, , November 27, 2006

<ph f="cmbx">Proper discs in </ph> <math xmlns="http://www.w3.org/1998/Math/MathML"> <mi>q</mi> </math> <ph f="cmbx">-convex manifolds</ph>

Theorem 1.1. Let

X

be a

q

-complete complex manifold of dimension

n \geq 2

with

1 \leq q \leq n - 1

. Given

p \in X

there is a proper holomorphic map

f : △ \to X

such that

f (0) = p

Theorem 1.2. Let

X

be a

(n - 1)

-convex complex manifold of dimension

n \geq 2

and assume that

ρ : X \to R

is an exhaustion function which is

(n - 1)

-convex in

{ρ > M}

for some

M \in R

. Let

d

be a complete metric on

X

which induces the manifold topology.

Given

ɛ > 0

0 < r < 1

, and a continuous map

f : \bar{△} \to X

such that

ρ (f (ζ)) > M

(ζ \in b △)

there is a proper holomorphic map

g : △ \to X

such that

(i) $d (f (ζ), g (ζ)) < ɛ$ for $| ζ | < r$ ,
(ii) $g (0) = f (0)$ .

Lemma 2.1. Let

X

be a complex manifold of dimension

n \geq 2

endowed with a Riemannian metric, which induces the distance function

d

X

. Assume that

Ω \subset \subset X

is an open set, which is biholomorphic to an open convex set in

C n

, and

Ω_{0} \subset \subset Ω

. Let

f

be a holomorphic map from a neighborhood of

\bar{△}

X

. Let

K

be a compact subset of

X

and

r

0 < r < 1

, such that

K \cap Ω = \emptyset

and

f (\bar{△} \ r △) \cap K = \emptyset

. Assume that

H : b △ \times \bar{△} \to X

is a continuous map with the following properties

(i) for each $ζ \in b △$ the map $η \mapsto H (ζ, η)$ is holomorphic on $△$ ,
(ii) $H (ζ, 0) = f (ζ)$ $(ζ \in b △)$ ,
(iii) if for some $ζ \in b △$ we have $f (ζ) / \in Ω_{0}$ then $H (ζ, η) = f (ζ)$ $(η \in \bar{△})$ ,
(iv) if for some $ζ \in b △$ we have $f (ζ) \in Ω_{0}$ then $H (ζ, η) \in Ω$ $(η \in \bar{△})$ .

Given

ɛ > 0

there is a continuous map

g : \bar{△} \to X

, holomorphic on

△

, such that

(i') $d (H (ζ, b △), g (ζ)) < ɛ$ $(ζ \in b △)$ ,
(ii') $d (f (ζ), g (ζ)) < ɛ$ $(ζ \in r \bar{△})$ ,
(iii') $g (\bar{△} \ r △) \cap K = \emptyset$ ,
(iv') $f (0) = g (0)$ .

Proof. Choose a biholomorphic map

Φ

from

Ω

to a convex subset of

C n

and choose an open set

Ω_{1}

such that

Ω_{0} \subset \subset Ω_{1} \subset \subset Ω

Φ (Ω_{1})

is convex and

H (f^{- 1} ({\bar{Ω}}_{0}), \bar{△}) \subset Ω_{1}

. Denote by

B

the unit ball in

C n

There is

α > 0

so small that

\begin{matrix} Φ (H (f^{- 1} ({\bar{Ω}}_{0}), \bar{△})) + α B \subset Φ (Ω_{1}), and \end{matrix}

(1)

\begin{matrix} if z, z^{'} \in Φ (Ω_{1}), | z - z^{'} | < α, then d (Φ^{- 1} (z), Φ^{- 1} (z^{'})) < \frac{ɛ}{3} . \end{matrix}

(2)

By decreasing

ɛ > 0

if necessary we may assume that

\begin{matrix} d (K, f (\bar{△} \ r △)) > ɛ and d (K, Ω_{1}) > ɛ . \end{matrix}

(3)

By a slight perturbation of

{\bar{Ω}}_{0}

one can assume that the set

f^{- 1} ({\bar{Ω}}_{0}) \cap b △

is at most finite union of disjoint closed arcs. Denote these arcs by

{I_{j}}_{j \in J}

where

J

is finite and where

I_{j}

are pairwise disjoint. For each

I_{j}

one can find a smooth simple closed curve

Γ_{j} \subset \bar{△} \ r \bar{△}

such that

Γ_{j} \cap b △

is a neighborhood of

I_{j}

b △

and

Γ_{j}

are pairwise disjoint. Choose an open arc

J_{j}

such that

I_{j} \subset J_{j} \subset Γ_{j} \cap b △

. Each

Γ_{j}

bounds a domain

D_{j} \subset △ \ r \bar{△}

, which is conformally equivalent to the unit disc. Since

f (I_{j}) \subset Ω_{0}

one can choose

Γ_{j}

in such a way that, in addition to the above, we have

f ({\bar{D}}_{j}) \subset Ω_{1}

(j \in J)

Choose a homeomorphic map

h_{j}

from

\bar{△}

{\bar{D}}_{j}

, which is holomorphic on

△

, and let

V_{j} = \bar{△} \ {t h_{j}^{- 1} (ζ); ζ \in I_{j}, t \in [0, 1]}

. Let

χ_{j} : \bar{△} \to [0, 1]

be a smooth function such that

s u p p d χ_{j} \subset h_{j} (V_{j})

χ_{j} | I_{j} \equiv 1

and

χ_{j}

equals

0

in some neighborhood of

Γ_{j} \ J_{j}

\bar{△}

By assumption there is

ρ > 1

such that

f

is holomorphic on

ρ △

. Choose

ρ^{'}

1 < ρ^{'} < ρ

. Define the map

\tilde{f} : ρ △ \to X \times C

\tilde{f} (ζ) = (f (ζ), ζ)

Note that

\tilde{f}

is a holomorphic embedding (not proper), so there is an open neighborhood

{\tilde{Ω}}_{2}

\tilde{f} (ρ^{'} △)

X \times C

and a biholomorphic map from

{\tilde{Ω}}_{2}

onto a bounded open subset of

C^{n + 1}

(see [Roy] ,[LS] ,[Ro2,Lemma1.1] ). One can choose a compact neighborhood

{\tilde{K}}_{2}

\tilde{f} (\bar{△})

{\tilde{Ω}}_{2}

. Let

{\tilde{Ω}}_{1} = Ω \times 3 △

and

{\tilde{K}}_{1} = {\bar{Ω}}_{1} \times 2 \bar{△}

. By construction the sets

{\tilde{Ω}}_{1}

and

{\tilde{Ω}}_{2}

are biholomorphic to open sets in

C^{n + 1}

. For each

j \in J

one can choose a neighborhood

D_{j}^{'}

{\bar{D}}_{j}

ρ △

such that

D_{j}^{'}

are pairwise disjoint and

f (D_{j}^{'}) \subset Ω_{1}

. Denote by

U_{1}

the union

\cup_{j} D_{j}^{'}

and by

U_{2}

the set

ρ^{'} △ \ \bar{\cup_{j} D_{j}}

. Note that open sets

U_{1}

and

U_{2}

satisfy

\bar{△} \subset U_{1} \cup U_{2}

We endow the manifold

X \times C

by the Riemannian metric

〈 (t_{x}, t_{z}), (t_{x}^{'}, t_{z}^{'}) 〉 = 〈 t_{x}, t_{x}^{'} 〉 + 〈 t_{z}, t_{z}^{'} 〉 (t_{x}, t_{x}^{'} \in T_{x} X, t_{z}, t_{z}^{'} \in T_{z} C) .

This Riemannian metric induces a distance function

\tilde{d}

X \times C

such that

d (x, x^{'}) \leq \tilde{d} ((x, z), (x^{'}, z^{'}))

for

x, x^{'} \in X, z, z^{'} \in C

We call a map

\tilde{u} : △ \to X \times C

restricted if

\tilde{u} (U_{j} \cap △) \subset {\tilde{K}}_{j}

(j = 1, 2)

By [Ro1,Proposition1] there is

δ > 0

such that for any restricted map

\tilde{u} : △ \to X \times C

with

| \bar{\partial} \tilde{u} (ζ) | \leq δ

there is a holomorphic map

\tilde{v} : △ \to X \times C

such that

\tilde{d} (\tilde{u} (ζ), \tilde{v} (ζ)) < \frac{ɛ}{3}

(ζ \in △)

and

\tilde{u} (0) = \tilde{v} (0)

For each

j \in J

define the map

H_{j} : Γ_{j} \times \bar{△} \to X

H_{j} (ζ, η) = H (ζ, η)

(ζ \in I_{j}, η \in \bar{△})

and

H_{j} (ζ, η) = f (ζ)

(ζ \in Γ_{j} \ I_{j}, η \in \bar{△})

. Property (iii) implies that

H_{j}

is continuous, and by (i)

H_{j}

is holomorphic in the second variable. By ( 1 ) and as

f ({\bar{D}}_{j}) \subset Ω_{1}

there is an

α_{j}

0 < α_{j} < α

, so small that

\begin{matrix} Φ (H_{j} (Γ_{j}, \bar{△})) + α_{j} B \subset Φ (Ω_{1}) . \end{matrix}

(4)

Let

c_{j} = max {| \bar{\partial} χ_{j} (ζ) |; ζ \in \bar{△}} \cdot max {| | (Φ^{- 1})^{'} (z) | |; z \in Φ (Ω_{1})},

where

| | (Φ^{- 1})^{'} (z) | |

denotes the operator norm of the differential of

Φ^{- 1}

in the given pair of metrics. Using [Glo,Lemma5.1] for the map

G_{j} (ζ, η) = Φ (H_{j} (ζ, η)) - Φ (f (ζ))

((ζ, η) \in Γ_{j} \times \bar{△})

we get the map

P_{j}

{\bar{D}}_{j}

such that the map

φ_{j} : {\bar{D}}_{j} \to C n

, defined by

φ_{j} (ζ) = P_{j} (ζ) + Φ (f (ζ))

(ζ \in {\bar{D}}_{j})

, is continuous, holomorphic on

D_{j}

, and

\begin{matrix} φ_{j} (ζ) \in Φ (H_{j} (ζ, b △)) + α B (ζ \in Γ_{j}), \end{matrix}

(5)

\begin{matrix} φ_{j} (ζ) \in Φ (H_{j} (Γ_{j}, \bar{△})) + α_{j} B (ζ \in {\bar{D}}_{j}), \end{matrix}

(6)

\begin{matrix} | φ_{j} (ζ) - Φ (f (ζ)) | < \frac{δ}{c_{j}} (ζ \in h_{j} (V_{j})) . \end{matrix}

(7)

Note that by ( 4 ) and ( 6 ) we get

φ_{j} ({\bar{D}}_{j}) \subset Φ (Ω_{1})

. Since

Φ (Ω_{1})

is convex, we can glue these maps and the map

Φ \circ f

together by taking

u (ζ) = {\begin{matrix} Φ^{- 1} (χ_{j} (ζ) φ_{j} (ζ) + (1 - χ_{j} (ζ)) Φ (f (ζ))) & ζ \in {\bar{D}}_{j}, \\ f (ζ) & ζ \in \bar{△} \ \cup_{j} D_{j} . \end{matrix}

By the properties of

χ_{j}

and

D_{j}

the map

u

is a smooth map from the closed unit disc to

X

. Note that

u (D_{j}^{'} \cap △) \subset Ω_{1}

. For

ζ \in △ \ \cup_{j} D_{j}

the map

u

equals

f

and since

f

is holomorphic, it follows that

\bar{\partial} u (ζ) = 0

. For

ζ \in {\bar{D}}_{j} \cap h_{j} (V_{j})

we have

| \bar{\partial} u (ζ) | \leq c_{j} | φ_{j} (ζ) - Φ (f (ζ)) |,

which is less than

δ

by ( 7 ). Take

ζ \in {\bar{D}}_{j} \ h_{j} (V_{j})

. Since

d χ_{j} (ζ) = 0

and the maps

φ_{j}

and

f

are holomorphic, it follows that

\bar{\partial} u (ζ) = 0

. Therefore we have obtained

| \bar{\partial} u (ζ) | \leq δ

(ζ \in △)

. Let

\tilde{u} (ζ) = (u (ζ), ζ)

(ζ \in △)

. It is easy to see that

\tilde{u}

is a restricted map from

△

X \times C

. It follows by the above that

| \bar{\partial} \tilde{u} (ζ) | \leq δ

(ζ \in △)

. Therefore there is a holomorphic map

\tilde{g}

from

△

X \times C

such that

\tilde{d} (\tilde{u} (ζ), \tilde{g} (ζ)) < \frac{ɛ}{3}

(ζ \in △)

and

\tilde{u} (0) = \tilde{g} (0)

. Denote by

Π : X \times C \to X

the canonical projection to the first factor. As it was noted above we get

d (u (ζ), (Π \circ \tilde{g}) (ζ)) < \frac{ɛ}{3}

(ζ \in △)

Since

u

is uniformly continuous on

\bar{△}

we may choose

R

r < R < 1

, so close to

1

that

d (u (R ζ), u (ζ)) < \frac{ɛ}{3}

for each

ζ \in \bar{△}

. The map

g (ζ) = (Π \circ \tilde{g}) (R ζ)

(ζ \in \bar{△})

is continuous, holomorphic on

△

, and by the above we obtain

\begin{matrix} d (g (ζ), u (ζ)) \leq d ((Π \circ \tilde{g}) (R ζ), u (R ζ)) + d (u (R ζ), u (ζ)) < \frac{2 ɛ}{3} (ζ \in \bar{△}) . \end{matrix}

(8)

To prove (i') choose

ζ \in b △

. If

ζ \in I_{j}

then it follows by the properties of

χ_{j}

and by ( 5 ) and ( 2 ) that

d (u (ζ), H (ζ, b △)) = d (Φ^{- 1} (φ_{j} (ζ)), H_{j} (ζ, b △)) < \frac{ɛ}{3}

. If

ζ \in J_{j} \ I_{j}

then

H (ζ, η) = f (ζ)

(η \in \bar{△})

and by ( 5 ) it holds that

| χ_{j} (ζ) φ_{j} (ζ) + (1 - χ_{j} (ζ)) Φ (f (ζ)) - Φ (f (ζ)) | = | χ_{j} (ζ) (φ_{j} (ζ) - Φ (f (ζ))) | \leq α

and by ( 2 ) we obtain

d (u (ζ), H (ζ, b △)) < \frac{ɛ}{3}

. If

ζ \in b △ \ \cup_{j} J_{j}

then

d (u (ζ), H (ζ, b △)) = d (f (ζ), f (ζ)) = 0

. So we have

d (u (ζ), H (ζ, b △)) < \frac{ɛ}{3}

(ζ \in b △)

, which by ( 8 ) proves (i'). As

f

equals

u

outside the sets

D_{j}

, which all lie in

\bar{△} \ r △

, ( 8 ) implies (ii'). For

ζ \in \bar{△} \ (\cup_{j} D_{j} \cup r △)

we have

u (ζ) = f (ζ)

; therefore ( 8 ) implies that

d (f (ζ), g (ζ)) < ɛ

which by ( 3 ) gives

g (ζ) / \in K

. Take

ζ \in D_{j}

for some

j \in J

. Then

u (ζ) \in Ω_{1}

and ( 8 ) together with ( 3 ) implies that

g (ζ) / \in K

. This proves (iii'). It is easy to see that (iv') holds and the proof is complete. □

Definition 3.1.

[U, ρ, φ, D]

will be called a convex

q

-convex configuration in

C n

if the following hold:

(i) $U$ is open convex in $C n$ , and $φ : U \to R$ is a convex $C^{2}$ function such that the sublevel set ${z \in U; φ (z) < 0}$ is nonempty and relatively compact in $U$ ,
(ii) $ρ : U \to R$ is strictly convex in the first $(n - q + 1)$ coordinates, i.e., for any $p \in U$ the restriction of $ρ$ to the slice ${z \in U; z_{n - q + 2} = p_{n - q + 2}, z_{n - q + 3} = p_{n - q + 3}, \dots, z_{n} = p_{n}}$ is strictly convex,
(iii) $d ρ \land d φ \neq 0$ for $z \in U$ with $ρ (z) = φ (z) = 0$ , and
(iv) $D = {z \in U; ρ (z) < 0, φ (z) < 0}$ is neither empty nor equal to ${z \in U; φ (z) < 0}$ .

Lemma 3.2. Let

ρ

be a

q

-convex function on a complex manifold

X

. Then for any point

p \in X

, which is a regular point of

ρ

, there exist holomorphic coordinates

Φ : V \to C n

in some neighborhood of

p

such that

Φ (V)

is the open unit ball,

Φ (p) = 0

, and for

0 < r < 1

[Φ (V), ρ \circ Φ^{- 1}, | z |^{2} - r^{2}]

defines a convex

q

-convex configuration.

Definition 3.3. Given domains

A_{1}

A_{2}

V

X

we say that

[A_{1}, A_{2}, V]

is a convex

q

-convex extension element if

(i) $A_{1} \subset A_{2}$ ,
(ii) $A_{2} \ A_{1} \subset \subset V \subset \subset X$ ,
(iii) there exist convex $q$ -convex configurations

[U_{j}, ρ_{j}, φ_{j}, D_{j}]

C n

(j = 1, 2)

such that

U = U_{1} = U_{2}

φ = φ_{1} = φ_{2}

and for some biholomorphic map

Ψ

from

U

to a neighborhood of

\bar{V}

where

V = Ψ ({z \in U; φ (z) < 0})

we have

V \cap A_{j} = Ψ (D_{j})

(j = 1, 2)

Lemma 3.4. Let

X

be a complex manifold of dimension

n \geq 2

and let

1 \leq q \leq n - 1

. Let

D \subset \subset G \subset \subset X

be two domains with

b D

compact and let

ρ

be a

q

-convex function without degenerate critical points in a neighborhood

U

\bar{G \ D}

such that

D \cap U = {z \in U; ρ (z) < 0}

and

G \cap U = {z \in U; ρ (z) < 1}

and

ρ

does not have local minima in

U

Then there exists a finite number of domains

D = A_{0} \subset A_{1} \subset \dots \subset A_{m} = G

such that

A_{j}

can be obtained from

A_{j - 1}

by means of a

q

-convex extension element.

Moreover, for any open covering

{U_{i}}_{i \in I}

X

these domains

A_{j}

can be in addition chosen such that for each

j

1 \leq j \leq n

, there are an index

i \in I

and

V \subset \subset U_{i}

such that

[A_{j - 1}, A_{j}, V]

is a

q

-convex extension element.

Lemma 3.5. Let

X

be a complex manifold of dimension

n \geq 2

, equipped with some metric

d

and let

1 \leq q \leq n - 1

. Let

[A_{1}, A_{2}, V]

be a convex

q

-convex extension element and let

K

be a compact subset of

A_{1}

such that

V \cap K = \emptyset

. Assume that

f : \bar{△} \to X

is a continuous map, holomorphic on

△

, such that

f (b △) \cap A_{1} = \emptyset

, and

r

0 < r < 1

, such that

f (\bar{△} \ r △) \cap K = \emptyset

. Given

ɛ > 0

there is a continuous map

g : \bar{△} \to X

, holomorphic on

△

, with the following properties

(i) $g (b △) \cap A_{2} = \emptyset$ ,
(ii) $d (f (ζ), g (ζ)) < ɛ$ $(| ζ | \leq r)$ ,
(iii) $g (\bar{△} \ r △) \cap K = \emptyset$ ,
(iv) $f (0) = g (0)$ .

Proof. Note that we may assume that

f

is holomorphic in a neighborhood of

\bar{△}

Since

[A_{1}, A_{2}, V]

is a convex

q

-convex extension element we have that

A_{1} \subset A_{2}

A_{2} \ A_{1} \subset \subset V \subset \subset X

and there exist convex

q

-convex configurations

[U_{j}, ρ_{j}, φ_{j}, D_{j}]

C n

(j = 1, 2)

such that

U = U_{1} = U_{2}

φ = φ_{1} = φ_{2}

and for some biholomorphic map

Ψ

from

U

to a neighborhood of

\bar{V}

where

V = Ψ ({z \in U; φ (z) < 0})

we have

V \cap A_{j} = Ψ ({z \in U; φ (z) < 0, ρ_{j} (z) < 0})

(j = 1, 2)

. We can choose

λ > 0

so small that

f (b △) \cap Ψ ({z \in U; φ (z) < 0, ρ_{1} (z) < λ}) = \emptyset

and the set

V_{0} = Ψ ({z \in U; φ (z) < 0, ρ_{1} (z) > λ, ρ_{2} (z) < λ})

is relatively compact in

V

Fix

μ

0 \leq μ \leq 1

. The function

μ (ρ_{1} - λ) + (1 - μ) (ρ_{2} - λ)

is defined on

U

and it is strictly convex in the first two coordinates. Choose a point

q \in {z \in U; φ (z) < 0, ρ_{1} (z) > λ, ρ_{2} (z) < λ}

and write

q = (q_{1}, q_{2}, q^{''})

There is exactly one

μ

0 \leq μ \leq 1

, such that

μ (ρ_{1} (q) - λ) + (1 - μ) (ρ_{2} (q) - λ) = 0

. Denote by

M_{μ, q^{''}}

the set

{(z_{1}, z_{2}, q^{''}) \in U; μ (ρ_{1} (z_{1}, z_{2}, q^{''}) - λ) + (1 - μ) (ρ_{2} (z_{1}, z_{2}, q^{''}) - λ) = 0}

. Note that

M_{μ, q^{''}}

is a real submanifold of dimension

3

C 2 \times {q^{''}}

. Denote by

T_{q} M_{μ, q^{''}}

its real tangent space at

q

. The intersection

E_{q} = T_{q} M_{μ, q^{''}} \cap ı T_{q} M_{μ, q^{''}}

is a complex line. By strict convexity the intersection of

{q} + E_{q}

with

{z \in U; ρ_{2} (z) = λ}

is a bounded connected convex subset of

{q} + E_{q}

therefore it is conformally equivalent to the unit disc. If we vary

q

smoothly these convex sets vary smoothly. Therefore with a proof similar to the proof of [Glo,Lemma4.1] we get the following.

Sublemma. Let

I

be a closed arc. Given a continuous map

γ : I \to {z \in U; ρ_{1} (z) > λ, ρ_{2} (z) \leq λ}

there is a continuous map

H_{γ} : I \times \bar{△} \to X

such that

(a) for each $ζ \in I$ the map $H_{γ} (ζ, η)$ is holomorphic in $η$ ,
(b) $H_{γ} (ζ, 0) = γ (ζ)$ $(ζ \in I)$ ,
(c) $H_{γ} (I, \bar{△}) \subset {z \in U; ρ_{1} (z) > λ, ρ_{2} (z) \leq λ}$ ,
(d) $ρ_{2} (H_{γ} (ζ, η)) = λ$ $(ζ \in I, η \in b △)$ .

We note that if

ρ_{2} (γ (ζ)) = λ

then

H_{γ} (ζ, η) = γ (ζ)

(η \in \bar{△})

. By patching up the maps

Ψ \circ H_{γ}

one gets a continuous map

H : b △ \times △ \to X

such that

(a') for each $ζ \in b △$ the map $H (ζ, η)$ is holomorphic in $η$ ,
(b') $H (ζ, 0) = f (ζ)$ $(ζ \in b △)$ ,
(c') if for some $ζ \in b △$ we have $f (ζ) / \in V$ then $H (ζ, η) = f (ζ)$ $(η \in \bar{△})$ ,
(d') if for some $ζ \in b △$ we have $f (ζ) \in V$ then $H (ζ, \bar{△}) \subset V$ .

Take

ɛ_{0}

0 < ɛ_{0} < ɛ

, so small that

\begin{matrix} d (Ψ ({z \in U; ρ_{2} (z) = λ, ρ_{1} (z) \geq λ}), Ψ ({z \in U; ρ_{2} (z) \leq 0})) > ɛ_{0} . \end{matrix}

(9)

Now we use Lemma 2.1 to get the map

g

such that

(i') $d (H (ζ, b △), g (ζ)) < ɛ_{0}$ $(ζ \in b △)$ ,
(ii') $d (f (ζ), g (ζ)) < ɛ_{0}$ $(ζ \in r \bar{△})$ ,
(iii') $g (\bar{△} \ r △) \cap K = \emptyset$ ,
(iv') $f (0) = g (0)$ .

Properties (ii), (iii) and (iv) follow from (ii'), (iii') and (iv'). By ( 9 ), (d) and (i') we get (i). □

Lemma 4.1. Let

X

be a complex manifold of dimension

n \geq 3

and let

1 \leq q \leq n - 1

. Let

U \subset \subset X

and let

ρ : U \to R

be a

(n - q + 1)

-convex function such that

{z \in U; a \leq ρ (z) \leq b} \subset \subset U

. Assume that

ρ

has at most one critical point in

{z \in U; a \leq ρ (z) \leq b}

and if

p

is a critical point of

ρ

then

a < ρ (p) < b

. Let

K

be a compact subset of

X

such that

K \cap U = \emptyset

. Assume that

f : \bar{△} \to X

is a continuous map, holomorphic on

△

, such that

f (b △) \subset {z \in U; ρ (z) > a}

and choose

r

0 < r < 1

, such that

f (\bar{△} \ r △) \cap K = \emptyset

. Given

ɛ > 0

there exists a continuous map

g : \bar{△} \to X

, holomorphic on

△

, with the following properties

(i) $g (b △) \subset {z \in U; ρ (z) > b}$ ,
(ii) $d (f (ζ), g (ζ)) < ɛ$ $(| ζ | \leq r)$ ,
(iii) $g (\bar{△} \ r △) \cap K = \emptyset$ ,
(iv) $f (0) = g (0)$ .

Proof. If ρ does not have any critical points in the set { z ∈ U ; a ≤ ρ ( z ) ≤ b } then we proceed as follows. By Lemma 3.4 and by the remark following the lemma there exist finitely many domains { z ∈ U ; ρ ( z ) < a } = A 0 ⊂ A 1 ⊂ ⋯ ⊂ A m = { z ∈ U ; ρ ( z ) < b } such that A j can be obtained from A j − 1 by means of a convex q -convex extension element. Now we obtain the map g by using Lemma 3.5 m times where each time we push the boundary of the disc to the complement of the set A j .
Now assume that p is a critical point of ρ with a < ρ ( p ) < b . It is enough to prove the lemma in the case q = n − 1 . By a small perturbation of ρ in the neighborhood of p we may in addition assume that p / ∈ f ( b △ ) . If p is a local minimum then the boundary of the disc already lies above the critical level set and we proceed in the same way as if there were no critical points.
Therefore from now on we may assume that p is not a local minimum. We can choose a biholomorphic change of coordinates Φ from a neighborhood of p to a neighborhood V of 0 in C n such that Φ ( p ) = 0 and ρ ∘ Φ − 1 | V is strictly plurisubharmonic in the first two coordinates. Denote by L the complex 2 -dimensional subspace generated by z 1 , z 2 . Let V 0 ⊂ ⊂ V be a neighborhood of 0 . Then for every small perturbation L ′ of L and for each z ∈ V 0 the map ρ ∘ Φ − 1 is strictly plurisubharmonic on ( z + L ′ ) ∩ V 0 . Therefore one can choose three complex linear subspaces L 1 , L 2 and L 3 in C n such that L 1 ∩ L 2 ∩ L 3 = { 0 } and for each j , 1 ≤ j ≤ 3 , and for z ∈ V 0 the map ρ ∘ Φ − 1 is strictly plurisubharmonic on ( z + L j ) ∩ V 0 . There is a δ > 0 so small that { z ∈ C n ; d i s t ( z , L j ) < δ , 1 ≤ j ≤ 3 } ⊂ ⊂ V 0 . Denote this set by V 1 . By taking smaller δ if necessary we may assume that f ( b △ ) ∩ V 1 = ∅ .
For each w ∈ { z ∈ U ; a ≤ ρ ( z ) ≤ b } \ Φ − 1 ( V 0 ) we can choose a coordinate neighborhood U w , a biholomorphic map Φ w : U w → Φ w ( U w ) ⊂ C n and a convex map φ w : Φ w ( U w ) → R such that [ Φ w ( U w ) , ρ ∘ Φ w − 1 , φ w ] is a convex ( n − 1 ) -convex configuration. We can further assume that U w ⊂ ⊂ U \ Φ − 1 ( V 1 ) .
For each w ∈ Φ − 1 ( V 0 \ { z ∈ C n ; d i s t ( z , L j ) < δ 2 , 1 ≤ j ≤ 3 } ) there is j , 1 ≤ j ≤ 3 , such that d i s t ( Φ ( w ) , L j ) ≥ δ 2 . By the above ρ ∘ Φ − 1 is strictly plurisubharmonic on ( Φ ( w ) + L j ) ∩ V 0 . Hence there is a biholomorphic change of coordinates on Φ ( w ) + L j near Φ ( w ) such that in the new coordinates ρ ∘ Φ − 1 is strictly convex on Φ ( w ) + L j near Φ ( w ) . Since strict convexity is preserved by small perturbations it follows that there is a neighborhood U w of w in X such that ρ ∘ Φ − 1 is strictly convex in the new coordinates on ( Φ ( z ) + L j ) ∩ Φ ( U w ) for each z ∈ U w . By construction the small tangent discs corresponding to Φ ( z ) along which we lift in Lemma 3.5 lie in Φ ( z ) + L j .
Denote by U p the set Φ − 1 ( { z ∈ C n ; d i s t ( z , L j ) < δ 2 , 1 ≤ j ≤ 3 } ) . Now we use Lemma 3.4 together with the remark following the lemma for the covering { U p } ∪ { U w ; w ∈ U , a ≤ ρ ( w ) ≤ b , w / ∈ U p } to get a finite number of domains { z ∈ U ; ρ ( z ) < a } = A 0 ⊂ A 1 ⊂ ⋯ ⊂ A m = { z ∈ U ; ρ ( z ) < b } such that for each j , 1 ≤ j ≤ m , there is a set U w from the covering such that A j \ A j − 1 ⊂ U w and if w ≠ p then A j can be obtained from A j − 1 by means of a convex q -convex extension element.
We construct the map g inductively. At each step we construct a continuous map f j : △ ¯ → U , holomorphic on △ , with the following properties
- (a) $f_{j} (\bar{△} \ r △) \cap K = \emptyset$ ,
- (b) $f_{j} (b △) \cap ({\bar{U}}_{p} \cup A_{j}) = \emptyset$ ,
- (c) $d (f_{j} (ζ), f (ζ)) < \frac{j}{m} ɛ$ $(| ζ | \leq r)$ ,
- (d) $f_{j} (0) = f (0) .$
Let f 0 = f and note that f 0 satisfies all the properties. Assume that we have already constructed the map f j with the properties (a), (b), (c) and (d) for some j , 0 ≤ j ≤ m − 1 . If A j + 1 \ A j ⊂ U p then we put f j + 1 = f j . Then f j + 1 obviously satisfies (a), (c) and (d). The property (b) follows from the fact that A j + 1 \ A j ⊂ U p and that the map f j satisfies (b). Otherwise, if A j + 1 \ A j ⊂ U w , w ≠ p , then we use Lemma 3.5 for the convex q -convex extension element [ A j − 1 , A j , U w ] to get the map f j + 1 . The fact that f j + 1 ( b △ ) misses U ¯ p follows from the properties of the covering; if Φ ( U w ) misses V 1 then obviously f j + 1 ( b △ ) misses U ¯ p . Otherwise, there is j , 1 ≤ j ≤ 3 , such that d i s t ( Φ ( U w ) , L j ) ≥ δ 2 . By (b) the boundary of the disc does not intersect U ¯ p therefore at each boundary point ζ ∈ b △ we have d i s t ( Φ ( f j ( ζ ) ) , L j ) > δ .
The family of small tangent discs along which we lift the boundary of the disc f j lies at each boundary point f j ( ζ ) in the local coordinates in the plane Φ ( f j ( ζ ) ) + L j . Therefore if the perturbation constants are small enough it holds that d i s t ( Φ ( f j + 1 ( ζ ) ) , L j ) > δ ( ζ ∈ b △ ) . This proves (b).
The properties (a), (b) and (d) are easily satisfied. The construction is finished. The map g = f m has all the required properties and the proof is complete. □

Proof of Theorem 1.2 . By Morse theory ([HL,Observation4.15] and [HL,Proposition0.5] ) we get an exhaustion function

ρ

of class

C^{\infty}

without degenerate critical points and

M^{'}

such that

ρ

(n - 1)

-convex on

{ρ > M^{'}}

and such that for

ζ \in b △

it holds that

ρ (f (ζ)) > M^{'}

. We may additionally assume that there is only one critical point on each critical level set.

Choose an increasing sequence

a_{j}

of regular values of

ρ

, converging to

\infty

and such that

ρ (f (ζ)) > a_{1}

(ζ \in b △)

and for each

j \in N

there is at most one critical value on

(a_{j}, a_{j + 1})

. Choose a decreasing sequence

ɛ_{j} > 0

such that

\begin{matrix} z \in X, ρ (z) \leq a_{j}, w \in X, d (z, w) \leq ɛ_{j} then | ρ (z) - ρ (w) | < 1 . \end{matrix}

(10)

Using Lemma 4.1 one can construct inductively a sequence of continuous maps

f_{n} : \bar{△} \to X

, holomorphic on

△

, and an increasing sequence

r_{n}

of positive numbers converging to

1

, such that for each

n

(a) $ρ (f_{n} (ζ)) > a_{n}$ $(ζ \in b △)$ ,
(b) $ρ (f_{n} (ζ)) > a_{n - 1} - 1$ $(r_{n} \leq | ζ | \leq 1)$ ,
(c) $d (f_{n} (ζ), f_{n - 1} (ζ)) < \frac{ɛ_{n}}{2^{n}}$ $(| ζ | \leq r_{n})$ ,
(d) $f_{n} (0) = f (0)$ .

By (c) the sequence

f_{n}

converges uniformly on compacta on

△

and the limit map

g

is holomorphic on

△

. By (d) we obtain that

g (0) = f (0)

. Suppose

n \in N

and

r_{n} \leq | ζ | \leq r_{n + 1}

. Then we get by (c) that

| f_{n} (ζ) - g (ζ) | \leq | f_{n} (ζ) - f_{n + 1} (ζ) | + | f_{n + 1} (ζ) - f_{n + 2} (ζ) | + \dots

\leq \frac{ɛ_{n + 1}}{2^{n + 1}} + \frac{ɛ_{n + 2}}{2^{n + 2}} + \dots < ɛ_{n} .

ρ (g (ζ)) \leq a_{n}

then this together with (b) and ( 10 ) implies that

ρ (g (ζ)) > a_{n - 1} - 2

(r_{n} \leq | ζ | \leq r_{n + 1})

. Since

{lim}_{n \to \infty} r_{n} = 1

and since

{lim}_{n \to \infty} a_{n} = \infty

, it follows that

g

is a proper map. This completes the proof. □

November 27, 2006

Barbara Drinovec Drnovšek