π 1 o r b ( C , m ) ↠ Z . \chi=\chi_C\circ f^{\mathrm{orb}},\qquad\chi_C:\pi_1^{\mathrm{orb}}(C,\mathbf{m})\twoheadrightarrow\mathbb{Z}. χ = χ C ∘ f orb , χ C : π 1 orb ( C , m ) ↠ Z .
We further transfer this non-finitely generated property into a topological condition.
Proposition 3.5. — Let M M M be a compact connected smooth manifold, and let
χ : G : = π 1 ( M ) ↠ Z \chi:G:=\pi_1(M)\twoheadrightarrow\mathbb{Z} χ : G := π 1 ( M ) ↠ Z be a group epimorphism. Denote the coverings
M ~ → q M ^ → p M \widetilde{M}\xrightarrow{q}\widehat{M}\xrightarrow{p}M M q M p M where M ^ \widehat{M} M is the regular Z \mathbb{Z} Z -covering and M ~ \widetilde{M} M is the universal covering. Take N = ker ( χ ) N=\ker(\chi) N = ker ( χ ) , and u u u be the primitive function of χ \chi χ on M ^ \widehat{M} M . If N N N is infinitely generated, then, for every regular value t t t of u u u , the level set of q ∗ u q^*u q ∗ u
M ~ t : = { x ∈ M ~ : u ∘ q ( x ) = t } \widetilde{M}_t:=\{x\in\widetilde{M}:u\circ q(x)=t\} M t := { x ∈ M : u ∘ q ( x ) = t } has infinitely many connected components.
Proof. — By Sard theorem, we can find a regular value t t t of u u u , such that the conclusion of this lemma is not vacuous. Due to M M M is compact and u ( T . x ) = u ( x ) + 1 u(T.x)=u(x)+1 u ( T . x ) = u ( x ) + 1 , the function u : M ^ → R u:\widehat{M}\to\mathbb{R} u : M → R is proper. Thus the level set M ^ t : = { x ∈ M ^ : u ( x ) = t } \widehat{M}_t:=\{x\in\widehat{M}:u(x)=t\} M t := { x ∈ M : u ( x ) = t } is compact, and we decompose it into the finite union of connected compact components:
M ^ t = M 1 ⊔ M 2 ⊔ ⋯ ⊔ M k . \widehat{M}_t=M_1\sqcup M_2\sqcup\cdots\sqcup M_k. M t = M 1 ⊔ M 2 ⊔ ⋯ ⊔ M k .
Fix a component M i M_i M i , and write G i = im [ π 1 ( M i ) → π 1 ( M ) ] G_i=\operatorname{im}[\pi_1(M_i)\to\pi_1(M)] G i = im [ π 1 ( M i ) → π 1 ( M )] . Due to M i M_i M i is compact and smooth, G i G_i G i is finitely generated. If [ N : G i ] < ∞ [N:G_i]<\infty [ N : G i ] < ∞ , then N N N would be finitely generated, contradicting the assumption. Hence we have [ N : G i ] = ∞ [N:G_i]=\infty [ N : G i ] = ∞ , which equals the number of connected components of q − 1 ( M i ) ⊂ M ~ t q^{-1}(M_i)\subset\widetilde{M}_t q − 1 ( M i ) ⊂ M t . This shows that M ~ t \widetilde{M}_t M t has infinitely many connected components.
The proof of Theorem 3.1 for regular level sets, together with the proof of Theorem 3.3 , will be given in sections 3.1 to 3.5. The extension of Theorem 3.1 to arbitrary level sets will be given in section 3.6. The main ideas of the proof are as follows:
First, the character χ \chi χ factors through the quasi-Albanese morphism. We construct an associated morphism
to a semi-abelian variety A A A such that the induced character does not vanish on any positive-dimensional connected algebraic subgroup of A A A . In the rational case, the torus part of A A A has dimension at most one. We then take a natural projective compactification of A A A whose boundary consists of two disjoint divisors, and use Deligne’s logarithmic de Rham theory to choose a logarithmic 1 1 1 -form ω \omega ω whose real part representing χ \chi χ , with nonzero purely imaginary residues along the boundary. This is carried out in sections 3.1 and 3.2.
When dim ψ ( X ) ≥ 2 \dim\psi(X)\geq2 dim ψ ( X ) ≥ 2 , the Lefschetz hyperplane theorem allows us to reduce to a smooth quasi-projective surface Y ⊂ X Y\subset X Y ⊂ X such that π 1 ( Y ) ≅ π 1 ( X ) \pi_1(Y)\cong\pi_1(X) π 1 ( Y ) ≅ π 1 ( X ) and ψ ∣ Y \psi|_Y ψ ∣ Y is generically finite onto its image. After resolving indeterminacies and the boundary, we choose smooth projective compactifications Y ‾ ⊂ Y ‾ \overline{Y}\subset\overline{Y} Y ⊂ Y , A ⊂ A ‾ A\subset\overline{A} A ⊂ A such that ψ ∣ Y \psi|_Y ψ ∣ Y extends to a morphism ψ ‾ : Y ‾ → A ‾ \overline{\psi}:\overline{Y}\to\overline{A} ψ : Y → A , and the relevant boundary divisors have simple normal crossings. We define the polar boundary by P = ( ψ ‾ − 1 ( D A ) ) r e d P=(\overline{\psi}^{-1}(D_A))_{\mathrm{red}} P = ( ψ − 1 ( D A ) ) red , and let Q Q Q be the union of the remaining irreducible components of Y ‾ ∖ Y \overline{Y}\setminus Y Y ∖ Y . Thus Y ‾ ∖ Y = P ∪ Q \overline{Y}\setminus Y=P\cup Q Y ∖ Y = P ∪ Q , and denote Y ∘ = Y ‾ ∖ P Y^\circ=\overline{Y}\setminus P Y ∘ = Y ∖ P . The pullback ξ = ψ ‾ ∗ ω \xi=\overline{\psi}^*\omega ξ = ψ ∗ ω is holomorphic on Y ∘ Y^\circ Y ∘ , and the nonzero residues of ω \omega ω give uniform control near P P P . We implement this control by passing to
the real-oriented blowup M = Bl P r o ( Y ‾ ) M = \operatorname{Bl}^{\mathrm{ro}}_{P}(\overline{Y}) M = Bl P ro ( Y ) , on which the angular variables remain well defined along the boundary; another way to see this is to use a Poincaré-type metric. This is carried out in section 3.3.
(3) The nonpolar boundary Q Q Q gives rise to two types of critical neighborhoods. Write Q = Q ′ ∪ Q ′ ′ Q = Q' \cup Q'' Q = Q ′ ∪ Q ′′ , where ξ ∣ Q j ∖ P ≡ 0 \left.\xi\right|_{Q_j\setminus P} \equiv0 ξ ∣ Q j ∖ P ≡ 0 for components Q j ⊂ Q ′ Q_j \subset Q' Q j ⊂ Q ′ , while ξ ∣ Q j ∖ P ≠ 0 \left.\xi\right|_{Q_j\setminus P} \ne0 ξ ∣ Q j ∖ P = 0 for components Q j ⊂ Q ′ ′ Q_j \subset Q'' Q j ⊂ Q ′′ . We put Q ′ Q' Q ′ together with the zero set Z ( ξ ) Z(\xi) Z ( ξ ) and construct Type I critical neighborhoods, following the local construction of Simpson [38 ] . For Q ′ ′ Q'' Q ′′ , we introduce Type II critical neighborhoods around the points where the tangential restriction of ξ \xi ξ vanishes, as well as around the transverse intersections of components of Q ′ ′ Q'' Q ′′ which cannot be handled by a globally tangent vector field. When components of Q ′ ′ Q'' Q ′′ meet a Type I neighborhood, we incorporate them into the corresponding mixed Type I situation. We then establish the required local 1-connectedness statements for all these neighborhoods, always after deleting the relevant part of Q Q Q , denoted as W i ∗ W_i^* W i ∗ , W i \mathcal{W}_i W i , W i † W_i^\dagger W i † for the type I, type II and mixed type respectively. This is carried out in section 3.4.
(4) We then pass to the regular Z \mathbb{Z} Z -covering p : M ^ → M p:\widehat{M}\to M p : M → M corresponding to χ \chi χ , and write Y ^ = p − 1 ( Y ) \widehat{Y}=p^{-1}(Y) Y = p − 1 ( Y ) . The primitive function u : M ^ → R u:\widehat{M}\to\mathbb{R} u : M → R , d u = p ∗ Re ξ \mathrm{d}u=p^*\operatorname{Re}\xi d u = p ∗ Re ξ , is proper in the rational case. We construct a uniformly controlled flow on M ^ \widehat{M} M which increases u u u outside the lifted critical neighborhoods, is tangent to the polar boundary, and preserves the complement of the lifted nonpolar boundary. Given a path in Y ^ \widehat{Y} Y whose endpoints lie in a level set L t ⊂ Y ^ L_t\subset\widehat{Y} L t ⊂ Y , the flow moves the portions outside the critical neighborhoods toward L t L_t L t . Inside the critical neighborhoods, the local 1-connectedness in Theorems Theorems 3.15, 3.17 , 3.19 and 3.21 allow us to replace the corresponding portions by paths in the same relative homotopy class while preserving the required control of u u u . The construction is arranged throughout so that the deformed path never crosses either P P P or Q Q Q , and hence remains in Y ^ \widehat{Y} Y . This is carried out in section 3.5.
This proves that, when dim ψ ( X ) ≥ 2 \dim\psi(X)\ge2 dim ψ ( X ) ≥ 2 , every regular level set of u u u in Y ^ \widehat{Y} Y is connected and its inclusion induces a surjection onto π 1 ( Y ^ ) \pi_1(\widehat{Y}) π 1 ( Y ) . Using π 1 ( Y ) ≅ π 1 ( X ) \pi_1(Y)\cong\pi_1(X) π 1 ( Y ) ≅ π 1 ( X ) and the corresponding induced coverings, we transfer this conclusion from Y ^ \widehat{Y} Y to X ^ \widehat{X} X , obtaining condition (ii) of Theorem 3.1 for regular levels. Additional local 1-connectedness results (Theorems 3.24 to 3.26), together with the same flow argument, extend the conclusion to arbitrary level sets in section 3.6. The resulting statement is then transferred from Y ^ \widehat{Y} Y to X ^ \widehat{X} X in the same way. On the other hand, when dim ψ ( X ) = 1 \dim\psi(X)=1 dim ψ ( X ) = 1 , the morphism ψ \psi ψ yields the factorization in condition (i).
On semi-Albanese map determined by a real character
This section is the quasi-projective analogue of Simpson’s construction of the Albanese map determined by a holomorphic one-form in [38 ] . We work instead with the quasi-Albanese map and semi-abelian varieties. We refer the reader to [17 , 16 ] for the relevant properties used below.
Let X X X be a smooth connected quasi-projective variety over C \mathbb{C} C , and let a : X → A 0 a:X\to A_0 a : X → A 0 be its quasi-Albanese map. The induced map a ∗ : H 1 ( X ; Z ) / Tor → H 1 ( A 0 ; Z ) a_*:H_1(X;\mathbb{Z})/\operatorname{Tor}\to H_1(A_0;\mathbb{Z}) a ∗ : H 1 ( X ; Z ) / Tor → H 1 ( A 0 ; Z ) is an isomorphism; see, for example, [17 Lemma 3.12] . Thus there exists a unique integral class λ 0 ∈ H 1 ( A 0 ; Z ) \lambda_0\in H^1(A_0;\mathbb{Z}) λ 0 ∈ H 1 ( A 0 ; Z ) such that a ∗ λ 0 = χ a^*\lambda_0=\chi a ∗ λ 0 = χ .
Let B ⊂ A 0 B\subset A_0 B ⊂ A 0 be the sum of all connected algebraic subgroups of A 0 A_0 A 0 where the restriction of λ 0 \lambda_0 λ 0 vanishes, which is a connected algebraic subgroup of A 0 A_0 A 0 and λ 0 ∣ B = 0 \lambda_0|_B=0 λ 0 ∣ B = 0 . Set A = A 0 / B A=A_0/B A = A 0 / B , and q : A 0 → A q:A_0\to A q : A 0 → A is the natural quotient map. Since B B B is connected, q ∗ : π 1 ( A 0 ) → π 1 ( A ) q_*:\pi_1(A_0)\to\pi_1(A) q ∗ : π 1 ( A 0 ) → π 1 ( A ) is surjective, and we have a unique (surjective) λ ∈ H 1 ( A ; Z ) \lambda\in H^1(A;\mathbb{Z}) λ ∈ H 1 ( A ; Z ) such that q ∗ λ = λ 0 q^*\lambda=\lambda_0 q ∗ λ = λ 0 . Set ψ = q ∘ a : X → A \psi=q\circ a:X\to A ψ = q ∘ a : X → A be the quasi-Albanese map associated to χ \chi χ . Then χ = ψ ∗ λ \chi=\psi^*\lambda χ = ψ ∗ λ .
X → a A 0 ψ ↓ ↓ q A → λ 0 Z λ ↓ Z \begin{CD}
X @>{a}>> A_0 \\
@V{\psi}VV @VV{q}V \\
A @>{\lambda_0}>> \mathbb{Z} \\
@V{\lambda}VV \\
\mathbb{Z}
\end{CD} X ψ ↓ ⏐ A λ ↓ ⏐ Z a λ 0 A 0 ↓ ⏐ q Z
We claim that λ \lambda λ is nonzero on every positive-dimensional connected algebraic subgroup of A A A . Suppose that H ⊂ A H\subset A H ⊂ A connected of positive-dimension, and λ ∣ H ≡ 0 \lambda|_H\equiv0 λ ∣ H ≡ 0 , then its inverse image H ~ = q − 1 ( H ) \widetilde{H}=q^{-1}(H) H = q − 1 ( H ) is a connected algebraic subgroup of A 0 A_0 A 0 containing B B B , and λ 0 ∣ H ~ = 0 \lambda_0|_{\widetilde{H}}=0 λ 0 ∣ H = 0 ; we have B ⊊ H ~ B\subsetneq\widetilde{H} B ⊊ H due to dim H > 0 \dim H>0 dim H > 0 , contradicting the definition of B B B . Thus we have shown
Lemma 3.6. — For every positive dimensional connected algebraic subgroup H ⊂ A H\subset A H ⊂ A , the restriction of λ \lambda λ to H H H is not identically zero.
Moreover, for the semi-abelian variety A A A , we have the following decomposition:
1 → T → A → A a b → 1 1\to T\to A\to A_{\mathrm{ab}}\to1 1 → T → A → A ab → 1
where T ≅ ( C ∗ ) ℓ T\cong(\mathbb{C}^*)^\ell T ≅ ( C ∗ ) ℓ is an algebraic torus and A a b A_{\mathrm{ab}} A ab is an abelian variety. For λ ∣ T ∈ H 1 ( T ; Z ) \lambda|_T\in H^1(T;\mathbb{Z}) λ ∣ T ∈ H 1 ( T ; Z ) , if it is identically zero, then by Theorem 3.6, we get ℓ = 0 \ell=0 ℓ = 0 ; otherwise, under the natural isomorphism H 1 ( T ; Z ) ≅ Hom ( π 1 ( T ) ; Z ) H^1(T;\mathbb{Z})\cong\operatorname{Hom}(\pi_1(T);\mathbb{Z}) H 1 ( T ; Z ) ≅ Hom ( π 1 ( T ) ; Z ) , assume that λ ∣ T \lambda|_T λ ∣ T is
λ ∣ T : ( a 1 , ⋯ , a ℓ ) ∈ Z ℓ ↦ a 1 n 1 + ⋯ + a ℓ n ℓ , ( n 1 , ⋯ , n ℓ ) ∈ Z ℓ not all zeros . \lambda|_T:(a_1,\cdots,a_\ell)\in\mathbb{Z}^{\ell}\mapsto a_1n_1+\cdots+a_\ell n_\ell,\qquad(n_1,\cdots,n_\ell)\in\mathbb{Z}^{\ell}\text{ not all zeros}. λ ∣ T : ( a 1 , ⋯ , a ℓ ) ∈ Z ℓ ↦ a 1 n 1 + ⋯ + a ℓ n ℓ , ( n 1 , ⋯ , n ℓ ) ∈ Z ℓ not all zeros .
Consider the associated algebraic character τ : T → C ∗ \tau:T\to\mathbb{C}^* τ : T → C ∗ given by τ ( z 1 , … , z ℓ ) = z 1 n 1 ⋯ z ℓ n ℓ \tau(z_1,\ldots,z_\ell)=z_1^{n_1}\cdots z_\ell^{n_\ell} τ ( z 1 , … , z ℓ ) = z 1 n 1 ⋯ z ℓ n ℓ , then τ ∗ : π 1 ( T ) → Z \tau_*:\pi_1(T)\to\mathbb{Z} τ ∗ : π 1 ( T ) → Z equals λ ∣ T \lambda|_T λ ∣ T . If ℓ > 1 \ell>1 ℓ > 1 , the identity component H = ( ker τ ) 0 H=(\ker\tau)^0 H = ( ker τ ) 0 is a positive-dimensional connected algebraic subgroup of dimension ℓ − 1 \ell-1 ℓ − 1 , and λ ∣ H ≡ 0 \lambda|_H\equiv0 λ ∣ H ≡ 0 , contradicting Theorem 3.6. We therefore obtain:
Lemma 3.7. — The torus part of A A A has dimension ℓ ≤ 1 \ell\leq1 ℓ ≤ 1 .
Choice of logarithmic holomorphic one-form
In case ℓ = 1 \ell=1 ℓ = 1 , we can write A = L × A=L^\times A = L × with L L L a topological trivial line bundle (due to its translation-invariance) with fiber C ∗ \mathbb{C}^* C ∗ on A a b A_{\mathrm{ab}} A ab . Set
A ‾ = P A a b ( O A a b ⊕ L ) , D A : = A ‾ ∖ A = D 0 ⊔ D ∞ (4) \overline{A}=\mathbb{P}_{A_{\mathrm{ab}}}(\mathcal{O}_{A_{\mathrm{ab}}}\oplus L),\qquad D_A:=\overline{A}\setminus A=D_0\sqcup D_\infty \tag*{(4)} A = P A ab ( O A ab ⊕ L ) , D A := A ∖ A = D 0 ⊔ D ∞ (4)
with D 0 D_0 D 0 and D ∞ D_\infty D ∞ are zero and infinity sections respectively. For ℓ = 0 \ell=0 ℓ = 0 , we take A ‾ = A \overline{A}=A A = A and D A = ∅ D_A=\varnothing D A = ∅ .
Recall that the mixed Hodge structure [10 ] on H 1 ( A ; C ) H^1(A;\mathbb{C}) H 1 ( A ; C ) has only Hodge types ( 1 , 0 ) (1,0) ( 1 , 0 ) , ( 0 , 1 ) (0,1) ( 0 , 1 ) , and ( 1 , 1 ) (1,1) ( 1 , 1 ) . The weight-one part is pulled back from the abelian quotient A a b A_{\mathrm{ab}} A ab , while the weight-two part is contributed by the torus T T T . Consequently the real-part map
F 1 H 1 ( A ; C ) → H 1 ( A ; R ) , α ↦ Re α , F^1H^1(A;\mathbb{C})\to H^1(A;\mathbb{R}),\qquad\alpha\mapsto\operatorname{Re}\alpha, F 1 H 1 ( A ; C ) → H 1 ( A ; R ) , α ↦ Re α ,
is surjective. We may now apply Deligne’s logarithmic de Rham theory [10 ] to choose a logarithmic holomorphic one-form whose real part represents λ \lambda λ on A A A and has controlled behavior along the boundary divisor D A D_A D A .
Lemma 3.8. — (a) There is a closed algebraic one-form ω \omega ω on A A A such that [ Re ω ] = λ [\operatorname{Re}\omega]=\lambda [ Re ω ] = λ , and ω \omega ω extends logarithmically to the standard projective compactification of A A A with purely imaginary residues.
(b) For every nonconstant algebraic map u : C → A u:\mathbb{C}\to A u : C → A from a smooth algebraic curve, u ∗ ω ≢ 0 u^*\omega\not\equiv0 u ∗ ω ≡ 0 .
Proof. — (a) In case ℓ = 0 \ell=0 ℓ = 0 , we take ω ∈ H 0 ( A ; Ω A 1 ) \omega\in H^0(A;\Omega_A^1) ω ∈ H 0 ( A ; Ω A 1 ) a holomorphic 1 1 1 -form such that [ Re ω ] = λ [\operatorname{Re}\omega]=\lambda [ Re ω ] = λ . In case ℓ = 1 \ell=1 ℓ = 1 , consider
1 → C ∗ → A → π A a b → 1 , 1\to\mathbb{C}^*\to A\xrightarrow{\pi}A_{\mathrm{ab}}\to1, 1 → C ∗ → A π A ab → 1 ,
we have W 1 H 1 ( A ; C ) = p ∗ H 1 ( B ; C ) W_1H^1(A;\mathbb{C})=p^*H^1(B;\mathbb{C}) W 1 H 1 ( A ; C ) = p ∗ H 1 ( B ; C ) and Gr 2 W H 1 ( A ; C ) ≅ C [ − 1 ] \operatorname{Gr}_2^W H^1(A;\mathbb{C})\cong\mathbb{C}[-1] Gr 2 W H 1 ( A ; C ) ≅ C [ − 1 ] . Consider the residue sequence
0 → Ω A ‾ 1 → Ω A ‾ 1 ( log D A ) → Res O D 0 ⊕ O D ∞ → 0 , 0\to\Omega_{\overline{A}}^1\to\Omega_{\overline{A}}^1(\log D_A)\xrightarrow{\operatorname{Res}}\mathcal{O}_{D_0}\oplus\mathcal{O}_{D_\infty}\to0, 0 → Ω A 1 → Ω A 1 ( log D A ) Res O D 0 ⊕ O D ∞ → 0 ,
then the toric part C ∗ C^* C ∗ is detected by the residue map
Res : H 0 ( A ‾ , Ω A ‾ 1 ( log D A ) ) → { ( c , − c ) ∈ C 2 : c ∈ C } . \operatorname{Res}: H^0\left(\overline{A},\Omega^1_{\overline{A}}(\log D_A)\right)\to\{(c,-c)\in\mathbb{C}^2:c\in\mathbb{C}\}. Res : H 0 ( A , Ω A 1 ( log D A ) ) → {( c , − c ) ∈ C 2 : c ∈ C } .
It is standard that there exists a global logarithmic 1-form η \eta η on A A A extending logarithmically to A ‾ \overline{A} A and satisfying
Res D 0 η = 1 , Res D ∞ η = − 1. \operatorname{Res}_{D_0}\eta=1,\qquad\operatorname{Res}_{D_\infty}\eta=-1. Res D 0 η = 1 , Res D ∞ η = − 1.
Locally near a point of D 0 D_0 D 0 , if z z z is local defining function of D 0 D_0 D 0 , we can write η = d z z + β \eta=\frac{dz}{z}+\beta η = z d z + β , with β \beta β holomorphic.
Let γ 0 \gamma_0 γ 0 be a small positively oriented loop around D 0 D_0 D 0 , and set λ ( γ 0 ) = n ∈ Z \lambda(\gamma_0)=n\in\mathbb{Z} λ ( γ 0 ) = n ∈ Z , we have
∫ μ n 2 π − 1 η = n = λ ( μ ) \int_{\mu}\frac{n}{2\pi\sqrt{-1}}\eta=n=\lambda(\mu) ∫ μ 2 π − 1 n η = n = λ ( μ )
and λ − [ Re n 2 π − 1 η ] \lambda-\left[\operatorname{Re}\frac{n}{2\pi\sqrt{-1}}\eta\right] λ − [ Re 2 π − 1 n η ] has trivial image in Gr 2 W ( H 1 ( A ; C ) ) \operatorname{Gr}^{W}_{2}(H^1(A;\mathbb{C})) Gr 2 W ( H 1 ( A ; C )) , thus there exists λ a b ∈ H 1 ( A a b ; R ) \lambda_{\mathrm{ab}}\in H^1(A_{\mathrm{ab}};\mathbb{R}) λ ab ∈ H 1 ( A ab ; R ) such that
λ − [ Re n 2 π − 1 η ] = π ∗ λ a b . \lambda-\left[\operatorname{Re}\frac{n}{2\pi\sqrt{-1}}\eta\right]=\pi^*\lambda_{\mathrm{ab}}. λ − [ Re 2 π − 1 n η ] = π ∗ λ ab .
Choose a holomorphic one form ω a b \omega_{\mathrm{ab}} ω ab such that [ Re ω a b ] = π ∗ λ a b [\operatorname{Re}\omega_{\mathrm{ab}}]=\pi^*\lambda_{\mathrm{ab}} [ Re ω ab ] = π ∗ λ ab , then
ω = n 2 π − 1 η + ω a b (5) \omega=\frac{n}{2\pi\sqrt{-1}}\eta+\omega_{\mathrm{ab}} \tag*{(5)} ω = 2 π − 1 n η + ω ab (5)
is a logarithmic one form on A A A such that [ Re ω ] = λ [\operatorname{Re}\omega]=\lambda [ Re ω ] = λ , and ω \omega ω has purely imaginary residues at divisors D 0 ⊔ D ∞ D_0\sqcup D_\infty D 0 ⊔ D ∞ :
Res D 0 ω = n 2 π − 1 , Res D ∞ ω = − n 2 π − 1 . \operatorname{Res}_{D_0}\omega=\frac{n}{2\pi\sqrt{-1}},\qquad\operatorname{Res}_{D_\infty}\omega=-\frac{n}{2\pi\sqrt{-1}}. Res D 0 ω = 2 π − 1 n , Res D ∞ ω = − 2 π − 1 n .
(b) Assume that u ∗ ω = 0 u^*\omega=0 u ∗ ω = 0 . Up to translation, the universal property of the quasi-Albanese morphism [17 Theorem 1.1] gives a factorization u = ϕ ∘ a C u=\phi\circ a_C u = ϕ ∘ a C , where ϕ : Alb ( C ) → A \phi:\operatorname{Alb}(C)\to A ϕ : Alb ( C ) → A is a homomorphism and a C a_C a C is the quasi-Albanese morphism of C C C . Let H = im ( ϕ ) H=\operatorname{im}(\phi) H = im ( ϕ ) . Since u u u is nonconstant, H H H is a positive-dimensional connected semi-abelian subvariety of A A A . The map Alb ( C ) → H \operatorname{Alb}(C)\to H Alb ( C ) → H induces a surjection on H 1 ( − ; R ) H_1(-;\mathbb{R}) H 1 ( − ; R ) , and a C a_C a C also induces a surjection on H 1 H_1 H 1 . Since u ∗ λ = [ Re ( u ∗ ω ) ] = 0 u^*\lambda=[\operatorname{Re}(u^*\omega)]=0 u ∗ λ = [ Re ( u ∗ ω )] = 0 , it follows that λ ∣ H = 0 \lambda|_H=0 λ ∣ H = 0 , contradicting Theorem 3.6.
Remark 3.9. — Whenever we use the expression eq. (5) , we have n ≠ 0 n\ne0 n = 0 . Indeed, in the case ℓ = 1 \ell=1 ℓ = 1 , the torus T ≃ C ∗ T\simeq\mathbb{C}^* T ≃ C ∗ is a positive-dimensional connected algebraic subgroup of A A A , and hence λ ∣ T ≠ 0 \lambda|_T\ne0 λ ∣ T = 0 by Theorem 3.6. Thus the corresponding residues of ω \omega ω along D 0 D_0 D 0 is nonzero.
Remark 3.10. — A good choice such as a logarithmic holomorphic one-form is essential for the argument below, since an arbitrary holomorphic 1-form on a smooth quasi-projective variety may have zeros accumulating toward infinity. For example, the holomorphic one-form sin ( 1 / z ) d z \sin(1/z)\,dz sin ( 1/ z ) d z on C ∗ \mathbb{C}^* C ∗ has zeros at z = 1 / ( k π ) z=1/(k\pi) z = 1/ ( k π ) , k ∈ Z ∖ { 0 } k\in\mathbb{Z}\setminus\{0\} k ∈ Z ∖ { 0 } , which accumulate at 0.
Such accumulation prevents us, in general, from choosing critical neighborhoods with a uniform positive width in § 3.4 , and therefore creates a serious difficulty for the flow argument used in § 3.5 . The same difficulty also obstructs a direct extension of Simpson’s global picture and patching argument to the noncompact setting, since that construction likewise requires uniform control of the critical neighborhoods.
Reduction to surfaces and real-oriented blowup
As in [38 ] , we consider the following two cases: dim ψ ( X ) = 1 \dim\psi(X)=1 dim ψ ( X ) = 1 and dim ψ ( X ) ≥ 2 \dim\psi(X)\ge2 dim ψ ( X ) ≥ 2 . They will be corresponding to (i) and (ii) of Theorem 3.1 correspondingly. For now, we consider first the case dim ψ ( X ) ≥ 2 \dim\psi(X)\ge2 dim ψ ( X ) ≥ 2 .
We may first reduce to the case where X X X is a smooth quasi-projective surface as follows. By taking a general complete intersection of hyperplanes, due to the Lefschetz hyperplane theorem (for example, see [HT85, Theorem 1.1.3]) we obtain a smooth quasi-projective surface Y ⊂ X Y\subset X Y ⊂ X such that the inclusion induces an isomorphism:
π 1 ( Y ) → ≃ π 1 ( X ) . (6) \pi_1(Y)\xrightarrow{\simeq}\pi_1(X). \tag*{(6)} π 1 ( Y ) ≃ π 1 ( X ) . (6)
Moreover, since dim ψ ( X ) ≥ 2 \dim\psi(X) \ge2 dim ψ ( X ) ≥ 2 , the hyperplanes may be chosen sufficiently generally so that, at a general point x ∈ Y x \in Y x ∈ Y at which d ψ x d\psi_x d ψ x has maximal rank, T x Y ∩ ker ( d ψ x ) = 0 T_xY \cap\ker(d\psi_x) = 0 T x Y ∩ ker ( d ψ x ) = 0 . Thus dim ψ ( Y ) = 2 \dim\psi(Y) = 2 dim ψ ( Y ) = 2 and ψ ∣ Y \psi|_Y ψ ∣ Y is generically finite onto its image. More precisely, we have
Lemma 3.11. — Let X X X be a smooth connected complex quasi-projective variety, and let ψ : X → A \psi: X \to A ψ : X → A be an algebraic morphism with dim ψ ( X ) ≥ 2 \dim\psi(X) \ge2 dim ψ ( X ) ≥ 2 . For every finite subset S ⊂ X S \subset X S ⊂ X , there exists a smooth connected quasi-projective surface Y ⊂ X Y \subset X Y ⊂ X containing S S S such that
π 1 ( Y ) → ∼ π 1 ( X ) and dim ψ ( Y ) = 2. \pi_1(Y) \xrightarrow{\sim} \pi_1(X) \quad\text{and} \quad\dim\psi(Y) = 2. π 1 ( Y ) ∼ π 1 ( X ) and dim ψ ( Y ) = 2.
We omit the standard proof, which uses a suitable very ample line bundle together with Bertini’s theorem and the quasi-projective Lefschetz hyperplane theorem.
We may choose smooth projective compactifications Y ⊂ Y ‾ Y \subset\overline{Y} Y ⊂ Y and A ⊂ A ‾ A \subset\overline{A} A ⊂ A such that ψ : Y → A \psi: Y \to A ψ : Y → A extends to a morphism ψ ‾ : Y ‾ → A ‾ \overline{\psi} : \overline{Y} \to\overline{A} ψ : Y → A , and such that
D A = A ‾ ∖ A , D Y = Y ‾ ∖ Y , P = ( ψ ‾ − 1 ( D A ) ) r e d D_A = \overline{A} \setminus A,\quad D_Y = \overline{Y} \setminus Y,\quad P = \left(\overline{\psi}^{-1}(D_A)\right)_{\mathrm{red}} D A = A ∖ A , D Y = Y ∖ Y , P = ( ψ − 1 ( D A ) ) red
are simple normal crossings divisors, with P P P a union of irreducible components of D Y D_Y D Y . Let Q Q Q be the reduced union of the remaining irreducible components of D Y D_Y D Y , so that D Y = P ∪ Q D_Y = P \cup Q D Y = P ∪ Q .
Since ψ : Y → A \psi: Y \to A ψ : Y → A is generically finite onto its two-dimensional image, the extension ψ ‾ : Y ‾ → A ‾ \overline{\psi} : \overline{Y} \to\overline{A} ψ : Y → A is also generically finite onto its image. Let Y ∘ = Y ‾ ∖ P Y^\circ= \overline{Y} \setminus P Y ∘ = Y ∖ P and ψ ∘ = ψ ‾ ∣ Y ∘ \psi^\circ= \overline{\psi}|_{Y^\circ} ψ ∘ = ψ ∣ Y ∘ . Let E E E denote the union of the positive-dimensional fibers of ψ ∘ : Y ∘ → A \psi^\circ: Y^\circ\to A ψ ∘ : Y ∘ → A . Then E E E is a finite union of irreducible curves in Y ∘ Y^\circ Y ∘ , and it contains both the closures in Y ∘ Y^\circ Y ∘ of the exceptional curves of ψ \psi ψ in Y Y Y and any irreducible component of Q Q Q contracted by ψ ‾ \overline{\psi} ψ . We call any irreducible curve C ⊂ E C \subset E C ⊂ E an exceptional curve , and call E E E the exceptional set .
Due to Theorem 3.8(b), we have
Lemma 3.12. — For any irreducible curve C ⊂ Y ∘ C \subset Y^\circ C ⊂ Y ∘ such that C ⊄ E C \not\subset E C ⊂ E ,
ψ ‾ ∗ ω ∣ C ≢ 0. \overline{\psi}^{*}\omega|_C \not\equiv0. ψ ∗ ω ∣ C ≡ 0.
Recall that for the logarithmic holomorphic 1-form ω \omega ω on A A A defined in section 3.2, locally near a smooth point of a divisor D 0 ⊂ D A D_0 \subset D_A D 0 ⊂ D A written as { z = 0 } \{z = 0\} { z = 0 } , it has the form
ω = n 2 π − 1 d z z + d h \omega= \frac{n}{2\pi\sqrt{-1}}\frac{dz}{z} + dh ω = 2 π − 1 n z d z + d h
where h h h is non-vanishing holomorphic and n ≠ 0 n \ne0 n = 0 (see Theorem 3.9).
For x 0 ∈ A x_0 \in A x 0 ∈ A fixed, define a function on A A A :
Φ A ( x ) = exp ( 2 π − 1 ∫ x 0 x Re ω ) , ∀ x ∈ A , \Phi_A(x) = \exp\left(2\pi\sqrt{-1}\int_{x_0}^{x}\operatorname{Re}\omega\right),\quad\forall x \in A, Φ A ( x ) = exp ( 2 π − 1 ∫ x 0 x Re ω ) , ∀ x ∈ A ,
which is well defined because λ = [ Re ω ] ∈ H 1 ( A ; Z ) \lambda= [\operatorname{Re}\omega] \in H^1(A;\mathbb{Z}) λ = [ Re ω ] ∈ H 1 ( A ; Z ) is integral. Usually, this function cannot be extended to A ‾ \overline{A} A , for near D 0 D_0 D 0 we have
Re ω = Re ( n 2 π − 1 ( d r r + − 1 d θ ) + d h ) = n 2 π d θ + Re ( d h ) , \operatorname{Re}\omega= \operatorname{Re}\left(\frac{n}{2\pi\sqrt{-1}}\left(\frac{dr}{r}+\sqrt{-1}d\theta\right)+dh\right)=\frac{n}{2\pi}d\theta+\operatorname{Re}(dh), Re ω = Re ( 2 π − 1 n ( r d r + − 1 d θ ) + d h ) = 2 π n d θ + Re ( d h ) ,
where z = r e − 1 θ z = r e^{\sqrt{-1}\theta} z = r e − 1 θ and x 0 x_0 x 0 is fixed (here we take 1 for example), and
Φ A ( x ) = exp ( − 1 n θ ) ⋅ exp ( 2 π − 1 ∫ x 0 x Re ( d h ) ) . \Phi_A(x) = \exp(\sqrt{-1}n\theta)\cdot\exp\left(2\pi\sqrt{-1}\int_{x_0}^{x}\operatorname{Re}(dh)\right). Φ A ( x ) = exp ( − 1 n θ ) ⋅ exp ( 2 π − 1 ∫ x 0 x Re ( d h ) ) .
When r r r tends to 0, the second factor tends to a limit while the first factor is not well defined on A ‾ \overline{A} A . To remedy this, we use the real-oriented blowup; then the function Φ A \Phi_A Φ A extends to the boundary. In short, the real-oriented blowup allows the angular coordinate θ \theta θ to remain well defined even when r = 0 r = 0 r = 0 . Of course, the resulting manifold is no longer a complex manifold in general. This construction is also the Kato–Nakayama space [21 (1.2.3)] associated with the log structure defined by the SNC divisor P P P . For simplicity, in our special case we describe the local construction and gluing process explicitly below. If the torus part is trivial in A A A , i.e., ℓ = 0 \ell= 0 ℓ = 0 , then A ‾ = A \overline{A} = A A = A , D A = P = ∅ D_A = P = \varnothing D A = P = ∅ , and M = Y ‾ M = \overline{Y} M = Y . For the following, we assume ℓ = 1 \ell= 1 ℓ = 1 .
Since P ⊂ Y ‾ P \subset\overline{Y} P ⊂ Y is a simple normal crossing divisor, we define its real-oriented blowup
M = Bl P r o ( Y ‾ ) , β : M → Y ‾ M = \operatorname{Bl}^{\mathrm{ro}}_P(\overline{Y}),\quad\beta: M \to\overline{Y} M = Bl P ro ( Y ) , β : M → Y
locally as follows. If P = { z 1 ⋯ z k = 0 } P=\{z_1\cdots z_k=0\} P = { z 1 ⋯ z k = 0 } , k = 1 , 2 k=1,2 k = 1 , 2 , write z j = r j e − 1 θ j z_j=r_j e^{\sqrt{-1}\theta_j} z j = r j e − 1 θ j , with r j ≥ 0 r_j\ge0 r j ≥ 0 , and replace each normal coordinate z j z_j z j by the pair ( r j , θ j ) (r_j,\theta_j) ( r j , θ j ) . Thus locally
M ≅ ( R ≥ 0 × S 1 ) k × C 2 − k , M\cong(\mathbb{R}_{\ge0}\times S^1)^k\times\mathbb{C}^{2-k}, M ≅ ( R ≥ 0 × S 1 ) k × C 2 − k ,
where S 1 S^1 S 1 is the unit circle in C \mathbb{C} C . These local models glue canonically. Indeed, on the overlap of two SNC coordinate charts, after possibly reordering, their defining coordinates satisfy z ~ j = u j z j \widetilde z_j=u_jz_j z j = u j z j for nowhere-vanishing holomorphic functions u j u_j u j . Hence, writing z ~ j = r j ′ e − 1 θ j ′ \widetilde z_j=r'_j e^{\sqrt{-1}\theta'_j} z j = r j ′ e − 1 θ j ′ ,
r j ′ = ∣ u j ∣ r j , e − 1 θ j ′ = u j ∣ u j ∣ e − 1 θ j ∈ S 1 . r'_j=|u_j|r_j,\qquad e^{\sqrt{-1}\theta'_j}=\frac{u_j}{|u_j|}e^{\sqrt{-1}\theta_j}\in S^1. r j ′ = ∣ u j ∣ r j , e − 1 θ j ′ = ∣ u j ∣ u j e − 1 θ j ∈ S 1 .
These transition maps extend smoothly to r j = 0 r_j=0 r j = 0 , and therefore the local oriented blowups glue to a manifold with boundary. The boundary with corners can be described as follows. In a coordinate neighborhood near { z 1 z 2 = 0 } \{z_1z_2=0\} { z 1 z 2 = 0 } , the boundary is of the form { r 1 = 0 } ∪ { r 2 = 0 } \{r_1=0\}\cup\{r_2=0\} { r 1 = 0 } ∪ { r 2 = 0 } , and the corner C C C is of the form { r 1 = r 2 = 0 } \{r_1=r_2=0\} { r 1 = r 2 = 0 } and C ≅ ( S 1 ) 2 C\cong(S^1)^2 C ≅ ( S 1 ) 2 .
Composing with ψ ˉ : Y ˉ → A ˉ \bar\psi:\bar Y\to\bar A ψ ˉ : Y ˉ → A ˉ , we get Φ : Y ∘ → S 1 \Phi:Y^\circ\to S^1 Φ : Y ∘ → S 1 defined by
Φ ( x ) = exp ( 2 π − 1 ∫ x 0 x Re ψ ˉ ∗ ω ) . \Phi(x)=\exp\left(2\pi\sqrt{-1}\int_{x_0}^{x}\operatorname{Re}\bar\psi^*\omega\right). Φ ( x ) = exp ( 2 π − 1 ∫ x 0 x Re ψ ˉ ∗ ω ) .
On Y Y Y , this is simply
Φ ( x ) = exp ( 2 π − 1 ∫ x 0 x Re ψ ∗ ω ) . \Phi(x)=\exp\left(2\pi\sqrt{-1}\int_{x_0}^{x}\operatorname{Re}\psi^*\omega\right). Φ ( x ) = exp ( 2 π − 1 ∫ x 0 x Re ψ ∗ ω ) .
Proposition 3.13. — The S 1 S^1 S 1 -valued map Φ : Y ∘ → S 1 \Phi:Y^\circ\to S^1 Φ : Y ∘ → S 1 extends smoothly to the boundary of M M M . Moreover, its restriction to each boundary component β − 1 ( P ) \beta^{-1}(P) β − 1 ( P ) and each boundary corner of M M M is a submersion.
Proof. — For the case ℓ = 0 \ell=0 ℓ = 0 and M = Y ˉ M=\bar Y M = Y ˉ , there is nothing to prove. For the case ℓ = 1 \ell=1 ℓ = 1 , by eq. (4) , take x ∈ P x\in P x ∈ P with ψ ˉ ( x ) ∈ D A \bar\psi(x)\in D_A ψ ˉ ( x ) ∈ D A ; without loss of generality, assume ψ ˉ ( x ) ∈ D 0 \bar\psi(x)\in D_0 ψ ˉ ( x ) ∈ D 0 . Let t t t be a local coordinate around ψ ˉ ( x ) \bar\psi(x) ψ ˉ ( x ) such that D 0 = { t = 0 } D_0=\{t=0\} D 0 = { t = 0 } and { z 1 , z 2 } \{z_1,z_2\} { z 1 , z 2 } be local coordinates around x x x , with P = { z 1 ⋯ z k = 0 } P=\{z_1\cdots z_k=0\} P = { z 1 ⋯ z k = 0 } , k = 1 , 2 k=1,2 k = 1 , 2 . Then locally we have
ψ ˉ ∗ t = v z 1 n 1 ⋯ z k n k \bar\psi^*t=vz_1^{n_1}\cdots z_k^{n_k} ψ ˉ ∗ t = v z 1 n 1 ⋯ z k n k
with v v v a non-vanishing holomorphic function, and n j n_j n j , 1 ≤ j ≤ k 1\le j\le k 1 ≤ j ≤ k , positive integers. Recall that from eq. (5) , locally we write
ω = ± n 2 π − 1 d t t + ω a b , n ≠ 0 , \omega=\pm\frac{n}{2\pi\sqrt{-1}}\frac{dt}{t}+\omega_{ab},\qquad n\ne0, ω = ± 2 π − 1 n t d t + ω ab , n = 0 ,
with ± \pm ± for D 0 D_0 D 0 and D ∞ D_\infty D ∞ , respectively. Then
ψ ˉ ∗ ω = ± n 2 π − 1 ( d v v + ∑ j = 1 k n j d z j z j ) + ψ ˉ ∗ ω a b . (7) \bar\psi^*\omega=\pm\frac{n}{2\pi\sqrt{-1}}\left(\frac{dv}{v}+\sum_{j=1}^{k}n_j\frac{dz_j}{z_j}\right)+\bar\psi^*\omega_{ab}. \tag*{(7)} ψ ˉ ∗ ω = ± 2 π − 1 n ( v d v + j = 1 ∑ k n j z j d z j ) + ψ ˉ ∗ ω ab . (7)
The part ± n 2 π − 1 d v v + ψ ˉ ∗ ω a b \pm\frac{n}{2\pi\sqrt{-1}}\frac{dv}{v}+\bar\psi^*\omega_{ab} ± 2 π − 1 n v d v + ψ ˉ ∗ ω ab is a closed holomorphic 1 1 1 -form on Y ˉ \bar Y Y ˉ near x x x , which can be written as d g dg d g locally with g g g a holomorphic function. Then
Φ = c exp ( − 1 ∑ j = 1 k m j θ j ) exp ( 2 π − 1 Re g ) , (8) \Phi=c\exp\left(\sqrt{-1}\sum_{j=1}^{k}m_j\theta_j\right)\exp(2\pi\sqrt{-1}\operatorname{Re}g), \tag*{(8)} Φ = c exp ( − 1 j = 1 ∑ k m j θ j ) exp ( 2 π − 1 Re g ) , (8)
with m j = ± n n j ≠ 0 m_j=\pm nn_j\ne0 m j = ± n n j = 0 and c ∈ S 1 c\in S^1 c ∈ S 1 a constant depending on the base point x 0 x_0 x 0 . This is a smooth function in the polar coordinates ( r 1 , θ 1 ; … ; r k , θ k ) (r_1,\theta_1;\ldots;r_k,\theta_k) ( r 1 , θ 1 ; … ; r k , θ k ) , k = 1 , 2 k=1,2 k = 1 , 2 , with r j ≥ 0 r_j\ge0 r j ≥ 0 , and therefore extends smoothly to the boundary of M M M .
It remains to show that the extended map does not have any critical point on the boundary β − 1 ( P ) ⊂ M \beta^{-1}(P)\subset M β − 1 ( P ) ⊂ M . Away from its corner, we can write its local equation as r j = 0 r_j=0 r j = 0 , then we have
∂ ∂ θ j Re ( g ) = 1 2 ( ∂ g ∂ z j − 1 z j + ∂ g ˉ ∂ z ˉ j ( − − 1 z ˉ j ) ) = Re ( ∂ g ∂ z j − 1 z j ) = 0 , when r j = 0 , \frac{\partial}{\partial\theta_j}\operatorname{Re}(g)=\frac{1}{2}\left(\frac{\partial g}{\partial z_j}\sqrt{-1}z_j+\frac{\partial\bar g}{\partial\bar z_j}(-\sqrt{-1}\bar z_j)\right)=\operatorname{Re}\left(\frac{\partial g}{\partial z_j}\sqrt{-1}z_j\right)=0,\qquad\text{when }r_j=0, ∂ θ j ∂ Re ( g ) = 2 1 ( ∂ z j ∂ g − 1 z j + ∂ z ˉ j ∂ g ˉ ( − − 1 z ˉ j ) ) = Re ( ∂ z j ∂ g − 1 z j ) = 0 , when r j = 0 ,
thus
∂ ∂ θ j ( arg Φ ) = m j ≠ 0. \frac{\partial}{\partial\theta_j}(\arg\Phi)=m_j\ne0. ∂ θ j ∂ ( arg Φ ) = m j = 0.
Thus Φ \Phi Φ is a submersion on each boundary away from the corner. Near the corner locally defined by r 1 = r 2 = 0 r_1=r_2=0 r 1 = r 2 = 0 , we have
∂ ∂ θ 1 ( arg Φ ) = m 1 ≠ 0 , ∂ ∂ θ 2 ( arg Φ ) = m 2 ≠ 0. \frac{\partial}{\partial\theta_1}(\arg\Phi)=m_1\ne0,\qquad\frac{\partial}{\partial\theta_2}(\arg\Phi)=m_2\ne0. ∂ θ 1 ∂ ( arg Φ ) = m 1 = 0 , ∂ θ 2 ∂ ( arg Φ ) = m 2 = 0.
Any one of these two equations implies that Φ \Phi Φ restricted to the corner is a submersion.
Remark 3.14. — The use of the real-oriented blowup along the polar divisor P P P is convenient but not essential for the argument. One may instead work directly on Y ∘ Y^\circ Y ∘ and equip it with a complete metric of Poincaré type near P P P . In local coordinates ( z 1 , z 2 ) (z_1,z_2) ( z 1 , z 2 ) , if P = { z 1 = 0 } P=\{z_1=0\} P = { z 1 = 0 } , a Poincaré-type metric can be written as
g P ∼ ∣ d z 1 ∣ 2 ∣ z 1 ∣ 2 ( log ∣ z 1 ∣ ) 2 + ∣ d z 2 ∣ 2 . g_P\sim\frac{|dz_1|^2}{|z_1|^2(\log|z_1|)^2}+|dz_2|^2. g P ∼ ∣ z 1 ∣ 2 ( log ∣ z 1 ∣ ) 2 ∣ d z 1 ∣ 2 + ∣ d z 2 ∣ 2 .
Hence one may choose the vector field used in § 3.5 to have uniformly bounded norm and to satisfy d arg Φ ( v ) = 2 π d\arg\Phi(\mathbf{v})=2\pi d arg Φ ( v ) = 2 π away from the critical neighborhoods defined in § 3.4 . Since a uniformly bounded smooth vector field on a complete Riemannian manifold has complete flow ([41 Proposition 1.4.4(v)] ), no integral curve can reach the divisor P P P in finite time. Thus the role played above by the submersion of Φ \Phi Φ on the boundary of β − 1 ( P ) \beta^{-1}(P) β − 1 ( P ) can be replaced by completeness with respect to a Poincaré-type metric.
Construction of two types of critical neighborhoods and local 1-connectedness
3.4. Construction of two types of critical neighborhoods and local 1-connectedness. — Set
ξ : = ψ ˉ ∗ ω ∣ Y ∘ = : ( ψ ∘ ) ∗ ω , \xi:=\bar{\psi}^{*}\omega|_{Y^\circ}=:(\psi^\circ)^*\omega, ξ := ψ ˉ ∗ ω ∣ Y ∘ =: ( ψ ∘ ) ∗ ω ,
which is a holomorphic 1-form on Y ∘ Y^\circ Y ∘ . We now consider its zero set Z ( ξ ) ⊂ Y ∘ Z(\xi)\subset Y^\circ Z ( ξ ) ⊂ Y ∘ . By Theorem 3.12, every positive-dimensional irreducible component of Z ( ξ ) Z(\xi) Z ( ξ ) must be contained in E E E . Hence Z ( ξ ) ∖ E Z(\xi)\setminus E Z ( ξ ) ∖ E is zero-dimensional, which is a finite set of points. Write
Z ( ξ ) = A 1 ⊔ ⋯ ⊔ A n , Z(\xi)=A_1\sqcup\cdots\sqcup A_n, Z ( ξ ) = A 1 ⊔ ⋯ ⊔ A n ,
where each A i A_i A i is either a point or a connected union of irreducible curves contained in E E E .
Since we are now dealing with the non-compact case, we need to be extremely careful about the boundary. Recall from § 3.3 that
Y ˉ ∖ Y = P ∪ Q and Y ∘ = Y ˉ ∖ P = Y ∪ ( Q ∖ P ) . \bar{Y}\setminus Y=P\cup Q\qquad\text{and}\qquad Y^\circ=\bar{Y}\setminus P=Y\cup(Q\setminus P). Y ˉ ∖ Y = P ∪ Q and Y ∘ = Y ˉ ∖ P = Y ∪ ( Q ∖ P ) .
So far in the present section, we have been working on Y ∘ Y^\circ Y ∘ , while our ultimate goal is to obtain the corresponding conclusion on Y Y Y . Thus, in the flow construction later in § 3.5 , we need to ensure that the flow preserves the complement of Q Q Q , so that a trajectory starting in Y Y Y remains in Y Y Y (or for the suitable Z Z Z -coverings). This cannot always be achieved simply by requiring the flow to be tangent to Q Q Q . Write Q = Q 1 ∪ ⋯ ∪ Q l Q=Q_1\cup\cdots\cup Q_l Q = Q 1 ∪ ⋯ ∪ Q l as the union of its irreducible components. For each Q j Q_j Q j , there are two possibilities:
We need to deal with these two cases separately. For the first case, this was considered by Simpson in [38 Theorem 1] (defined as the deleted subset D D D there). For this part, we shall put these components together with Z ( ξ ) Z(\xi) Z ( ξ ) and construct the critical neighborhoods as done by Simpson; we will call these critical neighborhoods of type I . For the second case, the zero set of ξ ∣ Q j \left.\xi\right|_{Q_j} ξ ∣ Q j is discrete. Around each of these finitely many points, we choose a suitable neighborhood such that outside these neighborhoods we can choose the flow tangent to Q Q Q . We call these critical neighborhoods of type II .
Denote the union of the Q j Q_j Q j in the first case by Q ′ Q' Q ′ , and the union of those in the second case by Q ′ ′ Q'' Q ′′ . Set
Σ = Z ( ξ ) ∪ Q ′ \Sigma=Z(\xi)\cup Q' Σ = Z ( ξ ) ∪ Q ′
and decompose it into connected components as Σ = ⋃ i = 1 N Σ i \Sigma= \bigcup_{i=1}^{N} \Sigma_i Σ = ⋃ i = 1 N Σ i . Note that for every Q j ⊂ Q ′ Q_j \subset Q' Q j ⊂ Q ′ , the restriction ψ ˉ ∣ Q j ∩ Y ∘ \bar{\psi}|_{Q_j \cap Y^\circ} ψ ˉ ∣ Q j ∩ Y ∘ is constant by Theorem 3.12. Hence Q j ⊂ E Q_j \subset E Q j ⊂ E and ψ ˉ ( Q j ) = { a j } \bar{\psi}(Q_j) = \{a_j\} ψ ˉ ( Q j ) = { a j } for some a j ∈ A a_j \in A a j ∈ A . In particular, Q ′ ∩ P = ∅ Q' \cap P = \varnothing Q ′ ∩ P = ∅ , and the first condition is reduced to ξ ∣ Q j ≡ 0 \xi|_{Q_j} \equiv0 ξ ∣ Q j ≡ 0 .
Since Theorem 3.13 implies that ψ ˉ ∗ ω \bar{\psi}^*\omega ψ ˉ ∗ ω has no zero in a neighborhood of the polar divisor P P P , the zero locus of ξ = ψ ˉ ∗ ω ∣ Y ∘ \xi= \bar{\psi}^*\omega|_{Y^\circ} ξ = ψ ˉ ∗ ω ∣ Y ∘ is contained in a compact subset of Y ∘ Y^\circ Y ∘ . Thus the problem of constructing critical neighborhoods of type I around Σ \Sigma Σ reduces to the compact situation considered in [38 ] .
For each Σ i \Sigma_i Σ i , let a i = ψ ∘ ( Σ i ) ∈ A a_i = \psi^\circ(\Sigma_i) \in A a i = ψ ∘ ( Σ i ) ∈ A . Choose a sufficiently small simply connected holomorphic coordinate neighborhood B i B_i B i of a i a_i a i in A A A . Since ω \omega ω is closed, there exists a holomorphic function h i h_i h i on B i B_i B i such that d h i = ω dh_i = \omega d h i = ω and h i ( a i ) = 0 h_i(a_i) = 0 h i ( a i ) = 0 . Choose pairwise disjoint relatively compact neighborhoods V i ′ V'_i V i ′ of Σ i \Sigma_i Σ i such that
V i ′ ‾ ⊂ ( ψ ∘ ) − 1 ( B i ) and V i ′ ‾ ∩ Σ = Σ i . \overline{V'_i} \subset(\psi^\circ)^{-1}(B_i) \quad\text{and} \quad\overline{V'_i} \cap\Sigma= \Sigma_i. V i ′ ⊂ ( ψ ∘ ) − 1 ( B i ) and V i ′ ∩ Σ = Σ i .
Define f i = h i ∘ ψ ∘ : V i ′ → C f_i = h_i \circ\psi^\circ: V'_i \to\mathbb{C} f i = h i ∘ ψ ∘ : V i ′ → C . Then d f i = ξ ∣ V i ′ df_i = \xi|_{V'_i} d f i = ξ ∣ V i ′ and f i ∣ Σ i = 0 f_i|_{\Sigma_i} = 0 f i ∣ Σ i = 0 .
Choose a smooth nonnegative function ρ i \rho_i ρ i on V i ′ V'_i V i ′ such that ρ i − 1 ( 0 ) = Σ i \rho_i^{-1}(0) = \Sigma_i ρ i − 1 ( 0 ) = Σ i and its sufficiently small sublevel sets are compactly contained in V i ′ V'_i V i ′ .1 Since all the critical points of f i f_i f i in V i ′ V'_i V i ′ belong to Σ i \Sigma_i Σ i , the set f i − 1 ( 0 ) ∖ Σ i f_i^{-1}(0) \setminus\Sigma_i f i − 1 ( 0 ) ∖ Σ i is smooth. By Sard's theorem, we may choose ε > 0 \varepsilon> 0 ε > 0 sufficiently small to be a regular value both of ρ i \rho_i ρ i and of its restriction to f i − 1 ( 0 ) ∖ Σ i f_i^{-1}(0) \setminus\Sigma_i f i − 1 ( 0 ) ∖ Σ i , simultaneously for all i i i , 1 ≤ i ≤ N 1 \leq i \leq N 1 ≤ i ≤ N .
Let U i U_i U i be the connected component of ρ i − 1 ( [ 0 , ε ] ) \rho_i^{-1}([0,\varepsilon]) ρ i − 1 ([ 0 , ε ]) containing Σ i \Sigma_i Σ i . Then U i U_i U i is a compact smooth manifold with smooth boundary, and f i − 1 ( 0 ) f_i^{-1}(0) f i − 1 ( 0 ) intersects ∂ U i \partial U_i ∂ U i transversely whenever their intersection is nonempty. Moreover, U i ∩ U j = ∅ U_i \cap U_j = \varnothing U i ∩ U j = ∅ when i ≠ j i \ne j i = j , and Σ i \Sigma_i Σ i is the unique connected component of Σ \Sigma Σ contained in U i U_i U i , for each 1 ≤ i ≤ N 1 \leq i \leq N 1 ≤ i ≤ N .
We take a small number δ > 0 \delta> 0 δ > 0 and define V i = f i − 1 ( Δ ( 0 ; δ ) ) V_i = f_i^{-1}(\Delta(0;\delta)) V i = f i − 1 ( Δ ( 0 ; δ )) , where Δ ( 0 ; δ ) \Delta(0;\delta) Δ ( 0 ; δ ) is a disk in C \mathbb{C} C of radius δ \delta δ . By choosing δ \delta δ sufficiently small, we may assume that f i − 1 ( v ) f_i^{-1}(v) f i − 1 ( v ) intersects ∂ U i \partial U_i ∂ U i transversely for every v ∈ Δ ( 0 ; δ ) v \in\Delta(0;\delta) v ∈ Δ ( 0 ; δ ) , whenever ∂ U i ∩ f i − 1 ( 0 ) ≠ ∅ \partial U_i \cap f_i^{-1}(0) \ne\varnothing ∂ U i ∩ f i − 1 ( 0 ) = ∅ . If ∂ U i ∩ f i − 1 ( 0 ) = ∅ \partial U_i \cap f_i^{-1}(0) = \varnothing ∂ U i ∩ f i − 1 ( 0 ) = ∅ , we may further arrange that the connected component of V i V_i V i containing Σ i \Sigma_i Σ i is contained in int ( U i ) \operatorname{int}(U_i) int ( U i ) . Then U i ∩ V i U_i \cap V_i U i ∩ V i is a smooth compact manifold with corners. We denote by W i W_i W i the unique connected component of U i ∩ V i U_i \cap V_i U i ∩ V i containing Σ i \Sigma_i Σ i . Write
T i = ∂ U i ∩ V i ∩ W i = ∂ U i ∩ f i − 1 ( Δ ( 0 ; δ ) ) ∩ W i . T_i = \partial U_i \cap V_i \cap W_i = \partial U_i \cap f_i^{-1}(\Delta(0;\delta)) \cap W_i. T i = ∂ U i ∩ V i ∩ W i = ∂ U i ∩ f i − 1 ( Δ ( 0 ; δ )) ∩ W i .
and
R i = ∂ V i ∩ U i ∩ W i = f i − 1 ( { z : ∣ z ∣ = δ } ) ∩ U i ∩ W i . R_i = \partial V_i \cap U_i \cap W_i = f_i^{-1}(\{z : |z| = \delta\}) \cap U_i \cap W_i. R i = ∂ V i ∩ U i ∩ W i = f i − 1 ({ z : ∣ z ∣ = δ }) ∩ U i ∩ W i .
Their intersection R i ∩ T i R_i \cap T_i R i ∩ T i is a smooth corner. We call these W i W_i W i critical neighborhoods of type I ; see Figure 1 .
Figure 1. Critical neighborhoods of Types I and II (with possible mixed cases).
There are now two mutually exclusive possibilities:2
(A) For some i i i , f i − 1 ( 0 ) ∩ ∂ U i ∩ W i = ∅ f_i^{-1}(0) \cap\partial U_i \cap W_i = \varnothing f i − 1 ( 0 ) ∩ ∂ U i ∩ W i = ∅ ;
(B) For each i i i , f i − 1 ( 0 ) ∩ ∂ U i ∩ W i ≠ ∅ f_i^{-1}(0) \cap\partial U_i \cap W_i \ne\varnothing f i − 1 ( 0 ) ∩ ∂ U i ∩ W i = ∅ ;
In section 3.3, under the assumption dim ψ ( Y ) ≥ 2 \dim\psi(Y) \geq2 dim ψ ( Y ) ≥ 2 , we reduced to the case dim ψ ˉ ( Y ) = 2 \dim\bar{\psi}(Y) = 2 dim ψ ˉ ( Y ) = 2 . We claim that Case (A) cannot occur. Assume not, up to shrinking δ > 0 \delta> 0 δ > 0 if necessary, we have f i − 1 ( Δ ( 0 ; δ ) ) ∩ W i f_i^{-1}(\Delta(0;\delta)) \cap W_i f i − 1 ( Δ ( 0 ; δ )) ∩ W i is compactly contained in int ( U i ) \operatorname{int}(U_i) int ( U i ) . Hence, for every sufficiently small nonzero regular value v v v of f i f_i f i , each connected component C ⊂ f i − 1 ( v ) ∩ W i C \subset f_i^{-1}(v) \cap W_i C ⊂ f i − 1 ( v ) ∩ W i is a compact smooth complex curve. Since the exceptional set E E E has only finitely many irreducible components and f i f_i f i is constant on each irreducible component of E ∩ W i E \cap W_i E ∩ W i , the set f i ( E ∩ W i ) f_i(E \cap W_i) f i ( E ∩ W i ) is finite. Choose a nonempty open subset V ⊂ Δ ∗ ( 0 ; δ ) ∖ f i ( E ∩ W i ) V \subset\Delta^*(0;\delta) \setminus f_i(E \cap W_i) V ⊂ Δ ∗ ( 0 ; δ ) ∖ f i ( E ∩ W i ) consisting of regular values of f i f_i f i . Then f i − 1 ( V ) ∩ W i ⊂ E = ∅ f_i^{-1}(V) \cap W_i \subset E = \varnothing f i − 1 ( V ) ∩ W i ⊂ E = ∅ . For any v ∈ V v \in V v ∈ V and any connected component C ⊂ f i − 1 ( v ) ∩ W i C \subset f_i^{-1}(v) \cap W_i C ⊂ f i − 1 ( v ) ∩ W i , we therefore have C ⊄ E C \not\subset E C ⊂ E . By Theorem 3.12, the restriction ( ψ ˉ ∣ C ) ∗ ω ≠ 0 (\bar{\psi}|_C)^*\omega\ne0 ( ψ ˉ ∣ C ) ∗ ω = 0 . On the other hand, since d f i = ξ = ψ ˉ ∗ ω df_i=\xi=\bar{\psi}^{*}\omega d f i = ξ = ψ ˉ ∗ ω , we have ξ ∣ C = 0 \xi|_C=0 ξ ∣ C = 0 , a contradiction. Therefore Case (A) cannot occur.
From now on, we are under case (B), and we have the following local 1-connectedness property which is [38 Lemma 8] . We include its proof here for the convenience of readers.
Proposition 3.15 (Simpson). — Assume that Case (B) holds. Up to shrinking δ \delta δ , assume that f i − 1 ( z ) ∩ W i f_i^{-1}(z)\cap W_i f i − 1 ( z ) ∩ W i meets ∂ U i \partial U_i ∂ U i transversely for each z ∈ Δ ( 0 ; δ ) z\in\Delta(0;\delta) z ∈ Δ ( 0 ; δ ) . Let
W i ∗ : = W i ∖ Q ′ , W_i^*:=W_i\setminus Q', W i ∗ := W i ∖ Q ′ , and let f i − 1 ( v ) f_i^{-1}(v) f i − 1 ( v ) be a nearby fiber with v ≠ 0 v\ne0 v = 0 . Then pair
( W i ∗ , f i − 1 ( v ) ∩ W i ∗ ) = ( W i ∗ , f i − 1 ( v ) ∩ W i ) (W_i^*,f_i^{-1}(v)\cap W_i^*)=(W_i^*,f_i^{-1}(v)\cap W_i) ( W i ∗ , f i − 1 ( v ) ∩ W i ∗ ) = ( W i ∗ , f i − 1 ( v ) ∩ W i ) is 1-connected.
Proof. — Set
W i ∗ ∗ = W i ∖ f i − 1 ( 0 ) = W i ∗ ∖ f i − 1 ( 0 ) . W_i^{**}=W_i\setminus f_i^{-1}(0)=W_i^*\setminus f_i^{-1}(0). W i ∗∗ = W i ∖ f i − 1 ( 0 ) = W i ∗ ∖ f i − 1 ( 0 ) .
Since f i − 1 ( 0 ) f_i^{-1}(0) f i − 1 ( 0 ) is a complex analytic hypersurface in the connected manifold W i W_i W i , its complement W i ∗ ∗ W_i^{**} W i ∗∗ is connected. Moreover, the inclusions W i ∗ ∗ ↪ W i W_i^{**}\hookrightarrow W_i W i ∗∗ ↪ W i and W i ∗ ∗ ↪ W i ∗ W_i^{**}\hookrightarrow W_i^* W i ∗∗ ↪ W i ∗ both induce surjections on fundamental groups. By compactness of W i W_i W i and transversality along its boundary, the restriction
f i : W i ∗ ∗ ⟶ Δ ∗ f_i:W_i^{**}\longrightarrow\Delta^* f i : W i ∗∗ ⟶ Δ ∗
is a proper smooth submersion. By Ehresmann’s fibration theorem
f i ∣ W i ∗ ∗ : W i ∗ ∗ ⟶ Δ ∗ , (9) f_i|_{W_i^{**}}:W_i^{**}\longrightarrow\Delta^*, \tag*{(9)} f i ∣ W i ∗∗ : W i ∗∗ ⟶ Δ ∗ , (9)
is a locally trivial fibration.
Consider the following long exact sequence on homotopy groups of the fibration (9) :
π 1 ( f i − 1 ( v ) ∩ W i ) ⟶ π 1 ( W i ∗ ∗ ) ⟶ ( f i ) ∗ π 1 ( Δ ∗ ) ⟶ π 0 ( f i − 1 ( v ) ∩ W i ) ⟶ π 0 ( W i ∗ ∗ ) = 1 (10) \pi_1(f_i^{-1}(v)\cap W_i)\longrightarrow\pi_1(W_i^{**})\mathop{\longrightarrow}^{(f_i)_*}\pi_1(\Delta^*)\longrightarrow\pi_0(f_i^{-1}(v)\cap W_i)\longrightarrow\pi_0(W_i^{**})=1 \tag*{(10)} π 1 ( f i − 1 ( v ) ∩ W i ) ⟶ π 1 ( W i ∗∗ ) ⟶ ( f i ) ∗ π 1 ( Δ ∗ ) ⟶ π 0 ( f i − 1 ( v ) ∩ W i ) ⟶ π 0 ( W i ∗∗ ) = 1 (10)
with 0 < ∣ v ∣ < δ 0<|v|<\delta 0 < ∣ v ∣ < δ .
By Case (B), choose y ∈ f i − 1 ( 0 ) ∩ T i y\in f_i^{-1}(0)\cap T_i y ∈ f i − 1 ( 0 ) ∩ T i , we have y ∉ Σ i y\notin\Sigma_i y ∈ / Σ i and f i − 1 ( 0 ) f_i^{-1}(0) f i − 1 ( 0 ) meets ∂ U i \partial U_i ∂ U i transversely at y y y . Hence we can choose a regular point x ∈ f i − 1 ( 0 ) ∩ int ( W i ) ∖ Σ x\in f_i^{-1}(0)\cap\operatorname{int}(W_i)\setminus\Sigma x ∈ f i − 1 ( 0 ) ∩ int ( W i ) ∖ Σ sufficiently close to y y y ; in particular, x ∈ W i ∗ x\in W_i^* x ∈ W i ∗ . By the implicit function theorem, there exists a small holomorphic disc σ : Δ ( 0 , δ 1 ) → W i ∗ \sigma:\Delta(0,\delta_1)\to W_i^* σ : Δ ( 0 , δ 1 ) → W i ∗ with 0 < δ 1 < δ 0<\delta_1<\delta 0 < δ 1 < δ such that f i ∘ σ = id f_i\circ\sigma=\operatorname{id} f i ∘ σ = id , thus im ( σ ) ∩ f i − 1 ( 0 ) = { x } \operatorname{im}(\sigma)\cap f_i^{-1}(0)=\{x\} im ( σ ) ∩ f i − 1 ( 0 ) = { x } . Let α \alpha α be the image under σ ∗ \sigma_* σ ∗ of a sufficiently small circle { ∣ v ∣ = δ 2 } \{|v|=\delta_2\} { ∣ v ∣ = δ 2 } with δ 2 < δ 1 \delta_2<\delta_1 δ 2 < δ 1 in this disc, then α \alpha α is a loop in W i ∗ ∗ W_i^{**} W i ∗∗ such that ( f i ) ∗ [ α ] (f_i)_*[\alpha] ( f i ) ∗ [ α ] generates π 1 ( Δ ∗ ) ≅ Z \pi_1(\Delta^*)\cong\mathbb{Z} π 1 ( Δ ∗ ) ≅ Z , and α \alpha α is null-homotopic in W i ∗ W_i^* W i ∗ . In particular, ( f i ) ∗ : π 1 ( W i ∗ ∗ ) → π 1 ( Δ ∗ ) (f_i)_*:\pi_1(W_i^{**})\to\pi_1(\Delta^*) ( f i ) ∗ : π 1 ( W i ∗∗ ) → π 1 ( Δ ∗ ) is surjective, then (10) implies that
every fiber f i − 1 ( v ) ∩ W i , 0 < ∣ v ∣ < δ , is connected. (11) \text{every fiber }f_i^{-1}(v)\cap W_i,\ 0<|v|<\delta,\text{ is connected.} \tag*{(11)} every fiber f i − 1 ( v ) ∩ W i , 0 < ∣ v ∣ < δ , is connected. (11)
Besides, for any [ β ] ∈ π 1 ( W i ∗ ∗ ) [\beta]\in\pi_1(W_i^{**}) [ β ] ∈ π 1 ( W i ∗∗ ) , let n = ( f i ) ∗ [ β ] ∈ π 1 ( Δ ∗ ) n=(f_i)_*[\beta]\in\pi_1(\Delta^*) n = ( f i ) ∗ [ β ] ∈ π 1 ( Δ ∗ ) , then [ β ] [ α ] − n ∈ ker ( f i ) ∗ [\beta][\alpha]^{-n}\in\ker(f_i)_* [ β ] [ α ] − n ∈ ker ( f i ) ∗ , and hence belongs to the image of π 1 ( f i − 1 ( v ) ∩ W i ) \pi_1(f_i^{-1}(v)\cap W_i) π 1 ( f i − 1 ( v ) ∩ W i ) . Since α \alpha α is null-homotopic in W i ∗ W_i^* W i ∗ and π 1 ( W i ∗ ∗ ) → π 1 ( W i ∗ ) \pi_1(W_i^{**})\to\pi_1(W_i^*) π 1 ( W i ∗∗ ) → π 1 ( W i ∗ ) is surjective, the inclusion f i − 1 ( v ) ∩ W i ↪ W i ∗ f_i^{-1}(v)\cap W_i\hookrightarrow W_i^* f i − 1 ( v ) ∩ W i ↪ W i ∗ induces a surjection on fundamental groups:
π 1 ( f i − 1 ( v ) ∩ W i ) ⟶ π 1 ( W i ∗ ) . (12) \pi_1(f_i^{-1}(v)\cap W_i)\longrightarrow\pi_1(W_i^*). \tag*{(12)} π 1 ( f i − 1 ( v ) ∩ W i ) ⟶ π 1 ( W i ∗ ) . (12)
Remark 3.16. — Similarly, we have a locally-trivial fibration
f i ∣ T i : T i ⟶ Δ ( 0 ; δ ) . (13) f_i|_{T_i}:T_i\longrightarrow\Delta(0;\delta). \tag*{(13)} f i ∣ T i : T i ⟶ Δ ( 0 ; δ ) . (13)
Next, consider an irreducible component Q i ⊂ Q ′ ′ Q_i\subset Q'' Q i ⊂ Q ′′ . Set
Q i ∘ : = Q i ∖ P . Q_i^\circ:=Q_i\setminus P. Q i ∘ := Q i ∖ P .
Since ξ ∣ Q i ∘ ≢ 0 \xi|_{Q_i^\circ}\not\equiv0 ξ ∣ Q i ∘ ≡ 0 and ξ ∣ Q i ∘ \xi|_{Q_i^\circ} ξ ∣ Q i ∘ extends meromorphically across the finite set Q i ∩ P Q_i\cap P Q i ∩ P , its zero set on Q i ∘ Q_i^\circ Q i ∘ is finite. Let D i D_i D i denote the finite set of points p ∈ Q i ∘ p\in Q_i^\circ p ∈ Q i ∘ such that
ξ ( p ) ≠ 0 , ξ ∣ T p Q i = 0. \xi(p)\ne0,\qquad\xi|_{T_pQ_i}=0. ξ ( p ) = 0 , ξ ∣ T p Q i = 0.
Away from D i D_i D i and the Type I critical neighborhoods, we may choose the vector field constructed in § 3.5 to be tangent to Q i Q_i Q i (or its lift) and to satisfy Re ξ ( v ) = 1 \operatorname{Re}\xi(v)=1 Re ξ ( v ) = 1 , so that its flow preserves Q i Q_i Q i and hence also its complement.
We therefore need additional local neighborhoods around the points of D i D_i D i , as well as around the finitely many transverse intersection points of distinct components of Q ′ ′ Q^{\prime\prime} Q ′′ .
( α ) (\alpha) ( α ) If p ∈ Q i ∩ Q j p\in Q_i\cap Q_j p ∈ Q i ∩ Q j and ξ \xi ξ vanishes on both T p Q i T_pQ_i T p Q i and T p Q j T_pQ_j T p Q j , then p ∈ Z ( ξ ) p\in Z(\xi) p ∈ Z ( ξ ) , and this point is already contained in a type I critical neighborhood.
( β ) (\beta) ( β ) If ξ \xi ξ vanishes on only one of the two tangent directions, then we can take a type II critical neighborhood.
( γ ) (\gamma) ( γ ) If neither tangential restriction vanishes, we still include p p p among the type II critical neighborhoods.
The argument for case ( γ ) (\gamma) ( γ ) is given in Theorem 3.18; in this case f l ∣ Q i f_l|_{Q_i} f l ∣ Q i and f l ∣ Q j f_l|_{Q_j} f l ∣ Q j are both submersions near p p p , so over the punctured disc the intersections with Q i ∪ Q j Q_i\cup Q_j Q i ∪ Q j vary as two distinct unramified branches. The same local 1-connectedness argument as for type II neighborhoods applies. The case ( β ) (\beta) ( β ) is analogous, with W l ∗ = W l ∖ ( Q i ∪ Q j ) \mathcal{W}_l^*=\mathcal{W}_l\setminus(Q_i\cup Q_j) W l ∗ = W l ∖ ( Q i ∪ Q j ) .
At each point p l ∈ D i p_l\in D_i p l ∈ D i , choose a sufficiently small simply connected neighborhood V l ′ ⊂ Y ∘ V_l'\subset Y^\circ V l ′ ⊂ Y ∘ . Since ξ \xi ξ is closed, the holomorphic Poincaré lemma gives a holomorphic function f l f_l f l on V l ′ V_l' V l ′ such that
d f l = ξ , f l ( p l ) = 0. df_l=\xi,\qquad f_l(p_l)=0. d f l = ξ , f l ( p l ) = 0.
Since ξ ( p l ) ≠ 0 \xi(p_l)\ne0 ξ ( p l ) = 0 , after shrinking V l ′ V_l' V l ′ we may assume that ξ \xi ξ is nowhere vanishing on V l ′ V_l' V l ′ . Thus f l f_l f l is a holomorphic submersion. Choose local holomorphic coordinates ( z 1 , z 2 ) (z_1,z_2) ( z 1 , z 2 ) centered at p l p_l p l such that f l = z 1 f_l=z_1 f l = z 1 . Since Q i Q_i Q i is smooth, the curve Q i Q_i Q i can be written locally as
Q i = { z 1 = ϕ ( z 2 ) } . Q_i=\{z_1=\phi(z_2)\}. Q i = { z 1 = ϕ ( z 2 )} .
where ϕ ( 0 ) = ϕ ′ ( 0 ) = 0 \phi(0)=\phi'(0)=0 ϕ ( 0 ) = ϕ ′ ( 0 ) = 0 and ϕ \phi ϕ is not identically zero. After a further holomorphic change of the coordinate z 2 z_2 z 2 , leaving z 1 z_1 z 1 unchanged, we may therefore write ϕ ( z 2 ) = z 2 m l \phi(z_2)=z_2^{m_l} ϕ ( z 2 ) = z 2 m l for some integer m l ≥ 2 m_l\ge2 m l ≥ 2 . Thus, locally,
Q i = { z 1 = z 2 m l } . Q_i=\{z_1=z_2^{m_l}\}. Q i = { z 1 = z 2 m l } .
Choose ε l > 0 \varepsilon_l>0 ε l > 0 sufficiently small and then choose 0 < δ l < ε l m l 0<\delta_l<\varepsilon_l^{m_l} 0 < δ l < ε l m l . Define
W l = { ( z 1 , z 2 ) : ∣ z 1 ∣ ≤ δ l , ∣ z 2 ∣ ≤ ε l } . \mathcal{W}_l=\{(z_1,z_2):|z_1|\le\delta_l,\ |z_2|\le\varepsilon_l\}. W l = {( z 1 , z 2 ) : ∣ z 1 ∣ ≤ δ l , ∣ z 2 ∣ ≤ ε l } .
such that W l \mathcal{W}_l W l is a smooth compact manifold with smooth corners. Its boundary consists of
T l = { ∣ z 1 ∣ ≤ δ l , ∣ z 2 ∣ = ε l } , R l = { ∣ z 1 ∣ = δ l , ∣ z 2 ∣ ≤ ε l } . \mathcal{T}_l=\{|z_1|\le\delta_l,\ |z_2|=\varepsilon_l\},\qquad
\mathcal{R}_l=\{|z_1|=\delta_l,\ |z_2|\le\varepsilon_l\}. T l = { ∣ z 1 ∣ ≤ δ l , ∣ z 2 ∣ = ε l } , R l = { ∣ z 1 ∣ = δ l , ∣ z 2 ∣ ≤ ε l } .
By the choice δ l < ε l m l \delta_l<\varepsilon_l^{m_l} δ l < ε l m l , we have
Q i ∩ T l = ∅ . Q_i\cap\mathcal{T}_l=\varnothing. Q i ∩ T l = ∅ .
Thus Q i Q_i Q i exits W l \mathcal{W}_l W l only through R l \mathcal{R}_l R l . After shrinking these neighborhoods if necessary, we may assume that the W l \mathcal{W}_l W l are pairwise disjoint. We call the resulting neighborhoods W l \mathcal{W}_l W l the critical neighborhoods of Type II , see Figure 1 .
For a critical neighborhood of Type II, we have a local 1-connectedness property which is considerably simpler than Theorem 3.15. The essential point is that ξ \xi ξ itself is nowhere vanishing on W l \mathcal{W}_l W l . Set
W l ∗ = W l ∖ Q i and W l ∗ ∗ = f l − 1 ( Δ ∗ ( 0 ; δ l ) ) ∩ ( W l ∖ Q i ) . \mathcal{W}_l^*=\mathcal{W}_l\setminus Q_i\qquad\text{and}\qquad
\mathcal{W}_l^{**}=f_l^{-1}\bigl(\Delta^*(0;\delta_l)\bigr)\cap(\mathcal{W}_l\setminus Q_i). W l ∗ = W l ∖ Q i and W l ∗∗ = f l − 1 ( Δ ∗ ( 0 ; δ l ) ) ∩ ( W l ∖ Q i ) .
Proposition 3.17. — With the above notations, for every sufficiently small nonzero v ∈ Δ ( 0 ; δ l ) v\in\Delta(0;\delta_l) v ∈ Δ ( 0 ; δ l ) , the pair
( W l ∗ , f l − 1 ( v ) ∩ W l ∗ ) \left(\mathcal{W}_l^*,f_l^{-1}(v)\cap\mathcal{W}_l^*\right) ( W l ∗ , f l − 1 ( v ) ∩ W l ∗ ) is 1-connected.
Proof. — By the local description above, for every 0 < ∣ v ∣ < δ l 0<|v|<\delta_l 0 < ∣ v ∣ < δ l , the fiber f l − 1 ( v ) ∩ W l ∗ f_l^{-1}(v)\cap\mathcal{W}_l^* f l − 1 ( v ) ∩ W l ∗ is a disc with the m l m_l m l points satisfying z 2 m l = v z_2^{m_l}=v z 2 m l = v removed. In particular, it is connected.
Since the map z 2 ↦ z 2 m l z_2\mapsto z_2^{m_l} z 2 ↦ z 2 m l is unramified over Δ ∗ ( 0 ; δ l ) \Delta^*(0;\delta_l) Δ ∗ ( 0 ; δ l ) , the m l m_l m l deleted points vary locally as m l m_l m l disjoint holomorphic branches. Hence
f l : W l ∗ ∗ → Δ ∗ ( 0 ; δ l ) f_l:\mathcal{W}_l^{**}\to\Delta^*(0;\delta_l) f l : W l ∗∗ → Δ ∗ ( 0 ; δ l )
is a locally trivial fibration. Its homotopy exact sequence gives
π 1 ( f l − 1 ( v ) ∩ W l ∗ ) → π 1 ( W l ∗ ∗ ) → π 1 ( Δ ∗ ( 0 ; δ l ) ) → 1. \pi_{1}(f_l^{-1}(v) \cap\mathcal{W}_l^{*}) \to\pi_{1}(\mathcal{W}_l^{**}) \to\pi_{1}(\Delta^{*}(0; \delta_l)) \to1. π 1 ( f l − 1 ( v ) ∩ W l ∗ ) → π 1 ( W l ∗∗ ) → π 1 ( Δ ∗ ( 0 ; δ l )) → 1.
Moreover, the inclusion W l ∗ ∗ ↪ W l ∗ \mathcal{W}_l^{**} \hookrightarrow\mathcal{W}_l^{*} W l ∗∗ ↪ W l ∗ induces a surjection on fundamental groups, since W l ∗ ∗ \mathcal{W}_l^{**} W l ∗∗ is obtained by removing the central complex hypersurface from W l ∗ \mathcal{W}_l^{*} W l ∗ .
Choose a ∈ C a \in\mathbb{C} a ∈ C such that δ l 1 / m l < ∣ a ∣ < ϵ l \delta_l^{1/m_l} < |a| < \epsilon_l δ l 1/ m l < ∣ a ∣ < ϵ l . Then s ( v ) = ( v , a ) s(v) = (v,a) s ( v ) = ( v , a ) defines a section of f l f_l f l over the disc Δ ( 0 ; δ l ) \Delta(0;\delta_l) Δ ( 0 ; δ l ) whose image is disjoint from Q i Q_i Q i . For a loop γ \gamma γ representing a generator of π 1 ( Δ ∗ ( 0 ; δ l ) ) \pi_1(\Delta^{*}(0;\delta_l)) π 1 ( Δ ∗ ( 0 ; δ l )) , the loop s ( γ ) s(\gamma) s ( γ ) lies in W l ∗ ∗ \mathcal{W}_l^{**} W l ∗∗ and is null-homotopic in W l ∗ \mathcal{W}_l^{*} W l ∗ .
Therefore, for any class in π 1 ( W l ∗ ) \pi_1(\mathcal{W}_l^{*}) π 1 ( W l ∗ ) , we may deform it to a class in π 1 ( W l ∗ ∗ ) \pi_1(\mathcal{W}_l^{**}) π 1 ( W l ∗∗ ) , and then multiply by a suitable power of [ s ( γ ) ] [s(\gamma)] [ s ( γ )] so that the resulting class lies in the kernel of
π 1 ( W l ∗ ∗ ) → π 1 ( Δ ∗ ( 0 ; δ l ) ) . \pi_1(\mathcal{W}_l^{**}) \to\pi_1(\Delta^{*}(0;\delta_l)). π 1 ( W l ∗∗ ) → π 1 ( Δ ∗ ( 0 ; δ l )) .
thus lies in the image of π 1 ( f l − 1 ( v ) ∩ ( W l ∗ ) ) \pi_1(f_l^{-1}(v) \cap(\mathcal{W}_l^{*})) π 1 ( f l − 1 ( v ) ∩ ( W l ∗ )) . Since [ s ( γ ) ] [s(\gamma)] [ s ( γ )] is trivial in π 1 ( W l ∗ ) \pi_1(\mathcal{W}_l^{*}) π 1 ( W l ∗ ) , it follows that
π 1 ( f l − 1 ( v ) ∩ ( W l ∗ ) ) ↠ π 1 ( W l ∗ ) . \pi_1(f_l^{-1}(v) \cap(\mathcal{W}_l^{*})) \twoheadrightarrow\pi_1(\mathcal{W}_l^{*}). π 1 ( f l − 1 ( v ) ∩ ( W l ∗ )) ↠ π 1 ( W l ∗ ) .
Together with the connectedness of the nearby fiber, this proves that the pair is 1-connected.
Remark 3.18. — For a type II critical neighborhood, Ehresmann’s fibration theorem is not directly applicable, since after deleting Q i Q_i Q i the resulting map need not be proper. Instead, the explicit local description above, in which the m l m_l m l branches are unramified away from p l p_l p l , gives the required locally trivial fibration.
In the case ( γ ) (\gamma) ( γ ) , let p ∈ Q i ∩ Q j ⊂ Q ′ ′ p \in Q_i \cap Q_j \subset Q^{\prime\prime} p ∈ Q i ∩ Q j ⊂ Q ′′ . Since neither ξ ∣ T p Q i \left.\xi\right|_{T_pQ_i} ξ ∣ T p Q i nor ξ ∣ T p Q j \left.\xi\right|_{T_pQ_j} ξ ∣ T p Q j vanishes, we may choose local coordinates ( z 1 , z 2 ) (z_1,z_2) ( z 1 , z 2 ) centered at p p p such that f = z 1 f=z_1 f = z 1 and
Q i = { z 2 = ϕ i ( z 1 ) } , Q j = { z 2 = ϕ j ( z 1 ) } , Q_i = \{z_2 = \phi_i(z_1)\}, \qquad Q_j = \{z_2 = \phi_j(z_1)\}, Q i = { z 2 = ϕ i ( z 1 )} , Q j = { z 2 = ϕ j ( z 1 )} ,
where ϕ i ( v ) ≠ ϕ j ( v ) \phi_i(v) \ne\phi_j(v) ϕ i ( v ) = ϕ j ( v ) for v ≠ 0 v \ne0 v = 0 sufficiently small. Taking
W ∗ = W ∖ ( Q i ∪ Q j ) , \mathcal{W}^{*} = \mathcal{W} \setminus(Q_i \cup Q_j), W ∗ = W ∖ ( Q i ∪ Q j ) ,
the fibers of f : W ∗ → Δ f:\mathcal{W}^{*} \to\Delta f : W ∗ → Δ away from 0 0 0 are discs with two points removed. The same section and fibration argument as above therefore gives the required local 1-connectedness, so this case may also be treated as a Type II critical neighborhood. ( β ) (\beta) ( β ) is proved similarly, thus we shall not repeat.
The two types of critical neighborhoods may intersect, as illustrated on the left in Figure 1 . Recall that the Q ′ Q' Q ′ -part is already contained in Σ \Sigma Σ . Thus, any non-polar component of Q Q Q shown intersecting Σ \Sigma Σ in the picture on the left belongs to Q ′ ′ Q'' Q ′′ . We use the notation Q i ′ ′ Q''_i Q i ′′ to emphasize this in the picture. The same convention applies to the picture on the right.
This kind of mixed situation can be remedied as follows. Fix Σ i \Sigma_i Σ i . There are only finitely many irreducible components of Q ′ ′ Q'' Q ′′ intersecting Σ i \Sigma_i Σ i ; relabel them as Q 1 i , … , Q J i i Q^i_1,\ldots,Q^i_{J_i} Q 1 i , … , Q J i i . Since Q j i ⊂ Q ′ ′ Q^i_j\subset Q'' Q j i ⊂ Q ′′ , we have ξ ∣ Q j i ≢ 0 \left.\xi\right|_{Q^i_j}\not\equiv0 ξ ∣ Q j i ≡ 0 . Hence Q j i Q^i_j Q j i is not contained in Σ i \Sigma_i Σ i , and therefore Q j i ∩ Σ i Q^i_j\cap\Sigma_i Q j i ∩ Σ i is a finite set. Define the union of these Q j i Q^i_j Q j i as Q i ′ ′ Q''_i Q i ′′ .
We can moreover arrange the parameter δ \delta δ in the construction of the type I neighborhood so that each Q j i Q^i_j Q j i exits through R i R_i R i , and not through T i T_i T i . Indeed, since d f i = ξ df_i=\xi d f i = ξ and ξ ∣ Q j i ≢ 0 \left.\xi\right|_{Q^i_j}\not\equiv0 ξ ∣ Q j i ≡ 0 , the restriction f i ∣ Q j i f_i|_{Q^i_j} f i ∣ Q j i is nonconstant. Hence its zero set is discrete. By shrinking the neighborhood of Σ i \Sigma_i Σ i and choosing ε \varepsilon ε sufficiently small, we may assume that
Q j i ∩ ∂ U i ∩ f i − 1 ( 0 ) = ∅ Q^i_j\cap\partial U_i\cap f_i^{-1}(0)=\varnothing Q j i ∩ ∂ U i ∩ f i − 1 ( 0 ) = ∅
for every 1 ≤ j ≤ J i 1\leq j\leq J_i 1 ≤ j ≤ J i . For each j j j such that Q j i ∩ ∂ U i ≠ ∅ Q^i_j\cap\partial U_i\neq\varnothing Q j i ∩ ∂ U i = ∅ , set
c j = min z ∈ Q j i ∩ ∂ U i ∣ f i ( z ) ∣ > 0. c_j=\min_{z\in Q^i_j\cap\partial U_i}|f_i(z)|>0. c j = z ∈ Q j i ∩ ∂ U i min ∣ f i ( z ) ∣ > 0.
We then further require, in the construction of the type I neighborhood, that δ < min j c j \delta<\min_j c_j δ < min j c j , where the minimum is taken over those j j j for which Q j i ∩ ∂ U i ≠ ∅ Q^i_j\cap\partial U_i\neq\varnothing Q j i ∩ ∂ U i = ∅ . It follows that Q j i ∩ T i = ∅ Q^i_j\cap T_i=\varnothing Q j i ∩ T i = ∅ for every j j j . Thus the components Q j i Q^i_j Q j i can meet the boundary of W i W_i W i only through R i R_i R i , and not through T i T_i T i . In the mixed case, we always choose the type I neighborhoods with this additional property.
After decreasing δ \delta δ further if necessary, we may also assume that f i ∣ Q j i f_i|_{Q^i_j} f i ∣ Q j i has no critical point over Δ ∗ ( 0 ; δ ) \Delta^*(0;\delta) Δ ∗ ( 0 ; δ ) for every j j j . Thus, over v v v in the punctured disc, the finite set f i − 1 ( v ) ∩ Q i f_i^{-1}(v)\cap Q^i f i − 1 ( v ) ∩ Q i varies locally as a finite collection of unramified branches, as in the case of type II critical neighborhoods.
Set
W i † = W i ∖ ( Q ′ ∪ Q i ′ ′ ) and W i † † = W i † ∩ f i − 1 ( Δ ∗ ( 0 ; δ ) ) . W_i^\dagger=W_i\setminus(Q'\cup Q''_i)\quad\text{and}\quad W_i^{\dagger\dagger}=W_i^\dagger\cap f_i^{-1}(\Delta^*(0;\delta)). W i † = W i ∖ ( Q ′ ∪ Q i ′′ ) and W i †† = W i † ∩ f i − 1 ( Δ ∗ ( 0 ; δ )) .
Since Q ′ ∩ W i ⊂ f i − 1 ( 0 ) Q'\cap W_i\subset f_i^{-1}(0) Q ′ ∩ W i ⊂ f i − 1 ( 0 ) , we have
W i † † = W i ∗ ∗ ∖ Q i ′ ′ . W_i^{\dagger\dagger}=W_i^{**}\setminus Q''_i. W i †† = W i ∗∗ ∖ Q i ′′ .
The map
f i : W i ∗ ∗ ⟶ Δ ∗ ( 0 ; δ ) f_i:W_i^{**}\longrightarrow\Delta^*(0;\delta) f i : W i ∗∗ ⟶ Δ ∗ ( 0 ; δ )
is a locally trivial fibration by Theorem 3.15. Since the intersections with the components of Q i ′ ′ Q''_i Q i ′′ vary locally as unramified branches, the restriction
f i : W i † † ⟶ Δ ∗ ( 0 ; δ ) f_i:W_i^{\dagger\dagger}\longrightarrow\Delta^*(0;\delta) f i : W i †† ⟶ Δ ∗ ( 0 ; δ )
is again a locally trivial fibration. Its fiber over v ≠ 0 v\neq0 v = 0 is F v † : = f i − 1 ( v ) ∩ W i † F^\dagger_v:=f_i^{-1}(v)\cap W_i^\dagger F v † := f i − 1 ( v ) ∩ W i † . It is obtained from the connected smooth complex curve f i − 1 ( v ) ∩ W i f_i^{-1}(v)\cap W_i f i − 1 ( v ) ∩ W i by removing finitely many points, namely f i − 1 ( v ) ∩ Q i ′ ′ f_i^{-1}(v)\cap Q''_i f i − 1 ( v ) ∩ Q i ′′ , and is therefore connected.
Consider the associated homotopy exact sequence:
π 1 ( F v † ) ⟶ π 1 ( W i † † ) ⟶ π 1 ( Δ ∗ ( 0 ; δ ) ) ⟶ π 0 ( F v † ) = 1. \pi_1(F^\dagger_v)\longrightarrow\pi_1(W_i^{\dagger\dagger})\longrightarrow\pi_1(\Delta^*(0;\delta))\longrightarrow\pi_0(F^\dagger_v)=1. π 1 ( F v † ) ⟶ π 1 ( W i †† ) ⟶ π 1 ( Δ ∗ ( 0 ; δ )) ⟶ π 0 ( F v † ) = 1.
By the construction above, both Q ′ Q' Q ′ and the components Q j i ⊂ Q i ′ ′ Q^i_j\subset Q''_i Q j i ⊂ Q i ′′ are disjoint from T i T_i T i . Hence, as in the proof of Theorem 3.15, we may choose a regular point
x ∈ f i − 1 ( 0 ) ∩ ( int ( W i ) ∖ ( Q ′ ∪ Q i ′ ′ ) ) x\in f_i^{-1}(0)\cap\left(\operatorname{int}(W_i)\setminus(Q'\cup Q''_i)\right) x ∈ f i − 1 ( 0 ) ∩ ( int ( W i ) ∖ ( Q ′ ∪ Q i ′′ ) )
sufficiently close to T i T_i T i . Up to shrinking δ \delta δ , there is a holomorphic section σ : Δ ( 0 ; δ ) → W i † \sigma:\Delta(0;\delta)\to W_i^\dagger σ : Δ ( 0 ; δ ) → W i † with f i ∘ σ = id f_i\circ\sigma=\operatorname{id} f i ∘ σ = id . Hence the loop obtained by applying σ \sigma σ to a generator of π 1 ( Δ ∗ ( 0 ; δ ) ) \pi_1(\Delta^*(0;\delta)) π 1 ( Δ ∗ ( 0 ; δ )) is null-homotopic in W i † W_i^\dagger W i † . Finally, the inclusion W i † † ↪ W i † W_i^{\dagger\dagger}\hookrightarrow W_i^\dagger W i †† ↪ W i † induces a surjection on fundamental groups. The same argument as in the proof of Theorem 3.15 therefore gives
π 1 ( F v † ) ⟶ π 1 ( W i † ) . \pi_1(F^\dagger_v)\longrightarrow\pi_1(W_i^\dagger). π 1 ( F v † ) ⟶ π 1 ( W i † ) .
Thus the same local 1-connectedness conclusion holds in the mixed case. In conclusion, we have proved
Proposition 3.19. — In the mixed case, we still have the pair
( W i † , F v † ) (W_i^\dagger,F_v^\dagger) ( W i † , F v † ) is 1-connected, for any v ≠ 0 v \ne0 v = 0 in Δ ( 0 ; δ ) \Delta(0;\delta) Δ ( 0 ; δ ) .
We deduce one more fact from the above 1-connectedness results that will be used later. The critical neighborhood can be of type I, type II, or mixed type. After decreasing δ > 0 \delta> 0 δ > 0 if necessary, we assume that the same δ \delta δ works for all type I, type II, and mixed-type critical neighborhoods.
Proposition 3.21. — For any − δ < η < δ -\delta< \eta< \delta − δ < η < δ ,
(a) For a type I critical neighborhood W i W_i W i , denote
S η ∗ = W i ∗ ∩ { η ≤ Re f i } , S_\eta^* = W_i^* \cap\{\eta\le\operatorname{Re} f_i\}, S η ∗ = W i ∗ ∩ { η ≤ Re f i } , then S η ∗ S_\eta^* S η ∗ is connected and π 1 ( S η ∗ ) → π 1 ( W i ∗ ) \pi_1(S_\eta^*) \to\pi_1(W_i^*) π 1 ( S η ∗ ) → π 1 ( W i ∗ ) is surjective.
(b) For a type II critical neighborhood W l \mathcal{W}_l W l , denote
S η ∗ = W l ∗ ∩ { η ≤ Re f l } , \mathcal{S}_\eta^* = \mathcal{W}_l^* \cap\{\eta\le\operatorname{Re} f_l\}, S η ∗ = W l ∗ ∩ { η ≤ Re f l } , then S η ∗ \mathcal{S}_\eta^* S η ∗ is connected and π 1 ( S η ∗ ) → π 1 ( W l ∗ ) \pi_1(\mathcal{S}_\eta^*) \to\pi_1(\mathcal{W}_l^*) π 1 ( S η ∗ ) → π 1 ( W l ∗ ) is surjective.
(c) For a mixed-type critical neighborhood W i † W_i^\dagger W i † , denote
S η † = W i † ∩ { η ≤ Re f i } , S_\eta^\dagger= W_i^\dagger\cap\{\eta\le\operatorname{Re} f_i\}, S η † = W i † ∩ { η ≤ Re f i } , then S η † S_\eta^\dagger S η † is connected and π 1 ( S η † ) → π 1 ( W i † ) \pi_1(S_\eta^\dagger) \to\pi_1(W_i^\dagger) π 1 ( S η † ) → π 1 ( W i † ) is surjective.
A similar conclusion holds if we consider intersection with { Re f i ≤ η } \{\operatorname{Re} f_i \le\eta\} { Re f i ≤ η } instead.
More generally, let − δ < η 1 ≤ η 2 < δ -\delta< \eta_1 \le\eta_2 < \delta − δ < η 1 ≤ η 2 < δ , with either η 1 < η 2 \eta_1 < \eta_2 η 1 < η 2 or η 1 = η 2 ≠ 0 \eta_1 = \eta_2 \ne0 η 1 = η 2 = 0 . Then the same conclusions hold for the corresponding bands W i ∗ ∩ { η 1 ≤ Re f i ≤ η 2 } W_i^* \cap\{\eta_1 \le\operatorname{Re} f_i \le\eta_2\} W i ∗ ∩ { η 1 ≤ Re f i ≤ η 2 } , W l ∗ ∩ { η 1 ≤ Re f l ≤ η 2 } \mathcal{W}_l^* \cap\{\eta_1 \le\operatorname{Re} f_l \le\eta_2\} W l ∗ ∩ { η 1 ≤ Re f l ≤ η 2 } , and W i † ∩ { η 1 ≤ Re f i ≤ η 2 } W_i^\dagger\cap\{\eta_1 \le\operatorname{Re} f_i \le\eta_2\} W i † ∩ { η 1 ≤ Re f i ≤ η 2 } .
Proof. — For (a)-(c), we abuse W i W_i W i for critical neighborhood of type I or II. Choose v ∈ Δ ( 0 ; δ ) v \in\Delta(0;\delta) v ∈ Δ ( 0 ; δ ) such that v ≠ 0 v \ne0 v = 0 and Re v ≥ η \operatorname{Re} v \ge\eta Re v ≥ η . Then
f i − 1 ( v ) ∩ W i ∗ ⊂ S η ∗ ⊂ W i ∗ . f_i^{-1}(v) \cap W_i^* \subset S_\eta^* \subset W_i^*. f i − 1 ( v ) ∩ W i ∗ ⊂ S η ∗ ⊂ W i ∗ .
By Theorems 3.15 and 3.17 , π 1 ( f i − 1 ( v ) ∩ W i ∗ ) → π 1 ( W i ∗ ) \pi_1(f_i^{-1}(v) \cap W_i^*) \to\pi_1(W_i^*) π 1 ( f i − 1 ( v ) ∩ W i ∗ ) → π 1 ( W i ∗ ) is surjective. Hence the factorization
π 1 ( f i − 1 ( v ) ∩ W i ∗ ) → π 1 ( S η ∗ ) → π 1 ( W i ∗ ) \pi_1(f_i^{-1}(v) \cap W_i^*) \to\pi_1(S_\eta^*) \to\pi_1(W_i^*) π 1 ( f i − 1 ( v ) ∩ W i ∗ ) → π 1 ( S η ∗ ) → π 1 ( W i ∗ )
implies that π 1 ( S η ∗ ) → π 1 ( W i ∗ ) \pi_1(S_\eta^*) \to\pi_1(W_i^*) π 1 ( S η ∗ ) → π 1 ( W i ∗ ) is surjective.
For the connectedness statements, in each of the three cases, the corresponding punctured superlevel set is a locally trivial fibration with connected fibers over { z ∈ Δ ∗ ( 0 ; δ ) : Re z ≥ η } \{z \in\Delta^*(0;\delta) : \operatorname{Re} z \ge\eta\} { z ∈ Δ ∗ ( 0 ; δ ) : Re z ≥ η } , which is connected. Hence the punctured superlevel set is connected. If η > 0 \eta> 0 η > 0 , this punctured superlevel set is already the desired superlevel set. If η ≤ 0 \eta\le0 η ≤ 0 , the desired superlevel set is its closure in W i ∗ W_i^* W i ∗ , W l ∗ \mathcal{W}_l^* W l ∗ , or W i † W_i^\dagger W i † , respectively, obtained by adjoining the remaining part of the central fiber. Hence S η ∗ S_\eta^* S η ∗ , S η ∗ \mathcal{S}_\eta^* S η ∗ , and S η † S_\eta^\dagger S η † are connected.
The sublevel and band assertions follow by almost the same argument; which we shall not repeat.
From now on, we include the mixed-type critical neighborhoods among the type I critical neighborhoods. Thus, unless otherwise specified, a type I critical neighborhood may refer either to an ordinary type I neighborhood W i ∗ W_i^* W i ∗ or to a mixed-type neighborhood W i † W_i^\dagger W i † .
Construction of a flow on the infinite cyclic covering
In this section, we introduce and modify the flow used by Zhou in [Zho22] to deform paths whose endpoints lie in a level set into that level set. Under the assumption dim ψ ( X ) ≥ 2 \dim\psi(X) \ge2 dim ψ ( X ) ≥ 2 , we have shown that case (A) cannot occur, so every critical neighborhood of type I (if any) satisfies case (B). We will show that this leads to (ii) of Theorem 3.1 .
The main point is to control the flow away from the critical neighborhoods so that it remains in Y Y Y , equivalently, so that it does not cross the boundary P ∪ Q P \cup Q P ∪ Q . Near the polar divisor P P P , this will follow from the boundary behavior established in § 3.2 . Along Q ′ ′ Q^{\prime\prime} Q ′′ , away from the critical neighborhoods, we choose the flow tangent to Q ′ ′ Q^{\prime\prime} Q ′′ , so that it preserves its complement. Inside critical neighborhoods of type I, type II, or mixed type, such a flow can still be constructed, but it may fail to have the desired increasing property for the primitive function, then we replace the relevant portions of the path using the local 1-connectedness results established in Theorems 3.15, 3.17 , 3.19 and 3.21.
We now lift all the objects constructed in § 3.4 to the regular Z \mathbb{Z} Z -cover
p : M ^ ⟶ M , p:\widehat{M}\longrightarrow M, p : M ⟶ M ,
corresponding to χ \chi χ , which can be defined explicitly as
M ^ : = M × S 1 R = { ( x , t ) ∈ M × R : Φ ( x ) = e 2 π − 1 t } . (14) \widehat{M}:=M\times_{S^{1}}\mathbb{R}=\{(x,t)\in M\times\mathbb{R}:\Phi(x)=e^{2\pi\sqrt{-1}t}\}. \tag*{(14)} M := M × S 1 R = {( x , t ) ∈ M × R : Φ ( x ) = e 2 π − 1 t } . (14)
where M M M is the real-oriented blowup of Y ‾ \overline{Y} Y along the polar divisor P P P . To simplify notation, we also use P P P to denote the polar boundary β − 1 ( P ) ⊂ M \beta^{-1}(P)\subset M β − 1 ( P ) ⊂ M . We write T T T for a fixed generator of the deck transformation group of M ^ \widehat{M} M . For any connected subset B ⊂ M B\subset M B ⊂ M , we use B ^ \widehat{B} B to denote one chosen connected component of p − 1 ( B ) p^{-1}(B) p − 1 ( B ) . Its other connected lifts, when distinct, are obtained from B ^ \widehat{B} B by applying suitable powers of T T T . Thus, throughout the following discussion, a hat indicates a chosen connected component of the inverse image under p p p .
We first consider the critical neighborhoods of the two types. On every critical neighborhood under consideration, the restriction of ξ \xi ξ admits a single-valued holomorphic primitive: namely, ξ = d f i \xi=df_i ξ = d f i on a type I neighborhood W i W_i W i , and ξ = d f l \xi=df_l ξ = d f l on a type II neighborhood W l \mathcal{W}_l W l . Hence Re ξ \operatorname{Re}\xi Re ξ is exact on each such neighborhood, so ∫ γ Re ξ = 0 \int_\gamma\operatorname{Re}\xi=0 ∫ γ Re ξ = 0 for every loop γ \gamma γ contained in it. Therefore the character χ \chi χ vanishes on the fundamental group of every critical neighborhood. It follows that the inverse image of each critical neighborhood in the Z \mathbb{Z} Z -cover is a disjoint union of copies of that neighborhood. More precisely,
p − 1 ( W i ) = ∐ n ∈ Z W ^ i , n = ∐ n ∈ Z T n W ^ i , 0 , p^{-1}(W_i)=\coprod_{n\in\mathbb{Z}}\widehat{W}_{i,n}=\coprod_{n\in\mathbb{Z}}T^n\widehat{W}_{i,0}, p − 1 ( W i ) = n ∈ Z ∐ W i , n = n ∈ Z ∐ T n W i , 0 ,
where each W ^ i , n \widehat{W}_{i,n} W i , n is mapped homeomorphically onto W i W_i W i ; similarly, for type II,
p − 1 ( W l ) = ∐ n ∈ Z W ^ l , n = ∐ n ∈ Z T n W ^ l , 0 , p^{-1}(\mathcal{W}_l)=\coprod_{n\in\mathbb{Z}}\widehat{\mathcal{W}}_{l,n}=\coprod_{n\in\mathbb{Z}}T^n\widehat{\mathcal{W}}_{l,0}, p − 1 ( W l ) = n ∈ Z ∐ W l , n = n ∈ Z ∐ T n W l , 0 ,
where each W ^ l , n \widehat{\mathcal{W}}_{l,n} W l , n is mapped homeomorphically onto W l \mathcal{W}_l W l .
Finally, we consider the lifting of the polar boundary P P P . Recall that, near a component P i ⊂ P P_i\subset P P i ⊂ P , the logarithmic form ψ ∗ ω \psi^*\omega ψ ∗ ω has nonzero residue along P i P_i P i . More precisely, on a punctured tubular neighborhood of P i P_i P i , we have ξ = ψ ∗ ω \xi=\psi^*\omega ξ = ψ ∗ ω , and for a small positively oriented loop γ \gamma γ around P i P_i P i ,
∫ γ Re ξ = m i ≠ 0. \int_\gamma\operatorname{Re}\xi=m_i\ne0. ∫ γ Re ξ = m i = 0.
The lift of a simply connected local component of P i P_i P i (along its tubular neighborhood N i N_i N i ) contains m i m_i m i -many components by standard theory of covering space; indeed, we have ∫ Re ξ : π 1 ( N i ) → Z \int\operatorname{Re}\xi:\pi_1(N_i)\to\mathbb{Z} ∫ Re ξ : π 1 ( N i ) → Z has image m i Z m_i\mathbb{Z} m i Z , and its lift to M ^ \widehat{M} M has ∣ Z / m i Z ∣ = m i |\mathbb{Z}/m_i\mathbb{Z}|=m_i ∣ Z / m i Z ∣ = m i many connected components, labeled as P ^ i j \widehat{P}_i^j P i j , 1 ≤ j ≤ m i 1\le j\le m_i 1 ≤ j ≤ m i . We emphasize that this is a local picture. For a tubular neighborhood of the entire component P i P_i P i , loops in P i P_i P i may contribute additional ξ \xi ξ -periods. Thus the image of ∫ Re ξ : π 1 ( N i ) → Z \int\operatorname{Re}\xi:\pi_1(N_i)\to\mathbb{Z} ∫ Re ξ : π 1 ( N i ) → Z may be d i Z d_i\mathbb{Z} d i Z for some d i ∣ m i d_i\mid m_i d i ∣ m i . Correspondingly, the inverse image of the global tubular neighborhood, as well as P i P_i P i , has d i d_i d i -many connected components.
Figure 2. Lifting of polar boundary
The fig. 2 shows the case when d i = m i = 3 d_i=m_i=3 d i = m i = 3 , with red dashed arrows meaning the orbit of a point on P ^ i \widehat{P}_i P i under action of T T T ; in particular, T T T does not need to preserve each connected component. Those γ ^ i \widehat{\gamma}_i γ i corresponds to loops with ξ \xi ξ -period 0 0 0 , and is a model of the lift of two types critical neighborhoods; each D i D_i D i means the fundamental domain.
Remark 3.22. — This explains why we work on the Z \mathbb{Z} Z -covering rather than on the universal covering. Our argument only needs to control the real part of a holomorphic 1 1 1 -form ξ \xi ξ corresponding to χ \chi χ . Passing to the ker χ \ker\chi ker χ -cover kills exactly the periods of Re ξ \operatorname{Re}\xi Re ξ detected by χ \chi χ , so that
d u = p ∗ Re ξ du=p^*\operatorname{Re}\xi d u = p ∗ Re ξ
for a globally defined primitive function u u u , while the deck group records precisely the translations detected by χ \chi χ . Passing instead to the universal covering would introduce the full fundamental group as deck group and substantially more information than is needed for the one-dimensional dynamic behavior determined by ξ \xi ξ .
Now we take care of the 1 1 1 -connectedness of the level sets of the primitive function
u : M ^ → R , u:\widehat{M}\to\mathbb{R}, u : M → R ,
defined by
d u = p ∗ Re ξ , u ( T x ) = u ( x ) + 1 , ∀ x ∈ M ^ . du=p^*\operatorname{Re}\xi,\qquad u(Tx)=u(x)+1,\qquad\forall x\in\widehat{M}. d u = p ∗ Re ξ , u ( T x ) = u ( x ) + 1 , ∀ x ∈ M .
With Φ \Phi Φ defined downstairs in eq. (8) and M ^ \widehat{M} M defined in eq. (14) , we have
e 2 π − 1 u = Φ ∘ p . e^{2\pi\sqrt{-1}u}=\Phi\circ p. e 2 π − 1 u = Φ ∘ p .
Recall that Y Y Y is a dense open subset of M M M . By abuse of notation, we continue to denote the oriented boundary β − 1 ( P ) \beta^{-1}(P) β − 1 ( P ) by P P P , and also denote the proper transform of Q Q Q in M M M by Q Q Q . Then
M ∖ Y = P ∪ Q = ∂ M ∪ Q . M\setminus Y=P\cup Q=\partial M\cup Q. M ∖ Y = P ∪ Q = ∂ M ∪ Q .
Write Y ^ = p − 1 ( Y ) \widehat{Y}=p^{-1}(Y) Y = p − 1 ( Y ) and, for t ∈ R t\in\mathbb{R} t ∈ R regular value of u u u , put
L t : = u − 1 ( t ) ∩ Y ^ . L_t:=u^{-1}(t)\cap\widehat{Y}. L t := u − 1 ( t ) ∩ Y .
We will show that any smooth path γ : [ 0 , 1 ] → Y ^ \gamma:[0,1]\to\widehat{Y} γ : [ 0 , 1 ] → Y whose endpoints p 0 = γ ( 0 ) p_0=\gamma(0) p 0 = γ ( 0 ) , p 1 = γ ( 1 ) p_1=\gamma(1) p 1 = γ ( 1 ) lie in L t L_t L t can be deformed relative to its endpoints into a path entirely contained in L t L_t L t . This shows that L t L_t L t is connected and that π 1 ( L t ) ↠ π 1 ( Y ^ ) \pi_1(L_t)\twoheadrightarrow\pi_1(\widehat{Y}) π 1 ( L t ) ↠ π 1 ( Y ) is surjective.
First, after perturbing γ \gamma γ slightly relative to its endpoints, we may assume that it meets L t L_t L t transversely, including at its endpoint germs. Then γ \gamma γ intersects L t L_t L t at finitely many points, and we can divide γ \gamma γ into finitely many subpaths accordingly, whose endpoints lie in L t L_t L t and whose interiors are disjoint from L t L_t L t ; in other words, each subpath lies entirely either below or above L t L_t L t , except for its two endpoints.
We may therefore treat these finitely many subpaths separately. It suffices to consider the case in which γ ( 0 ) , γ ( 1 ) ∈ L t \gamma(0),\gamma(1)\in L_t γ ( 0 ) , γ ( 1 ) ∈ L t , and u ( γ ( a ) ) < t u(\gamma(a))<t u ( γ ( a )) < t for all a ∈ ( 0 , 1 ) a\in(0,1) a ∈ ( 0 , 1 ) ; the case u ( γ ( a ) ) > t u(\gamma(a))>t u ( γ ( a )) > t is treated in the same way using the negative flow. For γ \gamma γ below the regular level set, let
r 0 = min a ∈ [ 0 , 1 ] u ( γ ( a ) ) < t . (15) r_0=\min_{a\in[0,1]}u(\gamma(a))<t. \tag*{(15)} r 0 = a ∈ [ 0 , 1 ] min u ( γ ( a )) < t . (15)
Next, we construct a vector field v ^ \hat{\mathbf{v}} v ^ which is the lift of a suitable smooth vector field v \mathbf{v} v on M M M . The flow defined by v ^ \hat{\mathbf{v}} v ^ will push γ \gamma γ upwards until it lies entirely inside the regular level set L t L_t L t .
For the critical neighborhoods of both types, let δ > 0 \delta>0 δ > 0 be the uniform constant chosen above. Choose c > 0 c>0 c > 0 sufficiently small such that 5 c < δ 5c<\delta 5 c < δ . We moreover choose c c c so that, for every type I critical neighborhood, all values z z z with 0 < ∣ z ∣ ≤ 5 c 0<|z|\leq5c 0 < ∣ z ∣ ≤ 5 c are regular values of f i f_i f i , and, in the mixed case, the restrictions of f i f_i f i to the components of Q ′ ′ Q'' Q ′′ have no critical points over this punctured disc. For a type II critical neighborhood, f l f_l f l is already a submersion, so the corresponding regularity is automatic.
For each type I critical neighborhood W i W_i W i , choose relatively compact neighborhoods of Σ i \Sigma_i Σ i :
W i ′ ′ ′ ⋐ W i ′ ′ ⋐ W i ∩ f i − 1 ( Δ ( 0 ; c ) ‾ ) ⋐ W i ′ ⋐ W i , W_i''' \Subset W_i'' \Subset W_i\cap f_i^{-1}\bigl(\overline{\Delta(0;c)}\bigr)\Subset W_i'\Subset W_i, W i ′′′ ⋐ W i ′′ ⋐ W i ∩ f i − 1 ( Δ ( 0 ; c ) ) ⋐ W i ′ ⋐ W i ,
where W i ′ = W i ∩ f i − 1 ( Δ ( 0 ; 3 c ) ) W_i'=W_i\cap f_i^{-1}\bigl(\Delta(0;3c)\bigr) W i ′ = W i ∩ f i − 1 ( Δ ( 0 ; 3 c ) ) . Define
T i ′ = T i ∩ W i ′ and R i ′ = W i ∩ { ∣ f i ∣ = 3 c } . T_i'=T_i\cap W_i'\qquad\text{and}\qquad R_i'=W_i\cap\{|f_i|=3c\}. T i ′ = T i ∩ W i ′ and R i ′ = W i ∩ { ∣ f i ∣ = 3 c } .
Similarly, for each type II critical neighborhood W l \mathcal{W}_l W l , choose relatively compact neighborhoods of p l p_l p l :
W l ′ ′ ′ ⋐ W l ′ ′ ⋐ W l ∩ f l − 1 ( Δ ( 0 ; c ) ‾ ) ⋐ W l ′ ⋐ W l , \mathcal{W}_l'''\Subset\mathcal{W}_l''\Subset\mathcal{W}_l\cap f_l^{-1}\bigl(\overline{\Delta(0;c)}\bigr)\Subset\mathcal{W}_l'\Subset\mathcal{W}_l, W l ′′′ ⋐ W l ′′ ⋐ W l ∩ f l − 1 ( Δ ( 0 ; c ) ) ⋐ W l ′ ⋐ W l ,
where W l ′ = W l ∩ f l − 1 ( Δ ( 0 ; 3 c ) ) \mathcal{W}_l'=\mathcal{W}_l\cap f_l^{-1}\bigl(\Delta(0;3c)\bigr) W l ′ = W l ∩ f l − 1 ( Δ ( 0 ; 3 c ) ) . Define
T l ′ = T l ∩ W l ′ and R l ′ = W l ∩ { ∣ f l ∣ = 3 c } . \mathcal{T}_l'=\mathcal{T}_l\cap\mathcal{W}_l'\qquad\text{and}\qquad\mathcal{R}_l'=\mathcal{W}_l\cap\{|f_l|=3c\}. T l ′ = T l ∩ W l ′ and R l ′ = W l ∩ { ∣ f l ∣ = 3 c } .
In the mixed case, which from now on is included among the type I critical neighborhoods, the same construction is used. By our previous choice of the type I neighborhoods, every component of Q ′ ′ Q'' Q ′′ is disjoint from T i T_i T i and exits W i W_i W i only through the R i R_i R i -side. The same property therefore holds for the smaller neighborhood W i ′ W_i' W i ′ with R i ′ R_i' R i ′ in place of R i R_i R i .
Set α = Re ξ \alpha=\operatorname{Re}\xi α = Re ξ . We choose a smooth vector field v \mathbf{v} v on M M M such that:
(a) v \mathbf{v} v vanishes in the neighborhoods W i ′ ′ ′ W_i''' W i ′′′ and W l ′ ′ ′ \mathcal{W}_l''' W l ′′′ for each i i i and $l.
(b) Outside ⋃ i = 1 m W i ′ ′ ∪ ⋃ l = 1 m W l ′ ′ \bigcup_{i=1}^{m}W_i''\cup\bigcup_{l=1}^{m}\mathcal{W}_l'' ⋃ i = 1 m W i ′′ ∪ ⋃ l = 1 m W l ′′ , we have
d arg ( Φ ) ( v ) = 2 π ⟺ α ( v ) = 1. d\arg(\Phi)(\mathbf{v})=2\pi\Longleftrightarrow\alpha(\mathbf{v})=1. d arg ( Φ ) ( v ) = 2 π ⟺ α ( v ) = 1.
(c) v \mathbf{v} v is tangent to every boundary component P i P_i P i of the real-oriented blowup M M M . For example, in the local coordinates used in section 3.3 3.3 3.3 , it may be chosen to be 2 π m i ∂ ∂ θ i \frac{2\pi}{m_i}\frac{\partial}{\partial\theta_i} m i 2 π ∂ θ i ∂ due to Theorem 3.13; around the corner, we can mix them to take v \mathbf{v} v of the form
2 π m 1 2 + m 2 2 ( m 1 ∂ ∂ θ 1 + m 2 ∂ ∂ θ 2 ) . \frac{2\pi}{m_1^2+m_2^2}\left(m_1\frac{\partial}{\partial\theta_1}+m_2\frac{\partial}{\partial\theta_2}\right). m 1 2 + m 2 2 2 π ( m 1 ∂ θ 1 ∂ + m 2 ∂ θ 2 ∂ ) .
Consequently, its flow preserves the boundary and M ∖ ∂ M M\setminus\partial M M ∖ ∂ M .
(d) On each boundary T i T_i T i of a type I critical neighborhood W i W_i W i , v \mathbf{v} v is tangent to T i T_i T i and satisfies α ( v ) = 1 \alpha(\mathbf{v})=1 α ( v ) = 1 .
(e) On each boundary T l \mathcal{T}_l T l of a type II critical neighborhood W l \mathcal{W}_l W l , v \mathbf{v} v is tangent to T l \mathcal{T}_l T l and satisfies α ( v ) = 1 \alpha(\mathbf{v})=1 α ( v ) = 1 .
(f) On Q ′ ′ Q'' Q ′′ away from the type I and type II critical neighborhoods, v \mathbf{v} v is tangent to Q ′ ′ Q'' Q ′′ and satisfies α ( v ) = 1 \alpha(\mathbf{v})=1 α ( v ) = 1 .
(g) On the annular regions W i ∩ f i − 1 ( Δ ( 0 ; 5 c ) ) ∖ W i ′ ′ W_i\cap f_i^{-1}\bigl(\Delta(0;5c)\bigr)\setminus W_i'' W i ∩ f i − 1 ( Δ ( 0 ; 5 c ) ) ∖ W i ′′ and W i ∩ f i − 1 ( Δ ( 0 ; 5 c ) ) ∖ W i ′ ′ \mathcal{W}_i\cap f_i^{-1}\bigl(\Delta(0;5c)\bigr)\setminus\mathcal{W}_i'' W i ∩ f i − 1 ( Δ ( 0 ; 5 c ) ) ∖ W i ′′ , the vector field may be chosen so that
d f i ( v ) = 1 ⟺ d Re f i ( v ) = 1 , d Im f i ( v ) = 0. df_i(\mathbf{v})=1\Longleftrightarrow d\operatorname{Re}f_i(\mathbf{v})=1,\ d\operatorname{Im}f_i(\mathbf{v})=0. d f i ( v ) = 1 ⟺ d Re f i ( v ) = 1 , d Im f i ( v ) = 0.
Proof. — Conditions (a)–(g) can be satisfied simultaneously as follows. For each critical neighborhood W W W (denoting either W i W_i W i or W l \mathcal{W}_l W l ), the bad loci are contained in W ′ ′ ′ W''' W ′′′ . Away from these loci, choose local fields tangent to Q Q Q and the prescribed boundary pieces T T T , with α ( v ) = 1 \alpha(\mathbf{v})=1 α ( v ) = 1 . On W ′ ∖ W ′ ′ ′ W'\setminus W''' W ′ ∖ W ′′′ , the local primitive and its restrictions to the T T T (or Q ′ ′ Q'' Q ′′ outside of critical neighborhoods) along which tangency is required are submersions. Hence these fields can also be chosen to satisfy d f i ( v ) = 1 df_i(\mathbf{v})=1 d f i ( v ) = 1 there, consistently with α ( v ) = 1 \alpha(\mathbf{v})=1 α ( v ) = 1 since d Re f i = α d\operatorname{Re}f_i=\alpha d Re f i = α . Multiply these fields by smooth cutoff functions that vanish on W ′ ′ ′ W^{\prime\prime\prime} W ′′′ and equal 1 1 1 on W ∖ W ′ ′ W \setminus W^{\prime\prime} W ∖ W ′′ . The fields therefore remain unchanged on W ′ ∖ W ′ ′ W^\prime\setminus W^{\prime\prime} W ′ ∖ W ′′ , preserving (g). Patching these fields with the normalized fields outside the critical neighborhoods by a partition of unity preserves the prescribed tangency and normalization conditions. The resulting field is smooth on M M M , tangent to every component of Q ′ ′ Q^{\prime\prime} Q ′′ and T T T , vanishes on the sets W ′ ′ ′ W^{\prime\prime\prime} W ′′′ , and satisfies α ( v ) = 1 \alpha(v)=1 α ( v ) = 1 outside the union of the sets W ′ ′ W^{\prime\prime} W ′′ .
Let v ^ \widehat{v} v denote the lift of v v v to M ^ \widehat{M} M . Equip M M M with a Riemannian metric and M ^ \widehat{M} M with its lifted metric. Since M M M is compact, v v v has bounded length, and hence so does v ^ \widehat{v} v . Moreover, the lifted metric on M ^ \widehat{M} M is complete, and v ^ \widehat{v} v is tangent to the lifted boundary
∂ M ^ = ⋃ P ^ i j . \partial\widehat{M}=\bigcup\widehat{P}^{\,j}_{i}. ∂ M = ⋃ P i j .
Therefore v ^ \widehat{v} v is complete, and its flow, denoted by Θ s \Theta_s Θ s , is defined for all s ∈ R s\in\mathbb{R} s ∈ R . Denote γ s ( a ) : = Θ s ( γ ( a ) ) \gamma_s(a):=\Theta_s(\gamma(a)) γ s ( a ) := Θ s ( γ ( a )) . Since
K γ : = { γ s ( a ) : 0 ≤ s ≤ c , 0 ≤ a ≤ 1 } (16) K_\gamma:=\{\gamma_s(a):0\leq s\leq c,\ 0\leq a\leq1\} \tag*{(16)} K γ := { γ s ( a ) : 0 ≤ s ≤ c , 0 ≤ a ≤ 1 } (16)
is compact, it intersects only finitely many lifts of the type I and type II critical neighborhoods. We relabel those which intersect K γ K_\gamma K γ as
W ^ 1 , … , W ^ n I ; W ~ 1 , … , W ~ n I I . \widehat{W}_1,\ldots,\widehat{W}_{n_{\mathrm I}};\qquad\widetilde{W}_1,\ldots,\widetilde{W}_{n_{\mathrm{II}}}. W 1 , … , W n I ; W 1 , … , W n II .
Denote their unions by W ^ \widehat{W} W and W ~ \widetilde{W} W , respectively. We use the same convention for W ^ i ′ \widehat{W}^{\prime}_i W i ′ , W ^ i ′ ′ \widehat{W}^{\prime\prime}_i W i ′′ , W ^ i ′ ′ ′ \widehat{W}^{\prime\prime\prime}_i W i ′′′ , R ^ i \widehat{R}_i R i , T ^ i \widehat{T}_i T i , et cetera.
Put Y ^ ≤ t : = Y ^ ∩ { u ≤ t } \widehat{Y}_{\leq t}:=\widehat{Y}\cap\{u\leq t\} Y ≤ t := Y ∩ { u ≤ t } . For the flow Θ s \Theta_s Θ s , we have the following.
Proposition 3.23. — (i) A trajectory cannot enter a critical neighborhood of type I, say W ^ i \widehat{W}_i W i , through T ^ i \widehat{T}_i T i .
(ii) A trajectory cannot enter a critical neighborhood of type II, say W ~ i \widetilde{W}_i W i , through T ~ i \widetilde{T}_i T i .
(iii) For every s ≥ 0 s\geq0 s ≥ 0 and x ∈ Y ^ ≤ t ∖ ( W ^ ∪ W ~ ) x\in\widehat{Y}_{\leq t}\setminus(\widehat{W}\cup\widetilde{W}) x ∈ Y ≤ t ∖ ( W ∪ W ) such that Θ [ 0 , s ] ( x ) ∩ ( W ^ ∪ W ~ ) = ∅ \Theta_{[0,s]}(x)\cap(\widehat{W}\cup\widetilde{W})=\varnothing Θ [ 0 , s ] ( x ) ∩ ( W ∪ W ) = ∅ , we have
Θ s ( x ) ∈ Y ^ \Theta_s(x)\in\widehat{Y} Θ s ( x ) ∈ Y and
u ( Θ s ( x ) ) = u ( x ) + s . (17) u(\Theta_s(x))=u(x)+s. \tag*{(17)} u ( Θ s ( x )) = u ( x ) + s . (17) In particular, the same eq. (17) holds for x x x belonging T ^ i \widehat{T}_i T i and T ~ i \widetilde{T}_i T i .
Proof. — Statements (i) and (ii) follow from (d) and (e), respectively, since the vector field is tangent to T ^ i \widehat{T}_i T i and T ~ i \widetilde{T}_i T i . For (iii), by (c), the flow preserves the lift of the polar boundary p − 1 ( P ) p^{-1}(P) p − 1 ( P ) and hence also its complement. Thus a trajectory starting in Y ^ \widehat{Y} Y cannot cross the polar boundary. Moreover, Q ′ Q^\prime Q ′ is contained in the type I critical neighborhoods, while all transverse intersection points of distinct components of Q ′ ′ Q^{\prime\prime} Q ′′ which require special treatment are contained in mixed type I or type II critical neighborhoods. Away from these critical neighborhoods, Q ′ ′ Q^{\prime\prime} Q ′′ is smooth and, by (f), the vector field is tangent to Q ′ ′ Q^{\prime\prime} Q ′′ . Hence a trajectory satisfying the hypothesis in (iii) cannot cross the lift of Q Q Q either. Therefore, Θ s ( x ) ∈ Y ^ \Theta_s(x)\in\widehat{Y} Θ s ( x ) ∈ Y .
For eq. (17) , recall that on the cyclic cover 2 π d u = p ∗ d arg Φ 2\pi\,du=p^*d\arg\Phi 2 π d u = p ∗ d arg Φ . Outside W ^ ∪ W ~ \widehat{W}\cup\widetilde{W} W ∪ W , we have d arg Φ ( v ) = 2 π d\arg\Phi(v)=2\pi d arg Φ ( v ) = 2 π , and therefore d d s u ( Θ s ( x ) ) = d u ( v ^ ) = 1 \frac{d}{ds}u(\Theta_s(x))=du(\widehat{v})=1 d s d u ( Θ s ( x )) = d u ( v ) = 1 . Integrating from 0 0 0 to s s s gives u ( Θ s ( x ) ) = u ( x ) + s u(\Theta_s(x))=u(x)+s u ( Θ s ( x )) = u ( x ) + s . The same computation applies along T ^ i \widehat{T}_i T i and T ~ i \widetilde{T}_i T i by (d) and (e).
The same short-time control applies to critical neighborhoods of both types. Suppose first that a trajectory enters a type I critical neighborhood through R ^ i \widehat{R}_i R i . Let s 0 s_0 s 0 be the first time at which it meets R ^ i \widehat{R}_i R i , and put y = Θ s 0 ( x ) y=\Theta_{s_0}(x) y = Θ s 0 ( x ) . Thus ∣ f i ∘ p ( y ) ∣ = 3 c |f_i\circ p(y)|=3c ∣ f i ∘ p ( y ) ∣ = 3 c . For 0 ≤ s ≤ c 0\leq s\leq c 0 ≤ s ≤ c , it follows that
c < 2 c ≤ ∣ f i ∘ p ( Θ s ( y ) ) ∣ ≤ 4 c < 5 c . c<2c\leq\lvert f_i\circ p(\Theta_s(y))\rvert\leq4c<5c. c < 2 c ≤ ∣ f i ∘ p ( Θ s ( y ))∣ ≤ 4 c < 5 c .
Thus the trajectory remains in the region c < ∣ f i ∣ < 5 c c<|f_i|<5c c < ∣ f i ∣ < 5 c for 0 ≤ s ≤ c 0\leq s\leq c 0 ≤ s ≤ c , and in particular cannot enter W ^ i ′ ′ = { ∣ f i ∣ ≤ c } \widehat{W}^{\prime\prime}_i=\{|f_i|\leq c\} W i ′′ = { ∣ f i ∣ ≤ c } during this time; thus we have f i ∘ p ( Θ s ( y ) ) = f i ∘ p ( y ) + s f_i\circ p(\Theta_s(y))=f_i\circ p(y)+s f i ∘ p ( Θ s ( y )) = f i ∘ p ( y ) + s . Exactly the same argument applies if a trajectory enters a type II critical neighborhood through R ^ i ′ \widehat{\mathcal{R}}'_i R i ′ .
Moreover, on each connected lift W ^ i \widehat{W}_i W i or W ^ i \widehat{\mathcal{W}}_i W i , there is a real constant such that Re ( f i ∘ p ) \operatorname{Re}(f_i\circ p) Re ( f i ∘ p ) or Re ( f i ∘ p ) \operatorname{Re}(f_i\circ p) Re ( f i ∘ p ) differs from u u u by that constant. Hence in either case, u ( Θ s ( y ) ) = u ( y ) + s u(\Theta_s(y))=u(y)+s u ( Θ s ( y )) = u ( y ) + s for 0 ≤ s ≤ c 0\leq s\leq c 0 ≤ s ≤ c . In particular, for y ∈ R ^ i ′ y\in\widehat{\mathcal{R}}'_i y ∈ R i ′ or y ∈ R ^ i ′ y\in\widehat{\mathcal{R}}'_i y ∈ R i ′ ,
u ( Θ s ( y ) ) = u ( y ) + s , 0 ≤ s ≤ c . (18) u(\Theta_s(y))=u(y)+s,\qquad0\leq s\leq c. \tag*{(18)} u ( Θ s ( y )) = u ( y ) + s , 0 ≤ s ≤ c . (18)
We now consider the finitely many lifted critical neighborhoods intersecting K γ K_\gamma K γ . Since the same argument applies to both types, we write W ^ i \widehat{W}_i W i for either type, with f i f_i f i , R ^ i ′ \widehat{\mathcal{R}}'_i R i ′ , and T ^ i ′ \widehat{\mathcal{T}}'_i T i ′ denoting the corresponding objects. After a small perturbation, assume that γ \gamma γ is transverse to their boundaries.
Let γ ∣ [ a , b ] \gamma|_{[a,b]} γ ∣ [ a , b ] be a component lying in W ^ i ′ \widehat{W}'_i W i ′ , with x = γ ( a ) x=\gamma(a) x = γ ( a ) and y = γ ( b ) y=\gamma(b) y = γ ( b ) . Then x , y ∈ R ^ i ′ ∪ T ^ i ′ x,y\in\widehat{\mathcal{R}}'_i\cup\widehat{\mathcal{T}}'_i x , y ∈ R i ′ ∪ T i ′ . By eqs. (17) and (18) , both points may be flowed for time c c c with desired u u u -increase:
u ( Θ c ( x ) ) = u ( x ) + c , u ( Θ c ( y ) ) = u ( y ) + c . u(\Theta_c(x))=u(x)+c,\qquad u(\Theta_c(y))=u(y)+c. u ( Θ c ( x )) = u ( x ) + c , u ( Θ c ( y )) = u ( y ) + c .
The flowed subpath Θ c ( γ ∣ [ a , b ] ) \Theta_c(\gamma|_{[a,b]}) Θ c ( γ ∣ [ a , b ] ) is homotopic relative to its endpoints to
σ x − 1 ∗ γ ∣ [ a , b ] ∗ σ y \sigma_x^{-1}*\gamma|_{[a,b]}*\sigma_y σ x − 1 ∗ γ ∣ [ a , b ] ∗ σ y
where σ x ( s ) = Θ s ( x ) \sigma_x(s)=\Theta_s(x) σ x ( s ) = Θ s ( x ) and σ y ( s ) = Θ s ( y ) \sigma_y(s)=\Theta_s(y) σ y ( s ) = Θ s ( y ) . This path lies in the corresponding critical neighborhood. Since on W ^ i \widehat{W}_i W i we have Re ( f i ∘ p ) = u + c i \operatorname{Re}(f_i\circ p)=u+c_i Re ( f i ∘ p ) = u + c i for some c i ∈ R c_i\in\mathbb{R} c i ∈ R , Theorem 3.21 allows us to replace it, relative to its endpoints, by a path on which
r 0 + c ≤ u ≤ t . r_0+c\leq u\leq t. r 0 + c ≤ u ≤ t .
We first apply Θ c \Theta_c Θ c to the whole path γ \gamma γ , and then perform this replacement on every portion passing through a critical neighborhood. Outside these neighborhoods, Theorem 3.23 gives an increase of u u u by c c c . Hence the resulting path is homotopic to γ \gamma γ and has minimum u u u -value at least r 0 + c r_0+c r 0 + c . Meanwhile, outside the critical neighborhoods, the choice of vector field prevents the flow from crossing P ^ \widehat{P} P or Q ^ \widehat{Q} Q , while inside the critical neighborhoods the replacements are carried out entirely in the corresponding neighborhoods with Q Q Q deleted, namely W ^ i ∗ \widehat{W}_i^* W i ∗ , W ^ i ∗ \widehat{\mathcal{W}}_i^* W i ∗ , or W ^ i † \widehat{W}_i^\dagger W i † . Hence the resulting path remains in Y ^ \widehat{Y} Y .
To deform γ \gamma γ into L t L_t L t without ever pushing any point above the level t t t , we allow the flow time to vary with the initial point. For x ∈ Y ^ ≤ t x\in\widehat{Y}_{\leq t} x ∈ Y ≤ t , define τ c ( x ) = min { c , t − u ( x ) } \tau_c(x)=\min\{c,t-u(x)\} τ c ( x ) = min { c , t − u ( x )} . Thus 0 ≤ τ c ( x ) ≤ c 0\leq\tau_c(x)\leq c 0 ≤ τ c ( x ) ≤ c , and τ c ( x ) = 0 \tau_c(x)=0 τ c ( x ) = 0 for x ∈ L t x\in L_t x ∈ L t . On the portions of the path where the flow is controlled, we deform by
H ( s , a ) = Θ s τ c ( γ ( a ) ) ( γ ( a ) ) , 0 ≤ s , a ≤ 1 , H(s,a)=\Theta_{s\tau_c(\gamma(a))}(\gamma(a)),\qquad0\leq s,a\leq1, H ( s , a ) = Θ s τ c ( γ ( a )) ( γ ( a )) , 0 ≤ s , a ≤ 1 ,
so that u ( H ( 1 , a ) ) = u ( γ ( a ) ) + τ c ( γ ( a ) ) = min { u ( γ ( a ) ) + c , t } u(H(1,a))=u(\gamma(a))+\tau_c(\gamma(a))=\min\{u(\gamma(a))+c,t\} u ( H ( 1 , a )) = u ( γ ( a )) + τ c ( γ ( a )) = min { u ( γ ( a )) + c , t } . After the local replacements inside the critical neighborhoods, denote the resulting path by γ c \gamma_c γ c , we obtain
min a ∈ [ 0 , 1 ] u ( γ c ( a ) ) ≥ min { r 0 + c , t } , u ( γ c ( a ) ) ≤ t for all a ∈ [ 0 , 1 ] . \min_{a\in[0,1]}u(\gamma_c(a))\geq\min\{r_0+c,t\},\qquad u(\gamma_c(a))\leq t\quad\text{for all }a\in[0,1]. a ∈ [ 0 , 1 ] min u ( γ c ( a )) ≥ min { r 0 + c , t } , u ( γ c ( a )) ≤ t for all a ∈ [ 0 , 1 ] .
Consequently, for a fixed constant c > 0 c>0 c > 0 , after finitely many iterations the whole path lies in L t L_t L t , and the endpoints are fixed throughout. During this process, at successive iterations we consider the compact sets K γ , K γ c , K γ 2 c , … K_\gamma,K_{\gamma_c},K_{\gamma_{2c}},\ldots K γ , K γ c , K γ 2 c , … ; although the lifted critical neighborhoods intersecting these compact sets may vary from one iteration to the next, only finitely many occur at each step, and we can apply the replacement in Theorem 3.21 for finitely many times.
Now recall that u : M ^ → R u:\widehat{M}\to\mathbb{R} u : M → R is proper. For a regular value t t t , the full level u − 1 ( t ) ⊂ M ^ u^{-1}(t)\subset\widehat{M} u − 1 ( t ) ⊂ M is therefore compact. Moreover, by the local description of the lift of P ∪ Q P\cup Q P ∪ Q , the level L t = u − 1 ( t ) ∩ Y ^ L_t=u^{-1}(t)\cap\widehat{Y} L t = u − 1 ( t ) ∩ Y has finite topological type: it is obtained from the compact level u − 1 ( t ) u^{-1}(t) u − 1 ( t ) by deleting the lifted boundary P ^ \widehat{P} P and Q ^ \widehat{Q} Q . Although the lifted boundary may have infinitely many components, these form a locally finite family, so only finitely many meet L t L_t L t . The compact pair ( L t ‾ , L t ‾ ∩ ( P ^ ∪ Q ^ ) ) (\overline{L_t},\overline{L_t}\cap(\widehat{P}\cup\widehat{Q})) ( L t , L t ∩ ( P ∪ Q )) admits a finite compatible triangulation. Consequently, L t = L t ‾ ∖ ( P ^ ∪ Q ^ ) L_t=\overline{L_t}\setminus(\widehat{P}\cup\widehat{Q}) L t = L t ∖ ( P ∪ Q ) has the homotopy type of a finite CW complex. In particular, π 1 ( L t ) \pi_1(L_t) π 1 ( L t ) is finitely generated. We can now prove Theorem 3.3 .
Proof of Theorem 3.3 . — Suppose first that dim ψ ( X ) ≥ 2 \dim\psi(X) \ge2 dim ψ ( X ) ≥ 2 . By the preceding argument, for every regular value t t t the inclusion L t ↪ Y ^ L_t \hookrightarrow\widehat{Y} L t ↪ Y induces a surjection on fundamental groups. Since π 1 ( L t ) \pi_1(L_t) π 1 ( L t ) is finitely generated, so is π 1 ( Y ^ ) ≅ ker χ \pi_1(\widehat{Y}) \cong\ker\chi π 1 ( Y ) ≅ ker χ , contradicting the assumption. Hence dim ψ ( X ) ≤ 1 \dim\psi(X) \le1 dim ψ ( X ) ≤ 1 . Since χ ≠ 0 \chi\ne0 χ = 0 , the image of ψ \psi ψ cannot be a point, and therefore dim ψ ( X ) = 1 \dim\psi(X) = 1 dim ψ ( X ) = 1 .
Let Z = ψ ( X ) ‾ Z a r ⊂ A Z = \overline{\psi(X)}^{\mathrm{Zar}} \subset A Z = ψ ( X ) Zar ⊂ A and let C 0 → Z C_0 \to Z C 0 → Z be the normalization. Then ψ \psi ψ factors through C 0 C_0 C 0 . Taking the Stein factorization after passing to proper compactifications and then restricting back to X X X , we obtain a surjective algebraic morphism f : X → C f : X \to C f : X → C onto a smooth quasi-projective curve C C C with connected general fiber, together with a homomorphism χ C : π 1 o r b ( C ) → Z \chi_C : \pi_1^{\mathrm{orb}}(C) \to\mathbb{Z} χ C : π 1 orb ( C ) → Z such that χ = χ C o r b ∘ f ∗ o r b \chi= \chi_C^{\mathrm{orb}} \circ f_*^{\mathrm{orb}} χ = χ C orb ∘ f ∗ orb .
We claim that the orbifold curve associated with f f f is hyperbolic. Let g g g be the genus of a smooth projective completion of C C C , let r r r be the number of points removed, and let m i m_i m i be the multiplicities of the multiple fibers (see section 2). Since χ ≠ 0 \chi\ne0 χ = 0 factors through f ∗ o r b f_*^{\mathrm{orb}} f ∗ orb , the orbifold fundamental group admits a nontrivial homomorphism to Z \mathbb{Z} Z , and in particular the underlying curve has nontrivial first homology. Thus 2 g + r − 1 ≥ 1 2g+r-1 \ge1 2 g + r − 1 ≥ 1 . Hence χ o r b ( C ) ≤ 0 \chi^{\mathrm{orb}}(C) \le0 χ orb ( C ) ≤ 0 .
Suppose that χ o r b ( C ) = 0 \chi^{\mathrm{orb}}(C) = 0 χ orb ( C ) = 0 , then eq. (3) implies that π 1 o r b ( C ) \pi_1^{\mathrm{orb}}(C) π 1 orb ( C ) is free abelian, and hence ker χ C \ker\chi_C ker χ C is finitely generated. Moreover, the orbifold homotopy exact sequence eq. (1) shows that ker ( f ∗ o r b ) \ker(f_*^{\mathrm{orb}}) ker ( f ∗ orb ) is a quotient of the finitely generated fundamental group of a general fiber. Thus ker χ \ker\chi ker χ , being an extension of ker χ C \ker\chi_C ker χ C by ker ( f ∗ o r b ) \ker(f_*^{\mathrm{orb}}) ker ( f ∗ orb ) , is finitely generated, a contradiction. Therefore χ o r b ( C ) < 0 \chi^{\mathrm{orb}}(C) < 0 χ orb ( C ) < 0 , and the orbifold curve associated with f f f is hyperbolic. □
We now prove (ii) of Theorem 3.1 for each regular level set under the assumption that (i) does not hold. If dim ψ ( X ) = 1 \dim\psi(X) = 1 dim ψ ( X ) = 1 , the preceding argument gives precisely the factorization in (i). Hence we must have dim ψ ( X ) ≥ 2 \dim\psi(X) \ge2 dim ψ ( X ) ≥ 2 .
We now pass from the surface case to X X X . Throughout this argument, we keep the form α = Re ( ψ ∗ ω ) \alpha= \operatorname{Re}(\psi^*\omega) α = Re ( ψ ∗ ω ) on X X X fixed. Let p : X ^ → X p : \widehat{X} \to X p : X → X be the covering associated with ker χ \ker\chi ker χ , and choose a primitive u : X ^ → R u : \widehat{X} \to\mathbb{R} u : X → R satisfying d u = p ∗ α \mathrm{d}u = p^*\alpha d u = p ∗ α . For t ∈ R t \in\mathbb{R} t ∈ R , put L t = u − 1 ( t ) ⊂ X ^ L_t = u^{-1}(t) \subset\widehat{X} L t = u − 1 ( t ) ⊂ X .
Let x 1 , x 2 ∈ L t x_1,x_2 \in L_t x 1 , x 2 ∈ L t . By Theorem 3.11, we may choose a smooth connected surface Y ⊂ X Y \subset X Y ⊂ X containing p ( x 1 ) p(x_1) p ( x 1 ) and p ( x 2 ) p(x_2) p ( x 2 ) such that π 1 ( Y ) → ∼ π 1 ( X ) \pi_1(Y) \xrightarrow{\sim} \pi_1(X) π 1 ( Y ) ∼ π 1 ( X ) and dim ψ ( Y ) = 2 \dim\psi(Y) = 2 dim ψ ( Y ) = 2 . Then Y ^ = p − 1 ( Y ) \widehat{Y} = p^{-1}(Y) Y = p − 1 ( Y ) is connected, contains x 1 , x 2 x_1,x_2 x 1 , x 2 , and its inclusion induces an isomorphism π 1 ( Y ^ ) → ∼ π 1 ( X ^ ) \pi_1(\widehat{Y}) \xrightarrow{\sim} \pi_1(\widehat{X}) π 1 ( Y ) ∼ π 1 ( X ) . The primitive used on Y ^ \widehat{Y} Y is the restriction u ∣ Y ^ u|_{\widehat{Y}} u ∣ Y .
The conclusion established for regular levels of u ∣ Y ^ u|_{\widehat{Y}} u ∣ Y above shows that L t ∩ Y ^ L_t \cap\widehat{Y} L t ∩ Y is path-connected. Thus x 1 x_1 x 1 and x 2 x_2 x 2 can be joined inside L t L_t L t , proving that L t L_t L t is path-connected. Moreover, choosing a base point in L t ∩ Y ^ L_t \cap\widehat{Y} L t ∩ Y , the composite π 1 ( L t ∩ Y ^ ) → π 1 ( Y ^ ) → ∼ π 1 ( X ^ ) \pi_1(L_t \cap\widehat{Y}) \to\pi_1(\widehat{Y}) \xrightarrow{\sim} \pi_1(\widehat{X}) π 1 ( L t ∩ Y ) → π 1 ( Y ) ∼ π 1 ( X ) is surjective. Since this composite factors through π 1 ( L t ) \pi_1(L_t) π 1 ( L t ) , the inclusion induces a surjection
π 1 ( L t ) ⟶ π 1 ( X ^ ) . \pi_1(L_t) \longrightarrow\pi_1(\widehat{X}). π 1 ( L t ) ⟶ π 1 ( X ) .
Therefore ( X ^ , L t ) (\widehat{X},L_t) ( X , L t ) is 1-connected for every t ∈ R t \in\mathbb{R} t ∈ R regular.
From regular level set to any level set
To pass from regular level sets to arbitrary level sets, the missing ingredient is an analogue of [38 Lemma 9] , and we include its proof here for completeness. This is precisely the remaining case of Theorem 3.21, namely η 1 = η 2 = 0 \eta_1 = \eta_2 = 0 η 1 = η 2 = 0 .
Proposition 3.24. — For a critical neighborhood W i W_i W i of type I, set
E i ∗ = W i ∗ ∩ { Re f i = 0 } . E_i^* = W_i^* \cap\{\operatorname{Re} f_i = 0\}. E i ∗ = W i ∗ ∩ { Re f i = 0 } . Then the pair ( W i ∗ , E i ∗ ) (W_i^*, E_i^*) ( W i ∗ , E i ∗ ) is 1-connected.
Proof. — Let I = { i s : − δ < s < δ } ⊂ Δ ( 0 , δ ) I = \{is : -\delta< s < \delta\} \subset\Delta(0,\delta) I = { i s : − δ < s < δ } ⊂ Δ ( 0 , δ ) be the imaginary axis, and write I = I − ∪ { 0 } ∪ I + I = I^- \cup\{0\} \cup I^+ I = I − ∪ { 0 } ∪ I + , where I + I^+ I + means the positive/negative imaginary segment. Then E i ∗ = f i − 1 ( I ) ∩ W i ∗ E_i^* = f_i^{-1}(I) \cap W_i^* E i ∗ = f i − 1 ( I ) ∩ W i ∗ . By Theorem 3.15, the fibers f i − 1 ( z ) ∩ W i ∗ f_i^{-1}(z) \cap W_i^* f i − 1 ( z ) ∩ W i ∗ are connected for every z ≠ 0 z \ne0 z = 0 . Since
f i : W i ∗ ∗ ⟶ Δ ∗ f_i : W_i^{**} \longrightarrow\Delta^* f i : W i ∗∗ ⟶ Δ ∗
is a locally trivial fibration, it follows that f i − 1 ( I + ) ∩ W i ∗ f_i^{-1}(I^+) \cap W_i^* f i − 1 ( I + ) ∩ W i ∗ and f i − 1 ( I − ) ∩ W i ∗ f_i^{-1}(I^-) \cap W_i^* f i − 1 ( I − ) ∩ W i ∗ are connected.
We first show that f i − 1 ( 0 ) ∩ W i ∗ f_i^{-1}(0) \cap W_i^* f i − 1 ( 0 ) ∩ W i ∗ is contained in the closure of f i − 1 ( I + ) ∩ W i ∗ f_i^{-1}(I^+) \cap W_i^* f i − 1 ( I + ) ∩ W i ∗ . Let x ∈ f i − 1 ( 0 ) ∩ W i ∗ x \in f_i^{-1}(0) \cap W_i^* x ∈ f i − 1 ( 0 ) ∩ W i ∗ . If x x x lies in the interior of W i W_i W i , then the germ of f i f_i f i at x x x is nonconstant. @page 26/41 starts_mid=0 ends_mid=1 Proof of Theorem 3.3 . — Suppose first that dim ψ ( X ) ≥ 2 \dim\psi(X) \ge2 dim ψ ( X ) ≥ 2 . By the preceding argument, for every regular value t t t the inclusion L t ↪ Y ^ L_t \hookrightarrow\widehat{Y} L t ↪ Y induces a surjection on fundamental groups. Since π 1 ( L t ) \pi_1(L_t) π 1 ( L t ) is finitely generated, so is π 1 ( Y ^ ) ≅ ker χ \pi_1(\widehat{Y}) \cong\ker\chi π 1 ( Y ) ≅ ker χ , contradicting the assumption. Hence dim ψ ( X ) ≤ 1 \dim\psi(X) \le1 dim ψ ( X ) ≤ 1 . Since χ ≠ 0 \chi\ne0 χ = 0 , the image of ψ \psi ψ cannot be a point, and therefore dim ψ ( X ) = 1 \dim\psi(X) = 1 dim ψ ( X ) = 1 .
Let Z = ψ ( X ) ‾ Z a r ⊂ A Z = \overline{\psi(X)}^{\mathrm{Zar}} \subset A Z = ψ ( X ) Zar ⊂ A and let C 0 → Z C_0 \to Z C 0 → Z be the normalization. Then ψ \psi ψ factors through C 0 C_0 C 0 . Taking the Stein factorization after passing to proper compactifications and then restricting back to X X X , we obtain a surjective algebraic morphism f : X → C f : X \to C f : X → C onto a smooth quasi-projective curve C C C with connected general fiber, together with a homomorphism χ C : π 1 o r b ( C ) → Z \chi_C : \pi_1^{\mathrm{orb}}(C) \to\mathbb{Z} χ C : π 1 orb ( C ) → Z such that χ = χ C o r b ∘ f ∗ o r b \chi= \chi_C^{\mathrm{orb}} \circ f_*^{\mathrm{orb}} χ = χ C orb ∘ f ∗ orb .
We claim that the orbifold curve associated with f f f is hyperbolic. Let g g g be the genus of a smooth projective completion of C C C , let r r r be the number of points removed, and let m i m_i m i be the multiplicities of the multiple fibers (see section 2). Since χ ≠ 0 \chi\ne0 χ = 0 factors through f ∗ o r b f_*^{\mathrm{orb}} f ∗ orb , the orbifold fundamental group admits a nontrivial homomorphism to Z \mathbb{Z} Z , and in particular the underlying curve has nontrivial first homology. Thus 2 g + r − 1 ≥ 1 2g+r-1 \ge1 2 g + r − 1 ≥ 1 . Hence χ o r b ( C ) ≤ 0 \chi^{\mathrm{orb}}(C) \le0 χ orb ( C ) ≤ 0 .
Suppose that χ o r b ( C ) = 0 \chi^{\mathrm{orb}}(C) = 0 χ orb ( C ) = 0 , then eq. (3) implies that π 1 o r b ( C ) \pi_1^{\mathrm{orb}}(C) π 1 orb ( C ) is free abelian, and hence ker χ C \ker\chi_C ker χ C is finitely generated. Moreover, the orbifold homotopy exact sequence eq. (1) shows that ker ( f ∗ o r b ) \ker(f_*^{\mathrm{orb}}) ker ( f ∗ orb ) is a quotient of the finitely generated fundamental group of a general fiber. Thus ker χ \ker\chi ker χ , being an extension of ker χ C \ker\chi_C ker χ C by ker ( f ∗ o r b ) \ker(f_*^{\mathrm{orb}}) ker ( f ∗ orb ) , is finitely generated, a contradiction. Therefore χ o r b ( C ) < 0 \chi^{\mathrm{orb}}(C) < 0 χ orb ( C ) < 0 , and the orbifold curve associated with f f f is hyperbolic. □
We now prove (ii) of Theorem 3.1 for each regular level set under the assumption that (i) does not hold. If dim ψ ( X ) = 1 \dim\psi(X) = 1 dim ψ ( X ) = 1 , the preceding argument gives precisely the factorization in (i). Hence we must have dim ψ ( X ) ≥ 2 \dim\psi(X) \ge2 dim ψ ( X ) ≥ 2 .
We now pass from the surface case to X X X . Throughout this argument, we keep the form α = Re ( ψ ∗ ω ) \alpha= \operatorname{Re}(\psi^*\omega) α = Re ( ψ ∗ ω ) on X X X fixed. Let p : X ^ → X p : \widehat{X} \to X p : X → X be the covering associated with ker χ \ker\chi ker χ , and choose a primitive u : X ^ → R u : \widehat{X} \to\mathbb{R} u : X → R satisfying d u = p ∗ α \mathrm{d}u = p^*\alpha d u = p ∗ α . For t ∈ R t \in\mathbb{R} t ∈ R , put L t = u − 1 ( t ) ⊂ X ^ L_t = u^{-1}(t) \subset\widehat{X} L t = u − 1 ( t ) ⊂ X .
Let x 1 , x 2 ∈ L t x_1,x_2 \in L_t x 1 , x 2 ∈ L t . By Theorem 3.11, we may choose a smooth connected surface Y ⊂ X Y \subset X Y ⊂ X containing p ( x 1 ) p(x_1) p ( x 1 ) and p ( x 2 ) p(x_2) p ( x 2 ) such that π 1 ( Y ) → ∼ π 1 ( X ) \pi_1(Y) \xrightarrow{\sim} \pi_1(X) π 1 ( Y ) ∼ π 1 ( X ) and dim ψ ( Y ) = 2 \dim\psi(Y) = 2 dim ψ ( Y ) = 2 . Then Y ^ = p − 1 ( Y ) \widehat{Y} = p^{-1}(Y) Y = p − 1 ( Y ) is connected, contains x 1 , x 2 x_1,x_2 x 1 , x 2 , and its inclusion induces an isomorphism π 1 ( Y ^ ) → ∼ π 1 ( X ^ ) \pi_1(\widehat{Y}) \xrightarrow{\sim} \pi_1(\widehat{X}) π 1 ( Y ) ∼ π 1 ( X ) . The primitive used on Y ^ \widehat{Y} Y is the restriction u ∣ Y ^ u|_{\widehat{Y}} u ∣ Y .
The conclusion established for regular levels of u ∣ Y ^ u|_{\widehat{Y}} u ∣ Y above shows that L t ∩ Y ^ L_t \cap\widehat{Y} L t ∩ Y is path-connected. Thus x 1 x_1 x 1 and x 2 x_2 x 2 can be joined inside L t L_t L t , proving that L t L_t L t is path-connected. Moreover, choosing a base point in L t ∩ Y ^ L_t \cap\widehat{Y} L t ∩ Y , the composite π 1 ( L t ∩ Y ^ ) → π 1 ( Y ^ ) → ∼ π 1 ( X ^ ) \pi_1(L_t \cap\widehat{Y}) \to\pi_1(\widehat{Y}) \xrightarrow{\sim} \pi_1(\widehat{X}) π 1 ( L t ∩ Y ) → π 1 ( Y ) ∼ π 1 ( X ) is surjective. Since this composite factors through π 1 ( L t ) \pi_1(L_t) π 1 ( L t ) , the inclusion induces a surjection
π 1 ( L t ) ⟶ π 1 ( X ^ ) . \pi_1(L_t) \longrightarrow\pi_1(\widehat{X}). π 1 ( L t ) ⟶ π 1 ( X ) .
Therefore ( X ^ , L t ) (\widehat{X},L_t) ( X , L t ) is 1-connected for every t ∈ R t \in\mathbb{R} t ∈ R regular.
From regular level set to any level set
To pass from regular level sets to arbitrary level sets, the missing ingredient is an analogue of [38 Lemma 9] , and we include its proof here for completeness. This is precisely the remaining case of Theorem 3.21, namely η 1 = η 2 = 0 \eta_1 = \eta_2 = 0 η 1 = η 2 = 0 .
Proposition 3.24. — For a critical neighborhood W i W_i W i of type I, set
E i ∗ = W i ∗ ∩ { Re f i = 0 } . E_i^* = W_i^* \cap\{\operatorname{Re} f_i = 0\}. E i ∗ = W i ∗ ∩ { Re f i = 0 } . Then the pair ( W i ∗ , E i ∗ ) (W_i^*, E_i^*) ( W i ∗ , E i ∗ ) is 1-connected.
Proof. — Let I = { i s : − δ < s < δ } ⊂ Δ ( 0 , δ ) I = \{is : -\delta< s < \delta\} \subset\Delta(0,\delta) I = { i s : − δ < s < δ } ⊂ Δ ( 0 , δ ) be the imaginary axis, and write I = I − ∪ { 0 } ∪ I + I = I^- \cup\{0\} \cup I^+ I = I − ∪ { 0 } ∪ I + , where I + I^+ I + means the positive/negative imaginary segment. Then E i ∗ = f i − 1 ( I ) ∩ W i ∗ E_i^* = f_i^{-1}(I) \cap W_i^* E i ∗ = f i − 1 ( I ) ∩ W i ∗ . By Theorem 3.15, the fibers f i − 1 ( z ) ∩ W i ∗ f_i^{-1}(z) \cap W_i^* f i − 1 ( z ) ∩ W i ∗ are connected for every z ≠ 0 z \ne0 z = 0 . Since
f i : W i ∗ ∗ ⟶ Δ ∗ f_i : W_i^{**} \longrightarrow\Delta^* f i : W i ∗∗ ⟶ Δ ∗
is a locally trivial fibration, it follows that f i − 1 ( I + ) ∩ W i ∗ f_i^{-1}(I^+) \cap W_i^* f i − 1 ( I + ) ∩ W i ∗ and f i − 1 ( I − ) ∩ W i ∗ f_i^{-1}(I^-) \cap W_i^* f i − 1 ( I − ) ∩ W i ∗ are connected.
We first show that f i − 1 ( 0 ) ∩ W i ∗ f_i^{-1}(0) \cap W_i^* f i − 1 ( 0 ) ∩ W i ∗ is contained in the closure of f i − 1 ( I + ) ∩ W i ∗ f_i^{-1}(I^+) \cap W_i^* f i − 1 ( I + ) ∩ W i ∗ . Let x ∈ f i − 1 ( 0 ) ∩ W i ∗ x \in f_i^{-1}(0) \cap W_i^* x ∈ f i − 1 ( 0 ) ∩ W i ∗ . If x x x lies in the interior of W i W_i W i , then the germ of f i f_i f i at x x x is nonconstant. Hence there is a holomorphic disc through x x x on which f i f_i f i is nonconstant, and the open mapping theorem gives points mapping to I + I^+ I + arbitrarily close to x x x . If x ∈ T i x \in T_i x ∈ T i , the same conclusion follows from the local triviality of f i ∣ T i : T i → Δ ( 0 , δ ) f_i|_{T_i}: T_i \to\Delta(0,\delta) f i ∣ T i : T i → Δ ( 0 , δ ) . Since the central fiber does not meet R i R_i R i , these are all possibilities. Thus
f i − 1 ( 0 ) ∩ W i ∗ ⊂ f i − 1 ( I + ) ∩ W i ∗ . f_i^{-1}(0) \cap W_i^* \subset f_i^{-1}(I^+) \cap W_i^*. f i − 1 ( 0 ) ∩ W i ∗ ⊂ f i − 1 ( I + ) ∩ W i ∗ .
It follows that
( f i − 1 ( I + ) ∪ f i − 1 ( 0 ) ) ∩ W i ∗ \left(f_i^{-1}(I^+) \cup f_i^{-1}(0)\right) \cap W_i^* ( f i − 1 ( I + ) ∪ f i − 1 ( 0 ) ) ∩ W i ∗
is connected.
It remains to connect the negative part to the positive part. Choose y ∈ f i − 1 ( I − ) ∩ T i y \in f_i^{-1}(I^-) \cap T_i y ∈ f i − 1 ( I − ) ∩ T i . Such a point exists because f i ∣ T i f_i|_{T_i} f i ∣ T i is a locally trivial fibration with nonempty fibers. Lifting a segment in I I I joining f i ( y ) f_i(y) f i ( y ) to a point of I + I^+ I + gives a path in T i ∩ E i ∗ T_i \cap E_i^* T i ∩ E i ∗ joining the negative part to the positive part. Hence E i ∗ E_i^* E i ∗ is connected.
Finally, choose z ∈ I + z \in I^+ z ∈ I + . Then f i − 1 ( z ) ∩ W i ⊂ E i ∗ f_i^{-1}(z) \cap W_i \subset E_i^* f i − 1 ( z ) ∩ W i ⊂ E i ∗ , and by Theorem 3.15 the composite π 1 ( f i − 1 ( z ) ∩ W i ) → π 1 ( E i ∗ ) → π 1 ( W i ∗ ) \pi_1(f_i^{-1}(z) \cap W_i) \to\pi_1(E_i^*) \to\pi_1(W_i^*) π 1 ( f i − 1 ( z ) ∩ W i ) → π 1 ( E i ∗ ) → π 1 ( W i ∗ ) is surjective. Therefore π 1 ( E i ∗ ) ↠ π 1 ( W i ∗ ) \pi_1(E_i^*) \twoheadrightarrow\pi_1(W_i^*) π 1 ( E i ∗ ) ↠ π 1 ( W i ∗ ) . Thus ( W i ∗ , E i ∗ ) (W_i^*, E_i^*) ( W i ∗ , E i ∗ ) is 1-connected.
The same argument gives the corresponding statements for type II and mixed type I critical neighborhoods.
Proposition 3.25. — For a critical neighborhood W l \mathcal{W}_l W l of type II, put
E l ∗ = W l ∗ ∩ { Re f l = 0 } . E_l^* = \mathcal{W}_l^* \cap\{\operatorname{Re} f_l = 0\}. E l ∗ = W l ∗ ∩ { Re f l = 0 } . Then the pair ( W l ∗ , E l ∗ ) (\mathcal{W}_l^*, E_l^*) ( W l ∗ , E l ∗ ) is 1-connected.
Proposition 3.26. — For a mixed type I critical neighborhood W i W_i W i , put
E i † = W i † ∩ { Re f i = 0 } . E_i^\dagger= W_i^\dagger\cap\{\operatorname{Re} f_i = 0\}. E i † = W i † ∩ { Re f i = 0 } . Then the pair ( W i † , E i † ) (W_i^\dagger, E_i^\dagger) ( W i † , E i † ) is 1-connected.
Now let t t t be a singular value of u u u , and still put L t : = u − 1 ( t ) ∩ Y ^ L_t := u^{-1}(t) \cap\widehat{Y} L t := u − 1 ( t ) ∩ Y . We show that every path in Y ^ \widehat{Y} Y with endpoints in L t L_t L t can be deformed, relative to its endpoints, into a path contained in L t L_t L t .
In contrast to the regular case, a path may intersect L t L_t L t infinitely many times, so we cannot simply perturb it to be transverse to L t L_t L t . Instead, we argue as follows: after a homotopy relative to the endpoints inside finitely many holomorphic coordinate balls in Y ^ \widehat{Y} Y , we may assume that γ \gamma γ is piecewise real-analytic. Since u u u is locally the real part of a holomorphic function, on each such piece u ∘ γ − t u \circ\gamma- t u ∘ γ − t either vanishes identically or has only finitely many zeros. Thus γ − 1 ( L t ) \gamma^{-1}(L_t) γ − 1 ( L t ) is a finite union of points and closed intervals. Leaving the portions contained in L t L_t L t fixed, we can therefore divide the remaining portions into finitely many subpaths whose endpoints lie in L t L_t L t and whose interiors lie entirely either below or above L t L_t L t . We treat these subpaths separately. We can thus argue on each subpath as done in section 3.5, with Theorems 3.21 and 3.24 to 3.26 used whenever needed.
We can also argue using a varying flow time on the whole path, with a sign depending on which side of L t L_t L t the point lies. This will also be used in section 4.2 for the general case. For x ∈ Y ^ x \in\widehat{Y} x ∈ Y , put
τ c ( x ) = min { c , ∣ t − u ( x ) ∣ } \tau_c(x) = \min\{c, |t-u(x)|\} τ c ( x ) = min { c , ∣ t − u ( x ) ∣ }
and define the signed flow time
σ c ( x ) = sgn ( t − u ( x ) ) τ c ( x ) , \sigma_c(x) = \operatorname{sgn}(t-u(x))\tau_c(x), σ c ( x ) = sgn ( t − u ( x )) τ c ( x ) ,
where sgn ( 0 ) = 0 \operatorname{sgn}(0)=0 sgn ( 0 ) = 0 . Thus σ c \sigma_c σ c is continuous (near L t L_t L t it is in fact smooth), it is positive on { u < t } \{u<t\} { u < t } and negative on { u > t } \{u>t\} { u > t } , and it vanishes on L t L_t L t . We deform the whole path by
H ( s , a ) = Θ s σ c ( γ ( a ) ) ( γ ( a ) ) , 0 ≤ s , a ≤ 1. H(s,a) = \Theta_{s\sigma_c(\gamma(a))}(\gamma(a)), \qquad0 \leq s,a \leq1. H ( s , a ) = Θ s σ c ( γ ( a )) ( γ ( a )) , 0 ≤ s , a ≤ 1.
As in section 3.5, the choice of vector field ensures that the flow preserves Y ^ \widehat{Y} Y ; we use this without further comment below. Since σ c \sigma_c σ c is continuous and the flow is smooth, this is a continuous homotopy fixing every point of L t L_t L t . On portions whose trajectories stay outside critical neighborhoods, Theorem 3.23(iii) and its negative-flow analogue give
u ( Θ s ( x ) ) = u ( x ) + s , u(\Theta_s(x)) = u(x) + s, u ( Θ s ( x )) = u ( x ) + s ,
and therefore ∣ t − u ( H ( 1 , a ) ) ∣ = max { ∣ t − u ( γ ( a ) ) ∣ − c , 0 } |t-u(H(1,a))|=\max\{|t-u(\gamma(a))|-c,0\} ∣ t − u ( H ( 1 , a )) ∣ = max { ∣ t − u ( γ ( a )) ∣ − c , 0 } .
Since
K γ : = { Θ s σ c ( γ ( a ) ) ( γ ( a ) ) : 0 ≤ s ≤ 1 , 0 ≤ a ≤ 1 } K_\gamma:= \{\Theta_{s\sigma_c(\gamma(a))}(\gamma(a)) : 0 \le s \le1,\ 0 \le a \le1\} K γ := { Θ s σ c ( γ ( a )) ( γ ( a )) : 0 ≤ s ≤ 1 , 0 ≤ a ≤ 1 }
is compact, it intersects only finitely many lifted critical neighborhoods. Inside these neighborhoods the same flow is applied, but the preceding equality for ∣ t − u ∣ |t-u| ∣ t − u ∣ need not hold. After the local replacements inside the critical neighborhoods thanks to Theorems 3.21 and 3.24 to 3.26, we have
min { r 0 + c , t } ≤ u ( γ c ( a ) ) ≤ max { t , R 0 − c } , for all 0 ≤ a ≤ 1 , \min\{r_0+c,t\}\le u(\gamma_c(a))\le\max\{t,R_0-c\},\qquad\text{for all }0\le a\le1, min { r 0 + c , t } ≤ u ( γ c ( a )) ≤ max { t , R 0 − c } , for all 0 ≤ a ≤ 1 ,
where r 0 = min a ∈ [ 0 , 1 ] u ( γ ( a ) ) ≤ t r_0=\min_{a\in[0,1]}u(\gamma(a))\le t r 0 = min a ∈ [ 0 , 1 ] u ( γ ( a )) ≤ t and R 0 = max a ∈ [ 0 , 1 ] u ( γ ( a ) ) ≥ t R_0=\max_{a\in[0,1]}u(\gamma(a))\ge t R 0 = max a ∈ [ 0 , 1 ] u ( γ ( a )) ≥ t . We denote the resulting path by γ c \gamma_c γ c .
We repeat the procedure for the resulting path, using at each stage the corresponding compact set K γ , K γ c , K γ 2 c , … K_\gamma,K_{\gamma_c},K_{\gamma_{2c}},\ldots K γ , K γ c , K γ 2 c , … . Although the lifted critical neighborhoods meeting these compact sets may vary from one step to the next, only finitely many occur at each step. As in the regular-level argument, after finitely many iterations the whole path lies in L t L_t L t . Since σ c = 0 \sigma_c=0 σ c = 0 on L t L_t L t , the endpoints remain fixed throughout. Hence every path in Y ^ \widehat{Y} Y with endpoints in L t L_t L t can be deformed, relative to its endpoints, into L t L_t L t .
The passage from Y ^ \widehat{Y} Y to X ^ \widehat{X} X follows the argument at the end of § 3.5 . Although formulated there for regular level sets, this argument applies verbatim to arbitrary level sets using the surface statement established above.
The compact and quasi-projective cases: a comparison
This section discusses the differences between the compact and noncompact settings for Theorem 1.2 and Theorem 3.3 , with particular emphasis on the differences in their proofs.
For the theorem of Napier–Ramachandran (Theorem 1.2), it was firstly stated and proved in [NR01], where the proof builds on the potential-theoretic methods developed in [29 ] . Roughly speaking, they use Evans–Selberg potential theory to construct pluriharmonic functions on complete (noncompact) Kähler manifolds under suitable various conditions. By constructing pluriharmonic functions with prescribed asymptotic behavior along different ends (for example, [29 Theorem 2.6] ) and studying the holomorphic foliations induced by their differentials, they use a cup-product lemma to obtain proper holomorphic mappings onto Riemann surfaces with connected fibers. This process is further developed in [33 ] using the notion of “filtered ends”. In particular, they proved:
Theorem 3.27 ([33 ] , Theorem 4.3). — A connected complete Kähler manifold of bounded geometry with at least three filtered ends admits a proper holomorphic mapping onto a Riemann surface.
For the infinite cyclic covering associated with an Z \mathbb{Z} Z -valued character whose kernel is not finitely generated, the covering has exactly two ends but infinitely many filtered ends, which is a restatement of Theorem 3.5. Thus, three filtered-ends theorem with an argument of descending to the Z \mathbb{Z} Z -covering provides another perspective on Theorem 1.2 .
In the previous sections (§ 3.1 to § 3.5 ), we provide a proof of this theorem in the quasi-projective setting. The reader may also use the argument in § 3.1 to § 3.5 to recover a proof of Theorem 1.2 in this compact case.
Remark 3.28. — Theorem 1.2 can be proved using other methods, at least for the compact Kähler case.
(a) One may use the min–max argument of Delzant–Py to prove Theorem 1.2 ; see [34 Chapter 5] . In fact, [34 Theorem 5.1] , applied to the primitive function u u u on M ^ \widehat{M} M , gives the conclusion. As pointed out there, this argument applies only to the rational case. (b) One may also modify the argument of [Zho22, Lemma 3] to give another proof of Theorem 1.2 , working instead on the universal covering and using Theorem 3.5. This approach applies to arbitrary real characters. Readers interested in this direction may similarly modify the argument of section 3 to work on the universal covering. We indicate the main idea. In this setting, it is unnecessary to keep track of the relative homotopy class of the path. If a level set in the universal covering is disconnected, then the Mayer–Vietoris sequence for reduced homology implies that either the corresponding sublevel set or the superlevel set is disconnected, or both. Suppose, for instance, that the superlevel set is disconnected. One then only needs to choose a path joining two of its components of the superlevel set and push the portion of the path lying in the sublevel set into the superlevel set, which produces a contradiction. See [Zho22, pp. 769–773].
Here, we point a difference between the compact Kähler case and the (noncompact) smooth quasi-projective case: how to explain conditions ( A ) (A) ( A ) and ( B ) (B) ( B ) in section 3.4. In the compact Kähler case, consider the singular holomorphic foliation F \mathcal{F} F defined by the closed holomorphic one-form ξ \xi ξ whose real part represents χ : π 1 ( X ) ↠ Z \chi:\pi_1(X)\twoheadrightarrow\mathbb{Z} χ : π 1 ( X ) ↠ Z . Condition ( A ) (A) ( A ) produces a compact regular leaf. The compact-leaf theorem (for example, see [30 , 1 , 13 ] ) then implies that all singular leaves are compact, and the corresponding leaf space gives a holomorphic fibration onto a compact Riemann surface with connected fibers. Here singular leaves are understood as the equivalence classes generated by connected fibers of local holomorphic primitives of ξ \xi ξ . This compact-leaf argument does not extend directly to arbitrary smooth quasi-projective varieties, as the following example shows.
Example 3.29. — Let X = ( P 1 × C ∗ ) ∖ { ( ∞ , 1 ) } X=(\mathbb{P}^1\times\mathbb{C}^*)\setminus\{(\infty,1)\} X = ( P 1 × C ∗ ) ∖ {( ∞ , 1 )} and let ξ = 1 2 π i d z 2 z 2 \xi=\frac{1}{2\pi i}\frac{dz_2}{z_2} ξ = 2 π i 1 z 2 d z 2 . Then Re ξ \operatorname{Re}\xi Re ξ represents an epimorphism π 1 ( X ) → Z \pi_1(X)\to\mathbb{Z} π 1 ( X ) → Z , and the leaves of the foliation defined by ξ \xi ξ are the fibers of the projection X → C ∗ X\to\mathbb{C}^* X → C ∗ . For z ≠ 1 z\ne1 z = 1 , the leaf P 1 × { z } \mathbb{P}^1\times\{z\} P 1 × { z } is compact, whereas the leaf over z = 1 z=1 z = 1 is ( P 1 ∖ { ∞ } ) × { 1 } ≅ C (\mathbb{P}^1\setminus\{\infty\})\times\{1\}\cong\mathbb{C} ( P 1 ∖ { ∞ }) × { 1 } ≅ C . Thus the union of the compact leaves is not closed in X X X .
With the function ψ \psi ψ defined in section 3.1, for compact Kähler case, we can use the above foliation argument to get the following implication, when the set of (type I) critical neighborhoods are not empty:
– Condition ( A ) (A) ( A ) ⟹ dim ψ ( X ) = 1 ⟺ \Longrightarrow\dim\psi(X)=1\Longleftrightarrow ⟹ dim ψ ( X ) = 1 ⟺ condition (i) Theorem 3.1 .
– Condition ( B ) (B) ( B ) ⟸ dim ψ ( X ) ≥ 2 ⟹ \Longleftarrow\dim\psi(X)\ge2\Longrightarrow ⟸ dim ψ ( X ) ≥ 2 ⟹ condition (ii) Theorem 3.1 .
While for noncompact quasi-projective manifolds, the same implications are from using Theorem 3.8(b) and Theorem 4.3(b) instead.
In one sentence, for compact Kähler manifolds the alternative may be detected either through the existence of a compact leaf, as in [29 , 30 , 13 Zho22, 22 ] , or through whether dim ψ ( X ) = 1 \dim\psi(X)=1 dim ψ ( X ) = 1 or dim ψ ( X ) ≥ 2 \dim\psi(X)\ge2 dim ψ ( X ) ≥ 2 , as in [38 ] ; for noncompact quasi-projective manifolds, however, the latter dimension criterion is the only one available to us.
Remark 3.30. — Note that in [38 ] , different components of Σ \Sigma Σ with the same ψ \psi ψ -values are put in a common critical neighborhood; while the critical neighborhood (of type I) in section 3.4 is chosen around the individual connected components Σ i \Sigma_i Σ i .
In the compact Kähler case, assume Σ ≠ ∅ \Sigma\ne\varnothing Σ = ∅ and interpret conditions ( A ) (A) ( A ) and ( B ) (B) ( B ) using Simpson’s critical neighborhoods, then:
– Condition ( A ) ⟺ dim ψ ( X ) = 1 ⟺ (A)\Longleftrightarrow\dim\psi(X)=1\Longleftrightarrow ( A ) ⟺ dim ψ ( X ) = 1 ⟺ (i) of Theorem 3.1 .
– Condition ( B ) ⟺ dim ψ ( X ) ≥ 2 ⟹ (B)\Longleftrightarrow\dim\psi(X)\ge2\Longrightarrow ( B ) ⟺ dim ψ ( X ) ≥ 2 ⟹ (ii) of Theorem 3.1 .
The alternatives (i) and (ii) may overlap, see Theorem 3.2. In particular, (ii) does not imply dim ψ ( X ) ≥ 2 \dim\psi(X)\ge2 dim ψ ( X ) ≥ 2 in general. If we further require in (i), the orbicurve C C C is hyperbolic, then (i) and (ii) are exclusive, and
– Condition (i) in Theorem 1.1 with C C C hyperbolic. ⟺ ker ( χ ) \Longleftrightarrow\ker(\chi) ⟺ ker ( χ ) is not finitely generated.
– Condition (ii) in Theorem 1.1 ⟺ ker ( χ ) \Longleftrightarrow\ker(\chi) ⟺ ker ( χ ) is finitely generated.
For rational characters in the smooth quasi-projective setting, the same equivalences hold with (ii) understood for the prescribed logarithmic representative α \alpha α ; see section 5.3. The rationality assumption and the restriction on the representative are essential here.
From the rational case to the general case
In this section, we modify the argument in section 3 to prove the Lefschetz alternative for an arbitrary nonzero real character in the smooth quasi-projective setting. We will explicitly identify the points at which essential modifications are required and explain how the relevant properties differ from those in the rational case.
Let X X X be a connected smooth complex quasi-projective variety, let G = π 1 ( X ) G = \pi_1(X) G = π 1 ( X ) , and let χ : G → R \chi: G \to\mathbb{R} χ : G → R be a nonzero character. Denote by r r r the Z \mathbb{Z} Z -rank of χ ( G ) ⊂ R \chi(G) \subset\mathbb{R} χ ( G ) ⊂ R . When r = 1 r = 1 r = 1 , this is the rational case discussed in section 3; when r ≥ 2 r \ge2 r ≥ 2 , we are in the irrational case, which is the main subject of present section.
The main difference between the two settings is that the primitive function u : X ^ → R u : \widehat{X} \to\mathbb{R} u : X → R is not proper in the irrational case. Thus the finite-generation argument used in the rational case no longer applies: the level sets need not have the homotopy type of a finite CW complex. We explain this further in section 5.3.
Choose an identification χ ( G ) ≅ Z r \chi(G) \cong\mathbb{Z}^r χ ( G ) ≅ Z r . By abuse of notation, we also denote by χ \chi χ the induced epimorphism G → Z r G \to\mathbb{Z}^r G → Z r through which the original character factors:
G → χ Z r χ ↓ ↓ R = R (19) \begin{CD}
G @>{\chi}>> \mathbb{Z}^r \\
@V{\chi}VV @VVV \\
\mathbb{R} @= \mathbb{R} \tag*{(19)}
\end{CD} G χ ↓ ⏐ R χ Z r ↓ ⏐ R (19)
We now prove the following.
Theorem 4.1. — Let X X X be a smooth connected complex quasi-projective variety and let χ : π 1 ( X ) → R \chi: \pi_1(X) \to\mathbb{R} χ : π 1 ( X ) → R be a nonzero group homomorphism. Let p : X ^ → X p : \widehat{X} \to X p : X → X be the regular covering corresponding to ker ( χ ) \ker(\chi) ker ( χ ) , so that π 1 ( X ^ ) = ker ( χ ) \pi_1(\widehat{X}) = \ker(\chi) π 1 ( X ) = ker ( χ ) . Then one of the following holds:
(i) There exists a surjective algebraic morphism with connected general fiber f : X → C f : X \to C f : X → C onto a smooth quasi-projective orbicurve such that χ \chi χ factors through π 1 o r b ( C ) \pi_1^{\mathrm{orb}}(C) π 1 orb ( C ) .
(ii) There exists a smooth closed 1 1 1 -form α \alpha α on X X X representing χ \chi χ such that, for a primitive function u : X ^ → R u : \widehat{X} \to\mathbb{R} u : X → R satisfying d u = p ∗ α du = p^*\alpha d u = p ∗ α , the pair ( X ^ , u − 1 ( t ) ) (\widehat{X},u^{-1}(t)) ( X , u − 1 ( t )) is 1 1 1 -connected for every t ∈ R t \in\mathbb{R} t ∈ R .
Modifications before lifting
For section 3.1, we still take the semi-abelian variety A 0 A_0 A 0 , the quasi-Albanese morphism a : X → A 0 a : X \to A_0 a : X → A 0 associated to X X X , and write λ 0 ∈ H 1 ( A 0 ; R ) \lambda_0 \in H^1(A_0;\mathbb{R}) λ 0 ∈ H 1 ( A 0 ; R ) such that a ∗ λ 0 = χ a^*\lambda_0 = \chi a ∗ λ 0 = χ . Let B B B be the sum of connected positive dimensional algebraic subgroups where λ 0 \lambda_0 λ 0 vanishes, and ψ : X → A : = A 0 / B \psi: X \to A := A_0/B ψ : X → A := A 0 / B be the quasi-Albanese map associated to χ \chi χ . Theorem 3.6 still holds by our construction, while Theorem 3.7 is certainly not true. In fact, due to the same argument, we now have ℓ ≤ r \ell\le r ℓ ≤ r . To avoid any possible confusion, we restate the lemmas from § 3 that remain valid in the present setting, at the risk of some repetition.
Lemma 4.2. — For every positive dimensional connected algebraic subgroup H ⊂ A H \subset A H ⊂ A , the restriction of λ \lambda λ to H H H is not identically zero.
Now we make the modification of § 3.2 . There is still a natural compactification of A A A as follows. Recall that we have a short exact sequence of algebraic groups
1 → T = ( C ∗ ) ℓ → A → π A a b → 1. 1 \to T = (\mathbb{C}^{*})^{\ell} \to A \xrightarrow{\pi} A_{\mathrm{ab}} \to1. 1 → T = ( C ∗ ) ℓ → A π A ab → 1.
Consider the action of T T T on P ℓ \mathbb{P}^{\ell} P ℓ defined by
( t 1 , … , t ℓ ) ⋅ [ z 0 : z 1 : ⋯ : z ℓ ] = [ z 0 : t 1 z 1 : ⋯ : t ℓ z ℓ ] . (t_1,\ldots,t_{\ell}) \cdot[z_0:z_1:\cdots:z_{\ell}] = [z_0:t_1z_1:\cdots:t_{\ell}z_{\ell}]. ( t 1 , … , t ℓ ) ⋅ [ z 0 : z 1 : ⋯ : z ℓ ] = [ z 0 : t 1 z 1 : ⋯ : t ℓ z ℓ ] .
The open orbit satisfying z 0 z 1 ⋯ z ℓ ≠ 0 z_0z_1\cdots z_{\ell}\ne0 z 0 z 1 ⋯ z ℓ = 0 is isomorphic to T ⊂ P ℓ T\subset\mathbb{P}^{\ell} T ⊂ P ℓ . Treating A A A as a principal T T T -bundle over A a b A_{\mathrm{ab}} A ab , we take the corresponding P ℓ \mathbb{P}^{\ell} P ℓ -bundle over A a b A_{\mathrm{ab}} A ab , which is smooth projective, denoted as A ‾ \overline{A} A . Then A ‾ ∖ A \overline{A}\setminus A A ∖ A consists of finitely many divisors:
D A = ⋃ 0 ≤ i ≤ ℓ D i : = ⋃ 0 ≤ i ≤ ℓ { z i = 0 } . D_A=\bigcup_{0\leq i\leq\ell}D_i:=\bigcup_{0\leq i\leq\ell}\{z_i=0\}. D A = 0 ≤ i ≤ ℓ ⋃ D i := 0 ≤ i ≤ ℓ ⋃ { z i = 0 } .
Unlike in the rank-one case, these boundary divisors may intersect; nevertheless, their union is a simple normal crossings divisor. Under the same notation as in § 3.2 , we have
A ‾ = P A a b ( O A a b ⊕ L 1 ⊕ ⋯ ⊕ L ℓ ) , \overline{A}=\mathbb{P}_{A_{\mathrm{ab}}}\left(\mathcal{O}_{A_{\mathrm{ab}}}\oplus L_1\oplus\cdots\oplus L_{\ell}\right), A = P A ab ( O A ab ⊕ L 1 ⊕ ⋯ ⊕ L ℓ ) ,
where L i ∈ Pic 0 ( A a b ) L_i\in\operatorname{Pic}^{0}(A_{\mathrm{ab}}) L i ∈ Pic 0 ( A ab ) is the line bundle corresponding to the i i i -th coordinate character of T = ( C ∗ ) ℓ T=(\mathbb{C}^{*})^{\ell} T = ( C ∗ ) ℓ .
Take γ k \gamma_k γ k be the k k k -th generator of π 1 ( T ) ≅ Z ℓ \pi_1(T)\cong\mathbb{Z}^{\ell} π 1 ( T ) ≅ Z ℓ and take a k = λ ( γ k ) ∈ R a_k=\lambda(\gamma_k)\in\mathbb{R} a k = λ ( γ k ) ∈ R . For each 1 ≤ k ≤ ℓ 1\leq k\leq\ell 1 ≤ k ≤ ℓ , it is standard that there exists a global logarithmic 1 1 1 -form η k \eta_k η k on A A A , extending logarithmically to A ‾ \overline{A} A , such that
Res D j η k = { 1 , j = k , − 1 , j = 0 , 0 , j ≠ 0 , k . \operatorname{Res}_{D_j}\eta_k=
\begin{cases}
1, & j=k,\\
-1, & j=0,\\
0, & j\ne0,k.
\end{cases} Res D j η k = ⎩ ⎨ ⎧ 1 , − 1 , 0 , j = k , j = 0 , j = 0 , k .
Locally, after trivializing L k L_k L k , the form η k \eta_k η k is given by d log ( w k / w 0 ) d\log(w_k/w_0) d log ( w k / w 0 ) up to a holomorphic 1 1 1 -form pulled back from the base A a b A_{\mathrm{ab}} A ab .
Then
λ − [ Re ∑ k = 1 ℓ a k 2 π − 1 η k ] \lambda-\left[\operatorname{Re}\sum_{k=1}^{\ell}\frac{a_k}{2\pi\sqrt{-1}}\eta_k\right] λ − [ Re k = 1 ∑ ℓ 2 π − 1 a k η k ]
has trivial image in Gr 2 W H 1 ( A ; R ) \operatorname{Gr}^{W}_{2}H^{1}(A;\mathbb{R}) Gr 2 W H 1 ( A ; R ) , and we can find λ a b ∈ H 1 ( A a b ; R ) \lambda_{\mathrm{ab}}\in H^{1}(A_{\mathrm{ab}};\mathbb{R}) λ ab ∈ H 1 ( A ab ; R ) such that
λ = ∑ k = 1 ℓ [ Re a k 2 π − 1 η k ] + π ∗ λ a b . \lambda=\sum_{k=1}^{\ell}\left[\operatorname{Re}\frac{a_k}{2\pi\sqrt{-1}}\eta_k\right]+\pi^{*}\lambda_{\mathrm{ab}}. λ = k = 1 ∑ ℓ [ Re 2 π − 1 a k η k ] + π ∗ λ ab .
Let ω a b \omega_{\mathrm{ab}} ω ab be a holomorphic one-form on A a b A_{\mathrm{ab}} A ab such that [ Re ω a b ] = λ a b [\operatorname{Re}\omega_{\mathrm{ab}}]=\lambda_{\mathrm{ab}} [ Re ω ab ] = λ ab and
ω = ∑ i = 1 ℓ a i 2 π − 1 η i + π ∗ ω a b . (20) \omega=\sum_{i=1}^{\ell}\frac{a_i}{2\pi\sqrt{-1}}\eta_i+\pi^{*}\omega_{\mathrm{ab}}. \tag*{(20)} ω = i = 1 ∑ ℓ 2 π − 1 a i η i + π ∗ ω ab . (20)
is the logarithmic one-form on A A A with [ Re ω ] = λ [\operatorname{Re}\omega]=\lambda [ Re ω ] = λ , and ω \omega ω has purely imaginary residues at each divisor in D A D_A D A . Due to Theorem 4.2, each a k ≠ 0 a_k\ne0 a k = 0 and a 1 + ⋯ + a ℓ ≠ 0 a_1+\cdots+a_{\ell}\ne0 a 1 + ⋯ + a ℓ = 0 . In conclusion, the residue of ω \omega ω along each irreducible component of D A D_A D A is purely imaginary and nonzero. Also, due to the same reason as in Theorem 3.8(b), we have
Lemma 4.3. — (a) There is a closed algebraic one-form ω \omega ω on A A A such that
[ Re ω ] = λ , [\operatorname{Re}\omega]=\lambda, [ Re ω ] = λ , and ω \omega ω extends logarithmically to the standard projective compactification of A A A with nonzero purely imaginary residues at each boundary divisor.
(b) For every nonconstant algebraic map u : C → A u:C\to A u : C → A from a smooth algebraic curve, u ∗ ω ≢ 0 u^*\omega\not\equiv0 u ∗ ω ≡ 0 .
For section 3.3, there is no essential change, except that Φ A \Phi_A Φ A introduced there is no longer well-defined. Also, from now on we assume that dim ψ ( X ) ≥ 2 \dim\psi(X)\ge2 dim ψ ( X ) ≥ 2 . With the notation introduced above, let ψ ˉ : Y ˉ → A ˉ \bar{\psi}:\bar{Y}\to\bar{A} ψ ˉ : Y ˉ → A ˉ be the extension to smooth compactifications with SNC divisors and put P = ( ψ ˉ − 1 ( D A ) ) r e d P=(\bar{\psi}^{-1}(D_A))_{\mathrm{red}} P = ( ψ ˉ − 1 ( D A ) ) red . Write P = ⋃ ν P ν P=\bigcup_\nu P_\nu P = ⋃ ν P ν for its irreducible components and set ξ = ψ ˉ ∗ ω \xi=\bar{\psi}^*\omega ξ = ψ ˉ ∗ ω on Y ∘ = Y ˉ ∖ P Y^\circ=\bar{Y}\setminus P Y ∘ = Y ˉ ∖ P .
Near a point of P ν P_\nu P ν , choose local coordinates such that
P = { z 1 ⋯ z k = 0 } , k = 1 , 2. P=\{z_1\cdots z_k=0\},\qquad k=1,2. P = { z 1 ⋯ z k = 0 } , k = 1 , 2.
After choosing a local trivialization of the P ℓ \mathbb{P}^{\ell} P ℓ -bundle A ˉ → A a b \bar{A}\to A_{\mathrm{ab}} A ˉ → A ab , let t i = w i / w 0 t_i=w_i/w_0 t i = w i / w 0 , 1 ≤ i ≤ ℓ 1\le i\le\ell 1 ≤ i ≤ ℓ , denote the corresponding torus characters. Then locally
ψ ˉ ∗ t i = v i z 1 n i 1 ⋯ z k n i k , \bar{\psi}^*t_i=v_i z_1^{n_{i1}}\cdots z_k^{n_{ik}}, ψ ˉ ∗ t i = v i z 1 n i 1 ⋯ z k n ik ,
where v i v_i v i is a nonvanishing holomorphic function and n i j ∈ Z n_{ij}\in\mathbb{Z} n ij ∈ Z . Hence
ψ ˉ ∗ d t i t i = d v i v i + ∑ j = 1 k n i j d z j z j . \bar{\psi}^*\frac{dt_i}{t_i}=\frac{dv_i}{v_i}+\sum_{j=1}^{k}n_{ij}\frac{dz_j}{z_j}. ψ ˉ ∗ t i d t i = v i d v i + j = 1 ∑ k n ij z j d z j .
Absorbing all holomorphic terms into a locally defined holomorphic function g g g , we obtain
ξ = ψ ˉ ∗ ω = ∑ j = 1 k c j 2 π − 1 d z j z j + d g , (21) \xi=\bar{\psi}^*\omega=\sum_{j=1}^{k}\frac{c_j}{2\pi\sqrt{-1}}\frac{dz_j}{z_j}+dg, \tag*{(21)} ξ = ψ ˉ ∗ ω = j = 1 ∑ k 2 π − 1 c j z j d z j + d g , (21)
where c j = ∑ i = 1 ℓ a i n i j ∈ R c_j=\sum_{i=1}^{\ell}a_i n_{ij}\in\mathbb{R} c j = ∑ i = 1 ℓ a i n ij ∈ R . Writing z j = r j e − 1 θ j z_j=r_j e^{\sqrt{-1}\theta_j} z j = r j e − 1 θ j , we get
α : = Re ξ = ∑ j = 1 k c j 2 π d θ j + Re d g . (22) \alpha:=\operatorname{Re}\xi=\sum_{j=1}^{k}\frac{c_j}{2\pi}d\theta_j+\operatorname{Re}dg. \tag*{(22)} α := Re ξ = j = 1 ∑ k 2 π c j d θ j + Re d g . (22)
We claim that c j ≠ 0 c_j\ne0 c j = 0 for every 1 ≤ j ≤ k 1\le j\le k 1 ≤ j ≤ k . For a fixed j j j , the vector ( n 1 j , … , n ℓ j ) ∈ Z ℓ (n_{1j},\ldots,n_{\ell j})\in\mathbb{Z}^{\ell} ( n 1 j , … , n ℓ j ) ∈ Z ℓ is nonzero. It therefore determines a nonconstant one-parameter subgroup
ρ j : C ∗ → T = ( C ∗ ) ℓ , s ↦ ( s n 1 j , … , s n ℓ j ) . \rho_j:\mathbb{C}^*\to T=(\mathbb{C}^*)^\ell,\qquad s\mapsto(s^{n_{1j}},\ldots,s^{n_{\ell j}}). ρ j : C ∗ → T = ( C ∗ ) ℓ , s ↦ ( s n 1 j , … , s n ℓ j ) .
If γ i \gamma_i γ i denotes the i i i -th standard generator of π 1 ( T ) \pi_1(T) π 1 ( T ) , then ( ρ j ) ∗ ( 1 ) = ∑ i = 1 ℓ n i j γ i (\rho_j)_*(1)=\sum_{i=1}^{\ell}n_{ij}\gamma_i ( ρ j ) ∗ ( 1 ) = ∑ i = 1 ℓ n ij γ i , and hence λ ( ( ρ j ) ∗ ( 1 ) ) = ∑ i = 1 ℓ a i n i j = c j \lambda((\rho_j)_*(1))=\sum_{i=1}^{\ell}a_i n_{ij}=c_j λ (( ρ j ) ∗ ( 1 )) = ∑ i = 1 ℓ a i n ij = c j . If c j = 0 c_j=0 c j = 0 , then ρ j ∗ λ = 0 \rho_j^*\lambda=0 ρ j ∗ λ = 0 , and therefore λ \lambda λ vanishes identically on the positive-dimensional connected algebraic subgroup im ρ j ⊂ A \operatorname{im}\rho_j\subset A im ρ j ⊂ A . This contradicts Theorem 4.2. Thus c j ≠ 0 c_j\ne0 c j = 0 .
Lemma 4.4. — In the local form eq. (4.3) of ξ \xi ξ , we have c j ≠ 0 c_j\ne0 c j = 0 for each 1 ≤ j ≤ k 1\le j\le k 1 ≤ j ≤ k , where k = 1 k=1 k = 1 or 2 2 2 .
Now take the same real-oriented blowup, so that the angular coordinates remain well-defined when the radial coordinates vanish. Near a point where
P = { z 1 ⋯ z k = 0 } , z j = r j e − 1 θ j , k = 1 , 2 , P=\{z_1\cdots z_k=0\},\qquad z_j=r_j e^{\sqrt{-1}\theta_j},\qquad k=1,2, P = { z 1 ⋯ z k = 0 } , z j = r j e − 1 θ j , k = 1 , 2 ,
we have eq. (4.4) for α \alpha α . Consider the boundary corresponding to P j P_j P j , given by r j = 0 r_j=0 r j = 0 . Since g g g is holomorphic across P j P_j P j , its restriction to r j = 0 r_j=0 r j = 0 is independent of the angular variable θ j \theta_j θ j . Hence ∂ ∂ θ j Re g = 0 \frac{\partial}{\partial\theta_j}\operatorname{Re}g=0 ∂ θ j ∂ Re g = 0 along r j = 0 r_j=0 r j = 0 . Therefore
α ( ∂ ∂ θ j ) = c j 2 π ≠ 0. \alpha\left(\frac{\partial}{\partial\theta_j}\right)=\frac{c_j}{2\pi}\ne0. α ( ∂ θ j ∂ ) = 2 π c j = 0.
It follows that α \alpha α does not vanish on the boundary of the real-oriented blowup. Since ∂ M \partial M ∂ M is compact, for any fixed smooth Riemannian metric on M M M , there exists a constant ε 0 > 0 \varepsilon_0>0 ε 0 > 0 such that ∥ α x ∥ ≥ ε 0 \lVert\alpha_x\rVert\ge\varepsilon_0 ∥ α x ∥ ≥ ε 0 for every x ∈ ∂ M x\in\partial M x ∈ ∂ M . By continuity, after shrinking to a collar neighborhood N N N of ∂ M \partial M ∂ M , we may assume that
∥ α x ∥ ≥ ε 0 2 for every x ∈ N . (23) \lVert\alpha_x\rVert\ge\frac{\varepsilon_0}{2}\qquad\text{for every }x\in N. \tag*{(23)} ∥ α x ∥ ≥ 2 ε 0 for every x ∈ N . (23)
This is the uniform bound we will use to construct a flow later.
The argument in § 3.4 goes through verbatim in the general setting, since we are working downstairs and nothing related to the rationality of χ \chi χ is used there. In conclusion, we still have finitely many Type I and Type II critical neighborhoods, with the mixed case classified as Type I.
The flow construction in the general case
4.2. The flow construction in the general case. — This is the part where we need a major revision of § 3.5 .
Since α \alpha α is closed on M M M , it defines
χ M : π 1 ( M ) → R , χ M ( [ γ ] ) = ∫ γ α . \chi_M:\pi_1(M)\to\mathbb{R},\qquad\chi_M([\gamma])=\int_\gamma\alpha. χ M : π 1 ( M ) → R , χ M ([ γ ]) = ∫ γ α .
Under the identification π 1 ( M ) ≅ π 1 ( Y ∘ ) \pi_1(M)\cong\pi_1(Y^\circ) π 1 ( M ) ≅ π 1 ( Y ∘ ) , the inclusion j : Y ↪ Y ∘ j:Y\hookrightarrow Y^\circ j : Y ↪ Y ∘ induces a surjection on π 1 \pi_1 π 1 , whose kernel is normally generated by small positively-oriented loops around Q Q Q . Since α \alpha α extends across Q Q Q , these loops have zero α \alpha α -period, and hence χ = χ M ∘ j ∗ \chi=\chi_M\circ j_* χ = χ M ∘ j ∗ and im χ M = im χ ≅ Z r \operatorname{im}\chi_M=\operatorname{im}\chi\cong\mathbb{Z}^r im χ M = im χ ≅ Z r . Let p : M ^ → M p:\widehat{M}\to M p : M → M be the covering corresponding to ker χ M \ker\chi_M ker χ M . Then p ∗ α = d u p^*\alpha=du p ∗ α = d u for some u : M ^ → R u:\widehat{M}\to\mathbb{R} u : M → R , and Y ^ : = p − 1 ( Y ) \widehat{Y}:=p^{-1}(Y) Y := p − 1 ( Y ) is precisely the covering of Y Y Y corresponding to ker χ \ker\chi ker χ .
We first consider the lifts of the two types of critical neighborhoods. For Type I, it can be argued as before that
p − 1 ( W i ) = ∐ γ ∈ Z r W ^ i , γ = ∐ γ ∈ Z r γ W ^ i , 0 . p^{-1}(W_i)=\coprod_{\gamma\in\mathbb{Z}^r}\widehat{W}_{i,\gamma}=\coprod_{\gamma\in\mathbb{Z}^r}\gamma\widehat{W}_{i,0}. p − 1 ( W i ) = γ ∈ Z r ∐ W i , γ = γ ∈ Z r ∐ γ W i , 0 .
Indeed, on W i W_i W i we have a local primitive f i f_i f i such that ξ = d f i \xi=df_i ξ = d f i , and hence α = d Re f i \alpha=d\operatorname{Re}f_i α = d Re f i . Thus, for any loop γ \gamma γ in W i W_i W i , we have
χ ( [ γ ] ) = ∫ γ α = ∫ γ d Re f i = 0 , \chi([\gamma])=\int_\gamma\alpha=\int_\gamma d\operatorname{Re}f_i=0, χ ([ γ ]) = ∫ γ α = ∫ γ d Re f i = 0 ,
i.e., π 1 ( W i ) ⊂ ker ( χ ) \pi_1(W_i)\subset\ker(\chi) π 1 ( W i ) ⊂ ker ( χ ) , which gives the conclusion above. For a Type II domain, the situation is the same, and we have
p − 1 ( W l ) = ∐ γ ∈ Z r W ^ l , γ = ∐ γ ∈ Z r γ W ^ l , 0 . p^{-1}(\mathcal{W}_l)=\coprod_{\gamma\in\mathbb{Z}^r}\widehat{\mathcal{W}}_{l,\gamma}=\coprod_{\gamma\in\mathbb{Z}^r}\gamma\widehat{\mathcal{W}}_{l,0}. p − 1 ( W l ) = γ ∈ Z r ∐ W l , γ = γ ∈ Z r ∐ γ W l , 0 .
The mixed case is still classified as Type I and has the same lifting property.
Now consider the vector field v \mathbf{v} v with properties as in § 3.5 . Precisely, take
W i ′ ′ ′ ⋐ W i ′ ′ ⋐ W i ∩ f i − 1 ( Δ ( 0 ; c ) ‾ ) ⋐ W i ′ ⋐ W i W_i^{\prime\prime\prime}\Subset W_i^{\prime\prime}\Subset W_i\cap f_i^{-1}\left(\overline{\Delta(0;c)}\right)\Subset W_i'\Subset W_i W i ′′′ ⋐ W i ′′ ⋐ W i ∩ f i − 1 ( Δ ( 0 ; c ) ) ⋐ W i ′ ⋐ W i
as shown there, and similarly
W l ′ ′ ′ ⋐ W l ′ ′ ⋐ W l ∩ f l − 1 ( Δ ( 0 ; c ) ‾ ) ⋐ W l ′ ⋐ W l . \mathcal{W}_l^{\prime\prime\prime}\Subset\mathcal{W}_l^{\prime\prime}\Subset\mathcal{W}_l\cap f_l^{-1}\left(\overline{\Delta(0;c)}\right)\Subset\mathcal{W}_l'\Subset\mathcal{W}_l. W l ′′′ ⋐ W l ′′ ⋐ W l ∩ f l − 1 ( Δ ( 0 ; c ) ) ⋐ W l ′ ⋐ W l .
Choose a smooth vector field v \mathbf{v} v on M M M such that
(a)∗ ^* ∗ v \mathbf{v} v vanishes in the neighborhood W i ′ ′ ′ W_i^{\prime\prime\prime} W i ′′′ for each i i i and W l ′ ′ ′ \mathcal{W}_l^{\prime\prime\prime} W l ′′′ for each l l l .
(b)∗ ^* ∗ Outside Type I critical neighborhoods ⋃ i = 1 M W i ′ ′ \bigcup_{i=1}^{M}W_i^{\prime\prime} ⋃ i = 1 M W i ′′ and Type II critical neighborhoods ⋃ l = 1 N W l ′ ′ \bigcup_{l=1}^{N}\mathcal{W}_l^{\prime\prime} ⋃ l = 1 N W l ′′ , we have
α ( v ) = 1. \alpha(\mathbf{v})=1. α ( v ) = 1.
(c)∗ ^* ∗ v \mathbf{v} v is tangent to every boundary face P P P of the real-oriented blowup M M M and satisfies α ( v ) = 1 \alpha(\mathbf{v})=1 α ( v ) = 1 . Such a choice is possible by the uniform lower bound in eq. (23) ; in fact, locally near a smooth point of P j = { r j = 0 } P_j=\{r_j=0\} P j = { r j = 0 } , we can simply take v = 2 π c j ∂ ∂ θ j \mathbf{v}=\frac{2\pi}{c_j}\frac{\partial}{\partial\theta_j} v = c j 2 π ∂ θ j ∂ . Near a corner, the corresponding choices can be interpolated within the tangent bundle of the boundary while preserving α ( v ) = 1 \alpha(\mathbf{v})=1 α ( v ) = 1 .
(d)∗ ^* ∗ On each boundary T i T_i T i of a Type I critical neighborhood W i W_i W i , v \mathbf{v} v is tangent to T i T_i T i and satisfies α ( v ) = 1 \alpha(\mathbf{v})=1 α ( v ) = 1 .
(e)∗ ^* ∗ On each boundary T l \mathcal{T}_l T l of a Type II critical neighborhood W l \mathcal{W}_l W l , v \mathbf{v} v is tangent to T l \mathcal{T}_l T l and satisfies α ( v ) = 1 \alpha(\mathbf{v})=1 α ( v ) = 1 .
(f)∗ ^* ∗ On Q ′ ′ Q^{\prime\prime} Q ′′ away from Type I and Type II critical neighborhoods, v \mathbf{v} v is tangent to Q ′ ′ Q^{\prime\prime} Q ′′ with α ( v ) = 1 \alpha(\mathbf{v})=1 α ( v ) = 1 . (g)* On the annular regions W i ∩ f i − 1 ( Δ ( 0 ; 5 c ) ) ∖ W i ′ ′ W_i \cap f_i^{-1}(\Delta(0;5c)) \setminus W_i^{\prime\prime} W i ∩ f i − 1 ( Δ ( 0 ; 5 c )) ∖ W i ′′ and W i ∩ f i − 1 ( Δ ( 0 ; 5 c ) ) ∖ W i ′ ′ \mathbb{W}_i \cap f_i^{-1}(\Delta(0;5c)) \setminus\mathbb{W}_i^{\prime\prime} W i ∩ f i − 1 ( Δ ( 0 ; 5 c )) ∖ W i ′′ , the vector field may be chosen so that
d Re f i ( v ) = 1 , d Im f i ( v ) = 0. d\operatorname{Re} f_i(\mathbf{v})=1,\ d\operatorname{Im} f_i(\mathbf{v})=0. d Re f i ( v ) = 1 , d Im f i ( v ) = 0.
Let v ^ \widehat{\mathbf{v}} v be the lift of v \mathbf{v} v to M ^ \widehat{M} M . Since M M M is compact and v \mathbf{v} v is tangent to ∂ M \partial M ∂ M , the vector field v \mathbf{v} v is complete, and hence so is v ^ \widehat{\mathbf{v}} v . Denote by Θ ^ s \widehat{\Theta}_s Θ s the flow generated by v ^ \widehat{\mathbf{v}} v . For t ∈ R t \in\mathbb{R} t ∈ R , set
L t : = { x ∈ Y ^ : u ( x ) = t } . L_t:=\{x\in\widehat{Y}:u(x)=t\}. L t := { x ∈ Y : u ( x ) = t } .
We now use the second method sketched in section 3.6. For c > 0 c>0 c > 0 , define the continuous function on M ^ \widehat{M} M :
σ c ( x ) = { min { c , t − u ( x ) } , if u ( x ) < t , 0 , if u ( x ) = t , − min { c , u ( x ) − t } , if u ( x ) > t . \sigma_c(x)=
\begin{cases}
\min\{c,t-u(x)\}, & \text{if }u(x)<t,\\
0, & \text{if }u(x)=t,\\
-\min\{c,u(x)-t\}, & \text{if }u(x)>t.
\end{cases} σ c ( x ) = ⎩ ⎨ ⎧ min { c , t − u ( x )} , 0 , − min { c , u ( x ) − t } , if u ( x ) < t , if u ( x ) = t , if u ( x ) > t .
For a path γ : [ 0 , 1 ] → Y ^ \gamma:[0,1]\to\widehat{Y} γ : [ 0 , 1 ] → Y , define
H ( s , a ) : = Θ ^ σ c ( γ ( a ) ) ( γ ( a ) ) , 0 ≤ s , a ≤ 1. H(s,a):=\widehat{\Theta}_{\sigma_c(\gamma(a))}(\gamma(a)),\qquad0\leq s,a\leq1. H ( s , a ) := Θ σ c ( γ ( a )) ( γ ( a )) , 0 ≤ s , a ≤ 1.
Since σ c \sigma_c σ c is continuous and v ^ \widehat{\mathbf{v}} v is smooth, H H H is a continuous homotopy. Moreover, if γ ( a ) ∈ L t \gamma(a)\in L_t γ ( a ) ∈ L t , then σ c ( γ ( a ) ) = 0 \sigma_c(\gamma(a))=0 σ c ( γ ( a )) = 0 , and hence H ( s , a ) = γ ( a ) H(s,a)=\gamma(a) H ( s , a ) = γ ( a ) for all s ∈ [ 0 , 1 ] s\in[0,1] s ∈ [ 0 , 1 ] . Thus the points of γ \gamma γ already lying in L t L_t L t remain fixed throughout the deformation.
Denote the compact subset
K γ = { Θ ^ s σ c ( γ ( a ) ) ( γ ( a ) ) : 0 ≤ s ≤ 1 , 0 ≤ a ≤ 1 } . K_\gamma=\{\widehat{\Theta}_{s\sigma_c(\gamma(a))}(\gamma(a)):0\leq s\leq1,\ 0\leq a\leq1\}. K γ = { Θ s σ c ( γ ( a )) ( γ ( a )) : 0 ≤ s ≤ 1 , 0 ≤ a ≤ 1 } .
Although the inverse images of the Type I and Type II critical neighborhoods may have infinitely many connected components in M ^ \widehat{M} M , these components form a locally finite collection, since the deck group Z r \mathbb{Z}^r Z r acts properly discontinuously on M ^ \widehat{M} M . Consequently, the compact set K γ K_\gamma K γ meets only finitely many of them, relabeled as
W ^ 1 , … , W ^ n I ; W ^ 1 , … , W ^ n I I . \widehat{W}_1,\ldots,\widehat{W}_{n_I};\widehat{\mathbb{W}}_1,\ldots,\widehat{\mathbb{W}}_{n_{II}}. W 1 , … , W n I ; W 1 , … , W n I I .
Similarly, the inverse images of the polar boundary P P P and the nonpolar boundary Q Q Q may have infinitely many lifts. In the irrational case, these faces may moreover occur at arbitrarily close u u u -levels, since the subgroup χ ( G ) ⊂ R \chi(G)\subset\mathbb{R} χ ( G ) ⊂ R need not be discrete. Nevertheless, the lifted boundaries form a locally finite family by the proper discontinuity of the deck action. Consequently, the compact subset K γ K_\gamma K γ meets only finitely many of them.
Away from these critical neighborhoods, we have from Theorem 3.23 that
d u ( v ^ ) = p ∗ α ( v ^ ) = α ( v ) = 1. du(\widehat{\mathbf{v}})=p^*\alpha(\widehat{\mathbf{v}})=\alpha(\mathbf{v})=1. d u ( v ) = p ∗ α ( v ) = α ( v ) = 1.
Hence, as long as the corresponding flow segment remains outside the critical neighborhoods W ′ ∪ W ′ W^\prime\cup\mathbb{W}^\prime W ′ ∪ W ′ ,
u ( Θ ^ s ( x ) ) = u ( x ) + s . u(\widehat{\Theta}_s(x))=u(x)+s. u ( Θ s ( x )) = u ( x ) + s .
The flow moves points toward the level L t L_t L t ; more precisely, outside of the critical neighborhoods, we have
min { r 0 + c , t } ≤ u ( γ c ( a ) ) ≤ max { t , R 0 − c } , for all 0 ≤ a ≤ 1. (24) \min\{r_0+c,t\}\leq u(\gamma_c(a))\leq\max\{t,R_0-c\},\qquad\text{for all }0\leq a\leq1. \tag*{(24)} min { r 0 + c , t } ≤ u ( γ c ( a )) ≤ max { t , R 0 − c } , for all 0 ≤ a ≤ 1. (24)
where
r 0 = min a ∈ [ 0 , 1 ] u ( γ ( a ) ) ≤ t , R 0 = max a ∈ [ 0 , 1 ] u ( γ ( a ) ) ≥ t . r_0=\min_{a\in[0,1]}u(\gamma(a))\leq t,\qquad R_0=\max_{a\in[0,1]}u(\gamma(a))\geq t. r 0 = a ∈ [ 0 , 1 ] min u ( γ ( a )) ≤ t , R 0 = a ∈ [ 0 , 1 ] max u ( γ ( a )) ≥ t .
Whenever the deformation encounters one of the finitely many lifted Type I or Type II critical neighborhoods, we use Theorem 3.21 to replace the corresponding portion of the path by one satisfying simultaneously eq. (24) . If both endpoints lie in L t L_t L t , corresponding to the real level { Re f i = η } \{\operatorname{Re} f_i=\eta\} { Re f i = η } , we use Theorems 3.15, 3.17 and 3.19 when η ≠ 0 \eta\ne0 η = 0 , and Theorems 3.24 to 3.26 when η = 0 \eta=0 η = 0 . These replacements are performed in the same homotopy class relative to the endpoints. Moreover, the flow preserves Y ^ \widehat{Y} Y by ( c ) ∗ (c)^* ( c ) ∗ and ( f ) ∗ (f)^* ( f ) ∗ , and the local replacements take place in the corresponding neighborhoods with Q Q Q deleted, say W i ∗ , W i ∗ W_i^*,\mathbb{W}_i^* W i ∗ , W i ∗ and W i + W_i^+ W i + .
After applying the homotopy H H H and making the local replacements described above, denote the resulting path by γ c \gamma_c γ c . At each iteration, the corresponding compact subsets K γ , K γ c , K γ 2 c , ⋯ K_\gamma, K_{\gamma_c}, K_{\gamma_{2c}}, \cdots K γ , K γ c , K γ 2 c , ⋯ meets only finitely many lifted Type I and Type II critical neighborhoods. Whenever such a neighborhood is encountered, we make the same local replacement as above. The collection of critical neighborhoods involved may vary from one iteration to the next, but it is finite at every step. Moreover, by ( c ) ∗ (c)^* ( c ) ∗ and ( f ) ∗ (f)^* ( f ) ∗ , together with the construction of the local replacements, the resulting path remains entirely in Y ^ \widehat{Y} Y throughout the deformation.
Thus, unless a point has already reached L t L_t L t , its distance in the u u u -coordinate from t t t decreases by at least c c c at each step. Since c > 0 c > 0 c > 0 is fixed, after finitely many iterations we obtain a path homotopic to γ \gamma γ , relative to its endpoints, which lies entirely in L t L_t L t . This proves the pair ( Y ^ , L t ) (\widehat{Y}, L_t) ( Y , L t ) is 1-connected.
By the same argument as at the end of section 3.5, the 1-connectedness of ( Y ^ , L t ∩ Y ^ ) (\widehat{Y}, L_t \cap\widehat{Y}) ( Y , L t ∩ Y ) transfers to ( X ^ , L t ) (\widehat{X}, L_t) ( X , L t ) . Consequently, if dim ψ ( X ) ≥ 2 \dim\psi(X) \ge2 dim ψ ( X ) ≥ 2 , then alternative (ii) of Theorem 4.1 holds. If dim ψ ( X ) = 1 \dim\psi(X) = 1 dim ψ ( X ) = 1 , we argue as in the proof of Theorem 3.1 to obtain alternative (i).
Remark 4.5. — Notice that, when r ≥ 2 r \ge2 r ≥ 2 , the primitive function u : M ^ → R u : \widehat{M} \to\mathbb{R} u : M → R is not proper. In particular, unlike in the rational case, the level sets of u u u , even regular ones, need not have finite CW type or finitely generated fundamental group. We postpone the interpretation of this phenomenon in terms of the BNS invariant to section 5.3.
Applications on quasi-projective groups
Results on BNS invariants
The application of Simpson’s Lefschetz theorem to the BNS invariant of Kähler groups was given by Delzant in [12 ] . Using our generalization of the Lefschetz theorem to smooth quasi-projective varieties, we obtain an analogous description of the BNS invariant for quasi-projective groups. We refer the reader to [5 , 6 , 39 ] for the general results on BNS invariants used in this section. Throughout this section, we assume that H 1 ( G ; R ) ≠ 0 H^1(G;\mathbb{R}) \ne0 H 1 ( G ; R ) = 0 ; otherwise, the character sphere S ( G ) S(G) S ( G ) is empty and there is nothing to consider.
Let G G G be a finitely generated group and χ : G → R \chi: G \to\mathbb{R} χ : G → R be a nontrivial group homomorphism. Denote G χ = { g ∈ G : χ ( g ) ≥ 0 } G_\chi= \{g \in G: \chi(g) \ge0\} G χ = { g ∈ G : χ ( g ) ≥ 0 } . For any finite generating set G \mathcal{G} G , denote the associated Cayley graph of G G G by C : = Cayley ( G ; G ) C := \operatorname{Cayley}(G; \mathcal{G}) C := Cayley ( G ; G ) . We define C χ C_\chi C χ to be the full subgraph of C C C spanned by G χ G_\chi G χ . We consider whether C χ C_\chi C χ is connected and record those χ \chi χ for which this happens. Notice that this property does not depend on the generating set G \mathcal{G} G (see [34 Proposition 11.1] ) and is invariant under multiplication of χ \chi χ by a positive real constant. Denote by [ χ ] [\chi] [ χ ] the equivalence class of χ \chi χ under multiplication by positive real constants, and write S ( G ) = ( Hom ( G ; R ) ∖ { 0 } ) / R + S(G) = (\operatorname{Hom}(G;\mathbb{R}) \setminus\{0\})/\mathbb{R}^+ S ( G ) = ( Hom ( G ; R ) ∖ { 0 }) / R + for the collection of all such equivalence classes. For a subgroup H ≤ G H \le G H ≤ G , define
S ( G , H ) : = { [ χ ] ∈ S ( G ) : χ ∣ H = 0 } . S(G,H) := \{[\chi] \in S(G) : \left.\chi\right|_H = 0\}. S ( G , H ) := {[ χ ] ∈ S ( G ) : χ ∣ H = 0 } .
Definition 5.1. — With the notation above, for a finitely generated group G G G , we define its BNS invariant by
Σ 1 ( G ) = { [ χ ] ∈ S ( G ) : C χ is connected } . \Sigma^1(G) = \{[\chi] \in S(G) : C_\chi\text{ is connected}\}. Σ 1 ( G ) = {[ χ ] ∈ S ( G ) : C χ is connected } .
Note that Σ 1 ( G ) \Sigma^1(G) Σ 1 ( G ) is an open subset of S ( G ) S(G) S ( G ) ; see [5 Theorem A] . Moreover, we give a list of results we need on the BNS invariant:
Proposition 5.2. — Let G G G be a finitely generated group.
(a) ([5 Theorem H] ) Let π : G → Q \pi: G \to Q π : G → Q be a group epimorphism. Then
π ∗ ( ( Σ 1 ( Q ) ) c ) ⊂ ( Σ 1 ( G ) ) c . \pi^*\left(\left(\Sigma^1(Q)\right)^{\mathrm{c}}\right) \subset\left(\Sigma^1(G)\right)^{\mathrm{c}}. π ∗ ( ( Σ 1 ( Q ) ) c ) ⊂ ( Σ 1 ( G ) ) c . (b) ([5 Theorem B1] ) Let π : G → A \pi: G \to A π : G → A be a group epimorphism with A A A abelian. Then ker ( π ) \ker(\pi) ker ( π ) is finitely generated if and only if
S ( G , ker ( π ) ) ⊂ Σ 1 ( G ) . S(G,\ker(\pi)) \subset\Sigma^1(G). S ( G , ker ( π )) ⊂ Σ 1 ( G ) .
(c) If G G G is abelian, then
Σ 1 ( G ) = S ( G ) . \Sigma^1(G)=S(G). Σ 1 ( G ) = S ( G ) .
In [25 ] , building on a result of Suciu [40 ] , we described the BNS invariant of a Kähler group in terms of the integral tropicalization of its rank-one jump ideal. More precisely, if X X X is a compact Kähler manifold with G = π 1 ( X ) G=\pi_1(X) G = π 1 ( X ) , then
Σ 1 ( G ) c = S ( Trop Z ( J 1 ( G ; Z ) ) ) . \Sigma^1(G)^\mathrm{c}=\mathop{\mathrm{S}}\nolimits(\operatorname{Trop}_{\mathbb{Z}}(\mathcal{J}^1(G;\mathbb{Z}))). Σ 1 ( G ) c = S ( Trop Z ( J 1 ( G ; Z ))) .
Here S S S denotes the projection from H 1 ( G ; R ) ∖ { 0 } H^1(G;\mathbb{R})\setminus\{0\} H 1 ( G ; R ) ∖ { 0 } onto the character sphere S ( G ) S(G) S ( G ) , and the tropicalization over Z \mathbb{Z} Z is defined in [25 ] , Proposition 3.9 using the trivial, mod-p p p , and p p p -adic valuations. The analogous description in terms of orbifold fibrations also holds for quasi-projective groups, by the following result.
Theorem 5.3. — Let X X X be a connected smooth complex quasi-projective variety with G = π 1 ( X ) G=\pi_1(X) G = π 1 ( X ) . Then
Σ 1 ( G ) c = S ( ⋃ f im ( f ∗ : H 1 ( C , m ; R ) → H 1 ( X ; R ) ) ) . \Sigma^1(G)^\mathrm{c}
=
S\left(\bigcup_f \operatorname{im}\left(f^*:H^1(C,\mathbf{m};\mathbb{R})\to H^1(X;\mathbb{R})\right)\right). Σ 1 ( G ) c = S f ⋃ im ( f ∗ : H 1 ( C , m ; R ) → H 1 ( X ; R ) ) . where the union runs over all hyperbolic orbifold fibrations f : X → ( C , m ) f:X\to(C,\mathbf{m}) f : X → ( C , m ) . Moreover, this is a finite union.
Proof. — Set T T T to be the right hand side of above equation, and we prove the two inclusions.
First, let f : X → ( C , m ) f:X\to(C,\mathbf{m}) f : X → ( C , m ) be a hyperbolic orbifold fibration and put Q = π 1 o r b ( C , m ) Q=\pi_1^{\mathrm{orb}}(C,\mathbf{m}) Q = π 1 orb ( C , m ) . The induced homomorphism f ∗ : G ↠ Q f_*:G\twoheadrightarrow Q f ∗ : G ↠ Q is surjective. For a hyperbolic Riemann orbifold, we have Σ 1 ( Q ) c = S ( Q ) \Sigma^1(Q)^\mathrm{c}=S(Q) Σ 1 ( Q ) c = S ( Q ) by Theorem 2.3. By Theorem 5.2(a), we obtain
f ∗ S ( Q ) = f ∗ Σ 1 ( Q ) c ⊆ Σ 1 ( G ) c . f^*S(Q)=f^*\Sigma^1(Q)^\mathrm{c}\subseteq\Sigma^1(G)^\mathrm{c}. f ∗ S ( Q ) = f ∗ Σ 1 ( Q ) c ⊆ Σ 1 ( G ) c .
Taking the union over all hyperbolic orbifold fibrations gives T ⊆ Σ 1 ( G ) c T\subseteq\Sigma^1(G)^\mathrm{c} T ⊆ Σ 1 ( G ) c ; this union is finite due to Theorem 2.2 .
Conversely, let [ χ ] ∈ Σ 1 ( G ) c [\chi]\in\Sigma^1(G)^\mathrm{c} [ χ ] ∈ Σ 1 ( G ) c . Suppose that dim ψ ( X ) ≥ 2 \dim\psi(X)\ge2 dim ψ ( X ) ≥ 2 , and take the logarithmic representative α = Re ω \alpha=\operatorname{Re}\omega α = Re ω constructed in section 4. Let K = Bl P ∪ Q r o ( X ‾ ) K=\operatorname{Bl}^{\mathrm{ro}}_{P\cup Q}(\overline{X}) K = Bl P ∪ Q ro ( X ) be the full real-oriented blowup of an SNC compactification of X X X . Since the logarithmic form ω \omega ω extends holomorphically across Q Q Q and has purely imaginary residues along P P P , its real part α \alpha α extends smoothly to K K K . Moreover, K K K is a compact manifold with corners whose interior is X X X . Since the inclusion X ↪ K X\hookrightarrow K X ↪ K is a homotopy equivalence, we have π 1 ( K ) ≅ G \pi_1(K)\cong G π 1 ( K ) ≅ G .
By our construction, every level set of primitive function u u u in the χ \chi χ -cover of X X X is connected and has π 1 \pi_1 π 1 surjects onto ker χ \ker\chi ker χ . Its inverse image in the maximal abelian cover of K K K is therefore connected and has π 1 \pi_1 π 1 surjects onto the fundamental group of that cover. Consequently, the inclusion of the (unique) unbounded superlevel component of u u u on the maximal abelian cover of K K K also induces a surjection on fundamental groups. Applying [5 ] , Theorem G and §5 on the compact model K K K gives [ χ ] ∈ Σ 1 ( G ) [\chi]\in\Sigma^1(G) [ χ ] ∈ Σ 1 ( G ) , a contradiction. Hence dim ψ ( X ) = 1 \dim\psi(X)=1 dim ψ ( X ) = 1 , and the Stein factorization yields an algebraic orbifold fibration in (i).
We claim that ( C , m ) (C,\mathbf{m}) ( C , m ) is hyperbolic. Indeed, the same argument as in the proof of Theorem 3.3 applies. If ( C , m ) (C,\mathbf{m}) ( C , m ) were not hyperbolic, then, since χ C ≠ 0 \chi C\ne0 χ C = 0 , the only possibilities are 2 g + r = 2 2g+r=2 2 g + r = 2 . Moreover, by (2.2), there are no multiple fibers and Q Q Q is free abelian. By Theorem 5.2(b)–(c), ker ( χ C ) \ker(\chi_C) ker ( χ C ) is finitely generated. Since the kernel of the orbifold fibration is also finitely generated, the exact sequence
1 ⟶ N ⟶ ker ( χ ) ⟶ ker ( χ C ) ⟶ 1 1\longrightarrow N\longrightarrow\ker(\chi)\longrightarrow\ker(\chi_C)\longrightarrow1 1 ⟶ N ⟶ ker ( χ ) ⟶ ker ( χ C ) ⟶ 1
would imply that ker ( χ ) \ker(\chi) ker ( χ ) is finitely generated. By Theorem 5.2(b), this gives S ( G , ker χ ) ⊆ Σ 1 ( G ) S(G,\ker\chi)\subseteq\Sigma^1(G) S ( G , ker χ ) ⊆ Σ 1 ( G ) , contradicting [ χ ] ∈ Σ 1 ( G ) c [\chi]\in\Sigma^1(G)^\mathrm{c} [ χ ] ∈ Σ 1 ( G ) c . Therefore ( C , m ) (C,\mathbf{m}) ( C , m ) is hyperbolic. Hence Σ 1 ( G ) c ⊆ T \Sigma^1(G)^\mathrm{c}\subseteq T Σ 1 ( G ) c ⊆ T , which proves the desired equality. □ \square □
Corollary 5.4. — The BNS invariant and its complement for the fundamental group of a connected smooth complex quasi-projective variety are both antipodally symmetric.
Application on quasi-projective groups
Recall that a quasi-projective group is a group that can be realized as the fundamental group of a smooth complex quasi-projective variety. Not every finitely presented group is quasi-projective; see, for example, [28 , 20 ] for known obstructions.
Combining the antipodal symmetry in Theorem 5.4 with the argument of Theorem 11 of [23 ] , we obtain the following obstruction to a group being quasi-projective.
Theorem 5.5. Any group admitting a properly ascending or descending HNN decomposition with finitely generated base group is not a quasi-projective group.
Indeed, the canonical character of such an HNN extension is integral, and Theorem 3.1 alone is sufficient to prove this. Hence, the proof of Theorem 11 of [23 ] applies verbatim, and we omit the details. The corresponding result for Kähler groups is due to Napier–Ramachandran [33 ] . Their proof, however, does not directly extend to the quasi-projective setting.
Next, we consider Thompson–Stein groups. Let n 1 , … , n k ≥ 2 n_1,\ldots,n_k \ge2 n 1 , … , n k ≥ 2 and
ℓ ∈ Z [ 1 n 1 ⋯ n k ] , ℓ > 0. \ell\in\mathbb{Z}\left[\frac{1}{n_1\cdots n_k}\right], \qquad\ell> 0. ℓ ∈ Z [ n 1 ⋯ n k 1 ] , ℓ > 0.
We denote by ⟨ n 1 , … , n k ⟩ = { n 1 a 1 ⋯ n k a k : a 1 , … , a k ∈ Z } \langle n_1,\ldots,n_k\rangle= \{n_1^{a_1}\cdots n_k^{a_k}: a_1,\ldots,a_k \in\mathbb{Z}\} ⟨ n 1 , … , n k ⟩ = { n 1 a 1 ⋯ n k a k : a 1 , … , a k ∈ Z } the multiplicative subgroup of R > 0 \mathbb{R}_{>0} R > 0 generated by n 1 , … , n k n_1,\ldots,n_k n 1 , … , n k . The Thompson–Stein group
F : = F ( ℓ , Z [ 1 n 1 ⋯ n k ] , ⟨ n 1 , … , n k ⟩ ) F := F\left(\ell,\mathbb{Z}\left[\frac{1}{n_1\cdots n_k}\right],\langle n_1,\ldots,n_k\rangle\right) F := F ( ℓ , Z [ n 1 ⋯ n k 1 ] , ⟨ n 1 , … , n k ⟩ )
is the group of all orientation-preserving piecewise-linear homeomorphisms f : [ 0 , ℓ ] → [ 0 , ℓ ] f : [0,\ell] \to[0,\ell] f : [ 0 , ℓ ] → [ 0 , ℓ ] having finitely many breakpoints, such that all breakpoints and their images belong to
Z [ 1 n 1 ⋯ n k ] , \mathbb{Z}\left[\frac{1}{n_1\cdots n_k}\right], Z [ n 1 ⋯ n k 1 ] ,
and every slope of f f f belongs to ⟨ n 1 , … , n k ⟩ \langle n_1,\ldots,n_k\rangle ⟨ n 1 , … , n k ⟩ . This group is known to be finitely presented. By [5 ] , Theorem 8.1 (which is true for any ℓ \ell ℓ as noticed by Fried–Vidussi in [18 Appendix] ), we have
Σ 1 ( F ) c = { [ log λ ] , [ log ρ ] } (25) \Sigma^1(F)^{\mathrm{c}} = \{[\log\lambda],[\log\rho]\} \tag*{(25)} Σ 1 ( F ) c = {[ log λ ] , [ log ρ ]} (25)
where λ \lambda λ and ρ \rho ρ are given by derivatives at the left and right end-points of [ 0 , ℓ ] [0,\ell] [ 0 , ℓ ] respectively:
λ : F → R + , λ ( f ) = d f d t ( 0 + ) \lambda: F \to\mathbb{R}^{+}, \qquad\lambda(f) = \frac{df}{dt}(0+) λ : F → R + , λ ( f ) = d t df ( 0 + )
ρ : F → R + , ρ ( f ) = d f d t ( ℓ − ) \rho: F \to\mathbb{R}^{+}, \qquad\rho(f) = \frac{df}{dt}(\ell-) ρ : F → R + , ρ ( f ) = d t df ( ℓ − )
Denote r = rank Z ⟨ n 1 , … , n k ⟩ ≤ k r = \operatorname{rank}_{\mathbb{Z}}\langle n_1,\ldots,n_k\rangle\le k r = rank Z ⟨ n 1 , … , n k ⟩ ≤ k . When r ≥ 2 r \ge2 r ≥ 2 , we can see that log ( λ ) \log(\lambda) log ( λ ) and log ( ρ ) \log(\rho) log ( ρ ) are both irrational. Indeed, for f ∈ F f \in F f ∈ F , write its initial uniquely as
f ′ ( 0 + ) = n 1 a 1 ⋯ n k a k f'(0+) = n_1^{a_1}\cdots n_k^{a_k} f ′ ( 0 + ) = n 1 a 1 ⋯ n k a k
then define characters χ i 0 : F → Z \chi_i^0 : F \to\mathbb{Z} χ i 0 : F → Z defined as f ↦ a i f \mapsto a_i f ↦ a i , 1 ≤ i ≤ k 1 \le i \le k 1 ≤ i ≤ k . Then
log ( λ ) = ∑ i = 1 k χ i 0 ⋅ log ( n i ) \log(\lambda) = \sum_{i=1}^{k} \chi_i^0 \cdot\log(n_i) log ( λ ) = i = 1 ∑ k χ i 0 ⋅ log ( n i )
If log λ \log\lambda log λ is rational, choose element f i ∈ F f_i \in F f i ∈ F such that the initial slope is n i n_i n i , then we need log ( n i ) / log ( n j ) ∈ Q \log(n_i)/\log(n_j) \in\mathbb{Q} log ( n i ) / log ( n j ) ∈ Q for each $1 \le i \ne j \le k$, or equivalently, n i a = n j b n_i^a = n_j^b n i a = n j b for some integers a a a and b b b . If r ≥ 2 r \ge2 r ≥ 2 , then we know log λ \log\lambda log λ is irrational; similarly, log ρ \log\rho log ρ is also irrational. Moreover, these two points cannot be antipodal. Indeed, one can explicitly construct an element f ∈ F f \in F f ∈ F whose initial slope is $1$ while its terminal slope is nontrivial (see [5 §8] or [7 ] ). Hence log λ \log\lambda log λ and log ρ \log\rho log ρ cannot be proportional; in particular, these two cannot be antipodal.
Now, Theorem 5.4 yields the following:
Proposition 5.6. Any Thompson–Stein group is not quasi-projective.
For the case k = 1 k = 1 k = 1 , write
F m : = F ( ℓ , Z [ 1 m ] , ⟨ m ⟩ ) . F_m := F\left(\ell,\mathbb{Z}\left[\frac{1}{m}\right],\langle m\rangle\right). F m := F ( ℓ , Z [ m 1 ] , ⟨ m ⟩ ) .
When m = 2 m = 2 m = 2 , this recovers the ordinary Thompson group F F F , which was shown not to be Kähler by Napier–Ramachandran [32 ] . More generally, Friedl–Vidussi proved in [18 Theorem 3.3] proves that all Thompson–Stein groups are not Kähler, using rank-gradient arguments together with the BNS invariants. The same antipodal-symmetry argument also gives another proof of [18 Theorem 3.3] . We also remark that for the case k = 1 k = 1 k = 1 , the proof of Theorem 5.6 only requires the rational Lefschetz alternative, say Theorem 3.1 .
Lefschetz alternative revisited via the BNS invariant
With the above explanation via the BNS invariant, especially Theorem 5.4 , we are able to understand the relationship between Theorem 4.1 and Theorem 3.3 .
When χ : G → R \chi: G \to\mathbb{R} χ : G → R has image of rank 1 1 1 , up to rescaling, we can assume that χ : G ↠ Z \chi: G \twoheadrightarrow\mathbb{Z} χ : G ↠ Z . To determine whether ker χ \ker\chi ker χ is finitely generated, by Theorem 5.2(b), it is the same to determine whether
S ( G , ker χ ) ⊂ Σ 1 ( G ) . S(G,\ker\chi) \subset\Sigma^1(G). S ( G , ker χ ) ⊂ Σ 1 ( G ) .
But now S ( G , ker χ ) = { [ χ ] , [ − χ ] } S(G,\ker\chi) = \{[\chi],[-\chi]\} S ( G , ker χ ) = {[ χ ] , [ − χ ]} . Also, by Theorem 5.4 , these two points are either both contained in Σ 1 ( G ) \Sigma^1(G) Σ 1 ( G ) or both contained in Σ 1 ( G ) c \Sigma^1(G)^c Σ 1 ( G ) c . Hence, to determine whether ker χ \ker\chi ker χ is finitely generated, one only needs to check whether [ χ ] ∈ Σ 1 ( G ) [\chi] \in\Sigma^1(G) [ χ ] ∈ Σ 1 ( G ) . Now, with the complement as explained in Theorem 5.3 , we have the following dichotomy:
– ker χ \ker\chi ker χ is not finitely generated ⟺ [ χ ] ∈ Σ 1 ( G ) c ⟺ \Longleftrightarrow[\chi] \in\Sigma^1(G)^c \Longleftrightarrow ⟺ [ χ ] ∈ Σ 1 ( G ) c ⟺ condition (i) of Theorem 3.1 with C o r b C_{\mathrm{orb}} C orb hyperbolic.
– ker χ \ker\chi ker χ is finitely generated ⟺ [ χ ] ∈ Σ 1 ( G ) ⟺ \Longleftrightarrow[\chi] \in\Sigma^1(G) \Longleftrightarrow ⟺ [ χ ] ∈ Σ 1 ( G ) ⟺ condition (ii) of Theorem 3.1 for α \alpha α the real part of a suitable chosen logarithmic one-form as in section 3.2.
The above consideration also explains why we cannot expect analogues of Theorem 3.3 or Theorem 1.2 to hold for irrational characters. More precisely, we can find an irrational character χ : G → R \chi: G \to\mathbb{R} χ : G → R such that ker χ \ker\chi ker χ is not finitely generated, while χ \chi χ does not factor through any holomorphic fibration, even in the compact Kähler case.
The process of finding such an example is straightforward. We first choose a Kähler group G G G for which both Σ 1 ( G ) \Sigma^1(G) Σ 1 ( G ) and Σ 1 ( G ) c \Sigma^1(G)^c Σ 1 ( G ) c are nonempty. Since Σ 1 ( G ) \Sigma^1(G) Σ 1 ( G ) is open, we can choose a rational character φ \varphi φ such that [ φ ] ∈ Σ 1 ( G ) [\varphi] \in\Sigma^1(G) [ φ ] ∈ Σ 1 ( G ) . On the other hand, Σ 1 ( G ) c \Sigma^1(G)^c Σ 1 ( G ) c is a union of positive-dimensional rational subspaces arising from holomorphic fibrations; see [12 ] for the compact Kähler case, while Theorem 5.3 gives the corresponding statement in the more general smooth quasi-projective setting. Hence Σ 1 ( G ) c \Sigma^1(G)^c Σ 1 ( G ) c also contains a nonzero rational character, and we choose a rational character ψ \psi ψ such that [ ψ ] ∈ Σ 1 ( G ) c [\psi] \in\Sigma^1(G)^c [ ψ ] ∈ Σ 1 ( G ) c . Now consider
χ = φ + a ψ \chi= \varphi+ a\psi χ = φ + a ψ
with a a a an irrational number sufficiently small so that [ χ ] ∈ Σ 1 ( G ) [\chi] \in\Sigma^1(G) [ χ ] ∈ Σ 1 ( G ) , due to openness of Σ 1 ( G ) \Sigma^1(G) Σ 1 ( G ) . By our construction, the equality χ ( g ) = 0 \chi(g) = 0 χ ( g ) = 0 implies φ ( g ) = ψ ( g ) = 0 \varphi(g) = \psi(g) = 0 φ ( g ) = ψ ( g ) = 0 . Hence
ker ( χ ) = ker ( φ ) ∩ ker ( ψ ) ⊂ ker ( ψ ) . \ker(\chi) = \ker(\varphi) \cap\ker(\psi) \subset\ker(\psi). ker ( χ ) = ker ( φ ) ∩ ker ( ψ ) ⊂ ker ( ψ ) .
In particular,
[ ψ ] ∈ S ( G , ker χ ) ∩ Σ 1 ( G ) c , [\psi] \in S(G,\ker\chi) \cap\Sigma^1(G)^c, [ ψ ] ∈ S ( G , ker χ ) ∩ Σ 1 ( G ) c ,
and therefore ker χ \ker\chi ker χ is not finitely generated by Theorem 5.2(b). On the other hand, [ χ ] ∈ Σ 1 ( G ) [\chi] \in\Sigma^1(G) [ χ ] ∈ Σ 1 ( G ) , so by Theorem 5.3 , χ \chi χ cannot factor through any hyperbolic holomorphic fibration. It cannot factor through a nonhyperbolic one either, with the same argument as in the proof of Theorem 5.3 .
This example also reminds us that in the proof of Theorem 4.1 , the level set constructed there, say L t ⊂ Y ^ L_t \subset\widehat{Y} L t ⊂ Y , need not have the homotopy type of a finite CW complex. Indeed, the example above gives an irrational character χ \chi χ satisfying condition (ii) of Theorem 4.1 for which ker ( χ ) \ker(\chi) ker ( χ ) is not finitely generated. Since
π 1 ( L t ) ↠ π 1 ( Y ^ ) = ker ( χ ) , \pi_{1}(L_{t}) \twoheadrightarrow\pi_{1}(\widehat{Y}) = \ker(\chi), π 1 ( L t ) ↠ π 1 ( Y ) = ker ( χ ) ,
it follows that π 1 ( L t ) \pi_{1}(L_{t}) π 1 ( L t ) is not finitely generated. In particular, L t L_{t} L t cannot have the homotopy type of a finite CW complex.
Acknowledgments
The author would like to thank Chen Zhou for helpful discussions concerning his paper and for pointing out the example appearing in Theorem 3.10. The author would also like to thank Yongqiang Liu for many useful discussions; the collaboration with him provided an important motivation for writing this paper. The author is partially supported by the National Natural Science Foundation of China (Grant No. 12501105).
References