Journal of Algebra 243, 448᎐472 Ž2001.

doi:10.1006rjabr.2001.8857, available online at http:rrwww.idealibrary.com on

Decomposition Numbers for Distant Weyl Modules

Anton Cox 1

Mathematics Department, City Uni¨ ersity, Northampton Square, London,

EC1V 0HB, England

Communicated by G. D. James

Received January 10, 2001

Consider a semisimple, connected, simply-connected algebraic group G

over an algebraically closed field k of characteristic p ) 0. One can
construct for each dominant weight ␭ a Weyl module ⌬Ž ␭. with that
highest weight whose character is given by Weyl’s character formula.
Although not in general simple, ⌬Ž ␭. has a simple head LŽ ␭., and all
simple modules arise in this manner.
Knowledge of the decomposition numbers d␭␮ s w ⌬Ž ␭. : LŽ ␮ .x for ␭
and ␮ ‘‘small’’ Ži.e., p-restricted. is equivalent to calculating the characters
of the corresponding simple modules Žand hence by Steinberg’s tensor
product theorem to determining the characters of all the simples.. Conse-
quently, much work has been undertaken to try to determine these
numbers, concentrating mainly on the case when p is large enough to be
able to consider the Lusztig conjecture. Indeed, for sufficiently large
primes the d␭␮ are now known by the work of Andersen et al. w1x.
Although in principle all decomposition numbers can be determined
from those for p-restricted weights Žvia character calculations using the
tensor product theorem and Weyl’s character formula. this is not straight-
forward in practice. Further, it is often more convenient to know decom-
position numbers rather than characters; for example, when relating
representations of the general linear and symmetric groups via Ringel
duality only the former can be translated between the two categories.
We shall consider the situation where ␭ is ‘‘large’’ and give an elemen-
tary algorithm for calculating decomposition numbers given those for all
p 2-restricted weights. If we regard Steinberg’s tensor product theorem as
an algorithm for determining large characters from smaller ones, then this

1
Supported by EPSRC Grant M22536 and EC Grant FMRX-CT97-0100.

448
0021-8693r01 $35.00
Copyright 䊚 2001 by Academic Press
All rights of reproduction in any form reserved.
 DISTANT WEYL MODULES 449

is an analogous result for decomposition numbers. Our algorithm can

easily be inverted, and we discuss an application of this to the representa-
tion theory of the symmetric group using Ringel duality.
There is another, similar, recursive character formula for Weyl modules
due to Jantzen w13x. Away from the boundary of the dominant region this
corresponds to a filtration of ⌬Ž ␭.. This is obtained by considering repre-
sentations of certain induced modules for infinitesimal subgroups Gr T of
G related to the Frobenius kernels. Doty and Sullivan w9x have given an
algorithm for determining decomposition numbers for these induced mod-
ules and have shown how the corresponding result for Weyl modules can
be deduced from this.
In order to describe our algorithm, we introduce certain sets of virtual
decomposition factors with multiplicities. Although these arise naturally in
our argument, this is essentially a combinatorial procedureᎏand hence it
is not immediately clear that such sets have any representation-theoretic
interpretation Žor even that the associated multiplicities are non-negative..
However, we shall show that they are precisely the composition factors
Žwith multiplicities. of certain modules studied by Lin w16x arising as lifts of
modules from corresponding quantum groups at a primitive p 2 root of
unity. More generally, Lin considers the lifts of modules from quantum
groups at p r roots of unity, and our algorithm can also be used to
determine the decomposition numbers for these modules.
In the light of these results, it is natural to ask if our algorithm
corresponds to successive refinements of some filtration of the Weyl
module. This seems to be related to a conjecture of Humphreys w11x
concerning filtrations of Weyl modules, which we briefly discuss. We then
consider evidence for such a structural interpretation, arising from results
of Doty w8x on the submodule structure of the symmetric powers and of
¨
Kuhne-Hausmann w15x on the structure of suitably ‘‘generic’’ Weyl mod-
ules for SL 3 .
Finally, we conclude by noting that an appropriate analogue of our
algorithm can also be derived for the quantum general linear group at a
root of unity in positive characteristic.

1. PRELIMINARIES

In this section we shall briefly review those basic results that will be
required later, mainly so as to fix our notation. All of this material can be
found in w14, Section II, Chaps. 1᎐6x. Towards the end we shall also prove
an elementary proposition on the geometry of lattice points in facets that
will be needed in the next section.
450 ANTON COX

We fix a maximal torus T ; G and hence the lattice of weights X ŽT ..

The pair Ž G, T . determines a root system R, inside which we choose a set
of positive roots Rq. The corresponding set of simple roots we denote by
S. The Weyl group W and associated affine Weyl groups Wp i act on the
space E s X ŽT . m⺪ ⺢.
More precisely, let ␣ k be the coroot associated to ␣ in X ŽT .U and
² ᎐, ᎐ : the usual bilinear form on X ŽT . = X ŽT .U . For each ␣ g R we
denote by s␣ the reflection on X ŽT . given by s␣ ␭ s ␭ y ² ␭, ␣ k: ␣ . This
action extends to the whole of E. Then W is just the group generated by
these reflections. For i G 1 we define Wp i , the affine Weyl group, to be the
semidirect product of W with the group p i ⺪ R Žacting by translations
on E ..
Let ␳ s 12 Ý ␣ g Rq ␣ , an element of X ŽT . m⺪ ⺡. It is easy to verify that
the dot action w. ␭ s w Ž ␭ q ␳ . y ␳ of W Žor Wp i . on E maps X ŽT . into
itself. Henceforth we shall use this action without further comment. As G
is semisimple and simply-connected, the set  ␣ k: ␣ g S4 is a basis for
X ŽT .U , and there is a corresponding basis  ␻␣ : ␣ g S4 of the fundamental
weights for X ŽT ., such that ² ␻␣ , ␤ k: s ␦␣ ␤ for all simple roots ␣ and ␤ .
This further implies that ␳ s Ý ␣ g S ␻␣ g X ŽT ..
The action of Wp i on E defines a system of p i-facets; these are sets of
the form

¡ ²␭ q ␳, ␣ : s n pk
␣for all ␣ g R
i q
Ž F. ¦
~ ¥
0
k
F s ␭ g E: Ž n y 1 . p - ² ␭ q ␳ , ␣ : - n p
i i

¢ ␣
for all ␣ g R
␣
q
1 Ž F .§
for suitable integers n␣ and a disjoint decomposition Rqs Rq 0 F j
Ž .
qŽ . qŽ .
R1 F . A facet F is called an alcove if R 0 F s ⭋ and a wall if
< Rq0 F s 1. The closure F of any alcove F is a fundamental domain for
Ž .<
Wp i on E, and Wp i permutes the alcoves simply transitively. Similarly,
F l X ŽT . is a fundamental domain for Wp i on X ŽT .. Thus it will often be
sufficient to study just the standard alcove Ci , where n␣ s 1 for all ␣ . We
will also need to consider the set of p i-restricted weights
X i Ž T . s  ␭ g X Ž T . : 0 - ² ␭ q ␳ , ␣ k: F p i for all ␣ g S 4 .
Clearly, X ŽT . is a disjoint union of translates of this set by the p i-weight
lattice p i X ŽT ., and X i ŽT . is a union of sets of the form X ŽT . l F for
certain p i-facets F of Wp i . Any weight ␭ can be uniquely written in the
form ␭ s ␭X q p i␭Y with ␭X g X i ŽT ., and any decomposition of ␭ in this
way is to be assumed to be of this form.
Key to our arguments will be the notion of scaling. For each weight ␭
and p i-facet F, there is at most one element of Wp i . ␭ in X ŽT . l F. Thus
 DISTANT WEYL MODULES 451

it is enough to identify the p i-facet in which a weight lies and its orbit
under Wp i to determine the weight itself. Let ␧ i : E ª E be the map
x ¬ p iy1 Ž x q ␳ . y ␳ . Note that ␧ i is a bijection taking p-facets to p i-facets
and that under this bijection the Wp-orbit of x corresponds to the Wp i-orbit
of ␧ i Ž x ..
We may identify Wp i and Wp j via the isomorphism induced from the
obvious isomorphism between p i ⺪ R and p j ⺪ R. Given an element w g Wp
we may denote its image in Wp i under this identification by w Ž i.. It is now
easy to verify that ␧ i Ž w Ž j.. ␭. s w Ž iqjy1.. ␧ i Ž ␭. and in particular that ␧ i
commutes with the dot action of W. We also have that ␧ i Ž ␭ q ␮ . s ␧ i Ž ␭.
q p iy1␮ for all weights ␭ and ␮.
We will often regard p i-facets as though they are p-facets Žby means of
y1 .
␧i and use certain combinatorics of p-facets associated to Wp to
determine a new family of p-facets and hence of p i-facets via ␧ i . By the
remarks above, given a weight ␭ in our original p i-facet, this will unam-
biguously determine a corresponding set of weights in the p i-facets thus
obtained. We shall refer to the identification of p i-facets with p-facets
Žand vice versa. via ␧ i as scaling, and given a p-facet F we shall call ␧ i Ž F .
the p i-facet corresponding to F.
For our later work, it will be important to know when the intersection of
a p i-facet with the weight lattice is non-empty. We shall abuse notation
and say that such facets are non-empty. Set h s max² ␳ , ␤ k: q 1: ␤ g
Rq4 . As G is connected this is just the Coxeter number of R. It is well
known that a p i-alcove contains a lattice point if and only if p i G h, and as
G is simply connected the same is also true for walls Žsee w14, Section II
6.3Ž1.x.. We will need the following easy generalisation of these results
Žour proof follows that in w14, Section II 6.3Ž1.x..

PROPOSITION 1.1. Suppose that p G h. A p-facet F is non-empty if and

only if the corresponding p i-facets Ž under scaling . are non-empty Ž for all i ..

Proof. It is clear that if F contains a lattice point, then so does ␧ i Ž F .

for all i. Thus it is enough to show that if F does not contain a lattice
point, then neither does ␧ i Ž F . for any i. The action of Wp on p-facets
corresponds under scaling to the action of Wp i on p i-facets and hence, by
the conjugacy of alcoves under Wp , we may assume that F ; C1.
For each wall in C1 there is a unique reflection that fixes it. As noted in
w14, Section II 6.3x, the set of such reflections consists of all s␣ , with ␣
simple, and s␤ , p , where ␤ is the longest short root of R. Here s␤ , p is the
reflection that fixes those ␭ satisfying ² ␭ q ␳ , ␤ k: s p. Now every facet
F in C1 can be identified with a distinct subset of these reflections by
setting FixŽ F . to be the set of those reflections which fix F pointwise. For
example, FixŽ C1 . is the empty set.
452 ANTON COX

Any element of X ŽT . can be written in the form ␭ s Ý ␣ g S m␣ ␻␣ and

so
² ␭ q ␳ , ␣ k: s m ␣ q 1

for all simple roots ␣ Žby the explicit expression for ␳ in terms of the ␻␣ ..
Also, if we write the coroot associated to the longest short root ␤ in the
form ␤ ks Ý ␣ g S b␣ ␣ k then for any other root ␥ with ␥ ks Ý ␣ g S c␣ ␣ k
we have c␣ F b␣ for all ␣ g S Žas ␤ k is the maximal Žlong. root in the
dual root system R k. .
First suppose that our facet is not fixed by s␤ , p . Then to contain a lattice
point we require that there exists ␭ s Ý ␣ g S m␣ ␻␣ such that ² ␭ q ␳ , ␣ k:
is zero for all ␣ g FixŽ F . and strictly between 0 and p for all other roots
not in the linear span of FixŽ F .. Consider ␭ s Ý ␣ g FixŽ F . y ␻␣ . Clearly
² ␭ q ␳ , ␣ k: s 0 for all ␣ g FixŽ F ., and for all roots ␥ not in the linear
span of FixŽ F . we have

0 - ² ␭ q ␳ , ␥ k: F ² ␳ , ␥ k: F ² ␳ , ␤ k: - h F p.

Thus every facet not fixed by s␤ , p contains a lattice point.

So it only remains to consider those facets F fixed by s␤ , p . Let SFixŽ F .
be the set of reflections in FixŽ F . not equal to s␤ , p . For F to contain a
lattice point we require that there exists a ␭ s Ý ␣ g S m␣ ␻␣ such that
² ␭ q ␳ , ␣ k: is zero for all ␣ g SFixŽ F ., equal to p when ␣ s ␤ , and
strictly between 0 and p for all other roots not in the linear span of
FixŽ F ..
Arguing as in the last paragraph, it is easy to see that it is enough to
solve the equation

Ý Ž m␣ q 1 . b␣ s p Ž 1.
␣ gS _SFix Ž F .

for some integers m␣ satisfying 0 F m␣ - p y 1 for each ␣ g S _SFixŽ F .,

as setting m␣ s y1 for all ␣ g SFixŽ F . will then give the desired ␭.
The proof now reduces to a case-by-case examination of the possible
values of the b␣ for each root system. Using the tables given in w2,
Planches I᎐IXx it is easy to verify for each root system that there is a
solution of Ž1. for p G h whenever the highest common factor of the b␣
for ␣ g S _SFixŽ F . is 1. Thus in these cases there is always a lattice point
in F. By inspection, when the highest common factor is greater than 1 it
must be less than h Žand hence less than p ., and so in this case there is no
lattice point in F. However, in this case exactly the same argument holds
for the p i-facets Žas there is still no solution to Ž1. when we replace p with
p i ., and so the result now follows.
 DISTANT WEYL MODULES 453

We conclude this section by recalling the basic properties of simple and

Weyl modules that we shall require. Given a Borel T ; B ; G we can
define the modules H i Ž ␭. s R i ind GB k␭ , where k␭ is the one-dimensional
B-module of weight ␭ and R i ind GB is the ith right derived functor of
induction. We set ␹ Ž ␭. s Ý iG 0 Žy1. i ch H i Ž ␭..
By choosing B appropriately, we may arrange that H 0 Ž ␭. is non-zero
precisely when ␭ is dominant, and for these weights ␹ Ž ␭. s ch H 0 Ž ␭. by
Kempf’s vanishing theorem. The Weyl module ⌬Ž ␭. is the contravariant
dual of H 0 Ž ␭. and has the same character, which is given by Weyl’s
character formula.
We will frequently use the following properties of ␹ Žsee w14, Section
II 5.8-9x..
LEMMA 1.2. For all ␭ g X ŽT ., w g W, and Ý␮ aŽ ␮ . eŽ ␮ . g ⺪w X ŽT .xW
we ha¨ e

␹ Ž ␭. Ý aŽ ␮ . e Ž ␮ . s Ý aŽ ␮ . ␹ Ž ␭ q ␮ . ,
␮ ␮

lŽ w .
␹ Ž w. ␭ . s Ž y1 . ␹ Ž ␭. .

Note that for each element ␭ either ␹ Ž ␭. s 0 or there exists a unique

element w␭ g W such that w␭ . ␭ is a dominant weight. When ␹ Ž ␭. s 0, we
set w␭ s 1.
Finally, we note that each Weyl module ⌬Ž ␭. has a simple head LŽ ␭.
Žwhose character is W-invariant . and that all simple modules can be
obtained in this manner. We will often abuse notation and refer to weights
as composition factors by identifying ␭ with the module LŽ ␭.. Any
dominant weight ␭ can be uniquely written in the form ␭ s Ý iG 0 ␭ i p i
with ␭ i g X 1ŽT . for all i. Then by Steinberg’s tensor product theorem we
have LŽ ␭. ( mi LŽ ␭ i . F , where F is the Frobenius morphism.
i

2. THE MAIN THEOREM

Throughout this section we shall assume that the decomposition num-

bers for Weyl modules with highest weight in the set of p 2-restricted
weights are known. In examples we shall only consider Weyl modules
whose highest weight lies in the interior of an alcove: however, our main
result holds for all weights without restriction.
Henceforth we will assume that p G h. This will allow us to appeal to
the translation principle w14, Section II 7.17, Corollaryx, and note that the
facets containing composition factors of a Weyl module ⌬Ž ␭. depend only
on the facet in which ␭ lies. Under this hypothesis, Proposition 1.1 will
454 ANTON COX

also ensure that the p-facet corresponding to a non-empty p i-facet will

also be non-empty. We will repeatedly make use of both of these proper-
ties without further comment. By considering the Steinberg weight Ž p i y
1. ␳ , it is easy to see that for p G h we also have X i ŽT . : Ciq1 , for all
i ) 0.
We will associate to each weight ␭ a set of i-virtual composition factors
Žwith multiplicities.. As this is a somewhat lengthy process, we will proceed
in several stages. We begin by associating to each non-empty p-facet F
containing a dominant weight a decomposition diagram. This is defined to
be a set of facets H with multiplicities d F H defined by picking an arbitrary
weight ␭ g F and determining which facets contain composition factors of
⌬Ž ␭.. These are the facets of the decomposition diagram, and their
multiplicities are just those of the corresponding composition factors.
Next consider the set of p 2-facets inside the set of p 2-restricted weights.
For each such facet ␧ 2 Ž F . that is non-empty we define a p 2-decomposition
diagram in the following manner. Under scaling, ␧ 2 Ž F . corresponds to the
Žnon-empty. p-facet F, which has an associated decomposition diagram.
The p 2-decomposition diagram associated to ␧ 2 Ž F . is just the set of
p 2-facets Žwith multiplicities. corresponding to this diagram under scaling.
To each non-empty p-facet F in the set of p 2-restricted weights we now
associate a ¨ irtual decomposition diagram. We proceed by induction on the
p 2-facets below the p 2-facet containing F Žusing the partial ordering on
facets induced by the usual dominance ordering on weights.. The set of
p-facets in the virtual decomposition diagram for F are just those E for
which the virtual decomposition number

cF E s d F E y Ý d H I c J E / 0, Ž 2.
I-H

where F lies in the p 2-facet ␧ 2 Ž H . and J is the image of F under Wp 2 in

␧ 2 Ž I .. The multiplicity of such a facet E is just c F E .
It is possible for some of the facets J in Ž2. to lie outside X 2 ŽT .. Thus
for our inductive definition to make sense we also need to define virtual
decomposition numbers c J E for such facets. Any such J can be uniquely
written in the form J s J X q p 2␶ , where J X is a p-facet in X 2 ŽT . and
␶ g X ŽT .. The virtual decomposition numbers c J X K are already defined by
induction, and we set c JE s c J X E X , where E s EX q p 2␶ .
Given a p 2-restricted weight ␭, the set of virtual composition factors
associated to ␭ is just the set of elements of Wp . ␭ lying in some p-facet of
the virtual decomposition diagram, with the corresponding multiplicities.
Before giving the definition of i-virtual composition factors, we shall
illustrate the above definitions with a few examples concerning alcoves.
For SL 2 , there is only one p 2-alcove in the set of p 2-restricted weights.
 DISTANT WEYL MODULES 455

This corresponds under scaling to the unique p-restricted alcove C1 , whose

associated decomposition diagram is just C1. Thus the p 2-decomposition
diagram associated to ␧ 2 Ž C1 . is just ␧ 2 Ž C1 ., and the virtual decomposition
diagram associated to a p-alcove F in the set of p 2-restricted weights
consists of those E for which

c F E s d F E / 0.

More generally, for any group G, the p 2-alcove C2 is its own p 2-decom-
position diagram, and hence for any p-facet F in C2 we have c F E s d F E .
Thus, for any weight in C2 , the virtual composition factors are just the
usual composition factors of the associated Weyl module.
For a non-trivial example, consider SL 3 , with p s 5, and a weight ␭ in
an alcove just above the lowest p 2-alcove Žas shown in Fig. 1a.. Now ␭ lies
in ␧ 2 Ž D ., where D is the upper alcove in the set of p-restricted weights.
The decomposition diagram associated to D is just D and C1 , each with
multiplicity one. Thus for the p-alcove F containing ␭ Žand any p-facet E .
we have
c F E s d F E y d D C 1 c JE s d F E y c J E s d F E y d JE ,

where J is the image of F in ␧ 2 Ž C1 . under Wp 2 . For ␭ in F as above, the

p-alcove J is that containing ␮ in Fig. 1b. The decomposition diagrams for
F and J are given in Figs. 1a and 1b respectively, and so the virtual
decomposition diagram associated to F is that given in Fig. 1c. ŽAll
multiplicities are 1 unless otherwise indicated.. The virtual composition
factors associated to ␭ are just those weights in Wp . ␭ lying in this final
diagram.
Returning to our definitions, we next associate to each p iq1-restricted
weight ␭ a set of i-virtual composition factors Žwith multiplicities.. When
i s 1 these will just be the virtual composition factors defined above.
Regard the p i-facets as p-facets by scaling. Then the p i-facet ␧ i Ž F .
containing ␭ corresponds to the p-facet F in the set of p 2-restricted
weights. We have already associated to F a virtual decomposition diagram.

FIGURE 1
456 ANTON COX

By scaling, we obtain a corresponding set of p i-facets with multiplicities.

This is the i-¨ irtual decomposition diagram associated to ␧ i Ž F ., and the
i-¨ irtual composition factors associated to ␭ are just those weights in Wp i . ␭
that lie in these p i-facets. We shall denote the corresponding multiplicity
of such a weight ␮ by c␭␮ i
.
Finally, we shall associate a set of i-virtual composition factors to an
arbitrary dominant weight ␭. Any such weight can be uniquely written in
the form ␭ s ␭X q p iq1␭Y with ␭X g X iq1ŽT .. Now the i-virtual composition
factors associated to ␭ are just those weights of the form ␮ s ␮X q p iq1␭Y ,
where ␮X runs over the set of i-virtual composition factors of ␭X , and
i
c␭␮ s c␭i X␮X .
We will show that, for p G h, the following algorithm completely deter-
mines the decomposition numbers for a given Weyl module ⌬Ž ␭..
ALGORITHM 2.1. Ž1. Let i be maximal such that ␭ does not lie in Ci
and let cfŽ ␭, i q 1. s  ␭4 . ŽThus ␭ lies in the set of p iq1-restricted weights..
Ž2. If i s 0 then we are done; otherwise continue with Step 3.
Ž3. For each weight ␮ in cfŽ ␭, i q 1. Žand keeping track of multiplic-
ities determine the set of i-virtual composition factors associated to ␮.
.
Ž4. Let cfŽ ␭, i . be the disjoint union of all the sets of virtual
composition factors Žwith multiplicities. obtained during Step 3.
Ž5. Set i s i y 1 and repeat from Step 2.
To illustrate the above algorithm we shall consider an example for SL 3 .
Let p s 5 and consider the Weyl module ⌬Ž181, 0, 0. for GL 3 Žwhere we
use the usual partition labelling for polynomial dominant weights.. This
Weyl module is just the contravariant dual of a symmetric power of the
natural module, and so its composition factors can be calculated using the
results in w8x.
The various iterations of the algorithm and the final result are shown in
Fig. 2. Note that the dotted lines indicate p-walls while the thicker lines
indicate higher powers of p. As each p-alcove can contain a unique
composition factor of any module, we merely indicate the alcove in which
each factor lies. The small diagrams indicate the various virtual decompo-
sition diagrams that arise during each iteration of the algorithm. For
future reference we have labelled the composition factors that arise; for
example, ␭ corresponds to the label d Ž i ..
We begin by considering the p 3-facets. After scaling, the p 3-facet
containing ␭ corresponds to the upper shaded p-alcove shown in Iteration
1. Thus the first iteration of the algorithm produces ␭ and the element of
Wp 3 . ␭ lying just above the lowest p 3-alcove Žindicated by a solid arrow..
After the second iteration, the virtual decomposition diagrams given in
 DISTANT WEYL MODULES 457

FIGURE 2

Iteration 2 arise, and after scaling Žand translation for those associated to
␭, since this is not a p 3-restricted weight. these give weights in the alcoves
indicated by dotted arrows. The final iteration uses the virtual decomposi-
tion diagrams shown in Iteration 3 to produce the set of shaded p-alcoves
shown in the central diagram.
Our main result is

THEOREM 2.2. Suppose that p G h. Gi¨ en a dominant weight ␭, the set of

composition factors of ⌬Ž ␭. Ž counted with multiplicities. is precisely the set
cfŽ ␭, 1. obtained from Algorithm 2.1.

Proof. First suppose that ␭ g X 2 ŽT .. We will show that the sum of the
characters of the virtual composition factors obtained from the algorithm
is ␹ Ž ␭., as required. We first note that if ␭ g C2 , then the set of virtual
composition factors associated to ␭ is precisely the full set of composition
factors of ⌬Ž ␭. Žby our earlier remarks., and hence the result is immediate.
458 ANTON COX

If ␭ g X 2 ŽT . does not lie in C2 then as noted at the start of the section

we must have ␭ g C3 . Thus there are two iterations when the algorithm is
applied to ␭, and the set cfŽ ␭, 2. is just the set of composition factors
arising from the p 2-decomposition diagram. After the second iteration the
multiplicity of ␮ in cfŽ ␭, 1. is just

Ý d H I c␶␮ s c␭␮ q Ý d H I c␶␮ ,

␶ ␶- ␭

where ␭ g ␧ 2 Ž H ., ␶ g Wp 2 . ␭, and ␶ g ␧ 2 Ž I .. Comparing this with Ž2., we

see that this equals d␭␮ as required.
Now suppose that ␭ is a p iq1-restricted weight with ␭ f Ci . We set
j s i y 1 if ␭ g X i ŽT ., and j s i otherwise. The first iteration of our
algorithm begins at level i, and cfŽ ␭, i q 1. s  ␭4 . We claim that, by
Steinberg’s tensor product theorem, it is enough to show that

Ž ch L Ž ␮Y . F . X
j
␹ Ž ␭. s Ý ž ␹Ž ␮ . q Ý a␮j X ␶ ␹ Ž ␶ . / Ž 3.
X
␮ gcfŽ ␭ , j . ␶- ␮

where ␮ s ␮X q p j␮Y and the a␮j X ␶ are defined in the following manner.
For E and F p-facets in X 1ŽT ., we define a1F E by choosing ␭ g F and
solving

ch L Ž ␭ . s Ý a1F E ␹ Ž ␮ E .
EFF

where ␮ E s ŽWp . ␭. l E. For weights ␮X g ␧ i Ž F . and ␶ g ␧ i Ž E . in X i ŽT .,

we define

a1F E if ␶ g Wp i . ␮X
a␮i X ␶ s ½ 0 otherwise.

For the claim, note that for each ␮ on the right-hand side of Ž3. we have
by our choice of j that ␮X - ␭. Thus by induction our algorithm gives the
character of ␹ Ž ␮X . Žand of ␹ Ž␶ . for all ␶ - ␮X ., but possibly starting at
level j. The effect of starting at level j y 1, as is the case when calculating
for ␮ g cfŽ ␭, j ., is to lose the elements descended from those elements of
cfŽ ␮ , j . not equal to ␮. After scaling, it can be seen that this corresponds
to calculating decomposition numbers for the simple module in the corre-
sponding p-alcove in X 2 ŽT . rather than of the Weyl module. Hence by the
induction hypotheses and the definition of the a␮j X ␶ it is enough to show Ž3..
 DISTANT WEYL MODULES 459

Let ch LŽ ␮Y . s Ý␯ m␮Y␯ eŽ ␯ .. Then we wish to show that

Ý Ž ch L Ž ␮Y . F . Ý
j

ž a␮j X ␶ ␹ Ž ␶ . /
X
␮ ␶F ␮

s Ý Ý m␮ ␯ e Ž p j␯ .
ž Y
/ž Ý a X
j
␮X ␶ ␹ Ž␶ . /
␮ ␯ ␶F ␮

s Ý Ý m␮ ␯ Ý Y
ž a␮j X ␶ ␹ Ž ␶ q p j␯ . /
␮ ␯ ␶F ␮X

l Ž w␶␯ .
s Ý Ý m␮ ␯ Ý Y
ž a␮j X ␶ Ž y1 . ␹ Ž w␶␯ Ž ␶ q p j␯ . . /
X
␮ ␯ ␶F ␮

is equal to ␹ Ž ␭. Žwhere we write w␶␯ for w␶qp i␯ for brevity.. This expres-
sion is of the form

␹ Ž ␭. q Ý f␭␪ ␹ Ž ␪ .
␪- ␭

for some coefficients f␭␪ Žwhere all the weights ␪ are dominant..
Now when j s 1 all these coefficients are zero by the calculation above
and the linear independence of the characters of Weyl modules. But in
general the f␭␪ depend only on the p j␯ , reflections about the boundaries
of the dominant region, and the combinatorics of p j-facets regarded as
p-facets. Thus, after fixing an appropriate power of the Frobenius mor-
phism, these coefficients depend only on the combinatorics of the p j-facets
regarded as p-facets and the weights of the LŽ ␮Y .. As for j s 1 all the f␭␪
are zero; the same must be true for all j by scaling.
There is another recursive formula which can be used for determining
the composition factors of Weyl modules, due to Jantzen w13, 3.1 Satzx. For
any weight ␭ g X ŽT . we have
Fr
␹ Ž ␭. s Ý Ý ˆ r Ž p r␮Y q ␮X . ␹ Ž ␮Y .
Zˆr Ž ␭ . : L ch Ž L Ž ␮X . . ,
Y X
␮ gX ŽT . ␮ gX r ŽT .

Ž 4.

where the Zˆr Ž ␭. are certain coinduced modules for the Jantzen subgroup
Gr T of G and the L ˆ r Ž ␭. are the corresponding simple modules Žsee w14,
Section II, Chap. 9x for details.. Suitably far away from the walls of the
dominant region all of the weights in this sum are dominant, and the
equality corresponds to a filtration of H 0 Ž ␭. with factors of the form
LŽ ␮X . m H 0 Ž ␮Y . F Žsee w14, Section II 9.11, Propositionx.. However, near
r
460 ANTON COX

the boundary of the dominant region the ␮Y will not all be dominant, and
although Ž4. can be modified using Lemma 1.2 the coefficients will now no
longer all be positive.
Although the virtual composition factors associated to a single weight ␮
arising during Algorithm 2.1 may also Žin principle. have negative multi-
plicities, we have
LEMMA 2.3. For any dominant weight ␭, the multiplicities of the elements
of cfŽ ␭, i . obtained after each iteration of Algorithm 2.1 are all positi¨ e.
Proof. Suppose there is some ␭ for which the lemma fails, and let i be
as in Algorithm 2.1. Then there exists some j F i such that the set F of
j-virtual composition factors obtained when considering p j-facets includes
some negative multiplicities. Choose ␭X such that it lies in the p-facet
corresponding to the p iyjq1-facet containing ␭ Žunder scaling.. Then the
set of composition factors of H 0 Ž ␭X . obtained using the algorithm will
correspond Žunder scaling. to those in F and have the same multiplicities.
But these multiplicities are all positive, which gives the desired contradic-
tion.
For there to be any possibility of obtaining a filtration of H 0 Ž ␭.
associated to our algorithm, it is clearly necessary that all the virtual
composition multiplicities associated to a given weight are positive. We
shall return to this question in Section 4. First, however, we shall exploit
the easy invertibility of our algorithm.

3. RINGEL DUALITY AND REPRESENTATIONS OF THE

SYMMETRIC GROUP

In this section we restrict our attention to the case where G is the

general linear group GL n . Although not itself semisimple, its representa-
tion theory can be deduced easily from that of SL n , and so the results from
the previous section apply. Associated to G are the Schur algebras
S s SŽ n, d ., and in w10x Erdmann showed how their representation theory
can be related to that of the symmetric group ⌺ d by Ringel duality. We
shall very briefly review this relationship Ždetails and further references
can be found in w10x. and show how the invertibility of Algorithm 2.1 allows
us to deduce certain results concerning representations of the symmetric
group. Throughout this section we shall assume that p ) n.
The category of S Ž n, d .-modules is naturally equivalent to the category
of SL n-modules all of whose composition factors LŽ ␭. satisfy ␭ s Ý i a i ␻ i
with Ý i iai s d y jn for some j G 0. As SŽ n, d . is a quasi-hereditary
algebra, there exists a certain characteristic module T for S, and we call
 DISTANT WEYL MODULES 461

the endomorphism algebra SX s End S ŽT . the Ringel dual of S. In fact Žup

to Morita equivalence., we can identify SX precisely:
THEOREM 3.1. Suppose that p ) n. Then the Ringel dual of SŽ n, d . is
Morita equi¨ alent to a certain Ž known. quotient of k ⌺ d , the group algebra of
the symmetric group on d symbols.
Proof. This is a special case of the first part of w10, Theorem 4.4x.
Let ⌳qŽ n, d . be the set of n-part partitions of d. To each ␭ g ⌳qŽ n, d .
we can associate a corresponding permutation module M ␭ for k ⌺ d and
certain explicitly defined submodules S ␭ of M ␭, called Specht modules.
The indecomposable direct summands of M ␭ are called Young modules,
and we define Y ␭ to be the unique such summand of M ␭ containing S ␭.
It is shown in w6, Ž2.6.x that Young modules have a Specht module
filtration, and we shall denote the multiplicity of S ␮ in some such filtration
of Y ␭ by Ž Y ␭ : S ␮ .. It is easy to see Žcf. w20, 4.10 Corollaryx. that if ␭ has at
most r non-zero parts, so also must ␮ for any S ␮ arising in such a
filtration of Y ␭. Indeed, by general results on Ringel duals and the explicit
identifications made in w10x, we have
PROPOSITION 3.2. Suppose that p ) n. Then for all ␭, ␮ g ⌳qŽ n, d . we
ha¨ e

Ž Y ␭ : S ␮ . s ⌬ Ž ␮ . : L Ž ␭. .

Proof. See w10, 4.4 Theoremx and w7, Lemma A4.6x.

Thus to determine the Specht modules arising in a given Young module
Y ␭ for k ⌺ d , it is enough to determine all Weyl modules containing the
simple module LŽ ␭. for SL n with highest weight in a certain bounded set
of weights. Here n can be taken to be the number of non-zero parts of ␭.
The advantage of our algorithm for computing decomposition numbers is
that it can easily be run in reverse. Starting with a given weight ␭ and the
initial data on Weyl modules corresponding to p 2-restricted weights, it is
easy to determine those weights Žwith multiplicities. which give rise to ␭
after one iteration of Algorithm 2.1. Iterate this procedure by determining
for each weight obtained at the ith stage a corresponding set Žwith
multiplicities. of new weights in a similar way Žby regarding p i-facets as
p-facets.. Thus Žas h s n. we obtain
PROPOSITION 3.3. Suppose that p ) n. Gi¨ en a weight ␭ g ⌳qŽ n, d ., we
can in¨ ert Algorithm 2.1 to gi¨ e an algorithm for determining Ž Y ␭ : S ␮ . for all
␮ g ⌳qŽ n, d ., from the decomposition numbers for Weyl modules for SL n
with p 2-restricted weights.
462 ANTON COX

4. LIFTING FROM THE QUANTUM GROUP

Although they arise naturally in the algorithm of Section 2, it is not yet

clear that the sets of virtual composition factors have any representation-
theoretic interpretation. In particular, it is not even clear that they have
non-negative multiplicities. In this section we shall realize them as the sets
of composition factors associated with G-modules obtained by lifting from
a corresponding quantum groupᎏat least when p is large enough for the
Lusztig conjecture to hold. We then discuss a possible connection with a
long-standing conjecture of Humphreys on the structure of Weyl modules.
The modules we require arise in the work of Lin w16x and are generalisa-
tions of certain modules considered by Lusztig. The constructions in this
section are based on w16, Section 2x, to which we refer the reader for
further details.
We begin by defining the various quantum algebras that we require. Let
Uq be the quantised enveloping algebra over ⺓Ž q . corresponding to the
root system R. If we set A s ⺪w q, qy1 x ; ⺓Ž q ., then Lusztig w19x has
constructed a certain A-form UA of Uq . For ␰ a fixed primitive p r th root
of unity, ⺓ becomes an A-algebra via the homomorphism taking q to ␰ .
We denote by U␰ the corresponding algebra ⺓ mA UA .
Setting Br to be the localisation of ⺪w ␰ x at the ideal Ž ␰ y 1., we have
that Br is a discrete valuation ring and UBr s Br mA UA is a Br-form
inside U␰ . Finally, for k an algebraically closed field of characteristic p, the
natural homomorphism A ª k taking q to 1 factors through the homo-
morphism B ª k taking ␰ to 1. We obtain an isomorphism of k-algebras
Uk [ k mA UA ( k mBr UBr .

We next wish to define various modules for each of these algebras,

following Lusztig w17᎐19x. For each dominant weight ␭ there exists a
unique finite-dimensional irreducible Uq-module L q Ž ␭. of type 1 with
highest weight ␭. If we fix some vector ¨␭ generating this module, then
L A Ž ␭. s UA ¨␭ is a UA -invariant A-lattice in L q Ž ␭.. We set ⌬ p r Ž ␭. s ⺓ mA
L A Ž ␭., the quantum Weyl module for U␰ . This has a unique simple
quotient, which we denote by L p r Ž ␭.. We denote the image of our
generating vector ¨␭ in this quotient by ¨␭ .
Now L BrŽ ␭. s UBr¨␭ is a Br-lattice in L p r Ž ␭., and so L p r Ž ␭ . s k mBr
L Br Ž ␭. is a Uk-module. Indeed, Lusztig has shown that this is even a
Žrational. G-module. The Weyl module for G can also be constructed
using Uq ; we have
⌬ Ž ␭ . ( k mA L A Ž ␭ . ( k mBr ⌬ BrŽ ␭ . ,

where ⌬ BrŽ ␭. s UBr¨␭ is a Br-lattice inside ⌬ p r Ž ␭..

 DISTANT WEYL MODULES 463

Our construction gives that L p r Ž ␭ . is a quotient of ⌬Ž ␭., and we wish to

understand the structure of this module. By w16, Theorem 2.7x, we have
that
Fr
L p r Ž ␭ . ( L p r Ž ␭X . m ⌬ Ž ␭Y . ,

where ␭ s ␭X q p r␭Y with ␭X g X r ŽT . and F denotes the usual Frobenius

morphism. Thus we first need to understand the structure of L p r Ž ␭ . for all
p r-restricted weights.
In w16x, Lin has investigated the decomposition patterns of these mod-
ules for suitably ‘‘generic’’ weights ␭. ŽRoughly, this means that ␭ is
suitably far away from any p i-walls for certain i; see w16, p. 286x for the
precise definition.. For arbitrary dominant weights ␭ and ␮ we have

⌬ Ž ␭. : L Ž ␮ . s Ý ⌬ p r Ž ␭. : L p r Ž ␯ . Lp r Ž ␯ . : LŽ ␮ . . Ž 5.
␯

It is this identity that will allow us to relate our algorithm to the decompo-
sition patterns of the L p r Ž ␭ . .
For the rest of this section, we shall assume that Lusztig’s conjecture
Žfor algebraic groups. holds for our choice of G and p and that p G 2 h y
2. For fixed G this is known to be the case if we take p to be sufficiently
large by the results in w1x.
A dominant weight ␭ satisfies the Jantzen condition if ² ␭ q ␳ , ␣ k: F
pŽ p q 2 y h. for all ␣ g Rq. For such ␭ we have

⌬ p Ž ␭. : L p Ž ␮ . s ⌬ Ž ␭. : L Ž ␮ .

and we see by induction using Ž5. that L p Ž ␭ . ( LŽ ␭. is irreducible.

More generally, we shall say that a dominant weight ␭ satisfies the ith
Jantzen condition if

² ␭ q ␳ , ␣ k: F p i Ž p q 2 y h .

for all ␣ g Rq and denote the set of such ␭ by Ji ŽT .. Note that ␭ satisfies
the Jantzen condition if and only if ␧ i Ž ␭. satisfies the ith Jantzen condi-
tion. For ␭ g Ji ŽT . we have

dFE if ␮ g Wp i . ␭
⌬ p i Ž ␭. : L p i Ž ␮ . s ½ 0 otherwise,
Ž 6.

where ␭ g ␧ i Ž F . and ␮ g ␧ i Ž E .. As p G 2 h y 2 we have that X i ŽT . :

Ji ŽT ., and hence for p 2-restricted weights the p 2-decomposition diagrams
defined in Section 2 are just the decomposition diagrams for the corre-
sponding quantum group at a p 2 root of unity.
464 ANTON COX

We next define some more sets of weights associated with Algorithm 2.1.
It will be convenient to set cfŽ ␭, i . s  ␭4 whenever ␭ g Ci , extending our
earlier notation. Given a weight ␭ and an integer i G 1, we define
DescŽ ␭, i ., the set of descendents of ␭ at le¨ el i, in the following manner.
First note that for any weight ␭ and integer i we have, by construction,
␭ g cfŽ ␭, i ., with multiplicity one, and all other weights ␮ g cfŽ ␭, i . satisfy
␮ - ␭. We define DescŽ ␭, i . by induction on ␭ via
ⴢ
cf Ž ␭ , 1 . s D Desc Ž ␮ , i . ,
␮ gcfŽ ␭ , i .

where as usual we run over elements of the index set counted with
multiplicities.
As an example, consider the weight denoted ␭ in Fig. 2. For i G 4 we
have Desc Ž ␭ , i . s cf Ž ␭ , 1 . , the set of composition factors
of ⌬Ž ␭.. The set DescŽ ␭, 3. is precisely the set of those weights with labels
of the form d Žy., eŽy., or f Žy., while DescŽ ␭, 2. s  ␭ s d Ž i ., d Ž ii .4 and
DescŽ ␭, 1. s  ␭4 .
The main result of this section is
THEOREM 4.1. Suppose that p G 2 h y 2 is such that the Lusztig conjec-
ture is satisfied for G. If ␭ g Ji ŽT . then the set of composition factors, with
multiplicities, of L p i Ž ␭ . equals DescŽ ␭, i ..
Proof. We proceed by induction on ␭. If ␭ lies in Ci , then cfŽ ␭, i . s  ␭4
and hence DescŽ ␭, i . s cfŽ ␭, 1.. But in this case, as w ⌬ p i Ž ␭. : L p i Ž ␮ .x s ␦␭␮
we see from Ž5. that L p i Ž ␭ . s ⌬Ž ␭., and so we are done by Theorem 2.2.
Now suppose that ␭ f Ci . As Ji ŽT . : Ciq1 , we see that Algorithm 2.1
begins at level i and that cfŽ ␭, i q 1. s  ␭4 . After scaling by ␧ i , the p i-facet
containing ␭ corresponds to some p-facet F in C2 . By our remarks before
Algorithm 2.1, this implies that the virtual decomposition diagram associ-
ated to F is just the usual decomposition diagram for F. Hence after
scaling by ␧ i we see that

dFE if ␮ g Wp i . ␭
i
c␭␮ s ½ 0 otherwise,
where ␭ g ␧ i Ž F . and ␮ g ␧ i Ž E .. But by Ž6. this equals w ⌬ p i Ž ␭. : L p i Ž ␮ .x.
Hence we have by Theorem 2.2 that the composition factors of ⌬Ž ␭. are
given by cfŽ ␭, 1., where
ⴢ
cf Ž ␭ , 1 . s D ⌬ p i Ž ␭ . : L p i Ž ␮ . Desc Ž ␮ , i . ,
␮
 DISTANT WEYL MODULES 465

where ␮ runs over the set of dominant weights. By induction, we see by

comparing with Ž5. that DescŽ ␭, i . is precisely the set of composition
factors of L p i Ž ␭ . , as required.
Recall that for p G 2 h y 2 we have X i ŽT . ; Ji ŽT .. Thus, for suitably
large p, our algorithm gives a means of calculating the composition factors
of L p i Ž ␭ . for all p i-restricted weights, given the decomposition numbers
for ⌬Ž ␮ . for p 2-restricted weights ␮. In particular, for any p 2-restricted
weight ␭ we have that DescŽ ␭, 2. is just the set of virtual composition
factors associated to ␭, and hence we obtain
COROLLARY 4.2. Suppose that p G 2 h y 2 is such that the Lusztig con-
jecture is satisfied for G. Then, for any p 2-restricted weight ␭, the ¨ irtual
decomposition numbers c␭␮ are all non-negati¨ e.
Recall from Section 1 that for any weight ␭ we may define modules
H j Ž ␭. by considering the right derived functors of ind GB . In w11x, Humphreys
has conjectured that any Weyl module ⌬Ž ␭. should have a filtration with
i
quotients of the form H j Ž ␮ . m LŽ ␯ . F , where ␮ is W-conjugate to a
weight in X i ŽT . and ␯ is dominant. On the level of characters, this would
imply that ⌬Ž ␭. has a filtration where each quotient has composition
factors clustered around a certain translate of X i ŽT . Žrelated to the weight
␯ defined above..
Now suppose that there exists a filtration of ⌬Ž ␭. associated to the ith
level in Algorithm 2.1; that is, a filtration whose successive quotients have
sets of composition factors of the form DescŽ ␮ , i . for elements ␮ g cfŽ ␭, i ..
On the level of characters, this would imply that ⌬Ž ␭. has a filtration
where each quotient has composition factors clustered around a certain
p i-alcove Žrelated to the weight ␮ .. In proving Ž5., Lin constructs an
associated filtration of the Weyl module, which by Theorem 4.1 is compati-
ble with the first iteration of Algorithm 2.1.
If we consider the case of SL 3 , with ␭ suitably ‘‘generic’’ in the lowest
p 3-alcove Žas in w11, Fig. 1x., it is easy to verify that the clusters of
composition factors arising from the final iteration of our algorithm
correspond to the Žconjectural. filtrations of ⌬Ž ␭. given by Humphreys in
w11, pp. 2672᎐2674x. Indeed, we have
THEOREM 4.3. Suppose that ␭ is a dominant weight for SL 3 such that all
composition factors of Zˆ1Ž ␭. lie in the same p 2-alco¨ e. Then there is a
filtration of ⌬Ž ␭. corresponding to the final iteration of Algorithm 2.1.
¨
Proof. This is an easy consequence of a theorem of Kuhne-Hausmann
w15, Kapitel VI, Satz 2x.
¨
More generally, Kuhne-Haussman has calculated the submodule struc-
ture of all Weyl modules for SL 3 which are multiplicity-free. For the
466 ANTON COX

examples given in w15, pp. 174᎐176x ᎏwhich do not satisfy the hypotheses
of Theorem 4.3ᎏit is easy to verify that there is still a filtration associated
to our algorithm.
If we assume that Humphreys’ conjecture holds, the above remarks give
some evidence that there may be a refinement of his filtration associated
to the corresponding level of Algorithm 2.1. In the next section we shall
consider further evidence for the existence of such a filtration.

5. A SOCLE SERIES FOR CERTAIN INDUCED MODULES

In this section we shall show how the facet combinatorics introduced in

Section 2 can be used to give a new description of certain results of Doty
in w8x. In particular, we shall give a filtration Žand for weights in alcoves a
description of the socle series. of symmetric powers of the natural module
for SL 2 and SL 3 . These are isomorphic to induced modules of the form
H 0 Ž d ␻ 1 ., which in turn are just the contravariant duals of the correspond-
ing Weyl modules Žand have the same composition factors.. Thus this
filtration will provide an interpretation in terms of module structure of the
character-theoretic result in Section 2.
These results could also be deduced from w3x for SL 2 and from w15,
Kapitel VI, Satz 1x for SL 3 . However, we prefer to consider the methods of
Doty as these are both simpler and more accessible to generalisation to
groups of higher rank.
We begin by recalling the combinatorial framework outlined in w8x. Let
S d Ž V . denote the dth symmetric power of the natural representation V of
GL n . We write d s Ý iMG 0 d i p i, with 0 F d i F p y 1 for all i, and d M / 0.
Given a monomial x b s x 1b 1 ⭈⭈⭈ x nb n g S d Ž V . we write bi s Ý jG 0 bi j p j with
0 F bi j F p y 1 for all i and j. Then we can associate to each monomial
x b an M-tuple of non-negative integers Ž c1Žb., . . . , c M Žb.., where the ele-
ments c k Žb. are defined by the equations

Ý Ý bi j p j s c k Ž b. p k q Ý d j p j.
iG0 j-k j-k

An equivalent set of defining equations is given by

Ý bi k s d k q pckq1 y c k . Ž 7.
iG0

We let C Ž d . be the set of M-tuples arising from some x b g S d Ž V .. This

set can be given a lattice structure by setting c F cX if and only if c k F cXk
for all k. Then the main result in w8x is
 DISTANT WEYL MODULES 467

THEOREM 5.1. There is a lattice isomorphism between the lattice of order

closed subsets of C Ž d . and the G-submodules of S d Ž V . Ž ordered by inclusion..
In particular, the composition factors of S d Ž V . are in one-to-one correspon-
dence with the elements of C Ž d ..
Further, by considering Ž7., we can determine explicitly the highest
weight vector of the composition factor corresponding to c as follows. Let
x b be the highest weight vector corresponding to c. Then b is given by the
set of equations

bi k s max  min  d k q p Ž c kq1 q 1 y i . y c k q i y 1, p y 1 4 , 0 4 , Ž 8.

where the d k and bi k are as defined above. Thus, given the set C Ž d .
corresponding to d, we can determine the composition factors of S d Ž V ..
The final result from w8x that we shall need is the following criterion for
recognising elements of C Ž d ..
LEMMA 5.2. An M-tuple c of non-negati¨ e integers is an element of C Ž d .
if and only if it satisfies the equations

0 F ck F Ý d j p jyk
jGk

and
0 F d k q pc kq1 y c k F n Ž p y 1 .

for all k, where we set c j s 0 for all j F 0 and j ) M.

We begin by analysing the structure of the lattice C Ž d .. First we note
that by induction Žusing Lemma 5.2. it is clear that we must have
0 F c k F n y 1 for each k. Given c g C Ž d ., we set <c < s Ý iG 0 c i . We shall
say that such an element c lies in the jth socle layer Soc j Ž C Ž d .. of C Ž d . if
the longest chain of elements strictly below c is of length j y 1.
LEMMA 5.3. An element c g C Ž d . lies in the jth socle layer if and only
if <c < s j.
Proof. Any chain of elements strictly below c must have length at most
<c <, and so it is enough to show that a chain of this length in fact exists. For
this, it will be enough to show that for each c / 0 there exists some cX - c
in C Ž d . such that for some k 0 we have cXk 0 s c k 0 y 1 and that cXi s c i for
all i / k 0 .
We first claim that if c / 0 then there exists some k such that 0 F d k q
pc kq 1 y c k - nŽ p y 1. and c k / 0. For otherwise take k 1 minimal such
that c k 1 / 0. Then we must have d k 1 q pc k 1q1 y c k 1 s nŽ p y 1. G p, and
hence that c k 1q1 ) 0. In a similar manner we deduce that c k ) 0 for all
468 ANTON COX

k G k 1 Žand that for all such k we must have d k q pc kq1 y c k s nŽ p y 1...

But this contradicts the condition that c Mq 1 s 0, and so the claim follows.
Taking k 0 to be minimal satisfying the above hypotheses, it is now an
easy exercise using the inequalities in Lemma 5.2 to verify that the
element cX defined above is in C Ž d . as required.
Most of the rest of this section will be devoted to proving
PROPOSITION 5.4. For GL 2 and GL 3 , the modules S d Ž V . ha¨ e filtra-
tions corresponding to the ¨ irtual decomposition factors obtained at any gi¨ en
stage in Algorithm 2.1. Moreo¨ er, for weights in the interior of alco¨ es, the
socle series for S d Ž V . can be constructed in the course of implementing the
algorithm.
Proof. To each composition factor LŽ ␭. we have associated a carry
pattern cŽ ␭.. We shall show that the iteration of Algorithm 2.1 correspond-
ing to p i-facets changes the value only of the first i elements of any c
obtained so far and that each virtual decomposition factor obtained at this
stage from a given weight corresponds to a different value of c i . In fact, for
weights in the interior of an alcove we shall show that at this stage the first
i elements of any carry pattern obtained so far are all equal. By Lemma
5.3, we shall thus obtain complete information on the socle series of the
symmetric powers corresponding to weights in the interior of an alcove.
We begin with an easy example to illustrate this result for SL 2 . ŽWe
postpone analysis of the socle series for the SL 3 example given in Fig. 2
until later in this section.. Let p s 3 and d s 43. On applying Algorithm
2.1 we obtain the set of composition factors labelled a through f in Fig. 3.
For SL 2 , the p 2-restricted weights are precisely those in the upper
closure of the lowest p 2 alcove. Thus the virtual composition factors
associated to a weight are just the ordinary composition factors of the
corresponding Weyl module. This is simple for weights on walls or in the
lowest p-alcove and has two composition factors corresponding to the
highest weight and its reflection about the p-wall immediately below it for
weights in the remaining p-alcoves. As in this case the various c i can only
be 0 or 1; the corresponding socle series predicted by our result will be as

FIGURE 3
 DISTANT WEYL MODULES 469

FIGURE 4

shown in Fig. 4a Žwith the actual submodule lattice obtained using the
results in w8x given in Fig. 4b..
We now return to the proof of Proposition 5.4. The submodule structure
for the symmetric powers has already been given in w8x, and we merely
verify that the results given there can be converted into the required form.
For this we need to know the relative positions of the virtual composition
factors for certain p 2-restricted weights. It is for this reason that we
restrict ourselves to considering SL 2 and SL 3 , although we conjecture that
the result should hold for SL n without restriction on n.
In order to convert the results from w8x for GL n into a form compatible
with the facet geometry, we use the usual change of coordinates ␭ s
Ž ␭1 , . . . , ␭ n . ¬ ␭ s Ž ␭1 , . . . , ␭ ny1 ., where we set ␭ i s ␭ i y ␭ iq1. This now
gives the coordinates of the corresponding SL-weight in terms of the basis
of fundamental weights.
We begin by considering the SL 2 case. Let b s Ž b1 , b 2 . be the highest
weight vector in some composition factor of S d Ž V .. By the preceding
remarks and Theorem 2.2 this is either Ž d, 0. or b is a reflection of some bX
about a p i-wall for some weight bX generated at an earlier stage in the
algorithm Žwhere we take i to be maximal with this property.. Let the
corresponding elements of C Ž d . be c and cX respectively.
It will be enough to show that c k s 1 y cXk when 1 F k F i and c k s cXk
otherwise. Suppose that bX s ap i y 1 q b with 0 - b - p i and a k 0
mod p. Then it will suffice to show that the weight x corresponding to the
desired value of c satisfies x q bX s 2 ap i y 2 Žas then x must equal b.. By
considering the various possible values of bXi k arising from Ž8. we see that

¡d k if Ž c k , c kq1 . s Ž 0, 0 .

bXk s bX1 k y bX2 k s ~p y d k y2 if Ž c k , c kq1 . s Ž 0, 1 .

Ž 9.
dk y 1 if Ž c k , c kq1 . s Ž 1, 0 .
¢p y 1 y d k if Ž c k , c kq1 . s Ž 1, 1 . .

Clearly, we must have Ž c 0 , c1 ., Ž cX0 , cX1 .4 s Ž0, 0., Ž0, 1.4 , while for 1 F t
F i y 1 we have Ž c t , c tq1 ., Ž cXt , cXtq1 .4 s Ž0, 0., Ž1, 1.4 . Further, by our in-
470 ANTON COX

ductive hypothesis we must have Ž c i , c iq1 ., Ž cXi , cXiq1 .4 s Ž0, 1., Ž1, 1.4 or
Ž1, 0., Ž0, 0.4 . Using Ž9. it is now easy to see that
iy1
Ý Ž bXk q x k . p k s p y 2 q Ý Ž p y 1. p t q Ž 2 a y 1. p i s 2 ap i y 2,
k ts1

as required. This completes the proof in the SL 2 case.

The proof for SL 3 proceeds in a similar manner. We shall first consider
the case where the initial weight lies in the interior of an alcove and show
that, when considering p t-alcoves, the set of virtual composition factors
associated to a given weight contains at most one of each of the three
types of alcove shown in Fig. 5a. Further, we show that the image of the
given weight under Wp t in each of these alcoves corresponds to the carry
pattern whose first t elements are equal to the integer labeling that
diagram and that the remaining elements of the carry pattern are fixed for
these alcoves.
Using Ž4. and the known decomposition numbers for the Zˆ1Ž ␭. for SL 3
w12x, it is easy to show that the various possible patterns of virtual
composition factors associated to weights outside the lowest p-alcove are
as shown in Fig. 5b.
It just remains to consider the various possible cases, calculate the
weights corresponding to the predicted carry patterns, and verify that the
differences between them correspond to the relative positions of the
various weights shown in each case. This is a routine Žif lengthy. exercise
using the expression for the bi k given in Ž8..
If the initial weight does not lie in an alcove, then the strong version of
the inductive hypothesis Žasserting the precise form of the corresponding
carry patterns. is no longer satisfied. However, by similar arguments one

FIGURE 5
 DISTANT WEYL MODULES 471

FIGURE 6

shows that at every stage the virtual composition factors correspond to

distinct values of an appropriate c t and that the c tX Žfor tX ) t . remain
constant. This completes the proof of Proposition 5.4.
To illustrate the SL 3 case, we return to the example considered in Fig. 2.
In Fig. 6a we give the socle series constructed in the proof of Proposition
5.4, while in Fig. 6b we give the corresponding submodule lattice calcu-
lated using w8x. The labels on the left are those used in Fig. 2, while on the
right the corresponding elements of C Ž d . are given.

6. THE QUANTUM MIXED CASE

To conclude, we consider the quantum general linear group q-GLŽ n, k .,

defined by Dipper and Donkin w5x, in the case when q is a primitive lth
root of unity and k has characteristic p ) 0. Algorithm 2.1 can easily be
modified to give a corresponding algorithm in this context by replacing
p i-facets with lp iy1-facets and taking as the initial dataset the decomposi-
tion numbers for all lp-restricted weights.
Now the same arguments as those in Sections 1 and 2 can be applied to
show that this gives the composition factors of the quantum Weyl module
corresponding to ␭. For the results on the geometry of facets it is
sufficient to require that both l and p are at least as big as the Coxeter
number, which in this case is n. The general theory reviewed in Section 1
is given in the quantum case in w7x and w4x.
Again, this has applications via Ringel duality; for the results in Section
3, we must replace the symmetric group with the Hecke algebra: the basic
theory needed in this case can be found in w7x and w20x. Finally, the results
on the submodule structure of the symmetric powers used in Section 5
have been generalised to the quantum setting in w21x.
472 ANTON COX

ACKNOWLEDGMENTS

I thank Paul Martin and Alison Parker for several useful discussions and Steen Ryom-Han-
sen for sharing certain tilting module calculations that helped in part to motivate this project.
I am also very grateful to Jens Jantzen for bringing the work of Lin to my attention.

REFERENCES

1. H. H. Andersen, J. C. Jantzen, and W. Soergel, Representations of quantum groups at a

p-th root of unity and of semisimple groups in characteristic p: Independence of p,
´
Asterisque 220 Ž1994..
`
2. N. Bourbaki, ‘‘Groupes et algebres de Lie,’’ Chaps. 4᎐6, Hermann, Paris, 1968.
3. R. Carter and E. Cline, The submodule structure of Weyl modules for groups of type A1 ,
in ‘‘Proceedings of the Conference on Finite Groups, Park City, UT, 1975,’’ pp. 303᎐311,
Academic Press, New York, 1976.
4. A. G. Cox, The blocks of the q-Schur algebra, J. Algebra 207 Ž1998., 306᎐325.
5. R. Dipper and S. Donkin, Quantum GL n , Proc. London Math. Soc. Ž 3 . 63 Ž1991.,
165᎐211.
6. S. Donkin, On Schur algebras and related algebras II, J. Algebra 111 Ž1987., 354᎐364.
7. S. Donkin, ‘‘The q-Schur Algebra,’’ London Mathematical Society Lecture Notes Series,
Vol. 253, Cambridge University Press, Cambridge, UK, 1998.
8. S. Doty, Submodules of symmetric powers of the natural module for GL n , Contemp.
Math. 88 Ž1989., 185᎐191.
9. S. Doty and J. B. Sullivan, Filtration patterns for representations of algebraic groups and
their Frobenius kernels, Math. Z. 195 Ž1987., 391᎐407.
10. K. Erdmann, Symmetric groups and quasi-hereditary algebras, in ‘‘Finite Dimensional
Algebras and Related Topics’’ ŽV. Dlab and L. L. Scott, Eds.., pp. 123᎐161, Kluwer,
Dordrecht, 1994.
11. J. E. Humphreys, On the structure of Weyl modules, Comm. Algebra 12 Ž1984., 2665᎐2677.
¨
12. J. C. Jantzen, Uber das Dekompositionsverhalten gewisser modularer Darstellungen
halbeinfacher Gruppen und ihrer Lie-Algebren, J. Algebra 49 Ž1977., 441᎐469.
13. J. C. Jantzen, Darstellungen halbeinfacher Gruppen und ihrer Frobenius-Kerne, J. reine
angew. Math. 317 Ž1980., 157᎐199.
14. J. C. Jantzen, ‘‘Representations of Algebraic Groups,’’ Academic Press, New York, 1987.
¨
15. K. Kuhne-Hausmann, Zur Untermodulstruktur der Weylmoduln fur ¨ SL 3 , Bonner Math.
Schriften 162 Ž1985..
16. Z. Lin, Highest weight modules for algebraic groups arising from quantum groups, J.
Algebra 208 Ž1998., 276᎐303.
17. G. Lusztig, Modular representations and quantum groups, in ‘‘Classical Groups and
Related Topics,’’ Contemporary Mathematics, Vol. 82, pp. 59᎐77, American Mathemati-
cal Society, Providence, RI, 1989.
18. G. Lusztig, Finite-dimensional Hopf algebras arising from quantized universal enveloping
algebras, J. Amer. Math. Soc. 3 Ž1990., 257᎐296.
19. G. Lusztig, Quantum groups at roots of 1, Geom. Dedicata 35 Ž1990., 89᎐113.
20. A. Mathas, ‘‘Iwahori᎐Hecke Algebras and Schur Algebras of the Symmetric Groups,’’
University Lecture Series, Vol. 15, American Mathematical Society, Providence, RI, 1999.
21. L. Thams, The subcomodule structure of the quantum symmetric powers, Bull. Austral.
Math. Soc. 50 Ž1994., 29᎐39.