You are on page 1of 15

Geom Dedicata (2016) 182:117–131

DOI 10.1007/s10711-015-0131-3

ORIGINAL PAPER

Hypersurfaces with constant higher order


mean curvature in Euclidean space

Luis J. Alías1 · Josué Meléndez1

Received: 15 July 2015 / Accepted: 20 November 2015 / Published online: 26 November 2015
© Springer Science+Business Media Dordrecht 2015

Abstract In this paper we derive sharp estimates for the infimum and for the supremum
of the squared norm of the second fundamental form of complete oriented hypersurfaces
of Euclidean space with constant higher order mean curvature and having two principal
curvatures, one of them simple. Besides, we characterize those hypersurfaces for which any
of these bounds is attained. Our results will be an application of a purely geometric result on
the principal curvatures of the hypersurface, the so called principal curvature theorem, given
by Smyth and Xavier (Invent Math 90:443–450, 1987).

Keywords Hypersurface · Mean curvature · Scalar curvature · Higher order mean


curvature · Second fundamental form · Principal curvature theorem

1 Introduction and statement of the main results

In [2, Theorem 11], the first author jointly with García-Martínez derive a sharp estimate for
the supremum of the scalar curvature Scal of a complete hypersurface in Euclidean space
with constant mean curvature and two distinct principal curvatures. Obviously, from the
basic relation between Scal and the squared norm of the second fundamental form A of
the hypersurface, that is |A|2 = n 2 H 2 − Scal [see Eq. (4)], such estimate can be stated
equivalently in terms of the infimum of |A|2 , and then we have the following statement.

Luis J. Alías was partially supported by MINECO/FEDER project MTM2012-34037, Spain. Josué
Meléndez was supported by CONACYT (México) Grant 234865.

B Luis J. Alías
ljalias@um.es
Josué Meléndez
josuems@ciencias.unam.mx
1 Departamento de Matemáticas, Universidad de Murcia, 30100 Espinardo, Murcia, Spain

123
118 Geom Dedicata (2016) 182:117–131

Theorem 1 Let M n be a complete oriented hypersurface of Rn+1 (n ≥ 3) with constant


mean curvature H and two distinct principal curvatures, one of them being simple. Then
sup Scal ≥ 0 or, equivalently, inf |A|2 ≤ n 2 H 2 . (1)
M M

Moreover, the equality holds in (1) if and only if M is either a circular cylinder Rn−1 × S1 (ρ),
with ρ = 1/n|H |, if H  = 0, or a higher dimensional catenoid, if H = 0.
The proof of this result, given in [2, Section 4], uses a nice geometric argument based
on the so called principal curvature theorem, due to Smyth and Xavier [11]. In this paper,
and following the approach introduced in [2], we will extend Theorem 1 to the case of
hypersurfaces with constant higher order mean curvature as follows.
Theorem 2 Let n ≥ 3 and 1 < r < n. Let M n be a complete oriented hypersurface of Rn+1
with constant r -th mean curvature Hr > 0 and two distinct principal curvatures, one of them
being simple. Then
 
n Hr 2/r
inf |A|2 ≤ (n − 1) . (2)
M n −r
Moreover, the equality holds in (2) if and only if M is isometric to the cylinder R × Sn−1 (ρ)
 
n − r 1/r
with ρ = .
n Hr
Observe that Theorem 2 does not hold for r = 1. When r = 1 and H > 0 the right estimate
for the infimum of |A|2 is the one given in Theorem 1, and the equality characterizes the
circular cylinder Rn−1 × S1 (ρ), with ρ = 1/n H . On the other hand, regarding the supremum
of |A|2 we obtain the following result, which holds even for r = 1.
Theorem 3 Let n ≥ 3 and 1 ≤ r < n. Let M n be a complete oriented hypersurface of Rn+1
with constant r -th mean curvature Hr > 0 and two distinct principal curvatures, one of them
being simple. Then
 
n Hr 2/r
sup |A|2 ≥ (n − 1) . (3)
M n −r
Moreover, the equality holds in (3) if and only if M is isometric to the cylinder R × Sn−1 (ρ)
 
n − r 1/r
with ρ = .
n Hr
In particular, when r = 1 the estimate (3) becomes
n2 H 2 n 2 (n − 2)H 2
sup |A|2 ≥ or, equivalently, inf Scal ≤ .
M n−1 M n−1
This estimate coincides with the one given by Alías and García-Martínez in [1, Corollary 4]
for more general complete hypersurfaces with constant mean curvature. Besides, the charac-
terization of the equality given here improves the one in [1] since we do not need to assume
that the infimum of Scal (or, equivalently, the supremum of |A|2 ) is attained at some point
of M.
It is worth pointing out that our main results improve, using a completely different tech-
nique, recent results given by Cheng [4], Wei [12], Wu [13] and Shu and Han [10] and others.
In fact, in all of these references the technique used by the authors is based on Otsuki’s idea
[7] of reducing the problem to the study and analysis of the solutions of an ODE. On the

123
Geom Dedicata (2016) 182:117–131 119

contrary, our approach is completely different and it is based on a purely geometric result on
the principal curvatures of the hypersurface, the so called principal curvature theorem, given
by Smyth and Xavier [11] (see Sect. 5 for further details).

2 Preliminaries

Let M n be a connected oriented hypersurface isometrically immersed into the Euclidean


space Rn+1 , and let us denote by A : X(M) → X(M) its second fundamental form (or shape
operator), with respect to a globally defined normal unit vector field N . As is well known, A
defines a self-adjoint linear operator on each tangent hyperplane T p M, and its eigenvalues
κ1 ( p), . . . , κn ( p) are the principal curvatures of the hypersurface. Associated to the second
fundamental form there are n algebraic invariants given by
Sr ( p) = σr (κ1 ( p), . . . , κn ( p)), 1 ≤ r ≤ n,
where σr : Rn → R is the elementary symmetric function in Rn , that is,

σr (x1 , . . . , xn ) = x i 1 . . . x ir .
i 1 <···<ir

The r -th mean curvature Hr of the hypersurface is then defined by


 
n
Hr = Sr , 1 ≤ r ≤ n.
r
In particular, when r = 1

1
n
1
H1 = κi = tr(A) = H
n n
i=1

is nothing but the mean curvature of M, which is the main extrinsic curvature of the hyper-
surface. On the other hand, H2 defines a geometric quantity which is related to the (intrinsic)
scalar curvature of M. Indeed, it follows from the Gauss equation of M that its Ricci curvature
is given by
Ric(X, Y ) = n H AX, Y  − AX, AY , X, Y ∈ X(M),
and then the scalar curvature of M is
 n 2
 
n
Scal = tr(Ric) = n H − |A| =
2 2
κi −
2
κi2 = n(n − 1)H2 . (4)
i=1 i=1

Then, H2 = R is nothing but the normalized scalar curvature of M. In general, when r is odd
the curvature Hr is extrinsic (and its sign depends on the chosen orientation), while when r
is even the curvature Hr is intrinsic and its value does not depend on the chosen orientation.
The key of the proofs of our results is a purely geometric result on the principal curvatures
of the hypersurface, the so called principal curvature theorem, given by Smyth and Xavier
[11]. It is stated as follows.
Theorem 4 (The principal curvature theorem) Let M n be a complete immersed orientable
hypersurface in Rn+1 , which is not a hyperplane, and let A denote its second fundamental
form with respect to a global unit normal field. Let  ⊂ R be the set of nonzero values
assumed by the eigenvalues of A and let ± =  ∩ R± .

123
120 Geom Dedicata (2016) 182:117–131

(i) If + and − are both nonempty, then inf M + = sup M − = 0.


¯ of  is connected.
(ii) If + or − is empty then the closure 

In this paper we are interested in the case of hypersurfaces in Euclidean space Rn+1 having
two distinct principal curvatures, which we will denote by λ and μ. In this case, the r -th mean
curvature of M is given by
  r   
n n − m m r −i i
Hr = λ μ, (5)
r r −i i
i=0
 
where n − m and m are the multiplicities of λ and μ, respectively, and as usual ij = 0
if i < j. As is well known since the pioneering work by Otsuki [7], if both principal
curvatures have multiplicity greater than 1, then the distributions of the space of principal
vectors corresponding to each principal curvature are completely integrable and each principal
curvature is constant on each of the integral leaves of the corresponding distribution (see
Theorem 2 and Corollary in [7]). In particular, if Hr curvature is constant it follows from
(5) that λ and μ are also constant and the hypersurface is an isoparametric hypersurface
with exactly two constant principal curvatures λ and μ, with multiplicities n − m and m,
1 < m < n − 1. Then, by the classical results on isoparametric hypersurfaces in Euclidean
space [6,9] the hypersurface must be an open piece of the standard product embedding
Rm × Sn−m (ρ) → Rn+1 with ρ > 0 (see also Theorem 2.1 in [12]).
As a consequence of the reasoning above, under the condition of having two distinct
principal curvatures, the interesting case for studying hypersurfaces in Rn+1 with constant
higher order mean curvature is the case where one of the principal curvatures is simple, that
is, with multiplicity 1. In this case, if λ and μ are the two distinct principal curvatures of M
with multiplicities (n − 1) and 1, respectively, then (5) reduces to

n Hr = λr −1 ((n − r )λ + r μ). (6)

As a first application of our method we can prove the following result as an easy consequence
of the principal curvature theorem. Observe that we do not need to assume here that Hr is
constant. Observe also that the result is meaningless if r is odd because in that case the sign
of Hr depends on the chosen orientation.

Proposition 5 If r is even, there exists no complete hypersurface M in Rn+1 (n ≥ 3) with


Hr∗ = sup M Hr < 0 and two distinct principal curvatures, one of them being simple.

Proof Assume that M is a complete hypersurface M in Rn+1 (n ≥ 3) with Hr∗ = sup M Hr <
0 and two distinct principal curvatures, one of them being simple. Let λ and μ be the two
distinct principal curvatures of M with multiplicities (n − 1) and 1, respectively, so that Hr
is given by (6). In particular, since Hr < 0 on M and r − 1 ≥ 1 > 0, then λ  = 0 on M. Since
r is even we may assume without loss of generality that λ > 0 and then + is non-empty. In
fact, the sign of Hr does not depend on the choice of the orientation, so that we may always
choose the orientation on M for which λ > 0. From (6) we obtain
n Hr n −r n Hr∗ n −r
μ(λ) = −1
− λ ≤ r −1 − λ,
rλ r r rλ r
with Hr∗ < 0. The global maximum of the function
n Hr∗ n −r
f (λ) = −1
− λ
rλ r r

123
Geom Dedicata (2016) 182:117–131 121

for λ > 0 is attained at


 1/r
n(r − 1)|Hr∗ |
λ0 = ,
(n − r )
with
max f (λ) = f (λ0 ) < 0.
λ>0
Therefore,
μ(λ) ≤ f (λ) ≤ f (λ0 ) < 0,
and − is non-empty with sup M − ≤ f (λ0 ) < 0. But this is a contradiction with the
principal curvature theorem because + is also non-empty.

In particular, when Hr is constant we obtain from Proposition 5 the part (2) of Theorem
5.4 in [13], using a completely different technique.
Corollary 6 If r is even, there exists no complete hypersurface in Rn+1 (n ≥ 3) with constant
negative r -th mean curvature and two distinct principal curvatures.
Proof Let M be a complete hypersurface in Rn+1 (n ≥ 3) with constant r -th mean curvature
and two distinct principal curvatures. If the multiplicities of the two distinct principal curva-
tures are both greater than 1, we know from the reasoning above (see also Theorem 2.1 in
[12]) that the hypersurface must be a standard product embedding Rm × Sn−m (ρ) → Rn+1
with ρ > 0. In this case, if 1 ≤ r ≤ n − m then Hr is the positive constant given by
   
n n−m 1
Hr = ,
r r ρr
while Hr = 0 if r > n − m. On the other hand, if one of the multiplicities is simple we know
from Proposition 5 that Hr cannot be a negative constant.

3 Proof of Theorem 2

Proof of Theorem 2 Let λ and μ be the two distinct principal curvatures of M with multi-
plicities (n − 1) and 1, respectively. We have that the r -th mean curvature of M is given
by
n Hr = λr −1 ((n − r )λ + r μ), (7)
and
|A|2 = (n − 1)λ2 + μ2 .
Since Hr is a positive constant and r > 1, λ  = 0. Thus
n Hr n −r
μ(λ) = − λ, (8)
r λr −1 r
and
 
n Hr (n − r )λ 2
|A|2 = |A|2 (λ) = (n − 1)λ2 + −
r λr −1 r
 2
n(r − 2r + n) 2
2 n Hr 2n(n − r )Hr −(r −2)
= λ + λ−2(r −1) − λ . (9)
r2 r r2

123
122 Geom Dedicata (2016) 182:117–131

When r is even we may assume without loss of generality that λ > 0. In fact, since the
principal curvature λ does not vanish on M and the sign of Hr does not depend on the choice
of the orientation, we may always choose the orientation on M for which λ > 0. On the other
hand, when r is odd the case λ < 0 cannot occur. Actually, if r is odd it follows from (8) that
the global minimum of μ(λ) for λ < 0 is attained at
 
n(r − 1)Hr 1/r
λ0 = − ,
(n − r )
with

min μ(λ) = μ(λ0 ) > 0,


λ<0

see Fig. 1. Therefore, if λ < 0 on M we get that + and − are both non-empty with
inf M + ≥ μ(λ0 ) > 0, which contradicts the principal curvature theorem.
Therefore, we may always assume that λ > 0 on M. It is easy to see from (9) that the
1/r
global minimum of |A|2 (λ) for λ > 0 is attained at λ = Hr , with
1/r 2/r
min |A|2 (λ) = |A|2 (Hr ) = n Hr .
λ>0

Observe that, independently of the value of Hr > 0, the equation


 
n Hr 2/r
|A|2 (λ) = (n − 1)
n −r
has exactly two positive roots λ1 and λ2 satisfying
1/r
0 < λ1 < Hr < λ2

(see Fig. 2). In fact, we can compute


 1/r
n Hr
λ2 = (10)
n −r
and observe that
μ(λ2 ) = 0. (11)

Fig. 1 Graph of the function μ(λ) given in (8) with r > 1, for r odd (on the left) and r even (on the right)

123
Geom Dedicata (2016) 182:117–131 123

Hr 2/r
Fig. 2 Graph of the function |A|2 (λ) given in (9) and the function constant |A|2 ≡ (n − 1) nn−r , for
1/r
r > 1. Note that the global minimum of |A|2 for λ > 0 is attained at λ = Hr

Suppose now that (2) is false, so that


 2/r
n Hr
|A| (λ) ≥ inf |A| > (n − 1)
2 2
. (12)
M n −r
This implies that either
0 < λ < λ1 ,
or
λ > λ2 .
1/r
If 0 < λ < λ1 , since μ(λ) is decreasing when λ > 0, we obtain μ(λ) > μ(λ1 ) > μ(Hr ) =
1/r
Hr . Therefore,
1/r
0 < λ < λ1 < Hr < μ(λ1 ) < μ(λ). (13)
Thus, we have that − is empty and, by the principal curvature theorem, we get  ¯ is con-
1/r
nected. On the other hand, also from (13), there exists ε > 0 such that intervals (−ε, Hr −ε)
1/r
and (Hr + ε, +∞) are disjoint and
¯ ⊂ (−ε, Hr1/r − ε) ∪ (Hr1/r + ε, +∞)

¯ is connected. Thus, this case cannot occur.
which contradicts that 
On the other hand, if λ > λ2 , then from (11) we have
μ(λ) < μ(λ2 ) = 0.
Therefore, + and − are both non-empty with inf M + ≥ λ2 > 0. But this contradicts
the principal curvature theorem. Thus, this case cannot occur neither. As a consequence, it
must be
 
n Hr 2/r
inf |A| ≤ (n − 1)
2
.
M n −r

2/r
n Hr
Suppose now that inf M |A|2 = (n − 1) n−r . This implies that either

0 < λ ≤ λ1 ,

123
124 Geom Dedicata (2016) 182:117–131

or
 1/r
n Hr
λ ≥ λ2 = .
n −r

Observe that the first case cannot happen. Actually, if 0 < λ ≤ λ1 then we would also have
1/r
0 < λ ≤ λ1 < Hr < μ(λ1 ) ≤ μ(λ). (14)

¯ would
Thus, we would have that − is empty and, by the principal curvature theorem, 
1/r
be connected. But from (14), there exists ε > 0 such that intervals (−ε, Hr − ε) and
1/r
(Hr + ε, +∞) are disjoint and

¯ ⊂ (−ε, Hr1/r − ε) ∪ (Hr1/r + ε, +∞)




¯ is connected. Therefore, it must hold necessarily


which contradicts that 
 1/r
n Hr
λ ≥ λ2 = .
n −r

From (11), since μ(λ) is decreasing when λ > 0, we obtain μ(λ) ≤ 0. This implies that
inf M + ≥ λ2 > 0 and hence, again by the principal curvature theorem, − must be empty,

1/r
n Hr
which means that μ = constant = 0. Hence, λ = constant = n−r is also constant
and, by the classical results on isoparametric hypersurfaces in Euclidean space [6,9] M is

1/r
isometric to R × Sn−1 (ρ), with ρ = n−r n Hr .

4 Proof of Theorem 3

For the proof of Theorem 3, in addition to the principal curvature theorem we will need
the following result, which is a consequence of the solution given by Perelman [8] to the
soul conjecture of Cheeger and Gromoll. Specifically, we will use the following topological
structure result for complete and non-compact Riemannian manifolds with non-negative
sectional curvature.

Lemma 7 Let M n be a complete, connected and non-compact Riemannian manifold with


non-negative sectional curvature containing a point where all the sectional curvatures are
strictly positive in all sections directions. Then the soul of M n is a point; equivalently M n is
diffeomorphic to Rn .

On the other hand, we will need also the following algebraic lemma for the sectional
curvatures at a point p of a hypersurface with two distinct principal curvatures in Euclidean
space, one of them being simple. Although it is possible known by the specialists, since we
do not know of any specific reference where finding it and for the readers’ convenience, we
include here a detailed proof.

Lemma 8 Let M be a hypersurface of Rn+1 with two distinct principal curvatures λ and μ
with multiplicities (n − 1) and 1 such that λ( p), μ( p) ≥ ε > 0 for some p ∈ M. Then all
sectional curvatures K M ( p) of M at p satisfy K M ( p) ≥ ε 2 .

123
Geom Dedicata (2016) 182:117–131 125

Proof Let E 1 , . . . , E n be a local orthonormal frame on a neighborhood of p such that AE i =


λE i for i = 1, . . . , n − 1 and AE n = μE n , where A is the second fundamental form of M.
We consider

n 
n
U= u i Ei V = vi E i
i=1 i=1

with |U | = |V | = 1 and U, V  = 0. It is well known that the curvature tensor R M of a


hypersurface M immersed in Rn+1 is given by the Gauss equation
R M (X, Y )Z = AX, Z AY − AY, Z AX.
Thus, the sectional curvature K M (U ∧ V ) of the plane generated by U, V is given by
K M (U ∧ V ) = R(U, V )U, V  = AU, U AV, V  − AU, V AU, V 
From a straightforward calculation we get
 n ⎛ n ⎞  2
  
n
K M (U ∧ V ) = u i λi ⎝
2
vjλj⎠ −
2
u i vi λi . (15)
i=1 j=1 i=1

Since λi = λ for i = 1, . . . , n − 1, λn = μ and U, V are orthonormal, we have



n 
n−1
u i2 λi = λ u i2 + μu 2n = λ(1 − u 2n ) + μu 2n = (μ − λ)u 2n + λ. (16)
i=1 i=1

Analogously, we can obtain



n 
n
vi2 λi = (μ − λ)vn2 + λ, and u i vi λi = (μ − λ)u n vn . (17)
i=1 i=1

We substitute (16) and (17) in (15) and simplify to find that


K M (U ∧ V ) = ((μ − λ)u 2n + λ)((μ − λ)vn2 + λ) − (μ − λ)2 u 2n vn2
= λ2 + λ(μ − λ)(u 2n + vn2 ). (18)
We evaluate now at the point p and consider two cases. We suppose first that λ( p) < μ( p).
Thus,
K M (U ∧ V ) = λ2 ( p) + λ( p)(μ( p) − λ( p))(u 2n + vn2 ) ≥ λ2 ( p) ≥ ε 2 , (19)
since λ( p) ≥ ε > 0. On the other hand, using the method of Lagrange multipliers we can
easily proof the following algebraic inequality:
n n n
If i=1 u i vi = 0 and i=1 u i =
2
i=1 vi = 1 then u i + vi ≤ 1 for every
2 2 2

i = 1, . . . , n.
Applying this fact in the case λ( p) > μ( p) we have
K M ((U ∧ V ) = λ2 ( p) + λ( p)(μ( p) − λ( p))(u 2n + vn2 )
≥ λ2 ( p) + λ( p)(μ( p) − λ( p)) = λ( p)μ( p)ε 2 (20)
when λ( p) > μ( p). Therefore, from (19) and (20), K M ( ) ≥ ε 2 > 0 for every 2-plane
⊂ T p M if λ( p), μ( p) ≥ ε.

We are now ready to give the proof of our second main result.

123
126 Geom Dedicata (2016) 182:117–131

Proof of Theorem 3 When r > 1 and since Hr > 0, we know from the reasoning done at
the beginning of the proof of Theorem 2 that λ > 0 on M, and μ and |A|2 can be written in
terms of λ as follows
n Hr n −r
μ(λ) = r −1 − λ, (21)
rλ r
and  2
n(r 2 − 2r + n) 2 n Hr 2n(n − r )Hr −(r −2)
|A|2 = λ + λ−2(r −1) − λ . (22)
r2 r r2
In particular, independently of the value of Hr > 0, the equation
 
n Hr 2/r
|A|2 (λ) = (n − 1)
n −r
has exactly two positive roots λ1 and λ2 satisfying
1/r
0 < λ1 < Hr < λ2

with  1/r
n Hr
λ2 = (23)
n −r
and
μ(λ2 ) = 0. (24)
In the case r = 1 one cannot deduce directly that the same must happen when H1 = H >
0. The problem is that Eq. (7) with r = 1 does not imply λ  = 0. However in this case (r = 1)
we can reach the same conclusion if we assume that (3) is false, that is,
n2 H 2
|A|2 ≤ sup |A|2 < . (25)
M n−1
In this case (22) becomes simply

|A|2 = (n − 1)λ2 + μ2
= n(n − 1)λ2 − 2n(n − 1)H λ + n 2 H 2 ,

and the equation


n2 H 2
|A|2 (λ) =
n−1
has again exactly two positive roots λ1 and λ2 given by
(n − 2)H nH
λ1 = and λ2 = .
n−1 n−1
Therefore, by (25) we have (see Fig. 3) that
n2 H 2
|A|2 (λ) <
n−1
if and only if
(n − 2)H nH
0< <λ< on M.
n−1 n−1

123
Geom Dedicata (2016) 182:117–131 127

Fig. 3 Graph of the function |A|2 for r = 1

Observe that in the case r = 1, once we know that λ > 0, Eq. (21) is still valid and reduces
to
μ(λ) = n H − (n − 1)λ. (26)
Therefore, we also have in this case
μ(λ2 ) = 0.
Suppose now that r ≥ 1 and (3) is false, so that
 2/r
n Hr
|A| (λ) ≤ sup |A| < (n − 1)
2 2
(27)
M n −r
with λ > 0 on M. This implies that
0 < λ1 < λ < λ2 on M,
where λ1 and λ2 are the two positive roots of
 2/r
n Hr
|A| (λ) = (n − 1)
2
.
n −r
1/r 1/r
It is worth pointing out that λ1 < Hr < λ2 and λ( p)  = Hr for every p ∈ M. In fact, if
1/r 1/r
λ( p0 ) = Hr at a point p0 ∈ M then we would have from (7) that μ( p0 ) = Hr = λ( p0 ),
which contradicts that M has two principal curvatures. This implies that either
1/r
0 < λ1 < λ( p) < Hr for every p ∈ M, (28)
or
1/r
Hr < λ( p) < λ2 for every p ∈ M. (29)
In the case (28) it follows from (26) if r = 1, and from (21) if r > 1, that
1/r 1/r
0 < μ(Hr ) = Hr < μ( p) < μ(λ1 ) for every p ∈ M. (30)
Similarly, in the case (29) it follows from (26) if r = 1, and from (21) if r > 1, that
1/r 1/r
μ(λ2 ) = 0 < μ( p) < μ(Hr ) = Hr for every p ∈ M. (31)

123
128 Geom Dedicata (2016) 182:117–131

Even more, in the case (29) we have in fact


Hr < λ( p) ≤ λ∗2 < λ2
1/r
for every p ∈ M, (32)

where λ∗2 ∈ Hr , λ2 satisfies


1/r

 2/r
n Hr
|A| 2
(λ∗2 ) = sup |A| (λ) < |A| (λ2 ) = (n − 1)
2 2
.
M n −r
Then, in this case we also have that μ( p) ≥ μ(λ∗2 ) > μ(λ2 ) = 0. Therefore, in both cases
there exists ε > 0 such that λ( p), μ( p) ≥ ε > 0 for every p ∈ M. Applying now Lemma 8
we obtain that all sectional curvatures K M of M satisfy K M ( p) ≥ ε 2 > 0 for every p ∈ M
and, by Bonnet–Myers’ theorem, the complete hypersurface M is compact.
In particular, using either (28) and (30) or (29) and (31), we obtain from the compactness
of M that [λ∗ , λ∗ ] and [μ∗ , μ∗ ] are disjoint with
 = [λ∗ , λ∗ ] ∪ [μ∗ , μ∗ ]
closed, where
λ∗ = min λ, λ∗ = max λ, μ∗ = min μ, μ∗ = max μ.
M M M M

¯ =  is disconnected, with
Thus,  −
= ∅. But this contradicts the principal curvature
theorem. Therefore, the inequality (27) is impossible and it must be
 
n Hr 2/r
sup |A|2 ≥ (n − 1) .
M n −r
Suppose now that equality holds,
 2/r
n Hr
sup |A|2 = (n − 1) .
M n −r
As above, by continuity, we have two possibilities:
1/r 1/r
(a) 0 < λ1 ≤ λ( p) < Hr and 0 < Hr < μ( p) ≤ μ(λ1 ), or
 
1/r n Hr 1/r 1/r
(b) Hr < λ( p) ≤ λ2 = and μ(λ2 ) = 0 ≤ μ( p) < Hr
n −r
for every p ∈ M.
Note that using Bonnet–Myers’ theorem in case (a) we deduce as above that M must
be compact. Hence, reasoning as above we obtain the same contradiction with the principal
curvature theorem. Therefore case (a) cannot occur. On the other hand, it happens exactly
the same if we assume in case (b) that M is compact. Therefore, it only remains to consider
the case (b) when M is non-compact. Our objective is to prove that it must be
 
n Hr 1/r
λ = constant = .
n −r

1/r
n Hr
Suppose on the contrary that λ( p0 ) < n−r , for some p0 ∈ M, so that μ( p0 ) > 0. Thus,
by Lemma 8, we have that all sectional curvatures of M are positive at p0 , and by Lemma 7,
we get that M n is diffeomorphic to Rn .
On the other hand, we know that λ > 0 with λ − μ  = 0 on M. Thus, all the conditions
of Theorem 4.2 in [5] are satisfied, and therefore, our hypersurface is a complete rotational

123
Geom Dedicata (2016) 182:117–131 129

hypersurface. Recall that rotational hypersurfaces in Rn+1 are constructed taking the orbit
of a curve α, called the profile curve, under the orthogonal transformations of Rn+1 leaving
fixed an axis. It is then immediate that a rotational hypersurface of Rn+1 can be parametrized
by
f (t1 , . . . , tn−1 , s) = (x1 (s)ϕ1 , . . . , x1 (s)ϕn , xn+1 (s))
where s ∈ R is the arc length of the curve α(s) = (x1 (s), 0, . . . , 0, xn+1 (s)) and
ϕ(t1 , . . . , tn−1 ) = (ϕ1 , . . . , ϕn ) is an orthogonal parametrization of a unit sphere Sn−1 ;
for details, see equation (2.11) in [5]. Thus, we can give a natural homeomorphism between
a complete rotational hypersurface and the product R × Sn−1 ,
(x1 (s)ϕ1 , . . . , x1 (s)ϕn , xn+1 (s))  → (s, ϕ1 , . . . , ϕn ).
But this is a contradiction with the fact that M n is diffeomorphic to Rn .

1/r
n Hr
As a consequence, λ = constant = n−r , and hence μ = constant = 0. In
other words, M is an isoparametric hypersurface, and by the classical results on isopara-
metric hypersurfaces in Euclidean space [6,9] M is isometric to R × Sn−1 (ρ), with

1/r
ρ = n−rn Hr .

5 Further applications

It is worth pointing out that when r is even Theorems 2 and 3 remain true if one replaces
the condition Hr > 0 by Hr  = 0, because there exists no complete hypersurface in Rn+1
(n ≥ 3) with constant negative Hr and two distinct principal curvatures, one of them being
simple (see Corollary 6). In particular, when r = 2 we know from (4) that H2 = R is nothing
but the normalized scalar curvature of M. In this case Theorems 2 and 3 can be re-written in
terms of R as follows.
Theorem 9 Let M n be a complete oriented hypersurface of Rn+1 (n ≥ 3) with constant
(normalized) scalar curvature R  = 0 and having two distinct principal curvatures, one of
them being simple. Then R > 0 and
n(n − 1)R
inf |A|2 ≤ ≤ sup |A|2 . (33)
M n−2 M

Moreover, any two of the


 equality holds in (33) if and only if M is isometric to the cylinder
n−2
R × Sn−1 (ρ) with ρ = .
nR
The estimate for sup M |A|2 given in (33), when written in terms of the total umbilicity tensor
of the hypersurface , coincides with the one given by Alías, García-Martínez and Rigoli
in [3, Theorem 2] for more general complete hypersurfaces with constant scalar curvature.
On the other hand, the characterization of the equality given here improves the one in [3]
since we do not need to assume that the supremum of |A|2 (or, equivalently, the supremum
of | |2 ) is attained at some point of M.
Remark 10 Observe that, using (4) we can rewrite (33) in terms of the mean curvature of the
hypersurface as follows,
(n − 1)2 R
inf H 2 ≤ ≤ sup H 2 . (34)
M n(n − 2) M

123
130 Geom Dedicata (2016) 182:117–131

In particular, and as a direct application of Theorem 9, we obtain the following consequence


which extends Theorem 4.1 in [4].
Corollary 11 Let M n be a complete oriented hypersurface of Rn+1 (n ≥ 3) with constant
(normalized) scalar curvature R  = 0 and having two distinct principal curvatures, one of
them being simple. Then R > 0. If either
n(n − 1)R n(n − 1)R
|A|2 ≥ or |A|2 ≤
n−2 n−2

n−2
then M is isometric to the cylinder R × Sn−1 (ρ) with ρ = .
nR
More generally, Theorem 2 gives the following consequence.
Corollary 12 (Theorem 4.1 in [12]) Let n ≥ 3 and 1 < r < n. Let M n be a complete
oriented hypersurface of Rn+1 with constant r -th mean curvature Hr > 0 and two distinct
principal curvatures, one of them being simple. If
 
n Hr 2/r
|A|2 ≥ (n − 1) , (35)
n −r
 
n − r 1/r
then M is isometric to the cylinder R × Sn−1 (ρ) with ρ = .
n Hr
On the other hand, the following is a direct consequence of our Theorem 3.
Corollary 13 (Part (1) of Theorem 5.4 in [13] and Theorem 1.7 in [10]) Let n ≥ 3 and
1 ≤ r < n. Let M n be a complete oriented hypersurface of Rn+1 with constant r -th mean
curvature Hr > 0 and two distinct principal curvatures, one of them being simple. If
 
n Hr 2/r
|A|2 ≤ (n − 1) , (36)
n −r
 
n − r 1/r
then M is isometric to the cylinder R × S (ρ) with ρ =
n−1 .
n Hr

Acknowledgments The authors would like to thank the anonymous referee for his/her valuable suggestions
and corrections which contributed to improve this paper. The second author is very grateful to Luis J. Alías
and the Department of Mathematics of the University of Murcia for the hospitality and support.

References
1. Alías, L.J., García-Martínez, S.C.: On the scalar curvature of constant mean curvature hypersurfaces in
space forms. J. Math. Anal. Appl. 363, 579–587 (2010)
2. Alías, L.J., García-Martínez, S.C.: An estimate for the scalar curvature of constant mean curvature hyper-
surfaces in space forms. Geom. Dedic. 156, 31–47 (2012)
3. Alías, L.J., García-Martínez, S.C., Rigoli, M.: A maximum principle for hypersurfaces with constant
scalar curvature and applications. Ann. Glob. Anal. Geom. 41, 307–320 (2012)
4. Cheng, Q.-M.: Complete hypersurfaces in a Euclidean space Rn+1 with constant scalar curvature. Indiana
Univ. Math. J. 51, 53–68 (2002)
5. do Carmo, M., Dajczer, M.: Rotational hypersurfaces in spaces of constant curvature. Trans. Am. Math.
Soc. 277, 685–709 (1983)
6. Levi-Civita, T.: Famiglia di superfici isoparametriche nell’ordinario spazio Euclideo. Att. Accad. Naz.
Lincie Rend. Cl. Sci. Fis. Mat. Natur. 26, 355–362 (1937)

123
Geom Dedicata (2016) 182:117–131 131

7. Otsuki, T.: Minimal hypersurfaces in a Riemannian manifold of constant curvature. Am. J. Math. 92,
145–173 (1970)
8. Perelman, G.: Proof of the soul conjecture of Cheeger and Gromoll. J. Differ. Geom. 40, 209–212 (1994)
9. Segre, B.: Famiglie di ipersuperficie isoparametriche negli spazi euclidei ad un qualunque numero di
dimensioni. Att. Accad. Naz. Lincie Rend. Cl. Sci. Fis. Mat. Natur. 27, 203–207 (1938)
10. Shu, S., Han, A.Y.: Hypersurfaces with constant k-th mean curvature in a unit sphere and Euclidean space.
Bull. Malays. Math. Sci. Soc. 35, 435–447 (2012)
11. Smyth, B., Xavier, F.: Efimov’s theorem in dimension greater than two. Invent. Math. 90, 443–450 (1987)
12. Wei, G.: Complete hypersurfaces in a Euclidean space Rn+1 with constant mth mean curvature. Differ.
Geom. Appl. 26, 298–306 (2008)
13. Wu, B.-Y.: On hypersurfaces with two distinct principal curvatures in Euclidean space. Houst. J. Math.
36, 451–467 (2010)

123

You might also like