MANIFOLDS WITH NONNEGATIVE RICeI CURVATURE ALMOST EVERYWHERE
SEONG-HUN PAENG
ABSTRACT. Under the condition of RicM ;:::-en -l)k, injM ;::: io, we prove the existence of an€>0 such that on the region of volume
€ >0 the curvature condition of splitting theorem can be weakened.
1. Introduction
It is an important problem in Riemannian geometry to classify the complete Riemannian manifolds by curvature conditions. Splitting the- orems are concerned about the non-compact manifolds with nonnega- tive curvature. Using splitting theorem, we can prove that some finite cover of a compact Riemannian manifold with nonnegative Ricci curva- ture can be splitted toTk xN,whereN is a compact simply connected manifold and Tk is a k-torus [3]. In [6, 8], we prove sphere theorems under weaker curvature conditions t4an the standard Ricci curvature condition RicM ~ n - 1, Le., Ricci curvature and injectivity radius bounded below and RicM ~ n - 1 on M - A where diameter ofA, diam(A) or volume of A, vol(A) are sufficiently small. Similarly to sphere theorems, we will prove a splitting theorem of compact space also holds even if the Ricci curvature conditions are not satisfied on the region of small volume.
LetMio,k be the class of n-dimensional complete Riemannian man- ifolds with RicM ~ -(n-1)kand the injectivity radius injM ~io. We obtain the following theorem;
THEOREM 1.1. Let M E M?o,k and diam(M) S d. Then there exists anE> 0 depending only on io, n,k,d such thatifRicM ~ -(n-
Received May 9, 1998.
1991 Mathematics Subject Classification: Primary 53C20; Secondary 53C21.
Key words and phrases: Ricci curvature, splitting theorem.
Partially supported by KlAS and in part supported by BSRI and GARC- KOSEF.
1)f on M - A, vol(A) ::; E, then M is diffeomorpmc to Tk x N up to finite cover, where N is a simply connected space.
This theorem is a generalization of the following theorem due to Cai
[2].
THEOREM 1.2. Let M E M(io,n,k) and diam(M) ::; d. Then there exists an E(io,n,k,d) > 0 depending only on n,io,k,d such that
ifRicM ~
-en -
1)E, then M is difIeomorphic to Tk x N up to finite cover, where N isa simply connected space.We use the compactness theorem due to Anderson and Cheeger [1]
and the rigidity result in a fixed small ball[7, 4]. The following notation will be used; if wefix81, ... ,8n ,thenr(81 , . . . ,8nlE) ~ 0andr(f) ~0 asf -+ O.
We would like to express our gratitude to Professor Hong-JongKim for much kind and helpful advice.
2. Preliminaries
In Preliminaries, we show that the conditions of RicM ~
-en
-1)kand injM ~ io make exponential map almost isometric on a ball of fixed small radius which depends only onn,k,io.
By Broclcs' estimate on the Laplacian of the distance function, we obtain the following Jacobi field estimate on io/2-ballB(p,io/2) [4, 7, 8]. LetM be a complete Riemannian manifold with RicM ~ -(n-1)k, injM ~io and ,et) be a minimal geodesic starting from p and yet) is a Jacobi field along, such that Y(O) = 0, (Y'(O),,'(0)) = O. Let db) be the distance from p to the cut point on ,. Then io ::; db).
Define A := VVr = Hess r, so trA = c:::'r and Y' = AY. Write A(t) =B(t)+lit. We know
iT
IIBll::;D(io,n,k)r1/2,for some constant D(io, n,k) [4, 7]. Then we have
for some constant D = D(io, n, k) ifr < io/2 [4, 7, 8]. In Euc1idean space, we know that D = O. For any f >0,. we can choose a uniform
ro > 0 which is depending only on n,k,io such that Dr1/2 < € for r < ro. Then
. {" dexp(v)11 }
1 - €~ mm Ilvll IvE TxM, x E B(p, ro)
{ Ildexp(v) 11 }
~ max IIvll /v ETxM, x E B(p,ro) ~ 1+€
on ro-ball by the above inequality. So we can find a uniform ro > 0 such that the exponential map is almost isometric on ro-ball.
Furthermore, we know that
(2.1) -(n -l)kcothk(d(-y) - t) ~ f}.r(-y(t)) ~ (n -l)kcothkt.
For the proof, see [I). Then by the Riccati equation, we get the follow- ing estimate for r ~ d('"Y) - 8,
Then
1:/21IA,/2
~ -trA(r)+trA(io/2) +(n -l)kd(-y)
2(n - 1)
~ (n -l)kcothk(d(-y) - r)+C(io,n,k) + . +(n -l)kd(-y) Zo
2(n- 1)
~ (n-1)kcothk8+C(io,n, k)+ . +(n -l)kd(-y) Zo
~ F(io,n, k, 8,d(-y)),
for some constant F depending only on io, n, k, 8, d(-y) since cothx is decreasing function. Then
iT
IIAII ~ (d(-y)iT
IIA/l2))1!2i o/2 io/2
~ Vdh)F(io,n,k,8,d(-y))1/2 = H(io,n,k,8,d(-y)),
for some constant FI .
r
IIBII =to/
2+IT
IIBIIlo lo io/2
I
T 1S. D(io, n, k)Ji o/2+ (IIAII+ -)
io/2 t
S. D(io,n, k)Jio/2+F1(io,n, k, 8,d('"Y))+to~
S.F2(io,n,k,8,d('"'()),
for some constantF2 • Then by the same method as above, we also get a uniform lower bound and upper bound of the IIYII depending only onn,k,io,8,d('"'() ont < d('"'() - 8. .
3. Estimate of the volume of the bad part
For the simplicity of argument, we only consider the sequence(Mj , 9j) inM~,ksatisfying the conditionsdiam(Mj )S.d,RicMj 2:0 on Mj - Aj
where vol(Aj) :::; f.j and f.j --+ O. Then we know that the universal cov- ering space ofMj , Mj converges to a compact subset of CQ-Riemannian manifold X on compact subset. Shortly we use Aj instead of thelift- ing of Aj, Aj . Fix a point Pj E Mj . We may assumePj E Mj . Let '"Ye(t) = expPjto and J.L is the measure on '"Ye. We use the following notations;
dj(0) = the distance fromPj to the cut point in the direction0, where 0E sn-l C T pjMj,
~ ={O E sn-l C TpjMj IJ.L('"'(e([O,dj(O)]) nAj ) < f.}, Sf,8(O) = inf{s Is> 8, 0E (~)C, JL('"Ye([8,s]) nAj )> f.}.
LEMMA 3.1. For any fixed D > 0, limj-+oo vole{exppjto I 0 E
(~)C, S;,8(O) s. t <min(dj(O), D)}) = O.
We can consider {exppjtoI0E (~)C, Sf,8(O) s. t < dj(O)} as a bad
part for applying the Bishop-Gromov theorem. We want to show that this bad part can be ignored.
Proof Assume thatM isan element ofMf0,k' Let{u,el,··· ,en-I}
be an orthonormal basis for the tangent space at some point q E M and let ti(t) be a Jacobi field along'"re(t) = exptO such that ti(O) =
0, ¥i'(0) =ei for i =1, .. ' ,n - l.
Define
J(u,t) = {t
o
-<n-l)(det 9(ti,l'j))! if t ~dj(O) if t> dj(O).
Then for a region A in the unit sphere of the tangent spaceTqM,
vol{expqto Ih <t <t2, () E A}= lA(
l
t2J(u, t)tn-1dtdu.tl
Let Jj and J-kbe the J ofMj and the space form with constant curva-
. . 1 - 1
ture -k, respectively andb1(u,t) =P(u,t)n=It, b(u,t) =J-k(U,t)n-1t.
. . . b(u, r)
In the proof of the Blshop--Gromov mequality, we see that b(u, a) ~
~«u,r ))' if r > a. Simply, we use J and b instead of Jj and bi, respec- b u,a
tively. We define
Ct2 ._ {b(U,r) I [ ]}
tl . -max b(u,s) r,s E tl,t2 .
If vol«8{)C) ---4 0 then there are nothing to prove by the Bishop-- Gromov theorem. So we assume that limj~oovol«8{)C) > O. Let
1 / 2 ' . 1/2
Then vol(Aj ) > Cj vol(cPn. So vol(cPn < Cj ---4 0 as j ---4 00 which isa contradiction. .
Now we may assume that for every direction ()E (E>~Y,
Then we know that for any E > 0, there exists a c > 8 such that
E~/2
bn-1(0,c) < _J_ and JL(-Yo[8,c] nAj ) < E. From this fact, we know
E
that c:::; 8!,6" Then by the above inequality, we get
(3.1)
for 81,6 < r < D. Sobn-1(0,r) < 7(8,E/€i) for r >81,6.
Let
Consequently,vol(..4j (D))(8, E) = J(e~)CJiL bn-1(0,t)dtdIJ --0 asj --
00. This completes the proof. 0
REMARK 3.2. By the above proof, we know that if Pi =
Ey4,
thenPi --0 and
.lim vol({expp_to I0E (~p'_)C, 8pj_6(0) :::;t < min(D,dj(O))}) =O.
J--+OO 3 3 3'
This value Pj will be used in following sections.
4. Proof of Theorem
We will prove that the limit space X = Rk X N whereN contains no line. Then we can prove Theorem 1.1 by a contradiction.
Shortly, we use Ai instead of the lifting ofAi, Ai. Let 'Yi be a line inMi and 'Yi(O) = Pi. Then'Yi converge to'Yin X and we may assume that Pi --p. We follow the proof in [9].
We know that vol(Mi ) ~ vl(io,n,k) and vol(B(-Yi(ti),2ti)) :::;
v2(n, k, ti) for some VI,V2. Then the number of fundamental domains in B(-Yi(ti),2ti) is bounded byni =nl(io,n,k,ti) :=V2/Vl. Let
as Remark 3.2 wheretiwill be chosen below. Let 'Yo",(0) = x. Now we use the similar notation as previous section;
ei(x) ={Ox E sn-l C TxMi IJ.Lbo",([0, cf(Ox))) nAi) < Pi}, S~(Ox) =inf{s Is> a, Ox E (8i(x)r, J.Lbo",([a,sD nAi)> Pi}.
We use ei instead ofeibi(ti)).
Let ~(a)be a a-tubular neighborhood of the cut locus of'Yi(ti)' We also know that the volume form b'Yi(t.)(t) has a uniform lower bound Ho(io, n, k,a,t) > 0 on~(a)Cas we see in Preliminaries.
Put D to be a bounded domain with smooth boundary in X andFi
be a diffeomorphism from D to a domain Di in Mi. We consider Di as D with a metric Ft9i where9i is the metric onMi. Define
and
B±(x)= .!im Bt(x).
t-too
Itis an essential part of the proof of :Theorem 1.1 that .6.B+ ==0, Le., B+ is a harmonic function. Ifwe assume the almost nonnegativity of Ricci curvature, we need not check that B+ isa harmonic function [2, 10]. But in our case, we must show that B+ is harmonic.
Now we chooseti such that
P·e10(n-l)kti
t - 0
Ho(io,n, k,a,ti )
and ti - 00 as i - 00. It is possible to find such ti by passing to a subsequence if necessary since we know ti - O. We will show that the limit ofBt has properties as Busemann function. Let df(O) be the distance from'Yi(ti) to the cut point in the direction(J E sn-l C T'i(ti)Mi .
Then we also know that Di C B(-Yi(ti),2ti) as i --t 00 and the volume of bad part
converges to 0 as i --t 00 by the choice of ti, Le.,
,i
meets at most n1 (io, n, k, ti ) fundamental domains so we may consider the volume of lifting of Ai as n1(io,n,k,ti)€i --t O. Precisely we get that for some constant H1(n, k,8)(4.1) bn- 1 (r) <(c2ti )n-1( 0 0)1/2n~€~ <H (n k 8)poe2(n-1)kti
'V.(t.) _ lj - 1 "~ ,
" • Pi
as (3.1) since the exponential growth rate for Cris less than (n-1)k.
Then
by the choice ofti, whereWn-1 is the volume of the standard(n - 1)- sphere. Ifi is sufficiently large, b.Bt (x) has an upper bound (n - 1)cothkd(-Yi(ti),x) ~ H2(n,k) for some constant H2(n,k). Since the volume ofWi converges to 0, we may consider the integration only on {exP'Yi(ti)(tein for computing the upper bound of liIDi-+ooIDi b.Bt.
Let
,9
C Mi be a geodesic from q in direction 0 and q> : [0,00) X sn-1 --t Mi such that q>(r,O) = expq(rO). Setq>*vgi = a(O, r)drdO and b= an=I,1 wherevgi isthe volume form ofMi . Then we mow thatb"+Ric(,',,')b < o.
n-1 -
For the proof, see [5}. In the case of the space of constant curvatures, the equality holds. So for Rn,
b" = o.
In this section, we only consider the direction0 E ei. If,9(r) E Af,
then
(b"b - b"bl(r) = (b"b)(r) ~0,
(4.2)
(4.4)
and if"Yo(r) E Ai andr :::; di(O) then
(b"b - b"b)(r) =«b"+b)b)(r) :::; (k+l)b(r)b(r) :::; C1(io,n,k)e2(n-1)kT,
for some constant Cl depending only on io,n,k. From this, we get (b'b - b'b)(r) = iT b"b - b"b:::; C1(io,n,k)pie2(n-1)kr.
Then we have (4.3)
(
b' _ ~) r = f;b"b-b"b < piC1(io,n,k)e2(n-1)kr
b b () b(r)b(r) - b(r)b(r) ,
b' b' p·C (i n k)e2(n-1)kT n - 1 C e4(n-1)ktip.
_ < -=- + t 1 0, , <2 - - +2_1 t - t 0
b - b b(r)b(r) - ti ti Ho
onR;.(b)C asi - t 00 by the choice ofti and we may assume ti/2 < r <
2ti.
Combining (4.2), (4.3) with the fact (n - 1)b'b'Yi(ti) =6Bt, we get
'Yi(ti)
[ 6B7-dV <
(1
+ [ ) 6B7-dVlUi t - Ui-Ri(O) l ui nRi(6) t
:::;1
Ui-Ri(O)6BtdV+H2(n,k) [lui nRi(6) dV<T(bl€i) +T(b),
for any fixed U cD. We can choose b> 0 arbitrarily small.
LEMMA 4.1. B+(x) 2: 1R;n fB(x,R)B+ for fixed small io/2 >
Wn -1
R>0 andx E D.
Proof. Let (r,O) be the normal coordinate system centered. at x.
The metric gi ofDi can be expressed. as gi = dr2+r2glkdOldOk, 1 :::;
l,k :::; n - 1. Let Gi = det(glk)' Itis known that 6r= n -1 + 8Iog.../G.
r Br
We know that H4 S VG(r) S Hs for r S i o/2 from Preliminaries.
Using this fact with (4.3) we obtain
1 ~!(Pi n 1 C (. n k) 2(n-l)kr lim u_v_v· = lim boor _
--=-
< Pi 1 to, , e ---t 0i-too & Or i-too 2 r - b(r)b(r) ,
onVi := {exptei(x) Its<f(Bx )}. Furthermore onVi,
(4.5)
~ !(Pi ~!(Pi 2t C (. k) 3(n-l)kr lim B7-_U_V _v · < tim 2to_U_V_vv < iPi 2 to,n, e ---tO.
i-too 2 Or - i-too .2 Or - b(r)b(r)
by the choice of ti and.JGiS C3(io,n,k)e(n-l)kr for some constants C2,C3 •
Take4>~ E C~(R;.(c5» such that 114>~11S2 and lllnk-too4>~ = 1 in the distribution sense and 4>f= 0 on the cut locus ofri(ti),where C~(D) is the class of coo-functions with compact support inD. ThenBt4>~
.is a smooth function. SinceBt4>f---t Bt in distribution sense, we also have bo(Bt4>D ---t b.Bt in distribution sense. By (4.4) and divergence theorem,
(4.6)
where 0< t < R. We know that
for some constant H6 since
/lbor - --11n-l S C(io,n,k) r
by [4]. Also we have
tivol(vtnB(x,t» S HltiPie2(n-l)kti ---t 0
(4.7)
(4.8)
from (4.1). So we obtain
lim
r
f B7-4>~8&dOdti-+oo}O }8B(x,t)nv.c ~ ~ Or
= lim
r
f B7-4>~_1_av75 ..JGidOdt= O.i-+oo}O }8B(x,t)nv,t ~ ~ & Or
Using (4.6),
lim lim f B7-4>~0 & dO
i-+ook-+OO}8B(x,t) ~ ~ Or
~.lim~-+ook-+oolim tn~1 }fB(x,t)6i(Bt4>f)
+ lim lim f B7"4>~a N dO
i-+ook-+oo }8B(x,t) ~ ~ Or
>limlim(f +f )
- i-+ook-+oo }8B(x,t)-~ }8B(:<)n~
( 8(BtOr4>f\fGi+(B7"~ 'I'~A~)ON) dOOr
= lim lim .!!:.- f B7"4>~v75dO.
i-+ook-+oodtJ8B(:&,t) 1. ~ Then we get
B + - I·· B+ - li li B+",k
Wn-l - .lillWn-l i - . m m Wn-l i 'l'i
~-+oo ~-+ook-+oo
in distribution sense and from (4.7) and (4.8), .lim lim wn - 1Bt4>f(x)
~-+ook-+oo
( 1
L
+ kiT 1 + kON) Od
> lim lim - B- 4>. - B· 4>- - - d t
- i-+ook-+oo tn - 1 8B(:&,T) ~ ~ 0 8B(:&,t) ~ ~ Or
~ Hm lim _1_
r
Bt4>~.i-+ook-+ootn-1 }8B(x,r)
Integrating this over (0,R), we obtain that
B+(x) ~ lim Hm _1_ { B74>~= _1_ ( B+
i-+ook-+ootn - 1 JB(x,r) t t tn- 1 JB(x,r)
in distribution sense. SinceB+ is continuous, we get Lemma 4.1. 0 We easily get that 6B+ ~ 0, Le., for any nonnegative Clf-function h and any8>0,
{ h6B+ = { ~hB+ = .lim { ~ihBt =.lim { h~iBt
lD lD t-+OOlDi t-+OOlDi
=i-+oolim ( { + ( ) h~iB:r
lDi -Ri (5) lDin~(5) . t ::; 7(8)+ .lim 7(8ki) = 7(8),
t-+oo
sincei --+ gin L1,p or Ca-norm so the coefficients of Laplacian oper- ator for harmonic coordinates converges in Ca-norm [1]. Then by the same argument as (4.4), we get the above inequality. Also we know that ~iBt--+ ~B+in distribution sense by the above argument.
By the same argument as [3, 9], we get ~(B+ +B-) == O. From
~B+, ~B- ::; 0, we obtain ~B+==0 soB+ is harmonic.
In[3],from~B+ == 0we get that Hess(B+) = 0and"VB+isa paral- lel vector field. It is also an important step to show thatHess(B+) = 0 in our case. We will follow Cai's proof. Define bi by the following Dirichlet problem;
~ibi=0 bilaD= B+.
Then we can prove Lemma3.1 in [2J.
LEMMA 4.2. l"Vibil2 converges, inthe strong Ll,q-topology (for1 <
q< p), to I"VB+12.
Proof. The only difference of proof occurs when we apply the Bochner- Weitzenbock formula. But we only need the integration of~il"VibiI2 so there are no obstruction to prove this lemma. 0
From this lemma, we get Hess(B+) = 0 in Lq and can prove the remainder of the proof by the same argument as [2J.
Now we obtain that there exists an€ > 0 depending only on n,k,iQ,d such that if RicM 2': 0 on M - A where vol(A) ::; € then M is diffeo- morphic to Tk x N up to finite cover, where N is a compact simply connected space. But the complete proof of Theorem 1.1 is the same as the above argument.
References
[1] M. Anderson andJ. Cheeger, Ca-compactness for manifolds with Ricci curva- ture and injectivity radius bounded below,J. Diff. Geom. 35 (1992), 265-28l.
[2] M. Cai, A splitting theorem for manifolds of almost nonnegative Ricci curva- ture, Ann. Global Anal. and Geom. 11 (1993),373-385.
[3] J. Cheeger and D. Gromoll, The splitting theorem for manifolds of nonnegative Ricci curvature,J. Diff. Geom. 6 (1971), 119-128.
[4] X. Dai, Z. Shen and G. Wei, Negative Ricci curvature and isometry group, Duke Math. J. 76 (1994), 59-73.
[5] S. Gallot, D. Hullin andJ. Lafontain, Riemannian geometry, Springer-Verlag, 1980.
[6] S.-H. Paeng, A sphere theorem under a curvature perturbation, Kyushu J.
Math. 50 (1996), 459-470.
[7] , Topological entropy for geodesic flows under a Ricci curvature condi- tion, Proc. Amer. Math. Soc. 125 (199'7), 1873-1879.
[8] , A sphere theorem under a curvature perturbation11, Kyushu J. Math.
52 (1998), 439-454.
[9] R. Schoen and S. -T. Yau, Lectures on Differential Geometry, International Press, 1994.
[10] Z. Shen, G. Wei, On Riemannian manifolds of almost nonnegative curvature, Indiana Univ. Math. J. 40 (1991), 551-565.
Korea Institute for Advanced Study (KIAS) 207-43 Cheongryangri-dong, Dongdaemun-gu Seoul 130-012, Korea
E-mail: [email protected]