A CHARACTERIZATION OF INOUE SURFACES

MARCO BRUNELLA
arXiv:1011.2035v1 [math.CV] 9 Nov 2010

Abstract. We give a characterization of Inoue surfaces in terms of

automorphic pluriharmonic functions on a cyclic covering. Together
with results of Chiose and Toma, this completes the classification of
compact complex surfaces of Kähler rank one.

In this paper we shall prove a conjecture proposed in [C-T]:

Theorem 0.1. Let S be a compact connected complex surface of algebraic

π
dimension 0. Suppose that there exists an infinite cyclic covering Se → S
e and a nonconstant
(with covering transformations generated by ϕ ∈ Aut(S))
e
positive pluriharmonic function F on S such that

F ◦ϕ =λ·F

for some λ ∈ R+ . Then S is a (possibly blown up) Inoue surface.

The class of Inoue surfaces was discovered by Inoue (and independently

Bombieri) around 1972 [Ino] [Nak]. They are special (and explicit) compact
quotients of H × C, and they enjoy the following properties:
- the first Betti number is 1, the second Betti number is 0;
- they admit holomorphic foliations;
- they do not contain compact complex curves.
Conversely, Inoue proved in [Ino] that any compact connected complex sur-
face with the above properties is a Inoue surface.
Our proof of Theorem 0.1 will be ultimately a reduction to Inoue’s the-
orem. The pluriharmonic function F naturally induces a holomorphic (and
possibly singular) foliation F on S. By a “topological” study of such a foli-
ation we will be able to understand some topological structure of S, and in
particular to show that c2 (Smin ) = 0 or c21 (Smin ) = 0 (where Smin denotes
the minimal model of S). From this vanishing of Chern numbers, and re-
sults of Kodaira and Inoue, the conclusion will be immediate. Remark that,
conversely and by construction, every Inoue surface satisfies the hypotheses
of Theorem 0.1, which therefore gives a precise characterization of Inoue
surfaces.
Together with the results of [C-T], Theorem 0.1 allows to complete the
classification of compact surfaces of Kähler rank one. Recall [H-L] [C-T]
that a compact connected complex surface S has Kähler rank one if it is
not Kählerian but it admits a closed semipositive (1, 1)-form, not identically
1
 A CHARACTERIZATION OF INOUE SURFACES 2

vanishing (this is not the original definition of [H-L], but it is equivalent to

it by the results of [C-T], see also [Lam] and [Tom]).
Corollary 0.2. ([C-T] + Theorem 0.1). The only compact connected com-
plex surfaces of Kähler rank one are:
(1) Non-Kählerian elliptic fibrations;
(2) Certain Hopf surfaces, and their blow-ups;
(3) Inoue surfaces, and their blow-ups.
In the case of a Inoue surface, a closed semipositive (1, 1)-form is given
by (dF/F ) ∧ (dc F/F ), with F as in Theorem 0.1.

1. Geometric preliminaries
Let S be a surface as in Theorem 0.1. Without loss of generality, we may
assume that S is minimal, since the hypotheses are clearly bimeromorphi-
cally invariant. The assumption a(S) = 0 implies that S belongs to the class
VII◦ [BPV, p. 188], that is b1 (S) = 1 and kod(S) = −∞: the existence of a
positive nonconstant pluriharmonic function on some covering of S excludes
the case of tori and K3 surfaces. For the same reason, S cannot be a Hopf
surface.
We claim that, in order to prove Theorem 0.1, it is sufficient to prove that
c2 (S) = 0
or
c21 (S) = 0.
Indeed, we firstly observe that these two conditions are equivalent, by Noether
formula and χ(OS ) = 0 (which follows from S ∈ VII◦ ). Then, c2 (S) = 0 and
b1 (S) = 1 imply b2 (S) = 0. By a classical result of Kodaira [Nak, Th. 2.4],
S contains no compact complex curve, otherwise it would be a Hopf surface.
Since S also admits a holomorphic foliation (see below), all the hypotheses
of Inoue’s theorem [Ino] are satisfied and we get that S is a Inoue surface.
The automorphic function F on Se induces a real analytic map

f = log F : S −→ S1 = R Z · log λ.
The regular fibers of f are smooth Levi-flat hypersurfaces in S, because F
is pluriharmonic. However, f could have also some singular fibers, corre-
sponding to critical points of F . In fact, our aim is precisely to show that
these singular fibers do not exist at all, since this is clearly equivalent to
c2 (S) = 0. In order to avoid some cumbersome statement, we shall sup-
pose that the fibers of f are connected. The general case requires only few
straightforward modifications of the proof below.
The holomorphic 1-form ω = ∂F ∈ Ω1 (S) e descends to S to a holomorphic
1
section (still denoted by ω) of Ω (S) ⊗ L, where L is a flat line bundle (the
one defined by the cocycle λ ∈ R+ ⊂ C∗ = H 1 (S, C∗ )). This twisted closed
holomorphic 1-form induces a holomorphic foliation F on S, which is tangent
to the fibers of f .
 A CHARACTERIZATION OF INOUE SURFACES 3

In the following it will be important to distinguish between the singular-

ities of F, Sing(F), and the zeroes of ω, Z(ω). The former are only isolated
points, since (as customary) we like to deal with “saturated” foliations. The
latter, on the contrary, may contain some compact complex curves. Remark
also that Z(ω) coincides with the set of critical points of f , Crit(f ).
The foliation F has a normal bundle NF and a tangent bundle TF [Br1],
which are related to the canonical bundle KS of S by the adjunction type
relation
NF ⊗ TF = KS−1 .
Because F is generated by ω ∈ Ω1 (S) ⊗ L, we have
X
NF = L ⊗ O(− m j Cj )
where {Cj } are the curves contained in Z(ω) (if any) and {mj } are the
respective vanishing orders.
We shall prove below that Z(ω) is at most composed by isolated points,
giving by the previous formula the flatness of NF = L. Then we shall prove
that either c2 (S) = 0 or TF is also flat. But in this second case we therefore
get that KS is flat too, hence c21 (S) = 0.

2. The structure of the regular fibers

Here we consider a regular fiber of f ,
Mϑ = {f = ϑ} , ϑ regular value,
and prove that it has the expected structure.
Proposition 2.1. The leaves of F|Mϑ are either all isomorphic to C, or all
isomorphic to C∗ . In the first case, Mϑ is diffeomorphic to T3 , and F|Mϑ
is a linear totally irrational foliation. In the second case, Mϑ is a S1 -bundle
over T2 , and F|Mϑ is the pull-back of a linear irrational foliation on T2 .
The first case will lead to Inoue surface of type SM , and the second case
(+) (−)
to those of type SN,p,q,r;t or SN,p,q,r [Ino].
Proof. The foliation Fϑ = F|Mϑ is defined by the closed and nonsingular
1-form β = dc F |Mϑ (which is well defined on a neighbourhood of any fiber,
up to a multiplicative constant). We may use some classical results of Tis-
chler [God, I.4] concerning the structure of (real) codimension one foliations
defined by closed 1-forms. According to those results, the foliation can be
smoothly perturbed to a fiber bundle over the circle with fiber Σg , the (real)
oriented compact surface of genus g ≥ 1. Note that, since a(S) = 0, the
leaves of Fϑ cannot be all compact, and so they are all dense in Mϑ . More-
over, by using the flow of a smooth vector field v on Mϑ such that β(v) ≡ 1,
and the closedness of β, we see that the leaves are all diffeomorphic to the
same abelian covering of Σg .
The above flow of v sends leaves to leaves, but of course it does not need
to preserve the complex structure of the leaves, that is it does not need
 A CHARACTERIZATION OF INOUE SURFACES 4

to realize a conformal diffeomorphism between the leaves. However, the

compactness of Mϑ implies, at least, that such a diffeomorphism is quasi-
conformal. In particular, all the leaves have the same (conformal) universal
covering: either they are all parabolic, or all hyperbolic. For our purposes,
it is sufficient to prove that the leaves are parabolic: since they are abelian
coverings of Σg , this implies that g = 1, and the rest of the statement is
standard [God, I.4].
We can associate to Fϑ a closed positive current Φ ∈ A1,1 (S)′ , by inte-
gration along the leaves against the transverse measure defined by β [Ghy]:
if η ∈ A1,1 (S), we define
Z
Φ(η) = β ∧ η.
Mϑ
Obviously this current does not charge compact complex curves, hence by
results of Lamari [Lam] [Tom, Rem. 8] it is an exact positive current. As a
consequence of this, its De Rham cohomology class [Φ] has vanishing product
with the Chern class of TF :
c1 (TF ) · [Φ] = 0.
Let us show that this implies the parabolicity of the leaves (this is a par-
ticularly simple instance of the foliated Gauss-Bonnet theorem, see [Ghy]).
In the opposite case, we may put on the leaves of Fϑ their Poincaré metric,
which can be seen as a hermitian metric on TF |Mϑ . It is a continuous metric
[Ghy], and it can be regularized by a smooth hermitian metric on TF |Mϑ
whose curvature along the leaves is still strictly negative (for instance, with
the help of the above vector field v). We then extend this hermitian metric
to the full TF , on the full S, in any way. The curvature form Θ ∈ A1,1 (S)
clearly satisfies Z
Φ(Θ) = β ∧ Θ < 0.
Mϑ
This is in contradiction with the vanishing of c1 (TF ) · [Φ]. 
Remark 2.2. Let us stress a subtle detail of the previous proof. The current
Φ can be also considered as a current on the real threefold Mϑ . As such,
however, it is not exact. Thus, in order to get the vanishing of c1 (TF )·[Φ], we
used also the fact that the tangent bundle TFϑ extends to the full S, or more
precisely that its Chern class in H 2 (Mϑ , R) extends to S, which is obvious
in our case since we have a global foliation on S. Now, one can imagine a
more general situation, in which we have a Levi-flat hypersurface M in a
class VII◦ surface, such that the Levi foliation is given by a closed 1-form
(or, more generally, admits a transverse measure invariant by holonomy). Is
it still true that the leaves of this Levi foliation are parabolic?

3. The structure of the singularities

In order to study Sing(F) and Z(ω), we need a general lemma on critical
points.
 A CHARACTERIZATION OF INOUE SURFACES 5

Let U e be a smooth complex surface and let D ⊂ U e be a compact con-

nected curve (with possibly several irreducible components). Suppose that
the intersection form on D is negative definite, so that D is contractible to
one point [BPV, p. 72]. After contraction, we get a normal surface U and
a point q ∈ U , image of D; we do not exclude that q is a smooth point. Let
now H e be a holomorphic function on U e , vanishing on D, such that
e = D.
Crit(H)
After contraction, we thus get a holomorphic function H on U with (at
most) an isolated critical point at q. If B is a small ball centered at q, then
H0 = {H = 0} ∩ B is a collection of k discs H01 , . . . , H0k passing through q,
whereas Hε = {H = ε} ∩ B (ε small and not zero) is a connected curve with
k boundary components. The topological type of Hε does not depend on ε
(small and not zero), it is the so-called Milnor fiber of H at q.
Lemma 3.1. Under the previous notation, suppose that the genus of the
Milnor fiber of H at q is zero. Then:
(1) q is a smooth point of U ;
(2) either q is a regular point for H, or it is a critical point of Morse
type.
Proof. The hypothesis means that the Milnor fiber is a sphere with k holes.
By a standard construction, we may glue to W = ∪|ε|<r Hε (r > 0 small)
a collection of k bidiscs in such a way that we obtain a normal complex
surface V and a proper holomorphic map
G : V −→ D(r)
such that:
(i) W ⊂ V and G|W = H;
(ii) Gε = G−1 (ε) is a smooth rational curve for every nonzero ε ∈ D(r);
(iii) G0 = G−1 (0) is a collection of k rational curves G10 , . . . , Gk0 passing
through q, with Gj0 ∩ W = H0j for every j.
Remark that all the components Gj0 of G0 have multiplicity 1, i.e. G
vanishes along Gj0 \ {q} at first order only. On the other hand, we may
blow-up q to the original D, and we get a smooth complex surface Ve and a
map
Ge : Ve −→ D(r)
b1 ∪ . . . ∪ G
whose fiber over 0 is G bk ∪ D, with G bj the strict transform of Gj .
0 0 0 0
By construction, we have
mult(G bj ) = 1
0
for every j and
mult(C) ≥ 2
e = D.
for every irreducible component C of D, since Crit(H)
 A CHARACTERIZATION OF INOUE SURFACES 6

Recall now [BPV, p. 142] that such a Ve can be also blow-down to the
e is sent
trivial fibration D(r)×CP 1 , in such a way that the singular fiber of G
1
to the regular fiber {0}×CP . In other words, that singular fiber is obtained
from a regular fiber by a sequence of monoidal transformations. It is then
easy to see that D necessarily contains a (−1)-curve: the reason is that a
monoidal transformation at a point belonging to an irreducible component
of multiplicity m creates a new irreducible component whose multiplicity
will be not less than m. By iterating this principle, we see that D contracts
to a regular point, whence the first part of the lemma.
Moreover, after this contraction the singular fiber becomes a curve (the
fiber G0 in the now smooth surface V ) still dominating a regular fiber,
hence in particular it has only normal crossings. Since all the components
of G0 pass through q, we get k = 1 (G0 is a single smooth rational curve of
selfintersection 0) or k = 2 (G0 is a pair of two smooth rational curves of
selfintersection −1). In the first case q is a regular point, and in the second
case it is a Morse type critical point. 
Remark 3.2. If H : C2 → C has an isolated critical point whose Milnor
fiber has genus zero, the the critical point is of Morse type: it is a particular
case of the previous lemma, but it is also a consequence of classical formulae
estimating the genus of the Milnor fiber. However, some care is needed when
C2 is replaced by a singular surface. For instance, take the function zw on C2
and quotient by the involution (z, w) 7→ (−z, −w). We get a normal surface
U and a holomorphic function H on U with an isolated critical point whose
Milnor fiber has genus zero. This kind of examples (and more complicated
ones) do not appear in Lemma 3.1 because, when we take the resolution
Ue → U , the critical set of H
e is not the full exceptional divisor D.
We can now return to our compact complex surface S.
Proposition 3.3. The zero set Z(ω) is composed only by isolated points, all
of Morse type. In particular, the normal bundle NF coincides with the flat
line bundle L.
Proof. Let D be a connected component of Z(ω). If it is a curve, then it is a
tree of rational curves with negative definite intersection form: this follows
from results of Nakamura on the possible configurations of curves on VII◦
surfaces [Nak], and the absence of elliptic curves and cycles of rational curves
[Tom] [C-T]. In particular, D is simply connected, and so the (twisted)
closed 1-form ω is exact on a neighbourhood U e of D: ω = dH e and Crit(H)e =
D. We therefore are in the setting of Lemma 3.1, and we have just to verify
the genus zero hypothesis.
Now, D is contained in a singular fiber Mϑ0 , which can be approximated
by regular ones, on which we already know that the foliation has leaves C
or C∗ . It follows obviously that the Milnor fiber has genus zero, and so by
Lemma 3.1 the contraction of D produces a smooth point. But we are also
assuming since the beginning that S is minimal, hence such a contraction
 A CHARACTERIZATION OF INOUE SURFACES 7

cannot exist and so Z(ω) is composed only by isolated points. By a similar

argument, and again Lemma 3.1, all these points are of Morse type. 

4. The planar case

Let Cv(f ) denote the critical values of f : S → S1 , and let J be a con-
nected component of S1 \ Cv(f ). On f −1 (J), the foliation F is nonsingular,
and it is given by the transverse intersection of the Kernels of the closed
1-form df = dF/F and the integrable 1-form dc F/F . It follows that the
differentiable type of F|Mϑ does not depend on ϑ ∈ J. We shall say that J
is of type C (resp. type C∗ ) if the leaves of F on f −1 (J) are isomorphic to
C (resp. C∗ ), according with Proposition 2.1. In this section we shall prove
that the existence of a component J of type C implies that S is a Inoue
surface of type SM .
Let us firstly recast Proposition 3.3 in the context of uniformisation of
foliations [Br2]. Recall that the leaves of a singular foliation are not simply
the leaves outside the singular points: generally speaking, and in order to
get a workable definition, we need to compactify some separatrices, the so-
called vanishing ends [Br2, p. 732]. However, the presence of a vanishing
end implies the existence of a rational curve on the surface, invariant by
the foliation, and over which the tangent bundle of the foliation has strictly
positive degree [Br2, p. 733]. We claim that this cannot happen in our
context, and so the leaves of our F are just equal to the leaves of F|S\Sing(F ) .
Indeed, if C ⊂ S is a rational curve, invariant by F, then c1 (TF ) · C =
2 − Z(F, C), where Z(F, C) is the number of the singularities of F along C
[Br1]. On the other hand, since these singularities are all of Morse type (CS
residue equal to −1), we also have C 2 = −Z(F, C). Hence
c1 (TF ) · C = 2 + C 2 ≤ 0
since, by minimality, C 2 ≤ −2.
We shall also use the main result of [Br2], which says that, since S is not
a Hopf surface nor a Kato surface, the foliation F is uniformisable, i.e. does
not have vanishing cycles (the reader is invited to give a simple proof of this
result, in our very special context).
Proposition 4.1. Let J ⊂ S1 \ Cv(f ) be a component of type C. Then
J = S1 , and S is a Inoue surface of type SM .
Proof. Suppose, by contradiction, that J 6= S1 , and let Mϑ0 be a fiber in the
boundary of f −1 (J). Such a fiber must contain a singular point p ∈ Sing(F).
Let L be the leaf corresponding to one of the two separatrices of p, and let
γ ⊂ L be a cycle close to p and turning around it. This cycle, which has
no holonomy, can be slightly deformed to a cycle γ ′ ⊂ L′ , where L′ is a leaf
contained in Mϑ , ϑ ∈ J close to ϑ0 . But L′ is simply connected, hence γ ′ is
homotopic to zero in L′ and so γ is homotopic to zero in L, by the absence
of vanishing cycles.
 A CHARACTERIZATION OF INOUE SURFACES 8

It follows that L is isomorphic to C, and L ∪ {p} is a rational curve C

which contains a unique singularity of the foliation. By the formulae above,
C 2 = −1, contradicting the minimality of S.
Therefore J = S1 , and so S is a T3 -bundle over the circle, without singu-
larities. By [Ino], it is a Inoue surface of type SM . 
Remark 4.2. It is worth observing that, in our quite special context, the
proof of Inoue’s theorem can be highly simplified. Indeed, and by our pre-
vious results, on the universal covering Sb of S the foliation is given by a
submersion π : Sb → H (with Im(π) coming from the pluriharmonic function
e all of whose fibers are isomorphic to C. The key point is to prove
F on S)
that such a universal covering is a product:
Sb = H × C.
Indeed, once we know this fact, it remains to study the action of Γ = π1 (S)
on H × C. But we already know a lot of properties of such an action (for
instance, Γ is a semidirect product of Z and Z3 , which acts on the H-factor
in a special affine way, etc.), and using that knowledge it is easy to conclude
that S is a Inoue surface of type SM .
In order to prove that Sb is a product, it is sufficient to show that π
is a locally trivial fibration, i.e. that every z ∈ H has a neighbourhood
Uz such that π −1 (Uz ) = Uz × C. By a classical theorem of Nishino [Nis],
this is equivalent to show that Vz = π −1 (Uz ) is Stein. By an argument of
Ohsawa [Ohs], the Steinness of Vz follows from the existence of a smooth
(not holomorphic!) foliation H on Vz whose leaves are holomorphic sections
of π over Uz (i.e., Vz is trivialisable by a smooth foliation with holomorphic
leaves).
Now, in our case such a foliation H is easy to construct. On any fiber Mϑ
we can take a real analytic foliation by real curves transverse to F|Mϑ . By
complexifying, we get, on a neighbourhood of Mϑ , a real analytic foliation
by complex curves, transverse to F. Using the special form of F, it is easy
b as the required property (here Uz is an
to see that this foliation, lifted to S,
horizontal strip in H).

5. The cylindrical case

Assume now that every J ⊂ S1 \ Cv(f ) is of type C∗ .
Lemma 5.1. Every leaf of F is isomorphic to C∗ .
Proof. The same argument used in Proposition 4.1 shows that, if L is a leaf
in a singular fiber Mϑ0 corresponding to a separatrix of some singular point
p, then L must be diffeomorphic to the cylinder R × S1 . Indeed, L cannot be
simply connected by the minimality of S, and cannot have a fundamental
group larger than Z by the absence of vanishing cycles (and of holonomy).
Remark now that, since the foliation is defined by a closed 1-form, every
leaf in Mϑ0 is either dense or properly embedded in the complement of the
 A CHARACTERIZATION OF INOUE SURFACES 9

singularities: this follows from the fact that the holonomy pseudogroup of
the foliation is composed only by translations on R. The second possibility
occurs only when the leaf is a “double separatrix”, i.e. a cylinder with both
ends converging to singular points, in which case the leaf is isomorphic to
C∗ and its closure is a rational curve of selfintersection −2.
Consider now an arbitrary dense leaf L′ in Mϑ0 . By the absence of van-
ishing cycles (and of holonomy), we get again that its fundamental group
cannot be larger than Z. On the other hand, this leaf accumulates to the
separatrix L, hence it cannot simply connected otherwise L would be simply
connected too. In conclusion, we see that every leaf in Mϑ0 is a cylinder.
To find the conformal type of the leaves, observe that an end of a leaf is
either convergent to a singular point (in which case it is obviously parabolic),
or it intersects a small ball B around a singular point p along infinitely
many annuli. These annuli are not homotopic to zero, by the previous
considerations. Moreover, we can extract among them infinitely many ones
with bounded modulus (i.e., all isomorphic to {r < |z| < 1} with r varying
in a compact subset of (0, 1)). It follows that the end is parabolic, and the
leaf is isomorphic to C∗ . 
Lemma 5.2. The line bundle TF⊗2 admits a continuous section on S \
Sing(F) which is nowhere vanishing.
Proof. The complex curve C∗ admits a “almost canonical” holomorphic vec-
∂
tor field: the vector field z ∂z , which can be almost uniquely characterized
as a complete holomorphic vector field whose flow is 2πi-periodic. There is
∂
however a minor ambiguity, since also the vector field −z ∂z (conjugate to
the previous one by the inversion z 7→ 1/z, which exchanges the two ends)
is complete and 2πi-periodic. This ambiguity can be removed when we take
∂ ⊗2 ∂ ⊗2
the square: (z ∂z ) = (−z ∂z ) . This means that, given any foliation F
with leaves isomorphic to C∗ , we get a canonical nonvanishing section of
TF⊗2 on S \ Sing(F), by the previous recipe. The point to be proved is that
such a section is (at least) continuous.
This is equivalent to prove the following. Let T ⊂ S be a local transversal
to F, isomorphic to a disc, and let VT be the corresponding holonomy tube
[Br2, p. 734]. It is a complex surface, equipped with a submersion QT :
VT → T , all of whose fibers are isomorphic to C∗ , and a section qT : T → VT .
For every t ∈ T we have a unique isomorphism it from Q−1 ∗
T (t) to C , sending
qT (t) to 1 (really, there is again a Z2 -ambiguity, which however can be easily
removed by prescribing an homotopy class). Therefore we get a trivialising
map
u : VT −→ T × C∗

u|Q−1 (t) = (t, it )

T

and the continuity of the above canonical section of TF⊗2 is clearly equivalent
to the continuity of u (for every transversal T ).
 A CHARACTERIZATION OF INOUE SURFACES 10

As shown in [Ghy, p. 78] (see also [Nis, I.2]), the continuity of u readily
follows from Koebe’s Theorem. Let us recall the argument, for completeness.
Take a compact K ⊂ Q−1 −1
T (t0 ) and an exhaustion of QT (t0 ) by relatively
compact open subsets {Ωn }n∈N . By a standard argument (e.g. Royden’s
Lemma), each Ωn can be holomorphically deformed to the nearby fibers
Q−1T (t), t ∈ Un = a neighbourhood of t0 in T . Thus, the maps it , t ∈ Un ,
can be seen as all defined on the same domain Ωn . By Koebe’s Theorem,
the distorsion of it on K ⊂ Ωn is uniformly bounded by a constant which
tends to zero as n → ∞, since Q−1 T (t0 ) is parabolic. We get in this way that
it |K uniformly converge to it0 |K as t → t0 , and since K was arbitrary we
get the continuity of u. 

Remark that, a posteriori, the above map u will be even holomorphic, as

well as the canonical section of TF⊗2 .
It is now easy to complete the proof of Theorem 0.1.
Proposition 5.3. If every component of S1 \Cv(f ) is of type C∗ , then there
(+)
is only one component, equal to S1 , and S is a Inoue surface of type SN,p,q,r;t
(−)
or SN,p,q,r .

Proof. By Lemma 5.2, the line bundle TF⊗2 is topologically trivial, i.e. it
is flat. From Proposition 3.3 and KS−1 = NF ⊗ TF it follows that KS is
flat too, and so c21 (S) = 0. As explained at the beginning, this is the same
as c2 (S) = 0, the foliation is nonsingular, and S is a Inoue surface (of the
claimed type). 

As in the planar case, also in the cylindrical case we do not need the full
strength of Inoue’s theorem, since we can directly prove that a covering of
S is isomorphic to H × C∗ .

References
[BPV] W. Barth, C. Peters, A. Van de Ven, Compact complex surfaces, Ergebnisse der
Mathematik (3)4, Springer (1984)
[Br1] M. Brunella, Foliations on complex projective surfaces, in Dynamical systems. Part
II, Pubbl. Scuola Norm. Sup. Pisa (2003), 49–77
[Br2] M. Brunella, Nonuniformisable foliations on compact complex surfaces, Mosc. Math.
J. 9 (2009), 729-748
[C-T] I. Chiose, M. Toma, On compact complex surfaces of Kähler rank one, preprint
arXiv (2010)
[Ghy] É. Ghys, Laminations par surfaces de Riemann, in Dynamique et géométrie com-
plexes, Panor. et Synth. 8, Soc. Math. France (1999), 49–95
[God] C. Godbillon, Feuilletages. Études géométriques, Progress in Math. 98, Birkhäuser
(1991)
[H-L] R. Harvey, H. B. Lawson, An intrinsic characterization of Kähler manifolds, Invent.
Math. 74 (1983), 169-198
[Ino] M. Inoue, On surfaces of Class VII◦ , Invent. Math. 24 (1974), 269-310
[Lam] A. Lamari, Le cône kählérien d’une surface, J. Math. Pures Appl. 78 (1999), 249-263
 A CHARACTERIZATION OF INOUE SURFACES 11

[Nak] I. Nakamura, Towards classification of non-Kählerian complex surfaces, Sugaku

Expositions 2 (1989), 209–229
[Nis] T. Nishino, Nouvelles recherches sur les fonctions entières de plusieurs variables
complexes. II. Fonctions entières qui se réduisent à celles d’une variable, J. Math.
Kyoto Univ. 9 (1969), 221-274
[Ohs] T. Ohsawa, A note on the variation of Riemann surfaces, Nagoya Math. J. 142
(1996), 1-4
[Tom] M. Toma, On the Kähler rank of compact complex surfaces, Bull. Soc. Math. France
136 (2008), 243-260

Marco Brunella, Institut de Mathématiques de Bourgogne – UMR 5584 –

9 Avenue Savary, 21078 Dijon, France