j.

differential geometry
74 (2006) 69-92

ONE-SIDED COMPLETE STABLE MINIMAL

SURFACES

Antonio Ros

Abstract
We prove that there are no complete one-sided stable minimal
surfaces in the Euclidean 3-space. We classify least area surfaces in
the quotient of R3 by one or two linearly independent translations
and we give sharp upper bounds of the genus of compact two-
sided index one minimal surfaces in non-negatively curved ambient
spaces. Finally we estimate from below the index of complete
minimal surfaces in flat spaces in terms of the topology of the
surface.

1. Introduction
Stable minimal surfaces in 3-manifolds minimize area up to second
order for any compactly supported smooth deformation. They play
a fundamental role in minimal surface theory. Do Carmo and Peng
[6], Fischer-Colbrie and Schoen [11] and Pogorelov [26], independently,
proved that complete stable two-sided minimal surfaces immersed in the
euclidean 3-space are flat. The results of [11] apply to 3-manifolds with
nonnegative scalar curvature. In these ambient spaces, Fischer-Colbrie
[10] proved that two-sided complete stable minimal surfaces with com-
pact boundary have finite finite total curvature. Recently Meeks and
Rosenberg [19] classified two-sided stable complete minimal surfaces in
3-manifolds of the type M × R, M being a closed Riemannian surface.
In the one-sided case the stability condition (and the area minimiza-
tion problem in general) becomes more subtle. This case should not be
considered as a minor or secondary one. On the contrary, it presents
a rich and interesting geometry, even in flat or positively curved 3-
manifolds, which is far from being understood. For instance, while
two-sided stable complete minimal surfaces in flat 3-manifolds must be
planar, Ross [33] has proved that the classical P and D Schwarz mini-
mal surfaces are stable when considered as nonorientable surfaces with
total curvature −4π in suitable 3-tori. As another example we point out

Partially supported by MCYT-FEDER research project MTM2004-02746.

Received 11/16/2004.

69
70 A. ROS

that it is a trivial fact that there are no two-sided compact stable mini-
mal surfaces in positively curved 3-manifolds. However, for any positive
integer n, there are lens spaces (metric quotients of the round 3-sphere)
which admit one-sided closed surfaces with Euler characteristic smaller
than −n minimizing area in its isotopy class, see [35].
In this paper we study one-sided complete stable minimal surfaces
(and related ones) in flat (or nonnegatively curved) 3-manifolds. When
the ambient space is euclidean, we will prove that
there are no nonorientable complete stable minimal surfaces
in R3 .
This theorem was previously known only in the finite genus case, Ross
[34] (see also [15]). As an important consequence, we extend to one-
sided surfaces the Schoen’s curvature estimate for stable minimal sur-
faces in 3-manifolds, [37].
We also consider quotients R3 /Γ of Euclidean space, where Γ is a
discrete group of translations of R3 . We prove that if rank (Γ) = 1
(resp. rank (Γ) = 2), then the only complete stable minimal surfaces in
R3 /Γ are planar surfaces and the Helicoid (resp. Scherk doubly periodic
surfaces) with total curvature −2π. This result was proved by Ross
and Schoen [36] assuming the surface has finite genus, see also [3, 27].
The topology of the Helicoid and the Scherk surfaces above is the one
of the projective plane once or twice punctured, respectively. In the
embedded case, the second surface exists only when the planar lattice
Γ is rectangular.
An important particular case of stable surfaces are area minimizing
surfaces in various natural classes of surfaces (integer homology, Z2 -
homology, isotopy, homotopy, etc). It is well-known that the plane is
the only complete area minimizing surface in R3 . The classification
of least area surfaces in general flat 3-manifolds is an interesting open
problem in classical minimal surface theory. The two-sided case gives
necessarily planar surfaces, but the list of one-sided area minimizing
surfaces in flat 3-manifolds is not known at the present. The result
below contains the first progress on this problem.
The Helicoid and the Scherk surfaces above are the only prop-
erly embedded nonflat area minimizing surfaces mod 2 in the
quotients of R3 by one or two linearly independent transla-
tions, respectively.
When the rank of Γ is 3 we are able to determine exactly the topology of
stable surfaces. We prove that closed stable minimal surfaces in flat 3-
tori are either planar 2-tori or Klein bottles with a handle. Planar 2-tori
are all stable. In fact they are (the unique) area minimizing surfaces in
their integer homology class. Klein bottles with a handle are one-sided,
their total curvature is −4π and their orientable two-sheeted covering
 ONE-SIDED COMPLETE STABLE MINIMAL SURFACES 71

is a genus 3 surface. Meeks has parameterized these surfaces as a 5-

dimensional family, see [17]. The moduli spaces of stable and unstable
minimal Klein bottles with a handle are, both, noncompact, see [34].
We do not have, at the moment, any bound on the topology of one-sided
stable minimal surfaces in flat 3-manifolds obtained by discrete groups
other than translations groups.

Besides stable minimal surfaces, minimal surfaces with index one have
received a large amount of attention, see the examples constructed in
[4, 24, 25]. If the ambient 3-manifold has nonnegative curvature, then
we can control the topology of these surfaces. Partial results of this
type have been proved by various authors [7, 16, 20, 28, 29, 41]. In
this paper we will prove the right sharp estimate for the first time (we
remark that the P Schwarz minimal surface of genus 3 has index one in
the cubic 3-torus, [33]):

Closed index one orientable minimal surfaces immersed in

an orientable 3-manifold with nonnegative Ricci curvature
have genus ≤ 3.

Finally, we give estimates of the index of a complete minimal surface

in term of its topology and we get a sharp estimate for the genus of stable
constant mean curvature surfaces in nonnegatively curved 3-manifolds
(this family of surfaces includes the solutions of the isoperimetric prob-
lem): improving previous results, we obtain that compact orientable
stable constant mean curvature surfaces, immersed in a 3-manifold with
nonnegative Ricci curvature, have genus ≤ 3.

We conclude this introduction with some comments about the strat-

egy of the proofs of our results. If we are considering stability problems
on minimal surfaces, or upper bounds for the eigenvalues of Schrödinger
operators on surfaces, we need to dispose of simple test functions (or
more generally test sections) with controlled energy. Among these we
have the ones that rely on the complex or algebraic geometry of the
surface. In particular, meromorphic maps (i.e., conformal maps be-
tween the surface and the 2-sphere) were used by Yang and Yau [39]
to obtain an upper bound of the first eigenvalue of the Laplacian on
a compact surface in terms of the area and the genus of the surface
(generalizing an earlier theorem of Hersch [13] for the genus zero case)
and then by several authors in other related contexts, see for instance
[7, 16, 20, 28, 29, 41]. Other natural test objects which have been
considered in various problems are harmonic forms, see [23, 38], and
more generally holomorphic sections in complex bundles, see for instance
[21, 8]. In this paper we will use test functions constructed using both
meromorphic maps and harmonic forms.
72 A. ROS

2. Preliminaries
Let Σ be a connected surface immersed in a Riemannian 3-manifold
M . If a unit normal vector field N is globally defined on Σ, we say
that the immersion is two-sided. Otherwise the immersion is said to be
one-sided. When the ambient manifold M is orientable two-sidedness is
equivalent to the orientability of Σ.
If Σ is a two-sided minimal surface, then any compactly supported
function u ∈ C 1 (Σ) determines a normal deformation of the surface with
infinitesimal variation uN . The second variation formula of the area for
this deformation is given by
Z
(1) Q(u, u) = |∇u|2 − (Ric (N ) + |σ|2 )u2 ,
Σ

where σ is the second fundamental form of the immersion, Ric (v) is the
Ricci curvature of M in the direction of the vector v and N the unit
normal vector along Σ. The associated linear operator ∆+Ric (N )+|σ|2 ,
∆ Rbeing the Laplacian of Σ, is called the Jacobi operator and satisfies
− Σ u(∆v + (Ric (N ) + |σ|2 )v) = Q(u, v). Using the Gauss equation we
get

(2) Ric (N ) + |σ|2 = Ric (e1 ) + Ric (e2 ) − 2K,

where e1 , e2 is orthonormal basis in the tangent plane of the surface and

K is its Gauss curvature. The surface Σ is stable if

(3) 0 ≤ Q(u, u), for all compactly supported u ∈ C 1 (Σ).

If Σ is compact stable and Ric ≥ 0, then the immersion is totally

geodesic. It follows from Fischer-Colbrie and Schoen [11] that complete
two-sided stable minimal surfaces in flat 3-manifolds are planar.
If the minimal surface Σ is one-sided, then the normal bundle is non-
trivial and the stability condition is expressed in a different form. It is
customary to work on the two-sheeted two-sided covering Σ e → Σ. If τ
denotes the change of sheets involution and N is the unit normal vector
e then N ◦ τ = −N and infinitesimal normal deformations of Σ
on Σ,
correspond to functions u on Σ e satisfying u ◦ τ = −u. Therefore, if Q
is the second variation formula of the area on Σ,e then the immersion of
Σ is stable if and only if
e with compact support and u ◦ τ = −u.
(4) 0 ≤ Q(u, u), ∀ u ∈ C 1 (Σ)

In the rest of this section we explain some properties of meromorphic

maps and harmonic 1-forms defined on a Riemann surface. These ob-
jects will be used as test functions to deduce geometric consequences
from stability and related hypothesis.
 ONE-SIDED COMPLETE STABLE MINIMAL SURFACES 73

2.1. Harmonic forms. Let Σ be an orientable Riemannian surface

and H 1 (Σ, R) the space of harmonic 1-forms. Recall that a 1-form ω on
Σ is harmonic if and only if it is closed, i.e., (∇ω)(x, y) = (∇ω)(y, x)
for all tangent vectors x and y (∇ω being the covariant derivative of
ω), and coclosed, that is ∇ω(e1 , e1 ) + ∇ω(e2 , e2 ) = 0 (e1 , e2 being an
orthonormal base in the tangent plane of Σ). Let us denote by K and
∆ the Gauss curvature and the rough Laplacian of Σ, respectively. In
particular, ∆ acts on 1-forms as ∆ω(x) = ∇2 ω(e1 , e1 , x)+∇2 ω(e2 , e2 , x).
It is well-known that if ω is harmonic, then
(5) ∆ω − Kω = 0.
Therefore, given ω, ω ′ ∈ H 1 (Σ, R), we obtain
(6) ∆hω, ω ′ i − 2Khω, ω ′ i = 2h∇ω, ∇ω ′ i.
We also recall that, in the case that Σ is compact, then ω is harmonic
if and only if it satisfies equation (5). Moreover, if Σ has genus g, then
the space of harmonic 1-forms H 1 (Σ, R) is 2g-dimensional.
If Σ is an orientable nonplanar minimal surface immersed in R3 , then
the differentials of the linear functions of the immersion dx1 , dx2 , dx3
are harmonic and generate a 3-dimensional subspace L(Σ) of H 1 (Σ, R).
The conjugate harmonic functions generate another 3-dimensional sub-
space L∗ (Σ) = span{dx∗1 , dx∗2 , dx∗3 } ⊂ H 1 (Σ, R). If we denote by N =
(N1 , N2 , N3 ) and σ the unit normal vector and the second fundamental
form of the immersion, respectively, we have that
(7) (∇dxj )(x, y) = σ(x, y)Nj and (∇dx∗j )(x, y) = σ(Jx, y)Nj ,
J being the complex structure of Σ and j = 1, 2, 3. Given ω ∈ H 1 (Σ, R),
we consider the dual tangent vector field (viewed as a R3 -valued func-
tion) Xω : Σ → R3 , defined by
(8) Xω = (hω, dx1 i, hω, dx2 i, hω, dx3 i).
We will need the following basic properties of the second fundamental
form:
i) σ is a trace zero symmetric tensor.
ii) ∇σ is symmetric in all its three arguments (Codazzi equation).
iii) σ(Jx, y) = σ(x, Jy), and so, σ ∗ (x, y) = σ(Jx, y) is also a trace zero
symmetric
P tensor. Moreover σ and σ ∗ are (pointwise) orthogonal.
iv) 2
i σ(x, ei )σ(ei , y) = −Khx, yi and |σ| = −2K (Gauss equation).

Lemma 1. Let Σ be an orientable nonflat minimal surface immersed

in R3 and ω a harmonic 1-form on Σ. Then
(9) ∆Xω + |σ|2 Xω = 2h∇ω, σiN.
Moreover h∇ω, σi ≡ 0 if and only if ω ∈ L∗ (Σ).
74 A. ROS

Proof. Using (6), (7) and iv), we obtain (9).

As planar points are isolated, to prove the last assertion of the lemma
we can assume that σ 6= 0 at every point of Σ. Clearly, the covariant
derivative of any harmonic 1-form is a linear combination of σ and σ ∗ .
Thus, h∇ω, σi ≡ 0 implies that there is a function h on Σ such that

(10) ∇ω(x, y) = hσ(Jx, y).

By taking covariant derivative with respect to a tangent vector z, we

get
∇2 ω(z, x, y) = ∇h(z) σ(Jx, y) + h∇σ(z, Jx, y).
From the curvature identity of Ricci, using ii) and iii), we obtain

−ω (R(z, x)y) = ∇2 ω(z, x, y) − ∇2 ω(x, z, y)

= ∇h(z) σ(Jx, y) − ∇h(x) σ(Jz, y),

where R(z, x)y = K(hx, yiz − hz, yix) is the curvature operator of the
surface. The contraction x = y = ei transforms the above identity into
X
Kω(z) = ∇h(ei )σ(Jz, ei ),
i
P
and, if we choose z = σ(Jek , v)ek from iv), we have
X X
K ω(ek )σ(Jek , v) = ∇h(ei )σ(Jek , ei )σ(Jek , v) = −K∇h(v).
k ik

In conclusion, we have shown that the pair (ω, h) is a solution of the

first order total linear differential equation

(11) ∇ω(x, y) = hσ(Jx, y) ∇h(x) = −σ(JXω , x).

Note that, as each solution is determined by its value at a point p ∈ Σ,

the set of solutions of (11) is a vector space of dimension at most three.
On the other hand, for each vector a ∈ R3 we have the explicit solution
of
ω(x) = hJx, ai h = hN, ai Xω = N ∧ a.
Therefore {ω ∈ H 1 (Σ, R) | h∇ω, σi = 0} = L∗ (Σ) and the lemma is
proved. q.e.d.

Palmer [23] used the tangent vectors fields Xω to study the index, as
a harmonic map, of the Gauss map of a closed constant mean curvature
surface in R3 .
 ONE-SIDED COMPLETE STABLE MINIMAL SURFACES 75

2.2. Meromorphic maps. The following construction is very useful in

the study of stability and index problems for minimal and constant mean
curvature surfaces: let φ : Σ → S 2 (1) be a nonconstant meromorphic
map on a compact orientable Riemannian surface (viewed as a R3 -valued
map by using a conformal identification between C and the unit 2-
sphere). The quadratic form
Z
(12) Qφ (u, u) = |∇u|2 − |∇φ|2 u2 , u ∈ C 1 (Σ),
Σ
does not depend on the metric but only on its conformal class. In
particular, the kernel and the index (i.e., the number of negative eigen-
values) of the associated operator ∆ + |∇φ|2 are conformal invariants.
Henceforth, when considering these invariants we can work without be-
ing precise about the metric in the conformal class we have chosen. The
linear coordinates of φ lie in the kernel, that is, ∆φ + |∇φ|2 φ = 0. The
first eigenvalue of the operator is always negative. So the index one case
is the simplest possibility in this context. If ϕ1 is the first eigenfunction
of ∆ + |∇φ|2 the operator has index one if and only if
Z
1
Qφ (u, u) ≥ 0 for any u ∈ C (Σ) such that uϕ1 = 0.
Σ
Note that ϕ1 is positive. The next lemma was first proved by Hersch
[13] in case φ is the identity map and then adapted to more general
situations in [39, 14].
Lemma 2. Given a nonconstant meromorphic map φ : Σ → S 2 (1) ⊂
R3 ,Σ being a compact Riemannian surface, and a positive continuous
function ϕR on Σ, there exists a Moebius transformation t : S 2 −→ S 2
such that Σ (t ◦ φ)ϕ = 0.
Montiel and Ros [20] proved the following result, see also [16].
Lemma 3. Let φ : Σ → S 2 (1) be a nonconstant meromorphic map
from a compact Riemannian surface such that the operator ∆ + |∇φ|2
has index one. Then the degree of any other nonconstant meromorphic
map φ′ is larger that or equal to the degree of φ, and the equality holds
if and only if there is a Moebius transformation t : S 2 −→ S 2 such that
φ′ = t ◦ φ.
This lemma, combined with known results from Algebraic Geometry,
allows us to conclude the following theorem; compare with [20, 28].
Theorem 4. Let Σ be a compact Riemann surface of genus g and
φ : Σ → S 2 (1) a nonconstant meromorphic map of degree d. Assume
the operator ∆ + |∇φ|2 has index one. Then
hg i
d ≤ + 1,
2
where [z] denotes the integer part of z.
76 A. ROS

Proof. The theorem follows directly from Lemma 3 and the properties
below. From the Brill-Noether theory, see [1], we have that any closed
Riemann surface admits a nonconstant meromorphic map of degree d ≤
[ g+1
2 ] + 1. That proves the theorem when g is even. If g is odd and d =
g+1
[ 2 ] + 1, then the Brill-Noether theorem asserts that the meromorphic
map is not unique up to Moebius transformations. So, in the case g is
odd we have d ≤ [ g+1
2 ]. q.e.d.

Particularizing the theorem to small genus cases we get that, if

∆ + |∇φ|2 has index one, then
g = 0 ⇒ d = 1,
g 6= 1,
g = 2, 3 ⇒ d = 2,
g = 4, 5 ⇒ d ≤ 3.
For nonorientable minimal surfaces, stability can be also reformu-
lated in terms of meromorphic maps as follow: besides the orientable
Riemannian surface Σ and the meromorphic map φ, we consider a fixed
point free anticonformal involution τ on Σ such that φ ◦ τ = −φ (the
nonorientable surface here is Σ/τ ). The relevant condition to be con-
sidered now is
(13) Qφ (u, u) ≥ 0 for all u ∈ C 1 (Σ) such that u ◦ τ = −u.
The following lemma has been used several times in the study of the
stability of nonorientable minimal surfaces. It was obtained by Ross
[34], who adapted the arguments of [16].
Lemma 5. Let Σ be a compact orientable Riemannian surface, τ
an anticonformal involution without fixed points and φ : Σ → S 2 (1) a
meromorphic map satisfying φ ◦ τ = −φ such that the quadratic form
Qφ verifies (13). Then any meromorphic map φ′ on Σ with φ′ ◦ τ = −φ′
satisfies degree (φ′ ) ≥ degree (φ) and the equality holds if and only if φ
and φ′ differ in a Moebius transformation.
Particularizing to the case genus ≤ 1, we conclude
Theorem 6 ([34]). In the hypothesis of the Lemma:
a) If genus (Σ) = 0, then the property (13) implies degree (φ) = 1.
b) If genus (Σ) = 1, then (13) does not hold.

Proof. In the genus zero case we can assume that τ is the antipodal
map. So we can take φ′ as the identity and we get a). To prove b)
we use the conformal automorphisms of the torus to find meromorphic
maps φi , i = 1, 2, of degree two satisfying φi ◦ τ = −φi and which are
not related by a conformal transformation of the sphere. q.e.d.
 ONE-SIDED COMPLETE STABLE MINIMAL SURFACES 77

3. Stability of one-sided minimal surfaces

If we have a nonorientable Riemannian surface Σ, we can consider
e → Σ and the change of sheets in-
the 2-sheeted orientable covering Σ
volution τ : Σe → Σ. e The space of harmonic 1-forms H 1 (Σ,
e R) de-
1 e + e − e
composes into two subspaces H (Σ, R) = H (Σ) ⊕ H (Σ) defined by
e = {ω | τ ∗ (ω) = ±ω}, and the Hodge star operator of Σ
H ± (Σ) e trans-
+ − e
forms H into H . Hence, when Σ is a compact surface of genus g,
then we get that dim H + = dim H − = g.
e ⊂ H + (Σ)
Note that, if Σ is minimally immersed in R3 , then L(Σ) e
∗
and L (Σ)e ⊂ H (Σ).
− e

Theorem 7. Let Σ be a nonorientable compact stable minimal sur-

face immersed in a flat three torus T 3 . Then Σ has the topology of a
Klein bottle with a handle.
Proof. As L(Σ)e ⊂ H + (Σ)e and the surface is nonflat, we deduce that
g ≥ 3 (this fact can be also proved by looking at the Gauss map of Σ, e
see Meeks [17]).
e then Xω ◦ τ = −Xω (in the same way, ω ∈ H + ⇒
If ω ∈ H − (Σ)
Xω ◦ τ = Xω ). Hence, we can use (the linear coordinates of) X = Xω
as a test function in the stability inequality (4), and we conclude that
X3 Z
0≤ Q(Xj , Xj ) = Q(X, X) = − h∆X + |σ|2 X, Xi
e
Σ
j=1
Z
= −2 h∇ω, σihX, N i = 0,
e
Σ
where the last equality follows because X is tangent and N is normal
to the surface. This implies that X lies in the kernel of Q, that is
∆X + |σ|2 X = 0. From Lemma 1 we deduce that ω ∈ L∗ (Σ) e and,
therefore, g = 3. So it follows that Σ is a Klein bottle with a handle
and that proves the theorem. q.e.d.

Minimal Klein bottles with a handle immersed in a flat 3-torus x :

Σ → T have total curvature −4π and they have been considered by
Meeks in [17]. He proves that there exists a minimal embedding x′ :
Σ → T ′ in some three torus and a covering map π : T ′ → T such that
x = π ◦ x′ . Moreover the conjugate minimal surface defines a second
isometric minimal embedding of Σ in another 3-torus T ′′ .
The following basic fact will be used in the proof of the main theorem
below.
Lemma 8. Let Σ be a nonorientable surface and Σ e → Σ the two-
e
sheeted orientable covering. If Σ has finite genus, then there is a com-
pact set F ⊂ Σ such that Σ − F is orientable.
78 A. ROS

Proof. Denote by τ the deck transformation of Σ e given by change of

sheet. As Σ is nonorientable, there exists a Jordan curve γ1 in Σ such
that its pullback image γ e is connected and its complement Σ
e1 ⊂ Σ e −γ e1
has one or two connected components. Moreover, in the second case τ
permutes these components and so Σ − γ1 is orientable. If Σ − γ1 is
nonorientable, then we can choose a second Jordan curve γ2 ⊂ Σ − γ1
with connected pullback image γ e2 disjoint with γ e1 , such that either
e − (e
Σ γ1 ∪ γe2 ) is connected or Σ − (γ1 ∪ γ2 ) is orientable. If Σ − (γ1 ∪ γ2 )
is nonorientable, we apply the argument again. As Σ e has finite genus,
after a finite number of times we get pairwise disjoint Jordan curves
γ1 , . . . , γg in Σ such that Σ − (γ1 ∪ · · · ∪ γg ) is orientable, as we claimed.
q.e.d.

Theorem 9. Let Σ be a complete stable minimal surface in R3 . Then

Σ is a plane.

Proof. In the orientable case the theorem was proved by Do Carmo

and Peng [6], Fisher-Colbrie and Schoen [11] and Pogorelov [26], in-
dependently. So we assume that Σ is non orientable. In this case, the
result has been proved by Ross [34], under the aditional assumption
that the surface has finite total curvature. Consider the 2-sheeted ori-
entable covering Σe → Σ and the change of sheets involution τ . We
discuss two situations separately.

Case 1: genus (Σ) e ≥ 2 (including the case genus = ∞). From

the classic theory of Riemann surfaces, see [9] p. 42, we know that
dim H 1 (Σ,e R) ∩ L2 (Σ)
e ≥ 4: in fact, each nonseparating Jordan curve
e
γ ⊂ Σ produces, in a canonical way, a square integrable harmonic 1-
e
form ωγ which is exact on Σ−γ and such that the period of ωγ along the
′ ′
curve γ is equal to 1, γ being a closed curve intersecting γ transversally
just at one point. Thus, if we take two disjoint Jordan curves γ and
η in Σe such that Σ e − γ ∪ η is connected, then the harmonic 1-forms
∗
ωγ , ωη , ωγ and ωη are linearly independent (here ω ∗ denotes the image
∗

of ω by the Hodge star operador of Σ). e In particular we deduce that

− e 2 e
dim H (Σ) ∩ L (Σ) ≥ 2.
Take ω ∈ H − (Σ) e ∩ L2 (Σ)
e and X = Xω . Take also compactly sup-
ported smooth maps Y : Σ e → R3 and ϕ : Σ e → R with Y ◦ τ = −Y ,
ϕ ◦ τ = ϕ where ϕ is a cutoff function satisfying 0 ≤ ϕ ≤ 1, ϕ ≡ 1 on a
neighborhood of the support of Y and |∇ϕ| ≤ 1. Using Xt = ϕX +tY =
ϕ(X + tY ), t ∈ R, as test function for the stability condition we have
Q(Xt , Xt ) ≥ 0, for each t. As the above is a quadratic expression in t
and Q(X, Y ) = Q(ϕX, Y ), this is equivalent to say that

(14) Q(X, Y )2 ≤ Q(Y, Y )Q(ϕX, ϕX).

 ONE-SIDED COMPLETE STABLE MINIMAL SURFACES 79

From equation (9), after several integration by parts, we obtain

Z
Q(ϕX, ϕX) = |∇(Xϕ)|2 − |σ|2 |Xϕ|2
e
ZΣ
= −h∆(Xϕ), Xϕi − |σ|2 |Xϕ|2
e
ZΣ
= −h∆X + |σ|2 X, Xiϕ2 − ϕ∆ϕ|X|2 − 2h∇ϕX, ϕ∇Xi
e
Σ
Z
1
= −ϕ∆ϕ|X|2 − h∇ϕ2 , ∇|X|2 i
e 2
ZΣ
= −ϕ∆ϕ|X|2 + ϕ∆ϕ|X|2 + |∇ϕ|2 |X|2
ZΣ
= |∇ϕ|2 |X|2 .
e
Σ

e the last integral converges to zero when we take the

As |X| ∈ L2 (Σ),
region {ϕ = 1} larger and larger and so, we obtain Q(X, Y ) = 0 using
(14). Therefore, from (9) we deduce
Z Z
−2 h∇ω, σihN, Y i = −h∆X + |σ|2 X, Y i = Q(X, Y ) = 0
e
Σ e
Σ

e ∩ L2 (Σ).
for all anti-invariant Y . Thus h∇ω, σi ≡ 0 and then ω ∈ L∗ (Σ) e
This argument gives us two linear functions of the immersion Σ e → R3 ,
say x1 and x2 , with finite energy, and so
Z Z
1≤ |∇x1 |2 + |∇x2 |2 < ∞,
e
Σ e
Σ

which contradicts the well-known fact that the area of a complete min-
imal surface in R3 is infinite, see the final remarks in Yau [40].
e ≤ 1. Now we follow Ross [34], see also [15]: Using
Case 2: genus (Σ)
Lemma 8 we have that Σ is orientable outside of a compact set. From
Fischer-Colbrie [10] we get that Σ has finite total curvature. Therefore
e is conformally equivalent to a finitely punctured compact Riemann
Σ
surface Σ of genus ≤ 1 and the Gauss map extends to a meromorphic
map N : Σ → S 2 (1). Moreover N ◦ τ = −N and the stability con-
dition can be rewritten in the compactified surface
R (using a metric on
Σ compatible with the conformal structure) as Σ |∇u|2 − |∇N |2 u2 ≥ 0
for any anti-invariant function u on Σ. As consequence of Lemma 5 we
conclude that genus(Σ) = 0 and degree(N ) = 1. Then it is well-known,
see [22], that the immersion of Σe gives either the Catenoid or the En-
neper surface, none of which factorizes through a nonorientable surface.
This contradiction finishes the proof of the theorem. q.e.d.
80 A. ROS

Using Theorem 9, we can extend the curvature estimate of Schoen

[37] to one-sided stable minimal surfaces.
Corollary 10. There is a constant c > 0 such that any stable mini-
mal surface (either one-sided or two-sided ) Σ immersed in R3 satisfies
the curvature estimate
|σ| ≤ c r−1 ,
where σ is the second fundamental form of the immersion and r is the
intrinsic distance to the boundary of Σ.
Proof. The corollary follows from the uniqueness of the plane among
complete stable minimal surfaces in R3 combined with a blow-up argu-
ment, see [5], [18] and [31]. In fact, assuming that the estimate does
not hold, we can produce after rescaling and suitable choosing of sub-
surfaces, a sequence of minimal surfaces Σn immersed in R3 and points
xn ∈ Σn such that, if we denote by |σn | and rn the length of second fun-
damental form and the distance to the boundary on Σn , respectively,
we have the following:
i) |σn |(xn ) = 1, for each n,
ii) rn (xn ) goes to infinity, and
iii) for each n, the function rn |σn | : Σn → R attains its maximum at
xn .
Passing to the limit we obtain a complete minimal surface immersed
in R3 which is stable and nonflat. This contradicts Theorem 9 and the
corollary is proved. q.e.d.

The above arguments can be adapted to give curvature estimates

for stable minimal surfaces in 3-manifolds M with injectivity radius
bounded from below and bounded curvatures and derivatives of curva-
tures. In particular, we have the following result.
Corollary 11. Let M be a complete 3-manifold with bounded ge-
ometry. Then there is a constant c > 0, depending on a lower bound
of the injectivity radius of M and bounds of the curvature of M and
its derivatives, such that any (either one-sided or two-sided ) complete
stable minimal surface Σ immersed in M satisfies |σ| ≤ c.
Now we consider the stability condition for one-sided complete mini-
mal surfaces in some quotients of R3 . In the two-sided case, this is the
same as the stability problem in the euclidean space itself (it is proved
in [11] that if a two-sided minimal surface is stable, then its universal
covering is stable, too). On the contrary, in the one-sided case differ-
ent quotient spaces present different kinds of complete stable minimal
surfaces.
 ONE-SIDED COMPLETE STABLE MINIMAL SURFACES 81

Theorem 12. Let f : Σ → M = R3 /Γ be a complete noncompact

and nonorientable minimal surface immersed in the quotient of the Eu-
clidean space by a discrete group of translations Γ. If Σ is stable, then
it is either (a quotient of ) the Helicoid or a (quotient of the) double
periodic Scherk’s surface. In the first case the surface is a Moebius strip
and in the second one a twice punctured projective plane. In both cases
the total curvature equals −2π.
Proof. We keep the notation of the proof of Theorem 9 and we discuss
the same cases as there.
The case genus (Σ) e ≥ 2 is shown to be impossible just as in the proof
above: although now the linear coordinates are not univalued, the fact
we need in the argument is only that the differential of the immersion
(dx1 , dx2 , dx3 ) is globally well-defined. So we get that Σ has finite area.
RAs Corollary 11 says that Σ has bounded curvature, we also have that
Σ |K| < ∞. These two finiteness conditions are clearly incompatible
for complete noncompact minimal surfaces and we conclude that this
case is impossible.
In the case genus (Σ) e ≤ 1, arguing as in the proof of Theorem 9 we get
that the genus is 0 and the immersion of Σ e has total curvature −4π. The
Cohn-Vossen inequality implies that Σ e has at most four ends. These
surfaces can be studied by looking at its Weierstrass representation.
They are parameterized by its Gauss map and, because of the existence
of the nonorientable quotient, we get that the ends consist of one or two
pairs of antipodal points. Under these assumptions we obtain either
the Helicoid or the doubly periodic Scherk surface or a surface in the
one parameter family of isometric surfaces associated to the two above.
Again the existence of the orientation reversing translation excludes all
the surfaces but the Helicoid and the doubly periodic Scherk surfaces,
itself. In fact, if ψd and ψs denote the double and the single periodic
Scherk surfaces respectively (these surfaces are multivalued and ψs is
the conjugate of ψd ), then the anticonformal involution τ induces a
translational period on ψd and a screw motion on ψs . The associated
surfaces are given by cos θ ψd + sin θ ψs and so the only way to have a
translation period on the associated surface is to take sin θ = 0. The
same argument applies to the Helicoid. q.e.d.

Let x0 : Σ → R3 /Γ0 be the standard embedding of the nonorientable

doubly periodic Scherk surfaces. Then Γ0 is an horizontal rectangular
planar lattice and the ends are asymptotic to the vertical planes contain-
ing the diagonals of the rectangle. The moduli of these surfaces coincides
with the one of the rectangles and so it comes in a 1-parameter family.
The immersions obtained in the Theorem are of the form x = π ◦ x0
where π is the linear projection of R3 /Γ0 onto R3 /Γ and Γ0 ⊂ Γ. Embed-
dedness holds only for the case Γ = Γ0 . In the same way the immersions
82 A. ROS

of the Helicoid consist of the standard imbedding of the nonorientable

Helicoid followed by a linear projection.

Now we consider properly embedded minimal surfaces Σ in a complete

3-manifold M which are area minimizing, mod 2. This is the same as
saying that each smooth compact subdomain Ω of Σ is area minimizing
among compact surfaces (both orientable and nonorientable) with the
same boundary. The plane is well known to be the only proper area
minimizing surface in R3 . As an interesting consequence of Theorem 12,
we obtain a complete classification of global area minimizing surfaces
(mod 2) in some flat 3-manifolds.

Theorem 13. Let Γ be a discrete group of translations of R3 with

rank (Γ) ≤ 2 and Σ ⊂ R3 /Γ a proper nonplanar area minimizing surface
mod 2.
1) If Γ is generated by a single translation, then Σ is a nonorientable
(quotient of the) Helicoid with total curvature −2π.
2) If rank (Γ) = 2, then Γ is generated by two orthogonal translations
and Σ is a nonorientable (quotient of the) doubly periodic Scherk
surface with total curvature −2π.

Proof. From Theorem 12 we only need to show that both the Helicoid
and the Scherk surface are area minimizing. Consider the compact flat
3-manifolds M = R3 /G, where G is the group generated by two linearly
independent horizontal translations v1 = (1, 0, 0) and v2 = (0, 0, a) and
the glide reflection (x, y, z) 7→ (−x, y + b, z), with a, b > 0 (this is the
group P c in international crystallographic notation). Note that M is
the metric product of a Klein bottle and a circle and that G contains
an index two lattice spanned by v1 , v2 and (0, 2b, 0).

Claim. There is a nonzero homology class α ∈ H2 (M, Z2 ) which

cannot be represented by an embedded planar surface.

Consider the fundamental region of the space M given in Figure 1.

The faces defined by the vertices 1, 4, 10, 7 and 3, 6, 12, 9 are identi-
fied by a translation which preserves the order of the vertices and the
same holds for the faces 1, 3, 6, 4 and 7, 9, 12, 10. The faces 1, 3, 9, 7 and
6, 4, 10, 12 are identified by a glide reflexion with mirror plane 2, 5, 11, 8
and translation vector the one connecting 2 and 5. The second homol-
ogy group of M is H2 (M, Z2 ) = Z2 ⊕Z2 ⊕Z2 and a basis of it is given by
α1 = [1, 3, 6, 4], α2 = [1, 3, 9, 7] and α3 = [3, 6, 12, 9] (note that the first
surface is a Klein bottle while the two other ones are tori). Other classes
that can be represented by planar surfaces are α1 + α2 = [7, 9, 6, 4] and
α2 + α3 = [2, 5, 11, 8] (these surfaces are a Klein bottle and a torus
respectively).
 ONE-SIDED COMPLETE STABLE MINIMAL SURFACES 83

Figure 1. Fundamental region of the Klein bottle times

circle.

Each (nonnecessarily connected) embedded planar surface in M is

either parallel or perpendicular to 3, 6, 12, 9. In the first case its homol-
ogy class lies in span{α2 , α3 }. In the second one the class of the surface
must be linear combination of α1 and α2 . Therefore the classes α1 + α3
and α1 + α2 + α3 do not contain any planar surface, as we claimed.
Let Σ ⊂ M be an area minimizing surface in the class α. Then Σ
is embedded, free of singularities and nonflat (in particular, Σ must
be one-sided). Up to a translation of the glide plane, we can assume
that Σ passes through the origin and that the maximum of the absolute
Gaussian curvature of Σ is attained at that point, |K|(0) = c2 . By
the maximum principle we can deduce that Σ has an embedded tubular
neighborhood in M or any radius r < 1/c, see [30]. The argument goes
as follows: the parallel surface at distance r, Σ(r), is a regular immersed
surface which defines a two sheeted covering of Σ. Moreover, it has
nonnegative mean curvature when we consider the outward pointing
unit normal vector (if the principal curvatures of Σ are (λ, −λ), then the
λ −λ
ones of Σ(r) are ( 1−rλ , 1+rλ )). For r > 0 small, Σ(r) is embedded and is
the boundary of the tubular neighborhood Ω(r) = {p ∈ M/dist(p, Σ) ≤
r}. If there exists the smaller r < 1/c such that Σ(r) has a double
point, then this surface has a selfcontact at its mean convex side, and
the maximum principle would imply that Σ(r) is minimal, which is
impossible. This proves the assertion.
As a consequence we obtain the estimate c ≥ 1: to see that, take p ∈ Σ
with normal vector v1 . From the existence of the embedded tubular
84 A. ROS

neighborhood, we get that the segment {p + tv1 | − 1/c < t < 1/c} is
embedded in M , which gives c ≥ 1. On the other hand, Corollary 11
implies that c is bounded from above independently of the parameters
a and b, if we assume a, b ≤ 1.
Now we take limits when a and b go to infinity. Up to a subsequence,
the least area surface Σ converges (with multiplicity one, because of the
existence of embedded tubular neighborhood) to a nonflat Z2 -area min-
imizing surface Σ∞ in the quotient manifold R3 /hv1 i. From Theorem
12, the only possibility for Σ∞ is the Helicoid with total curvature −2π.
To prove that the Scherk surfaces are area minimizing, we argue in
the same way fixing the parameter a and letting b going to infinity.
q.e.d.

The combination of Theorems 12 and 13 allows us to classify area

minimizing surfaces in any natural classes of surfaces (isotopy, homo-
topy, etc) in R3 /Γ, with rank(Γ) ≤ 2.

4. Index one minimal surfaces.

First we will improve Theorem 4 in the genus four case. This improve-
ment will allow us to give a sharp bound on the topology of two-sided
index one minimal surfaces in nonnegatively curved ambient spaces in
Theorem 15. Not surprisingly, the proof of Theorem 14 depends on an
argument involving minimal surfaces.
Theorem 14. Let Σ be a compact Riemann surface of genus 4 and
φ : Σ → S 2 (1) a nonconstant meromorphic map of degree d. If the
operator ∆ + |∇φ|2 has index one, then d = 2.
Proof. According to Theorem 4, it is enough to exclude the case
d = 3. Note that we cannot do that only by using Lemma 3, as there
are closed nonhyperelliptic Riemann surfaces of genus 4 which admit a
unique (up to Moebius transformations) meromorphic map of degree 3.
Reasoning by contradiction, assume that ∆ + |∇φ|2 has index one and
d = 3. From Lemma 3 we have that Σ is not hyperelliptic and that any
meromorphic map of degree 3 on Σ is related with φ by a conformal
transformation of the 2-sphere.
Claim. The meromorphic map φ : Σ → S 2 is the Gauss map of an
unbranched multivalued conformal minimal immersion x from Σ to R3
whose differential dx is globally well-defined on Σ.
The proof depends on some special properties of compact Riemann
surfaces of genus four which we present briefly; for complete proofs and
comments see [9, pp. 108-110]. As the space of holomorphic quadratic
differentials on Σ has dimension 9, if ω1 , . . . , ω4 is a basis of the space
of holomorphic differentials, then the 10 quadratic differentials ωi ωj ,
 ONE-SIDED COMPLETE STABLE MINIMAL SURFACES 85

1 ≤ i ≤ j ≤ 4,Pare linearly dependent. So, there are complex numbers

aij such that ij aij ωi ωj = 0. The rank of this quadratic form is 3 or
4 and by a change of basis this expression transforms into
(15) ω12 + ω22 + ω32 = 0 or ω12 + ω22 + ω32 + ω42 = 0.
If the rank is 4, then Σ admits two degree 3 meromorphic maps which are
not related by a Moebius transformation and, so, this case is discarded.
Then the rank of the quadratic expression is 3. Moreover, the unique
meromorphic map of degree √ 3 on Σ (up to Moebius transformations) is
given by g = ω3 /(ω1 − −1 ω2 ) : Σ → C.
If the assertion in the claim holds for a meromorphic map φ, then
it also holds for any meromorphic map related with φ by a Moebius
transformation, see for instance [20, 22]. So it is enough to √ prove the
claim for the meromorphic map g. If we consider ω = ω1 − −1 ω2 , we
can write
√
1 2 −1
ω1 = (1 − g ) ω, ω2 = (1 + g 2 ) ω, ω3 = g ω.
2 2
Note that if g has a pole of order m at a point, then the order of the
zero of ω at that point is at least 2m. As g has three poles and any
holomorphic differential on Σ has exactly 6 zeros, it follows that the
zeros of ω coincide with the poles of g and that the order of the zero is
twice the order of the pole at these points. Therefore, g and ω are the
Weierstrass data of a multivalued minimal immersion x from Σ to R3 ,
the differential of x being the real part of ω1 , ω2 and ω3 , see [22]. That
proves the claim.
From now on we work with the metric induced by x. In particular
|∇φ|2 = |σ|2 , σ being the second fundamental form of x, and the sec-
ond variation formula of the area Q coincides with quadratic form Qφ
associated to ∆ + |∇φ|2 . The differential (dx1 , dx2 , dx3 ) of the immer-
sion x is globally well-defined on Σ, which in turn implies that the map
ω ∈ H 1 (Σ, R) 7→ Xω : Σ → R3 introduced in (8) isR well-defined, too.
The dimension of the space V = {ω ∈ H 1 (Σ, R)| Σ ϕ1 Xω = 0}, ϕ1
being the first eigenfunction of ∆ + |σ|2 , is at least 2 × 4 − 3 = 5. As in
the proof of Theorem 7 we get that each ω ∈ V satisfies Q(Xω , Xω ) = 0,
which gives ∆Xω + |σ|2 Xω = 0, and, thus, ω ∈ L∗ (Σ). So V ⊂ L∗ (Σ)
and this contradiction shows that in the case genus = 4 and d = 3 we
cannot have index one. q.e.d.

Now we consider the case where Σ is a closed two-sided minimal

surface whose second variation formula quadratic form has index one.
If ϕ1 is the first eigenfunction of ∆ + |σ|2 , then the index one condition
can be written as
Z
(16) 0 ≤ Q(u, u), for all u ∈ C 1 (Σ) such that uϕ1 = 0.
Σ
86 A. ROS

When Σ is a compact two-sided minimal surface immersed in a 3-

manifold with nonnegative Ricci curvature, then one can use meromor-
phic functions as test functions, combined with Theorem 3, to control
the topology of Σ; see [41, 7, 20, 29, 28]. However, the bounds ob-
tained in these works are not sharp. As there are genus 3 index one
compact minimal surfaces in some flat 3-tori, see [33], the result below
gives the right estimate.

Theorem 15. Let Σ be a compact orientable two-sided minimal sur-

face immersed in a 3-manifold M with nonnegative Ricci curvature
Ric ≥ 0. If Σ has index one, then genus (Σ) ≤ 3.

Proof. Suppose Σ is a connected surface of genus g. From the Brill-

Noether theorem, see [1], we know there is a nonconstant meromorphic
map, φ : Σ → S 2 (1) of degree smaller than or equal to ≤ [(g + 1)/2] + 1,
where [z] denotes the integer part of z. Using Lemma 2 we can assume
that the R3 -valued function φ is orthogonal to the first eigenfunction
ϕ1 of the Jacobi operator of Σ (which can be assumed to be positive).
Therefore, we can use the linear components of φ as test functions in
the stability condition (3). Adding these inequalities and taking into
account (2), we get
Z
0 ≤ Q(φ, φ) = |∇φ|2 − (Ric (e1 ) + Ric (e2 ) − 2K)|φ|2 .
Σ

As φ is conformal, the first term transforms into 8π degree (φ). Using

that φ is of length one and the Gauss-Bonnet theorem, we obtain
Z
0 ≤ 8π([(g + 1)/2] + 1) + 4π(2 − 2g) − Ric (e1 ) + Ric (e2 ).
Σ

From this inequality, we conclude that g ≤ 5. Moreover, in the cases

g = 4 or 5 the equality holds. This implies that Ric (e1 ) + Ric (e2 ) = 0,
the Jacobi operator is ∆ − 2K and φ lies in its kernel, ∆φ − 2Kφ = 0.
As any meromorphic map is harmonic, we also have ∆φ + |∇φ|2 φ = 0.
Therefore |∇φ|2 = −2K and so the operator ∆ + |∇φ|2 has index one.
Moreover, the degree of φ is d = 3 (in the case g = 4) or d = 4 (for
g = 5). As this contradicts Theorem 4 and Theorem 14, we conclude
that g ≤ 3. q.e.d.

In the case Σ is complete and not compact in a manifold with non-

negative Ricci curvature, Ritoré [28] proved that if Σ has index one
(i.e., each relatively compact domain in Σ has index one), then g ≤ 3
and r ≤ 4 − g, g and r being the genus and the number of ends of Σ,
respectively. If Σ is compact, then he shows that g ≤ 4.
 ONE-SIDED COMPLETE STABLE MINIMAL SURFACES 87

5. Estimates of the index of minimal surfaces

The arguments of the above sections can be easily adapted to obtain
estimates of the index of complete minimal surfaces in R3 or in quotients
of R3 . In this section we present briefly the results which can be obtained
in this way. See [3, 12, 20] for other results about the index of minimal
surfaces.
Theorem 16. Let Σ be a nonflat compact minimal surface in a flat
3-torus T 3 . Then
2g − 3
index (Σ) ≥ , if Σ is orientable, and
3
g−3
index (Σ) ≥ , if Σ is nonorientable,
3
where g is either the genus of Σ, when Σ is orientable, or the genus of
the orientable two-seeted covering, when Σ is nonorientable.
Proof. Suppose first that Σ is orientable. Denote the index and the
genus of Σ by n and g, respectively, and consider the eigenfunctions of
the Jacobi operator ϕ1 , . . . , ϕn with negative eigenvalues. The dimen-
sion of the space
Z
V = {Xω | Xω ϕk = 0, k = 1, . . . n, ω ∈ H 1 (Σ, R3 )}
Σ
is at least 2g − 3n. If Xω ∈ V , then the index assumption gives
Q(Xω , Xω ) ≥ 0. On the other hand, we know that in fact the equality
holds in this inequality and we can deduce as in the proof of Theorem
7, that ω ∈ L∗ (Σ) and so dim V ≤ 3. Joining the two estimates we have
for the dimension of V , we conclude that 2g − 3n ≤ 3. That proves the
first assertion of the theorem.
If Σ is nonorientable, we denote its two-seeted orientable covering by
e
Σ. If the index of Σ is n, then there are eigenfuntions ϕ1 , . . . , ϕn of
the Jacobi operator of Σ e with ϕk ◦ τ = −ϕk (τ being the change of
sheet involution) and negative eigenvalues. The argument used in the
orientable case applies if we consider test functions in the space
Z
V = {Xω | e R3 )}
Xω ϕk = 0, k = 1, . . . n, ω ∈ H − (Σ,
Σ
e and we
whose dimension is at least g − 3n, g being the genus of Σ,
conclude as above that g − 3n ≤ 3. q.e.d.
Theorem 17. Let Σ be a complete nonflat minimal surface in R3 .
Then Σ has finite index if and only if it has finite total curvature. More-
over
2g
index (Σ) ≥ , if Σ is orientable and
3
g
index (Σ) ≥ , if Σ is nonorientable.
3
88 A. ROS

In the orientable case g is the genus of Σ and in the nonorientable one

g is the genus of the two-sheeted orientable covering of Σ.
Proof. Suppose first that Σ is orientable.
R Fisher-Colbrie [10] proved
2
that index (Σ) < ∞ if and only if Σ |σ| < ∞. If the index of Σ is
n, then there are eigenfunctions ϕ1 , . . . , ϕn ∈ L2 (Σ) of ∆ + |σ|2 with
negative eigenvalues and such that any compactly supported smooth
function ϕ L2 -orthogonal to ϕ1 , . . . , ϕn satisfies Q(ϕ, ϕ) ≥ 0, see Propo-
sition 2 in [10]. If the index estimate in the statement of the theorem
does Rnot hold, then there is X = Xω 6= 0, with ω ∈ L2 (Σ) ∩ H 1 (Σ, R)
and Σ Xϕk = 0, for k = 0, . . . n. After that property we can deduce,
as in the proof of Theorem 9, that ω lies in L∗ (Σ): In fact, we have
that Q(Xt , Xt ) ≥ 0 for the R3 -valued compactly supported function
Xt = ϕ(X + tY + v1 ϕ1 + · · · + vn ϕn ), where t is a real number, ϕ
is 3
R a cut-off function, Y : Σ → R has compact support and satisfies 3
Σ Y ϕk = 0, k = 0, . . . , n, and the vectors vk = vk (t, Y, ϕ) ∈ R are
2
choosen such that Xt is L -orthogonal to the eigenfunctions ϕ1 , . . . , ϕn .
Now we follow, step by step, the proof of Theorem 9 and we conclude
that ω ∈ L2 (Σ) ∩ L∗ (Σ). As Σ has finite total curvature, this means
that one of the linear coordinates of Σ is constant. This contradiction
proves the theorem in the orientable case.
In the case Σ is nonorientable, if follows from [10] that finite total
curvature implies finite index. To prove the converse result, using again
[10], it is enough to prove that finite index implies that the orientable
covering has finite genus. Now it follows that both the finiteness of the
genus and the bound of the g in terms of the index can be obtained by
the same kind of arguments used in the orientable case adapted to the
nonorientable context (as in the proof of Theorem 9). We remark that,
in particular, Proposition 2 in [10] is true in the nonorientable case.
q.e.d.
Theorem 18. Let Σ be a complete noncompact minimal surface in
R3 /Γ, Γ being a discrete subgroup of translation of R3 . Then Σ has
finite index if and only if it has finite total curvature and, in this case,
we have
2g − 1
index (Σ) ≥ , if Σ is orientable, and
3
g−1
index (Σ) ≥ , if Σ is nonorientable,
3
g being defined as in Theorem 17.
Proof. The only difference with the proof of Theorem 17 above is that
now we can have dim L∗ (Σ) ∩ L2 (Σ) = 1 (for instance, the Riemann
minimal examples are singly periodic minimal surfaces satisfying this
property). The proof of Theorem 12 gives that this dimension cannot
be larger than 1. q.e.d.
 ONE-SIDED COMPLETE STABLE MINIMAL SURFACES 89

6. Stability of constant mean curvature surfaces

The understanding of stable constant mean curvature surfaces is one
of the key steps in the study of the isoperimetric problem, see [32]. Let
Σ be a compact (two-sided) surface with constant mean curvature H
immersed in a Riemannian 3-manifold M . The surface Σ is said to be
stable (for the isoperimetric problem) if the second variation formula of
the area is nonnegative for any volume preserving deformation of Σ.
This is the sameR as saying that Q(u, u) ≥ 0 for all smooth functions
u ∈ C 1 (Σ) with Σ u = 0, where Q is defined as in (1), see Barbosa and
Do Carmo [2],
Z
Q(u, u) = |∇u|2 − (Ric(N ) + |σ|2 )u2 .
Σ
The mean value zero condition is the infinitesimal corresponding of the
volume constraint. When Σ is a minimal surface, we can use two differ-
ent notions of stability. When necessary to avoid ambiguity, we speak of
stability for the Plateau problem or stability for the isoperimetric prob-
lem. If Ric ≥ 0 and Σ is stable for the isoperimetric problem and
nonconnected, then Σ is totally geodesic. Note also that, opposite to
the minimal case, the two-sided assumption is the natural one in the
context of constant mean curvature surfaces. This justifies the hypoth-
esis of the theorem below.
Theorem 19. Let Σ be a compact connected orientable two-sided con-
stant mean curvature surface, immersed in a 3-manifold M with non-
negative Ricci curvature. If Σ is stable (for the isoperimetric problem),
then genus (Σ) ≤ 3.
Proof. The first part of the proof is parallel to the one of Theorem 15:
Instead of (2), in the constant mean curvature case the Gauss equation
gives Ric (N )+|σ|2 = Ric (e1 )+Ric (e2 )+4H 2 −2K. Using meromorphic
maps φ of small degree and mean value zero as test functions in the
stability quadratic form Q, we obtain that the genus of Σ is smaller
than or equal to 5. Moreover, if the genus is 4 or 5, then Σ is a minimal
surface of index one (as minimal surface). This contradicts Theorem 15
and, so, the theorem is proved. q.e.d.

References
[1] E. Arbarello, M. Cornalba, P.A. Griffiths, & J. Harris, Geometry of algebraic
curves, Vol. I, Grundlehren der Math. Wiss., 267, Springer-Verlag, New York,
1985, MR 0770932, Zbl 0559.1401.
[2] J.L. Barbosa & M. do Carmo, Stability of hypersurfaces with constant mean
curvature, Math. Z. 185 (1984) 339–353, MR 0731682, Zbl 0513.53002.
90 A. ROS

[3] J. Choe, Index, vision number and stability of complete minimal surfaces, Arch.
Rational Mech. Anal. 109 (1990) 195–212, MR 1025170, Zbl 0695.53045.
[4] T.H. Colding & C. De Lellis, The min–max construction of minimal surfaces,
Surveys in differential geometry, 8, Lectures in Geometry and Topology held
in honor of Calabi, Lawson, Siu and Uhlenbeck, 2003, 75–107, MR 2039986,
Zbl 1051.53052.
[5] T.H. Colding & W.P. Minicozzi II, Estimates for parametric elliptic integrands,
Int. Math. Res. Not. 6 (2002) 291–297, MR 1877004, Zbl 1002.53035.
[6] M. do Carmo & C.K. Peng, Stable complete minimal surfaces in R3 are planes,
Bull. Amer. Math. Soc. (N.S.) 1 (1979) 903–906, MR 0546314, Zbl 0442.53013.
[7] A. El Soufi & S. Ilias, Immersions minimales, première valeur propre du
laplacien et volume conforme, Math. Ann. 275 (1986) 257–267, MR 0854009,
Zbl 0675.53045.
[8] N. Ejiri, The index of minimal immersions of S 2 into S 2n , Math. Z. 184 (1983)
127–132, MR 0711733, Zbl 0517.53056.
[9] H.M. Farkas & I. Kra, Riemann surfaces, Lecture Notes on Mathematics, 71,
Springer-Verlag, 1980, MR 1139765, Zbl 0764.30001.
[10] D. Fischer-Colbrie, On complete minimal surfaces with finite Morse index in
three-manifolds, Invent. Math. 82 (1985) 121-132, MR 0808112, Zbl 0573.53038.
[11] D. Fischer-Colbrie & R. Schoen, The structure of complete stable minimal sur-
faces in 3-manifolds of nonnegative scalar curvature, Comm. Pure Appl. Math.
33 (1980) 199–211, MR 0562550, Zbl 0439.53060.
[12] A. Grigor’yan, Yu. Netrusov & S.-T. Yau Eigenvalues of elliptic operators and
geometric applications, Surveys in Differential Geometry IX (2004) 147–218,
Zbl 1061.58027.
[13] J. Hersch, Quatre propriétés isopérimétriques des membranes sphériques ho-
mogènes, C.R. Acad. Sci. Paris Ser. A-B 270 (1970) A1645–A1648,
Zbl 0224.73083.
[14] P. Li & S.-T. Yau, A new conformal invariant and its applications to the Will-
more conjecture and the first eigenvalue of compact surfaces, Invent. Math. 69
(1982) 269–291, MR 0674407, Zbl 0503.53042.
[15] I.C. Lima & A.M. Da Silveira, Stability of complete nonorientable minimal
surfaces in R3 , An. Acad. Brasil. Cinc. 68 (1996) 515–518, MR 1751245,
Zbl 1010.53006.
[16] F.J. López & A. Ros, Complete minimal surfaces with index one and stable
constant mean curvature surfaces, Comment. Math. Helv. 64 (1989) 34–43,
MR 0982560, Zbl 0679.53047.
[17] W.H. Meeks III, The theory of triply periodic minimal surfaces, Indiana Univ.
Math. J. 39 (1990), 877–936, MR 1078743, Zbl 0721.53057.
[18] W.H. Meeks III & J. Pérez, Conformal properties in classical minimal surface
theory, Surveys in Differential Geometry, IX, Eigenvalues of Laplacians and
other geometric operators (A. Grigor’yan and S.-T. Yau, eds.), International
Press, 2004, 275–335.
[19] W.H. Meeks III & H. Rosenberg, Stable minimal surfaces in M ×R, J. Differential
Geom. 68 (2004) 515–534, MR 2144539.
[20] S. Montiel & A. Ros, Schrödinger operators associated to a holomorphic map,
Global differential geometry and global analysis (Berlin, 1990), 147-174, Lecture
Notes in Math., 1481, Springer, Berlin, (1991), MR 1178529, Zbl 0744.58007.
 ONE-SIDED COMPLETE STABLE MINIMAL SURFACES 91

[21] S. Montiel & F. Urbano, Second variation of superminimal surfaces in self-

dual Einstein four-manifolds, Trans. Amer. Math. Soc. 349 (1997) 2253–2269,
MR 1422905, Zbl 0879.53044.
[22] R. Osserman, A survey of minimal surfaces, Second edition, Dover Publications,
Inc., New York, 1986, MR 0852409, Zbl 0209.52901.
[23] B. Palmer, Index and stability of harmonic Gauss maps, Math. Z. 206 (1991)
563–566, MR 1339977, Zbl 0845.53008.
[24] J.T. Pitts, Existence and regularity of minimal surfaces on Riemannian man-
ifolds, Mathematical Notes, 27, Princeton University Press, Princeton, 1981,
MR 0626027, Zbl 0462.58003.
[25] J.T. Pitts & J.H. Rubinstein, Equivariant minimax and minimal surfaces in geo-
metric three-manifolds, Bull. Amer. Math. Soc. 19 (1988) 303–309, MR 0940493,
Zbl 0665.49034.
[26] A.V. Pogorelov, On the stability of minimal surfaces, Dokl. Akad. Nauk SSSR
260 (1981) 293–295, MR 1473134, Zbl 0495.53005.
[27] M. Ritoré, Stable periodic projective planes, Proc. Amer. Math. Soc. 124 (1996)
3851–3856, MR 1363182, Zbl 0926.53008.
[28] , Index one minimal surfaces in flat three-space forms, Indiana Univ.
Math. J. 46 (1997) 1137–1153, MR 1631568, Zbl 0922.53022.
[29] M. Ritoré & A. Ros, Stable Constant Mean Curvature Tori and the Isoperimetric
Problem in Three Space Forms, Comment. Math. Helvet. 67 (1992) 293–305,
MR 1161286, Zbl 0760.53037.
[30] , The spaces of index one minimal surfaces and stable constant mean
curvature surfaces embedded in flat three manifolds, Trans. Amer. Math. Soc.
348 (1996) 391–410, MR 1161286, Zbl 0867.53007.
[31] A. Ros, The Gauss map of minimal surfaces, Differential Geometry, Valen-
cia 2001, Proceedings of the conference in honour of Antonio M. Naveira, O.
Gil-Medrano and V. Miquel ed., World Scientific, 2002, 235–252, MR 1922054,
Zbl 1028.53008.
[32] , The isoperimetric problem, Proceedings of the Clay Mathematical In-
stitute MSRI summer school on Minimal Surfaces, Global theory of minimal
surfaces, 175–209, Clay Math. Proc., 2, Amer. Math. Soc., Providence, RI,
2005, MR 2167260.
[33] M. Ross, Schwarz’ P and D surfaces are stable, Differential Geom. Appl. 2
(1992) 179–195, MR 1245555, Zbl 0747.53010.
[34] , Complete nonorientable minimal surfaces in R3 , Comment. Math. Helv.
67 (1992) 64–76, MR 1144614, Zbl 0779.53006.
[35] , The second variation of nonorientable minimal submanifolds, Trans.
Amer. Math. Soc. 349 (1997) 3093–3104, MR 1422909, Zbl 0890.53047.
[36] M. Ross & C. Schoen, Stable quotients of periodic minimal surfaces, Comm.
Anal. Geom. 2 (1994) 451–459, MR 1305713, Zbl 0843.58026.
[37] R. Schoen, Estimates for stable minimal surfaces in three-dimensional manifolds,
Seminar on minimal submanifolds, 111-126, Ann. of Math. Stud., 103, Princeton
Univ. Press, Princeton, NJ, 1983, MR 0795231, Zbl 0532.53042.
[38] F. Urbano, Index of Lagrangian submanifolds of CP n and the Laplacian of 1-
forms, Geom. Dedicata 48 (1993) 309–318, MR 1254100, Zbl 0802.53013.
92 A. ROS

[39] P. Yang & S.-T. Yau, Eigenvalues of the Laplacian of compact Riemann surfaces
and minimal submanifolds, Ann. Scuola Norm. Sup. Pisa Cl. Sci. (4) 7 (1980)
55–63, MR 0577325, Zbl 0446.58017.
[40] S.-T. Yau, Isoperimetric constants and the first eigenvalue of a compact Rie-
mannian manifold, Ann. Sci. cole Norm. Sup. 8 (1975) 487–507, MR 0397619,
Zbl 0325.53039.
[41] , Nonlinear analysis in geometry, L’enseignement Math. 33 (1987) 109–
158, MR 0896385, Zbl 0631.53002.

Departamento de Geometrı́a y Topologı́a

Facultad de Ciencias - Universidad de Granada
18071 Granada
Spain
E-mail address: aros@[Link]