1 Tracial moment problems related to Connes’ embedding problem Sabine Burgdorf École polytechnique fédéral de Lausanne San Diego January 11, 2013 2 Classical moment problem I Let K ⊆ Rn be closed. Moment problem Let L : R[x] → R be linear, L(1) = 1. Does there exist a probability measure µ with supp µ ⊆ K such that for all f ∈ R[x]: Z L(f ) = f (a) dµ(a)? 2 Classical moment problem I Let K ⊆ Rn be closed. Moment problem Let L : R[x] → R be linear, L(1) = 1. Does there exist a probability measure µ with supp µ ⊆ K such that for all f ∈ R[x]: Z L(f ) = f (a) dµ(a)? Theorem (Riesz, Haviland) Let K ⊆ Rn be non-empty and closed, L ∈ R[x]∨ . There exists a measure µ supported on K such that Z L(f ) = f (a) dµ(a) for all f ∈ R[x] if and only if L(p) ≥ 0 for all p ∈ R[X ] that are positive on K . 3 NC polynomials I X = (X1 , . . . , Xn ) non-commuting variables I ChX i unital associative algebra freely generated by X I NC polynomials: f = X w fw w, where w ∈ hX i, fw ∈ K 3 NC polynomials I X = (X1 , . . . , Xn ) non-commuting variables I ChX i unital associative algebra freely generated by X I NC polynomials: f = X fw w, where w ∈ hX i, fw ∈ K w I (anti-)involution ∗ : ChX i → ChX i s.t. I Xi self-adjoint is complex conjugation on C I ∗ 3 NC polynomials I X = (X1 , . . . , Xn ) non-commuting variables I ChX i unital associative algebra freely generated by X I NC polynomials: f = X fw w, where w ∈ hX i, fw ∈ K w I (anti-)involution ∗ : ChX i → ChX i s.t. I Xi self-adjoint is complex conjugation on C I ∗ I Evaluation in self-adjoint matrices I A = (A1 , . . . , An ) ∈ (SCs×s )n I f (A) = f1 1s + fX1 A1 + fX2 A2 + ... + fX 2 X3 X 3 A21 A3 A32 + ... 1 2 4 NC hypercube and quadratic module I The vN hypercube K is the set of all n-tuples of self-adjoint contractions in a finite vN algebra I The matricial hypercube is Kmat = {A | A ∈ (SCs×s )n for some s ∈ N, kAi k ≤ 1, i = 1 . . . , n} 4 NC hypercube and quadratic module I The vN hypercube K is the set of all n-tuples of self-adjoint contractions in a finite vN algebra I The matricial hypercube is Kmat = {A | A ∈ (SCs×s )n for some s ∈ N, kAi k ≤ 1, i = 1 . . . , n} I SChX i = {f ∈ ChX i | f ∗ = f } symmetric part of ChX i Definition The quadratic module of the hypercube is X X M := {f ∈ ChX i | f = hj∗ hj + pij∗ (1 − Xi2 )pij , hj , pij ∈ ChX i}. j I i,j If f ∈ M then f (A) 0 for all A ∈ K 4 NC hypercube and quadratic module I The vN hypercube K is the set of all n-tuples of self-adjoint contractions in a finite vN algebra I The matricial hypercube is Kmat = {A | A ∈ (SCs×s )n for some s ∈ N, kAi k ≤ 1, i = 1 . . . , n} I SChX i = {f ∈ ChX i | f ∗ = f } symmetric part of ChX i Definition The quadratic module of the hypercube is X X M := {f ∈ ChX i | f = hj∗ hj + pij∗ (1 − Xi2 )pij , hj , pij ∈ ChX i}. j I I i,j If f ∈ M then f (A) 0 for all A ∈ K P Let U = {f ∈ ChX i | f = j (pj qj − qj pj ), pj , qj ∈ ChX i} 5 Connes’ embedding problem Conjecture (Connes, 1976) If ω is a free ultrafilter on N and F is a II1 factor with separable predual, then F can be embedded into the ultrapower Rω . 5 Connes’ embedding problem Conjecture (Connes, 1976) If ω is a free ultrafilter on N and F is a II1 factor with separable predual, then F can be embedded into the ultrapower Rω . Theorem (Klep, Schweighofer, 2008) Connes’ conjecture holds iff the following is equivalent for all f ∈ ChX i: (i) Tr(f (A)) ≥ 0 for all A ∈ Kmat , (ii) ∀ε ∈ R>0 : f + ε ∈ M + U. 6 CEP using convex cones I Convex cones: M + U and P = {f ∈ ChX i | Tr(f (A)) ≥ 0 for all A ∈ Kmat } 6 CEP using convex cones I Convex cones: M + U and P = {f ∈ ChX i | Tr(f (A)) ≥ 0 for all A ∈ Kmat } I CEP is equivalent to P = (M + U) 6 CEP using convex cones I Convex cones: M + U and P = {f ∈ ChX i | Tr(f (A)) ≥ 0 for all A ∈ Kmat } I CEP is equivalent to P = (M + U) I This is equivalent to P ∨ = (M + U)∨ I C ∨ = {L : KhX i → K | L linear, L(p) ≥ 0 for all p ∈ C} 6 CEP using convex cones I Convex cones: M + U and P = {f ∈ ChX i | Tr(f (A)) ≥ 0 for all A ∈ Kmat } I CEP is equivalent to P = (M + U) I This is equivalent to P ∨ = (M + U)∨ I I C ∨ = {L : KhX i → K | L linear, L(p) ≥ 0 for all p ∈ C} Counter example for CEP? Find p ∈ P and L ∈ (M + U)∨ s.t. L(p) < 0 7 How to describe P ∨ ? Theorem (Riesz, Haviland) ∨ . There exists a Let K 0 ⊆ Rn be non-empty and closed, L ∈ R[x] R measure µ supported on K 0 such that L(f ) = f (a) dµ(a) for all f ∈ R[x] if and only if L(p) ≥ 0 for all p ∈ R[X ] that are positive on K 0. 7 How to describe P ∨ ? Theorem (Riesz, Haviland) ∨ . There exists a Let K 0 ⊆ Rn be non-empty and closed, L ∈ R[x] R measure µ supported on K 0 such that L(f ) = f (a) dµ(a) for all f ∈ R[x] if and only if L(p) ≥ 0 for all p ∈ R[X ] that are positive on K 0. I Let L := {L : ChX i → C | L linear, L(U) = {0}} I Mmat := {L ∈ L | ∃s ∈ N, measure µ : supp µ ⊆ K ∩ (SCs×s ) : R L(f ) = Tr(f (A) dµ(A) for all f ∈ ChX i} MvN := {L ∈ L | ∃ a finite vN algebra N and A ∈ K ∩ N : L(f ) = τ (f (A)) for all f ∈ ChX i} 7 How to describe P ∨ ? Theorem (Riesz, Haviland) ∨ . There exists a Let K 0 ⊆ Rn be non-empty and closed, L ∈ R[x] R measure µ supported on K 0 such that L(f ) = f (a) dµ(a) for all f ∈ R[x] if and only if L(p) ≥ 0 for all p ∈ R[X ] that are positive on K 0. I Let L := {L : ChX i → C | L linear, L(U) = {0}} I Mmat := {L ∈ L | ∃s ∈ N, measure µ : supp µ ⊆ K ∩ (SCs×s ) : R L(f ) = Tr(f (A) dµ(A) for all f ∈ ChX i} MvN := {L ∈ L | ∃ a finite vN algebra N and A ∈ K ∩ N : L(f ) = τ (f (A)) for all f ∈ ChX i} Question Do we have P ∨ = Mmat or P ∨ = MvN ? 8 The tracial moment problem (Matricial) tracial moment problem For which L ∈ L exist some s ∈ N and a probability measure µ with supp µ ⊆ K ∩ (SCs×s )n such that for all f ∈ ChX i: Z L(f ) = Tr(f (A)) dµ(A)? 8 The tracial moment problem (Matricial) tracial moment problem For which L ∈ L exist some s ∈ N and a probability measure µ with supp µ ⊆ K ∩ (SCs×s )n such that for all f ∈ ChX i: Z L(f ) = Tr(f (A)) dµ(A)? Tracial vN-moment problem For which L ∈ L exist a finite vN algebra N and A ∈ KvN ∩ N such that for all f ∈ ChX i L(f ) = τ (f (A))? 8 The tracial moment problem (Matricial) tracial moment problem For which L ∈ L exist some s ∈ N and a probability measure µ with supp µ ⊆ K ∩ (SCs×s )n such that for all f ∈ ChX i: Z L(f ) = Tr(f (A)) dµ(A)? Tracial vN-moment problem For which L ∈ L exist a finite vN algebra N and A ∈ KvN ∩ N such that for all f ∈ ChX i L(f ) = τ (f (A))? I s = 1: Classical moment problem on Rn 8 The tracial moment problem (Matricial) tracial moment problem For which L ∈ L exist some s ∈ N and a probability measure µ with supp µ ⊆ K ∩ (SCs×s )n such that for all f ∈ ChX i: Z L(f ) = Tr(f (A)) dµ(A)? Tracial vN-moment problem For which L ∈ L exist a finite vN algebra N and A ∈ KvN ∩ N such that for all f ∈ ChX i L(f ) = τ (f (A))? I s = 1: Classical moment problem on Rn Question Do we have P ∨ = Mmat or P ∨ = MvN ? 9 Intermission: Tracial Hankel matrix I We can describe (M + U)∨ using tracial Hankel matrices: I Associate to L ∈ L the sesquilinear form BL : ChX i × ChX i, (f , g) 7→ L(f ∗ g). 9 Intermission: Tracial Hankel matrix I We can describe (M + U)∨ using tracial Hankel matrices: I Associate to L ∈ L the sesquilinear form BL : ChX i × ChX i, (f , g) 7→ L(f ∗ g). Definition The tracial Hankel matrix M(L), indexed by u, v ∈ hX i, is given by M(L)u,v := L(u ∗ v ). For g ∈ SChX i and L ∈ L the localizing matrix M[g, L] is given by M[g, L]u,v := L(u ∗ gv ). I (M + U)∨ = {L ∈ L | M(L) 0, M[1 − Xi2 , L] 0 for all i} 10 One tracial Hankel matrix Example Consider KhX , Y i with basis (1, X , Y , X 2 , XY , YX , Y 2 , . . . ); 2 L(1) L(X ) L(Y ) L(X ) ... L(X ) L(X 2 ) L(XY ) L(X 3 ) . . . M(L) = L(Y ) L(YX ) L(Y 2 ) L(YX 2 ) . . . .. . .. . .. . .. . .. . L(1 − X 2 ) L(X − X 3 ) L(Y − X 2 Y ) ... 3 2 4 L(X − X ) L(X − X ) L(XY − X 3 Y ) . . . M[1 − X 2 , L] = L(Y − YX 2 ) L(YX − YX 3 ) L(Y 2 − YX 2 Y ) . . . .. .. .. .. . . . . 11 Relation between Mmat and MvN Theorem (B) Let L ∈ L with (i) M(L) 0 (ii) M[1 − Xi2 , L] 0 for all i = 1, . . . , n (iii) rank M(L) = s < ∞. Then L has a representing measure on K ∩ (SCs×s )n . 11 Relation between Mmat and MvN Theorem (B) Let L ∈ L with (i) M(L) 0 (ii) M[1 − Xi2 , L] 0 for all i = 1, . . . , n (iii) rank M(L) = s < ∞. Then L has a representing measure on K ∩ (SCs×s )n . Corollary (B) Let L ∈ MvN . Then L ∈ Mmat if M(L) is of finite rank. 12 Relation of P ∨ and M? Question Do we have P ∨ = Mmat or P ∨ = MvN ? 13 Relation of P ∨ and M I PvN := {p ∈ ChX i | τ (p(A)) ≥ 0 for all A ∈ K } Theorem (B) Let L ∈ L. Then L ∈ MvN if and only if L(p) ≥ 0 for all p ∈ PvN , i.e. ∨ . MvN = PvN 13 Relation of P ∨ and M I PvN := {p ∈ ChX i | τ (p(A)) ≥ 0 for all A ∈ K } Theorem (B) Let L ∈ L. Then L ∈ MvN if and only if L(p) ≥ 0 for all p ∈ PvN , i.e. ∨ . MvN = PvN Theorem (Klep, Schweighofer) (M + U) = PvN I ∨ (M + U)∨ = PvN I (M + U)∨ = MvN 14 Relation of P ∨ and M I Ps := {p ∈ ChX i | Tr(p(A)) ≥ 0 for all A ∈ K ∩ (SCs×s )n } I Ms := {L ∈ Mmat | supp µ ⊆ K ∩ (SCs×s )n } Theorem (B, Klep) Let L ∈ L. Then L ∈ Ms if and only if L(p) ≥ 0 for all p ∈ Ps . 14 Relation of P ∨ and M I Ps := {p ∈ ChX i | Tr(p(A)) ≥ 0 for all A ∈ K ∩ (SCs×s )n } I Ms := {L ∈ Mmat | supp µ ⊆ K ∩ (SCs×s )n } Theorem (B, Klep) Let L ∈ L. Then L ∈ Ms if and only if L(p) ≥ 0 for all p ∈ Ps . Question P= \ Ps and Mmat = s∈N Do we have P ∨ ⊆ Mmat ? [ s∈N Ms . 15 Summary of relations I ∨ =M PvN vN I Ps∨ = Ms for all s ∈ N I ∨ =M Ps∨ ⊆ P ∨ ⊆ PvN vN I Mmat ⊆ MvN ∩ {L ∈ L | rank M(L) < ∞} I Reminder I I I I I I P = {p ∈ ChX i | Tr(p(A)) ≥ 0 for all A ∈ Kmat } PvN := {p ∈ ChX i | τ (p(A)) ≥ 0 for all A ∈ K } Ps = {p ∈ ChX i | Tr(p(A)) ≥ 0 for all A ∈ K ∩ (SCs×s )n } MvN = {L ∈ L | ∃ a finite vN algebra N and A ∈ K ∩ N : L(f ) = τ (f (A)) for all f ∈ ChX i} s×s Mmat = {L ∈ L | ∃s ∈ N, measure R µ : supp µ ⊆ K ∩ (SC ) : L(f ) = Tr(f (A) dµ(A) for all f ∈ ChX i} Ms = {L ∈ Mmat | supp µ ⊆ K ∩ (SCs×s )n } 16 Back to CEP I M = {f ∈ ChX i | f = I I P U = {f ∈ ChX i | f = j hj∗ hj + P j (pj qj Pr i=1 Pni ∗ j=1 pij (1 − Xi2 )pij } − qj pj ), pj , qj ∈ ChX i} P = {f ∈ ChX i | Tr(f (A)) ≥ 0 for all A ∈ Kmat } I Kmat = {A ∈ (SCs×s )n | s ∈ N, kAi k ≤ 1, i = 1 . . . , n} I CEP is equivalent to P = (M + U) I CEP is equivalent to P ∨ = (M + U)∨ 16 Back to CEP I M = {f ∈ ChX i | f = I I P U = {f ∈ ChX i | f = j hj∗ hj + P j (pj qj Pr i=1 Pni ∗ j=1 pij (1 − Xi2 )pij } − qj pj ), pj , qj ∈ ChX i} P = {f ∈ ChX i | Tr(f (A)) ≥ 0 for all A ∈ Kmat } I Kmat = {A ∈ (SCs×s )n | s ∈ N, kAi k ≤ 1, i = 1 . . . , n} I CEP is equivalent to P = (M + U) I CEP is equivalent to P ∨ = (M + U)∨ I Strategy 1 I Find p ∈ P, L ∈ (M + U)∨ s.t. L(p) < 0 16 Back to CEP I M = {f ∈ ChX i | f = I I P U = {f ∈ ChX i | f = j hj∗ hj + P j (pj qj Pr i=1 ∗ j=1 pij (1 − Xi2 )pij } − qj pj ), pj , qj ∈ ChX i} P = {f ∈ ChX i | Tr(f (A)) ≥ 0 for all A ∈ Kmat } I Kmat = {A ∈ (SCs×s )n | s ∈ N, kAi k ≤ 1, i = 1 . . . , n} I CEP is equivalent to P = (M + U) I CEP is equivalent to P ∨ = (M + U)∨ I Strategy 1 I I Pni Find p ∈ P, L ∈ (M + U)∨ s.t. L(p) < 0 Strategy 2 I I I Show (M + U)∨ ⊆ Mmat (then CEP is true and P ∨ = Mmat .) Know, if L ∈ (M + U)∨ then L ∈ MvN If additionally rank M(L) < ∞, then L ∈ Mmat . 17 Positivstellensätze (1) I CEP is equivalent to each of the following statements 3 (Klep, Schweighofer) Let f ∈ SChX i with Tr(f (A)) ≥ 0 for all contractive A ∈ (SCs×s )n , s ∈ N. Then f ∈M +U i.e. for all ε > 0, f + ε ∈ M + U. loc , 17 Positivstellensätze (1) I CEP is equivalent to each of the following statements 3 (Klep, Schweighofer) Let f ∈ SChX i with Tr(f (A)) ≥ 0 for all contractive A ∈ (SCs×s )n , s ∈ N. Then f ∈M +U loc , i.e. for all ε > 0, f + ε ∈ M + U. 4 (Juschenko, Popovych) Let f ∈ SChX∞ i with Tr(f (A)) ≥ 0 for all A ∈ (UCs×s )n , s ∈ N. Then f ∈ Σ2 + U loc . 18 Positivstellensätze (2) I CEP is equivalent to each of the following statements 1 (Hadwin) Let f ∈ SChT i (Ti ’s are contractive self-adjoints) with Tr(f (A)) ≥ 0 for all contractive A ∈ (SCs×s )n , s ∈ N. Then f ∈ Σ2 + U w∗ . 18 Positivstellensätze (2) I CEP is equivalent to each of the following statements 1 (Hadwin) Let f ∈ SChT i (Ti ’s are contractive self-adjoints) with Tr(f (A)) ≥ 0 for all contractive A ∈ (SCs×s )n , s ∈ N. Then f ∈ Σ2 + U 2 w∗ . (Radulescu) Let P Can hX i be the ring generated by {f ∈ ChhX ii | w |fw |R deg w < ∞, ∀R > 0}. Let f ∈ SCan hX i with Tr(f (A)) ≥ 0 for all A ∈ (SCs×s )n , s ∈ N. Then f ∈ Σ2 + U w∗ . 19 General structure I All 4 statements have the same structure: Let A be a ∗-algebra with involution, let f ∈ A be symmetric and have positive trace on all finite dimensional ∗-representations of A. Then f is in some closure of a convex cone C. I A can be ChX i, ChX∞ i, Can hX i I C can be Σ2 + U or M + U 19 General structure I All 4 statements have the same structure: Let A be a ∗-algebra with involution, let f ∈ A be symmetric and have positive trace on all finite dimensional ∗-representations of A. Then f is in some closure of a convex cone C. I A can be ChX i, ChX∞ i, Can hX i I C can be Σ2 + U or M + U I All statements are proved by the same method: Proposition f ∈ closure(C) if f has positive trace on all ∗- representations into finite von Neumann algebras. I For all statements we can define a corresponding tracial moment problem as above