ebook img

Quantifying separability in virtually special groups PDF

0.81 MB·English
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 Quantifying separability in virtually special groups

QUANTIFYING SEPARABILITY IN VIRTUALLY SPECIAL GROUPS MARKF.HAGENANDPRIYAMPATEL Abstract. Wegiveanew, effectiveproofoftheseparabilityofcubicallyconvex-cocompactsubgroupsof specialgroups. Asaconsequence, weshowthatifGisavirtuallycompactspecialhyperbolicgroup, and Q ≤ GisaK-quasiconvexsubgroup,thenanyg ∈ G−Qofword-lengthatmostnisseparatedfromQ byasubgroupwhoseindexispolynomialinnandexponentialinK.ThisgeneralizesaresultofBou-Rabee 6 andtheauthorsonresidualfinitenessgrowth[9]andaresultofthesecondauthoronsurfacegroups[34]. 1 0 2 b e Introduction F Early motivation for studying residual finiteness and subgroup separability was a result of the rel- 5 evance of these properties to decision problems in group theory. An observation of Dyson [18] and ] Mostowski [31], related to earlier ideas of McKinsey [30], states that finitely presented residually finite R groupshavesolvablewordproblem. Thewordproblemisaspecialcaseofthemembershipproblem, i.e. G theproblemofdeterminingwhetheragiveng ∈ GbelongstoaparticularsubgroupH ofG. Separability h. canproduceasolutiontothemembershipprobleminessentiallythesamewaythatasolutiontotheword t problemisprovidedbyresidualfiniteness(see,e.g.,thediscussionin[2]). AsubgroupH ≤ Gisseparable a m inGif,forallg ∈ G−H,thereexistsG(cid:48) ≤ GwithH ≤ G(cid:48) andg (cid:54)∈ G(cid:48). Producinganupperbound,in f.i. [ termsofgandH,ontheminimalindexofsuchasubgroupG(cid:48)iswhatwemeanbyquantifyingseparability ofH inG. Quantifyingseparabilityisrelatedtothemembershipproblem;seeRemarkDbelow. 2 Recently,separabilityhasplayedacrucialroleinlow-dimensionaltopology,namelyintheresolutions v 1 oftheVirtuallyHakenandVirtuallyFiberedconjectures[1,40]. Itsinfluenceintopologyisaconsequence 0 oftheseminalpaperofScott[37],whichestablishesatopologicalreformulationofsubgroupseparability. 0 Roughly, Scott’s criterion allows one to use separability to promote (appropriately construed) immer- 7 0 sions to embeddings in finite covers. In [1], Agol proved the Virtually Special Conjecture of Wise, an . outstanding component of the proofs of the above conjectures. Agol’s theorem shows that every word 1 0 hyperbolic cubical group virtually embeds in a right-angled Artin group (hereafter, RAAG). Cubically 5 convex-cocompact subgroups of RAAGs are separable [25, 22] and Agol’s theorem demonstrates that 1 word hyperbolic cubical groups inherit this property via the virtual embeddings (separability properties : v are preserved under passing to subgroups and finite index supergroups). In fact, since quasiconvex sub- i X groups of hyperbolic cubical groups are cubically convex-cocompact [22, 36], all quasiconvex subgroups r ofsuchgroupsareseparable. Inthispaper, wegiveanew, effectiveproofoftheseparabilityofcubically a convex-cocompactsubgroupsofspecialgroups. Ourmaintechnicalresultis: TheoremA. LetΓbeasimplicialgraphandletZ beacompactconnectedcubecomplex,basedata0–cube x,withabasedlocalisometryZ → S . Forallg ∈ A −π Z,thereisacubecomplex(Y,x)suchthat: Γ Γ 1 (1) Z ⊂ Y; (2) thereisabasedlocalisometryY → S suchthatZ → S factorsasZ (cid:44)→ Y → S ; Γ Γ Γ (3) anyclosedbasedpathrepresentingg liftstoanon-closedpathatxinY; (4) |Y(0)| ≤ |Z(0)|(|g|+1), where|g|isthewordlengthofg withrespecttothestandardgenerators. Date:February8,2016. 2010MathematicsSubjectClassification. 20E26,20F36. 1 Via Haglund-Wise’s canonical completion [23], Theorem A provides the following bounds on the sepa- rabilitygrowthfunction(definedinSection1)oftheclassofcubicallyconvex-cocompactsubgroupsofa (virtually) special group. Roughly, separability growth quantifies separability of all subgroups in a given class. CorollaryB. LetG ∼= π X,withX acompactspecialcubecomplex,andletQ betheclassofsubgroups 1 R representedbycompactlocalisometriestoX whosedomainshaveatmostRvertices. Then SepQR(Q,n) ≤ PRn G,S forallQ ∈ Q andn ∈ N,wheretheconstantP dependsonlyonthegeneratingsetS. Hence,lettingQ(cid:48) be R K theclassofsubgroupsQ ≤ GsuchthattheconvexhullofQx˜liesinNK(Qx˜)andx˜ ∈ X(cid:101)(0), SepQ(cid:48)K(Q,n) ≤ P(cid:48)gr (K)n, G,S X(cid:101) whereP(cid:48) dependsonlyonG,X(cid:101),S,andgr isthegrowthfunctionofX(cid:101)(0). X(cid:101) Inthehyperboliccase,wherecubicallyconvex-cocompactnessisequivalenttoquasiconvexity,weobtain aboundthatispolynomialinthelengthofthewordandexponentialinthequasiconvexityconstant: Corollary C. Let G be a group with an index-J special subgroup. Fixing a word-length (cid:107) − (cid:107) on G, S suppose that (G,(cid:107) − (cid:107) ) is δ-hyperbolic. For each K ≥ 1, let Q be the set of subgroups Q ≤ G such S K thatQisK-quasiconvexwithrespectto(cid:107)−(cid:107) . ThenthereexistsaconstantP = P(G,S)suchthatforall S K ≥ 0,Q ∈ Q ,andn ≥ 0, K SepQK(Q,n) ≤ Pgr (PK)J!nJ!, G,S G wheregr isthegrowthfunctionofG. G CorollaryCsaysthatifGisahyperboliccubicalgroup,Q ≤ GisK-quasiconvex,andg ∈ G−Q,then g isseparatedfromQbyasubgroupofindexboundedbyafunctionpolynomialin(cid:107)g(cid:107) andexponential S inK. The above results fit into a larger body of work dedicated to quantifying residual finiteness and sub- groupseparabilityofvariousclassesofgroups(see,e.g.,[10,11,27,14,34,33,35,8,12,29]). WhenGisthe fundamentalgroupofahyperbolicsurface,compareCorollaryCto[34,Theorem7.1]. Combiningvarious cubulationresultswith[1],thegroupscoveredbyCorollaryCincludefundamentalgroupsofhyperbolic 3–manifolds [6, 26], hyperbolic Coxeter groups [24], simple-type arithmetic hyperbolic lattices [5], hy- perbolicfree-by-cyclicgroups[21],hyperbolicascendingHNNextensionsoffreegroupswithirreducible monodromy [20], hyperbolic groups with a quasiconvex hierarchy [40], C(cid:48)(1) small cancellation groups 6 [38],andhencerandomgroupsatlowenoughdensity[32],amongmanyothers. In[9]Bou-Rabeeandtheauthorsquantifiedresidualfinitenessforvirtuallyspecialgroups,byworking in RAAGs and appealing to the fact that upper bounds on residual finiteness growth are inherited by finitely-generated subgroups and finite-index supergroups. Theorem A generalizes a main theorem of [9], andaccordinglytheproofisreminiscentoftheonein[9]. However, residualfinitenessisequivalent to separability of the trivial subgroup, and thus it is not surprising that quantifying separability for an arbitrary convex-cocompact subgroup of a RAAG entails engagement with a more complex geometric situation. Ourtechniquesthussignificantlygeneralizethoseof[9]. RemarkD(Membershipproblem). IfHisafinitely-generatedseparablesubgroupofthefinitely-presented group G, and one has an upper bound on Sep{H}(|g|), for some finite generating set S of G, then the G,S following procedure decides if g ∈ H: first, enumerate all subgroups of G of index at most Sep{H}(|g|) G,S usingafinitepresentationofG. Second,foreachsuchsubgroup,testwhetheritcontainsg;ifso,ignoreit, andifnot,proceedtothethirdstep. Third,foreachfinite-indexsubgroupnotcontainingg,testwhether 2 itcontainseachofthefinitelymanygeneratorsofH;ifso,wehaveproducedafinite-indexsubgroupcon- taining H but not g, whence g (cid:54)∈ H. If we exhaust the subgroups of index at most Sep{H}(|g|) without G,S finding such a subgroup, then g ∈ H. In particular, Corollary C gives an effective solution to the mem- bershipproblemforquasiconvexsubgroupsofhyperboliccubicalgroups,thoughitdoesnotappeartobe anymoreefficientthanthemoregeneralsolutiontothemembershipproblemforquasiconvexsubgroups of(arbitrary)hyperbolicgroupsrecentlygivenbyKharlampovich–Miasnikov–Weilin[28]. The paper is organized as follows: In Section 1, we define the separability growth of a group with respecttoaclassQofsubgroups,whichgeneralizestheresidualfinitenessgrowthintroducedin[7]. We alsoprovidesomenecessarybackgroundonRAAGsandcubicalgeometry. InSection2,wediscusscorol- lariestothemaintechnicalresult,includingCorollaryC,beforeconcludingwithaproofofTheoremAin Section3. Acknowledgments. We thank K. Bou-Rabee, S. Dowdall, F. Haglund, D.B. McReynolds, N. Miller, H. Wilton,andD.T.Wiseforhelpfuldiscussionsaboutissuesrelatedtothispaper. Wealsothankananony- mous referee for astute corrections. The authors acknowledge travel support from U.S. National Science FoundationgrantsDMS1107452,1107263,1107367"RNMS:GeometricStructuresandRepresentationVa- rieties"(theGEARNetwork)andfromgrantNSF1045119. M.F.H.wassupportedbytheNationalScience FoundationunderGrantNumberNSF1045119. 1. Background 1.1. Separability growth. Let G be a group generated by a finite set S and let H ≤ G be a subgroup. LetΩ = {∆ ≤ G : H ≤ ∆},anddefineamapDΩH : G−H → N∪{∞}by H G DΩH(g) = min{[G : ∆] : ∆ ∈ Ω ,g (cid:54)∈ ∆}. G H Thisisaspecialcaseofthenotionofadivisibilityfunctiondefinedin[7]anddiscussedin[11]. Notethat H isaseparablesubgroupofGifandonlyifDΩH takesonlyfinitevalues. G The separability growth of G with respect to a class Q of subgroups is a function SepQ : Q×N → G,S N∪{∞}givenby (cid:110) (cid:111) SepQ (Q,n) = max DΩQ(g) : g ∈ G−Q,(cid:107)g(cid:107) ≤ n . G,S G S If Q is a class of separable subgroups of G, then the separability growth measures the index of the sub- group to which one must pass in order to separate Q from an element of G − Q of length at most n. For example, when G is residually finite and Q = {{1}}, then SepQ is the residual finiteness growth G,S function. The following fact is explained in greater generality in [9, Section 2]. (In the notation of [9], SepQ (Q,n) = RFΩQ (n)forallQ ∈ Qandn ∈ N.) G,S G,S Proposition 1.1. Let G be a finitely generated group and let Q be a class of subgroups of G. If S,S(cid:48) are finitegeneratingsetsofG,thenthereexistsaconstantC > 0with SepQ (Q,n) ≤ C ·SepQ (Q,Cn) G,S(cid:48) G,S forQ ∈ Q,n ∈ N. Hencetheasymptoticgrowthrateof SepQ isindependentofS. G,S (Similar statements assert that upper bounds on separability growth are inherited by finite-index super- groupsandarbitraryfinitely-generatedsubgroupsbutwedonotuse,andthusomit,these.) 3 1.2. Nonpositively-curved cube complexes. We assume familiarity with nonpositively-curved and CAT(0) cube complexes and refer the reader to e.g. [19, 22, 39, 40] for background. We now make ex- plicit some additional notions and terminology, related to convex subcomplexes, which are discussed in greater depth in [4]. We also discuss some basic facts about RAAGs and Salvetti complexes. Finally, we willusethemethodofcanonicalcompletion,introducedin[23],andreferthereaderto[9,Lemma2.8]for theexactstatementneededhere. 1.2.1. Local isometries, convexity, and gates. A local isometry φ : Y → X of cube complexes is a locally injectivecombinatorialmapwiththepropertythat,ife ,...,e are1–cubesofY allincidenttoa0–cube 1 n y,andthe(necessarilydistinct)1–cubesφ(e ),...,φ(e )alllieinacommonn–cubec(containingφ(y)), 1 n then e ,...,e span an n–cube c(cid:48) in Y with φ(c(cid:48)) = c. If φ : Y → X is a local isometry and X is 1 n nonpositively-curved, then Y is as well. Moreover, φ lifts to an embedding φ˜ : Y(cid:101) → X(cid:101) of universal covers,andφ˜(Y(cid:101))isconvexinX(cid:101) inthefollowingsense. LetX(cid:101) beaCAT(0)cubecomplex. ThesubcomplexK ⊆ X(cid:101) isfullifKcontainseachn–cubeofX(cid:101) whose 1-skeletonappearsinK. IfK isfull,thenK isisometricallyembeddedifK∩(cid:84) H isconnectedwhenever i i {Hi} is a set of pairwise-intersecting hyperplanes of X(cid:101). Equivalently, the inclusion K(1) (cid:44)→ X(cid:101)(1) is an isometric embedding with respect to the graph-metric. If the inclusion K (cid:44)→ X(cid:101) of the full subcomplex K is a local isometry, then K is convex. Note that a convex subcomplex is necessarily isometrically embedded,andinfactK isconvexifandonlyifK(1) ismetricallyconvexinX(cid:101)(1). Aconvexsubcomplex K is a CAT(0) cube complex in its own right, and its hyperplanes have the form H ∩K, where K is a hyperplaneofX(cid:101). Moreover,ifK isconvex,thenhyperplanesH1∩K,H2∩K ofK intersectifandonly if H ∩ H (cid:54)= ∅. We often say that the hyperplane H crosses the convex subcomplex K to mean that 1 2 H ∩K (cid:54)= ∅andwesaythehyperplanesH,H(cid:48) crossiftheyintersect. Hyperplanesareanimportantsourceofconvexsubcomplexes,intworelatedways. First,recallthatfor all hyperplanes H of X(cid:101), the carrier N(H) is a convex subcomplex. Second, N(H) ∼= H ×[−1, 1], and 2 2 the subcomplexes H ×{±1} of X(cid:101) “bounding” N(H) are convex subcomplexes isomorphic to H (when 2 H isgiventhecubicalstructureinwhichitsn–cubesaremidcubesof(n+1)–cubesofX(cid:101)). Asubcomplex of the form H × {±1} is a combinatorial hyperplane. The convex hull of a subcomplex S ⊂ X(cid:101) is the 2 intersectionofallconvexsubcomplexesthatcontainS;see[22]. LetK ⊆ X(cid:101) beaconvexsubcomplex. ThenthereisamapgK : X(cid:101)(0) → K suchthatforallx ∈ X(cid:101)(0), thepointg (x)istheuniqueclosestpointofK tox. (ThispointisoftencalledthegateofxinK;gates K arediscussedfurtherin[17]and[3].) ThismapextendstoacubicalmapgK : X(cid:101) → K,thegatemap. See e.g.[4]foradetaileddiscussionofthegatemapinthelanguageusedhere;weuseonlythatitextendsthe map on0–cubes andhas the property thatfor allx,y, ifg (x),g (y)are separatedby a hyperplaneH, K K thenthesameH separatesxfromy. Finally, thehyperplaneH separatesxfromg (x)ifandonlyifH K separates x from K. The gate map allows us to define the projection of the convex subcomplex K(cid:48) of X(cid:101) ontoK tobeg (K),whichistheconvexhulloftheset{g (x) ∈ K : x ∈ K(cid:48)(0)}. Convexsubcomplexes K(cid:48) K K,K(cid:48) are parallel if g (K) = K(cid:48) and g (K(cid:48)) = K. Equivalently, K,K(cid:48) are parallel if and only if, for K(cid:48) K eachhyperplaneH,wehaveH ∩K (cid:54)= ∅ifandonlyifH ∩K(cid:48) (cid:54)= ∅. Notethatparallelsubcomplexesare isomorphic. Remark1.2. Weoftenusethefollowingfacts. LetK,K(cid:48)beconvexsubcomplexesofX(cid:101). Thentheconvex hullC ofK ∪K(cid:48) containstheunionofK,K(cid:48) andaconvexsubcomplexoftheformG (K(cid:48))×γˆ,where K G (K(cid:48)) is the image of the gate map discussed above and γˆ is the convex hull of a geodesic segment γ K joining a closest pair of 0–cubes in K,K(cid:48), by [4, Lemma 2.4]. A hyperplane H crosses K and K(cid:48) if and only if H crosses G (K(cid:48)); the hyperplane H separates K,K(cid:48) if and only if H crosses γˆ. All remaining K hyperplanes either cross exactly one of K,K(cid:48) or fail to cross C. Observe that the set of hyperplanes 4 separatingK,K(cid:48) containsnotripleH,H(cid:48),H(cid:48)(cid:48) ofdisjointhyperplanes,noneofwhichseparatestheother two. (Suchaconfigurationiscalledafacingtriple.) 1.2.2. Salvetti complexes and special cube complexes. Let Γ be a simplicial graph and let A be the corre- Γ spondingright-angledArtingroup(RAAG),i.e. thegrouppresentedby (cid:10) (cid:11) V(Γ) | [v,w], {v,w} ∈ E(Γ) , whereV(Γ)andE(Γ)respectivelydenotethevertex-andedge-setsofΓ. ThephrasegeneratorofΓrefers tothispresentation;wedenoteeachgeneratorofA bythecorrespondingvertexofΓ. Γ The RAAG A is isomorphic to the fundamental group of the Salvetti complex S , introduced in [16], Γ Γ whichisanonpositively-curvedcubecomplexwithone0–cubex,anoriented1–cubeforeachv ∈ V(Γ), labeledbyv,andann–torus(ann–cubewithoppositefacesidentified)foreveryn–cliqueinΓ. AcubecomplexX isspecialifthereexistsasimplicialgraphΓandalocalisometryX → S inducing Γ a monomorphism π1X → AΓ and a π1X-equivariant embedding X(cid:101) → S(cid:101)Γ of universal covers whose image is a convex subcomplex. Specialness allows one to study geometric features of π X by working 1 inside of S(cid:101)Γ, which has useful structure not necessarily present in general CAT(0) cube complexes; see Section 1.2.3. Following Haglund-Wise [23], a group G is (virtually) [compact] special if G is (virtually) isomorphictothefundamentalgroupofa[compact]specialcubecomplex. 1.2.3. Cubical features particular to Salvetti complexes. Let Γ be a finite simplicial graph and let Λ be an induced subgraph of Γ. The inclusion Λ (cid:44)→ Γ induces a monomorphism A → A . In fact, there is an Λ Γ injective local isometry S → S inducing A → A . Hence each conjugate Ag of A in A is the Λ Γ Λ Γ Λ Λ Γ stabilizerofaconvexsubcomplexgS(cid:101)Λ ⊆ S(cid:101)Γ. Afewspecialcaseswarrantextraconsideration. When Λ ⊂ Γ is an n-clique, for some n ≥ 1, then S ⊆ S is an n–torus, which is the Salvetti Λ Γ complexofthesub-RAAGisomorphictoZn generatedbynpairwise-commutinggenerators. Inthiscase, SΛ is a standard n-torus in SΓ. (When n = 1, SΛ is a standard circle.) Each lift of S(cid:101)Λ to S(cid:101)Γ is a standard flat; when n = 1, we use the term standard line; a compact connected subcomplex of a standard line is a standard segment. The labels and orientations of 1–cubes in SΓ pull back to S(cid:101)Γ; a standard line is a convexsubcomplexisometrictoR, allofwhose1–cubeshavethesamelabel, suchthateach0–cubehas oneincomingandoneoutgoing1–cube. WhenLk(v)isthelinkofavertexv ofΓ,thesubcomplexS isanimmersedcombinatorialhyper- Lk(v) planeinthesensethatS(cid:101)Lk(v) isacombinatorialhyperplaneofS(cid:101)Γ. Thereisacorrespondinghyperplane, whose carrier is bounded by S(cid:101)Lk(v) and vS(cid:101)Lk(v), that intersects only 1–cubes labeled by v. Moreover, S(cid:101)Lkv is contained in S(cid:101)St(v), where St(v) is the star of v, i.e. the join of v and Lk(v). It follows that S(cid:101)St(v) ∼= S(cid:101)Lk(v) ×S(cid:101)v, whereS(cid:101)v isastandardline. NotethatthecombinatorialhyperplaneS(cid:101)Lk(v) ispar- alleltovkS(cid:101)Lk(v) forallk ∈ Z. Likewise,S(cid:101)v isparalleltogS(cid:101)v exactlywheng ∈ AΛ,andparallelstandard lines have the same labels. We say S(cid:101)v is a standard line dual to S(cid:101)Lk(v), and is a standard line dual to any hyperplaneH suchthatN(H)hasS(cid:101)Lk(v) asoneofitsboundingcombinatorialhyperplanes. Remark 1.3. We warn the reader that a given combinatorial hyperplane may correspond to distinct hyperplaneswhosedualstandardlineshavedifferentlabels;thisoccursexactlywhenthereexistmultiple vertices in Γ whose links are the same subgraph. However, the standard line dual to a genuine (non- combinatorial)hyperplaneisuniquely-determineduptoparallelism. Definition 1.4 (Frame). Let K ⊆ S(cid:101)Γ be a convex subcomplex and let H be a hyperplane. Let L be a standardlinedualtoH. TheframeofH istheconvexsubcomplexH(cid:48)×L ⊆ S(cid:101)Γ describedabove,where H(cid:48) isacombinatorialhyperplaneboundingN(H). IfK ⊆ S(cid:101)Γ isaconvexsubcomplex,andH intersects K,thentheframeofH inK isthecomplexK∩(H×L). Itisshownin[9]thattheframeofH inK has 5 theform(H ∩K)×(L∩K),providedthatLischoseninitsparallelismclasstointersectK. Notethat theframeofH isinfactwell-defined,sinceallpossiblechoicesofLareparallel. 2. ConsequencesofTheoremA Assuming Theorem A, we quantify separability of cubically convex-cocompact subgroups of special groupswiththeproofsofCorollariesBandC,beforeprovingTheoremAinthenextsection. ProofofCorollaryB. Let Γ be a finite simplicial graph so that there is a local isometry X → S . Let Γ Q ∈ Q berepresentedbyalocalisometryZ → X. Thenforallg ∈ π X −π Z, byTheoremA,there R 1 1 is a local isometry Y → S such that Y contains Z as a locally convex subcomplex, and g (cid:54)∈ π Y, and Γ 1 |Y(0)| ≤ |Z(0)|(|g|+1). Applyingcanonicalcompletion[23]toY → SΓyieldsacoverS(cid:98)Γ → SΓinwhich Y embeds;thiscoverhasdegree|Y(0)|by[9,Lemma2.8]. LetH(cid:48) = π1S(cid:98)Γ∩π1X,sothatπ1Z ≤ H(cid:48),and g (cid:54)∈ H(cid:48),and[π X : H(cid:48)] ≤ |Z(0)|(|g|+1). Thefirstclaimfollows. 1 LetG ∼= π1X withX compactspecial,Q ≤ G,andtheconvexhullofQx˜inX(cid:101) liesinNK(QX˜). Then the second claim follows since we can choose Z to be the quotient of the hull of Qx˜ by the action of Q, and|Z(0)| ≤ gr (K). (cid:3) X(cid:101) Ingeneral,thenumberof0–cubesinZ iscomputablefromthequasiconvexityconstantofaQ-orbitin X(cid:101)(1) by[22,Theorem2.28]). Inthehyperboliccase,weobtainCorollaryCintermsofthequasiconvexity constant,withoutreferencetoanyparticularcubecomplex: ProofofCorollaryC. We use Corollary B when J = 1, and promote the result to a polynomial bound whenJ ≥ 1. LetQ ∈ Q andletg ∈ G−Q. K The special case: Suppose J = 1 and let X be a compact special cube complex with G ∼= π X. Let 1 Z → X be a compact local isometry representing the inclusion Q → G. Such a complex exists by quasiconvexity of Q and [22, Theorem 2.28], although we shall use the slightly more computationally explicit proof in [36]. Let A(cid:48) ≥ 1,B(cid:48) ≥ 0 be constants such that an orbit map (G,(cid:107)−(cid:107)S) → (X(cid:101)(1),d) is an (A(cid:48),B(cid:48))-quasi-isometric embedding, where d is the graph-metric. Then there exist constants A,B, dependingonlyonA(cid:48),B(cid:48)andhenceon(cid:107)−(cid:107) ,suchthatQxis(AK+B)-quasiconvex,wherexisa0–cube S inZ(cid:101) ⊂ X(cid:101). BytheproofofProposition3.3of[36],theconvexhullZ(cid:101)ofQxliesintheρ-neighborhoodof √ (cid:16) (cid:16) (cid:17) (cid:17) Qx,whereρ = AK+B+ dimX+δ(cid:48) csc 1 sin−1(√ 1 +1 andδ(cid:48) = δ(cid:48)(δ,A(cid:48),B(cid:48)). CorollaryB 2 dimX provides G(cid:48) ≤ G with g (cid:54)∈ G(cid:48), and [G : G(cid:48)] ≤ |Z(0)|(|g|+1). But |g|+1 ≤ A(cid:48)(cid:107)g(cid:107) +B(cid:48) +1, while S |Z(0)| ≤ gr (ρ). Thus[G : G(cid:48)] ≤ gr (ρ)A(cid:48)(cid:107)g(cid:107) +gr (ρ)B(cid:48)+gr (ρ),sothatthereexistsP suchthat X(cid:101) X(cid:101) S X(cid:101) X(cid:101) 1 SepQK(Q,n) ≤ P gr (P K)n G,S 1 X(cid:101) 1 forallK,Q ∈ Q ,n ∈ N,whereP dependsonlyonX. K 1 The virtually special case: Now suppose that J ≥ 1. We have a compact special cube complex X, and [G : G(cid:48)] ≤ J!, where G(cid:48) ∼= π X and G(cid:48) (cid:47)G. Let Q ≤ G be a K-quasiconvex subgroup. By Lemma 2.1, 1 thereexistsC = C(G,S)suchthatQ∩G(cid:48) isCJ!(K+1)-quasiconvexinG,andthusisP CJ!(K+1)- 2 quasiconvexinG(cid:48),whereP dependsonlyonGandS. 2 Letg ∈ G−Q. SinceG(cid:48)(cid:47)G,theproductQG(cid:48) isasubgroupofGofindexatmostJ!thatcontainsQ. Hence, ifg (cid:54)∈ QG(cid:48), thenwearedone. Wethusassumeg ∈ QG(cid:48). Hencewecanchoosealefttransversal {q ,...,q } for Q∩G(cid:48) in Q, with s ≤ J! and q = 1. Write g = q g(cid:48) for some i ≤ s, with g(cid:48) ∈ G(cid:48). 1 s 1 i Suppose that we have chosen each q to minimize (cid:107)q (cid:107) among all elements of q (Q ∩ G(cid:48)), so that, by i i S i Lemma2.3,(cid:107)q (cid:107) ≤ J!foralli. Hence(cid:107)g(cid:48)(cid:107) ≤ ((cid:107)g(cid:107) +J!). i S S By the first part of the proof, there exists a constant P , depending only on G,G(cid:48),S, and a subgroup 1 G(cid:48)(cid:48) ≤ G(cid:48) suchthatQ∩G(cid:48) ≤ G(cid:48)(cid:48),andg(cid:48) (cid:54)∈ G(cid:48)(cid:48),and [G(cid:48) : G(cid:48)(cid:48)] ≤ P gr (P P CJ!(K +1))(cid:107)g(cid:48)(cid:107) ≤ P gr (P P CJ!(K +1))(cid:107)g(cid:48)(cid:107) . 1 G(cid:48) 1 2 S 1 G 1 2 S 6 LetG(cid:48)(cid:48)(cid:48) = ∩s q G(cid:48)(cid:48)q−1,sothatg(cid:48) (cid:54)∈ G(cid:48)(cid:48)(cid:48),andQ∩G(cid:48) ≤ G(cid:48)(cid:48)(cid:48) (sinceG(cid:48) isnormal),and i=1 i i [G(cid:48) : G(cid:48)(cid:48)(cid:48)] ≤ (P gr (P P CJ!(K +1))(cid:107)g(cid:48)(cid:107) )s. 1 G 1 2 S Finally, letH = QG(cid:48)(cid:48)(cid:48). This subgroup clearly contains Q. Suppose that g = q g(cid:48) ∈ H. Then g(cid:48) ∈ QG(cid:48)(cid:48)(cid:48), i i.e. g(cid:48) = ag(cid:48)(cid:48)(cid:48) forsomea ∈ Qandg(cid:48)(cid:48)(cid:48) ∈ G(cid:48)(cid:48)(cid:48). Sinceg(cid:48) ∈ G(cid:48) andG(cid:48)(cid:48)(cid:48) ≤ G(cid:48), wehavea ∈ Q∩G(cid:48), whence a ∈ G(cid:48)(cid:48)(cid:48),byconstruction. Thisimpliesthatg(cid:48) ∈ G(cid:48)(cid:48)(cid:48) ≤ G(cid:48)(cid:48),acontradiction. HenceH isasubgroupofG separatingg fromQ. Finally, [G : H] ≤ [G : G(cid:48)(cid:48)(cid:48)] ≤ J![P gr (P P CJ!(K +1))((cid:107)g(cid:107) +J!)]J!, 1 G 1 2 S andtheproofiscomplete. (cid:3) Lemma2.1. LetthegroupGbegeneratedbyafinitesetS andlet(G,(cid:107)−(cid:107) )beδ-hyperbolic. LetQ ≤ G S be K-quasiconvex, and let G(cid:48) ≤ G be an index-I subgroup. Then Q∩G(cid:48) is CI(K +1)-quasiconvex in G forsomeC dependingonlyonδ andS. Proof. Since Q is K-quasiconvex in G, it is generated by a set T of q ∈ Q with (cid:107)q(cid:107) ≤ 2K +1, by [13, S LemmaIII.Γ.3.5]. Astandardargumentshows(Q,(cid:107)−(cid:107) ) (cid:44)→ (G,(cid:107)−(cid:107) )isa(2K+1,0)-quasi-isometric T S embedding. Lemma2.3showsthatQ∩G(cid:48) isI-quasiconvexin(Q,(cid:107)−(cid:107) ),since[Q : Q∩G(cid:48)] ≤ I. Hence T Q∩G(cid:48)hasageneratingsetmakingit((2I+1)(2K+1),0)-quasi-isometricallyembeddedin(G,(cid:107)−(cid:107) ). S ApplyLemma2.2toconclude. (cid:3) Thefollowinglemmaisstandard,butweincludeittohighlighttheexactconstantsinvolved: Lemma2.2. LetGbeagroupgeneratedbyafinitesetS andsupposethat(G,(cid:107)−(cid:107) )isδ-hyperbolic. Then S there exists a (sub)linear function f : N → N, depending on S and δ, such that σ ⊆ N (γ) whenever f(λ) γ : [0,L] → Gisa(λ,0)–quasigeodesicandσ isageodesicjoiningγ(0)toγ(L). Proof. Seee.g. theproofof[13,TheoremIII.H.1.7]. (cid:3) Lemma 2.3. Let Q be a group generated by a finite set S and let Q(cid:48) ≤ Q be a subgroup with [Q : Q(cid:48)] = s < ∞. Thenthereexistsalefttransversal{q ,...,q }forQ(cid:48) suchthat(cid:107)q (cid:107) ≤ sfor1 ≤ i ≤ s. HenceQ(cid:48) 1 s i S iss-quasiconvexinQ. Proof. Supposethatq = s ···s isageodesicwordinS ∪S−1 andthatq isashortestrepresentative k ik i1 k of q Q(cid:48). Let q = s ···s be the word in Q consisting of the last j letters of q for all 1 < j < k, k j ij i1 k and let q = 1. We claim that each q is a shortest representative for q Q(cid:48). Otherwise, there exists p 1 j j with (cid:107)p(cid:107) < j such that q Q(cid:48) = pQ(cid:48). But then s ···s pQ(cid:48) = q Q(cid:48), and thus q was not a shortest S j k j+1 k k representative. It also follows immediately that q Q(cid:48) (cid:54)= q Q(cid:48) for j (cid:54)= j(cid:48). Thus, q ,q ,...,q represent j j(cid:48) 1 2 k distinctleftcosetsofQ(cid:48) providedk ≤ s,andtheclaimfollows. (cid:3) Remark2.4(Embeddingsinfinitecovers). GivenacompactspecialcubecomplexX andacompactlocal isometry Z → X, Theorem A gives an upper bound on the minimal degree of a finite cover in which Z embeds; indeed, producing such an embedding entails separating π Z from finitely many elements in 1 π X. However,itisobservedin[9,Lemma2.8]theHaglund–Wisecanonicalcompletionconstruction[23] 1 producesacoverX(cid:98) → X ofdegree|Z(0)|inwhichZ embeds. 3. ProofofTheoremA Inthissection,wegiveaproofofthemaintechnicalresult. Definition 3.1. Let SΓ be a Salvetti complex and let S(cid:101)Γ be its universal cover. The hyperplanes H,H(cid:48) of S(cid:101)Γ are collateral if they have a common dual standard line (equivalently, the same frame). Clearly collateralism is an equivalence relation, and collateral hyperplanes are isomorphic and have the same stabilizer. 7 Being collateral implies that the combinatorial hyperplanes bounding the carrier of H are parallel to those bounding the carrier of H(cid:48). However, the converse is not true when Γ contains multiple vertices whose links coincide. In the proof of Theorem A, we will always work with hyperplanes, rather than combinatorialhyperplanes,unlessweexplicitlystatethatwearereferringtocombinatorialhyperplanes. ProofofTheoremA. Let x˜ ∈ S(cid:101)Γ be a lift of the base 0–cube x in SΓ, and let Z(cid:101) ⊆ S(cid:101)Γ be the lift of the universal cover of Z containing x˜. Since Z → SΓ is a local isometry, Z(cid:101) is convex. Let Z(cid:98) ⊂ Z(cid:101) be the convex hull of a compact connected fundamental domain for the action of π1Z ≤ AΓ on Z(cid:101). Denote by K theconvexhullofZ(cid:98)∪{gx˜}andletSbethesetofhyperplanesofS(cid:101)Γ intersectingK. Wewillforma quotientofK,restrictingtoZ(cid:98) → Z onZ(cid:98),whoseimageadmitsalocalisometrytoSΓ. Thesubcomplex(cid:98)Z(cid:98)(cid:99): LetLbethecollectionofstandardsegments(cid:96)inK thatmaptostandardcircles inSΓwiththepropertythat(cid:96)∩Z(cid:98)hasnon-contractibleimageinZ. Let(cid:98)Z(cid:98)(cid:99)beconvexhullofZ(cid:98)∪(cid:83)(cid:96)∈L(cid:96), sothatZ(cid:98) ⊆ (cid:98)Z(cid:98)(cid:99) ⊆ K. PartitioningS: WenowpartitionSaccordingtothevarioustypesofframesinK. First,letZbetheset ofhyperplanesintersectingZ(cid:98). Second,letNbethesetofN ∈ S−Zsuchthattheframe(N∩K)×(L∩K) of N in K has the property that for some choice of x0 ∈ N(0), the segment ({x0}×L)∩Z(cid:98) maps to a nontrivialcycleof1–cubesinZ. LetnN ≥ 1bethelengthofthatcycle. ByconvexityofZ(cid:98), thenumber n is independent of the choice of the segment L within its parallelism class. Note that N is the set of N hyperplanes that cross (cid:98)Z(cid:98)(cid:99), but do not cross Z(cid:98). Hence each N ∈ N is collateral to some W ∈ Z. Third, fixacollection{H ,...H } ⊂ S−Zsuchthat: 1 k (1) For1 ≤ i ≤ k−1,thehyperplaneHi separatesHi+1 from(cid:98)Z(cid:98)(cid:99). (2) For1 ≤ i < j ≤ k, ifahyperplaneH separatesH fromH , thenH iscollateraltoH forsome i j (cid:96) (cid:96) ∈ [i,j]. Similarly,ifH separatesH1 from(cid:98)Z(cid:98)(cid:99),thenH iscollateraltoH1,andifH separatesHk fromgx˜,thenH iscollateraltoH . k (3) Foreachi,theframe(H ∩K)×L ofH inK hasthepropertythatforeveryh ∈ H(0),theimage i i i i inZ ofthesegment({h}×Li)∩Z(cid:98)isemptyorcontractible. (Here,Li isastandardsegmentofa standardlinedualtoH .) i Let H be the set of all hyperplanes of S−Z that are collateral to H for some i. Condition (3) above i ensuresthatH∩N = ∅,whileH = ∅onlyifK = (cid:98)Z(cid:98)(cid:99). Finally,letB = S−(Z∪N∪H). Notethateach B ∈ BcrossessomeH . Figure1showsapossibleK andvariousfamiliesofhyperplanescrossingit. i N2 B2 H2 H3 N1 B1 H1 Figure1. HyperplanescrossingK (thedarkshadedareaontheleftisZ(cid:98)). 8 Mapping (cid:98)Z(cid:98)(cid:99) to Z: We now define a quotient map q : (cid:98)Z(cid:98)(cid:99) → Z extending the restriction Z(cid:98) → Z of Z(cid:101) → Z. NotethatifN = ∅,then(cid:98)Z(cid:98)(cid:99) = Z(cid:98),andqisjustthemapZ(cid:98) → Z. HencesupposeN (cid:54)= ∅andlet N ,...,N bethecollateralismclassesofhyperplanesinN,andfor1 ≤ i ≤ s,letN(cid:48) bethecollateralism 1 s i class of N in S, i.e. N together with a nonempty set of collateral hyperplanes in Z. For each i, let L i i i be a maximal standard line segment of (cid:98)Z(cid:98)(cid:99), each of whose 1–cubes is dual to a hyperplane in N(cid:48) and i which crosses each element of N(cid:48)i. For each i, let Ni ∈ Ni be a hyperplane separating Z(cid:98) from gx˜. Then Ni∩Nj (cid:54)= ∅fori (cid:54)= j,sinceneitherseparatestheotherfromZ(cid:98). WecanchoosetheLi sothatthereisan isometricembedding(cid:81)ki=1Li → (cid:98)Z(cid:98)(cid:99),sincewhetherornottwohyperplanesofS(cid:101)Γcrossdependsonlyon theircollateralismclasses. For each nonempty I ⊆ {1,...,k}, a hyperplane W ∈ Z crosses some U ∈ ∪ N(cid:48) if and only i∈I i if W crosses each hyperplane collateral to U. Hence, by [19, Lemma 7.11], there is a maximal convex subcomplex Y(I) ⊂ Z(cid:98), defined up to parallelism, such that a hyperplane W crosses each U ∈ ∪i∈IN(cid:48)i if and only if W ∩ Y(I) (cid:54)= ∅. Let A(I) be the set of hyperplanes crossing Y(I). By the definition of Y(I)and[19,Lemma7.11],thereisacombinatorialisometricembeddingY(I)×(cid:81)i∈ILi → (cid:98)Z(cid:98)(cid:99),whose image we denote by F(I) and refer to as a generalized frame. Moreover, for any 0–cube z ∈ (cid:98)Z(cid:98)(cid:99) that is notseparatedfromahyperplanein∪ N(cid:48)∪A(I)byahyperplanenotinthatset,wecanchooseF(I)to i∈I i containz (thisfollowsfromtheproofof[19, Lemma7.11]). Figures2, 3, 4, and5showpossibleN(cid:48)’sand i generalizedhyperplaneframes. N(cid:48) 1 N(cid:48) 2 Figure 2. Collateral families N(cid:48) andN(cid:48). Figure3. Y({1})×L . 1 2 1 Figure4. Y({2})×L Figure5. Y({1,2})×(L ×L ) 2 1 2 Tobuildq,wewillexpress(cid:98)Z(cid:98)(cid:99)astheunionofZ(cid:98)andacollectionofgeneralizedframes,defineqoneach generalizedframe,andcheckthatthedefinitioniscompatiblewheremultiplegeneralizedframesintersect. Let z ∈ (cid:98)Z(cid:98)(cid:99) be a 0–cube. Either z ∈ Z(cid:98), or there is a nonempty set I ⊂ {1,...,k} such that the set of hyperplanesseparatingz fromZ(cid:98)iscontainedin∪i∈IN(cid:48)i,andeachN(cid:48)i containsahyperplaneseparatingz 9 fromZ(cid:98). IfH ∈ A(I)∪(cid:83) N(cid:48) isseparatedfromzbyahyperplaneU,thenU ∈ A(I)∪(cid:83) N(cid:48),whence i∈I i i∈I i wecanchooseF(I)tocontainz. Hence(cid:98)Z(cid:98)(cid:99)istheunionofZ(cid:98)andafinitecollectionofgeneralizedframes F(I ),...,F(I ). 1 t Foranyp ∈ {1,...,t},wehaveF(I ) = Y(I )×(cid:81) L andweletY(I ) = im(Y(I ) → Z)and p p j∈Ip j p p let Lj = im(Lj ∩Z(cid:98) → Z) be the cycle of length nNj to which Lj maps, for each j ∈ Ip. Note that Z containsF(I ) = Y(I )×(cid:81) L andsowedefinethequotientmapq : F(I ) → Z astheproductof p p j∈IP j p p theabovecombinatorialquotientmaps. Namely,q (y,(r ) ) = (y,(r mod n ) )fory ∈ Y(I ) p j j∈IP j Nj j∈Ip p andr ∈ L . j j Toensurethatq (F(I )∩F(I )) = q (F(I )∩F(I ))foralli,j ≤ t,itsufficestoshowthat p p j j p j     (cid:89) (cid:89) (cid:89) F(Ip)∩F(Ij) := Y(Ip)× Lk∩Y(Ij)× L(cid:96) = [Y(Ip)∩Y(Ij)]× Lk. k∈Ip (cid:96)∈Ij k∈Ip∩Ij Thisinturnfollowsfrom[15,Proposition2.5]. Hence,thequotientmapsq arecompatibleandtherefore p defineacombinatorialquotientmapq : (cid:98)Z(cid:98)(cid:99) → Z extendingthemapsqp. Observe that if H = ∅, i.e. K = (cid:98)Z(cid:98)(cid:99), then we take Y = Z. By hypothesis, Z admits a local isometry toS andhasthedesiredcardinality. Moreover, ourhypothesisong ensuresthatg (cid:54)∈ π Y, butthemap Γ 1 q shows that any closed combinatorial path in S representing g lifts to a (non-closed) path in Z, so the Γ proofofthetheoremiscomplete. ThuswecanandshallassumethatH (cid:54)= ∅. Quotients of H-frames: To extend q to the rest of K, we now describe quotient maps, compatible with themapZ(cid:98) → Z,onframesassociatedtohyperplanesinH. AnisolatedH-frameisaframe(H ∩K)×L, whereH ∈ HandH crossesnohyperplaneofZ(cid:98)(andhencecrossesnohyperplaneof(cid:98)Z(cid:98)(cid:99)). Aninterfered H-frameisaframe(H∩K)×L,whereH ∈ HandH crossesanelementofZ. Equivalently,(H∩K)×L isinterferedifg (Z(cid:98))containsa1–cubeandisisolatedotherwise. N(H) DefinequotientmapsonisolatedH-framesbythesamemeansaswasusedforarbitraryframesin[9]: let(H ∩K)×LbeanisolatedH-frame. LetH betheimmersedhyperplaneinS towhichH issentby Γ S(cid:101)Γ → SΓ,andletH ∩K betheimageofH∩K. WeformaquotientYH = H ∩K×Lofeveryisolated H-frame(H ∩K)×L. NowwedefinethequotientsofinterferedH-frames. LetA(cid:98)= g (Z(cid:98))andletAbeimageofA(cid:98)under N(H) Z(cid:98) → Z.... There is a local isometry A → SΓ, to which we app.l.y. canonical c(cid:12)o.m.. pl(cid:12)etion to produce a finite coverSΓ → SΓ whereAembeds. By Lemma 2.8 of [9], deg(SΓ → SΓ) = (cid:12)(cid:12)S(Γ0)(cid:12)(cid:12) = (cid:12)(cid:12)A(0)(cid:12)(cid:12) ≤ (cid:12)(cid:12)Z(0)(cid:12)(cid:12). Let ... H ∩K = im(H ∩K → S ),andmaptheinterferedH-frame(H ∩K)×LtoY = H ∩K ×L. Γ H Constructing Y: We now construct a compact cube complex Y(cid:48) from Z and the various quotients Y . H AhyperplaneW inK separatesH1fromZ(cid:98)onlyifW ∈ N. EachH-hyperplaneframehastheform(Hi∩ K)×L = (H ∩K)×[0, m ],parametrizedsothat(H ∩K)×{0}istheclosestcombinatorialhyperplane i i i i in the frame to Z(cid:98). We form Y(cid:48)(1) by gluing YH1 to Z along the image of gZ(cid:98)((H1∩K)×{0}), enabled by the fact that the quotients of interfered H-frames are compatible with Z(cid:98) → Z. In a similar manner, formY(cid:48)(i)fromY(cid:48)(i−1)andY byidentifyingtheimageof(H ∩K)×{m }∩(H ∩K)×{0} Hi i−1 i−1 i inY ⊂ Y(cid:48)(i−1)withitsimageinY . LetY(cid:48) = Y(cid:48)(k). Hi−1 Hi LetK(cid:48) = (cid:98)Z(cid:98)(cid:99)∪(cid:2)(cid:83)Hi∈H(Hi∩K)×Li(cid:3). Since Hi ∩Hj = ∅fori (cid:54)= j, there exists amap (K(cid:48),x˜) → (Y(cid:48),x) and a map (Y(cid:48),x) → (S ,x) such that the composition is precisely the restriction to K(cid:48) of the Γ coveringmap(S(cid:101)Γ,x˜) → (SΓ,x). IfY(cid:48) → S failstobealocalisometry,thenthereexistsiandnontrivialopencubese ⊂ H ∩K × Γ i−1 {m } (or Z if i = 1) and c ⊂ H ∩K ×{0} such that S contains an open cube e¯×c¯, where e¯,c¯are i−1 i Γ the images of e,c under S(cid:101)Γ → SΓ, respectively. Moreover, since g (Hi∩K) ⊆ g (Hi−1∩K), we can Z(cid:98) Z(cid:98) 10

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.