<<

JOURNAL OF ALGEBRA 197, 506᎐520Ž. 1997 ARTICLE NO. JA977117

Cosemisimple Bialgebras and Discrete Quantum

Eduard Vaysleb

Department of , Uni¨ersity of California, Los Angeles, California 90095-1555

Communicated by Susan Montgomery

Received December 20, 1996

We show for a cosemisimple bialgebra that a standard *-operation making it into a discrete quantum must be unique. It may not exist: we prove such an

operation on a cosemisimple OqŽŽ..SL2 ރ exists if and only if the parameter q is real. We also conclude that discrete quantum groups form a more restrictive class than cosemisimple *-Hopf algebras. ᮊ 1997 Academic Press

INTRODUCTION

Let A s Ž.A, m,1,⌬,␧ be a bialgebra over ރ, i.e., a with a m, a multiplicative identity 1, a comultiplication ⌬, and a counit ␧. It is called cosemisimple wxSw, Sect. 14; Mon, Sect. 2.4 if as a it is a direct sum of simple subcoalgebras. d Recall that on the full dual algebra M s A of any bialgebra A the multiplication is definedwx Sw, Sect. 1.1 by ²:²xy, a s x m y, ⌬Ž.a :, where x, y g M; a g A. DEFINITION 1. Let M be an over ރ. We say it is a *-algebra if there is a conjugate linear involution x ª x*Ž called a *-oper- ation. on M such that 1* s 1, Ž.xy * s y*x*, Ž.x**sx for all x, y g M. The *-operation on M s Ž.M, ) is called standard Žor a standard involu- tion. if for x g M, x*x s 0 « x s 0. 506

0021-8693r97 $25.00 Copyright ᮊ 1997 by Academic Press All rights of reproduction in any form reserved. COSEMISIMPLE BIALGEBRAS 507

Followingwx ER we define a *-operation and a co-*-operation on a bialgebra.

DEFINITION 2. Let A be a bialgebraŽ. not necessarily cosemisimple . A conjugate linear involution ) : A ª A is called a *-operation if it is an algebra antihomomorphism and a coalgebra , i.e.,

Ž.i ŽA,) .is a *-algebra; Ž.ii ␧ Ž.a* s ␧ Ž.a , ⌬ Ž.a* s ŽŽ) m )( .⌬ .Ž.a for all a g A. A conjugate linear involution q : A ª A is called a co-*-operation if it is an algebra homomorphism and a coalgebra antihomomorphism, i.e.,

Ž.i1qs1, Ž.ab qqqs ab,Žaq.qsa; Ž.ii ␧ Žaqq .s ␧ Ž.a , ⌬ Ža .s Ž␴ ( Žqmq .(⌬ .Ž.a for all a, b g A; where ␴ : A m A ª A m A is the usual flip carrying a m b to b m a. A*- is defined as a bialgebra with a *-operation which has an antipode. In the case when A is a Hopf algebra with a bijective antipode S we have the followingwx ER, Lemma 7.1 : * is a *-operation if 2 and only if qs S() is a co-*-operation, in particular, Ž.S() s id. Recall that the antipode S of a cosemisimple Hopf algebra is bijective wxLar . So, in the cosemisimple situation a *-Hopf algebra can be also defined as Hopf algebra with a co-*-operation. Given a co-*-operation on a bialgebra A we see immediately that its d Ž.conjugate linear dual defined on M s A by

²:²:x*, a s x, aq ; x g M, a g A,1Ž. turns M into a *-algebra. We say that a co-*-operation on a cosemisimple bialgebra A is standard if its dual is standard in the sense of Definition 1.

DEFINITION 3. A discrete quantum semigroup is a cosemisimple bialge- bra with a standard co-*-operation. Inwx ER a discrete quantum is defined as a cosemisimple Hopf algebra A with a standard co-*-operation. In other words, a discrete quantum group is a discrete quantum semigroup which has an antipode. A basic nonclassical example of such is a dual of a compact quantum group Žseewx ER, Sect. 10. . Let us note that finite-dimensional discrete quantum groups are exactly finite-dimensional Kac algebrasŽ studied as ‘‘ groups’’ by George Kac and his students in the 1960swx KacPal. . We will give other finite-dimensional examples of discrete quantum semigroups in the next section. 508 EDUARD VAYSLEB

There are two natural questions asked inwx Eff : Does every cosemisimple Hopf algebra have a discrete quantum group structure? Is such a structure unique? In this paper we prove a more general uniqueness result: if the co-*-op- eration on a cosemisimple *-bialgebra is standard then it must be unique up to a bialgebra isomorphism. As a special case it implies uniqueness of a standard co-*-operation for a discrete quantum groupᎏessentially the result inwx And . In the finite-dimensional case we believeŽ and all the known examples support it. that there is no difference between cosemisimple Hopf algebras and discrete quantum groupsŽ. or Kac algebras . However, if a cosemisimple Hopf algebra is infinite-dimensional a ‘‘good’’ co-*-operation which would make it into a discrete quantum group may not exist. We prove this on the example of OqŽŽ..SL2 ރ : this Hopf algebra is cosemisimple whenever q is not a root of unitywx MMNNU ; 1 however, it has *-operations only for q g ޒ j S ; and only one of these Ž.with q real is standard. Our result also shows that cosemisimple bialge- bras form a larger class than discrete quantum semigroups.

1. EXAMPLES

EXAMPLE 1wx Sw, Sect. 3.2 . Given a semigroup G with the unit 1 we define in a usual way the cocommutative bialgebra A s ރwxG . It has a linear basis Ä4g: g g G ; in this basis the multiplication is defined as the semigroup multiplication in G; and we set up the cocommutative coalge- bra structure by writing ␧Ž.g s 1, ⌬ Ž.g s g m g for all g g G. Then d ރŽ. Ł ރ MsAcan be identified with the G s g g G , i.e., the algebra of all complex functions on G; its one-dimensional matrix blocks are generated by the delta-functions ␦g . Suppose there is an antipode S : A ª A. Then for every basis element ggGwe have ⌬Ž.g s g m g and the map S must satisfy Sgg Ž.sgS Ž. g s␧Ž.g1s1. Thus if some g in G does not have an inverse Ž i.e., G is not a group. , the bialgebra A does not have an antipode, and so A is not a Hopf algebra. The natural co-*-operation on A is given by gqs g, g g G.Itis Ž. U standard as its dual on M s ރ G is defined by ␦ggs ␦ . Indeed for Ž. Ž. xsxgg,ggGwe write x* s x , g g G and so for each g g G

²: 2 x*x,gs<

In the finite-dimensional situation A is a cosemisimple bialgebra if and d only if M s A is a semisimple bialgebraŽ see, e.g.,wx Mon, Sect. 5.1. . So we can directly define a discreteŽ. or, rather, finite quantum semigroup in this case as a semisimple bialgebra with a standard *-operation as in Defini- tion 1. Ž. EXAMPLE 2. Let us take M s C [ M2 ރ ᎏobviously it is a semisimple algebra with identity. Denote the elements of M as Ž.␭, x where ␭ g ރ Ž. Ž. Ž . Ž. and x g M2 ރ . Now set E s 1, 0 , x s 0, x , and let us have 1 s 0, 1 . Then the identity is I s E q 1. Define the counit on M by ␧␭Ž.,xs␭ and the comultiplication by

⌬Ž.E s E m E, ⌬Ž.x s E m x q x m E q x m 1. Then the axioms of bialgebra are easy to checkwx V . Note that this 5-dimensional bialgebra is nontrivialŽ i.e., neither commu- tative nor cocommutative. . As we knowwx Wi, Zh there is only one 5-dimensional Hopf algebra; it is both commutative and cocommutative and is given by the cyclic group of order 5 in the sense of Example 1.Ž It is also knownwx Wi that the first nontrivial cosemisimple Hopf algebra is the 8-dimensional example of Kac and Palyutkinwx KacPal .. t The usual matrix *-operation on M given by Ž.␭, x * s Ž␭, x .is of d course standard; and we see that A s M is a discrete quantum semi- group but not a discrete quantum group since it cannot be a Hopf algebra. As an aside let us note two differences between finite-dimensional cosemisimple bialgebras with and without antipode. Ž.1 We know fromwx LR that a finite-dimensional cosemisimple Hopf algebra over ރ is always semisimple as an algebra. This happens to be false d for bialgebras: the cosemisimple bialgebra A s M from Example 2 is not a semisimple algebra. To see it consider the dual basis of A consisting of Ä4 the elements ␧, eij , where ␧ : M ª ރ is the counit of M and thus the multiplicative unit of A, and ²: ²: E,eij s0, x, eijs x ij; i, j g Ä41,2 ŽŽ.. here, as before, x g M2 ރ . The multiplication on A is given by the formulas

²:²:²:²:²E,ab s E, aE,b; x,ab s E m x q x m E q x m 1, a m b : ²: for all a, b in A. We compute E, eeij kl s0, and ²:² :²: x,eeij kl s EmxqxmEqxm1, eijm e kls x ij1, e kl and is equal to xij if k s l and to zero otherwise. We also see that 510 EDUARD VAYSLEB

Ä4 Ä 4 eeij 12 seeij 21s0 for all i, j g 1, 2 . Denoting J s span e 12, e 21 we con- 2 clude that AJ s J, J s 0. Hence J is a nilpotent left ideal of A, and A is not semisimple. Ž.2 Unlike finite-dimensional Hopf algebras Ž see, e.g.,w Sw, Corollary 5.1.6x.Ž which always have one left integral and one right integral up to a scalar multiple. , the bialgebra M from Example 2 has more than one left integral and no right integrals at all.

d The axiom for the left integral ⌳ l g A s M is

Ž⌳ llm id .Ž.⌬ F s ⌳ Ž.FI, for each F g M. For the right integral ⌳ r the definition is symmetric:

Ž.Ž.Ž.id m ⌳⌬rrFs⌳FI, FgM.

Since ⌬Ž.E s E m E and E / I we get

Ž⌳ lllm id .Ž.⌬ E s ⌳ Ž.EEs⌳ Ž.EI, Ž. Ž. therefore ⌳ lrE s 0. The same argument gives ⌳ E s 0. Now take xsŽ.0, x g M and compute

Ž.Ž.Ž.Ž⌳ llm id ⌬ x s ⌳ m id E m x q x m E q x m 1.

s⌳lllŽ.xEq⌳ Ž.x1s⌳ Ž.ŽxEq1 .s⌳ l Ž.xI.

Thus any nonzero functional on M2Ž.ރ defines a left integral. On the other hand

Ž.Ž.Ž.Žid m ⌳⌬rrxsid m ⌳ E m x q x m E q x m 1.

s⌳rrŽ.xEq⌳ Ž.1x, Ž. Ž.Ž . which should be equal to ⌳ rrxIs⌳ xEq1 . Then for x / 1 we must Ž. Ž. Ž. Ž. have ⌳ rr1 x s ⌳ x 1, therefore ⌳ rr1 s ⌳ x s 0; and so ⌳ rs 0on M. Let us note that the same phenomena could also be exhibited in the cocommutative case of Example 1 with a finite semigroup G Žsee, e.g.,w CP, Sect. 5.2x. .

2. UNIQUENESS

A cosemisimple bialgebra A can be writtenŽ see, e.g.,wx ER. as a direct sum of simple matrix . That means

A s A␭, [␭g⌳ COSEMISIMPLE BIALGEBRAS 511

Ä ␭ 4 where ⌳ is a set of indices, and each A␭ s span eij; i, j s 1, 2, . . . , n␭ with

n␭ ␭ ␭␭ ␭ ⌬Ž.eij sÝeikme kj,␧Ž.eijs␦ ij. ks1

␭ The basis elements eij of a coalgebra A␭ satisfying the relations above are called comatrix units. Any subcoalgebra of a cosemisimple bialgebra A is completely reducible Sw, Lemma 14.0.1 ᎏand so is equal to A A , where ⌳ is a wx s [␥ g ⌳ 0 ␥ 0 subset of ⌳. ␭ For a comatrix units basis Äeij4 of a cosemisimple A let us define the Ä ␭4 d dual basis Eij of M s A as usual by

² ␣ ␤: Eij,e rs s␦␦␦␣␤ ir js.

␭ Then for a fixed ␭ the elements ÄEij4 are matrix units of a simple matrix algebra M␭. The dual algebra M can be identifiedŽ seewx ER, Sect. 3. with the direct product M Ł M , where each M M Ž.ރ as an algebra. s ␭ ␭␭( n␭ Ł Suppose M s ␭g ⌳ M␭ has a standard *-operation as in Definition 1. The standardness condition easily implies that for each ␭ g ⌳ the *-oper- ation maps the simple matrix algebra M␭ onto itself. Furthermore, Kaplan- sky showedwx Kap that for a standard involution on a simple matrix algebra Ž ␭. ␭ Ž M␭ there exists a choice of matrix units for which Eij* s E ji this explains the use of the word ‘‘standard’’ in Definition 1. . Conversely, this Ž.t formula defines the usual *-operation on M␭ given by x* s x , which is standard. The following result was suggested by E. Kirschberg.

LEMMA. Suppose a simple matrix algebra M M Ž.ރ has two standard ␭ s n␭ Ä4 Ž. in¨olutions:*and †. If for the matrix units Eij of M␭ we ha¨eEij*sE ji, then there is a positi¨e matrix C g M␭ such that

† y1 x s C x*C for all x g M␭.

Proof. The map ␺ s †() is an automorphism of a simple matrix algebra M␭, so by the Skolem᎐Noether theorem it has to be inner. Thus Ž. y1 ␺xsTxTfor some invertible T g Mn. That means that for all Ž.† y1† xgM␭ we have x* s TxTor, replacing x with x* we get x s Ty1 x*T. Since †(† s idŽ. † is an involution we see that

†††111 xsŽ.xs ŽTxy*T .sTTy*xT Žy .*T Ž.2 512 EDUARD VAYSLEB

Ž. y1 for all x g M␭. If we substitute x s T* into 2 we get T* s TT*T, hence TT* s T*T, and so T must be normal. Consider its polar decompo- 1 sition T s UC s CU with unitary U and positive C. Then Ty T* s y1 y1 y2 Ž. y22 2 UCCU*sU and 2 means x s UxUfor all x g M␭.SoU i␪ 1 must be a scalar; assume it equals i g S . Suppose U itself is not a scalar. Then it has two eigenvalues "ei␪ r2, and 1 2 U␰ H ␰ for some nonzero vector ␰. We take ␨ s Cy r ␰ and compute

1 2 1 2 Ž.T␨ N ␨ s Ž.CUCy r ␰ N Cy r ␰ s Ž.U␰ N ␰ s 0.

U Ä4 † Consider P␨s P␨␭g M ᎏthe orthogonal projection onto ␨ . Then PP␨␨ y1 sTPTP␨ ␨ s0 and † cannot be standardᎏa contradiction. † 1 1 1 Thus we see U is a scalar and x s Cy Uy x*UC s Cy x*C, i.e., † is implemented by a positive C. Ž. Remark 1. The inner automorphism ␾ : M␭ ª M␭ given by ␾ x s y1r21r2 Ž. CxCdefines the new matrix units Fijs ␾ E ij on M␭. Our new standard involution † acts on them as

† y1 y1r2 1r2 Fij s C Ž.C ECij *C

y1 1r2 y1r2 y1r2 1r2 sC C ECji CsC ECji sFji.

This demonstrates that the standard *-operation on M␭ is unique up to an algebra automorphismᎏso we have given aŽ different fromwx Kap. proof of the Kaplansky result quoted in the beginning of this section.

d Let us consider a simple matrix coalgebra A␭ s M␭ and look at the dual d map ⌽ s ␾ : A␭ ª A␭ given by

²:²:²y1r21r2: x,⌽Ž.as␾ Ž.x,asCxC,a; xgM␭,agA␭.

It follows from the general theory of bialgebraswx Sw, Sect. 3 that ⌽ is a coalgebra automorphism. Now we can prove uniquenessŽ. up to a bialgebra automorphism of a discrete quantum semigroup structure for any cosemisimple bialgebra.

THEOREM 1. Let A s [␭ A␭ be a cosemisimple bialgebra. Suppose it has two standard co-*-operations q : A ª A and ( : A ª A. Then there exists a bialgebra automorphism ⌽ of A such that ⌽Žaq.Ž.s ⌽Њ a for all a in A. Ž. d Ł Proof. As in 1 , on the dual algebra A s M s ␭g ⌳ M␭ we consider the dual *-operations given by

† ²:²:²:²:x*, a s x, aq ; x , a s x, aЊ ; x g M, a g A. COSEMISIMPLE BIALGEBRAS 513

Since the two given co-*-operations are standard we have that ) and † are standard involutions on M. Therefore ) : M␭ ª M␭␭␭and † : M ª M for every ␭ g ⌳. Taking the duals again we obtain the mappings q : A␭ ª A␭ and ( : A␭␭ª A for every ␭ g ⌳. ␭ Ž␭.␭ Assume the matrix units Eij are chosen such that Eijs E ji on each M␭. From the lemma we know that the standard involution † : M ª M can † y1 U be written on each M␭ as x␭␭␭␭s C x C , where the C␭are positive matrices in M␭. Let us consider the algebra automorphisms ␾␭␭of M y1r2 1r2 defined by ␾␭: x␭␭␭␭ª C xC . d By the previous discussion, the adjoint maps ⌽␭ s ␾␭␭on A are coalgebra automorphisms. We define the direct sum coalgebra automor- phism ⌽ s [␭ ⌽␭ on the whole A. We will also use the dual algebra d automorphism ⌽ s ␾ s Ł␭ ␾␭ of M. We need to show that ⌽ is in fact a bialgebra automorphism intertwin- ing the two involutions: q and (. Ž.i Let us start by showing that ⌽ is unital, i.e., for the multiplica- tive identity 1 g A we have ⌽Ž.1 s 1. Since 1 is a group-like element Ž that is, ⌬Ž.1 s 1 m 1 . , it generates a one-dimensional subcoalgebra. So there is a fixed index 0 g ⌳ such that no s 1. Then of course M0 ( ރ, and for an Ž. arbitrary x s x␭ ␭g ⌳ g M, ²: x,1 sx0 gރ. Ž. y1r21r2 Then we see that ␾00000x s CxCsx, and so ²:²: ²: x,⌽Ž.1s␾ Žx .,1 sx0s x,1 for all x g M. Ž.ii To prove ⌽ commutes with the multiplication of A it suffices to show that for arbitrary ␣, ␤, ␥ g ⌳, a g A␣ , b g A␤␥, x g M the follow- ing holds ²:²:x,⌽Ž.ab s x, ⌽ Ž.Ž.a ⌽ b .3Ž.

Let us fix indices ␣, ␤ g ⌳. Since A␣ , A␤ are subcoalgebras of the bialgebra A, the restriction of the multiplication map m : A m A 2 a m b ª ab g A to the coalgebra A␣ m A␤ is a coalgebra homomorphism. The image AA␣ ␤ of this restriction is a finite-fimensional subcoalgebra of A, which is a cosemisimple coalgebra. Bywx Sw, Lemma 14.0.1 , AA␣ ␤ is completely reducible, and so is equal to A for some finite [␥ g ⌳ 0 ␥ ⌳ 0 ; ⌳. Thus we have a coalgebra epimorphism: m : A A A .4Ž. ␣m ␤␥ª [ ␥g⌳0 514 EDUARD VAYSLEB

Ž. Ž. Ž.Ž. Note that ⌽ A␭ s A␭ for each ␭, hence ⌽ ab , ⌽ a ⌽ b are in A for all a A , b A . In particular if x M and ␥ ⌳ [␥g⌳0␥ g ␣␤g g ␥f 0 both sides ofŽ. 3 are equal to zero. Define the algebra monomorphism ⌫ as the dual map to the coalgebra epimorphismŽ. 4 ,

²:²: ⌫Ž.x,ambsx,ab ; a g A␣, b g A␤, x g M0; it acts from M M to M M . 0 s [␥g⌳0␥␣␤m ŽWe note that our ⌫ is a finite-dimensional truncation of the ‘‘comulti- plication’’ defined inwx ER on the whole algebra M as the dual to the multiplication on A.. We can rewriteŽ. 3 as

²:Ž⌫(␾ .Ž.x,ambs²: Ž␾m␾ .Ž.⌫ Ž.x,amb;

agA␣,bgA␤,xgM0.

Now in order to prove ⌽ commutes with the multiplication we need only to show that

⌫(␾ s Ž.␾ m ␾ (⌫ Ž.5 as maps from M M to M M . 0 s [␥g⌳0␥␣␤m Ž. iii Since each A␭ is closed under the two co-*-operations q and (, Definition 2 shows that the mapŽ. 4 commutes with q and (. This implies its dual map ⌫ commutes with the dual involutions ) and †. So the map ⌫ is an injective *-homomorphism between finite-dimen- Ž. sional C*-algebras. Also it is unital, i.e., ⌫ 1 s 1 m 1, because for a g A␣ , b g A␤ we can write

²:1,ab ␧ Ž.ab ␧ Ž.Ž.a ␧ b ²:²:1,a1,b. MM0s s s ␣␤M

Denoting by n␭ the size of M␭␣we fix some *-isomorphism M m M␤( MŽ.ރ. Then ⌫ can be described as follows ER, Proposition 8.3 : there nn␣ ␤ wx is a sequence of integer multiplicities m and a unitary matrix V M Ž.ރ ␥ g nn␣ ␤ Ž. such that for every x s x␥ g M0 we have

⌫Ž.x Vy1 x[m␥ V s ž/[␥ ␥g⌳0

Žthe exponents m␥ determine the ‘‘Bratteli diagram’’ of the map ⌫ wxBr. . COSEMISIMPLE BIALGEBRAS 515

Since the mapping ⌫ commutes with both ) and †, the algebraic automorphism of M given by

y1 ␺ s †() : Ž.x␭; ␭ g ⌳ ª Ž.C␭␭␭xC;␭g⌳ must also commute with ⌫. It implies

Ž.Ž.Ž.␺ m ␺ Ž.Ž.⌫ x s ⌫ ␺ x for all x g M0. Note that on M␣␤m M the map ␺ m ␺ acts by x␣␤m y ª Žy1y1.Ž .Ž . C␣ m Cx␤␣␤␣␤myCmC. Therefore we have the equality

1 yy1 [m␥ Ž.C␣mCV␤␥␣ž/[␥xVC Ž.mC␤

[m␥ y1y1 y1y1[m␥ sVž/[␥Ž.CxC␥␥␥ VsVCž/[␥xCV␥ ,

[m␥ [m␥ where the positive matrix C s [␥␥C␥ is in [ M␥ . Ž. Since this must hold for an arbitrary x s x␥ g M0 we conclude that the matrix D VCŽ.CVy1 Cy1is in the commutant of M[m␥. s ␣ m ␤␥[␥g⌳0 In particular D commutes with C. This implies

y1 y1 y1 y1 VCŽ.␣mCV␤␣CCsCV Ž. C m CV␤C ,

Ž.y1 meaning the positive matrices VC␣ mCV␤ and C commute, and D is a positive matrix. The automorphism ␾ could be viewed as the ‘‘square root’’ of ␺. Repeating the previous argument we see thatŽ. 5 is equivalent to having VCŽ 1r21CVCr2.y1y1r2commute with all x M[m␥. But we can ␣ m ␤␥g[g⌳0␥ write

1r2 1r21r2y1y1r2 y1y11r2 VCŽ.␣mCVC␤␣s ŽVCŽ.mCVC␤ .sD , and observe that the matrix D1r2 is in the commutant of M[m␥ [␥ g ⌳ 0 ␥ since the positive matrix D is in the commutant. Ž.iv From above, ⌽ is a bialgebra automorphism of A. Recall from Remark 1 that †(␾␭ s ␾␭ () for all ␭ g ⌳. We want to show that Ž q.Ž. ⌽as⌽Њ a on A. For arbitrary x g M␭, a g A,

²:²:²:qq† x,⌽Ž.as␾␭ Ž.x,as␾␭ Ž.x,a ²:²:²: s␾␭Ž.x*,as x*, ⌽ Ž.a s x, ⌽Њ Ž.a .

Since the pairing²: , : M = A ª ރ is nondegenerate the proof is completed. 516 EDUARD VAYSLEB

Remark 2. Suppose A is a Hopf algebra with an antipode S. Recall wxSw that S is the unique map satisfying the axiom

⌺aS12Ž. a s⌺Sa Ž. 12 as␧ Ž.a1 for all a g A. It follows that every bialgebra automorphism ⌽ of A must also be a Hopf automorphism, i.e., it must commute with S.Ž This is probably ‘‘folklore,’’ though it does not appear inwx Sw or w Mon x .. Indeed we write

⌬⌽Ž.Ž.a s Ž⌽m⌽⌬ . Ž. Ž.as⌺⌽ Ža12 .m ⌽ Ža ., and so

⌺⌽Ž.aS12Ž.⌽ Ž.a s⌺SŽ⌽ Ž. Ž.a 12⌽ Ž.a s␧Ž.⌽ Ž.a1s␧ Ž.a1. Taking ⌽y1 of this equality and using the result that ⌽y1 commutes with the multiplication on A we obtain

y1 y1 ⌺a12Ž.Ž.Ž.Ž.Ž.⌽ (S(⌽ a s ⌺⌽ (S(⌽ aa 12s␧a1,

1 and conclude that ⌽y (S(⌽ s S. Thus if a bialgebra A in Theorem 1 happens to be a Hopf algebra the bialgebra isomorphism ⌽ constructed in the proof automatically becomes Hopf isomorphism. This way the uniqueness of a standard involution for a discrete quantum group immediately follows from our result.

3. EXAMPLE OF NONEXISTENCE

In this section we show that cosemisimple Hopf algebras form a wider category than discrete quantum groups. It follows that, respectively, cosemisimple bialgebras form a wider cate- gory than discrete quantum semigroups. Let A s [␭ A␭ be a cosemisimple Hopf algebra with the antipode S. d On the dual algebra M s A consider the dual to S map ␬ : M ª M, defined by

²:²:␬Ž.x,asx,Sa Ž.; xgM,agA. The map S 2is a coalgebra automorphism of A, and, bywx Lar , S 2maps 2 each A␭ onto itself. Then its dual ␬ is an algebraic automorphism of M mapping every simple matrix algebra M␭ onto itself. So by the Skolem᎐Noether theorem there exists an invertible T␭ g M␭ such that for COSEMISIMPLE BIALGEBRAS 517 all x g M␭ we have

2 y1 ␬ Ž.x s T␭ xT␭. Inwx ER, Chap. 7Ž see alsowx V. it is proved that if A is a discrete quantum group then ␬ 2 must be implementable by positive matrices. d More precisely, in this case every simple matrix subalgebra M␭ of M s A has a basis of matrix units with respect to which

2 y1 ␬ Ž.x s C␭ xC␭␭for all x g M , where C␭ is a positive matrix in M␭. Ž. Ž Ž.. Let us consider the classical by now Hopf algebra Aq sOq SL2 ރ Žsee, e.g.,wx FRT .Ž.ᎏthe coordinate algebra of the quantized SL2 ރ .Asan associative algebra AqŽ.is generated by four elements a, b, c, d. The comultiplication on AqŽ.is defined by stating that its generators

ee11 12 ab s ž/ee21 22 ž/cd constitute comatrix units of a simple matrix coalgebra A2 . The antipode is given by

1 SaŽ. Sb Ž. d yqby s . ž/ScŽ. Sd Ž. ž/yqc a The structure of AqŽ.depends on the value of a nonzero parameter qgރ. Inwx MMNNU Masuda et al. proved AqŽ.is cosemisimple if q is not a root of unity. In more detail, they showed that

AqŽ.s Al, [lgގ where each Allis a simple matrix coalgebra of size n s l. When q is real Ž.q /"1 there is a co-*-operation which turns AqŽ. into a discrete quantum group corresponding to the compact quantum group SUqŽ.2 of WoronowiczŽ the correspondence is explained inw ER, . q Ä4 Chap. 10x . It is defined on the generators as eijs e ji; i, j g 1, 2 . The Woronowicz analog of the Peter᎐Weyl theorem for SUqŽ.2w Wor, Theorem Žl. 5.7x implies the following: one can choose comatrix units eij in each simple coalgebra Al such that

Žl.Žq l. Ž.eijse ji ;i,jgÄ41,2,...,l .

This means the co-*-operation q is standard; by Theorem 1 it is the only standard co-*-operation if the real value of q is fixed. 518 EDUARD VAYSLEB

We can prove that q g ޒ is also necessary for existence of a standard co-*-operation on AqŽ..

THEOREM 2. The cosemisimple Hopf algebra OqŽŽ..SL2 ރ has a discrete quantum group structure only if q is a real numberŽ. q / 0, " 1.

Proof. Consider the simple matrix coalgebra A22of size n s 2. In the Ž. Ä4 dual simple matrix algebra M22( M ރ with the dual basis Eij take the elements

q 0 y1 1 q 0 01 00 Ks , Kys , Es , Fs . ž/0qy1 ž/0 q ž/00 ž/ 10

Ž. Ž. Note that SA22sA, and so ␬ M2s M 2. Directly using the definition of ␬ as the dual of S we compute

"1 .1 1 ␬ Ž.K s K , ␬ Ž.E syqE, ␬ Ž.F syqy F; 2 "1 "1 2 2 2 2 ␬Ž.KsK,␬ Ž.Esyq E, ␬ Ž.F syqy F.

2 Thus in the basis Ä4Eij the inner automorphism ␬ is acting on M2 as

2 1 ␬ Ž.x sKxKy .

If A has a standard co-*-operation then in some other basis the same automorphism must be implemented by a positive diagonal matrix C s Ž. diag c12, c . Multiplying by a constant, if necessary, we can assume C s Žy1. diag c, c , c ) 0. Thus there exists an invertible matrix R g M2 such that

2 1 1 1 ␬ Ž.x s RCRy xRCy Ry

Ž. for all x g M2 ރ . Therefore for some nonzero constant ␣ we have 1 K s ␣RCRy . 2 Comparing determinants we get ␣ s 1, so ␣ s "1. Let

su Rs , st y u¨ / 0. ž/¨ t

Then write KR s "RC,or

q 0 su suc 0 s" . ž/0qy1ž/ž/¨t ¨tž/0cy1 COSEMISIMPLE BIALGEBRAS 519

Then

qs qu cs cy1 u 1 1 s " 1 . ž/qy ¨qty ž/c¨ cty

1 Since q / qy by assumption we conclude that either q s "c or q s "cy1.Soqmust be real. In fact one can compute directly all the possible co-*-operations on AqŽ.Žthe list is given inwx FRT , see the explicit calculation in wx V. . In the case when q is not a root of unity OqŽŽ..SL2 ރ has exactly three co-*- operations: two for a real q and one when <

ACKNOWLEDGMENT

I thank my adviser E. Effros for help and encouragement.

REFERENCES wxAnd N. Andruskiewitsch, Compact involutions of semisimple quantum groups, Czechoslo¨ak J. Phys. 44 Ž.1994 , 963᎐972. wxBr O. Bratteli, Inductive limits of finite dimensional C*-algebras, Trans. Amer. Math. Soc. 171 Ž.1972 , 195᎐234. wxCP A. H. Clifford and G. B. Preston, ‘‘The algebraic theory of semigroups,’’ in Math. Surveys, No. 7, Vol. 1, Amer. Math. Soc., Providence, 1961. wxEff E. G. Effros, Some quantizations and reflections inspired by the Gelfand᎐Naimark theorem, Contemp. Math. 167 Ž.1994 , 99᎐113. wxER E. G. Effros and Z.-J. Ruan, Discrete quantum groups. I. The Haar measure, Internat. J. Math. 5 Ž.1994 , 681᎐723. wxFRT L. D. Faddeev, N. Yu. Reshetikhin, and L. A. Takhtajan, Quantization of Lie groups and Lie algebras, Leningrad Math. J. 1 Ž.1990 , 193᎐225.w English translationx wxKacPal G. I. Kac and V. G. Palyutkin, Finite ring groups, Trans. Moscow Math. Soc. 15 Ž.1966 , 251᎐294.wx English translation wxKap I. Kaplansky, Dual rings, Ann. of Math. 49 Ž.1948 , 689᎐701. wxLar R. G. Larson, Characters of Hopf algebras, J. Algebra 17 Ž.1971 , 352᎐368. wxLR R. G. Larson and D. E. Radford, Finite dimensional cosemisimple Hopf algebras in characteristic 0 are semisimple, J. Algebra 117 Ž.1988 , 267᎐289. wxMMNNU T. Masuda, K. Mimachi, Y. Nakagami, M. Noumi, and K. Ueno, Representa- tions of quantum groups and a q-analogue of orthogonal polynomials, C. R. Acad. Sci. Paris. Ser´ . I Math. 307 Ž.1988 , 559᎐564. 520 EDUARD VAYSLEB wxMon S. Montgomery, ‘‘Hopf algebras and their actions on rings,’’ CBMS Lecture Notes, Vol. 82, Amer. Math. Soc., Providence, 1993. wxSw M. E. Sweedler, ‘‘Hopf Algebras,’’ Benjamin, New York, 1969. wxV E. Vaysleb, ‘‘*-Operations and *-Representations of Cosemisimple Hopf Alge- bras,’’ Ph.D. thesis, UCLA, 1997. wxWi R. Williams, ‘‘Finite Dimensional Hopf Algebras,’’ Ph.D. thesis, Florida State University, 1988. wxWor S. L. Woronowicz, Compact matrix pseudogroups, Commun. Math. Phys. 111 Ž.1987 , 613᎐665. wxZh Y. Zhu, Hopf algebras of prime , Internat. Math. Res. Notices 1 Ž.1994 , 53᎐59.