ebook img

Automatic Construction of Explicit R Matrices for the One-Parameter Families of Irreducible Typical Highest Weight (0|α) Representations of U_q[gl(m|n)] PDF

65 Pages·0.6 MB·English
Save to my drive
Quick download
Download
Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.

Preview Automatic Construction of Explicit R Matrices for the One-Parameter Families of Irreducible Typical Highest Weight (0|α) Representations of U_q[gl(m|n)]

Automatic Construction of Explicit R Matrices 1 0 for the One-Parameter Families of Irreducible 0 2 Typical Highest Weight (0˙ |α˙ ) Representations m n n a of U [gl(m|n)] J q 4 ] David De Wit A RIMS, Kyoto University, JAPAN Q . February 1, 2008 h t a m Abstract [ WedetailtheautomaticconstructionofRmatricescorrespondingto(the 1 tensorproductsof)the(0˙m|α˙n)familiesofhighest-weightrepresentations v of the quantum superalgebras Uq[gl(m|n)]. These representations are ir- 7 reducible, contain a free complex parameter α, and are2mn dimensional. 2 OurRmatricesareactually (sparse)rank4tensors,containingatotalof 0 24mn components, each of which is in general an algebraic expression in 1 the two complex variables q and α. 0 1 Although the constructions are straightforward, we describe them in 0 full here, to fill a perceived gap in the literature. As the algorithms are / generally impracticable for manualcalculation, wehaveimplemented the th entire process in Mathematica;illustrating our results with Uq[gl(3|1)]. a m 1 Introduction : v i X Broadly, R matrices are solutions to the various versions of the Yang–Baxter equation, and as such, are of great interest in mathematical physics and knot r a theory (see, e.g. [18]), both in their algebraic (i.e. “universal”) forms, and in their (matrix) representations (i.e. “quantum” forms), useful for explicit computations. Here,wewillbespecificallyconcernedwithquantumRmatrices associated with the quantum superalgebras U [gl(m|n)]. q Althoughmuchis knownaboutthe originandpropertiesofquantumsuper- algebraRmatrices(e.g. [19]providesuniversalRmatrices),explicitexamplesof theirquantumRmatricesarerareintheliterature,duelargelytothecomputa- tionaleffortinvolvedinobtainingthem. Thispaperdescribestheautomationof analgorithmtogenerateasuiteofexplicitquantumRmatricesforU [gl(m|n)]. q Asreadersofthis organmaynotbe familiarwiththesealgebraicstructures,we provide a full description of their details. 1 Specifically, we construct trigonometric R matrices Rˇm,n(u) corresponding to the α-parametric highest weight minimal representations labeled (0˙ |α˙ ) of m n the U [gl(m|n)]. These irreducible representations are 2mn dimensional, and q contain free complex parameters q and α; the real variable u is a ‘spectral’ pa- rameter. QuantumRmatricesRˇm,n areimmediatelyobtainableasthe spectral limits u→∞ of the Rˇm,n(u). OurRmatricesareinfactgraded,astheyarebasedongradedvectorspaces, hence they actually satisfy graded Yang–Baxter equations. However, it is a simple matter to remove this grading and transform them into objects that satisfy the usual Yang–Baxter equations. TheconstructionshavebeenimplementedinMathematica,andresultsob- tainedforn=1andm=1,2,3,4;weillustratethealgorithmsusingU [gl(3|1)]. q Full listings of all our R matrices have been announced in [6]. As they are solutions to Yang–Baxter equations, our R matrices are of im- mediate practicalinterest. Firstly, they are of physicalinterestin that they are applicabletotheconstructionofexactlysolvablemodelsofinteractingfermions. Corresponding to Rˇm,1(u), we may construct an integrable 2m state fermionic modelonalattice. ModelsassociatedwithU [gl(2|1)]andU [gl(3|1)]havebeen q q discussedin[14]and[13],respectively. TheU [gl(4|1)]casehasanelegantinter- q pretationintermsofa2-legladdermodelforinteractingelectrons: adiscussion of this is provided in [6]. Furthermore, corresponding to each Rˇm,n, we may obtain a polynomial ‘Links–Gould’ link invariant LGm,n [21], cf. the celebrated Jones polynomial. These LGm,n are two-variable,integer-coefficientLaurent polynomials,and are generallysubstantiallymorepowerfulthantheJonespolynomialindistinguish- ingknots. (LG1,1degeneratestothewell-knownAlexander–Conwaypolynomial in the single variable q2α (cf. [2]).) A fuller documentation of the suite LGm,n hasbeenprovidedbymyselfincollaborationwithLouisKauffmanandJonLinks in [4, 5, 7, 9]. Although the LGm,n are far from being complete invariants, as they can distinguish neither mutants nor inversion[5, 9], it turns out that even LG2,1 is infact morepowerfulthanthe well-knowntwo-variableHOMFLYand Kauffman invariants, being able to distinguish (including chirality) all prime knots of up to 10 crossings [5]. Their evaluation also involves automatic sym- boliccomputation,butthecomputationalaspectsarecomparativelypedestrian. Lastly, we mention explicitly that this paper contains no new theorems, although it does contain two new technical lemmas, proven in Appendix A. It is primarily intended to provide a proper foundation for the results presented in [6, 7], although it also serves as a tutorial on an application of symbolic computation. Whilst it specifically pertains to representations of U [gl(m|n)], q many of the algorithms have a much broader application. The following subsections provide a synopsis of the paper. 2 1.1 Algebraic overview Fixing m and n, we are initially interested in a 2mn dimensional vector space V that is a module for the U [gl(m|n)] minimal typical highest weight repre- q sentation Λ = (0˙ |α˙ ). The algebra contains a free complex variable q, whilst m n the representation π acting on V contains a free complex variable α. Our Λ V is actually (Z ) graded; this ensures compatibility with the (Z ) grading of 2 2 U [gl(m|n)]. q Using the properties of U [gl(m|n)], we apply a version of the Kac induced q module construction (KIMC) [16, 17] to establish a (weight) basis {|ii}2mn for i=1 V. This involves postulating |1i as a highest weight vector, and recursively acting on |1i with all possible distinct products of simple lowering generators Ea+1 todefinetheotherbasisvectors,normalisingaswego. Thisconstruction a requires a ‘Poincar´e–Birkhoff–Witt (PBW) lemma’ for U [gl(m|n)] [8, 28], i.e. q aset ofcommutationssufficientto transformany productofalgebragenerators into a normal form (see §4.2), together with a statement that the algebra is spanned by the set of all such normal forms. Where V has a graded weight basis {|ii}2mn, the tensor product module i=1 V ⊗ V has a natural 22mn dimensional basis {|ii ⊗ |ji}2mn , which inherits i,j=1 a weight system and a grading from V. For our particular representation, the orthogonaldecompositionofV ⊗V isknown[11],andcontainsnomultiplicities, viz: V ⊗V = V , k k M where the submodule V has highest weight λ , and these λ are known, and k k k all distinct. To build R matrices acting on V ⊗V, we require an alternative, orthonormal weight basis B= B for V ⊗V, corresponding to this decom- k k position, viz B is a basis for V . Again using the KIMC, the basis vectors k k of each B are derived as lineaSr combinations of the form θ (|ii⊗|ji), where k ij the coefficients θ are algebraic expressions in q and α. This process initially ij yieldsabasisB thatis notnecessarilyorthonormal,sowealsoapplyaGram– k Schmidt process to orthonormalise B into B . The desired R matrix is then k k a weighted sum of projectors onto these V , where the weights are eigenvalues k of the appropriate second order Casimir invariants. The algebraicstructureof U [gl(m|n)]is detailedin§2, andanintroduction q toitshighestweightrepresentationsisprovidedin§3. In§4,weprovideanormal orderingandaPBWlemmaforU [gl(m|n)]. Theconstructionofourparticular q (0˙ |α˙ ) representations is detailed in §5. §6 describes the construction of the m n bases B , and §7 describes the construction of projectors and R matrices. k 3 1.2 Implementation and results Explicit computations within the representation theory of quantum superalge- bras are tedious and error-prone when performed manually. The dimensions of representations are generally large, and in our case, we have the presence of the two variables q and α; these generally manifest themselves in complicated rationalalgebraicexpressions,whose symmetries must be continually identified and exploited to avoid the arising of intractable messes of algebra. The construction of the basis {|ii}2mn involves many applications of the i=1 PBW lemma to simplify long strings of algebra generators. This is compu- tationally expensive; firstly as the simplification involves a minimally-efficient sorting process, and second as it involves a geometric explosion in the number of terms being sorted. The construction of the weight space bases B is nontrivial, as each basis k vectorofeachB generallycontainsmanytermsoftheformθ (|ii⊗|ji),where k ij the coefficients θ are generally complicated rationalalgebraic expressions in q ij and α. (That said, we have avoided the more theoretically difficult situation of computing weight space decompositions in cases where there are weight multi- plicitiesintheunderlyingcarrierspaceV.) Althoughthe Rmatriceshave24mn components, Nature is kind to us in that most of these components are zero, and those that are not are generally simpler than the θ . ij To the best of our knowledge, computer implementation of the algebraic structures and algorithms described herein has not previously been achieved. We haveimplementedthe entireprocessasasuite of Mathematica functions; the thousands of lines of code perform algebraic computations that a human being could not ever realistically expect to perform correctly. From §2.3 onwards, we use U [gl(3|1)] to illustrate our results. These are q summarised in Appendix B; where we list the explicit matrix elements for the generators of the underlying 8 dimensional representation, orthonormal bases B for the 4 submodules V ⊂V ⊗V, the components of the associated 4 pro- k k jectors P onto the V , and finally, the trigonometric and quantum R matrices, k k Rˇ3,1(u) and Rˇ3,1, respectively. Whilsttherearenotheoreticallimitstomandn,acurrentpracticallimitfor computation is mn 6 4. This is convenient, as an immediate application [6] of the materialcritically requires Rˇ4,1(u). Although translationof the interpreted Mathematica code into a compiled language would increase the speed of the computations, storage requirements would still limit mn to perhaps 7 in the general case. Further discussion of implementational issues and results is provided in §8. 4 2 The quantum superalgebras U [gl(m|n)] q The algebraic structures labeled U [gl(m|n)] are quantum superalgebras,1 de- q scribed in many places, e.g. [1, 10, 24, 25, 26], and in the book [3, see §6.5]. For our purposes, m and n are positive integers, to be regarded as fixed, and q is to be regarded as a nonzero complex variable. As U [gl(m|n)] may be q unfamiliar to the readers of this organ, in §2.1 we introduce its phylogeny, and in §2.2, we provide a full description of its structure in terms of generatorsand relations. Beyondthat, in§2.3wedescribeits rootsystem,andin§2.4weshow how it may be regardedas a Hopf (super)algebra. 2.1 The phylogeny of U [gl(m|n)] q 1. Wherenisapositiveinteger,recallthattheLiealgebra gl(n)isequivalent totheusual(complex)vectorspaceofn×n(complex)matricesaugmented by a ‘vector multiplication’ operation which is the usual matrix multipli- cation. gl(n) is of course a unital algebra, and is of dimension n2 and rank n−1. The n2 generators {ea }n of gl(n) satisfy a commutation b a,b=1 relation: [ea ,ec ]=δcea −δaec , b d b d d b where[·,·]is the usualcommutator (bracket), definedforX,Y ∈gl(n)by: [X,Y],XY −YX. 2. Letting both m and n be positive integers, the Lie superalgebra gl(m|n) may be obtained from gl(m+n) by retaining the generators {ea }m+n, b a,b=1 butmodifyingthe definitionofthecommutatorbracketandcommutation relations to include some ‘parity factors’ of ±1. Specifically, we have the commutation relation: [ea ,ec ]=δcea −(−)[eab][ecd]δaec , (1) b d b d d b where [·,·] is now the graded commutator (bracket), defined for homoge- neous (see below) X,Y ∈gl(m|n) by: [X,Y] , XY −(−)[X][Y]YX, (2) and extended by linearity. In both (1) and (2), [X]∈ {0,1} refers to the grading ofthehomogeneouselementX. Forthisreason,Liesuperalgebras are sometimes called “gradedLie algebras”. From the gl(m+n) case, we see that gl(m|n) is of dimension (m+n)2 and rank m+n−1. 1These structures are sometimes called “quantum supergroups”, but they are actually (associative,noncommutative) algebras. 5 3. U[gl(m|n)] is then the usual universal enveloping algebra obtained from gl(m|n)byregardingthegl(m|n)generatorsasletters inwordscontained inU[gl(m|n)],wherethe(graded)commutatorbracketactsasarelationto reducethealgebrasomewhat. U[gl(m|n)]isinfinitedimensional,although of finite rank, viz, again m+n−1. 4. Thequantumsuperalgebra U [gl(m|n)]isthenaso-called‘q-deformation’2 q ofU[gl(m|n)],whichmaintainsitsviabilityasaHopf(super)algebrastruc- ture (see below) [12]. Roughly speaking, the deformation amounts to ‘exponentiation by q’; indeed U[gl(m|n)] may be recovered as the limit q → 1 of U [gl(m|n)]. U [gl(m|n)] is of course also infinite dimensional, q q and again of rank m+n−1. 2.2 Generators and relations for U [gl(m|n)] q Following Zhang [28, pp1237-1238],we providea full descriptionof U [gl(m|n)] q intermsofgeneratorsandrelations. ForvariousinvertibleX,wewillrepeatedly use the notation X ,X−1. 2.2.1 U [gl(m|n)] generators q Where I , {1,...,m +n} is the set of the gl(m|n) indices, we define a Z 2 grading [·]:I →Z : 2 0 if a6m even indices [a], 1 else odd indices. (cid:26) Throughout, we shall use dummy indices a,b ∈ I where meaningful. A set of (m+n)2 generators for U [gl(m|n)] is then: q K , m+n Cartan a Eb , a<b 1(m+n)(m+n−1) lowering . (3)  a 2  Eab, a<b 21(m+n)(m+n−1) raising  Let usnow introduce the notation, for any a∈I:  q ,q(−)[a], a For any power N, replacing q with qN immediately shows that (q )N =(qN) , a a so we may write qN with impunity, specifically, we will write q ≡ q−1. Next, a a a an equivalent notation for Ka is qaEaa; where the exponential is defined in the usual manner as an infinite sum, thus powers KN are meaningful; specifically, a we will often be working with N ∈ 1Z. Thus, under the mapping q 7→q, K is 2 a mapped to K , where we intend K ≡K−1. As expected, for arbitrary powers a a a M,N, we have: KMKN =KM+N where K0 ≡Id, a a a a where Id is the U [gl(m|n)] identity element. Apart from N ∈ N, powers (i.e. q products) of the non-Cartan generators (Ea )N for a6=b, are not meaningful. b 2Wemightsay‘quantumdeformation’here,6buttherelationtoquantummechanicsismore ofanalogythanofrigor. Onthe generatorswe define anaturalZ gradingintermsofthe gradingon 2 the indices: [KN],0, [Ea ],[a]+[b] (mod2), (4) a b wheretheformermaybeseenasaspecialcaseofthelatterbysettinga=band making the identification Ka ≡q(−)[a]Eaa. We use the terms “even” and “odd” forgeneratorsinthesamemanneraswedoforindices. ElementsofU [gl(m|n)] q are said to be homogeneous if they are linear combinations of generators of the same grading. The product XY of homogeneous X,Y ∈ U [gl(m|n)] has q grading: [XY],[X]+[Y] (mod2). (5) Thus, for example, inspection of (4) and (5) shows that we may cheerfully substitute [Ea ] for [Ea Eb ]. Further, we also have the following useful results c b c for a<b<c<d: [Ea ][Eb ]=[Eb ] and [Ea ][Eb ]=[Ea ][Ec ]=0. d c c b c b d 2.2.2 U [gl(m|n)] simple generators q The full set of generators (3) includes some redundancy; some can be regarded as simple in that the rest may be expressedin terms of them. We shall call the following subset of 3(m+n)−2 generators the U [gl(m|n)] simple generators: q K , m+n Cartan a Ea+1 , a<m+n m+n−1 simple lowering . a   Ea , a<m+n m+n−1 simple raising  a+1  The fact that there are m +n−1 simple lowering generators indicates that U [gl(m|n)] has rank m + n − 1. Note that there are only two odd simple q generators: Em+1 (lowering) and Em (raising). m m+1 2.2.3 U [gl(m|n)] nonsimple generators q In the gl(m|n) case, the remaining nonsimple (non-Cartan) generators satisfy the same commutation relations as the simple generators. The situation is verydifferentforU [gl(m|n)];the nonsimple generatorsdo not satisfy the same q commutationrelationsasdothesimplegenerators. Instead,theyarerecursively definedintermsofsumsofproductsofthesimplegenerators(see[27,p1971,(3)] and [28, p1238,(2)]). Strictly speaking, they are not explicitly required for the definitionofthealgebra;theirusecanhelpsimplifyotherwiselargeexpressions. Towhit,asetofU [gl(m|n)]nonsimplegeneratorsmaybedefinedrecursively q for a<b by: (a) Eb , Eb Ec −q Ec Eb nonsimple lowering a c a c a c (6) (b) Ea , Ea Ec −q Ec Ea nonsimple raising, b c b c b c (cid:27) where a<c<b; viz c is an arbitrary index, we do not intend a sum here. 7 In §5, we will have use for an alternative set of nonsimple generators,again defined recursively for a<b by: (a) Eb , Eb Ec −q Ec Eb alternative nonsimple lowering a c a c a c (7) (b) Ea , Ea Ec −q Ec Ea alternative nonsimple raising, b c b c b c (cid:27) where we intend Ea , Ea when Ea is simple, viz, for any |a−b|= 1. Note b b b that we use a boldface Ea where the original source [27] uses an overline Ea ; b b the use of the boldface notation saves the overline for indicating inverses. These definitions may be written more concisely with some more notation. Writing Sa ,sign(a−b), we may replace (6) and (7), for all a6=b, by: b (a) Ea = Ea Ec −qSbaEc Ea b c b c b c (8) (b) Eab = EacEcb−qScbaEcbEac. ) The two different sets of generators are in fact Hermitian conjugates. For all meaningful indices a,b, we have: (Ea )† =Eb , (Ea )† =Eb , (KN)† =KN, b a b a a a and these definitions ensure that (X†)† = X for all U [gl(m|n)] generators X. q Notethattheseareordinary,notZ gradedHermitianconjugates,meaningthat 2 we have (XY)† =Y†X†; expressly not (XY)† =(−)[X][Y]Y†X†. Lastly, we mention a result of Zhang [27, Lemma 3], which gives us a more efficient formula than (8b) for expanding the alternative nonsimple generators: Ea =Ea +Sa ∆ Ec Ea , (9) b b b c b c c X for any indices a 6= b, where the sum is over all c strictly between a and b. (If |a−b|=1, then the sum is ignored, and the result is trivial.) Note that in (9), we have introduced the following handy notation: ∆ , q−q, ∆ ,q −q =(−)[a](q−q)=(−)[a]∆ a a a (10) ∆ , (∆)−1 ∆a ,(∆a)−1. ) 2.2.4 The graded commutator The graded commutator [·,·] : U [gl(m|n)]×U [gl(m|n)] → U [gl(m|n)], is de- q q q fined for homogeneous X,Y ∈U [gl(m|n)] by (2), viz: q [X,Y],XY −(−)[X][Y]YX, andextendedbylinearity. Forcompleteness,wementionthatforassociative su- peralgebras,ofwhichU [gl(m|n)]iscertainlyanexample,wehavethefollowing q useful graded commutator identities: (a) [XY,Z] = X[Y,Z]+(−)[Y][Z][X,Z]Y (11) (b) [X,YZ] = [X,Y]Z+(−)[X][Y]Y[X,Z]. ) 8 2.2.5 U [gl(m|n)] relations q With this notation, we have the following U [gl(m|n)] relations: q 1. The Cartan generators all commute; for any powers M,N: KMKN =KNKM. (12) a b b a 2. The Cartan generators commute with the simple raising and lowering generators in the following manner: K Eb K =q(δba−δba±1)Eb . (13) a b±1 a a b±1 From (13), we have the following useful interchange: K Eb =q(δba−δba±1)Eb K . (14) a b±1 a b±1 a In Lemma 2 (proved in Appendix A.1), we show that (14) may be much strengthened to: KNEb =qN(δba−δca)Eb KN, (15) a c a c a for any meaningful indices b,c (viz b<c, b>c, and even b=c), and any power N. 3. The non-Cartan simple generators satisfy the following commutation re- lations (this is the really interesting part!): K K −K K [Ea ,Eb+1 ]=δa a a+1 a a+1. (16) a+1 b b q −q a a Alternatively, again employing the notation of (10), we may write this: [Ea ,Eb+1 ]=δa(−)[a] (−)[a]Ea −(−)[a+1]Ea+1 , (17) a+1 b b a a+1 q h i where we have introduced the q-bracket, defined for various invertible X ∈U [gl(m|n)], including scalars (well, scalar multiples of Id): q qX −qX [X] , , observe that lim[X] =X. (18) q q−q q→1 q Note that in (17), Zhang [27] replaces (−)[a]Ea −(−)[a+1]Ea+1 with a a+1 the more convenient expression h . a 9 In fact, (16) generalises to a more useful result (proven in [8]): K K −K K [Ea ,Eb ]= a b a b =∆ (K K −K K ), (19) b a a a b a b q −q a a viz: [Ea ,Eb ]=(−)[a] (−)[a]Ea −(−)[b]Eb . b a a b q h i We also have, for |a−b|>1, the commutations: Ea+1 Eb+1 =Eb+1 Ea+1 and Ea Eb =Eb Ea . (20) a b b a a+1 b+1 b+1 a+1 4. The squares of the odd simple generators are zero: (Em )2 =(Em+1 )2 =0. m+1 m In fact, we may show that this implies that the squares of nonsimple odd generators are also zero: (Ea )2 =0, [a]6=[b]. (21) b 5. Lastly, we have the U [gl(m|n)] Serre relations; their inclusion ensures q that the algebra is reduced enough to be simple. For a6=m: (Ea+1a)2Ea±1+a1±1−∇Ea+1aEa±1+a1±1Ea+1a+Ea±1+a1±1(Ea+1a)2 =0 (Eaa+1)2Ea±a1±1+1−∇Eaa+1Ea±a1±1+1Eaa+1+Ea±a1±1+1(Eaa+1)2 =0. where to save space, we have introduced the notation: ∇, q+q. These may be more succinctly expressed using nonsimple generators. Noting that for a6=m, we have q =q , the above become, again for a6=m: a a+1 (a) Ea+1 Ea+2 = q Ea+2 Ea+1 a a a a a (b) Ea Ea = q Ea Ea a+1 a+2 a a+2 a+1  (22) ((dc)) EEaa+−11a−1EEaa+1a == qqaEEaa+1aEEaa−+11a−1.  a+1 a+1 a a+1 a+1  10

See more

The list of books you might like

Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.