VI

Linear Connections

VI1 Affine Connections

It has been seen that at each point ofa differentiable manifold M
thereis an n-dimensional tangent space
T(M). Hence the tangent spaces T,M) and T,(M) at pointp and gare isomorphic, since they are both n.
dimensional. However, in order to obtain a definite isomorphism relating T, and T, it is necessary to
introduce some additional structure on the manifold called a connection. This connects the tangent
spaces at different points on the manifold.
Such aconnection would be obtain for example, by anon-singular linear mapping of abasis (e,) of T,
into a basis fe, of T,. The most suitable tool for developing this line of approach is the theory of fibre
bundles. However, the following treatment is most useful in the study of Riemannian geometry.
Suppose Mis a smooth manifold of dimension n. Then we have defined the tangent vector of the
manifold Mat a point p as a derivation
X:C(M) ’Rsuch that
Xf:p ’ Xf(p) = X,f.
vector fields on M.
In this section we introduce a similar operation on the set x(M) of all differentiable

Definition VI.1.1l: An affine connection ona manifold Mis a R-bilinear mapping

V:x(M) x x(M) ’ X(M),
written V(X, Y) = V,Y or Vx() and satisfies the following properties:
(1) Vay.B, Z= aVy Z + BV, Z,
(2) VagrZ =f VyZ+gV,Z,
(3) ,Y+gZ) = f,Y+(f)Y +gV Z+ (Xg)Z
for all a, Be R, f. ge C"(M), X, Y, ZeX(M). connection or Koszul connection. The
The affine connection V defined above is also called a linear
the vector field V,Ye X(M) Is
operator V,, is called the covariant differentiation with respect to X and
called the covariant derivative of Ywith respect to X. argument.
tensor inthe first
From property (2) it follows that V, Y=f,Y, which implies that is a
property (3), it follows
Thus if X,e T,M and Yez(M), then Vx,Ye T,M is defined. Again from argument, and hence
that V, (SY) =f,Y + (Xf) Y, which implies that Vis not a tensor in the second
Thus connections
Vx,YeT, Mdepends only on the values of Yin an arbitrary small neighbourhood ofp. =U0
and if either X
can be restricted to open subsets of M. It can be easily seen that if X, Ye Y(M)
INEAR CONNECTIONS 195

ye0 onan open subset Uc Msuch that isa connection satisfying propertiess(1)-(3),
( then the vector
field V,Y= 0on U. Hence if X, Xe z(M) suchthat X, = X, pe M, then for every vector field
r(,), =(r),
Donerty (1) implies thatV is R-linear in both variables as Xf= 0for f is aconstant function. But V
isC´(M-linear in the first variable but not in the second variable.
Amanifold Mwith a given affine connection is called an affinely connected space. We now give an
example ofaconnection.

Eample VI.1.1: Suppose M=R and X, Ye x (R").

Le Y= ) b where b'e C(R') and u', ..,. u" are the local coordinates of R".Given

X. Ye X(R), we define the mapping

V:x (R") xx (R") -x(R)

by V,r-(xo'))a'.
du' du'u'
/=l i,j=l
(1.1)

where X= a' ae C(R"). Then we shall show that V is a linear or affine connection
j=l

on R'ie., Vsatisfies the properties (I )-(3) in the definition of aconnection.

Suppose Z
=c e C"(R"). Then for a, Be R, we have
k=l

VaregyZ =Varsgre*

-~(a(xe')+ B(Ye'D:
k=l

atB(Yet)
k=l k=l

k=l. k=l

le., Var-gy 2Z = aV,Z + BV, Z, a, BeR.

 DIFFERENTIAL GEOMETRY'OF
196 MANIFOLDS
BVZ. Thus property (1) is verified
Similarly we can show that V,(aY + BZ) = a Vy?+
S,ge C*(R")
Again, Vg+pyZ = V

- )((x +g
k=l

k=l
du

+
k=l
duk k=l

=E(xe*) k=l k=l

=fV,Z+g. which is property (2).

Finally, :Y+ gZ )= V + g

+
k=l Ju' k=l

-)(x6'+ f(XO)xg
k=l Qu' )e' + g(Xe*) k=l
(by Leibniz's rule)
-) ( )B' du'
i=l
+
i=l du' -+ k=l
(Xg )e Eg(Xe*)
k=l

/=l i=l
dutt(Xg))c dutt8 (Xe).
i=l k=|

- (Xf)Y +fV,Y+ (Xg)Z +g VZ.

Thus VfY+ gz) =fVxY+ (Xf)Y+ gVyZ+ (Xg)Z,
which is property (3).Hence Vsatisfies all the properties of an affine connection. Consequently, Vdefined
by (1.1) is a linear or affine connection inR".
If ¼ is a connection on M and (U,; u', ., u") is a coordinate chart, then we set
LINEAR CONNECTIONS 197

and write
du'

(1.2)
k=l

(1.3)
ie., k=|

where are smooth functions on U, called the coeficients of the connection V with respect to thelocal
coordinates u'. The functions r ,EC (0) are said to be the Christoffel symbols ofthe linear connection
suchthat UDVto.
Pinthe given local coordinates u'. If (V, u') is another coordinate"system of M under
voidset, then it can be verified that the transformation formula for the connection coefficients
achange of coordinates is given by
n

+ (1.4)
p.q,r=l q=l
is not a tensor
where are the components of r# in the coordinate chart (V. ). This implies that
invariant property.
field on M and hence the vanishing of connection coefficients at a point in M is not an at a
connection coefficients are zero
In other words, there may exist a coordinate system such that the
then V, Y also vanishes at
given point of the manifold. If X, Ye x(M) and X vanishes at a point pe M,
that point. connection. It can be easily seen that
Now aquestion arises whether or not a manifold always admits a
However, any paracompact HausdorfY manifold
amanifold always not necessarily admits a connection.
always admits at least one affine connection.
M and f be a diffeomorphism
Definition VI.1.2 : Let V be an affine connection on a smooth manifold
be defined on Mby
of Monto itself.Then a new affine connection V can
X, YEz(M).
andf is called an affine
The affine connection is said to be invariant under f if V=V
transformation from one manifold onto another
transformation on M. Similarly, we can define an afYine
manifold.
field VyY. Suppose u',.., u" are the local
We now findout the local expression of the vector
X, Y can be expressed as
oordinates of a point pe M. Then the vector field
n
where X, YeC(M), i,j =1, 2, .., n.
X=)x du'
i=l
j=l
du'

Hence we have M
M.
VY = V
i=l
198 DIFFERENTIAL GEOMETRY OF
MANIFOLDS
by property (2), as X'e C"(M).
n n
a(y'))a
(3)]
du' Jou' lby property
+
i=l j=l

i,j=l k=l
by (1.3) and writing the dummy index
i,k=l

iby k, in the second term.

-+Sx
i=l du' duk

x'y'r) +) x du' du (1.5)

This shows that (,Y), depends on X(p), Y(p) and the derivatives of Y. The relation (1.5)
du'
is the local expression of the vector field V,Y. We now give another
tion : important example of linear connec

Example VI.1.2 : Let MCR" be a smoothly imbedded n-manifold. If pe M,

then canonically
T, (R")= R". Also
T, (R°) = T, MO T, (M).
where T, M =(Xe T,(R") : XI T, M}.
perpendicularity being defined by the Euclidean metric in R". Then
TM = T; Mc T(R)IM
pe M
is an (m -n)- plane bundle over Mand
T(R", = T(M) O T (M)
is a canonical direct sum bundle
decomposition. T (M) is called the normal bundle of M in R", Then the
canonical projection
p:T(R")|, T(M) T (M) ’ T(M)
is a surjective homomorphism of bundles.
Let X, YeX(M). Given pe M, there is a
neighbourhood U ofp in R" and extensions ,Y of
XI(UOM) and Y\un M, respectively to tields on U. Then (V,Ý), depends
only on = X, and on
P
. If y:(-E, e) ’ M is asmooth curve
such that y(0) =p, and ifwe represent X = [Yl., then
 199
LINEAR CONNECTIONS

dt dt l=0
l=0

ù. Therefore VY isa well defined

This dependsonly on Mum but not on the choice of extension of
elementof r(T(R"),) andit can be easily seen that it satisfies allthe properties
(1) - (3)ofa linear or
affine connection.

y is a smooth map Y: (a, b) ’ 1(M)

Let Y: (a, b) ’ M be a smooth curve. A vector field Yalong
such that the diagram commutes.
T(M)

(a,b)

M
Figure 44

Then the restriction of the vector field Ye x(M)

We denote by x(y)), the set of allvector fields along x

given by (Yly, = Yyo and the velocity field is Ý)= dy)

to y) is dt
associated covariant derivative or absolute
Suppose V is an affine connection on M. Then an
differential is an operator

dt
:x()) ’ x(r())
following properties :
defined for every smooth curve Yon M and satisfies the
VX VY
is R-linear i.e., (aX +BY)= a +B
dt
a,BeR, X, Ye X(r(),
dt dt dt

+f for all feC (a, b), and for all Xex (Y());
dt

(ü) if Ye x(M), then

(Y1,o =V,o()=V, ¬ly (M), a<t<b.

dt d
covariant
Then it can be seen that to each connection V on M, there is associated a unique
considered V. as the operator of covariant
derivative Hence in the Definition VI.6.1, we have
dt
 200 DIFFERENTIAL GEOMETRY OF
MANIFOLDS
denvaive along X. In terms of local coordinate sVstem u', it can be easily seen that along Y),

dt k=l dt
+
i,j=l
x'yr (1.6)

where
Jy (t)

and

Definition VI.6.3 : Suppose M is a manifold with a connection V, and y: (a, b) ’ M IS a smooth curve
on M given by : u'= u'(). Let X() be a tangent vector field defined on y
i.e., XE X(Y)) given by
x)-x i /y ()
Then we say that X() is parallel along yifits covariant derivative or
i.e., if absolute differential along y is zero,
VX
d
=0 on y.
Ifthe tangent vectors of acurve Yare parallel along y,
then we call y a self parallel curve ora geodesic.
If y: u'= u'(), then Ý() = du' and hence (1.6) reduces to
dt
VX
+
dt dt dt
i,j=|
If X() is parallel along y, then we have
du'
dt
-+X'T dt
=0
(1.7)
which is a system of first-order ordinary
on y gives rise to a parallel tangent differential equations. Thus a given tangent vector X at any point
vector field, called the parallel displacement or
Xalong the curve yY. parallel translation of
Definition VI.6.4': If V is a connection on a smooth
on M such that e (a, b) and manifold M and y : (a, b) M is a smooth curve
X,e1Yn, (M) 1.e., Xy 1s tangent to M at y (t, ) then there is a
parallel vector field XE x( ()) such that X()= Xo This vector field X) is called the unique
transport or parallel translation of X, =X(t,) along y. parallel
If y is a geodesic, then its tangent vector
LNEARCONNECTIONS 201

X() =

parallel:along y. For a geodesic curve Y we have

ia
V,X=0, where X() = Y(). Hence for a geodesic
Curve Y, we
must have (follows from (1.7)
d du k du du
dt dt dt
-r dt
= 0

d'ut du du ! (1.8)
d2 t = 0,
dt dt
here Einstein's summation convention have been used. This is a system of second-order otdinary
Aferential equations. Hence there exists a unique geodesic through a given point of M which is tangent
point.
toagiven tangent vector at that
For the Euclidean space E", we have =0 for all ii. k= 1. 2. ....n and hence (1.6) reduces
to

VX
n dyt
dt dt dk
the usual derivative.
Thus in Euclidean spaces the covariant derivative coincides with
constant if V, Y= 0 for all
Definition VI.6.5 : Avector field Y on a smooth manifold M is said to be
peM and X,e T,M.
on any small open subsets of M. If
Generally, there do not exist such vector fields even
7: (a, b) ’ M isa smooth curve on M
with connection then we have defined

VY
Vdro (Y) =Vyo,YeT,(M), where p=rl), a<t,<b.
dt dt

V Ye T, M. Thus
Then dy()| Hence the image of X, under the connection Vis given by
dt

asubmanifold, then for any tangent vector field Yex(M), at each point pe M, we have a
if Mc R" is
Imear mapping X, ’ VY of T,M’T,Msuch that
the map V: T,Mxx(M) ’ T, M given by V(X,, ) = VY is a R-linear mappingwhich satisfies
(@) for fe C"(M),
V,) = (X,DY, +f(p) V Y,
(ü) for X, Ye x(M), [X, Y} = -V,x.displacement along curves came first in studying difter.
Although, historically, the notion of parallel If
preferable to start from the notion of covariant derivative.
entiation on manifolds, but nowadays it is
and if X,El,M is given arbitrarily, then there
pY(0) is the initial point of the curve y(t), a<!<b, that X, has the given value. Therefore it
vector field X, along Y () such
ASS a unique constant
 202 DIFFERENTIAL GEOMETRY OF
MANIFOLDS
followsthat given a piecewise differentiable curve y(). there exists an isomorphism t, : T,M ’Ty, M
such that t,(X) is a parallel vector field along y ().
In Chapter lIl, we have already defined a tensor field K, oftype (0, r) at a point pe Mas a R-linear
mapping
K,:TMO ..... T M’R
such that K,(X,,
a, X,, , X,) = a,K,(X, .. X,, .. X)
...,

for all X,e T, M. a eR, i= 1, 2, .. r.

We now introduce the global definition of the tensor fields over the manifold.
Detinition VI.1.6 : A
tensor field K oftype (0. r) on the manifold M is a C(M)-linear mapping
K: 2(M) X X(M) X .... ...x x(M) ’ CM) such that
K(X,, f X,, ... X) =f, K(X,, .. x,, . X)
for all EC(M), X¬X(M), i =1, 2, .., :
Definition VI.1.7 : Atensor field K oftype (l, ) on the manifold M is a C(M)- linear mapping
K:x(M) x x(M)x XX(M) ’ X(M)
such that K(X,, ...f X,, .. X,) =f, K(X, . X, .., X,)
for all feC"(M), XE X (M), i, = 1, 2, .... r.
Any such mappings in the above definitions can also be considered as a tensor field of type (0, r) or
(l, r).

Defirifion VI.1.8 (Covariant Derivative of atensor field of type(, s)): Suppose M is a smooth mani
oldand K(X,, ... X, w,, .. w) is a tensor field of type (r s). Then the covariant derivative of the
tensor field K is a tensor field of type (r. s+ l), defined by
(V K) (X,, .... X,, ,, ., 0; X)/
= (V, K) (X, . X,.
= ,K (X,.

-}xex,. X,,. VyX,, X,.,, . X,, @, .., @, )

K (x,. ..., X,, 0,. .... @, V,.

jl ., Q,) (1.9)
for all XE x(M), 0,¬ X(M), i =l, 2, ..., r,j =l, 2, ... s. If VK = 0, then we say that K is
convariantly constant and conversely.
Again, for any fe C(M), we define its covariant derivative along X as
(1.10)
The covariant derivative of any tensor field K of type (1, s), which is a vector-valued tensor
field,
expressed as K(X,, . X, ), is avector-valued tensor field of type (1, s+) and defined as
(VK) (X,, . X,; X) =(, K)(X,. .. X)
=
V, K(X,. .. X) -)K (X,, .., X,. V, X,, X,., . .. X,) (1.11)
jH
 NEARCONNECTIONS 203

manifold with an affine X, 0.

Let Mbe a connection V. Suppose pe M, X, Yex(M) and to y()
fYO isan integral curve of Xthrough p = y(0) and t, is parallel translation from p
thej
wth
respecttothe curve Y, 1.e., t,:IM. ’Ty Msuch that ,(X,) is parallel vector field along y)
can shown that
it
then
(1.12)
(,), - lim
(’0t
{,Y-Y,}
of(1.12), we can write
Byvirtue
1 (1.13)
(9,K), = lim
(rKy-K,}
Kis atensor field of type (r, s) on the manifold M. Also we can put
1 (1.14)
= lim (flr ()) -fp)}.
(’0

of the tensor
pEM. Hence the covariant derivative operator V, can be extended to alinear mapping
dohra of mixed tensor satisfying the following properties:
is a derivation of the mixed tensor algebra
.commutes with
contractions.
by
fo is al-form on M and XeX(M), then the covariant derivative of o (X) is given (1.15)
(7,0)(Y) =V,w (Y) -0(7,), where Ye x(M).
derivative .Y and the Lie derivative
We now give the distinction between the covariant
4,Y= [X, ). In fact, we have
0) £gY*fE, Y but VY=fV,Y for feC(M)
) EggyZ # fEyZ + g£,Z, but V-yZ =first fV,Z + gV,Z,
this implies that V, is C(M)-linear in the argument while £, is not.
derivative the type of a tensor is not preserved as
3) £, is type preserving operator while in covariant
its covariant rank increased by one.
Example VI.6.2.(ii)
4) If X,YEX(M),then Lie derivatives satisfy the identity (see
=0 (as for example, for coordinate vector
where the right hand side means £,£, - t,£, Hence if [X, Y] the covariant derivative V. This failure
general for
fields) then £, and £,commute. But this result fails in
tensor of the manifold, which plays a key role in
is measured by a tensor field, known as curvature
differential geometry. In fact, generally,
V,V,- V,V,- Vx = R(X, ) 0
then
and hence ,,V, - V,7, * 0, even though ([X, Y] =0,
V,,Z- V,,Z-Vx , Z= R(X, )z [X, Y} =0, by our assumption.
LC., V,,z- ,7,2 = R(X, Y)Z 0, since
M2 Tersion Tensor of an Affine Connection
YEX(M), we have defined the connection on M as
Ouppose M is a smooth manifold. Then for any X,
 204 DIFFERENTIAL GEOMETRYYOF MANIFOLI,

VYEx(M). If we interchange Xand Yin the expression V,Y, then it gives rise to introducc avector
field T oftype (1,2) defined as follows:

Defnítion VIL2.1 : The mapping T:x(M) ×X(M) ’ X(M) given by

T(X, Y) = , Y- ,X-(X, Y). (2.1)
Then the vector field T(X, )EY(M) is called the torsion tensor field of the affine connection V, for all
X, Ye x(M).
From (2.1), it follows that T(X, Y) =-T(. ) and hence Tis skew-symmetric tensor ficld on MWe
shall show the following:
() T(X, Y) is R-bilinear,
ü) T(X, Y) is C"(M) -bilinear.
In fact, fora, BeR and X, Y, Ze x(M) we have by definition
T(aX + BY, Z) = Vak-pyZ - V,(aX+BY) - (aX+BY, Z
= a V, Z + B ,Z- a V,X- B,Y- [ax, Z] - [BY, Z]
= a (V,2 -V,X- [X, Z]) + B(V,2 - VY- [Y, ZJ)
= a T(X, Z) + BT(Y, Z).
Similarly, we can show that
T(X, aY + BZ) = aT(X, ) +BT(X, Z).
Again, forf,ge CM) and X, Y, Zex(M), we have
T(fX + gY, Z) = Vgy2 - V,(fX +gY)- ([X +gY, Z]
-fV,2 +g V,2 -(/ V,X +(2)X +gV,Y+ (Zg)) - fX, Z] - IgY, Z)
-f(V,2- ,X)+g(9,2- V,) - (4)X- (Zg)Y- f IX, Z) - (2)X)
- (s[Y, Z]- (Zg) Y) [by (6.24) ofchapter IV]
- f(,Z- V,x- (X, Z]) +g(V,Z - VY- [Y, Z))
-f T(X, Z) + gT(Y, Z).
Again,
T(X, fY +gz) = ,Y+gZ) - -[X. f Y+gz).
-f Y+ (X)Y + g V,Z +(Xg)Z - f V,X-g V,X- [X, fY] - [X, gZ)
-f(9,Y - 9,X) + g(V, Z- V, X) + (X) Y+ (Xg)Z- f[X, Y]
+ (X) Y) - (g[X, Z] + (Xg)Z)
= fT(X, ) + gT(X, Z). [by (6.25) of chapterIV]
Thus T is C(M)-bilinear, although Vis C(M)-linear in the first variable only.
In terms of local coordinate system u,..,u" of M, the tensor Tis given by

T(X, ) = TX'Y/ .

where X =X and summation convention have been used, Then T', are called the
du'
components of the tensor T.
NEAR CONNECTIONS 205

DefinftionVI.2.2 : An affine connection Von asmooth manifold Mis said be symmetric ifits torsion
ensorvanishes identically i.e., if V,Y- 9,X = (X, Y] for allX, Ye X(M).
In this case
we say that the affine
connection is torsion-free. Atorsion-free affine connection
smooth manifold M.
alwaysexists on a
In terms of local coordinates u! u", the components T, of T is given by

du'
du' -Va du' du''du'

=(T,-r) since =0 in any coordinate chart (U, u'),

du' du!
where we have used Einstein's summation convention.

Since are linearly independent on U, we have

(2.2)
T = (T- r),
If the connection V defined on any open subset UCM is symmetric, then from (2.2) we get
(2.3)
Thus we may state the following:
if and only if, in every local
Theorem VI.2.1 : The affine connection Vis torsion free or symmetric
connection coefficients r is symmetric
coordinate chart (U, u', .., u"), the Christoffel symbols i.e,
with respect to subscripts, i.e., = ri for all i, j. k = 1, 2, ..,.n.

components T', of T in the coordinate chart

From (1.4) and (2.2), it can be easily seen that the
(V.v') of M is given by
(2.4)
T =TP
for the components of a (1, 2) -type tensor and
This shows that T' satisfies the transformation law
hence T is a tensor of type (1,2).
sum of
smooth manifold M can be decomposed into a
Theorem VI.2.2 : Any affine connection V on a
amultiple of its torsion tensor
and a torsion-free connection. 2043

manifold M with torsion Tand connection

Proof. Let be any affine connection on a smooth
coefficients T. Let
(2.5)
T,=c +r)
Then from (2.2), it follows that
Thus : is symmetric with respect to the
7-;--0, and hence
 206
DIFFERENTIAL GEOMETRY OF
MANIFOLDS
subscripts. Hence by Theorem VL.2.1, it follows that I are the coefficients of?isome connection
V
Vis torsion free or symmetric. Therefore from (2.2) and (2.5) and
we get

(2.6)
where Einstein's summation convention have been used.
Hence
V,Y = % T(X, Y) + V,Y,
(2.7)
where x= -, Y= Thus from (2.7), the theorem follows.

Now, the geodesic equation (1.8) is equivalent to

d'ut du' du'
=0.
dt dt (2.8)
Hence a connection Vand the corresponding torsion-free connection V have the same
geodesics. The
intrinsic meaning of V and ù have the same geodesics is that the difference A = V-V is a
skew .
symmetric tensor.

Theorem VI.2.3 : If P is a symmetric affine connection on M, then for any point p e

a local coordinate system u' such that the M there exists
corresponding connection coefficients vanish at p.
Proof. Suppose (V v) isa local coordinate system at pe M with connection
coefficients r . Let
u' = y' t 2 (p) (v- v(P)) (v'- v'(p))
(2.9)
Ou' a'u'
Then = 8, = (p). (2.10)

ou'
Thus the matrix is non-degenerate in a
neighbourhood of p. Hence (2.9) provides for a change
of local coordinates in a neighbourhood of p. From
(1.4) we see that the connection coefficients in
the new coordinate system u satisfy (p) =0, lsi, i, kSn.
This proves the theorem.
We now suppose that M is a affinely
connected space andy:I’M is a smooth curve on M. where
 207
INEAR CONNECTIONS

vectors
b)cR. Then a vector field on y is given by the field consisting of tangent
/=(a,
curve
dy) This field Y() = () is called the tangent vector field on y orthe natural lift of the
d vectors (is
parallel
yto
TM, The curve is called a geodesic if its tangent vector field consists ofgeodesic y is thatin
covariantly constant), i.e., if its natural lift is horizontal. The essential property of a
ofr motion along this curve, its tangent vectors Y() are transformed parallely, i.e., the curve
the process
curved. In this sense, geodesics area generalization of straight lines in affine geomery.
We have seen that geodesics are characterized analytically by the equation

dt
equations (1. 6). These are second order
and in terms of local coordinates they are characterized by the theory
Therefore by the
differential equations which are resolved with respect to higher order derivatives.
of ordinary differential equations, for any point peM and for any
vector XeT, M, there exists a maximum
geodesic y for which y(0) =p and
(not continued to any larger interval of the real line R)
equations, this geodesic
ö(0) = (0) = X and by the uniqueness of solution of ordinary differential
,() or r, (1, A),
is unique. We denote this unique maximal geodesic with the properties above by y,
Iel. or p(4).
The interval of the real line R on which the geodesic ... is defined is denoted by
certainly depend on the point p
The functions u() which give the geodesic y.. in local coordinates
variables, the variable t, n coordinates of the
and the vector X and, in fact, they are functions of 2n+]
theorem on the smooth dependence of
point p and n coordinates of the vector X. By a well-known
smooth functions of all 2n+1
solutions of differential equations on initial data, these functions are
the vector field
variables. Thus the unique maximal geodesic , y depends smoothly on the point p and
interval , ,.
X. But the following example shows that the similar assertions are not true for the
semi-circles x+y'= 1,
Example VI.2.1 : Let the manifold M be the (x, y)-plane R from which the two
y20 and x'+y' = 4, y s0, are removed. Then the geodesics ,, passing through the point
these
end points lie on semicircles.
pl0, 0) are rectilinear intervals with the directing vector X whose
assumed to be a unit vector
Thus if is the inclination angle to the abscissa of the vector X (which is
for definiteness), then
(-2,1) if 0<p<n
Ip.x = (-1,2)if n <Q<2
(-1,)if=0or p=I
Therefore, for X =(t1, 0), the function X’,, hasa discontinuity.
then the manifold M with the
If I,,= , (X) = R for any point p eM and for any vector Xe1, M,
afine connection is said to be geodesically complete. The manifold M in the above example is not
geodesically complete.
 DIFFERENTIAL GEOMETRY OF.
208 MANIFOLDS

Figure 45

Ifthe functions u'() satisfyequations (1.8), then the functions v'() = u'()also satisfy equations
(1.8) for any leR because of the quadratic dependence of the left hand sides of these equations on the
first derivatives. Since

dv' hdu
-(0) = -(0), it follows that
dt dt
Y,(A, X) =Y, (1, AX). (2.11)
This means that under the parameter change ’ , the geodesic Y, , transforms into the geodesic
Y, Ar Of course in this case
teI,(X) ue I, (X).
Definition VI.2.3 : Avector Xe T,M is said to be exponentiable ifthe geodesic Y, is defined for
t=1(i.e.,if le I,(X). For any exponentiable vector Xe T, M, the point
Exp,X= ,(1, X) (2.12)
iscalled the exponential of the vector X. The mapping Exp or Exp, :X’ Exp,X= Y,(1,X) is called the
exponential mapping.
We note that the zero vector O= 0, of the tangent space T,M is exponentiable and
Exp(O) =Exp, O-p.
The set of allexponentiable vectors of the vector space T,M (the domain of the mapping Exp,) is denoted
by O,. It contains the vector O, and implies from (2.11) that, it is star-shaped with respect to this point,
i.e., if Xe 0,, then AXE0, for any 4, 0sAs1. The relation O, =T,M holds for all pe Miffthe manifold
M is geodesically complete.

Definition VI.2.4 : Suppose Mis asmooth manifold with an affine connection V. Then a neighbour
hood U, of the vector Oin T,Mis called a normal neighbourhood if Xe U, , then 2Xe U, for 0sisl
and the mapping Exp is adiffeomorphismof U, onto an open neighbourhood U, of the point pe M.
The neighbourhood U, of thepoint pe Mis called anormal neighbourhood of p if U, =ExpU,
where U, is anormal neighbourhood of Oin T,M.
INEAR CONNECTIONS 209

i
If Mis an affinely connected space, then each point pe Mhas anormalIneighbourhood U, which
normalI neighbourhood of each of its points.
isa
The coordinates u', .., u" in a normal neighbourhood U, of any point pe Mare called normal
coordinatesifthee diffeomorphism Exp transforms them into linear coordinates on T,M(more precisely
on , ). characteristic property of normal coordinates is that geodesics passingthrough a
A point p have
the equations
u'= a't, i=1,2, . , n
where ' are constants. The nomal coordinates are sometimes calledthe geodesic
in these coordinates,
at
normal coordinates. In atorsion-free affinely connected space M, the coordinates u'. ... . u" centered
M.
p are
normal iff ru'u'=0, k = 1,2, ..,n at
p. I; are the connection coefficients of V on
For any point P, of an affinely connected space M, there exists anumber S>0 and a neighbourhood
is contained in
Uof the point p, such that for each point pe U, the neighbourhood Uofthe point P,
normal neighbourhood Us, of the point p. since UcU,, and U, is normal, there exists a vector
a
YeT,M such that Exp,X = q and the segment
(2.13)
Ygil’ Bxp, iX, 0stSI

Po
Usp

Figure 46

U,, We now state the following theorem

of the geodesic %. , connects the point p with the point g in 6.p

without proof, known as Whitehead Theorem :

manifold with an affine connection .
Theorem VI.2.4 (Whitehead Theorem): Suppose M is asmooth y.. is
qeU, the segment
Then each point p, of M has a neighbourhood U such that for any points p,
contained in U.
the geodesic %. , to the closed interval
The path t’Expp tX, 0 Stsl, is the restriction of
[0 1). This path is called a geodesic segment connecting the
point p with the point q= Exp X. Since
normal neighbourhood U ofthe pointD.
this geodesic segment is defined only for q belongs to a certain
it by Y,a We can also find that Y.. is the
we call it a normal geodesic segment from p to q and denote
q. Thus for each point p,e M,
unique nomal geodesic segment connecting the pointp with the pointexists a normal geodesic segment
there
there exists a neighbourhood U such that for any points p. q ¬ U,

connected space, is called a

Definition VI.2.5 : Aneighbourhood U of apoint P,E M, an affinely
210 DIFFERENTIAL GEOMETRY OF

is defined and is
MANIFOLDS
normal convex neighbourhood if for any points p.q ¬ U, the segment Vp4
contained in
Anormalconvex neighbourhood is a normalIneighbourhood of each of its points andiis
to the ballB*. Thus as astrong fom of the Whitehead theorem we can state the following : dif eomorphic
Theorem VI.2.5: Suppose Mis a smooth manifold with an affine connection V. Then each point p. of
Mhas a normal convex neighbourhood.
It can be proved that a torsion-free affine connection will be completely determined locally by the
curvature tensor.

VI.3 Curvature Tensor of an Affine Connection

If afunction f is a C"-function then in advanced calculus the second order partial derivatives are
independent of the order of differentiation, i.e.,

du'ou! ou'du'
But for functions on manifolds, i.e., for fe C´(M), and X, Ye X(M), the analogous property
does not
hold i.e., X(Yf) = Y(Xf), in general. In fact, X(Y) Y(X) = (X, Y]f. But [X, Y] =£,Y* 0, in
general. In asimilar manner, the operator V, and V does not commute to each other, that is.
V,V,- V, V, 0, i. .,(V,, V,]*0. Hence it determines a tensor field on M given by
V,V,Z- V,,Z- VxyZ= R(X, Y)Z,
known as the curvature tensor of the manifold, defined as follows:

Defjnition VI.3.1: Suppose M is a smooth manifold with an affine connection V. Then the curvature
sehsor field R of type (1,3) of the connection is defined by the mapping
R:x(M) xx (M) xx (M) ’ x (M)
given by R(X, Y)Z = V,V,z- V,VZ- Vxy Z (3.1)
for all X, Y, Zex(M).
In terms of local coordinates u' of any point pe M, the
components of R is given by
(3.2)

If we take X Y= Z=
du'
then

R(X,Y)Z-= R
(3.3)
From (3.1) we have
R(X, Y)Z= V,V
(3.4)
LINEAR CONNECTIONS 211

Since = 0
and VZ =fV,Z, we get

Hence (3.4) takes the form

R(X, Y)Z = V, (3.5)

Now we have by (1.3),

du"
[since V, (S) =fV,Y + (Xf)Y] .
h

k h + TT (3.6)

-+ r r h [Interchanging the dummy indices

h and in the 2nd term]
Similarly we can obtain

(3.7)
du du' du'

Using (3.6) and (3.7) in (3.5)we get by virtue of(3.3) that

r

Since are linearly independent we get

RR= (3.8)
where Einstein's summation convention have been used throughout the calculation. Hence the comp0
nents of R in terms of localcoordinates is given by (3.8). From (3.8) it can be easily seen that R' satisfies
the transformation law of atensor of type (1,3) and hence R is a tensor of type (1, 3).
212 DIFFERENTIAL GEOMETRY OF
MANIFOLDS
Note: It is very important to note that in (3.8) we have deduce the components ofR with
respect to an
atine connection and hence it is not symmetric in general. Thus in (3.8) we can never write
T= r for all i,j. But if the affine connection is symmetric, then by virtue of Theorem
Tollows that = r; for all i,j.
VI.2.1, it
However, in the next chapter i.e., in Chapter VIl, we shall see that in a
Kiemannian manifold, the relation - r for all ¿j always holds with respect to a Levi-Civita
connection as it is a torsion-free connection.
We now demonstrate the properties of R as follows :

Pheorem VI.3.1 : If M is a smooth manifold with an affine connection V, then its curvature tensor
satisfies the following:
(1) R(X + BY, Z)U = aR(X, Z) U+ BR(Y, Z) U, (3.9)
(2) R(X + gY, Z) U = fR (X, Z)U + gR(Y, Z)U (3.10)
(3) R(X, fY + gZ) U= f. R(X, Y)U+ g.R(X, Z)U (3.11)
(4) R(X, Y) GZ + gU) =f.R (X, Y)Z + g.R(X, Y)U (3.12)
for all X, Y, Z, Ue X(M), a, BeR,f. ge C"(M).

Proof. The proof of (3.9) readily follows from R-linearity of Vand similar expression may be obtained with
respect to second and third variable.
We shall prove (3.10) and (3.12) only, whereas the proof of (3. 11) follows immediately like (3.10).
Proof of (2). By definition we have
R GX + gY, Z)U= VxaVU- V,kgU - Vgr z,

-f,V,Ut gV,V,U-fVU-gV,V,U- (Zf) V,U

- (Zg) V,U - Vx s/Y. z) - (Ze)y U
=f(V,V,U - V,V,U- Vx.z,U) t g(V,V,U- V,U- zU) [Since
VHx.z -(4xU =fVxzU -(Z)V,U as for fe C(M), (Z) e C"(M)) .
=fR(X, Z)U+ gR(Y, Z)U, which is (3.6).

Proof of (4). We have

R(X, Y)(GZ + gU) = V,, (fZ+gU) - V, V,(fZ + gU) Vxn (SZ+gU)
V,(J9,Z + (Y)Z + gv, Z + (Yg)Z) - V,(SV,Z + (Xf)Z + g V,Z
+ (Xg) Z) - (f Vx yZt ((X, Y}f)Z +g Vy y Z + ([X, Y]g)Z)
=fVZ + (Xf) V,Z + gV,VZ+ (Xg) V,Z + X(Yf)Z + (Yf) V,Z
+ X(Yg)Z +
(Yg)VZ - fV,VZ- (Yf)V,Z - g,V,Z- (Yg) VZ
-Y(X))Z - (Xf) VZ - Y(Xg)Z- (Xg) V,Z - fxy,Z-gVx yZ
- ([X, YIf)Z - ([X, YJ)g)Z
IINEAR CONNECTIONS 213

= f(V,V,Z- 9,,Z- V, Z) t g(,9,2- V,,2 - Vx.2)

+ X(Yf) Z - Y(Xf)Z + X(Yg)Z - Y(Xg) Z - ([X, Y]f)Z - ([X, Y]g) Z
= fR(X, Y)Z + g R(X, Y) 2,
which is(3.12).

From the theorem, it follows that

RL, )Z =fR(X, Y)Z,
(KX.)Z =fR(X, Y)Z and
G R(X, Y)fZ =fR(X, Y)Z for any fe C" (M) and X, Y, Z ex(M).
From (3.1), it follows that
R(X, Y) = V,V,- V,V,- VN
(X. Y] (3.13)
=-(v,,- ,v,- Vy ) (Since [X, Y]=- [, XJ)
=-R(Y, X) for all X, YE X(M).
Thus R(X, Y) =- R(Y, X) and hence R is skew-symmetric with respect to the first two variables. From
(3.13), it follows that R(X, ) is a mapping i.e., R(X, Y): x(M) ’x(M); and R(X, Y) is called the
curvature operator, which can be uniquely extended up to a certain derivation of the algebra of tensor
fields on the manifold. Since the composition oftwoderivations is also a derivation, the right hand side of
(3.13) is the restriction ofthe derivation V,, - ,v, - of the algebra of tensorfields ón the
manifold M to x(M). Thus both sides of (3.13) can be treated as derivations of tensorfields on M. In
particular, ifK is atensor field oftype (1, r), then for any vector fields X, Y, X,, ..,X,ex (M) the formula
(R(X, Y)K) (X,, .., X) = R(X, Y)K(X,, .. X,)

-XK («, ..X, RX, Y)x, x,., ..x, )

i=l
(3.14)

holds. Also if K is a tensor field of type (0, r) on M, then the following formula holds :
(R(X, Y)K) (X,, .., X, )= R(X, Y)K(X,, ...,X,)

R(X, Y)X,, X,. . .., X, ). (3.15)

In (1.15) ifwe takes = 1,2, 3, then we have
(VK)(Y, X) = (, K)(Y) = VK(Y) K(V,),
(VK)(Y, Z; X) = (9, K)(Y, Z) = V,K(Y, Z) -K(9, Y, Z) - K(Y, V,Z)
(VK) (Y, Z, U; X) = (, K)(Y, Z, U) = _K(Y, Z, U) K(V,Y, Z, U)
- K(Y, V,Z, U) K(Y, Z, V,U).
In the first of these formulas, the symbols V, K, K and V, denote operators on xM). The first
summand in the right hand side is the result of applying the operator K and then the operator V, to the
field Y, and the second one is the result of applying the operator V, and then the operator K. Therefore,
USing the sign or the composition of operators in order to avoid confusion, we can write this formula in the
form of the relation between operators
V,K= VoK- Ko Vg.
 214 DIFFERENTIAL GEOMETRY OF MANIFOLDe

By definition, this means that the operator V,K is the commutator (9, K] of theoperators V, and K:
V,K- (, K]. (3.16)
Thus if K is a tensor field oftype (1.3) then we can write
(V,K)(Y, Z) |V,. K(Y, Z)) - K(V, Y, Z) - K(Y, V,Z) (3.17)
where the tensor field Koftype (1, 3) is identified by the lincar operator
K(X, ): x(M) ’ x(M), Z ’ K(X, Y, Z), for all X, Y, ZE X(M).
Hence for r = 1, the formula (3. 14) can be written as
R(X, Y)K = (R(X, Y), K]J.
For r =3, (3.14) can also be written as
(R(X, Y). K)(U, V) = (R(X, Y), K(U, V))- K(R(X, Y)U, V) K(U, R(X, Y) V), (3.18)
where K is an arbitrary tensor field oftype (1. 3) and hence K(U, V), U, VE X(M) is therefore a certain
C"(M) -linear operator from x(M) ’ X(M).
Ifin particular, M= R", then from(3.4), it follows that R=0, as r,=0 for all 7, ij in an Euclidean
space. Consequently R(X, Y)Z = 0in E", and hence in any manifold M, R can be think of as a wayof
measuring how much M deviates from being Euclidean.

VI.4 Some Exercises

Exercise VI4.1 : If Vis alinear connection of thesmooth manifold M, prove that the connection of
M defined by
7,Y= V, X+ [X, Y],X,Ye X(M)
is a linear connection and find the coefficients of V in terms of V.

Exercise VI4.2: Let V be a linear connection and A be a tenser field oftype (1, 2) on a smoothmanifold
M. Then prove that the connection ù defined by
,Y=V, Y+ A(X, ), X, Ye x(M)
isa linear connection of M.

Exercise VI.4.3: If V and ù are inear connection on a smooth manifold M, then prove that
-(-)+tù is alinear connection of M for each E/0, 1].
Exercise VI. 4.4: Let :M’ M be a diffeomorphism and be alinear connection of M such
that = () be defined by
VY = (Vo'y Y) for all x'. Y'e X(M).
Then prove the following:
) is a linear connection of M
(ü) If p is the flow of a vector field Xe x(M) such that p(V) = V for all te
R, then
X, YE X(M).
Exercise VI.4.5: If is a linear connection of the smooth manifold Msuch that we
define a linear
LINEAR CONNECTIONS 215

connection V=7(VyX+ ), where ù,Y= V, X+ [X, Y], (sometimes called the conjugate
connection of V). X, YEX(M), then prove the following:
) is torsion free.
i) If V is torsion free, then = ù = 7.

Exercise VI.4.6: If w is a differential r-form on a smooth manifold M equipped with a torsion free linear
connection V, then prove that

(r+) do(X, X,) =2-)'

j=0
, o)(g:... , X,,...... .x,)

X....x, ¬ X(M), where X, denotes that the corresponding vector field is omitted.

Exercise VI.4.7: If Visa symmetric or torsion free linear connection of asmooth manifold M, prove
that
2do(X, Y) = (V,o)(Y) - (,o) (X), X, Ye z(M). wE A'(M).
Exercise VI.4.8 : Let M and N be smooth manifolds with linear connections and ù respectively.
Then a C*- map f: M’ N issaid to be connection preserving if

holds for allpe M, where X, Yare f- related to X.ø respectively. If f is adiffeomorphism, then prove
that
) f(R(X, Y)Z) = RfX, fY) fZ,
where R and R are the curvature tenser fields of Vand respectively.
() f(T(X, Y)) = TGX, fY).
where Tand T are torsion tensor of V and respectively.

Exercise VI.4.9: Suppose Vand ¼are two linear connection ofa smooth manifold Mwith coefficients
r and r respectively such that they are related by
F= r+ 28, 0,
where is a differential 1-form, 8 denotes the Kronecker delta. Then find the value of
R--Rk, where R and R are their respective curvature tensor fields.

Exercise VI.4.10: Suppose V is the linear connection of R´ = {(<', x)} with components =0
Cxcept :=1. If Y () is the curve given by
Y) = ('(), y'()) = (-2e+ 4, 1+),
then compute the vector field obtained by parallel transport along yof its tangent vector at y(0). Is Ya
geodesic curve ?
216 DIFFERENTIAL GEOMETRY OF MANIFOLDS

Exercise VI.4.11: Suppose Visa linear connectionof R' - ((*. x)} with components I, =0 excep1
T» =2 and y() is the curve given by y) = (y'), y')) -(e +5, 3t+7). Then compute the
vector field obtained by parallel transport alongyof its tangent vector at y( 2). 1s Ya geodesic curve?
an
Exercise VI.4.12 : If Tand R are the torsion and curvature respectively ofa linear connection Von
afinely connected space M, then show that for all X, Y, Ze x(M),
() T(T(X, Y), Z) = TV,Y Z) + T(2, , X) - T([X, Y), Z)
(ii) R(T(X, Y), Z) = R(V, Y, Z) + R(Z, V, X) - R([X, Y], Z).

Exercise VI.4.13 : If Vis an affine connection on an affinely connected space M and is a bilinear
function such that
Y= Y-T(X, Y),
and T =-7, where T and T are
for all X, Y, ZE x(M), then show that ù is a linear connectionon M
the torsion tensors corresponding to V and ¼ respectively.
always
Exercise VI.4.14 : Prove that a linear connection defined by a parallelization on a manifold has
on a non-parallelizable
zero curvature. Do you think that connection with zero curvature can never occur
manifold ? Justify.
Exercise VI.4.15: Show that a paracompact manifold admits a linear connection.