Borel complexity and automorphisms of C*-algebras

Transcription

1 Martino Lupini York University Toronto, Canada January 15th, 2013

2 Table of Contents 1 Auto-homeomorphisms of compact metrizable spaces 2 Measure preserving automorphisms of probability spaces 3 Automorphisms of tracial von Neumann algebras 4 Automorphisms of C*-algebras

3 Borel reducibility E and F are equivalence relations on standard Borel spaces X and Y E B F iff f : X Y Borel such that xey iff f (x)f f (y) Borel reduction allows one to compare the complexity of eq. relations This gives a hierarchy in the complexity of classification problems.

4 Hierarchy Some equivalence relations can be used as benchmarks of complexity: the identity relation id X on a standard Borel space X ; the orbit equivalence relation ES X induced by some S X ; the orbit equivalence relation EG X induced by some G X. An equivalence relations E is: smooth if E B id X for some X ; classifiable (by countable structures) if E E S X for some S X ; orbitable if E EG X for some G X.

5 Conjugacy of actions Suppose that Γ is a countable discrete groups, and X is a compact metrizable space. Define Homeo(X ) to be the group of auto-homeomorphisms of X An action Γ φ X is a group homomorphism φ : Γ Homeo(X ) Two actions φ, φ are conjugate if there is α Homeo(X ) such that α φ (γ) α 1 = φ (γ) for every γ Γ Problem: Classify the actions of Γ on X up to conjugacy

6 Z-actions Consider the particular case Γ = Z Z-actions on X correspond to single auto-homeomorphisms of X. Conjugacy of actions correspond to conjugacy in Homeo(X ) Problem: Classify elements of Homeo(X ) up to conjugacy

7 Polish group of auto-homeomorphisms Homeo(X ) is a Polish group endowed with the compact-open topology This is the topology of uniform convergence w.r.t. any compatible metric We want to study the complexity of the relation of conjugacy inside Homeo(X ) E Homeo(X )

8 The zero-dimensional case If X is zero-dimensional then Homeo(X ) = Aut(B X ) where B X is the countable Boolean algebra of clopen subsets of X Moreover Aut(B X ) S It follows that E Homeo(X ) B E X S for some S X and hence E Homeo(X ) is classifiable

9 The Cantor space Theorem (Camerlo-Gao, 2001) If X = 2 ω is the Cantor space, then is Borel complete E Homeo(2 ω ) E Homeo(2 ω ) has thus maximal complexity among classifiable e.r. This is obtained of the corollary of Theorem (Camerlo-Gao, 2001) Isomorphism of countable Boolean algebras is Borel complete Equivalently, by (a Borel version of) Stone s duality, homeomorphism of Boolean spaces is Borel complete

10 Interval and square Consider now Homeo ([0, 1]) where [0, 1] is the unit interval Theorem (Hjorth, 2000) E Homeo([0,1]) is classifiable The proof involves a suitable coding of auto-homeomorphisms of [0, 1] by countable structures On the other hand increase of dimension entails an increase of complexity Theorem (Hjorth, 2000) E Homeo([0,1]2 ) is not classifiable This is one of the earliest application of Hjorth s theory of turbulence

11 (Pre)turbulent Polish group action Suppose that G X is a Polish group action V is an open neighborhood of 1 in G U is an open subset of X x is an element of U The local orbit O (x, U, V ) is the set of points of U that can be reached from x by applying elements of V without ever leaving U The action G X is preturbulent if 1 every orbit is dense 2 every local orbit is somewhere dense The action G X is turbulent if moreover it has meager orbits

12 Hjorth s turbulence theorem Theorem (Hjorth, 2000) If then E X G G X is preturbulent, and S Y is an S -action on a Polish space is generically E Y S -ergodic: For any Borel map f : X Y sending G-orbits into S -orbits, C invariant comeager set such that f [C] is contained in a single orbit. In particular if G X is turbulent then E X G is not classifiable

13 Consider the Polish group H = N Z with the product topology the subgroup G of H defined by (x n ) G iff the action G H by translation x n n 0 The action G H is turbulent. ( Moreover any element h of H codes φ h Aut [0, 1] 2) such that Gh = Gh iff φ h and φ h are conjugate. This shows that In particular E Homeo([0,1]2 ) E H G B E Homeo([0,1]2 ) is not classifiable

14 C*-algebras A unital C*-algebra A is a complex unital algebra with involution x x endowed with a norm making it a Banach space and satisfying and xy x y x x = x 2. All C*-algebras are henceforth assumed to be unital and separable Suppose that X is a compact metrizable space Let C(X ) be the space of complex-valued continuous functions on X C(X ) is a C*-algebra with pointwise operations and sup norm Abelian C*-algebras are exactly the ones of this form (Gelfand-Neimark)

15 Automorphisms groups Suppose that A is a C*-algebra Automorphisms of A are bijections α : A A preserving all the operations Automorphisms of A automatically preserve the norm The group of automorphisms of A is denoted by Aut(A) Aut(A) is a Polish group with the topology of pointwise norm convergence

16 Abelian case In the abelian case A = C(X ) for some compact metrizable space X There is an exact correspondence between auto-homeomorphisms of X and automorphisms of C(X ) Precisely is an isomorphism of Polish groups Homeo(X ) Aut (C(X )) φ (f f φ) In general Aut(A) can be thought as the group of auto-homeomorphisms of a noncommutative space

17 Automorphisms of standard probability spaces Suppose that (Y, ν) is a standard probability space Aut(Y, ν) is the group of measure preserving transformations of Y The weak topology on Aut(Y, ν) is the weakest topology such that T ν (T [A] B) measurable, for every A, B X Borel Aut(Y, ν) is a Polish group endowed with the weak topology Problem: Study the Borel complexity of conjugacy in Aut(Y, ν)

18 The atomless standard probability space Suppose that (X, µ) is the atomless standard probability space A (X, µ) = {T Aut (X, µ) : T is aperiodic} T Aut (X, µ) is aperiodic if T n 1 for every n 1 Lemma A (X, µ) is a conjugation-invariant dense G δ in Aut (X, µ) Theorem (Foreman-Weiss 2004, Kechris 2010) The action of Aut (X, µ) on A (X, µ) is turbulent Corollary The elements of Aut (X, µ) are not classifiable up to conjugacy.

19 Meager conjugacy classes The shortest argument is due to Rosendal (2009) Consider for V neighborhood of 1 in Aut (X, µ) and I N infinite N(V, I ) = {T Aut (X, µ) : T n V for some n I } By Rohlin s lemma N(V, I ) is dense open in Aut (X, µ). It follows that N (I ) = V N(V, I ) is a conjugation-invariant dense G δ subset of Aut(X, µ) If C is a nonmeager orbit then C N (I ) for every I N infinite. Thus C I N infinite N (I ) = {1}

20 The Rosendal property A similar argument works in general. Suppose that: G is a Polish group; V is a neighborhood of 1 in G; I N is infinite. As before set N (I, V ) = {g G : g n V for some n I } Definition G has the Rosendal property if N (I, V ) is dense for every V and I

21 Lemma If G H are Polish groups, G has the Rosendal property, then the relation on G of conjugacy in H does not have a comeager class Corollary Under the assumptions of the theorem, if moreover the action of G on itself by conjugation is (generically) turbulent, then the elements of H are not classifiable up to conjugacy.

22 Not purely atomic standard probability space Suppose that (Y, ν) is a not purely atomic standard probability space. There is non-null X Y such that (X, ν X ) is the atomless standard probability space, where for A X ν(a) = 1 ν(x ) ν(a) Since Aut(X, ν X ) Aut(Y, ν) it follows that conjugacy in Aut(Y, ν) is not classifiable

23 L spaces Define L (Y, ν) = {f : Y C Borel, bounded, up to mesure zero} This is a C*-algebra with respect to pointwise operations, with norm The function f = inf {α R + : f α a.e.} τ : L (Y, ν) C f fdν defines the norm f τ = τ (f f ) 1 2 = ( ) 1 f 2 2 dν (L (Y, ν), τ) is an example of tracial von Neumann algebra

24 Traces A trace τ on a C*-algebra A is a function τ : A C such that 1 τ is linear 2 τ(1) = 1 3 τ (a a) 0 4 τ (ab) = τ (ba) τ defines a the τ-norm a τ = τ (a a) 1 2

25 Tracial von Neumann algebras A tracial von Neumann algebra (M, τ) is given by: 1 a C*-algebra M 2 a trace τ on M such that the unit ball of M is complete in τ-norm The abelian tracial vn algebras are exactly the ones of the form L (Y, ν) Tracial von Neumann algebras will be assumed to be separable in τ-norm

26 Automorphism groups Suppose that (M, τ) is a tracial von Neumann algebra Denote by Aut(M, τ) the group of automorphisms of M that preserve τ Aut(M, τ) with the topology of pointwise τ-norm convergence is Polish Aut (L (Y, ν), τ) = Aut(Y, ν) (as Polish groups) In general Aut(M, τ) can be thought as the group of measure preserving transformations of a noncommutative probability space.

27 Finite factors Suppose that (M, τ) is a tracial von Neumann algebra The center of M is Z(M) = {a M : ab = ba for all b M} A tracial von Neumann algebra M is a finite factor if the center is trivial Every tracial vn algebra is isomorphic to a direct integral of factors The trace in a finite factor is unique Any automorphism of a finite factor is automatically trace-preserving

28 II 1 factors A matrix algebra M n with the usual normalized trace is a finite factor All finite dimensional finite factors are isomorphic to matrix algebras Definition A II 1 factor is an infinite dimensional finite factor II 1 factors play a key role in the theory of von Neumann algebras They were first studied in the 1930s by Murray and von Neumann

29 The hyperfinite II 1 factor Consider the increasing sequence of matrix algebras regarded as finite factors, where (M 2 n) n ω Define M 2 n M 2 n+1 ( ) A 0 A = 0 A R = n ω M 2 n (completion w.r.t. τ ) This is the hyperfinite II 1 factor It is the Fraïssé limit of the class of matrix algebras (as finite factors)

30 McDuff II 1 factors The hyperfinite II 1 factor plays a key role in the study of II 1 factors. It is the first example of II 1 factor considered by Murray and von Neumann R is self-absorbing, i.e. R R = R If M is any II 1 factor, then R M Definition A II 1 factor M is McDuff if M R = M Being McDuff is equivalent to many other regularity properties Goal: Show that the automorphisms of M are not classifiable Inspired by the commutative case, let us consider first the case M = R

31 Inner automorphisms We need to identify the right analog of aperiodic transformation Definition An element u of R is unitary if uu = u u = 1 Any unitary element defines an automorphism of R Ad(u) : a uau Automorphisms of this form are called inner Remark: In the abelian case there are no nontrivial inner automorphisms Lemma Inner automorphisms are dense in Aut(R)

32 Aperiodic automorphisms Definition An automorphism of R is aperiodic if none of its powers is inner This is the analogue of aperiodic element of Aut (X, µ) Denote by A(R) the set of aperiodic elements of Aut(R) Lemma A(R) is a conjugation-invariant dense G δ subset of Aut(R) Theorem (Kerr-Li-Pichot, 2010) The action of Aut(R) on A (R) is turbulent Corollary Automorphisms of R are not classifiable up to conjugacy

33 Nonclassification for McDuff factors In order to show that the conjugacy classes in Aut(R) are meager, one in fact shows that Aut(R) has the Rosendal property Therefore whenever G is a Polish group such that G Aut(R), the relation of conjugacy in G is not classifiable In particular if M = M R is McDuff then Aut(M) = Aut (M R) Aut (1 R) = Aut (R) Thus the automorphism of any McDuff II 1 factor are not classifiable The same conclusion holds for the free group factors L (F n )

34 Open problems Suppose that M is an arbitrary II 1 factor Question Are the automorphisms of M not classifiable up to conjugacy? Theorem (Kerr-Li-Pichot, 2010) The action of Aut(M) on itself by conjugation is generically preturbulent To answer the question affirmatively, it would be enough to show that Aut(M) has meager conjugacy classes

35 C*-algebras (joint work with David Kerr and Chris Phillips) A is a C*-algebra, and Aut(A) is the Polish group of automorphisms of A Problem: Determine the Borel complexity of conjugacy in Aut(A) The Jiang-Su algebra Z plays here an analog role as R for II 1 factors Z was defined by Jiang and Su in 1999 in a groundbreaking paper Z was constructed as direct limit of suitable building blocks Z is in fact the Fraïssé limit of these building blocks Z is self-absorbing, i.e. Z Z = Z (more in fact is true...)

36 Z-absorbing C*-algebras The analogue of McDuff factors are Z-absorbing C*-algebras A C*-algebra A is Z-absorbing if A Z = A Conjecture (Toms-Winter) If A is simple nuclear, the following statements are equivalent 1 A is Z-absorbing; 2 A has finite nuclear dimension; 3 A has strict comparison. Nuclear dimension is the right noncommutative generalization of the notion of dimension for compact metrizable spaces The Toms-Winter conjecture has been verified in many interesting cases

37 Strongly aperiodic automorphisms Z is endowed with a unique trace τ The completion Ẑ of Z with respect to the τ-norm is isomorphic to R Any α Aut(Z) induces an automorphism α of Ẑ = R Definition α is strongly aperiodic if α Aut(R) is aperiodic Define A(Z) to be the set of strongly aperiodic automorphisms Lemma (Sato) A(Z) is a conjugation-invariant dense G δ subset of Aut(Z) Theorem The action of Aut(Z) on A (Z) is turbulent

38 Nonclassification In order to show that Aut(Z) has meager conjugacy classes, one in fact shows that Aut(Z) is a Rosendal group Therefore whenever G is a Polish group such that G Aut(Z) the relation of conjugacy in G is not classifiable In particular if A is a Z-absorbing C*-algebra then Aut(A) = Aut (A Z) Aut (1 Z) = Aut(Z) and the automorphisms of A are not classifiable up to conjugacy The same conclusion holds when A = A K (consequence of a result of Kechris and Sofronidis about the unitary group)

39 Open problem Suppose that A is an infinite-dimensional simple C*-algebra Question Is it true that the automorphisms of A are not classifiable up to conjugacy The assumption that A is simple is necessary There is a nonsimple A for which automorphisms are classifiable An example is given by A = C (2 ω ), as we have seen

40 Outer conjugacy Consider the relation E of outer conjugacy on Aut(Z) defined by αeβ iff Ad(u) γ α γ 1 = β for some γ Aut(Z) and u Z unitary Question Is E classifiable by countable structures? All strongly aperiodic automorphisms of Z are E-equivalent (Sato, 2009) Since strongly aperiodic automorphisms form are a comeager set, the theory of turbulence seems not applicable to this situation