Approximate Homotopy of Homomorphisms from C(X) into a Simple C -algebra ∗ 8 0 0 Huaxin Lin 2 n a Current address: Department of Mathematics, University of Oregon, Eugene, J OR 97403,USA 8 E-mail address: [email protected] 2 ] A O . h t a m [ 6 v 5 2 1 2 1 6 0 / h t a m : v i X r a 2000 Mathematics Subject Classification. Primary : 46L05,46L35 Abstract. LetX beafiniteCWcomplexandleth1,h2:C(X)→Abetwo unitalhomomorphisms,whereAisaunitalC∗-algebra. Westudytheproblem when h1 and h2 are approximately homotopic. We present a K-theoretical necessary and sufficient condition for them to be approximately homotopic undertheassumptionthatAisaunitalseparablesimpleC∗-algebra oftracial rank zero, or A is a unital purely infinite simpleC∗-algebra. When they are approximatelyhomotopic,wealsogiveaboundforthelengthofthehomotopy. Theseresultsarealsoextended tothecasethat h1 andh2 areapproximately multiplicativecontractive completelypositivelinearmaps. Suppose that h:C(X)→Ais amonomorphism and u∈Ais aunitary (with [u] = {0} in K1(A)). We prove that, for any ǫ > 0, and any compact subset F ⊂C(X), there exists δ>0andafinite subset G ⊂C(X) satisfying the following: if k[h(f),u]k < δ and Bott(h,u) = {0}, then there exists a continuous rectifiablepath {ut:t∈[0,1]}suchthat u0=u, u1=1A and k[h(g),ut]k<ǫ for all g∈F and t∈[0,1]. (e0.1) Moreover, Length({ut})≤2π+ǫ. (e0.2) We show that if dimX ≤ 1, or A is purely infinite simple, then δ and G are universal(independentofAorh). InthecasethatdimX=1,thisprovidesan improvementoftheso-calledtheBasicHomotopyLemmaofBratteli,Elliott, Evans and Kishimoto for the case that A is mentioned above. Moreover, we showthatδandG cannotbeuniversalwheneverdimX≥2.Nevertheless,we alsofoundthatδcanbechosentobedependentonameasuredistributionbut independent of A and h. The above version of the so-called Basic Homotopy isalsoextended tothecasethatC(X)isreplacedbyanAH-algebra. Wealsopresentsomegeneralversionsofso-calledSuperHomotopyLemma. Contents Chapter 1. Prelude 1 1. Introduction 1 2. Conventions and some facts 4 3. The Basic Homotopy Lemma for dim(X) 1 12 ≤ Chapter 2. The Basic Homotopy Lemma for higher dimensional spaces 25 4. K-theory and traces 25 5. Some finite dimensional approximations 30 6. The Basic Homotopy Lemma — full spectrum 38 7. The Basic Homotopy Lemma — finite CW compleces 44 8. The Basic Homotopy Lemma — compact metric spaces 49 9. The constant δ and an obstruction behind the measure distribution 54 Chapter 3. Purely infinite simple C∗-algebras 63 10. Purely infinite simple C∗-alegbras 63 11. Basic Homotopy Lemma in purely infinite simple C∗-algebras 69 Chapter 4. Approximate homotopy 77 12. Homotopy length 77 13. Approximate homotopy for homomorphisms 85 14. Approximate homotopy for approximately multiplicative maps 93 Chapter 5. Super Homotopy 99 15. Super Homotopy Lemma — purely infinite case 99 16. Super Homotopy Lemma — finite case 106 Chapter 6. Postlude 115 17. Non-commutative cases 115 18. Concluding remarks 124 Bibliography 127 v CHAPTER 1 Prelude 1. Introduction Let A be a unital C∗-algebra and let u A be a unitary which is in the ∈ connected component U (A) of the unitary group of A containing the identity. 0 Then there is a continuous path of unitaries in U (A) starting at u and ending at 0 1. It is known that the path can be made rectifiable. But, in general, the length of the path has no bound. N. C. Phillips ([41]) proved that, in any unital purely infinite simple C∗-algebra A, if u U (A), then the length of the path from u to 0 ∈ the identity can be chosen to be smaller than π+ǫ for any ǫ >0. A more general result proved by the author shows that this holds for any unital C∗-algebras with real rank zero ([20]). Let X be a path connected finite CW complex. Fix a point ξ X. Let A be ∈ a unital C∗-algebra. There is a trivial homomorphism h : C(X) A defined by 0 → h (f) = f(ξ)1 for f C(X). Suppose that h : C(X) A is a unital monomor- 0 A ∈ → phism. When can h be homotopic to h ? When h is homotopic to h , how long 0 0 could the length of the homotopy be? This is one of the questions that motivates this paper. Let h ,h : C(X) A be two unital homomorphisms. A more general ques- 1 2 → tion is when h and h are homotopic? When A is commutative, by the Gelfand 1 2 transformation, it is a purely topological homotopy question. We only consider noncommutative cases. To the other end of noncommutativity, we only consider the case that A is a unital simple C∗-algebra. To be possible and useful, we ac- tually consider approximate homotopy. So we study the problem when h and h 1 2 are approximately homotopic. For all applications that we know, the length of the homotopy is extremely important. So we also ask how long the homotopy is if h 1 and h are actually approximately homotopic. 2 Let X be a path connected metric space. Fix a point ξ . Let Y = X ξ . X X X \ Let x X and let L(x,ξ ) be the infimum of the length of continuous paths from X ∈ x to ξ . Define X L(X,ξ )=sup L(x,ξ ):x X . X X { ∈ } We prove that, for any unital simple C∗-algebra of tracialrank zero, or any unital purely infinite simple C∗-algebra A, if h : C(X) A is a unital monomorphism → with [h ]= 0 in KL(C (Y ),A), (e1.1) |C0(YX) { } 0 X then, for any ǫ >0 and any compact subset C(X), there is a homomorphism F ⊂ H :C(X) C([0,1],A) such that → π H h on , π H =h and 0 ǫ 1 0 ◦ ≈ F ◦ Length( π H ) L(X,ξ ), t X { ◦ } ≤ 1 2 1. PRELUDE where h (f) = f(ξ )1 for all f C(X) and π : C([0,1],A) A is the point- 0 X A t ∈ → evaluation at t [0,1],and where Length( π H ) is appropriatelydefined. Note t ∈ { ◦ } that [h ] = 0 . Thus the condition (e1.1) is necessary. Moreover, the 0|C0(YX) { } estimate of length can not be improved. Suppose that h ,h : C(X) A are two unital homomorphisms. We show 1 2 → that h and h are approximately homotopic if and only if 1 2 [h ]=[h ] in KL(C(X),A), 1 2 undertheassumptionthatAisaunitalseparablesimpleC∗-algebra oftracialrank zero, or A is a unital purely infinite simple C∗-algebra. Moreover, we show that the length of the homotopy can be bounded by a universal constant. Bratelli,Elliott, Evans andKishimoto ([2]) consideredthe following homotopy question: Letuandvbetwounitariessuchthatualmostcommuteswithv.Suppose that v U (A). Is there a rectifiable continuous path v : t [0,1] with v = v, 0 t 0 ∈ { ∈ } v =1 suchthatthe entire pathalmostcommutes withu?They foundthat there 1 A is an additional obstacle to prevent the existence of such path of unitaries. The additional obstacle is the Bott element bott(u,v) associated with the pair u and v. They proved what they called the Basic Homotopy Lemma: For any ǫ > 0 there exists δ > 0 satisfying the following: if u,v are two unitaries in a unital separable simple C∗-algebra withrealrankzeroandstablerankone,orina unitalseparable purelyinfinite simpleC∗-algebra A, ifv U (A) andsp(u)is δ-denseinS1 except 0 ∈ possibly for a single gap, uv vu <δ and bott (u,v)=0, 1 k − k thenthere exists a rectifiable continuouspathofunitaries v :t [0,1] suchthat t { ∈ } v =v, v =1 and uv v u <ǫ 0 1 A t t k − k for all t [0,1]. Moreover, ∈ Length( v ) 4π+1. t { } ≤ Bratteli, Elliott, Evans and Kishimoto were motivated by the study of classifi- cation of purely infinite simple C∗-algebras. The Basic Homotopy Lemma played animportantroleintheirworkrelatedtotheclassificationofpurelyinfinitesimple C∗-algebras and that of [9]. The renewed interest of this type of results is at least partly motivated by the study of automorphism groups of simple C∗-algebras (see [18]). ItisalsoimportantinthestudyofAF-embeddingofcrossedproducts([39]). We now replace the unitary u in the Basic Homotopy Lemma by a monomor- phism h :C(X) A. We first replace S1 by a path connected finite CW complex → X. A bott element bott (h,v) can be similarly defined. We proved that, with the 1 assumption A is a unital simple C∗-algebra of real rank zero and stable rank one, or A is a unital purely infinite simple C∗-algebra the Basic Homotopy Lemma holds for any compact metric space with dimension no more than one. Moreover, in the case that K (A)= 0 , the constant δ does not depend on the spectrum of 1 { } h (so that the condition on the spectrum of u in original Homotopy Lemma can be removed). The proof is shorter than that of the original Homotopy Lemma of Bratteli,Elliott, Evans andKishimoto. Furthermore,we areable to cut the length of homotopy by more than half (see 3.7 and 11.3). For more general compact metric space, the simple bott element has to be replaced by a more general map Bott(h,v). Even with vanishing Bott(h,v) and 1. INTRODUCTION 3 withAhavingtracialrankzero,weshowthatthesamestatementisfalsewhenever dimX 2. However, if we allow the constant δ not only depends on X and ǫ but ≥ also depends on a measure distribution, then the similar homotopy result holds (see 7.4) for unital separable simple C∗-algebras with tracial rank zero. On the other hand, if A is assumed to be purely infinite simple, then there is no such measure distribution. Therefore, for purely infinite simple C∗-algebras, the Basic HomotopyLemmaholdsforanycompactmetricspacewithshorterlengths. Infact our estimates on the lengths is 2π+ǫ (for the case that A is purely infinite simple as well as for the case that A is a unital separable simple C∗-algebra with tracial rank zero). Several other homotopy results are also discussed. In particular, a version of Super Homotopy Lemma (of Bratelli, Elliott, Evans and Kishimoto) for finite CW complex X is also presented. The presentation is organizedas follows: In section 2, we provide some conventions and a number of facts which will be used later. In section 3, we present the Basic Homotopy Lemma for X being a compact metricspacewithcoveringdimensionno morethanoneunderthe assumptionthat A is a unital simple C∗-algebra of real rank zero and stable rank one, or A is a unitalpurelyinfinite simpleC∗-algebra. The improvementismadenotonlyonthe bound of the length but is also made so that the constant δ does not depend on the spectrum of the homomorphisms. In section 4 and section 5, we present some results which are preparations for later sections. In section 6, 7 and 8, we prove a version of the Basic Homotopy Lemma for general compact metric space under the assumption that A is a unital separable simple C∗-algebra of tracial rank zero. The lengthy proof is due partly to the complexity caused by our insistence that the constant δ should not be dependent on homomorphisms or A but only on a measure distribution. In section 9, we show why the constant δ can not be made universal as in the dimension 1 case. A hidden topological obstacle is revealed. We show that the originalversion of the Basic Homotopy Lemma fails whenever X has dimension at least two for simple C∗-algebras with real rank zero and stable rank one. In section 10, we present some familiar results about purely infinite simple C∗-algebras. In section 11, we show that the Basic Homotopy Lemma holds for general compact metric spaces under the assumption that A is a unital purely infinite simple C∗-algebra. In section 12, we discuss the length of homotopy. A definition related to Lips- chitz functions is given there and some elementary facts are also given. Insection13,weshowtwohomomorphismsareapproximatelyhomotopicwhen theyinducethesameKLelementundertheassumptionthatAisaunitalseparable simple C∗-algebra of tracial rank zero, or A is a unital purely infinite simple C∗- algebra. We also give an estimate on the bound of the length of the homotopy. In section 14, we extend the results in section 13 to the maps which are not necessary homomorphisms. Insection15,wepresentaversionoftheso-calledSuperHomotopyLemmafor unital purely infinite simple C∗-algebra A. 4 1. PRELUDE In section 16, we show that same version of the Super Homotopy Lemma is valid for unital separable simple C∗-algebra of tracial rank zero. In section 17,we show that the Basic Homotopy Lemma in section 8 and 11 is valid if we replace C(X) by a unital AH-algebra. In section 18, we end this paper with a few concluding remarks. Acknowledgment Themostofthisresearchwasdoneinthesummer2006whentheauthorwasin East China Normal University where he had a nice office and necessary computer equipments. It is partially supported by Shanghai Priority Academic Disciplines. The work is also partially supported by a NSF grant (00355273). 2. Conventions and some facts 2.1. Let A be a C∗-algebra. Using notation introduced by [8], we denote K(A)= (K (A) K (A,Z/nZ)). i=0,1 i n=1 i ⊕ ⊕ Letm 1beaninteger. DenotebyC aMcommutativeC∗-algebra withK (C )= m 0 m Z/mZ≥and K (C )=0. So K (A,Z/mZ)=K (A C ), i=0,1. 1 m i i m ⊗ A theorem of Dadarlat and Loring ([8]) states that Hom (K(A),K(B))=KL(A,B), Λ ∼ if A satisfies the Universal Coefficient Theorem and B is σ-unital (see [8] for the definition of Hom (K(A),K(B))). We will identify these two objects. Λ Let m 1 be an integer. Put ≥ F K(A)= (K (A) K (A,Z/kZ)). m i=0,1 i k|m i ⊕ ⊕ We will also identify these two objectsM. 2.2. Let B be a sequence of C∗-algebras. Denote by l∞( B ) the product n n { } of B , i.e., the C∗-algebra of all bounded sequences a : a B . Denote by n n n n { } { ∈ } c ( B )thedirectsumof B ,i.e,theC∗-algebra ofallsequences a :a B 0 n n n n n { } { } { ∈ } for which lim a = 0. Denote by q ( B ) = l∞( B )/c ( B ) and by n→∞ n ∞ n n 0 n k k { } { } { } q :l∞( B ) q ( B ) the quotient map. n ∞ n { } → { } 2.3. Let A be a C∗-algebra and let B be another C∗-algebra. Let ǫ > 0 and A be a finite subset. We say that a contractivecompletely positive linear map G ⊂ L:A B is δ- -multiplicative if → G L(ab) L(a)L(b) <δ for all a,b . k − k ∈G Denote by P(A) the set of projections and unitaries in M (A˜) M (A^C ). ∞ ∞ m ∪ ⊗ m≥1 [ We also use L for the map L id : A C M B C M , ⊗ Mk⊗Cm ⊗ m ⊗ k → ⊗ m ⊗ k k = 1,2,...,. As in 6.1.1 of [28], for a fixed p P(A), if L is δ- -multiplicative ∈ G with sufficiently small δ and sufficiently large , L(p) is close to a projection (with G the norm of difference is less than 1/2) which will be denote by [L(p)]. Note if two projections are both close to L(p) within 1/2, they are equivalent. IfL:A B isδ- -multiplicative,thenthereisafinite subset P(A), such → G Q⊂ that [L](x) is well defined for x , where is the image of in K(A), which ∈ Q Q Q 2. CONVENTIONS AND SOME FACTS 5 means that if p ,p and [p ]=[p ], then [L(p )] and [L(p )] defines the same 1 2 1 2 1 2 ∈Q element in K(B). Moreover, if p ,p ,p p , [L(p p )]= [L(p )]+[L(p )] 1 2 1 2 1 2 1 2 ⊕ ∈Q ⊕ (see 0.6 of [24] and 4.5.1 and 6.1.1 of [28]). This finite subset will be denoted Q by . Let K(A). We say [L] is welldefined, if . In what follows, δ,G P δ,G Q P ⊂ | Q ⊃P whenever we write [L] , we mean that [L] is well defined (see also 2.4 of [6] for P P | | further explanation). The following proposition is known and has been implicitly used many times. Proposition 2.4. Let A be a separable C∗-algebra for which K (A) is finitely i generated (for i=0,1), and let K(A) be a finite subset. Then, there is δ >0 P ⊂ andafinitesubset Asatisfying thefollowing: IfB is aunitalC∗-algebra andif G ⊂ L:A B is a δ- -multiplicative contractive completely positive linear map, there → G is an element κ Hom (K(A),K(B)) such that Λ ∈ [L] =κ . (e2.1) P P | | Moreover, thereisafinitesubset K(A)suchthat,if[L] iswelldefined, PA ⊂ |PA there is a unique κ Hom (K(A),K(B)) such that (e2.1) holds. Λ ∈ Proof. Since K (A) is finitely generated (i=0,1), by 2. 11 of [8], i Hom (K(A),K(B))=Hom (F K(A),F K(B)) Λ Λ m m for some m 1. Thus it is clear that it suffices to show the first part of the ≥ proposition. Suppose that the first part of the lemma fails. One obtains a finite subset K(A), a sequence of σ-unital C∗-algebras B , a sequence of positive n numberPs ⊂d with ∞ δ < , a finite subsets A with ∞ is dense { n} n=1 n ∞ Gn ⊂ ∪n=1Gn in A, and a sequence of δ - -multiplicative contractive completely positive linear n n P G maps L : A B such that there exists no κ Hom (K(A),K(B)) satisfying n n Λ → ∈ (e2.1). DefineΦ:A l∞( B )byΦ(a)= L (a) fora AanddefineΦ¯ :A n n q ( B )by→Φ¯ =π{Φ,w⊗hKer}eπ :l∞( B { ) } q ( ∈B )isthequotie→nt ∞ n n ∞ n { ⊗K} ◦ { ⊗K} → { ⊗K} map. Thus weobtainanelementα Hom (F K(A),F K(q ( B )))such Λ m m ∞ n that [Φ¯]=α. Since K (A) is finitely∈generated (i=0,1), by 2.11{of [8⊗],Kth}ere is an i integer m 1 such that ≥ Hom (K(A),K(q ( B )))=Hom (F K(A),F K(q ( B )) and Λ ∞ { n⊗K} ∼ Λ m m ∞ { n⊗K} Hom ((K(A),K(B ))=Hom (F K(A),F K(B )). Λ n ∼ Λ m m n By applying 7.2 of [31] and the proof of 7.5 of [31], for all larger n, there is an element κ Hom (F K(A),F K(B )) n Λ m m n ∈ such that [L ] =κ . n P n P | | This contradicts the assumption that the first part of the lemma fails. (cid:3) The following is well known and follows immediately from the definition. Proposition2.5. LetAbea unitalamenable C∗-algebra. Forany finitesubset K(A), there exists δ >0 and a finitesubset A satisfying the following: for P ⊂ G ⊂ 6 1. PRELUDE any pair of δ- -multiplicative contractive completely positive linear maps L ,L : 1 2 G A B (for any unital C∗-algebra B), → [L ] =[L ] 1 P 2 P | | provided that L L on . 1 δ 2 ≈ G 2.6. Let B be a C∗-algebra and C = C([0,1],B). Define π : C B by t → π (f)=f(t) for all f C. This notation will be used throughout this article. t ∈ Thefollowingfollowsimmediatelyfrom2.5andwillbeusedfrequentlywithout further notice. Proposition2.7. Let A and B be two unital C∗-algebras and let L:A B be → a contractive completely positive linear map. Let P(A) be a finite subset. Sup- Q⊂ pose that, for some small δ >0 and a large finite subset , L is δ- -multiplicative G G and . Put = in K(A). Suppose that H : A C([0,1],B) is a δ,G Q ⊃ Q P Q → contractive completely positive linear map such that π H = L and π H is 0 t ◦ ◦ δ- -multiplicative for each t [0,1]. Then, for each t [0,1], G ∈ ∈ [π H] =[L] . t P P ◦ | | The following follows immediately from 2.1 of [35]. Lemma 2.8. Let B be a separable amenable C∗-algebra. For any ǫ > 0 and any finite subset B there exists a finite subset B and δ > 0 satisfy- 0 1 F ⊂ F ⊂ ing the following: Suppose that A is a unital C∗-algebra, φ : B A is a unital → homomorphism and u A is a unitary such that ∈ [φ(a),u] <δ for all a . (e2.2) 1 k k ∈F Then there is an ǫ- S-multiplicative contractive completely positive linear map 0 F ⊗ ψ :B C(S1) A such that ⊗ → φ(a) ψ(a) <ǫ and ψ(a g) φ(a)g(u) <ǫ (e2.3) k − k k ⊗ − k for all a and g S, where S = 1 ,z and z C(S1) is the standard 0 C(S1) ∈ F ∈ { } ∈ unitary generator of C(S1). 2.9. Let A be a unital C∗-algebra. Denote by U(A) the group of all unitaries in A. Denote by U (A) the path connected component of U(A) containing 1 . 0 A Denote by Aut(A) the group of automorphisms on A. If u U(A), denote by ∈ adu the inner automorphism defined by adu(a)=u∗au for all a A. ∈ Definition 2.10. Let A and B be two unital C∗-algebras. Let h:A B be a → homomorphism and v U(B) such that ∈ h(g)v =vh(g) for all g A. ∈ Thus we obtain a homomorphism h¯ : A C(S1) B by h¯(f g) = h(f)g(v) for ⊗ → ⊗ f A and g C(S1). The tensor product induces two injective homomorphisms: ∈ ∈ β(0) : K (A) K (A C(S1)) (e2.4) 0 1 → ⊗ β(1) : K (A) K (A C(S1)). (e2.5) 1 0 → ⊗ ThesecondoneistheusualBottmap. Note,inthisway,onewriteK (A C(S1))= i ⊗ K (A) β(i−1)(K (A)). We use β(i) :K (A C(S1)) β(i−1)(K (A)) for the i i−1 i i−1 ⊕ ⊗ → the projection to β(i−1)(K (A)). i−1 d