Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
of
Manif0Ids
PURE A N D APPLIED MATHEMATICS
A Series of Monographs and Textbooks
Edited by
1964
ACADEMIC PRESS
NEW YORK AND LONDON
0 1964,
COPYRIGHT BY PRESSINC.
ACADEMIC
ALL RIGHTS RESERVED.
NO PART OF THIS BOOK MAY BE REPRODUCED I N ANY FORM,
BY PHOTOSTAT, MICROFILM, OR ANY OTHER MEANS, WITHOUT
WRITTEN PERMISSION FROM THE PUBLISHERS.
LIBRARY
OF CONGRESS CARDNUMBER:64-20317
CATALOG
Preface . . . . . . . . . . . . . . . . v
CHAPTER 1
Manifolds
CHAPTER 2
Lie Groups
CHAPTER 3
Fibre Bundles
CHAPTER 4
Differential Forms
4.1 Introduction . . . . . . . . . . . . . 53
4.2 Classical Notion of Differential Form . . . . . . . . 53
4.3 Grassmann Algebras . . . . . . . . . . . . 54
vii
...
Vlll CONTENTS
CHAPTER 5
Connexions
CHAPTER 6
Affine Connexions
6.1 Definitions . . . . . . . . . . . . . . 89
6.2 T h e Structural Equations of an Affine Connexion . . . . . 99
6.3 T h e Exponential Maps . . . . . . . . . . . 108
6.4 Covariant Differentiation and Classical Forms . . . . . . 111
CHAPTER 7
Riemannian Manifolds
CHAPTER 8
CHAPTER 9
Riemannian Curvature
CHAPTER 10
CHAPTER 11
BIBLIOGRAPHY . . . . . . . . . . 260
SUBJECT
INDEX . . . . . . . . . . 265
This Page Intentionally Left Blank
Geometry
of
Manifolds
This Page Intentionally Left Blank
CHAPTER 1
M an i fo Ids
+
If is a map of M into N andJ! , I a map of P into T , then I,!J o will
denote their composition, that is, J!,I o is + +followed by 4. Here
M , N , P, T are any sets and we understand that the domain of
+ +
I,!J o is #-l(P) n M (in particular, $J c may have an empty domain).
T h e same sort of convention, namely, that the domain is the largest
meaningful set, will be used in the formation of sums, products, and
other combinations of maps. If U c M we use +ILI for the restriction
+
of to u.
FIG. 1.
1
2 1. MANIFOLDS
exp(-I/.) if x > O
if x 0.<
Let {rm}be an ordering of the rational numbers and
22 3 .
W
).(g = - rJ.
n=l
FIG. 2.
Examples
I Rd+'
FIG. 3
1.2. Definition of a Manifold 5
tively. Then Vo = {+, +} is a basis for a C" structure on Sd. [Stereo-
graphic projection: +(.) is the point where the straight line from
(0, ..., 0, 1) through x intersects ud+,-l(0) = Rd.]
(5) Real projective space. Let Pa be real projective d-space, that
is, the collection of straight lines through the origin in Rd+'. T h e
+
natural covering map : Sd + P d , which takes x into the line through
x,induces a C" structure on Pa,that is, there is a unique C" structure
+
on Pa such that is a C" map with local C" inverses. (See problem 4
below.)
(6) Low dimensions. Essentially the only 1-dimensional connected
paracompact manifolds are R1 and S1. If the restriction of para-
compactness is omitted, then this is no longer true, as an example
of a long line shows-this is obtained by connecting together an
uncountable well-ordered collection of half-closed intervals [39,
pp. 55 and 561.
A 2-dimensional manifold is called a surface. T h e objects tradi-
tionally called "surfaces in 3-space" can be made into manifolds in
a standard way. T h e compact surfaces have been classified as spheres
or projective planes with various numbers of handles attached [80].
(7) Product manifolds. Let M , N be C" manifolds with structures
%, 93 and dimensions d, e, respectively. Let p , : M x N -+ M ,
p , : M x N + N be the projections. Then Vo = {(+ o p , , I) o p z ) =
( ~ 1 0 **., uao + o P , , ~ +
1 0O P ,
9 ueo $ 0 ~ 2 )I + E V , I ) E ~ }
is a basis for a Cm-structure on M x N. T h e same structure could
also be obtained by using only a basis for V, 9 in their places. More
specific examples are:
Cylinder = R x S1
2-dimensional (ordinary) torus = S1 x S1= T 2
d-dimensional torus = S1 x x S1 ( d factors) = T d .
Note also that Rd+ecan be canonically identified with Rd x Re.
(8) Non-Hausdorfl manifold. T h e following example shows that
the Hausdorff property of a manifold does not follow from the
existence of a C" structure. T h e underlying point set consists of
the interval (0, 3) with topology described in terms of neighborhoods
as follows.
T h e neighborhood of a point in (0, 1) u (1, 2) u (2, 3) would be
as in the topology induced from the reals.
6 1. MANIFOLDS
O ----.,.1 2 3
-2
1
->
FIG.4.
1.3. Tangent Space 7
+
Problem 4. Prove that if : M + N is a covering map and N has
a C" structure, then there is a unique C" structure on M such that
M is a C" covering of N.
+
Problem 5. Prove that if : M - t N is a covering map and M has
a C" structure such that for every Ui, U j open sets in M on which
q5 is a homeomorphism and +( Ui) = +( U j ) we have (4 o 4 is
a C" map on U i ,then N has a unique C" structure such that M
is a C" covering of N .
Problem 6. Prove that if N is a connected C" manifold, then there
exists an essentially unique simply connected C" covering of N .
(Part of the problem is to show that manifolds satisfy a condition
sufficient for the existence of simply connected topological coverings.)
Problem 7. A d-dimensional complex manifold is a topological space
locally homeomorphic with complex space Cd, and these homeo-
morphisms are complex analytically related. Make this precise and
show that a complex manifold is an even-dimensional (real) manifold.
Examples of complex manifolds are Cd itself, the Riemann surface
of a complex analytic function of one complex variable, complex
d-dimensional projective space CPd, and the set of all nonsingular
linear transformations of Cd, denoted by Gl(d, C).
FIG.5.
8 1. MANIFOLDS
FIG.6.
lacobian matrices. +
If : M -t N , m E M , (xl , ..., x d ) is a coordinate
system at m, (yl , . . . , y e )is a coordinate system at +(m), then the
matrix of d+ with respect to the bases { D J m ) } and {Duj+(m))} is
the Jacobian (Dzj(yio +)(m)).
FIG. 1.
Chain rule. If + : M - t N , i,/J : N-+ P are C", then d($ o +) = di,/Jo d+.
We can rephrase our previous definition of the tangent y * ( t ) to
a C" curve y as: y,(t) = dy(D(t)). (D = D , = d/du.) Later (3.3)
we shall give the collection of all tangents to M , T ( M ) ,a C" structure,
so that y* will become a C" curve in T ( M ) .
FIG. 8.
[X,
YI = 2(fiD&
1.i
- g,D,,f,) DZj.
FIG. 9.
FIG. 10.
FIG.1 1 .
Noticing that d T - , ( Y ( y ( t ) ) ) f = Y ( y ( t ) ( f oT P t ) we
, see that we must
consider derivatives with respect to Y , and hence we introduce S,
the one-parameter group associated with Y . Then if we define
G : V -+ R, where V is a neighborhood of (0,O) in R2, by
FIG.12.
= +
Y2f(m) 2 X Y f ( m ) - 2Y2f(m) - 2 X Y f ( m )
+
X2f(m) - 2 Y A p ) - 2X2f(m)
+ + +
Y 2 f ( m ) 2 X Y f ( m ) X2f(m)
= 2XYf(m) - 2YXf(m). QED
Problem 20. Let h be a C" curve such that h,(O) = 0, and define
g on small positive numbers by g(P) = h(t), and let m = h(0).
Show that:
(1) T h e map defined on F ( M , m) by f + ( f c h)"(O) is a tangent
at m.
(2) T h e same tangent is 2 liml+o+g*(t),where the meaning of lim
is the same as above.
If we call this the 2nd order tangent to h at a point where the
1st order tangent vanishes, generalize (1) so as to define an nth order
tangent to h at a point where the 1st order, 2nd order, ..., (n - 1)st
order tangents vanish, and prove a result analogous to (2).
20 1. MANIFOLDS
1.5 Submanifolds
Problem 26. Let t+b : N + M be C" with dim M < dim N , and
let S be the set of measure zero defined in problem 11. Prove that
if m 4 S then P = $-l(m) with the topology induced from N has a
unique C" structure so that it becomes a submanifold of N under
the inclusion map [88].
+
Problem 27. For $ = uZ3 u13 - 3u1u2 : R2 ---f R find what S is
and show why +-l(m) is not a submanifold in the cases where m E S.
When t+b is a real-valued function on R3 the submanifolds obtained
as in problem 26 are the traditional surfaces in 3-space.
Problem 28. Let f : P -+ M be C", i : N -+ M a submanifold, such
that f ( P ) C i ( N ) . Let g = i-' o f : P 4N .
(a) If g is continuous show that g is C".
(b) Find an example for which g is not continuous.
Lie Groups
G x G -+ G given by (g, h) - t g h
G -+ G given by g +g-' are Cc.
Lie group admits a Lie group structure such that the covering map
is C" and a homomorphism. In particular, every connected Lie
group is locally isomorphic to a simply connected Lie group.
A Lie subgroup of a Lie group is a subgroup which is also a Lie
group and a submanifold.
(d) Use this formula for Xij to verify directly that the brackets
are preserved under the identification of g with gI(d, R ) [25].
2.4 Homomorphisms
FIG. 13.
I7
then the diagram
dj
G-H
Proof. Assuming for the moment that exp is C" we first show that
it is a diffeomorphism in a neighborhood of 0. I t is sufficient, according
to the inverse function theorem, to prove that d exp is onto at 0.
If s E G, , then there is a left invariant vector field X such that
X(e) = s. By definition, exp takes the ray generated by X (that is,
the curve in g given by r -+ r X ) into the integral curve of X through
e . Hence d exp takes the tangent to this ray at 0 into the tangent to the
integral curve of X at e, that is, into s.
We blame the differentiability of exp on theorems in differential
equations which say that solutions depend in a C" manner on para-
meters entering in a C" manner into the functions defining a system
of differential equations (see appendix). I n this case we have the system
of differential equations determined by a left invariant vector field
X and X depends linearly on a system of linear coordinates of g.
QED
Let X , , ..., X, be a basis of g; then we identify g with Rd as a
manifold by the correspondence CciXi tf (c, , ..., c,). I n a sufficiently
small neighborhood U of 0 in g, exp is a diffeomorphism and hence
can be taken as a coordinate map. T h e coordinates and coordinate
neighborhood obtained in this way are designated as canonical.
" 1
eA = z g A k (Ao= I ) .
0
G4d, R )
o.'(s) = (01*(4 +
4)) ( 4 ) 1 u,*(s)>.
2.6 Representations
Proposition. Ad is a representation of G on g.
Xi = UiDi + ~ $ 2 + ~ $ 3 +u~D,
9
Fibre Bundles
-1
G
Corollary. If X , Y ~ g the
, Lie algebra of a Lie group G, then
ad(X) Y = [X,Y].
Proof. Let G act to the left on itself as follows: g E G, g : G + G is
defined by g(h) = hg-l; that is, the transformation g is just R,-1,
right multiplication by g-l. Since Ad g is dL, o dR,-1 , and X is
t Here we are assuming the action is a left action. In the case of a right action,
obvious modifications are necessary in the definitions, although the statement of
the theorem is the same.
3.2. Principal Bundles 41
invariant under left translations dL, for every X E ~we, may apply
the theorem with 9 = g and j = Ad. We have to determine, for
X E g, the vector field on G. For m E G= M , we have that m : G+M
as above is given by m(g) = g(m) = mg-l, so m =L, o J , where
J is the inversing map. Thus x ( m ) = dL, o dJ(X(e)) = dL,(-X(e))
= - X ( m ) , by lemma 2.1. Therefore k = - X . Since dj = d(Ad) =
ad, the theorem now gives the result.
Problem 3 . Prove as a corollary to this: if G is Abelian, then g is
Abelian.
Proof of theorem. Assume the action is to the left. Let X Eg, so
y : t+ exp t X is a one-parameter group of diffeomorphisms of M ,
so there is a vector field 2 on M which arises from differentiating
functions along orbits. For a real-valued function f on M , and m E M
we have
d
Zf(m)= ( O ) ( f @ 0 Y) = dm(r*(O))f,
so Z(m) = dm(X(e)) = X ( m ) .
On the other hand, since the diagram above is commutative
exp ( t d j ( X ) ) = exp (dj(tX))= jexp ( t X ) .
U
FIG.14.
P is called the bundle space, M the base space, and G the structural
group. For m E M , d ( m ) is called the jibre over m. T h e fibres are
diffeomorphic to G, in a special way via the map p : G +r - I ( n ( p ) ) C P,
defined by p ( g ) = R,p. We note that in terms of the F , , the right
action of G on P is given by right translation, that is, if p 4( ~ ( p , )
F,(p)), then pg + (n(p),F&) g ) . This follows from the fact that
F d P d = F d P )g .
We now give some examples of bundles.
Trivial (product) bundle. If G is a Lie group, M a manifold, then
P =M x G provided with the right action of G on itself in the second
factor, that is, ( m , g) h = ( m , gh), is the bundle space of a principal
bundle, the trivial bundle. A bundle is isomorphic to a trivial bundle
if and only if there is a C" cross section of n, that is, a C" map
K : M - t P such that n o K is the identity on M (cf. [85], pp. 25
and 36).
3.2. Principal Bundles 43
Bundle of bases. Let M be a C" manifold and B ( M ) the set of (df 1)-
tuples (m, el , ..., ed), where m E M and el , ...,ed is a basis of M,,
and let r : B ( M ) + M be given by r ( m , e l , ..., ed) = rn. T h e n
GZ(d, R ) acts to the right on B ( M ) by: let g E GZ(d, R), viewed as a
matrix, g = (gij); let (m, el , ..., ed) E B ( M ) , and define R,(m, el, ...,
ed) = ( m , C gilei , ..., C giaei). If m E M , (xl , ..., x d ) a coordinate
system defined in a neighborhood U of m, then we define F , by:
if m' E U , Fu(m',fl , ...,fa)= (dxj(fi)) = (gij) E GZ(d, R). Thus the
functions yi = xi o 7-r and yii = xii o F , give a coordinate system
on r - l ( U ) , where xii are the standard coordinates on Gl(d, R )
(see 2.2).
Using the C" structure given to B ( M ) by the local product represen-
tation (IT, F,), we see that B ( M ) is the bundle space of a principal
bundle, called the bundle of bases of M .
It is sometimes convenient to view B ( M ) as the set of nonsingular
linear transformations of Rd into the tangent spaces of M , that is,
we identify p = (m, el, ..., ea) with the map p : ( r l , ..., r d ) -+c riei.
When this is done it is natural to consider GZ(d, R ) as the nonsingular
linear transformations of Rd, for we have: pg(rl, ..., r d ) =
N P
Gji’(m) = F j ( ~ ) ( F i ( p ) ) -=
l Gji(m).
Examples
(1) Let (P, G, M ) be as above, and let G act on itself by left transla-
tion. Then (P, G, M ) is the bundle associated to itself with fibre G.
(2) Tangent bundle. Consider the bundle of bases B (M). (Note
that we shall often denote a bundle by its bundle space alone.) Now
by definition GZ(d, R ) is the group of nonsingular linear transforma-
tions of Rd,and hence acts on Rd to the left. The bundle space of the
associated bundle with fibre Rd is denoted by T ( M ) , and the bundle
is called the tangent bundle to M . T ( M ) can be identified with the
space of all pairs ( m , t ) , where m E M , t E M , [the “m” in the pair
3.3. Associated Bundles 47
Y d S h t ) = d%(t)
of different sorts. For example, this is the case when M has an almost
complex structure (cf. problem 11).
(4) Vector bundles. These are bundles in which the fibre is a
vector space and the structural group is a subgroup of the general
linear group of that vector space. T h e tensor bundles are a special
case. They are frequently defined with no explicit mention made of
the structural group by giving the bundle space as the union of vector
spaces, all of the same dimension, each associated to an element of the
base space, and defining the manifold structure via smooth, linearly
independent, spanning cross sections over a covering system of coordin-
ate neighborhoods. For example, we essentially did this for T ( M )
when we exhibited the coordinates: in that case the cross sections
were the coordinate vector fields D,*.
Another example of this type is the quotient space bundle of an
imbedding, which usually is considered in the case of Riemannian
manifolds as the normal bundle (the Riemannian metric, defined in
Chapter 7, is employed to get a uniform choice of representatives for
the quotient spaces). This vector bundle may be defined as follows:
Let i : N+ M be the imbedding of the submanifold N in M . T h e
fibre over n E N is the quotient space Mi(Jdi(Nn), and the bundle
space is the union of these fibres, so we may consider the bundle space
as the collection of pairs (n, t + di(N%)), where t E Mi(%). T o get
coordinates we first take a coordinate system in M at i(n), say
xl, ..., xd, and we may assume that xj o i = yi, j = 1, ..., d‘ give a
coordinate system at n, and that Xj(n) = D,.i(n)) +
di(Nn) are
linearly independent for j = d‘ + 1, ..., d. Then for some neighbor-
hood U of n, Xj(n’), j = d‘ +1, ..., d are still linearly independent,
so there is a dual basis Vi(n’). Then for (n‘, X ) E ~’-l( U ) , where n’
is the projection from the bundle space to N , we define coordinates
z1, ..., zd by:
(i) J z = -identity.
(j) Compute J in terms of real coordinate vector fields which come
from the real and imaginary parts of a complex coordinate system.
(k) J is defined on T ( M ) and is a bundle map.
An almost complex structure on a manifold M is a bundle map
J : T ( M ) + T ( M ) such that
(1) J ( M m ) = M , for every m E M .
(2) J z = -identity on each Mm.
T h e J obtained in the above problem is called the complex structure
of the complex manifold M . An almost complex structure will be
called a complex structure only if it is obtained in this way.
Problem 11. (a) If M has an almost complex structure, then M has
even dimension.
(b) M has an almost complex structure if and only if the group
of the bundle of bases can be reduced to Gl(d/2, C) represented in
Gl(d, R) as matrices of the form (g ):- (which corresponds to
+
( A iB)E Gl(d/2, C)).
Every 2-dimensional orientable differentiable manifold admits a
complex structure, so, of course, every 2- dimensional manifold
which admits an almost complex structure admits a complex structure.
T h e latter result is not true for higher dimensional manifolds [21].
+
If M has an almost complex structure then Mm iMnlhas a direct
+
sum decomposition Z m Slnsuch that H m= all x + iJx for
x E Mm. We say that C" vector field X belongs to Z if X ( m ) E Z m
for every m, and then write X E 2. I t is not in general true that if
X E 8, Y E H then [X, Y ]E 8. However, if J is a complex structure
the X v l agrees with our previous definition and [Z,Z]c 2 ;in
fact, this latter condition has been shown to be also a sufficient condi-
tion that J be a complex structure [64].
Problem 12. T h e maximal compact subgroups of R* and C* are
(1, -1} = So and {eie} = S1,respectively. Show that the reduction
52 3. FIBREBUNDLES
of the principal bundles of examples (1) and (3) in 3.2 to these sub-
S1,CP).
groups gives principal bundles: ( S d ,So,P d ) and (SZd+l,
Carry out the same construction to get principal bundles over
quaternionic projective spaces QPd:(S4d+3, S3,QPd).
In the case d = 1 the bundle projections become the Hopf maps
S3 -+ S 2 = CP1, S7+ S4 = QP'.
CHAPTER 4
Differential Forms
4.1 Introduction
In the last chapter the concept of a tensor was briefly mentioned,
and differential forms, the subject of the present chapter are just
special types of tensors. However, we shall initially introduce different-
ial forms here by means of the more explicit formulation in terms of
Grassmann algebras and shall then return briefly to the tensorial
approach (4.5).
I f f is a C" function on M , then we notice that to every m E M
there corresponds the differential of f at m, 1.3, which is a linear
functional on M , , and this correspondence is smooth in the following
sense. Let X be a C" vector field on M , then df(X)(m)= df,X(m)
=Xf(m) defines a C" function Xf on M . Such a smooth assignment
of linear functions is called a differential 1-form, although not every
differential I-form arises as the differential of a C" function. However,
before pursuing this subject further, we must develop some machinery,
namely, Grassmann algebras.
is over all ordered subsets (il, ..., i,) of (1, ..., d } , and the ail,,,i, are
real numbers.
So a strict definition will involve some kind of multiplication of
differentials and then linear combination of these products. Thus
M,* will be imbedded in an algebraic system which has both multi-
plication and vector operations. Furthermore, this shall be done so
that if (yl, ...,y d )is a second coordinate system at m then the expression
C bil.,.ipdyi;.. dyiD,obtained from the other by the usual rules for
change of variable in multiple integrals, will be just the expansion
of the same algebraic object in terms of the new basis.
T h e required algebraic system for this is a Grassmann algebra,
which we now discuss.
0 = (u + V)Z = u2 + + vu + v2 = uv + vu.
11v
Property (iv) is a short but poor way of stating that there are no more
relations among the elements of G than those which follow from
(iii).
T o each basis e l , ..., ed of V there corresponds a basis of G . T h e
elements of this basis of G are in one-to-one correspondence with the
subsets of {I, ..., d } as follows:
(a) If the subset is+ = empty set, we let e4 = e .
(b) If the subset is s = {il , ..., i,} with i, < < ip, let
e, = e, 1 ..- eig*
4.3. Grassmann Algebras 55
Note that there are exactly 2d subsets, so that in order to show that
the e,'s are a basis it is enough to prove the:
(2) Any two Grassmann algebras over V are essentially the same,
that is, if G and H are two such then there is an isomorphism of G
onto H which leaves each element of V fixed. In fact we need only
choose a basis of V and map the e,’s for G into the corresponding
ones for H .
(3) If T is a linear transformation of a vector space V into a vector
space W , and G ( V ) , G ( W ) are the Grassmann algebras over V , W ,
respectively, then there is a unique extension of T to a homomorphism
of G ( V ) into G ( W ) , denoted by Th , and Th satisfies T,(G(V)P)C
G( W)p.
Proof. Uniqueness follows immediately from the facts that T(e)
must equal e, and that G( V ) is generated by e and V . By linearity, Th
can be extended to all of G( V ) after defining
T,(q . * * v p )= T(v,) ... T(v,) for v1 , ..., v p E V ,
and such a Th is clearly a homomorphism.
Problem 5. If T : V - t W , S : W-+ X are linear transformations
of vector spaces, show that ( S T ) , = S,T,.
Problem 6 . If T : V -+ V is a linear transformation, then it admits
a unique extension to a derivation of G( V ) into G( V ) ,that is, a map
T, such that if g, h E G(V ) then T,(gh) = T,(g) h + gT,(h).
Problem 7. If T , S : V -+ V , show that [S, , T,] is a derivation and
that [ S , TId = [ S , , T J , that is, ( S T ) , - ( T S ) , = S,T, - TdSd.
Give an example to show that S,T, is not always a derivation.
If V is a d-dimensional vector space and G its Grassmann algebra,
then Gd is one dimensional. A linear transformation of Gd into itself
is simply scalar multiplication by an element of F. If S is the transfor-
mation then we denote this scalar by k ( S ) .
4.4.Existence of Grassmann Algebras 57
that is, f is linear in each variable when the others are fixed,
FIG. 15.
4.4. Existence of Grassmann Algebras 59
Proposition 1. fg E Gp+q.
Corollary. If ZI E V , then v 2 = 0.
Finally, addition, scalar multiplication, and the identity are defined
in the obvious ways.
Dimension and Generation by { e , V } . We have already shown that
G = CGi satisfies axioms (i)-(iii) for a Grassmann algebra. It remains
to prove (iv) and (v).
where ~ ( i ,...,
, i,) is the permutation needed to put i, , ..., ip in
ascending order.
Applying this formula with f = e, gives
Hence f = Xsspf(ws)e,.
Problem 10. Prove that the shuffle permutations are a cross section
and use them to write out all possible products for p , q 2. <
Problem 11. Suppose we define another product on the collection
of multilinear alternating functions on W by: if f E Gp, g E G*,
p , q 2 1, then f*g = apqfg, where a,,, is a nonzero element of F , fg
is the old product. I n order that G = CGi be a Grassmann algebra
over V with * as multiplication we must have upq = a and a condi-
?*.
tion equivalent to the associative law. Derive this condition and use it
to express up* in terms of a,, , q 2 , ..., u ~ , - ~Conversely,
. show that
aij may be assigned arbitrarily, and that when we let ulj = j +1 the
formula for f*g is
plus the (g) functions w, , s E P , dual to the basis {dx,} of GnLPat each
point of V .
Alternatively, if Rd* is the dual of Euclidean d-space Rd, then by
dualizing the action of Gl(d, R ) on Rd we see that GZ(d,R ) acts on
Rd* by taking the transpose of the inverse and hence also on G(Rd*)p
by extending t o homomorphisms. Gp is then the bundle associated
to the bundle of bases B ( M ) by this action [see 3.3(3)].
A dzflerential p-form 0 on B C M is then a cross section of Gp over
E, that is, a map 0 : E - t Gp such that T O 0 = identity. 0~ C" if
0 is a C" map.
Orientation. If E is a vector bundle over M we let E, = the set of
nonzero vectors of E. E, is an open submanifold of E.
If M is connected then Gd, is either connected or has exactly two
components. This follows easily from the facts that any path in M can
be lifted to a path in Gd, and each fibre of Gd, has exactly two com-
ponents (d = dimension of M ) .
M is orientable if Gdohas two components. Each component of
Gd, is in this case called an orientation of M . M is nonorientable if
Gd, is connected.
Lemma 3. A paracompact manifold M is orientable if and only if
M admits a continuous, nonvanishing, globally defined d-form.
Proof. Suppose 0 is a continuous, everywhere defined d-form with
values in Gd,. Let y be any curve in Gd, which begins and ends in the
same fibre GmdO= nonzero elements of Grnd.Since Grid is one-
dimensional for each n E M , we may write y ( t ) = a ( t ) O(T o y ( t ) ) ,
where a is a continuous real-valued function. Now a ( t ) is never zero,
so a ( t ) is of constant sign. Hence, the initial and final points of y are
in the same component of G,,,d,. Thus there is no curve connecting
the two components of Grndo,so Gd, is not connected and M is
orientable.
Conversely, suppose that 11.1 is orientable. Let {Ui} be a locally
finite covering of M by connected coordinate neighborhoods, {fi}
an associated C" partition of unity. Choose an orientation of M . Then
for any connected coordinate neighborhood U with associated coordin-
+
ates xl,..., xd, = dx, is a continuous cross section of U in Gag,
+
so either or -4 is in the chosen orientation. Thus we get +i defined
on U , with values in the chosen orientation and 0 = C fi+i defines
the desired nonzero C" d-form. Q E D
64 4. DIFFERENTIAL
FORMS
d(dx,) = 2 (-
i=l
9
1)j-1 dXi, . . . d(dXij) ... axip
= 0.
Thus we get d0 = XEp Cj”=l DZja,d x j dx, as the unique expression
for do.
There is one point that has been passed over; namely, when we
wrote (l), we should actually have put for the left-hand side
46 I(domaln o n domain { z i j ) )!
A similar remark holds for the right-hand side of (1). Thus we still
must show that if 0 and C$ agree on an open set U , then d0 and dC$ also
agree on U . T o do this we let f be the function with domain U and
value 1 at every point of U. Then f 0 and f + are equal and have the
same domain U. By (i) df = 0, so by (iv)
de lu = f d e = d ( f e ) = q+)
= f d + = d+ iU,
as desired. This concludes the proof of the theorem.
T h u s we have demonstrated the form which d must take in a
coordinate system. In order to show that d exists we would have to
show consistency in overlapping coordinate systems. This is actually
quite easy, since we can verify (i)-(v) for a coordinate neighborhood,
and then the uniqueness and consistency under restriction to smaller
domain will give the consistency on the intersections of coordinate
neighborhoods. Now we develop an intrinsic formulation of d which
will be used extensively later.
B(vl , ..., v,,,) = 2 (- 1)i-l v,e(v,, ..., vi-l , v,+~, ..., v9+,)
a
66 FORMS
4. DIFFERENTIAL
d0 =
SEP
2 D,,
i
a , dxj dx,.
= 2 D,,f 2 s g n b ) dxk(T/,l)dxs(Vnv,,
k RGK
> * * a 9 V37n(p+1,),
Problem 16. Write out in full the intrinsic expression for d0 when
0 is a 1-form and a 2-form. In the latter case use skew-symmetry
to put it in the form of a sum over the cyclic permutations of 1, 2, 3.
68 4. DIFFERENTIAL
FORMS
+
grad cfg) = g gradf fgrad g
curl (fX) = (gradf) x X +
fcurl X
+
div (fX)= (gradf) . X f div X
div(X x Y ) = (curl X ) . Y - X . curl Y
+*(W,,
..., b) = @(&t l , .*.,4 tn).
Notice that on the space of 1-forms at a point +* is just the usual
dual linear transformation to dd.
$* df = 4f o d+ = d ( f o 4) = d($*f) and
d(+* df) = d ( d ( f 0 4) = 0 = +*(O) = +*(d(df)). QED
Example. If 8 = u2 du, + u, du, + u, du, and rj : R2 + R3 by
4 = (sin u1 cos u 2 , sin u, sin u 2 , cos u,), then
= 2 (- i ) q q t l , ...,
i.1
tj-l , tj+l , ..., t P )ei(tj)
=o
since t j E D,, = annihilator of En&.
Conversely, if w ( t l , ..., t,) = 0 for every t , , ..., t, ED,, let
8k+l, ..., Od be a completion of e l , ..., 8, to a basis of G,,1, and let
e l , ..., ed be a basis dual to 0, , ..., ed. Then e k + l , ..., ed span D,,
and if we write w = CsEpa,8,,we have, if k < j, < <j , ,
4.9. Vector-Valued Forms and Operations 71
Hence each term of w must have a Oi with i < k, that is, w E W,.
Frobenius' Theorem [29]. Let E be a C" k-dimensional codistribution
on M . Let W be the function which assigns to each rn E M the ideal
W, generated by Em in G,. Then E is completely integrable if and
only if W is invariant under the exterior derivative operator, that is,
d( W )C W .
We shall show that this theorem is equivalent to theorem 1.6,
which says that a C" distribution is completely integrable if and only
w>
if it is involutive. First notice that the condition d( C Wis equivalent
to the assertion that for each m E M there is a local basis of 1-forms
8, , ..., 8, in E such that do, E W.
We have E is completely integrable if and only if the associated
( d - k)-distribution D is completely integrable if and only if D is
involutive if and only if Oi(Vj) = 0, i = 1, ..., k; j = 1, 2 implies
Oi([V,, V,]) = 0, i = 1, ..., k if and only if Oi(Vj)= 0, i = 1, ..., k;
j = 1, 2 implies dO,(V,, V,) = 0, i = 1, ..., k, (theorem 2) if and
only if do, E W , i = 1, ..., k (lemma 7). QED
Problem 21. Derive the integrability condition given in problem
1.29 for the 1-codistribution spanned by the 1-form P du, +
Q du, +
R du,. (Hint: Use problem 4.)
Problem 22. Let w be a p-form with p <
d - k. Prove that w is in
the ideal generated by linearly independent 1-forms e l , ..., 8k if
and only if we1 ... Ok = 0.
On the other hand, if p > d - k, then w is always in the ideal
generated by 8,, ..., $ k , so this shows: W is the space annihilated
by multiplication by 8, ... $k.
+
If is a g-valued p-form, w a g-valued q-form, where g is a Lie
+
algebra, then [+, w ] is the g-valued ( p q)-form given by
+
tangent spaces A,,,, F m, a n d z , , . Since F m = d m i A n L, J may
be extended to be a complex linear endomorphism of F m ,still
satisfying J z = -1. T h e dual of J extends to a derivation J* of the
complex Grassmann algebra G(F,n*).
w E G P + Q ( F m * ) is called an element of type ( p ,q) if J*w = ( p - q ) iw.
By allowing m to move over M , we define complex differential forms
of type ( p , q). We denote the set of these objects byXP,Q(M). Elements
of %'pq0 are called holomorphic p-forms.
Problem 25. For f a complex-valued C" function on M , define df and
show that it is a complex differential I-form.
Problem 26. For a complex coordinate system (z, , ..., zd), show that
the dzi , dZi at m, with 1, generate G (F m * ) .Express a form of type
( p , q) in terms of these generators, and show that
G'(Fm*) = C GP'q(Fm*).
p+q=r
Connexions
) O
d ~ ( X ( p )=) d ~ d p ( X ( e )=
14
5.1. Definitions and First Properties 75
FIG. 16.
+
If t E P, , we write t = V t H t , where V t E V , , H t E H,.
If X is a vector field, then we write also (VX)(p) = V(X(p)),
(HX)(p) = H(X(p)). Elements of H , are called horizontal.
Problem 1. Prove that X E C" if and only if V X and HX E C".
Since dn lx, is one-to-one, drr: H , w Mn(p, by dimensionality.
Hence, to every vector field X on M there corresponds a unique
vector field 2 on P, called the (horizontal) lift of X, with the properties
that for everyp E P, x ( p ) E H, and drr T ( p ) = X(rr(p)). From problem
76 5 . CONNEXIONS
H [ X , PI = [ X ,Y].
where xi
E g. Then the f i E C" [25, prop. 1 , p. 881, s o $ ( X ) = Cif,Xi
E C", proving (iv).
T h e properties (ii)-(iv) characterize 1-forms of connexions in the
following sense.
4 is its
1-form. Hence, there is a one-to-one correspondence between
such 1-forms on P and connexions on P.
Proof. {t E P, I +(t) = O}. Then for t E P,, let Vt
Let H p =
= 7b(t)(p),so t Vt E H,, which proves property (ii) in the definition
-
FIG. 17.
FIG. 18.
Now
Dw = dw o H .
W X , Y ) = X + ( Y )- Y+V)- +([X,YI)
= X ( Y ’ ) - Y ( X ’ )- [X’,
Y’]
= - H+>+l(X,Y )
= - Q [+,+ l W , Y ) + @ ( X ,Y ) , (4.9)
as desired, since @ is horizontal. Note that X ( Y ’ ) = X(constant
g-valued function on P ) = 0, and Y ( X ’ )= 0 similarly.
(ii) X E g, Y horizontal. Let X‘ E g be as in (i):
& (X, Y ) = X + ( Y ) - Y+(W - N X ,YI)= 0,
82 5. CONNEXIONS
@(Z
Y) = 4 ( X , Y ) = X W ) - Y 4 V ) - N X , Yl)
= -MX, Yl);
so H is involutive if and only if [ X , yl is horizontal if and only if
@ ( X , Y ) = - + ( [ X , yl) = 0, which implies @ = 0, as asserted.
Problem 5. Let H be a closed subgroup of a Lie group G and consider
the principal bundle (G, H , GIH). Let C$ be a connexion and show
5.4. Existence of Connexions 83
= (-I)~Y,+,W(Y , ...,
~ Y,) since w is horizontal
= (-l)P(hX) W(Yl , ..., Y,)
= (-l)p+l ad X w ( Y l ,..., Y,)
= - 14,WI(Y1,.*.,y , , YV+I),
Afine Connexions
6.1 Definitions
using the invariance of the connexion under right action, that this
transformation is independent of the choice of b over m.
dn = 2
i?i,k
(g;luj)(gkiek),
so
wi(dRst)= Z<&zj.
Hence, w(dR,t) = g-lw(t), as asserted.
T h e proof of (i) is direct. Let yi , y j k be product coordinates on a
coordinate neighborhood of B ( M ) (3.2), and we have to show that
w(Dzl,,) and w(Dy,) are C". By (ii), W ( D ~ , =~0,
) so we need only
consider w(Dy,). But
d7r D J b ) = z y > ' ( b ) ej (problem 3.4)
j
BfMl
FIG. 20.
6.1.4 Torsion
T h e torsion form Q of an affine connexion H on B ( M ) is the Rd-
valued 2-form
S2 = Dw = dw o H.
94 6. AFFINECONNEXIONS
[W,E(Y)l = --h@(E(x), W )- W W x ) ,W ) ) ;
that is, the curvature and torsion give the vertical and horizontal
components of the bracket of two basic vector fields.
Proof. We show that w and 4 applied to each side yield the same
function, and this will be sufficient since these forms are parallelizing,
that is, are dual to a set of parallelizing vector fields. T h e right-hand
side is easy to calculate since AX is vertical and E ( x ) is horizontal:
+(right side) = -+(A@(E(x), E ( y ) ) )
= -@(W,
E(y)),
w(right side) = --w(E(Q(E(x), E ( y ) ) )
= -Qn(E(x), E(Y)).
T o apply 4 and w to the left-hand side we use the intrinsic formulas
for the exterior derivatives d$,dw (see 4.6).
= Q ( W ,W ) .
6.1. Definitions 95
is horizontal.
Proof. Notice first that
f v X DXi(4)= Ei(X(4).
Hence, if the image is horizontal we shall have on the image, and hence
everywhere by equivariance, [Ei, E j ] = 0, which implies that curva-
ture and torsion are zero, by the above lemma.
Conversely, if these forms vanish, then by theorem 5.5 there is a
horizontal manifold N , and E l , ..., Ed are vector fields with trivial
brackets. Pulling them down to M , we have by theorem 1.5 that
there is a coordinate system xl, ..., xd on M such that
Dxi = d r E i IN
This coordinate system will then have the required property.
Problem 1. T h e property of having torsion zero is invariant under
combinations of connexions. Using the existence proof in Chapter
5 show that every paracompact manifold has affine connexions with
torsion zero.
T h e proof that T,, is well defined is similar to the proof for R,,.
Also, for vector fields X and Y we have a vector field T x y .
6.1.6 Geodesics
Let y be C" curve in M . A vectorfield along y is a cross section of
y in T ( M ) ,the tangent bundle of M . For example, y* defines a vector
field along y . A vector field X along y is said to be a parallel
vector field along y if, for any u, v , X ( u ) is the parallel translate of
X ( v ) along y from y ( u ) to y ( v ) . In terms of the connexion on the as-
sociated bundle T ( M ) (see 5.4) this says that X defines a horizontal
curve in T ( M ) .
A C" curve y in M is called a geodesic if its tangent vector field,
y* , is parallel along y .
Notice that a geodesic is a parametrized curve and not just a point
set. And the only reparametrization of a geodesic which will again
give a geodesic is a linear change in the parameter.
6.1. Definitions 97
in the way that we did in theorem 1.4. So by that theorem the curve y
generated by the end of these lifts, which is above u, will have second
order tangent 2 [ X , Y ] ( b ) , with the first order tangent 0. Thus, by
lemma 5 the horizontal component will be a lift of 2 T s t ,which proves
the first assertion.
Similarly, the vertical component of 2 [ X , Y ] ( b )measures the second
order deviation of y from parallel translation, so A has second deriva-
tive equal to the corresponding transformation on Mm , which is
2R,, by lemma 5 .
w'=Y* /'
+r
,----
Y
. \
\
'
Rd
FIG. 21.
6.2. T h e Structural Equations of an Affine Connexion 99
Theorem 4. We have
dw = - 4 w + Q,
4 = - 4[4,41 + @
These are called the first and second structural equations of the
connexion. (See 4.9 for definitions of the forms &J, [+, 43.)
Proof. T h e second structural equation is simply the structural
equation for a connexion on a principal bundle (see theorem 5.3).
T o prove the first structural equation we prove the following more
general result.
Combining this with the Bianchi identity (theorem 5.4) gives the
afine Bianchi indentities: D@ = 0, DSZ = @u.
For the vector fields E i ,Ejk this theorem gives the formula
[Ejk,
Ei] = 8ikEj.
We remark that this formula also follows directly from theorem
3.1 and hence can in turn be used to prove the first structural equation.
Problem 3. Give an alternate proof of theorem 7 by applying the
formula LA, = i(hX) d + &(AX) to the solder form w and evaluating
on the vector field E ( x ) . (Compare with problem 5.4.)
since II, has torsion zero. So (c$- II,)w = Q, and hence T must satisfy
the equation
-TW = s.
T h e last part of the theorem is now immediate. T o show that such a
T exists, we simply define, for b E B ( M ) , s E B ( M ) , , x E Rd,
-T(S)(X) = 8 G(S, E(X)(b)).
I t is not hard to verify that T is a C", equivariant, horizontal 1-form.
Further, for b E B ( M ) , s, t E B ( M ) , ,
-TU(s, t)= -T(S) U(t) + T(t) W(s)
Y w ’ ( Y ) = 0.
Thus the condition for y to be a geodesic is that the left side of this
equation is 0, and this becomes, along points of cr,
d
- W’(U*)
du
+ +yo*) w ’ ( u * ) = 0.
Inserting the meaning of w ‘ , +‘ we have
Theorem 9. +
Let andJ! , I be two connexion forms on B ( M ) , T their
difference form. Then these two connexions have the same geodesics
in M if and only if for every b E B ( M ) ,s E ,
T(S) W(S) 0.
Proof. If the two connexions have the same geodesics, then given
sE B(M), there is a geodesic y such that a lift 7 of y satisfies 7,(0)
= s.
(If s is vertical y = constant.)
Then by lemma 9
FIG. 22.
6.3. The Exponential Maps 109
e G b t = G ( l ,b, b, t ) . QED
6.3.1 Completeness
An afine connexion is called complete if all geodesics can be in-
finitely extended, that is, if each exponential map is defined on the
whole tangent space. This is equivalent to saying that the local group
of transformations of B ( M ) generated by any basic vector field E ( x )
can be extended to a global one-parameter group of transformation
of B ( M ) .We shall see in Chapter 8 that completeness of a Riemannian
connexion is equivalent to the completeness of the Riemannian metric.
= -+(Y - H Y ) w ( X ) + +(X)
w(Y - H Y )
= -+(Y) w ( X ) .
6.4. Covariant Differentiation and Classical Forms 113
Evaluating at b we get
H L ( X ) = L(X) + +(t)w(X(6)).
Applying b to H i w ( x ) gives V , X by (iii), which proves the formula.
Comparing with the result of lemma 9 easily yields:
=f(n6) X(b).
114 6 . AFFINECONNEXIONS
That is
fi = (fon)X
Hence
it@) = i ( f 0 n) X ( 6 ) + f 0 n ( b ) ix
= (tf)X ( 6 ) + f(.rrb) iX,
and applying b to both sides yields
Generalize this.
6.4. Covariant Differentiation and Classical Forms 115
= (tf)X ( m ) + f ( m ) vtx.
V,(fX)
Sometimes it is convenient to turn the covariant derivative around,
that is, for each C" vector field X defined on an open set U of M we
consider linear transformation T(X), defined on each Mm with m E U ,
by the formula 7(X) t = V , X . A connexion is then given by hypothes-
izing the existence of such T(X)'S which satisfy the following condition,
corresponding to (ii): ~(fx) t = f ( m ) T(X)t +
( t f )X(m)(see [65]).
We now give a direct relation between covariant differentiation of
vector fields and the connexion form on B ( M ) . It will depend on a
local cross section of M in B ( M ) . Let X,,..., X, be vector fields
defined on an open set U C M such that
x :m - (m,- J a m ) ,
is a cross section U + B ( M ) . For any vector fields Y
...I X&))
= C, fixiand
X on U we define
L(XY)= 2
(XfJ xi.
1
6.4.4 Coordinates
Then we have
d
vx,xi= 2 rijkxk,
k=l
and
For this reason connexions with torsion zero are referred to as sym-
metric connexions. T h e notation used is that given by Nomizu [66],
except for a change in the sign of T and R.
Finally, we define the functions Ti: and Rimik directly in terms of
the connexion. Consider the coordinate cross section x of B ( M )
defined by
x(n) = (n,X1(n), .-.,X,(n)>.
Then
R i w i i k = @irn(dx Xj , dx X k ) ,
and
rijk= + k j ( d x Xi),
where and QinL are the components of these gI(d, R)-valued forms.
118 6. AFFINECONNEXIONS
Thus derive the equation given for Rimikin terms of the rijk
by using
the pulled-down structural equation
d , + $ - L2 [ A *I + YJ.
Problem 18. Pull down the formula d@ = -[4, @] (corollary, 5.5)
to get the coordinate form of the Bianchi identity
FIG.23.
I)y = Ay + x,
X
for y E Rd.
T~ : GZ(d,R ) + A ( d , R )
by q(b) = (b, 0), qc(A)= (t :), and show that this gives a bundle
map of B ( M ) into A ( M ) which induces the identity map on the base
space M .
Let w be the solder form on B ( M ) and consider a connexion c j on
B ( M ) , with @ and 1;2 as the curvature and torsion forms.
T h e Lie algebra of A ( d , R ) may be considered to be
6.4. Covariant Differentiation and Classical Forms 121
6
Define an a(d, R)-valued form on ? ( B ( M ) )with the property that
y*$ = 6
(8 :), and show that may be extended by right translation
to a connexion form on A ( M ) ,also denoted by+. If 3 is the correspond-
ing curvature form, show that
Ri ern a n n i an M an i fo Ids
I Y I = f l l aY * I I l
x(y(t)) = st
0
[exp (-l/sz) sin 1/sI2ds, y ( y ( t ) )= 1'' exp (- l/s2) sin l / s ds.
' 0
Show that for this curve y o f-' is not a broken C" curve so that y
cannot be reparametrized with respect to arc length so as to remain
broken C".
We define a function p : M x M -+ R v {+ a} by
for (a,, ..., a d , m) E Sd-l x 0. Then by the fact that the expressions
in (1) are homogeneous in the q ' s , we have that (1) holds for
( a l , ..., a d , m) E Rd x 0. Now let y be any broken C" curve from
FIG. 24.
7.1. Definitions and First Properties 125
FIG.25.
as follows:
ei = z ( 4 U h el I *..,ed))ijvj(n)
fi = 2 ( Q / ( n , f * > ..., f d ) ) i f v M
i = 1, ..., d, where(n, e l , ..., ed) E B ( M ) , (n,fl , ...,fd) E F ( M ) . Pro-
ceeding similarly for a covering of M by coordinate neighborhoods and
defining right action exactly as in 3.2, we see that we have defined
principal bundle structures on B ( M ) and F ( M ) , and the local product
representations are compatible with the differential structure. In
particular, F ( M ) is a manifold and has structural group O(d),while the
A, being clearly C", show that the bundle structure defined on
B ( M ) is the same as that introduced previously. Further, we have
immediately from 3.4 that F ( M ) represents a reduction of the group
GZ(d, R) of B ( M ) to O(d), and F ( M ) is a submanifold of B(M).
M = M
FIG. 26.
This is true because the second plus the third term of the right-hand
side above is invariant under interchange of x and y , while the first
term changes sign.
When T is not horizontal then using TW in place of SZ in defining K ,
it is an automatic verification that 2(T(X) ~ ( y )~, ( z )is) given as above
in terms of K , so TW determines T. QED
Remark. Lemma 2 may be proved by showing the map T + TW is
one-to-one and then applying a dimensionality argument. T h e
lemma also holds if the forms T are taking values in a subalgebra
consisting of all linear transformations which are skew-symmetric
with respect to some non-degenerate symmetric bilinear form in Rd.
that is, Ql = TW +
Q. Thus by choosing r w = -Q we get a con-
nexion +o with torsion 0. If we started with +o instead of we +
would have Q, = T W , that is, the correspondence would then be
between connexions and torsion forms. QED
Problem 6 . Find a formula for +o in terms of w and dw using the
above lemma and the fact that +ow = -dw.
Problem 10. Take a circle of radius r and center (s, 0, 0), r < s, and
lying in the plane uz = 0 and rotate it about the u,-axis. Realize this
surface as the range of an imbedding of T 2 ,the 2-torus. Compute the
Riemannian metric.
4. The Riemannian d-sphere of radius r , Sd = {x E Rd+' I I I x j 1 = r } ,
is obtained by giving it the induced Riemannian metric from Rd+l.
Rotations are all isornetries, so S d has O ( d +
1) as a transitive group
of isometries.
Problem 11. Let xo , ..., xdPl be linearly independent points of
Rd+l all of which lie on S d .Show that there exists a unique isometric
imbedding of Sk,the k-sphere with radius r, in Sd whose range
contains xo , ..., xk , k = 1, ..., d - 1. This gives a chain
S1 c S2 C ... C Sd-l C Sd. Prove that any such chain can be mapped
on any other by an element of S O ( d 1). +
Problem 12. Prove the following facts about Sd:
(a) For any m E S d ,there is a unique isometry j , leaving m fixed and
such that dj, = - identity on Sdlll.
(b) For each nonzero t E Sd, , there is a unique great circle F ( t )
tangent to t , and S1(t)is invariant under j , , that is, S1(t)= j , ( S 1 ( t ) )
= Sl(-t).
(c) If X is a parallel vector field along S1(t),then a!jmX== - X .
(d) If m, n, p are equally spaced along S 1 ( t )and X ( m ) = t , use j , to
show that X ( p ) = a tangent to S 1 ( t )at p , and hence S 1 ( t )is a geodesic.
5 . Riemannian products. If M' and M" are Riemannian manifolds,
then M = M f x M" becomes a Riemannian manifold when the inner
products are defined as
(s' + s", t' + t " ) = (s', t')' + (S'', t")'',
that is, tangents to one manifold are regarded as being perpendicular
to those of the other manifold. M is then called the Riemannian
product of M' and M". For example, Rd is the d-fold Riemannian
product of R .
Problem 13. Show that the Riemannian connexion of the Riemannian
product of Riemannian manifolds is the affine product of the
Riemannian connexions of the manifolds. (See problem 6.22.)
6. Flat tori. T h e Riemannian product of S1with itself d times is
called the $at d-dimensional torus T d . No distinction is usually made
134 7. RIEMANNIAN
MANIFOLDS
between S1of different radii, and in a given torus all of the radii need
not be the same.
Problem 14. Show that the flat d-dimensional torus T dis flat.
7. Covering manifolds. If i : N + M is a covering map, M a
Riemannian manifold, then i may be considered as an immersion, and
when N is given the induced Riemannian metric, (i, N ) is called a
Riemannian covering of M . For example, the simply connected
Riemannian covering space of a flat torus is R d , since this is true for
d = 1, and the relation is preserved under products.
Problem 15. Prove the last remark, namely, that if i’ : N‘ + M ’ ,
i“ : N” + m’ are Riemannian coverings, then the Riemannian
product N’ x N” is the Riemannian covering of M‘ x M ‘ with
covering ma pi’ x i”.
Problem 16. There is a reverse process, also. If i : N + M is a
covering map and if N has a Riemannian structure for which the deck
transformations are isometries, then there is a natural induced
Riemannian structure on M which makes (i, N ) into a Riemannian
covering of M . For example, real projective space P a is covered by
S d and can be given a structure from that on the Sd of example 4.
8. Parallelixable manifolds. If X I , ..., X, are parallelizing vector
fields on a manifold M , then ( X iXi) , = (constant functions)
defines a corresponding Riemannian metric on M . Rd and T d are
special cases; more generally, any Lie group has such a metric.
Problem 17. Show that the Riemannian connexion of this metric is
the same as the direct or opposite connexions if and only if
[Xi , Xi] = 0 for every i , j .
Problem 18. T h e Riemannian connexion of a parallelization is the
torsion free connexion (problem 6.5) if and only if for all constant
linear combinations X,Y of the parallelizing fields ([X, Y], X) = 0,
which is equivalent to the condition
I +
([Xi Xjl, x,> <Xj [Xi > Xkl)
7 = 0
map J : B ( M ) -
Problem 19. Show that a homometry f : M-+ N , when extended to a
B ( N ) ( J h , el , e d ) = ( f ( m ) ,df(e,), ... df(e,))),
a * * ,
FIG. 27.
on
(z1 , z2) - (z1 exp 2 4 P , z2 exp 2 7 4 P )
S3= {(zl , ~ I
2 )zi E C , I z1 I' + I z2 1' z= I},
given by i(m, t, , ..., td) = (m,Rt, , ..., Rtd , It, , ...,Itd) is a well-
defined imbedding of the real manifold C F ( M ) and is consistent with
the previously defined inclusion of C B ( M ) in B ( M) . We shall write
C F ( M ) for i(CF(M)). T h e following problems show that the
Riemannian connexion on F ( M ) does not in general reduce to a
connexion on CF(M).
Problem 37. Prove that the following assertions are equivalent:
(a) T h e Riemannian connexion on F ( M ) reduces to a connexion
on CF(M),that is, if b E CF(M),then Hb is tangent to CF(M).
(b) T h e parallel translate of a holomorphic tangent is holomorphic.
(c) Parallel translation commutes with the projection P. (See p. 51.)
(d) Parallel translation commutes with J .
(e) Covariant derivatives of J vanish.
(f) If X , Y are vector fields on M , then V,(JY) = JV,(Y).
Problem 38. For s, t E .A,$, define Q(s, t ) = -( Js, t ) . Show that Q
is a differential 2-form on M and that
Q(S, t) = 21(PS, P t )
= *I(?, t).
(4 (4 =- (b) - (4.
Recall problems 6.13, 6.14 and the vector field formula for the
exterior derivative.]
Problem 39. A direct proof that M is Kahler if and only if the
Riemannian connexion on F ( M ) reduces to C F ( M ) is outlined in this
problem.
Let & , w!, be the components of the connexion and solder forms
on F ( M ) . Let J?= Q o d r . Define a linear transformation field onp
7.4. Examples and Problems 143
Use this to show that there exists a real closed form 1;2 of type (1, 1) on
CPd such that on Ui
Q = z d'd''jj. (Problem 4.28.)
- '
Show that the symmetric form associated with 1;2 is positive definite
and admits J as an orthogonal transformation, and hence show that
CPd is a Kahler manifold with respect to the corresponding Hermitian
structure.
A result of the preceding two problems is that every nonsingular
projective variety admits a Kahler structure.
CHAPTER 8
8.1 Geodesics
@=E$+@
~
$ = exp,* W,
d#i = 2
k=l
eki$k
d
doij = -
k=l
OikOkj + @ij.)
I I d expms I l2 =
= <#(s),
#(t)>
I I #(s) I l2 = 2 #N.
= +i(t)
(4 qS)= 0.
8.1. Geodesics 147
This says that the lengths of radial vectors are preserved under
d exp,, and that tangents to horizontal curves in F ( M ) are horizontal
tangents.
A C" rectangle in M is a map Q of a rectangle [a, b] x [c, d ] in R2
into M which can be extended to a C" map of a neighborhood of the
rectangle. Q(a, c) is the initial corner of Q , while the base of Q is the
(a, d ) /b,d)
FIG. 28.
Dz(wQ(D1), w Q ( W = 0.
f = r o expi]
Xi%,
T = 2 , the radial unit vector field defined on B - (0).
+
If q E M , , s E (M,,Jq, then we write s = sT sN where sT is a
multiple of T(q) and sN 1T(q).For t E M,, , b E B‘, we write similarly
+
t = tT t N ,where t T is a multiple of T ( b ) and t N _L T(b). We have
+
d exp,,s = d expmsT d explnsN, and from the corollary above,
d expl,,sN 1d exp,sT . Hence,
(i) d exp,sT = (d exp,s), . Also, from (b), p. 146,
(ii) I I d expnis, I I = I I S T I I .
Let [a, b] be the interval on which T is defined, and let c E [a,b] be
the smallest number such that P ( T ( c ) ) = r(p) = I a I. Now define a
curve 7 on [a, c] by
7
7)=ao---
IIPII Or.
150 8. GEODESICS
AND COMPLETE
RIEMANNIAN
MANIFOLDS
1
-
-
__ d a o dr o d exp,-'(T*)
II P II
hence,
FIG. 30.
FIG. 31.
pactness of E , = 8,,
the m k have a limit point m,and by passing to a
subsequence we may assume {m,} converges to m. Now
+
Now suppose y is a curve for which { y } 12 = I y I, 12 > 0. Then by
splitting y in half, we must have that the discrepancy on one of the
halves is at least k / 2 . By repeatedly halving we get a nested sequence
of parameter values s, < u, such that the discrepancy of y restricted
to [s,, u,] is at least k/2", and u, - s, = c/2,, where c = b - a,
[a, b] is the interval of definition of y. Let t be the common limit of
s, and u, , and let t , be a number in [s, , u,] such that
g(4 = P ( Y ( 4 , r(O))/ls I
and
d u n ) un - As,) sn = g(Un)(un - s n ) + ( d u n ) - d-4)sn,
we get
Riemannian Curvature
(a) then follows from the fact that 0 is a 2-form, and hence is
alternating, while (b) follows since @ is o(d)-valued, o(d) consisting
of skew-symmetric transformations of Rd.
In order to prove (c), we first notice that for a, b, c E Rd,
,.(g +
y , x, Y) ,
.(g y , x, Y ) +,.(g y , z, Y ) 4-,.(g Y , z, Y ) = 0
and hence,
g(., y , z, y ) +,.(g y , x, Y) = 0.
F(M) x O(4
p , 4 identification maps
4 1=7 , 1.
From problem 7.8 it follows that only the identity in H acts trivially
on m,and, hence, if X E ~X, # 0, then ad XI, # 0. Using this and the
fact that ad X is skew-symmetric with respect to ( , ) on g, show that
k( , ) is negative definite on 5. When M is irreducible with h # 0
this shows that K ( , ) is nondegenerate on g, from which it easily
follows that g has no proper Abelian ideals, that is, g is semisimple.
I t can be shown that if k( , ) is negative definite, or h < 0 in this case,
then G is compact (see [33], p. 122).
Problem 12. Pick f E F ( M ) such that ~ ( f )= 0, and consider the
corresponding imbedding of G as a closed submanifold of F ( M ) (see
problem 6.21). Show that G is a subbundle of F ( M ) with group H , and
hence that H may be viewed as a subgroup of O(d).
Problem 13. G acts on F ( M ) and so it makes sense to speak of
invariant vector fields on F ( M ) . Show that basic and fundamental
vector fields are invariant.
Problem 14. Let X E m, Y E b. Show that X is the restriction to G
of a basic vector field, and hence that the Riemannian connexion on
F ( M ) reduces to a connexion on G. This proves that the holonomy
group of M with respect to f is contained in the isotropy group H . T h e
converse is also true [28].Show that Y is the restriction to G of a
fundamental vector field, and in fact Y = hY, where is viewed as a
subalgebra of o(d). If X = E ( x ) Ic , x E Rd,show that
ad Y ( X )= E ( Y x ) IG.
Use this to derive a formula for the curvature of a plane section of M,,
in terms of the Lie algebra structure of g and the inner product ( , ).
This then serves to determine the curvature everywhere on M .
Problem 16. Assume that M is irreducible. Show that if X = 0, then
curvature is 0 ; otherwise, curvature is nonnegative or nonpositive
according as h is negative or positive, respectively.
Problem 17. Calculate the curvature (constant) of the d-dimensional
Riemannian sphere of radius r .
0
exp(X) = exp (x* 1 H is one-to-one.
Since V is defined only on 0,we shall often write V' for Vu*V.So the
Jacobi equation becomes V" = R,*u* .
174 9. RIEMANNIANCURVATURE
= - 2 @id(Q), G ( t ) )e i w (6.1.5)
U ( t ) = u*(O) + tV’(O),
and this corresponds to
FIG. 32.
w = p* , u = u* , K ( V )= K ( U , V )=
Corollary 2. If K( V ) <
0 for all Jacobi fields along u, then geodesics
from m near u pull away from u as compared to the corresponding
rays in M , .
m & Ks 0
m 6 K>O
FIG. 33.
(c) Define A(s) = J i L ( u )du, the “area” of the “circle,” and obtain
the formula
K = (12/T) ljz ( T S 2 - A(S))/S4.
Remark. I?. Cartan gives a generalization which tells the area of small
spheres, volume of small 3-spheres, etc. [ I 7, p. 2521.
Problem 24. Let M = Sd of radius r . Using problem 7.12 show that
the “circles” of “radius s in Sd” are circles of radius r sin sir in Rdfl.
Hence, calculate L(s) and prove that K = l/r2 for all plane sections.
Problem 25. Use the above to calculate explicitely the Jacobi fields
on Sd as follows: let y(s) be a geodesic with s arc length. Let x,
y E Sd,(,,, be tangents perpendicular to y*(O). Then identifying
tangents to S d with tangents to Rd+l,we may write
Y = 2 b,~,(y(O)).
Show that the formula for the Jacobi field X along y with X ( 0 ) = x,
X(0) = y is
X ( S )= cos S/Y 2 a i ~ , ( y ( s )+) sin s/r 2 ~JI~(~(s)).
Let M , ‘ M be d-dimensional Riemannian manifolds, m E M ,
’m E ’ M , f = ( m , f i , ...,f d ) E F ( M ) ,’f = (’m, ’fi , ..., ’ f d ) EF(’M),and
define as in 8.1
- -
8 = expJ*+ ’8 = exp,,*‘q5
- -
0 = expf*@ ‘0 = exp,,*’@
- -
I#= expJ*w = expef*’w.
9.5. Theorems Involving Curvature 181
FIG. 34.
by (a) and the remark before the statement of the theorem describing
(xBy P)’(O).
0
(c) and (d) #(x)
and ’+(’x)
o L satisfy the same initial conditions
along any ray. They will be the same if we can show that the second
derivatives are the same. But rays are integral curves of T , so we must
show that T 2 + ( X )and T*(’#(’X)o L) coincide. Deferring the proof
that T 2 # ( 2 )= @(T , x)
#( T ) , we have
TZ(’#(’X)0 L ) = ( ’ P ’ # ( ’ X ) )n L
= (’O(’T,’X)’#(’T)) 0L
= (’@(IT,’ X ) 0 L)(’#(’T)0 L )
10.1 Definitions
185
186 10. IMMERSIONS
AND THE FUNDAMENTAL
SECOND FORM
(a) Let M be the orbit of X,under O(d). Show that the isotropy
group which leaves X , fixed is O(r) x O(d - r ) , so M is diffeo-
morphic to Gd.,.
(b) If we introduce the metric ( , ) on S defined by ( X , X ’ )
= tr XX‘, then S is isometric to flat RD space, D = d(d +
1)/2, and
O(d) acts as isometries. T h e metric induced on M is thus invariant
under O(d), so M , with this metric, is a Riemannian symmetric space.
Problem 2. If we let O(d) act on S as in problem 1, and X E S has n
distinct characteristic values with multiplicities dl , ..., d, , show that
the orbit is the flag manifold FZ(d; d, , ...,, d,). (See problem 7.25.)
Establish the relation between the inner products on the blocks mij
and the characteristic values of X .
+
the difference form - +, which is horizontal and equivariant. Thus
T = r*+ - + +J is also horizontal and equivariant. T is essentially
the second fundamental form of the immersion, although the formal
definition will be in terms of a related object.
We now use 4, to define a connexion on F(M). Notice that
satisfies
(1) +a(ds-l(O)) = 0
(2) = ,for g E Id O(e).
From (1) and (2) it follows that there exists a unique form +o on
F ( M )such that
+d = +o 0 ds.
dT’(F(N,q b ) c MY(*)
and
= - $dW‘ - TW’.
Now, dw’ and -+do’ are Rd-valued, while TW’ is Re-valued, since
w’ is Rd-valued. Hence,
dW‘ = -+dw’
TW‘ = 0.
10.3 Curvature
dT = - [ d d + de 9 T] + r-
190 10. IMMERSIONS
AND THE SECOND
FUNDAMENTAL
FORM
T,y = b ~ ( 3b-’(y),
)
( E x Y ) ( m )= w3*)(~))*
( D x Y ) ( m )= w a Y ) ( b ) ) >
192 10. IMMERSIONS
AND THE FUNDAMENTAL
SECOND FORM
where ~ ( b=
) m. But
so
= b-’(TXTyx ~ TYTxx).
Hence,
TxT,x = - 2 H,(x,y ) T x z z .
a
Now
(Txzz Y )
? = Hz(x, Y ) .
Therefore,
(TxT,x,y> = - 2 Hz(x,Y Y .
2
Similarly,
T,x = - H*(X,X ) ~ i ,
2
so
- ( T , Txx,Y > = 2
a
K ( X , x ) HdY ,Y 1.
194 10. IMMERSIONS
AND THE FUNDAMENTAL
SECOND FORM
FIG. 35.
Proof. If x" is a horizontal lift, then this follows from the proposition
above. If x" is not horizontal, then let x' be the horizontal lift, and
notice that dG,(x" - x') is tangent to Mnti, so that its inner product
10.7. Hessians of Normal Coordinates of N 197
H f ( X , Y ) = r(xf).
Since df, = 0, it is easy to verify that H f ( x ,y) does not depend on
the choice of X and further that H , is symmetric.
Problem 9. Verify these facts about H f without using coordinates.
We now return to an immersion I : M -+ N. T h e second fun-
damental form will be related to the Hessians of certain functions on
M (see theorem 3 below).
Let V be a normal coordinate neighborhood in N at n = I ( m ) , and
let v , , ..., vUf be normal coordinate functions on V so that
V,(n) = (a/&,)(.) are an orthonormal basis of N , . Let ui= vi o 1.
Then a linear combination u = Xi aiui has a critical point at m if and
only if
so
Ifu(%,
x) = x x u = x(g, 0 I ) = dZ(x)g, = yg, ,
wherey = Y(n).
198 10. IMMERSIONS
AND THE FORM
SECONDFUNDAMENTAL
By proposition 1 we have
T,x = D,X - E,X
= D,X since 0 is a geodesic in M.
= D,Y
This last step, D, Z,g,(n) Vi= 0, is true because the vi are normal
coordinates in N so that Zigi(n) Viis the tangent field to the geodesic
in the direction of y.
Thus
fux, 4 = YE1
= (Txx, Vdn)>
= -H,(x, x), z = V,(n).
We have proved:
g = 2 qg, ,
a
dg(n) = z
i
ba dVi(?2), u = z
i
bpi , f = g lM! .
Show that H, = - H , , both being defined on M, = M,'
Problem 11. If S, , ..., S, are symmetric d x d matrices, then show
that Rd-+ Rf given by x + (x, -(S,x, x), ..., -(Sex,x)) is an
imbedding such that S,, ..., S, are the matrices of the second
fundamental form with respect to the basis D@), i <
d, of Rd,,
and basis Di(0), i > d , of the normal space (only at this one point).
Problem 12. Let f = uluz + +
u 3 u 4 ,g = u1u3 uzu4:R4--t R. Then
the intersection of f-l(l) ng-l(O) with a neighborhood of (1, 1, 0, 0)
is a 2-dimensional submanifold of R4.Find the curvature of this
submanifold at the point (1, 1, 0, 0).
is defined.
10.8. A Formulation of the Immersion Problem 201
i
0-P
I I d 4 x ) I I = I I d r , dZ’(f) I I
= I I d r , dp2 4 d J ( f ) I I
+
= I I W ’ dj d J ( f ) I I I I W” dj dl(f) I I
= I1 d d j d J ( f ) I1+ 0
= I I dnl 4 d J ( f ) II
= II dr1(f) I I
= IIXII.
-d e x d matrix
and
4” o(e)-valued.
d7 = -+’T - 74’’ + K w ’ ~ ’ ’ ~
&” = - +”2 + kw ” w ” ~ ,
Let M be connected and simply connected, N complete.
(4) d$o” = -u ~ u ~ .
+o(Yo*) = 0, +o”(ro*) = 0,
204 10. IMMERSIONS
AND THE SECOND
FUNDAMENTAL
FORM
and if
uOij = z
k
SOijkeOk Y
then
the functions sOijko yo are continuous and hence bounded, since the
domain of yo is a closed interval. T h e lift of yo to P, which we shall
call y, must satisfy
d$%Y* = E ( x ) + 2
i,k<d.i>d
xk(sijk y pZ)Fij = x-
Integral curves of such an X clearly exist by the condition of com-
pleteness of N , so the lift y exists.
It remains to show that r is 1-1. To this end we note that Q is a
principal bundle with group G and some base manifold M‘ and that
r : Q -+ B is a bundle map. Since curves can be lifted from B t o Q
they can also be lifted from M to M‘. Thus the induced map M’ + M
is a covering map, is 1-1 because M is simply connected, M’ connected.
This implies that r is also 1-1. QED
Remarks. (a) T o find local imbeddings of M into N it suffices to
construct forms such as uo , on a cross section of M into B,
because they can then be extended by equivariance to the remaining
part of the fibre; instead of a cross section we can also use M itself.
(b) Equations (3) and (4) are the structural equations for uo and
I,!I~’’. Equation (4) says that the o(e)-valued part of the curvature on B
is -uoluo. T o interpret (3) as a structural equation we consider the
adjoint action of G on its orthogonal complement in o ( f ) . Then
(-:otp) has values in that G-module, is horizontal, and satisfies an
10.8. A Formulation of the Immersion Problem 205
invariance of the type in lemma 5.3, part (ii). Thus (3) may be restated
as
Duo = 0, (3')
where D is as in 5.3 for the connexion
40 + 40".
(c) Equations (1) and (2) are algebraic at each point. Cartan [I61
has shown that they possess many solutions provided f 3 d(d l), 4 +
so that (1) and (2) are solvable locally. T h e solutions which Cartan
obtained are of a sort which he calls regular, and for analytic
Riemannian manifolds this regularity implies the existence of local
solutions of (3) and (4) also; this is the statement of the Cartan-
Kahler theorem [15, 431 in our context. Analyticity is definitely
required because the Cartan-Kahler theorem leans on the Cauchy-
Kowalevsky theorem, which is true only in the analytic case [34,
p. 81; 431.
Problem 13. T h e algebraic problem (1) and (2) is equivalent to
finding second fundamental form matrices Si, i = 1, ...,e, such that
the curvature K ( x , y ) of orthonormal vectors x, y E M , is given by
e
~ ( xy ), =
i=l
((sex, +
x> (siyt Y > - (Six,Y > ~ ) A.
Problem 14. Assuming Cartan's solution of (1) and (2), which only
depends on the curvature identities, show that if R is a tensor (not a
tensor field) satisfying those identities then there is a Riemannian
manifold having R as curvature tensor at one point. (Hint: let k = 0
consider the imbedding of problem 1 1 .)
T h e following problem gives an outline of the solution of (1) and (2),
but withf larger by d than Cartan's solution.
Problem 15. Let U = 4-linear functions on Rd satisfying the
curvature identities, S = symmetric d x d matrices, P = symmetric
4-linear functions on Rd, and T = 4-linear functions on Rd satisfying
the identities: for t E T , x , y , z, w E Rd, t ( x , y , z, w ) = t ( y , x, z , w )
= t ( z , w , x,y ) . P is a subspace of T ; T may be considered to be the
space of symmetric bilinear forms on V = the symmetric part of
Rd @ Rd. Let D = d(d - 1)/2, E = D +
d = dimension V .
When we speak of t E T being positive definite we mean it as a
symmetric bilinear form on V .
206 10. IMMERSIONS
AND THE FUNDAMENTAL
SECOND FORM
andG: T - Uby
G(t)(x,y , z, w)= t ( x , z, y , 4 - t ( x , w , y , 4.
10.9 Hypersurfaces
m,4 = ("(m),V d @ ) 4.
We say that M x Se is immersed as a manifold of partial rotation.
If M is a curve in R2 we say ] ( M x S e ) is a hypersurface of rotation.
We give M the metric induced by I , M x Se the metric induced
by I.
(a) Show that M x {x} C M x Se is a totally geodesic submanifold
for every x E Se and is isometric to M .
(b) Show that { m } x Se C M x Seis a totally geodesic submanifold
if and only if m is a critical point of vd , and in that case {m}x Se is
isometric to an e-sphere of radius vd(m).
(c) J is an imbedding if and only if I is also.
Problem 25. Products of spheres as hypersurfaces. If we let M = Sd
in problem 24, M imbedded in Rdfl as a translate of the usual Sd
so as to have v d > 0, then J gives us an imbedding of S d x Se as a
hypersurface.
(a) Find the curvature K ( X , Y ) in the cases:
(i) X , Y both tangent t o S d ;
210 10. IMMERSIONS AND THE SECOND FUNDAMENTAL
FORM
We now take u p the study of the relation between the lengths of the
longitudinal curves of a rectangle and the associated vector field.
If the base curve is a geodesic and the initial and final angles of the
associated vector field satisfy certain reasonable conditions, then the
derivative of the length with respect to the transverse parameter
is zero. Hence, in this case it is the second derivative which determines
how nearby curves compare in length with the base curve.
In general, these first and second derivatives of the lengths of
longitudinal curves are given by differentiating the length integral
under the integral sign with respect to the transverse parameter.
Thus we get what are called the Jirst and second variations of arc
length along the curve of a rectangle in the form of integrals.
A broken C" rectangle Q is a map [a, b] x [c, d ] -+ M such that there
is a finite partition u,, = a < u1 < ... < u,-~ < u, = b of [a, b]
for which the restriction of Q to ui] x [c, d ] is a C" rectangle,
i = 1, ,.., n.
213
214 11. SECOND OF ARC LENGTH
VARIATION
so
and
= 4Q(WwQ(D1);
(2) DAQ(D2)- D24Q(Dl)= DAQ(D2)
= -[CQ(D1),4Q(D2)l + OQ(D1 D2) 9
= @Q(D,, D2).
T h e inner products of the longitudinal and transverse fields of Q
are given by the corresponding inner products of wQ(D,)and wQ(D2).
Lemma 1 . Let Q be a C" rectangle, and assume that the tangent
to the base curve has constant length C. Then
1 b
ja
l'(0) = - - (wQ(D2),DlwQ(Dl)>(u,0 ) du
C
+ c1 {<wQ(D2),wQ(D1)>(b,0) - (wQ(D2),w Q ( 4 ) > ( a ,0)).
Proof. According to the last remark above,
D2 I I d W 1 ) I I (u, 0) = ~ 2 < w Q ( D 1 ) ,wQ(Dl)>""%0 )
1
=c <D2WQ(D1),wQ(D1)>(u,0 )
= c1 <DPJQ(D2) 4Q(D2)WQ(Dl)>wQ(D1)>(u, 0 )
-
= c1 <D1wQ(D2),wQ(D1)>(u,0 )
= c1 D, (wQ(D2), 0)
wQ(D1)>(.,
1
- c (wQ(D2), GwQ(D1)>(u, 0 );
we have used (1) and the fact that +O(D,) is skew-symmetric.
Therefore,
1 b
J
Z'(0) = - - (wQ(D2),D1wQ(D1))(u,
C a
0 ) du
+ c1 (wQ(D2),wQ(D1)>(.,0)li 9
FIG. 36.
N and P. Then a curve T has the property that for all such rectangles
with base curve T, Z'(0) = 0 if and only if T is a geodesic from N to P
which is perpendicular to both N a n d P.
Proof. T h e idea is that we are able to get rectangles with suffi-
ciently arbitrary wQ(D,) that the formula for Z'(0) = 0 will yield
DlwQ(Dl)(u,0) = 0, and then that T is smooth at the breaks, per-
pendicular to N a n d P at the ends.
11.1. First and Second Variation of Arc Length 217
and that the terms in the sum of corollary 1 are all 0. Thus I'(0) = 0
gives
S:f(u) / I DIWQ(D1)1 1 2 ( u , 0) du = 0,
<DPQ(D2),wQ(Dl)>z= (Dl<UQ(~2),
wQ(Dl)>)2 = 0,
because the associated field is perpendicular to T.
FIG. 37.
need only show that V and V' are perpendicular to T * at one point,
which we have already for V(O),hence for S,*o,V(0).Thus it suffices
to show E,,,,X is perpendicular to T,(O) = X(0). But
11.3 Focal Points and Conjugate Points [37,57, 59,60, 61, 731
FIG. 38.
FIG. 39.
so that V’ = W + 2. Then
230 11. SECOND OF ARC LENGTH
VARIATION
(f) I ( Y ) = <Z(b)- S r , ( b ) v ( b ) ,V b ) ) .
Hence by corollary 1, Y - Y , = 0.
Problem 15. If V , , ..., V , are C" vector fields along T and are
independent except at T(O), and V is a broken C" vector field along
T , with V = ZfiVi the expression for V in terms of the Vivalid on
I2 = ~ ( T ( c ) ~, ( b ) )Then
. +
I( V ) = 11( +
w,
V l [ O , C l = y I[O.Cl , I‘ I[C,bl = 0. Let I = I ( N , T ( b ) ) , I1 = T(C)),
V ) I,( V ) = 11(Y ) I,(O) = 0 by
corollary 2 to theorem 4.
Now choose intermediate manifolds for an infinitesimal deformation
F so that c is not one of the ui . Then F( V ) is smooth at c, while V has
a break at c, hence F( V ) # V and I(F( V ) ) < I( V ) = 0. By theorem
3 there are shorter curves from ~ ( bto) N .
FIG. 40.
and we do not know a priori that the number of focal points is finite.
Thus there are fields YiE 9, i = 1, ..., h_ , such that Yil i o , c , l is an
N-Jacobi field, YiIl c , b l = 0, and for each c E (0, b), {Yi1 c = ci} are
independent. Let %- be the linear space spanned by these Yi, and
+
let X = 2- 3,, where Z ois the null space of I .
Problem 17. Prove that the Yiare independent, so the dimension of
3- is h- . Moreover, the sum%- +
%,, is direct.
FIG. 41.
4 > Bu :I ( U J -
B,(z), where
I(u)
are linear transformations. A , and B, are continuous in u since
YY,#is continuous. However, there is a unique solution for
x of A,(y) + B,(x) = 0, so that Bil exists and is continuous. Thus the
solution Yu’(un+)= x = -B;’A,(y) is continuous in u, and finally so
is I,( Y,).
Now let .X+ be a maximal subspace of X , on which I , is positive
definite and let S+ be the unit sphere of X + with respect to I! . Then
a(t) = codim X+.Then f : ( u , Y ) -+I,(Y,) is a continuous function
on U x S+ which is positive on {t} x S+. Since S+ is compact, there
is a neighborhood U‘ of t such that f is positive on U‘ x S+ (see
lemma 6 below). Thus for u E U‘, I , is positive-definite on the space
H,(X+) spanned by H,(S+), and a(.) <
a(t). But a is nondecreasing,
so a(t+) = a(t).
T h e same argument applies to a maximal negative definite subspace
of I , on X , to show that i(u) 2 i(t) for u in some neighborhood of t .
Thus i(t-) = i(t).
T h e n f,,,
and fM are continuous.
236 11. SECONDVARIATION
OF ARC LENGTH
Theorem 9. f is continuous.
Proof. For {ti}, a convergent sequence in & ( M ) ,we wish to show that
limf(ti) = f(1im ti). Let yi = exp uti ,pi = yi(0), mi= f(ti) = yi(ci),
t = lim ti , y = exp ut, p = y(O), and m = f(t) = y(c). I t will be
sufficient to show lim ci = c, for then by continuity of exp,
lim m, = lirn exp citi = exp lim c,t, = exp (lim ci lim ti) = exp ct = m.
A closed geodesic is a geodesic segment for which the initial and final
points coincide; a smooth closed geodesic is one for which the initial
and final tangents coincide.
I t is intuitively clear that a homotopy class of loops based at p
should have a closed geodesic for its minimal-length representative,
242 1 1 . SECOND
VARIATION
OF ARC LENGTH
FIG.42.
FIG.43
FIG. 44.
1 1.7. Closed Geodesics 24 5
Y
FIG.45.
FIG.46.
sphere S(m, r ) of radius r < r,, about m satisfies the convexity coil-
dition: if y is a geodesic tangent to S(m, r ) at n = y(O), then for
sufficiently small u, p(m, y(u)) 3 r . If B(m, ro) is locally convex then
exp, must be one-to-one on B(0, y o ) c M n l ,for otherwise there would
be in B(m, ro) points on the minimum locus of m ; if y were perpendi-
cular to a geodesic T from m at a point ~ ( r beyond ) the minimum
point, then p(m, y(u)) < r for small u, since p(m, ~ ( r )< ) r.
T h e relation between the concepts of convexity and local convexity
is not as simple as it is in Euclidean spaces. For example, on a flat
cylinder a normal coordinate ball with diameter greater than half the
circumference of the cylinder will be locally convex but will not be
convex because it will contain opposite points, which have two
minimal segments. On the other hand, if the convexity condition fails
for S ( m , r ) , then B(m, r ) is not convex. T o show this let y be a geodesic
tangent to S(m, r ) at n = y(0) and having p = y(u) near n inside
S(m, r ) . Then a Jacobi field along y which points outward at p and
vanishes at n will have a corresponding rectangle having longitudinal
geodesics, of which only y is tangent to S ( m , r ) . T h u s there will be
11.8. Convex Neighborhoods 247
segments which start near p , pass outside S(m, r ) , and return inside
S(m, r ) at n. However, it might happen that B(m, r’) is convex for
some r’ > r .
T h e example of a flat cylinder shows that the following is the best
possible result of its kind.
When all such b, are positive definite we call B(m,r,,) strongly locally
convex, abbreviated SLC.
Problem 28. Let B(m,ro) be a normal coordinate ball, 7 a geodesic
from m,and z = ~ * ( r )where, r E (0, r o ) .Then the values of m-Jacobi
fields { V ( r )I V E X } form the tangent space to S(m, r ) at ~ ( r )Show
.
that:
(a) X is the space of S(m, r)-Jacobi fields along T , so m is a focal
point of S(m, Y ) of order d - 1.
(b) T h e second fundamental form H , of S(m, r ) is essentially
the same as 6 , .
Problem 29. Show by continuity considerations that B(m,r ) is SLC
for sufficiently small r . If B(m,r ) is locally convex but not SLC, then
B(m,rl) is not S L C for rl > r . Find examples to show that B(m,r )
can be normal for all r , SLC for Y E (0, a), locally convex for r E [a, b ] ,
and not locally convex for r > b, where a, b are arbitrary except for the
restriction a < <6 03.
Problem 30. A minimal submanifold is one for which the second
fundamental forms all have trace zero, that is, for every normal z,
tr S , = 0. Show that a compact minimal submanifold cannot be
immersed in a strongly locally convex ball.
Problem 31. Let B(m,ro) be a strongly locally convex ball and let N
be a compact manifold which is immersed in B(m,ro), giving N the
+
induced metric. Suppose 1 2 d i m N > d = dim M . Use the
following theorem of Otsuki [68,69] to show that there is a plane
section P such that K,(P) > K,(P).
Let V be a real vector space of dimension n. Suppose Q1, ...,Qk ,
k < n, are symmetric bilinear real-valued forms on V such that
for all u, ZI E V .
Then there is at least one nonzero vector U E V such that
Qi(u, u ) =: 0 for allj.
Prove this theorem in the case k = 1.
Problem 32. Show that every ball in Euclidean d-space is strongly
locally convex. Then apply the preceding problem to show that a flat
11.8. Convex Neighborhoods 249
+ s”
0
((V’, V ’ ) - ( S V , V ) )du.
FIG. 47
Proof. Let V be the m-Jacobi field such that V(b) = d expms, W the
n- Jacobi field such that W(b)= d exp,t, f = ( V , V ) , andg = ( W , W ) .
We wish to show thatf(b) 3 g(b). It will be sufficient to prove
(a) ( f i g ) ( O + ) = 1,
(b) f‘/f2 g‘ig on (0, b], because then fig will be nondecreasing.
T o prove (a) we note that there are constant vector fields A and B
on the rays to x a n d y in Mr,Land N , such that A ( x ) = s, B ( y ) = t ,
V = u X , and W = u Y , where X = d expmA, Y = d exp,B. Then
fig = ( X , X > / ( Y , y>.But ( X , X)(O) = ( A , A ) = (s, s) = ( t , t )
= ( Y , Y)(O).
T o prove (b), suppose first that there is no conjugate point of m on
o((0,Y]). Thenf(r) f 0, so we may let X = V/f(~)l/~; similarly, let
Y = W/g(r)1/2.Then X and Y are Jacobi fields, so
2 52 11. SECOND OF ARC LENGTH
VARIATION
= ( Y , Y>’(r)
= (g‘/g)(r)s
It now follows that f ( r ) 3 g ( r ) as long as there is no conjugate point
of m on ~ ( ( 0r ,] ) .However, this was only needed to enable us to divide
byf(r)li2, so the inequality and continuity now give that f ( r ) >, g ( r )
for r E (0, b].
which follows from the fact that the basis of #-vectors generated by an
orthonormal basis of vectors is orthonormal. [Express Yi’(r)in terms
of an orthonormal basis which includes Y l ( r ) ,..., Y,(r) and expand.]
Using this to differentiatef2 yields the relation
(yi’,yi> ( r ) < f0
((g’)2 - ~ ( y >* E i ) g 2 ) du.
so the result now follows from the fact that J ( r , x) (urisin ~ r ) ~ is- l
monotonic in r , and similarly for the other case with b.
Remark. T h e same conclusion obtains if curvature has range [a2,b2],
which is a result of Gunther [32].
T h e reader may find it instructive to translate (i) and (ii) below into
their coordinate forms and verify that essentially the same theorems are
found in standard references [26].
Let F be a C" map, F : Un x Urn-+ T(Rn),where Un and Urnare
open subsets of Rn and Rm,respectively, T(Rn)the tangent bundle to
Rn, and such that F(u, u') E Runfor every (u, u') E U n x Urn.Thus F
gives a C" vector field on Un for every u' E Urn; or, in classical
terminology, a system of n first order differential equations in n
unknowns, depending in addition on a parameter u' E Urn.
(b) rj(&rj(4 u, 4, +
u’) = 4(s t , u , 4.
Bibliography