Contents lists available atScienceDirect
Advances in Mathematics
www.elsevier.com/locate/aim
Morse-Novikov cohomology of almost nonnegatively curved manifolds
Xiaoyang Chen
SchoolofMathematicalSciences,InstituteforAdvancedStudy,TongjiUniversity, Shanghai,China
a r t i c l e i n f o a b s t r a c t
Articlehistory:
Received9July2019
Receivedinrevisedform26April 2020
Accepted21May2020 Availableonline9June2020 CommunicatedbytheManaging Editors
Keywords:
Morse-Novikovcohomology Almostnonnegativecurvature
LetMnbeaclosedmanifoldofalmostnonnegativesectional curvature and nonzero first de Rham cohomology group.
Using a topological argument, we show that the Morse- NovikovcohomologygroupHp(Mn,θ) vanishesforanypand [θ]∈ HdR1 (Mn),[θ] = 0. Based on a new integral formula, wealsoshowthatasimilarresultholdsforaclosedmanifold of almostnonnegative Riccicurvature underthe additional assumptionthatitscurvatureoperatorisuniformlybounded frombelow.
©2020ElsevierInc.Allrightsreserved.
1. Introduction
LetMnbeasmoothmanifoldandθarealvaluedclosedoneformonMn.SetΩp(Mn) thespaceofrealsmoothp-formsanddefinedθ: Ωp(Mn)→Ωp+1(Mn) asdθα=dα+θ∧α forα∈Ωp(Mn).Thenwehaveacomplex
· · · →Ωp−1(Mn)−→dθ Ωp(Mn)−→dθ Ωp+1(Mn)→ · · ·
E-mailaddress:xychen100@tongji.edu.cn.
https://doi.org/10.1016/j.aim.2020.107249 0001-8708/©2020ElsevierInc. Allrightsreserved.
whose cohomologyHp(M,θ)=Hp(Ω∗(Mn),dθ) iscalled thep-th Morse-Novikovcoho- mologygroupofMnwithrespecttoθ.Ifθ1,θ2aretworepresentativesinthecohomology class [θ],then Hp(M,θ1)Hp(M,θ2).HenceHp(M,θ) only dependsonthe deRham cohomology class of θ. This cohomology shares many propertieswith the ordinary de Rham cohomology.See[11,18,19] andsection2fordetails.
If [θ] = 0,the Novikov cohomology group Hp(M,θ) is isomorphic to the de Rham cohomology group HdRp (Mn).There arelots of workrelating de Rham cohomology to curvature properties of Riemannian manifolds. See for example [20]. In particular, a celebrated theorem of Gromov says that the Betti number of a closed manifold with almost nonnegativesectionalcurvatureis boundedabovebyaconstantdependingonly the dimension of the manifold [10]. Here we saythat a Riemannianmanifold Mn has almostnonnegativesectionalcurvatureifitadmitsasequenceofRiemannianmetricsgi
suchthat
sec(gi)≥ −1 i D(gi)≤1,
where sec(gi) isthesectionalcurvatureofgi andD(gi) isthediameterofgi.
However,therearequitefewworkdiscussingtherelationshipbetweenMorse-Novikov cohomology Hp(M,θ) and curvature when [θ] = 0. This paper is trying to make an attempttowardsthisdirection.Ourfirstresultisthefollowingtheorem.
Theorem 1.1.LetMn beaclosed Riemannianmanifoldof almostnonnegativesectional curvature andnonzero first de Rham cohomology group, then theMorse-Novikov coho- mology Hp(M,θ)= 0forany p(includingp= 0)and any[θ]∈HdR1 (Mn),[θ]= 0.
From the workin[8,15],we knowthataclosed Riemannianmanifold Mn of almost nonnegative sectional curvature is an almost nilpotent space. Namely, there is afinite cover of Mn, denoted by Mˆn, such that π1( ˆMn) is a nilpotent group that operates nilpotently on πk( ˆMn) for every k ≥ 2. Recall that an action by automorphisms of a group G on an abelian group V is called nilpotent if V admits a finite sequence of G-invariantsubgroups
V =V0⊃V1⊃V2⊃ · · · ⊃Vk = 0
suchthattheinducedactionofGonVj/Vj+1istrivialforanyj.NowTheorem1.1isa consequenceofthefollowingtopological result.
Theorem 1.2. Let Mn be a smooth manifold with nonzero first de Rham cohomol- ogy group. If Mn is an almost nilpotent space, then the Morse-Novikov cohomology Hp(M,θ)= 0 forany pandany [θ]∈HdR1 (Mn),[θ]= 0.
ByTheorem 2.1 in section2, we see thatn
p=0(−1)pdimHp(Mn,θ) isequal to the EulercharacteristicnumberofMn. Hencewegetthefollowing
Corollary 1.3. Let Mn be a smooth manifold with nonzero first de Rham cohomology group.IfMn isanalmostnilpotentspace,thenitsEulercharacteristicnumbervanishes.
Corollary 1.3implies thataclosed Riemannianmanifold of almost nonnegative sec- tional curvature and nonzero first de Rham cohomology group has vanishing Euler characteristic number. This result has previously been proved by Yamaguchi in [23]
usingcollapsingtheory.
Theorem 1.1 fails for closed manifolds of almost nonnegative Ricci curvature. Re- call thataRiemannianmanifold hasalmost nonnegative Riccicurvature ifit admitsa sequenceofRiemannianmetricsgi suchthat
Ric(gi)≥ −n−1 i D(gi)≤1,
where Ric(gi) is the Riccicurvature of gi and D(gi) is the diameterof gi. LetM4 be themanifoldperformingsurgery alongameridiancurve inT4,i.e., removingatubular neighborhood of the curve and attaching acopy of D2×S2. In [1], Anderson showed thatM4admits asequenceofRiemannianmetricsgi suchthat
|Ric(gi)| ≤ n−1 i D(gi)≤1.
Moreover,itsfundamentalgroupisisomorphictoZ3anditsEulercharacteristicnumber isnonzero.Forany[θ]∈HdR1 (M4),[θ]= 0,byTheorem2.1andTheorem2.3insection2, we getHp(M4,θ)= 0 forp= 2 and H2(M4,θ)= 0.However, thesectional curvature of gi constructed by Anderson can nothave a uniform lower bound. Otherwise, there will be also an upper bound of the sectional curvature and by Theorem 1 in[22], M4 will fiberover S1 which is impossible bythe construction. Inparticular, the curvature operatorofgicannothaveauniformlowerbound.BythefollowingTheorem1.4andits Corollary1.5, M4 infactcannotadmit asequenceof Riemannianmetricsgi ofalmost nonnegativeRiccicurvaturewithcurvatureoperatoruniformlybounded frombelow.
Theorem 1.4. Let Mn be a closed Riemannian manifold with nonzero first de Rham cohomologygroupand admitsasequenceof Riemannianmetricsgi suchthat
Ric(gi)≥ −n−1 i D(gi)≤1.
If the curvature operator of gi is uniformly bounded from below by −Id, then for any [θ]∈HdR1 (Mn),[θ]= 0,there existssome t∈R,t= 0 suchthat Hp(M,tθ)= 0 forany p,whereHp(M,tθ)istheMorse-Novikovcohomologygroup withrespecttotθ.
Corollary 1.5.Let Mn be a closed Riemannian manifold with nonzero first de Rham cohomologygroup.IfMnadmitsasequenceofRiemannianmetricsofalmostnonnegative Ricci curvature with curvature operator uniformly bounded from below, then the Euler characteristic numberof Mn vanishes.
ForaclosedRiemannianmanifold(Mn,gi) withalmost nonnegativeRiccicurvature and nonzerofirst deRham cohomologygroup,Theorem 1 in [22] alsoimpliesthatMn hasvanishingEuler numberifthesectionalcurvature ofgi hasauniformupperbound.
Theorem1 in[22] wasprovedbycollapsingtheoryandisquitedifferentfromourmethod inthispaper.
It has been known that the fundamental group of a closed manifold M of almost nonnegativeRiccicurvatureisalmostnilpotent[3,16].ByTheorem2.3,H1(M,θ)= 0 for any[θ]= 0 withoutanyadditionalassumption.See[14] forrelatedworkonnoncollapsed almost Ricciflatmanifolds.
Finally, wepointoutthatforaclosed RiemannianmanifoldM ofnonnegative Ricci curvature andnonzerofirstdeRhamcohomologygroup,thentheMorse-Novikovcoho- mology Hp(M,θ)= 0 foranypand[θ]∈HdR1 (M),[θ]= 0.Infact,byCheeger-Gromoll splitting theorem [5], afinite coverof M is diffeomorphicto aproduct of a torus and a simplyconnected manifold. ByTheorem 2.1 andExample 1, wesee thatthe Morse- Novikovcohomology Hp(M,θ)= 0 for anypand[θ]∈HdR1 (M),[θ]= 0.
The proofof Theorem 1.2is basedonCartan-Lerayspectral sequence onequivalent homology[4].Bypassingtoafinitecover,wecanassumethatMn isanilpotentspace.
The closedoneform θ onMn defines alinearrepresentationof thefundamentalgroup of Mn:
ρ:π1(Mn)→GL(1,C) =C∗,[γ]→e
γθ
.
TherepresentationρdefinesacomplexrankonelocalsystemCρoverMn[6].Wedenote byHp(Mn,Cρ) thep-thcohomologygroupofMnwithcoefficientsinthislocalsystem.
ByTheorem2.2 insection2,foranyp,wehave
Hp(Mn, θ)Hp(Mn,Cρ).
By duality, it suffices to show that Hp(Mn,Cρ) = 0, where Hp(Mn,Cρ) is the p-th homology groupof Mn with coefficientsinthis local system.Letπ:Mn →Mn be the universalcoverofMn.BytheCartan-Lerayspectralsequence[4],wehave
Ekl2 =Hk(π1(Mn), Hl(Mn,C))⇒Hk+l(Mn,Cρ), (1.1)
whereHk(π1(Mn),Hl(Mn,C)) isthek-th homologygroupofπ1(Mn) with coefficients intheπ1(Mn)-moduleHl(Mn,C).Then weprovebyinductionto getthevanishing of Hp(Mn,Cρ).
Theproof of Theorem 1.4 is basedon Hodgetheory of Morse-Novikov cohomology.
Let d∗ be the formal L2 adjoint of d with respect to the Riemannian metric gi. We canalso define an operator d∗θ as the formal L2 adjoint of dθ with respect to gi. Fur- ther,Δθ=dθd∗θ+d∗θdθ isthecorrespondingLaplacian.Theseoperatorsarelower-order perturbations of thecorresponding operators inthe usual Hodge-de Rham theory and therefore havemuch thesameanalytic properties. Forexample, the usualproof of the Hodgedecompositiontheoremgoesthrough,andoneobtainsanorthogonaldecomposi- tion
Ωp(Mn) =Hp(Mn)⊕dθ(Ωp−1(Mn))⊕d∗θ(Ωp+1(Mn)),
whereHp(Mn) isthespaceofΔθharmonicforms,whichisisomorphictoHp(Mn,θ).
By Hodge theory, for each i we can choose a harmonic form θi inthe cohomology class[θ].LetV(gi) bethevolumeof(Mn,gi),dVithevolumeformofgiandXithedual vectorfield ofθi definedby gi(Xi,Y)=θ(Y).Set ti = ( V(gi)
M n|Xi|2dVi)1/2>0.Choose a Δtiθi harmonicform αi inHp(Mn,tiθi).Theideaistoshowthatαi≡0 forsufficiently largei,whichreliesonthefollowingcrucial integralinequalityprovedinCorollary4.3.
Mn
t2i|Xi|2|αi|2dVi≤Cn
Mn
(ti|∇Xi|+ti|div(Xi)|)|αi|2dVi (1.2)
forsomeconstantCn dependingonlyonn.
AsRic(gi)≥ −n−i1,applyingBochnerformulato Xi,weget
Mn
|∇Xi|2dVi ≤n−1 i
Mn
|Xi|2dVi. (1.3)
Combining (1.2) and(1.3),forsufficientlylargeiwewillshow
Mn
|αi|2dVi≤ 1 2
Mn
|αi|2dVi.
Henceαi≡0.Seesection5fordetails.
Acknowledgments
Theauthor ispartially supportedby NationalNaturalScienceFoundation of China No. 11701427 and the Fundamental Research Funds for the Central Universities. He thanksProfessorBinglongChen,JohnLottandAndreiPajitnovforhelpfuldiscussions.
2. Basicproperties ofMorse-Novikovcohomology
Inthissection wecollectsomebasicpropertiesofMorse-Novikovcohomology.
Theorem2.1. LetMn beacompactn-dimensionalmanifoldandθ aclosedoneformon Mn.Then:
(1) If θ =θ+df,f ∈C∞(Mn,R),then for any p,we have Hp(Mn,θ) Hp(Mn,θ) and theisomorphismisgivenby themap[α]→[efα];
(2) If [θ] = 0 and Mn is connected and orientable, then H0(Mn,θ) and Hn(Mn,θ) vanish. Moreover, theintegration
:Hp(Mn,θ)×Hn−p(Mn,−θ),(α,β)→
Mnα∧β induces anisomorphism Hp(Mn,θ)(Hn−p(Mn,−θ))∗.
(3) n
p=0(−1)pdimHp(Mn,θ)isequaltotheEuler characteristicnumberof Mn; (4) IfNd bead-dimensionalmanifoldandγbe aclosed oneformon Nd,thenwehave Hk(Mn×Nd,π1∗θ+π∗2γ)
p+q=kHp(Mn,θ)
Hq(Nd,γ),where π1:Mn×Nd→ Mn,π2:Mn×Nd→Nd are theprojection maps.
(5) If π : Mn → Mn is a covering space with finite sheet, then π∗ : Hp(Mn,θ) → Hp(Mn,π∗θ) isinjectiveforany p.
Proof. Seepage 476-480 in[11] and Proposition1.2 in[18] for the proof of parts 1-4.
Forpart5,byTheorem 2.2,wehave
Hp(Mn, θ)Hp(Mn,Cρ),
where Cρ isthecomplexrankonelocalsystemdefinedbythelinearrepresentation ρ:π1(Mn)→GL(1,C) =C∗,[γ]→e
γθ
and Hp(Mn,Cρ) is the p-th cohomology group of Mn with coefficients in this local system.
As π:Mn →Mn is acoveringspace with finitesheet,onecanconstruct atransfer map (seee.g.[9,12]) h:Hp(Mn,π∗Cρ)→Hp(Mn,Cρ) such that hπ∗ =kId, where k is the degreeof π. It follows that π∗ : Hp(Mn,θ)Hp(Mn,Cρ)→ Hp(Mn,π∗Cρ) Hp(Mn,π∗θ) isinjective.
As acorollaryofTheorem 2.1,weget
Example 1.Let Mn be n-dimensionaltorus,thenHp(Mn,θ)= 0 forany pand[θ]= 0 byTheorem 2.1.
LetθbeaclosedoneformonMn.Considerthefollowinglinearrepresentationofthe fundamentalgroupofMn:
ρ:π1(Mn)→GL(1,C) =C∗,[γ]→e
γθ
.
TherepresentationρdefinesacomplexrankonelocalsystemCρoverMn[6].Wedenote byHp(Mn,Cρ) thep-thcohomologygroupofMn withcoefficientsinthislocalsystem.
Theorem2.2. Hp(Mn,θ)Hp(Mn,Cρ)forany p.
Proof. The proof is contained in [19]. For the convenience of the reader, we provide the details here. Let π : Mn → Mn be the universal cover of Mn. The cohomology groupsHp(Mn,Cρ) areisomorphicto Hρp(Mn),thecohomologygroupsofthecomplex Ω(Mn,ρ),consisting oftheρ-equivariantdifferentialforms onMn relative tothe usual differential(theproof isanalogous to thesheaf-theoreticproof ofde Rham’s theorem).
Let h be afunction on Mn such that dh = π∗θ. We give amapping F : Ω∗(Mn) → Ω∗(Mn,ρ) by the formula F(w) = ehπ∗w. It is easy to see that F is one-to-one and commuteswith thedifferentials.Hence
Hp(Mn, θ)Hρp(Mn)Hp(Mn,Cρ).
Theorem2.3. LetMn be an-dimensionalmanifold andθ a closed oneform on Mn.If thefundamental groupofMn has afinitelygenerated nilpotentsubgroup offinite index, thenH1(Mn,θ)=Hn−1(Mn,θ)= 0 forany [θ]= 0.
Proof. Let G⊆π1(Mn) be afinitely generated nilpotent subgroup of finite index and π: Mn →Mn the coveringspace of Mn with π1(Mn) G. The closed oneform π∗θ definesalinearrepresentationof G:
ρ:G=π1(Mn)→GL(1,C) =C∗,[γ]→e
γπ∗θ
.
TherepresentationρdefinesacomplexrankonelocalsystemCρoverMn.Wedenoteby Hp(Mn,Cρ) thep-thcohomologygroupofMn withcoefficientsinthelocal systemCρ. LetK(G,1) bethetopologicalspacesuchthatπ1(K(G,1))=G,πi(K(G,1))= 0,i≥2 andLρthecomplexrankonelocalsystemoverK(G,1) definedbyρ.Sincetheclassifying mapMn →K(G,1) inducesoverQacohomologyisomorphismindegreeone,weget
H1(Mn,Cρ)H1(K(G,1),Lρ).
Asπ:Mn→Mnisafinitecover,[θ]= 0 impliesthat[π∗θ]= 0.ThenLρisanontrivial localsystemoverK(G,1).AsGisafinitelygenerated nilpotentgroup,byTheorem2.2 in[17], foranyp,wehave
Hp(K(G,1),Lρ) = 0.
Inparticular,
H1(Mn,Cρ)H1(K(G,1),Lρ) = 0.
ByTheorem2.1 andTheorem2.2,we have
H1(Mn, π∗θ) = 0 H1(Mn, θ) = 0
Hn−1(Mn, θ)H1(Mn,−θ) = 0.
For a smooth manifold which is not an almost nilpotent space, its Morse-Novikov cohomology doesnotnecessarilyvanishas thefollowing exampleshows.
Example 2.[15] Leth:S3×S3→S3×S3 bedefinedby h: (x, y)→(xy, yxy).
This mapisadiffeomorphismwithinversegivenby h−1: (u, v)→(u2v−1, vu−1).
LetM bethemappingtorus ofh.ThenM hasthestructureofafiberbundle:
S3×S3→M→S1.
Theinducedmap h∗,3 onHdR3 (S3×S3) isgivenbythematrix Ah= 1 1
1 2
(2.1) Notice thatthe eigenvaluesof Ah are different from 1 inabsolute value. HenceMn is not an almost nilpotent space. Letλ be aeigenvalue ofAh with λ=e−t,t= 0,t ∈R and θ ageneratorof HdR1 (M). WeclaimthatH3(M,tθ)= 0. Tosee this, observe that tθ defines alinearrepresentationofthefundamentalgroupofM:
ρt:π1(M)→GL(1,C) =C∗,[γ]→et
γθ
.
The representation ρt defines a complex rank one local system Cρt over Mn [6]. We denote byHp(Mn,Cρt) thep-thcohomologygroupofMn withcoefficientsinthislocal system.ByTheorem2.2insection2,foranyp,wehave
Hp(M, tθ)Hp(Mn,Cρt).
On the other hand,by Wang’sexactsequence in Proposition6.4.8 in[6] page 212,we have
dimCHp(Mn,Cρt) = dimCker(h∗,p−e−tId) + dimCcoker(h∗,p−1−e−tId),
where h∗,p :Hp(S3×S3,C)→Hp(S3×S3,C) isthelinearmap inducedby h. Ase−t isaneigenvalueof h∗,3,weseethatdimCker(h∗,3−e−tId)>0 andH3(M,tθ)= 0.
3. Cartan-Lerayspectralsequence
In this section we apply Cartan-Leray spectral sequence to proveTheorem 1.2. By passingto a finite cover, wecan assumethat Mn isa nilpotent space. Theclosed one formθ inducesalinearrepresentation ofG=π1(Mn):
ρ:π1(Mn)→GL(1,C) =C∗,[γ]→e
γθ
.
ByTheorem 2.2,foranyp,wehave
Hp(Mn, θ)Hp(Mn,Cρ),
where Cρ is the complex rank one local system over Mn defined by ρ. By duality, it suffices to prove the vanishing of Hp(Mn,Cρ), which is the homology group of Mn with coefficients in the local system Cρ. Let Mn be the universal cover of Mn. The representationρtogetherwiththeGactiononMn bydecktransformationinduces the diagonalactiononHl(Mn,C)Hl(Mn,Z)⊗C.BytheCartan-Lerayspectralsequence (Theorem7.9,page173in[4]),wehave
Ekl2 =Hk(G, Hl(Mn,C))⇒Hk+l(Mn,Cρ),
where Hk(G,Hl(Mn,C)) is the k-th homology groupof G with coefficients in the G- moduleHl(Mn,C).See[4] formoredetailsofhomologyofgroups.Forus, weonlyneed thefollowinglongexactsequence(Proposition6.1,page71in[4]).
Lemma3.1. Foranyshortexact sequence0→M→M →M→0ofG-modules,there isthefollowinglongexact sequence:
· · · →Hi(G, M)→Hi(G, M)→Hi(G, M)→Hi−1(G, M)→Hi−1(G, M)→ · · ·
→H1(G, M)→H1(G, M)→H1(G, M)→H0(G, M)→H0(G, M)
→H0(G, M)→0.
As Mn is a nilpotent space, then G = π1(Mn) is a nilpotent group that operates nilpotentlyonπm(Mn) foreverym≥2.ByLemma2.18in[13],Goperatesnilpotently onHl(Mn,Z) foreveryl,thatisV =Hl(Mn,Z) admitsafinitesequenceofG-invariant subgroups
V =V0⊇V1⊇. . . Vk = 0
such thatthe inducedactionofGonVj/Vj+1 istrivialforany j.Therepresentation ρ ofGinducesadiagonalactiononVj⊗C andwehavethefollowingshortexactsequence of Gmodules:
0→Vj+1⊗C→Vj⊗C→Vj/Vj+1⊗C→0.
Wenow proveHk(G,Vj⊗C)= 0 foranyj byinduction.It isclearthatHk(G,Vk⊗ C)= Hk(G,0) = 0. As [θ] = 0, we see thatρ is a nontrivial representation of G. By assumption,theinduced actionofGonVj/Vj+1 is trivialforanyj. Thenthediagonal action of GonVj/Vj+1⊗C is nontrivial.As Gisafinitely generated nilpotent group, byTheorem 2.2in[17],weget
Hk(G, Vj/Vj+1⊗C) = 0.
ByLemma3.1andinduction,foranyj,weget Hk(G, Vj⊗C) = 0.
Inparticular,
Hk(G, Hl(Mn,C)) =Hk(G, V0⊗C) = 0.
BytheCartan-Lerayspectralsequence[4],wehave
Ekl2 =Hk(G, Hl(Mn,C))⇒Hk+l(Mn,Cρ).
Henceforanyk,l≥0,wehave
Hk+l(Mn,Cρ) = 0.
Then wegetHp(Mn,θ)= 0 foranypand[θ]= 0.
4. Anintegral formulaofΔθharmonicforms
In sectionwe derive anintegralformula ofΔθ harmonicforms whichwill be crucial intheproofofTheorem1.4.
Let (Mn,g) bea closedRiemannian manifold and θ aclosed real oneform on Mn. Define dθ : Ωp(Mn)→Ωp+1(Mn) as dθα=dα+θ∧αfor α∈Ωp(Mn).Let d∗ be the formalL2adjointofdwithrespecttog.Wecanalsodefineanoperatord∗θastheformal L2adjointofdθwithrespecttog.Further,Δθ=dθd∗θ+d∗θdθisthecorrespondingLapla- cian. These operators are lower-order perturbations of the corresponding operators in theusualHodge-deRhamtheoryandthereforehavemuchthesameanalyticproperties.
For example, the usual proof of the Hodge decomposition theorem goes through, and oneobtainsanorthogonaldecomposition
Ωp(Mn) =Hp(Mn)⊕dθ(Ωp−1(Mn))⊕d∗θ(Ωp+1(Mn)),
whereHp(Mn) isthespaceofΔθharmonicforms,whichisisomorphictoHp(Mn,θ).
LetdV bethevolumeformofgandX thedualvectorfieldofθdefinedbyg(X,Y)= θ(Y).ChooseaΔθ harmonicformαinHp(Mn,θ).Then
dθα=dα+θ∧α= 0 d∗θα=d∗α+iXα= 0.
The following integral formula and its Corollary 4.3 will be crucial in the proof of Theorem1.4.
Theorem4.1.
Mn
|X|2|α|2dV =1 2
Mn
α∧[LX,∗]α,
where[LX,∗]α=LX∗α− ∗LXαandLXαistheLiederivative ofαinthedirectionX.
Remark 4.2.When θ is exact and X = ∇f for some smooth function f on Mn, we believethat the integral formula inTheorem 4.1 is the sameas [7]. It is also possible toadaptthemethodin[7] toproveTheorem4.1.However,wepresentadifferent proof here.
Corollary4.3.
Mn
|X|2|α|2dV ≤Cn
Mn
(|∇X|+|div(X)|)|α|2dV
forsomeconstant Cn depending only onn.
Proof. TheRiemannianmetricgonMninducesalinearmapbetweenT MnandT∗Mn definedby
g:T Mn→T∗Mn
< g(X), Y >=g(X, Y),∀X, Y ∈T Mn.
Let g−1 be the inverse of the above map g and h the endomorphism of the bundle T∗Mn→Mn by
h=LXg◦g−1.
ThederivationoftheGrassmannalgebraΛT∗Mninducedbyhisdenotedbyi(h).This is alinearmapsuchthat,ifγ∈T∗Mn, theni(h)(γ)=h(γ),and
i(h)(ω1∧ω2) = (i(h)ω1)∧ω2+ω1∧(i(h)ω2) (4.1) forany ω1,ω2∈ΛT∗Mn. Thefollowing formulaisprovedin[21].
[LX,∗]ω= (i(h)−1
2T rh)∗ω (4.2)
forany ω∈ΛT∗Mn.
Letdiv(X) bethedivergenceofX withrespectto g.As (LXg)(Y, Z) = g(∇YX, Z) + g(Y,∇ZX)
forallY,Z∈T Mn,we seethatT rh= 2div(X).ThenbyTheorem4.1,we get
Mn
|X|2|α|2dV ≤Cn
Mn
(|∇X|+|divX|)|α|2dV
forsomeconstantCn depending onlyonn.
Now weproveTheorem 4.1.Wefirstlyneedthefollowinglemmas.
Lemma 4.4.Forany pformω,wehave
∗iXω= (−1)p−1θ∧ ∗ω, (4.3) where ∗istheHodgestaroperatorwith respecttog.
Proof. Foranyp−1 formξ,wehave
Mn
ξ∧ ∗iXω=
Mn
g(ξ, iXω)dV
=
Mn
g(θ∧ξ, ω)dV =
Mn
θ∧ξ∧ ∗ω
= (−1)p−1
Mn
ξ∧θ∧ ∗ω.
Hence
∗iXω= (−1)p−1θ∧ ∗ω.
Lemma4.5. Letβ=∗α,then
dβ−θ∧β= 0.
Proof. Asd∗α= (−1)n(p+1)+1∗d∗αand d∗α+iXα= 0,weget (−1)n(p+1)+1∗d∗α+iXα= 0.
Hence
(−1)n(p+1)+1∗ ∗d∗α+∗iXα= 0.
ByLemma4.4,wehave
∗iXα= (−1)p−1θ∧ ∗α.
Itfollowsthat
(−1)pd∗α+ (−1)p−1θ∧ ∗α= 0 So
dβ−θ∧β= 0.
Nowweproceed toproveTheorem 4.1.Asdα+θ∧α= 0,weget iXdα+iX(θ∧α) = 0.
So
iXdα∧β+|X|2α∧β−θ∧iXα∧β = 0. (4.4) Ontheotherhand,asdβ−θ∧β= 0,we get
iXdβ−iX(θ∧β) = 0.
So
iXdβ∧α− |X|2β∧α+θ∧iXβ∧α= 0.
Then
α∧iXdβ− |X|2α∧β+ (−1)pθ∧α∧iXβ = 0. (4.5)
By(4.4),(4.5),weget
−iXdα∧β+α∧iXdβ−2|X|2α∧β+θ∧iXα∧β+ (−1)pθ∧α∧iXβ = 0. (4.6) Combined with
θ∧iXα∧β+ (−1)pθ∧α∧iXβ=θ∧iX(α∧β)
=|X|2α∧β−iX(θ∧α∧β) =|X|2α∧β, we get
−iXdα∧β+α∧iXdβ=|X|2α∧β. (4.7) Since
d(iXα∧β) =diXα∧β+ (−1)p−1iXα∧dβ, we get
Mn
iXα∧dβ= (−1)p
Mn
diXα∧β. (4.8)
Ontheother hand,wehave
0 =iX(α∧dβ) =iXα∧dβ+ (−1)pα∧iXdβ. (4.9) Combining (4.8),(4.9),weget
Mn
α∧iXdβ=−
Mn
diXα∧β. (4.10)
From (4.7),(4.10),weget
Mn
|X|2α∧β=−
Mn
iXdα∧β−
Mn
diXα∧β =−
Mn
LXα∧β
=−
Mn
LX(α∧β) +
Mn
α∧LXβ=
Mn
α∧LXβ. (4.11)
As β=∗α,weget
Mn
α∧LXβ =
Mn
α∧LX∗α=
Mn
α∧ ∗LXα+
Mn
α∧[LX,∗]α. (4.12)
Moreover,
Mn
α∧ ∗LXα=
Mn
LXα∧ ∗α
=
Mn
LX(α∧ ∗α)−
Mn
α∧LX∗α=−
Mn
α∧LX∗α
=−
Mn
α∧ ∗LXα−
Mn
α∧[LX,∗]α.
Hence
Mn
α∧ ∗LXα=−1 2
Mn
α∧[LX,∗]α. (4.13)
By(4.11),(4.12),(4.13),weget
Mn
|X|2|α|2dV =1 2
Mn
α∧[LX,∗]α.
5. ProofofTheorem1.4
Inthis sectionwe giveaproof ofTheorem 1.4. Theproof isbasedon Corollary4.3.
Anothercrucialtoolis thefollowing Poincaré-Sobolevinequality([2],page397).
Theorem5.1. Let(Mn,g)be a closed smoothRiemannian manifoldsuchthat forsome constant b>0,
rmin(g)D2(g)≥ −(n−1)b2,
whereD(g)isthediameterof g,Ric(g) istheRiccicurvature ofg and
rmin(g) =inf{Ric(g)(u, u) :u∈T M, g(u, u) = 1}. LetR= bC(b)D(g),whereC(b)istheuniquepositiveroot oftheequation
x b
0
(cht+xsht)n−1dt= π
0
sinn−1tdt.
Thenforeach 1≤p≤n−qnq ,p<∞and f ∈W1,q(Mn),we have f− 1
V(g)
Mn
f dVp≤Sp,qdfq
fp≤Sp,qdfq+V(g)1/p−1/qfq,
where V(g)isthevolumeof (Mn,g),S(p,q)= (V(g)/vol(Sn(1))1/p−1/qRΣ(n,p,q) and Σ(n,p,q)is theSobolevconstant of thecanonicalunitsphereSn defined by
Σ(n, p, q) =sup{fp/dfq:f ∈W1,q(Sn), f = 0,
Sn
f = 0}.
Let p = n−22n ,q = 2 in Theorem 5.1 and apply Theorem 3and Proposition6 in [2]
pages395-396, thenwegetthefollowingmeanvalueinequality.
Theorem 5.2.Let n ≥ 3 and (Mn,g) be a closed n-dimensional smooth Riemannian manifold suchthatforsome constant b>0,
rmin(g)D2(g)≥ −(n−1)b2.
If f ∈W1,2(Mn)isa nonnegative continuousfunction suchthat fΔf ≥ −cf2 (here Δ is anegativeoperator)inthesense of distribution forsomepositivenumber c,then
maxx∈Mn|f|2(x)≤Bn(σnRc1/2)
Mnf2dV V(g) ,
where σn =vol(Sn)1/nΣ(n,n−22n ,2) andBn :R+→R+ isa functiondefinedby
Bn(x) = ∞ i=0
(xνi(2νi−1)−1/2+ 1)2ν−i, ν = n n−2. The functionBn satisfiestheinequalities
Bn(x)≤exp(2x√ ν/(√
ν−1)),0≤x≤1 Bn(x)≤Bn(1)x2ν/(ν−1), x≥1.
In particular,limx→0+Bn(x)= 1 andBn(x)≤Bn(1)xn forx≥1.
Let Mn be aclosed Riemannian manifold with nonzero first de Rham cohomology groupandadmits asequenceofRiemannianmetricsgi suchthat
Ric(gi)≥ −n−1 i D(gi)≤1.
Moreover,thecurvatureoperatorofgiisuniformlyboundedfrombelowby−Id.Forany [θ]∈HdR1 (Mn),[θ]= 0, we aregoing to provethatthereexists somet ∈R,t= 0 such
thatHp(Mn,tθ)= 0 forany p. Ifn= 2,sincethefirstBetti numberofM2isbounded by2 (seee.g.[2]),thegenusofM2isatmost1 andHp(M2,tθ)= 0 byExample1.Now weassumethatn≥3.Letd∗ betheformalL2 adjointofdwithrespecttogi.ByHodge theory,wecanchooseaharmoniconeformθiinthecohomologyclass[θ].Then
dθi= 0 d∗θi = 0
θi= 0.
Let ti = ( V(gi)
M n|Xi|2dVi)1/2 > 0, where V(gi) is the volume of (Mn,gi), dVi is the volume form of gi, |Xi|2 = gi(Xi,Xi) and Xi is thedual vector field of θi defined by gi(Xi,Y) = θ(Y). We claim that for sufficiently large i, Hp(Mn,tiθi) = 0 for any p.
ChooseaΔtiθi harmonicform αi inHp(Mn,tiθi).Then dαi+tiθi∧αi= 0 d∗αi+itiXiαi= 0.
Thegoalis to provethatαi = 0. ByTheorem 2.1, wecanassume that1≤deg(αi)≤ n−1.AsRic(gi)≥ −n−1i ,applyingBochnerformula toXi [20], weget
1
2Δ|Xi|2=|∇Xi|2+Ric(gi)(Xi, Xi)≥ |∇Xi|2−n−1
i |Xi|2, (5.1) whereΔ istheLaplacianactingonfunctionswhichisanegativeoperator.Then
Mn
|∇Xi|2dVi ≤n−1 i
Mn
|Xi|2dVi. (5.2)
Letdiv(Xi) bethedivergenceofXi withrespect togi.Asθiisaharmoniconeform, weseediv(Xi)= 0 (seee.g.Proposition31in[20] page206).ByCorollary4.3,wehave
Mn
t2i|Xi|2|αi|2dVi≤Cn
Mn
ti|∇Xi||αi|2dVi, (5.3)
for someconstantCn depending only onn. Applying Hölder’s inequalityon (5.3) and using(5.2),weget
Mn
t2i|Xi|2|αi|2dVi≤Cn
Mn
ti|∇Xi||αi|2dVi
≤Cn(
Mn
t2i|∇Xi|2dVi)12(
Mn
|αi|4dVi)12