ABSENCE OF CARTAN SUBALGEBRAS IN CONTINUOUS CORES OF FREE PRODUCT VON NEUMANN ALGEBRAS 4 1 0 YOSHIMICHIUEDA 2 y a Abstract. We show that the continuous core of any type III free product factor has no M Cartan subalgebra. This is a complement to previous works due to Houdayer–Ricard and Boutonnet–Houdayer–Raum. 5 1 ] 1. Introduction and Statement A O It is known, see [5, Theorem 1], that a given von Neumann algebra comes from an orbit equivalence relationif andonly if it has aCartansubalgebra(i.e., aMASA withnormalcondi- . h tionalexpectation,whosenormalizergeneratesthewholealgebra). HencethesearchforCartan t a subalgebrasinagivenvonNeumannalgebraisthoughtofasthatforhiddendynamicalsystems m producing the algebra. Therefore, it is important in view of ergodic theory to seek for Cartan [ subalgebras in a given von Neumann algebra. The aim of this short note is to complement two recent important works on free product 2 vonNeumannalgebras,due toHoudayer–Ricard[8]andBoutonnet–Houdayer–Raum[1]estab- v 9 lishing, among others, that any free product von Neumann algebra has no Cartan subalgebra. 8 The work [1] generalizes Ioana’s previous important work [9] on type II factors to arbitrary 1 4 factors. In the 90s Voiculescu [22] first proved the absence of Cartan subalgebras (or more 6 generally diffuse hyperfinite regular subalgebras) in free group factors by free entropy. In the . 1 late 00s Ozawa and Popa [12] succeeded in proving its various improved assertions by defor- 0 mation/rigidity and intertwining techniques, and a couple of years later Popa and Vaes made 4 an epoch-making work [14] in the direction. The works [8],[9],[1] (and thus this note too) were 1 done under the influence of these two breakthroughs [12],[14]. See [1, §1] for further historical : v remarks in the direction. i X LetM ,M betwonon-trivial(i.e.,6=C)vonNeumannalgebraswithseparablepreduals,and 1 2 r ϕ ,ϕ be faithful normalstates onthem, respectively. Denote by (M,ϕ)=(M ,ϕ )⋆(M ,ϕ ) a 1 2 1 1 2 2 their free product (see e.g. [19, §§2.1]) throughout this note. By [19, Theorem 4.1] the free product von Neumann algebra M admits the following general structure: M =M ⊕M with d c finite dimensional M and diffuse M such that M can explicitly be calculated with possibly d c d M = M , and moreover, such that if (dim(M ),dim(M )) 6= (2,2), then M becomes a full c 1 2 c factor of type II or III (λ 6= 0) and the T-set T(M ) does the kernel of the modular action 1 λ c t ∈ R 7→ σϕ = σϕ1 ⋆σϕ2 ∈ Aut(M) itself; otherwise M = L∞[0,1]⊗¯ M (C). This note is t t t c 2 mainly devoted to establishing the following: Theorem 1. If Mc is of type III, then its continuous core Mc = Mc⋊σϕc R (i.e., the crossed product of M by the modular action t ∈ R 7→ σϕc ∈ Aut(M )) with ϕ := ϕ↾ does never c t fc c Mc possess any Cartan subalgebra. SupportedbyGrant-in-AidforScientificResearch(C)24540214. AMSsubjectclassification: Primary: 46L54;secondary:46L10. Keywords: Cartansubalgebra,Freeproduct,TypeIIIfactor,Continuous core. 1 2 Y.UEDA Our motivations are as follows. It is well-known that if A is a Cartan subalgebra in a von Neumann algebra N, then so is A⋊ R = A⊗¯ L(R) in the continuous core N = N ⋊ R σψ σψ with a faithful normal state ψ = ψ◦E on N, where E denotes the unique normal conditional e expectation from N onto A. The work [1] actually shows only the absence of such special Cartansubalgebrasinthe continuouscore M (evenwhenboth M ,M are hyperfinite). Hence 1 2 Theorem 1 is seemingly strongerthan the originalone [1, TheoremA]. On the other hand, the f former work [8] shows the absence of generalCartan subalgebras in the continuous core of any typeIII freeAraki–Woodsfactor;thusthequestionthatTheorem1answersaffirmativelywasa 1 simpletestoftheproblem[19,§§5.4]askingwhetherornotM fallsintotheclassoffreeAraki– c Woods factors introduced by Shlyakhtenko [16] when both M ,M are hyperfinite. (See §3 for 1 2 other tests.) Next, it is known that the continuous core M is an amalgamated free product von Neumann algebra over a diffuse subalgebra (see [18, Theorem 5.1]), and amalgamated f freeproducts overdiffuse subalgebrasusuallybehavequite differently fromplainfreeproducts. Lastly,thestructuretheoryfortypeIIIfactors(see[17,Ch.XII])suggeststhatapreferableway ofstudy oftype III factorsis toregardtheircontinuouscores(ormorepreferablytheir discrete cores if possible) with canonical group actions as main objects rather than their associates. Theorem1and[20, §§2.4]altogethershowthatthecontinuous/discretecoresofany‘typeIII free product factor’ have no Cartan subalgebra. Consequently, this note completes the study ofprovingtheabsenceofCartansubalgebrasforarbitraryfreeproductvonNeumannalgebras, though what is new is the combination of technologies provided in [8],[1] with Proposition 2 provided below. 2. Proof Keep the notation in §1. As emphasized in §1 the question here is about an amalgamated free product von Neumann algebra over a diffuse subalgebra. The amalgamated free product in question arises from the inclusions Mi⋊σϕi R ⊇C1⋊σϕi R, i = 1,2, see [18, Theorem 5.1]. Hence the next simple observation, which itself is of independent interest, plays a key rˆole in our discussion below. In what follows, for a given (unital) inclusion P ⊇ Q of von Neumann algebraswedenotebyN (Q)thenormalizerofQinP,i.e.,allunitariesu∈P withuQu∗ =Q. P Proposition 2. Let N be a von Neumann algebra with separable predual and ψ be a faithful normal positive linear functional on it. Then the normalizer NNe(C1⋊σψ R) of C1⋊σψ R = C1⊗¯ L(R) in N = N ⋊σψ R sits inside Nψ ⋊σψ R = Nψ⊗¯ L(R), and hence NNe(C1⋊σψ R) is exactly the unitary group of N ⊗¯ L(R). In particular, if the centralizer N is trivial, then e ψ ψ C1⋊ R is a singular MASA in N. σψ Proof. Thediscussionbelowfollowsetheideaoftheproofof[11,Theorem2.1],butthekeyisthe so-called modular condition instead. Let ρ : R y L2(R) be the ‘right’ regular representation, i.e., ρt =λ−t, t∈R, with the usual ‘left’ regular representationλ:RyL2(R). It is standard, seee.g.[3,Theorem3.11],thatN⋊ R⊇C1⋊ Risidenticalto(N⊗¯ B(L2(R)))(σψ⊗¯Adρ,R) ⊇ σψ σψ C1⊗¯ L(R), which is conjugate to (N⊗¯ B(L2(R)))(σψ⊗¯Adv,R) ⊇C1⊗¯L∞(R) (1) by taking the Fourier transform on the second component, where L∞(R) acts on L2(R) by multiplication and the v , t∈R, are the unitary elements in L∞(R) defined to be v (s):=eits, t t s∈R. Hence it suffices to work with the inclusion (1) instead of the original inclusion. Let u ∈ (N⊗¯ B(L2(R)))(σψ⊗¯Adv,R) be a unitary element such that u(C1⊗¯ L∞(R))u∗ = C1⊗¯ L∞(R). By [4, Appendix IV] (together with a trick used in the proof of [10, Theorem 17.41] if necessary) one can choose a non-singular Borel bijection α : R → R in such a way CARTAN SUBALGEBRAS IN CONTINUOUS CORES 3 that u(1 ⊗ f)u∗ = 1 ⊗ (f ◦ α−1) = 1 ⊗ u fu∗ for every f ∈ L∞(R), where (u g)(s) = α α α [(dm◦α−1/dm)(s)]1/2g(α−1(s)), g ∈L2(R) with the Lebesgue measure m(ds)=ds. Set w := u(1⊗u∗),aunitaryelementin(N⊗¯ B(L2(R)))∩(C1⊗¯ L∞(R))′ =N⊗¯ L∞(R)by[17,Theorem α IV.5.9; Corollary IV.5.10], since L∞(R) is a MASA in B(L2(R)). Since (σψ⊗¯ Adv )(u) = u, t t for every t∈R one has w=(σψ⊗¯ id)(w)(1⊗eit((·)−α−1(·))); hence t (σψ⊗¯ id)(w)=(1⊗eit(α−1(·)−(·)))w. (2) t Since the standard Hilbert space H := L2(N) is separable (see e.g. [23, Lemma 1.8]), we can appeal to the disintegration ⊕ ⊕ H⊗¯ L2(R)= H(s)ds, N⊗¯ L∞(R)= N(s)ds (3) Z Z R R with the constant fields H(s) = H, N(s) = N (see e.g. [4, Part II, Ch. 3, §4; Corollary of ⊕ Proposition3]). Thus we can write w= w(s)ds and choose s7→w(s) as a measurable field R ofunitaryelementsinN (seee.g.[4,PartRII,Ch.2,p.183]). Bytheidentification(3)weobserve that ⊕ ⊕ (σψ⊗¯ id)(w)=(∆it⊗1)w(∆−it⊗1)= ∆itw(s)∆−itds= σψ(w(s))ds, t ψ ψ Z ψ ψ Z t R R where ∆ is the modular operator associated with ψ. Therefore, the identity (2) is translated ψ into ⊕ ⊕ σψ(w(s))ds = eit(α−1(s)−s)w(s)ds. Z t Z R R This implies, by e.g. [4, Part II, Ch. 2, §3, Corollary of Proposition 2], that there exists a co- null subset S of R so that for every s∈S one has σψ(w(s))=eit(α−1(s)−s)w(s) for all rational t numbers t and hence for all t ∈ R by continuity. Therefore, for all s ∈ S t 7→ σψ(w(s)) has t an entire extension σψ(w(s)) := eiz(α−1(s)−s)w(s) so that by the modular condition, e.g. [17, z Exercise VIII.2.2], we have ψ(1)=ψ(w(s)∗w(s))=ψ(σψ(w(s))w(s)∗)=e(s−α−1(s))ψ(w(s)w(s)∗)=e(s−α−1(s))ψ(1), i implying that α−1(s) = s and w(s) ∈ N . Thanks to [4, Part II, Ch. 3, §1, Theorem 1] we ψ conclude that ⊕ ⊕ u=w= w(s)ds∈ N (s)ds=N ⊗¯L∞(R) Z Z ψ ψ R R with the constant field N (s)=N . This immediately implies the desired assertion. (cid:3) ψ ψ Let us start proving Theorem 1. In what follows, Tr stands for the canonical trace on M = M ⋊σϕ R, which the so-called dual action scales. (See [17, Ch. XII].) We need to recall a central notation of intertwining techniques, initiated by Popa,in the present setup. Let P,Q f be (not necessarilyunital) vonNeumann subalgebrasof M suchthat Tr(1 ) is finite andTr↾ P Q still semifinite. We write P (cid:22)MfQ if there exist a non-zefro projection q ∈Q with Tr(q) finite, a natural number n, a (possibly non-unital) normal ∗-homomorphismπ from P into the n×n matrices over qQq, and a non-zero partial isometry y as a 1×n matrix over 1 Mq such that P xy = yπ(x) holds for every x ∈ P. See [8, Lemma 2.2],[1, Lemma 2.3] (due to Vaes) for its f equivalent conditions. The main, necessary ingredients from [8],[1] are in order. (I) Assume that both M ,M are hyperfinite. As pointed out in [8, §§5.2] Theorem 5.2 of 1 2 the same paper still holds true in the present setup. The proof is basically same, and finally arrives at the main argument in the proof of [7, Theorem 3.5] re-organizing several arguments 4 Y.UEDA from [12],[13]. However, one has to replace [8, Theorem A] and [8, Theorem 4.3] (with the ‘free malleable deformation’) by [15, Theorem 4.8] and the proof of [2, Theorem 4.2] (with Ioana–Peterson–Popa’s original malleable deformation), respectively. Hence the consequence becomesasfollows. Letp∈C1⋊σϕRbe a non-zeroprojectionwith Tr(p)finite. LetP ⊆pMp be a (not necessarily unital) hyperfinite von Neumann subalgebra. If P 6(cid:22)Mf Mi = Mi⋊σϕfi R (֒→ M canonically) for all i, then N f (P)′′ is hyperfinite. (This statemenft itself does not 1PM1P use the full power of the assumption that both M ,M are hyperfinite.) Then, [1, Proposition f 1 2 2.7] with the hyperfiniteness of M1,M2 gives the necessary consequence: If N1PMf1P(P)′′ has no hyperfinite direct summand, then P (cid:22)MfC1⋊σϕ R. (II) Assume that either M1 or M2 has no hyperfinite direct summand. Let p∈C1⋊σϕ R be a non-zero projection with Tr(p) finite, and let P be a unital regular hyperfinite von Neumann subalgebra in pMp. Then, [1, Proposition 2.8, Theorem 5.1, Lemma 5.2] altogether show that P (cid:22)MfC1⋊σϕ Rf. As explained in [20, §§2.1] we may and do assume M = M after cutting M by a suitable c central projection of either M or M if necessary. Moreover, when M is not hyperfinite, the 1 2 i same trick enables us to assume that M indeed has no hyperfinite direct summand. i Suppose, on the contrary, that there exists a Cartan subalgebra Q in M. Let q ∈C1⋊σϕ R be a non-zero projection with Tr(q) finite. Since Tr↾ must be semifinite (see e.g. [17, Lemma Q f V.7.11]) and Q diffuse, we may and do assume, by conjugating Q by a unitary, that q falls in Q. Remark that C := Qq is also a Cartan subalgebra in qMq (see e.g. [21, Lemma 4.1 (i)] and [8, Proposition 2.7]). Here, we have known by [19, Theorem 4.1] that M = M is a f c non-hyperfinite factor of type III, and hence it is standard, see e.g. [1, Proposition 2.8], that qMq has no hyperfinite direct summand. Applying the above (I) or (II) to p:=q and P :=C wfe have C (cid:22)Mf C1⋊σϕ R. Then a contradiction will occur once we prove the next general lemma, which holds true for arbitrary von Neumann algebras M and enables us to avoid the ‘case-by-case’proof of [8, Theorem D (1)] (see the proof of Proposition 4). Lemma 3. Let p ∈ M be a non-zero projection with Tr(p) finite, and let A be a MASA in pMp. If either the cenftralizer Mϕ is diffuse or NpMfp(A)′′ has no type I direct summand, then Af6(cid:22)MfC1⋊σϕ R. Proof. Choose a MASA D in M . It is plain to see that if M is diffuse, then so is D. By ϕ ϕ [8, Proposition 2.4] (or Proposition 2) B := D ⋊σϕ R = D⊗¯L(R) becomes a MASA in M. Suppose,onthe contrary,thatA(cid:22)MfC1⋊σϕR (⊆B). Since AandB areMASAs inpMpafnd M, respectively, [8, Proposition 2.3] with its proof ensures that there exists a non-zerofpartial ifsometry v ∈ M such that vv∗ ∈ A, v∗v ∈ B, v∗Av = Bv∗v and Avv∗ (cid:22)Mf C1⋊σϕ R. Choose a maximal,orthogonalfamily of minimal projections e ,e ,... of D (n.b., this family is empty f 1 2 whenM is diffuse). Sete :=1− e , andthenDe mustbe diffuse ore =0. Thentwo ϕ 0 k≥1 k 0 0 possibilities occur; namely v(e ⊗1P)6=0 for some k ≥1 or not. We first prove that the former k case is impossible thanks to Proposition 2, while the latter case can easily be handled thanks to [8, Proposition 5.3]. Assumethat w:=v(ek⊗1)6=0forsome k ≥1. Inthis case,NpMfp(A)′′ hasno type Idirect summand. Remarkthatek⊗1∈D⊗¯L(R)=B. Thusw∗Aw =Bw∗w⊆w∗w(M⋊σϕR)w∗w = w∗w(ekMek⋊σϕek R)w∗w, wherewe define ϕek :=ϕ↾ekMek so thatσtϕek =σtϕ↾ekMek for every t ∈ R because ek ∈ D ⊆ Mϕ. Note that Bw∗w = (Dek ⋊σϕek R)w∗w = (Cek⊗¯ L(R))w∗w = (Cek ⋊σϕek R)w∗w. By the hypothesis here Nww∗Mfww∗(Aww∗)′′ = ww∗(NpMfp(A)′′)ww∗ (see CARTAN SUBALGEBRAS IN CONTINUOUS CORES 5 e.g.[8,Proposition2.7])isnon-commutative,andhencethenormalizerof(Cek⋊σϕek R)w∗w in w∗w(ekMek⋊σϕek R)w∗w does not sit inside (Cek⋊σϕek R)w∗w itself. By e.g. [21, Lemma 4.1 (i)] one observes that (e Me ) = (Ce )′∩(e Me ) = (De )′∩(e Me ) = (De )′∩ k k ϕek k k k ϕek k k k ϕek k ekMϕek =(D′∩Mϕ)ek =Dek =Cek. Therefore, by Proposition 2, Cek⋊σϕek R is a singular MASA in ekMek⋊σϕek R, and thus by e.g. [8, Proposition 2.7] again so is (Cek⋊σϕek R)w∗w in w∗w(ekMek⋊σϕek R)w∗w, a contradiction. We then treat with the remaining case; namely v(e ⊗ 1) = 0 for all k ≥ 1, that is, k v∗v ≤ e ⊗1. This case was essentially treated in the proof of [8, Theorem D (1)]. One has 0 (De0⊗¯C1L(R))v∗v ⊆ Bv∗v = v∗Av ∼=Adv Avv∗ (cid:22)Mf C1⋊σϕ R so that (De0⊗¯ C1L(R))v∗v (cid:22)Mf C1⋊σϕ R holds. Set N := De0 ⊕e⊥0Me⊥0 of M, a diffuse von Neumann subalgebra of M. Since v∗v ≤ e ⊗1 and v∗v ∈ B = D⊗¯L(R), one easily sees that v∗v falls into N′ ∩M via 0 the canonical embedding N ⊆ M ֒→ M ⋊σϕ R = M. By [8, Proposition 5.3] we obtainfthat (De0⊗¯C1L(R))v∗v(=Nv∗v) (cid:14)MfC1⋊σϕ R, a contrfadiction. Hence we are done. (cid:3) 3. Remarks The next free product counterpart of [8, Theorem D (1)] holds thanks to Lemma 3. Proposition 4. If both M ,M are hyperfinite, M is of type III and e∈M a non-zero finite 1 2 c 1 c projection, then the normalizer of any MASA in eM e generates a hyperfinite von Neumann c f subalgebra. f Proof. WeworkinsideM,andnotethatMcisadirectsummandofM with1Mfc =1Mc⊗1L(R) ∈ Z(M). Suppose, onthefcontrary,that thfere existsa MASA C ineMfce suchthatNeMfce(C)′′ is notfhyperfinite. ThecenterofNeMfce(C)′′ sitsinsideC,andthusrepflacingewithasmallernon- zeroprojectioninC ifnecessarywemayanddoalsoassumethatN f (C)′′ hasnohyperfinite eMce direct summand. Since Mc is a factor, C1Mc ⋊σϕc R diffuse and Tr↾C1Mc⋊σϕcR semifinite, the projection e is equivalentfto a non-zero projection 1Mc ⊗e0 ∈C1Mc ⋊σϕ R; hence we may and dofurther assumethatthereexistsa projectionf =1⊗e0 ∈C1⋊σϕR suchthatTr(f)is finite and e=1Mc⊗e0 =1Mfcf. The knownfact summarizedas (I) in the proofof Theorem1 shows C (cid:22)MfC1⋊σϕ R, a contradiction due to Lemma 3. (cid:3) Proposition2 and [8, Remark 5.4] precisely show that the von Neumann subalgebras gener- atedbythequasi-normalizerandthe(groupoid)normalizerofN⋊ R⊃C1⋊ Raredifferent σψ σψ in general. Thus one may expect thatthe continuouscore ofany ‘type III free productfactor’ 1 has no regular diffuse hyperfinite von Neumann subalgebra. This is unclear at the moment of this writing, but the next weaker assertion follows directly from [8],[1]. Proposition 5. If M is of type III and e ∈ M a non-zero finite projection, then eM e has c 1 c c no regular, hyperfinite type II von Neumann algebra. 1 f f Proof. We mayandassumethatM =Mc, e∈C1⋊σϕR,andthatMi hasnohyperfinitedirect summand as long as it is not hyperfinite. If P is a regular hyperfinite type II von Neumann 1 algebra of eMe, then as in Theorem 1 one gets P (cid:22)MfC1⋊σϕ R, which is impossible. (cid:3) A free profduct counterpart of [6, Theorem 1.2] was also implicitly shown in [2] as follows. Proposition 6. If both M ,M are hyperfinite, M is of type III and e∈M a non-zero finite 1 2 c 1 c projection, then the relative commutant of any type II von Neumann subalgebra in eM e is 1 f c hyperfinite. f 6 Y.UEDA Proof. Suppose, on the contrary, that there exists a type II von Neumann subalgebra N of 1 eM e=eMe (since M is a directsummandof M) suchthatN′∩eMe is nothyperfinite. One c c canchooseanon-zeroprojectionz ∈Z(N′∩eMe)sothatP :=(Nz)′∩zMz =(N′∩eMe)z(see f f f f f e.g.[21,Lemma4.1(ii)])hasnohyperfinitedirfectsummand. By[2,Theofrem4.2]P′∩fzMz (cid:22)Mf Mi0 for some i0, but Nz ⊆P′∩zMz (cid:22)MfC1⋊σϕR is impossible; hence by [1, Propositfion2.7] Pg⊆NzMfz(P′∩zMz)′′ (cid:22)MfMi0,fa contradiction. (cid:3) f g Acknowledgement The most crucial part of this work was carried out during the author’s stay at Hokkaido University in Dec. 4–7, 2013. We thank Reiji Tomatsu for his hospitality. References [1] R. Boutonnet, C. Houdayer and S. Raum, Amalgamated type III free product factors with at most one Cartansubalgebra,Compos. Math.,150(2014), 143–174. [2] I. Chifan and C. Houdayer, Bass–Serre rigidity results in von Neumann algebras, Duke Math. J., 153 (2010), 23–54. [3] A. van Daele, Continuous Crossed Products and Type III von Neumann Algebras, London Mathematical SocietyLectureNoteSeries,31,CambridgeUniversityPress,1978. [4] J. Dixmier, Von Neumann Algebras, North-Holland Mathematical Library, 27. North-Holland Publishing Co.,1981. [5] J. Feldman and C.C. Moore, Ergodic equivalence relations, cohomology, and von Neumann algebras. II, Trans. Amer. Math. Soc.,234(1977), 325–359. [6] C. Houdayer, Structural results for free Araki–Woods factors and their continuous cores, J. Inst. Math. Jussieu,9(2010), 741–767. [7] C.Houdayer andD.Shlyakhtenko, StronglysolidII1 factors withanexotic MASA,Int. Math. Res. Not., Volume2011,1352–1380. [8] C. Houdayer and E´. Ricard, Approximation properties and absence of Cartan subalgebra for free Araki– Woods factors,Adv. Math.,228(2011), 764–802. [9] A.Ioana, Cartansubalgebrasofamalgamated freeproductII1 factors,arXiv:1207.0054v2. [10] A.Kechris,Classical Descriptive Set Theory,GTM156,Springer-Verlag,1995. [11] S. Neshveyev and E. Størmer, Ergodic theory and maximal abelian subalgebras of the hyperfinite factor, J. Funct. Anal.,195(2002), 239-261. [12] N.OzawaandS.Popa,OnaclassofII1 factorswithatmostoneCartansubalgebra,Ann. of Math.,172 (2010), 713–749. [13] N.OzawaandS.Popa,OnaclassofII1 factorswithatmostoneCartansubalgebra,II,Amer. J. Math., 132(2010), 841–866. [14] S. Popa and S. Vaes, Unique Cartan decomposition for II1 factors arising from arbitrary actions of free groups,Acta Math.,212(2014), 141–198. [15] E´. Ricard and Q. Xu, Khintchine type inequalities for reduced free products and applications, J. Reine Angeq. Math.,599(2006), 27–59. [16] D.Shlyakhtenko, Freequasi-freestates,Pacific J. Math.,177(1997), 329–368. [17] M.Takesaki,TheoryofOperatorAlgebras,I,II,III,EncyclopediaofMathematicalSciences,124,125,127, OperatorAlgebrasandNon-commutativeGeometry, 5,6,8,Springer,2002,2003,2003. [18] Y.Ueda,AmalgamatedfreeproductoverCartansubalgebra,Pacific J. Math.,191(1999), 359–392. [19] Y.Ueda,Factoriality, type classificationandfullnessforfreeproductvonNeumannalgebras, Adv. Math., 228(2011), 2647–2671. [20] Y.Ueda,DiscretecoresoftypeIIIfreeproductfactors,arXiv:1207.6838v3. [21] B.J.Woeden, NormalcyinvonNeumannalgebras,Proc. London Math. Soc.(3),27(1973), 88–100. [22] D.Voiculescu,TheanaloguesofentropyandofFisher’sinformationmeasureinfreeprobabilitytheory,III: TheabsenceofCartansubalgebras,Geom. Funct. Anal.,6(1996), 172–199. [23] S.Yamagami,Notesonoperator categories,J. Math. Soc. Japan, 59(2007), 541–555. GraduateSchoolof Mathematics,Kyushu University,Fukuoka,819-0395,Japan E-mail address: [email protected]