Scribd Logo
Upload

Welcome to Scribd!

  • 14 views

    (University of Minnesota Technical Report 82-101) Peter J. Olver - Differential Hyperforms I (-)

    Uploaded by

    abel

    Copyright:

    © All Rights Reserved
    Download as PDF or read online from Scribd
    Flag for inappropriate content

    You might also like

    (University of Minnesota Technical Report 82-101) Peter J. Olver - Differential Hyperforms I (-)

    Uploaded by

    abel
    14 views118 pages

    Original Title

    (University of Minnesota Technical Report 82-101) Peter J. Olver - Differential Hyperforms I (_)

    Copyright

    © © All Rights Reserved

    Available Formats

    PDF or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    Report this Document

    Copyright:

    © All Rights Reserved
    Download as PDF or read online from Scribd
    Flag for inappropriate content
    Download as pdf
    14 views118 pages

    (University of Minnesota Technical Report 82-101) Peter J. Olver - Differential Hyperforms I (-)

    Uploaded by

    abel

    Copyright:

    © All Rights Reserved
    Download as PDF or read online from Scribd
    Flag for inappropriate content
    Download as pdf
    of 118
    University of Minnesota Mathematics Report &2-101 Differential Hyperforms I. Peter J. Olver™ School of Mathematics University of Minnesota Minneapolis, MN 55455 Abstract, A theory of higher order differential forms, called differential hyperforms, based un the theory of Schur functors, is constructed over Euclidean space. Generalizations of the deRham complex lead to the notions of a hypercomplex and hypercohonology theories based on higher order derivatives. Applications include the systematic derivation of higher order divergence identities for hyperJacobians and a wide variety of interesting higher order Pfaffian systems with integrability criteria. New, explicit formulae in the algebraic theory of Schur functors are also presented. * jeheseargh supported in part by National Science Foundation Grant NSF mcs 81-00786. 1. Introduction. 6 ee eee 2. Shapes and tableaux... es. eee ee 3. Schur functors. 2 ee eee eee 4. Functorial maps... +e ee eee 5. The Pleri maps. oe ee ee ee ee 6. Polarization maps. se ee ee ee ee ee T. Hypercomplexes. eee eee ee ee ee eT 8. The Schur hypercomplex. 6 eee ee ee ee ee MB 9. Duality and interior products...) ee ee ee + 88 10. The Littlewood-Richardson Rule... ee ee ee eB lL. Algebraic differentials... +... eee. ee eee 12, Products and differentials for fat shapes... 2... 1 oe Th 13. The differential hypercomplex. 6... 2. ee eee ee ee UW. Leib: S Rules. ee ee ee eee 15. Hyperjacobian identities... 2... eee eee eee + 92 16. Exactness and integrability... 2... ...-..5 roduet ion This is the first * two papers in which a new theory of higher order differential forms, or "hyperforms," is introduced, based on the recently developed and, in my view, fundamentally important theory of Schur functors. This paper is devoted to the construction of hyperforms over Euclidean space, but with no consideration of their behavior under changes of coor- dinates. The sequel will do the more difficult constructions over arbitrary smooth manifolis. Although a regrettably extensive amount of algebraic machi ery must be developed before these hyperforms can be properly considered, I hope that the intrinsic beauty of the identities which can be found, as well as the range of potential applications in both di rential geometry and partial differential equations will make the reader's effort in assimilating the material worthwhile. When I first learned about differential forms and their applications in differential geometry, especially deRham's theorem, [26], I was always struck by the fact that these constructions only involved first order der- ivat: jes - either of the coefficients of the forms, or the functions giving changes of coordinates. Indeed, the only places in which higher order deri- vatives make their appearance in differential geometry are a) in jet bundles and overdetermined systems of differential equations, [14], [26], b) the more or less eq alent, but far less sophist ated theory of extensors, ef. [21] and references therein, c) the theory of higher order frame bundles or tangent vectors, eg. [16], [28, § 1.26]. Although extremely interesting and useful, none of the above mentioned theories has any of the flavor of differential forms with differentials, exact sequences and cohomology in any truly higher order sense. I propose that the theory of differential hyperforms fills this long neglected gap, and provides a correct setting for the systematic development of a differential geometry based on higher order derivatives with many potential applications. In particular, higher order cohcmology, characteristic classes, etc., are among the ideas that remain to be developed, all of which must be dealt with in later investigations. Be that es it may, the origi motivation for my development of this theory came from a question raised in the study of variational problems of interest in nonlinear elasticity, [3], on nat homogencous differential polynomials could be written as higher order divergences, the first non- trivial example being the curvature identity 2a pt _ pe uf) # DpyCayy,) ~ Dec w), a nie the D's demoting derivatives. This problem was solved in [23]; the most general such polynomial is a linear combination, of polynomials T called hyperjacobians as they were higher order analogues of the classical. Jacobian determinants - see section 15. Tt was also of interest to actually construct e explicit identities for these hyperjacobians, and to do this a some what unsophisticated version of differential hyperforms was proposed. At the sane time as [3], [23] were being written the concept of a Schur functor, or shape functor having been introduced by Towber, [27], and Lascoux, [17], was being developed by Akin, Buchsbaum and Weyman, (1], [2], [30], for the purpose of resolving certain determinantal ideals. Although results.in [3], [23] required detailed Imowleage of certain pro- perties of these determinatal ideals I was unaware of the existence of Schur functors until T heard David Buchsbaum lecture in Minnesota in January, 1981. I realized that my primitive algebraic constructions of hyperforms in [23] were just @ special case of a mich grander theory of Schur functors, thereby considerably gaining in power and range of applicability. This paper is the first fruit of thie marriage of the powerful new algebraic tenc! ues to Aifferential geometric ideas. It is, in my opinion, just the beginning of the application of these fundamentally important functors to a wide range of geometric, topological and even physical problems. ‘The basic idea behind differential hyperforms is ad follows. On the P Euclidean space M-= RP , corresponding to each shape (Young diagram) 0 is a hyperfon bundle ™, obtained by applying the Schur functor I, point- the cotangent bundle TM ; smooth sections of =, are called ) ~ nyperfome. For each shape yO there fs « eifverontial dt taking A -nyperforms to y-hyperforms, so that the coefficients are alfterentiatea |u/A] times. the aitferentials df commute in the obvious sense, ana, moreover di = 0 if u/A has to or more boxes in any colum. For aetk q single column, =o At, and af cor werk! 4s tne ordinary differential. Thus the deRhan complex forms a small part of the much larger differential nypercomplex foruea by the hype In this paper, since a only involves |y/%| order differentiations, ‘orm bundles and differentials. and no lower order, it cannot be invariant under changes of coordinates, which is why only Euclidean spaces are considered here. ‘The differential hypercomplex is also exact in a certain sens simplest manifestation of exactness occurs for shapes ACUCY with w/ and v/y each consisting of a single row of boxes, in consecutive ross, and which overlap in precisely one columns for instance %=(3,1) , (3,3) » v= (3,3,2) + Then a given u-hyperform 7] satisfies age (1-2) for some A-hyperform £ if and only if y aqtao- (1-3) » Meco x g v/ul = (Note a Qs= a 8-0 forall §.) rf lu/al =, |[v/nl =2 , then (1.2) forms a large system of k-th order nonhomogeneous linear constant coefficient partial differential equations in the coefficient functions of & , end (1.2) constitute the full set of £-th order integrability conditions for their solution. By suitable choice of 4,4, v one can find systems of any desired order Kk wi tegrability conditions of some other predetermined order &. Often these systems are nontrivial - see section 16. For more general AC , the system (1.2) will still be of order |u/A| , but the integrability conditions will consist of several systems of the form (1.3) corresponding to different v,>y with possibly different orders |v,/ul . Again, any desired orders can be found by a suitable choice of 1, y - The exactness also includes syst: ans of mixed order with integrability conditions of various orders. These systems can be viewed as true, nontrivial higher order generalizat ns of the classical Pfaff systems, [12], [25]. They could be valuable as specific, nontrivial examples for further investigations into the Spencer-Gold- schmidt theory of overdetermined systems of differential equations, [Us], [26]. For manifolds, the lack of exactness of the hypercomplex will introduce new types of cohomology, but we ill defer this to the second paper. (A similar type of hypercomplex has been introduced by Delong, [7], in his study of Killing tensors. It is of great interest to study the connection between our hyperforms and his differentials, which require a Riemannian connection.) The Schur functors required for these constructions only involve real or complex vector spaces. This restriction to characteristic zero throughout leads to several simplifications in the underlying Schur functor theory, the most notable being the lack of distinction between the symmetric and divided power algebras and hence between Schur and co-Schur functors, ef. [2] - Actually, our defi ions are a slightly hybridized version of co-Schur functors, but for simplicity the extraneous co- has been dropped. For a vector space V, the Schur s 1, V have been around for a long time; they are just the irreducible representation modules of the general linear group GL(V) , [4], [20], [29]. Towber, [27], uses a second definition in terms of tensor powers of the symmetric algebra of V modulo an ideal of relations. Yet a th rd definitior used by Akin, Buch+ sbaum and Weyman, [2], is as the image of certa: maps on these tensor powers. Each defin. on has its own particular advantage - the first, in conjuction with functoriality and Schur's lemma reduces otherwise impossible identities to the computation of a single constant; the second is best for specific computational examples; the advantage in the development of the underlying theory. All three are used at one point or another here, and I see no reason to prefer one over the other. Although a substantial amount of vitel theoretical groundwork has ven laid by Lascoux, [17], and Akin, Bucksbaum and Weyman, [2], I have, for the most part, found their constructions of maps between gcwur spaces far too unwieldy to work with when it comes to specific computations, many of which are vital in the development of the theory. This is especially true in the so-called Pieri formula for decomp sing certain tensor products. Sections 5 and 6 redo this important result with explicit, easily eam putable expressions for the relevant maps which have not appeared so far. (This redevelopment is one expl: nation for the somewhat inordinate length of this paper.) In essence, the Pieri formula provides a product between avector veV anda byperform w€L,V to give 2 hyperform v* aL, V for any w2A with |y/A|=1. This Pieri product lends to an algebraic version of the differential hypercomplex, which we call the Schur hyper- complex, and which has similar properties of commbativity, closure and exactness - see Secti 8. It is also possible to define products between hyperforms, but here uns into the problem that, according to the Littlewood-Richardson rule, the Schur space IV may occur with multiplicity greater than one in the tensor product f4¥ @ 1, V, and hence there may be more then one type of invariantly defined product between a A-hyperform and a p- hyperform giving a v-hyperform. For this reason, except for one special case, we have not investigated the complete “shape algebra" in any great detail. If, however, Aju,v are all of the form ntr for fixed n, O0} . Given a, p=a+Jj denotes the diagram obtained by adding a single box to the j-th rov of a, so 8, =a, rts, By=G,+1. In the case of shapes, the relation w=A+J presures that the resulting diagram is actually a shape, eA, Dh ya hs sar "g (i,k) positive, let 6 =a/(j,k+) be the diagram obtained by deleting J>1. Given all boxes in the j-th row of q in all colums beyond (and including) the k th colum, so ainfa, 2). for i¢3,8 The diagrams, and hence the shapes, are partially ordered so aGp Af @, , w= (4,3,2,2) Dual shape: ’ X= (3,210) Hook hy, in (filled in boxes) 4,308) =5 Tableau of shape 2% S = (2,2,3,4} aah: f Standard tableau of snape 1, came 8 - fe & Simple and canonical : E eobienus of sbepe 4 uhh feb] 1 Z simple canonical Figure 1. Given a shape 4 , define {Afacw , u/% has no colum with two or nore boxes} lu) = Bu) ~ 82) ayu) = su) ng,in) ts) = {vlnce} , ey eS e 1 = (vluey, [v/ul = 1, a £ 1 {vlucy , v/u has no colwm with vo or nore boxes} Bu) ~ ay) , a = maximal if no other shape contains 44 the set of maximal shapes is denoted S°(u,v) . Similarly, given 2 = [27,...,.29} c 34 PQx) = Fw) ns2a4yn... Ashape vED4(A,u) is minimal if no other shape in J°(,h) is containea in v; 3°(4,n) denotes the set of all minimal shapes. Note that for acs*(u) , 8°(u,3°(S.n)) =D but, in general, for yo3"(y) 3°08 (uy) w) FE However, equality will hold if v, =u, for all v&y. The sets Rokay will be called mi: A= 8%u,y) and y= 3° (Ryu) If A= {a} or y={v} consists of a single shape, then we will drop the bold-face and write 8'(y,v) , ete. In particular, shapes A,u,v are min-max related if and only if »/2 and v/u both are single rowed, lying in consecutive rows and overlepping in a single column; in other words, Rew/(sryrt) swe v/(d+), for come i>0, j>0- Example 2.1 Let 4 = (3,2,1) . If v= (353.2) , then d= 3°%u,v) = £(2,2,2) , (3,252) -13- and 44, are min-max related. If y={(3,3,151) , (3,2,2,1)} , then = 8%u.y) = (3,2) 5 (22,2) and again },u,y are min-max related. fema 2.2 Let 3,u,2 be mincmex related. If p €3'(u) with pZv for all v€y, then there exists 2€A with pest(h} - Proof If p/w consists only of boxes in the first row, then any X will do. Otherwise 8'(x,yUfp}) ¢ S\(u,y) ds nonempty, so let A, be an element. Let 2€} contain 4, . rf p #3) , then r€8%(n,yule]) , so 2 cannot be w,y-maxinal. Q.E.D. Tema 2.3 Let A,i.y be min-max relat ‘Then y Bweu souU rH), (2.2) vEL vey ana 43) TAN Tv) =f. For XE, vey. (2.2) Proof The secondsformula is obvious since if p>v for vey then pe€d(X) for all AEX. To prove (2.1), if 9 ET (u) , then either pes'(v) for som vey (p can't ve in Tv) for any v), or pv forall vey, so, by Lema 2.2, 9 ETA) for seme REX - Sometimes the exponential notation my Ae kt ee > i > Ry > s+ >, > 0 wil be used for a shape with m, rows of length k, , My rows of length k, , ete. A shape is n-bounded if 25 mn) and & is the difference product -16- Byer) = MG =)» ip , th =0-) Theoren 2.5 Let |a| =n. The number of standard tableau of shape X with entries in {1,...jn) , each entry occuring precisely once, is > (2.5) Me where 3+ Sch wr functors In this section the basic theory of Schur functors in characteristic zero is outlined. Proofs of the statements can be found in the basic references [2], [17], [27]. let V be a fixed (real or complex) veetor space, The tensor, symmetric and exterior algebras of V are denoted by @, = VY , = OV, A= AV respectively. Fach of these ic a graded algebra, 80 @,V =, 0 8,V, ete. Given a graded vector space W=@W, , and an element c@€W” , define the tensor product ay = Way OH, Por + For a shape } , define the map. Br @ 0 + Op ey composing the tensor product of diagonal maps with the tonsor products of projections % J i remembering that the various copies of V occurring in the intermediate tensor product are labelled according to their positions in the diagram A. The sel ace 14 =I,V is then defined to be the image of 8 - This is the definition of Schur functor favored by Lascoux, [17], Nielsen, (22] and Akin, Buchsbaum and Weyman, [1], [2], [30]. -18- Adifferent construction was proposed by Towber, [27]. Let J denote the two-sided ideal in @,0, generated by relations of the form pel Eo xpe@ (x, Oz) = 0, GBD “1 called Young symmetries, for xy5++-4%,,, €V, 260, ,V, and pda, where y= ¥,0 04 OH) O OX,» Then ker 8, = A= INO, hence the Schur space % % / oh is the quotient space. For computational purposes it is useful to know the following consequential relstions, ef. [27]. Theorem U1 Let x= x40 ++. Ox, €0,,¥E9,,2€ 0, with p>atr. Then the relations x @ (yz) = (-1)9 B (x, 9 y) @ (x, 0 2) (3-2) are in the ideal g. In (3-2) the sum is over all multi-indices I,J with TUS =(2,..,p),1Nd=f and a] =a, s0 ox ete. The Schur space I, can be identified with the representation space of the irreducible representation p, of the general linear group GL(V) corresponding to the shape 4. This interpretation, which can be inferred from the standard works on representation theory, e.g. ion of L, in terms % (4], [20], [29], leads to yet another def: Young symmetrizers. If |A| =n, let 1, be the canonical tableau of shape X. The Young syrmetrizer for the chape 2 is, ef. (2.5), E sen(x)m 5 3-3) ‘the cum being over ali permutations x and p of {1,...n] with Preserving the columns and p the rows of 7, ef. [hs page 101], [295 page 12k]. This induces a map A eVray, where the permutations 1,9 in g, permute the various copies of V in the tensor product. With the above normali ation, 0, is a projection: we imo, 21, » In fact, 0, clearly restricts to a projecti 2% % 71, for wh we use the sane notation, A decomposable clement of @0, is of the form l a n Br GLP OHO. COO... O20), where each 5 €V, and can thus be identified with a tableau of shape A with values in V , the entries being the 5 . This in turn defines @ decomposable element of 1, , which we also write as x , from now on omitting the maps 8, or @, when there is NO poi ibility of confusion. a £ V has basis (eys.++5@,] 5 then for each tal In particular, au of shape 4 with values in {1,...,p} , there is a corresponding decomposable element 2, of @0, , and hence 1, , obtained by substituting the vector e, for the integer i wherever it occurs in 2. (For the standard tableau of Figure 1, e, = @. ( a - row standard t: y= (©, 9,08, 0e,) @ (e,0e,) @ a, -) The row standard ta 3°83 provide a basis ¢, of @,0, , but these elements are no longer independent aux T when viewed as elements of I, . The following theorem, [2], [8] provides a basis for the Schur space ly - em 3.2 Tet dimv=p, and [. sje.) be a basis. esoty Dea basis. Then the elements e, for T standard tableaux of shape 2 form a basis A). for efore the dimension of is NP, as given in for I, . Therefore the dimension of L,V MPs as g n (2 Given a nonstandard tableau T , it is useful to know how to rewrite the element e, elements of Iy 5 this is knowm as a straightening 1 Rearranging the symmetric products in ¢, we may assume that T is at least row standard. Let the last non-standard entry of T occur at position (i,j) . In other words, | t. pat ty, 4, i, or i and j'> Jj. ‘hen use theorem 3.1 on the i-lst and i-th rows of e, T fron the (i-l)st row of Ty the product of the first J entries of x being the symmetri produ 21 the i-th row and 2 any remaining entries i instance, for the nonstandard tableau of figure 1, 50 we re-express e, as a Lineer combination of elements corres- ponding to the tableaux fe BEy 2]2]3]" FEEL FE 2 » & » bf > pile IE A the first and last with coefficients +1 , the other two with coef- ficients +2 . (For tableau with repeated entries, this step must sometimes be wodified slightly by rearranging the equation if the original tableau appears on both sides of the identity.) Of the resulting tableaux, those which are not standard are again subjected to this procedure. The entire process i guarenteed to terminate in a finite number of steps, t each stage the lexicographic order, obtained by writing out the entries of the tableau in reverse o: i.e. right to left, bottom to top, increases! See Doubillet - Rota - stein, (8), or DeConeini - Eisenbud - Procesi, [6], for details. Finally, we remark that if 2 consists of one row, then Ly is the symmetric power , . If A= 1" consists of one column, then I is Just the wedge power A, More generally, the same methods can be used to define Schur functors L /,V for skew shapes u/h, cf. [2]. Our only interest in these move general functors is in the following decomposition theorem for direct sums. 22+ Theorem 3-3 Suppose V=W@Z. Then L (a) nv=eL) (yz) , . AY on, W,2) (3-5) wnere each u)(W,2) has a natural , with associated graded vector space L(y) ~ 9 Wek AZ, (3-6) rES Cu) but the direct isomorphism is not natural. ‘The proof of this theorem can be inferred from [2] or [17]- We are primarily interested in the special case dim z= 1 , when 4 1 Oe rw EB (u) , fo} ne Bu) - Since each symmetric power 0,% is also one-dimensional, we get the following corollary Corollary 3.4 Suppose V=W®Z, and dimZ=1. Then =o 0k) Lv =o Lw (3-7) where cach L(")w is filtered, with associated graded vector space is ea eS aad AW (3-8) 1, resi (u) corollery will be of key importance in proving an algebraic version of the Poincaré lema for hypercomplexes. -23- The category @ is that of finite dimensional (real or complex) vector spaces, the morphisms being linear maps between vector spaces. A funet ial map, or natural, ransfo} ion, between two functors # and § on @ is a linear map go WW for V any vector space, which comuutes with the morphisms: if ADVoOW is linear then 9° 5A f tare (2) For a given pair of functors .,% , the set of all functorial maps ot 5-H forms a vector space itself under the obvious operations. (see {aK].) of particular interest are functors 3 consisting of combinations of tensor, symmetric or wedge powers, or, more generally, Schur functors. Any such algebraic functor applied to V, 3V , decomposes into a direct sum of irreducible representations of GL(V) , i.e. a direct sum of Schur spaces IV - This decomposition will be "functorial” : independent of the particular vector space V provided dim V {s ‘sufficiently large, sono I,V is zero. In particular, the multiplicities of the various 1,V in Sv are independent of the particular vector space V+ Schur's lemma then proves the following basic result. =2hke Theoren b.1 Let 3 be a combination of tensor, symmetric, wedge Powers, or Schur functors. Then the vector space of functo 1 maps : F+L, for any given Schur functor I, has dimension equal to the sultiplicity of I, in J, te, the miltiplicity of Iv in sv for any Vowith av fo. These maltiplicities are only mown in general for fairly simple B's , e-g- the tensor product of two Schur functors, see section 10- More general functors lead into the difficult area of plethysm of which only Limited results are thus far known, [15], [18}, [30], [32]. (ala except the third are written in terms of Schur functions). 5. For a specific instance of the previous theorem, we consider the Pieri Formula, [17], [30] oO, @L, @ L (5-2) 1 Oy 7 . wexz(a) decomposing the tensor product of a symmetric power with a Schur functor. Combining with projections onto the various summands we obtain functorial maps Se POL hs (5.2) As a corollary of (5.1) and theorem .1 we The Pieri maps are uniquely determined up to a constant mltiple (it dependent of any particular vector space). It is helpful to have explicit formulae for the Pieri maps, but, for simplicity, we restrict attention to the case L=1. Let 2 be fixed, and let w= A+K be the shape obtained by adding a single box to the k-th row of 1. The Pieri map bh -y i VOW h is then constructed as follows: First consider the linear map @0,~0 a p20 © SG given by t(x@y) = E xg@(y Ox, for x= ¥,0++. 0x, deconpositie. The (1,4) transposition ry J is the linear map on @,0, which is the identity on all factors except the ith and jth, where it agrees with +: 7: 0 @ Q, (1f i> the oraer of the factors is reversed), If A; Note that im7,,C@, 0, where XI, is the diagram obtained from a3 ys Om ig A by taking one box from the i-th row and adding onto the j-th row. (Ay; 4s not necessarily a shape.) Multiple transpositions corres- ling to J = (§yo-+-od. defined b; gto T= (Jyrsesydy) are defined by 7 teat 6-4) ven k dg? > where p=#J can range from 2 to m-k+l. Let br VO8,% “8, 1% 78% OV be the interchange map r(v@a) =wev, and identify the latter space with @, 0, , where A, Finally define rm - oh: V8 e0, ~ 8,0, by b= Eot nyse, (5.5) he sum being over att se, . me coettictente are 2g > 6-6) a pe=#J, ice. the product of the hook lengths of all hooks whose initial box is in row k , and whose terminal box is in a roy indexed by J (except for mi). P Theorem 5.2 Let por+k. If gf is as defined in (5.6), x ‘then Ave) oy: Therefore g induces the Pieri map a - oh: VOL, -% (the tvo “aifterent" meanings of of should not cause any confusion.) Before proving this theorem, it is helpful to con F some examples. Example 5+ Suppose A= (p,q) is & two-rowed shape. For w= (ptl,q) , ohlv @ (x@y)) = (xe) ey + (xoy@(y, eM) 5-7 for x€a,, 0...0y €0 ,vEV. The verification of xEO, Yo ¥y EF % > ¥ theorem 5.2 is fairly routine in this example. Even simpler are the cases 4 = (p,qtl) or uw = (p,G,2} , where b x@(y Ov) # = (B,a+1) ahlv@ (xey)) = x@y@y » = (2,9,2) In fact, if p= A+k , then & only affects the i-th rows of x €@,0, for idk. For X= (p,a,r) , u = (prl,a,r) , we have the more complicated formula G(v@(x@ye2)) = (x0v) 8y@z 4+ 2 2x Oy.) O(y, Ov) Ge + p-a + SS Blo. 024) @y 8 (29 0¥) + wusery JF r9ug) 0(y, 929) O02, 00) The general formula (5-5) can now be fathomed. Jemma 5.h Let ifj,kft- For C6@%, G,-aj¢ Ursjtig lO = | ng (0) ° otherwise . -29- (Here [f,gl=f*g-gef for linear fg The proof is trivial. Proof of theoren 5.2 Mote firot that the relations (3.2) defining the ideal 4, span the image of in @,0, , for 1 Theil = ° 6.8) for all such 2. There are three cases depending on the relative magnitudes of k and 2. ide = ices in i it i O Consider the case k=£. Indices in ak, split into tw classes: T= (6K) JY = (02,8) for some m>44+2. Then by lemma 5-4 , m where K€ AT) +t, ‘nk? "ke, kt k "mk? whereas C J oa QO, FMT e a? yin! = Te Mata * Teta jko Tk ket! ‘The latter coefficient is just the (k,kt1) hook length of 2 , hence using (5-6) , “1 at Ty? Tye) 2 O -30- Summing over K proves (5.8). Next consider the case k<%. Now the indices in Dan split into three classes: J = (I,2,K) , J” = (If41,4) , (T2,2h1,K) « ko, nm 3 for TE Ay, Jt41. Again by lemma 5.4, yt aa) (yet aaa) =o > Te where "hi I° ‘Therefore since as can easily be verified. The remaining case % reversing the direction of the Pi rimaps. (See also (31]-) In the special case that 1, (and hence 1, ) are symmetric pavers, these agree with the classical partial polarizations of synmetri polynomials, hence the x will be called polarization maps. It is easily shom that x is uniguely determined up to a constant multiple, and, moreover, when compos: 4 with gf gives @ multiple of the identity map 41 on L, Weynan vases the cansteaction of the 38 on the samosition of where A is a tensor product of diagonal maps A,., ~V@A, on the factors Pp 6 aps Nyy 5 ‘a corresponding to columns with boxes in p/A , followed by symmetric multi- plication on the £ extra copies of V. Weedless to say, this form- ulation is not of great help when one is interested in readily computable rmulae. hoe Again, for simplicity we restrict atten ion to the case p= Atk y first define te 9, -¥849, by 33+ x = Baye, > (6.2) the sum being over all J € Ay, or xt Fa, the copy of V in the image ng given row index 0. The modified constants are os pel e+e yo Csla) = Ts te) -D) - (6.2) Note that Ly -l= je) 2 = hy Theorem 6.1 Let p=\tk. With x defined as in (6.1) we have Ruevess hence %{ induces the partial polarization a. % The proof is very s. wre. lar to that of theorem 5.2, so we leave it to the reader and content ourselves with a couple of easy examples. Example 5.2 Let (p,q) be a two rowed shape. If p>q, sO A= (p-1,q) shape, thea a Pp xe @y)= E x, @x ey » isl * i for 9...0x,€0 ,y=y, 0... Oy, € 0, whereas for TEL pe pV EM Yq Sq (e, a Z oy, @x®y - E x,O(xp Oy) Oye. oO, gen 2. Og 9 ¥5) Oyg h, xyleen) = ¥ isl i id Similarly, if w=(p,a,r) , 4=(p,a,r-1) , Bn @xeyez - x * & a x,(x@¥@2) Fe EMO, 04) Oyen, k i = e 5 , Berry 7% (3,019) 0,02) 22, + Then Lo (6.3) where Ay alu) -2 on = (ytn-%) aa a (6.4) p=#d, then Tom 1 Ty0) = CDP ac + & for some Bea. For JEAQ, fo =p, KE at 5 ; ° dgy++odq) € AQ + Then tte rl) for some § Ef, . Substituting (5-5), (6.2) into (6.3), ve iunediately see that Ke ECP y cay 7a), ane 16gF) Sy et rep k _ the sum being over all J€ A’ andell KEAL, , p= #K. The proof of (6.K) therefore reduces to the following series of easy combinatorial lemmas. emma 6.4 The following formulae hold: i) Gyr? (6.5) 4,6(4) -2 . it) E (-1)Fo4(2) no ay > (6.6) KE at k) Closure The resulting maps a are identically zero whenever v/h has two or more boxes in any colin, i.e. X €8°(v) in the notation of section 2. -38- Within any hypercomplex, there are a number of interesting "sub- complexes". First we generalize the ni ion of a complex as used in homological algebra, [5]. Definition 7.2 An n-complex is given by a set of vector spaces Wy indexed by nonnegative integers j , and linear maps subject to the condition that the ntl -fold iteration Phen See TES EWS Waa vanishes identically for all j . A complex in the ordinary sense is thus a l-complex. Of course, any n-complex is also an nil-conplex. To exclude this triviel case, the term n-complex will usually imply that n-fold iterations fyuney ' or1 7 fy ae not all identically zero. Within any hypercomplex, there are distinguished n-complexes- Nanely, consider only the n-fat shapes, ordered by their weight, end the maps fg PMA TH, for Ay nofat, |u/X[=2. The closure condition on the hyperconplex implies that these are indeed n-complexes. There are also a large number of ordinary complexes. For any shape hy define shapes yu), with uo=a,ur=ari, vl iss, = a2. d, adi, ‘Then it is easy to see that 39" form a complex - fy, £; = 0 ~ called the \ - subcomplex of the given ce hypercomplex. For example, if X= (3,1) , then the \~subcomplex is fo_, a > H —, f » FAA » f 1—> o EP ~ We now turn to the definition of exactness for a hypercomplex. where we have suppressed the W's « Definition 7.3 A hypercomplex is exact tor every shape uf 0, every 26 8'(u) , veo Uy) with A,u,y min-max related (see section 2) fo ker f= 5 img. (7.2) vey # axey * Note: the sum is not necessarily direct. The motivation for this definition comes from the exactness of the Schur hypercomplex, to be discussed in the following section. Note that (1-1), by closure, the left hand side always contains the right hand side. ‘Thus we can define the cohomology of a hypercomplex; Definition 7. Given @ hypercomplex, and A,p,y min-max related, the Aju,y - cohomology is hay vey rer / vey nh: We will not pursue the investigation of this cohomology theory here in any detail. To prove exactness of a hypercomplex, it actually suffices to check exactness of the h- subcomplexes, and in fact, only special cases ital simplification in the exactness proof's Theorem 7.5 Given a hypercomplex fh Wy“, 5 suppose that for all min-max related shapes A,u,v with [v/u|=1 (or, alternatively, with [uy /.|=1) ker % = ind, (7-2) Then the entire hypercomplex is exact. Proo! We do the case |v/y|=1, the other case being similar. Let Au.¥ be min-max related. Note that no shape in y contains any other shape therein. Let v* be the smallest shape containing all the shapes in y. Let m(v* /y)>1 be the last row in which vt contains boxes not in H, and let k= vE-, be the number of boxes in this last row. (Note xr #1 for the min-max relation to hold. The proof proceeds by induction, first on k, then ©. For the induction on *,if k>2, let * . For each v'€y , let o° be the shape obtained by deleting the box (rjc) from v’ , if it occurs. The g* are all different. Let g=({o'}, so rlo*/y) =r ot -w = k-1- Let p = 8°(u,0) , and note that p,u,d, are min-max related and there is a one-to-one correspondence between 449 € 2 and ele. with plant, Consider the sequence of maps okie are direct sums of f’ respectively, F,(@ x,) = pi, ot ‘J i HEX Oy) x, €W FA) 89 () EW . By the induction ja 345753 8 bh hypothesis, ker an(F + Fy) ker in(F, * Fy) 8°(o*,v4) < 8%u,0), as can easily be checked. Therefore ner(fy + Fy) = Sa Father F, Fot in(P ° But this is precisely (7.1) for }uu,y - To do the induction step fron r-1 to r we need an easy "diagram - chasing" lems: Lema 7.6 Consider a commutative diagram of linear maps ROOF www * al B BYE, aba with the top and bottom rows exact. Then the conditions i) (P, * Fy) = ker Fy+ker 4) ker(F, +P) = ker Fy+ker Fy 43) am(n, + Fy) = im BLM im Fg are equivalent. In particular, if either a) F, is aurjective, or b) is injective, 5 then both i) and ii) hold. Now suppose v* has a single box in the last row r>2not in p, in colum veewpth. Let x=ner,aee/(r-lct+) so {ol = s°(u,x) . (In this proof, o,8,¥,9 will all denote shapes.) Given vVoey, set oF =W/(r,c+) 3p, co abe pd. tet 2 3%ug) ,g= fo}. note that R= pUlal. tet x) = Cri} = 3%, por oF fv) , set pla /(r-1,04) , so 8%o%,v4) = [p9} . Note that pl>a. For r>2, set {65} = 8%p%,0%) , so 6% aieeors fron p! in the (r-2)nd vow. Finally, if pi’ fa, ret yi = Cyl} = B°(a,p!) . (See Figure 2 for illustrational purposes.) Note that each 1! aiters from ¢% in at most one row, labelled It is not too difficult to show that if j is such that ol fv) ana pi) fa, then y! can be constructed from 7) by setting PET (2-104) , rye She Jk =r-l, 43 tnd ignoring any TK with ry =. tet f= (rin sea Consider the commutative diagram, where the F, are constructed J as in the induction step on k: (See figure 2 for an representative example.) The only nontrivial question of comutativity corresponds to 77* with on no corresponding y"* exists. But 7° €5°(5%) since the x and r , since r-dst rovs in 19*/g) overlap, 50 that part of Fg °F, vanishes by closure. By the induction hypothesis, or (7.2), each short sequence consisting of two consecutive maps in the same row or column is exact, e.g. ker FQ = imF 9 8 (for FyyFy » note that sf of = vi, pl, {o}oW , is injective. For 3 o FoF if Baa, Fy g 7 CO) Ae surgeetive.) oo Figure 2 0 r=5 eal A = 1(3,3,2,2), (4525252), (453,251), (453,204 # = (4,3,2,1) vos (4,4,3,2,1) mn = (4,3,2,2,1) o (454,352) 2 £(3,3,2,2) 5 (4,2,2,1), (4,3,2,2)} r= Cb 452) 5 (4546353) 5 445352,2)) o (4,352) 6 (44,3) Bobb), (44,353) F e (sh, 2) (the W's have been omitted leaving only the relevant shapes in the dlagran.) 45 - I claim that for each j the subdiagram yl satisfies condition i) of lemma 7.6, i.e. Jo pd ker FZ ° F3 ly kor od ker P+ ker FS (7-3) we claw , or a= jal since FI; 7 : ial sin FZ: Lo] ® se is injective. Otherwise, note that 8°(69,3°) = Cao}, parts of 3d conrespondine os ase fe, hence contribute nothing to the above kerne: Therefore, by the induction hypothesis ker(FA * +inFl, so by exactness, the claim is proven. Now use lemma 7.6 on boxes B and A in the large diagram. Since we have just shown condition i) holds for box A , by summing (7.3) over condition ii) holds for box B. A second application of the lemma, ‘his time to boxes B end C shows that condition i) holds for box C¢ - But this is just Jj, -46- ker(Fy, ¢ Fg) = ker Fy + ker Fyo + im Fy 5 7 by exactness, which is precisely what we need to establish. completes the induction step on r , and hence the proof of the theorem. One point to note in the definition of exactness for hypercomplexes is that there is no compatible notion solely for n-complexes. For instance, u = (3,3,1) , v = (3,3,2) are t th j-fat, but in an exact hypercomplex ker £¥ = im i" x where A= (3,2,1) is no longer fat. This indicates that one must consider the full hypercomplex, and not just n-subcomplexes, to yet any meaningful cohonological results. ‘The notions of hypercomplex, etc. can be dualize: Definition 7.7. A cohypercomplex is given by spaces W¥ indexed x by shapes % , and Linear maps do gh oy Gy By defined whenever pA, |ul = [Al +2, subject to conditions of a) Comutativity: The diagram of maps is commutative, which Llows us to unarbiguously define x doe Bt Wy ow. for any v4 by iteration. -u7- b) Closure: gh = 0 whenever 4 € 8°(v) Note that the a » for u,X n-fet, again form an n-complex. Definition 7.8 A cohypercomplex is exact if for every J,y,y min-max related, » Z img =f ker a vey YoREL There is an analogous theorem to theorem 7.5 for checking exactness of cohypercomplex, which, for brevity, we do not, state or prove explicitly. Similarly one can define the homology of a cohypercomplex. Finally we note that for finite - dimensional vector spaces, the notions of hypercomplex and cohypercomplex are precisely dual to each 7:9 Tf each W, in a hypercomplex is a finite - dimensional vector space, then the dual spaces WM, and dual maps gt = a cohypercomplex, and vice-versa. Exactness of one of these implies (s)" tom ss of the other. -48- 8. The Schur Hypercomplex. The Pieri maps of theorem 5.2 will now be used to a ne an algebraic hypercomplex between the Schur spaces of a vector space V. For each vVE€V the Pieri map lve -) defines a linear function from Ly to b, for any HO, [ul = |X[+1, uniquely determined up to constant multiple by functoriality. We now fix this constant. Definition 8.1 Let Ati. The roduet, api V@L— 3b, is defined by = vaca ved ,vev, Ce, : (8.1) where ch is the Pieri map in theorem 5.2, and (8.2) where L(f,k) is the total hook length of % mod % , as defined in section 2. The subscript j on the Pieri product will be dropped when there ts no ambiguity. set aM = cf gt. The precise motivation for the choice (8.2) of the constants ch will be discussed later: The fundametal result to be proved is Theoren 8.2 Fix O#vEV- Then the Pieri products ve: Ino hy oho. form an exact hypercomplex, called the Schur hypercomplex associated with v. ‘The corresponding partial polarizations will be denoted a Prep h— veu with ARG 4, (6.3) Byt ay so that by theorem 6.2, Woe Wont mn. The remainder of this section is devoted to @ proof of théorem 8.2. The three aspects of commutativity, closure and exactness are treated in turn. For the time being, define the Pieri product by formula (8.1), but leave the precise choice of the constants ey op Lema 8.3 Let the Pleri product be defined by (8-1). Jj let w=4d ut = AFK, veutk=pl+J be slaves. Then for vev, Ce, ve (v4.0) = RLV (Y* 0) 5 (8-4) where K fs a constant depending only on Av and the af. # is easy to see that the compositions v* j(v*,C) or v¥ Gre 50) induce functorial maps 0) @ y—> Ly since {v Ov|v€v} spans ®5 + By Pierits formula and theorem k.1, any such functorial map is uniquely determined p to constant multiple, hence, provided the left hand side of (8.4) is not identically zero (which we show subsequently) the lenma is proven. To explicitly determine the constant K in (8.4), by functoriality we need only consider vector spaces V of sufficiently large dimension. Thus if m= m(A) , let e),...,¢,,, be Minearly independent in vy with vee). Let ¢ correspond to the simple tableau 7, of ° shape 2. Fron the formula (5.5) it is easy to check that every summand e, occurring in v¥ JG and v *(v*,6) corresponds to a standard tableau 1, hence ve need not worry about cancellations among terms. Thus to determine K , we need only compare the coefficients of one of these terms in the two products, which, for ease of computation, we choose to be ep > where T is the tableau of shape v coinciding with T, on A, and ith m+ in the remaining two boxes in v/\ . For definiteness, assume J 1, (8.1) for v>2, |v| = [A +k, by iterating the maps }(v@ +) and using the fact that (v“ =v... Ov|v€ ¥} spans OV. By Plerits formula and theorem 4.1, these maps must be identically zero whenever v/% has two or more boxes in any column. This, therefore proves the closure of the Schur hypercomplex. 5h Suppose 2 €8(v) . Suppose v. €V. Then yp aMg Ev tv iti Fv, EHO, for any 5 € 1, . The sum is over all permtations x of {1, and the Pieri products * are taken over any series of shapes gfe alert. This follows immediately from the fact that L, does not appear = Ch, = a Re uy Shy Chg ee Suv vite | in the Pierd decomposition (5.1) of 0, @1, - We now turn to a proof of exactness. Fix of vEV, let Z be the one-dimensional subspace of Y spanned by v and let W be any complementary subspace, 0 V=W@Z. We use the decomposition of Tyv given in corollary 3.4. Tenma 8.7 let O#vEV,2XESL(u) . then the Linear nap abit e- " aor edit Why is an injection. Moreover, suming over 2 €8(u) , ® yw» 09 w ryt e OG Lh is an isomorphism (cf. (3.8).) Proof Let J = (Jys-++53,) ve @ nondecreasing sequence of integers 1s <3 Sa, 8° Sp Sme= mu), with Hild = 2) dly) in the lexicographic ordering. By induction, assume z yk(¥e.) JQ) >a ° is an injection. Then if Eg e@x) = 0 aQ) aa) ON for gy EIgW , the only decomposable terms y appear: with J(y) = J, are those ari from with 3(,) By, appear with nonzero multiple. Thus x, = 0, and the induction ts step i completed. -56- Theorem 8.8 Let Of VEV,pON,k=lv/p| . Then Hore g =o Af and only if ces im Ao.) d= [a/ul 2€8°u,») Proof Using the direct sum decomposition (3.7), and the previous lema, ceE E Ho @ 9, J nest) (y) j jek! Again by lemma 8.6, #(C) = 0 if and only if 1 2 = O for all VE 85(4) 85,() - Therefore ce o£ im HW’ @ +) x €s"(u) 3%) -57- Finally, s re 8h(u)y nsw) A As not wav - maxit with ROA, and im (vw) @ +) = im yf ai jo gM eat HO em) =O ee en) ‘This completes the proof. Corollary 8.9 The Schur hypercomplex is exact. This follows directly from theorem 8.8 and theorem 7.5. (Note that theoren 8.8 is stronger than exactness since v, does not have to equal by «) Example 8.9 Let u = (3,2,1) , v= (3,3,2) , so 8°(u,v) = {(2,2,1),(3,2,0)) - The symbols x,y,z,w,u,v denote linearly independent elements of V , ordered in the manner 4 wae LV, with x,y,2,,u €W. Then from the definition of Pieri cated. We restrict attention to the sumand products and the straightening formalae, (omitting the symmetric product symbol for plicity) 58- W804 @ (yy) @ (aw) @ a) <2 Guy) (awe) @ (wr) + E Gav) © (am) @ Co) 2 F (y2) @ (aw) @ (ww) + % Gye) @ (zw) @ (we) - F Gyy) @ (aw) @(w) , Wyle? @ (yz) @ (ww) @ al = - Z (xy2) @ (war) @ (wv) Wl? @ (xyz) @ (wu) @ v] = (xyz) @ (wav) @ (vv) « Therefore ker Pe.) is spanned by elements of the forms Gav) © (2x) @ w+ Z [Gy2) © (we) @ us (xy) @ (av) @ uj + + § Gye) ee) v4 § (C92) © (ru) @ v + (ye) @ (zu) @ v} and (xyz) © (we) © us % (xyz) @ (wu) Ov. v @ (xy) @ (aw) @u) , A= (2,2,1) , Beklv e (xyz) @w eu), a = (3,20, respectively, hence we have YW? @-) = imt(ve -) + int{ve +) ker 4 (v" @+) = imi (v@ -) + imtikve-) . in accordance with the theorem. Note that the sum is not direct, since 1, (xyv) @ (zv) @ w + 3 (xy) @ (2H) @ ¥ - - § G2) @ (we) ov + b (yw) 9 (av) Ov Lies in both. Indeed, the above element is just Me ” ayelv’ @ Oy) @z2e@w) ats (2,1,D) 5 which clearly factors throu ga both Iyw and Iy.W . ‘The Schur hypercomplex sits inside a larger hypercomplex, whose properties will be useful when we come to prove exactness of the differential hypercomplex. Given 2Cy, |w/2| =k, define the map Yet yer, as the composition © 0 0 8 PF, 20, en HPD a, oer 84, o be where A: 0,70, @0, is the diagonal map on the symhetric algebra of V, Hr O% 7, is the (modified) projection iv) =0,wee, wy wEO, , (note the factor ki) and Oo Ot 7h, Ss the Plert mp. For ACucyv, vo Vv 1. ‘g Qewed. (8-10) Proof Since {v"|v € Vv} generates ©, , it suffices to check (8.10) on vi@E for E€1, arbitrary. Wow if |u/d| =e, then at AOP es) & f Keotvee) , (8.12) ~60- n k+4 (and is zero follows easily from the commutativity of the Schur hygercomplex. Theorem 8.11 The polynomisl hypercomlex YW: 0, @1, 7 0, @ hy is exact. Proof Comutativity and closure follow directly from (8.10) and the corresponding properties of the Schur hypercomplex. The proof of. exactness rests on the following lemma: 1 Lemna 8.12 Suppose YCy, |u/A| =k, and suppose p E Sy (A) Tee) « Then the diegram commutes. Here c= (kt)! /21 (The top map x is really the restriction ot of WY to O,, 81, .) Proof By Schur's lemma, the bottom map mist be a multiple of the identity as all maps commute with the representation of GL(V) . To compute ¢ , by (8.12) P(g @ & hoo os) = ~61- proving the lemma. Clearly, to prove exactness of the polynomial nypercomplex it suffices to Look at homogeneous pieces. Let A,u,v be min-max related with |u/r| =k, [v/p| = 4%. Using the Pieri forma (5.1), we have, for n2>0, vi yw EN o eL Saeret OT > Sore © hy —P eH Q nN X (8.22) ae a @ a Hye ise uy n 1, p é€vt ee oad x OMICS) arts a 8.12, 3 x Ao Cw). thei By Schur's lemma, and lemma 8.22, if p EDL, (A) Th (a). then alt, 1, ~%, 4s 2 nonoro multiple of the identity; otherwise ak | E, vanishes. But by lemma 2.3, 1 1 z hy) 32 u) = UsA) 5 nee) n 1 Boren Tv) =f , so the lower sequence is trivially exact. For -k (ey then the relation (3.1) requires that (le, @e,) Be. 2 08 Ge.) @6,) = -1. ( a? Ma Th other words, che only way to make [ey] , [eq] dual ds to make the pairing between 1,V and 1,V" depend on the order in which the basis elements of V are written. Amore natural way to proceed is to induce the pairing between T,¥ and ,V" from that on @yA,V, Gav". Namely define wo") = <5 (Bo , 80") (0 * * for @€ VY, €6,0,V", and 6, is the map in the definition 6h of the Schur spaces. It is easy to check that ( , ) induces a non- trivial pairing between 1,V, LV - lowing lemma is a direct consequence of (9-1). Letma 9.1 Let 1,S be tableaux of shape 4, let X(T,8) denote the set of all triples of permutations (p,,x,p5) acting on the diagram 4 with p, preserving rows and x preserving colums, such that p,xp,(T) = 6. Then i C 2) Ceprtg) =r Egy MO - (9.2) In particular, (egs¢g)=0 if and & have different contents. le, in 3 Thus, for example, in Lip y)RF , K(e, 0 es) @e, , (ey, Oe) Ge) =1, (ey 02) @ ey 5 ey (oP @e 2 Fee, eed = 2, etc. More generally, if 7 is any tableau with content ¢(1) <2 for each i. (i.e. each symbol appears at most once in T), then ke, prea. (9.3) At the other extreme, if T, is the simple tableau of shape 2, then fen 3, ) Sat. (9-4) V induce dual maps, the interior products, on L,V -65- Definition 9.2 Let y=A+j- The interior product 4.:V8 we =u, (9.5) is defined so that (Gv od aw ¥ 56,0 (9-6) for all v€V,CELV, w€LV « An explicit formula for the interior product can be given in terms of the partial polarization maps: Theorem 9.3 For w= 245, mu) em, » voor retvexo), (9-7) for vEV,w€ ay . Here x@4: veVe ww - uv" is the pairing between V and V" , and wet MEE ty? ‘ oe Hy Eyl (9.8) > 3 ks YK Proof By (9.1), (9-7) is equivalent to the relation res) = of Wwe ska) (9-9) x x OX for all § €1,V, where the right hand ( , ) is the pairing between ve 1,V and We Lv". By functoriality, we need only check (9.9) a pand €= es pan ef for the specific corresponding to the simple tableaux of shape \, u respectively. Now by (6.1) » pte xe) bye, @e jd where the omitted terms are all of the form ¢,@a for k j V5! oe (9.10) H A. Ale ood) = Substituting these last two identities into (9.9) and using (8.1), completes the proof of the theorem. Example let men Whe, hence ) Ly ent ¢ 95) = - FZ [(e,0e, 0€,) ce, + (6, 0€, 06) OF, ] whereas ese, 06,08,) O(e,, 06.) = (c, 0e,08,) OF, ‘The dual hypercomplex corresponding the Schur hypercomplex is based on these interior products. As © direct result of lemma 7.9 and theorem 8.2 we have ~67- Theorem 9.5 Fix Of v€V. Then the interior products vaiuv is form an exact cohypercomplex, called the Schur cohypercomplex. There is also a dual cohypercomplex corresponding to the poly- nomial hypercomplex. Namely, for w>2 let A * * * * * 1 ov ety = av" Ly ORT OEE aT OO denote the dual map to Yj. In the special case |b /A| 1, (9.9) ~toud ON" @ 5, * wo vweveniv even, where the second arrow is the mitiplication map 0,V" @.V" ~ ov" . We immediately have Theorem 9.6 The polynomial cohypercomplex x is exact. ‘The polynomial cohypercomplex can be characterized in a slightly different way. Name: ifwe identify 0,V" with the space @ of polynomial functions on V , then by theorem 9.3, Revd een een, (9-32) » where v,€V"@V is the diagonal element given by ® Ee, @e i for dual bazes on, ation of (9.12) to the reader. -68- ittlewood - Richard: More generally, we may consider the decomposition of the tensor product of two Schur functors into irreducible representations. The basic result was formated by Littlewood and Richardson, [18], in the context of Schur functions. To state this, we need the concept of a word of Yamanouchi. This is a sequence of positive integers VypeVos subject to the condition that in any subsequence Yy,Vp5+--5¥,_ any integer i oceurs at least as often as any larger integer j>i . Thus 1,2,1,3,2,1 is a word of Yamanouchi, whereas 1,2,1,2,2 is not. ‘Theorem 10.1 For any vector space Y, the tensor product decomposition =eny . . LOL -Om tn, : (10.1) holds. Here Re is the number of standard tableau T of shape v/% with entries in (1,...,|ul} , of content wo and sequence obtained by listing the ent: sof T row by row, but from right to left in each roy, forms a word of Yamanouchi. For 2 proof see [2]. Note that Ny O unless vDA,u and [vl = [al + |u|. m practice, one obtains Xu by finding all possible ways to fill in v\2 with wy 1's, p, 2's, ete. in such a way that the resulting tableau is standard, and listing the rows from right to left in order forms a word of Yamanouchi. ‘Thus, for example, if jt) and v = 3,2,1, then Ne 2 since there are two possible ways to place 2 1's and 1 2 in v/X to forma word of Yamanouchi dew -69- see [151], [32], [32] for further examples. Thus the space of functorial maps ¥= HI, 8h,; 1) is of dimension 1 4 Pieri - type product + Each such map p in % can be used to define a G T,0%, 71, but now, unless Ny’, Repeated applications of the Littlewood Richardson rule lead to = 1, there is much more freedom in the definition. more general decompositions QL, =b,8 OL (10.2) for tensor products of Schur functors, and hence Wj - dimensional spaces 1, , L,) , each elenent of which defines a r-fold product between Schur spaces. Problems arise, however, if one tries to choose such products in some consistent fashion so as to make them (at the very least) associative. aed $+ (ed) EAL, 01, OF Inteed, if 9 €H1,@L,,,) 5 & CHL OL)» » 1) then Ly) , but unless this latter space has dimension 1, this in general will not agree (even up to multiple) with a com position 40 (41 @4) for ¥ EWI, Olyn 5 Bale Fen, 92,, , 1,) - In other words, no anount of Juggling coeffi will make the partial product G(@(E@T) @C) and @(E@¥(NSC)) agree, -70- as the reader can check in simple examples. We thus leave aside problems of defining products between elements of Schur spaces except in one important special case, which we discuss in section 12. 11. Algebraic Di Given a functorial map on a tensor product of Schur spaces, there are induced maps on ightly larger tensor products called the algebraic aifterentials of the given functorial map. These will be important when we establish a Leibnitz-type rule for differential hyperforms in section 1s. Tn this cection ve introduce this concept and discuss a few elementary algebraic properties. Let 2= (a4,...,0) be an ordered x-tuple of shapes with @ Ly denoting the corresponding tensor product of Schur spaces, cf- (10-2). Let T, cccur with positive miltiplicity in 1, , and suppose (provided 494k is a shape). Further let vou, [v/pl =1- of y is the functorial map Yost Seg By given by the composition te vy. * Srey, 7 Ve @L, “3 % where the last map i che Pieri proaict ¢Y . the Jjketh algebrate differential of 4 is the map aaa ba Bea? given by composing Yg , with the partial polarization P “L ev, Ly wee ad cf. (8.3). Tf a= (i) , ,wedencte ¥ , by just #.. vd YP a Example 11.2 Let w= (2,2) ,v 3,1). Consider the map P:Vevev—1 ie given by W(x@y@z) = (x Oy) Oz. Then vu : O,8VOV 7 Ly is # (Gx Ox,)@¥@z) = 2 (x, ox, 0y) Oz - F Oy Oxy 02) By , ” 2 3 2 3 2 whereas V3: Veveo, =, is the different functorial map W328 ¥8(z, 025) Commutativity of the hypercomplex immediately gives as above, henna 11.3 For 4 nes tes et Venas* "es e5 **e(5,5") (a3) and a. =e. , = . (21.2) syd ge VS adhe dS es whenever both make sense. for yas above, wok ; a. %65 ‘ay (22.3) the sum being over all k such that 2+k is a shape. ~The 22. In this section we restrict attention to the n-subcomplexes of the Schur hypercomplex given by the n-fat shapes X= nr , O (x-I)n. (22.38) -15- Proof We prove case ii), the proof of case i) being similar, but easier to Implement. Assuse, by induction, thet the tensor product 1 9... 5 decomposes into a direct sum of ,'s corresponding to shaves of the following three types. a) # n-fat. This oceurs with multiplicity 1. bd) pts with pw >nel ¢) u's which are n-bounded, but have 2

    You might also like