ebook img

Growth of regulators in finite abelian coverings PDF

22 Pages·2013·0.34 MB·English
by  
Save to my drive
Quick download
Download
Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.

Preview Growth of regulators in finite abelian coverings

Algebraic & Geometric Topology13(2013)2383–2404 msp Growth of regulators in finite abelian coverings THANG TQ LÊ Weshowthattheregulator,whichisthedifferencebetweenthehomologytorsionand thecombinatorialRay–Singertorsion,offiniteabeliancoveringsofafixedcomplex hassub-exponentialgrowthrate. 54H20;57Q10,37B50,37B10 1 Introduction 1.1 Basedfreecomplexovergroupringanditsquotients Suppose (cid:25) isafinitelypresentedgroupand ZŒ(cid:25)(cid:141) isthegroupringof (cid:25) overthering Z ofintegers. Let C beafinitelygeneratedbasedfree ZŒ(cid:25)(cid:141)–complex @ @ @ @ 0!Cm(cid:0)!m Cm(cid:0)1 (cid:0)m!(cid:0)1Cm(cid:0)2(cid:0)!(cid:1)(cid:1)(cid:1)(cid:0)!2 C1(cid:0)!1 C0!0: Here “based free” means each C is a free ZŒ(cid:25)(cid:141)–module equipped with a preferred k base. For a normal subgroup (cid:128) C (cid:25) let C(cid:128) WD ZŒ(cid:25)=(cid:128)(cid:141)˝ZŒ(cid:25)(cid:141)C. Assume that the index Œ(cid:25) W (cid:128)(cid:141) is finite. Then C(cid:128) is a finitely generated based free Z–complex, where the preferredbaseof ZŒ(cid:25)=(cid:128)(cid:141)˝ZŒ(cid:25)(cid:141)Ck isdefinedusingtheoneof Ck inanaturalway. A prototypical case is the following. Suppose Xz ! X is a regular covering with (cid:25) the group of deck transformations and X a finite CW–complex. Choose a lift in z z X of each cell of X. Then the CW–complex C of X induced from that of X is a finitelygeneratedbasedfree ZŒ(cid:25)(cid:141)–complex. Foranormalsubgroup (cid:128) C(cid:25), C(cid:128) isthe CW–complex ofthecovering X(cid:128),corresponding tothegroup (cid:128),and Hk.C(cid:128)/ isthe kth homologyofthecovering X(cid:128). Usually,interestinginvariantsdonotdependonthe choiceoftheliftsofcellsof X. Published: 2July2013 DOI:10.2140/agt.2013.13.2383 2384 ThangTQLê 1.2 Twotorsions Wecandefinetwotorsionsofthequotientcomplex C(cid:128),thehomologytorsion (cid:28)H.C(cid:128)/ andthecombinatorialRay–Singertorsion (cid:28)RS.C(cid:128)/,asfollows. Thehomologytorsion is (cid:18)Y(cid:3) (cid:19)(cid:0)1 (cid:28)H.C(cid:128)/WD jtorZ.Hk.C(cid:128)//j 2RC; k where tor .M/ isthe Z–torsionpartofthefinitelygeneratedabeliangroup M,and Z Q(cid:3) a isthealternatingproduct k k Y(cid:3)a DYa.(cid:0)1/k: k k k k TheRay–Singertorsionof C(cid:128) is (cid:28)RS.C(cid:128)/DY(cid:3)det0.@k/2RC k 0 where det isthegeometricdeterminantoflinearmapsbetweenbasedHermitianspaces. 0 Werecallthedefinitionof det inSection2. 1.3 Comparison: generalquestion We want tocompare theasymptotics ofthe two torsionsas (cid:128) becomes “thinnerand thinnerin (cid:25)”,sothat (cid:25)=(cid:128) approximates (cid:25) inthefollowingsense. Afiniteset S of generatorsof (cid:25) definesawordlengthfunction l (andhenceametric)on (cid:25). Define S h(cid:128)iWDminfl .x/jx2(cid:128)nfegg: S Here e istheunitof (cid:25). Inallthatfollows,statementsdonotdependonthechoiceof thegeneratorset S,sincethemetricsoftwodifferentgeneratorsetsarequasi-isometric. We are interested in the following question: Suppose C is L2–acyclic (see eg Lück [12]). Underwhatconditionsdoesitholdthat (1) lim ln.(cid:28)H.C(cid:128)//(cid:0)ln.(cid:28)RS.C(cid:128)// D0? j(cid:25) W(cid:128)j h(cid:128)i!1; j(cid:25)W(cid:128)j<1 Themotivationofthisquestioncomesfromthequestion[12]: canoneapproximate L2–torsionsbyfinite-dimensionalanalogs? Insomefavorableconditions,oneexpects thatthegrowthrateofeachof (cid:28)H and (cid:28)RS isthe L2–torsion,hencetheymustbethe same. Algebraic & Geometric Topology,Volume13(2013) Growthofregulatorsinfiniteabeliancoverings 2385 Remark 1.1 (a) If f(cid:128) ;n D 1;2;:::g is a sequence of exhausting nested normal n subgroupsof (cid:25),ie, (cid:128)nC1(cid:26)(cid:128)n and Tn(cid:128)nDfeg,then limn!1h(cid:128)niD1. Thelimit in(1)ismoregeneral(stronger)thanthelimitofanexhaustingnestedsequence,aswe donothavethe“nested”property. (b) Thereexistsasequence (cid:128)nC(cid:25) suchthat limn!1h(cid:128)niD1 ifandonlyif (cid:25) is residuallyfinite. Hence,thelefthandsideof(1)makessenseonlywhen (cid:25) isresidually finite. (c) Define tr(cid:25).x/Dıx;e for x2(cid:25). Thisfunctionaltraceisthebaseforthedefinition ofmanycombinatorial L2–invariants. Forafixed x2(cid:25),wehave (2) lim tr(cid:25)=(cid:128).x/Dtr(cid:25).x/: h(cid:128)i!1 Thisisthereasonwhyoneexpectsthatas h(cid:128)i!1,many L2–invariants(undersome technicalconditions)canbeapproximatedbythecorrespondinginvariantsof (cid:25)=(cid:128). 1.4 Mainresults Themainresultofthepapertreatsthecase (cid:25) DZn. Theorem1 Suppose C isan L2–acyclicfinitelygeneratedbasedfree CŒZn(cid:141)–complex. Then(1),with (cid:25) DZn,holdstrue. Wewillnotgivethedefinitionof L2–acyclicity. Instead,for (cid:25) DZn,wewillusean equivalent definition (Elek [4], Lück [12]): the L.2/ homology H.2/.C/ vanishes if k and only if Hk.C˝ZŒZn(cid:141)F/D0. Here F is the fractional field of the commutative domain ZŒZn(cid:141). Remark1.2 (a) Ourresultdoesnotimplythat lim ln.(cid:28)H.C(cid:128)// D lim ln.(cid:28)RS.C(cid:128)// ; j(cid:25) W(cid:128)j j(cid:25) W(cid:128)j h(cid:128)i!1; h(cid:128)i!1; j(cid:25)W(cid:128)j<1 j(cid:25)W(cid:128)j<1 as we cannot prove the existence of each of the limits. For (cid:25) D Z, it was known thatbothlimitsexistandareequaltothe L2–torsionof C;seeGonzález-Acuñaand Short [6], Riley [17], and Lück [12]. Even for the case where (cid:25) D Z2 and C is a 2–termcomplex 0!C !C !0 (sothatonly H .C/ isnon-trivial),thereisstill 1 0 0 no proof of the conjecture that the L2–torsion is equal to either of the above limits. Forresultsanddiscussionsofthisandrelatedconjectures,seeLück[12;11],Bergeron andVenkatesh[1],Lê[10;9],FriedlandJackson[5],andSilverandWilliams[20]. Algebraic & Geometric Topology,Volume13(2013) 2386 ThangTQLê (b) Itshouldbenotedthattheexactcalculationofthetorsionpartofthehomologyof finitecoverings,evenintheabeliancase,isverydifficult;seeHillmanandSakuma[7], MayberryandMurasugi[13],andPorti[15]forsomepartialresults. 1.5 Refinement Suppose (cid:25) is residually finite and the L2–homology H.2/.C/D0 for some k. For k anynormalsubgroup (cid:128) C(cid:25) offinite index, thehomologygroup Hk.C(cid:128)/ isa finitely generated abelian group. Because Hk.2/.C/ D 0 one should expect that Hk.C(cid:128)/ is negligible. Infact,atheoremofLück[11](andKazhdanforthiscase)saysthat lim rkZHk.C(cid:128)/ D0: j(cid:25) W(cid:128)j h(cid:128)i!1; j(cid:25)W(cid:128)j<1 Thismeansthefreepart Hk.C(cid:128)/free of Hk.C(cid:128)/ issmallcomparedtotheindex. There is another measure of the free part Hk.C(cid:128)/free, denoted by Rk.C(cid:128)/ and called the regulator, or volume; see [1] and Section 3. Another expression of the fact that Hk.C(cid:128)/free is small compared to the index is expressed in the following statement, whichcomplementstheresultofKazhdanandLück. Theorem2 Suppose C isafinitelygeneratedbasedfree ZŒ(cid:25)(cid:141)–complexwith (cid:25) DZn and H.2/.C/D0 forsomeindex k. Then k (3) lim lnvol.Hk.C(cid:128)/free/ D0: j(cid:25) W(cid:128)j h(cid:128)i!1; j(cid:25)W(cid:128)j<1 Remark1.3 Thequestion(andsomeformoftheconjecture)aboutthegrowthrateof regulatorswasfirstraisedin[1],where,amongotherthings,aspecialcaseofTheorem2 was established: It was proved that if (cid:25) D Zn and (cid:128) runs the set of sublattices of the form kZn, then (3) holds. The proof there can be modified to include the case when (cid:128) runsthesetofuniformsublattices,asdefinedin[16]. Ourresultremovesany restrictionon (cid:128). 1.6 Ontheproofs Fortheproofsweusetoolsincommutativealgebraandalgebraicgeometry. Inparticular, we make essential use of the theory of torsion points in Q–algebraic set (a simple versionoftheManin–Mumfordprinciple). Wehopethatthemethodsandresultscan beadaptedtothecaseofelementaryamenablegroups. Algebraic & Geometric Topology,Volume13(2013) Growthofregulatorsinfiniteabeliancoverings 2387 Acknowledgements IwouldliketothankMBaker,NBergeron,HDao,WLück,AThom,UZannierand theanonymousrefereeforhelpfuldiscussions/comments. Theauthorissupportedin partbyanNSFgrant. 1.7 Organizationofthepaper InSection2werecallthenotionsofgeometricdeterminantandvolume. Wediscuss therelationbetweenhomologyandRay–SingertorsionsinSection3. Anoverviewof thetheoryoftorsionpointsinalgebraicsetisgiveninSection4. Section5containsa crucialgrowthestimatewhich isneededintheproofsofthemain theorems,givenin Section6. 2 Geometric determinant, lattices and volume in based Her- mitian spaces Inthissectionwerecallthedefinitionofgeometricdeterminantandbasicfactsabout volumesoflatticesinbasedHermitianspaces. 2.1 Geometricdeterminant Foralinearmap fW V !V ,whereeach V isafinite-dimensionalHermitianspace, 1 2 i thegeometricdeterminant det0.f/ isthe product ofallnon-zero singular values of f . Recall that x 2R is singular value of f if x (cid:21)0 and x2 is an eigenvalue of f(cid:3)f . Byconvention, det0.f/D1 if f isthe0map. Thuswealwayshave det0.f/>0. Sincethemaximalsingularvalueof f isthenorm kfk,wehave (4) det0.f/(cid:20)kfkdimV2 if f isnon-zero. Remark2.1 Thegeometricmeaningof det0f isthefollowing. Themap f restricts toalinearisomorphism f0 from Im.f(cid:3)/ to Im.f/;eachisaHermitianspace. Then det0f D jdet.f0/j, where the ordinary determinant det.f0/ is calculated using or- thonormalbasesoftheHermitianspaces. 2.2 BasedHermitianspaceandvolume Suppose W is a finite-dimensional based Hermitian space, ie, a C–vector space equipped with an Hermitian product .(cid:1);(cid:1)/ and a preferred orthonormal basis. The Z–submodule (cid:127)(cid:26)W spannedbythebasisiscalledthefundamentallattice. Algebraic & Geometric Topology,Volume13(2013) 2388 ThangTQLê Fora Z–submodule(alsocalledalattice) ƒ(cid:26)W with Z–basis v ;:::;v ,define 1 l vol.ƒ/Dˇˇdet(cid:0).vi;vj/li;jD1(cid:1)ˇˇ1=2: Byconvention,thevolumeofthe0spaceis1. If ƒ(cid:26)(cid:127),wesaythat ƒ isanintegral lattice. Itisclearthat vol.ƒ/(cid:21)1 if ƒ isanintegrallattice. Fora C–subspace V (cid:26)W ,thelattice V.Z/WDV \(cid:127) iscalledthe Z–supportof V . Wedefine vol.V/WDvol.V.Z//: Alattice ƒ(cid:26)(cid:127) isprimitiveifiscutoutfrom (cid:127) bysomesubspace,ie, ƒDV.Z/ for somesubspace V (cid:26)W . Bydefinition,anyprimitivelatticeisintegral. As usual, we say that a subspace V (cid:26)W is definedover Q if it is defined by some linear equationswith rational coefficients(using the coordinatesin the preferredbase). Itiseasytoseethat V isdefinedover Q ifandonlyifitisspannedbyits Z–support. Suppose V ;V aresubspacesof W definedover Q,and fW V !V isa C–linear 1 2 1 2 map. Wesaythat f isintegralif f.V.Z//(cid:26)V.Z/. 1 2 Wesummarizesomewell-knownpropertiesofvolumesoflattices(seeegBertrand[2]). Proposition 2.1 Suppose V ;V are subspaces of W defined over Q of a based 1 2 Hermitianspace W and fW V !V isanintegral,non-zero C–linearmap. Then 1 2 (5) vol.V CV /(cid:20)vol.V / vol.V /; 1 2 1 2 (6) vol.kerf/volŒf.V.Z//(cid:141)Ddet0.f/ vol.V /: 1 1 Foradetaileddiscussionof(6)anditsgeneralizationstolatticesinZŒZn(cid:141),seeRaimbault [16]. 3 Regulator, homology torsion and Ray–Singer torsion Inthissectionweexplaintherelationbetweenthehomologytorsionandthecombina- torialRay–Singertorsion. Proposition3.1ofthissectionwillbeusedintheproofof maintheorems. Throughoutthissectionwefixafinitelygeneratedbasedfree Z–complex E 0!Emd(cid:0)m!(cid:0)1Em(cid:0)1d(cid:0)m!(cid:0)1Em(cid:0)2(cid:0)!(cid:1)(cid:1)(cid:1)(cid:0)d!2 E1(cid:0)d!1 E0!0: DefineaHermitianproducton E ˝ C suchthatthepreferredbaseisanorthonormal k Z base. Now E ˝ C becomesabasedHermitianspace. k Z Algebraic & Geometric Topology,Volume13(2013) Growthofregulatorsinfiniteabeliancoverings 2389 Weusethenotation Zk Dkerdk; Bk DImdkC1; Bk D.Bk˝ZC/\Ek: Let dk(cid:3)W Ek(cid:0)1!Ek beadjointof dk and DkW Ek !Ek bedefinedby Dk Ddk(cid:3)dkCdkC1dk(cid:3)C1: 3.1 Ray–Singertorsionandhomologytorsion DefinetheRay–Singertorsionandthehomologytorsionof E by (cid:28)RS.E/DY(cid:3)det0.dk/2RC; k (cid:18)Y(cid:3) (cid:19)(cid:0)1 (cid:28)H.E/D jtor .H .E//j : Z k k Remark3.1 TheRay–Singertorsionandthehomologytorsioncanbedefinedthrough theclassicalReidemeistertorsionasfollows. Let hz beanorthonormalbasisof ker.D /˝ CDH .E˝ C/. Withthebases fhz g k k Z k Z k ofthehomologyof E˝ C,onecandefinetheReidemeistertorsion (cid:28)R.E˝ C;fhz g/, Z Z k defineduptosigns(seeegTuraev[21]). Itisnotdifficulttoshowthat (cid:28)RS.E/Dˇˇ(cid:28)R.E˝ZC;fhzkg/ˇˇ: Both B and Z areprimitivelatticesin E ,and B (cid:26)Z . Thereisacollection h k k k k k k ofelementsof Z (cid:26)E thatdescendtoabasisofthegroup Z =B ,thefreepartof k k k k H .E/. Sinceh isabasisofH .E˝ C/,theReidemeistertorsion(cid:28)R.E˝ C;fh g/ k k k Z Z k isdefined. ItisnotdifficulttoprovethefollowinggeneralizationoftheMilnor–Turaev formula[14;21]: (cid:28)H.E/Dj(cid:28)R.E˝ C;fh g/j: Z k 3.2 Regulators By definition, H .E/DZ =B . The Z–torsion of H .E/ is B =B , and the free k k k k k k part H .E/ is isomorphic to Z =B . For this reason, we define the volume k free k k vol.H .E/ / tobe k free vol.Z / R .E/WD k : k vol.B / k Algebraic & Geometric Topology,Volume13(2013) 2390 ThangTQLê Herewefollowthenotationof[1],where R iscalledtheregulator. UsingIdentity k (6),onecanprove(see[1,Formula2.2.4]) (cid:18) (cid:19) Y(cid:3) (7) (cid:28)RS.E/D(cid:28)H.E/ R .E/ : k k Wewillusethefollowingestimateoftheregulator. Proposition3.1 Let Rz WDvol.kerD /. Forevery k,onehas k k Rz (cid:21)R (cid:21) 1 : k k z R k Proof Let W betheorthogonalcomplementof B ˝ C in Z ˝ C and pW Z ˝ k Z k Z k Z C!W betheorthogonalprojection. Then R Dvol.p.Z //: k k ByHodgetheory(forfinitelygenerated Z–complexes), ker.D /DE \W DW.Z/: k k Itfollowsthat ker.D /(cid:26)p.Z /,andhence vol.p.Z //(cid:20)vol.ker.D //,or k k k k (8) R (cid:20)Rz : k k By[2,Proposition1(ii)], vol.Z / Œ.W \Z(cid:3)/WZ (cid:141) (9) R D k D k k ; k vol.B / Rz k k where Z(cid:3) isthe Z–dualof Z in Z ˝ C undertheinnerproduct. Notethat Z(cid:3) is k k k Z k alsotheorthogonalprojectionof E onto Z ˝ C. k k Z Since the numerator of (9) is (cid:21) 1, we have R (cid:21) 1=Rz , which, together with (8), k k provestheproposition. 4 Abelian groups, algebraic subgroups of .C(cid:3)/n and torsion points WereviewsomefactsaboutrepresentationtheoryoffiniteabeliangroupsinSection4.1 andthetheoryoftorsionpointsonrationalalgebraicsets(asimpleversionofManin– Mumfordprinciple)inSections4.2and4.3. Algebraic & Geometric Topology,Volume13(2013) Growthofregulatorsinfiniteabeliancoverings 2391 4.1 Decompositionofthegroupringofafiniteabeliangroup Suppose A is a finite abelian group. The group ring CŒA(cid:141) is an A–module (the regularrepresentation)andisa C–vectorspaceofdimension jAj. Equip CŒA(cid:141) with a Hermitian product so that A is an orthonormal basis. This makes CŒA(cid:141) a based Hermitianspace,with ZŒA(cid:141) thefundamentallattice. LetAyDHom.A;C(cid:3)/,knownasthePontryagindualofA,bethegroupofallcharacters (cid:3) of A. Here C is the multiplicative group of non-zero complex numbers. We have jAyjDjAj. Thetheoryofrepresentationsof A over C iseasy: CŒA(cid:141) decomposesasadirectsum ofmutuallyorthogonalone-dimensional A–modules: M (10) CŒA(cid:141)D Ce(cid:31); (cid:31)2Ay where e(cid:31) istheidempotent (11) e(cid:31) D 1 X(cid:31).a(cid:0)1/a: jAj a2A The vector subspaces Ce(cid:31) are not only orthogonal with respect to the Hermitian structure,butalsoorthogonalwithrespecttotheringstructureinthesensethat e(cid:31)e(cid:31)0D 0 if (cid:31)¤(cid:31)0. Each Ce(cid:31) isanidealofthering CŒA(cid:141). Fromthetraceidentity(seeegSerre[19,Section2.4])wehave,forevery a2A, X (cid:26)0 ifa¤e; (12) (cid:31).a/D jAj ifa¤e: (cid:31)2Ay Here e2A isthetrivialelement. 4.2 Algebraicsubgroupsof .C(cid:3)/n andlatticesin Zn 4.2.1 Algebraicsubgroupsof.C(cid:3)/n Analgebraicsubgroupof.C(cid:3)/n isasubgroup thatisclosedintheZariskitopology. Foralattice ƒ,ie,asubgroup ƒ of Zn,notnecessarilyofmaximalrank,let G.ƒ/ be thesetofall z2Cn suchthat zkD1 forevery k2ƒ. Herefor kD.k ;:::;k /2Zn 1 n and zD.z ;:::;z /2.C(cid:3)/n weset zkDQ zki. 1 n i i Itiseasytoseethat G.ƒ/ isanalgebraicsubgroup. Theconverseholdstrue: Every algebraic subgroup is equal to G.ƒ/ for some lattice ƒ; see Schmidt [18]. If ƒ is primitive,then G.ƒ/ isconnected,andinthiscaseitiscalledatorus. Algebraic & Geometric Topology,Volume13(2013) 2392 ThangTQLê 4.2.2 Automorphismsof .C(cid:3)/n Anexampleofatorusofdimension l isthestan- dard l–torus T D.C(cid:3)/l (cid:2)1n(cid:0)l (cid:26).C(cid:3)/n,whichis G.„n(cid:0)l/,where „n(cid:0)l Df.k1;:::;kn/2Znjk1D(cid:1)(cid:1)(cid:1)Dkl D0g: The following trick shows that each torus is isomorphic to the standard torus. For details,see[18]. For matrix K 2GL .Z/ with entries .K /n , one can define an automorphism n ij i;jD1 ' of .C(cid:3)/n by K (cid:18) n n n (cid:19) ' .z ;z ;:::;z /D YzK1j;YzK2j;:::;YzKnj : K 1 2 n j j j jD1 jD1 jD1 For any lattice ƒ(cid:26)Zn, ' .G.ƒ//DG.K.ƒ//. When ƒ is a primitive lattice of K rank n(cid:0)l, there is K 2GLn.Z/ such that K.ƒ/D„n(cid:0)l. Then 'K.G.ƒ// is the standard l–torus. 4.2.3 Algebraicsubgroupsandcharactergroups Fixgenerators t ;:::;t of Zn. 1 n Wewillwrite Zn multiplicativelyandusetheidentification ZŒZn(cid:141)DZŒt˙1;:::;t˙1(cid:141). 1 n Suppose (cid:128) (cid:26)Zn is a lattice. Every element z2G.(cid:128)/ defines a character (cid:31) of the z quotientgroup A(cid:128) WDZn=(cid:128) via (cid:31) .tk1(cid:1)(cid:1)(cid:1)tkn/Dzk; where kD.k ;:::;k /: z 1 n 1 n Conversely,everycharacterof A(cid:128) arisesinthisway. Thusonecanidentify G.(cid:128)/ with thePontryagindual Ay(cid:128) via z!(cid:31)z. Wewillwrite ez fortheidempotent e(cid:31)z,andthedecomposition(10),with (cid:128) having maximalrank,nowbecomes M (13) CŒA(cid:128)(cid:141)D Cez: z2G.(cid:128)/ 4.3 Torsionpointsin Q–algebraicsets 4.3.1 Torsionpoints Withrespecttotheusualmultiplication, C(cid:3)WDCnf0g isan abeliangroup,andsoisthedirectproduct .C(cid:3)/n. Thesubgroupoftorsionelements of C(cid:3), denoted by U,is thegroupof rootsofunity,and Un isthe torsionsubgroup of .C(cid:3)/n. If (cid:128) (cid:26)Zn isalatticeofmaximalrank,then G.(cid:128)/ isfinite,and G.(cid:128)/(cid:26)Un. Algebraic & Geometric Topology,Volume13(2013)

Description:
Hence, the left hand side of (1) makes sense only when . It is easy to see that V is defined over Q if and only if it is spanned by its Z–support. [3] B Clair, K Whyte, Growth of Betti numbers, Topology 42 (2003) 1125–1142.
See more

The list of books you might like

Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.