Complete nonsingular holomorphic foliations on Stein manifolds

We embed the Stein manifold X as a closed complex submanifold in a Euclidean space \mathbb{C}^N; see [Forstneric2017E, Theorems 2.4.1 and 9.3.1] for a survey of the classical embedding theorems. For every i\in\mathbb{N} the set X_i=X\cap i \overline \mathbb{B}^N is compact and \mathscr{O}(X)-convex, i.e., holomorphically convex in X. Pick a number c_i>0 such that

Here, |v| denotes the Euclidean length of a tangent vector v\in T_xX\subset T_x\mathbb{C}^N\cong \mathbb{C}^N. In every spherical shell S_i=\{z\in\mathbb{C}^N: i<|z|<i+1\} (i\in\mathbb{N}) we choose a labyrinth \Gamma_i satisfying Lemma 2.1 with a constant M_i>0, chosen such that

We explain the basic case with k=n-1. Using the Oka–Weil type approximation theorem for noncritical holomorphic functions (see [Forstneric2003AM, Theorem 2.1]) inductively, we find a holomorphic function f:X\to\mathbb{C} without critical points such that condition (?) holds for all i\in \mathbb{N} with \Gamma_i replaced by \Gamma'_i:=\Gamma_i\cap X. To be precise, choose constants C_i>0 (i\in \mathbb{N}) with \lim_{i\to\infty} C_i=+\infty. In the induction step, we are given a noncritical holomorphic function f_i on a neighbourhood of X_i in X. Next, we define f_i on a neighbourhood of \Gamma'_{i} in X to be an arbitrary noncritical holomorphic function satisfying \min_{\Gamma'_i}|f_{i}| >C_i. Since X_i\cup \Gamma'_i is a compact \mathscr{O}(X)-convex set, we can apply [Forstneric2003AM, Theorem 2.1] to find a noncritical holomorphic function f_{i+1} on a neighbourhood of X_{i+1} in X which approximates f_i on X_i\cup \Gamma'_i. Assuming that the approximation is close enough at every step, the sequence f_i converges uniformly on compacts in X to a noncritical holomorphic function f:X\to \mathbb{C} satisfying \min_{\Gamma'_i} |f| > C_i for every i\in\mathbb{N}. Pick a divergent path \gamma:[0,1)\to X contained in a level set of f and choose i_0\in\mathbb{N} such that \gamma(0)\in i_0\mathbb{B}^N. Note that \gamma also diverges in \mathbb{C}^N since X is a closed noncompact submanifold. We see as before that there is an integer i_1\ge i_0 such that \gamma avoids the labyrinth \Gamma'_i, and hence also the labyrinth \Gamma_i\subset\mathbb{C}^N for all i\ge i_1. From (Proof) and (Proof) it follows that

Hence, every nonempty level set f^{-1}(c) is a closed complete complex hypersurface in X, which is smooth since f has no critical points.

The same argument gives nonsingular \mathfrak{g}-complete holomorphic foliations of X of any dimension k\ge \left[\frac{n}{2}\right]. In this case, one inductively uses the Oka-Weil type approximation theorem for holomorphic submersions (see [Forstneric2003AM, Theorem 2.5]) in the above proof; we leave out the obvious details. The same holds for 1\le k< \left[\frac{n}{2}\right] if we assume the existence of a surjective complex vector bundle map TX\to X\times \mathbb{C}^q with q=n-k > \left[\frac{n+1}{2}\right].

Every irreducible component of a nonempty fibre f^{-1}(z) (z\in \mathbb{C}^q) is a closed complex subvariety of X of some dimension d\in \{1,\ldots,\dim X-1\}. In particular, the foliation does not have any zero dimensional leaves. Given a compact set K\subset X we clearly have that

Since f is nonconstant, the function f_j:X\to \mathbb{C} is nonconstant for some j, hence an open map. Thus, f_j(X) is an open subset of \mathbb{C} containing the compact subset f_j(K). Choose a nonempty open subset U\subset f_j(X)\setminus f_j(K). The set X\setminus f_j^{-1}(f_j(K)) then contains f_j^{-1}(U), hence is nonempty and not relatively compact. By (Proof) the same is true for X\setminus \mathcal{F}(K).

Let \mathbb{B} denote the open unit ball of \mathbb{C}^n. Set r_1=1, B_1=r_1\overline\mathbb{B}=\overline\mathbb{B}, \mathcal{F}_1=\mathcal{F}, and E_1=\mathcal{F}_1(B_1); see (1.2). Choose a sequence M_j>0 such that

Pick a closed ball B'_1\subset \mathbb{C}^n centred at 0 and containing B_1 in its interior. By Lemma 2.1 and Remark 2.2 there is a labyrinth \Gamma_1\subset \mathring B'_1\setminus B_1 which enlarges the \mathfrak{g}-distance by M_1>0 such that B_1\cup \Gamma_1 is polynomially convex. Fix a number 0<\epsilon_1<1. Lemma 3.1 furnishes an automorphism \phi_1\in\mathrm{Aut}(\mathbb{C}^n) such that

1. (a_1) \phi_1(E_1)\cap \Gamma_1=\varnothing, and

2. (b_1) |\phi_1(z)-z|<\epsilon_1 for all z\in B_1.

In the second step, we choose a number r_2>2 such that \phi_1(\overline\mathbb{B}) \subset (r_2-1) \mathbb{B} and set

Pick a slightly bigger ball B'_2\supset B_2 containing B_2 in its interior. By Lemma 2.1 there is a labyrinth \Gamma_2 \subset \mathring B'_2\setminus B_2 which enlarges the \mathfrak{g}-distance by M_2>0. Given \epsilon_2>0, Lemma 3.1 furnishes an automorphism \phi_2\in\mathrm{Aut}(\mathbb{C}^n) such that

1. (a_2) \phi_2(E_2)\cap \Gamma_2=\varnothing, and

2. (b_2) |\phi_2(z)-z|<\epsilon_2 for all z\in B_2.

Continuing inductively, we obtain an increasing sequence of numbers r_j>0, balls

labyrinths \Gamma_j\subset \mathring B_{j+1}\setminus B_j, constants \epsilon_j>0, automorphisms \phi_j\in\mathrm{Aut}(\mathbb{C}^n), and foliations \mathcal{F}_j of \mathbb{C}^n such that, setting

the following conditions hold for every for j=1,2,\ldots:

1. \phi_j(E_j)\cap \Gamma_j=\varnothing.

2. |\phi_j(z)-z|<\epsilon_j for z\in B_j.

3. 0<\epsilon_{j+1} < \frac12 \min\{\epsilon_{j}, \mathrm{dist}(\phi_j(E_j),\Gamma_j)\}. (Note that the set \phi_j(E_j) is closed and \Gamma_j is compact, so these sets are at positive distance by (ii_j).)

4. \Phi_j(j\overline \mathbb{B}) \cup B'_j \subset (r_{j+1}-1) \mathbb{B}.

5. The labyrinth \Gamma_j enlarges the \mathfrak{g}-distance by M_j and B_j\cup\Gamma_j is polynomially convex.

Assuming that we have obtained these quantities for indices \le j, the induction step goes as follows. Let \mathcal{F}_{j+1} and E_{j+1} be given by (?). Pick a number \epsilon_{j+1} satisfying (iii_j) and a number r_{j+1}>r_j satisfying (iv_j), and set B_{j+1}=r_{j+1}\overline\mathbb{B}. Choose a ball B'_{j+1}\supsetneq B_{j+1}. Lemma 2.1 gives a labyrinth \Gamma_{j+1}\subset \mathring B'_{j+1}\setminus B_{j+1} satisfying (v_{j+1}). Then, Lemma 3.1 gives an automorphism \phi_{j+1}\in\mathrm{Aut}(\mathbb{C}^n) satisfying (i_{j+1}) and (ii_{j+1}), closing the induction.

We claim that the sequence \Phi_j\in\mathrm{Aut}(\mathbb{C}^n) converges uniformly on compacts in \mathbb{C}^n to an automorphism \Phi\in \mathrm{Aut}(\mathbb{C}^n). Indeed, since

and \sum_{j=1}^\infty \epsilon_j<\infty (see (ii_j)–(iv_j)), the sequence \Phi_j converges uniformly on compacts in the domain \Omega=\bigcup_{j=1}^\infty \Phi_j^{-1}(B_j)\subset\mathbb{C}^n to a biholomorphic map \Phi:\Omega\to\mathbb{C}^n onto \mathbb{C}^n (see [Forstneric2017E, Corollary 4.4.2]). Furthermore, condition (iv_j) ensures that j\overline \mathbb{B}\subset \Omega for every j, so \Omega=\mathbb{C}^n and hence \Phi\in\mathrm{Aut}(\mathbb{C}^n). Hence, the sequence of foliations \mathcal{F}_{j+1}=(\Phi_j)_*\mathcal{F} converges uniformly on compacts in \mathbb{C}^n to a limit foliation \mathcal{G}=\Phi_*(\mathcal{F}).

It remains to show that the foliation \mathcal{G} is \mathfrak{g}-complete. We must show that for any divergent path \gamma:[0,1)\to \mathbb{C}^n contained in a leaf of \mathcal{F}, the path \tilde \gamma =\Phi\circ\gamma:[0,1)\to\mathbb{C}^n has infinite \mathfrak{g}-length. (Since the foliation \mathcal{F} is assumed to have closed leaves of positive dimension, every divergent path in a leaf of \mathcal{F} is also divergent in \mathbb{C}^n. Note that \tilde \gamma is a divergent path in \mathbb{C}^n contained in a leaf of \mathcal{G}.) For every k\in\mathbb{N} let \gamma_k=\Phi_k\circ \gamma:[0,1)\to \mathbb{C}^n; this is a divergent path contained in a leaf of the foliation \mathcal{F}_{k+1}=(\Phi_k)_*\mathcal{F}. Pick j\in\mathbb{N} such that \gamma(0) \in j\mathbb{B}. By (iv_j) we have that \gamma_j(0)=\Phi_j(\gamma(0)) \in (r_{j+1}-1) \mathbb{B}\subset B_{j+1}, and hence \gamma_j ([0,1))\subset E_{j+1} (see (?)). Conditions (i_j)–(iii_j) imply that for every k>j we have that \gamma_k(0)\in B_{j+1}=r_{j+1}\overline \mathbb{B} and the trace of \gamma_k avoids the labyrinths \Gamma_{i} for i=j+1,\ldots,k. Since \gamma_k is a divergent path in \mathbb{C}^n, its \mathfrak{g}-length is at least \sum_{i=j+1}^k M_i by (i_i). As k\to\infty, it follows that \gamma_k converges to a divergent path \tilde \gamma =\Phi\circ\gamma:[0,1)\to\mathbb{C}^n which has infinite \mathfrak{g}-length in view of (Proof).

The proof will follow that of Theorem 1.5 if we find labyrinths \Gamma in the given Stein manifold X which increase the length of divergent paths avoiding \Gamma by a given amount and whose pieces are holomorphically contractible in X. (Note that the labyrinths used in the proof of Theorem 1.1, which are obtained by intersecting labyrinths in \mathbb{C}^N whose pieces are balls in affine real hyperplanes with the embedded submanifold X\subset\mathbb{C}^N, need not satisfy this property.) To this end, we shall prove the following lemma whose first part generalizes Lemma 2.1, as well as [CharpentierKosinski2020, Lemma 2.4] by Charpentier and Kosiński, while the second part is an analogue of Lemma 3.1 adjusted to this situation. As before, \mathbb{B} denotes the unit ball of \mathbb{C}^N.

With Lemma ? in hand, we obtain Theorem 1.7 by following the proof of Theorem 1.5, and we leave the obvious details to the reader.

Let \mathcal{F}_0 be a proper holomorphic foliation on a Stein manifold X with the density property, and let \Omega be a pseudoconvex Runge domain in X. By using Lemma ? inductively, we find a normal exhaustion \Omega_1\Subset \Omega_2\Subset\cdots\subset \bigcup_{i=1}^\infty\Omega_i=\Omega by open relatively compact pseudoconvex Runge domains and for each i=1,2,\ldots a labyrinth \Gamma_i\subset \Omega_{i+1}\setminus \overline \Omega_i having the properties (a)–(c) in Lemma ? for a given constant M_i>0 chosen such that \sum_i M_i=+\infty. Furthermore, we can ensure that the compact set \Gamma_{i+1}\cup\overline \Omega_i is \mathscr{O}(\Omega)-convex (and hence also \mathscr{O}(X)-convex) for every i\in\mathbb{N}. Since the components of the labyrinth \Gamma=\bigcup_{i=1}^\infty \Gamma_i are holomorphically contractible, we can apply the proof of Theorem 1.5 to inductively twist the foliation \mathcal{F}_0 by a sequence of automorphisms \Phi_i\in\mathrm{Aut}(X) (i=1,2,\ldots), chosen so that they converge uniformly on compacts in \Omega, and the leaves of the foliations \mathcal{F}_i=(\Phi_i)_*\mathcal{F}_{i-1} avoid more and more components of \Gamma as i\to\infty. The limit holomorphic foliation \mathcal{F}=\lim_{i\to\infty}\mathcal{F}_i on \Omega is such that every leaf avoids all but finitely many labyrinths \Gamma_i, and hence it is \mathfrak{g}-complete. Furthermore, the construction ensures that every leaf is a pseudoconvex Runge domain in a leaf of the initial foliation \mathcal{F}_0 on X. The details can be found (for the case X=\mathbb{C}^n) in [AlarconGlobevnikLopez2019Crelle] and [AlarconForstneric2020MZ].

Theorem 1.1 furnishes a holomorphic submersion f_0:X\to \mathbb{C}^q with \mathfrak{g}-complete leaves, where q=N-k\le \left[\frac{n+1}{2}\right]. Clearly, f_0 extends to a holomorphic submersion U\to \mathbb{C}^q from an open neighbourhood U of X in \mathbb{C}^N. By [Forstneric2003AM, Theorem 2.5] there is a holomorphic submersion f:\mathbb{C}^N\to\mathbb{C}^q which agrees with f_0 to the second order along X. The foliation on \mathbb{C}^N determined by f then clearly satisfies Theorem 2.3. Furthermore, given a Riemannian metric \mathfrak{h} on \mathbb{C}^N, one can choose \mathcal{F} to be \mathfrak{h}-complete, transverse to X, and such that its trace on X is \mathfrak{g}-complete. We leave further details to the reader.

Pick a compact neighbourhood K_0 of B and compact convex neighbourhoods K_j of \Gamma_j for j=1,\ldots,m such that the sets K_0,\ldots,K_m are pairwise disjoint and K=\bigcup_{j=0}^m K_j is polynomially convex. (The existence of such sets follows from standard results on polynomial convexity; see Stout [Stout2007].) Let \Psi_0=\mathrm{Id}\in\mathrm{Aut}(\mathbb{C}^n) be the identity and set K'_0=K_0. For every j=1,\ldots,m we choose an automorphism \Psi_j\in \mathrm{Aut}(\mathbb{C}^n) such that the compact sets K'_j:=\Psi_j(K_j) are pairwise disjoint, we have that

and the union \bigcup_{j=0}^m K'_j is polynomially convex. Such \Psi_j are obtained by squeezing K_j (j=1,\ldots,m) by a dilation into a small neighbourhood of an interior point, then translating the images into small pairwise disjoint balls around some points in \mathbb{C}^n\setminus (E\cup K'_0) (this set is open and nonempty by the assumption on E), and applying [ForstnericRosay1993, Theorem 1.1] to approximate the final map by an automorphism of \mathbb{C}^n. By [ForstnericRosay1993, Theorem 2.3] (see also [Forstneric2017E, Corollary 4.12.4]), given \delta>0 there is an automorphism \Psi\in\mathrm{Aut}(\mathbb{C}^n) satisfying

Let \Theta=\Psi^{-1}. If \delta>0 is small enough then condition (b) holds, and we have that \Psi(\Gamma_j) \subset \mathring K'_j and hence \Gamma_j\subset \Theta(\mathring K'_j) for every j=1,\ldots,m, which by (Proof) also yields (a).