proceedings of the

american mathematical society

Volume 110, Number 1, September 1990

FLAT SURFACES WITH MEAN CURVATURE VECTOR

OF CONSTANT LENGTH IN EUCLIDEAN SPACES

KAZUYUKI ENOMOTO

(Communicated by Jonathan M. Rosenberg)

Abstract. Complete flat surfaces in Rn are studied under the condition that
the normal connection is flat and the length of the mean curvature vector is
constant. It is shown that such a surface must be the product of two curves of
constant geodesic curvature.

In this paper we prove the following theorem.

Theorem. Let M be a complete C°° flat surface in Rn. Suppose that the normal
connection of M is flat and the length of the mean curvature vector is constant.
Then there exist curves of constant geodesic curvature C, z'zzRr and C2 in M.""
( 1 < r < n— 1) such that M is congruent to the Riemannian product of Cx and
C2.

A surface is called flat if the Gaussian curvature is zero at every point.

Let M be a surface in R" . We denote the standard inner product and
the covariant differentiation of R" by ( , ) and D respectively. For tangent
vector fields X, Y, and a normal vector field Ç of M, we write DXY =
DXY + B(X, Y) and DXÇ = -A(X + D¡¿i, where DXY (resp. -A(X) is
the tangential component of DXY (resp. DXÇ), and B(X, Y) (resp. DXÇ ) is
the normal component of DXY (resp. DXÇ). Let {e¡} (i = 1, 2) be a local
orthonormal frame field of the tangent bundle TM of M and {ea} (a =
3, ... , zz) be a local orthonormal frame field of the normal bundle T M of
M. We define coAB(X) = (DxeA ,eB) (A, B = 1, ... , n) for X in TM. We
say that a point p in M is umbilical with respect to a normal vector £ at p
if A? is proportional to the identity transformation of the tangent space T M
of M at p. If K denotes the Gaussian curvature of M, the Gauss equation
is given by

(1) K = (B(ex, ex),B(e2,e2)) - (B(ex, e2),B(ex, e2)).

Received by the editors July 28, 1989 and, in revised form, October 19, 1989.
1980 Mathematics Subject Classification (1985 Revision). Primary 53C42; Secondary 53A07.
Key words and phrases. Flat surface, mean curvature vector, product surface.

©1990 American Mathematical Society

0002-9939/90 $1.00+$.25 per page

211

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 212 KAZUYUKI ENOMOTO

The Codazzi equation is given by

(2) (DxB)(Y,Z)-(DYB)(X,Z) =0
for tangent vector fields X , Y, and Z .
Lemma I. If M is flat and the normal connection of M is flat, then there exists
an orthonormal basis {ex, e2} of T M for each p in M such that B(ex , e2) =
B(e2,ex) = 0 and (B(ex ,ex), B(e2, e2)) = 0.
Proof. Since the normal connection is flat, all ^ 's (£ G Tp M) are simul-
taneously diagonalizable. Hence there exists an orthonormal basis {ex , e2}
of T M such that (A(ex,e2) = (A*e2,ex) = 0 for all £,. Then we have
B(ex, e2) = B(e2, ex) = 0. Since the Gaussian curvature is zero, it follows
from (1) and B(ex, e2) = 0 that (B(ei,e]), B(e2, e2)) = 0.
Let M be a surface in R" which satisfies the conditions in the theorem.
Since the length of the mean curvature vector is constant, we have \\B(ex, ex)\\ +
1 2
\\B(e2, e2)\\ —c for some constant c, where {ex, e2} is the orthonormal basis
of T M given in Lemma 1.
If c = 0, then M is totally geodesic. In the following argument, we assume
c^O.
Let 7i: M —>M be the universal covering of M. Since M is complete and
flat, M is isometric to R equipped with the standard flat metric.
Lemma 2. There exists a global C°° orthonormal frame field {èx,ê2} on M
such that, for any p in M, dn(ëx) and dn(ê2) are eigenvectors of A* for all
Ç in the normal space of M at n(p).
Proof. Since c / 0, there does not exist a point on M which is umbilical
with respect to all normal vectors. Hence for each point p of M there exists a
simply connected neighborhood U and a C°° normal vector field £ defined on
U such that every point in U is not umbilical with respect to £. Let {ex, e2}
be an orthonormal frame field of TM\V which consists of eigenvectors of A..
By a result in [6], {e^ , e2} is a C°° frame field on U. For q in M, let V,
*1>if » fj) nave me same properties as U , £,, {ex, e2} respectively. Suppose
that U n V is nonempty and connected. Since the normal connection of M
is flat, A, and A have common eigenvectors on U n V . Hence we can take
{f\ » fii so tnat e\ = f\ and e2 = fi on ^n ^ • Since M is simply connected,
this continuation method and the standard monodromy argument allows us to
define a global C°° orthonormal frame field {èx, ë2} on M which has the
desired property.
By Lemma 2, B(dn(eA , d7i(èi)) (i = 1, 2) is a C°° normal vector field of
M which satisfies \\B(dn(èx), dn(èx))\\ + \\B(dn(è1), dn(è2))\\ - c~. We use
the Codazzi equation (2) to obtain

(3) D^(B(ex ,ex)) = coX2(ex)(B(ex, ex) - B(e2,e2))

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 FLAT SURFACES WITH MEAN CURVATURE VECTOR 213

and

(4) D^(B(e2,e2)) = co2x(e2)(B(e2, e2) - B(ex ,ex)),

where et — dn(ëA for i = 1,2. We define subsets MQ, Mx, M2 of M by

M. = {peM: B(dn(êi), dn(ët)) = 0} for i = 1 , 2 and M0 = M\(MX U M2).
For p in M let ¿r((Z) be the maximal integral curve of ët with ¿f(.(0)= /z,.

Lemma 3. If p lies in Mx (resp. M2 ), then d2 (resp. ox ) is entirely contained

in Mx (resp. M2 ). Moreover, d2 (resp. ox ) is a geodesic of M.
Proof. Suppose p lies in Mx . Set

f(t) = \\B(d7l(ëx(d2(t))), d7Z(ëx(d2(t))))\\2.

Using (3), we see that f(t) satisfies a differential equation

(5) df/dt = v(t)f(t),

where if/(t) = 2(Dè ëx, ë2)(à2(t)). Since, by Lemma 2, y/ is a C°° function
defined for all t and /(0) = 0,(5) implies that /(*) = 0 for all t. Thus d2(t)
is contained in Mx for all t.
Since \\B(dn(è2), dn(è2))\\2 attains its maximum (= c ) along d2(t), we
have

(6) ëx(\\B(d7t(ê2),d7i(è2))\\2) = 0
at every point of ö2(t). Equations (4) and (6) yield co2x(e2) = 0, which implies
that â2(t) is a geodesic of M. The proof is similar when p lies in M2.

We define continuous unit vector fields Yx and Y2 on M by

Yx= (\\B(d7t(èx),d7i(êx)\\/c)êx - (\\B(dn(è2),d7i(ê2))\\lc)ë2

and
Y2 = (\\B(d7i(èx),dn(êx))\\/c)êx + (\\B(d7i(ë2),dn(ë2))\\/c)ë2.
Yx and Y2 are C°° on M0. Moreover, using (3) and (4) together with
\\B(dn(êx), d7t(êx))\\2 + \\B(dn(ê2), dn(è2))\\2 = c2, one can show that D~Yi =
0 for z = 1, 2 . Thus we have
Lemma 4. Every integral curve of Y¡\M0 (i - 1 , 2) is geodesic.

Let p be a point in M0 and let f((t) be an integral curve of Yt\MQ with

7,(0) =p.
Lemma 5. y¡ is definedfor all t e R.
Proof. Suppose y( is defined for all t e [0,T) but not for t = 7\ Then
¿7= lim^y. y.(/) is contained in either Mx or M2. Suppose q lies in Mx . By

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 214 KAZUYUKI ENOMOTO

Lemma 4, y¡([0, T)) is a geodesic segment on M such that \imt^T(dyj/dt) =

lim,^^ Y^y^t)) — ±è2(q). By Lemma 3, the geodesic of M which passes
through q e M, and has the tangent vector ±ê2 at q is entirely contained in
Mx . This implies that y(([0, T)) is contained in Mx , which is contradiction.
Proof of theorem. If c = 0, M is totally geodesic, i.e., a two-dimensional affine
subspace in R" .
Suppose c ^¿ 0 and MQ / 0. By Lemmas 4 and 5, every integral curve
of Y^Mq (i = 1, 2) is a complete geodesic. Since M is isometric to the
Euclidean plane, all integral curves of F;|A/0 for each i must be parallel lines.
This implies that M0 = M and that the angle between Y, and Y2 is constant
on M. Then, by the definition of Yi, we see that all integral curves of ëi are
parallel lines on M for each i. In the case when MQ = <f>(i.e., M = Mx or
M = M2 ), we use Lemma 3 to observe that all integral curves of êj are parallel
lines on M. Therefore, there is a Cartesian coordinate system (ux, u2) on M
(= R ) such that d/dü¡ = ëi for z = 1,2. Let X(ux, u2) be the position
vector of the point on M whose coordinate is (ux, u2). Then we have

d2X/düxdü2 = Dëë2 = Dee2 = De e2 + B(ex, e2) = 0.

Hence there exist R"-valued functions C(ux) and C2(ü2) suchthat X(üx, ü2)
—C,(ù,) + C2(^2) • Since ët - dX/dü¡ = dC¡/dü¡ and (<?,,ë2) —0, we have

(7) (dCx/düx,dC2/dü2} = 0
for any (ux , u2) in R2 .
Let Pi be the affine subspace of the lowest dimension which contains C .
Equation (7) implies that Px and P2 are orthogonal. Since X(ux , u2) =
Cx(ux) + C2(u2), M is the Riemannian product of Cx and C2. Since M
is complete, C; is either a curve of infinite length without endpoints or a cov-
ering of a closed curve Ct. Then M is the Riemannian product of C, and
C2. (We set C¡ = C¡ if C; is not a covering of a closed curve.) The geodesic
curvature of Cl is given by \\B(dn(ej), dn(ëj))\\, which is constant since the
angle between Yx and Y2 is constant on M. Hence C, and C2 have constant
geodesic curvatures.

Corollary. Let M be a compact flat surface in R4. // the normal connection

of M is flat and the length of the mean curvature vector is constant, then M is
congruent to the Riemannian product of two circles, Sl (rx) x Sl(r2).
Remark 1. Our theorem is global (i.e., completeness is necessary), as we see in
the following example: Let S - {(x, y, z) = (reos6, rsind, r): 1 < r < 2,
0 < 6 < 2n). S is a flat surface in R whose mean curvature is \/2\/2r. Let
4>(s) = (<t>x(s),4>2(s)) be a curve in R parameterized by arc-length s whose

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 FLAT SURFACES WITH MEAN CURVATURE VECTOR 215

curvature is y 2(1 -s 2). Define an isometric immersion <P: R —>R by

<S>(x,y, z) = (x,y, 4>x(z), cf>2(z)). Then M = 0(5) is a flat surface in R4
having a mean curvature vector of constant length and flat normal connection.
But clearly, M is not a Riemannian product of two curves.
Remark 2. If the mean curvature vector is parallel with respect to the normal
connection, the length of the mean curvature vector is constant. Moreover, the
normal connection is flat if M is in R and not minimal. A flat surface in R4
with parallel mean curvature vector is locally the Riemannian product of two
circles [4].
Remark 3. There are many compact flat surfaces in R with flat normal con-
nection which are not congruent to the Riemannian product of two plane curves
[3, 5].
Remark 4. A similar problem is studied in [ 1] for surfaces of positive constant
curvature.

References

1. K. Enomoto, Umbilical points on surfaces in WN, Nagoya Math. J. 100 (1985), 135-143.
2. _, The Gauss image of flat surfaces in K4 , Kodai Math. J. 9 (1986), 19-32.
3. _, Global properties of the Gauss image of flat surfaces in K , Kodai Math. J. 10 ( 1987),
272-284.
4. D. A. Hoffmann, Surfaces of constant mean curvature in manifolds of constant curvature, J.
Differential Geom. 8 (1973), 161-176.
5. U. Pinkall, Hopf tori in S3 , Invent. Math. 81 (1985), 379-386.
6. H. Reckziegel, Completeness of curvature surfaces of an isometric immersion, J. Differential
Geom. 14 (1979), 7-20.

Faculty of Industrial Science and Technology, Science University of Tokyo, Os-

hamanbe, hokkaido, 049-35 japan

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use