Mh. Math. 127, 1±14 (1999)
Real Hypersurfaces in Complex Two-Plane Grassmannians
By
JuÈrgen Berndt1; andYoung Jin Suh2;
1University of Hull, United Kingdom
2Kyungpook National University, Taegu, South Korea (Received 13 November 1996; in revised form 3 March 1997)
Abstract.The complex two-plane GrassmannianG2 Cm2in equipped with both a KaÈhler and a quaternionic KaÈhler structure. By applying these two structures to the normal bundle of a real hypersurfaceMinG2 Cm2one gets a one- and a three-dimensional distribution onM. We classify all real hypersurfacesMinG2 Cm2,m53, for which these two distributions are invariant under the shape operator ofM.
1. Introduction
In Riemannian geometry, the general problem in submanifold theory is to determine or describe the submanifolds of a given Riemannian manifold which satisfy certain geometrical data. Submanifold theory in Euclidean spaces is classical, in spaces of constant curvature well-established. In more general ambient spaces one encounters two basic problems. On the one hand, the fundamental equations for submanifolds become rather complicated and are dif®cult to handle.
On the other hand, it is not easy to decide what data are natural to consider, they should somehow be related to the geometrical structure of the ambient space.
To give an example, consider a real hypersurface M in a complex projective spaceCPm. The Codazzi equation is rather complicated in the generic case, but simpli®es considerably when requiring that the one-dimensional distribution J ?MonMobtained by applying the KaÈhler structure JofCPm to the normal bundle ?M of M is invariant under the shape operator A of M. The condition AJ ?M J ?M appears to be rather natural, and in fact there is a well- established theory for such hypersurfaces. Any tube around a complex sub- manifold inCPm satis®es this geometrical condition. T. E. CECIL and P. J. RYAN
proved in [4] that these tubes are essentially characterized by this feature. Here the word essentially refers to some additional condition on the focal map. So, roughly speaking, the theory of real hypersurfaces inCPmwithAJ ?M J ?Mis the theory of tubes around complex submanifolds inCPm.
1991 Mathematics Subject Classi®cation: 53B25; 53C15, 53C35
Key words: Complex Grassmannians, real hypersurfaces, tubes, shape operator, KaÈhler structure, quaternionic KaÈhler structurePartially supported by DFG Grant Be 1811/1-1Partially supported by BSRI-97-1404 and TGRC-KOSEF
The analogous question in the quaternionic projective space HPm leads to a surprise. The corresponding geometrical feature is that the three-dimensional distribution J ?Mon M, which is obtained by applying the almost Hermitian structures in the quaternionic KaÈhler structureJofHPmto?M, is invariant under A. In fact, every tube around a quaternionic submanifold of HPm has this geometrical feature. (Note that by a result of D. V. ALEKSEEVSKII [1] such a quaternionic submanifold is necessarily totally geodesic.) But the converse is not true. The ®rst author proved in [2] that also every tube around a totally geodesic CPm inHPm satis®es AJ ?M J ?M, and that there are no other ones. So the real hypersurfaces inHPm withAJ ?M J ?M are precisely the tubes around totally geodesicHPk;k2 f0;. . .;mÿ1g, andCPm.
In this paper we study the analogous question in the complex Grassmann manifold G2 Cm2 of all two-dimensional linear subspaces in Cm2. This Riemannian symmetric space has a remarkable geometrical structure. It is the unique compact, KaÈhler, quaternionic KaÈhler manifold with positive scalar curvature. So, inG2 Cm2we have the two natural geometrical conditions for real hypersurfaces that J ?M and J ?M are invariant under the shape operator.
The main result of this paper is the classi®cation of all real hypersurfaces in G2 Cm2satisfying both conditions.
Theorem 1.Let M be a connected real hypersurface in G2 Cm2;m53. Then both J ?MandJ ?Mare invariant under the shape operator of M if and only if (1) M is an open part of a tube around a totally geodesic G2 Cm1 in G2 Cm2, or
(2) m is even, say m2n, and M is an open part of a tube around a totally geodesicHPn in G2 Cm2:
Any tube aroundG2 Cm1has four (resp. three for a particular radius) distinct constant principal curvatures and might also be regarded as a tube around the focal set ofG2 Cm1in G2 Cm2, which is a totally geodesicCPm. Any tube around HPn has ®ve distinct constant principal curvatures, and the other focal set of the tube is a complex hypersurface inG2 Cm2which is a Riemannian homogeneous space isomorphic to Sp n1= U 2 Sp nÿ1. The two families of tubes together with their focal sets are just the orbits of the isometric actions of the subgroups S U 1 U m1and Sp n1of SU m2, respectively.
In [3] the ®rst author proved that any tube as in Theorem 1 satis®es the geometrical hypotheses in that theorem. So in this paper we are concerned with the proof of the converse statement, which is much more complicated. The paper is organized as follows. In Section 2 we recall basic facts aboutG2 Cm2. In Section 3 we study thoroughly the Codazzi equation for real hypersurfaces inG2 Cm2.
By using the hypotheses this leads to two cases which will be studied separately in Sections 4 and 5.
2. Riemannian Geometry ofG2 Cm2
In this section we summarize basic material about G2 Cm2, for details we refer to [3]. ByG2 Cm2we denote the set of all complex two-dimensional linear
subspaces inCm2. The special unitary groupGSU m2acts transitively on G2 Cm2 with stabilizer isomorphic to K S U 2 U m G. Then G2 Cm2 can be identi®ed with the homogeneous space G=K, which we equip with the unique analytic structure for which the natural action ofGonG2 Cm2 becomes analytic. Denote bygandkthe Lie algebra ofGandK, respectively, and bymthe orthogonal complement ofkingwith respect to the Cartan-Killing form Bofg. Thengkmis an Ad(K)-invariant reductive decomposition of g. We put oeK and identify ToG2 Cm2 with m in the usual manner. Since B is negative de®nite ong, its negative restricted to mmyields a positive definite inner product onm. By Ad(K)-invariance ofBthis inner product can be extended to a G-invariant Riemannian metric g on G2 Cm2. In this way G2 Cm2 becomes a Riemannian homogeneous space, even a Riemannian symmetric space.
For computational reasons we normalize g such that the maximal sectional curvature of G2 Cm2;g is eight. Since G2 C3 is isometric to the two- dimensional complex projective space CP2 with constant holomorphic sectional curvature eight we will assume m52 from now on. Note that the isomorphism Spin 6 ' SU 4 yields an isometry between G2 C4 and the real Grassmann manifoldG2 R6of oriented two-dimensional linear subspaces ofR6.
The Lie algebrak has the direct sum decomposition ksu m su 2 R, whereR is the center of k. Viewingk as the holonomy algebra ofG2 Cm2, the center R induces a KaÈhler structure J and the su(2)-part a quaternionic KaÈhler structure J on G2 Cm2. If J1 is any almost Hermitian structure in J, then J J1 J1J, and J J1 is a symmetric endomorphism with J J12I and tr J J1 0. This fact will be used frequently throughout this paper.
A canonical local basisJ1;J2;J3 ofJconsists of three local almost Hermitian structures J in J such that JJ1 J2 ÿJ1J, where the index is taken modulo three. SinceJis parallel with respect to the Riemannian connectionr of G2 Cm2;g, there exist for any canonical local basisJ1;J2;J3 ofJthree local one-formsq1;q2;q3 such that
rXJ q2 XJ1ÿq1 XJ2
for all vector ®eldsXonG2 Cm2. Also this fact will be used frequently.
Let p2G2 Cm2 and W a subspace of TpG2 Cm2. We say that W is a quaternionic subspace ofTpG2 Cm2ifJ1W W for all J1 2 Jp. And we say thatW is a real subspace of TpG2 Cm2 ifJW ? W. A quaternionic resp. real submanifold of G2 Cm2 is a submanifold all of whose tangent spaces are quaternionic resp. real subspaces of the corresponding tangent spaces of G2 Cm2.
The Riemannian curvature tensorR ofG2 Cm2is locally given by R X; YZg Y;ZXÿg X;ZYg JY;ZJXÿg JX;ZJYÿ2g JX;YJZ
X3
1
g JY;ZJXÿg JX;ZJYÿ2g JX;YJZ
X3
1
g JJY;ZJJXÿg JJX;ZJJY;
whereJ1;J2;J3 is any canonical local basis ofJ. A non-zero tangent vectorXof G2 Cm2is said to be singular ifXis tangent to more than one ¯at ofG2 Cm2:
In G2 Cm2 there are two types of singular tangent vectors X which are characterized by the propertiesJX?JXandJX2JX. We will have to compute explicitly Jacobi vector ®elds along geodesics whose tangent vectors are all singular. For this we need the eigenvalues and eigenspaces of the Jacobi operator RX :R ; XX. LetXbe a unit vector tangent toG2 Cm2. IfJX?JXthen the eigenvalues and eigenspaces ofRX are
0 RXJJX 1 HCX? 4 RJXJX;
where HCXRXRJXJXJJX. If JX 2 JX, there exists an almost Hermitian structure J1 in J such that JXJ1X. Then the eigenvalues and eigenspaces ofRX are
0 RX fYjY ?HX;JY ÿJ1Yg 2 C?X fYjY?HX;JYJ1Yg 8 RJX;
whereCXandHXdenote the complex and quaternionic span ofX, respectively, andC?X is the orthogonal complement ofCXin HX.
3. The Codazzi Equation for Real Hypersurfaces inG2 Cm2 In this section we derive some basic formulae from the Codazzi equation for a real hypersurface inG2 Cm2.
LetMbe a real hypersurface ofG2 Cm2, that is, a hypersurface ofG2 Cm2 with real codimension one. The induced Riemannian metric on M will also be denoted by g, and r denotes the Riemannian connection of M;g. Let N be a local unit normal ®eld ofMandAthe shape operator ofMwith respect toN. The KaÈhler structureJ ofG2 Cm2 induces onM an almost contact metric structure ; ; ;g. Furthermore, letJ1;J2;J3 be a canonical local basis ofJ. Then eachJ
induces an almost contact metric structure ; ; ;g onM. Using the above expression forR, the Codazzi equation becomes
rXAYÿ rYAX XYÿ YXÿ2g X;Y
X3
1
XYÿ YXÿ2g X;Y
X3
1
XYÿ YX
X3
1
X Y ÿ Y X:
The following identities can be proved in a straightforward manner and are used frequently in subsequent calculations:
1 ÿ2; 12; ; X X:
LetDbe the maximal quaternionic subbundle of the tangent bundleTMofMand D?the orthogonal complement ofDinTM. Now we assume thatAD D. Then D?is also invariant underA, and by a suitable choice ofJ1;J2;J3the vector ®elds are principal curvature vectors everywhere, sayA for1;2;3. The Codazzi equation then implies
2 X Y ÿ2 Y X ÿ2g X;Y
21 X2 Y ÿ21 Y2 X ÿ2g X;Y
21 X2 Y ÿ21 Y2 X
g rXAYÿ rYAX;
g rXA;Y ÿg rYA;X
X Y ÿ Y X
ÿ1 q2 X1 Y ÿq2 Y1 X
ÿ ÿ2 q1 X2 Y ÿq1 Y2 X
g AAX;Y ÿ2g AAX;Y:
PuttingX in this equation yields
Y Y ÿ1q2 1 Y ÿ ÿ2q1 2 Y
ÿ4 Y 2 11 Y 2 22 Y:
Inserting this and the corresponding equation forXinto the previous one implies 2 X Y ÿ2 Y X ÿ2g X;Y
21 X2 Y ÿ21 Y2 X ÿ2g X;Y
21 X2 Y ÿ21 Y2 X
2 1 1 X Y ÿ1 Y X
2 2 2 X Y ÿ2 Y X
4 Y X ÿ X Y
ÿ1 q2 1 X Y ÿ1 Y X
q2 X1 Y ÿq2 Y1 X
ÿ ÿ2 q1 2 X Y ÿ2 Y X
q1 X2 Y ÿq1 Y2 X
g AAX;Y ÿ2g AAX;Y:
This implies
Lemma 1. If A and X 2 D with AXX, then 0 2ÿAXÿ 2X
ÿ 21 1 X 22 2 X ÿ4 X
ÿ ÿ1q2 X1 ÿ2q1 X2
ÿ2 Xÿ2 Xÿ2 X22 X1ÿ21 X2: Now we assume that A instead of AD D and denote by ? the orthogonal complement of the real span ofin TM. Taking inner product of the Codazzi equation withyields
ÿ2g X;Y 2X3
1
X Y ÿ Y X ÿg X;Y
g rXAYÿ rYAX;
g rXA;Y ÿg rYA;X
X Y ÿ Y X g AAX;Y ÿ2g AAX;Y:
PuttingX implies
Y Y ÿ4X3
1
Y:
Inserting this and the corresponding equation for X into the previous equation gives
ÿ2g X;Y 2X3
1
X Y ÿ Y X ÿg X;Y
4X3
1
X Y ÿ Y X g AAX;Y
ÿ2g AAX;Y:
From this we easily derive
Lemma 2. If Aand X 2? with AXX, then 0 2ÿAXÿ 2X
2X3
1
2 Xÿ Xÿ X ÿ X:
Next, we assume again AD D. The 1-component of the equation in Lemma 1 gives
0 ÿ1q2 X 4 1 X 2 1 X ÿ2 2 X;
whereas the2-component leads to
0 ÿ2q1 X ÿ4 2 X ÿ2 2 X ÿ2 1 X:
Both equations hold for any index. Raising the index of the second equation by one and then combining with the ®rst equation yields
2 2 X 1 X ÿ 1 X:
We now assume in addition thatis a principal curvature vector ofMeverywhere, sayA. We may write
XX ZZ
with suitable unit vector ®eldsX2DandZ2D?. SinceDandD? are invariant underA, we have AX2DandAZ2D?. Thus
XX ZZA XAX ZAZ
implies X 0, Z 0, or AXX and AZ Z. Suppose that the last possibility holds. Without loss of generality we may assume that the canonical local basis is chosen in such a way thatZ3. Then 1 0 2, and we get
2 Z X 2 3 X 12 X ÿ 21 X 0:
Therefore we necessarily have X 0 or Z 0. But this just means thatis either a vector ®eld inD? or inD. We summarize this in
Proposition 1.Let M be a real hypersurface of G2 Cm2;m53. Suppose that AD Dand A. Then is tangent toDor to D? everywhere.
From now on we assume that M is a connected real hypersurface of G2 Cm2;m53, withAD DandA. According to Proposition 1 we have to consider two cases. This will be done separately in the next two sections.
4. The Case:Is Tangent toD
In this section we assume that is tangent toD. Then the unit normalN is a singular tangent vector ofG2 Cm2of typeJN ?JN and the vectors; 1; 2; 3; 1; 2; 3 are orthonormal.
Putting X in Lemma 2, yields 0 2ÿAÿ 4. If 2 ;then 04 422, which is impossible. Hence 2ÿ60.
InsertingXin Lemma 1 gives
0 2ÿAÿ 4ÿ ÿ1q2 1
ÿ2q1 2:
Since is perpendicular to1 and2, we may now conclude Lemma 3. For each we have
A with 4
2ÿ4
2ÿ: In particular, we have either4 0or:
Next, from Lemma 1, we obtain withXthe equation 0 2ÿAÿ 2
ÿ ÿ1q2 1 ÿ2q1 2
2;ÿ22;121;2:
Putting this implies 0 2ÿAÿ 4. If 2, then 04 422, which is impossible. Thus, comparing withAwe get
Lemma 4. For each we have
4
2ÿ:
Taking as index1 we obtain 0 21ÿA2ÿ12: If 21, then 01 and therefore 0. Lemma 3 and 0 impliesÿ=2 and hence0, which contradicts2= 60 in Lemma 4.
Thus, we have 21ÿ 60 and therefore Lemma 5. For each we have
2 1
21ÿ:
Using successively Lemma 3, Lemma 5, and again Lemma 3, we derive 42
22ÿ2 1
21ÿ 41
821ÿ21
and therefore
2ÿ8 12 2212
412ÿ8 12 ÿ32:
This formula holds for any index. Subtracting the corresponding equation obtained by replacingby1 gives 0 28 1ÿ2. As above,20 leads to a contradiction. Therefore, 1; 2; 3 must be equal, say 12 3 :. Lemma 3 shows that also 1; 2; 3 are equal, say 123:. From Lemma 5 we then get ÿ 0. First we suppose that. According to Lemma 3 we have either or 40. In the
®rst case we have, which contradicts Lemma 4. In the second case we get 0 from Lemma 3, which contradicts 40. Therefore 6
and hence0:We summarize this in
Lemma 6. For each we have A and A 0, where is determined by40:
Next, letXbe a principal curvature vector ofMorthogonal toRJJJ, sayAXX. Note thatXorthogonal toRJJJjust meansX 2 HC?. Then Lemma 1 implies for eachthe equation 0 2ÿAXÿ 2X:
If 2, then 02222, which is impossible. Hence 2ÿ60
and we obtainAjJXidJX with 22ÿ. ReplacingX by, for instance, 1X yields AjJ1X22ÿidJ1X. Since 2X2JX\J1X, we get 22ÿ; and sinceX2J1Xthis implies and hence22ÿ:From this we conclude
Lemma 7.Aj HC? has at most two distinct eigenvalues, each of which is a solution of x2ÿxÿ10. The corresponding eigenspaces are quaternionic.
For X as above we derive from Lemma 2 the equation 0 2ÿAXÿ ÿ 2X. One easily sees that 2ÿ60, whence AXX with 22ÿ:By means of Lemma 7 we know that2ÿÿ10. If, then the preceding formula implies2ÿÿ10, and so60 and, which contradicts 04in Lemma 6. Therefore we must have6, and together with Lemma 7 we conclude
Lemma 8. Aj HC? has two distinct eigenvalues and , which are the solutions of x2ÿxÿ10. The corresponding eigenspaces T and T are quaternionic and real, that is,
JTT; JTT; JTT:
In particular, the quaternionic dimension m of G2 Cm2is even, say m2n.
Since is tangent to D, we know from a formula obtained for the proof of Lemma 2 that
grad :
So for the Hessian ofwe get
hess X;Y g rXgrad;Y g rX ;Y
X Y g rX;Y
X Y g AX;Y:
By symmetry of the Hessian this implies
0 X Y ÿ Y X g AAX;Y:
For X this gives Y Y. Inserting this and the corresponding equation forXinto the previous one yields
0 g AAX;Y:
Suppose thatAAvanishes at some pointpofM. Then, atp, forX 2 HC? withAXXwe getAX ÿX. But, sinceis a solution ofx2ÿxÿ10 and 60, the number ÿ cannot be a solution of that equation also, which contradicts Lemma 8. Hence we must haveAA60 at each point ofM. This yields0 and hence grad 0, that is, is constant onM. It follows from Lemmata 6 and 8 that all principal curvatures ofMare constant. Since 04 we have 60. Without loss of generality we may assume >0. Then there exists some r 20; =4 so that 2 cot 2r. Then 04 gives ÿ2 tan 2r. The solutions of x2ÿxÿ10 are then cot r and ÿtan r. We summarize this discussion in
Proposition 2.Let M be a connected real hypersurface of G2 Cm2. Suppose that AD D, A, andis tangent toD. Then the quaternionic dimension m of G2 Cm2 is even, say m2n; and M has ®ve distinct constant principal curvatures
ÿ2 tan 2r; 2 cot 2r; 0; cot r; ÿtan r
with some r20; =4. The corresponding multiplicities are
m 1;m 3m ;m 4nÿ4m and the corresponding eigenspaces are
TR ;TJJ ;TJ ;T;T; where
TT HC?; JTT;JTT; JT T:
For p 2 M denote by cp the geodesic in G2 Cm2 with cp 0 p and _
cp 0 Np, and byFthe smooth map
F:M!G2 Cm2; p7!cp r:
Geometrically,Fis the displacement ofMat distancerin direction of the normal
®eldN. For eachp2Mthe differentialdpFofFatpcan be computed by means of Jacobi vector ®elds by
dpF X ZX r:
Here, ZX is the Jacobi vector ®eld along cp with initial values ZX 0 X and ZX0 0 ÿAX. Using the explicit descriptions of the Jacobi operator RN for the caseJN ?JN in Section 2 and of the shape operatorAofMin Proposition 2 we get
ZX r cos 2r ÿ 2 sin 2r
EX r; ifX 2 Tand 2 f; g cos r ÿsin rEX r; ifX 2 Tand 2 f; g
EX r; ifX 2 T;
8>
<
>:
whereEXdenotes the parallel vector ®eld alongcpwith theEX 0 X. This shows that the kernel of dF is TT JNT and that F is of constant rank dim TTT 4n. So, locally, F is a submersion onto a 4n-dimensional submanifoldBofG2 Cm2. Moreover, the tangent space ofBatF(p) is obtained by parallel translation of TTT p HT p, which is a quater- nionic and real subspace ofTpG2 Cm2. Since bothJandJare parallel alongcp, also TF pB is a quaternionic and real subspace of TF pG2 Cm2. Thus B is a quaternionic and real submanifold ofG2 Cm2. SinceBis quaternionic, it is totally geodesic inG2 Cm2[1]. The only quaternionic totally geodesic submanifolds of G2 Cm2;m2n54, of half dimension are G2 Cn2 and HPn [3]. But only HPnis embedded inG2 Cm2as a real submanifold. So we conclude thatBis an open part of a totally geodesic HPn in G2 Cm2. Rigidity of totally geodesic
submanifolds ®nally implies thatMis an open part of the tube with radiusraround a totally geodesicHPn in G2 Cm2. Thus we have proved
Theorem 1A.Let M be a connected real hypersurface of G2 Cm2;m53. If AD D; A, andis tangent toD, then m2n and M is an open part of a tube around a totally geodesicHPn in G2 Cm2.
5. The Case: Is Tangent toD?
We now assume thatis tangent toD?. Then the unit normalN is a singular tangent vector ofG2 Cm2of typeJN 2 JN. So there exists an almost Hermitian structureJ1 2 J such thatJN J1N. Then we have
1; 1; 2 ÿ3; 32; D D:
In the following the indices andwill be either 2 or 3 and distinct from each other.
Inserting X in the equation of Lemma 2 yields 2 ÿA
4: Since 2ÿ0 implies 04 422, which is impossible, we get 2ÿ60 and A 42ÿ. Using A; 2 ÿ3 and3 2 we obtain42ÿ and hence
Lemma 9. 223ÿ 23 ÿ40:
Next, letX 2 DwithA XX. TheD-component of the equation in Lemma 1 is 0 2ÿAXÿ 2X. If 2ÿ 0, then 02
222, which is impossible. Thus we get Lemma 10. If X 2 Dwith AXX, then
2ÿ 60 and AXX with 2
2ÿ:
We now replaceXby2X. Then theD-component of the equation in Lemma 1 with index one is 0 223ÿ 23 ÿ23Xÿ22X. Applying 2 to this equation yields 0 223ÿ 23 ÿ21X2X. Thus we obtain
Lemma 11. If X 2 Dwith AXX, then either
X1X and 223ÿ 23 0 or
X ÿ1X and 223ÿ 23 ÿ40:
Now supposeXsatis®esX1X. Then theD-component of the equation in Lemma 1 with index one is 2ÿA1X 41X. If 2ÿ0, then 04422, which is impossible. Hence we obtain
Lemma 12. If X 2 Dwith AXX andX1X, then 2ÿ60 and A1X4
2ÿ1X:
From Lemma 10 we derive
A1XA23X 3223X 423 2 2ÿ3 423 ÿ2 2ÿ31X and
A1X ÿA32X ÿ 2332X 423 ÿ2 2ÿ3 423 2 2ÿ31X:
Comparing these two equations leads to 423 0 or2 3. We ®rst assume 423 0. Then Lemma 9 shows that 60, and 2 and 3 are the two solutions of the quadratic equation x212xÿ40. If X satis®es X ÿ1X, then 2X satis®es 2X12X. Thus we may choose X so that X1X. From Lemma 12 and the subsequent equation we derive
ÿ1
1XA1X4 2ÿ1X:
Therefore,is a solution of the quadratic equationx26xÿ0. It follows that 2is a solution ofx212xÿ40 and must therefore coincide with2or 3, which is a contradiction to Lemma 10. Thus we have proved
Lemma 13. 2 3 ::
Lemma 9 implies that is a solution of the quadratic equation x2ÿxÿ20. Let X 2 D with AX X. The equation after Lemma 12 shows thatA1X1X. So the eigenspaces ofAjDare1-invariant. Moreover, if X satis®es X1X, then Lemma 12 yields that is also a solution of x2ÿxÿ20. On the other hand, from Lemmata 10 and 13 it follows2 3: Thus we may put:23. According to Lemma 11,satis®es ÿ 0 and hence 2 f0; g. If , then Lemma 10 implies that is a solution of x2ÿx20, which is impossible for0 and contradicts2ÿÿ20 for. Therefore,andare the distinct solutions ofx2ÿxÿ20, that is
; 2 1
2
28
p
; 6:
So ÿ2, and using again Lemma 10 proves that0. We summarize this in Lemma 14.AjDhas exactly two distinct eigenvalues andwith the same multiplicities2mÿ2, and the corresponding eigenspaces T and T satisfy
T fXjX ? H ;JXJ1Xg;
T fXjX ? H ;JX ÿJ1Xg:
We have 0, and and are the distinct solutions of the equation x2ÿxÿ20.
The above inclusions are in fact equalities, because all these sets are vector spaces of the same dimension. As in the case whereis tangent toDwe deduce in an analogous way
grad and 0 g AAX;Y:
Suppose that AA0 at some point p 2 M. Then, at p, we use the 1- invariance of T to derive XAX ÿAX ÿX. This implies 0, which is impossible by means of Lemma 14. Thus AA is non-zero everywhere and we conclude that 0 everywhere. Thus is constant onM, and Lemma 14 implies thatM has constant principal curvatures.
From ÿ2 we see thatand have different sign. We may choose the unit normalNin such a way thatis positive, say
p2
cot
p2
rwith some r 20; =
p8
. We use Lemma 14 to compute
p8
cot
p8
r and ÿ2= ÿ
p2
tan
p2
r. Altogether we have thus proved
Proposition 3.Let M be a connected real hypersurface of G2 Cm2. Suppose that AD D;A, and is tangent to D?. Let J1 2 J be the almost Hermitian structure such that JNJ1N. Then M has three if r=2
p8
or four (otherwise) distinct constant principal curvatures
p8
cot
p8
r;
p2
cot
p2
r; ÿ
p2
tan
p2
r; 0 with some r 20; =
p8
. The corresponding multiplicities are m 1; m 2;m 2mÿ2m and for the corresponding eigenspaces we have
TRRJN; T C?C?N;
TfXjX?H ;JXJ1Xg;
TfXjX?H ;JX ÿJ1Xg:
We de®necp,F,ZXandEXas in the preceding section. In the present situation we get
ZX r
cos
p8
r ÿ
8 p sin
p8 r
EX r; ifX 2 T
cos
p2
r ÿ
p2sin
p2 r
EX r; ifX 2 T and 2 f; g
EX r; ifX 2 T;
8>
>>
><
>>
>>
:
So the kernel of dF is T TRJN C?NJN. Hence F is of constant rank dim T T 4mÿ4 and locally a submersion onto a submanifoldBof G2 Cm2. As T T HN? is quaternionic, Bis a quaternionic hypersur- face ofG2 Cm2and hence totally geodesic. The only quaternionic hypersurfaces in G2 Cm2;m53, are open parts of a totally geodesic G2 Cm1 in G2 Cm2 [3]. As in the preceding section we may now conclude thatMis an open part of the tube with radiusraround a totally geodesicG2 Cm1inG2 Cm2, and we have proved
Theorem 1B. Let M be a connected real hypersurface of G2 Cm2;m53. If AD D; A, and is tangent to D?, then M is an open part of a tube around a totally geodesic G2 Cm1in G2 Cm2.
Theorem 1 now follows from Theorems 1A and 1B and Proposition 1.
References
[1] ALEKSEEVSKIIDV (1968) Compact quaternion spaces. Funct Anal Appl2: 106±114
[2] BERNDTJ (1991) Real hypersurfaces in quaternionic space forms. J Reine Angew Math419: 9±26 [3] BERNDTJ (1997) Riemannian geometry of complex two-plane Grassmannians. Rend Sem Mat
Univ Politec Torino55: 19±83
[4] CECILTE, RYANPJ (1982) Focal sets and real hypersurfaces in complex projective space. Trans Amer Math Soc269: 481±499
J. BERNDT Y. J. SUH
Department of Mathematics Department of Mathematics
University of Hull Kyungpook National University
Hull HU6 7RX Taegu 702-701
United Kingdom South Korea