Zero-dimensional Gorenstein Algebras with the Action of the Symmetric Group
HIDEAKIMORITA(*) - AKIHITOWACHI(**) - JUNZOWATANABE(***)
1. Introduction.
Suppose that A is a Gorenstein K-algebra with a strong Lefschetz element l2A. Put L l2End(A). Then it is possible to construct a degree 1 map D2End(A) such that fL;D;Hg is an sl(2)-triple, H[L;D], and the weight space decomposition coincides with the natural grading decomposition. This means that the eigenspaces ofHare precisely the homogeneous parts ofA. Suppose moreover that the symmetric group Sk acts onAas permutation of the variables and that the invariant linear formlx1x2 xn is a strong Lefschetz element. Then, as is ob- vious, the vector space KerLis fixed, hence the spaces KerL\Aiare fixed by the action ofSk. Thus in such a case an irreducible decomposition ofA can be constructed by first decomposing the kernel of the multiplication map
(x1 xk):A!A 1
into irreducible spaces and then applying the map D repeatedly to the constituents of the decomposition of KerL. (Or equivalently, first decom- pose KerDand then applyL.)
An obvious example of an Artinian Gorenstein algebra in which lx1 xk is a strong Lefschetz element is the equi-degree mono-
(*) Indirizzo dell'A.: Muroran Institute of Technology, Muroran 050-8585, Japan.
E-mail: morita@muroran-it.ac.jp
(**) Indirizzo dell'A.: Division of Comprehensive Education, Hokkaido Institute of Technology, Sapporo 006-8585, Japan.
E-mail: wachi@hit.ac.jp
(***) Indirizzo dell'A.: Department of Mathematics, Tokai University, Hiratsu- ka 259-1292, Japan.
E-mail: watanabe@sm.u-tokai.ac.jp
mial complete intersection
A(n;k)K[x1; ;xk]=(xn1; ;xnk):
It is possible to identifyA(n;k) with the tensor space (Kn)k
as (GL(n)Sk)-modules. The purpose of this paper is to understand the strong Lefschetz property ofA(n;k) withla strong Lefschetz element in the context of the Schur-Weyl duality.
LetYT be the Young symmetrizer inK[Sk] corresponding to a Young tableauT. One might ask first what is the Hilbert series of
SkYT(A(n;k));
the maximal isotypic submodule ofA(n;k) corresponding to the shape ofT.
It turns out that it equals the q-analog of the Weyl dimension formula.
Since it does not seem to be well known and since we have been unable to find a good reference, we give a proof for it in section 6. This is stated in Proposition 27.
Proposition 27 is not a consequence of the strong Lefschetz property of A(n;k), but Corollary 36, which states that theq-analog of the Weyl di- mension formula is unimodal and symmetric, may be thought of as a consequence of the factlis a strong Lefschetz element.
Terasoma-Yamada [8] considered the coinvariant algebra ofSk, AK[x1; ;xk]=(e1; ;ek);
2
whereei is the elementary symmetric polynomial of degreei. Note that this is also a Gorenstein algebra with the action ofSk. They remark that the Hilbert series ofSkYT(A) is obtained as theq-analog of the hook-length formula. Thus our situation is similar to theirs, but it is not true that the Hilbert series of SkYT(A) is unimodal and symmetric. The algebra A above has the strong Lefschetz property, but no symmetric linear form is a strong Lefschetz element, which causes the difference.
In section 7, we consider the irreducible representation rl:GL(n)!GL(Wl)
of the general linear groupGL(n) on a finite dimensional vector spaceWl and show how the Jordan matrix with a single eigenvaluea,
J(a;n)2GL(n)
is transformed byrl. This is done again by identifying the Artinian algebra
A(n;k) with the tensor space (Kn)k as (GL(n)Sk)-modules so that the Schur-Weyl duality applies to A(n;k). Here the consideration of the be- havior oflas a multiplication operator is indispensable. The two extremal cases ofA(n;k) treated in Sections 3 and 4 should be regarded as examples which show that our method is a new way to understand some classical results.
One other result of this paper is Theorem 22 in Section 5. We apply Theorem 19 to determine the minimal number of generators of the ideal
(x21; ;x2k):(x1 xk)
in the polynomial ring. By analyzing Specht polynomials involving only square-free monomials, it is possible to determine a minimal generating set of the ideal.
The authors thank the referee very much for helpful comments and suggestions.
2. Preliminaries.
2.1 ±A list of notational conventions.
Here is a list of notation which we are going to fix throughout the paper.
Details follow the list.
K denotes an algebraically closed field of characteristic 0.
l(k1;k2; ;kr)`kindicates thatlis a partition of a positive in- tegerk. The same notation is used to indicate a Young diagram of sizek. It is assumed thatk1k2 kr>0. The lengthroflis denoted byl(l).
IfJ2End(V) is nilpotent, we call the partitionl(J)(k1; ;kr) the conjugacy class of J, indicating the sizes of Jordan blocks in the Jordan decomposition ofJ.
A Young tableauTis a Young diagramlwith a numbering of boxes with integers 1;2; ;k. In this caselis the shape ofT.
YTdenotes the Young symmetrizer defined by the Young tableauT.
It is an element of the group algebraK[Sk].
Suppose thatTis a Young tableau andlis the shape ofT. IfV is an Sk-module, then Yl(V) denotes SkYT(V). Thus Yl(V) is the maximal submodule ofVconsisting of modules isomorphic to the Specht moduleVl. We may regard Yl( ) as a functor to extract the isotypic component be- longing tol.
DTdenotes the Specht polynomial defined by a Young tableauT.
2.2 ±The equi-degree monomial complete intersection as a tensor space.
LetRK[x1; ;xk] be the polynomial ring. The partial degree of a homogeneous polynomialf 2Ris the maximum degree offwith respect to a single variable. ByA(n;k) we denote the vector subspace ofRconsisting of polynomials of partial degree at most n 1. Let I be the ideal I(xn1; ;xnk) ofR. Then as vector spaces we have the decomposition
RA(n;k)I:
The vector spaceA(n;k) may be regarded as the tensor space (Kn)k. Since A(n;k)R=Ithe vector spaceA(n;k) has the structure of a commutative ring. PutAA(n;k). A basis ofAcan be the set of monomials of partial degree at most (n 1):
fxi11 xikkj0i1; ;ikn 1g An element ofAis expressed uniquely as
XF(i1;i2; ;ik)xi11xi22 xikk:
With the identificationA(Kn)kthe general linear groupGL(n) acts on the vector spaceAas the tensor representation. Let
F:GL(n)!GL(A) 3
be the representation. Explicitly, ifg(gab)2GL(n) then
F(g)(xj11 xjkk) Xn 1
a0
gaj1xa1
!
Xn 1
a0
gajkxak
! : 4
Here the indicesa;bof the matrix entries forg(gab)2GL(n) range over 0;1; ;n 1. At the same time the symmetric groupSkacts onAas the permutation of the variables.
We are interested in the decomposition ofAinto irreducibleSk-mod- ules. According to the Schur-Weyl duality it will give us an irreducible decomposition ofAas (GL(n)Sk)-modules.
2.3 ±Young tableaux andSpecht polynomials.
A partition of a positive integer kis a way to express kas a sum of positive integers. If we say thatl(k1; ;kr) is a partition ofk, it means
that kk1 krandk1 kr>0. A partition ofkis identified with a Young diagram of size kin a well known manner. Thus the same notationl(k1; ;kr) denotes a Young diagram of sizeP
kiwith rows of kiboxes,i1; ;r, aligned left. A Young tableauTis a Young diagraml whose boxes are numbered with integers 1; ;kin any order. In this case we say thatlis the shape ofT. A Young tableau is standard if every row and column is numbered increasingly.
A Young tableauTdefines a Specht polynomial, denotedDT, as follows:
PutI f1;2; ;kgand
Ij fa2Ija is in thejth column of Tg:
Define
DjY
a;b2a5bIj
(xa xb);
and finally,
DT Y
j
Dj; 5
wherejruns over all columns.DTis the Specht polynomial defined by the Young tableauT.
2.4 ±Nilpotent matrices andJordan bases.
Let M(k) denote the set of kk matrices with entries in K. Let J2M(k) be a nilpotent matrix. Let
nirankJi rankJi1 fori0;1; ;p;
wherepis the least integer such thatJp1O. Thenbl:(n0;n1; ;np) is a partition ofk. We denote the dual partition ofblbyl(J). (cf. Definition 1 below.)
Now suppose thatTis a Young tableau of sizek. Using the numbering ofTdefine the matrixJ(aij)2M(k) by
aij 1 if jis next to and on the right of iinT;
0 otherwise:
( 6
It is easy to see that the matrixJis nilpotent andl(J) is the shape ofT.
We call any matrix defined as above for a Young tableau TaJordan ca- nonical form(of a nilpotent matrix).
Let V be a vector space of dimension k. If a basis of V is fixed, we may identify M(k) and End(V). Suppose that J2End(V) is nil- potent. A Jordan basis for J is a basis of V on which J is in a Jordan canonical form. (According to the definition of a Jordan basis just defined above, any permutation of basis elements of a Jordan basis is a Jordan basis.) Since K is algebraically closed, there exists a Jordan basis. Note that if two nilpotent elements J and J0 are conjugate, then l(J)l(J0). We make a definition of the ``conjugacy class'' of a nilpotent endomorphism, with a slight abuse of language, as follows.
DEFINITION1. LetVbe ak-dimensional vector space overK. Suppose thatJ2End(V) is nilpotent. We say that a partition (k1; ;kr) ofkis the conjugacy classofJif the Jordan canonical form ofJconsists of the Jordan blocks of sizesk1; ;kr. We denote byl(J) the conjugacy class ofJ.
REMARK2. Note that the notationl(J) coincides with the previously definedl(J) for a nilpotent matrix. In fact, if we putnidim ImJi=ImJi1, then the sequencebl(n0;n1; ;np) is a partition ofk. One sees easily that the dual partitionltoblis the conjugacy class ofJ.
LetJ2End(V) be nilpotent with the conjugacy classll(J). Suppose thatBVis a Jordan basis forJ. Then it is possible to place the elements ofBinto the boxes of the Young diagraml(J) bijectively in such a way that it satisfies the following conditions:
e;e02Bande0Je,e0is next to and on the right of e:
e2KerJ,eis at the end of a row.
( 7
(cf. Equation (6).)
REMARK 3. As explained above, by choosing a bijection between the elements of a Jordan basis for J and the boxes of the Young diagram l(J), it is possible to identify a Jordan basis Bfor J and the Young diagram l(J). With this identification the rightmost boxes of l(J) form a basis for KerJ. Also the boxes of the first column of l(J) coincide with fb2Bjb62ImJg. Once l(J) is known, KerJ\B de- termines B. Similarly the diagram l(J) and the subset BnImJB determine B.
2.5 ±The strong Lefschetz property.
LetV Lc
i0Vibe a finite dimensional graded vector space. The Hilbert series ofVis the mapi7!dimVi, which we usually write as the polynomial P( dimVi)qi. (For convention we let dimVi0 for i<0 ori>c.) Let J2End(V) be a map of degree one so J consists of the graded pieces JjVi:Vi!Vi1. ThenJis nilpotent. We say that J2End(V) is a strong Lefschetz elementif the restricted map
Jc 2ijVi:Vi!Vc i
is bijective for alli0;1; ;[c=2]. (Such an endomorphism exists only if the Hilbert series ofV is symmetric and unimodal.)
DEFINITION4. Let ALc
i0Ai be an Artinian graded K-algebra. De- note by:A!End(A) the regular representation ofA. (I.e.,a(b)ab for a;b2A.) We say thatA has the strong Lefschetz property, if there exists a linear form l2A such that l2End(A) is a strong Lefschetz element. We call such a linear forml2Aa strong Lefschetz element ofA as well asl2End(A).
PROPOSITION5. Suppose that ALc
i0Aiis a graded Artinian K-algebra with a symmetric Hilbert seriesP
hiqi. Let l be a linear form of A. Thenl is a strong Lefschetz element if andonly ifl(l)is the dual partition to (h00;h01; ;h0c), which is a permutation of (h0;h1; ;hc) put in the de- creasing order.
PROOF. See [3] Proposition 18. p
PROPOSITION 6. With the same notation as above, suppose that J2End(A)is a strong Lefschetz element. Then any homogeneous basis of KerJ can be extended uniquely to a homogeneous Jordan basis for J.
PROOF. Let P
hiqi be the Hilbert series of A. Since J is a strong Lefschetz element, we have
dim (KerJ\Ai) 0 if hihi1; hi hi1 if hi>hi1: (
8
Letsbe the greatest integer such thaths 1<hsand let mihi hi 1; fori0;1;2; ;s:
Sincehihc ifor 0ic, we may rewrite the equation (8) as dim (KerJ\Ac i) mi fori0;1;2; ;s;
0 otherwise.
(
Now letBbe a homogeneous basis of KerJgiven arbitrarily. We have to find a basis ofAcontainingBon whichJis in the Jordan canonical form. Put Bc iB\Ac ifori0;1; ;s. ThenBc iis a basis of KerJ\Ac i. Since the restricted mapJc 2i:Ai!Ac iis bijective, there is a finite setBiAi such that#Bimiand such thatJc 2i(Bi)Bc i. It is easy to show that the set
B:e Gs
i0
fJj(Bi)jj0;1;2; ;c 2ig
is linearly independent and hence is a basis ofA. It is easy to see thatBeis a desired basis. The uniqueness is obvious since the choice of the finite setBi
is unique fori0;1; ;s: p
The following result is proved in [9] and plays an important role in this paper.
THEOREM7. Let AA(n;k)be the same as defined inSection 1. Then A has the strong Lefschetz property andx1x2 xk is a strong Lef- schetz element of A.
3. A(n;2), the two fold tensor ofKn.
If k2, then the Gorenstein algebra AK[x1; ;xk]=(xn1; ;xnk) takes the form
AK[x1;x2]=(xn1;xn2):
Write x;y for x1;x2. Recall that we identify AA(n;2) as a vector subspace ofRK[x;y] andRA(xn;yn). Denote by:A!End(A) the regular representation of the Artinian algebraA. Put J (xy).
Since J is a strong Lefschetz element, we may use Proposition 6 to construct a Jordan basis for J. First we would like to construct a homogeneous basis for KerJ.
The Hilbert series ofAis the following sequence.
degree 0 1 2 n 2 n 1 n 2n 3 2n 2
dim 1 2 3 n 1 n n 1 2 1
SinceJis a strong Lefschetz element,J:Ai!Ai1is either injective or surjective. Hence we have
dim(KerJ\Ai) 0; if i5n 1;
1; if n 1i2n 2:
(
Since dim(KerJ\Ai) is at most one, a homogeneous basis of KerJ is uniquely determined up to constant multiple. For dn 1;
n;n1; ;2n 2, put bdXd
j0
( 1)jxd jyj; andbdbdmod (xn;yn):
Then sincebd(xy)xd1yd1, we havebd2KerJfordn 1. Thus we have
KerJ\Ad hbdi for dn 1;n; ;2n 2:
9
BecauseJ is a strong Lefschetz element, there is an elementai2Aifor each in 1 such that J2n 2 2i(ai)b2n 2 i. Note that Jj(ai) are all symmetric ifiis even and are alternating ifiis odd. Now we have proved the following.
THEOREM8. The set G
n 1
i0
fJj(ai)jj0;1;2; ;2n 2 2ig
is a homogeneous Jordan basis for J2End(A). The basis element Jj(ai)is symmetric if i is even and alternating if i is odd. The conjugacy class of J is given by
l(J)(2n 1;2n 3; ;3;1):
PROOF. The first part was treated more generally in Proposition 6. The second part follows immediately from the definition ofb2n 2 iandai. The
third statement follows from Proposition 5. p
Now we consider the representation F:GL(n)!GL(A)
as introduced in Section 2.2. (For the definition ofFsee (3) and (4).) Recall that the special linear group SL(2) has a unique irreducible module of dimensionifor eachi>0. We denote it byV(i 1). Fixn>0 and let
C:SL(2)!GL(n) 10
be the irreducible representation corresponding to the moduleV(n 1).
We may considerAA(n;2) as anSL(2)-module via the composition SL(2)!C GL(n)!F GL(A):
11
PROPOSITION9. With the same notation above A decomposes into SL(2)- modules as
AV(2n 2)V(2n 4) V(0):
PROOF. Abbreviate rFC, so r:SL(2)!GL(A). The group homomorphismrinduces a Lie algebra homomorphism
dr:sl(2)!gl(A):
It is well known that the irreducible decomposition ofris determined by that ofdr. Now recall that the Lie algebrasl(2) is the vector space spanned by three elementse;f;hwith the bracket relations
[e;f]h;[h;e]2e;[h;f] 2f:
It is easy to see that to decomposeAinto irreduciblesl(2)-modules is to decomposedr(e) into Jordan blocks (cf. [9]). Notice thatdC(e) is nilpotent and may be considered as a single Jordan block by conjugation sinceCis irreducible. Consequentlydr(e)2gl(A) may be considered as the multi- plication map(xy) by definition ofranddr. Hence the assertion follows
from Theorem 8. p
REMARK10. The isomorphism in Proposition 9 is known as the Clebsch- Gordan decomposition of the tensor productV(n 1)V(n 1) assl(2)- modules. We may also regard it as a consequence of Pieri's formula as follows. Let rl denote the irreducible representation of GL(n) corre- sponding to a Young diagram l with l(l)n. Pieri's formula (e.g. [6, I
(5.16)]) applied to the character proves that rmr(r)M
l
rl;
where the sum is taken over all partitionslsuch thatl=mis a horizontalr- strip. Whenn2 andm(r) and allrlare restricted toSL(2), it gives us the decompostion of Proposition 9 withrreplacingn 1.
With the same notation as above let WV(n 1)V(n 1). Let W WsWa be the decomposition of W into the symmetric and alter- nating tensors respectively. (It is well known that the spacesWs andWa are irreducible GL(n)-modules, which we take for granted.) Via the re- presentation (10) the spacesWs andWaare also SL(2)-modules. The fol- lowing proposition shows how Ws and Wa decompose into irreducible SL(2)-modules.
PROPOSITION11. If n is even, then
Ws V|{z}(2n 2)V(2n 6) V(2)
n=2
;
and
WaV(2n|{z}4)V(2n 8) V(0)
n=2
:
If n is odd, then
Ws V|{z}(2n 2)V(2n 6) V(0)
(n1)=2
;
and
WaV(2n|{z}4)V(2n 8) V(2)
(n 1)=2
:
PROOF. Identify WA(n;2). Then Ws is the space spanned by the symmetric polynomials inAandWathe alternating polynomials. Hence the assertion follows immediately from Theorem 8 and Proposition 9. p
REMARK12. LetrFCbe the composite SL(2)!C GL(n)!F GL(W)
as defined in (11). Letrrsrabe the decomposition corresponding to WWsWa. Lethdiag q 1;q
2SL(2). The plethysm formula (e.g.
[6, (I.8) Example 9)]) can be used to obtain the diagonal form for the ma- tricesrs(h) andra(h) and the Jordan decomposition for
dr 0 0
1 0
L:
As is conceived, it gives us the same decomposition as Proposition 11. In particular, it provides us with another proof for the fact that xy is a strong Lefschetz element in the Artinian algebraR=(xn;yn).
4. A(2;k), thek-fold tensor ofK2.
Throughout this section we fixAA(2;k). SoAis the subspace of the polynomial ring K[x1; ;xk] spanned by square-free monomials. At the same timeAis endowed with the algebra structure with the identification:
AK[x1; ;xk]=(x21; ;x2k):
We putL (x1x2 xk) and D @
@x1 @
@xk:
We think of L and D as operating on the polynomial ring R
K[x1; ;xk]. Recall thatRAI, whereI(x21; ;x2k). We denote by DjA the restricted mapDonA. Similarly by LjA we denote the map R=I!R=I induced by L. Thus LjA;DjA2End(A). Let H be the com- mutatorH[LjA;DjA]. SoH2End(A).
PROPOSITION13. (a)hLjA;DjA;Hiis ansl(2)-triple.
(b)There exists a Jordan basis B for LjA such that BnIm (LjA) is a basis ofKer (DjA).
PROOF. (a) LetJ 0 0
1 0
;J 0
1 0 0
. Then the three elements hJ;J ;[J;J ]i, where [J;J ] 1
0 0
1
, is ansl(2)-triple. This proves the case k1. Let E2 be the 22 identity matrix. Then one sees that, using square free monomials as a basis ofA, the mapLjAis represented by
the matrix
Xk
i1
E2 E2
|{z}
i 1
JE|{z}2 E2 k i
and similarlyDjAby Xk
i1
E2 E2
|{z}
i 1
J E|{z}2 E2 k i
:
We induct on k to show that the three matrices LjA;DjA, and H:[LjA;DjA] satisfy the required relations [H;LjA]2LjA, and [H;DjA] 2DjA.
(b) See Humphreys ([4] pp. 31-34). p
The following theorem enables us to construct a Jordan basis forLjA. THEOREM14. For i0;1;2; ;[k=2], the vector space(KerD)\Aiis spannedby the Specht polynomial of degree i. The Specht polynomials arising from the standard Young tableaux form a basis of(KerD)\A.
Proof of Theorem 14 is postponed to the end of Proposition 18.
LEMMA15. KerDK[fxi xjj1i;jkg].
PROOF. Recall that RK[x1; ;xk] and KerD ff 2RjDf 0g.
Since KerDis a subalgebra ofR, we have
KerDK[fxi xjj1i;jkg]:
The right hand side is isomorphic to the polynomial ring in (k 1) variables.
Noticing that Dis surjective of degree 1, it is easily verified that they
coincide by comparing the Hilbert series. p
LEMMA16. Put V (KerD)\A. Then dimViMax k
i
k i 1
;0
;
where ViV\Ai.
PROOF. First note that the Hilbert series ofAis given byP k i qi.
The graded vector spaceAhas the strong Lefschetz property withLjA a strong Lefschetz element. By Proposition 13, the mapsDjAand LjA are alike except that DjA is a degree 1 map. ThusDjA:Ai!Ai 1 is either injective or surjective. Thus the assertion follows. p LEMMA17. Let T be a Young tableau with the shapel. Suppose thatl has size k. LetDTbe the Specht polynomial defined by T.
(1) DT1()lhas one row.
(2) DT2A andDT61()lhas two rows.
(3) Suppose thatDT2A. Then the degree ofDTis equal to the sec- ondterm ofl.
(4) DT2Ker [DjA:A!A]\Ai if l(k i;i).
PROOF. (1), (2) and (3) are immediate from the definition of DT. (4)
follows from Lemma 15. p
We need one more proposition to prove Theorem 14.
PROPOSITION18. Suppose thatl(k i;i)is a Young diagram.
(a) The number of standard Young tableaux of shape l is k
i
k i 1
.
(b) The set of Specht polynomials defined by the standard Young tableaux of a fixedshapelis linearly independent.
PROOF. (a) Suppose thatTis a standard Young tableau. If the boxkis removed fromTit is a Standard Young tableau of sizek 1. Thus the in- duction works. (Details are left to the reader.) (b) Suppose thatTis stan- dard. Then one notices easily that the head term ofDT in the reverse lex- icographic monomial order is the product of monomials in the second row.
Thus the proof is complete. p
PROOF OFTHEOREM14. Immediate by Lemmas 15, 16, 17 and Propo-
sition 18. p
Now our main theorem of this section is stated as follows.
THEOREM19. Let JLjAandVd(KerDjA)\Ad. Put h[k=2], and mi k
i
k i 1
for i0;1;. . .;h.
(1) The conjugacy classl(J)of J is given by
l(J)
(k|{z}1; ;k1
m0
;k|{z}1; ;k 1
m1
;k|{z}3; ;k 3
m2
; ;1;|{z} ;1
mh
); if k is even;
(k|{z}1; ;k1
m0
;k|{z}1; ;k 1
m1
;k|{z}3; ;k 3
m2
; ;2;|{z} ;2
mh
); if k is odd:
8>
>>
<
>>
>:
(2) For0dh, and0ik 2d, the vector space Ji(Vd)
is an irreducible Sk-module of isomorphism typel(k d;d).
(3) For any Specht polynomialDT2Vd, the vector space hDT;J(DT);J2(DT) ;Jk 2d(DT)i
is an irreducible GL(2)-module of isomorphism typel(k d;d).
(4) An irreducible decomposition of A as Sk-modules is given by AMh
d0
M
k 2d
i0
Ji(Vd)
! :
In particular the irreducible Sk-module of type l(k d;d)occurs (k1 2d)times, andthe irreducible GL(2)-module of type(k d;d) occurs mdtimes.
PROOF. (1) See Proposition 5. (2) In Lemma 17 and Proposition 18 we showed that the spaceVdis spanned by the Specht polynomials defined by the standard Young tableau of shape (k d;d). It is well known that this is irreducible. (Cf. [8].) Alsoli(Vd) is isomorphic toVdunless it is trivial be- causex1 xkisSk-invariant. (3) Letrbe the composition
SL(2)!C GL(2)!F GL(A);
where C is the natural injection and Fis the tensor representation. To decomposeAinto irreducibleGL(2)-modules is the same as to decompose it as SL(2)-modules. This is obtained by decomposing the Lie algebra re- presentation:
dr:sl(2)!gl(A):
Now the assertion follows from Lemma 13. (4) Clear from (1), (2) and (3).
p
EXAMPLE 20. Let k4. Let l (x1x2x3x4)2End(A(2;4)).
We exhibit the Jordan basis ofl. The Hilbert series ofAis (1; 4; 6; 4; 1).
The derived sequence is (1; 3; 2).
1. The Specht polynomial of degree 0 is 1.
2. The Specht polynomials of degree 1 (corresponding to the standard Young tableau with shape l(3;1)) are a:x1 x2;b:x1 x3 andc:x1 x4.
3. The Specht polynomials of degree 2 (corresponding to the standard Young tableau with shape l(2;2)) are f :(x1 x2)(x3 x4) andg:(x1 x3)(x2 x4).
The bases for the irreducible decomposition of AA(2;4) as GL(2)- modules are:
1. h1;l;l2;l3;l4i
3. ha;la;l2aiandhb;lb;l2biandhc;lc;l2ci 3. hfiand hgi
We have 24513312. When li is expanded, all terms which contain a square of a variable should be regarded zero. With this conventionliis equal to theith elementary symmetric polynomial multi- plied byi!.
5. Application to the theory of Gorenstein rings.
Put AR=(x21; ;x2k) and lx1x2 xk 2A. Using the notation of Proposition 13, we have lLjA. Since we have obtained a Jordan basis for DjA:A!A, it gives us a basis for 0:l as well as for Ker [D:A!A]. (cf. Proposition 13.) However, it does not necessarily a minimal ideal basis for 0:l. In this section we would like to exhibit a minimal basis of the ideal 0:l. Denote by ( )?:A!Athe ``Hodge dual'' ofA.
Namely, define
M? (x1 xk)=M
for a monomialM2A. By linearity this is extended to define the dual map A!A. The following lemma is easy to see, and proof is omitted.
LEMMA21. Letl(r;s)be a Young diagram, with rsk, andletD be a Specht polynomial of shapel. Then for any integer i,1ik,
(1) @
@xiDis either0or a Specht polynomial of shape(r1;s 1).
(2) @
@xiF
?
xi(F?)for any F2A:
(3)
@
@x1 @
@xk
F
?
(x1 xk)(F?)for any F2A:
In the next theoremDT denotes the Specht polynomial defined by the Young tableau of some shapel.
THEOREM22. In the polynomial ring R, put I(x21; ;x2k):(x1 xk):
Then the minimal number of generators of the ideal I is k k
h
k h1
:
Here h is such that hk=2or h(k1)=2according as k is even or odd.
If k is even, then
I(x21; ;x2k) fD?Tjshape(T)(h;h)gR;
and if k is odd,
I(x21; ;x2k) fD?Tjshape(T)(h;h 1)gR:
(If k is even, it is the same if?is dropped.)
PROOF. It is enough to prove the second part. Put V
Ker [D:A!A], and letVibe the homogeneous part of degreeiso that we have the decompositionVL
Vi. First notice that
@
@x1Vi @
@xkViVi 1 12
by Lemma 21 (1). This is equivalent to
x1Vi? xkVi?Vi?1 13
by Lemma 21 (2). We have already proved that the vector space V is spanned by the Specht polynomials. Hence by Lemma 21 (3), we have that 0:lis spanned byD?Tfor variousT. The above equality (13) shows that 0:l
is, as an ideal, generated byVh?. p
COROLLARY23. Let AK[x1; ;xk]=(x21; ;x2k), andlet l be the linear element lx1 xkof A. Then the minimal number of generators of the ideal(0:l)is
k h
k h1
:
Here h is as inTheorem 22.
PROOF. Immediate by Theorem 22. p
THEOREM24. As before let AR=(x21; ;x2k), andlet y2A be a gen- eral linear form of A. Then the Macaulay type of A=(y)is the hth Catalan number 1
h1 2h
h . Here h is as inTheorem 22. Equivalently if we put I(x21; ;x2k;Y), where Y is a general linear form of the polynomial ring andif we write a minimal free resolution of R=I as
0!Fk!Fk 1! !F1!R!R=I!0;
then we haverank Fk 1 h1
2h h .
PROOF. The Macaulay type ofA=(y) is equal to the minimal number of generators of 0:y. It is well known that this is also equal to the last rank of the minimal free resolution of 0:y. Now it suffices to notice that
k h
k h1
1 h1
2h h ;
wherehk=2 or (k1)=2 as in Theorem 22. p
6. The Hilbert series of the ring of invariants ofA(n;k).
In this section we letAK[x1; ;xk]=(xn1; ;xnk) wherenandkare arbitrary positive integers. Let G:Sk act onA by permutation of the variables. In the next theorem we would like to exhibit the ring of in- variantsAGand the Hilbert series ofAG.
THEOREM25.
AGK[e1; ;ek]=(pn;pn1; ;pnk 1):
Here edis the elementary symmetric polynomial of degree d and pdis the power sum pdxd1 xdk. Hence the Hilbert series of AGis:
hAG(q)(1 qn)(1 qn1) (1 qnk 1) (1 q1)(1 q2) (1 qk) PROOF. Consider the exact sequence
0!(xn1; ;xnk)!R!A!0:
Since chK0, we have the exact sequence
0!(xn1; ;xnk)G!RG!AG!0:
Note that RGK[e1; ;ek]. The socle degree of A is nk k, and Ank k h(ek)n 1i. Sinceenk 1is fixed under the action ofG, this shows that the maximum degree of elements ofAGisnk k.
Put A0RG=(pn;pn1; ;pnk 1). Obviously we have a natural sur- jection:
A0!AG!0 14
which we would like to prove to be an isomorphism. First note that the rational series in the statement of this theorem is the Hilbert series ofA0. (This can be obtained using the factRGK[e1;e2; ;ek].) This shows that A0andAGhave the same socle degree, which is equal tonk k. SinceA0is an Artinian Gorenstein ring, the one dimensional vector space of the maximum degree is the unique minimal ideal of the ring A0. This shows that the map (14) cannot have a non-trivial kernel. This completes the
proof. p
REMARK26. 1. Since limq!1
(1 qn)(1 qn1) (1 qnk 1)
(1 q1)(1 q2) (1 qk) nk 1 k
; we have dimAGhAG(1) nk 1
k
. This is expected sinceAGis the irreducible GL(n)-module corresponding to the trivial l, which is the symmetric tensor space.
2. One may conceive that the Hilbert series of Yl(A) where Yl is a Young symmetrizer corresponding to l(k1; ;kr)`k should be ob- tained as a q-analog of the dimension formula of the irreducibleGL(n)- module in the decomposition of (Kn)k. This can be proved using the q-
dimension formula ([5] Proposition 10.10). We indicate an outline of proof in the next proposition.
Recall that we have the isomorphismA(n;k)(Kn)kas vector spaces.
The symmetric groupSkacts onA(n;k) as permutations of variables. Also the general linear group GL(n) acts on A(n;k) as the tensor repre- sentation. Recall that Yl(A(n;k))SkYT(A(n;k)), where the map YT:A(n;k)!A(n;k) is a Young symmetrizer with shapel`k with at most n parts. It is well known that Yl(A(n;k)) is an irreducible
GL(n)Sk
-module and every irreducible GL(n)-module is obtained as an irreducible component ofYl(A(n;k)) for somel. (For details see, e.g., [2] pp. 336-339.)
PROPOSITION27. Forl(k1;. . .;kr)`k with rn, the Hilbert series of the module Yl(A(n;k))as a graded subspace of A(n;k)is
hYl(A(n;k))(q)qn(l) Y
1i5jn
[ki kjj i]
[j i] : Here n(l):P
j1(j 1)kj,[a]denotes1 qa
1 q for any positive integer a, and kj0for j>r.
Before we give a proof we review theq-dimension formula [5, § 10.9,
§ 10.10] forGL(n). LetWbe a finite dimensional irreducibleGL(n)-module with highest weightl, andW L
m Wmits weight space decomposition. The q-dimension ofW is defined by
dimqW X
m:weight ofW
( dimWm)qhl m;di; 15
whereh;idenotes the standard inner product onRn, anddis ``the half sum of the positive roots'' defined by (n 1;n 3;. . .; n1)=22Rn. Note that the exponent hl m;diis the number of the simple roots oc- curring inl m, sinceha;di 1 for any simple roota.
Theq-dimension formula [5, Proposition 10.10] says dimqW Y
1i5jn
[ki kjj i]
[j i] ; 16
where W is a finite dimensional irreducible GL(n)-module with highest weightl(k1;. . .;kn).
PROOF OF PROPOSITION 27. - Let GL(n) act on the n-dimensional K-vector space K[x]=(xn) through the vector representation with the basis 1;x;x2;. . .;xn 1. Then xj 1 is a weight vector with weight
ej(0;. . .;1;. . .;0) (only jth entry is 1). Therefore the weight of a
weight vector decreases by a single simple root (like ej ej1), as its degree increases by one. This principle holds also for A(n;k), since A(n;k)K[x1;. . .;xk]=(xn1;. . .;xnk) is isomorphic to the tensor product K[x1]=(xn1)K KK[xk]=(xnk) as GL(n)-modules. In particular, the homogeneous polynomial with the least degree in Yl(A(n;k))A(n;k) is the highest weight vector.
It follows from the principle above and the note after (15) that the Hilbert series and the q-dimension of Yl(A(n;k)) differ only by a power of q. Notice that the q-dimension dimqYl(A(n;k)) starts with q0 followed by higher terms, while the Hilbert series hYl(A(n;k))(q) starts with qd, where d is the degree of the monomial of the highest weight vector. Therefore it suffices to show that the degree of a monomial with weight l(k1;. . .;kr)`k (rn) is equal to n(l).
Since the weight of 1 ise1(1;0;. . .;0) in theGL(n)-moduleK[x]=(xn), the weight of 1 inA(n;k)'(K[x]=(xn))kiske1(k;0;. . .;0). The degree of a monomial with weightlis
hke1 l;di h(k k1; k2; k3;. . .; kr);di
k22k3 (r 1)kr:
We thus have proved the assertion. p
7. The multiplicative group of the Artinian algebras.
We need the following proposition from commutative algebra. Very important to us is the corollary that follows, which says that the generic form of degreedhas the largest rank among all elements of orderd. The proof for the case where M is principal is given in [10, Proposition A in Appendix]. Since the proof goes verbatim, we omit the proof.
PROPOSITION28. Let(R;m)be an Artinian local ring with residue field K andM a finite R-module. Let m1; ;msbe arbitrary elements ofm, and let lx1m1x2m2 xsms with xi2R. Let X1; ;Xs be in- determinates over R, let QR[X1; ;Xs]be the polynomial ring andlet
Me MRQ=Y(MRQ):
where Y X1m1 Xsms. Let U be the set of non-zero-divisors of Q andletabe the ideal(X1 x1; ;Xs xs)Q. Then we have
lengthQU(MeU)lengthQ=a(Me QQ=a):
17
Furthermore, there exists a radical idealbK[X1;. . .;Xs]such that the inequality(17)is the equality if andonly if the ideal(X1 x1;. . .;Xs xs) does not containb.(xidenotes the image of xiin K.)
COROLLARY29. Suppose that ALc
i0Aiis a standard graded K-alge- bra, where K is an infinite field. Let MLb
iaMi be a finite graded A- module. Letmbe the maximal ideal of A. Let d be any positive integer.
Then the minimal value
MinfdimM=yMg;
where y ranges overmd, is attainedby a homogeneous element y of degree d. If M has the strong Lefschetz property, and if z2A1is a strong Lefschetz element for M, then we have
dimM=zdMMinfdimM=yMjy2mdg;
or equivalently, the rank of the linear mapzd:M!M is greater than or equal to the rank ofy:M!M for any y2md.
PROOF. Since mdL
jdAj and since md=md1Ad, any represen- tative of aK-basis ofmd=md1is a minimal generating set of the idealmd and vice versa. Thus the first part is a direct consequence of Proposition 28.
(We apply the proposition by lettingm1; ;ms be a set of minimal gen- erating set of md.) To see the second part suppose that y2md is any homogeneous element. The linear mapy:M!Mdecomposes into the sum of piece-wise linear maps y:Mi!Mid. Hence the rank of the linear mapy:M!Mdoes not exceed
Xb d
ia
MinfdimMi;dimMidg:
18
(Here we have setMbi0 fori>0.) Now ifzis a strong Lefschetz ele- ment and ifyzd, then the rank ofy:M!Mis equal to the integer given in (18) above. In fact the linear map zd:Mi!Mid is either in- jective or surjective as is easily deduced from the definition. Thus we have shown that the rank of the linear mapzd:M!Mis greater than or equal
to the rank ofy0:M!Mfor anyy02md. This is equivalent to the second
assertion of the corollary. p
Let ALc
i0Aibe an Artinian standard gradedK-algebra. We denote by A the multiplicative group ofA. Obviously it is an Abelian algebraic group with the underlying algebraic set
(a0;a1; ;ac)2AM
Aijai2Ai;a060 : We may regard it as an algebraic subgroup ofGL(A).
EXAMPLE30. LetAK[x]=(xn). Thenf 2Ainduces an automorph- ismf :A!A. If we write
f a0a1xa2x2 an 1xn 1; thenf is represented by the matrix
a0 a1 a2 an 2 an 1
a0 a1 an 2
a0 an 3
... .. . ...
.. . ...
ao 0
BB BB BB B@
1 CC CC CC CA :
This way we regard the groupAas the algebraic subgroup ofGL(n). p LetV be ann-dimensional vector space and let
F:GL(V)!GL(Vk)
be the tensor representation. Now letV K[x]=(xn) be as in Example 30 above and letAA(n;k)K[x1; ;xk]=(xn1; ;xnk). Define the map
F0:V!A by
f(x)7!f(x1)f(x2) f(xk):
Then it is easy to see thatF0is a group homomorphism. As described above we may identifyVas a subgroup ofGL(V) andAa subgroup ofGL(Vk).
Note thatF0is nothing but the restriction ofF. In other words we have the commutative diagram:
19
where the vertical maps are natural inclusions.
ByJ(a;n) we denote thennmatrix
It is an elementary fact that if a matrixMhas a single eigenvaluea, thenM decomposes into a direct sum of such blocks as follows:
This is known as the Jordan canonical form of M. We denote this matrix by
J(a;n1) J(a;nr):
DEFINITION 31. Let VPb
iaVi be a graded K-vector space with the Hilbert series
h(q):hV(q)Xb
ia
hiqi;
where we assume that Va60andVb60. Suppose that h(q)is symmetric andunimodal. Then thedual Hilbert seriesof V, or of h(q), is defined to be the descending sequence of positive integers
u1;u2; ;us which satisfy
h(q) 1 1 q
Xr
i1
q(ab1 ui)=2 q(ab1ui)=2 : 20
REMARK32. With the same notation as above, the Hilbert seriesh(q) of V gives a partition of the positive integer dimV. In fact dimV
haha1 hb. (If one prefers, the summands may be arranged in the increasing or decreasing order.) The dual Hilbert series of V is nothing but the dual partition of the Hilbert series. For example if h(q)qa3qa16qa27qa36qa43qa5qa6, then the dual Hilbert series of V is 7;5;5;3;3;3;1, independent of the shift bya.
Now we may state the main theorems of this section.
THEOREM33. LetF:GL(n)!GL((Kn)k)be the tensor representation of the general linear group GL(n). Let MJ(a;n). Then akis the unique eigenvalue ofF(M)andthe Jordan canonical form ofF(M)is given by
J(ak;u1) J(ak;ur);
where u1; ;ur is the dual Hilbert series of the polynomial h(q)
(1q qn 1)k.
PROOF. Put AA(n;k)K[x1; ;xk]=(xn1; ;xnk). Then, since (1 qn 1)kis the Hilbert series ofA, it suffices to prove that (1)akis the unique eigenvalue of F(M) and (2) F(M) ak has the same Jordan canonical form as that of a strong Lefschetz element ofA. (See Proposition 6.) PutV K[x]=(xn). We have the commutative diagram (19) as shown above. With the natural embedding V!GL(n) the element ax corresponds to the matrix J(a;n). Thus F0(ax)F(M). Now put l0F0(ax) ak. Then we have
l0(ax1) (axk) akak 1(x1 xk)polynomial inxi of degree2:
Putlx1 xk. Then we have 1
ak 1l0
i
li modmi1; i1;2; :
Thus by Proposition 28 we have that the rank of (l0i) is equal to the rank of (li) for alli1. This shows thatF0(a x) akandlhave the same Jordan canonical
form. Now the proof is complete. p
THEOREM34. Letl`k with at most n parts andlet Wlbe an irreducible GL(n)-module corresponding to l. Denote by fl:GL(n)!GL(Wl) the corresponding irreducible representation of GL(n). Put
A(n;k)K[x1; ;xk]=(xn1; ;xnk):
Suppose that the sequence (f|{z}1; ;f1
m1
;f|{z}2; ;f2 m2
; ;f|{z}s; ;fs ms
)
is the dual Hilbert series of Yl(A(k;n)). Then the Jordan canonical form of fl(J(a;n))is given by
J(ak;f1) J(ak;f1)
|{z}
m1=m
J(a|{z}k;f2) J(ak;f2)
m2=m
J(a|{z}k;fs) J(ak;fs)
ms=m
;
where m is the number of the standard Young tableaux of shapel.
Before proving this theorem, we collect some basic facts in a lemma.
LEMMA35. Put AA(n;k)K[x1; ;xk]=(xn1; ;xnk). Letl`k and let T be a standard Young tableau with shapel. We identify A with(Kn)k. Then
(1) Yl(A)A is an irreducible GL(n) Sk-module.
(2) The number of times that the irreducible GL(n)-module Wl occurs in Yl(A)is equal to the dimension of the irreducible Sk-module that occurs in Yl(A).
(3) Suppose that W Yl(A)is an irreducible GL(n)-module. Let m be the dimension of any irreducible Sk-module that occurs in Yl(A). Then hYl(A)(q)mhW(q).
(4) Let lx1 xk. Then l is a strong Lefschetz element for Yl(A). Hence the endomorphisml2End(Yl(A))decomposes as
J(0;u1) J(0;ur);
where u1; ;uris the dual Hilbert series of Yl(A).
PROOF. (1) and (2) are immediate from the Schur-Weyl duality. Let WYl(A) be an irreducible GL(n)-module and letT1; ;Tm be all the standard Young tableaux with shapel. Then we have
Yl(A)SkWYT1W YTmW:
Thus (3) follows. We prove (4). WriteLfor the linear mapl2End(A). Let D2End(A) be the degree 1 map such thatD;L;[D;L] is ansl(2)-triple.
We note thatDis compatible with the action ofSkas well asL. LetZbe any homogeneous basis for 0:l. Then the set S
i0Di(Z) is the Jordan basis forL, and the subsetYl(A)\ S
i0Di(Z)
is a Jordan basis ofLjYl(A). This shows
(4). (cf. Proposition 7.) p
PROOF OFTHEOREM34. - We use the notation of the lemma above. It is well known thatWlappears as a submodule of (Kn)k. First recall that the dimension of an irreducibleSk-module inYl(A) is equal to the number of the standard Young tableaux of shape l. To prove the assertion of the theorem it suffices to show that (1)F0(ax) restricted on the submodule Yl(A) has the unique eigenvalueakand (2)l0:F0(ax) akdecomposes into Jordan blocks in the same way as the multiplicationl2End(Yl(A)) decomposes into Jordan blocks. (It should be noticed beforehand that lYl(A)Yl(A) andl0WlWl, which is easy to see.) By (4) of the lemma above,lx1 xkis a strong Lefschetz element forYl(A). Hence the
proof is complete. p
COROLLARY36. The graded vector space Yl(A)has a unimodal sym- metric Hilbert series for anyl`k.
PROOF. The assertion is an immediate consequence of the fact that l2Ais a strong Lefschetz element forYl(A). (cf. [5, Exercise 10.12].) p
REFERENCES
[1] W. FULTON,Young Tableaux, London Mathematical Society Student Text 35, Cambridge University Press, 1997.
[2] R. GOODMAN - N. R. WALLACH, Representations andInvariants of the Classical Groups, Cambridge University Press, 1998.
[3] T. HARIMA- J. WATANABE,The finite free extension of an Artinian K-algebra with the strong Lefschetz property, Rend. Sem. Mat. Univ. Padova,110(2003), pp. 119-146.
[4] J. E. HUMPHREYS,Introduction to Lie Algebras and Representation Theory, Springer-Verlag, 1972.
[5] V. G. KAC, Infinite-dimensional Lie algebras, third ed., Cambridge Uni- versity Press, Cambridge, 1990.
[6] I. G. MACDONALD, Symmetric functions andHall Polynomials, 2nd ed., Oxford Science Publications, London, 1995.
[7] T. MAENO, Lefschetz property, Schur-Weyl duality and a q-deformation of Specht polynomial, Comm. Algebra,35(2007), pp. 1307-1321.
[8] T. TERASOMA- H. -F. YAMADA,Higher Specht polynomials for the symmetric group, Proc. Japan Acad.,69(1993), pp. 41-44.
[9] J. WATANABE,The Dilworth number of Artinian rings andfinite posets with rank function, Adv. Stud. Pure Math.,11(1987), pp. 303-312.
[10] J. WATANABE,m-full ideals, Nagoya Math. J.,106(1987), pp. 101-111.
[11] H. WEYL, Classical groups, their invariants andrepresentations, 2nd Edition, Princeton, 1946.
Manoscritto pervenuto in redazione il 20 giugno 2007.