THE HIGHER TRANSVECTANTS ARE REDUNDANT ABDELMALEKABDESSELAMANDJAYDEEPCHIPALKATTI 8 0 0 2 n ABSTRACT. Let A,B denote generic binary forms, and letur = (A,B)r denote their a r-thtransvectant inthesenseofclassicalinvariant theory. Inthispaperweclassifyall J thequadraticsyzygiesbetweenthe ur .Asaconsequence,weshowthateachofthe 0 { } higher transvectants ur : r 2 isredundant inthe sense thatitcan becompletely 1 { ≥ } recovered fromu0 andu1. Thisresultcanbegeometrically interpreted intermsofthe ] incomplete Segre imbedding. The calculations rely upon the Cauchy exact sequence G of SL2-representations, and the notion of a 9-j symbol from the quantum theory of A angularmomentum. . h We give explicit computational examples for SL3,g2 and S5 to show that this result at haspossibleanaloguesforothercategoriesofrepresentations. m [ MathematicsSubjectClassification(2000): 13A50,22E70. Keywords: angular momentum in quantum mechanics, binary forms, Cauchy exact 1 v sequence,9-jsymbols,representations ofSL2,transvectants. 3 3 5 CONTENTS 1 1. Introduction 1 . 1 2. TheCauchyexactsequence 6 0 8 3. TheincompleteSegreimbedding 15 0 4. SL -representations 18 : 3 v 5. Thestandardrepresentationofg 21 2 i X 6. ThestandardrepresentationofS 24 d r 7. Wignersymbols 27 a References 36 1. INTRODUCTION TransvectantswereintroducedintoalgebramoreorlessindependentlybyCay- ley and Aronhold (see [11, 13]). The German school of classical invariant theo- rists used them dexterously in the symbolical treatment of algebraic forms (for instances,see [23, 41]). In theirmodernformulation,they encodethe decompo- sition of the tensor productof two finite-dimensionalSL -representationsover a 2 fieldofcharacteristiczero. 1 2 ABDESSELAMANDCHIPALKATTI We begin by giving an elementary definition of transvectants. In §1.3-1.5 we describetheirreformulationinthelanguageofSL -representations.Anoutlineof 2 themainresultsisgivenin§1.9(onpage5)aftertherequirednotationisavailable. We will use [21, 24] as standard references for classical invariant theory, and inparticularthesymboliccalculus. Modernaccountsofthissubjectmaybefound in[15,31,35]. Thereaderisreferredto[19,Lecture6],[40,Ch.3]and[42,Ch.4] forthebasictheoryofSL -representations. 2 1.1. Let m n m n A = a xm ixi, B = b xn ixi; i i 1 − 2 i i 1− 2 i=0 (cid:18) (cid:19) i=0 (cid:18) (cid:19) X X denote binary forms of orders m,n in the variables x = x ,x . (The coeffi- 1 2 { } cientsareassumedtobeinafieldofcharacteristiczero.) Letr denoteaninteger suchthat0 r min(m,n). Ther-thtransvectantofAandB isdefinedtobe ≤ ≤ thebinaryform (m r)!(n r)! r r ∂rA ∂rB (A,B) = − − ( 1)i (1) r m!n! − i ∂xr i∂xi ∂xi ∂xr i i=0 (cid:18) (cid:19) 1− 2 1 2− X oforderm+n 2r. Inparticular(A,B) istheproductofA,B,and(A,B) is 0 1 − (uptoamultiplicativefactor)theirJacobian. Byconstruction, (B,A) = ( 1)r(A,B) . (2) r r − Theprocessoftransvectioncommuteswithachangeofvariablesinthefollow- α β ingsense. Letg = denoteamatrixofindeterminates.Define γ δ (cid:18) (cid:19) m m A = a (αx +βx )m i(γx +δx )i, ′ i 1 2 − 1 2 i i=0 (cid:18) (cid:19) X andsimilarlyforB etc. Thenwehaveanidentity ′ (A,B ) = (detg)r[(A,B) ]. ′ ′ r r ′ Inclassicalterminology,(A,B) isajointcovariantofA,B. r 1.2. NowletA,B denotegenericformsofordersm,n,thatistosay,theircoef- ficients are assumed to be independentindeterminates. Write u = (A,B) for r r the r-th transvectant.1 Broadlyspeaking,the main resultof this paperis thatthe highertransvectants u : r 2 are redundantin the sense that each of them r { ≥ } canberecoveredfromtheknowledgeofu andu . Webeginwithanillustration. 0 1 1 ‘Uberschiebung’inGerman. HIGHERTRANSVECTANTS 3 Example1.1. Assumem = 5,n = 3. Thenwehaveanidentity 21 21 315 (u ,u ) + (u ,u ) + u2 = u u , (3) 8 0 0 2 16 0 1 1 256 1 0 2 which gives a formula for u in terms of u ,u . (This is an instance of general 2 0 1 formulaetobeprovedbelow.) Similarly, 20 20 25 (u ,u ) + (u ,u ) + u u = u u , (4) 0 1 2 0 2 1 1 2 0 3 3 9 14 which indirectly expresses u in terms of u ,u . Our result shows the existence 3 0 1 ofsuchformulaeingeneral. Theorem1.2. Assumem,n,r 2. Withnotationasabove,thereexistconstants ≥ c Qsuchthatwehaveanidentity i,j ∈ 1 u = c (u ,u ) . r i,j i j r i j u0 −− 0 i j<r ≤X≤ Since the right hand side depends only on u ,...,u , it follows by in- 0 r 1 { − } duction that u ,u determine the rest of the higher transvectants. In fact, more 0 1 generally we will exhibit explicit formulae for all the quadratic syzygies between the u ,ofwhich(3)and(4)arespecialcases. i { } The titleof thepapershouldnotbeunderstoodto meanthat‘highertransvec- tion’ is redundant. Notice, for instance, that the formula for u itself involves 2 (u ,u ) . 0 0 2 1.3. SL -representations. Throughoutthispaperweworkoveranarbitraryfield 2 k of characteristic zero. Let V denote a two-dimensional k-vector space with basis x = x ,x . For m 0, the symmetric power S = SymmV is the 1 2 m { } ≥ space of binary m-ics, with an action of the linearly reductive group SL(V) = ϕ End(V) : det ϕ = 1 . The S : m 0 areacompletesetofirreducible m { ∈ } { ≥ } SL(V)-representations, and any finite-dimensional representation decomposes asa directsumofirreducibles. By Schur’slemma,ifalinearmapS S is m m −→ SL(V)-equivariant,thenitisnecessarilyascalarmultiplication. Henceforth, V will not be explicitly mentioned if no confusion is likely; for in- stance,S (S )willstandforSymm(SymnV)etc. m n 1.4. Itwillbeconvenienttointroduceseveralpairsofvariables y = (y ,y ), z = (z ,z ),... 1 2 1 2 all on equal footing with x. Then, for instance, an elementof the tensor product S S can be representedas a bihomogeneousform F(x,y) of orders m,n m n ⊗ inx,yrespectively.DefineCayley’sOmegaoperator ∂2 ∂2 Ω = , xy ∂x ∂y − ∂x ∂y 1 2 2 1 4 ABDESSELAMANDCHIPALKATTI andthepolarisationoperator ∂ ∂ y∂ = y +y . x 1 2 ∂x ∂x 1 2 Ifc standsforthesymboliclinearformc x +c x ,then x 1 1 2 2 m! (y∂ )ℓcm = cm ℓcℓ. x x (m ℓ)! x− y − The operators Ω , y∂ etc. are similarly defined. The symbolic bracket (xy) xz z standsforx y x y ,andlikewisefor(xz)etc. 1 2 2 1 − 1.5. Wehaveadirectsumdecompositionofthetensorproduct min(m,n) S S S , (5) m n m+n 2r ⊗ ≃ − r=0 M usuallycalledtheClebsch-Gordandecomposition.Let π :S S S r m n m+n 2r ⊗ −→ − denotetheprojectionmap,whichactsbytherecipe F(x,y) πr f(m,n;r)[Ωr F(x,y)] ; (6) xy y x −→ → where (m r)!(n r)! f(m,n;r)= − − . m!n! Wehavewritteny xforthesubstitutionofx ,x fory ,y respectively,sothat 1 2 1 2 → therighthandsideof(6)isoforderm+n 2r inxasrequired. − InparticularifA(x) S ,B(x) S ,thenastraightforwardbinomialexpan- m n ∈ ∈ sionshowsthattheimageπ (A(x)B(y))coincideswiththetransvectant(A,B) r r asdefinedin(1). Insymbols,ifA = αm,B = βn,thenwehavetheformula x x (A,B) = (αβ)rαm rβn r. (7) r x− x− Theinitialscalingfactorin(6)issochosenthat(7)hasthesimplestpossibleform. 1.6. The map π is a splitsurjection,letı : S S S denoteits r r m+n 2r m n − −→ ⊗ section. Forcm+n 2r S ,itisgivenby x − m+n 2r ∈ − cm+n 2r ır g(m,n;r)(xy)rcm rcn r, x − x− y− −→ where m n g(m,n;r) = r r . (8) m+n r+1 (cid:0) (cid:1)r−(cid:0) (cid:1) Define (cid:0) (cid:1) (m+n 2r+1)! h(m,n;r) = f(m,n;r)g(m,n;r) = − . (9) (m+n r+1)!r! − HIGHERTRANSVECTANTS 5 Nowobservethatbytheformulaon[24,p.54], 1 Ωr [(xy)rcm rcn r] = cm+n 2r, { xy x− y− }y→x h(m,n;r) x − whichverifiesthatπ ı is theidentitymaponS (alsosee[17]and[30, r r m+n 2r ◦ − §18.2]). 1.7. Angular momenta. There is a process analogous to transvection in the quantumtheoryof angularmomentum. In brief, the eigenvectors(of the Casimir element for the Lie algebra su ) can exist in any of the states j labelled by the 2 nonnegativehalf-integers 0,1/2,1,3/2,... . The coupling of two states j ,j 1 2 { } producesafinitesetofangularmomentumstates j j , j j +1, j j +2,...,j +j . 1 2 1 2 1 2 1 2 | − | | − | | − | If we letm = 2j ,n = 2j , then thisreducesto the Clebsch-Gordandecompo- 1 2 sition. (Thestandardaccountofthis theorymaybefoundin [6, 16].) Ata crucial placeinourstudyoftransvectantsyzygieswewillneedtheconceptof9-jsymbol whicharises from the possiblecouplingsoffour angularmomentumstates. This is further explained in §7, where an introductionto the relevant notions from the quantumtheoryofangularmomentumwillbegiven. 1.8. Self-duality. ThemapS S kestablishesacanonicalisomorphism m m ⊗ −→ of S with its dual representationS = Hom(S ,k). It identifiesA S with m m∨ m ∈ m thefunctional S k, B (A,B) . m m −→ −→ Consequently,everyfinite-dimensionalSL -representationiscanonicallyisomor- 2 phic to its own dual.2 We have a canonical trace element in S S which m m ⊗ correspondstotheform(xy)m. 1.9. Results. We can now state the main results of this paper. Let the u be i { } as in §1.2. For an integer r such that 2 r min(m,n), define a (quadratic) ≤ ≤ syzygyofweightrtobeanidentity ϑ (u ,u ) = 0, ϑ Q (10) i,j i j r i j i,j −− ∈ X wherethesummationisquantifiedoverallpairs(i,j)suchthat 0 i j, i+j r. ≤ ≤ ≤ Noticethatonlyonesummandin(10)involvesu ,namelyϑ u u . r 0,r 0 r 2ThisisnolongertrueofSLN-representationswhenN >2. Insomecontextsthisself-duality leadstosimplification,andinsomeotherstoconfusion. 6 ABDESSELAMANDCHIPALKATTI Let K(m,n;r) denote the vector space of weight r syzygies. In §2.3–2.4 we will show that there is a natural isomorphism of K(m,n;r) with the space of equivariantmorphisms Hom (S , 2S 2S ). SL(V) 2(m+n r) m n − ∧ ⊗∧ ThiswillimplythatK(m,n;r)hasabasiswhichisinnaturalbijectionwiththeset ofintegralpoints r 2 Π(m,n;r) = (a,b) N2 :a+b − . { ∈ ≤ 2 } Since (a,b) = (0,0) is such as point, there exist nontrivial syzygies of any (p) weight r 2. For an arbitrary p = (a,b) Π(m,n;r), let ϑ denote the ≥ ∈ i,j correspondingsyzygycoefficients. (p) In§2.10wewillgiveanexplicitformulafortherationalnumberϑ . Itwillfollow i,j (p) thatifwespecialisetop = (0,0),thenϑ = 0. Wecanthenrewriteidentity(10) 0,r 6 intheform (p) 1 ϑ i,j u = (u ,u ) , r i j r i j u0 −ϑ(p) −− 0,r X and thereby complete the proof of Theorem 1.2. In Theorem 3.1 we prove the thematicallyrelatedresultthatthemorphism PS PS P(S S ) m n m+n m+n 2 × −→ ⊕ − whichsendsapairofforms(A,B)to(AB,(A,B) ),isanimbeddingofalgebraic 1 varieties. Of courseit wouldbe of interestto findsimilarredundancytheoremsfor other categoriesof representations. In sections 4,5 and 6, we give one example each ofthisphenomenonrespectivelyforrepresentationsofSL ,g andS . 3 2 5 2. THE CAUCHY EXACT SEQUENCE Inthissectionweestablishthebasicset-upwhichleadstothecharacterisation ofquadraticsyzygiesbetweentransvectants. 2.1. Given any two finite-dimensionalvector spaces W ,W , we have a short 1 2 exactsequence(see[4,§III.1])ofGL(W ) GL(W )-representations 1 2 × 0 2W 2W δ S (W W ) ǫ S (W ) S (W ) 0, (11) 1 2 2 1 2 2 1 2 2 −→ ∧ ⊗∧ −→ ⊗ −→ ⊗ −→ C which we may call the Cauchy exact sequence. (The correspondingformula on | {z } charactersisduetoCauchy–see[19,AppendixA].) Letthedotstandforsymmetrisedtensorproduct,i.e.,wewriteg hinsteadof · 1(g h+h g). Withthisnotation,ǫisthe‘regrouping’map 2 ⊗ ⊗ (g g ) (h h ) (g h ) (g h ), 1 2 1 2 1 1 2 2 ⊗ · ⊗ −→ · ⊗ · HIGHERTRANSVECTANTS 7 andδ isthemap (g h ) (g h ) (g g ) (h h ) (g h ) (h g ). 1 1 2 2 1 2 1 2 1 2 1 2 ∧ ⊗ ∧ −→ ⊗ · ⊗ − ⊗ · ⊗ Theexactnessof(11)isaninstanceofageneralresultaboutSchurfunctors(see loc. cit.), but it is elementary to check in this case. Indeed, it is immediate that ǫ δ = 0, implyingimδ kerǫ. Now write w = dimW , and observethat the i i ◦ ⊆ dimensionsofthefirstandthethirdvectorspaceadduptothesecond: w w w +1 w +1 w w +1 1 2 1 2 1 2 + = , 2 2 2 2 2 (cid:18) (cid:19)(cid:18) (cid:19) (cid:18) (cid:19)(cid:18) (cid:19) (cid:18) (cid:19) henceimδ = kerǫ. 2.2. ConsidertheSegreimbedding PS PS P(S S ), [(A,B)] [A B] m n m n × −→ ⊗ −→ ⊗ with image X, and ideal sheaf . Since X is projectively normal, we have an X I exactsequence 0 H0( (2)) g H0( (2)) h H0( (2)) 0. (12) X P X −→ I −→ O −→ O −→ Letusintroduceaseriesofgenericforms m n m n A= a zm kzk, B = b zn kzk, (13) k k 1 − 2 k k 1− 2 k=0 (cid:18) (cid:19) k=0 (cid:18) (cid:19) X X ofordersm,n,and m+n 2ℓ − m+n 2ℓ Uℓ = k− qk,ℓ z1m+n−2ℓ−kz2k, (14) k=0 (cid:18) (cid:19) X of orders m + n 2ℓ for 0 ℓ min(m,n). (That is to say, the a,b,q are − ≤ ≤ assumedtobesetsofdistinctindeterminates.)Considerthepolynomialalgebras Q =k[ q ], R = k[a ,...,a ;b ,...,b ]. k,ℓ 0 m 0 n { } The former is graded by N, and the latter by N N. If we write U = (A,B)z × ℓ ℓ (where the transvectant is taken with respect to z variables) and equate coeffi- cientsinz,theneachq isgivenbyapolynomialexpressionina,b. Thisdefines k,ℓ a ring morphismQ R. Now, we have isomorphismsof graded(respectively −→ bigraded)rings Q ∼ Se([Sm Sn]∨), −→ ⊗ e 0 M≥ R −∼→ Se(Sm∨)⊗Se′(Sn∨) e,e′ 0 M≥ definedasfollows: observethat ( 1)k (U ,zm+n 2ℓ kzk)z = q , − × ℓ 2 − − 1 m+n−2ℓ k,ℓ 8 ABDESSELAMANDCHIPALKATTI hence we identify q with the functional in [S S ] which sends the biform k,ℓ m n ∨ ⊗ αmβn S S to x y m n ∈ ⊗ ( 1)k ((αβ)ℓαm ℓβn ℓ,zm+n 2ℓ kzk)z . − × z − z− 2 − − 1 m+n−2ℓ This extends to give an isomorphism of Q with the symmetric algebra on the space [S S ] . The second isomorphism is defined similarly. The induced m n ∨ ⊗ map Q R on vector spaces may be naturally identified with the map h 2 2,2 −→ from(12). 2.3. Consideraformalexpression Ψ = ϑ (U ,U )z , i,j i j r i j −− i,j X where ϑ are arbitrary elements in Q. We should like to determine whether Ψ i,j correspondstoaweightr syzygy. Now,thedatumΨisequivalenttoamorphism ofSL(V)-representations f :S Q , H(z) (H(z),Ψ)z . Ψ 2(m+n−r) −→ 2 −→ 2(m+n−r) Thisistobeinterpretedasfollows: Ψ,H(z)arebothformsoforder2(m+n r) − inthez-variables.Henceaftertransvectiontherighthandsidehasnoz-variables remaining,andwegetaquadraticexpressioninthe q . k,ℓ { } Now Ψ represents a bona fide weight r syzygy iff the following condition is satisfied: if we substitute (A,B) for U , then Ψ vanishes. This is equivalent to i i the requirement that h f = 0, i.e., f factor through kerh. Hence we have Ψ Ψ ◦ provedthefollowing: Proposition 2.1. The vector space K(m,n;r) of weight r syzygies is naturally isomorphictoHom (S ,H0( (2))). (cid:3) SL(V) 2(m+n r) X − I 2.4. Now,byspecialising(11)wehavetheexactsequence 0 2S 2S δ S (S S ) ǫ S (S ) S (S ) 0. (15) m n 2 m n 2 m 2 n −→ ∧ ⊗∧ −→ ⊗ −→ ⊗ −→ C D E Byself-du|ality(s{eze§1.8})weca|nide{nztifyH}0(P(S|m Sn),{zP(2))a}ndH0( X(2)) ⊗ O O respectivelywith and ,inducinganisomorphismofH0( (2))with . X D E I C 2.5. Wehaveisomorphisms m−1 ⌊ 2 ⌋ 2S S (S ) S , m 2 m 1 2(m 1) 4a ∧ ≃ − ≃ − − a=0 M andsimilarlyfor 2S . Hence,foreachpairp = (a,b)intheset n ∧ Π(m,n;r) = (a,b) N2 : 2(a+b+1) r , (16) { ∈ ≤ } HIGHERTRANSVECTANTS 9 wehaveamorphismφ definedtobethecomposite a,b S θ1 S S θ2 S (S ) S (S ) 2(m+n r) 2(m 1) 4a 2(n 1) 4b 2 m 1 2 n 1 − −→ − − ⊗ − − −→ − ⊗ − θ3 2S 2S . m n −→ ∧ ⊗∧ Hereθ is dualto the(r 2a 2b 2)-th transvectantmap,θ is dualtothe 1 2 − − − tensorproductof2a-thand2b-thtransvectantmaps,andθ isanisomorphism. 3 By construction the φ : (a,b) Π form a basis of the space of SL(V)- a,b { ∈ } equivariantmorphisms S . Let K(a,b) denote the corresponding 2(m+n r) − −→ C weightr syzygy,writtenas κ (u ,u ) = 0, (17) i,j i j r i j −− X wherethesumisquantifiedoverpairs(i,j)suchthat0 i,j randi+j r. ≤ ≤ ≤ (We havenotyetimposedtheconditioni j.) In orderto extractthe coefficient ≤ κ ,wewillconstructasequenceofmorphisms i,j S (S S ) η1 (S S ) 2 η2 ( S ) ( S ) 2 m n m n ⊗ m+n 2i m+n 2j ⊗ −→ ⊗ −→ − ⊗ − i j M M η3 S S η4 S , m+n 2i m+n 2j 2(m+n r) −→ − ⊗ − −→ − whereη isthenaturalinclusion 1 1 v v (v v +v v ), 1 2 1 2 2 1 · −→ 2 ⊗ ⊗ η isanisomorphism,η isthetensorproductofnaturalprojections,andη isthe 2 3 4 (r i j)-thtransvectionmap. − − In §2.6 – 2.7 below, we will give precise symbolic formulae for these maps. Oncethisisdone,thefollowingpropositionisimmediate. Proposition2.2. Foranyp = (a,b) Π(m,n;r),theendomorphism ∈ η η η η δ θ θ θ :S S 4 3 2 1 3 2 1 2(m+n r) 2(m+n r) ◦ ◦ ◦ ◦ ◦ ◦ ◦ − −→ − ξ | (a{,zb) } isthemultiplicationbyκ . i,j 2.6. In order to describe θ we will realise S as the space of order 1 2(m+n r) − 2(m+n r)formsinz,andS S asthespaceofbihomogeneous 2m 2 4a 2n 2 4b − − − ⊗ − − formsoforders(2m 2 4a,2n 2 4b)inx,y respectively.Then − − − − θ :f(z) 1 −→ (xy)r 2a 2b 2 − − − [(x∂ )2m 2a+2b r(y∂ )2n+2a 2b rf(z)]. z − − z − − (2m+2n 2r)! − 10 ABDESSELAMANDCHIPALKATTI We realiseS (S ) S (S )as the spaceof quadrihomogeneousforms of 2 m 1 2 n 1 − ⊗ − orders(m 1,m 1,n 1,n 1)respectivelyinp,q,u,v,whicharesymmetric − − − − inthevariablepairsp,qandu,v. Then (pq)2a(uv)2b θ :g(x,y) 2 −→ (2m 4a 2)!(2n 4b 2)! × − − − − [(p∂ )m 2a 1(q∂ )m 2a 1(u∂ )n 2b 1(v∂ )n 2b 1g(x,y)]. x − − x − − y − − y − − 2.7. Now realise S (S S ) as the space of forms of orders (m,n,m,n) 2 m n ⊗ respectively in p,u,q,v which are symmetric with respect to the simultaneous exchange of variable pairs p q,u v. Inside this space, the image of δ ↔ ↔ consists of those forms which are antisymmetric in each of the pairs p,q and u,v. Then δ θ : h(p,q,u,v) (pq)(uv)h(p,q,u,v). 3 ◦ −→ RealisingS S as biforms in x,y, the compositemorphismη m+n 2i m+n 2j 3 − ⊗ − ◦ η η sendsQ(p,u,q,v) to 2 1 ◦ h(m,n;i)h(m,n;j)[Ωi Ωj Q(p,u,q,v)], pu qv followed by the substitutions p,u x and q,v y. The multiplier h is as → → in§1.6. Finally, η :R(x,y) 4 −→ h(m+n 2i,m+n 2j;r i j)[Ωr i jR(x,y)] . x−y− x,y z − − − − → 2.8. The h factors are introduced to ensure that if Ψ = (u ,u ) , then the i j r i j −− map(see§2.3) η η f : S S 4 1 Ψ 2(m+n r) 2(m+n r) ◦···◦ ◦ − −→ − is the identity. By contrast, the normalising factors appearing in θ are not so i crucial;their purposeis merelyto simplifysome intermediateexpressions. Their omissionwouldhavetheharmlesseffectofmultiplyingeachsyzygycoefficientby thesamefactor. 2.9. Torecapitulate,foreach(a,b) Π(m,n;r),theendomorphismofS 2(m+n r) ∈ − definedbythecomposite