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!
+ 2<x,P.PW) ()> 1
1
= 4 <x,x> S..1) Thus 0 is horizontally conformal with dilation A given by A2 (x) = 4Ix 12, for all
xERm. Write P1 in matrix form as (Pb) 1 a,b-1,...,m , for each i = 1, ... , n.. Then ,
fi (x) = E a,b
P.ab
xaxb
Thus
Of.1 = Ea P.aa 1 147
Since P.2 = 1, the eigenvalues of P. are +1 and -1. If the eigenspaces of +1,-1 are E+(Pand E_(P.) respectively, then since P.P. + P.P.i = 0, i j, we see that i l i i l Thus dim E + (P.t ) P interchanges the eigenspaces E+(Pi ) and E- ( Pi of P.. P. i dim E (P.) = 1, whence
E Paa = 0 a
1
LI and f.1 is harmonic for each i . 1, ... , n . Examples of harmonic morphisms defined by Clifford systems, include as special
cases Examples 8.2.1. and 8.2.2. From Theorem 8.1.5, to each such example we can associate a harmonic Riemannian submersion onto a sphere. Amongst the family of isoparametric functions (8.4. 1), some are inhomogeneous ; furthermore in some cases the focal varieties are inhomogeneous [ 17 1. Thus we can associate harmonic Riemannian submersions from compact inhomogeneous spaces
onto spheres. We can modify the proof of Theorem 8.1.5 to obtain the following.
Theorem 8.4.2 Let 0: Rm -. Rn be a harmonic morphism defined by homogeneous polynomials of degree p, with dilation a given by X2 (x) = p2 Ix 12p-2. Let F: Em -- E be defined by F(x) I 0(x)12 , and write f = F ISm-1 , then 0 IM : Mc -- c2 Sn-1 is a harmonic homothetic submersion, where Mc = f-1(c) c and c # 0.
Proof Since c X 0, the projection p c : Mc -- I', which maps down normal geodesics,
is well-defined. Therefore 0 Mr -'c2. Sn-1 is onto, since it factors through IM : : c 2 S n-1 - Sn-1 ; i. e. the folpc , O I :F -- Sn-1 and the projection map c Jr 1
c2
lowing diagram commutes:
0
i 0 148
Ir
In order to prove the theorem, the crucial point to observe is that the proof of Theorem 8.1.5 will go through provided that the following claim is true.
For x c M c
Claim 8.4.3
(VA2 (x)) lies in the plane spanned by VX2 (x) and
is the unit normal vector field to M c For then the mean curvature of the fibre of 0 over O (x) would be perpendicular to M at x, and the rest of the proof goes through as before. x
where
c
Proof of Claim 8.4.3
Write 0 = (O1,
x c- Mc is spanned by f V Oi (X))
F=
... , n'
i=1
0n) , then the horizontal space through Now
n
2 E 0 k=l k
so
n
VF = 2 E
k=1
0k 00k
therefore
< VF,V0i > = 201 a2 Also X2 (x) = p2 Ix V X2
I2p-2
, therefore
= 2p2 (p- 1) Ix 12p-4
.
k
xk Vxk
Now k
< xk Vxk, V O > = p. Oi
by the homogeneity of 0i. Thus <
VA2, V Oi > =
2 p3(p - 1)0 .
Hence the horizontal projection of Vat is proportional to the horizontal . projectiop of VF. Since is proportional to the projection of VF onto the claim is proved. on Sm-1
TSm-1
Example 8.4.4 Consider Example 8.2.1 , Case (2). Here 0: R8 0 =(01,02) is given by 01(x,Y) =
Ix12
-
R2,
Iy12
02(X,Y) = 2 <x,Y >
,
149
where x,y are quaternions. The level hypersurfaces are parametrized by the sets Ms = { ei9 (toss. x + isins.y) ; 0 E [0,2 Tl], (x,y) E S3,21 , where s E [0, IT/4 ]. If z E M s then z can be expressed in the form
z = eiO(coss.x + isins.y) (cos0toss.x - sin6sins. y) + i(sin 0toss.x +cos0sins. y) Then
01(z) = Ix12- Iy"12 =
cost B costs +sin2B sin2s - sing 0 cos2s - cost 0 sin2s
= cos28.cos2s
,
and
02 (z) = 2 (cos B toss sin 0 toss - sin O sins cos g sins)
= sin 2 0 . cos 2 s . Thus 0 IM : Ms s
2s. S1 ; 0 (z) = cos 2s. a
2i6
.
Theorem 8.4.2 allows us to construct many equivariant maps x : Sm-1 -. Sn associated to a harmonic polynomial morphism 0: Rm - Rn with dilation A given by
X2
(x) = p2 Ix 12p-2 . We simply define X by
X (z) = (cos2a(s)(OIIIo p(z)), sin2 a(s))
,
where F = 10 12 is parametrized such that F = cosps, F= s-1(0), p is the projection down normal geodesics onto I' and a (0) = 0, a(TT/p) = TT/2 . In particular all the harmonic morphisms associated to a Clifford system give such equivariant maps, and hence we have examples of equivariant maps between spheres
with respect to isoparametric functions, where one of the isoparametric functions has non-homogeneous hypersurfaces. This then justifies our use of isoparametric hypersurfaces as opposed to homogeneous hypersurfaces. Furthermore, when the same isoparametric function Rives; rise to two distinct harmonic Riemannian submersions via Theorem 8.1.5, we expect tq be able to construct examples of equivariant maps similar to Example 5.3.4. Indeed the two Riemannian submersions of Example 8.2.1 given by Cases (2) and (3) give rise to 150
Example 5.3.4. Similarly Cases (1) and (4) of Example 8.2.1 generate the Hopf map from S7 to S4. In fact, suppose 0: Rm -- Rn is a harmonic polynomial morphism, satisfying
m - 2 = p(n-2) where p is the degree of the homogeneous polynomials defining position 8. 1. 1 we see that 10 (x) 12 =
0.
Then from Pro-
Ix 12p
for all x E Rm. Let 0 = (01, ..., 0n), and define harmonic polynomial morphisms p and a by
p = (01, ..., 0p) a = (0p+1, ... , 0n) Call such p and a complementary. Let Ms be a level hypersurface of the isoparametric function f defined by f(x) = Ip(x) 12 IS m_1 ,
1=
Ip12 +
IaI2
for each x E SThen since ,
must also be a level hypersurface of the isoparametric function g defined by g (x) = lc(x)12 I Sm-1 , for all x E S m-1. From Theorem 8.4.2, we obtain harmonic homothetic submersions p INl: Ms - a(s) SP-1, c IM : Ms -- b(s) Sq-1 M
s
s
s
p +q = n, for some functions a(s) and b(s). Since 1 = Ip 12 + 1x12 , we have a(s)2 + b(s)2 = 1, so we can choose a(s) = cos ps and b(s) = sin ps (by writing Ms = f 1(cos2ps)). Given two harmonic polynomial maps g 1: S p-1 - S r-1 and g2 : Sq-1 - S s-1, we can now define an equivariant map from S m-1 to S r+s-1 as follows. Let x E Sm-1, then( p (x) , a (x)) E SP - 1 * S q-1 ; we then compose with g1 * g2 , to obtain the point gl *
g2(p(x),(#(x)) E
Sr-1 * Ss-1 = Sr+s-1. The map so defined is clearly equi-
variant since the map of level hypersurfaces is harmonic of constant energy density
by Theorem 8.4.2. The Smith maps of Section 1.3 can be seen to arise in this way as follows. Consider 151
the harmonic polynomial morphism 0: Rm - Rm
given by
0(x1,...,xm) _ (x1, ...,xm) . Define the complementary maps o and a by
p (xl, ... ,xm) = (xl, ... , xp) and
c(x
,
1
...,xm
(x
p+1
, ...,x m ).
Then f (x) = Ip (x) 12 IS m_ 1 is isoparametric of degree 2. persurface given by f(x) = cos2 s. Then g(x) = Ir (x) 12
Let Ms be the level hy-
is given by g(x) = sin2s on M. For x ESm-1 ; (p (x), r(x) C SP-1 Sq-1, p + q = m. We now compose with harmonic polynomial maps g 1: S p-1 -. S r-1 and g2 : Sq-1 -. S 3-1 as above, to obtain the Smith map from Sm-1 to Sn-1 , r + s = n.
152
ISm_1
9 Deformations of metrics
9. 1
Deformations of the metric for harmonic morphisms
Let 0: (M , g) -- (N , h) be a horizontally conformal map between Riemannian manifolds with dilation X. Let U C M be an open set upon which d0 is non-zero. Let
x e U, then for horizontal vectors X, Y e`'x , we have 2g(X,Y) = 0*h(X,Y) . Thus we can decompose g as
g = (1/X).0 *h+k ,
(9.1.1)
over U, where (I/X2 ) 0 * h represents the horizontal part of the metric, and k the vertical part. The stress-energy tensor of 0 is given by
S0 = e(0)g- 0*h anA2.g - X2(g-k)
2'(n-2)X2.g+A2.k
(9.1.2)
,
where n = dim N. Thus
V S0 = '2 (n - 2).d(A2) + V *(A2k)
.
(9.1.3)
We therefore have
If 0 : (M,g) - (N.h) is horizontally conformal with dilation X and 0 is a submersion almost everywhere, then 0 is harmonic (and so a harmonic morphism) if and only if Lemma 9. 1. 1
21(2 - n). d X2 = V*(h2k)
,
(9.1.4)
where k represents the fibre metric (where defined) (assume both sides of (9.1.4) are zero when X = 0). 153
Define a new metric g on M by
g = (1/A2 c2).¢*h + (1/p2).k
(9.1.5)
,
where o2, p2 : M - R are smooth functions. The new metric g may not be welldefined everywhere if c2, p2 have zeros , and we remove such points to obtain the Riemannian manifold (ii , g ). Denote by V the associated Levi-Civita connection. The map 0: (M,g) - (N,h) induces a new map -0: (M,g) -- (N,h). If -0 is a sub-
mersion almost everywhere then, from Lemma 9.1.1,
9
is a harmonic morphism
if and only if
2'(2 - n).d(A2o2) _
*(A2(T 2.k/p2)
(9.1.6)
.
Our aim is to reformulate and solve equation (9.1.6), given that equation (9.1.4) is satisfied. The connection coefficients of p are described in terms of Proposition 9. 1.2 those of V by the following formulae. Use the following ranges of indices:
1 < i,j, ... < n = dim N; n+1 < r,s, ... < m = dim M; 1 < a,b, ... < m ; and let ( e a)1< a< m
=
(ei, e
r ) denote a local orthonormal basis with respect
g over a subset U of M where d0 # 0', and ei ,e r are horizontal, vertical, for each i , r respectively. Let 06a (e it er (cei , p e ) denote the corresponding r orthonormal basis with respect tog over U . Then to
g(et,V e es) = Pg(et,De e s ) + iP(g(et,dP(er)es) + g(esdp(et)er)
r
r
- g(er,dp(es)et ))
(9.1.7)
;
g(ei,V er es) = c2g(ei, a es) + (o2/2p)g(es,dp(ei)er) ;
(9.1.8)
r
g(ek, e e.) + i
)
o(g(ek,do(ei)e,) + g(e,,dc(ek)ei) I
)
- g(ei,d o(ej)ek)); Z(g(er,(a2+p2)
r a iej) - g(e,(a2
(9.1.9)
- p2) a ei) ]
+(o2/o)g(ej,da(er)ei)); 154
(9.1.10)
g(er, Veies) = p cg(er, V e s
+
s ag(e,dp(ei) e) r
(9.1.11)
i
g(el,Verei)= oag(ej, ere.) + za(1-(a2/P))g(er, [ei,e11) + (p/a)g(e,,da(er)ei));
g(e,V
r
(9.1.12)
-(p/u)g(ei,ve es)
(9.1.13)
r
Proof We use the fundamental formula for an orthonormal frame X, Y, Z E (see for example [ 14 1) :
g(Z,VXY) = a(g(Z, IX,Y1) + g(Y, IZ,X1) - g(X, IY,Z1))
.
TM
(9.1.14)
Similarly
g(Z,79-XY) = 2(g(Z, [X, Y1) + 9(X, [Z,X1) -g(X,
(9.1.15)
Let Y' denote Lie derivation. First of all we prove equation (9.1.7). From equation (9.1.15) , letting k, k denote the vertical part of the metric g, j respectively,
g (et, ve s) _ r
es) + k(esY er) - k (ergs et) )
z (k (et,
t
r
s
21(Pk(et,9e es) + k(et,dp(er)es) + ... ) r z(Pg(et, a es) + g(et,dP(er)es + ... ) r On the other hand
g(et, ve es) = k (e t, Ve es) r r _ (1/P)k(e,Vs) t r (1/p)g(et,ve es
r
Therefore
g(et,De es) = P2g(etV es) + iP(g(et,dP(er)es) + ...
r
)
.
r
The proof of equation (9.1.9 ) is similar. We prove one more, say equation (9.1.10) ; 155
the others use similar arguments. From equation (9.1.15), and writing H,H for the horizontal part of the metrics g,g respectively
gr,vele)) = z(k(er, ele)
ereil - H(ei,eer))
+ H(e)y
((1/P)k(er,-
e
ae.) +(o/a)H(e., a (a ei) r - H(ei, e. per)) l
21((0
2
/P)k(er, r e.) + PH(e., a e.) r
i
- p H(ei, a er) + (o/a) H(e,, do(er) ei)) . l
Now
H(e., 2' e - V er e r e.) i - H(ei,.e e r ) = g(ej, V er i e.
)
,
l
- g(ei, Ve,e r - v e.) er 7
l
= g(ej, a er) - g(ei, a er) = g(e , V e. + V e.).
Thus
g(er,
r
2
z (g(er, (a /p) (V
ei 1
-
ej
I
e ei
e ei )) + g(er,P (V V
+ V
j
g( ei . ,
d o(er )e.))
But
g (er, vale;) = k (er, vale; ) = (1/P) k(er, ve, e. ) = (1/P) g(er' V_ e. i
)
,
giving equation (9.1.10) We now compare V ; k with V * k
Lemma 9.1.3
156
Using the notations of Proposition 9.1.2
.
J
))
Vk(ej) = P2 V*k(ej) + AP(m-n)dp(ej)
(9.1.16)
for all j = 1, ... , n. Proof dices)
The divergence with respect tog is given by (summing over repeated in-
V k(e.) = eak(e a,e.) - k(V_eaaa,e,)) - MeaV- e.) ea )
)
- k(ea, Ve e
a)
(1/a) (k(ea, V-aej) - a k (ea, d a (ea) a ,) ) l
(p/a) k(ea, Ve ej ) a
(p/a) g(er, Ve ej)
r
(p2la2)(a2g(ej,Ve er) + (a2 /2p)g(er)dp(ej)er ))
r
,
from equations (9.1.11) and (9.1.13), =
P2g(ej , Ve e r + i P(m-n) dp(e.)
r
.
On the other hand V * k(e.)
k (e a,
g(er,
Ve
V
r
ej )
e.)
g(ej, erer
Lemma 9.1.4
Using the notations of Proposition 9. 1.2;
V *k(es) = P2dp(es) - zn(p/c) (da(es) + p2V*k(es)
(9.1.17)
for all s = n+l, ... , m. Proof
The divergence with respect tog is given by (summing over repeated indices) 157
V; k(e s ) = e ak06a ,e s ) - k(V_ eaaa ,e s ) - k(e a , V_ eaes
= per(ok(er,es)) - g(ve ei,es) - g(ve r, es ) r
i
- pg(e
ve es ) r
pdp(es) - p2g(es, V ei) - (P2/2 a) g(ei,do(es)ei)
ei
-p2 g(es, e er) - zPg(erdp(es)er) r
- P((1/P)g(er,ve e s ) - g(e r dp(e r )es))
r
where we have used equations (9.1.7) and (9. 1. 10) ;
pdp(es) - p 2 g(es, e ei) - n(P 2 /2 o)d a(es) - P 2 g(es, a er i r
a e s ) - pd P(e - z(m - n)pdp(e s - p2g(e, rp r + 4(m - n)Pdo(es) + p2dp(es) using equation (9.1.7) again ;
o2dp(es) - 2n(p2/o)da(es) - p2(g(es, a ei) 1
r e e s)) s e e r +g(e, + g(e,v
r
r
.
LI
Theorem 9.1.5 Assume equation (9.1.4) is satisfied, i.e. 0 is a harmonic (N,h) is a harmonic morphism if and only if, morphism. Then using the notations of Proposition 9. 1.2,
(2-n)d(a2)(e.)
=
(2 - n)d(X2a2)(es)
z(o /P2) (m-n)d(o2)(e.) =
2P2d(a2x2/P2)(es) +02A2 (2dp(es)
- 21 n(1/a2)d(a2)(es) - n(1/X2)dX2(es))
158
(9.1.18)
(9.1.19)
if and only if
(2 - n)d(log a2)(ej) = 2(m - n)d(logp2)(e.)
(9.1.20)
n)d(logX2o2)(es)
(2 -
=
2d(log(a a2/P2)(es) +2dp(es) - 2nd(log a2)(es) - nd(logX2) (es)
(9.1.21)
,
for all j = 1, ..., n and all s = n + 1, ..., m.
Remark 9.1.6 It may be more natural to write (9. 1.18) and (9.1.19) become
v2
=
a A2 , whence equations
(2 - n)(d(logµ2) - d(loga2))(e.) = (m - n)dp(ej)
(9.1.22 )
(2-n)d(logp2)(es) = 2d(log (p2/P2))(es) +2dp(es) - nd(log(u/X))(es) - nd(logX2)(es) for all j = 1 ,
,
(9.1.23 )
..., n and s = n + 1, ..., m.
Proof of Theorem 9.1.5 O*(A2k) =
First note that (summing over repeated indices)
dX2(e)k(e
V*((X2a2/P2)k)
=
+
X2 V*k
(9.1.24)
,
d(a2X2/P2)(er)k(er) +(a A2/P2) v*k
.
(9.1.25)
Let equation (9.1.6) act on e.; ) 22 22
2(2-n)d(A a )(ej) _ (a x /p2 )V k(ej) (a a2 /P2)(P2V*k(e.) + 20 (m - n)dp(e.))
from Lemma 9.1.3, a22(2 - n)d7X2(ej) +(o a2/P)12(m - n)dp(ej) from equation (9.1.4), giving equation (9.1.18) .
Now let equation (9.1.6) act ones, to give 2(2 - n)d(X2a2)(es) = d(a2A2/P2)(er)k(er,es) + (o2 A2/P2) V*k(es) =
p2d(o A2/P2)(es) +(a A2/P2) (P2dp(e8)
Zn(P2/a)da(es) +p2p*k (es)) 159
Now
V*k(es) _ (1/A2)V*(A2k)(es) - (1/A2)dX2(es)
'(2-00A 2)dA2(es) - (1/a2)dX2(es Zn(1/A2)dX2(es)
,
giving equation (9.1.19) .
Remark 9.1.7 If we put a2 = p2 = 1, then equations (9. 1. 18) and (9. 1.19) are satisfied - similarly, if a2 and p2 are both constant then the equations are also satisfied.
Remark 9.1.18 There is a striking analogy of the above methods with the classical notion of a Backlund transformation. There one has a hyperbolic surface M in IIR3, which is parametrized by a function a: M - R. This function is in fact the angle between the asymptotic coordinates, and the Codazzi equation for M is equivalent to a satisfying a certain second order equation called the Sine-Gordon equation. Conversely to each solution of the Sine-Gordon equation one can construct a hyperbolic surface. The idea of Backlund was to write down a first order equation in two variables a and a, such that if a is a solution of the Sine-Gordon equation, and a satisfies the first order equation, then a is also a solution of the Sine-Gordon equation. We view Backlund's idea in a more general context as the following fundamental principal:
(i) we are given a second order problem parametrized by a set of functions
(a,p,...), (ii) there is a set of first order equations in two sets of parameters; (a,Q,... ) and (a, p , ... ) , which associates to a solution (a , p , ...) of (i) another solution
(a, p., ...) of (i) In the context of harmonic morphisms, the parameter is the dilation A 9.2
Examples
Given a particular harmonic morphism 0: (M,g) -. (N,h), we attempt to find non-trivial solutions to equations (9.1.18) and (9.1.19) . We also consider instances when there are no non-trivial solutions. 160
Example 9.2.1 If n > 2 and p is constant, then equation (9. 1.18) implies that V a2 is vertical. Example 9.2.2 If p is constant and V A2 , Vat are both vertical, then equation (9.1.18) is satisfied. Equation (9.1.19) now becomes 2
d(X2a2) (e s ) = o x do(es) + a2dx2(es)
,
for all s = n+1, ... , m. This is satisfied if and only if 2 d a(e
s
= d o(e s
)
for all s = n + 1, ... , m ; if and only if
do = 0
,
i.e. the function a is constant. Example 9.2.3 If o = 1, n = 2, VU2 is horizontal and VA 2 is vertical, then equations (9. 1.18) and (9.1.19) are both satisfied. For example, let 0: R3\{0{ -. S2 be defined by 0 (x) = x/ Ix 1, for all x s R3 \ { 01, . Then if o : S2 -. R is any smooth function whidi does not take on the value 0 E R ; o : R3 \ { 0; -- I R, e(x) = -5(0(x)) has
the property that Dal is horizontal. We thus obtain a new harmonic morphism
fd: R3\{01 _ S2 .
Example 9.2.4 Let 0: S3 _ S be the Hopf fibration; so x2 = 1. Put p = 1, then equation (9.1.19) is satisfied if V c2 is horizontal. For example, if a : S2 _ R is smooth and does not take on the value zero; define a: S3 R by a(x) = a (0(x) ), for all x e S3 , then V a2 is horizontal. Example 9.2.5 Let 0: R4 - R3 be the Hopf map of Example 7.2. 1. Then 0 is a harmonic morphism with dilation A given by X2 (x) = 4 Ix 12 , for all x e R4. The VA2 is fibres of 0 consist of great circles of Euclidean spheres of R4, and hence perpendicular to the fibres and hence horizontal. Equation (9. 1.20) becomes
-d(log a2)(e.) =
d(log p2)(e.)
for all j = 1, ... n . This is solved if a-2 = constant x p .
If we choose p2 such
that V p2 is horizontal - for example p2 (x) = 0 ( Ix I2 ) for some function P. then 161
equation (9. 1.21) is also satisfied. More generally we have the following.
Let 0: R4 - R3 be the Hopf map of Example 9.2.5, and let Tip : Rp - R 4, p > 4, be the projection map. Define x p : Rp - R3 , by x p = 4 2 0 o TI4 . Then x p is a harmonic morphism with dilation A given by X (x) _ 4 I T14 (x )12 , for all x E Rp . Thus 0X2 is horizontal. Equation (9. 1.20) becomes Example 9.2.6
- d(log a2)(e.)
=
2'(p-3)d(Iogp2)(e.)
for all j = 1, ... , n. This is satisfied if a-2
= constant x p
(p-3)
If we choose a2 such that V a2 is horizontal, for instance a2 (x) is a function of Ti (x )12 , then equation (9. 1.21) is also satisfied. I
4
Remark 9.2.7 When V a2 is horizontal, then a2 16 _ 1(y) = constant, for all y E N. Thus a2 can be thought of as a function on N, and we can view the change from the map 0 to the map 0 as changing the metric h on N to h = h/a2 . That is, equation (9.1.5) can be seen as
g = (1/X2)0*h+(1/P2)k= 6*h(a2X2) +(1/P2)k. In this way, by suitable choices of the functions p2 and a , it may be possible, by removing and adding certain points, to change the topology of both M and N. 9.3. Deformations of metrics for equivariant maps Suppose 0: M -- N is equivariant with respect to the isoparametric functions
s: M - R and t: N -- E, where we suppose M and N are space forms. Using the notations of Chapter 4, we consider the map 0
S't
: Ms
Nt
-
between level hypersurfaces of s and t. Away from the focal varieties we can express
the metric g of M as
g = ds2 +
gs
'
where gs is the induced metric on Ms. Similarly on N, we can write h as
h = d t2 + ht 162
,
where ht is the induced metric on Nt There exist four reasonable kinds of deformations of the metrics g and h which we can consider. (i) Define the new metric g on M* by
(1/µ (s)) ds2 + gs
(9.3.1)
,
wherep(s) is smooth and positive and such that the resulting metric g extends smoothly across the focal varieties. (ii) Define the new metric h on N* by h
= (1/v(t) )dt2 + ht
(9.3.2)
where v(t) is smooth and positive and such that the resulting metric h extends smoothly across the focal varieties. (iii) If Ms has principal curvatures A1(s) , ..., Ap(s) with corresponding eigen-
spaces S1, ... , Sp, then we can express gs as
gs = 91(s) + g2(s)
+
...
+gp(s)
,
where, for each x c Ms, and for all X,Y C T Ms, then X
gi(s)x(X,Y) =
gs(X,Y), if X,Y C Si(x) 0
,
if either X C S. (x), j
or
Y C S. (x), j
i i
Define a new metric g on M*by g = d s2 + gs, where
gs = (1/a1(s)2 )g1(s) +
where al ,
.
+ (1/ap(s)2)gp (s)
,
(9.3.3)
...
, ap are smooth functions, which are chosen such that the resulting metric g extends smoothly across the focal varieties. Essentially, if M is a sphere, we are changing the sizes of the principal curvature small spheres of Ms by a factor depending on S.
(iv) We perform a similar construction to case (iii) for the level hypersurface Nt, by expressing ht in its principal curvature components as
ht = h1+ ...
+ hq 163
and defining
ht = (1/v1(t)2)h1 + ... + (1/v (t)2)hq
(9.3.4)
for some suitably chosen functions vI(t) , ... , vq(t) . (v) We can perform various combinations of (i) ... (iv) always in such a way that the resulting metric extends smoothly over all of M (N) We consider each case in turn. (i) We first of all work out the connection coefficients for the new metric. Let x E M* , and locally about x choose an orthonormal basis (X ) with respect to g, a' such that if y E Ms for some s, then (X (y)) is an orthonormal basis for T M Now g = (1/p (sZZ )) d s2 +g I and so (Xa, T) _ (X p ) is an orthonormal basis with s a9
respect tog .
Proposition 9.3. 1 VX
If
= p VX4
V is the Levi-Civita connection with respect to -g, then
(9.3.5)
,
for all X C TM, s for s E int Is Proof We can adapt Proposition 9.1.2 to apply to our present situation, and if we write a2 = 1 and p2 = p2 , then equation (9. 1. 10) tells us that
VXX) = 2 g(S, VXx) p therefore
-g(VX ,X) = g(U,V X)
-µ _ 164
(VX E,X )
-fig (VX
X)
In the new metric g , V s no longer has unit length. We therefore reparametrize s into 9(s) , such that -g (V s , 0 s) =1. Clearly the new map 0: (M,g) - (N,h) (we assume that g extends smoothly to M) is harmonically equivariant with respect to s and t. We must therefore compute the equations (4.1.4) and (4.1.5) in the new variables . Note we are using Theorem 4.2. 1, since s is now a generalized isoparametric function. X1(s0) , ... , X (s0) are the principal curvatures of M. with respect to Suppose P the isoparametric function s, for some s0 E int Is. Then from Proposition 9.3.1, the principal curvatures of Ms with respect to the generalized isoparametric func0 _ tion s are given by X 1(s0) , ... , )gyp (s0) , where Ak(s0) = µ(s0)Ak(s0)
(9.3.6)
for k = 1, ... , p. Therefore
As = P As
(9.3.7)
,
where 2i is the Laplacian with respect to g . Since p. (t) and yk (s, t) remain unchanged, equation (4.1.4) become s lk
as(s)
p
+
k=1 µl k yk
(9.3,8)
0;
this is the reduction equation in the new metric g . Alternatively this equation can be written in terms of the variable s (from Lemma 5.1.1) as Ids l2 + O s as (s) + g
p
(9.3.9)
k I p. k yk = 0 .
Now Ids 12 = µ(s)2. Whilst g
p s = s (s) Id 1 22 + s'(s)Zs
(9.3.10)
g
But without loss of generality assume inf Is = 0, then s is given by
s(s) = f(1/µ(u))du. 0
Therefore s' (s) =A (s) and s"( s) = MI( s )p (s) . comes
Thus equation (9.3.12) becomes
165
p(s)2 a,t(s)
+(p '(s)p(s) +p(s) Z s)a' (s) p +
(9.3.11)
0.
k 1 pikyk
We therefore have, substituting equation (9.3.7) into equation (9.3. 11) ;
0 : (M,g) - (N.h) is equivariant with respect to isoparametric functions s and t. Then 0 is harmonic with respect to the metric g of (i) if Theorem 9.3.2
If
and only if
O"(s) +(pp((s) + &s)a'(s)+(1/p(s)2) E
k=1
p yk=0
(9.3.12)
ik
for all s s int I s (ii) Such a change in the metric h still preserves the fact that 0 is equivariant, for we are simply reparametrizing the isoparametric function t. We therefore have
Theorem 9.3.3 If 0: (M,g) -. (N,h) is equivariant with respect to the isoparametric functions s and t. Then 0 is harmonic with respect to the metric h of (ii) on N, if and only if p
a (s) + As a (s) + v E k=1
"
p
y
lk k
=0.
(9.3.13)
for all s c- int I s Proof We have simply changed the principal curvatures p1, ... , pq on the level hypersurfaces Nt by an amount given by equation (9.3.9) , that is (t)
= v(t)p. (t)
Since t is a generalized isoparametric function on (N,h ), the result now follows from equation (4.2. 1). (iii) First of all assume p = 2. Let mi = dim Si, i = 1,2, and consider a local frame field adapted to the principal curvature spaces ; (X ,) =
0
a
a=1,...,m-1
(Xi'XrMi=1,... , m1, r=m 1 +1,... ,ml +m2 =m-1 Proposition 9.3.4 166
If
(Xa, ) is the frame field defined above, then
(X
a,
0=
(Xi , Xr , V
) =(a 1 Xi , o2Xr, ) is an orthonormal frame field with respect to
If
is the Levi-Civita connection for g then the following formulae hold: g(Xi, V
o2g(Xi,V
=0
(9.3.14)
g(Xr,V) _
a2g(Xr,V U) = 0
(9.3.15)
0
(9.3.16)
)
g(Xt, V . Xs = a2g(Xt,vX Xs)
r
g
(9.3.17)
r
Xj) = 1 g (Xk, vX Xi
(Xk' V
(9.3.18)
g(X., VX Xs) = 2g(Xi, V Xs)
(9.3.19)
g(Xr,V Xj)
=
(9.3.20)
g(Xr, V- Xs)
= al a2 g(Xr,OX Xs)
(9.3.21)
g(Xj,VX Xi) =U g(X.,VX Xi) 1U2
(9.3.22)
r
r
a2 (Xr,VxXI)
r
r
g(Xi'VX.t )
= g(Xi,VX U) - dal( )g(X.,Xi)/gal
j
(9.3.23)
j
g(Xr,OX
)
g(Xr, VX)
) - da(U)g(Xr,Xs)/2a2
= g(Xr,OX
s
(9.3.24)
s
=0
(9.-3.25)
1
etc., for all
i,j,k=1,...,m1 and r,s,t=m1+1,...,m1+m2'
Similar formulae hold for any p, and we now assume p is arbitrary, and the prin-
cipal curvatures are al, ... , ap with multiplicities ml, ... , mp respectively. Corollary 9.3.5 The integral curves of with respect to g .
are affinely parametrized geodesics
Proof This follows from equations (9.3.14), (9.3.15) and (9.3.16).
167
Corollary 9.3.6 If X is a principal curvature vector of Ms , with principal i Xi is still a principal curvature vector of Ms curvature , then in the metric Aki
with principal curvature
k = Ak + uk
(s)/2 vk
where X. E Sk
i Proof This follows from equations (9.3.23), (9.3.24) and (9.3.25).
Corollary 9.3.7
With respect to the metric
s is a generalized isoparametri c
function.
Corollary 9.3.8
If E is the Laplacian with respect to p
As = As - E k=1
then
,
Mk ak (s)/2 ok
Proof This follows from Corollary 9.3.6 and Lemma 2.2.9 Let S denote the stress-energy tensor of 0 in the metric
.
again, and use the ranges of indices 1 < i,j, ... < ml; m1 + I < a,b, ... < m1 + m2 = m- 1.
< r,s, ...<m I +m2,
6
Assume p = 2
We make the following further assumption on the equivariant map 0 . One of the conditions for equivariance is that d O(Sk) C Tjk , for each k = 1, ..., p and some
jk = 1, ... , q . If whenever k / 1, then jk j 1; call 0 p - equivariant. For example, the map 0: Sm-1 - Sn-1; 0(coss.x, sins.y)) = (cos a (s) g1 (x), yESq-1, p+q=m,g1: Sp-1 _ Sr-1 , g2: sina(s)g2(y)), xESp-l'
S q-1 -- Ss -1, harmonic of constant energy, is p- equivariant. On the other hand
the map ¢: S3 _ S2 ; 0 (coss.x, sins y)) _ (cos a(s), sina(s).xky1), x,y E S1, is not p- equivariant. Proposition 9.3.9
V SO(Xk) for all
=
In the above notation, and provided 0
,
0 is p- equivariant, then
V *S0(Xr) = 0
k = 1,...,m1 and all r=ml+1""'
m1+m2.
Proof The divergence is given by (summing over repeated indices) 168
V *S0(Xk)
V*(6*ht) (Xk)
-Xa(0*ht(Xa,Xk))
0*ht(OX +
a
a'Xk)+
0*ht(
a'VXXk) a
}
_ -Xi(0*ht(Xi,Xk)) - Xr(0 ht(Xr,Xk)) + 0*ht(vX Xi,Xk) + 0*ht(VX Xr,Xk)
r + 0 ht(Xi, OX Xk) + 0*ht(Xr,o- Xk) . i r i
*
Now the 2nd term vanishes since 0 preserves the orthogonality of the principal curvature spaces. The 4th term vanishes from equation (9.3. 19) and the fact that the mean curvature of the small sphere .,/ is proportional to t (Proposition 2.2.1) . 2 The 6th term is equal to a2 a1 0 ht(Xr, oX Xk) . Now
r
g(Xs,V Xk) X
r
g(VX Xs,Xk)
r
But OX Xs points in the
r
t;
direction, since by Proposition 2.2.1, `12 is totally
geodesic in Ms, and so g(X Xs,Xk) = 0. Thus X Xk is either in S1 space or proportional to . Both are perpendicular to S2 space, and since 0 preserves this orthogonality; the 6th term vanishes also. We have now established that
V* S (X k ) 0
=
*
) + 0*h(X,V a3 (-X(0*h i k i t(Xi ,Xk )) + 0 ht (VX.X,X t i X.Xk))
But this is zero since the right hand side of equation (4. 1. 11) is zero. Similarly
V S0(Xr) =0 Similarly for arbitrary p, and provided 0 is p-equivariant, one can show V * 0 is proportional to . Theorem 9.3.10 If 0 : M, g) -- (N,h) is p- equivariant with respect to isoparametric functions s and t. Then 0 is harmonic with respect to the metric g of ( iii) if and only if
169
aiT(s)
°I (s)
p
+(As
p
- k 1 mk 20 k(s) ) as(s) - zdt(k 1 °k2
'k
p
- (1/2 a') E
k=1 ak °k1 yk
0
,
for s E int I
s
Proof As before
v*S0() = z(dsy+dty.a'(s)a"(s)a'(s) - v*(0 *ht) (
- O s ( a ' ( s ) )2
)
P
where y = E y , with yk = °k2 yk' k=1 k
Then
(d/(i s)y = k=1 2 E°k ok ky+ k=1 °2d y k s k E Also
V*(0*ht)(T)
E A*ht(Xa,VX a
a
)
E E °k0*ht(X. ,-XkX. ), where (Xi )i span Sk k ik k k kk E
2;
6*ht(Xik,Xik
°k2
kk
Xk
°k2 A kyk + kE °k2 ak (s) yk/2 °k Thus
P
zd
s
V * (0 *ht)
°k2
E
k=1
(2 ds yk - Xkyk) + 2 k °k 1c yk
°k °k yk ' from Proposition 4.1.12 M
2
k
The result now follows from Corollary 9.3.8 .
Remark 9.3.11 We have defined the notion of 0 being p-equivariant in order to carry out the deformation. We can replace this condition by another condition.
If 0: (M,g) - (N,h) is equivariant (M,N are space forms ) and 0 170
I.`fk:-
f. k
Ik
is harmonic of constant energy density, for each k = 1, ... , p, then call 0 '/equivariant with respect to s and t. We remark that all the examples of equivariant maps which we have considered are also 'f - equivariant. If 0 is '/ - equivariant and ik: fk - M is the inclusion map, then d(0 o i k ) = dO(Aik) + trace Vd0(dik, dik)
is proportional to n , and since dik is proportional to proportional to
r7
;
trace VdO(dik,dik) is
.
The new metric g has simply changed the sizes of the small spheres .'/ k' and we conclude that 0 ,t: Ms _N is still harmonic and of constant energy density in the S, t. t metric 'j. Thus 0 is equivariant with respect to the generalized isoparametric functions s and t, and Theorem 9.3.10 is true with "p-equivariant" replaced by "jequivariant "as a consequence of Theorem 4.2. 1 (iv) As in case (i), the integral curves of n = Vt are geodesics in the metric h , and the function t: N -- R is a generalized isoparametric function. If we assume that 0: (M,g) -- (N,h) is.,/-harmonically equivariant, then in the metric h , 0s t: Ms _N is harmonic of constant energy density. Thus 0: (M,g) -t (N,h) is equivariant with respect to the generalized isoparametric functions s and t. Theorem 4.2.1 applies, and we have Theorem 9.3.12 If 0: (M,g) -- (N,h) is y-equivariant with respect to the isoparametric functions s and t. Then 0: (M,g) -. (N,h) is harmonic, where h is defined in GO, if and only if
a (s) +as cx (s) + Ek (µ.
+
lk
v.
lk
/2 v )y /v. 2 k lk lk
=0
for all s E int I s . Proof
For each k = 1, ..., p, we have
yk(s,t) = E 0 * h. (Xi lk .
k
(1/v (t)2 ) yk(s,t)
Xi k
k
lk
where h. = h /v. (t)2 ). In the metric h the principal curvatures become 1
7
l = µ + v.-12 P. 171
for each j = 1, ... , q. The result now follows from equation (4.2.1). 9.4
Examples
Example 9.4.1 Consider a deformation of the kind given by case (i) of Section 9.3. The problem here is to define u(s) in such a way that g is a smooth metric. For example, suppose µ (s) = K, where K is some constant not equal to 1. Consider the join S2 = S1* S0 . Then, with respect to the metric g , S2 still has constant positive curvature [ 38, vol. 2, chapter 7, addendum I 1 equal to K2 , and so has the appearance of a rugby ball [ 13 1.
The metric g is no longer C1. However, we can use bump functions as follows. Define
µ(s) = 1 +l(s), t > -1
,
where t : Is -- R is smooth, and supp t c int Is. Then (M, g) is a smooth manifold, since the question of smoothness only arises across the focal varieties of s, and g = g in a neighbourhood of these focal varieties. Suppose we are given one of Smith's maps between two spheres. Consider the reduction equation (equation (1.3.7)) - we use the reparametrized equation to avoid singularities. Use the same reparametrization for equation (9.3.12), so that time varies between - oo and + oo. Then the asymptotic form of the equation as time u -- + '0 will be the same in the deformed case as in the undeformed case. Theorem 6.1.9 and Lemma 6.1.11 will still apply to yield a non-trivial solution. Therefore for the maps of Example 5.3.1, the existence of solutions will be unaffected provided It 12 is small enough. For maps from Euclidean space to sphere and from hyperbolic space to sphere, the existence of solutions will again be unaffected. However, we would expect the asymptotic behaviour as time u - + oo to change substantially with such deformations of the metric. 172
Example 9.4.2 Consider the join of two harmonic polynomial maps as described in Section 1.3 and Example 5.3.1 . We thus consider a map 0 : S m- i _ S n-1 of the form
0(coss x, sins y) = (cos a(s)gI(x) , sin o(s)g2(y))
,
where x SP-1 , y C S q-1 , gl : SP-1 _ S r- i and g2 : S q-1 -- S s-1 are harmonic polynomial maps with I dg. 12 = ? . constant, p + q = m and r + s = n. Express the Euclidean metric on S m-1 in the form g
= d s2
+
cost s dx2 + sings. dy2
where dx2 is the Euclidean metric for SP-1 and dy2 that for Sq-1. We now perform a deformation of the metric incorporating deformations of types (i) and (iv) in order to make Sm-i ellipsoidal. The deformed metric has the form
g = (a 2sin 2s + b 2cos2s) ds 2 + a 2cos2s dx 2 + b2sin2 sdy 2 The sphere has now become ellipsoidal with one set of axes of length a having multiplicity (p - 1), one set of length b with multiplicity (q - 1), and the other axis retaining its original length of 1. Using the notations of equation (9.3. 1) and equation (9.3.3) we have µ(s)
2
2 2 2 2 = 1/(a sin s + b cos s)
2
,
al
= 1/a
2 ,
2 a2
= 1/b
2
(9.4.1)
The reduction equation before the deformation has the form
a "(s) + As a '(s)
+
2 E µ k=1 ik
where as =(q - 1) cots - (p- 1) tans, yl = A1cos2(s)/cos2s, y2 = µ.
= tan a(s) and p
j2
cot a(s).
X2sin2(s)/sin2s,
Using Theorem 9.3.2 and Theorem 9.3.10 ,
after the deformation the reduction equation becomes a 1 '(s)
+
W µ ( (s)
+ A s)
a'(s) + ( 1/u(s))2 k2lµj akyk
(9.4.2)
k
We reparametrize equation (9.4.2) as before, defining a new variable u by eu = tan s. Note that care must be taken since equation (9.4.2) no longer has the form of equatica (5.1.1) and so Lemma 5.1.1 no longer applies. 173
A computation shows that
p '(s) _ -(a 2- b )2 sins coss/(a 2sin 2s + b 2cos2s) 3/2 or
(a2- b2) u -u e +e
p' (s(u)) =
b2e-u)
(a2eu + u
e +e
3/2
-u
This tends to 0 as u tends to both - -a and + oo . Also u -u e + ep(s(u )) 2 2
a e
u +b2e-u
which tends to 1/b2, 1/a2 as u tends to becomes, in the variable u , 1
eu + e-u
-Go, + 00
respectively. Equation (9.4.2)
[µ +(q-2)e-u - (p-2)eu] a' (u) +
(12
X e )
eu +e-u
1
sin a(u)cos a(u) C
a
2
u
A 2e
b
-u
2
= 0. (9.4.3)
If we abbreviate this in the form
ait(u) + h(u)a'(u) +g(u)sina(u) cosa(u) = 0
then as u -- - 0 h(u) -- q - 2
g(u) -- - X2
and as u
h(u) - -(p-2) g(u) -- X1. The damping conditions become
174
,
Thus the damping conditions are as for the Smith maps, given by equations (1.4. 1). The methods of Section 6.1 (in particular Theorem 6.1.9 and Lemma 6. 1. 10) show that, provided the damping conditions are satisfied, equation (9.4.3 ) has a solution
yielding a smooth harmonic map 0: S m-1 - S n-1, where S m-1 has the ellipsoidal metric described above. We therefore have Theorem 9.4.3 The join of two harmonic polynomial maps between spheres for which the damping conditions are satisfied, can be deformed into a harmonic map from the ellipsoid described above into the sphere. The ellipsoid has distinct eccen-
tricities given by a, b and 1, for any a, b > 0. Theorem 9.4.4.
The join of two harmonic polynomial maps between spheres for
which the damping conditions are not satisfied, cannot be rendered harmonic by an ellipsoidal deformation of the above kind on the domain sphere.
In contrast, however, we see in the next example that such ellipsoidal deformations on the range sphere do yield harmonic maps.
Example 9.4.5 Again consider the join of two harmonic polynomial maps as in the last example. We can deform the range sphere into an ellipsoid, by using deformations of types (ii) and (iv). That is, we express the Euclidean metric on Sn-1 in the form
h = d t2 + cos2 t du 2
+
sin2t d v2
,
where d u2 , d v2 are the Euclidean metrics for S r-1. S s-1 respectively, then the deformed metric has the form h = (a2 sin2 t + b2 cos2 t) d t2 + a2 cos2 t d u2 + b2 sin2 t d v2
In the notations of equation (9.3.2) and the equation (9.3.4), we have v(t)2
=
1/(a2sin2t
+
b2cos2t), v1 = 1/a2, v2
=
1/b2
(9.4.5)
From Theorem 9.3.3 and Theorem 9.3.12, the reduction equation becomes 2
a (s) + As a '(s) + v(a(s)) E "
k=1
-yk
L.
lk
2
=0.
V.
)k
175
Using the parameter u defined by eu = tan s, we obtain
a" (u) +
eu+e -u 1
(eu
((q - 2)e-u - (p -2 )eu) a' (u)
sin a (u) cos a (u) +e-u) (a2sin2 a(u) +b2cos2a(u))2
(a2a eu - b 2 A a-u) = 0 . 1
2
(9.4.5) This equation fits into the category of equations described in Theorem 6.1.9. The arguments of that section apply, and we are assured of a solution provided the damping conditions are satisfied. These damping conditions are that
(q-2)2 < 4bX2 (p-2)2 < 4a X1.
(9.4.6)
Given the join of two harmonic polynomial maps, we can always find a and b such
that (9.4.6) are satisfied. We therefore have Theorem 9.4.6 The join of any two harmonic polynomial maps can always be rendered harmonic by a suitable deformation on the range sphere. In particular we can apply Theorem 9.4.6 to Example 1.4. ]. to yield
Theorem 9.4.7 For each n = 1,2, ... , there exists a smooth metric on the range sphere Sn (depending on n), such that each class of Tin (Sn) = Z contains a harmonic representative. The deformed spheres are familiar ellipsoids whose eccentricities depend only on n and the degree.
176
References
1
2
3
4 5
R. Abraham and J.E. Marsden Foundations of Mechanics (Second Edition), Benjamin/Cummings, (1978). P. Baird and J. Eells A conservation law for harmonic maps, Springer L.N.M. vol. 894 (1981), 1 - 25. F. Brauer and J.A. Nohel Ordinary Differential Equations, Benjamin, (1967). F. Brauer and J. A. Nohel Qualitative Theory of Ordinary Differential Equations, Benjamin, (1969). Families de surfaces isoparametriques dans les espaces E. Cartan a courbure constante, Ann. Mat. Pura Appl. , 17 (1938), 177 - 191.
6
E. Cartan
Sur des families remarquabies d'hypersurfaces isoparametriques dans les espaces spheriques, Math. Z. , 45 (1939) , 335 - 367.
E. Cartan
Sur quelques families remarquables d'hypersurfaces, C.R. Congres Math. Liege, (1939), 30 - 40. Sur des families d'hypersurfaces isoparametriques des 8 E. Cartan espaces spheriques a 5 et a 9 dimensions, Revista Univ. Tucuman, Serie A 1, (1940), 5 - 22. Theory of Ordinary Differential Equations, 9 E. Coddington and N. Levinson McGraw-Hill, (1955). 10 J. Eells and L. Lemaire A report on harmonic maps, Bull. London Math. Soc., 10 (1978), 1 - 68. 7
ii J. Eells and L. Lemaire
On the Construction of Harmonic and Holomorphic Maps between Surfaces, Math. Ann. , 252 (1980), 27 - 52.
12 J. Eells and J.H. Sampson Harmonic mappings of Riemannian manifolds, Am. J. Matt.. , 86 (1964), 109 - 160. 13 L.P. Eisenhart Riemannian Geometry, Princeton Univ. Press, (1925). 14 G. F. R. E lli s and S. W. Hawking The large scale structure of space time, 15 R.H. Escobales
16 D. Ferus
Cambridge Monographs on Math. Physics (1973). Riemannian submersions from complex projective space,
J. Diff. Geom., 13 (1978), 93 - 107. Notes on Isoparametric Hypersurfaces, Preprint. 177
17 D. Ferus, H. Karcher and H.F. Munzner Cliffordalgebren and neue isoparametrische Hyperflachen , Math. Z. 177 (1981), 477 502.
18 The Feynman Lectures on Physics. Addison-Wesley (1964). Harmonic morphisms between Riemannian manifolds, 19 B. Fuglede Ann. Institut Fourier, 28 (1978) , 107 - 144. A criterion of non-vanishing differential of a smooth map, 20 B. Fuglede Bull. London Math. Soc 14 (1982), 98 - 102. 21 P. Hartman Ordinary Differential Equations, Wiley, (1964). Die Grundlagen der Physik, Nachr. Ges. Wiss.Gottingen 22 D. Hilbert (1915), 395 - 407 ; (1917), 53 - 76 ; Math. Ann. , 92
(1924), 1- 32. Differential Equations, Dynamical Systems and Linear Algebra, McGraw-Hill, (1974). 24 D. Hoffman and R. Osserman The geometry of the generalized Gauss map, Memoir A,M.S. 236 (1980). 25 W.-Y. Hsiang and H.B. Lawson Minimal submanifolds of low cohomogeneity, J. Diff. Geom. , 5 (1971), 1 - 38. Fiber Bundles, McGraw-Hill, (1966). 26 D. Husembller 27 T. Ishihara A mapping of Riemannian manifolds which preserves harmonic functions, J. Math. Kyoto Univ. , 19 (1979), 215 - 229. 28 L. Lemaire On the existence of harmonic maps, Thesis, Warwick Univ., (1977). 29 1-I. P. McKean The Sine-Gordon and Sinh-Gordon Equations on the Circle, Comm. in Pure and Appl. Math., Vol. XXXIV. , p. 197 (1981). 30 H.F. Munzner Isoparametrische Hyperflachen in Sph'aren I, Math. Ann. 251 (1980). 57 - 71. 31 M. Nagumo Uber die Differentialgleichung y"= f (x,y,y'), Proc. Phys.-Math. Soc. Japan (3) 19 (1937). 32 K. Nomizu Elie Cartan's work on isoparametric families of hyper-
23 Hirsch and Smale
11
surfaces, Proc. Symp. Pure Math., 27 (1975), 191 200.
33 H. Ozeki and M. Takeuchi On some types of isoparametric hypersurfaces in spheres, I, II, Tohoku Math. J. , 27 (1975), 515 - 559;
28 (1976), 7-55. 34 G. Sansone and R. Conti 178
Non-linear Differential Equations, Pergamon, (1964).
35 S. Sasaki
36 R. T. Smith
On the Differential Geometry of Tangent Bundles of Riemannian Manifolds, Tohoku Math. J., 10 (1958), 338 - 354.
Harmonic mappings of spheres, Thesis, Warwick Univ., (1972).
Harmonic mappings of spheres, Amer. J. Math., 97 (1975), 364 - 385. Differential Geometry, Vol. 2,3,4, (1975). 38 M. Spivak A class of hypersurfaces with constant principal curva39 R. Takagi tures in a sphere, J. Diff. Geom. , 11 (1976) , 225 - 233. 40 R. Takagi and T. Takahashi On the principal curvatures of homogeneous hypersurfaces in a sphere, Diff. Geom. in honour of K. Yano, Tokyo (1972) , 469 - 481. Composition Methods in Homotopy Groups of Spheres, 41 H. Toda A.M. Studies 49. Non-existence of solutions to certain Dirichlet problems 42 J. C. Wood for harmonic maps, Preprint (1981) . A note on harmonic polynomial maps, Preprint. 43 R. Wood
37 R.T. Smith
179
Index of definitions
Clifford system Critical set of a map Damping conditions
Degree of an isoparametric function Dilation
Energy density
Equivariant with respect to isoparametric functions Focal variety Generalized family of isoparametric hypersurfaces Generalized isoparametric function Harmonic map Harmonic morphism Hopf invariant Hopf map
Horizontally conformal
Isoparametric function Isoparametric hypersurface Join of a map Join of two spheres Laplacian Orthogonal multiplication
p- equivariant with respect to isoparametric functions Reduction
Reduction equation
Rendering problem S-equivariant with respect to isoparametric functions S-wavefront preserving with respect to isoparametric functions 180
Second fundamental form of a map
1, 7
Stiefel manifold
36
Stress-energy tensor Unit speed reparametrization of a generalized family of isoparametric hypersurfaces Wavefront preserving with respect to isoparametric functions
43
66 50
181