Estimates on the Metric Tensor
David Glickenstein
February 14, 2000
1. Intro duction. This is a summary of some easy estimates on the metric tensor of a
manifold if the sectional curvatures are b ounded. It generally follows [1, section 1.8].
2. De nitions. We will estimate derivatives of the metric in geo desic p olar co ordinates.
Supp ose wehave a Riemannian manifold M; g. We consider geo desic p olar co ordinates,
2 2 i j
and in these co ordinates the metric can b e written as g = dr + r g i j d d .
We also are going to need to take a geo desic ton M which is parametrized by arclength,
0
and we will say that P = 0, Q = r , and T t= t.
0
In the sequel, we shall talk ab out vector elds V along the curve . When we write V t,
it is understo o d to mean r V t, where r is the Riemannian connection on M .
T t
We shall also restrict ourselves to r such that r is in the ball around P where exp is
P
a di eomorphism inside the injectivity radius.
3. Relevant Theorems. We will use Jacobi elds and the Index Inequality. We rst
should de ne the Index of a vector eld.
De nition 3.1. The index at r of a vector eld V t along a curve is
Z
r
0 0
I V; V = [gV t;V t + g RT t;V tT t;V t] dt
r
0
d
where T t= is the tangential vector eld to .
dt
This is, I think, essentially the second variation of arclength.
The imp ortant theorem is the following:
Theorem 3.2 Index Inequality. See, for instance, [2]. Let t bea geodesic between
P = 0 and Q = r such that therearenoconjugate points between P and Q, and let
J t be a Jacobi eld along such that J 0 = 0 and V t be another smooth vector eld
d
, V 0=0, and J r = Vr, along such that both J and V are orthogonal to T t=
dt
then I J;J I V; V .
r r
In order to use the Index Inequality,we will need the following two prop ositions.
Prop osition 3.3 See [1], 1.46 . Q = exp X is a conjugate point if and only if exp
0
P P
is singular at X .
0
Thus if we stay within the injectivity radius on M ,we will not encounter any conjugate
p oints. We will also need to know: 1
2
Prop osition 3.4. The injectivity radius for a spaceofconstant curvature b is =b.
Finally,we will use the fact that Jacobi elds are the in nitesimal variations of geo desics.
In geo desic normal co ordinates at P , the geo desics through p are given by exp tv . Thus
p
if wehave a Jacobi eld along a geo desic such that J 0 = 0, then we see that in our
@
0
geo desic p olar co ordinates wehave Jacobi elds J t= tkJ 0k where kw k foravector
@
tv
p
w means g w; w. Thus we see that:
Prop osition 3.5. For a Jacobi eld J along a geodesic t, t 2 [0;r] with r =Q,
@
2 0 2
such that J 0=0 and J r = j , we have g J r ;Jr = r g kJ 0k .
Q
@
4. Lower Bounds for the Comp onents of the Metric Tensor. We shall prove the
following:
2
Theorem 4.1. If the sectional curvatureofMisboundedabove by b then we have the
fol lowing two estimates, provided r<=b:
p
@ 1
g b cotbr log
@r r
sin br
g r;
br
i
where = for any i as de ned in the polar representation of the metric.
2 2
Now supp ose the sectional curvature of M is b ounded ab ovebyb ,sogRJ;TT; J b ,
2
or g RT; J T; J b .
00
We furthermore observe that if J is a Jacobi eld, then, by de nition, J t=RT; J T ,
so we get
Z
r
0 0 00 0
I J;J= [gJ t;J t + g J t;Jt] dt = g J r ;Jr 4- 1
r
0
We are going to compare the metric on M with the metric on a space of constant curvature
2
b .To do this, we need to make a frame fe t; :::; e tg on M by taking an orthonormal basis
1 n
0
fe ; :::; e g of T M such that e = 0 and then parallel translating it along . Notice
1 n 1
0
0
that since is a geo desic and parametrized by arclength, e t= t. We also need to take
1
~
a geo desic ~ on another manifold M which is the same length as and do the same thing
i i
to get a frame fe~ t; :::; ~e tg along ~ .We then have the map : c te t 7! c t~e t, or
1 n i i
~
V 7! V . Notice the following:
~
Lemma 4.2. The map :M; g ! M; g~ has the fol lowing properties for vector elds
~ ~
V and W along and V and W along ~ :
~ ~
g V; W =g ~V;W
0 0
~ ~
gV ;W=g ~V ;W
0 0 0 0
~ ~
gV ;W =g ~V ;W
i i
Proof: Use the frames wehave already set up. Then if V = v e and W = w e where
i i
i i
v ;w are real-valued functions, then
X
i j i i i j
~ ~
g V; W =v w ge;e = v w =v w g~~e ; e~ =g ~V;W
i j i j 2
Notice that it is imp ortant that we use frames instead of coordinates. Furthermore, since
0 0 0
0 i i i 0 i
~
the frames are parallel, i.e. r e = 0,, we get V = v e + v r e = v e and V = v e~ so
T i i T i i i
the last items follow. 2
Now,
Z
r
0 0
I J;J = [gJ ;J +gRT; J T; J ] dt
0
Z
r
0 0 2
g J ;J b gJ;J dt
0
b
~ ~
= I J; J
b 2
where I is the index on a manifold of constant curvature b and where wehavechosen
a frame fe g along and a corresp onding frame fe~ g along some curve~ of the same length
i i
in the manifold of constant curvature.
In constant curvature, wehave a Jacobi eld whichvanishes at t = 0 and ends at the
sin bt
~ ~
vector J r given by J r .
sin br
P
i
What do we really mean by this? Supp ose J r = cer with our frame as ab ove.
i
sin bt
i
~ ~ ~ ~
Let our new eld b e de ned by V t= c r ~e t. Notice that V r =Jr, V 0=0,
i
sin br
cos bt
0 i
~ ~
V t=b c r ~e t since the frame is parallel, and V is orthogonal to ~ at every p oint
i
sin br
0
~
since J r is and wehave a parallel frame along a geo desic, soe ~ t= Tt= ~t for all t.
1
Now compute letting c = c r using the Index Inequality 3.2 on the manifold of constant
i i
2
curvature b whichwe can do since we are inside the injectivity radius by assumption that
r =b:
sinbt
b b i
~ ~
I J; J I c e~ t
i
r r
sinbr
Z
h i
r
X X
1
2 2 i 2 2 2 i 2
= b cos bt c b sin bt c dt
2
sin br
0
X
b
i 2
= sinbr cos br c
2
sin br
= b cotbr g J r ;Jr
Wehave just ab out proven an estimate on the metric tensor. We use 4- 1. Wenow know
that
0
g J r ;Jr b cotbr g J r ;Jr 4- 2
@
i
Nowifwe take J t to b e a Jacobi eld such that J r = j for = for some i, then
Q
@
2 0 2
g J r ;Jr = r g kJ 0k by Prop osition 3.5, so we can compute:
0
b cotbr g J r ;Jr g J r ;Jr
0
g J r ;Jr
b cotbr
g J r ;Jr
@
gJ r;Jr
@r
=
2g J r ;Jr 3
@
2 0 2
r g kJ 0k
@r
=
2 0 2
2r g kJ 0k
1 @
p
= + log g
r @r
And thus wehave our rst estimate on the metric tensor in p olar co ordinates:
@ p 1
log g b cotbr
@r r
Notice that
1 sinbr d sinbr d
b cotbr log = log = + log b
r dr r dr br
so if weintegrate from 0 to r we get
0
p
sin br
0
g r ; log log
0
br
0
@ @
2 2
since at 0, the metric is Euclidean, so g 0; = 1 since in R wehaveg ; =r and
@ @
sin br
lim =1.Thus wehave
s!0
br
sin br
g r;
br
n 1
for any , a co ordinate on S .
5. Lower Bounds on Determinant of Metric. We shall now use what we proved ab ove
to prove:
2
Theorem 5.1. If the sectional curvatureofMisboundedabove by b then we have the
fol lowing two estimates, provided r<=b:
p
@ n 1
log jg r; j n 1b cot br
@r r
p
n 1
sin br
jg r; j
br
where jg j = det g i j .
We rst take a bunch of Jacobi Fields J ; :::; J such that together with T at r they
1 n 1
i j
form a basis for T M .Wenow let jg j = det g and supp ose that in co ordinates wehave
Q
@
k
Jt= c and compute:
i
k
i
@x
p
@ 1 @
log jg j = tr log jg j
@r 2 @r
1 @
j k
i j
= tr g g
2 @r
@ gJ r;J r 1
i j
i j
g =
2
2 @r r
n 1
i j
0