Brody Curvres and Mean Dimension

JOURNAL OF THE
AMERICAN MATHEMATICAL SOCIETY

Volume 28, Number 1, January 2015, Pages 159182
S 0894-0347(2014)00798-0
Article electronically published on May 22, 2014
BRODY CURVES AND MEAN DIMENSION
SHINICHIROH MATSUO AND MASAKI TSUKAMOTO
1. Introduction

1.1. Main result. Let z = x + y 1 C be the standard coordinate in the
complex plane C. Let f = [f0 : f1 : : fN ] : C CP N be a holomorphic map
(fi : holomorphic function). We dene |df |(z) 0 by

1 2 2
|df | (z) :=
2
log(|f0 | + |f1 | + + |fN | )
2 2 2
:= + 2 .
4 x2 y
|df |(z) is classically called a spherical derivative. It evaluates the dilatation of the
map f with respect to the Euclidean metric on C and the FubiniStudy metric on
CP N . Namely, for a tangent vector u Tz C, the norm of df (u) Tf (z) CP N is
given by |df (u)| = |df |(z) |u|. See the equation (6) in Section 4.2 for more details.
A holomorphic map f : C CP N is called a Brody curve ([3]) if it satises
|df |(z) 1 for all z C (i.e. f is 1-Lipschitz). When N = 1, Brody curves are
meromorphic functions f : C C {} satisfying
|f (z)|
|df |(z) = 1.
(1 + |f (z)|2 )
The exponential function f (z) = ez satises this condition. If f (z) is a rational
function or elliptic function, then we can nd c > 0 such that f (cz) is a Brody curve.
Hence all rational/exponential/elliptic functions become Brody curves under some
scale changes. So Brody curves exhibit a quite wide variety of behaviors. Moreover
we cannot expect any symmetry for general Brody curves. Therefore it is dicult
to establish a deep structure theory of Brody curves.
In this paper we adopt a relatively new viewpoint (initiated by Gromov [11]) for
the study of Brody curves. We study the dynamical system consisting of Brody
curves. General Brody curves might have little structure, but there is a possibility
that the system of all Brody curves has much more structure than individual Brody
curves. The purpose of this paper is to show that it certainly has a beautiful
structure.
Let M(CP N ) be the space of Brody curves in CP N endowed with the compact-
open topology (the topology of uniform convergence on compact subsets). M(CP N )
is a compact metrizable space, and it becomes a dynamical system under the fol-
lowing continuous C-action:
M(CP N ) C M(CP N ), (f (z), a) f (z + a).
Received by the editors June 28, 2012 and, in revised form, January 6, 2014.
2010 Mathematics Subject Classication. Primary 32H30, 54H20.
Key words and phrases. Brody curve, mean dimension, deformation theory.
2014
c American Mathematical Society
159
Licensed to Stony Brook Univ. Prepared on Fri Jun 2 15:58:32 EDT 2017 for download from IP 129.49.120.232.
License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use
160 S. MATSUO AND M. TSUKAMOTO
We can easily prove that the dynamical system M(CP N ) is innite dimensional
and has innite topological entropy. So this is a very huge system. Gromov [11]
introduced the notion mean dimension for the study of this kind of huge dynam-
ical system. Mean dimension is an invariant of topological dynamical systems. We
review its denition in Section 2.1. Our main subject in this paper is to calculate
the mean dimension dim(M(CP N ) : C) of the system M(CP N ).
Let f : C CP N be a Brody curve. We dene an energy density (f ) by
setting

1
(1) (f ) := lim sup |df | dxdy .
2
R R2 aC |za|<R
This limit always exists; see Section 2.2. Let (CP N ) be the supremum of (f )
over f M(CP N ).
The main result of this paper is the following:
Theorem 1.1.
2(N + 1)(CP N ) dim(M(CP N ) : C) 4N (CP N ).
Corollary 1.2.
dim(M(CP 1 ) : C) = 4(CP 1 ).
The formula dim(M(CP 1 ) : C) = 4(CP 1 ) was conjectured in [21, p. 1643, (4)].
This formula is very surprising (at least for the authors) because the denitions of
the left-hand side and the right-hand side are totally dierent.
The upper bound dim(M(CP N ) : C) 4N (CP N ) was already proved in [18,
Theorem 1.5] by using the Nevanlinna theory. (Remark: A dierent denition of
the energy density was used in the papers [18, 21]. For this point, see Section 2.2.)
The subject of the present paper is to prove the lower bound dim(M(CP N ) : C)
2(N + 1)(CP N ).
1.2. Non-degenerate Brody curves. For a C and r > 0 we set Dr (a) := {z

C| |z a| r}. The following is a key notion of the paper. This notion was rst
introduced by Yosida [23]. Gromov [11, p. 399] also discussed it in a more general
situation. See also Eremenko [4, Section 4] and Remark 1.4 below.
Denition-Lemma 1.3. Let f : C CP N be a Brody curve. Then the following
two conditions are equivalent.
(i) Any constant curve does not belong to the closure of the C-orbit of f . In
other words, for any sequence of complex numbers {an }n1 , the sequence
of Brody curves {f (z + an )}n1 does not converge to a constant curve.
(ii) There exist > 0 and R > 0 such that for all a C we have
||df ||L (DR (a)) .
f is said to be non-degenerate if it satises one of (and hence both) the above
conditions.
Proof. The following argument is given in [23]. Suppose that the condition (ii)
fails. Then for any n 1 there is an C such that ||df ||L (D1 (an )) 1/n. Taking
a subsequence, we can assume that the sequence {f (z + an )}n1 converges to a
Brody curve g(z). Then ||dg||L (D1 (0)) = 0. This implies that g is a constant curve.
BRODY CURVES AND MEAN DIMENSION 161
Suppose the condition (ii) holds. Let {an }n1 be a sequence of complex numbers.
If {f (z + an )}n1 converges to g(z), then ||dg||L (DR (0)) . Hence g(z) is not a
constant curve. This proves the condition (i).
Remark 1.4. The above argument also proves that the conditions in Denition-
Lemma 1.3 are equivalent to the following:
(ii) For any R > 0 there exists > 0 such that for all a C we have
||df ||L (DR (a)) .
Yosida [23, Theorem 4] proved (i) (ii) for the case of N = 1. In [23] Brody
curves f : C CP 1 satisfying (i) are called meromorphic functions of 1st category.
In Eremenko [4, Section 4] Brody curves f : C CP N satisfying (i) are called
binormal curves. Gromov [11, p. 399] used the terminology uniformly nondegen-
erate.
Example 1.5. If f (z) M(CP 1 ) is a rational or exponential function, then it is a
degenerate (i.e. not non-degenerate) Brody curve. A non-constant elliptic function
f (z) M(CP 1 ) is a non-degenerate Brody curve.
In our viewpoint, non-degenerate Brody curves are non-singular points of the
system M(CP N ), and they behave very nicely for the calculation of the mean
dimension.
Theorem 1.6. Let f : C CP N be a non-degenerate Brody curve with
||df ||L (C) < 1. Then
dim(M(CP N ) : C) 2(N + 1)(f ).
Next we show that there are suciently many non-degenerate Brody curves.
Theorem 1.7. Let f : C CP N be a holomorphic map with ||df ||L (C) < 1.
Then for any > 0 there exists a non-degenerate Brody curve g : C CP N
satisfying ||dg||L (C) < 1 and (g) (f ) .
Proof of Theorem 1.1, assuming Theorems 1.6 and 1.7. The upper bound dim
(M(CP N ) : C) 4N (CP N ) was already proved in [18, Theorem 1.5]. Here we
prove the lower bound. Let f : C CP N be a Brody curve. Let 0 < c < 1
and set fc (z) = f (cz). Then |dfc |(z) = c|df |(cz) and (fc ) = c2 (f ). Since
||dfc ||L (C) c < 1, we can apply Theorem 1.7 to fc . Then for any > 0 there
exists a non-degenerate Brody curve g : C CP N satisfying ||dg||L (C) < 1 and
(g) (fc ) = c2 (f ) . By Theorem 1.6
dim(M(CP N ) : C) 2(N + 1)(g) 2(N + 1)(c2 (f ) ).
Let 0 and c 1. We get dim(M(CP N ) : C) 2(N + 1)(f ). Taking the
supremum over f M(CP N ), we get dim(M(CP N ) : C) 2(N + 1)(CP N ).
2. Some preliminaries
2.1. Review of mean dimension. In this subsection we review the denition of
mean dimension. For the detail, see Gromov [11] and LindenstraussWeiss [13].
For some related works, see also Lindenstrauss [12] and Gournay [68, 10].
Let (X, d) be a compact metric space, and let Y be a topological space. Let
> 0. A continuous map f : X Y is called an -embedding if Diamf 1 (y)
for all y Y . Here Diamf 1 (y) is the supremum of d(x1 , x2 ) over x1 , x2 f 1 (y).
We dene Widim (X, d) as the minimum integer n 0 such that there are an
n-dimensional polyhedron P and an -embedding f : X P .
For example, let X = [0, 1] [0, ] with the Euclidean distance. Then the pro-
jection : X [0, 1] is an -embedding, and we have Widim (X, Euclid) = 1.
The following example is very important in the later argument. This was given
by Gromov [11, p. 333]. For the detailed proof, see Gournay [10, Lemma 2.5] or
Tsukamoto [21, Appendix].
Example 2.1. Let V be a nite dimensional Banach space over R, and set Br (V ) :=
{x V | ||x|| r} for r > 0. For 0 < < r,
Widim (Br (V ), ||||) = dim V.
Here we consider the norm distance on Br (V ).
For a subset C and r > 0, we dene r as the set of a C satisfying

Dr (a) = and Dr (a) (C\) = . Let n (n 1) be a sequence of bounded
Borel subsets of C. It is called a Flner sequence if for all r > 0
Area(r n )
0 (n ).
Area(n )
For example, the sequence n := Dn (0) is a Flner sequence. The sequence n :=

[0, n] [0, n] is also Flner. We need the following OrnsteinWeiss lemma. For
the proof, see Gromov [11, pp. 336338].
Lemma 2.2. Let h : {bounded Borel subsets of C} R0 be a map satisfying the

following three conditions:
(i) If 1 2 , then h(1 ) h(2 ).
(ii) h(1 2 ) h(1 ) + h(2 ).
(iii) For any a C and any bounded Borel subset C, we have h(a + ) =
h() where a + := {a + z C|z }.
Then for any Flner sequence n (n 1) in C, the limit of the sequence
h(n )
(n 1)
Area(n )
exists, and its value is independent of the choice of a Flner sequence.
Suppose that the Lie group C continuously acts on a compact metric space X.
Here we dont assume that the distance is invariant under the group action. For a
subset C, we dene a new distance d on X by
d (x, y) := sup d(a.x, a.y).

a
It is easy to see that the map Widim (X, d ) satises the three conditions in
Lemma 2.2 for each > 0. So we dene a mean dimension dim(X : C) by

Widim (X, dn )
dim(X : C) := lim lim
+0 n Area(n )
where n (n 1) is a Flner sequence in C. The value of the mean dimension
dim(X : C) is independent of the choice of a Flner sequence, and it is a topological
invariant. (That is, it is independent of the choice of a distance on X compatible
with the topology.) For example, we have

Widim (X, dDR (0) )
dim(X : C) = lim lim
+0 R R2
(2)
Widim (X, d[0,R][0,R] )
= lim lim .
+0 R R2
2.2. Energy density. Here we explain some basic properties of the energy density
(f ) introduced in (1). Let f : C CP N be a Brody curve. Then the map

sup |df |2 dxdy
aC a+
clearly satises the three conditions in Lemma 2.2, where C is a bounded Borel
subset. Therefore we can dene the energy density (f ) by

1
(f ) := lim sup |df | dxdy ,
2
n Area(n ) aC a+
n
where n (n 1) is a Flner sequence in C. In particular, we have

1
(f ) = lim sup |df | dxdy
2
R R2 aC |za|<R
(3)
1
= lim 2 sup |df | dxdy .
2
R R a,bR [a,a+R][b,b+R]
From [19], the quantity (CP N ) = supf M(CP N ) (f ) satises
0 < (CP N ) < 1, lim (CP N ) = 1.

N
Moreover
2
2 dx
(CP 1 ) 1 10100 .
3 1 x3 1
See [21, Section 1.2] for the lower bound, and see [19, Section 5] for the upper
bound. Here
2
2 dx
= 0.6150198678198 .
3 1 x3 1
In the papers [18,21] a dierent denition of energy density was used. Let T (r, f )
be the NevanlinnaShimizuAhlfors characteristic function:
r
dt
T (r, f ) := |df | dxdy
2
.
1 |z|<t t
From the Brody condition |df | 1, we have T (r, f ) r 2 /2. We dene NSA (f ) by
2
NSA (f ) := lim sup 2 T (r, f ).
r r
It is easy to see NSA (f ) (f ). This quantity NSA (f ) was used in the papers
[18, 21]. (We used the notation e(f ) for NSA (f ) in [18, 21].) Let NSA (CP N ) be
the supremum of NSA (f ) over f M(CP N ). Trivially we have NSA (CP N )
(CP N ), but indeed we can prove the equality:
NSA (CP N ) = (CP N ).
This is proved in [22].
3. Proof of Theorem 1.6

In this section we prove Theorem 1.6 assuming Propositions 3.1 and 3.2 below.
Theorem 1.7 will be proved in Section 6. Let T CP N be the tangent bundle of CP N .
It naturally admits a structure of a holomorphic vector bundle. We consider the
FubiniStudy metric on it. Let f : C CP N be a Brody curve, and let f T CP N
be the pull-back of T CP N by f . f T CP N is a holomorphic vector bundle over the
complex plane C, and its Hermitian metric is given by the pull-back of the Fubini
Study metric. Let Hf be the space of holomorphic sections u : C f T CP N
satisfying ||u||L (C) < +. (Hf , ||||L (C) ) is a complex Banach space (possibly
innite dimensional). We set Br (Hf ) := {u Hf | ||u||L (C) r} for r 0.
Proposition 3.1. Let f : C CP N be a non-degenerate Brody curve with

||df ||L (C) < 1. Then there exist > 0 and a map
B (Hf ) M(CP N ), u fu ,
satisfying the following two conditions:
(i) f0 = f .
(ii) For all u, v B (Hf ) and z C
1
||u v||L (C) .
|d(fu (z), fv (z)) |u(z) v(z)||
8
Here d(, ) is the distance on CP N dened by the FubiniStudy metric, and |u(z)
v(z)| is the berwise norm of f T CP N .
Let R > 0 and C. is said to be an R-square if = [a, a + R] [b, b + R]
for some a, b R.
Proposition 3.2. Let f : C CP N be a non-degenerate Brody curve. Then for
any R-square C with R > 2 there exists a nite dimensional complex linear
subspace V Hf satisfying the following two conditions:
(i)
dimC V (N + 1) |df |2 dxdy Cf R.

Here Cf is a positive constant depending only on f (and independent of R,
).
(ii) For all u V we have ||u||L (C) 2 ||u||L () .
Propositions 3.1 and 3.2 will be proved later (Sections 4 and 5). Here we prove
Theorem 1.6, assuming them.
Proof of Theorem 1.6. We dene a distance on M(CP N ) by

1
dist(g, h) := n
sup d(g(z), h(z)), (g, h M(CP N )).
n=0
10 |z|n
Then |dist(g, h) d(g(0), h(0))| (1/9) supzC d(g(z), h(z)). Hence for C
1
(4) |dist (g, h) sup d(g(z), h(z))| sup d(g(z), h(z)).
z 9 zC
Let > 0 be the positive constant introduced in Proposition 3.1. Let C be
an R-square (R > 2). By Proposition 3.2, there exists V = V Hf satisfying
the conditions (i) and (ii) in Proposition 3.2. We investigate the map B (Hf )
M(CP N ), u fu , (given by Proposition 3.1) and its restriction to B (V ) :=
V B (Hf ).
From the condition (ii) of Proposition 3.1, for u, v B (Hf ), we have
supzC d(fu (z), fv (z)) (9/8) ||u v||L (C) . Hence (B (Hf ), ||||L (C) ) M(CP N )
is continuous. For u, v B (Hf )

dist (fu , fv ) sup |u(z) v(z)|
z

dist (fu , fv ) sup d(fu (z), fv (z)) + sup d(fu (z), fv (z)) sup |u(z) v(z)|

z z z
1 1
sup d(fu (z), fv (z)) + ||u v||L (C) (by Proposition 3.1 (ii) and (4))
9 zC 8
1
||u v||L (C) .
4
Thus
1
||u v||L (C) .
||u v||L () dist (fu , fv ) +
4
For u, v B (V ) = V B (Hf ), we have ||u v||L (C) 2 ||u v||L () by
Proposition 3.2 (ii). Hence
||u v||L (C) 4 dist (fu , fv ), (u, v B (V )).
Hence for < /4,
Widim (M(CP N ), dist )
Widim4 (B (V ), ||||L (C) )
= dimR V (by Example 2.1)

2(N + 1) |df |2 dxdy 2Cf R (by Proposition 3.2 (i)).

Since Widim (M(CP N ), dist ) = Widim (M(CP N ), dist[0,R][0,R] ), for < /4,
the quantity Widim (M(CP N ), dist[0,R][0,R] ) is bounded from below by

2(N + 1) sup |df |2 dxdy 2Cf R.

Here runs over all R-squares. Dividing this by R2 and letting R , we get

1
lim Widim (M(CP ), dist[0,R][0,R] ) 2(N + 1)(f ).
N
R R2
Here we have used (3). Let 0. Then dim(M(CP N ) : C) 2(N + 1)(f )

by (2).
Remark 3.3. The above argument also gives the lower bound on the local mean
dimension dimf (M(CP N ) : C). Local mean dimension is a notion introduced in
[15]. The readers can skip this remark.
Let f : C CP N be a non-degenerate Brody curve with ||df ||L (C) < 1. Let
Br (f )C M(CP N ) (r > 0) be the set of g M(CP N ) satisfying distC (f, g) r.
Since f0 = f , if (4/5)r then u B(4/5)r (Hf ) satises fu Br (f )C . Let C
be an R-square (R > 2). As in the above proof, for 4 < (4/5)r , we get

Widim (Br (f )C , dist ) 2(N + 1) |df |2 dxdy 2Cf R.

Hence

1
dimf (M(CP N ) : C) := lim lim lim sup Widim (Br (f )C , dist )
r+0 +0 R R2 :R-square
2(N + 1)(f ).
Then dimloc (M(CP N ) : C) := supf M(CP N ) dimf (M(CP N ) : C) satises
2(N + 1)(CP N ) dimloc (M(CP N ) : C) dim(M(CP N ) : C) 4N (CP N ).

The proof is the same as the proof of Theorem 1.1. In particular we get
dimloc (M(CP 1 ) : C) = dim(M(CP 1 ) : C).
4. Proof of Proposition 3.1

In this section we prove Proposition 3.1.
4.1. Analytic preliminaries. Let f : C CP N be a Brody curve. As in

Section 3, let T CP N be the tangent bundle of CP N with the natural holomor-
phic vector bundle structure, and let E := f T CP N be the pull-back of T CP N . E
is a holomorphic vector bundle over the complex plane C. Its Hermitian metric h is
given by the pull-back of the FubiniStudy metric. E is equipped with the unitary
connection dened by the holomorphic structure and the metric h.
Let 1 < p < be a real number, and k 0 be an integer. Let a Lpk,loc (0,i (E))
(i = 0, 1) be a locally Lpk -section of 0,i (E) (the C -vector bundle of (0, i)-forms
valued in E). For a subset C, we set
k 1/p

||a||Lp () := |n a|p dxdy .
k
n=0
We dene the Lpk -norm ||a|| Lp by

k
||a|| Lp := sup ||a||Lp (D1 (z)) .

k k
zC
Let Lpk (0,i (E)) be the Banach space of all a Lpk,loc (0,i (E)) satisfying
||a|| Lp < +.
k
Lemma 4.1. (i) For a L22,loc (0,i (E)),

||a||L (C) const ||a|| L2 .
2
Precisely speaking, if the right-hand side is nite then the left-hand side is also nite
and satises the inequality.
(ii) If a Lp2,loc (0,i (E)) with p > 2, then
||a||L (C) + ||a||L (C) constp ||a|| Lp .
2
Proof. Since M(CP N ) is compact, there are > 0 and constk > 0 (k 0) such that
for every z C there is a trivialization u of the holomorphic vector bundle E over a
neighborhood of D (z) such that u h = (h) (the Hermitian matrix representing

h under the trivialization u) satises hC k (D (z)) , h C k (D (z)) constk . Here

(h ) = (h)1 . Then the norms ||a||Lp (D (z)) and ||a||L (D (z)) are equivalent to
k
||u a||Lp (D (z)) and ||u a||L (D (z)) uniformly in z C respectively. (We consider
k
u a as a CN -valued (0, i)-form in D (z).) Hence the Sobolev embedding theorem
(GilbargTrudinger [5, Chapter 7.7]) implies
||a||L (D (z)) const ||a||L2 (D (z)) .
2
Here the important point is that const is independent of z C. Thus ||a||L (C)
const ||a|| L2 . (ii) can be proved in the same way.
2
Let : C R be a C -function satisfying ||||C k (C) < + for all k 0. We

set (a) := e (e a) for a 0,1 (E). Here is the formal adjoint of the
Dolbeault operator : 0 (E) 0,1 (E) with respect to the Hermitian metric h.
is the formal adjoint of with respect to the metric e h. We dene the operator
: 0,i (E) 0,i (E) by setting
a := a
(i = 0), a := a (i = 1).
Lemma 4.2. For a Lpk+2 (0,i (E)),

||a|| Lp constp,k, ||a|| Lp + || a|| Lp .
k+2 k
More precisely, if a Lpk+2,loc (0,i (E))

and the right-hand side of the above is nite
p
then a Lk+2 and satises the above inequality.
Proof. We use the trivialization u of E introduced in the proof of Lemma 4.1.
Since ||||C l (C) < + for all l 0, under the trivialization u, the operator is
represented as

= (1/2) + A +B +C
x y
over a neighborhood of D (z) where the C l -norms (l 0) of the matrices A, B, C
over D (z) are bounded uniformly in z C. Then from the Lp -estimate (Gilbarg
Trudinger [5, Chapter 9.5])

||a||Lp (D/2 (z)) constp,k, ||a||Lp (D (z)) + || a||Lp (D (z)) .
k+2 k
The desired estimate follows from this.
4.2. Perturbation of the Hermitian metric. Here we develop a perturbation

technique of a Hermitian metric (Lemma 4.5 below). Gromov also discussed it in
[11, p. 399]. Tsukamoto [21, Section 4.3] studied an easier situation.
Lemma 4.3. Let g : C R0 be a non-negative smooth function with ||g||C k (C) <
+ for all k 0. We suppose that the following non-degeneracy condition holds:
There exist > 0 and R > 0 such that for all p C we have ||g||L (DR (p)) .
Then there exists a smooth function : C R satisfying
( + 1) = g, ||||C k (C) < + (k 0), sup (z) < 0.
zC
2 2 2 2
Here = /x + /y .
Proof. We need the following sublemma.
Sublemma 4.4. Let : C R be a function of class C 2 ( Cloc 2
). Suppose that
the norms ||||L (C) and ||( + 1)||L (C) are both nite. Then
||||L (C) 4 ||( + 1)||L (C) .
Proof. Take z0 C such that |(z0 )| ||||L (C) /2. For simplicity, we suppose
z0 = 0. Moreover we suppose (0) 0. (If (0) < 0, then we apply the following
argument to .) We dene w : C R by
2
1
w(z) := e(x cos +y sin )/ 2 d.
2 0
w satises
( + 1/2)w = 0, min w(z) = w(0) = 1, w(z) + (|z| +).
zC
Then ( + 1)w = w/2 1/2. For > 0, set M := 2 ||( + 1)||L (C) + > 0.
(+1) (M w ) M/2(+1) = ||( + 1)||L (C) +/2(+1) /2.
Since the function M w is positive for |z| 1, the weak minimum principle
(GilbargTrudinger [5, Chapter 3.1, Corollary 3.2]) implies that this function is
non-negative everywhere. Hence
||||L (C) /2 (0) M w(0) = M = 2 ||( + 1)||L (C) + .
Let 0. We get
||||L (C) 4 ||( + 1)||L (C) .

Let n : C [0, 1] (n 1) be a cut-o function such that n = 1 over Dn (0)
and supp(n ) Dn+1 (0). We want to solve the equation ( + 1) = n g. The
following is a standard L2 -argument.
Let L21 (C) be the space of L2 -functions : C R satisfying /x, /y L2
with the inner product , L21 := , L2 + /x, /xL2 + /y, /
yL2 . Consider the bounded linear functional:
L21 (C) R, , n gL2 .
From the Riesz representation theorem, there uniquely exists n L21 (C) satisfying
, n L21 = , n gL2 for all L21 (C). This implies ( + 1)n = n g as a
distribution. From the local elliptic regularity, n is smooth and ||n ||L (C) < +.
Then we can apply Sublemma 4.4 to n and get
||n ||L (C) 4 ||n g||L (C) 4 ||g||L (C) < +.
By the local elliptic regularity, for every compact subset K C and k 0, the
sequence ||n ||C k (K) (n 1) is bounded. Then we can choose a subsequence n1 <
n2 < n3 < such that nk converges to some in C over every compact subset
of C. satises ( + 1) = g and ||||L (C) 4 ||g||L (C) . By the elliptic
regularity, ||||C k (C) < + for all k 0.
Note that we have not used the non-degeneracy condition of the function g so
far. We need it for the proof of the condition supzC (z) < 0.
Set M := supzC (z). There are zn C (n 1) such that (zn ) M . Set
n (z) := (z + zn ) and gn (z) := g(z + zn ). Then
( + 1)n = gn .
The sequences ||n ||C k (C) and ||gn ||C k (C) (n 1) are bounded for every k 0. Hence
by choosing a subsequence (denoted also by n and gn ), we can assume that n
and gn converge to and g respectively in C over every compact subset of C.
They satisfy
g 0, ( + 1) = g 0, (z) (0) = M.
From the non-degeneracy condition of g, the function g is not zero. Hence if
is a constant, then = g is a negative constant function and M < 0. If
is not a constant, then the strong maximum principle [5, Chapter 3.2, Theorem
3.5] implies that cannot achieve a non-negative maximum value. Hence M =
(0) = maxzC (z) < 0.
Recall that f : C CP N is a Brody curve and E = f T CP N . For a 0,1 (E)

we have the Weintzenbock formula:
1
(5) a = a + a,
2
where := [/z , / z] is the curvature operator. The crucial fact for the
analysis of this paper is that the holomorphic bisectional curvature of the Fubini
Study metric is positive. From this, there exists a positive constant c such that
h(a, a) c|df |2 |a|2 .
This means that the curvature operator is positive where |df | is positive. The non-
degeneracy condition of the map f enters into the argument through this point;
see the condition (ii) of Denition-Lemma 1.3. In the next lemma we will prove
that if f is non-degenerate then we can perturb the Hermitian metric h so that the
curvature is uniformly positive.
Lemma 4.5. Let f : C CP N be a non-degenerate Brody curve. There is a
smooth function : C R with ||||C k (C) < + (k 0) satisfying the following.
Let be the curvature of the Hermitian metric h := e h. Then there is c > 0
such that
h ( a, a) c |a|2h
for all a 0,1 (E).

Proof. We have a = + a for a 0,1 (E), and hence
4 a

2
h ( a, a) = e |a|h + h(a, a) e + c|df |2 |a|2h .
4 4
By the non-degeneracy of f and Lemma 4.3, there is a smooth function : C R
satisfying
( + 1) = 4c|df |2 , ||||C k (C) < + (k 0), sup (z) < 0.
zC
Then

h ( a, a) e (/4)|a|2h = (/4)|a|2h sup (z)/4 |a|2h .
zC

Hence c := supzC (z)/4 > 0 satises the statement.
In our convention, the FubiniStudy metric gij on CP N
is given by
1 2
gij = log(1 + |z1 |2 + + |zN |2 )
2 zi zj
over {[1 : z1 : : zN ]} CP N . The spherical derivative |df |(z) for a holomorphic
curve f : C CP N satises

(6) f 1 zj = |df |2 dxdy.
gij dzi d
The FubiniStudy metric gij satises the K
ahlerEinstein equation
2
Ricij = log(det(gkl )) = 2(N + 1)gij .
zi zj
From this, the curvature operator = [/z , / z] in (5) satises

1
(7) tr()dzd z = (N + 1)|df |2 dxdy
2

since tr()dzdz = f ( Ricij dzi d
zj ). The equation (7) will be used in the proof

of Proposition 5.1. Note that the form ( 1/2)tr()dzd z is the Chern form
representing c1 (E) although we have c1 (E) = 0 because H 2 (C; Z) = 0.
4.3. L -estimate. Let f : C CP N be a non-degenerate Brody curve, and let
: C R be a smooth function introduced in Lemma 4.5. Propositions 4.6 and
4.7 below essentially use the positivity of the curvature .
The following L -estimate was proved in [21, Proposition 4.2].
Proposition 4.6. Let a 0,1 (E) be an E-valued (0, 1)-form of class C 2 (a Cloc
2
).
Set b := a. If ||a||L (C) , ||b||L (C) < +, then
||a||L (C) constf, ||b||L (C) .
Proof. The proof is similar to the proof of Sublemma 4.4. For the detail, see [21, pp.
16481649].
Proposition 4.7. Let b L22,loc (0,1 (E)) and suppose ||b||L (C) < +. Then
there uniquely exists a L24,loc (0,1 (E)) satisfying
a = b, ||a||L (C) < +.
Moreover ||a||L (C) + ||a||L (C) constf, ||b||L (C) .
Proof. The uniqueness follows from Proposition 4.6. (Note the Sobolev embedding
L24,loc Cloc
2
in R2 .) So the problem is the existence. We have the Weintzenbock
formula: for a 0,1 (E)
1
a = a + a,
2
where is the unitary connection on E with respect to the metric h = e h.
satises the positivity condition in Lemma 4.5.
Let n : C [0, 1] be a cut-o function such that n = 1 over Dn (0) and
supp(n ) Dn+1 (0). From the positivity of the curvature, as in the proof of
Lemma 4.3, a standard L2 -argument shows that there is an L21 (0,1 (E)) (the
space of L2 -sections a of 0,1 (E) satisfying a L2 ) satisfying an = n b as
a distribution. For the detail, see [21, Lemma 5.3]. The local elliptic regularity
implies an L24,loc . By Lemmas 4.1 (i) and 4.2,

||an ||L (C) const ||an || L2 const ||an || L2 + || an || L2
2

const ||an ||L2 + ||n b||L (C) < +.
By Proposition 4.6 we have ||an ||L (C) constf, ||n b||L (C) constf, ||b||L (C) .
Then for any compact set K C the sequence ||an ||L2 (K) (n 1) is bounded. By
2
choosing a subsequence n1 < n2 < n3 < , the sequence ank converges to some
a weakly in L22 (DR (0)) (and hence strongly in L (DR (0))) for every R > 0. a
satises a = b, and ||a||L (C) supn1 ||an ||L (C) constf, ||b||L (C) . By the
local elliptic regularity a L24,loc . By Lemmas 4.1 (ii) and 4.2
||a||L (C) + ||a||L (C) const ||a|| L3 const (||a|| L3 + ||b|| L3 )
2
constf, ||b||L (C) .

4.4. Deformation theory. Let f : C CP N be a non-degenerate Brody curve
with ||df ||L (C) < 1. In this subsection we study a deformation of f and prove
Proposition 3.1. Gromov [11, pp. 399400, Projective interpolation theorem] stud-
ied a dierent kind of deformation theory. Our argument is a generalization of the
deformation theory of elliptic Brody curves developed in [21].
Consider the following map (see McDuSalamon [16, p. 40]):
: L2 (E) L2 (0,1 (E)), u Pu ( exp u) d
3 2 z.
Here exp u = expf (z) u(z) is dened by the exponential map of the FubiniStudy
metric, and

1
exp u := exp u + J exp u (J: complex structure of CP N ).
2 x y
Pu(z) : Texpf (z) u(z) CP N Tf (z) CP N is the parallel translation along the geodesic
expf (z) (tu(z)) (0 t 1).
is a smooth map between the Banach spaces. (0) = 0 and the derivative of
at the origin is equal to the Dolbeault operator:
d0 = : L2 (E) L2 (0,1 (E)).
3 2
Proposition 4.8. There is a bounded linear operator Q : L22 (0,1 (E))

L23 (E) satisfying Q = 1.
Proof. We will prove that the map

(8) = : L24 (0,1 (E)) L22 (0,1 (E))
is an isomorphism. ( : C R is a smooth function introduced in Lemma 4.5.)
Then Q := 1 : L22 (0,1 (E)) L23 (0,1 (E)) becomes a right inverse
of . The injectivity of the map (8) directly follows from the L -estimate in

Proposition 4.6.
On the other hand, by Proposition 4.7, for every b L22 (0,1 (E)) there is
a L L24,loc (0,1 (E)) satisfying a = b. By Lemma 4.2, a L24 . Thus the
map (8) is surjective.
Let Hf be the Banach space of all L -holomorphic sections of E introduced in

Section 3. Hf is equal to the kernel of the map : L23 (E) L22 (0,1 (E)) by
Lemmas 4.1 and 4.2. Moreover the norms |||| L2 (k 0) are all equivalent to the
k
norm ||||L (C) over Hf .
From Proposition 4.8 and the implicit function theorem, there are r > 0 and
a smooth map : {u Hf | ||u||L (C) < r} ImQ (ImQ L23 (E) is a closed
subspace) such that
(u + (u)) = 0, (0) = 0, d0 = 0.
The rst and second conditions imply that fu := expf (u + (u)) becomes a holo-
morphic curve with f0 = f . The third condition implies that for any > 0 there
exists 0 < < r such that if u, v Hf satisfy ||u||L (C) , ||v||L (C) , then
||(u) (v)||L (C) ||u v||L (C) .
Proof of Proposition 3.1. Since ||df ||L (C) < 1, if 1, the holomorphic curves fu
(u B (Hf )) satisfy ||dfu ||L (C) 1. We will prove that if 0 < < r is suciently
small, then the map
B (Hf ) u fu M(CP N )
satises the conditions in Proposition 3.1. The condition (i) (f0 = f ) is OK. So we
want to prove the condition (ii).
We choose 0 < < r suciently small so that all u, v B (Hf ) satisfy
||(u) (v)||L (C) (1/20) ||u v||L (C) ,
and that if v1 , v2 Tp CP N are two tangent vectors satisfying |v1 |, |v2 | 2, then
|d(exp(v1 ), exp(v2 )) |v1 v2 || (1/20)|v1 v2 |.
The former condition comes from d0 = 0, and the latter is just a standard property
of the exponential map. Then all u, v B (Hf ) satisfy
|d(exp(u + (u)), exp(v + (v))) |u + (u) v (v)||
(1/20)|u + (u) v (v)|
(1/20) ||u v||L (C) + (1/20) ||(u) (v)||L (C)
(1/20 + 1/400) ||u v||L (C) ,
and
||u + (u) v (v)| |u v|| |(u) (v)| (1/20) ||u v||L (C) .
These inequalities imply the condition (ii):

|d(exp(u + (u)), exp(v + (v))) |u v|| (1/8) ||u v||L (C) .

5. Study of Hf : proof of Proposition 3.2

In this section we prove Proposition 3.2. Let R > 0, and let = [a1 , b1 ]
[a2 , b2 ] C be an R-square (i.e. b1 = a1 + R and b2 = a2 + R). For 0 < r < R/2,
we set
r = {([a1 , a1 + r) (b1 r, b1 ]) [a2 , b2 ]} {[a1 , b1 ] ([a2 , a2 + r) (b2 r, b2 ])}.
This notation is used only in this section. It conicts with the notation r intro-
duced in Section 2.1. The following is a preliminary version of Proposition 3.2.
Proposition 5.1. Let f : C CP N be a Brody curve. Let > 0, and let C
be an R-square with R > 2. Then there exists a nite dimensional complex linear
subspace W 0 (E) (the space of C -sections of E = f T CP N ) satisfying the
following three conditions:
(i)

dimC W (N + 1) |df |2 dxdy C R,

where C is a constant depending only on . The important point is that it
is independent of R.
(ii) All u W satisfy u =0 outside of .

(iii) All u W satisfy u L (C)
||u||L (C) .
Proof. Set = [a1 , b1 ] [a2 , b2 ]. Let i : R R (i = 1, 2) be smooth functions

such that 0 i 1, i (x) = x over [ai + 1/2, bi 1/2], i (x) = i (ai + 1/4) over
x ai + 1/4 and i (x) = i (bi 1/4) over x bi 1/4. Moreover we assume that,
(k)
for k 1, |i | constk (depending only on k 1).

We dene a C -map f : C CP N by f(x+ 1y) := f (1 (x)+ 12 (y)). We
:= f T CP N be the
have |df|(z) := maxuTz C, |u|=1 |df(u)| 1 for all z C. Let E
N
pull-back of T CP by f . E is a complex vector bundle over C with the Hermitian
(the pull-back of the FubiniStudy metric) and the unitary connection
metric h
(the pull-back of the Levi-Civita connection on T CP ). From the denition of f,
N
the connection is at over 1/4 . Flat connections over 1/4 are classied by
their holonomy maps 1 (1/4 ) U (N ). Hence there is a bundle trivialization
(as a Hermitian vector bundle) g of E = d + A (A:
over 1/4 such that g()
connection matrix) satises
||A||C k (1/4 ) constk (k 0).
Here constk are universal constants depending only on k. The important point is
that they are independent of R. Let : [0, 1] be a cut-o function such that
= 1 over \1/5 , = 0 over 1/6 , and ||||C k () constk . We dene a unitary
connection on E over by := g 1 (d + A). We have = over \1/5 .

and the connection are equal to the
Under the trivialization g, the metric h
standard metric and the product connection of 1/6 CN over 1/6 .

Consider an elliptic curve T := C/(RZ + R 1Z), and let : C T be
the natural projection. We dene a complex vector bundle E over T as follows.
E = E over (\1/5 ) = \1/5 , and E |(1/4 ) is equal to the product bundle
(1/4 ) CN . We glue these by the map g. The metric h and the connection
naturally descend to the metric and connection on E (also denoted by h and ).
Let := [/z , / z] be the curvature of . From the denition, =
[/z , / z] over (\1/2 ) = \1/2 , and | | const (a universal constant)
all over T. Then by (7)

1
(9) c1 (E ) = tr( )dzd z (N + 1) |df |2 dxdy const R.
T 2 T

Let : (E ) (E ) be the Dolbeault operator over T twisted by the uni-
0 0,1
tary connection (i.e. the (0, 1)-part of the covariant derivative : 0 (E)

1 (E)). Let H be the space of u (E ) satisfying u = 0. From the
0 0
RiemannRoch formula and the above (9)

(10) dimC H 0
c1 (E ) (N + 1) |df |2 dxdy const R.
T
Lemma 5.2. For all u 0

H ,

|| u||L (T) K ||u||L (T) .
Here K is a universal constant (independent of f , R, ).
Proof. The connection has the following property: There is a universal constant
r > 0 such that for every p T there is a bundle trivialization v of a Hermitian
vector bundle E over Dr (p) satisfying v( ) = d + A with
||A ||C k (Dr (p)) constk (k 0).
Then the result follows from the elliptic regularity.
Let = () > 0 be a small number which will be xed later. We take points
p1 , . . . , pM (1 ) with M const R such that for every p (1 ) there is
pi satisfying d(p, pi ) . We dene V H 0
as the space of u H satisfying
0
u(pi ) = 0 for all i = 1, . . . , M . From (10),

M

(11) dimC V dimC H dimC
0
Epi (N + 1) |df |2 dxdy C R.
i=1
Let u V and p (1 ). Take pi satisfying d(p, pi ) . From u(pi ) = 0 and

Lemma 5.2,
|u(p)| || u||L (T) K ||u||L (T) .
We choose > 0 so that K < 1. Then the maximum of |u| is attained in T\(1 ).
Let : C [0, 1] be a cut-o such that = 1 over \1 , supp() is contained
in the interior of \1/2 , and |d| 10. For u V , we set u := u. Here
we identify the region \1/2 with (\1/2 ) where we have E = E, and we
consider u as a section of E over the plane C. Set W := {u |u V }. We have
||u ||L (C) = ||u||L (T) . Hence, by (11), we get the condition (i):

dimC W = dimC V (N + 1) |df |2 dxdy C R.

The condition (ii) is obviously satised. u =

u is supported in 1 .

u
10 ||u||L ((1 )) 10 K ||u||L (T) = 10 K ||u ||L (C) .
L (C)
We choose > 0 so that 10 K . Then the condition (iii) is satised.
Proof of Proposition 3.2. Let > 0 be a small number which will be xed later. By
Proposition 5.1, for this and any R-square (R > 2), there is a nite dimensional
complex linear subspace W 0 (E) satisfying the conditions (i), (ii), (iii) in
Proposition 5.1. By Proposition 4.7, there is a linear map
W 0,1 (E), u a,
such that

a = u,
a Cf u
Cf ||u||L (C) .
L (C) L (C)
Set u := u a. Then u
= 0 and ||u ||
L (C) (1 Cf ) ||u||L (C) . We choose

> 0 so that 1 Cf > 0. We set V := {u |u W }. Then V Hf and

dimC V = dimC W (N + 1) |df |2 dxdy C R.

For u W (recall supp(u) )

||u ||L (C) (1 + Cf ) ||u||L (C) = (1 + Cf ) ||u||L () ,
||u ||L () (1 Cf ) ||u||L () .

Hence
1 + Cf
||u ||L (C) ||u ||L () .
1 Cf
We choose > 0 so small that
1 + Cf
2.
1 Cf

6. Infinite gluing: proof of Theorem 1.7

We prove Theorem 1.7 in this section. Our method is gluing; we glue innitely
many rational curves to a (possibly degenerate) Brody curve f : C CP N , and
construct a non-degenerate one.
A kind of innite gluing construction is classically used for the proof of
Mittag-Leers theorem. Probably another origin of innite gluing construction
is the shadowing lemma in dynamical system theory (for example, see Bowen
[2, Chapter 3]). Angenent [1] developed a shadowing lemma for an elliptic PDE.
Gromov [11, p. 403] suggested an idea of gluing innitely many rational curves
to a (pseudo-)holomorphic curve. Macr`NolascoRicciardi [14] developed gluing
innitely many selfdual vortices. Gournay [6, 9] studied an innite gluing method
for pseudo-holomorphic curves. Tsukamoto [17, 20] studied gluing innitely many
YangMills instantons.
First we establish a result on gluing one rational curve.
Proposition 6.1. There are 0 > 0, R0 > 0 and K > 0 satisfying the following
statement. Let f : C CP N be a Brody curve. If f satises ||df ||L (DR (p)) < 0 for
some p C and R R0 + 1, then there exists a holomorphic curve g : C CP N
satisfying the following three conditions:
(i) 0 ||dg||L (DR (p)) 2/3.
(ii) ||dg|(z) |df |(z)| K/|z p|3 over |z p| > R.
(iii) d(f (z), g(z)) K/|z p|3 for z = p.
Proof. The proof is just a calculation. It may be helpful for some readers to consider
the case of N = 1 by themselves. Let > 0 be a suciently small number. 0 , R0 ,
K and will be xed later. Several conditions will be imposed on them through
the argument, but basically they need to satisfy
1
0 , R0 1, 4 .
R0 R0
Fix a > 0 so that the curve q : C CP N dened by q(z) := [1 : a/z 3 : : a/z 3 ]
satises ||dq||L (C) = 1/12. Here

3a N r 2
|dq|(z) = 6 (r = |z|).
(r + N a2 )
We can suppose ||dq||L (DR (0)) = 1/12 since we choose R0 1.
0
From the symmetry we can assume p = 0 and f (0) = [1 : 0 : : 0]. Let f (z) =
[1 : f1 (z) : : fN (z)] where fi (z) are meromorphic functions in C. Since |df | 0
over |z| R with R R0 + 1, if we choose 0 suciently small (0 /R0 ), we
have
(12) |fi (z)| , |fi (z)| (|z| R0 ).
Set gi (z) := fi (z) + a/z 3 , and we dene g : C CP N by g(z) := [1 : g1 (z) : :
gN (z)]. We will prove that this map g satises the conditions (i), (ii), (iii).
First we study the condition (iii). The FubiniStudy metric is given by
N
i=1 |dzi | + 1i<jN |zj dzi zi dzj |
2 2
2
ds = on {[1 : z1 : : zN ]}.
(1 + |zi |2 )2

|dzi |2 + 2( |zi |2 )( |dzi |2 ) 2(1 + |zi |2 ) |dzi |2 2
ds
2 |dzi |2 .
(1 + |zi |2 )2 (1 + |zi |2 )2

N
Hence ds 2/ i=1 |dzi | . Thus for f (z) = [1 : f1 (z) : : fN (z)] and
2
g(z) = [1 : f1 (z) + a/z : : fN (z) + a/z 3 ] we get

3

N
a 2N/
(13) d(f (z), g(z)) 2/ |a/z | =
3 2 .
i=1
|z|3
Next we study the conditions (i) and (ii). We have

|fi (z)|2 + i<j |fi (z)fj (z) fi (z)fj (z)|2
|df |(z) = ,
(1 + |fi (z)|2 )

|gi (z)|2 + i<j |gi (z)gj (z) gi (z)gj (z)|2
|dg|(z) = ,
(1 + |gi (z)|2 )
where
3a 3a a
gi = fi
, gi gj gi gj = (fi fj fi fj ) + 4 (fi fj ) + 3 (fi fj ).
z4 z z
Case 1: Suppose r := |z| R0 . We will prove 0 ||dg||L (DR (0)) 2/3.
0
From (12),
a 2a 3a 3a
|gi (z)| + 3 3 , |gi (z)| 4 4 .
r r r 2r
Here we have supposed min(a/R03 , 3a/(2R04 )). Then

N (3a/(2r 4 )) 3a N r 2 3a N r 2 |dq|(z)
|dg|(z) = = .
(1 + 4N a2 /r 6 ) 2 (r 6 + 4N a2 ) 8 (r 6 + N a2 ) 8
Hence ||dg||L (DR (0)) (1/8) ||dq||L (DR (0)) = 1/96 0 . Here we have supposed
0 0
0 1/96. On the other hand,
|dg|(z)

|3az 2 z 6 fi |2 + i<j|z 6 (fi fj fj fi ) + 3az 2 (fi fj ) + az 3 (fi fj )|2
= 6 .
(r + |a + z 3 fi |2 )
From (12),
a
|a + z 3 fi | a R03 , (here we suppose R03 a/2).
2
N a2 r 6 + N a2
r6 + |a + z 3 fi |2 r 6 + .
4 4
|3az 2 z 6 fi | 3ar 2 + r 6 r 2 (3a + R04 ) 4ar 2 , (we suppose R04 a).
|z 6 (fi fj fj fi ) + 3az 2 (fi fj ) + az 3 (fi fj )|
ar 2
r 2 (22 R04 + 6a + 2aR0 ) .
N
2

Here we have supposed 22 R04 + 6a + 2aR0 a/ N2 . Then

4ar 2 16N + 1 24ar 2 N 2
|dg|(z) 6 2
6 2
= 8|dq|(z) , (|dq| 1/12).
(r + N a ) (r + N a ) 3
Thus we get 0 ||dg||L (DR (0)) 2/3.
0
Case 2: Suppose |z| R0 . We will prove ||df |(z) |dg|(z)| K/r 3 for an
appropriate K > 0. We have
2
|fi | |gi |2 (|fi | + |gi |) |fi gi | (2|fi | + a/r 3 )(a/r 3 ) (2|fi | + a/R03 )(a/r 3 ),

|fi |2 |gi |2 a 2 Na 2a
|fi | + 1 + |fi |
r3 R03 r3
Na
(we suppose 3 2).
R0
If |fi | a/r 3 , then

3 2 |fi |2 a2 |fi |2 a2 x2
|gi | |fi | a/r
2
6 6, (x y)
2
y .
2
2 r 2 R0 2
If |fi | < a/r 3 , then

|fi |2 a2 |fi |2 a2
|gi |2 0 > 6 6.
2 r 2 R0
Therefore we always have |gi |2 |fi |2 /2 a2 /R06 .

N a2 1 1 N a2 1
1+ |gi |2 1 6 + |fi |2 1+ |fi |2 we suppose .
R0 2 2 R06 2
Hence

1 1 4a
3 (1 + |fi |) 4a N + 1 1 + |fi |2

1 + |gi |2 1 + |fi |2 (1 + |f |2 )2
r

i r 3 (1 + |fi |2 )2
(14)
4a N + 1 4a N + 1
= 3 .
r 3 (1 + |fi |2 )3/2 r (1 + |fi |2 )
Then, from gi = fi 3a/z 4 and the above (14),

|gi | |fi | |gi | |gi |

1 + |gk |2 1 + |fk | 2 1 + |gk | 2 1 + |fk | 2

|g | |fi |
+ i
1 + |fk | 2 1 + |fk | 2

4a N + 1(|fi | + 3a/r 4 ) 3a
+ 4 .
r 3 (1 + |fk |2 ) r (1 + |fk |2 )

From |df | 1, we have |fi |/(1 + |fk |2 ) . Hence the above is bounded by

4a N + 1 4a N + 1 3a
( + 3a/r ) + 3a/r
4 4
( + 3a) + 3 .
r3 r3 r

Here we have supposed r R0 1. Set Ka := 4a N + 1( + 3a) + 3a. Then

|gi | |fi | Ka
(15) .
1 + |gk |2 1 + |fk | 2 r3
From (14), for i < j,

|gi gj gj gi | |fi fj fj fi | |gi gj gj gi | |gi gj gj gi |

1 + |gk |2 1 + |fk |2 1 + |gk |2 1 + |fk |2

|gi gj gj gi | |fi fj fj fi |
+
1 + |fk |2 1 + |fk |2

4a N + 1|gi gj gj gi |

r 3 (1 + |fk |2 )
|(gi gj gj gi ) (fi fj fj fi )|
+ .
1 + |fk |2
From gi gj gj gi = (fi fj fj fi )+(3a/z 4 )(fi fj )+(a/z 3 )(fi fj ), this is bounded by

4a N + 1 |fi fj fj fi | 3a(|fi | + |fj |) a(|fi | + |fj |)
+ 4 +
r3 1 + |fk |2 r (1 + |fk |2 ) r 3 (1 + |fk |2 )
(16)
3a(|fi | + |fj |) a(|fi | + |fj |)
+ 4 + .
r (1 + |fk |2 ) r 3 (1 + |fk |2 )
From |df | 1,
|fi fj fj fi | |fi | + |fj |
, 2 .
1 + |fk |2 1 + |fk | 2
Since i < j,
|fi | + |fj | 2 |fi |2 + |fj |2
2.
1 + |fk | 2 1 + |fk | 2
Hence the above (16) is bounded by

4a N + 1 3a 2 2a 3a 2 2a
+ + + +
r3 r4 r3 r4 r3

4a N + 1 3a 2 2a
( + 3a 2 + 2a ) + + .
r3 r3 r3

Here r R0 1. Set Ka := 4a N + 1( + 3a 2 + 2a ) + 3a 2 + 2a . Then

|gi gj gj gi | |fi fj fj fi | Ka

1 + |gk |2 1 + |fk |2 r 3 .
From this and (15),

N Ka2 + N2 (Ka )2
N
||dg|(z) |df |(z)| (1/ ) N (Ka /r 3 )2 + (Ka /r 3 )2 = 3 .
2 r
Here we have used the inequality

x2 + + x2 y 2 + + y 2 (x1 y1 )2 + + (xl yl )2 .
1 l 1 l
Set

N
K := max a 2N/, N Ka2 + (Ka )2 / .
2
(This K satises the condition (iii) by (13).) Then
K
||df |(z) |dg|(z)| (r R0 ).
r3
Thus we have proved the condition (ii).
For R0 |z| R,
K 1 2
|dg|(z) ||df ||L (DR (0)) + 3 0 + ,
R0 2 3
where we have chosen R0 and 0 so that K/R03 1/2 and 0 1/6. In Case 1, we
proved 0 ||dg||L (DR (0)) 2/3. Thus we get the condition (i):
0
0 ||dg||L (DR (0)) 2/3.

Proof of Theorem 1.7. Let ||df ||L (C) 1 , (0 < 1). Let 0 , R0 , K be the

positive numbers introduced in Proposition 6.1. For > 0, we set := min(0 , ).
Let R = R(, ) R0 + 1 be a large positive number which will be xed later.
We index the elements of Z2 by natural numbers: Z2 = {(1 , 1 ), (
2 , 2 ),
(3 , 3 ), . . .}. For n 1, we set pn := 2R(n + 1n ) and n := {x + y 1
C| |x 2Rn | R, |y 2Rn | R}. The squares n (n 1) become a tiling of

the plane C.
We inductively dene the sequence of Brody curves fn : C CP N (n 0) as
follows. We set f0 := f . Suppose we have dened fn .
(1) If ||df ||L (n+1 ) , then we set fn+1 := fn .
(2) If ||df ||L (n+1 ) < and ||dfn ||L (n+1 ) 0 , then we set fn+1 := fn .
(3) If ||df ||L (n+1 ) < and ||dfn ||L (n+1 ) < 0 , then we apply Proposition
6.1 to fn and pn+1 (note DR (pn+1 ) n+1 ) and get a holomorphic map
fn+1 : C CP N satisfying the following (i), (ii), (iii).
(i) 0 ||dfn+1 ||L (DR (pn+1 )) 2/3.
(ii) ||dfn+1 |(z) |dfn |(z)| K/|z pn+1 |3 over |z pn+1 | > R.
(iii) d(fn (z), fn+1 (z)) K/|z pn+1 |3 for z = pn+1 .
For every n 1, by (i) and (ii)
K const K
|dfn |(z) max(1 , 2/3) + max(1 , 2/3) + .
|z pk |3 R3
k:|zpk |>R
Here const is a positive constant independent of n. We choose R so large that

the right-hand side is bounded by max(1 /2, 3/4) < 1. Then all fn : C
CP N become Brody curves, and we can continue the above inductive construction
innitely many times. Moreover, for all n 1,
(17) ||dfn ||L (C) max(1 /2, 3/4).
For any compact set C, by the condition (iii), there exists n() 1 such
that
K
sup d(fn (z), fn+1 (z)) < +.
z d(pk , )3
nn() k:d(pk ,)1
Hence the sequence fn converges to a holomorphic curve g : C CP N uniformly

over every compact subset of C. From (17) we have ||dg||L (C) max(1 /2, 3/4) <
1. We will prove that g is non-degenerate and (g) (f ) .
For proving the non-degeneracy of g, it is enough to show ||dg||L (n ) /2 for
all n 1. See the condition (ii) of Denition-Lemma 1.3.
Case 1: If |df |(z) for some z n , then
K const K
|dg|(z) .
|z pk |3 R3
k:k=n
We can choose R so large that ||dg||L (n ) /2.

Case 2: If |df |(z) < for all z n , then for some k {n 1, n} and w n
we have |dfk |(w) 0 . Hence
K const K
|dg|(w) 0 .
|w pl | 3 R3
l:l=n
We can choose R so large that ||dg||L (n ) /2.

We have proved that g is non-degenerate. Next we will prove (g) (f ) .
For this sake, it is enough to prove that for every n 1

1 1
(18) |dg| dxdy
2
|df |2 dxdy .
(2R)2 n (2R)2 n
Case 1: If ||df ||L (n ) , then for all z n

2K const K
|dg|2 (z) |df |2 (z) 2 ||dg|(z) |df |(z)|
|z pk |3 R3
k:k=n
for suciently large R. Hence (18) holds if we choose R suciently large.

Case 2: If ||df ||L (n ) < , then (recall = min(0 , ))

1
|df |2 dxdy 2 .
(2R)2 n
Hence (18) holds trivially.
Thus we have proved (g) (f ) .
Acknowledgements
The rst author was supported by a Grant-in-Aid for JSPS fellows (23149),
and the second author was supported by a Grant-in-Aid for Young Scientists (B)
(21740048).
References
[1] Sigurd Angenent, The shadowing lemma for elliptic PDE, Dynamics of innite-dimensional
systems (Lisbon, 1986), NATO Adv. Sci. Inst. Ser. F Comput. Systems Sci., vol. 37, Springer,
Berlin, 1987, pp. 722. MR921893 (89b:58165)
[2] Rufus Bowen, Equilibrium states and the ergodic theory of Anosov dieomorphisms, Second
revised edition, Lecture Notes in Mathematics, vol. 470, Springer-Verlag, Berlin, 2008. With
a preface by David Ruelle; Edited by Jean-Ren e Chazottes. MR2423393 (2009d:37038)
[3] Robert Brody, Compact manifolds and hyperbolicity, Trans. Amer. Math. Soc. 235 (1978),
213219. MR0470252 (57 #10010)
[4] Alexandre Eremenko, Normal holomorphic curves from parabolic regions to projective spaces,
available at arXiv:0710.1281.
[5] David Gilbarg and Neil S. Trudinger, Elliptic partial dierential equations of second or-
der, Classics in Mathematics, Springer-Verlag, Berlin, 2001. Reprint of the 1998 edition.
MR1814364 (2001k:35004)
[6] Antoine Gournay, Dimension moyenne et espaces dapplications pseudo-holomorphes,
Departement de Mathematiques dOrsay, 2008.
[7] Antoine Gournay, On a H older covariant version of mean dimension (English, with English
and French summaries), C. R. Math. Acad. Sci. Paris 347 (2009), no. 23-24, 13891392, DOI
10.1016/j.crma.2009.10.014. MR2588787 (2011b:46041)
[8] Antoine Gournay, A dynamical approach to von Neumann dimension, Discrete Contin. Dyn.
Syst. 26 (2010), no. 3, 967987, DOI 10.3934/dcds.2010.26.967. MR2600725 (2011d:37008)
[9] A. Gournay, Complex surfaces and interpolation on pseudo-holomorphic cylinder, available
at arXiv:1006.1775.
[10] Antoine Gournay, Widths of p balls, Houston J. Math. 37 (2011), no. 4, 12271248.
MR2875268
[11] Misha Gromov, Topological invariants of dynamical systems and spaces of holomorphic
maps. I, Math. Phys. Anal. Geom. 2 (1999), no. 4, 323415, DOI 10.1023/A:1009841100168.
MR1742309 (2001j:37037)
[12] Elon Lindenstrauss, Mean dimension, small entropy factors and an embedding theorem, Inst.

Hautes Etudes Sci. Publ. Math. 89 (1999), 227262 (2000). MR1793417 (2001j:37033)
[13] Elon Lindenstrauss and Benjamin Weiss, Mean topological dimension, Israel J. Math. 115
(2000), 124, DOI 10.1007/BF02810577. MR1749670 (2000m:37018)
[14] Marta Macr`, Margherita Nolasco, and Tonia Ricciardi, Asymptotics for selfdual vortices on
the torus and on the plane: a gluing technique, SIAM J. Math. Anal. 37 (2005), no. 1, 116,
DOI 10.1137/040619843. MR2176921 (2006g:58039)
[15] Shinichiroh Matsuo and Masaki Tsukamoto, Instanton approximation, periodic ASD con-
nections, and mean dimension, J. Funct. Anal. 260 (2011), no. 5, 13691427, DOI
10.1016/j.jfa.2010.11.008. MR2749431 (2012b:58019)
[16] Dusa McDu and Dietmar Salamon, J-holomorphic curves and symplectic topology, American
Mathematical Society Colloquium Publications, vol. 52, American Mathematical Society,
Providence, RI, 2004. MR2045629 (2004m:53154)
[17] Masaki Tsukamoto, Gluing an innite number of instantons, Nagoya Math. J. 188 (2007),
107131. MR2371770 (2009c:53035)
[18] Masaki Tsukamoto, Moduli space of Brody curves, energy and mean dimension, Nagoya
Math. J. 192 (2008), 2758. MR2477610 (2009i:32022)
[19] Masaki Tsukamoto, A packing problem for holomorphic curves, Nagoya Math. J. 194 (2009),
3368. MR2536526 (2010g:32024)
[20] Masaki Tsukamoto, Gauge theory on innite connected sum and mean dimension, Math.
Phys. Anal. Geom. 12 (2009), no. 4, 325380, DOI 10.1007/s11040-009-9065-z. MR2551659
(2011g:58014)
[21] Masaki Tsukamoto, Deformation of Brody curves and mean dimension, Ergodic Theory
Dynam. Systems 29 (2009), no. 5, 16411657, DOI 10.1017/S014338570800076X. MR2545021
(2011a:32033)
[22] Masaki Tsukamoto, Remark on energy density of Brody curves, Proc. Japan Acad. Ser. A
Math. Sci. 88 (2012), no. 8, 127131, DOI 10.3792/pjaa.88.127. MR2989063
[23] K. Yosida, On a class of meromorphic functions, Proc. Phys.-Math. Soc. Japan 16 (1934),
227235.
Department of Mathematics, Osaka University, Toyonaka, Osaka 560-0043, Japan

E-mail address: matsuo@math.sci.osaka-u.ac.jp
Department of Mathematics, Kyoto University, Kyoto 606-8502, Japan

E-mail address: tukamoto@math.kyoto-u.ac.jp

Brody Curvres and Mean Dimension

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Brody Curvres and Mean Dimension

Uploaded by

Copyright:

Available Formats

JOURNAL OF THE

AMERICAN MATHEMATICAL SOCIETY

BRODY CURVES AND MEAN DIMENSION

SHINICHIROH MATSUO AND MASAKI TSUKAMOTO

1.2. Non-degenerate Brody curves. For a C and r > 0 we set Dr (a) := {z

Widim (Br (V ), ||||) = dim V.

Here we consider the norm distance on Br (V ).

For a subset C and r > 0, we dene r as the set of a C satisfying

For example, the sequence n := Dn (0) is a Flner sequence. The sequence n :=

Lemma 2.2. Let h : {bounded Borel subsets of C} R0 be a map satisfying the

exists, and its value is independent of the choice of a Flner sequence.

d (x, y) := sup d(a.x, a.y).

where n (n 1) is a Flner sequence in C. In particular, we have

From [19], the quantity (CP N ) = supf M(CP N ) (f ) satises

0 < (CP N ) < 1, lim (CP N ) = 1.

3. Proof of Theorem 1.6

Proposition 3.1. Let f : C CP N be a non-degenerate Brody curve with

Proof of Theorem 1.6. We dene a distance on M(CP N ) by

Here we have used (3). Let 0. Then dim(M(CP N ) : C) 2(N + 1)(f )

Then dimloc (M(CP N ) : C) := supf M(CP N ) dimf (M(CP N ) : C) satises

2(N + 1)(CP N ) dimloc (M(CP N ) : C) dim(M(CP N ) : C) 4N (CP N ).

4. Proof of Proposition 3.1

4.1. Analytic preliminaries. Let f : C CP N be a Brody curve. As in

We dene the  Lpk -norm ||a|| Lp by

||a|| Lp := sup ||a||Lp (D1 (z)) .

Lemma 4.1. (i) For a L22,loc (0,i (E)),

Let : C R be a C -function satisfying ||||C k (C) < + for all k 0. We

Lemma 4.2. For a  Lpk+2 (0,i (E)),

More precisely, if a Lpk+2,loc (0,i (E))

The desired estimate follows from this.

4.2. Perturbation of the Hermitian metric. Here we develop a perturbation

Recall that f : C CP N is a Brody curve and E = f T CP N . For a 0,1 (E)

constf, ||b||L (C) .

Proposition 4.8. There is a bounded linear operator Q :  L22 (0,1 (E))

Proof. We will prove that the map

Let Hf be the Banach space of all L -holomorphic sections of E introduced in

These inequalities imply the condition (ii):

5. Study of Hf : proof of Proposition 3.2

Proof. Set = [a1 , b1 ] [a2 , b2 ]. Let i : R R (i = 1, 2) be smooth functions

RiemannRoch formula and the above (9)

Lemma 5.2. For all u 0

u(pi ) = 0 for all i = 1, . . . , M . From (10),

Let u V and p (1 ). Take pi satisfying d(p, pi ) . From u(pi ) = 0 and

The condition (ii) is obviously satised. u =

We choose > 0 so that 10 K . Then the condition (iii) is satised.

For u W (recall supp(u) )

||u ||L () (1 Cf ) ||u||L () .

6. Infinite gluing: proof of Theorem 1.7

g(z) = [1 : f1 (z) + a/z : : fN (z) + a/z 3 ] we get

Next we study the conditions (i) and (ii). We have

If |fi | < a/r 3 , then

Hence the above (16) is bounded by

0 ||dg||L (DR (0)) 2/3.

C| |x 2Rn | R, |y 2Rn | R}. The squares n (n 1) become a tiling of

Here const is a positive constant independent of n. We choose R so large that

Hence the sequence fn converges to a holomorphic curve g : C CP N uniformly

We can choose R so large that ||dg||L (n ) /2.

We can choose R so large that ||dg||L (n ) /2.

Case 1: If ||df ||L (n ) , then for all z n

for suciently large R. Hence (18) holds if we choose R suciently large.

Department of Mathematics, Osaka University, Toyonaka, Osaka 560-0043, Japan

Department of Mathematics, Kyoto University, Kyoto 606-8502, Japan

You might also like

We dene the Lpk -norm ||a|| Lp by

Lemma 4.2. For a Lpk+2 (0,i (E)),

Proposition 4.8. There is a bounded linear operator Q : L22 (0,1 (E))