This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
so Y = ,
-
Since Y is a unit vector, /? = ± - 1 2 ± V 2 2 V( +»' )[(l-»' ) +V'2]' If 5 denotes the shape operator associated to the immersion X, then: -SX2
=
dvY:
V^V
*^U
- 5 J f i = S „ y = <^ - V j 9 ' - v'f3 +
+ < - i-ip1
{y 1 — ip1-tp2
fx)
x + ln'
2^'V 2 /? (1-V2
\-
(V'V + V'2)/? l-^2
165 So, I(X2,X2)
= \{X2,SX2)
= -\{X2,dvY)
n(x1,x2) = i(x 1 ,5x 2 ^V'
= -p
x/T+^
v w
-l(XudvY) ,„' 2
2
(1 + V 2 ) § ( 1 - ^ 2 )
(l + v 2 ) i ( l - v 2 )
-r + (l + v2)5 (l + ^ 2 )5
V'2 + ( V 2 - 1 ) 2 c# (i + v 2 ) ( i - ^ ) In
= I(Xl,Xl)
= I{Xl,SXl)
-f>& - V>'P
=
--[{XhduY)
v'P
VV
1 -
\x) around x0 such that C(X, X, X) = 0 or C(Y, Y, Y) = 0 around x„. The proof of the lemma is completed. Let V be a connections satisfying the assumptions of Theorem 2.2, that is Ric is nondegenerate, I Vffic Iflic^ 0 and V can be realized on an affine sphere. Lemma 2. If Ric is definite, then around each point there is a Ric-isothermal coordinate system (u, v) such that VRic(U, U,U) = 0 and VRic{V, V, V) = -2e, where e = 1 or —1 depending on whether Ric is positive or negative definite. Proof of Lemma 2. Let X, Y be a ffic-orthonormal frame around a fixed point x0. By Lemma 1, we know that X,Y can be chosen in such a way that VRic(X, X, X) = 0 around x0. Then Vffic(Y, Y, Y) + 0 and VRic(X, Y, Y) = 0. We have that VXX = BY and V x Y = CX for some functions B, C. Let VYY = EX + FY. Since VRic{X, X, Y) = -VRic(Y, Y, Y), we get C = -B - 2F, i.e. VXY = -(B + 2F)X. Since VRic(Y, Y, X) = 0 and VRic(Y, X, X) = -VRic{Y, Y, Y), we have VYX = -FX-EY. (Nn,h) is called a biharmonic morphism if it pulls back germs of biharmonic functions on N to germs of biharmonic functions on M. Remark 2.2. 1). By using the local harmonic coordinates in N and the regularity of solu tions of elliptic partial differential equations we know that a biharmonic morphism is always smooth. 2). It is clear from Definition 2.1 that any constant map is a biharmonic morphism. When dimN = 1, say N = R, the biharmonic morphisms tp : (Mm,g) —> R are nothing but the biharmonic functions. Note that the notion of biharmonic functions differs from the notion of p-harmonic func tions. The latter is a special case of p-harmonic maps; these are critical points of the p-energy functional Ep{(j>,U) = - f \d<j>\"dx PJ a ) + Ui ° ) +(V v Q ,V ¥ /)fl(V,V v/ ) (10) (N, h) is a biharmonic morphism if and only if ).
222 Since VRic{Y, Y, Y) = -2eF and VRic{Y, Y, Y) ± 0, we have F ± 0. Compute now the Ricci tensor of V. We get Ric{X, X) = XE-B2 (2.7)
E2,
+ YB + FB-
Ric(X, Y) = -XF + 3EF, Ric(Y, Y) = XE + 2YF + YB - B2 - 6F 2 - E2 - 5BF
Hence B + F
(2-8)
YF
XF
E
=W
=W
Set U = F~*X, V = F~*Y. Using (2.8) we obtain [U,V] = 0. One also sees that S/Ric(V, V, V) = -2e. This completes the proof of Lemma 2. Lemma 3. / / Ric is indefinite, then around each point there is a coordinate system (u, v) such that Ric(U, U) = -Ric(V, V), Ric(U, V) = 0 and either VRic{U, U, U) = 0, VRic(V, V,V) = 2 and Ric(U, U) > 0, or VRic(U, U,U) = 0 X7Ric{V, V, V) = - 2 and Ric(V, V) < 0. Proof of Lemma 3. By Lemma 1, around a fixed point x0 can be chosen a smooth local ffic-orlhonormal frame X,Y such that VRic{X, X, X) = 0. Since | VRic \mc¥z 0, VRic(Y,Y,Y) / 0. Assume that Ric(X,X) > 0, i.e. Ric(X,X) = 1. Using the equalities 0 = VRic{X, X, X) = VRic(X, Y, Y) = VRic(Y, X, Y) = 0, VRic{Y, Y, Y) = VRic{Y, X, X) = Vffic(X, Y, X), we obtain
VXX
= BY,
VXY
= {B-2F)X,
VYX
= -FX + EY,
V F F = EX + FY
for some functions B, E, F where F / 0. Then VRic{Y, Y, Y) = IF. Computing the Ricci tensor we obtain
(2.9)
Ric(X, Y) = -XF
- SEF,
Ric(X, X) = -XE
+ YB + FB + B2 - E2,
Ric{Y, Y) = XE-YB
+ 2YF + 5BF - B2 - 6F2 + E2.
From these formulas we get XF -F~ = ~ ^
YF
„ „ JF=F~B-
Set U = F"3 X, V = F~*Y. One now easily computes that [U, V] = 0 and VRic(V, V, V) . 2. The case where Ric(X, X) < 0 can be treated analogously.
223 We can now complete the proof of the theorem. Assume first that Ric is definite. Take a coordinate system (u,v) as in Lemma 2. Set $ = Ric(U,U) = Ric(V,V). Let A, B, C, D, E, F be the Christoffel symbols of V relative to (u, v), as in (1.15). Using the symmetry of Vffic, the assumptions that tr jj; c Vffic(-, •, W) = 0, VRic(U, U, U) = 0 and VRic{V, V, V) = -2s, we obtain
(2.10)
VVU
= AU + BV,
VVV
= CU + AV,
yvV
=
-AU-BV,
and
A=l(Jn|*|) U ) (2.11)
B = -iin(|*|)„-J, C=ijn(|*|)„-J.
Using (1.16) and the above formulas we get
$ = Bv - Du + - | - = --[Jn(l $ |) u u + ln(\ $ |)„„] - ^ To prove the converse it is sufficient to check by a direct computation t h a t the connection given by (2.2), where $ satisfies (2.3) is affine spherical. Assume now that Ric is indefinite and (u, v) is a coordinate system as in Lemma 3. Set $ = Ric(U, U). Using the equalities 0 = VRic(U, U, U) = VRic{V, V, U) = VRic{U, V, V), VRic{V, U, U) = VRic{U, V, U) = VRic{V, V, V) = 2e we get
(2.12)
VVU
= AU + BV,
VVV
= CU + AV,
VVV
= AU + BV,
where A=
\(ln\$\)„,
* = i(Jn I * |)„ + J ,
224
C=i(ln|$|)„-|. Using now (1.16) and the above formulas, we obtain that $ satisfies the equality
$ = i[(Jn | $ | ) w - (In | $ |)u„] + ^ , which is equivalent to (2.5). The converse, as in case of definite Ric, can be checked by a direct computation. 3. Realizations on affine minimal surfaces. The following lemma is crucial for our considerations. Lemma 3.1. Let f : M2 —> R 3 be a Blaschke surface with nowhere flat induced connection V and affine metric h. Assume that the Ricci tensor Ric of V has constant non-zero rank. 1) If Ric is nondegenerate indefinite, then f is affine minimal if and only if the asymp totic distributions of Ric are h-orthogonal. 2) If Ric is nondegenerate definite, then f is minimal if and only if h is indefinite and asymptotic distributions of h are Ric-orthogonal. 3) If Ric has rank 1, then f is minimal if and only if h is indefinite and ker Ric is an asymptotic distribution of h. Proof. Consider 1). Assume that / is minimal. If / is convex, then for each x € M 2 there is an h-orthonormal basis X,Y of TXM2 such that SX = XX, SY = —XY, where A ^ 0. Hence Ric(X, Y) = 0, Ric(X, X) = -A and Ric(Y, Y) = A, i.e. Ric is indefinite. Moreover, U = X+Y, V = X-Y are asymptotic for Ric. We have SU = XV and SV = XU and consequently 0 = Ric(U, U) = -h(SU, U) = -Xh(U, V), that is, h(U, V) = 0. Consider now the case where / is minimal, h is indefinite, r k £ = 2 and S is nondiagonalizable. Let U, V be asymptotic for h. Set SU = aU + f3V, SV = yU - aV. Since S is non-diagonalizable, -y/3 < 0. In view of the Ricci equation we get a = 0. Hence Ric(U, U) = —/3h(U, V) and Ric(V, V) = -yh(U, V), which means that Ric is indefinite. Let now U, V be flic-asymptotic. Set SU = aU + fiV, SV = -yU - aV. We have a 2 + /?7 ^ 0. From the Ricci equation we obtain 2ah(U, V) = yh(U, U) - Ph{V, V). We also have 0 = -Ric{U, U) = h(SU, U) = ah(U, U) + fih(V, U), 0 = -Ric(V, V) = h(SV, V) = yh{U, V) - ah{V, V). Using the last three formulas one gets yPh(U, V) = aph(V, V) = a[yh{U, U) - 2ah(U, V)] = -y/3h{U, V) - 2a2h{U, V).
225 Therefore (1f] +
a2)h(U,V)=0,
that is, h(U, V) = 0. Assume now that Ric is nondegenerate indefinite and the asymptotic distributions of Ric are /i-orthogonal. Let U, V span the asymptotic distributions. Then Ric(U, V) ^ 0 and h(U, U) / 0, h(V, V) ^ 0. By the Gauss equation we have SV = -(h{U, U))-xRic{U, V)U,
SU = -{h{V, V))~lRic(U, V)V.
It follows that t r S = 0. Moreover, h is definite and S is diagonalizable if detS < 0, and h is indefinite and S is non-diagonalizable if det S > 0. 2) From the above considerations we know that if / is minimal and Ric is nondegenerate definite, then h is indefinite and S is diagonalizable. Hence h and S are simultaneously diagonalizable. Let X, Y be an ft-orthonormal frame diagonalizing S. Ten U = X + Y, V = X - Y are /i-asymptotic and SU = XV, SV = XU. We now have 0 = -h{SU, V) = Ric(U,V). Assume now that Ric is nondegenerate definite, h is indefinite and the ft-asymptotic distributions are ffic-orthogonal. Let X, Y be ^-asymptotic and SX = aX + /3Y, SY = "fX + 6Y. By the Ricci equation a = 6. We have 0 = Ric(X,Y)
=trSh{X,Y)-h{SX,Y)
= 2ah(X, Y) - ah(X, Y),
that is, t r S = 2a = 0. 3) If / is minimal and rkRic = 1, then rkS = 1 and S is non-diagonalizable. Hence im S = ker S. Moreover, since Ric = —h(S-,-), ker Ric = kerS*. Let X G ker S. By the Ricci equation we get h(X, SY) = 0 for all Y. Hence h(X, X) = 0, that is, ker Ric is fr-asymptotic. Conversely, let rkRic = 1 and ker Ric be h-asymptotic. Take X £ ker Ric and Y another h-asymptotic vector. Then h(X, Y) ^ 0. If we set SX = aX + /3Y, SY = yX + 5Y, then,using again the Ricci equation, we get a = S. Therefore 0 = Ric(X, Y) = 2ah(X, Y) - h(aX + /3Y, Y) = ah(X, Y), and consequently tr S = 0. The proof of Lemma 3.1 is completed. We shall now prove three theorems corresponding to the three cases in Lemma 3.1. Theorem 3.2. Let V be a torsion-free Ricci-symmetric connection on M2. If rk Ric = 1 on M2, then V is locally realizable on a minimal surface in R 3 if and only if V is projectively flat arid ker Ric is V -parallel. Proof. Assume that V is realizable on a minimal surface with the Blaschke metric h. By Lemma 3.1 we know that h is indefinite and ker Ric is an asymptotic distribution for h. Let (it, v) be a coordinate system such that U, V span the asymptotic distributions and U £ ker Ric. By (1.12), we have Tl2(U, V) = 0. On the other hand, S7Ric(W, U, 11) = 0 for every W. Therefore, by (1.11), we have VRic(U, U, V) = 0. It implies that Ric{VrrU, V) = 0, i.e. ker Ric is totally geodesic.
226 Let 9 = h(U, V) and A, B, C, D, E, F be the Christoffel symbols of V as in (1.15). By the apolarity, we get (In \ 9 \)u = A + D, (In | 9 |)„ = C + F. Hence Vh(U, V, U) = 9U- h(VvV, U) - h(VvU, V) = 9U-(D + A)9 = 0. By the first Codazzi equation we have Vh(V, U, U) = 0. Therefore D = 0, which means that ker Ric is parallel relative to V. Consequently VRic(V, V, U) = 0 Looking now again at 1Z2(U, V) = 0, one sees that VRic(U, V, V) = 0. We have proved that VRic(U, V, V) = VRic(V, V, U) = VRic(U, U, V) - VRic(V, U, U) = 0. In particular, VRic is symmetric. Assume now that ker Ric is V-parallel and V is projectively fiat. Let (u, v) be such a coordinate system that U spans kerffic and VuU — 0. We have VyV = CU and VVV = EU + FV for some functions C,E,F. Obviously VRic(W,U,U) = 0 for any W. Using the parallelity of ker Ric one sees that VRic(V, V, U) = 0. Hence, by the projective flatness of V, S7Ric(X, Y,Z) = 0 if at least one of the arguments X, Y, Z lies in ker Ric. It follows that 1Z2(U, V) = 0 and 1Zi(U, V) = 0. Let 0 be a volume element parallel relative to V. Set 9 = Q(U, V). Since A = D = 0, the condition V 6 = 0 implies that 6U = 0 and (ln\d\)v = C + F. Define h by h(U, U) = 0,
h(U, V) = 9,
h(V, V) = * ,
where * is any function satisfying the equation <SU = 2C9.
Since 1Zi(U, V) = 0 and H2(U, V) = 0, the equality (1.12) is satisfied. We also compute Vh(U, V,U) = 6U = 0 = Vh(V, U, U), Vh(U, V, V) = V>u-2Ce = $ = 9v-(C + F)9 = Vh(V, U, V). Obviously O^. = O. By Theorem 1.1, V, h can be realized on a Blaschke surface. By Lemma 3.1, the realization is minimal. Theorem 3.3. Assume that V is a torsion-free connection on M2 with symmetric nondegenerate indefinite Ricci tensor. Let (u, v) be a coordinate system spanning the Ricasymptotic distributions, 0 - a volume element parallel relative to V and A, B, C, D, E, F Christoffel symbols of V relative to (u,v), as in (1.15). Assume that the vector fields Tli(U,V) andTZ3(U,V) (given by (1.11)) are nowhere 0. The connection V is realizable on a minimal surface in R 3 if and only ifR,z(U, V) = 01Zi (U, V) for some function /3 and the following equations are satisfied: (3.1)
F-0B=±(ln\p\)v,
(3-2)
J ~ A = Vln I P D«-
227 If P > 0, then V is realizable on a non-convex surface with non-diagonalizable operator. If j3 < 0, then V is realizable on a convex surface.
shape
Proof. Assume first that V is an induced connection on a minimal surface. Set /in = h(U, U) and h22 = h(V, V), where h is the Blaschke metric on the surface. By Lemma 3.1 we know t h a t h{U, V) = 0. Hence, by (1.12), h22Tll(U, V) + hnTl3(U, V) = 0. If we define /3 by the equality K3{U, V) = PKi(U, V), then p ^ 0 everywhere and h22 = -fihn. Let 6 = eh(U,V). Then e2 =
(3.3)
h\l\p\.
From the apolarity condition we obtain (3.4)
A + D = {ln\6\)u,
C + F = (ln\ 8 \)v.
The first Codazzi equation yields
(In | hn
\)V = C + 0B,
ln{\ p \)u + {In \ / . „ | ) u = D + j .
Combining this with (3.3) and (3.4) we obtain the desired equalities (3.1) and (3.2). Conversely, let V be a connection on an abstract manifold satisfying the assumptions listed in the theorem. Let 0 be a volume element parallel relative to V. Set 9 = @(U, V). Let hn, h22 be functions satisfying the conditions h\x \ f) \= 82, h22 = ~phn. Define h by h(U,V) = 0, h{U,U)=hu, h(V,V)=h22. By a straightforward computation we check that V, h satisfy the assumptions of Theorem 1.1. By Lemma 3.1 we know that the realization is on a minimal surface. The proof is completed. T h e o r e m 3.4. Assume that V is a torsion-free connection with symmetric nondegenerate definite Ricci tensor on a two-dimensional manifold M2. Let (u,v) be an isothermal coordinate system relative to Ric and A,B,C,D,E, F Christoffel symbols o / V relative to (u,v), as in (1.15). Let 0 be a volume element parallel relative to V. Define (3.6)
(3.7)
e = e(U,V),
a = A-D,
$ =
b = -B-C,
Ric(U,U),
c = F-C,
d = D + E.
Assume moreover that the vector fields Tli(U, V) —TZz(U, V), TZ2(U, V) are nowhere zero. The connection V is realizable on a minimal surface if and only ifTZ2(U,V), TZi(U,V)7^3(U,V) are linearly dependent at each point, that is, (3.8)
(c+6)2
+
(a
_d)2
=
;
„(|Jj)2+(,n|£|)2
228 and
a* (3.9)
_
— = Pv]/ + Q#,
where (3.10)
cos(j>TZ2(U, V) - sin[ni{U, V) - 1l3(U, V)] = 0,
(3.11)
(3.12)
z = u + iv,
\t = 8(cos
P = hL{a + d)-i{c~b)},
Q=hi{a-d)-i{c
+ b)}.
Proof. We have ll^U, V) = ${-(C
+ B)U - [(In | $ |) u - 2A]V},
1Z2(U, V) = ${[(Jn | $ |)„ - 3D - E]U - [(In | $ |)„ - 3C - B]V}, ■R3(U, V) = ${[(ln | $ |)„ - 2F]t/ + (D + E)V}. Using also the condition VO = 0, that is, A + D = (ln\9\)u,
C+F =
(ln\6\)v,
we obtain
(3.13) n, (U, V) - Tl3(U, V) = ${[-(/n I | ) „ + (c + b)]U + [-(In | f |)„ + (a - d)]V),
(3.14)
H2(U, V) = *{[(/„ | £ |) u + (o - d)]U + [-(In | |
|)„ - (c + 6)] V}.
By virtue of (3.13) (3.14) it is clear that TZi — TZ3 and 7?. 2 are linearly dependent if and only if (3.8) is satisfied. Assume first that V is an induced connection on a minimal surface. By Lemma 3.1 we know that the Blaschke metric h is indefinite and its asymptotic distributions are ffic-orthogonal. Therefore, there is a function ip such that (3.15)
X = cosipU 4- sinibV Y = —sintpU + cosipV
229 are /i-asymptotic. By (1.12) we have U2(X,Y) = 0. Since 0 and Gh differ by a constant factor, we can assume that and h(X, Y) = 6. On the other hand, by a straightforward computation we obtain that 1l2(X, Y) = 0 if and only if (3.16)
cos2i,n2{U,V)
- sin2i>{'Rl{U,V)-ll3(U,V))
= 0.
Moreover, h(X, Y) = B and h(X, X) = h{Y, Y) = 0 if and only if h(U, V) + h{V, V) = 0, (3.17)
h(U,U) = -9sin2ip, h(U,V) = 8cos2ip.
Set / i u = h(U, U) and ft12 = h(U, V). The symmetry of Vh is equivalent to the equations Uh12 - Vhu + (B + Qhn
+ (D - A)hl2 = 0
(O.lo)
Vh12 + Uhu + (C-F)h12-(D + E)hu = 0 The last equations are equivalent to (3.9) if 2ifi = <j>. Conversely, let V be a connection on a manifold M2 satisfying the assumptions of the theorem. Take <j> defined by (3.10) and h defined by (3.17), where ip = \§. Then h is indefinite and X,Y defined by (3.16) are ft-asymptotic. Moreover, 0 = 0^. The equation (3.9) means that Vh is symmetric. Therefore, all the assumptions of Theorem 1.1 are satisfied and consequently V, h are realizable on a Blaschke surface. By Lemma 3.1, the realization is minimal. The proof is completed.
References. [NS] K. Nomizu, T. Sasaki, Affine Differential Geometry, (1994), Cambridge University Press. [LSZ] A. M. Li, U. Simon, G. Zhao, Global Affine Differential Geometry of Hypersurfaces, de Gruyter -Berlin; New York, 1993. [0]i B. Opozd&,Locally symmetric connections on surfaces, Results in Math., 1481 (1991), 185-191. [0] 2 B. Opozda, A class of protectively flat surfaces, Math. Z. 219 (1995), 77-92. [OJ3 B. Opozda, An intrinsic characterization of developable surfaces, Results Math. 27 (1995), 97-104. [0] 4 B. Opozda, A characterization of affine cylinders, Mh. Math. 121 (1996), 113-124. [OJ5 B. Opozda, On rigidity of affine surfaces, Proc. Amer. Math. Soc, 124 (1996), 2175-2184.
230 [SW] U. Simon , C.P. Wang, Local theory of affine 2-spheres, Proc. Symposia Pure Math ematics, 54 (1993), 585-598. [SI] W. Slebodziiiski, Sur quelques problemes de la theorie des surfaces de I'espace affine, Prace Mat. Fiz., 46 (1939), 291-345.
INSTYTUT MATEMATYKI UJ, UL. REYMONTA 4, 30-059 K R A K O W , POLAND
E-mail address: opozdaSim.uj.edu.pl RECEIVED FEBRUARY 14, 2000
Geometry and Topology of Submanifolds X eds. W. H. Chen et al. (pp. 231-239) © 2000 World Scientific Publishing Co.
231
BIHARMONIC MORPHISMS BETWEEN RIEMANNIAN MANIFOLDS YE-LIN OU* ABSTRACT. In this paper, we study biharmonic morphisms-the maps between Riemannian manifolds which pull back germs of biharmonic functions to germs of biharmonic functions. We obtain characterizations of such maps and some results concerning their links to har monic morphisms and biharmonic maps.
KEYWORDS: Biharmonic morphisms, Harmonic morphisms, Biharmonic maps MATHEMATICAL S U B J E C T CLASSIFICATION (1991): 58E20 , 53C42.
1.
B I H A R M O N I C F U N C T I O N S AND T H E I R A P P L I C A T I O N S
m
Let (M ,g) be a Riemannian manifold with A = -4=^- ("^S w 'gjj) > the Laplace-Beltrami operator on M. Definition 1.1. A function f on M is said to be polyharmonic of order k (an integer > 2) or k-harmonic if
A*/ = 0 fc
1
1
where A = A(A*" ) with A = A . A 2-harmonic function is also called a b i h a r m o n i c or a s t r e s s
function.
Clearly, any harmonic function is biharmonic while there are lots of biharmonic functions, e.g., / = xexpxcosy, x4 — 3x2y2, which are not harmonic. Biharmonic functions play a fundamental role in elasticity and hydrodynamics. It is wellknown [8] that to solve the plane stress problem, i.e., to determine the stress distribution in a two-dimensional problem one has to solve the equations of equilibrium together with the boundary and some compactibility conditions. These equations, when the weight of the body is the only body force, can be written as
dx d<Jy a + oy EP_ dx2
dy drxy a ox +P9 = ° cf_ dy2
where ox, oy and rxy are components of stress at a point of the thin plate. In order to integrate these equations, G. B. Airy in 1862 introduced the so-called stress function <j>(x, y) "Supported by "The Special Funding for the Young Talents" Guangxi, P. R, China (1998-2000). This paper is in final form and no version of it will be submitted for publication elsewhere.
232 such that
a =
°
w~p9y
d2
,
d*<j>
d4
d4<j>
Therefore, the solution of a two dimensional stress problem reduces to finding a biharmonic function that satisfies the boundary condition of the problem provided that the body force is the only force. Biharmonic functions soon turned out to attract so much attention that the Academie des Sciences posed as its annual prize question the biharmonic Dirichlet problem for clamped elastic plate and the problem was solved by Hadamard in his monumental 1908 Memoir. On the other hand, since the 1970's biharmonic functions have been found to have very interesting applications in the classification theory of Riemannian manifolds similar to the classification theory of Riemann surfaces based on the existence or non-existence of certain harmonic functions with various boundedness conditions (For more detail see [7]). In this paper we study maps between Riemannian manifolds which pull back germs of biharmonic functions to germs of biharmonic functions. We first give characterizations of such maps and then study their links to harmonic morphisms and biharmonic maps. 2. PRELIMINARIES
A harmonic morphism is a map tp : (Mm,g) —► (Nn, h) between two Riemannian mani folds which pulls back germs of harmonic functions on N to germs of harmonic functions on M. In another word, harmonic morphisms are maps between Riemannian manifolds which preserve Laplace's equation in the sense that for any harmonic function / : U —> R, defined on an open subset U of N with ip~l(U) non-empty, / o ip : <^_1(Z7) —> R is a harmonic function. Harmonic morphisms are characterized independently by Fuglede [2] and Ishihara [4] as horizontally weakly conformal harmonic maps. Therefore, locally, a harmonic morphism tp = (ip1,... , ifin) is a solution to the following over-determined system of nonlinear partial differential equations of 2nd order:
^v,A*
Wft.»
axi oxj where A2 : M —> [0, +oo) is a smooth function. In stochastic theory, harmonic morphisms are known as Brownian path preserving maps meaning that they send a Brownian motion on M to a Brownian motion on N. Much work has been done in constructing and classifying
233 harmonic morphisms between certain model spaces. For a bibliography and the "atlas" of harmonic morphisms we refer to [3]. Definition 2.1. A continuous mapping
Thus our notion of biharmonic morphisms is different from that of p-harmonic morphisms in [6]. The notion of biharmonic morphisms is also different from that of /i-harmonic morphisms in [2, 1], As a direct consequence of Definition 2.1 we have the following Corollary 2.3. The composition of biharmonic morphisms is again a biharmonic mor phism. In order to study biharmonic morphisms we need the following calculations concerning the bi-Laplace operator A 2 on a Riemannian manifold. Lemma 2.4. Let (Mm,g) be a Riemannian manifold. Then for any f,h,k have (1)
A2{hk) = 2Ag{SJh, V/fc) + 2j(VA/i, Vfc) + 2g(Vh, VAfc)
(2)
2
£ C°°(M), we
+ hA2k + k&2h + 2A/iAfc A (hkf) = Chkf{4g(Vh, V5(Vfc, V/)) + 2hl\g{Vk, V/) + 4g(Vk, V / ) A h + hkA2f + 2hAkAf + 2hg{Vk, V A / ) + 2hg(Vf, VAk)} where V/i denotes the gradient of of h, and C^fiAhkf}
= A^; + A^fh + A!hk
Proof. A direct computation based on the expression for the Laplace operator applied to the products of two or three functions. □ Lemma 2.5. Let ip : (Mm,g) —► {Nn, h) be a smooth mapping with local expression V(x)
=
{V\x),...,^(x))
234 Then for any smooth function f on N we have, A2(/oV)
(3)
= (fijkk o ¥>)$( V , V^)( V \ V?ft) + (fljk o
(fij o ^{AgiVip*, V ) + 2(V<^', V A ^ ) +
A^Atp*}
+ (/ j 0 V )AV Proof. A straightforward computation based on the following Laplace operator on the the composition and the product of functions: A ( / o
a 3. CHARACTERIZATIONS OF BIHARMONIC MORPHISMS
In this section, we will give two characterizations of biharmonic morphisms and study their relation to harmonic morphisms and biharmonic maps. Theorem 3.1. Let ip : (Mm,g) —y {Nn,h) be a smooth mapping between Riemannian manifolds. Then the following properties are equivalent: 1) f is a biharmonic morphism. 2) ip is a horizontally weakly conformal map with dilation A and, with respect to local harmonic coordinates (ya) in N
(Al^Q = 0
A ^ V ) = A4(A2„(j/V))°¥> {A2M (^
(4)
hold for any a,/3,fi= 1,2, • • • , n. 3) There exists a C°° function A2 on M such that A 2 M (/oy,) = A 4 ( A 2 v / ) o ¥ ,
(5) 4
for any C function f on N. In order to prove Theorem 3.1 we need the following Lemmas. Lemma 3.2. [4] Let k
(6)
Lf(X)=Y: E ' " ^ ( ^ ^ L ^ 1
dx*
■■■dxi»
be a linear partial differential operator of order k(> 2) and of elliptic type which is defined in the neighbourhood of the origin x = 0. Assume the coefficients are in Ba(Ro) for some a(0 < a < 1) and Ro(> 0). Then for any constants C^...^ (0 < ii H 1- in < k) which are symmetric in i\,- • ■ ,in and satisfy
(?)
J2 0
ail"!"(0)C,,...„=0
235 there exists a solution f to Equation (6) defined in the neighbourhood of the origin x = 0 and satisfying
°J
(8)
Lemma 3.3. Let Sk(Rn) be the space of all symmetric k-tensors on R (,) defined by
Then the product
{{atl..,k),{bn...h))=all...ik5^---S^bjl...h is an inner product which turns Sk(Rn) into an inner product space. Proof. The proof is straightforward checking of the definition of an inner product and is omitted. □ Lemma 3.4. Let (Vn, {,)) be an inner product vector space. If {a, c) = 0 implies {b, c) = 0. for all c G V, then a = Xb for some A e R. Proof. The given condition implies that o and b have the same orthogonal complement. Therefore if a is the zero vector then so is b, otherwise, they are orthogonal to the same nondegenerate hyperplane in Vn, which shows that a = Xb for some A £ R. □ Proof of Theorem 3.1. To prove the implication 1) => 2) we let p £ M and suppose that q = (f(p) € N. Let (x') and (ya) be local coordinates and local harmonic coordinates around p and q respectively. After some calculations based on Lemma 2.4, we can rewrite Equation (3) as (9)
AJ, (/ o V) = Aa^»faPal> o v + Aa^fafSa o
with symmetric coefficients 'A"*" = i [g{Vipa, V
+g{Vfa, V)p(VvA Vip")] { A°" = | [ A 2 M ( ^ V V ) - Z ^ A ^ V ) -2^A2M(VV) - 2^A2M(VV)]
A2M(VV)
Aa = A2M
On the other hand, a direct computation shows that with respect to local harmonic coordi nates in TV, the biharmonic equation can be written as (11)
A 2 N / = Baf)^fafiap + fft'fap,, + Ba?faf)
with symmetric coefficients
(12)
' B"^" = \ (h^h"" + ha"h^ + haoh^) B*"' = I [ A ^ d / V l / ' ) " 2Va^%(yfiv") -2y»AtN(y»y') - 2y'A»,(i/V)]
B°> = 5 A M /V)
236 Now let W = SA(Rn) 0 S3(Rn) © S2{Rn) denote the direct sum of the inner product spaces of symmetric tensors on Rn (see Lemma 3.3). then for any x 6 M, A(x) = (Aafi'"'(x),Aa^(x),Aafi(x)) and B o y ( i ) = B(y) = (Ba^P(y),Ba'3a(y),B'"i(y)) can be viewed as elements of W. With this viewpoint and noting that the biharmonic equation (11) is a linear partial differential equation of elliptic type of order 4 which satisfies the conditions of Lemma 3.2, we see that for any constant C = {Cttpap,Capa,Cap) e W with condition (7) now becoming (B(y),C) = 0, there exists a biharmonic function / defined in a neighbourhood of (ya) satisfying the condition (fafapiv),
fa0a(y),
fafiiv))
= {fa{l*p °
0 (p(x))
= {Capap-i Capa> Cap) If f is a biharmonic morphism, then / o tp is a biharmonic function in a neighbourhood of (x'). From these and Equation (9) we have (A(x),C) = 0. Now Lemma 3.4 applies to give A(x) = n{x)B o (p(x) for some constant n(x) depending on x. Writing it componentwise we have
f
A<*f>'i>(x) = n{x)Ba
for any a,/S,a,p = 1,2,-•• , n. Putting a = a = p = f3 in the first equation of (13) and using Equations (10) and (12) we have (g{Vtpa, V<pa))2 = n{x) (haa o tp)2 for a = 1,2, • • • , n. From this it follows that fi(x) > 0 and that g(V<pa, Vtp«) = s/JKx)haa
(14)
o
for a = 1,2, • • • , n. Putting a = a, p = /3 in the first equation of (13) and using Equations (10) and (12) we have 2g(Vtpa, Vtp»)g(Vtpa, V / ) + g(Vtpa, V ^ a ) p ( V / , V / ) =li{x) [2hafl o tphaii otp + h^o tph^ o ip] From this and (14) we have g{V
(15)
From this we can calculate that li{x) = ~\dtp(x)\4 = X\x) Inserting this into (15) we see that, either |d
A2M(<pa^V")(x) = A4 ( A ^ j / V ) ) ° V(x) Note that A|,<£ = 0 is clearly true since the coordinate functions ya are harmonic (hence biharmonic) and by the definition of biharmonic morphism, ipa = ya o tp is a biharmonic function. Thus we see that tp is horizontally weakly conformal with property (4), which a
237 completes the proof of the implication 1) => 2). Now we proceed to prove the implication 2) => 3). By using Lemma 2.4 and the fact that the coordinates ya are harmonic we can calculate that
(16)
AS,(!/Y) =2ANhafi i%<3'»
/
(17)
rlhaP\
rlh"^
AJKVW) =4 ( f c - ^ _ + ^ | r + a
+ 2 (y ANh^
^ ^ )
+ /AJV/I"" +
y"ANhafi)
By using these equations, together with the conformality of p> and the property (4), we can rewrite Equation (9) as A 2 M (/ O tp) = faB
+ JaBu, o p~-
Bh^
Aha"~
ay"
„ dhaf
+^ " ~
ay"
dhaP
-4- 4 V ^ ay"
O if
+ U ° ¥>y (2ANhaf>) o
+ 7h{Vh", VUP) + faBANha?>) O p
= A4 (A 2 „/) O tp which ends the proof of implication 2) => 3) The implication 3) =4- 1) is obvious and is omitted. For the case when the target manifold is Euclidean space we have Corollary 3.5. Let tp : (M,g) —> (#",(,)), ip(x) = (pl{x),--- ,pn{x)) be a smooth map into a Euclidean space with the standard metric. Then ip is a biharmonic morphism if and only if p is a horizontally weakly conformal biharmonic map with
A^V)
=o
A2M(V
=0
for any a, /3, \i = 1,2, • • • , n Proof. Noting that the orthogonal coordinates (ya) in Rn are harmonic we see that the first equation in (4) holds by the definition of a biharmonic morphism. This Equation, when the target manifold is a Euclidean space, means exactly that tp is a biharmonic map [5]. Now substituting h"P = 6aP into Equations (16) and (17) we have that A2N(yayP) = 0 and A2N{yay^ytl) = 0 for any a, /3, \x = 1, 2, • • • , n. From these and Theorem 3.1 we obtain the corollary. □ Corollary 3.6. Any harmonic morphism p : (M,g) —► (N,h) with constant dilation is a biharmonic morphism. Proof. If
= \i{Alf)oV
238 Therefore it follows from Theorem 3.1 that ip is a biharmonic morphism.
D
Example 3.7. It follows from Corollary (3.6) that any harmonic Riemannian submersion is a biharmonic morphism. The well-known Hopf fibrations S 2 n _ 1 —► Sn for n = 1, 2,4 or 8 are harmonic morphisms with constant dilations and hence they are biharmonic morphisms. For the relationship between harmonic morphisms and biharmonic morphisms in general, we have Theorem 3.8. A harmonic morphism ip : (M,g) —> (N,h) is a biharmonic morphism if and only if ip is horizontally homothetic with harmonic energy density, i.e. Ae(ip) = 0. Proof. To prove the necessity of the condition, let us suppose that
\i(ANf)o
for any / 6 C°°(N). A direct computation using local harmonic coordinates (ya) in N yields (18) A2M(foV) = \*(Aif)o") + /i a " o
+ft*ora(VA2,V/)]/*otJ
+
dhaP hafl o
fa/jOip
oy
Therefore it follows from Theorem 3.1 that the harmonic morphism ip is a biharmonic mor phism if and only if (19)
| [h* o
Qha/3
+ 2-j^-oVg{V\2^Va)
faBoV
= (\
holds for any / € C°°{N). By special choices of / in the above equations we have (20)
hal> o pg{V\2, V) + ha" o W (VA 2 , V / )
(21)
+ hfi" o
hold for any a, /3, p, = 1,2,
oy
, n. From Equation (20) we obtain (n + 2)g(VX2, Vy") = 0
for any \i, which gives 2 (22) 5(VA ,V^) = 0 for any yu. Thus tp is horizontally homothetic. Inserting Equation (22) into Equation (21) we obtain AMA2 = 0. Note that for horizontally weakly conformal ip, the energy density e(ip) = |A 2 . Thus AMe(
239 Conversely, if y is horizontally homothetic with AM?-(
weakly □
When the target manifold is Euclidean space we have Corollary 3.10. There exists no non-constant M is compact.
biharmonic morphism tp : (M,g) —^ Rn if
Proof. From Corollary 3.5 we know that a biharmonic morphism ip : (M,g) —► Rn is a special horizontally weakly conformal biharmonic map. If M is compact then it follows from [5] that the map ip is harmonic. It is well-known that a harmonic map from a compact manifold into Rn is necessarily constant. □ Acknowledgments. A part of this work was done while the author was in a two-month-visit to International Centre for Theoretical Physics, Trieste, Italy as a visiting mathematician. The work was completed while the author was visiting the Department de Mathematiques, Universite de Bretagne Occidentale, France. The author is grateful to the two institutions for their hospitality and generosity. The author would like to thank P. Baird and E. Loubeau for many very useful discussions during the preparation of this work. REFERENCES [1] P. Baird and S. Gudmunsson, p-Harmonic maps and minimal submanifolds, Math. Ann., 294 (1992), 611-624. [2] B. Fuglede, Harmonic morphisms between Riemannian manifolds, Ann. Inst. Fourier (Grenoble), vol 28 (1978), 107-144. [3] S. Gudmundsson, The Bibliography of harmonic morphisms, http://www.maths.lth.se/matematiklu/personal/sigma/harmonic/bibliography. html. [4] T. Ishihara, A mapping of Riemannian manifolds which preserves harmonic functions, J. Math. Kyoto Univ.,19(2) (1979), 215-229. [5] G. Jiang, 2-harmonic maps and their first and second variational formulas, Chin. Ann. Math., Ser. A 7 (1986), 389-402. [6] E. Loubeau, On p-harmonic morphisms, to appear in Diff. Geo.and its applications. [7] L. Sario, M. Nakai, C. Wang and L. Chung, Classification Theory of Riemannian manifolds, Lecture Notes in math. 605, Springer- Verlag, Berlin, 1977. [8] S.P. Timosheako and J.N. Goodier, Theory of Elasticity, McGraw-Hill, Inc. 1970. DEPARTMENT OF MATHEMATICS, GUANGXI UNIVERSITY FOR NATIONALITIES, NANNING 530006, P . R . C H I N A . E-MAIL: [email protected] R E C E I V E D D E C E M B E R 17,
1999
240
Geometry and Topology of Submanifolds X eds. W. H. Chen et al. (pp. 240-251) © 2000 World Scientific Publishing Co. QUADRIC REPRESENTATION OF A SUBMANIFOLD IN PSEUDO-EUCLIDEAN SPACE
CHONGZHEN OUYANG AND TIANZHEN Q I U ABSTRACT. Let x : Mn —> E™ be an isometric immersion of a pseudo-Riemannian manifold into a pseudo-Euclidean space. The map x = xxt (t denotes transpose) is called the quadric representation of M. We study map x in general, especially as related t o the condition of being of finite type or Ax = Bx + C, where B and C are two constant matrices. KEYWORDS: pseudo-Euclidean space, submanifold, quadric representation, Laplacian. 1991 MATHEMATICS S U B J E C T CLASSIFICATION: Primary 53C40; Secondary 53C42.
1. Introduction. Let x : M " —> Em be an isometric immersion of an n-dimensional Riemannian manifold M into the Euclidean space Em and SM(m) be the mxm real symmetric matrices space. SM(m) becomes the standard Euclidean space of ^m(m 4- 1) dimension when equipped with the metric g(P,Q) = \tr(PQ). Regarding x as a column matrix in Em, we call the map x = xxl : Mn —> SM(m), where t denotes transpose, the quadric representation of submanifold M([3]). x is said to be of finite type if it can be decomposed into a sum of finitely many eigenfunctions of Laplacian from different eigenspaces ([1],[2]). In [3], I. Dimitric established some general results about the quadric representation in particular those relative to the condition of x being of finite type. In [4], J. Lu gave some results on submanifolds which satisfy Ax = Bx + C, where B and C are two constant matrices. In the present paper wc shall discuss the submanifolds of a pseudo-Euclidean space and generalize the above results. SM(m) becomes a ^m(m + l)-dimensional pseudo-Euclidean space of index u(m — v) when equipped with the metric g(P,Q) = \tr{JPJQ), where J = (rjiSij), r]i = ■■■ = Vm—v
— 1) Wm—v+l
=■•••
— T)m =
1.
Let E™ be the pseudo-Euclidean space with the standard flat metric ds2 = YlT^i' ^x\ ~ Y^j=m-v+\ &x% where (x\, ■ • ■ ,xm) is a rectangular coordinate system of E™. S™~x(r) and H™~^{r) denote the pseudo-Riemannian sphere and pseudo-hyperbolic space resp.,
S?~\r) = {xeE?\
(x,x) = r2}; H^(r)
= {x € S™ | (x,x) = -r 2 }
where (,) denotes the pseudo-Euclidean metric of E™. If r = 1, 5™~1(1) is denoted by S™~\ and H^{1) is denoted by H ^ . ([2],[5]). Project supported by NSFC (Grant No. 19871038) and NSF of Jiangxi Province. This paper is in final form and no version of it will be submitted for publication elsewhere.
241 Let x : Mn —► E™ be an isometric immersion of an n-dimensional pseudo-Riemannian manifold M into the pseudo-Euclidean space E™ of index v. Regarding i as a column matrix in E™, we obtain the map x = xx* : Mn —> SM(m), where t denotes transpose. We call x the quadric representation of submanifold M. A smooth map x : Mn —> SM(m) is said to be of fc-type if we can globally write (1.1)
X=X0+Xi~{
i-Xk,
where £Q is a constant matrix in SM(m),and Axi = \Xi(i = 1, • • • , k), where Ai, • • ■ , A/t are k different eigenvalues of Laplacian of M , and A acts a vector function componentwise. x is said to be of infinite type if it can not be written as (1.1). Let x : Mn —> E™ be an isometric immersion of an n-dimensional pseudo-Riemannian manifold M into the pseudo-Euclidean space E™. In the present paper we first prove that if n > 1, x is an isometric immersion if and only if x(M) C S ™ - 1 , and that £ is a conformal map if and only if x(M) C ^ " ' ( r ) or H™Sil{r) (theorem 1). Next we establish some general results about the relation between the immersions x and x on some geometric objects (theorem 2). For the pseudo-Riemannian manifold M in E™, we obtain that if x is of 1-type, M is a totally geodesic submanifold of 5^* _ 1 (r) or H^S^ir) in E™ (theorem 3); and if the mean curvature vector of M in E™ vanishes, t h e quadric representation x must be of infinite type (theorem 4). For the non-degenerate submanifold M with the parallel mean curvature vector in E™, we claim that if the quadric representation x satisfy Ax = Bx + C,where B and C are m x m constant matrices, M is a totally geodesic submanifold or contained in a quadric hypersurface(theorem 5). Finally we obtain that there is no submanifold in E™ whose quadric representation satisfies Ax = Bx (theorem 6), and the only non-degenerate submanifold in E™ satisfying Ax = C is totally geodesic one (theorem 7). 2. Preliminaries. Let x : Mn —> E™ b e an isometric immersion of an n-dimensional pseudo-Riemannian manifold M into the pseudo-Euclidean space £7™. Suppose that e i , - - - , e m are local orthonormal vector fields along M such that the first n vector fields are tangent to M . From now on, we shall use the following index convention: \
l
n + l
n + l < / 3 , 7 < m —1.
At any point x e E™, for any column vector V in E™, we denote
i where EA = (BA^A)
r
= ±1-
Let w 1 , • • ■ ,u}m be the dual coframe on E™, and let o)g be the connection forms corre sponding to the connection V on £7™. The structure equations on E™ are duA
= -Y^^B B
AwB
>
£
AUA
+ £BUB
= 0.
242 Restricted to M, w|- are the connection forms of the connection V in M, and dw* = - ^
Oj'j A iJ,
SjUj + £iu{ = 0.
i
The second fundamental form of M is
h = ^2 uj ® ^ ® e r = 5Zft^'a,i® ^ ® eT"
(2.1)
The shape operator Aer in the direction er is (2.2)
ASr = - ^214 ® e3- = e r ^ c ^ w ' ® e,-.
We denote ASr by A,., and the corresponding matrix is A r = (er£J7iJ'-). The mean curvature vector of M in £™ is
(2.3)
ff
= ij>/^ r .
Let D be the normal connection of M in #™. Then we have the Gauss equation (2.4)
VxY
= VxY
+
h(X,Y),
and the Weingarten equation (2.5)
V X £ = -A({X)
+
Dxt
where X, V are tangent vector fields of M, and f is a normal vector field of M. The real symmetric matrices space SM(m) becomes a ^m(m + 1)-dimensional pseudoEuclidean space of index (m — v)v when equipped with the metric g(P, Q) = ^tr(JPJQ), where J = (VASAB), VI = ■ ■ ■ = Vm-v = l,»jm-„ + i = • • ■ = Vm = - 1 . Thus, for any two vectors X = (X1, • • - , X m )«, Y = ( Y \ • • • , Ym)< in E™, we have (X, Y) = ^ ^ X ^ Y - 4 = X*JY = t r ( X J Y ' ) . y4
We define the map * : £ ™ x £ „ m - t SM(m) by X * Y = XYl+YXf vectors X, Y in E™. Then * is bilinear and symmetric, and we have (2.6)
g(X*Y,V*W)
for any two column
= {X, W)(Y, V) + {X, V)(Y, W).
Regarding r a s a column matrix, we obtain a smooth map x : E™ —> SM(m), i 4 i = xx* = \x * x. Its tangent map is (2.7)
x, = dx : E™ -> SM(m),
X ^dx(X)
= X * x.
243 Let V be the connection on SM(m) (2.8)
with respect to ,then
V X ( Y * Z) = Vdi(x){Y
* Z) = VXY
*Z + Y*
VXZ,
where X, Y, Z are vectors in E™. From the symmetry of *, the Gauss equation and the Weingarten equation, we get 2_]ffi/i(ei,X) * e> = ^ J ^ E r ^ r ^ , ei)e r * e* = 2^erATX
(2.9)
i,r
i
Y,e^Vxei^ * ei = lL£iUi(X)ei
(2.10)
*er,
r
*ei = °>
i,3
i
where X is a tangent vector of M. Let A be the Laplacian of M, n
Af = 5>[(V e , e i )/ - eifo/)], V/ e C°°(M). i=l
Then we have
(2.ii)
zi(x * y) = (AX-) * y + x * (AY) - 2^ e i (v e i x) * (ve,y), vx, y e K 1 i
Therefore, we get (2.12)
Ai = -A(x
* x) = Ax * x - 2 j e ; V e i : E * V e j £ = - n i l * x - Y ^ e ^ * e;. i
i
3. Quadric r e p r e s e n t a t i o n of a submanifold in p s e u d o - E u c l i d e a n s p a c e . T h e o r e m 1. Let x : Mn —> E™ be an isometric immersion of an n-dimensional pseudoRiemannian manifold into the pseudo-Euclidean space E™, and x : M " —> SM(m) be its quadric representation. Then (i) Ifx(M) C S™-1, thenx is an isometric immersion. Moreover, ifn > 1, the converse also holds. (ii) If n > 1, then x is a conformal map if and only if x(M) is contained in S™~l(r) orH^\r). Proof. We'll first show (ii). Let X, Y be any two tangent vectors of M, and XLY. a conformal map, then,from (2.6) and (2.7), 0 =
~g(xt(X),it(Y))
=
(x,X)(x,Y).
If X,Y are unit vectors, we have the following cases: 1) if X,Y time-like, X + Y and X — Y are also perpendicular, then we get 0 = {x, X + Y){x, X-Y)
If x is
are both space-like or
= (x, X)2 - (x, y ) 2 ;
244 2) if there is a space-like vector and the other is time-like in X, Y, then 0 = {x,X + Y){x, X + Y) = (x, X)2 + (x, Y)2. Therefore, {x, X) = 0 for any tangent vector X of M, and hence {x, x) = ± r 2 = const., which shows that x(M) C 5™ _1 (r) or H™S\ (r). The converse is clear because g(xt(X),xt(Y)) = (x,x)(X,Y). Now we show (i). If n > 1 and x is an isometric immersion, from (ii) we know x(M) C 5™ _1 (r) or H™Si1(r), thus (x, x) = ±r 2 . For any tangent vector X of M, we have g{Xt.(X),x,(X))
= g(X*x,X*x)
= ±r2(X,
X),
which shows that if x is isometric immersion, x{M) C S1™-1. The converse is clear. D In the following theorem and its proof, we'll use symbols with "for the objects related to the immersion x, those with ' related to the immersion into S™ _1 and symbols without "or ' are related to the immersion into E™. Theorem 2. Letx : Mn —> S™~1 C E™ be an isometric immersion from an n-dimensional pseudo-Riemannian manifold M into a pseudo-Riemannian sphere S™ -1 which imbedded standard in the pseudo-Euclidean space E™, and x : Mn -¥ SM(m) be its quadric repre sentation. Then (i) g(h, h) = const. <=> (h, h) = const., g(H, H) = const. •» {H, H) = const.; (ii) AH = fil <=> AH = fil; (Hi) M = 0 » / i ' = 0 « W = 0 . Proof, (i) Since (3.1)
h(X,Y) =
Vx(Y*x)-VxY*x
= VXY * x + Y * X - VXY *x = h(X, Y) * x + Y * X, therefore, we get g(h,h) = (h,h) + n ( n + l), which shows that g(h, h) = const. <=> {h, h) = const. On the other hand, (3.2)
H = ~'^2£ih(ei>ei) i
= — ^2siei
*&i + H *X.
i
Therefore g(H,H) = 1 + (H,H) + (H,x)2 = ^ + (H,H), where we have used that H = H1 — x, which shows that g(H, H) = const. O {H, H) = const. (ii) From (i), for any tangent vector X of M, we have ~ ~ VxH = Vx(H*x)
1 + -J2
£iV x (ei * e{) = VXH *x + H*X i
2 +- ^ i
^ ( V ^ e , * et)
245 = {-AHX
+ DXH) * x + H * X + - J2si(Vxei
= -{dxVAjjX
+ DXH) +H*X
+ h{eu X)) * e*
+ -J^erArX
* er,
n *■—'
where we have used (2.4),(2.5),(2.9) and (2.10). Let e m = x. Then
VXH
= -xJAHX
+ ^-^X)
+ DXH'
* x + H' * X + -S"
n i—1
n
On the other hand, VXH = -AHX shows that (ii) is true.
ERARX * e«.
+ DXH, thus we have AHX = AHX + ^X,
which
(iii) From (ii) we have
(3.3)
DXH = DXH' *x + H'*X
-J2s/3APx*e/3
+ n
P
If DH = 0, then g(DxH,ei * ep) = ^(ApX,ei) + (H',ep){X,ei). Let X = eh j ± i, we have {Apej,et) = 0, i.e. h^ = 0. Let X = e<, we get 0 = ^eph^ + \ep ] [ \ h^, and, therefore, h{i = 0. Hence, h! = 0. If h' = 0, then H' = 0. From (3.3) we have DXH = 0, for any X e TM, which means that DH = 0. If Vh = 0, of course, DH = 0. Now we only need show that V/i = 0 when h' = 0. In fact, if h! = 0, we have h(X, Y) = -{X, Y)x, and therefore, from (3.1), we get h(X, Y) = -{X,Y)x*x + X*Y. Thus
(Vxh)(ei,ej)
= Vxei*ej
□
+ ei*Vxej
= Dx(h{ei,ej))
-h(Vx&i,
- h(e;, Vxe.,)
~(Vxei*ej - (V xe-i, ej)x*x) - (e, *S/xej - (Vxej, ei)x*x) = 0.
246 4. Submanifolds with finite type x. Theorem 3. Let x : Mn —► S™ be an isometric immersion of a pseudo-Riemannian manifold M into the pseudo-Euclidean space E™. If the quadric representation x is of 1-type, then M is a totally geodesic submanifold of a pseudo-Riemannian sphere S'^'~1(r) or a pseudo-hyperbolic space H™S\{r) in E™. Proof. Suppose that £ is a 1-type map, then we can write x = xo + i\, where XQ is a constant matrix in SM(m),and Aii = \x\. Thus Ai\ = \(x — XQ) and, from (2.12), we get X(x — XQ) + nH * x + YJ £;e; * ej = 0. i
Differentiating this relation along a vector field X of M, and using (2.5), (2.9) and (2.10), we obtain (4.1)
XX * x + n(DxH
- AHX) * x + nH * X + 2^2erArX
* er = 0 .
V
We apply g(e, * e r , —) to (4.1) and let X = e;. Then,multiplying by £; and summing on i, we obtain 0 = n\(x,er) + n{DXTH,er) - n ^ £ i ( A f f e i , e i ) ( a : , e r ) i 2
+n (H,er)
+ 2^£i(^rei,ei). i
If we multiply this relation by £ r e r and sum on r, we get (4.2)
(A - Y^ £i{AHeu ei))xN + DXTH + (n + 2)H = 0. i
Applying g(er * es, —) to (4.1), we have (DXH, er)(x, es) + (DXH, es){x, er) = 0. Therefore, (4.3)
xN = 0 or
DH = 0.
Applying g(e{ * e,-, - ) to (4.1), we get (4.4)
AX - nAHX = 0 or xT = 0-
Let U = {p e M : xT ^ 0 at p}; V = {p £ U : xN ^ 0 at p}. From (4.3),(4.4) we know DH = 0 and nAjf = A/ on V. From (4.2), we get H = 0 on V. Now we compute tr(A(Ji)) on V, tr{A(Jx))
= Atr(Jx) = A(x,x) = 2{Ax,x) - 2 ^ £ i ( e i , e i ) =
-In.
247 On the other hand, tr(A{Jx))
= tr{J\(x
- x0)) = \(tr(Jx)
- tr(Jx0))
= A((x, x) -
tr(Jx0)).
Therefore, A / 0 and (x,x) = tr(Jxo) — x = cons^-- Thus —2n = A{x,x) = 0, which is a contradiction. Consequently V = 0 , x = x-r on U. But —2n = \({x,x) — tr(Jxo)), so that (a;, a;) = tr(Jxo) — x = cons^-- Therefore xr = 0, which is contradict to x = XTTherefore we must have U = 0, x = x^, d{x,x) = 0. Thus (a;, a;) = cr2, where r is a positive constant number and c = ± 1 . When c = 1, x(M) C S'^ l _ 1 (r); when c = —1, x{M) C H™-?{r). We have H = H' - $x, where ff' is the mean curvature vector of M in S ™ _ 1 ( r ) or H™S[l{r). And from (4.3) we know DH = 0. Thus from (4.2) we get + (n + 2)H = {n+ 2)H' + (A - J2ei(AHei>
0 = (A - ^2ei(AHei,ei))x i
et) - n -
2)x,
i
and, therefore, H' = 0. Let e m = £a:, then from (4.1), we have (A -
2
^ " \ ^°)X r
Therefore A =
* x + 2Y^epApX P
*ep = Q,
Va; G T M .
and A^ = 0, i.e. M is a totally geodesic submanifold in S™~l(r)
'"t
or
The following theorem shows that if x is of finite type, then the mean curvature vector does not vanish. T h e o r e m 4. Let x : Mn —► £™ 6e an isometric immersion with zero mean curvature vector of a pseudo-Riemannian manifold into pseudo-Euclidean space. Then its quadric representation x is of infinite type. Proof. Suppose x is of fc-type, where k is a positive integer. Then we can write x = x0 +£\ H where XQ is a constant matrix in SM(m), Q{t) = tk + a ^ ' where ax = -J2i=i (4.5)
A
i>
hit,
A£i = \xi,i 1
= 1, • • • , k. Therefore there is
+■■■ + a f c _!i + ak,
a
2 = E , < j A ' ^ ' ' ' - • ,«fc = (-l) f c AiA 2 • • ■ Afc, such that Q{A)(x
- x0) = 0.
We have tr(J(Ax)) = A(tr(Jx)) = A(x,x) = 2{Ax,x) - 2 ^ i £ i ( e i , e i ) = - 2 n . Therefore tr(J(Alx)) = 0, i > 2, and from (4.5), we have tr{J(ak-\Ax + \k{£ - x0))) = 0. Thus (4.6)
ak((x,x)-tr(Jx0))
=
2nak_1.
248 If ak 4- 0, {x, x) = tr(Jxo) + 2n^= const, so x(M) c S™~l{r) or H^{r). But there is no submanifold with zero mean curvature vector in S™~l{r) and H™S\{r), which is a contradiction. In fact if M is contained in 5 " _ 1 ( r ) (or H™~i(r)), let H and H' be the mean curvature vector in E™ and S™~l{r) (or H^Z\(r)),respectively, we have H = H'± ^x, where (x, i?') = 0. Therefore H = 0. If Qfc = 0, then one eigenvalue, say Ai, must be zero. If k > 2, from (4.6) we have afc_! = 0. This implies that another eigenvalue, say A2, is zero, which is a contradiction because Aj ^ A2. If k = 1, then Ax = 0, and, therefore, tr(JAx) = 0, which is a contradiction because tr(JAx) = —2n. □ 5. Submanifold whose quadric representation satisfy Ax = Bx + C. Theorem 5. Let x : M" —¥ E™ be an isometric immersion of a pseudo-Riemannian manifold into pseudo-Euclidean space with parallel mean curvature vector. If its quadric representation x satisfies Ax = Bx + C, then M must be (a piece of) a totally geodesic submanifold or contained on a quadric hypersurf'ace where B, C are m x m matrices. Proof. Choosing that e m is parallel to the mean curvature vector H, we have H = aem, the mean curvature a = const., and u^ = 0. Differentiating Ax along a vector X of M, using (2.9),(2.10) and (2.12), we get ^ax(X){Ax)
= naAmX
*x — naem * x — 2 2_,erArX
* er.
r
On the other hand Va^x){Ax) (5.1)
2naAmX
= \{BX * x + Bx * X). Therefore
* x — 2naem * x — 4 \ J erArX *er = BX * x + Bx * X. r
Applying g(eT * e s , —) to (5.1), we have (BX,er){x,es)
+ (BX,es){x,er)
Therefore we have either (x, er) = 0 or (BX,er)
=0.
= 0, Vr.
Case 1. XM — 0, x = XT- Applying g(er * Y, -) to (5.1) we have (5.2)
(BX,er)(x,Y)
+ (Bx,er){X,Y)
= -2naer5rm(X,Y)
-
4{ArX,Y).
From (x, eT) = 0 we get 0 = Y(x, eT) = -(x, ArY) = —{ATx, Y), and, hence, Arx = 0. Let X = Y = x in (5.2). Then we obtain (5.3) {{Bx, er) +naerSrm){x,x) = 0. When a = 0, if (x,x) = 0, M is contained in a quadric cone; and if (x,x) ^ 0 then (Bx,e r ) = 0. Let Y = x in (5.2) we get (BX,ep) = 0, and, hence, ApX = 0, i.e. M is totally geodesic.
249 When a ^ 0, let X = Y = e< in (5.2), we have (n + l){Bx, e m ) = ~2n2aem -
4nasm.
From this and (5.3) we know (a;, x) = 0, i.e. M is contained in a quadric cone. Case 2. {BX)N = 0. Applying g(er * Y, —) to (5.1), we get (5.4)
{Bx,Y){x,er)
+ {Bx,er){X,Y)
=
2na{AmX, Y)(x, e r ) - 2naer6rm{X, Y) - 4{ArX, Y). When a = 0, (5.4) implies (BX,Y) = (BY,X). Applying g(Y * Z, - ) to (5.1), we have (BX)T + (BXT)T = 0, and, therefore, X{Bx,x) = 0. Thus, (Bx,x) = const., i.e. M is contained in a quadric hypersurface. When a ^ O w e have the following subcases: Subcase 1. (x,em) = 0. Let r = m and X = Y = e}, in (5.4). Multiplying this relation by Ej and summing on j , we obtain (Bx, em) = - 2 n a e m - 4asm. From this and (5.4) we have AmX = e m aX, and, therefore, 0 = X(x, e m ) = (x, - A m X ) = -Ema(x,
X).
Hence X(x,x) = 0 and (x,x) = const., i.e. M is contained in a quadric hypersurface. Subcase 2. {x,em) ^ 0. From (5.4) we get (BX,Y) = (BY,X). Let r = m, Y = xT in (5.4) we have (5.5)
2naem{X,x)
+ 4{AmX,x)
+ (Bx,em){X,x)
+
{BX,xT){x,em)
- 2na{AmX,x){x,em)
= 0.
Applying g(Y * Z, —) to (5.1), we have (5.6)
{BX, Y)(x, Z) + {BX, Z){x, Y) + {Bx, Y){X, Z) + {Bx, Z){X, Y) -2na{AmX,Y){x,
Z) - 2na{Am, Z){x, Y) = 0.
Let y = Z = Ci in (5.6). Multiplying this relation by gj and summing on i, we get (5.7)
{BX, xT) + {Bx, X) - 2na{AmX,x)
Because {BXT,X)
= {BX,XT),
(5.8)
= 0.
we have
BxT + {Bx)T - 2naAmxT
= 0.
Let X = Y = ti in (5.6). Multiplying this relation by e^, we get (5.9)
(^2si{Bei,el)
- 2n2o?em)xT
+ BxT + {n + \){Bx)T - 2naAmxT
= 0.
250 From (5.5) and (5.7), we obtain (5.10)
((Bx,em)
+ 2naem)xT - {x, em)(Bx)T
+ 4Amxr
= 0.
From (5.8)-(5.10) we get n({Bx,em)
+ 2naem)xT 4- {x, e m ) ( ^ £ i ( B e , , e i ) - 2n2a2em)xT
+ AnAmxr = 0.
i
Let r = m, X = Y = et in (5.4). Multiplying this relation by e, and summing on i, we have (5.11)
2n(n + 2)asm + (^2Ei(Bei>ei)
~ 2n 2 a 2 e m )(a;,e m ) + n{Bx,em)
= 0.
i
From the above two equations we have AmXT = aemXT- Using this and (5.7) we get X{Bx,x) = (BX,x) + (Bx,X) - 2na(AmX,x) = 2na(X,AmxT) = n2a2emX{x,x). 2 2 Therefore, ((B — na emI)x,x) = const.. If B = n(x £mI, from (5.11) we know ot — 0 which is a contradiction. Hence, {{B - na2emI)x,x) = a(a is constant), M is contained in a quadric hypersurface. □ The following theorem shows that C / 0. Theorem 6. There does not exist non-degenerate submanifold in E™ whose quadric rep resentation satisfies Ax = Bx. Proof. From (2.12) we know that if there is a submanifold M satisfying Ax = Bx, then Bx + nH * x + YJ £ i e i *Si = 0. i
Applying gifii * ej, —) to this equation, we have (Bx,ei)(x,ei)
+ 2EJ = 0.
We can choose e-i £ TXM such that (x,ej) = 0, which is a contradiction to the above equation. □ Theorem 7. The only n-dimensional non-degenerate submanifold in E™ satisfying Ai = C(C ^ 0) is totally geodesic, i.e. it is the n-dimensional linear subspace of E™. Proof. Differentiating Ax - C along a vector field X of M, using (2.9),(2.10) and (2.12), we have (5.12)
nVxH*x
+ nH*X
+ 2^2erArX
*er = 0.
r
If H = 0, applying g(Y * er, - ) to the above equation,we get (ArX, Y) = 0, VA", Y, r, and M is totally geodesic.
251 Suppose H ^ Q,H = ctem. Using the Weingarten equation and (5.12), we have (5.13)
n(Xa)em
* x + naDxem
* x — naAmX
* x + naem * X + 2 V ^ e r ArX
* eT = 0.
r
Case 1. Q = const..
Applying g(em * ei, —) to (5.13), we obtain
-na(AmX,ei)(x,em) and, therefore, (na{x,em) (5.14)
+ naEm{X,ei)
— 2)AmX AmX
= naemX.
= aemX,
+ 2(AmX,ei)
= 0,
Hence,
and
(x, em) =
n 4- 2 . net
Applying g(e; * Cj, —) to (5.13),we have {AmX,ei)(x,ej)
+ (AmX,
ej) {x, e{) = 0 ,
and, therefore, (5.15)
(x,ej)
= Q.
Now applying g(ep * ei, - ) to (5.13) by means of (5.14) and (5.15), we have 2{ApX, a) = na2em(X,
ei)(x, ep),
and, therefore, 0 = ^2,et{Apei,ei)
=
-£ma2(x,ep).
i
Thus (x, ep) = 0, and, hence, x = ! ^ e m e m . Now, (5.13) becomes S r w m ( ^ O e r * e m = 0, which implies uTm(X) = 0. Thus X = VYX = - ^ E m A , , ! = — s i ^ X , which is a contradiction. Case 2. a ^ const., i.e. l a / 0 for some X. Applying g(em * er, —) to (5.13), we have (x, er) = 0 which shows that x eTxM. Now applying g{et * e,, - ) to (5.13), we obtain (5.16)
(AmX,ei)(x,ej)
+ {AmX,e]){x,et)^0,
There is e 3 such that (x,ej) ^ 0. (5.16) implies (AmX,ei) which is a contradiction. □
VXeTx(M). = 0, and, therefore, a = 0
REFERENCES 1. B. Y. Chen, Total mean curvature and suhmanifolds of finite type, World Sci., 1984. 2. B. Y. Chen, Finite suhmanifolds in a pseudo-euclidean space and applications, Kodai Math. J. 8 (1985), 358-374. 3. I. Dimitric, Quadric representation of a submanifold, Proc. Amer. Math. Soc. 114 (1992), 201-210. 4. J. Lu, Suhmanifolds whose quadric representations satisfy Ax = Bx + C, Kodai Math. J. 20 (1997), 135-142. 5. J. A. Wolf, Spaces of constant curvature, McGraw Hill, New York, 1967. DEPARTMENT OF MATHEMATICS, NANCHANG
UNIVERSITY,
NANCHANG 330047, P . R. O F C H I N A R E C E I V E D S E P T E M B E R 20,
1999
252
Geometry and Topology of Submanifolds X eds. W. H. Chen et al. (pp. 252-258) © 2000 World Scientific Publishing Co.
THE SPECTRAL GEOMETRY OF THE DOLBEAULT LAPLACIAN WITH COEFFICIENTS IN A HOLOMORPHIC VECTOR BUNDLE FOR A HERMITIAN SUBMERSION
JEONGHYEONG PARKf
ABSTRACT. Let TV : Z —> Y be a Hermitian submersion. We study when the pull back of an eigenform of the Dolbeault Laplacian with coefficients in a holomorphic vector bundle E over Y is an eigenform of the Dolbeault Laplacian on Z with coefficients in TV*E. Keywords: Dolbeault Laplacian, Riemannian Submersion, Eigenvalues 1991 Mathematics Subject Classification: Primary 58G25
§1 INTRODUCTION
Let M be a compact complex manifold of complex dimension n with associated al most complex structure J. Let g be a J invariant Riemannian metric. We adopt the Einstein convention and sum over repeated indices. Let z = (z1,...,zn) be local holo morphic coordinates on an open subset U of M. Let 9j := -^ and 9; := -^. If / 6 C°°(M), let df := diiftdz1 and df := diifidz*. If / = {1 < ix < ... < ip"< n} and J = {1 < ji < ... < j q < n} are multi indices, let dz1 := dz'1 A • • • A dz'" and dz3 := dz'1 A • • • A dzi«. Let A(p'q) := spani/^pjj^gjd^ 7 A dzJ} be the bundle of forms of bidegree (p, q). We may extend d and 8 to invariantly defined operators d(fI}Jdzr 1
dtfudz
A dzJ) := d{fItJ) A dz1 A dzJ : C°°(A(M>) -> J
A dz ) := d{fu)
1
J
X,
Cco(A^+1'^) 00
A dz A dz : C" (A^'")) -> C (A
Let E be an auxiliary holomorphic vector bundle over M which is equipped with a unitary metric h. Let / 6 C°°(E® A' 0 ' 9 ') be an E valued form of bidegree (0,g). Let {sa} be a local holomorphic frame for E over a coordinate chart U; we expand / = s"®fa where fa e C 00 (A( 0 ' 9 '). The complex exterior derivative dE ■ C"x>(£,igiA(0>9)) -> C oo (£'®A( 0 ''' +1 )) with coefficients in E is defined by dE(sa « fa) = S" ® 9/a! fThe author wishes to acknowledge the financial support of the Korea Research foundation made in the program year of 1998, and BSRI-98-1425. This paper is in final form and no version of it will be submitted for publication elsewhere.
253 dE is invariantly defined and dE = 0. Let SE be the formal adjoint of the operator The E valued Dolbeault Laplacian is defined by A{°'q)
dE.
:= SEBE + dESE on C°°{E ® A(°'«>).
We denote the associated eigenspaces by E(\, A(°'q))
:= {/ G C°°{E®
A(°-«)) : A g ' ? ) / = A / } .
Let w : (Z, Jz, gz) —> (V, J y , C'x(Ez) so t h a t 7T*<9EY = BEZK*. In previous work [8] we showed that if the pull-back of an eigenform is an eigenform, then the eigenvalue can only increase. In this short note, we generalize these results to the setting of the E valued Dolbeault Laplacian. Let H be the horizontal tangent space of the submersion. We split H ® C = H1'0 ffi'H0,1 into the horizontal holomorphic and anti-holomorphic tangent vectors. We shall prove 1.1 T h e o r e m . Let IT : Z —> Y be a Hermitian (1)
5Ezn*
- IT*5EY
(2) A J ^ V - n'A^f
= 7T* ® {6ZK*
-
submersion.
-K'ZY).
= rr* ® ( A < ^ V -
n'A^).
This shows that the twisting bundle plays no role in the kinds of eigenvalue questions we investigate. Our results of [8] then extend immediately to this setting to yield the following result. 1.2 T h e o r e m . Let TT : Z —► Y be a Hermitian holomorphic bundle overY.
submersion.
Let ( E y , / i y ) be a
Hermitian
(1) / / 0 + $ e E(X, A{Ef) and J/TT** e E(n,A{Ef), then X < p.. Furthermore, q = 0, then X = \i. (2) Fix q with 0 < q < dime Y. The following conditions are equivalent:
if
i ) A ^ V = ,'A^>. ii) VA > 0, 30(A) > 0 so n*E(X, A<£?>) C E(»(\),
A%f).
iii) The fibers of ir are minimal. If q = 0, i/ien there is no further condition. q > 0, tten the holomorphic horizontal distribution riito is integrable.
If
Previously, we constructed examples where eigenvalues can change so the results cited above are non-trivial. We refer to [8] for the proof of the following result. 1.3 T h e o r e m . Let E — 1 be the trivial line bundle. Let 0 < A < /i < oo and let q > 2 be given. There exists a Hermitian submersion -K : V —> U and there exists 0 ^ $ € E(X, A(°'q)) so that TT*$ G E{n, A{°'q)). In §2, we give the proof of Theorem 1.1. There is a corresponding analogue in the real category that we state and prove in §3. It is a pleasant task to thank Professor P. B. Gilkey for stimulating conversations on this subject.
254 §2 T H E PROOF OF THEOREM
1.1
Let M be a holomorphic manifold, let J be the associate almost complex structure, and let g be a Riemannian metric which is J invariant. Let E be a holomorphic vector bundle over M and let ftbea fiber metric on E. We wish to express 5E in terms of the derivatives of h and the untwisted operator 5 relative to suitable local frames for E® A( 0,9 '. We adopt the following notational conventions. If £ is a real cotangent vector and if ui is a differential form, let ext(£)cj : = f A u denote exterior multiplication and let int(£) be the dual, interior multiplication. We extend ext and int to the complexified exterior algebra so the maps £ —> ext(£) and £ —> int(£) are complex linear. We then have the duality relationship: (2.1.a)
fl(ext(f)wi,w2)
=
g(ui,iat(g)u2).
Let {UJ1} be a local frame for Af0,9' over a coordinate chart U so dui1 = 0; for example, we could take u1 = dz1. Let {s°} be a local holomorphic frame for E over U. Let dvol = Qdx be the Riemannian volume element on U. Let hab give the inner product on E relative to this local frame and let hab be the inverse matrix. We have: (2.1.b)
hab := h(sa, sb), hab = ft6a, and habhbc = 8ac.
2.2 Lemma. Let f = s" ® /„ and F = sb ® F;, be compactly supported within U. We expand fa = fajuT. (1)
(f,SEF)L2(u)=(dEf,F)LHU) = -Ju h^fajG^diiSgid? A UJ1, Fb)}dvol ba Iug(fa,di{h }mtidz^Fbjdvol. (2) IfE = l, then (f,6F)L2{u) = {dj,F)L,{u) = -fv fid^diiggidz1 A u/, F)}dvol. (3) 8EF = sb® (6Fb - hbc mt(dhdc)Fd).
Proof. We prove the first assertion by integrating by parts; the sections / and F are assumed to have compact support within a coordinate chart and therefore there are no boundary contributions. We use equations (2.1.a) and (2.1.b) to compute (dE{sa®fa),sb®Fb)LHU) = fu di{fa,i}habGg(dzi = -fu fajdiih^Ggidz1
A UJ1, Fb)dx A u1, Fb)}dx
= -Ju h^fajG-'diiGgid?
A w1, Fb)}dvol
ba
- fv g(fa, di{h } intidz^F^dvol. We set h = 1 to derive assertion (2) from assertion (1). To prove the final assertion, we use assertions (1) and (2) and equation (2.1.b) to compute: (/, SEF)L2{U) = - Jv h^fa^g-'dilGgidz' A UJ1, Fb)}dvol - fu g(fa, di{hba} mtidz^F^dvol = (/, sb ® S{Fb})L,{u) - Jv g(fa, habhbc mt(dhdc)Fd)dvol = (/, sb
255 (sbY®Fb)
(5Bzir*-K*~d~EY)
= 4 ® {(5 Z T* - ir*5y)F6 " T'/ifcc intz(SZ7r*/idc)7r*Fd +7T*(/l()cmty(dy/ldc)F(j}. Since 7r* inty = int z 7r* and 7r*9y = BZK*, the additional terms involving ft cancel and Theorem 1.1 (1) follows. We prove assertion (2) by using assertion (1) and observation that pull back commutes with d to compute: = 7r* ® (8z5zx* = IT* ® (dzSzK*
dzn'dy)
— 7T*9y<5y)
^EZBEZ^* - T[*5EY9EY = (%"■* K*5EY)9EY = 7T* ® (5z7T* — 7T*5y)3y = 7T* (8 (<5Z9z7T* - 7T*5y9y). D §3 A REAL ANALOGUE P+1
Let d : C^A^M -> C°°A M be the exterior derivative. If V is the Levi-Civita connection, then we can define d = extoV; see, for example, [3, Lemma 1.5.3]. Let E be a smooth auxiliary coefficient bundle which is equipped with a smooth fiber metric h and a unitary connection V#. These structures define the notion of a geometric coefficient bundle. We use V'E and the Levi-Civita connection to covariantly differentiate tensors of all types and to define dE := (1 ® ext) o V : C°°{E ® A"M) -> C°°(F ® A P + 1 M). Let SE be the formal adjoint of dg. The Laplacian with coefficients in E is then defined by APE := 5EdE + dE6ELet 7r : Z —> Y be a Riemannian submersion. Let F z := 7r*Fy with the pull-back structures. Fix a point siG G V and choose a local orthonormal frame s for Fy so that Vs(jo) = 0. Then dEY ( s y ® fajdy^iyo)
= ( s y ® <*/„,/ A dy'){y0)
since the connection 1 form of V E vanishes at j/o- Let s | = 7r*sy and let 7r«o = 2/o- We then have similarly that dEzw*{s$,®fa,idyl)(z0)
= ( 4 ® dzn*faJ A dj/7)(z0)
=?r*(s y ® dy/ 0>7 A dyT){y0). This shows dEzn* — ir*dEY = 0 at j/o- Since ?/o was arbitrary, we have (3.1.a)
dE z 7T* = T>*dEY-
We can perform a similar computation of the adjoint operators; the twisting owing to the coefficient bundle plays no role since we can choose the structures so the connection 1 form
256 and the first order derivatives of the metric on the coefficient bundle play no role. We then have (3.1.b)
<$Bz7r* - TC*6EY = 7T* ® (5Z7T* -
IT'SY).
We use equations (3.1.a) and (3.1.b) t o see: A^TT* - 7T*A^ = 7T* ® (A* 7T* - ^* A?.). Results of [9] then yield 3.2 T h e o r e m . Let n : Z —> Y be a Riemannian auxiliary geometric coefficient bundle overY.
submersion.
Let {Ey^hy^Y)
be an
(1) IfO j= $ e E{X, AqEy) and ifn*
We have given a fairly computational proof of assertion (1) of Theorem 1.1 in §2 as the paper is not intended for experts in the field. However, a more abstract proof is available along the lines given in in the real case in §3. We define X := 5Ezir* - ■K'hy - 7T* ® (5Z7r* - 7r*Jy) on C ° ° ( £
= 5ab and dhab(P)
= 0.
We have X vanishes if h = 5 since the twisting bundle plays no role in that instance. The operator 8 does not involve h; h enters only in computing the adjoint. Thus X apriori is linear in the 1 jets of h as is evidenced by the computatations performed in §2.
257 But these jets vanish at P. Since X is natural, X vanishes identically and the desired intertwining formula follows. Thus Theorem 1.1 hold for arbitrary (p, q); we omit details in the interests of brevity. It is a pleasure to thank the referee for pointing this out to us. Normally only the Dolbeault operator Ag with coefficients in a holomorphic vector bundle is considered since E ® A<™> = {E ® A<*°>} ® A<°*> so A<™> = A < ^ ( „ 0 ) . However, in the present instance, it does make sense to consider the operator A^ for p / 0 separately. Set EY = E0 ® \b>M(Y). Then 7r*(£y) = n*(E0) 0 A^U* which is a holomorphic sub-bundle of 7r*(£o) ® A(p'°)(Z). Since the orthogonal complement of TT*(£;0) ® A&^H* is not in general a holomorphic sub-bundle of n*{E0) ® A^' 0 ', A^f does not split as a direct sum operator; the obstruction is, of course, the integrability tensor for H^o; if that vanishes and if the fibers are minimal, all the operators coincide. The arguments in the real setting are much easier than the corresponding arguments in the complex setting since we can choose structures which are flat to second order in the real setting but not in the complex setting. Never the less, no essential new phenomena arise by considering coefficient bundles in either the real or complex contexts. This is somewhat surprising since new phenomena do arise in the context of index theory. For example in the real context, the untwisted signature complex always has index zero if m = 2 mod 4; a non trivial index can be generated by choosing a suitable coefficient bundle. In the complex context, the study of meromorphic functions on a Riemann surface is essentially the study of the operator &L where L is the holomorphic line bundle defined by a suitable divisor class; the Riemann-Roch formula gives the index of this elliptic operator. REFERENCES [I] [2] [3] [4] [5] [6] [7] [8] [9] [10] [II] [12]
L. Berard Bergery and J. P. Bourguignon, Laplacians and Riemannian submersions with totally geodesic fibers, Illinois J. Math. 26 (1982), 181-200. F. Burstall, Non-linear functional analysis and harmonic maps, Ph D Thesis (Warwick). P. B. Gilkey, Tnvariance Theory, the Heat Equation, and the Atiyah-Singer Index theorem (2 edition), ISBN 0-8493-7874-4, CRC Press, Boca Raton, Florida, 1994. P. B. Gilkey, J. V. Leahy, and JH Park, The spectral geometry of the Hopf fibration, J. Phys. A. 29 (1996), 5645-5656. , Eigenvalues of the form valued Laplacian for Riemannian submersions, Proc. Amer. Math. Soc. 126 (1998), 1845-1850. , Eigenforms of the spin Laplacian and projectable spinors for principal bundles, J. Nucl. Phys. B. 514 [PM] (1998), 740-752. , Spectral Geometry, Riemannian Submersions, and the Gromov-Lawson Conjecture, ISBN 0-8493-8277-7, CRC Press, 1999. , The eigenforms of the complex Laplacian for a holomorphic Hermitian submersion, Nagoya Math. J. 156 (1999), 135-157. P. B. Gilkey and JH Park, Riemannian submersions which preserve the eigenforms of the Laplacian, Illinois J. Math. 40 (1996), 194-201. S. I. Goldberg and T. Ishihara, Riemannian submersions commuting with the Laplacian, J. Differ ential Geom. 13 (1978), 139-144. S. Gudmundsson, The Bibliography of Harmonic Morphisms, (This bibliography is available on the internet at http://www.maths.lth.se/matematiklu/personal/sigma/harrnonic/bibliography.html). D. L. Johnson, Kaehler submersions and holomorphic connections, J. Differential Geom. 15 (1980), 71-79.
258 [13] [14] [15] [16] [17]
Y. Muto, Some eigenforms of the Laplace-Beltrami operators in a Riemannian submersion, J. Ko rean Math. Soc. 15 (1978), 39-57. , Riemannian submersion and the Laplace-Beltrami operator, Kodai Math. J. 1 (1978), 329338. JH Park, The Laplace-Beltrami operator and Riemannian submersion with minimal and not totally geodesic fibers, Bull. Korean Math. Soc. 27 (1990), 39-47. B. Watson, Manifold maps commuting with the Laplacian, J. Differential Geom. 8 (1973), 85-94. , Almost Hermitian submersions, J. Differential Geom. 11 (1976), 147-165.
DEPARTMENT
OF
MATHEMATICS,
HONAM
UNIVERSITY,
SEOBONGDONG
KWANGJU, 506-714 SOUTH K O R E A
E-mail address: jhparkChonam.honam.ac.kr R E C E I V E D S E P T E M B E R 30,
1999
59,
KWANGSANKU,
Geometry and Topology of Submanifolds X eds. W. H. Chen et al. (pp. 259-265) © 2000 World Scientific Publishing Co.
259
Willmore Surfaces and Minimal Surfaces with Flat Ends Chia-Kuei Peng
Liang Xiao*
Abstract In this paper, the relations between Willmore surfaces and flat-type minimal sur faces in R3 are discussed, and the number of ends of genus zero flat-type minimal surfaces is determined. Key Words: Willmore surface, flat minimal surface, Weierstrass representation 1991 Mathematics Subject Classification: 53A05, 53A10, 53C40, 53C42.
1. Introduction R. Bryant [1] pointed out a very interesting relation between Willmore surfaces and minimal surfaces with flat ends. Let M be a compact surface in R3 and X : M <-> R3 be the inclusion immersion. Consider the functional
W(X)=
f (H2-K)dA,
(1)
J M
where H, K and dA are the mean curvature, the Gauss curvature and the volume element, resp., of the immersion X. The surface M is called Willmore surface if the first variation of the functional W(X) is stationary at X. In 1973, White [9] pointed out that the two-form (H2 — K)dA has the property of being invariant under conformal transformations of R3. It is a standard result that a conformal transformation of R3 can be decomposed into a product of Euclidean motions, constant scale transformations and inversions. The only non-trivial case is that to prove (H2 — K)dA is invariant under inversion [10]. The Euler-Lagrange equation of the functional W is AH + 2H{H2-K)=0
(2)
where A is the Laplacian of the induced metric on M. The equation (2) is also invariant under conformal transformations of R3. The general problem is to find all Willmore surfaces and study their geometric properties. It is interesting to notice that H = 0 is a solution of the equation (2), which means that minimal surfaces in R3 are candidates for Willmore surfaces. But the trouble is that •The project is supported by No. 19531050 of NSFC and CAS This paper is in final form and no version of it will be submitted for publication elsewhere.
260 minimal surfaces in R3 can never be compact. Fortunately, the equation (2) is invariant under conformal transformations, so if there exists some kinds of special minimal surfaces in R3 which can be transformed as smooth compact surface in R3 under inversion, then we find Willmore surfaces. This observation leads to the following definition. Let M be a complete minimal surface with finite total curvature and embedded ends in R3. Then after a rotation of the coordinates, each end of M can be written as follows (not simultaneously) X3 = a log{Xl + X2) + b + C ^ + g
+ o(|AT 2 )
(3)
for suitable constants o, b, c, d [8]. Definition. Let M be a complete minimal surface with finite total curvature and embedded ends. The end Ei of M is called flat if a = 0 at Ei. M is called of flat-type if all the ends of M are flat. Let M be a flat-type minimal surface in R3. Let X0 $ M and Ix0 be the inversion with center X0- Then IXo(M) is a Willmore surface [1], In [1] Bryant proved that if M ~ S2 then the converse is true in the following sense. Let M C R3 be a Willmore surface. Then there exists a unique point XQ on M such that IXo(M) is a flat-type minimal surface in R?. Due to Bryant's results, the classification of Willmore surfaces which are homotopic to S2, is reduced to the problem of the classification of flat-type minimal surfaces conformal to S2-{Pu...,Pk}. In the same paper [1], Bryant gave one example which has 4 ends. He mentioned that a similar situation holds for k = 2(n + 1), n > 2, without stating the examples. Furthermore, he said that he does not know whether there exist flat-type minimal surfaces with odd ends. In this paper, we will show the existence and nonexistence of flat-type minimal surfaces in R3. In fact, we will prove the following theorems. Theorem A. There exist genus 0 flat-type minimal surfaces with k ends, where k is any integer except 2,3 and 5. Theorem B. There do not exist genus 0 flat-type minimal surfaces with k ends, where fc = 2,3 or 5. We are grateful to professor S.S. Chern for his continuous guidance and encouragement.
2. Proof of Theorem A and B In this section, we will prove the theorems A and B stated in the introduction. For the sake of completeness, we first recall the classical theory of minimal surfaces in R3 and some results from [2] with some modifications. Let M be a complete minimal surface conformal to S2 — {Pi,..., Pk}. By taking isother mal parameters on M, the classical Weierstrass representation says that the coordinate func tions (xi,X2,x3) on M in R3 can be represented as Xj =Re J §jdz,
j = 1,2,3,
where * t = 1/(1 - 2), $ 2 = 5^1/(1 + g2), * 3 = /•
261 This representation allows us to translate the geometric properties of minimal surfaces in B? into analytic data for the functions $j. Now we list some without proof. (i) M is an immersion <=> {$,} have no common zero points. (ii) M has finite total curvature o {$;} are meromorphic functions on S2. (iii) The ends of M are embedded ■£> {$;} have poles of order at most 2. (iv) M is regular (i.e. single valued) •» {$;} have no real periods for any T € Hl(M, Z). From these relations it is not difficult to see that if M is a complete minimal surface conformal to S2 - {Pi,..., P/t}, then the ends of M are embedded if and only if the total curvature satisfies C(M) = —4ir(k — 1). By the definition of a flat-type minimal surface, we have Lemma 1. Let M be conformal to S 2 — {Pi,..., P/t}. Then M is a flat-type minimal surface if and only if the functions $;-, j = 1,2,3, satisfy (i) {$j} have poles of order < 2. (ii) {$jdz} are exact, i.e., the functions $3- have no periods. Lemma 2. $jdz,j = 1,2,3, are exact if and only if fdz, fgdz, fg2dz are exact. As we mentioned before, since M has finite total curvature and embedded ends, we can assume O P2 g = | and / = -k (4)
U(z-
atf
3=1
where P,Q are polynomials without common zero points and max(deg P, deg Q) = k — 1. From Lemma 2 it follows that the exactness of <&jdz is equivalent to the exactness of the following holomorphic differentials P2 ~k
Q2
a n d
' 1
II (z - ajY 3=1
U(z-
a3f
3=1
,
PQ 1
U(z-
,.,
2
■
(5)
a,-)
3=1
Now we give a proof of theorem A. First we will treat with the case that k is even which is much easier. Actually we will assume that Y[{z - a3) = z{z2n+1 - 1 ) , n > 1, 3=1
(6) 2n+1
z
+■
n+\ In order to show the exactness, we need some lemmas. Lemma 3. Let tj, j = 0,..., m — 1, be the m roots of zm = 1, then
262 L e m m a 4.
Assume that
(8) where F(z) is a polynomial and F(a,j) ^ 0 for all o3-. Then W{z)dz F'(ai) F(at)
=
^ 1 ^ ■
1,...
,k,
where F ' ( O J ) =
is exact if and only if dF dz
(9)
By Lemma 3 and 4 it is quite easy to check the exactness of all the terms in (5) defined by (6). For instance, we see that W(z) =
(z2n+1 + £l)2 z2{z2n+i-iy
Q2
2n+i 2 -i) z^{z
(10)
When z = 0, we have
Q2
d
o.
S ^ ^ - i ^ - i ) . When z = u, we have (2n + l)z 2 :
Q'
2n+l 1
~
1
.
,,1
C
C
Z
^_Ll
(11)
I
On the other hand, we have
•>+E. ■
1
(2n + l ) - l l
- +-
^
,
n
,l
= (n + l ) - .
(12)
For the other two terms in (5), it is similar, so we have P r o p o s i t i o n 1 For any even integer k = 2(ra + l ) , n > 1, there exist genus 0 flat-type minimal surfaces in Ft? with k ends. Now we will treat the other case that k is odd. In this case, we will assume Y[(z-aj)=z(zn-\)(zn-\),
n>4,
P = zm(zn-c),2<m
n + \^2m, n
Q = (*" - a)(z
- b),
where {A, a,b, c} are constants to be determined. L e m m a 5.
P2 2
n
z {z -i)(zn-\y
-dz
(13)
263 is exact if and only if c and A satisfy n m-l
n — 1 = — +
—
+
n
. s d4)
,
r r i
nX n - 1 nA m - 1+ = —r h • A-c 2 A-l' L e m m a 6.
91
dz
2
« (*"-l)(z»-A)2 is exact if and only if a, b and A satisfy n n n — 1 n 1 1-a + ^ 1 - 6 = - - 2^ + ^-T> 1-A nA A-o
nA A-6
n — 1
(15)
nA
1 + —2 ^ + A - l '
L e m m a 7. If n > 4 and 2 < m < n — l , n + l ^ 2m, then the equations (14) and (15) have a solution {A, a, b, c] such that A, a, 6, c are pairwise distinct and different from 0 and 1. The proofs of these lemmas are long but straight forward computations. We will leave them to the reader. L e m m a 8 [7]. There exist genus 0 flat-type minimal surfaces in B? with 7 ends. According to these lemmas, we have P r o p o s i t i o n 2. For any odd integer k = 2n + l , n > 3, there exist genus 0 flat-type minimal surfaces in R3 with k ends. By Proposition 1 and 2, we obtain Theorem A. We need a couple of simple lemmas to prove Theorem B. L e m m a 9 If
2
A, Li=i z ~ z' has no periods, then
m
A, —=0.
£
dz
(16)
J = l,..-,m,
(17)
3*1
and vice versa. Proof.
y^z-zj
2r{z_zi?^z^i{z_Z]){z^Zi) A
T
l (z~
| z 2
')
A
= y
2
*
z z 2
i( ~ i)
y
A,A
I
j+i (zi ~ z') \
l_
~ z3
z
~
Aj
12V i v i\U
1 z
z
i-
z
,)
z
-
z
i'
z
i
264 Lemma 9 follows from the fact that the exactness is equivalent to the vanishing of the coefficients of all the terms (z - 2|) _1 . Lemma 10 Suppose all the holomorphic differentials in (5) are exact and let A{a\, 02,..., a*) denote the matrix with entries ay such that 0, - J —,
if i = j , if i + 3-
at—a* '
'
J
Then rank A < k — 2. Proof. Write
_4
Q
^
k
t—i A _ „,' . *T ^ ^ E «=i"*~ >"*' fl -a. Uz-ai
(=1
/=i
Y[z-ai
7
a
x
B, l=lZ~a'
Then it follows from the exactness of all the holomorphic differentials in (5) and Lemma 9 that m
m
A
p
E ^ ^ = 0, £ - ^ - = 0, / = !,.. .,m. £? 03 - ai
^ a, - a,
Since P and Q have no common zero points, (A\,..., independent. Therefore rank A < k - 2.
Ak) and ( B i , . . . , B^) are linearly
Proof of Theorem B. Notice that for pairwise distinct complex numbers 01,02,03, o4 and 05, ranfc yl(oi,o2) = 2, ronfc J4(a1,02,a3) = 2 and ranfc A(ai, 02,03,04,05) = 4. Hence Theorem B follows from Lemma 9, 10 and the above fact. Remark. ends.
In [11], Xiao proved the nonexistence of genus 0 flat minimal surfaces with 5
An immediate consequence of Proposition 1 and 2 is the following Theorem C. There exist Willmore surfaces X : S2 —> R3 such that / H2dA = 4nk x
(18)
where k is any integer except 2,3 and 5. Proof. From the conformal invariance of (H2 — K)dA we have f(H2 - K)dA = f{H2 - K)da = -C(M)
= 4m{k - 1),
(19)
M
where da is the volume element of of the the surface surface M. M. Furthermore, Furthermore, from from the the Gauss-Bonnet Gauss-Bonnet Theorem j KdA = j KdA = 4TT. (20) x s2 Hence, we have /H 2 dA = 4-Kk. x
265 Remark. Li and Yau [6] showed that if tp : M ^-> R3 is any smooth immersion and k is the maximum number of preimages of a point in -R3 under tp, then f HldAj, > 4irk.
(21)
M
Our theorems show that equality in this inequality can be realized for Willmore surfaces of S2 in R? (except k = 2,3 and 5).
References [1] R. L. Bryant, A duality theorem for Willmore surfaces, J Diff Geo, 20(1984), 23-53 [2] L.G. Jorge and W.H. Meeks, The topology of complete minimal surfaces of finite total Gaussian curvature, Topology, 22(1983), 203-221. [3] B. Lawson , Lectures on minimal surfaces, IMPA (1973). [4] H.Z. Li, C.P. Wang and F.E. Wu, The Mobius minimal surfaces in Sn, preprint. [5] P. Li and S.T. Yau, A new conformal invariant and its applications to the Willmore conjecture and the first eigenvalue of compact surfaces, Invent. Math. 69(1982), 269-291. [6] R. Osserman, A survey of minimal surfaces. Van Nostrand Reinhold, New York, 1969. [7] C.K. Peng and L. Xiao, Existence of fiat-type minimal surfaces with 7 ends in ft3, J. of the Graduate School, Academia Sinica, 16(1999), 89-94. [8] R. Schoen, Uniqueness, symmetry and embeddedness of minimal surfaces, J Diff Geo, 18(1983), 791-809. [9] J.H. White, A global invariant of conformal mappings in space, Proc. Amer. Math. Soc, 38(1973), 162-164. [10] T.J. Willmore, Total curvature in Riemannian geometry, John Wiley & Sons, New York, (1982). [11] L. Xiao, On complete minimal surfaces with parallel and flat ends, Lectures Notes in Mathe matics 1369, 289-294. Chia Kuei Peng and Liang Xiao Department of Mathematics, Graduate School at Beijing, University of Science and Technology of China, P.O.Box 3908, Beijing 100039, P.R. China email: pengck(5)public.fhnet.cn.net [email protected] Received December 24, 1999
266
Geometry and Topology of Submanifolds X eds. W. H. Chen et al. (pp. 266-273) © 2000 World Scientific Publishing Co.
INTRINSIC PROPERTIES OF REAL H Y P E R S U R F A C E S I N COMPLEX SPACE FORMS
PATRICK J. RYAN A B S T R A C T . We give a simple argument for the nonexistence of Einstein hypersurfaces in the complex space forms CPn and C f f " which is valid for all dimensions n > 2. In addition, we survey classification results t h a t are stated in terms of intrinsic geometrical properties of the hypersurface. Many questions that have been settled for Hopf hypersur faces a n d / o r for dimensions n > 3 still remain open in the general case. K E Y W O R D S : real hypersurface, complex space form, Einstein hypersurface. 1991
M A T H E M A T I C S S U B J E C T CLASSIFICATION: 53B25
§0.
INTRODUCTION
In this paper we will be concerned with real hypersurfaces M 2 n _ 1 in the complex space forms of constant holomorphic curvature 4c ^ 0, namely the complex projective space C P n and the complex hyperbolic space CHn. In the first few papers on this topic (Lawson [L], R. Takagi [T], Okumura [O], Y. Maeda [Ma], Kon [Ko]) it turned out that for the standard examples, the structure vector field W was a principal vector. Conversely, classification arguments were greatly simplified when this property held. Later, Berndt [B] defined Hopf hypersurfaces to be those for which the integral curves of W are geodesies and showed that this was equivalent to specifying that W be principal vector. In 1982 the present author and T.E. Cecil [CR] found a geometric meaning for this condition in the case where the ambient space is CPn by proving (roughly speaking) that Hopf hypersurfaces are tubes of constant radius over complex submanifolds of CPn. Over the intervening years, more than one hundred papers have appeared which relate to the work cited above. The chapter in the Kuiper memorial volume [NR1] was an attempt to make these results more accessible by assembling them in a coherent fashion and by including complete proofs. However, limitations of space and time dictated that some results were merely stated and some topics (e.g. generalizations to quaternionic projective and hyperbolic spaces) had to be omitted entirely. In this paper, we focus more narrowly on theorems that are expressed in terms of intrinsic properties of the hypersurfaces. We will not attempt to give attributions for Partially supported by t h e Natural Sciences and Engineering Research Council of Canada. This paper is in final form and no version of it will be submitted for publication elsewhere.
267 all theorems quoted in what follows. For such detail, as well as other background and notation, we refer the reader to [NR1]. §1.
PRELIMINARIES
We begin with the standard examples of Takagi [T] in C P " and Montiel [Mo] in CHn which have become well-known and have even acquired a standard nomenclature. They consist of the following: In complex projective space, C P " : (Al) (A2) (B) (C) (D)
Geodesic spheres. Tubes over totally geodesic complex projective spaces CPk, 1 < k < n - 1. Tubes over complex quadrics and R P " . Tubes over the Segre embedding of C P 1 x C P m where 2m + 1 = n and n > 5. Tubes over the Pliicker embedding of the complex Grassmann manifold C?2,5These occur only for n = 9. (E) Tubes over the canonical embedding of the Hermitian symmetric space SO(W)/U(b). These occur only for n = 15.
In complex hyperbolic space, CHn: (AO) Horospheres. (Al) Geodesic spheres and tubes over totally geodesic complex hyperbolic hyperplanes. (A2) Tubes over totally geodesic CHk, 1 < k < n - 2. (B) Tubes over totally real hyperbolic space RHn. All these hypersurfaces are homogeneous, and it was this property that motivated Takagi to create his list. For more detailed information on these standard examples, see [NR1], in particular pp. 254-262. The fact that a Riemannian manifold occurs as a hypersurface in a complex space form imposes severe restrictions on its intrinsic geometry. We list several conditions that the curvature of a Riemannian manifold can satisfy: (1) constant sectional curvature (2) locally symmetric (3) Einstein (4) semisymmetric (5) parallel Ricci tensor (6) cyclic parallel Ricci tensor (7) Ricci semisymmetric (8) Ricci recurrent (9) harmonic curvature (10) cyclic Ryan The relationships among these conditions for a general Riemannian manifold include the following: (i): (1) =► (2), (3)
268 (ii): (iii): (iv): (v): (vi):
(2) (3) (4) (5) (7)
=> =► => => =►
(4), (5) (5) (7) (6), (7), (8), (9) (10) §2.
T H E EINSTEIN CONDITION
A Riemannian manifold is an Einstein space if its Ricci tensor S satisfies S = pi for some constant p. Einstein spaces cannot occur as hypersurfaces in C P " or CHn for n > 2. This non-existence question was first considered by Y. Maeda [Ma] who proved it for C P " when n > 3 subject to a certain lower bound on the scalar curvature. Kon [Ko] and Montiel [Mo], in papers devoted primarily to more general conditions, provided proofs covering C P " and CHn respectively. Cecil and Ryan [CR] also proved nonexistence as a corollary to their more general theorem. After that, researchers went on to investigate weaker conditions, attempting to find characterizations of the standard examples or to establish further nonexistence results. The case n = 2 was not explicitly addressed until the recent paper of Niebergall and the author [NR2] in which the nonexistence of Einstein hypersurfaces in C P 2 and CH2 was established. A closer look at the arguments in [CR] reveals a direct proof that a hypersurface satisfying the Einstein condition must in fact be Hopf. The relevant portion of the argument is valid for n = 2. This can be used to construct a simple proof of nonexistence that works for all n > 2. We now present this proof. The theorem is stated as follows: Theorem. Let M2n~1, where n > 2, be a real hypersurface in a complex space form of constant holomorphic sectional curvature 4c =^ 0. Then M cannot be an Einstein space. The key step is to prove the following lemma. Lemma. 7 / M 2 n _ 1 , where n > 2, is an Einstein hypersurface in a complex space form of constant holomorphic sectional curvature Ac ^ 0 ; then M is a Hopf hypersurface. Proof. The Ricci tensor is expressed in terms of the shape operator A by SX = (2n + l)cX - 3c {X, W) W + mAX - A2X where m = trace A is the mean curvature. Writing P = mA — A2 and assuming the Einstein condition S = pi, we get PX - 3c (X, W) W - aX = 0 where a = p — (2n + l)c. Then PW = (a + 3c)W while if X is any vector in Wx, PX = aX. Since c / 0, a + 3c has multiplicity 1 as an eigenvalue of P . On the other hand, every eigenvector of A is also an eigenvector of P . If an orthonormal basis of
269 principal vectors is chosen, one of these vectors must be ±W. This shows that W is a principal vector and that M is a Hopf hypersurface.
□ In order to complete the proof of the theorem, we proceed as follows. Using the fact that W is principal and hence W± is invariant by A, we have A2 — mA + al = 0 on Wx. The following basic facts [NR1, pp. 245-246] will be useful. L e m m a . Let M 2 ™ -1 , where n > 2, be a Hopf hypersurface in a complex space form of constant holomorphic sectional curvature Ac ^ 0. If AW = aW and tp is the tangential projection of the complex structure J, then (1) IfXeW-1 and AX = XX, then (\-%)A
(*)
while if X £ Wx is an eigenvector of A with eigenvalue A, then mA - A2 = a.
(**) There are three cases to consider.
Case 1: Tx is not (^-invariant. Then there is a second eigenvalue n such that AipX = fnpX. Of course, m/i — /J,2 = a, and hence A + \i = m. There can be no additional principal curvatures since all principal curvatures on Wx satisfy the quadratic equation t2 — mi + a = 0. Calculating the mean curvature yields m = a + (n — 1) (A + n) and hence (n - 2)m + a = 0. If n = 2 we immediately have a = 0, a = —3c = c from (*) and (2) of the lemma, respectively, contradicting c ^ 0. Proceeding with n > 3, the same two equations yield ma o / x 3c + — + c = 3c + a = ma — a = (n — ljma so that 0 > (n — | ) m a = 4c. If c > 0, this is a contradiction. If c < 0, we resort to the standard focal set calculation [Mo, p. 521]. Since M is a Hopf hypersurface with
270 its particular configuration of principal curvatures, we have X/J, = —c so that a = —c. Then (2) of the lemma gives 0 > ^ = —2c, a contradiction. Case 2: T\ is (^-invariant but T\ =fi W^. Then A2 = aX + c and there is a second principal curvature \i on W1- satisfying the same equation. (There can be no third principal curvature, for the same reason as in Case 1). Thus X + fi = a and X/J, = —c. On the other hand, since A and \x satisfy (**), we get a = m and c = —
In 1989 U-H. Ki proved the following theorem which was stated as Theorem 6.15 in [NR1, p. 278]. T h e o r e m . Let M 2 n _ 1 , where n > 3, be a real hypersurface in a complex space form of constant holomorphic sectional curvature 4c ^ 0. Then the Ricci tensor of M cannot be parallel everywhere. Ki did not include the dimensional restriction in his statement. However, his method of proof was to show that the Ricci-parallel condition implies the Einstein condition and then to apply Kon's and Montiel's results [Ki, p. 80]. Kon and Montiel, however, were explicit about assuming n > 3 (see [Ko, p. 351] and [Mo, p. 516]). In light of the result of the preceding section, we can now regard Ki's theorem as being proved for all n > 2. §4.
CYCLIC-PARALLEL RICCI TENSOR AND HARMONIC CURVATURE
In a 1978 paper [G], entitled Einstein-like manifolds which are not Einstein, A. Gray discussed two classes of Riemannian manifolds defined by properties of the Ricci ten sor. He showed how they were related to more familiar classes and produced examples to show that the classes were distinct. The two classes in question were those mani folds having cyclic-parallel Ricci tensor and those having harmonic curvature. We now consider these classes in the context of real hypersurfaces. The Ricci tensor of a Riemannian manifold is said to be cyclic-parallel if the cyclic sum
<(Vjr5)y, Z) + ((VYS)Z,
X) + ((VZS)X,
Y)=0
271 for all X, Y, and Z. After considering conditions that are too strong to permit immersion into a complex space form, we have finally arrived at one that actually occurs. Theorem. Let M 2 n _ 1 , where n > 2, be a Hopf hypersurface in a complex space form of constant holomorphic sectional curvature Ac > 0. If the Ricci tensor S is cyclic-parallel, then M is an open subset of a hypersurface from Takagi's list. There is a restriction on the radii that can occur. The "cyclic-parallel Ricci" condition is noteworthy in that it is an intrinsic property. On the other hand, the theorem itself is not an intrinsic one. The question of whether there is a wider class of examples when the Hopf condition is not imposed remains open. Note also that the theorem says nothing about hypersurfaces in CH". This theorem was stated as Theorem 6.21 in [NR1], p. 279. However, the statement there contains a typographical error. The word real should read Hopf. A Riemannian manifold is said to have harmonic curvature if its Ricci tensor S is a Codazzi tensor, i.e. (VxS)Y = (VyS')X. Concerning this condition, a non-existence theorem for Hopf hypersurfaces was stated as follows in [NR1]. Theorem. Let M2n~1, where n > 3, be a Hopf hypersurface in a complex space form of constant holomorphic sectional curvature 4c ^ 0. Then M cannot have harmonic curvature, that is, (VxS)Y — (VYS)X cannot vanish identically. The general existence question is still open. When n = 2, it is open even for Hopf hypersurfaces. §5.
SEMISYMMETRY AND RELATED CONDITIONS
If a tensor T is not parallel, i.e. VT / 0, it may be that the curvature operators R(X, Y) annihilate T. When a Riemannian manifold M satisfies R ■ R = 0, M is said to be semisymmetric while if R ■ S = 0 it is called Ricci-semisymmetric or a Ryan space [NR1, p. 283]. Again, these conditions are too strong for hypersurfaces in CPn and CHn, at least for n > 3. The following non-existence theorem occurs as Theorem 6.29 in [NR1]. Theorem. In a complex space form of constant holomorphic sectional curvature Ac / 0, there exists no real hypersurface M 2 " - 1 , n > 3, satisfying R ■ S = 0. For n = 2, there are no Hopf hypersurfaces satisfying R ■ S = 0. A Riemannian manifold M is said to be cyclic-Ryan if the cyclic sum over tangent vectors X, Y, and Z of (R(X,Y) ■ S)Z vanishes. It was proved by Ki, Nakagawa, and Suh [KNS] that this intrinsic condition implies the (non-intrinsic) Hopf condition, which leads to its equivalence to another non-intrinsic condition pseudo-Einstein, at least when n > 3. A hypersurface M in a complex space form is pseudo-Einstein if there are constants p and o such that S = pi on W1- but SW = (p + a)W. The classification of pseudo-Einstein hypersurfaces due to Kon and Montiel, together with
272 the work of Ki, Nakagawa, and Suh gives us the following intrinsic theorem (compare Theorems 6.1 and 6.30 of [NR1]). Theorem. Let M 2 ™ -1 , where n > 3, be a real hypersurface in a complex space form of constant holomorphic sectional curvature 4c ^ 0. Then M satisfies the cyclic-Ryan condition if and only if it is an open subset of one of the following: (1) a geodesic sphere in CP™ or CHn. (2) a tube of radius ur over a complex projective subspace CPk in CP™, 1 < k < n - 2, where 0 < u < n/2, cot2w = k/(n — 1 — k), and c = 1/r 2 . (3) a tube over a complex hyperbolic hyperplane in CHn. (4) a tube of radius ur over a complex quadric Q™-1 in CP™, where 0 < u < n/4, cot 2 2u = n - 2, and c = 1/r2. (5) a horosphere in CHn.
§6.
RECURRENCE CONDITIONS
A Riemannian manifold M is said to have recurrent curvature if there is a 1-form a such that VxR = a(X)R for all tangent vectors X. Further, M is said to be Riccirecurrent if the Ricci tensor satisfies a similar identity VxS = a(X)S. Recently, Hamada [H] has obtained a non-existence theorem for Ricci-recurrent Hopf hypersurfaces in CP™. The theorem is valid for all n > 2. However, no intrinsic theorem in known. In particular, the general existence question is still open. REFERENCES [B]
J. Berndt, Real hypersurfaces with constant principal curvatures in complex space forms, Ge ometry and Topology of Submanifolds, II (Avignon, 1988) (1990), World Sci. Publ., Teaneck, NJ, 10-19. [CR] T. E. Cecil and P. J. Ryan, Focal sets and real hypersurfaces in complex projective space, Trans. Amer. Math. Soc. 2 6 9 (1982), 4 8 1 ^ 9 9 . [G] A. Gray, Einstein-like manifolds which are not Einstein, Geom. Dedicata 7 (1978), 259-280. [H] T. Hamada, On real hypersurfaces of a complex projective space with recurrent Ricci tensor, Glasgow Math. J. 4 1 (1999), 297-302. [Ki] U-H. Ki, Real hypersurfaces with parallel Ricci tensor of a complex space form, Tsukuba J. Math. 13 (1989), 73-81. [KNS] U-H. Ki, H. Nakagawa and Y.J. Suh, Real hypersurfaces with harmonic Weyl tensor of a complex space form, Hiroshima Math. J. 20 (1990), 93-102. [Ko] M. Kon, Pseudo-Einstein real hypersurfaces in complex space forms, J. Differential Geometry 14 (1979), 339-354. [L] H. B. Lawson, Jr., Rigidity theorems in rank-1 symmetric spaces, J. Differential Geometry 4 (1970), 349-357. [Ma] Y. Maeda, On real hypersurfaces of a complex projective space, 3. Math. Soc. Japan 28 (1976), 529-540. [Mo] S. Montiel, Real hypersurfaces of a complex hyperbolic space, J. Math. Soc. Japan 3 7 (1985), 515-535. [NR1] R. Niebergall and P.J. Ryan, Real hypersurfaces in complex space forms, Tight and taut submanifolds (Berkeley, CA, 1994), 233-305, Math. Sci. Res. Inst. Publ., 32, Cambridge Univ. Press, Cambridge, 1997.
273 [NR2] R. Niebergall and P.J. Ryan, Semi-parallel and semi-symmetric real hypersurfaces in complex space forms, Kyungpook Math. J. 38 (1998), 227-234. [O] M. Okumura, On some real hypersurfaces of a complex projective space, Trans. Amer. Math. Soc. 212 (1975), 355-364. [T] R. Takagi, On homogeneous real hypersurfaces in a complex projective space, Osaka J. Math. 10 (1973), 495-506. D E P A R T M E N T OF M A T H E M A T I C S AND STATISTICS, M C M A S T E R U N I V E R S I T Y , H A M I L T O N , O N T A R I O , C A N A D A , L8S
4K1
E-mail address:
pjrQmail.cas.mcmaster.ca R E C E I V E D F E B R U A R Y 23,
2000
274
G e o m e t r y and T o p o l o g y of Submanifolds X eds. W . H. C h e n et at. (pp. 2 7 4 - 2 8 3 ) © 2 0 0 0 W o r l d Scientific P u b l i s h i n g C o .
On t h e Nonexistence of Stable Minimal Submanifolds in Positively Pinched Riemannian Manifolds
Y I - B I N G SHEN*
HUI-QUN XU
ABSTRACT. In this paper, it is proved that there are no compact stable minimal submanifolds in a compact simplyconnected 0.77-pinched Riemannian manifold, so that a conjecture of Lawson-Simons is supported. In particular, the conjecture is true for complete ^-pinched hypersurfaces in Rm+1 with m > 3. Moreover, the instability of a compact domain of a minimal submanifold in 5-pinched manifolds is considered. KEYWORDS: 5-pinched Riemannian manifold; minimal submanifolds; instability 1991 MATHEMATICS SUBJECT CLASSIFICATION: 53C40; 53C20
§1.
INTRODUCTION
A minimal submanifold M in a Riemannian manifold M is the critical point of its volume functional. M is said to be stable if the second variation of the volume is always nonnegative for any normal deformation of M in M with compact support; otherwise M is called unstable. It is well known that there are no compact stable minimal submanifolds in the Euclidean spheres. A Riemannian manifold M is said to be 5-pinched for 0 < 8 < 1 if the sectional curvature K{M) of M satisfies S < K(M)
< 1 everywhere. In [LS] Lawson
and Simons proposed the following C o n j e c t u r e . There are no compact stable minimal submanifolds M in a compact simply-connected j pinched Riemannian manifold M. Aminov showed in [Am] that the conjecture is true if M is homeomorphic to a two-sphere. In [Ho], [HW] and [01], several results support this conjecture; but it is far from solving the conjecture. As far as 1 know, the problem has made yet no headway up to now. In this article we would like to give a report on some of our recent work concerning the nonexistence of compact stable minimal submanifolds in positively pinched Riemannian manifolds. Firstly, it is proved that there are no compact stable minimal submanifolds in a compact simply-connected 0.77-pinched Riemannian manifold (Theorem 1). As a direct generalization, there are no stable currents (or stable varifolds) in a compact simply-connected 0.77-pinched Riemannian manifold. The proof method of the above result has been used in [02] to prove the instability of harmonic maps into simply-connected 0.83-pinched Riemannian manifolds. Next, we consider the positively curved Euclidean hypersurfaces. It is proved that there are no compact stable minimal submanifolds on the j-pinched complete hypersurface in Rm+l
with m > 3 (Theorem 2). It
may be a better partial answer to the Lawson-Simons conjecture. 'Supported by NNSFC and NSFZP This paper is in final form and no version of it will be submitted for publication elsewhere.
275 Finally, our consideration is to deal with the instability of a domain D of a minimal submanifold M in positively pinched Riemannian manifolds. Let M be a compact simply-connected (S-pinched Riemannian manifold with 5 > 0.77. Then there exists a constant C(5) depending only on 6 such that if Ai(D)
for a domain D C M with the compact closure D and the smooth boundary dD, where M is an n-dimensional minimal submanifold in M and Ai(D) denotes the first eigenvalue of the Laplacian of D with the Dirichlet boundary condition, then D is unstable (Theorem 3). It can be viewed as a partial answer to the question in Remark 4 of [01]. When i5 = 1 this is just Theorem B of [01]. §2.
PRELIMINARIES
Let M be an tn-dimensional compact simply-connected <5-pinched Riemannian manifold with 5 > 1/4. On putting E = TM © e(M) where e(M) is a trivial line bundle on M with the canonical metric, we may obtain a Euclidean vector bundle E over M. Let e be a cross-section of length one in e(M). Define a metric connection V" on E as follows^ 1,2 1: V'XY = VXY
1 + <S\ 1/2 - [ ^ - j <X,Y>e, l /
V i e = ( i ± i )
(2.1)
\ ,
(2,)
where X, Y are vector fields on JW, < , > and V are the Riemannian metric and Levi-Civita connection of M, respectively. Then there is a flat connection V close to V" such that (cf.[Gl] and [01] for detail) II V ' - V " ||
(2.3)
where V - V" ||:= max{|| VXY
-V'XY\\:
Xs
TM, || X ||= 1, Y € E, || Y ||= 1}
t i W = | ( 1 - S)S~l U + (61'hm^6-1^2rA
, 1/2
*2(«)
(2.4)
:
Assume now that M *-> M be an n-dimensional compact minimal submanifold isometrically immersed in M. Let N(M)
be the normal bundle of M in M and V be a cross-section in N(M)
with compact support.
Then the second variational formula of M for V is HV, V)=
i ( E l l V ^ H
2
- ^ f
-Y^
where dv is the volume element of M, V
2
''=1
(2.5) >}dv,
and B are the normal connection and the second fundamental
form of M in M respectively, R is the curvature tensor of M, and {e4} is a local orthonormal frame field on M. Now suppose that M is isometrically immersed in the Euclidean space Rm+l by D the flat connection of Rm+l.
as a hypersurface. Denote
Then the second fundamental form of M in Rm+1 h(X,Y)em+1=DxY-WxY
is defined as
276 for X, Y £ TM, where em+\ is the normal vector field to M in Rm+1.
The following proposition is well
known from [LS]. P r o p o s i t i o n 2 . 1 . Let M be an m-dimensional
compact hypersurface in RmJrl
with the second fundamen
tal form h. If at each point of M m
£
n
£ { 2 M e * , e r f - ft(e;, e J )/»(e r , er)} < 0
(2.6)
r=n+l<=1
/or any /oca/ orthonormal frame field {e*,e r } on M , where 0 < n < m, rTien t/tere are no compact stable minimal n-dimensional
submanifolds in M.
From now on we make use of the following convention on ranges of indices unless otherwise stated: 1 <&>.?>'••< n i
1 < a, /?,7, • ■ • < m;
n + 1 < r, s, • • ■ < m.
Let x € M be an arbitrary point and let Aa be principal curvatures of M corresponding to the principal directions {e Q } which form an orthonormal basis at x € M. Clearly, in such a basis, the second fundamental form h of M is diagonalized, i.e., /i(e Q , e^) = \a6ai3- Thus, from the Gauss equation of M in Rm+1 it follows that Rafiafi = A«A„
(a # P)
(2.7)
at i 6 M, where Raffaff = < R(e a ,e/j)e0,e a >. On putting
ea = Y,aeae0
(2.8)
for a special orthogonal matrix (a„), we have from (2.6) and (2.8) 53{2Me<,e r ) 2 - Met,e<)/i(er,e r )}
=5>A„
2( £ " > ? )
( £ o?o? ) - £ ( « ? ) V ) ' }
(2.9)
= £ ( A a ) 2 ( a ? ) 2 ( a ? ) 2 + F(n,m), where F(n,m) = £
AaA3 { 2 ^ a ? a f a > f - £ ( a ? ) 2 ( 0 2 } .
(2.10)
§3. POSITIVELY PINCHED RIEMANNIAN MANIFOLDS
In this section, we prove the following T h e o r e m 1. There are no compact stable minimal submanifolds in a compact simply-connected 0.77pinched Riemannian manifold M. Proof. Consider the vector bundle E over M constructed as in §2. Let W be a parallel cross-section of E with respect to V . Let W™ respectively. Set U = W™.
and
ff'M
be the TM-component and the e(M)-component of W,
Assume that M be an n-dimensional compact minimal submanifold in M with
277 n < m — dimM. Let UT and UN be the TM-component and the 7V(A/)-component of U, respectively. We take V = UN in (2.5). Thus we have V ^ V = ( V e . V ) " = (Ve,C/ T UnNI M
= (V^.W) '
- 1
-
VeiUT)N (VeiUT)N,
from which and (2.1) it follows that
E I ^y ii2 = £ I (v;w)™™w f+J2\\ (ve,uTr II2 i=i
i=i
i=i n
W
T N
-2Y,
(3.1)
= Eii(v^)™
nw M
2
s
' »ii +En ^.
f/T
)ii
2
where B is the second fundamental form of M in M. Since M is i-pinched, then n
- £ < f l ( e f , V ) V , e ; > < -«<5 || V || 2 .
(3.2)
Substituting (3.1) and (3.2) into (2.5) yields that J(V,V) < /
q(W)dv,
(3.3)
where
1(W) = £ || (V^)™nW<M> |» + £ || flfc,^) ||2 i=l
i=l n
- 2^
n
< V';(W, ( V ; > j ) w X
W,e, > - E
< *»(<*■ e;), V > 2
(3.4)
-n<5||y||2. Define W = {W e r ( £ ) : V ' W = 0}. Then W is isomophic to flm+1 and has a natural inner product. We define a quadratic form Q on W by
Q(W) = [ q(W)dv, JM
where q(W) is defined by (3.4). If we choose an orthonormal basis {WA} (A = 1, ■ ■ • , m + 1) of W and put UA = (WA)™ and VA = (C/A) N , then we have from (3.3) ro+1
m+1
Y,"y^vA)
-
= tTQ=
/ ( tr 9) d «-
(3-5)
Since the trace of q is independent of the choice of the orthonormal basis for each fiber of E, at any point l E M w e choose an orthonormal basis {WA} such that {Wi = ej} are tangent to M and {Wr} are normal
278 to M. Then, by using (3.4) and (2.3), at x we have m+l
n
m+l n
*«= E E ii «^)™™< M > if + E E ii *(«."!) II2 A=l i=l m+l
n
>< w ^ i >
- 2 E E < KWAWW A = I t,j=i m+l n
m+l
" E E < B{fii,es),VA >2 -n* E II V* II' A=l t,i=l m+l n m
A=l n
= E E E
< <WA,WT
>2 + || B f -2 E II (V^e,-)" II2
A=l i=l r = n + l
(3.6)
i,j=l
- ||B|j2 -n(m-n)<$ m+l
n
m
- 5 I 1 I H < WA,V'e';W>>2-n(m-n)<5 A=l t=l r~n+l n m :
E
\\K^r\\2-n(m-n)6
E
§4.
Let R
m+1
0.77-
POSITIVELY CURVED EUCLIDEAN HYPERSURFACES
be a Euclidean (m + l)-space and M a hypersurface in Rm+1.
We have the following
T h e o r e m 2. There are no compact stable minimal submanifolds on the ^-pinched complete hypersurface ~M inRm+l. Proof. First of all, the pinching condition for curvature in Theorem 2 implies that M is compact by the theorem of Bonnet-Meyers. Since M is j-pinched, then (2.7) yields that i
(a*P),
(4.1)
which implies that all of {AQ} are nonnegative and have the same sign. So, without loss of generality, we may assume that at a point x € M 0 < Ai < A2 < • ■ • < A m .
(4.2)
By (4.1) and (4.2), one can see that \
a
> - for a ^ 1; \
m
<2
and
Aa < 1 for a^m\ Ai > - 4
if m > 3. -
(4.3) (4.4)
279 Since (a£) in (2.8) is a special orthogonal matrix, then we have
( E ° ? a n + ( E < a n +2X>? a ? a M = <5a<3.
(4.5)
Set that
Ga,m = 2(j2a?aTj \ i
+ 2 ( £ < < C ) + E « ) 2 ( C ) 2 + E( a D 2 K) 2 .
/
\
r
/
i,r
(4-6)
i,r
which is nonnegative. By virtue of (4.1), (4.5) and (4.6), it follows from (2.10) that F(n,rn) = - E
A0AmGa>m
- E E A ° M Ea?°< a,0?ma?0
[ \ *
+ £a?a/
V r
+ E«)V)2 /
t,r
-J E £((£°?^Y + fe^vY+Ew)V)2} a,0?ma^0 — / _, I T
—
(\
i
/
V r
/
i,r
J
^a-^m J (ja,m
(4.7)
- \ E ( (E«>?Y + fe-rV) +£w) V)'l ajtp ( \ i
/
A
\ r
I
>,r
'
a
J
= E (J - ^ ~) «.- + \ E f 2E ?«f >? - £(«?) V ) 3 } ajim ^
G
a^0
[
a
i,r
t,r
J
+ 5 E ( 2 E<•?«?«? «e - E(a?)2(«?)2} - 1 E « ) 2 « ) 2 E ( j " A*A™) Go.-™ " J"(™ -") ~ \ E ( a ? ) 2 « ) 2 Inserting (4.7) into (2.9) and noting that n(m — n) > (m — 1), we obtain E
{2/i(ei,e r ) 2 - / i ( e i , e i ) / i ( e r , e r ) }
<
E (J - A*A™) GQ,m - i(m - 1) + E U. ~ \) («?)2(«?)2,
where G a > m is defined by (4.6). Moreover, by (4.5), we have
E
G
°,m = £ G ° , » - 2 +2E( a ™) 2 «) 2
a^m
= »£(a™)2 + (m-n)E(
i
>E(°?')2 + E(a?,)2 = L
i,r
(4.9)
280 Since (a£) is a special orthogonal matrix, then
from which it follows that
5>?afa?a£ < i« a/s .
(4.10)
i,r
In the following, fixed a point x € M, all calculation will be carried out at that point x. According to (4.2), (4.3) and (4.4), we separate the proof into four cases: (1) Ai > j and | < Am < 1; (2) \i > j and 1 < Am < 2; (3) Ai < i and i < Am < 1; (4) Ai < \ and 1 < Am < 2. Case (1): Ai > j and j < Am < 1. Then there is a number t € [2,3] such that | < A
m
< ^ .
(4.11)
It follows that A a A m > AmAi > -
for
a/m.
(4.12)
By (4.2), (4.9), (4.10), (4.11) and (4.12), we have from (4. 1
G m A 2 a 2 (4-8) < " I 5 " J) 4 E ».™ " V ~ !) + ( ™ " l ) E W) ( ?)
ft.
<
\\
-(f-J)>->*7{(^)"-J4 I
< 0
for
2< t <3
and
m > 3.
Case (2): Ai > j and 1 < Am < 2. Then there is a number t 6 [8,15] such that | < Am < ± ± ± .
(4.13)
It follows that for a ^ m
A»Aro > AiAm > - 1 ,
\a<J-
(4.14)
lo Am r By using the similar way as in the case (1), from (4.13) and (4.14) we have
<«><-(H)>->*H*-iWs:(*-J) a^m
< 0
for
8 < t < 15
and
m > 3.
Case (3): Ai < j and j < Am < 1. By noting that A a A m > j and Ai < | , we have directly from(4.8)
JE(«-i)
(4.8)<-i(m-l) + 4 <_I(m_i)
+
:
<-^(ra-l)<0.
281 Case (4): Ai < j and 1 < \
m
< 2. Then there is a number t 6 [16,31] such that
4
(4,5)
It follows that for a ^ m
A
A
^i
^i>c4
AQAro > AmA] > ~ .
< 416 > (4.17)
By using the similar way as in the case (2), we have from (4.15), (4.16) and (4.17)
(4 8)
- < - ( A -1) - \(m -i)+g 5 (A« -1) ^^ f
i\
i,
.,. i
322 ~ 4 / ~ 4
J _2 32 2 +
322 32
m
~ i
2 - +
a/1,?
l
l
1 I (t + X 4 l i ^ 6 ~ / ~4
44 I,V. 16 16 Jj ~ 16 16 "
< 0
for
+
m - 2 I /16\ 1 4 1 U J ~4
16\ I V' /
4
16 < t < 31
and
5 4
m > 3.
In sum, it has been proved that the inequality (2.6) holds at that point x. Since the point x G M is arbitrary, then Theorem 2 follows from Proposition 2.1 immediately.
□
Remark. From the proof of Theorem 2 we see that the same conclusion as in Theorem 2 holds for the complete hypersurface M immersed in a unit Euclidean sphere 5 m + ' , whose sectional curvature KJJ- satisfies 1 + S < Kw
< 2. §5.
T H E INSTABILITY OF A DOMAIN
We now consider the instability of a domain of a minimal submanifold M in M. T h e o r e m 3. Let M be a compact simply-connected S-pinched Riemannian manifold with S > 0.77. Then there exists a constant C(5) depending only on 5 such that if Xi(D)
for a domain D c M with the compact closure D and the smooth boundary 3D, where M is an n-dimensional minimal submanifold in M and Xi(D) denotes the first eigenvalue of the Laplacian of D with the Dirichlet boundary condition, then D is unstable. Proof. Let M be an n—dimensional minimal submanifold in M and let V C M be a domain with the compact closure V and the smooth boundary dV. Let / be the eigenfunction oiV corresponding to AifD). By using the same notation as in the proof of Theorem 1, we have (cf. [01]) HfV,fV)=
f JV
q(fW)dv,
282 where
q(fW) = | V/ |2|| V |p +f £ || iy^W)™^M)
f +f JT || (Veil7T)w f
«=1
i=l
+ 2 / £ > / ) < V,V£W > - 2 / X > / ) < V,Veif/T >
VlW,(V:iUT)N >-fJ2<
-2fJT< 1=1 n
B(eiiej),V >2
»,J=1
-/25]. •=i
In terms of the same method as in the proof of Theorem 1, at each point x £ V C M we choose an orthonormal basis {WA} such that {Wj} are tangent to M and {Wr} are normal to M. A straightforward computation gives m+l
tr , =(m - n) | V / | 2 + / 2 ^
£
|| (V» W^)™™<"> f
+ / 2 E E I (V..Dl)w II2 +2/ 5 ] £ ( « / ) < Vx.ViWx > 2
A=l m+l
t
A=\ m+l
i
A=l i
(5.1)
m+l
- / E D « / ) < ^ , VeiC/J > -2/ 2 53 ^ < Vi',110,, (Vi'.Ujr > 2
A~l i B
2
2
m+l
- / E E < (ei.ei),VA > -/ 53 53 < flfe,^)1^ > . .4=1 i,j
A=l i
By using (2.3) and noting that VA = (WA)™, m+l
_
/ 2 E E II K W ™ A=l
we have
n K ( M )
H2< jn(m - n)/ 2 (fc 2 (i)) 2 ,
i
m+l
m+l
/2 E E i (^.t/J) w n2= f I B II2, A-l
i
.4=1 i
m+l
m+l
/ E £(«/> < VA,veiui >= o, A=l
/ E £(«/) < K4, v» wA >= o,
i
/2 E E < *(e,,ei),K4 >2= /2 I B II2. A=l i,j
m+l
/ 2 E E < K ^ . (v'e'^D" >= f E II ( V ^ ) N ii2A=l
-f
* m+l
i,j
E E < ^(ei.KtJVA.ei >< -n(m-n)<S/ 2 .
A=l
>
Substituting these into (S.l) gives that tr q < (m - n) | V / | 2 +in(m - n)/2(fc2(5))2 + / 2 || B f
- 2/ 2 £ ii «*>)" ii2 -/ 2 ii B ii2 -»("»- « w 2 ij=l
< (m - n){| V / | 2 +n/ 2 (*2 W/2) 2 - n<5/2}
(5 2)
-
283 at x. Thus, we have m+l
-
< (m - n) / {| V / | 2 +nf(k2(5)/2)2
£ H/VAJVA) A=i
-
n6f}dv
Jv
= {m-n)
(5.3)
[ / 2 {Ai (V) Jv
nC(S)}dv,
where C(5) = S-(k2(6)/2)\
(5.4)
Clearly, we see from (2.4) that C(6) > 0 when 6 > 0.77. Hence, the right hand side of (5.3) is negative if Ai (D) < nC{6). The proof of Theorem 3 is proved.
□
Some values for C(5) in Theorem 3 are shown in the following table.
6
1 0.95
0.9
0.85
0.8
0.77
C{6)
1 0.93
0.84
0.70
0.43
0.12
REFERENCES
[Am] Aminov,J., On the instability of a minimal surface in an n-dimensional Riemannian space of positive curvature, Math. USSR Sb., 29(1976), 359-375. [Ho] Howard,R., The nonexistence of stable submanifolds, varifolds, and harmonic maps in sufficiently pinched simply connected Riemannian manifolds, Michigan Math. J., 32(1985), 321-334. [HW] Howard,R., Wei,S.W., On the existence and non-existence of stable submanifolds and currents in positively curved manifolds and topology of submanifolds in Euclidean spaces, A preprint. [Gl] Grove,K,, Karcher,H., Ruh,E.A., Jacobi fields and Finsler metrics on compact Lie groups with an application to differentiable pinching problems, Math. Ann., 211(1974), 7-21. [G2]
, Group actions and curvature, Inven. Math., 23(1974), 31-48.
[LS] Lawson,H.B. Jr., Sinmons,J., On stable currents and their application to global problems in real and complex geometry, Ann. of Math., (2) 98 (1973), 427-450. [01] Okayasu,T., On the instability of minimal submanifolds in Riemannian manifolds of positive curva ture, Math. Z., 201(1989), 33-44. [02]
, Pinching and nonexistence of stable harmonic maps, Tohoku Math. J., 40(1988), 213-220.
DEPARTMENT OF MATHEMATICS, HANGZHOU
E-mail
310028,
address:
WEST-BROOK
CAMPUS,
ZHEJIANG
P.R.CHINA
ybshanCdial.zju.edu.cn R e c e i v e d J a n u a r y 3 , 2000
UNIVERSITY,
284
Geometry and Topology of Submanifolds X eds. W. H. Chen et al. (pp. 284-293) © 2000 World Scientific Publishing Co.
Intrinsic and Extrinsic Geometry of Ovaloids and Rigidity Udo Simon*
Luc Vranckent
Changping Wang*
Martin Wiehe*
Abstract Our main result is that an ovaloid with nowhere dense umbilics and prescribed Weingarten operator and spherical volume form is rigid in Euclidean 3-space. In case of an ovaloid of revolution we can drop the assumption on the volume form. Keywords: ovaloids, global rigidity, Weingarten operator, surfaces of revolution. 2000 MS-Classification: 53C42, 53C24
Introduction Bonnet's uniqueness theorem for Euclidean hypersurfaces states that the first fundamental form I and the second fundamental form II together completely determine the hypersurface, that means that all geometric invariants can be derived from the two forms; thus I and II together completely describe the geometry of the hypersurface. Both forms are related by the Weingarten operator S:
n(i>,ui) = i(Su, v). It is a trival consequence of this relation that the pair {I, S} forms another fundamental system of geometric invariants. In a standard terminology one calls all invariants intrinsic which belong to the Riemannian geometry of the first fundamental form metric; all invariants which depend on the immersion of the hypersurface into the ambient space belong to the extrinsic geometry. In particular, the Weingarten operator S and its invariants describe the extrinsic curvature properties of the hypersurface. The integrability conditions of the structure equations give relations between the two fundamental forms and thus they admit the study of relations between the intrinsic and the extrinsic geometry. R e v i e w of s o m e i n t r i n s i c r e s u l t s . One of the most famous results in this direction is the theorema egregium of Gaufi. If we denote by k\,... ,kn the principal curvatures and "The authors are partially supported by a German-Chinese cooperation project of DFG and NSFC. * Partially supported by a research fellowship of the Alexander von Humbold Stiftung (Germany) The authors thank K. Voss for critical remarks. This paper is in final form and no version of it will be submitted for publication elsewhere.
285 by H\,... ,H„ their (normed) elementary symmetric functions (in particular: H := Hi denotes the mean curvature, K := Hn the GauB-Kronecker curvature) of an n-dimensional hypersurface in Euclidean (n + l)-space, n > 2, then this result states (see e.g. [K-N-II], p.43 and [HEIL]): Theorema egregium of GAUSS - extended version. (i) The curvature functions H2r, for 2 < 2r < n, are intrinsic invariants; (ii) the curvature functions {Hir+i)2, for 3 < 2r + 1 < n, are intrinsic invariants. In a rough terminology, the mean curvature is the only genuine extrinsic curvature invariant within the set of curvature functions Hi,... , Hn; this explains the particular interest in the study of the mean curvature. While Gaufi' result is true for any hypersurface, there are other results which need additional assumptions; recall the following two well known results, one a local theorem, the other one global; they describe conditions under which the first fundamental form determines the second fundamental form. We introduce the following notation: M denotes a connected, oriented C°°-manifold of di mension M = n > 2 and x, x*: M —> En+1 hypersurface embeddings into Euclidean space such that (p := x* ox-1 is a diffeomorphism between x(M) and x*(M). In an obvious nota tion 1,11,5 and I * , I * , 5 * denote the fundamental invariants of a; and x#, resp., mentioned in the beginning. Theorem of Beez (1876) - Killing (1885). Let dimM = n > 3 and x,x* be isometric hypersurfaces, i.e. 1 = 1* on M. 7/rank(5) > 3, then II* = ±H; thus x,x* are congruent in En+l. Theorem of Cohn-Vossen (1927). Let x,x*: M —> E3 be ovaloids (i. e. compact without boundary and with positive Gaufi curvature). Ifx,x* are isometric then they are congruent. Both results are a consequence of the integrability conditions; one proves that the metric form I uniquely determines the second fundamental form (modulo sign in the Beez-Killing theorem); then one applies Bonnet's rigidity result (see e.g. [K-N-II], p. 43, and [COHN-V]). Remarks. We would like to recall some modest extensions of the foregoing uniqueness results. (i) We weaken the assumption of the isometry and assume instead that only the LeviCivita connections coincide: V = V*. According to the Ricci-Lemma both metrics are parallel VI = 0 = VI*. If we additionally assume that x,x* are locally strongly convex, i.e. the Gaufi curva tures are positive, then the Riemannian spaces (M, I) and (M, I*) are irreducible; thus parallelity implies I* = cl for some positive constant c £ R. This gives: Extensions. Letx,x# be locally strongly convex and assume V = V*:
286 (a) Ifn>3
and rank(S) > 3, then x, x* are homothetic;
(b) ifn — 2 and x, x* are ovaloids, then x, x* are homothetic. (ii) Another extension of Cohn-Vossen's result was proved by Hsu [HSU]: If two ovaloids x, x*: M -> E3 satisfy K ■ I = K* ■ I* (K, K* Gaufi curvatures) at any p e M, then x,x* are homothetic (and K* = c • K with 0 < c € R). - Hsu's assumption is a particular conformal relation I* = q ■ I with q = K ■ K*~x, where the proof finally gives q = c - 1 . It is well known that the assertion is not any more true for a general conformal relation I* = q ■ I with 0 < ? £ C°°(M), but without further restrictions on 1Hsu's result generalizes the foregoing extension (i): V = V* implies equality of the Ricci tensors; in dimension n = 2 that gives K ■ I = Ric = Ric # = K* ■ I*. (iii) One might try other extensions of the foregoing results in (i). E.g., recall that a connection is determined by a natural parametrization of its autoparallel curves within the projective class. This raises, in particular, the question whether one can weaken the above assumption V = V* and assume instead that there exists a diffeomorphism of the two ovaloids preserving the autoparallel curves. Are both still homothetic? The answer is in the negative; this follows from the examples recently given independently in [MATV], [TABA, Theorem 6], and [VOSS]. Review of some extrinsic results. As far as we know it was E. Cartan [CARTAN] in 1943 who startet a systematic investigation of the role of the second fundamental form and proved some local existence and uniqueness theorems in terms of the extrinsic geometry. So far any attempt failed of proving a global "extrinsic" analogue to Cohn-Vossen's result assuming that the second fundamental forms of two ovaloids x, x* coincide: II = I*. But there exist several rigidity results under additional assumptions: 1.1. Theorem. Letx,x&: M —> E3 be ovaloids. (i) ([GROVE],1957). // H = I * and the Gaufi curvatures coincide, K = K*, then x,x* are congruent. (ii) (1970-73; see [HUCK et al] pp.59-61). If H = H# and F{H,K) = F(H*,K#) for some C1-function F : l x R - > 8 such that 6\F • d2F := | £ • | £ > 0 then x,x* are congruent. Theorem l.l(i) was generalized to dimension n > 2 by Gardner1 [GARDNER-I] in 1969, using a new integral formula. Erard (1968) extended Cartan's investigations and additionally studied infinitesimal deformations preserving the second fundamental form. See [ERARD] and sections 2.3.a.B; 2.3.b.B; 3.4.2; 3.8.2 in [HUCK et al] for related results. So far, all similar global results for surfaces in Euclidean space E3 have two extrinsic assumptions for proving an extrinsic uniqueness theorem. The situation is different for closed surfaces in a non-flat space form M. In [LIU-S-W] we recently proved a uniqueness theorem with only one extrinsic assumption. 'R. Gardner (f 1998) was a Ph.D. student of S.S. Chern
287 1.2. Theorem. Let x,x#: M2 —> M 3 be closed surfaces in a non-flat space form with positive definite Weingarten operators 5,5*. If the third fundamental forms coincide, I(5«, Sv) =: HI(u, v) = IH#(u, v) = I#(5#«, S*v) then x, x* are congruent. The proofs of the foregoing intrinsic and the extrinsic global results use the Codazzi equations as essential tool. All different proofs of the intrinsic Cohn-Vossen theorem consider the Codazzi equations for the difference tensor between the second fundamental forms D := H — II* of two ovaloids x,x# in Euclidean 3-space; one derives a linear equation for D. In case of the extrinsic rigidity results in Theorem 1.1 the Codazzi equations lead to a nonlinear elliptic equation for the difference tensor E := I - I* (see [HUCK et al], I.e.). It is the aim of this paper to present a new method of proof and the following new extrinsic result. Theorem A. Let x,x#: M —> E3 be ovaloids in Euclidean 3-space with nowhere dense umbilics and with the property that, at any p 6 M, the Weingarten operators S, 5 * and the spherical volume forms w(lH), w(m*) coincide:
s = s*,
o;(ni) = w(in#).
Then x,x* are congruent. 1.3. Corollary. Let x,x*\ M -> E3 be ovaloids such that S = S* and w(Iff) = OJ(IH # ). // x is analytic then x,x# are congruent up to a reparametrization. The basic idea for the proof of Theorem A (sections 2 and 3) is to consider the unique, I-selfadjoint, positive definite operator L defined by I*(fl, tu) =: l(Lv, Lw) and to study its algebraic and analytic properties. A second tool is to use the Codazzi equations for 5 = 5 * in terms of the two Levi-Civita connections V = V(I) and V* = V(I # ) to get relations for the symmetric (1.2) difference tensor (V — V*) between the connections which finally lead to PDEs for the operator L. If one follows the proof it seems that one might drop the assumption on the volume forms. We would like to state the following Conjecture. Let x, x* : M —» E3 be ovaloids with nowhere dense umbilics and with S = S* at corresponding points. Then x, x* are congruent. In section 4, we give another partial answer to this conjecture.
2
Codazzi operators in terms of different metrics
As before, let M be a connected, oriented C^-manifold of dimension n > 2. We recall the well known definition of a Codazzi operator ip with respect to an affine connection V on M
288 which we assume to be torsion free; if the pair {V,^} satisfies Codazzi equations
(v„v)™ = (v w v)«; we call {V, V>} a Codazzi pair. 2.1. Lemma. Let dinxM = n = 2 and consider two torsion free connections V, V* and an operator i/> on M; let {W,ip}, {V#,ip} be Codazzi pairs. Assume that ip has two real eigenvalue functions v\,v2 on M. If, at p € M, v\{p) / v2(p) then there exists a local parametrization [ul,u2) of a chart U around p s.t. for the associated Gaufi basis {81,82}: m)
= "A-
(2.1.1)
In such coordinates we have T*r = T\2 on U for r = 1,2. Proof. In local terminology, the Codazzi equations for {V, ^ } read:
Substract the analogous equation for {V*,^} : r
W,-it ) = #(r;.-r*r).
Equation (2.1.1) implies (1 = j / i = 2):
^i(^1-r#r) = ^(q 2 -rf/). This gives the assertion.
□
2.2. Calculation. Let V = V(g) be the Levi-Civita connection of a semi-Riemannian metric g on M, dimM = n = 2. Assume that g has the local representation
on a chart U on M. Then the Christoffel symbols rf • satisfy the relations 2T\X = diQngu);
2T\2 = 5 2 l n P u ; 2 1 ^ = - ( s n ) " 1 ^ ;
2r?x = -(g22)-1d29n;
2T2W = 8j \ng22; 1Y\2 = d2 ln(g22).
2.3. Calculation. Let dimM = 2 and consider two Riemannian metrics g, # on M. Then there exists a unique g-self-adjoint, positive definite operator L such that g*{u,v) = g{Lu,Lv)
(2.3.1)
for tangent vectors u,v. Denote the eigenvalue functions of L by Ai,A2 > 0 and consider a local chart U with Ai ^ A2 such that the Gaufi basis {d\,82} consists of eigendirections: L{8i) = \8i.
289 (i) Then locally the metrics ,* are represented by matrices
g:(9" \
°) 0
and g* . { %*»
22 /
\
° V
(2.3.2)
X22g22 )
0
(ii) Assume Ai / A2 on U. Then Ai, A2 are differentiable and the Christoffel symbols of the metrics satisfy
rfi 1 - r} x = d, in AX ;
rfi2 - r? 1= -\ {^A} r#i 1 22
_ r1 L
22 I
22
2
A2
r* 1 - r } 2 = 3 2 in Ai;
r # 2 - r ^ = d2 in A2;
(922T1 d2gn - |a 2 A t • {g22ylgil] * I (an)" 1 d\9ii - ^ | • 9iA2 • (911)"1322; r # 2 - r 2 2 = dt in A 2 .
2.4. Lemma. Let g,g* be Riemannian metrics on M satisfying (2.3.1); assume that there is an operator ?/> on M which is selfadjoint with respect to g and g# at the same time, denote the eigenvalues of i\) by vlt v2. Then, with the notation from 2.3: L2i> = i>L2;
(i)
(ii) // Ai / A2 and v\ ^ v2 then L and ip have the same eigenspaces at any point of M. Proof, (i) g(L2ipu,v) = g* {ipu,v) = 3* (u, ij>v) = g(L2u,ipv) = g(ipL2u,v) for all u,v; this implies (i); (ii) is an immediate consequence. □ 2.5. Corollary. Consider g, g* and ip as in 2.4 and assume that the two eigenvalues Ai, A2 of L differ at a point p G M. Then: (i) there exists a chart U around p with At yt A2 on U and with local coordinates (u 1 ,^ 2 ) s.t. the operators and the metrics have the following local matrix representations: Sii
0
0
9*
A23ii
0
0
322
\
2
A 22 j '
A! 0 0 A2
V°
V2
1
(ii) Ai, A2 are differentiable on U; (iii) the Christoffel symbols of g and g* satisfy the relations in (2.2) and (2.3).
290 2.6. P r o p o s i t i o n . Consider g,g& and ip as in (2.4) and assume additionally that tp and the Levi-Civita connections V := V(g) and V * := V(*) form two Codazzi pairs { V, ip} and {V*, tp}. If Ai ^ A2 and V\ ^ 1/2 at p, consider the local parametrization (u 1 , « 2 ) of a chart U as in (2.5) with Ax 7^ A2 and v\ ^ f2 on U. Then: (i) J4S before, Ai,A2 are differentiable on U and d2Ai = 0 = 5iA 2 . (ii) The relations for the Christoffel symbols in (2.3) 1
11
~
#2 1r
22 r1 # 2 11 1
22 r
■p# 1
12
x
11 2 22 r2 ~111 Lr
* 22 r
L
r
12
simplify:
= SilnAx; = 92lnA2; =
- ^ 2 ) - 2 ( A 2 - A22) ■ (g22r1di9li;
=
-i(Ai)-2(A2-A2).(Sll)-15l522;
= 0.
Proof, (i) follows from Lemma 2.1 and 2.3.(ii). Insert now (i) into the other relation in 2.3.(ii); this gives (ii). □
3
Proof of Theorem A
We consider the two ovaloids x,x& : M —» E3 with their metrics g = 1,5* := I*, the associated Riemannian volume forms u)(g) = to(g#) and the Weingarten operators 5 = 5 * . We proceed with the following steps of the proof. Step 1. Denote by TV the (closed) set of umbilics of x on M. N is nonempty for an ovaloid. S = S* implies N = TV*. M \ TV is dense and open in M. We set S = S* = tfi and adopt the notation from 2.1 - 2.6. Around any p £ M\N there exists a local chart U s.t. we have the matrix representations from 2.5; moreover, on U: d2Xi = 0 = 9iA 2 . From w(II) = coQS*) and K = K* we get u(g) - u(g#); together with (2.3.1) this gives d e t L = 1. Thus Ai ■ A2 = 1. Differentiation gives d^Ai = 0 = 32A2 on U, thus Ai = const, A2 = const, on U and on any connected component of M \ TV. As the eigenvalue functions A!,A2 are continuous on M and as TV is nowhere dense A^ A2 must be positive constants on M. Step 2. Assume that Ai 5^ A2 on M. Then there exist two orthogonal eigenvectors eX)ei at any point p € M, and we get a pair of differentiable, nowhere vanishing tangent vector fields on M. But this contradicts the fact that the genus of M is zero. P r o o f of Corollary 1.3. If a: is a sphere then x * must be a sphere of the same curvature and are therefore congruent up to a reparametrization. Otherwise the umbilics are isolated and the corollary follows from Theorem A. □
291
4
Ovaloids of revolution
In this section we want to prove the previously mentioned conjecture in the special case that x is an ovaloid of revolution. In particular we want to prove the following: 4.1. Theorem. Let x • M —> E3 be an ovaloid of revolution with nowhere dense umbilics and let x* : M —> E3 be another ovaloid with S = 5 * at corresponding points. Then x,x* are congruent. Proof. Since x : M —> E3 is an ovaloid of revolution we can write x(u1,u2) = (r(« 1 )cosa 2 ,r(tt 1 )sinit 2 , ^(u1)), 1
(4.1-1) 2
where 0 < u < A parametrizes the meridians as arc length parameter and 0 < u < 2w parametrizes the parallels of latitude with radius r(u 1 ); thus we have r(ul) > 0 and r'(u1)2 + s'(u1)2 = 1 for 0 < ul < A , and r(0) = r(A) = 0.
(4.1.2)
The two "poles" PM and ps (u1 = 0 and ul = A) are umbilics. It follows by a straightforward computation that 3 = 1 has the representation on M\{pjv,Ps} : ffn = 1, Si2 = 0, 322 =
r2,
and r and s satisfy Sd^ = (r's" - r"s')di
and
Si\ = ^c\.
Consider the two metrics g and g* to be related as in (2.3.1). The curvature lines for x and x* coincide on M\{PN,PS}, on this set we get the following representation for * = I*: „ # _ \2
„# _ n
. # _ \2 2.
besides at the northpole pn (u1 = 0) and the southpole p$ (u1 = A), we can write Ldi = \\d\, Ld2 = \id2, where the eigenvalue functions Ai and A2 are differentiable functions on M\{p^,ps} continuous on M. The differentiability of Ai, A2 implies that we can prove the PDE a2A1 = 91A2 = 0
and (4.1.3)
on A/yip^ps}, similar to Proposition 2.6. Now we look at what happens at p^ (and ps). Suppose that L at p^ is not a multiple of the identity. Then there exists a unique differentiable orthonormal frame in a neighborhood of PN such that LiJ\\ ~ A\yii:
LiJ\2 — A 2 ^ 2 -
Therefore, since L and S commute and the interior of the set of umbilic points is empty, it follows - if necessary after exchanging X\ and X2 - that in a neighborhood of pjv, except at PN-
X2 = (— sinu 2 ,cosi( 2 ,0).
292 Since the right hand side can not be differentiably extended to p^, a contradiction follows. Thus Ai = A2 at PMAgain we use the continuity of Ai and A2 on M, thus lirriui^o Ai and lim„i_>A A2 exist. Recall the equations (4.1.3) on M\{pN,ps}Considering lim u i_ +A A 2 , which exists and obviously must be independent of u2, it follows that A2 is constant on the whole of M. Now we introduce a function r such that r = p- — 1. The condition K = K* then implies that r satisfies the following differential equation almost everywhere: 0 = r"r +
-r'^-
implying that there exist a constant c such that (r') 2 r = c. Since r(0) = r(A) = 0, it follows that c has to vanish. The fact that x is an ovaloid then implies that r cannot be constant on an open set and therefore that r = 0. Consequently Ai = 1. Since at PN,M = A2, it follows that L = id and therefore x and x * are congruent. D R e m a r k . The condition that r' is somewhere zero is crucial in the proof of the previous theorem. Indeed, if we consider for example the paraboloid, parametrized by x(u,v)
= (ucosv,usinv,
\v?),
we find, using the same technique as in the previous theorem, that all hyperbolioids of the one parameter family parametrized by . . , u u J{\ + c) + cu2 1 Xc{u,v) = ( -——cost;, ^ - ^ s i n i ) , -* -== ), Vl + c VI + c c\/l + c c where c is a nonzero real number satisfying c > —1, have the same shape operator given by qJL
=
^
*
a
y/TTrfdv 1
J f l T - '
(1 + U2)2
The above example also shows that the conjecture can not remain true for complete surfaces with positive Gaussian curvature.
References [CARTAN]
Cartan, E.: Les surfaces qui admettent une seconde forme fondamentale donnee. Bull. Sci. math., II. S.67, 8-32 (1943). Zbl. 27, 425.
[COHN-V]
Cohn-Vossen, S.: Zwei Sdtze iiber die Starrheit der Eiflachen. Nachr. der Ges. der Wissensch. zu Gottingen, Math.-Phys. Kl. Jahrg. 1927, 125-134.
293 [ERARD]
Erard, P.J.: Uber die zweite Fundamentalform von Flachen im Raum. Disserta tion, ETH Zurich 1968.
[GARDNER-I]
Gardner, R.B.: An integral formula for immersions in Euclidean space. J. Diff. Geometry 3, 245-252 (1969). Zbl. 188, 263.
[GARDNER-II] Gardner, R.B.: The geometry of subscalar pairs of metrics. Proc. Carolina Conf. holomorphic mappings minimal surfaces, Chapel Hill 1970, 29-42 (1970). Zbl. 218, 341. [GROVE]
Grove, V.G.: On closed convex surfaces. Proc. Amer. Math. Soc. 8, 777-786 (1957). Zbl. 83, 372.
[HEIL]
Heil, E.: The "Theorema egregium" for hypersurfaces. Lecture Geometrie-Tagung Oberwolfach October 1992, Tagungsbericht 46 (1992), p.6.
[HSU]
Hsu, C.S.: Generalization of Cohn-Vossen's theorem. Proc. Amer. math. Soc. 11, 845-846 (1960). Zbl. 192, 271.
[HUCK et al]
Huck, H.; Roitzsch, R.; Simon, U.; Vortisch, W.; Walden, R.; Wegner, B.; Wendland, W.: Beweismethoden der Differentialgeometrie im Groflen. Lecture Notes in Mathematics 335, Springer-Verlag Berlin, Heidelberg, New York (1973).
[K-N-II]
Kobayashi, S.; Nomizu, K.: Foundations of Differential Geometry II. New York: Interscience Publ. 1969.
[LIU-S-W]
Liu, H.; Simon, U.; Wang, C.P.: Codazzi tensors and the topology of surfaces. Annals Global Analysis Geometry 16 (1998), 1-14.
[MATV]
Matveev, V.S.; Topalov, P.J.: Quantum integrability of Beltrami-Laplace operator as geodesic equivalence. Preprint.
[TABA]
Tabachnikov, S.: Protectively equivalent metrics, exact transverse line fields and geodesic flow on the ellipsoid. Comment. Math. Helv. 74 (1999), 306-321.
[VOSS]
Voss, K.: Geodesic Mappings of the Ellipsoid. In: Geometry and Topology of Submanifolds, X.
U. Simon, L. Vrancken, and M. Wiehe, FB Mathematik TU-Berlin, Germany. C.P. Wang, Department of Mathematics, Peking University, P.R. China. e-mail: [email protected] [email protected] [email protected] [email protected] Received February 16, 2000
294
Geometry and Topology of Submanifolds X eds. W. H. Chen et al. (pp. 294-302) © 2000 World Scientific Publishing Co.
Geodesic Mappings of the Ellipsoid Konrad Voss
Abstract There exists a 1-parameter family of nonhomothetic ovaloids in Euclidean 3-space such that the metrics are geodesically equivalent to the metric of the triaxial ellipsoid. Keywords: geodesic mapping, metrics on ovaloids, ellipsoid. 1991 Mathematics Subject Classification: 53 C 22, 53 A 05
1. Introduction We denote by x = (x\,X2,xz)
points in Euclidean space R3 and consider the ellipsoid
£ = L[ |l £a i + ^b + ^c = l )l ,
a>b>c>0.
x i-> y = x/\x\ defines a diffeomorphism between E and the standard sphere S2 . The problem of the behaviour of the geodesies on E has been investigated by many authors, starting with C.G.J. Jacobi in 1839 [5]. Of special interest is the question for the existence of closed geodesies. There are four umbilical points on E:
they are singularities in the net of curvature lines; for a picture see [4, p. 167]. The geodesies through one of the umbilics, say p, meet again in the opposite point —p, but they are not closed, with exception of the ellipse x2 = 0. We ask the following question: Are there ovaloids in R 3 , not homothetic with E, which can be mapped onto E, such that the geodesies on the two surfaces correspond? The problem, which pairs of surfaces allow such a geodesic mapping, has been solved by U. Dini in 1869 in the generic case [3]. It turns out, that the construction of Dini can be applied to the umbilic free part of E, but fails to work in the umbilics. In this paper we will prove: T h e o r e m 1 If g is the metric of the ellipsoid E, considered as a Riemannian metric on S2, then there is a 1 -parameter family of metrics g£ on S2, for all £ > 0, such that go = g and all ge have the same geodesies. 1
This paper is in final form and no version of it will be submitted for publication elsewhere.
295 Theorem 1 implies the existence of a family of ovaloids in R3 with geodesically equivalent metrics gc, at least for an interval 0 < e < £o, by applying Weyl's embedding theorem. The possibility of such a construction was first pointed out by Marie-Helene Bossel in her master's thesis (Diplomarbeit) at ETH Zurich in 1977, [1].
2. Geodesic diffeomorphisms A diffeomorphism / between Riemannian manifolds (Mn,g) and (M ,~g) is called geodesic (or projective) if / sends the geodesies of M to geodesies of M. We can identify M and M by virtue of / and consider two metrics g,~g on M . If g and ~g have the same geodesies, we say that the metrics are geodesically equivalent. Denote by V, V the Levi-Civita connections of g, g and by
VXY-VXY
= T(X,Y) = TXY
the difference tensor. Then the following condition is known, see H. Weyl [7], [8, p. 196]: The metrics g,~g are geodesically equivalent if and only if (2)
T(X,Y)=A{X)Y
+ A(Y)X
with a linear form A, given by A(X) = —^ trace Tx ■ In components with respect to local coordinates, (2) can be written as
(3)
7* = r* j -r*=tfA,- + ^A j ,
^ = - 1 ^ .
(3) represents a number of homogeneous linear relations between the components of T. In the case n = 2 there are four such relations: (4) (5)
Tl = 2T}2,
Tlx = 2T\
7* = 7 ^ = 0.
3. Liouville nets and the theorem of Dini Consider a curve net N on (M2,g) and local coordinates u,v such that N is the net of coordinate lines. N is called a Liouville net, if g - with respect to u,v - has the form (6)
g: ds2 = [U{u) - V{v)}[a{u)du2 + /]{v)dv2}
with functions U > V > 0 and a, /3 > 0.
296 The form (6) of g is preserved under transformations (7)
u = u{u),
v = v(v);
especially u = / \A*(U) du , v = / \/P(v) dv yields ds2 = [[/—VJfdS^+diJ2], i.e. a conformal mapping of a coordinate neighbourhood in (M2, g) into the Euclidean u, iJ-plane. Two metrics g,g~ on M2 are conformally equivalent at a point p € M2 if g at p is a multiple of g. If g, ~g have no point of conformity, then there is exactly one curve net on M2 which is orthogonal with respect to both metrics. The theorem of Dini now can be formulated as follows: Theorem 2 The metrics g, ~g without points of conformity are geodesically equivalent if and only if the common orthogonal net of g, g~ is a Liouville net with respect to both metrics and, if g is given by (6), then ~g is given by ds2 =
(8)
-du + -dv
Notice that a transformation (7) changes a and /3 , but does not change the relation between (6) and (8). Indication of the proof: We can assume ds2 = Edu2 + Gdv2, ds2 = Edu2 + Gdv2 and introduce E/E = if, G/G = ij>. Then (4) is equivalent to
[iog(^-2)L = o , V-v,"2 = [U(u)}3
,
[iog(^-2)]u = o 2
lpip-
= [V(v)}3;
U,V>0
Since U ^ V, we can assume U > V. Expressing E,G by E,G,U,V, the equations (5) are equivalent to
[E/(U-V)]v = 0,
[G/(U-V)l
it turns out that
= 0.
D
Let us assume that g is a Liouville metric, given by (6). The metric g in (8) is not uniquely determined. For, choosing arbitrary constants e > 0 and t £ l , we can write [E(U + t)-e(V
+ t)] -du* + -dv' e
e
According to (8) we get a 2-parameter family of geodesically equivalent metrics :
9t,e
ds2 = \
1
1
V +t
U+t
u+t
du2 + -
-dv2
where gtE and gtti only differ by the the factor e 3 , that means, we have a 1-parameter family up to homothety. We wish to insert g into this family. This can be done by taking t = 1/e. Since 1/(1 + eU) = 1 - (EU)/(1+EU) , 1/(1 +eV) = 1 - (eV)/(l + eV), we bring Dini's theorem into the following form:
297 Theorem 3 Every Liouville metric (6) can be embedded into a 1-parameter family gc of geodesically equivalent metrics (9)
ge : ds]
U 1+eU
V 1 + eV
* dv> 1+eV
1+EU
for arbitrary e > 0 .
4. Parametrization of the ellipsoid E We have to construct local coordinates, adapted to the problem, for a neighbourhood of an arbitrary point of E. At first we recall known formulas in elliptic coordinates, where the coordinate lines are the lines of curvature (see [6], [2, §123, §504]). Then we show how all points of E can be included, especially the umbilics (1). Case (a):
£12:2:13 / 0. The rational function
(10)
Q{t, x) = -lL
+
-A-
+
^L + l
(x fixed)
t—a t—b t—c has three simple poles with residues x\,x\,x\. If Q{0,x) = 0, i.e. x e E, then Q has zeros t = 0 , t = u € ]6, a[, t = v £ ]c, b[. Thus Q is equal to the quotient
(ID
Q(,x)-
*->(«-«>
{t - a){t - b){t - c) '
Comparing the residues in (10) and (11), we infer (
>
m2_ l
"
a{a-u)(a-v) ^ _ b(b-u)(b-v) 2 (a- b)(a - c) 'X2~ (b- a)(b - c) ' * 3
c(c - u)(c- v) ( c - o)(c- b) '
With the abbreviations Cl
^ (a-b){a-c),C2
\J (a-b)(b-c)'
Ci
/
V(a -c)(6 -0)
(12) can be written in the form (13)
xi = ± c 1 y / ( o - u ) ( a - w), x2 = ±c2\/{u - b){b- v), x3 = ±c3v/(w -c)(u - c),
where («, u) runs through the rectangle R = {(u,
v)\a>u>b>v>c}.
The functions x(u, v) from (13) are continuous at the boundary of R (but not differentiable); the boundary segments u = a and v = c correspond to the ellipses X\ = 0 and X3 = 0, respectively, the two boundary segments u = b and v = b to the ellipse £2 = 0 i («,w) = (6,6) yields the umbilics (1).
298 In elliptic coordinates (M, V) 6 R the metric of E is ds2 =
(15)
dv2 (a - v)(b— v)(v — c)
-du (a — u)(u — b)(u — c)
ds2 has rank 2 for (u, v) € R : u,v are admissible coordinates in case (a). Case (b):
x2 ^ 0. We can include the points u = a and v = c by the transformation
(16) Substituting (16) into (13), we can write (17) x\ = c\yja — c — v2 u, xi = ±c2y/(a - b — «2)(6 — c — S 2 ), x 3 = C3V0 — c — «2 0, where («, S) runs through the rectangle (18)
R = < (£, 5TI —y/a — b .
Inserting (16) into (15) yields the formula (19) ds2 = [a - c - (S2 + S2)]
rdu2 + ■
(a — b — tP)(a — c — u2)
-dV ' (a —c — v2)^ — c — v2)"
which has rank 2 in R: u,v are admissible coordinates in case (b) [which includes case (a)]. Case (c): xxx3 / 0. The subset xix3 / 0 of E consists of four connected components according to the signs of x\,x%\ each component contains one of the umbilics (1). We restrict ourselves to the part E+ : Xi > 0, x3 > 0 of E with the umbilic u = v = b [see (13) and (1)]. Substituting u = b + £2, v = b — r)2
(20) into (13), we get (21)
Xl
= clx/(a -b-?)(a
- b + n2), x2 = aft,
x3 = c3y/(b -c + ?)(b
-c-V2).
The functions (21) are real analytic in the rectangle R= {(£>»?) I -%/a - b < £ < y/a-b
, -y/b-c
< r\ < y/b - c\ .
R coincides with R from (18), since £2 = a — b — u2, rf = b — c — v2. (15) is transformed into (22)
ds2 = (e + r)2
b+e (a-b-ti*)(b-c
b-r,2 Jrf . drf ?d? ++e a — b + rj2) (b — c - rj2)
299 which has rank 2 in i?\{(0,0)} : S;,n are admissible coordinates for all points of E+ exception of the umbilic p : (£,??) = ( 0 , 0 ) . [Case (a) is again included]. (21) implies
with
x
\((,v) = z|(-e,-o) ■
+
Thus E — for (£,rf) 6 R - is covered twice with a branch point at p. The coordinate net in the cases (a),(b),(c) is the net of curvature lines of E, which has a singularity at p. Case (d): Coordinates near the umbilic p: (£, JJ) = (0,0). We restrict (£, rj) to the disk D(r) = {(£,T)) | f + r? < r2 = mm(a-b,b-c)}
C R
and apply the transformation (23)
Z+
iv
= ( ^ T
= p + i
<> = &■
D(r) is mapped onto D(\r2) as two-sheeted covering with branch point at r = 0. (21) can be written in terms of p,a: (24)
x\=c\
yj{a - b)2 — (a - b)2p - a2 , a;2 = c2o", x3 = c 3 y/(b — c)2 + (b — c)2p - a2 .
The vectors | f , ^ are linearly independent for (p, a) = (0, 0): p, a are admissible coordinates in case (d). R e m a r k s (i) The transformation (23) sends the coordinate lines of the (-plane to parabolas in the r-plane with common focus T = 0. (ii) The coordinate net in the cases (a),(b),(c) is a Liouville net as described in (6) [see (15), (19) and (22)]. Applying a suitable transformation of type (7), one could pass to the isothermic form with a = /3 = 1. We will not do this, since the inquired metric 5 anyhow belongs to another conformal structure.
5. Construction of geodesic mappings of the ellipsoid E The local coordinates, constructed in Section 4 for the parametrization of E, can be regarded as coordinates for the sphere S2 , and we can consider the metric g of E as a metric on S 2 . If Pj are the points on S2 corresponding to the umbilics of E, then g on S2\{pi,p2,P3,P4} = M is given by the formulas (15), (19) or (22). Thus formula (9) in Theorem 3 yields a 1-parameter family of geodesically equivalent metrics ge on M with g0 = g. The question is, whether ge for e > 0 can be extended to a metric on S2 , which automatically would be geodesically equivalent to g.
300 Observe that (22) has the form (25)
ds2 = [F(e) - F(-n2)} [ G ( 0 # 2 + G{-n2)drf]
with the special choice of the functions (26)
F(s)=b + s,
G(s) =
b+s a — b - s)(b — c + $)
Passing from (25), (26) to ge, we get metrics which again have the form (25), where now F, G are certain convergent power series oo
(27)
oo
F{s) = Y,aksk,
G{s)=J2bkSk;
k=0
a0,ai,b0>0.
fc=0
Theorem 1 is a consequence of Theorem 4 Let F,G be arbitrary real analytic functions of the form (27). Then there exists a real analytic Riemannian metric (ds')2 = A [B(TdT2 + rdf2) + 2CdTdr]
(28)
such that (28) is transformed into (25) by the pull back r = |C 2 with £ = £ + in . Proof: With help of the polynomials
m=0
m=0 ^
'
we define the metric (28), inserting the functions OO
00
A T
T
B T
,
^ ) = E ? /*fr )> ^ ) = E # wrf ■d k=l
2
fc=l oo C
i
(^)=E#^(T,T). fc=0
Notice that A(Q, 0) = |oi > 0 and C(0,0) = 60 > 0.
301 (25) can be written in terms of dQ,dQ:
ds2 = $ [*(d(2 + df) + 2Sd(dc] ,
(29) where
* = \ [Fif) - F(-v2)] , * = \ [G(e) - G(-r,2)] , 5 = 1 [G(£2) + G(-„4)] . With f2 = (^yM , —??2 = ( ^ )
j w e write for the sum and the difference oo
G(t2) ± G(-v2) = £ **2"2* [(C + 02k ± (C - 02*]
(c+c)2* ± (c - c)2* = £ (2fc) [i ± (-1)'] c-> ? ■ In the case of the difference, we have terms only for j odd: j = 2m + l , 0 < m < f e — 1, and there is a factor QQ. In the case of the sum, we have terms only for j even: j = 2m with 0 < m < k. Therefore we can write, regarding r = |(" 2 :
*(C, C) = (U(r, T), *(C, C) = d\B(T, T) , S(Q, C) = C(T, T) . Since Tdr2 + rdr2 = ±(QQ)2{dQ2 + dQ*) and drdf = (QdC.dC,, (28) is transformed into (29).
□ Remark At the umbilics all metrics g£ from Theorem 4 are conformally equivalent, see (28) with r = 0.
References [1] Bossel, M.-H.: Geodatische Abbildungen spezieller Flachen. Diplomarbeit, Eidgenossische Technische Hochschule Zurich, 1977 [2] Darboux, G.: Lecons sur la theorie generate des surfaces I-IV, Deuxieme edition, Gauthier-Villars et Cie., Paris, 1914 [3] Dini, U.: Sopra un problema che si presenta nella teoria generale delle rappresentazioni geografiche di una superficie su di un'altra. Annali di Matematica pura ed applicata, Serie II - Tomo III, 269-293 (1869) [4] Hilbert, D. und Cohn-Vossen, S.: Anschauliche Geometrie, Dover Publications, New York, 1944
302 [5] Jacobi, C.G.J.: Note von der geodatischen Linie auf einem Ellipsoid und den verschiedenen Anwendungen einer merkwiirdigen analytischen Substitution, (1839). Ge sammelte Werke, Band 2, 57-63, Verlag von G. Reimer, Berlin, 1882 [6] Jacobi, C.G.J.: Uber die Abbildung eines ungleichaxigen Ellipsoids auf einer Ebene bei welcher die kleinsten Theile ahnlich bleiben (1861). Gesammelte Werke, Band 2, 399-416. Verlag von G. Reimer, Berlin, 1882 [7] Weyl, H.: Zur Infinitesimalgeometrie: Einordnung der projektiven und konformen Auffassung. Nachrichten der Koniglichen Gesellschaft der Wissenschaften zu Gottingen, Mathematisch-physikalische Klasse, 99-112 (1921). Also printed in [7, p.195-207] [8] Weyl, H.: Gesammelte Abhandlungen, Band II, Springer-Verlag, Berlin-HeidelbergNew York, 1968
K. Voss, Departement Mathematik, ETH Zurich Private address: Wabergstr. 19 CH - 8624 Griit Switzerland e-mail: [email protected] Received February 10, 2000
Geometry and Topology of Submanifolds X eds. W. H. Chen et al. (pp. 303-314) © 2000 World Scientific Publishing Co.
303
T H E CLASSIFICATION OF H O M O G E N E O U S S U R F A C E S I N C P 2
CHANGPING WANG ABSTRACT. A surface M in C P 2 is called (locally) homogeneous, if for any two points p,q £ M there exists a transformation a 6 U(3) which takes a neighborhood of p 6 M to a neighborhood of q G M and takes p to q. Such surfaces automatically have constant curvature and constant Kaehler angle, but in general non-minimal. Minimal surfaces in C P n with constant curvature and constant Kaehler angle have been studied by many authors (see for example [B-W-2], [O]), and minimal homogeneous surfaces in C P 2 have been classified in [E-G-T]. In this paper we classify homogeneous surfaces in C P 2 without the assumption of minimality. We show that any (locally) homogeneous surface in C P 2 is E7(3)-equivalent to an open part of either C P 1 , or the Veronese surface, or R P 2 , or a standard flat torus in C P 2 . We also show that the Kaehler angle 6 of any compact oriented surface in CP has the property that there exists at least a point p G M such that either 6{p) — 0, 6(p) — n or 9(p) — n/2. Keywords: invariants for surfaces in C P 2 , homogeneous surface, classification 1991 Mathematics Subject Classification: 53C42, 53A10
§0. I n t r o d u c t i o n A surface M in C P " is called (locally) homogeneous, if for any two points p, q £ M there exists a transformation a £ U(n + 1) which takes a neighborhood of p £ M to a neighborhood of q £ M and takes p to q. Homogeneous surfaces in C P " form an interesting class of surfaces. They are surfaces with constant Gauss curvature and Kaehler angle, and in general non-minimal. In this paper we prove the following two theorems concerning surfaces in C P 2 . T h e o r e m A . Any (locally) homogeneous surface in C P 2 is U(3) -equivalent to an open part of either C P 1 , or the Veronese surface, or R P 2 , or the standard flat torus in C P 2 . T h e o r e m B . Let x : M —» C P 2 be an immersion of an oriented compact surface. Let 8 be the Kaehler angle of x. Then there exists at least one point p £ M such that either 0(p) = 0, 9(p) = 7T or 9(p) = TT/2 . This paper is organized as follows. In §1 we give invariants system of surfaces in C P 2 . In §2 we classify homogeneous holomorphic curves in C P 2 . In §3 we give structure equations for generic surfaces in C P 2 and prove Theorem B. In §4 we classify homogeneous surfaces in C P 2 which are neither holomorphic nor anti-holomorphic. Combining the results of §2 and §4 then yield Theorem A. Partially supported by NSFC19701003 , DFG466-CHV-II3/127/0 and Qiushi Award. This paper is in final form and no version of it will be submitted for publication elsewhere.
304 §1. Invariants for Surfaces in CP 2 Let CP 2 be the 2-dimensional complex projective space with the Fubini-Study metric ho of constant holomorphic sectional curvature 4. Then h = Re(ho) = ^(/io + ho) is a Riemannian metric on CP 2 . As well-known C/(3) acts on CP 2 as an isometry group: T([Z]):={TZ],
T€S£/(3), Z = (zuz^zrf
€ C 3 \{0}.
For any Z = (z\, z2,23)', W = (IUJ, u»2, IU3)' € C3 we define (1.1)
Z = (z~l, / 2 , z i ) ' ,
£ ■ W^ : = ZltUl + Z 2 1 « 2 + 23W3.
Then the inner product <, > on R6 = C3 can be written as (1.2)
Now let S 5 be the unit sphere in C3 and let n : S 5 —> CP 2 be the standard projection W -> [W]. For any local section Z : B4 -> S 5 of TV defined on an open ball B4 of CP 2 we have (1.3)
h0 = (dZ - (dZ -Z~)Z)® {d~Z - {dZ ■ Z)~Z).
We call (1.3) a local representation of the Fubini-Study metric. From (1.3) we know that the Riemannian metric ft on B4 C CP 2 is given by (1.4) = ^{(dZ-
h=^{h0
+ T0)
(dZ ■ 1)Z) ® (dZ - (dZ ■ Z)~Z) + {dZ - {dZ ■ Z)~Z) ® (dZ - {dZ ■ ~Z)Z)}.
For any Z € S 5 we define H z := {W e C 3 | W ■ ~Z = 0}. As is well-known we have Proposition 1.1: (i) dir : ( H z , < , > ) -> (T[z]CP2,/i) is an isometry; (ii) Let J be the complex structure on CP 2 , then dTr(iW) = J(d-K(W)) for all W € HzLet x : M —> CP 2 be an immersion of an oriented surface. Then x induces a metric g :— x*h on M. For any local section Z of 7r : S 5 -» CP 2 we can define a local lift y := Zox of the immersion x : M —> CP 2 . Such a local lift y of x exists around each point of M. Let y : U —> S 5 be a local lift of x defined on an open set U of M. Let z = u + iv be a complex coordinate for [/ with respect to g. We denote by dz~2^du
W ' 8i~
2(-du+%W
the Cauchy-Riemann operators and define
(1-5)
t~yz- {vz ■ y)v, v ■■= v% - (vz ■ y)v-
305 By (1.4) and (1.5) we have (1.6)
g =x*h =£-rjdz®dz
+ r)- ~£dz ® dz + -(||£|| 2 + |M|2)(dz ® dz + dz
where we use the notation \\Z\\2 := |zi| 2 + |z2|2 + |.Z3|2 forZ= (zi,z2,zi)t. Since z = u + iv is a complex coordinate for M with respect to g, we can find a function LO : U — ► R such that (1.7)
g = -e2"(dz ®dz + dz® dz) = e2uj(du ®du-\-dv®
dv).
From (1.5), (1.6) and (1.7) we get that (1.8)
e-»7=0, M 7 = 0, T)-y = 0;
(1.9)
e - 2 - ^ | | 2 + e- 2 w ||r,|| 2 : =a + 6 = l ,
where we have defined a and b respectively by a : = e - 2 1 | f | | 2 , b := e~2»\\V\\\
(1.10)
It is easy to check that o and b are independent of the choice of the local lift y and the complex coordinate z, they are globally defined invariants of x with respect to the isometry group SU(3) of (CP2, h). Since 0 < o, b < 1, we can define globally an invariant 9 : M -» [0, ?r] by (1.11)
e:=2arcsin(Va).
It is clear that 9 is a continuous function which is smooth around any point p & M with 0 < o(p) < 1. Thus we have a=(sin^)2, 6=(cos^)2.
(1.12)
Proposition 1.2: The invariant 6 defined by (1-11) is exactly the Kaehler angle ofx. Proof. Let z — u + iv be a complex coordinate on M with respect to g. Then by (1.7) we know that {ei,e2} := {e~lJdx(-J^),e~uldx(Jfc)} is an orthonormal basis for dx(TWl). Traditionally, the Kaehler angle 8K is defined by the formula h(ei,Je2) = cos#;j. Now let y : M —> S 5 be a local lift of x. Since for any smooth curve c(t) on M we have i r o t / o c(t) = c(t), thus we get dir(X(y)) = X for all X e TCP 2 , which implies that (d-n)~1(X) = X(y) — (X(y) ■ y)y, where (cfor)-1 is the inverse of the isometry dn in (i) of Proposition 1.1. Thus we have (1.13)
(dn)-\ei)
= e-»(yu - (yu ■ y)y) = e - " ( f + n); (rf7r)-1(e2) = i e -«(£ - r,).
306 Using Proposition 1.1 and (1-12) we get =< (dn)~1(ei),i(dTT)~1(e2)
COSOK = h(ei,Je2) 2
2
= -e- "(M\\ -\\v\\2)
= b-a
>
= coSe.
Therefore 9 is exactly the Kaehler angle of x. Q.E.D. Definition 1.3: A point p € M is called a holomorphic point (resp. anti-holomorphic point, real point) for x : M —> C P 2 if 6(j>) = 0 (resp. Q(p) = IT, 6(p) = \K). A point p e M t s called a complex point of x if it is holomorphic or anti-holomorphic. It is clear that if all points of a connected surface x : M —> C P 2 are complex, then x is a holomorphic or anti-holomorphic curve in C P 2 . To introduce other 5'Z7(3)-invariants we define the following complex 1-form $ and 3-form 9: (1.14)
(1.15)
$ := e _ 2 w 6 • fjdz = - e - 2 " ^ • rjsdz := ipdz; * := 6 ■ rjdz3 = -£. ■ %dz3 := ipdz3.
Using (1.8) one can easily verify that $ and \t are independent of the choice of local lift y of x and the complex coordinate z of ( M , ) and thus globally defined on M . Moreover, if x is transformed by an isometry T e £77(3), then £ and n are also transformed by the same T, thus by (1.14) and (1.15) we know t h a t $ and * are SU(3) invariant. It is not difficult to show t h a t x : M —> C P 2 is minimal if and only if $ = 0. By the definition of $ and \f we have P r o p o s i t i o n 1.4: The complex points of x are zeros o / $ and * . §2. H o m o g e n e o u s h o l o m o r p h i c curves i n C P 2 In this section we assume t h a t x : M —> C P 2 is a locally homogeneous holomorphic curve. Let y : V —> S 5 is a local lift of x, then we have (2.1)
„ , = \y
for some smooth function A : U —> C. We define (2-2)
F =
e-6"\[y,yz,yz,}\2,
where [■ • - ] is the determinant in C 3 . It is easy to check t h a t F is independent of the choice of the local lift y and the complex coordinate z and thus a globally defined 5 , J7(3)-invariant function on M. By the homogeneity of x we know t h a t F is a constant. From (2.1) we get V = Vz - (Uz ■ y)y = 0. Thus we know from (1.9) that {y, £} is a subbasis in C 3 . First we consider the case that F = 0. In this case we know from (2.1) and (2.2) that y and y* := yz satisfies the following P D E system: (2-3)
Vz = V*; Vz = *y; y*z=ay+/3y*;
y*z=\zy
+ \y*
307 for some smooth function a and /3. Now let v be a unit vector in C3 such that (2.4)
i».j/ = i / . p ' = 0
at one point p G U, then by the uniqueness of the initial problem of the PDE system (2.3) we know that (2.4) holds in all point of M. It follows that x is isometric to the plane (2.5)
CP 1 = {[(0, z, wf] G CP 2 | \z\2 + H 2 = 1}
in CP 2 , which is the orbit of the subgroup
j ft) z - w ) ||z| 2 + M 2 = l l of 5(7(3) acting on the point [(0,1,0)'] G CP 2 . Now we consider the case that F / 0 . We define
(2.6)
" = &-(k-f)iKir 2 *-
Then by (2.2) we have [y,£,v] = [y,yz,yzz] ^ 0. Using (1.8) and (2.1) we get
Thus {y,£, v\ is a local moving frame in C3 along M. From (2.2) we get that
Since ||£||2 = e 2 ", we have ||i/|| 2 = Fe4", thus for any function V : M -> C3 we have the following formula (2.7)
V = {V ■ y)y + e~2u{V • £)£ + F - 1 e " 4 w ( y • v)v.
From (1.5), (2.1) and the fact ||y||2 = 1 we deduce that (2.8)
yz = -Xy + £, y-z = Xy.
From (2.6), (2.8) and the identity yzz = yzz we find that (2.9)
& = «* + "> Z-z =
-e2"y^Xti,
where a = (£z • ?)||£||~ 2 and we also have used the formula £z ■ y = —£ ■ yl, which implies that (2.10)
Xz+Xz=-e 2ui
308 From (2.9) and the identity £zz = £zz we get (2.11)
vz = ( - a - 2wz + J)e2wy + (\z - az - e2l")£ + \v.
Since vz -y = -v -pj = 0 and vz-~l = -v-(,z = -Fe4u, (2.12)
we get from (2.10) and (2.11) that
a = 2uz - A, 2UJZZ + (2 - F)e2" = 0.
Since vz ■ y = — v ■ yz = 0 , vz ■ \ = — v ■ ^z = 0 and \\v\\2 = Fe 4 w , we get from (2.11) and (2.12) that (2.13)
uz = (4u^ - J)v, vz = -Fe2u£
+ \v.
From (2.13) and the identity vzz = vzz we get that (2.14) 4uzz + (1 + F)e2u = 0. It now follows from (2.12) and (2.14) that F = 1. Thus the Gauss curvature K = Ae~2uu!zz = 2, which means that our surface is isometric to (C, /x_J.u ny.i). Therefore we can find the complex coordinate z on the surface such that 2
(2.15)
(1 + kl 2 ) 2 '
Using this complex coordinate we get from (2.10) and (2.15) that (2.16)
( A - A 0 ) , + ( A - A 0 ) , = 0 , A0
l + l*l:
which implies that i{(\ — Ao)dz — (A — \o)dz} is a closed real l-form. Thus we can write locally this l-form as d# for some real function ■&. Since Ao = A 4- itf^, it is easy to check that if we change the local lift y : U —> S 5 to y := e'^y, then the function A defined by (2.1) will change to Ao- We note that all coefficients in the structure equations (2.8), (2.9) and (2.13) depend only on A. If we normalize the complex coordinate z by (2.15) and the local lift by y, then we get the structure equations for x by letting A = Ao: z
(2-17)
yz=
z | 2 3/ + g, Vz = -1 , | i2y> 1 + \zr 1 + \zr
z
2
l + p|
C1 + \z\ )
z 1 + |z|*
2 (1 + |z| 2 ) 2
z
1+
FI
^ 1+ F|2
309 The last three equations are integrable PDE system of {y, £, v}. It is clear that two surfaces satisfying this PDE system are SU(3)—equivalent. An easy (surface) solution to the system are given by (2-20)
y=TJ—(i,y/2z,z*)*.
(2-21)
^=(1+}z|2)2(-2^v/2(l-N2),2z)t;
(2.22)
,
=
_ J _ ( ,
2
, _ ^ , i r .
[y] defines the so called Veronese surface in CP 2 , which is an orbit of the subgroup
(2.23)
{( w2 \\-V2wz \\ -z2
V2wz \w\2-\z\2 -V2zw
z2 \ ) V2zw | \z\2 + \w\2 = 1 } w2 ) J
of SU(3) acting on the point [(1,0,0)']. Thus up to transformations of SU(3) the surface is an open part of the Veronese surface in CP 2 . As a summary to this section we have Theorem 2.1: Any local homogeneous holomorphic curve in CP2 is SU(3) — equivalent to an open part of the standard CP 1 or the Veronese surface. It is clear that the conjugated surface [y] of a homogeneous surface x = [y] in CP 2 is still homogeneous. Thus we have classified all homogeneous holomorphic or anti-holomorphic curves in CP 2 . §3. Structure equations for surfaces without any complex points in CP 2 Let x : M —y CP2 be a surface without any complex points. Let y : U —> S 5 be a local lift of x. Let £ and r) be as defined in equation (1.5). Then at each point of U {y,£,ri} is a basis for C 3 . Thus for any smooth function F : U —> C3 we have the formula (3.1))
V = (V-y)y+(V-
O a ^ e " 2 ^ + (V ■
rj^e-^jj.
From (1.5) and the fact that y ■ y = 1 we get (3.2)
!/z = Ay + e, y-z = -Jy + n,
where A = yz -y. From (1.8) and (3.2) we have £z ■ y = —f • yz = 0, thus by (1.14) and (3.1) we can find functions p such that Zz = pt + b-1e-2"iPn.
(3.3) By (3.2) and (3.3) we get that
6 ■ V = -Z- T* = ~ae2", 6 • £ = (||f||2),- - £ ■ Tz = (ae2u)-z ~ ae2"p,
310 thus from (1.10) and (3.1) we get £i = -ae2"y
(3.4)
+ ((log(ae 2 <")) z - p)£ + fc" W
It follows from (3.2) and the identity yzz = yzz that (3.5)
- ae2")y + ((log(oe 2 ")) z - - p + A)£ + (A +
r,z = (A, + \
Since by (3.2) and (1.10) we have r\z-y = —r\-yz — -be2w \-z+ \ z = (a - b)e2u,
(3.6)
b~ly)V.
and r\z ■ £ = —e2ujlp, thus we get
(log(ae 2 u )) 2 - p + A + o " V = 0.
By (3.2), (1.14), (3.3) and (3.5) we have % • y = -r) ■ Tz = 0, Vz ■ I = -»? ■ 6 = -*/> and % • r? = ( N | 2 ) z - - r, ■ Tz = (be2"h
- be2"{\ + 6" V ) ,
thus we have (3.7)
m = -ar^e-^M
+ ((log(he 2 w )), - A - 6" V h -
From (3.3), (3.4), (3.5) and the identity (,zz = £,zz we get (3.8)
(log(ae2")U = - ( a " V ) , - ( a " 1 ? ) , + ( a 6 ) - V 4 " M 2 -
(at)"V|2
+ (6 - 2o)e 2 w
(3.9)
i/-z + ( a " 1 -
ft-1)^
= e 2 w ( ^ - 2uz
t r V V
2
-e *V(log(a6)),; From (3.5), (3.6), (3.7) and the identity rjzz = r\zz we get (3.10)
(log(6e 2 ")) zz - ={b-\)-z + (b-lTp)z + (aby'e-^M2 + (o-26)e2w.
- (o6)"V|2
By (1.12) we have a/fc = ( t a n 0 / 2 ) 2 , a^+b'1
4
(sin9)2'
It follows from (3.8) and (3.10) that (3.11)
A log tan 0/2 = - 8 e - 2 " { ( — ^ — ) z - + {yJ^)z} (sine/)*1 (sint')' !
+ 6cos0.
We note that (3.11) holds at any point p e M with 9(p) ^ 0, TV. We define the real l-form O on M by
n:=-i—±-j(3-«). (sin#)^
311 Then we have
(3-12)
da 2e 2
= " "«(imV ) z " + ( ( i m V ) J " M '
where dM is the volume form on M . It follows from (3.11) and (3.12) that for any immersion of compact oriented surface M without complex point in C P 2 we have the integral formula (3.13)
/ cos 6dM = 0, M
which implies that there exists at least a point p e M such t h a t 0(p) = n / 2 . Thus we have proved Theorem B. §4. H o m o g e n e o u s surfaces w i t h o u t c o m p l e x p o i n t i n C P 2 Now we assume that x : M —t CP2 be a homogeneous surface without complex point. Then all the 5(7(3) invariant functions of £ are constant. In particular, the Kaehler angle 6, and thus the functions a and b in the formulas of §3, the Gauss curvature K = — 4e~2lJuizz of x are constant. Since $ = ipdz is a globally defined 5(7(3) invariant, by the homogeneity of x we know that either $ has no zero point or it vanishes identically. It is easy to check that the 2-form (ipz — 2uztp)dz2, the functions e - 2 l *'|^| 2 and e~2loipz are also globally defined 5(7(3) invariants. Thus we can find constants C\, 02,03 such t h a t (4.1)
e-2«>|2 =
Cl,
(Vz - 2wtV)
= c2
where we take C2 = 0 for the case $ = 0. Using (4.1) and the identities (e~2"\tp\2)z and tpzi = >fzz we get that (4.2)
cic 2 + c j = 0 ,
Kcx =
= 0
-A\c3\2.
T h e o r e m 4 . 1 : Let x : M —> C P 2 be a local homogeneous surface without complex point. If its Gauss curvature is positive, then up to a 5(7(3) transformation x is an open part of the standardMP2 in CP2 given by the immersion (xi,0:2,3:3) 6 S 2 - * [(2:1,2:2,£3)'] 6 C P 2 . Proof. From the second formula of (4.2) and the fact that K > 0 and c\ > 0 we get ci = 0. Thus $ = 0. It follows from (3.11) and (1.12) that Q = -K/2 and a = b, i.e. x is totally real. From (3.9) we know that ipz = 0. Since ||*|| 2 := e - 6 " ^ ! 2 is a globally defined 5!7(3) invariant, it must be a constant. If $ ^ 0, we get 0 = A l o g | | $ | | 2 = 4{log(e _ 6 "|V'| 2 )} z j = 6 ^ , a contradiction to the fact that K > 0. Thus we have $ = 0. It follows from (3.8) that K = —4e~2u,ujzz = 1. Thus we can find a complex coordinate z of x such that e 2 u = / t . u|a\a ■ Since x is totally real, we get from (3.6) that Xz + \ z = 0. We know that d(idy ■ y) = 0, thus we can find real function 1? such that idy ■ y = dft. Thus by taking the local lift y := e'®y if necessary we may assume that A = 0. In fact, such lift is a horizontal lift of x. Thus from (3.6) we get p = 2wz = — 1+ f , a . Therefore by (3.2), (3.3), (3.4), (3.5) and (3.7) we have the following structure equations for x: (4.3)
yz = f, yz = ??;
312 - _ ^
- _
(4-4)
& =
(«)
^ = -JTTWT2y' * = -TTwv-
|
6
=
Since two surfaces satisfying the same PDE system (4.3), (4.4) and (4.5) are SU(3) equiv alent, and we know an easy (surface) solution to this PDE system: z,i(z-z),l-\z\2)t-,
(4.6)
y=—l—.{z
(4-7)
$ = ^ - J - ^ ( l - z 2 , - i ( l + z2),-2^;
(4.8)
,, = __j_(i_^ )< (i
+
+ z ») ) _2*)«;
thus up to a transformation in SU(3) the homogeneous surface x is an open part of the surface [y] defined by (4.6). It is clear that [y] is a part of the surface RP 2 in CP 2 given by the immersion (2:1,2:2,£3) £ S 2 - » [(2:1,2:2,2:3)'] € CP 2 , parameterized by a stereographic projection, which is SU(3) equivalent to the orbit of the subgroup of SU(3) given by (2.23) acting on the point [(0,1,0)'] e CP 2 . Q.E.D. Theorem 4.2: Let x : M —► CP 2 be a locally homogeneous surface without complex point. If its Gauss curvature is flat, then up to a U(3) transformation x is an open part of one of the following homogeneous tori in CP2: (4.9) where (Ti,T2,T3)
{ [ ( T i e " - " , T 2 e T ' - f * , T g e ^ - ^ ) ' ] | z e C}, and (a,r,n)
are constant vectors in C3 satisfying
T1T2T3 ^ 0, of - TO + rjX - p,f + fxo - op, ± 0. Proof. From (4.2) we get C3 = 0 and C2 = 0. Since K = 0, we can choose a complex coordinate z such that ui = 0. With this coordinate the function
|i/>|2 = 1 + M 2 , cose = 6 - a = 0,
thus x is totally real. By (3.9) we get ipz = 0, thus the first equation of (4.10) yields i/> is constant. By the first equation of (3.6) we have A^ + \ z = 0. Then by taking the normalized local lift y = e'*j/, if necessary, where ■& is the real function defined by the equation cW = i(—\dz + \dz), we may assume that A = 0. Thus we get the following structure equation: (4-11)
Vz = £, Vz = V,
313 (4.12)
L =
(4.13)
T)z = -y-Tp£
& = -V-
+
+ tpr), % = -i/>£ - Wi
lt is an integrable PDE system with constants tp, ip £ C satisfying |V>|2 = 1 + \f\2- If we write E = (y, f, 7)), then this PDE system can be written as (4.14)
dE =
E(Adz-~Atdz),
where A is the matrix 0 1
(4.15)
0 v
o v
-l -Xp
Since AA = A A, we can find T € !7(3) such that TAT = diag(a, r, /x). Thus we get E = E(0)r'dioff (e ffz - ffJ , e " - " ,
(4.16)
e^~^)T.
Since T* := diaff(l, ^ 2 , ^2)^(0) e f (3), if we write T(1,0,0)' = ( T 1 , r 2 , r 3 ) t , we know from (4.16) that x = [y] is (7(3) equivalent to an open part of the surface (4.9) by the transformation T*. It is clear that the surface (4.9) is an orbit of the subgroup G = {diag(e"z-^, acting on the point [(T1,T2,T3y]
erz~f\
e^'^)
| z e C}
e CP 2 . Q.E.D.
Finally we consider the case that x : M —> CP 2 is a local homogeneous surface with negative Gauss curvature K. In this case we have $ ^ 0, otherwise we get from (3.9) that ^ is a holomorphic cubic form such that ||\F|| := e~6u\ip\2 = const., then from the formula A log H^ll2 = 6K we get K — 0, a contradiction to our assumption that K is negative. Since $ ^ 0, by the homogeneity of x we know that there exists a constant c4 such that * = c 4 $ 3 . From (4.1), (3.8), (3.10) and (3.9) we get that (4.17)
(4.18)
(4.19)
~K
~K
= -a-l(c3
+ c3) + (ah)-1|c4|2c? - (ab)'1 Ci + (b - 2a);
= b~l(c3 + c3) + (afc)-1|c4|2c1 - (ab)-la
3c4c3 + ( a - 1 - 6 _1 )c 4 ci = c2 + (6 _ 1 - a - 1 ) .
It follows from (4.17), (4.18) and (4.1) that (4.20)
+ (a - 26);
c3 + c3= 3ab{b - a) = -Cj (c2 + c 2 ).
314 Thus from (4.17) and (4.18) we get -K + 1 + 3oh(6 - a)2 = (a6) _1 ci(|c 4 | 2 c 2 - 1),
(4.21)
where K = —ci|c2|2 < 0, wh ich implies that |c4|2cf > 1 and c2 ^ 0. On the other hand we know from (4.19) and (4.1) that c 4 C!{-3c 2 + (afc)_1(6 - a)} = c2 + (o6) _1 (o - b).
(4.22)
Since by (4.20) i?e(c2) has the same sign as (a - b), we get from (4.22) that |c4|2cf < 1, a contradiction. This shows that there exists no local homogeneous surface with negative Gauss curvature. To summarize we have Theorem 4.3: Any locally homogeneous surface without complex point in CP2 is U(3)equivalent to an open part of RP2 or a standard flat torus in CP2 given by Theorem 4.2. Thus Theorem A follows from Theorem 2.1 and Theorem 4.3. REFERENCES [B-J-R-W] Bolton, J., Jensen, G.R., Rigoli, M., Woodward, L.M., On conformal minimal immersions of S2 into CP", Math. Ann. 279, 599-620 (1988). [B-W-l]
Bolton, J., Woodward, L.M., Congruence theorems for harmonic maps from a Riemann into CP" and Sn, J. London Math. Soc. (2) 45 363-376 (1992).
[B-W-2]
Bolton, J.,Woodward, L.M., Minimal surfaces in CPn with constant curvature and Kaehler angle, Math. Proc. Camb. Pil. Soc 112, 287-296 (1992).
[C-W]
Chern, S.S., Wolfson, J., Minimal surfaces by moving frames, Am. J. Math. 105, 59-83 (1983).
[E-G-T]
Eschenburg, J.H., Guadalupe, I.V., Tribuzy, R.A., The fundamental surfaces in C P 2 , Math. Ann. 270, 571-598 (1985).
[E-W]
Eells, J., Wood, J.C., Harmonic maps from surfaces to complex projective spaces, Adv. Math. 49, 217-263 (1983).
[J-R]
Jensen, G.R., Rigoli, M., A quantization theorem for harmonic S2 in CPn, Abstr. Am. Math. Soc. 6, 383 (1985).
[O]
Ohnita, Y., Minimal surfaces with constant curvature and Kaehler angle in complex space forms, Tsukuba J. Math 13, 191-207 (1989).
[R]
Rigoli, M., A rigidity result for holomorphic immersions Soc. 93, 317-320 (1985).
[W]
Wolfson, J.G., Harmonic sequences and harmonic maps of surfaces to complex manifolds, J. Differential Geom. 27, 161-178 (1988).
Changping
Wang:
Department
of Mathematics,
Peking
University,
Beijing
surface
equations of minimal
of surfaces in CPn, Proc. Am. Math.
100871,
E-MAIL: uangcp81pku.edu.cn
People's
F A X
Received January 7, 2000
.
0086
Republic 10
6
Grassmann
of
275i8oi
China
Geometry and Topology of Submanifolds X eds. W. H. Chen et al. (pp. 315-335) © 2000 World Scientific Publishing Co.
315
A semi-classical limit and its applications Yanlin Yu Abstract This paper deals with an example of a semi-classical consideration for heat equa tions. It shows that some geometric theorems are corollaries of such a semi-classical consideration. Key words: semi-classical limit, heat equation, Morse inequalities, Hopf index the orem. Math. Subject Classification: 58G10, 58G11, 58A10, 53C05.
1
An example of a semi-classical consideration
Let M be a compact, closed Riemannian manifold of dim n, and V a vector field without degenerate zeros on M. Let A*(M) be the space of differential forms on M, and D = d + S: A*(M)-> A*(M) be the de Rham-Hodge operator, which is an elliptic operator. Let us consider Witten's deformation of d + S. Dt = (d + S)+ t[V* A +i{V)} : A'(M) -> A*(M), where V* is a 1-form dual to the vector field V, and V*A means the exterior product by V*, while i(V) the interior product by V. It is easy to see that Dt is a self-conjugate operator. Let □ t = D\: A * ( M ) - > A * ( M ) , which is a family of Laplacian operators with a parameter t. If D t is thought as a deformation of a physical system, Witten ([3]) considered a limit situation of D t as t —> oo and called it a "semi-classical limit". More precisely he considered a marvelous number, which is the number of so-called "small eigenvalues" determined by the semi-classical limit, and related it with geometric problems. By Bismut's observation this number can be expressed as lim (5 — lim)tr e~ TD ' This paper is supported in part by National Natural Science Foundation of China and Funds of Chinese Academy of Sciences, and it is in final form and no version of it will be submitted for publication elsewhere.
316 for a non-degenerate gradient vector field V, where e TD| is a family of solution operators of the heat operators ^r + □(, and (s - lim) means a limit process as "r —> 0 and t —> oo and r i
is kept as a constant s."
-Ta
In this paper we examine (s — lim)tre '. Let {E\,..., (field) on M. The vector field V can be expressed as
En} be a local orthonormal frame
t
Define v,j by V i ^ = !><;£*, where y is the Levi-Civita connection. In general the matrix (vtj) is not symmetric. We define its absolute matrix as follows. Denote the matrix (vij) by B, and its transpose by B*. Let the absolute matrix of B be the arithmetic square root of a positive symmetric matrix BB", and denote it by |B|. It is clear that |B| is a positive symmetric matrix. And let 6 = 2s\B\. Recall we have defined the linear map [V* A +i(V)], now we define two maps
E?,Er by E+ = {E,)' A +i(Ej), E~ = (Ej)' A -i(Ej). The main theorem of this paper is as follows. Theorem 1 For the family of Laplacians Ot defined by using a vector field without degenerate zeros as above, and for a parameter s > 0, we have 1. (s - lim)tre~ Tn ' is finite. If we denote it by P(s), then P(s)=
£
(det^2(cosh0(p„)-l))
tvexpisJ^vME+E-^J.
Pc,eZero(v)
(It is worthy to note that if V = 0, then (s — lim)tre - T °' = oo.) 2. P(s) is a positive, decreasing function. 3. lim4_x) P(s) limj-Hx, P(s)
= oo. > lim inf^oo dim ker Ot.
4. Let a be the super symmetric isomorphism of A*(M) such that it is 1 on and - 1 on Aodd(M), and let stre- TD ' = tr • ae~Ta\ then stre
TD
' does not depend on r and t, it is the Euler number x{M).
Aeven(M),
317 The new point in the theorem is the first assertion. Let us describe a proof of it as follows. The proof is based on a discussion on the integral kernel of e~TOt. There exists a unique family of linear maps G(r,q,p,t) : A*p(M) ^ A'g(M) such that (e-ra't)(q)=
[ G(T,q,p,t)
V0.
JM
Such a family G(T, q, p, t) is called a fundamental solution of the heat operator Jj: + Dt. The fundamental solution can be determined by the following equations (£ + Ot)G(T,q,p,t)
=0
l i m / G{r,q,p,t)(j>(p)dp= <j>(q),
V<£,
where Ot acts on the indeterminate q. By using the fundamental solution we have the following expression tre- TDe = / trG(T,p,p,t)dp. JM
How to understand the fundamental solution G(T, q,p, t) is a very serious task. The difficulty comes from the fact that G(r,q,p,t) and even G{r,p,p,t) are defined globally. That means for an open set U in M we can not determine G(r, q, p, t) | u by using geometric data of U. A well-known asymptotic theorem only says that limT_joG(r, p, p, t) can be determined locally, i.e. the formal power series of G(r, p,p, t) at r = 0 can be determined by local data around p. Now in our case we do not know if (s — lim)G(r, p,p, t) can be determined locally, all we want to prove is that (s — lim) / tr G(r,p, p, t)dp JM
can be determined "locally". It means that we can introduce a special parametrix H(T, q,p,t), which is defined locally of course, and prove an equality (s — lim) / tr G(T, p,p,t)dp = (s — lim) / tr JM
H(r,p,p,t)dp.
JM
An evaluation of (s — lim) / tr H(T} p, p, t)dp in §4 will give a proof of the theorem. JM
2
Parametrix
In this and the next paragraphs, V may be degenerate. Choose a local orthonormal frame (field) {Ei,..., En} on M, let {w\,..., uin} be the coframes dual to {Ei,..., En}. From the definitions in §1, it follows that V* = Si^t^ii and V" A +i(V) = YliViEf. Therefore Dt = (d + 6) + £ tv.Et = YXEr V £ j + tviEt), i
i
318 and D( = D2 = (d + 6)2 + Ej,k(Ek^EktvjE+ + tVjEfE;VEk) + e Ej,k VjV.E+Et = (d {d + + S) 5)22 + + EEj, tvjjkkEE^Ef + EEj, tvj + EfE^)VEk + t2 £,- v-2 jtk k tv k~Ef + jtkk tVjiE^Ef " t„,'.. c + IT- _i_ A2'ir .,"2 = {d + S)22 _- ^Ej, k tVjkEfE; + t Ej v) (One may see [6] for the multiplication table of Ef,..., £+, E± ,..., E~.) Now for p e M, in a neighborhood of p we choose a normal coordinate system {j/i,..., yn} centering at p and an orthonormal moving frame {Ei,..., En}, which is parallel along geodesies passing through p and E
i(Py-7^7\P-
Of course the coordinates of p are (0,..., 0). Suppose the coordinates of q are (j/ x ,..., yn) = Y, for q near p. Ot can be written as D
< = - E ^ 2 + *2 E ( « * ( P ) + E " j t W - Y,tvjk(p)E+E£ + .... i
u
Vi
j
j,k
k
Define a parametrix H{r,q,p,t)
: h'p{M) ^ k*q{M)
by H(r, q,p, t) = *(r, Y,p, t)exp {£ rtvjk(p)E+E;}
•
where
*(r,F, P,t) =
ifay/det^)
e x p { - ^ y ^ H ^ F * - 2Tt2Yf^B>(p)a*(p)
-
2rt2a(p)0^^a'(p)},
in which (B{p))ij = %(P)> <*(P) = ("i(p), • • •, vn(p)), and 9 and G# are 2rt\B*(p)\ and 2ri|B(p)| respectively, Y = (yi,...,yn), and a* are the transposes of B, Y and o respectively. Moreover,
£+:
A;(M)
where B*, Y*
-> A;(M)
means a composition of £+ in A*(M) and a parallel translation from p to q. The definition of H(T, q,p, t) is based on a relation between
«
U!/
'
i
*
j,k
and $(T, V,p, f), which is proved in the following lemma. Lemma 1 It holds
(i: ~ E I s + ? E(»J(P) + E^(p)2/*)2)$(r, ,p, i) = 0. "T
i
"Hi
j
k
319 Proof Sometime we denote an (i,j)~element of a matrix C by Cy, and coshG by cosh, and cosh G* by cosh*. Note that ~ does not depend on r, hence §f{j) = 0. This fact helps in the following computations. First 6Y
then
j y
\ T sinh /
dry
\ T sinh /
4
r 6V \ sinh /
/SfcfcT */**(A) = i(d««(A))"'*(*«(A))
^
—
ItrU 2
e (( V1 ^TsinhJ
-^- f 6 ^ 9T VTsinheJy
l f rU / r s i n h e f 2 V S T (,
=
cosh 6 ^ sinh 2 T / ^
1.4-j. /* Q cosh \ _
2 l r ^rsinhj '
And then by 02 T sinh 2 '
G d /cosh\ T 8T \ sinh j
T d /cosh—1\ 0 9r \ sinh J
2
cosh —1 sinh2 '
we get _
2 i
2ycosh-i sinh 2
.
B
. _
2t
2
cosh*-l sinh*
»| J
Again,
y \ i£r fr — -I- f ^-"8yf
y +2f
W,r fecoshNl I 2T "
V sinh /
2yCosh(cosh-l)B<.a.
T +
iv(e<x*to\2Y"
IT7
(, sinh /
4r2i4oB
'
(igl-d) 2 B-V
} .
Therefore S-HI:*
" Z« $»*) = " 4 ^ y e 2 y * - 2t2YB'a* - 2 t 2 a £ ^ f i a : = -47yy© 2 y* -2t2YB*a*
-lt2a^^-a*
- t2a ( g f e ) 2 G#V
sinh* 2
Ve#sinh*/
= - t r B * B y - 2t YB*a* - t2aa* = -t2{YB,
2
+ a)(BY, + a*).
The lemma is proved. Before comparing G(r,q,p,t) with H(r,q,p,t),
we first need to bound
320 because of the equality (g^ + at)G(r,q,p,t)
= 0.
For the estimation we introduce Q(T,q,p) = ^ = * e x p \-^lf-
) : A;{M)
AT
-+ A'q(M),
where p(q,p) is the distance between p and q. Lemma 2 For s0 > 0 there exist c0, C! > 0 such that for rt < s0 we have \H(T, q,p, t)\ < C0Q{CIT, q,p)exp I
Proof
u(p)2 I .
First note that |exp i,TtVjk(p)E^E^\
| < const.,
and then we have *(*-,«,
P,t)
=
^ ^ d e t ( ^ e x p f - i y ^ ^ y
exp{-£(YW +^aB^^W2 #
1
) ^ ^ + 4 s a B ^ W"
s „ sinh6 „*1 "Ta0#coshe#a J
<
^
d
e
t
(a^e)«p{-^ag^H,a-},
and
*(r l9 , P.*) = ^ V / d e t ( ^ ) 2
exp{-2ri :p{-2rt2 (aW (oTF+ + :i y ^ l B ' r
1
) ^ + ly^iB*^-1)
1 yO(coBh+l)yt I 8T sinh * J
<
i „ ,/dctC
— vw V
9
Icxpf
v sinhe /
lyefcosh+Dy.-.
^
8T
sinh
■''
where cosh* —1 6 cosh 0 , W =. sinh0 \|e#sinh#' they are conjugate to each other. Therefore W ■■
321 The lemma is proved. Lemma 3 For so > 0 there exist Co, C\ such that for any r, t with 0 < rt < so we have o
(
I (fr + °t)H Proof
,2
(r> «> P. *)l < c o(v^* + 1)<9(CIT, g, p)exp | - ^ - « ( P ) 2
First we recall some notations and facts in [5], let H,j, T* , Rijki be defined by
Ei = H'>i:, V^Ej
= T*Ek,
Riiki = - < ( V E i V E . - VEj V E j - V[Bi,Ej])Ek, Ei >, and let (Hij) be the inverse of (77 u ). Corollary 8 in [5] claims a Taylor's expansion Hijiv) = <% + - Y Riktj{p)Vkyi + ■■■■ ° k,l As usual Ao is defined by Ao = £ ( V £ , V E j - V F i ) , i
where Ft = VE.Ei. The Weitzenboeck formula reads (d + S)2 = - A 0 + R, where fi
=«E
RjkiE'EJ
E~£ Ef + - £ itijij -
Thus D t = - Ao - Y
tojkEtEk
+t2Y
jk
tf
+ R-
j
Denote exp < £ rtvjk{p)E+E^
\ (j>(q, p)
by 17, then H = $ [ 7 By using Lemma 1, Weitzenboock's formula and the well known formula A 0 ( W ) = (Ao*)C/+ 2 £ ( V E j $ ) ( V £ i r j ) + $(A 0 £7), i
we have ( & + at)($U)
=
-[(A0 - E , ^ ) * ] t / - 2 E,(VEi$)(VEirj) +7 1 + 72 + 73 + 74,
where A
=*(£U-Et»jkE}EZU)
h = t2[E,«? - Ei(«i(p) + £* ^ ( P M W /3 = £*£/, 74
=-*A0t/.
322 Note that Lemma 2 still holds if we replace H by $. We write the right-hand side of the inequality in Lemma 2 as COQ(CIT, q,p)exp j -^-v{j>f
\ = VTCOQ(2CIT,
q,p)exp < -^{p)2
\ (,,
where
'{^
f = exp \ —^zP{q,P)
ri - -£TV(P)
If we can prove \^-1(^- + at){$U)\ < const.(v^t + 1), OT
then
|(& + ni)(«OI = l*-x(£ + °«)(«0H*l < l ? * " 1 ^ + □ ( )(*?7)|v^ n coQ(2 C l r, g ,p)exp{-^(p) 2 } < const.c 0 Q(2cir,9,p)exp{-gijj(p) 2 }( x /ft 4-1), which implies Lemma 3. So we check 6 terms in the expression of ( ^ + D()($!7) along this way. Note that, for any mi,m2 > 0, P2{l,P) r
(rtMriT^C
where the constant c depends only on wii,m2. So we have \t*~lh\
< t2\ Efe(p) + «,■*(?)» + 0(M 2 )) 2 - E(»i(p) + 2
2
vjk{p)ykm
3
Vk ~ 2rt 2 { ^ )
j h
(B'a%)
323 it follows l ^ - 1 * , ! < const. ( M + Tt2\v{p)\\ £. Note that <j>(q,p) = 1 for q , which is near p, and exp < YlTivjk(p)E^Ek on y, so due to |Ff | < const.\y\ we have \Ui\ < const.\y\, hence |^$ _1 $ii7i| < const. (—
+ Tt2\v(p)\\y\)t<
Let
then
+T?iWaAj}, and
+r?,/p%}. By using fl^' = <% + - Y, Rikij{p)ykyi + ■■■,
and (^■+rt2\v{p)\),
\Aj\ < const. we have
2
\&-\9« - I5^*)| < const. 2 ay,
Therefore the lemma is proved.
const.
\ does not depend
324
3
Levi iteration
Definition Km(r,q,p,t),
Suppose H{r,q,p,t) is given. By the following procedure we construct K(r,q,p,t), and G(r,q,p,t) as follows K0(r,q,p,t)
Km+i{r,q,p,t) K(r,q,p,t)
=(£
+
=1
dv I K0(T - v,q,z,i)Km(v,z,p,t)dz,
=^(-l)
nt)H(r,q,p,t),
m+1
Vm > 0,
^m(r,g,p,<),
m=0
G(r,q,p,t)
=H(r,q,p,t)+
dv \ H(r — v,q,z,t)K(v,z,p,t)dz. Jo JM The above procedure is called Levi iteration. The first question for the Levi iteration is whether the series, which defines K(r,q,p,t), converges, and whether G(r,q,p,t) defined here is the fundamental solution. And the second question is how K(T, q, p, t) inherits a kind of bounded properties of ( ^ + Dt)H(r, q,p, t). The second question is a key point. If we solve it well, the first question can be solved easily by a routine trick described in the book [4], and the difference between G(r,q,p,t) and H{r,q,p,t) can be estimated as well. It may be difficult to prove that K(T, q,p, i) satisfies Lemma 3 with K0(r, q,p, t), however we can prove that they satisfies Lemma 4, which is shown later. In order to do that we need the following Lemma A and Lemma B. In Euclidean space of dim n there is a well known convolution formula / Q(r-v,q,z)Q{v,z,p)dz = Q{T,q,p), which can be extended to Riemannian manifolds, i.e. there yields /
G(r-v,q,z)G(v,z,p)dz
= G(T,q,p),
in which G(r,p,q) is a Brownian motion or a fundamental solution of the heat equations. However in general, G(r,p,q) is different from Q(r,p,q) defined above. Due to the lack of a precise formula for G(r,p,q), we can not use it to estimate geometric quantities. We are reluctant to use Q(T, p, q), and prove the following lemma instead of the well known formula about the Brownian motion G(r,p,q). Lemma A In a Riemannian manifold M for a fixed sufficient small e > 0 and for any ci,c 2 > 0 with c\ < c>i, there exists a constant c = c(ci,c2,e) such that, for any q, p € M, v, T > 0 with 0 < v < r, we have
/
<3(CI(T - v),q,z)Q{c2v,z,p)dz
< cQ(c2r,q,
p(q,z)
Proof
It is equivalent to prove an inequality
/ . r - exp\-A(v,T,q,z,p)\dz< const., Jp(q,z)<e V [T - V)V 14 J where const, means a constant which does not depend on u, r, q, and p, and where
A(v,T,q,z,p) = - 4 ^ 4 - ^ ^ + Ci(T-l/)
CiV
C2T
P2M1
-
325 Now we prove the above inequality in three separate cases. (i) If p(q,p) > 4e and ^ > | , then due to p(q,z) < e, p{z,P) > p{q,P) ~ p{q,z)>At-t
r
= 3e,
-p(z,p) - T-p(q,z) > ^P(Z,P) - 2i:p{q,z) > T- -2 o = e> r v ,r rvu ' " ' ~ 2rK r' ' ~ 2
and thus A(i/,T,g,z,p)
< —e^Ml _ £!lMl + fi!tol < _ _ £ % * ! _ P 2 (*»P) _i_ (p(g>*)+p(*»p))2
= -^F^F(7P(?.^)-I^P(Z.P))2 <
I
e2
Therefore — e 2 } < const.
exp— < . / - — ' — — exp{—-— (r — v)v
4
y (r — v)v
4C2(T -
v)v
(ii) If p(g,p) > 4e and 2 ^ < i,then 7 > 5 and thus ,/rZ7"exp4
<\/
= 1 : : n
exp{-(i-i)|^4 + i ( - 4
k
4
P2(*J>) _l_ p 2 ( g j » h ) C2f
< ^ e x
C2T
P
'J
{ - ( i - £)$«#- -£-5-(*,(,,*)-
^p(z,p))2)
< ^ e x p { - ( i - i ) ^ } , and
■W)<= V ( r - v)v
^ J <
■/»(».*)«; V ( r const.,
v
)
I
4
CO(T-^)J
■ \ —) X ) 1-.i Therefore the lemma is true in this case. where c0 = (-— (iii) Now we consider the case where p(q, p) < 4e. Let .
V
A = -, T
/u =
T — V T
.
By using an argument as in the proof of case (i), Xp2{q,z) + W2(z,p)
- Xp,p2(q,p)
> Xp2{q,z) + p,p2(z,p)
- Xp,(p(q,z) + p(z,p))2
= {Mq,z)~
2
>0, we can let
PP{Z>P))
326 Let o be a point on the geodesic joining p and q such that
p{q,p) Due to p(q,z) < t and p(q,p) < 4e and the fact, which claims W2 = p2(o,z) in Euclidean space, we have W2>\p2{o,z) for a sufficiently small e > 0. Then < ^/(7J^"exp{- 8 c 2 ( T T _^p 2 (o,z)}. and
/ \
i——rexPTd2:
7 y (r — f)i/
4
< const. / . / r— exp{--—-. —p2(Q,z)\dz < const.. i | ( r - i / ) v H l 8c2 (r - vy K ' " ~
Therefore we finish the proof of Lemma A. Lemma B Let e be small enough. For any s<s > 0, c > 0 there exists a constant h(s0, c) such that, for any c > h(so,c) and t,Ti,T2 > 0 with T\t,T
nt2 A j T2t2 J / r2<2 2\ —v(pY > exp i —r~v(pY \ < exp I —j-v{qY \,
where /> = p(q,p). Proof It is equivalent to prove exp { - ^ - - ^(P)2}
exp \^f(v(qY
- « b ) 2 ) } < 1,
or - ^ - ^ P )
2
+ ^WS)2-KP)2)
From Taylor expansion of v(g)2 «(?)2 = «J(P)2 + 2w(p)«/(p)p + . . . , it follows that there exists a constant k, which does not depend on p and q, such that v(qY-v(p)2
+ p2).
327 So
- & -T-fv{v? + ^(tf <-& -
^(P)2
-
+^ ( ^ )
"f^Pf + ^(MP^P + P2
+ £r) + ^ £ 7
Therefore if we choose h(so,c) > (kso + Aksl)c, then we have
- ^ - - ^ ( P )
2
+ ^M9)2-«(P)2)
So Lemma B is true. Now we prove new estimates for H(r,q,p,t),(-^ + Ot)H(r,q,p,t), which are different from Lemma 2 and 3. Lemma 4 Choose e properly such that the supports of <j>(q,p),H(r,q,p,t) and ( ^ + Ot)H(r,q,p,t) are contained in {(T,q,p,t)\p(q,p) < e}. Then, for a sufficient small e and fixed To, s 0 > 0, there exist Co, cx such that, for any r, t with 0 < r < TO and 0 < rt < so, we have \H{r,q,p,t)\ ^CoQ^T.g^expj-^te)2}, Idr + D ^ f f ^ ^ f t i J I ^ c o ^ t + ^ Q ^ T . g . r t e x p j - ^ ^ ) 2 } . Proof The lemma is trivial due to Lemma 2, Lemma 3 and Lemma B. Now it is time to prove that K(r,q,p,t) analogously satisfies: Lemma 5 The righthand side of the expression, which defines K(r,q,p,t) in the Levi iteration in section 3, is an absolute convergent series. And for sufficiently small t and fixed To, «o > 0 there exist Co, c\ such that, for any r, t with 0 < r < T 0 and 0 < rt < so, we have \K(r,q,p,t)\
f rt2 < c o ( / r t + l)Q( C l T,g,p)exp < — — v{q)'
Proof Choose co,ci as in the Lemmas 3, 4. Let c\ > Max{/i(so,2ci),2ci,e}, and c = c(2ci,6i), where /i(so,2ci,e),c(2ci,Ci) are given as in the Lemmas B, A. We prove the following inequalities by induction on m. \Km(r,q,p,t)\
< (^Y \C\J
c{^c0c)m{^t
+
m\
\)m+l—O{clT,q,p)^v\'-T^{q) ( c\
where m > 0. For m = 0 the above inequality is true due to Lemma 4. Let
d) 2c °' \Ci,
and define
328 We are going to check the following equalities < am(V^t + l)m+1—Q(clT,q,p)exp I- — % ) 2 . ml [ Ci J Suppose the inequality is true for m, by using an inequality in Lemma 4 and the fact \Km(T,q,p,t)\
SnppoTt(K0{r,q,p,t))
C {{r,q,p,t)\p{q,p)
< e},
we have \Km+1\ = 1/ / K0(T -v,q,z,t)Km(v,z,p,t)dzdv\ J J fr ¥ = 1/ / K0{T v,q,z,t)Km{v,z,p,t)dzdv\ JO
<
Jp(z,q)<e
/
c0am(Vr~r^t
/
+ l)(^t
+ l)m+1 —rQdzdv
JO Ji>(z,q)<e
<
/
ml
c 0 a m ( v / ff + ir + 2 —rQdzdv,
/
J0 Jp{z,q)<e
m!
where Q = Q{C\{T - v),q,z)exp \
\ vt2 v(z)2 \ Q(clu,z,p)exp >\ \ ——v -^v{z)'-
-
By Lemma B we have Q =
VrQ(2cl(r-u),q,z)Q(clu,z,p)exp{-£^L-^v(zr} exp{-^v(z)2-fv(z)2}
<
VrQ(2c1(r-v),q,z)Q(c^,z,p)exp{-^L-^v(z)2} exp{-fv(z)2} - v),q,z)Q(c^,z,p)exp{-€-v(q)2}
< VTQ(2Cl(r
.
By Lemma A we also have /
Qdz < VTcQiciT, q,p)exp\——v(q)2
Jp{z,q)<e
(
C\
\. J
Therefore, tr
|iWl
<
Vm
n
/ (c0cV?)am(Vr't
m
7/1.
+ l)
q,p)exp
I
{-£v(q)2}
Tf2 C\
.
This finishes the induction. Due to the inequality Z%=o\Km(T,q,P,t)\
< (%)* cod/Ft+ l)exp{VTcoc(^s0 Q(ciT,q,p)exp{-€-v(q)2} ,
1
\—Lv(q)2\
+ l) + —dvQ(c1T,q,p)exp
•JO
= am+l(^Ft
(
2
+T0)}
I
329 the lemma is true. Lemma 6 Let R(T,q,p,t)
=
di>
Jo
JM
H(T —
v,q,z,t)K(i/,z,p,t)dz,
then, for sufficiently small e and fixed r 0 , S(, > 0, there exist c0, C\ such that, for any r, t with 0 < r < r 0 and 0 < rt < s0, we have \R{r,q,p,t)\
l~—v(q)2\.
< c0T(^/rt + l)Q(clr,g,p)exp
The proof of Lemma 6 is similar to that of Lemma 5, so we omit it. By a routine method it is easy to see that G(r,q,p,t), which is defined in the Levi iteration, i.e. G(T, q, p, t) = H(r,q, p, t) +
R(r,q,p,t),
is just the fundamental solution of the heat operator £ + □(.
4
Proof of Theorem 1
The main part of Theorem 1 is the following Theorem 2 and 3. Theorem 2 For a vector field V without degenerate zeros, let G(T,q,p,t),H(T}q,p,t) be the fundamental solution and the parametrix defined above, then we have lim)/
G(r,p,p,t)dp
= (s - l i m ) /
Proof
H(r,p,p,t)dp.
JJM M
JM
equalitj Theorem 2 is equivalent to the following equality lim) / R(r,p,p,t)dp
= :0. 1
JM
' JM By Lemma 6 l ( s - l i m ) / R{T,p,p,t)dp\ JM
<(*-lim)/ . ,.
C o T ( /
^?^1)exp(-— v(p)2} dp C
JM ^/4-KC\T f C 0 (V/TS + T ) r.
n exv\
f
{ s2
I
Let Nc be an e-neighborhood of the zero set of V, and let 6=
Mm{\Vfr)\;p$Nc},
then hm T ^o
/ —,, n exp < v(p) } dp JM-N, y/4irciT ( C\T J < limT_»0 /
n
n
exp<
6i\dp
= Q,
J
2
\
v(j>Y\dp.
330 thus \(s - Urn) /
R{r,p,p,t)dp\
^fS
< lim /
+ ]
{ - — v ( p ) 2 } dp.
,! exp
Suppose pa 6 Zero(V), choose a normal coordinate system centering at pa, and an orthonormal moving frame as in §2. Let the coordinates of p be Y = ( j / i , . . . ,yn)', that means
From V(p) =
Vi^Etip)
Vii{p)=
+ ■■■ +
vn(p)En(p)
{EjV^+VkWik
= ^(p)+°(M), it follows that Viip) = =
Vi(pa) + ^{pa)Vj + o(\y\) ««(Pa)yj + o(||/|),
so we have
(
«u(Po)
"nl(Pa)
\
:
«ln(Po)
=
••■
: ••■
+•••
«nn(Pa) y
yjB'(p,,) + . . . .
Let 7Va be a connected component in 7Ve, which contains pa, then ..
f CO(V/TS + T ) VjV, ^ T T C i T
= EPa hnv-jo / JNa
\ {
s2 CiT
—,.
2\
J
n exp 4
y/4-KCiT
< (const.) ZPa linv-^o f
( C
v(p) CXT
\ dp J
° ( / ^ l r T ) e x p ( - — Y B ' ( p a ) B ( p a ) y * } dy.
p a is non-degenerate of, i.e. the the positive symmetric matrix B*(pa)B(pa) thus the integral
is non-singular,
is bounded, so
■ / M^±^exp{-il,( P ) 2 U P = 0. The lemma is proved. Theorem 3 For a vector field V without degenerate zeros, the following holds ( s - l i m ) JMH(T,p,p,t)dp
= Y^ (det ^(cosh9(pa)
- 1))
tr e x p { s J > i ; , ( p a ) £ t £ - | P t > } .
331 Proof
By Lemma 4 \H(r,p,p,t)\
<
nexp\
v{pf
\.
For p£Nt, \H(r,p,p,t)\
< -j=nexp y/AnciT
(-— A . { CiT J
The righthand side of the above inequality goes to zero as r -> 0, therefore |(s-lim)
/
JM-N,
H(r,p,p,t)dp\
<(s-lim)/ (s-lim)
/
S2 \ dp = 0,
° „exp^
H(r,p,p,t)dp
JM
(s - lim) / +(s - lim) f JN,
= (s -lim) /
JM-N,
H(r,p,p,t)dp
J N,
= (s - lim) £
/
H{T,p,p, t)dp.
JNc
In Na, let us choose the orthonormal frame and the normal coordinates as in the Theorem 2, thus
H (r,P,P,i) = ^ i 7 r y d e t ( s ^ ) «i(p)
• exp I -2T?(VI{P),
N
+
• • •, vn(p))-
MP)
TtY.vij(p)E?E-\,
)
where 9 = 2rt\B(p)\. and choose (vi{p),..., vn(p)) = (vi{yu ...,y„),...,
vn(yu ..., yn)).
Let (wu . . . , » „ ) = V&s(vi(y),...,
vn{y))
cosh 8 — 1 61 sinh 61 '
then H(r,p,p,t)
=-^v/det(^) • e x p { - £ K + • • • + w2n) + sE«y(p)S+£;-}
Note that
det(^"--"'J) =det(fl<mi--""J) det(ffi^l) = det ( V W r a s r ) ■ det(B(p„) + ...),
.
332 so by the fact that B(pa) is non-degenerate, (d(w ,...,w )\ detfr'-' ;m^o, 1
n
(w\,.. •, wn) is a coordinate system in Na, moreover dp=
=
{1 + o(\y\))dyi ■ ■ ■ dyn
(1 + o(|y|))| det(B(p„) + °(\y\))\-ldet
(y8Syj!f^yl
dw, ■■■dwn
Therefore /
J
Na
trH(r,p,p,t)dp
- \
JN„
4>{wu ... ,wn)
—„
\Z4TTT
■ e x P { - ^ ( w i + ' ■ • + wl)} dwi ■ ■ ■ dwn, where 4>(wu ...,wn)
= ^ d e t ( a ^ ) ( l + odj/l))! det(B(p„) + o(\y\))\-1
det
tr ex
(V^V^T)" 1 P {*£%(?)£+£/} Under the semi-classical limit
trH(r,p,p,t)dp
•
= <j)(0,... ,0).
JNa
Then the lemma follows from the following computations
*(°> ■ • ■. °) = \/det{*h)\
^t(B(Pa))\-'
■
tr
«P
{sEvii(pa)E;E-}
{sEVij{pa)E?Ey}
= det ( ^ ( c o s h S - l ) ) _ 1 t r exp{sEvij(pa)E^E-}
,
in which 0 = 0(pa), EtEr = E+E~\Pa. Proof of Theorem 1 The first assertion of Theorem 1 is trivial because of Theorem 2 and Theorem 3. For Si < s2, due to p( Sl ) = = > =
(si -lim)tre- T D ' 1^(^00 tre~+ a < limt^ootre - "? 0 ' (s2-lim)tre-TD'=p(s2),
the second assertion is true. Due to tre" T D ' > dimkerD (
333 and P(s) = (s-lim)tre- T D ', the third assertion is true. The existence of Dt implies that str e~TD' does not depend on r. Using the heat equation method we compute limstre" TDt T->0
and find that it does not depend on t; this gives a proof of the fourth assertion.
5
Applications
If V is a gradient vector field of a non-degenerate Morse function / , then Vtj is symmetric and hence Dt can be restricted to the m—forms for m > 0, i.e. we have Dt : Am(M) —-> Am(M). For this new Dt, denoted by O)m\ the first three assertions of Theorem 1 still hold. In this case Witten has proved that dimker DJm' = bm, where bm is the m—th Betti number of M. Now let us compute PW(S) = U m E ( d e t v / 2 ( c o s h % a ) - l ) ) \
r
exp{s£%(pQ).E+£-U.
Without loss of generality we may assume Vij(Pa)) = M i r Then Tl
det v /2(cosh6'(p a )-l) = ]\ Ve2s*< + e"25« - 2, i=l
and J
KJ
i$J
where J is any subset of m elements in { 1 , . . . , n}, and further tr exp{s£ Vij(pa)E^E-\pJ
= Zj e x p { s ( £ m - Y,IM)}
Note that Km ^ = ( ; • * * * o, s->oo e»M [ 0, if Hi < 0. It is easy to see that, if the set {fj,\,..., /i n } contains exactly m negative numbers, in other words if the index of V is m, then
lim Ej(n^e^ni 6J e-
aw
)
334 is 1, otherwise it is zero. After checking hm
detv/2(coshfl(p„-l)) IlLi Ve2s"< + e" 2s « - 2 ———;—: = hm ———r—; = 1,
we know that Um (det ^ 2 ( c o s h 0 ( p o ) - l ) )
tr e x p { s ^ ( p * ) ^ ^ , , }
is 1 or 0, depending on the fact whether p a is a zero of index m or not. Let cm be the number of zeros of V with index m, then hm £ (det ^ 2 ( c 0 B h % o ) - l ) )
tr e x p { s £ v ^ E f E ; \ P a }
= Cm.
Thus by Theorem 1 we get an inequality
which is the Morse inequality. Using Bismut's inequalities m
m
£(-ir+*tre-^
> ^(-i)»+*6 4 ,
fc=0
k=0
we can prove the strong Morse inequalities similarly. In the remaining case where V is a vector field without degenerate zeros, we do not have □| . Let us compute liniiT, ( d e t ^ c o s h f ^ j O - l ) ) s
a
^
str
exp{*'ZlVii(pa)E?ET\Pti}.
'
When s goes to zero, det ^2(coshtf(p„)-l) = (2s)"| det(«y(p 0 ))| + °(s" +1 ), and n1 By Theorem 1 and by Proposition 3 in [6] i s t r ( 8 £ vjk(pa)E+E; j
= (2s)" • det(vjk(p)),
we get str e-TO( = lims_>0 E« (det ^ ( c o s h 6{pa) - 1)) ~ str exp{s E % {pa)EtEj P<,6zto(v)l
det
^^))l'
It gives an analytic proof of the Hopf index theorem.
\Va }
335
References [1] Bismut, J.-M., The Witten complex and the degenerate Morse inequalities, J. Diff. Geom., 23(1986), 207-240. [2] Cycon, H., Froese, R., Kirsch, W. and Simon, B., Schrodinger operator, Springer-Verlag, (1987). [3] Witten, E., Supersymmetry
and Morse theory, J. Diff. Geom., 17(1982), 661-692.
[4] Yu, YL, Index theorem and heat equation method (in Chinese), Shanghai Scientific and Technical Publishers (1996), 1(1985), 135-160 [5] Yu, YL., Local index theorem for Dirac operator, Acta Mathematica Sinica, New series, 3(1987), 152-169. [6] Yu, YL., Local index theorem for signature operators. Acta Mathematica Sinica, New series, 3(1987), 363-372. [7] Yu, YL., Semi-classical limit and Hopf theorem, Proceedings of the international con ference on manifolds and singularities, Chinese Mathematical Society-National Natural Science Foundation of China-Topology and Geometry Research Center, (1995), 113-122. Institute of Mathematics Academia Sinica Beijing 100080 P R China e-mail address: [email protected] Received January 18, 2000
336
Geometry and Topology of Submanifolds X eds. W. H. Chen et al. (pp. 336-345) © 2000 World Scientific Publishing Co.
^-invariants and the Poincare-Hopf index formula Weiping Zhang*
Abstract We present an analytic proof of the Poincare-Hopf index theorem. Our proof makes use of an old idea of Atiyah and works for the case where the isolated zeros of the vector field can be degenerate. Key words: Euler characteristic, elliptic boundary problem, spectral flow, ^-invariant. Math. Subject Classification: 58G10.
0
Introduction
Let M be an even dimensional oriented smooth closed manifold. Let V G T(TM) smooth tangent vector field on M. We assume that the singularity set
be a
B(V) := {x G M : V(x) = 0} consists of isolated points. Then for any x G B(V) one can define an integer deg y (:r), which we call the degree of V at x, as follows: let Ux be a sufficiently small open neighborhood of x such that V is nowhere zero on Ux\ {x} and that the closure of Ux is diffeomorphic to the standard closed ball in the dim M dimensional Euclidean space, then V induces a map v from 8UX, which is diffeomorphic to the standard sphere 5 d i m M - 1 ( l ) , to SdimM-l(l) in the following manner: for any y G 8UX, V(y) may be viewed as a unit vector in the Euclidean space containing Ux, which determines the point v(y) on 5 d l m M _ 1 ( l ) . We define deg y (x) to be the Brouwer degree of this induced map. Let x(-W) denote the Euler characteristic of M. The Poincare-Hopf index formula (cf. [6] Theorem 11.25) can be stated as follows. T h e o r e m 1 The following identity holds: x ( ^ 0 = Y,xeB(v) deg^(x). In this paper, we present an analytic proof of this classical result by developing an old idea of Atiyah [1] on manifolds with boundary. In doing so, we reduce the problem to a calculation of variations of ^-invariants on the spheres around the zeros of V. The main point here is that we do not deform the vector field V to make its zeros nondegenerate. Thus in particular, our proof works for the case where V has degenerate zeros. This is different from the analytic proof proposed by Witten [10]. We hope that the ideas involved This paper is in final form and no version of it will be submitted for publication elsewhere. 'Partially supported by the NSFC, MOEC and the Qiu Shi Foundation.
337 in this proof may be useful in other situations; in particular we hope this will yield a deeper analytic understanding of some of the results in the paper of Atiyah-Dupont [2], where further generalizations of the Poincare-Hopf index formula have been studied extensively. Here is a brief outline of the paper. In Sections 1 and 2, we reduce our proof to compu tations of spectral flows of Dirac type operators on spheres around zeros of V. In Section 3, we compute these spectral flows via variations of ^-invariants.
1
Splitting of the index
1.1
An analytic interpretation of x(M)
Let j b e a Riemannian metric on M. Let A"(T*M) denote the exterior algebra bundle of the cotangent bundle T*M. Let d* be the formal adjoint of the exterior differential operator d with respect to the standard L2 inner product on the space of smooth differential forms: n*(M) :=r(A*(T*M)). Let D denote the de Rham-Hodge operator defined by D = d + d' :fi*(M)-»fi'(M). Let De/o be the restriction of D to fieven/odd(M) respectively. Then De : « everl (M) -+ Qodd(M) is a first order elliptic differential operator whose formal adjoint is D0 : f2odd(M) —> J2even(M). Furthermore, by the Hodge decomposition theorem (cf. [9] Corollary III.5.6), one has (1) X (M) = indD e .
1.2
de Rham-Hodge operator on manifolds with boundary
We adopt the following notational conventions. We use the Riemannian metric g to identify the tangent bundle TM with the cotangent bundle T*M; if e e TM is a tangent vector, let e* £ T*M be the corresponding dual cotangent vector. Let ext and int denote exterior and interior multiplications respectively. Let c(e) and c(e) be the Clifford operators acting on A"(T*M) given by c(e) := ext(e') — int(e*) and c(e) := ext(e*) + int(e'). Choose S > 0 be small enough so that the balls Bs(x) :={yeM:
d(x, y) < 6} for x G B(V)
are disjoint. We choose the Riemannian metric to be flat on these balls; these balls are then isometric to the ball of radius 5 in Euclidean space. To simplify subsequent notation, we let B{x) := Bs/2(x). Let M be the closure of the complement of UX€B(V)B(X) in M: M := Closure{M \
\Jx&B{v)B{x)}.
338 We now study the elliptic boundary problems of the type of Atiyah-Patodi-Singer [3] for the de Rham-Hodge operator on M- The point here is that since we have assumed that the Riemannian metric g is flat on Ux€B(v)B(x), one has to deal with the situation where the metric near the boundary dM is not of product nature. For this, we will make use of the more refined analysis developed in the paper of Gilkey [8]. Let DM (resp. DMfi/0) be the restriction of D (resp. De/0) to M. Let e i , . . . ,edimM be an oriented orthonormal basis for TM and let VA*'T*M' be the canonical Euclidean connection on A*(T*M) lifted from the Levi-Civita connection V ™ of g. Then one has that (cf. [8] (5.2)) dimM
DM = Y, e(ei)V£*(r*M) : n*(M)|„ -> Q'[M)\M-
(2)
»=i
Let 5 be the inward unit normal vector field on dM and let / i , . . •, /dimM-i be an oriented orthonormal basis of TdM. Let Ljk = (V™e*,n), 1 < j , k < dimM — 1, be the second fundamental form of the isometric embedding dM <-*• M. Following [8] Lemma 2.2, we define the tangential operators acting on Cleven/°d
-i dimM—1
J W „ = -c(fi) E
L
»
(3)
-> Sl™/°M(M)\aM
(4)
respectively. To be more precise, for any 1 < j < dim M — 1, set cUi) = -c(n)c(/ J ) : n°™t°M(M)\dM and ^__ V ^
= V
A-(T-M)
-1 dimM—1 £ L .
t 5 ( A ) : n
even/odd
( M ) | 8 A / )
^neven/odd
( M ) | a A <
( g )
Then one verifies the following analogue of [8] Lemma 2.2(d): dimM—1
A>AW»=
E
~c{h)Vij:U^laA\M)\aM^Q,^l°Ai{M)\aM.
(6)
i=i
Let PaM,>o,e (resp. PS.M,>O,O! resp. PB.M,>O,O) be the orthogonal projection from the L2completion space of ( ^ ( A f ) ^ (resp. aoii{M)\aM, resp. QoM(M)\aM) to its sub-Hilbert space obtained from the orthogonal direct sum of nonnegative (resp. nonnegative, resp. positive) eigenspaces of DaMfi (resp. DaM,o, resp. I>a.M,o)Let (DaM,eio,PdM,>o,0fi)) be the realizations of the operators AM,e/o (resp. DaMfi) with respect to the the Atiyah-Patodi-Singer type boundary conditions given by Pe.M,>o,e/o (resp. PaM,>o,o) respectively (cf. [3] and in particular [8]). These bound ary value problems are all elliptic. Moreover, (DaM,0, Pajw,>o,o) is adjoint to (DaMfi, PsAi,>o,e) ([8] Theorem 2.3(a)). The above strategy can also be developed for each B(x), x € B(V), with similar notation.
339
1.3
A splitting formula for x(^0
We state the main result of this section as follows. Proposition 2 We have that x(M) = ind(D/K]01Pavf,>o,e)Proof. By (3), one sees that on each 8B(x), x e B(V), one has DdM,e = -DdB(x),eWe can then apply [8] Theorem 6.4(g), which generalizes the Atiyah-Patodi-Singer index theorem [3] to the case where the metric near the boundary is not of product nature, to M, M and B(x), x G B(V), to get mdDe = md{DMte,PgMiy0ie}+
ind
Y,
(DB(x),e, PaB(x),>o,e) +
i€B(V)
JZ dim(kerL>aB(l)ie). (7)
xeB(V)
Now since each ball B(x), x € B(V), is a standard ball in an Euclidean space, one sees that the operators DB(x^e/0 are the standard Dirac operators twisted by a trivial vector bundle on B(x). In view of (6) and [8] Lemma 4.1, one then sees that the induced operators DdB(x),e/o on the boundary are the standard Dirac operators twisted by trivial vector bundles with trivial twisted connections.1 If dim M > 4, then the scalar curvature on each dB(x) is positive and one uses the Lichnerowicz formula to see that there are no harmonic spinors on dB(x). If dim M = 2, one computes directly that the 'bounding' spin structure is the Mobius spin structure and hence there are no harmonic spinors on the boundary. Thus the kernels of the boundary operators are trivial, i.e., for any x £ B(V), dim(ker DdB{x)fi/o) = 0.
(8)
On the other hand, by using Green's formula (cf. [8] (2.28)) as well as the fact that the metric on B(x) is flat, one deduces easily that for any s S fieven^odd(M)|B(a!), x £ B(V), one has j.
e
/
(De/0s, De/os)dvB^
1
=-
JB(x)
(s, DdB(x)fij0s)dv9B{x) JdB(x)
/•
- -
dimM—1
(s,
6 JdB(x)
+ E L (^:{T'M)s,^:{T'M)s)dvB(x), dimM i=1
r
JB(x)
JZ
L
r
,_1
o)
where e\,..., edimM is an orthonormal basis of TB(x), and dvB(x) (resp. dvgB(x)) is the volume form on B(x) (resp. 8B{x)) induced by the Riemannian metric g. Now since the mean curvature — £ d ™ M _ 1 Ljj of the isometric embedding 8B(x) c-> B(x) is positive, one verifies directly from (9) that ind ( % ) , „ PdB(x),>o,e) = 0,
for any x £ B(V).
The proposition now follows from equations (1), (7), (8) and (10). □ ^ee Section 3.2 for a more detailed explanation.
(10)
340 Remark 3 It is important to note that the reason that i?ae(x),e/o are not equivalent to the de Rham-Hodge operators on dB(x) is due to the fact that here we are dealing with the Atiyah-Patodi-Singer type boundary problems in situations where one does not assume that the metric near the boundary is of product structure. Remark 4 One can also prove the splitting formula (7) alternatively without using the index theorem ([8] Theorem 6.4(g)) for manifolds with boundary. To this order, one deforms the Riemannian metric g near the hypersurface dM in M so that the metric near dM is of product nature. One also deforms the de Rham-Hodge operator on M to a Dirac type operator such that near dM, it is of product nature with the induced tangential operators on dM given by DgM,e/0- One then applies the splitting formula for this deformed Dirac type operator, which can be proved by using the Bojarski theorem (cf [5] Theorem 24-1), to get (7). We leave the detaib to the interested reader.
2
E u l e r c h a r a c t e r i s t i c a n d t h e s p e c t r a l flow
2.1
Review of the definition of the spectral flow
The concept of the spectral flow was introduced by Atiyah-Patodi-Singer in [4] for a curve of self-adjoint elliptic differential operators. Let D(u), 0 < u < 1, be a smooth curve of self-adjoint first order elliptic differential operators on a compact smooth manifold, then the spectral flow of the family {D(u)}oo (resp. Pi,>o) be the orthogonal projection from the L2-completion space of the domain of D(0) (resp. D(l)) to its sub-Hilbert space obtained from the orthogonal direct sum of nonnegative eigenspaces of D(0) (resp. D(l)). Then the operator T(P0,>o, Pi,>o) = Po,>oPi,>o : Im(Pi,>0) -> Im(P0,>0)
(11)
is a Fredholm operator. Furthermore, by a result of Dai and Zhang [7] Theorem 1.4, one has sf{D{u), 0 < u < 1} = indT(P 0i > 0 , Pi,>„).
2.2
(12)
Euler characteristic and the spectral flow
We normalize V to assume that \V\g = l Set
on
M.
_ DdM,e = c(V)DdMt0c(V) : ne™(M)\dM -> ne™(M)\dM
(13)
From equations (3)-(6) and (13), one deduces that ^^
dimM-l
DdM,e = DaMte + C(V) £
c(/,-)c(V£MV).
(14)
341 We use this equation to see that the boundary problem (DM,e, c(V)PaM,>o,0c(V)) is elliptic. Moreover, as was already observed in [1] (2.2), the following operator is a 0th order operator: A(V) = DM, 0 + c(V)DM,ec(V) : a°M(M)
-► aoM(M).
One uses equation (8) and [8] Theorem 2.3(a) to see that: ind {DMfi, c(V)PgM,>0iOc(V)) = ind (c(V)DMfic(V), PaM,>o,o) = ind {-DMja + A(V), P9M,>o,o) = ind (DMi0 - A(V), PaM,>o,o) = ind (DM,o, PaM,>o,o) = -ind (DMto,e)-
(15)
Clearly, PdM,>o,e '■= c(V)PgM,>o,oC(V) is the orthogonal projection mapping from the L2-completion space of the domain of DgM,e to its sub-Hilbert space obtained from the orthogonal direct sum of nonnegative eigenspaces of DgM,e- Thus, from (15) and (11) one has ind {DMfi,
PaM,>0,e) = r ( i n d (DM,e, =
PdM,>0,e) ~ ind (DM,e,
^indT(P8^1?>0)e, PaM,>o,e),
PdM,>0,e))
(16)
where the last equality follows from the variation formula for Dirac type operators with global elliptic boundary condition (cf. [5] Proposition 2.14).2 Now for any u 6 [0,1], set DdM,e(u)
= (1 ~ u)DaMie
+ uDdMfi.
(17)
Then {DdMte(u)}0
= -si{DaMie(u)
: 0 < u < 1}.
For 0 < u < 1, let -Dae(x),e(w) be the restrictions of -DaM>e(u) to each boundary sphere around x 6 B(V). As dM = — U i e B dB(x), where one also takes account the orientations, one gets the following formula, which is the main result of this section, X{M) = - \ Z
£
s f { D 9 B ( x ) » : 0 < u < 1}.
(18)
X€B(V)
Remark 5 The deduction (15) has been inspired by an idea of Atiyah [1] which was used to show that the Euler characteristic of a closed manifold admitting a nowhere zero vector field is zero. 2 The book [5] only deals with the situation of product nature near the boundary. However, one can use deformations as was indicated in Remark 4 to reduce our problem to the case of product nature near the boundary.
342
3
A computation of the spectral flow
In this section, we compute the spectralflowsappearing in the right hand side of (18) through variations of ^-invariants.
3.1
^-invariants and spectral flow
For any u € [0,1], following [3], let rj(DgB(x)Au)i s) be the ^-function of DgB{x)Au) defined for s e C with Re (s) > dim M + 1, v{DdB(x)Au)>s)=
E
-fTTT-
A€Spec(DaBW,«(u))\{0} l A l
It can be extended to a meromorphic function on C which is holomorphic at s = 0 (cf. [4]). The value of rj{DgB(x),e{u)->s) at s = 0, denoted by 77(£>8B(x),e(u))> is the ^-invariant of DdB{x),e{v) in the sense of Atiyah, Patodi and Singer [3]. Let rj{DsB{x)Au)) be the reduced 77-invariant of DgB(x)te(u), which was also defined in [3]: -ir> 1 \\ V{L>aB(x)Au)) —
dim keTD
9B(x)Au))+v{DgB(x)Au)) 2 '
(
By (14) and (17), one sees that for any u £ [0,1], ^DgB^x)fi{u) is a bounded operator. By standard results for heat kernel asymptotics, one has the following asymptotic expansion as t -> 0 + , n^DaB{x)Au)
ex P (-t(D aBW , e («)) 2 )] = ^
+ ■ ■ • + °-0- +
Off"),
where k = dim M — 1 and C-k/2, • • • > C-1/2 are smooth functions of u € [0,1]. The following well-known result (cf. [11] Proposition 3.6) illustrates the relations between spectral flow and variations of reduced ^-invariants. Proposition 6 For any s e [0,1], one has sf{A>s(*),e("), 0
=j
-j^du
+ rj(DdB(x)As)) ~ ??(A)S(x),e(0)).
Now, from (3), (13) and (17), one verifies that DSBWM)
= c(V)D9Mt0c(V)
= -c(n)c(V)D 8A1 , e (0)c(y)c(ri).
Thus, one finds that rj{D9B{x)Al)) =5?(A)B(x),e(0)). From Proposition 6 and equation (19), one gets sf{DBB{x)te(u), 0 < u < 1} = f1 ^jg-du. J0
V7T
(19)
(20)
343
3.2
Evaluation of the spectral flow
We compute in this subsection the right hand side of (20). As we have noted in the proof of Proposition 2, the operator DgB(x),e is the standard Dirac operator twisted by a trivial vector bundle. We first make this more precise. Let S(TB(x)) := S+(TB[x)) © S_(TB(x)) be the Z2-graded bundle of spinors over B(x) associated to g\B(x) (which is the standard Euclidean metric). We have Aeven(T*B(a;)) = S+(TB(x)) ® S+{TB(x)) © S„(TB{x)) ® S_(TB(x)), S°M{T'B{x)) = S+(TB{x)) ® S.(TB{x)) © S_(TB(z)) ® S+{TB{x)). In order to avoid confusion, we will use the symbol ' to indicate the twisted spinor bundles (That is, the second factors in the tensor products in the right hand sides of the above equations). Then one sees directly that S'±(TB(x)) are trivial vector bundles on which V ™ lifts to the trivial flat connections. Let r denote the Z2-grading operator of the splitting S'(TB{x}) = S'+{TB{x)) © S'_(TB(x)), that is, T\S'±(TB(X)) = ±Id. Then one has i
A
dim M 2
c(ei)---c(e dimM ). (21) -1 From equation (6) and [8] Lemma 2.2, one gets the identification of differential operators -c(n)D 8e ( x ) !e c(n) = rDgB^0 as maps from T(S+(TB(x)) ® S'_(TB(x)))\eB{x) to T(S+{TB{x)) ® S'_(TB(x)))\aB{x).
(22)
Now denote the canonical Dirac operator on dB(x) with twisted coefficient the trivial vector bundle S'(TB(x))\gB^ by DaB{x) : T(S+{TB(x)) ® S'(TB(x)))\aB{x)
-> T(S+(TB(x)) ®
S'(TB(x)))\aB{x).
Set for any 0 < u < 1 that Aw<«)(«) = (1 - u)DaB[x) + uc(V)DaB{x)c(V).
(23)
From equations (6), (13), (17), (22), (23^and by proceeding as in [8] Sect. 3, one sees that the two families {DsB(i),e(«)}o
(25)
344 where Trs is the notation of the supertrace of operators on S'(TB(x)) with respect to the Z2-grading operator r. Now by (21), one verifies that ' M E c(V)c(VTf>MV)f;rk+l}
= 0 if k* \ - 1,
(26)
J=l
while
Trs[(E e(v)5(v^y)/;)- 1 ] = ^ r j ^ 1 * A ' ' " A |
B
F* A (VlMV)'
^
A • • • A ( V ^ V)",
(27)
where JB V A (V£MV)* A • • • A (VjJ^ V)* is the function on 3B(x) such that V*A(Vj l M yrA---A(Vj n M i n* = / ' A - - - A / : _ 1 A ( - n r / B y * A ( V ™ y ) * A - - - A ( V ™ V ) * . (28) Let v : 9B(a;) —> 5" 1(1) denote the canonical map induced by Vlas^), let u be the volume form on S n - 1 (l). Then by (28) one verifies directly that
/* A • • • A fn-i f v A (vThMvy A • • • A (vi^vy = »*w.
(29)
From (20), (24)-(27) and (29), one deduces that rf{D«K,),(«),
0 < u < 1} = - - L ^
(2 _ i ) ! t ,. w .
(30)
On the other hand, since the volume of S"-1^) equals to 27r" / ' 2 /(| —1)!, from the standard differential geometric interpretation of the Brouwer degree, one has (S-1V t 2 W 2 JaB(x)
& v w
which, together with (30), gives that sf{D88(*),e(u),0 < u < 1} = -2deSv(x).
(31)
The Poincare-Hopf index formula then follows from (18) and (31). □
A c k n o w l e d g e m e n t s The author is grateful to the referee for his critical reading and very helpful suggestions.
345
References [1] M. F. Atiyah, Vector fields on manifolds. Arbeitsgemeinschaft fur Forschung des Landes Nordrhein-Westfalen, Diisseldorf 1969, 200 (1970), 7-24. [2] M. F. Atiyah and J. L. Dupont, Vector fields with finite singularities. Ada Math. 128 (1972), 1-40. [3] M. F. Atiyah, V. K. Patodi and I. M. Singer, Spectral asymmetry and Riemannian geometry I. Math. Proc. Cambridge Philos. Soc. 77 (1975), 43-69. [4] M. F. Atiyah, V. K. Patodi and I. M. Singer, Spectral asymmetry and Riemannian geometry III. Math. Proc. Cambridge Philos. Soc. 79 (1976), 71-99. [5] B. Booss and K. P. Wojciechowski, Elliptic Boundary Problems for Dirac Operators. Birkhauser, 1993. [6] R. Bott and L. Tu, Differential Forms in Algebraic Topology. Graduate Text in Math. 82, Springer-Verlag, 1982. [7] X. Dai and W. Zhang, Higher spectral flow. J. Fund. Anal. 157 (1998), 432-469. [8] P. B. Gilkey, On the index of geometrical operators for Riemannian manifolds with boundary. Adv. in Math. 102 (1993), 129-183. [9] H. B. Lawson and M.-L. Michelsohn, Spin Geometry. Princeton Univ. Press, 1989. [10] E. Witten, Supersymmetry and Morse theory. J. Diff. Geom. 17 (1982), 661-692. [11] W. Zhang, Analytic and topological invariants associated to nowhere zero vector fields. Pacific J. Math. 187 (1999), 379-398. Nankai Institute of Mathematics, Nankai University, Tianjin 300071, P. R. China E-mail: [email protected] Received January 29, 2000.