Journal of Algebra 412 (2014) 51–86

Contents lists available at ScienceDirect

Journal of Algebra
www.elsevier.com/locate/jalgebra

Jordan quadruple systems ✩

Murray Bremner ∗ , Sara Madariaga
Department of Mathematics and Statistics, University of Saskatchewan, Canada

a r t i c l e i n f o a b s t r a c t

Article history: We define Jordan quadruple systems by the polynomial

Received 6 March 2014 identities of degrees 4 and 7 satisfied by the Jordan tetrad
Available online 17 May 2014 {a, b, c, d} = abcd + dcba as a quadrilinear operation on
Communicated by Jean-Yves Thibon
associative algebras. We find further identities in degree 10
To our friend and colleague Luiz which are not consequences of the defining identities. We
Antonio Peresi on his retirement introduce four infinite families of finite dimensional Jordan
quadruple systems, and construct the universal associative
MSC: envelope for a small system in each family. We obtain
primary 17C05 analogous results for the anti-tetrad [a, b, c, d] = abcd − dcba.
secondary 17A42, 17C50, 17C55,
Our methods rely on computer algebra, especially linear
18D50
algebra on large matrices, the LLL algorithm for lattice
basis reduction, representation theory of the symmetric
Keywords:
Jordan tetrad group, noncommutative Gröbner bases, and Wedderburn
Polynomial identities decompositions of associative algebras.
Gröbner bases © 2014 Elsevier Inc. All rights reserved.
Universal associative envelopes
Wedderburn decompositions

1. Introduction

In this paper we study the quadrilinear operations {a, b, c, d} = abcd + dcba and
[a, b, c, d] = abcd − dcba in associative algebras. The first is the Jordan tetrad which

✩
Murray Bremner was supported by a Discovery Grant from NSERC, the Natural Sciences and
Engineering Research Council. Sara Madariaga was supported by a Postdoctoral Fellowship from PIMS,
the Pacific Institute for the Mathematical Sciences.
* Corresponding author.
E-mail addresses: [email protected] (M. Bremner), [email protected] (S. Madariaga).

http://dx.doi.org/10.1016/j.jalgebra.2014.05.001
0021-8693/© 2014 Elsevier Inc. All rights reserved.
52 M. Bremner, S. Madariaga / Journal of Algebra 412 (2014) 51–86

plays an important role in the structure theory of Jordan algebras [26,29]. The second
is the anti-tetrad, which seems not to have been studied until now.

1.1. Motivation

In an associative algebra for n ≥ 2 we define the n-tad to be this n-ary multilinear

operation: {a1 , . . . , an } = a1 · · · an + an · · · a1 . For n = 2 we obtain the Jordan product
{a, b} = ab + ba satisfying commutativity and the Jordan identity:
    
{a, b} ≡ {b, a}, {a, a}, b , a ≡ {a, a}, {b, a} .

There are further “special” identities satisfied by the Jordan product in every associative
algebra which do not follow from the defining identities; the simplest occur in degrees 8
and 9 and are called the Glennie identities [17,18]. A Jordan algebra is “special” if it can
be represented as a subspace of an associative algebra closed under the Jordan product;
otherwise, it is “exceptional”. If a special Jordan algebra is finite dimensional then its
universal associative enveloping algebra is also finite dimensional. A survey of the role
of identities in Jordan theory has been given by McCrimmon [24]. For the structure and
representation theory of finite dimensional Jordan algebras, see Jacobson [19]. For the
modern theory including infinite dimensional algebras, see McCrimmon [25].
For n = 3 we obtain the Jordan triple product abc + cba; in every associative algebra,
this operation satisfies identities which define Jordan triple systems (JTS):

{a, b, c} ≡ {c, b, a},

       
{a, b, c}, d, e ≡ {a, d, e}, b, c − a, {b, e, d}, c + a, b, {c, d, e} .

In contrast to the Jordan identity these identities are multilinear. There are special
identities in higher degree: identities satisfied by the Jordan triple product in every
associative algebra but which do not follow from the defining identities [22,23]. For the
classification of finite dimensional JTS, see [20,27,28] and for their universal associative
envelopes, see [21].
Closely related to Jordan triple systems are the anti-Jordan triple systems (AJTS),
see [16]. Finite dimensional simple AJTS have been classified [1]. These systems are de-
fined by identities satisfied by the anti-Jordan triple product abc−cba in every associative
algebra:

{a, b, c} + {c, b, a} ≡ 0,
       
{a, b, c}, d, e ≡ {a, d, e}, b, c + a, {b, e, d}, c + a, b, {c, d, e} .

Universal associative envelopes for one infinite family of simple AJTS have been con-
structed [15].
 M. Bremner, S. Madariaga / Journal of Algebra 412 (2014) 51–86 53

At the next step n = 4 we obtain the Jordan tetrad abcd + dcba, which arises in the
study of symmetric elements of associative algebras. Let An be the free unital associative
algebra on generators x1 , . . . , xn over a field F. The involution ∗ : An → An defined
on monomials by (xi1 · · · xid )∗ = xid · · · xi1 satisfies (ab)∗ = b∗ a∗ and (a∗ )∗ = a. The
subspace Hn = {a ∈ An | a∗ = a} of symmetric elements is a Jordan algebra under
the Jordan product. Let SJn be the free special Jordan algebra: the Jordan subalgebra
of Hn generated by x1 , . . . , xn . Cohn [12,13] has shown that SJn = Hn for n ≤ 3; but
for n ≥ 4, SJn = Hn and Hn is generated as a Jordan algebra by x1 , . . . , xn and the
tetrads xi1 xi2 xi3 xi4 + xi4 xi3 xi2 xi1 for 1 ≤ i1 < i2 < i3 < i4 ≤ n. Since An is unital, Hn
is generated by x1 , . . . , xn using the tetrad as a quadrilinear operation. From the tetrad
we recover the Jordan product by setting two arguments to 1, and so this operation
provides the natural algebraic structure on Hn for n ≥ 4.

Definition 1.1. The tetrad and anti-tetrad are respectively these quadrilinear operations
on associative algebras:

{a, b, c, d} = abcd + dcba, [a, b, c, d] = abcd − dcba.

1.2. Outline

In Section 2 we recall basic results from the representation theory of the symmet-
ric group, emphasizing a computational point of view, with a focus on applications to
polynomial identities. In Section 3 we use computer algebra to determine a complete
set of generators for the multilinear polynomial identities of degrees 4 and 7 satisfied by
the tetrad in every associative algebra. These identities define Jordan quadruple systems
(JQS). In Section 4 we use representation theory to show that there are further “special”
identities in degree 10 satisfied by the tetrad in every associative algebra which do not
follow from the defining identities for JQS. We use the LLL algorithm for lattice basis
reduction to obtain five explicit nonlinear special identities. In Section 5 we introduce
four infinite families of finite dimensional JQS. For one system in each family, we use
noncommutative Gröbner bases to construct its universal associative envelope; in each
case the envelope is finite dimensional, and we use the Wedderburn decomposition of
an associative algebra to classify the finite dimensional irreducible representations. In
Sections 6–8 we describe analogous results for the anti-tetrad. Throughout the paper we
suggest a number of open problems as possible directions for further research.

1.3. Conventions

Unless otherwise indicated, all computations are performed with the computer algebra
system Maple using arithmetic over either the ring Z of integers or the field Q of rational
numbers. To save computer memory, we often use arithmetic over the finite field Fp for
54 M. Bremner, S. Madariaga / Journal of Algebra 412 (2014) 51–86

some prime p, followed by rational reconstruction to recover results over Z. We use the
symbol ≡ to indicate that an equation holds for all values of the arguments.

2. Computational methods for polynomial identities

In this section we review rather informally the structure theory of the group algebra
QSn from an algorithmic point of view, with a focus on applications to polynomial
identities. For a more detailed exposition, see [9, §5].

2.1. Representation matrices

If F is a field of characteristic 0 or p > n, then the group algebra FSn is semisimple,

and decomposes as the direct sum of simple two-sided ideals, each isomorphic to a full
matrix algebra:

≈
R : FSn −−
→ Mdλ (F).
λ

The sum is over all partitions λ of n, and we write Rλ : FSn → Mdλ (F) for the projec-
tion onto component λ. The dimension dλ of the irreducible representation [λ] can be
computed from the Young diagram of λ using the hook formula. Given any permutation
σ ∈ Sn , the dλ ×dλ representation matrix Rλ (σ) in the natural representation has entries
in {0, 1, −1} and can be efficiently computed using the methods of [11] and [9, Fig. 1].
We study multilinear polynomial identities for nonassociative quadrilinear operations.
Monomials in such an operation have degrees n ≡ 1 (mod 3). By an association type
in degree n we mean a placement of operation symbols in a sequence of n arguments
(without specifying the arguments). By the identity monomial for an association type
in degree n we mean the monomial with the identity permutation a1 · · · an of the argu-
ments. In general, a monomial in degree n consists of an association type applied to a
permutation of the arguments. If there are t = t(n) distinct association types in degree n,
totally ordered in some way, then any multilinear polynomial I of degree n can be writ-
ten as a sum of t components I1 + · · · + It ; in each component, the terms differ only by
the permutation of the arguments. We can therefore regard I as an element of the direct
sum of t copies of the group algebra, (FSn )t , on which Sn acts by left multiplication.
For each partition λ, we apply the projection Rλ to each component I1 , . . . , It to obtain
a sequence of t matrices of size dλ × dλ which we combine horizontally into a matrix of
size dλ × tdλ . This is the representation matrix for the component of I in partition λ.
The row canonical form (RCF) of this matrix is the normal form of I for partition λ.
(We assume that zero rows have been removed from the RCF, so that a matrix in RCF
always has full rank.) Each row of the RCF generates a submodule of (FSn )t isomorphic
to [λ]. Hence the rank of this matrix is the multiplicity of [λ] in the submodule of (FSn )t
generated by I.
 M. Bremner, S. Madariaga / Journal of Algebra 412 (2014) 51–86 55

More generally, we consider a sequence I (1) , . . . , I (s) of s multilinear identities in degree

n and the sdλ × tdλ matrix in which the (i, j) block is the image under Rλ of the terms
of the i-th identity in the j-th association type. The rows of the RCF of this matrix
provide a canonical set of generators for the isotypic component of type [λ] generated by
I (1) , . . . , I (s) in (FSn )t .

2.2. Polynomial identities

The process of finding a complete set of generators for the Sn -module of multi-
linear identities in degree n satisfied by the tetrad {a1 , a2 , a3 , a4 } consists of three
steps. (For the anti-tetrad, the process is similar, but we must also keep track of sign
changes resulting from reversal.) These computations take place in the multilinear sub-
spaces Quad(n) of degree n in the free quaternary algebra with one operation satisfying
{a1 , a2 , a3 , a4 } − {a4 , a3 , a2 , a1 } ≡ 0. Since Quad(n) is also a left Sn -module (permuta-
tions act on the subscripts), we can regard it as the degree n component of the symmetric
operad Quad generated by one quaternary operation for which the action of S4 is given
by the symmetry in degree 4; that is, the generating space Quad(4) is the quotient of
the regular S4 -module QS4 by the left ideal generated by a1 a2 a3 a4 − a4 a3 a2 a1 .

Step 1. We distinguish two types of identities satisfied by the tetrad in degree n which
are consequences of known identities in lower degrees.
• Type 1: Symmetries. Since the tetrad satisfies symmetry in degree 4, we use this to
reduce the number of association types in degrees n > 4. For example, in degree 7 we
reduce the number of association types from 4 to 2 as follows:
   
a1 , a2 , {a3 , a4 , a5 , a6 }, a7 −→ a7 , {a3 , a4 , a5 , a6 }, a2 , a1 ,
   
a1 , a2 , a3 , {a4 , a5 , a6 , a7 } −→ {a4 , a5 , a6 , a7 }, a3 , a2 , a1 .

This does not eliminate all the identities in degree n which follow from symmetry in
degree 4, since for each association type there remain identities of the form ι−τ ≡ 0 where
ι is the identity monomial and τ is a monomial in the same association type obtained
from ι by a single application of symmetry. For example, for one of the association types
in degree 10 we have these identities:
   
{a1 , a2 , a3 , a4 }, a5 , a6 , {a7 , a8 , a9 , a10 } − {a4 , a3 , a2 , a1 }, a5 , a6 , {a7 , a8 , a9 , a10 } ≡ 0,
   
{a1 , a2 , a3 , a4 }, a5 , a6 , {a7 , a8 , a9 , a10 } − {a1 , a2 , a3 , a4 }, a5 , a6 , {a10 , a9 , a8 , a7 } ≡ 0,
   
{a1 , a2 , a3 , a4 }, a5 , a6 , {a7 , a8 , a9 , a10 } − {a7 , a8 , a9 , a10 }, a6 , a5 , {a1 , a2 , a3 , a4 } ≡ 0.

Symmetries of the association types generate a submodule Symm(n) ⊂ Quad(n).

• Type 2: Liftings. We assume by induction that we have already determined a set of
Sn−3 -module generators for the multilinear identities of degree n − 3 satisfied by the
56 M. Bremner, S. Madariaga / Journal of Algebra 412 (2014) 51–86

tetrad. (We do not include the symmetry in degree 4 in this process, so n − 3 > 4.) Let
I(a1 , . . . , an−3 ) ≡ 0 be such an identity. We consider n − 3 substitutions of a tetrad for
an argument of I, and two embeddings of I into a tetrad:
 
I {a1 , an−2 , an−1 , an }, a2 , . . . , an−3 , . . . ,
   
I a1 , . . . , {ai , an−2 , an−1 , an }, . . . , an−3 , . . . , I a1 , a2 , . . . , {an−3 , an−2 , an−1 , an } ,
   
I(a1 , a2 , . . . , an−3 ), an−2 , an−1 , an , an−2 , I(a1 , a2 , . . . , an−3 ), an−1 , an .

These multilinear polynomials clearly vanish, and form a (possibly redundant) set of
Sn -module generators for the multilinear identities in degree n which are consequences
of I. We repeat this process for every generator in degree n − 3. Iteration of this process
produces a set of (possibly redundant) generators for the Sn -module of identities for the
tetrad in degree n which are consequences of the known identities of lower degree. The
liftings of the identities of lower degree generate a submodule Lift(n) ⊂ Quad(n).

Definition 2.1. The submodule Old(n) = Symm(n) + Lift(n) ⊂ Quad(n) consists of

identities in degree n which are consequences of known identities of lower degree.

We use the methods of Section 2.1 to find canonical generators for Old(n). For each λ,
we obtain the matrix RCFλ (Old(n)) whose rows are a set of independent generators for
the simple summands [λ] in the isotypic component of type λ in Old(n).

Step 2. We determine all the multilinear identities satisfied by the tetrad in degree n.
By Section 2.1, these identities are elements of Quad(n) = (FSn )t , where t = t(n) is the
number of association types in degree n. Each monomial in degree n can be expanded
by applying the definition {a, b, c, d} = abcd + dcba to each occurrence of the operation
symbol; the result is a multilinear associative polynomial of degree n: an element of FSn .
This process extends linearly to Quad(n), giving the expansion map En : Quad(n) →
FSn , which is an Sn -module morphism. The kernel of En is the submodule of Quad(n)
consisting of those multilinear polynomials which vanish after expansion into the free
associative algebra.

Definition 2.2. The submodule All(n) ⊂ Quad(n) is the kernel of En : Quad(n) → FSn :
all the multilinear identities in degree n satisfied by the tetrad.

We use the methods of Section 2.1 to find canonical generators for All(n). For each λ,
we calculate the matrix RCFλ (All(n)) whose rows are a set of independent generators
for distinct simple summands of type [λ] in All(n). We restrict En to the corresponding
isotypic component of its domain and codomain, obtaining the map Enλ : Mdλ (F)t →
Mdλ (F). The matrix representing Enλ has size dλ × tdλ , and the i-th block consists of
the representation matrix of the expansion of the i-th identity monomial. We compute
a canonical basis for the nullspace of Enλ as follows:
 M. Bremner, S. Madariaga / Journal of Algebra 412 (2014) 51–86 57

• we calculate RCF(Enλ ), its rank q, and its nullity tdλ − q;

• we set the free variables (columns without leading 1s) to the standard basis vectors
in Ftdλ −q and solve for the leading variables;
• we put the resulting basis vectors into a matrix of size (tdλ − q) × tdλ ;
• we compute the RCF of this matrix, which is RCFλ (All(n)).

Step 3. We compare the results of Steps 1 and 2 to determine whether there exist new
multilinear identities satisfied by the tetrad in degree n.

Definition 2.3. The quotient module New(n) = All(n)/Old(n) consists of the new iden-
tities for the tetrad in degree n: a complete set of representatives for the equivalence
classes of all identities modulo old identities.

To obtain a canonical set of generators for New(n), for each λ we compare the ma-
trices RCFλ (Old(n)) and RCFλ (All(n)). If these two matrices have the same rank, they
must be equal; this indicates that every identity in the isotypic component λ follows
from identities of lower degree. If the ranks are not equal, then oldrank(λ), the rank
of RCFλ (Old(n)), must be strictly less than allrank(λ), the rank of RCFλ (All(n));
moreover, the row space of RCFλ (Old(n)) must be a subspace of the row space of
RCFλ (All(n)). The difference allrank(λ) − oldrank(λ) is the multiplicity of [λ] in
New(n). To find Sn -module generators for the isotypic component λ in New(n), we
identify the positions (i1 , j1 ), . . . , (ioldrank , joldrank ) of the leading 1s in RCFλ (Old(n)),
and the analogous positions (i1 , j1 ), . . . , (iallrank , jallrank

) in RCFλ (All(n)). Comparing
these two sets of positions, we obtain
   
J = {j1 , . . . , joldrank }  j1 , . . . , jallrank

= J , J  \ J = jk 1 , . . . , jk allrank−oldrank .

Rows ik1 , . . . , ikallrank−oldrank of RCFλ (All(n)) are the canonical generators of New(n).

2.3. Linear algebra over Q, Fp and Z

In general, we prefer to do all computations over the field Q of rational numbers.

However, it is well known that, even if the original matrix is very sparse and its entries
belong to {0, ±1}, standard algorithms for computing the RCF produce exponential
increases in the matrix entries. Even if enough computer memory is available to store
the intermediate results, the calculations can take far too much time. It is therefore
often convenient to use modular arithmetic, so that each matrix entry uses a fixed small
amount of memory. This leads to the problem of rational reconstruction: recovering the
correct results over Q or Z from the known results over Fp .
In general, rational reconstruction is extremely ill-defined: we want to compute an
inverse for a partially-defined infinity-to-one map. Therefore it is only effective when
we have a good theoretical understanding of the arithmetical nature of the expected
58 M. Bremner, S. Madariaga / Journal of Algebra 412 (2014) 51–86

results. In our computations, we may assume that the correct rational coefficients have
a common highly composite denominator. The reason is that the vector spaces we study
are all modules over the symmetric group Sn , and in the Wedderburn decomposition of
the group algebra QSn into a direct sum of full matrix algebras, the matrix units are
linear combinations of permutations in which the coefficients have n! as their common
denominator. Moreover, FSn is semisimple whenever F has characteristic 0 or p > n,
so formulas giving the Wedderburn decomposition in rational arithmetic also apply to
modular arithmetic with a prime larger than the degree of the multilinear polynomial
identities. For a more precise statement of this fact, see [8, Lemma 8].
If we use a large enough prime, say p > n! rather than p > n, then we can recog-
nize the common denominator d of the rational coefficients from the distribution of the
congruence classes modulo p: the modular coefficients are clustered near the congruence
classes representing i/d for 1 ≤ i ≤ d − 1. This allows us to recover the rational coef-
ficients; we then multiply by the LCM of the denominators to get integer coefficients,
and finally divide by the GCD of the coefficients to get the integer vector with least
Euclidean length which is a scalar multiple of the original coefficient vector. Once we
have a realistic conjecture for the correct integer coefficients, we can perform a much
simpler computation using rational arithmetic to verify the results.
Most of our computations involve finding a basis of integer vectors for the nullspace of
a matrix with integer entries. In some cases, the modular methods described in the pre-
vious paragraphs give good results, meaning that the basis vectors have small Euclidean
lengths. In other cases, we obtain much better results using the Hermite normal form
(HNF) of an integer matrix (the analogue over Z of the RCF) together with the LLL
algorithm for lattice basis reduction. If A is an s × t matrix over Z then computing the
HNF of the transpose produces two matrices over Z: a t × s matrix H and a t × t matrix
U with det(U ) = ±1 such that U At = H. If rank(A) = r then the bottom t − r rows
of U form a lattice basis for the left integer nullspace of At , which is the right integer
nullspace of A. We then apply the LLL algorithm with increasing values of the parameter
to this basis in order to obtain shorter basis vectors. We define a precise measure of the
size of a lattice basis consisting of integer vectors v1 , . . . , vk by the number of decimal
digits in the product of the Euclidean lengths of the vectors:


k
log10 vi . (1)
i=1

For a more detailed discussion of the application of HNF and LLL to polynomial iden-
tities, including algorithms in pseudocode, see [6, §3] and the monograph [3].

3. Defining identities for Jordan quadruple systems

In this section we determine a complete set of Sn -module generators for the multilinear
polynomial identities satisfied by the tetrad in degrees 4 and 7.
 M. Bremner, S. Madariaga / Journal of Algebra 412 (2014) 51–86 59

Lemma 3.1. Every multilinear identity in degree 4 satisfied by the tetrad in every asso-
ciative algebra is a consequence of the symmetry {a, b, c, d} − {d, c, b, a} ≡ 0.

Proof. Consider two copies of the group algebra QS4 ; the first, denoted Q, is the multilin-
ear subspace of degree 4 in the free quaternary algebra with one operation {−, −, −, −};
the second, denoted A, is the multilinear subspace of degree 4 in the free associative alge-
bra. Bases of Q and A are the sets {{aσ , bσ , cσ , dσ } | σ ∈ S4 } and {aσ bσ cσ dσ | σ ∈ S4 },
ordered lexicographically. We initialize the 24 × 24 expansion matrix E in which the
(i, j) entry is the coefficient of the i-th associative monomial in the expansion of the j-th
quaternary monomial, and compute its RCF, which has rank 12, and hence nullity 12.
We obtain a basis for the nullspace by setting the free variables equal to the standard
basis vectors in Q12 and solving for the leading variables. We put these basis vectors
into a 12 × 24 matrix and compute its RCF; every row represents a permutation of the
stated symmetry. 2

Theorem 3.2. Every multilinear identity in degree 7 satisfied by the tetrad in every asso-
ciative algebra is a consequence of the symmetry and these three identities in degree 7:
       
{a, b, c, d}, e, f, g + {a, b, f, e}, d, c, g + {d, c, f, e}, a, b, g − g, {b, a, d, c}, f, e
   
− g, {b, a, e, f }, c, d − g, {c, d, e, f }, b, a ≡ 0,
       
{a, b, c, d}, e, f, g − {a, b, g, f }, e, c, d + {a, b, d, c}, e, g, f − {a, b, f, g}, e, d, c
     
+ {a, e, c, d}, b, g, f − {a, e, g, f }, b, d, c + {a, e, d, c}, b, f, g
     
− {a, e, f, g}, b, c, d − a, {b, c, d, e}, f, g + a, {b, g, f, e}, c, d
   
− a, {b, d, c, e}, g, f + a, {b, f, g, e}, d, c ≡ 0,
       
{a, b, c, d}, e, f, g − {a, f, g, c}, b, e, d + {c, b, a, d}, e, g, f + {f, b, c, e}, g, a, d
     
− {f, g, a, e}, c, b, d − {f, g, a, d}, b, c, e − {f, g, e, d}, a, b, c
     
+ {g, b, a, e}, f, c, d − {g, f, c, e}, a, b, d − {g, f, c, d}, b, a, e
     
− {g, f, e, d}, c, b, a + {e, a, b, d}, c, f, g + {e, c, b, d}, a, g, f
     
+ a, {b, c, g, f }, e, d + c, {b, a, f, g}, e, d − f, {b, c, e, g}, a, d
   
− g, {b, a, e, f }, c, d + e, {a, g, f, c}, b, d ≡ 0.

Proof. By symmetry, we need to consider only two association types in degree 7,

   
{−, −, −, −}, −, −, − , −, {−, −, −, −}, −, − ,

each having 7!/2 = 2520 distinct multilinear monomials, for a total of 5040. A basis of
Quad(7) consists of these monomials ordered by association type and then by lex order of
the permutation. A basis of the group algebra QS7 , the multilinear subspace in the free
60 M. Bremner, S. Madariaga / Journal of Algebra 412 (2014) 51–86

associative algebra, consists of the permutations in lex order. We record the expansion
of the identity monomial in each association type:

 
{a, b, c, d}, e, f, g = abcdef g + dcbaef g + gf eabcd + gf edcba,
 
a, {b, c, d, e}, f, g = abcdef g + aedcbf g + gf bcdea + gf edcba.

The other expansions are obtained by permutation of the arguments, since the expansion
map E : Quad(7) → QS7 is an S7 -module homomorphism.
Using modular arithmetic (p = 101) we initialize a 5040 × 5040 matrix in which the
(i, j) entry is the coefficient of the i-th associative monomial in the expansion of the
j-th quaternary monomial. We compute the RCF and find that the rank is 2520, and
hence the nullity is 2520. As in the proof of Lemma 3.1, we find a basis of the nullspace;
every coefficient belongs to {0, 1, 2, 50, 51, 99, 100}. For each vector, we multiply by 2
if 50 or 51 is a coefficient, and do nothing otherwise. Since the rows are coefficient
vectors of polynomial identities, we can multiply by nonzero scalars to obtain equivalent
identities with simpler coefficients. Reducing modulo p using symmetric representatives,
the coefficients belong to {−4, −2, −1, 0, 1, 2, 4}. We regard these as integers, and sort
the vectors by increasing Euclidean length.
The next step is to extract from the linear basis of 2520 vectors a much smaller set of
generators for the nullspace as an S7 -module. Using modular arithmetic, we initialize a
zero matrix consisting of upper and lower 5040 × 5040 blocks. For each vector, regarded
as the coefficients of a polynomial identity, we:

• check to see if the identity belongs to the row space of the matrix;
• if not, apply all permutations of the arguments to the identity, and store the results
in the rows of the lower block;
• compute the RCF of the matrix; if the rank increases, record the identity as a gen-
erator (at this point, the lower block is again zero).

Only four identities increase the rank, and one belongs to the submodule generated by
the others, which are independent (none is a consequence of the other two). These are
the three identities in the statement of this theorem. 2

Remark 3.3. The identities of Theorem 3.2 can be checked by hand. For example, the
expansion of the first identity into the free associative algebra produces

abcdef g + gf eabcd + dcbaef g + gf edcba + abf edcg + gcdabf e

+ ef badcg + gcdef ba + dcf eabg + gbadcf e + ef cdabg + gbaef cd
− gbadcf e − ef badcg − gcdabf e − ef cdabg − gbaef cd − dcbaef g
− gf eabcd − dcf eabg − gcdef ba − abcdef g − gf edcba − abf edcg = 0.
 M. Bremner, S. Madariaga / Journal of Algebra 412 (2014) 51–86 61

Table 1
S7 -module multiplicities for the tetrad in degree 7.
λ dλ Symmetries Expansions New
rows cols symm rows cols rank null
7 1 3 2 0 2 1 1 1 1
61 6 18 12 4 12 6 3 9 5
52 14 42 28 12 28 14 8 20 8
512 15 45 30 16 30 15 6 24 8
43 14 42 28 12 28 14 7 21 9
421 35 105 70 36 70 35 18 52 16
413 20 60 40 24 40 20 10 30 6
32 1 21 63 42 20 42 21 9 33 13
322 21 63 42 20 42 21 12 30 10
3212 35 105 70 36 70 35 17 53 17
314 15 45 30 16 30 15 9 21 5
23 1 14 42 28 12 28 14 7 21 9
22 13 14 42 28 12 28 14 6 22 10
215 6 18 12 4 12 6 3 9 5
17 1 3 2 0 2 1 0 2 2

We need computer algebra to prove that these identities are a complete set of S7 -module
generators for the tetrad identities in degree 7.

Definition 3.4. A Jordan quadruple system (or JQS) is a vector space Q over a field F of
characteristic 0 or p > 7 with a quadrilinear map {−, −, −, −} : Q4 → Q satisfying the
symmetry of Lemma 3.1 and the identities of Theorem 3.2.

Lemma 3.5. Over a field of characteristic 0 or p > 7, column “New” of Table 1 gives the
multiplicity of the S7 -module [λ] in the kernel of the expansion map for each partition λ.

Proof. Table 1 was computed using the methods of Section 2.1. Columns 1 and 2 give
the partitions λ and the dimensions dλ . In degree 7 we consider only the symmetries of
the association types, since there are no liftings of identities from lower degrees. There
are two association types each with one symmetry:
   
{a1 , a2 , a3 , a4 }, a5 , a6 , a7 − {a4 , a3 , a2 , a1 }, a5 , a6 , a7 ≡ 0,
   
a1 , {a2 , a3 , a4 , a5 }, a6 , a7 − a1 , {a5 , a4 , a3 , a2 }, a6 , a7 ≡ 0.

Under “Symmetries”, columns 3 and 4 (rows, cols) give the size of the representation
matrix for the symmetries, and column 5 (symm) gives its rank. Under “Expansions”,
columns 6 and 7 give the size of the expansion matrix, column 8 gives its rank, and
column 9 gives the nullity of its transpose. (For an explanation of using the transpose,
see [5, pp. 446–447].) Column “New” is the multiplicity of the module [λ] (null minus
symm) in the kernel of the expansion map. 2

Remark 3.6. The data in Table 1 show that there is no point in searching for possibly
simpler nonlinear identities whose linearizations are equivalent to the identities of The-
orem 3.2. If I ≡ 0 is a nonlinear identity of degree 7, then every monomial of I contains
62 M. Bremner, S. Madariaga / Journal of Algebra 412 (2014) 51–86

k ≥ 2 occurrences of some variable, say a. In L(I), the linearization of I, the variable

a is replaced by k distinct variables a1 , . . . , ak and L(I) is invariant under all permuta-
tions of these variables. To obtain the consequences of L(I) in partition 17 , we evaluate
the alternating sum over all permutations of the variables in L(I), but this is clearly 0,
contradicting the multiplicity 2 for the module [17 ].

4. Special identities in degree 10 for the tetrad

In this section we show that there are identities for the tetrad which do not follow from
the defining identities for JQS. (We have not included these identities in Definition 3.4,
so that the operad for JQS is quadratic and admits a Koszul dual.)

Definition 4.1. Let A be an associative algebra and let JQS(A) be the underlying vector
space of A with the tetrad as the operation. If Q is a JQS, we call Q special if there is
an embedding Q → JQS(A) for some A; otherwise we call Q exceptional. If I ≡ 0 is a
polynomial identity which is satisfied by the tetrad in every associative algebra but which
does not follow from the defining identities for JQS, then we call I a special identity for
the tetrad.

Open Problem 4.2. Do there exist exceptional JQS?

Theorem 4.3. For the tetrad, and for each partition λ of 10, columns “symm”, “symmlift”,
“null” and “New” in the corresponding row of Table 2 contain respectively the multiplicity
of [λ] in the modules Symm(10), Old(10), All(10) and New(10).

Proof. The symmetry in degree 4 implies that every association type for a quadrilinear
operation in degree 10 is equivalent to one of the following:
     
{−, −, −, −}, −, −, − , −, −, − , −, {−, −, −, −}, −, − , −, −, − ,
       
−, {−, −, −, −}, −, −, − , −, − , −, −, {−, −, −, −}, −, − , −, − ,
    (2)
{−, −, −, −}, {−, −, −, −}, −, − , {−, −, −, −}, −, {−, −, −, −}, − ,
   
{−, −, −, −}, −, −, {−, −, −, −} , −, {−, −, −, −}, {−, −, −, −}, − .

These 8 types have respectively 1, 1, 1, 1, 2, 2, 3, 3 symmetries ι − τ ≡ 0 where ι is the

identity monomial and τ is a monomial with a permutation of order 2 of the variables.
We list the corresponding monomials τ for each type:
             
{dcba}ef g hij , a{edcb}f g hij , a {edcb}f gh ij , a b{f edc}gh ij ,
       
{dcba}{ef gh}ij , {abcd}{hgf e}ij , {dcba}e{f ghi}j , {abcd}e{ihgf }j ,
       
{dcba}ef {ghij} , {abcd}ef {jihg} , {ghij}f e{abcd} , a{edcb}{f ghi}j ,
   
a{bcde}{ihgf }j , j{f ghi}{bcde}a .
 M. Bremner, S. Madariaga / Journal of Algebra 412 (2014) 51–86 63

Table 2
S10 -module multiplicities for the tetrad in degree 10.
# λ dλ Symmetries and liftings Expansions New
rows cols symm symmlift rows cols rank null
1 10 1 9 8 0 7 8 1 1 7 ·
2 91 9 81 72 30 68 72 9 4 68 ·
3 82 35 315 280 140 260 280 35 20 260 ·
4 812 36 324 288 166 272 288 36 16 272 ·
5 73 75 675 600 342 565 600 75 35 565 ·
6 721 160 1440 1280 784 1200 1280 160 80 1200 ·
7 713 84 756 672 442 628 672 84 44 628 ·
8 64 90 810 720 416 670 720 90 50 670 ·
9 631 315 2835 2520 1594 2365 2520 315 155 2365 ·
10 622 225 2025 1800 1150 1680 1800 225 120 1680 ·
11 6212 350 3150 2800 1878 2630 2800 350 170 2630 ·
12 614 126 1134 1008 704 941 1008 126 66 942 1
13 52 42 378 336 204 320 336 42 16 320 ·
14 541 288 2592 2304 1456 2160 2304 288 144 2160 ·
15 532 450 4050 3600 2330 3380 3600 450 220 3380 ·
16 5312 567 5103 4536 3008 4244 4536 567 291 4245 1
17 522 1 525 4725 4200 2808 3939 4200 525 260 3940 1
18 5213 448 4032 3584 2464 3357 3584 448 224 3360 3
19 515 126 1134 1008 708 945 1008 126 60 948 3
20 42 2 252 2268 2016 1282 1880 2016 252 136 1880 ·
21 4 2 12 300 2700 2400 1582 2260 2400 300 140 2260 ·
22 432 210 1890 1680 1092 1580 1680 210 100 1580 ·
23 4321 768 6912 6144 4032 5760 6144 768 384 5760 ·
24 4313 525 4725 4200 2802 3933 4200 525 265 3935 2
25 423 300 2700 2400 1562 2239 2400 300 160 2240 1
26 422 12 567 5103 4536 3022 4260 4536 567 276 4260 ·
27 4214 350 3150 2800 1870 2616 2800 350 180 2620 4
28 416 84 756 672 446 631 672 84 40 632 1
29 33 1 210 1890 1680 1080 1569 1680 210 110 1570 1
30 3 2 22 252 2268 2016 1302 1900 2016 252 116 1900 ·
31 32 212 450 4050 3600 2322 3368 3600 450 230 3370 2
32 3 2 14 225 2025 1800 1164 1693 1800 225 105 1695 2
33 323 1 288 2592 2304 1456 2159 2304 288 144 2160 1
34 322 13 315 2835 2520 1588 2358 2520 315 160 2360 2
35 3215 160 1440 1280 784 1198 1280 160 80 1200 2
36 317 36 324 288 162 268 288 36 20 268 ·
37 25 42 378 336 192 309 336 42 26 310 1
38 2 4 12 90 810 720 428 680 720 90 40 680 ·
39 2 3 14 75 675 600 336 559 600 75 40 560 1
40 2 2 16 35 315 280 146 265 280 35 15 265 ·
41 218 9 81 72 28 67 72 9 5 67 ·
42 110 1 9 8 2 8 8 1 0 8 ·

For each association type, each symmetry reduces the number of multilinear monomials
by a factor of 2, so the total number of monomials is


1 1 1 1 1 1 1 1
10! + + + + + + + = 9 979 200.
2 2 2 2 4 4 8 8

This number is so large that we must decompose the computation into smaller pieces
using the representation theory of S10 . (To do the following computation without repre-
sentation theory would require a 3 628 800 × 9 979 200 matrix.)
64 M. Bremner, S. Madariaga / Journal of Algebra 412 (2014) 51–86

The symmetry implies that each identity I(a, b, c, d, e, f, g) from Theorem 3.2 produces
9 liftings in degree 10:
   
I {a, h, i, j}, b, c, d, e, f, g , I a, {b, h, i, j}, c, d, e, f, g , ...,
 
I a, b, c, d, e, f, {g, h, i, j} ,
   
I(a, b, c, d, e, f, g), h, i, j , h, I(a, b, c, d, e, f, g), i, j .

For each partition λ with module [λ] of dimension dλ , we construct a matrix M of size
9dλ × 8dλ (“rows” and “cols” under “Symmetries and liftings”) consisting of dλ × dλ
blocks. To process an identity, we compute the representation matrices of its terms in
each association type [9], store these matrices in the last row of blocks, and compute the
RCF of M , so the last row of blocks becomes zero. In this way, we process the symmetries
of the association types and compute the rank of M (column “symm”). We retain these
results in M and process the liftings; the rank of M increases (column “symmlift”) to
reach the multiplicity of [λ] in the module Old(10). When we are done, the rows of
RCF(M ) form a canonical set of generators for the isotypic component of [λ] in Old(10).
For each partition λ, we construct a matrix X of size 8dλ ×dλ (“rows” and “cols” under
“Expansions”) consisting of dλ × dλ blocks; the i-th block contains the representation
matrix for the terms of the expansion of the identity monomial in the i-th association
type. We compute the RCF of the transpose X t and find its rank (column “rank”) and
nullity (column “null”); we then extract a basis for the nullspace by setting the free
variables to the standard basis vectors and solving for the leading variables. We put the
nullspace basis into the rows of a matrix N and compute the RCF. The rows of RCF(N )
form a canonical set of generators for the isotypic component of [λ] in All(10).
If the multiplicity of [λ] in Old(10) coincides with its multiplicity in All(10), then
there are no new identities for partition λ. (In this case, we check the consistency of
the computations by verifying that the two submodules are equal, which amounts to
verifying that the two matrices in RCF are equal.) Otherwise, there are new identities
in degree 10 for the tetrad corresponding to partition λ. 2

Open Problem 4.4. Determine a minimal set of multilinear identities in degree 10 for the
tetrad which generate New(10) as an S10 -module.

In the rest of this section we present some nonlinear special identities in degree 10
for the tetrad. These identities correspond to the partitions λ = s1t (s + t = 10) for
which column “New” in Table 2 is nonzero, namely λ = 614 , 515 , 416 . The corresponding
Young diagrams have a long first row and a long tail. For these λ, the structure theory
of QS10 allows us to assume that the identity is a symmetric function of the variables in
the first row and an alternating function of the variables in the tail, and this reduces the
computational problem to a manageable size. We first recall the notion of linearization
matrices, called operators in [7, §2].
 M. Bremner, S. Madariaga / Journal of Algebra 412 (2014) 51–86 65

Definition 4.5. Let n = n1 + · · · + nk (n ≥ n1 ≥ · · · ≥ nk ≥ 1) be a partition and

μ = μ(an1 1 , . . . , ank k ) a quaternary monomial of multidegree (n1 , . . . , nk ). For i = 1, . . . , k
set Ni = {n1 +· · ·+ni−1 +1, . . . , n1 +· · ·+ni } so that N1 ∪N2 ∪· · ·∪Nk is a set partition
of {1, . . . , n} with |Ni | = ni . The basic linearization of μ is the multilinear monomial
(μ) obtained from μ by first replacing ai by ai for i = 1, . . . , k and then replacing the
ni occurrences of ai by aj for j ∈ Ni with j increasing from left to right. The symmetric
and alternating linearizations of μ are

 
L+ (μ) = ··· σ1 · · · σk (μ),
σ1 ∈P1 σk ∈Pk
 
L− (μ) = ··· i (σ1 ) · · · k (σk )σ1 · · · σk (μ),
σ1 ∈P1 σk ∈Pk

where Pi is the symmetric group on Ni and i : Pi → {±1} is the sign.

Remark 4.6. Over a field of characteristic 0 or p > n, the identity I holds if and only if
L+ (I) holds; the same is not true for L− (I).

Definition 4.7. Let λ be another partition of n and let Rλ : QSn → Md (Q) be the
corresponding irreducible representation of Sn of dimension dλ . The symmetric and
alternating linearization matrices for partitions n1 , . . . , nk and λ are

 
L+
n1 ,...,nk ;λ = Rλ (σ), L−
n1 ,...,nk ;λ = (σ)Rλ (σ).
σ∈P1 ×···×Pk σ∈P1 ×···×Pk

Linearization matrices make it much easier to compute the representation matrix for
the linearization of a nonlinear monomial μ: we can use the basic linearization (μ) of
the nonlinear monomial instead of using all the terms in L+ (μ).

Lemma 4.8. For all partitions n1 , . . . , nk and λ and all monomials μ, we have

       
n1 ,...,nk ;λ · Rλ (μ) ,
Rλ L+ (μ) = L+ Rλ L− (μ) = L−
n1 ,...,nk ;λ · Rλ (μ) .

Proof. These equations follow directly from the definitions, the linearity of the maps,
and the homomorphism property of a representation. 2

Theorem 4.9. The multihomogeneous identities in Fig. 1 are satisfied by the tetrad in
every associative algebra but are not consequences of the defining identities for JQS.
The sums are over all permutations σ of the non-repeated variables and (σ) is the
sign.
66 M. Bremner, S. Madariaga / Journal of Algebra 412 (2014) 51–86

  σ σ σ σ  σ σ σ σ 
(σ) a, a, a, b , a, a, c , d , a, e − a, a, a, b , a, a, c ,d ,e ,a
σ∈S4
 σ  σ  σ σ   σ  σ σ σ 
+ a, a, b , a , c , a, a , d , e , a + a, a, b , a , c , a, d , a, e , a
 σ σ σ  σ  σ σ σ  σ 
+ a, b , a, c , a, d , a , a, a, e − a, b , a, c , a, d , a , a, e , a
  σ  σ σ σ   σ σ σ σ 
− a, a, a, b , a , a, c , d , a, e − a, a, b , a, c , a, d , a, e , a
  σ  σ σ σ   σ σ σ  σ
+ a, a, a, b , a , a, c , d , a, e − a, a, a, b , c , d , a, a , a, e
  σ σ σ  σ    σ σ σ  σ 
+ a, a, a, b , c , d , a, a , e , a + a, a, b , a, c , a, d , a , e , a ≡ 0
  σ σ σ  σ σ  σ σ σ  σ σ
(σ) a, a, b , c , a, d , a , a, e , f − a, a, b , c , a, d , a , e , a, f
σ∈S5
 σ σ σ  σ σ  σ σ σ  σ σ
− a, a, b , c , d , a, a , a, e , f + a, a, b , c , d , a, a , e , a, f
 σ σ σ  σ σ  σ σ σ  σ σ
− a, b , a, c , a, d , a , a, e , f + a, b , a, c , a, d , a , e , a, f
 σ σ σ  σ σ  σ σ σ  σ σ
+ a, b , a, c , d , a, a , a, e , f − a, b , a, c , d , a, a , e , a, f
 σ σ σ σ  σ   σ σ σ σ  σ 
− b , a, c , d , a, e , a , a, f , a − b , a, c , d , e , a, a , f , a, a
 σ σ σ σ  σ   σ σ σ σ  σ 
− b , c , a, d , a, e , a , f , a, a − b , c , a, d , e , a, a , a, f , a ≡ 0
  σ σ σ  σ σ  σ σ σ  σ σ 
(σ) a, a, a, b , c , d , a , e , a, f − a, a, a, b , c , d , a , e , f , a
σ∈S5
 σ σ σ σ  σ  σ σ σ σ  σ 
+ a, b , a, c , d , e , a , a, a, f − a, b , a, c , d , e , a , a, f , a
  σ σ σ σ σ    σ σ σ  σ σ
− a, a, a, b , c , a, d , e , f , a − a, a, a, b , c , d , a , e , a, f
  σ σ σ  σ σ    σ σ σ σ σ 
+ a, a, a, b , c , d , a , e , f , a + a, a, b , a, c , a, d , e , f , a
  σ σ σ σ  σ   σ σ σ σ  σ 
− a, b , a, c , d , e , a , a, a, f − a, b , c , a, d , e , a , a, f , a
  σ σ σ σ σ    σ σ σ σ σ 
+ a, a, a, b , c , a, d , e , f , a − a, a, b , a, c , a, d , e , f , a ≡ 0
  σ  σ σ σ σ   σ σ σ σ σ 
(σ) a, a, b , a , c , a, d , e , f , a − a, b , a, c , a, a, d , e , f , a
σ∈S5
 σ σ σ  σ σ   σ σ σ σ σ
− a, b , a, c , a, d , a , e , f , a + a, b , a, c , d , a, e , a, a, f
 σ σ σ σ  σ   σ σ σ  σ σ
+ a, b , a, c , d , e , a , a, a, f + a, a, a, b , c , d , a , e , a, f
  σ σ σ σ σ   σ σ σ  σ σ 
+ a, a, b , a, c , a, d , e , a, f + a, a, b , a, c , d , a , e , f , a
  σ σ σ σ σ   σ σ σ σ σ 
− a, b , a, c , d , a, e , a, a, f − a, b , c , a, d , a, e , a, f , a
 σ  σ  σ  σ σ   σ σ σ σ σ
+ b , a, a, c , a , d , a , e , a, f − a, a, a, b , c , d , a, e , a, f
  σ σ σ σ σ   σ σ σ σ σ 
− a, a, b , a, c , a, d , e , a, f − a, a, b , a, c , a, d , e , f , a
  σ σ σ σ σ 
− a, a, b , a, c , d , a, e , f , a ≡ 0
  σ σ σ σ σ σ  σ σ σ σ σ σ
(σ) b , a, c , d , a, a, e , a, f , g − b , a, c , d , a, a, e , f , a, g
σ∈S6
 σ σ σ σ σ σ  σ σ σ σ σ σ
+ b , a, c , d , e , a, f , a, a, g − b , c , a, d , a, a, e , a, f , g
 σ σ σ σ σ σ  σ σ σ σ σ σ 
+ b , c , a, d , a, a, e , f , a, g + b , c , a, d , a, e , f , a, a, g ≡0

Fig. 1. Some special identities for the tetrad in degree 10.

 M. Bremner, S. Madariaga / Journal of Algebra 412 (2014) 51–86 67

Proof. The Young tableaux for partitions λ = 614 , 515 , 416 are

For each λ, the primitive idempotents in the Young basis of the two-sided ideal in the
group algebra correspond to standard tableaux; each idempotent is the product of the
symmetric sum over row permutations and the alternating sum over column permuta-
tions. Hence an identity corresponding to a partition λ with first row of size s and tail
of size t is symmetric in the first s variables and alternating in the last t variables. This
allows us to reduce the number of monomials by considering only those in which the
non-repeated variables appear in increasing order, and identifying these monomials with
the corresponding alternating sums. We write in detail the proof for λ = 614 ; the others
are similar.
Let J be the multihomogeneous subspace with variables a6 bcde in the free quaternary
algebra with one operation {−, −, −, −} satisfying symmetry in degree 4. For the eight
association types (2) we have respectively 134, 134, 134, 134, 86, 86, 49, 52 monomials,
and so dim J = 809 (recall that the non-repeated variables appear in increasing order).
Let A be the corresponding multihomogeneous subspace in the free associative algebra;
we have dim A = 5040. We construct the 809 × 5040 expansion matrix E in which the
(i, j) entry is the (integer) coefficient of the j-th associative monomial in the expansion
of the i-th quaternary monomial. Using modular arithmetic, we compute RCF(E) and
obtain rank 110 and nullity 699; the entries of the RCF are 0, ±1, ±2, ±4. We identify
the columns which contain leading 1s; these columns j1 , . . . , j110 form a basis of the
column space. Using integer arithmetic, we combine these columns of E into a much
smaller 809 × 110 matrix E  . We compute H = HNF(E  ) and confirm that the rank and
nullity have not changed in characteristic 0; this computation also provides an 809 × 809
unimodular matrix U such that U E  = H. The last 699 rows of U form a lattice basis for
the integer nullspace of E  ; they have already been reduced using the LLL algorithm, and
the entries are 0, ±1, ±2. Using formula (1), this basis has size ≈ 607. To get identities
as simple as possible, we apply the LLL algorithm again, with standard parameter 3/4;
this new basis has size ≈ 537. We sort these 699 vectors by increasing length.
We proceed as in the proof of Theorem 4.3 using modular arithmetic (p = 101) but
considering only partition λ = 614 . The matrix has size 1134 × 1008. We first compute
the rank of the symmetries (704) and the symmetries with the liftings (941). The usual
next step would be to process the linearizations of the 699 nonlinear identities obtained
in the previous paragraph; this however produces multilinear identities with very large
numbers of terms. In order to reduce the size of the computation, we use the equations
of Lemma 4.8. We require a combination of symmetric linearization over the first 6
variables (in the first row of the Young diagram) with alternating linearization over
68 M. Bremner, S. Madariaga / Journal of Algebra 412 (2014) 51–86

the last 4 variables (in the tail). This involves a slight modification of the linearization
matrices in Definition 4.7. Our combined linearization matrix is the sum of 6!4! = 17 280
terms, but it factors as the product of two partial linearization matrices with respectively
6! terms (permutations σ ∈ S6 acting on {1, . . . , 6}) and 4! terms (permutations τ ∈ S4
acting on {7, . . . , 10}), for a total of only 6! + 4! = 744 terms:

   
(τ )Rλ (στ ) = (τ )Rλ (σ)Rλ (τ ) = Rλ (σ) · (τ )Rλ (τ ).
σ∈S6 , τ ∈S4 σ∈S6 , τ ∈S4 σ∈S6 τ ∈S4

This 126 × 126 matrix is surprisingly simple: it has rank 1 and 21 nonzero entries. Since
the rank of all identities for λ = 614 is 942, we expect exactly one new identity. After
processing the 699 nonlinear identities, we find that only one increases the rank. This is
the identity in the statement of the theorem. 2

5. Four families of finite dimensional Jordan quadruple systems; examples of universal

associative envelopes

In this section we consider some small examples of special JQS, and determine the
structure of their universal associative envelopes. We first define four infinite families An ,
Bn , Cpqr , Dpq of finite dimensional JQS which are subspaces of matrix algebras which
are closed under the tetrad.

• An consists of all n × n matrices; the dimension is n2 .

• Bn consists of all n × n symmetric matrices; the dimension is 12 n(n + 1).

For the remaining systems, we recall the classification [10] of simple associative n-ary
algebras in the case n = 4. We choose integers p, q, r ≥ 1 and consider block matrices of
size p + q + r, where Mpq is a matrix of size p × q:

⎡ ⎤
0 0 Mpr
⎣ Mqp 0 0 ⎦. (3)
0 Mrq 0

This space is closed under the associative quadruple product of matrices.

• Cpqr (p ≥ q, r) consists of all matrices (3); the dimension is pq + qr + rp.

t
• Dpq consists of all matrices (3) with q = r where Mpq = Mqp and Mqq is symmetric;
1
the dimension is pq + 2 q(q + 1).

We consider the smallest non-trivial system J in each family: A2 , B2 , C111 , D11 with di-
mensions 4, 3, 3, 2 respectively. To construct the universal associative envelope U (J), we
use noncommutative Gröbner bases in free associative algebras [4]: U (J) is the quotient
 M. Bremner, S. Madariaga / Journal of Algebra 412 (2014) 51–86 69

F B/I(G) of the free associative algebra F B on a basis B of J by the ideal I(G)

generated by the following set of relations:
  
G = abcd + dcba − {a, b, c, d}  a, b, c, d ∈ B .

We find a Gröbner basis of I(G) and identify the monomials in F B which do not have
the leading monomial of any Gröbner basis element as a subword. The cosets of these
monomials form a basis for U (J). If U (J) is finite dimensional, then we can determine its
structure using the Wedderburn decomposition of associative algebras [2]. We consider
the four smallest systems by increasing dimension.

Proposition 5.1. We have U (D11 ) ∼

= Q ⊕ M3 (Q), and hence up to isomorphism, D11 has
only two finite dimensional irreducible representations, the 1-dimensional trivial repre-
sentation and the 3-dimensional natural representation.

Proof. In the natural representation by 3 × 3 matrices, D11 has this basis:

⎡ ⎤ ⎡ ⎤
0 0 1 0 0 0
⎣
a= 1 0 0⎦, ⎣
b= 0 0 0⎦.
0 0 0 0 1 0

The only nonzero quadruple products are {a, b, a, a} = a and {b, a, a, b} = b. Hence
U (D11 ) = F a, b/I(G) where I(G) is generated by the following self-reduced set
{g1 , . . . , g10 } in lex order of leading monomials with a ≺ b:

a4 , aba2 + a2 ba − a, ab2 a, ba3 + a3 b, ba2 b − b,

baba + abab, b2 a2 + a2 b2 , b2 ab + bab2 , b3 a + ab3 , b4 .

We make the Gröbner basis algorithm deterministic by considering the pairs (gi , gj ) in
lex order of the subscripts (i, j); whenever their composition is reducible, we reduce the
highest term in deglex order using the generator whose leading monomial is lowest in
deglex order. For example, (g2 , g1 ) produces this composition:
 
g2 a2 − abg1 = aba2 + a2 ba − a a2 − aba4 = aba4 + a2 ba3 − a3 − aba4 = a2 ba3 − a3 .

We reduce this composition in three steps using g2 , g2 , g1 as follows:

 
a2 ba3 − a3 − ag2 a = a2 ba3 − a3 − a aba2 + a2 ba − a a = −a3 ba2 ,
 
−a3 ba2 + a2 g2 = −a3 ba2 + a2 aba2 + a2 ba − a = a4 ba − a3 ,
a4 ba − a3 − g1 ba = a4 ba − a3 − a4 ba = −a3 .

We cannot reduce further, so we record a3 as a (monic) composition. This process gives

the following set {h1 , . . . , h24 } of distinct nonzero reduced compositions:
70 M. Bremner, S. Madariaga / Journal of Algebra 412 (2014) 51–86

Table 3
Structure constants for U (D11 ).
1 a b a2 ab ba a2 b aba ba2 a2 ba
2 2 2
1 1 a b a ab ba a b aba ba a2 ba
a a a2 ab 0 a2 b aba 0 a2 ba −a2 ba + a 0
b b ba 0 ba2 0 0 b 0 0 ba
a2 a2 0 a2 b 0 0 a2 ba 0 0 a2 0
ab ab aba 0 −a2 ba + a 0 0 ab 0 0 aba
ba ba ba2 0 0 b 0 0 ba 0 0
a2 b a2 b a2 ba 0 a2 0 0 a2 b 0 0 a2 ba
aba aba −a2 ba + a 0 0 ab 0 0 aba 0 0
ba2 ba2 0 b 0 0 ba 0 0 ba2 0
a2 ba a2 ba a2 0 0 a2 b 0 0 a2 ba 0 0

a3 , ab2 , bab, b2 a, b3 , a3 b, a2 b2 , abab, ab3 , bab2 , a3 ba,

a3 b2 , a2 bab, a2 b3 + b2 , abab2 , bab3 , a3 bab, a2 bab2 − a3 b3 − ab2 ,
a2 bab2 − bab, a2 bab2 , abab3 , a3 bab2 − 2a2 b2 , a2 bab3 , a2 bab3 − ab3 .

We sort {g1 , . . . , g10 , h1 , . . . , h24 } by deglex order of leading monomials, and compute
the normal form of each element with respect to the preceding elements. Many normal
forms are 0, showing that I(G) is generated by these five elements:

b2 , a3 , bab, aba2 + a2 ba − a, ba2 b − b.

All compositions of these elements reduce to 0, so we have a Gröbner basis for I(G).
Only d = 10 monomials in F a, b are not divisible by any of the leading monomials:

1, a, b, a2 , ab, ba, a2 b, aba, ba2 , a2 ba.

The cosets of these monomials modulo I(G) form a basis for U (D11 ). We calculate the
structure constants for U (D11 ) by computing the normal forms of the products of these
monomials with respect to the Gröbner basis; see Table 3.
The radical of U (D11 ) is the nullspace of the Dickson matrix [4, Corollary 12]: the
d × d matrix Δ defined in terms of the structure constants ckij for U (D11 ):
⎡ ⎤
10 · · · · · 3 3 3 ·
⎢ · · ·
⎢ · 3 3 · · · ·⎥⎥
⎢ · · ·
⎢ 3 · · · · · ·⎥⎥
⎢ ⎥
⎢ · · 3 · · · · · · ·⎥  
⎢ ⎥
⎢ · 3 · · · · · · · 3⎥ 
d 
d
Δ=⎢ ⎥ Δij = ckji ck .
⎢ · 3 ·
⎢ · · · · · · ·⎥⎥
⎢ 3 · · ·⎥
k=1 =1
⎢ · · · 3 · · ⎥
⎢ ⎥
⎢ 3 · · · · · · 3 · ·⎥
⎢ ⎥
⎣ 3 · · · · · · · 3 ·⎦
· · · · 3 · · · · ·
 M. Bremner, S. Madariaga / Journal of Algebra 412 (2014) 51–86 71

This matrix has full rank, so U (D11 ) is semisimple. The center Z(U (D11 )) is the nullspace
of the d2 × d matrix from [4, Corollary 15]; the RCF of this matrix is

⎡ ⎤
· 1 · · · · · · · ·
⎢·
⎢ · 1 · · · · · · ·⎥⎥
⎢·
⎢ · · 1 · · · · · ·⎥⎥
⎢·
⎢ · · · 1 · · · · ·⎥⎥
⎢ ⎥.
⎢· · · · · 1 · · · ·⎥
⎢ ⎥
⎢·
⎢ · · · · · 1 · −1 ·⎥⎥
⎣· · · · · · · 1 −1 ·⎦
· · · · · · · · · 1

Hence a basis for Z(U (D11 )) consists of (the cosets of) 1 and c = a2 b + aba + ba2 . We
calculate c2 = c, so x(x − 1) is the minimal polynomial of c. Hence a basis for Z(U (D11 ))
consisting of orthogonal primitive idempotents is {c, 1 − c}.
We compute the two-sided ideals in U (D11 ) generated by c and 1 − c. We find that
dim 1 − c = 1 so that 1 − c ∼ = F with basis 1 = 1 − a2 b − aba − ba2 . Moreover,
dim c = 9, so we expect that c ∼ = M3 (F). To obtain an isomorphism, we compute
a basis for c, and find that it consists of the leading monomials of the Gröbner basis
(excluding 1). The coset of b generates a 3-dimensional left ideal in c with basis e1 = b,
e2 = ab, e3 = a2 b. We identify these with the standard basis vectors in F3 and determine
the elements of c corresponding to the matrix units Eij (1 ≤ i, j ≤ 3). For each i, j we
determine c1 , . . . , c9 ∈ Q so that the coset of the generic element,

E = c1 a + c2 b + c3 a2 + c4 ab + c5 ba + c6 a2 b + c7 aba + c8 ba2 + c9 a2 ba,

acts like the matrix unit Eij ; that is, Eij ek = δjk ei . From Table 3 we obtain:

Eb = c8 b + c1 ab + c3 a2 b,
Eab = c5 b + c7 ab + (c1 + c9 )a2 b,
Ea2 b = c2 b + c4 ab + c6 a2 b.

From this we obtain linear systems which determine c1 , . . . , c9 for each Eij , and solving
these systems gives this isomorphism c ∼
= M3 (F):

[E11 , E12 , E13 , E21 , E22 , E23 , E31 , E32 , E33 ]

 
←→ ba2 , ba, b, a − a2 ba, aba, ab, a2 , a2 ba, a2 b .

The set {1, Eij | 1 ≤ i, j ≤ 3} is a basis for U (D11 ) which reflects the decomposition into
simple two-sided ideals, and proves the proposition. 2
72 M. Bremner, S. Madariaga / Journal of Algebra 412 (2014) 51–86

Open Problem 5.2. Determine the structure of U (Dpq ) for all p, q. In particular, do we
always have the isomorphism U (Dpq ) ∼
= Q ⊕ Mn (Q) where n = pq + 12 q(q + 1)?

Proposition 5.3. We have U (C111 ) ∼ = Q ⊕ 2 M3 (Q), and hence up to isomorphism,

C111 has only three finite dimensional irreducible representations: 1-dimensional triv-
ial, 3-dimensional natural, and another 3-dimensional.

Proof. Our basis for C111 consists of matrices defining the natural representation:
⎡ ⎤ ⎡ ⎤ ⎡ ⎤
0 0 0 0 0 0 0 0 1
a = ⎣1 0 0⎦, b = ⎣0 0 0⎦, c = ⎣0 0 0⎦.
0 0 0 0 1 0 0 0 0

The nonzero quadrilinear products are: {a, c, b, a} = a, {b, c, a, b} = b, {c, b, a, c} = c.

The original generating set for I(G) contains 45 elements and is already self-reduced. We
obtain 290 distinct non-trivial compositions; the resulting set of 335 generators collapses
to a self-reduced set of only 13 elements, which is a Gröbner basis:

a2 , b2 , c2 , aba, aca, bab, bcb, cac, cbc,

acba + abca − a, bcab + bacb − b, cbac + cabc − c, cabca − ca.

Only 19 monomials in F a, b, c do not have a leading monomial of a Gröbner basis

element as a subword:

1, a, b, c, ab, ac, ba, bc, ca, cb, abc, acb, bac, bca, cab, cba, abca, bacb, cabc.

Hence U (C111 ) has dimension 19; as in the proof of Proposition 5.1 we find that U (C111 )
is semisimple, and its center has dimension 3. The cosets of these elements form a basis
of the center consisting of orthogonal primitive idempotents:

1 − abc − acb − bac − bca − cab − cba, abc + bca + cab, acb + bac + cba.

These elements generate simple two-sided ideals of dimensions 1, 9, 9 respectively. We

omit the isomorphisms of each simple ideal with a full matrix algebra. 2

Open Problem 5.4. Determine the structure of the universal associative envelope of Cpqr
for all p, q, r. Do we always have the isomorphism U (Cpqr ) ∼ = Q ⊕ 2 Mn (Q) where
n = pq + qr + rp?
√
Proposition 5.5. We have U (B2 )⊗Q K ∼
= K⊕3 M2 (K) where K = Q(β) with β = 1+ −3.
Hence over this quadratic extension field, B2 has only four finite dimensional irreducible
representations up to isomorphism: 1-dimensional trivial, 2-dimensional natural, and
another two 2-dimensional.
 M. Bremner, S. Madariaga / Journal of Algebra 412 (2014) 51–86 73

Proof. Our basis for B2 consists of matrices defining the natural representation:
     
1 0 0 0 0 1
a= , b= , c= .
0 0 0 1 1 0

There are 19 nonzero quadrilinear products of these generators (which we omit). The
original generating set for I(G) contains 45 elements and is already self-reduced. We
obtain 533 distinct non-trivial compositions; the resulting set of 578 generators collapses
to a self-reduced set of only 8 elements, which is a Gröbner basis:

ab, ba, ca − bc, cb − ac, c2 − b2 − a2 , a4 − a, b4 − b, b3 c + a3 c − c.

Only 13 monomials in F a, b, c do not have a leading monomial of a Gröbner basis

element as a subword:

1, a, b, c, a2 , ac, b2 , bc, a3 , a2 c, b3 , b2 c, a3 c.

Hence U (B2 ) has dimension 13; as before, we find that U (B2 ) is semisimple. The center
has dimension 4 and the following basis: 1, a + b, a2 + b2 , a3 + b3 . These elements are
not orthogonal primitive idempotents; unlike the previous two cases, to split the center
we must extend the base field.

• First splitting: set e = 1 and x = a + b. The minimal polynomial of x is t4 − t =

t(t3 − 1), so we set y = x and z = x3 − e. Then y generates a 3-dimensional central
ideal with identity element a3 + b3 , and ζ1 = −z is a primitive idempotent.
• Second splitting: set e = a3 + b3 and x = a + b. The minimal polynomial of x is
t3 − 1 = (t − 1)(t2 + t + 1), so we set y = x − e and z = x2 + x + e. Then y generates a
2-dimensional central ideal with identity element − 13 (a + b) − 13 (a2 + b2 ) + 23 (a3 + b3 ),
and ζ2 = 13 z is a primitive idempotent.
• Third splitting: set e = − 13 (a + b) − 13 (a2 + b2 ) + 23 (a3 + b3 ) and x = (a + b) − (a3 + b3 ).
√
The minimal polynomial of x is t2 + 3t + 3 = (t − α)(t − ᾱ) where α = 12 (−3 + −3 ).
We set y = x − αe and z = x − ᾱe; each generates a 1-dimensional central ideal. We
get the idempotents ζ3 = √−1 −3
y and ζ4 = √1−3 z.

Using the fact that ab = ba = 0 in U (B2 ) since these monomials are in the Gröbner
basis, we obtain the central basis of orthogonal primitive idempotents:

1 
ζ1 = 1 − a3 − b3 , a + b + a2 + b2 + a3 + b3 ,
ζ2 =
3
1    
ζ3 = − β(a + b) + β̄ a2 + b2 − 2 a3 + b3 ,
6
1    
ζ4 = − β̄(a + b) + β a2 + b2 − 2 a3 + b3 ,
6
74 M. Bremner, S. Madariaga / Journal of Algebra 412 (2014) 51–86

√
where β = 1 + −3. These four elements generate simple two-sided ideals of dimensions
1, 4, 4, 4 respectively. We omit the calculation of the isomorphisms of each simple ideal
with a full matrix algebra. 2

Open Problem 5.6. Determine the structure of the universal associative envelope of Bn
for all n. In particular, after a quadratic extension K of Q, do we always have the
isomorphism U (Bn ) ⊗Q K ∼ = K ⊕ 3 Mn (K)?
√
Proposition 5.7. We have U (A2 ) ⊗Q K ∼
= K ⊕ 6 M2 (K) where K = Q(β), β = 1 + −3.
Hence over this quadratic extension, A2 has exactly seven finite dimensional irreducible
representations up to isomorphism: 1-dimensional trivial, 2-dimensional natural, and
another five 2-dimensional.

Proof. Our basis for A2 consists of these matrices in the natural representation:
     
1 0 0 1 0 0 0 0
a= , b= , c= , d= .
0 0 0 0 1 0 0 1

There are 31 nonzero quadrilinear products of these generators (which we omit). The
original generating set of I(G) contains 136 elements and is already self-reduced. We ob-
tain 2769 distinct non-trivial compositions; the resulting set of 2905 generators collapses
to a self-reduced set of 25 elements, which is a Gröbner basis:

ad, b2 , bd − ab, c2 , cd − ac, da, db − ba, dc − ca, d2 − cb − bc + a2 ,

aba, aca, acb + abc − a3 , bab, bca − abc, bcb − ba2 − a2 b, cac,
cba + abc − a3 , cbc − ca2 − a2 c, a4 − a, ba3 + a3 b − b, ba2 b, ca3 + a3 c − c,
ca2 b + ba2 c − d, ca2 c, cabc + a3 c − c.

Only 25 monomials in F a, b, c, d do not have a leading monomial of a Gröbner basis

element as a subword:

1, a, b, c, d, a2 , ab, ac, ba, bc, ca, cb, a3 , a2 b,

a2 c, abc, ba2 , bac, ca2 , cab, a3 b, a3 c, a2 bc, ba2 c, a3 bc.

Hence U (A2 ) has dimension 25; as before, we find that U (A2 ) is semisimple. The center
has dimension 7 and the following basis:

1, a + d, bc + cb, a3 − abc + bac, abc + cab,

1 
a − a2 bc + ba2 c, − a2 + bc + a3 bc.
2
 M. Bremner, S. Madariaga / Journal of Algebra 412 (2014) 51–86 75

These elements are not orthogonal primitive idempotents; as in the previous case, to
split the center we must extend the base field. We obtain the following central basis
consisting of orthogonal primitive idempotents:

1 − a3 − bac − cab,
1 
d − a2 + cb + abc + cab + a2 bc − ba2 c + 2a3 bc ,
3
1 
a + a2 + bc + a3 − abc + bac − a2 bc + ba2 c − 2a3 bc ,
3
1 
− βa + β̄a2 + β̄bc − 2a3 + 2abc − 2bac − βa2 bc + βba2 c − 2β̄a3 bc ,
6
1 
− βd − β̄a2 + β̄cb − 2abc − 2cab + βa2 bc − βba2 c + 2β̄a3 bc ,
6
1 
− β̄a + βa2 + βbc − 2a3 + 2abc − 2bac − β̄a2 bc + β̄ba2 c − 2βa3 bc ,
6
1 
− β̄d − βa2 + βcb − 2abc − 2cab + β̄a2 bc − β̄ba2 c + 2βa3 bc ,
6
√
where β = 1+ −3. These seven elements generate simple two-sided ideals of dimensions
1, 4, 4, 4, 4, 4, 4 respectively. We omit the calculation of the isomorphisms of each simple
ideal with a full matrix algebra. 2

Open Problem 5.8. Determine the structure of the universal associative envelope for An
for all n. In particular, after making a quadratic extension K of Q, do we always have
the isomorphism U (An ) ⊗Q K ∼ = K ⊕ 6 Mn (K)?

6. Defining identities for anti-Jordan quadruple systems

In this and the next two sections we present analogous results for the anti-tetrad
[a, b, c, d] = abcd − dcba. Since the methods are very similar, we omit most details.

Lemma 6.1. Every multilinear identity in degree 4 satisfied by the anti-tetrad is a conse-
quence of the skew-symmetry [a, b, c, d] + [d, c, b, a] ≡ 0.

Theorem 6.2. Every multilinear identity in degree 7 satisfied by the anti-tetrad is a con-
sequence of skew-symmetry in degree 4 and these two identities in degree 7:
       
[a, b, c, d], e, f, g − [a, b, f, e], d, c, g + [d, c, f, e], a, b, g + g, [b, a, d, c], f, e
   
− g, [b, a, e, f ], c, d + g, [c, d, e, f ], b, a ≡ 0,
       
[a, b, c, d], e, f, g − [a, f, g, c], b, e, d + [c, a, b, d], g, e, f − [c, b, a, f ], g, e, d
       
− [c, g, d, b], a, e, f − [c, g, e, f ], a, b, d + [c, g, e, d], b, a, f − [b, a, c, f ], e, d, g
       
− [b, c, f, g], a, e, d + [b, e, d, g], a, c, f − [g, f, e, d], c, b, a + [g, d, e, f ], c, a, b
76 M. Bremner, S. Madariaga / Journal of Algebra 412 (2014) 51–86

Table 4
S7 -module multiplicities for the anti-tetrad in degree 7.
λ dλ Symmetries Expansions New
rows cols sym rows cols rank null
7 1 3 2 2 2 1 0 2 0
61 6 18 12 8 12 6 3 9 1
52 14 42 28 16 28 14 8 20 4
512 15 45 30 14 30 15 6 24 10
43 14 42 28 16 28 14 7 21 5
421 35 105 70 34 70 35 18 52 18
413 20 60 40 16 40 20 10 30 14
32 1 21 63 42 22 42 21 9 33 11
322 21 63 42 22 42 21 12 30 8
3212 35 105 70 34 70 35 17 53 19
314 15 45 30 14 30 15 9 21 7
23 1 14 42 28 16 28 14 7 21 5
22 13 14 42 28 16 28 14 6 22 6
215 6 18 12 8 12 6 3 9 1
17 1 3 2 2 2 1 0 2 0

       
− [f, a, b, d], e, g, c − [f, c, e, d], g, a, b − a, [b, c, g, f ], e, d − c, [a, b, d, g], e, f
       
+ c, [b, a, f, g], e, d + b, [a, c, g, d], e, f − b, [a, g, f, c], e, d + b, [a, g, d, e], c, f
 
+ d, [c, b, a, e], f, g ≡ 0.

Proof. For the anti-tetrad the rank is 2519 and the nullity is 2521. 2

Definition 6.3. An anti-Jordan quadruple system (AJQS) is a vector space Q over a field
F of characteristic 0 or p > 7 with a quadrilinear map [−, −, −, −] : Q4 → Q satisfying
the skew-symmetry of Lemma 6.1 and the identities of Theorem 6.2.

Lemma 6.4. Over a field of characteristic 0 or p > 7, for each partition λ column “New” of
Table 4 gives the multiplicity of [λ] in the kernel of the expansion map for the anti-tetrad.

Since the multiplicity for the last partition is 0, we expect to find simpler nonlinear
identities whose linearizations generate the S7 -module of all multilinear identities for the
anti-tetrad in degree 7 (compare Remark 3.6). This computation consists of two parts.

Part 1. For each partition of 7 except the last, we construct the corresponding multiset
of variables:

a7 , a6 b, a5 b2 , a5 bc, a4 b3 , a4 b2 c, a4 bcd, a3 b3 c,
a3 b2 c2 , a3 b2 cd, a3 bcde, a2 b2 c2 d, a2 b2 cde, a2 bcdef.

For each partition λ, we generate all permutations of the multiset; we write Aλ for the
number of these associative monomials. We then apply each of the two association types
to all the permutations, and straighten them using the skew-symmetry in degree 4, to
 M. Bremner, S. Madariaga / Journal of Algebra 412 (2014) 51–86 77

obtain a set of monomials in normal form; we write Qλ for the number of these quater-
nary monomials. For each λ, the nonlinear expansion matrix Eλ has size Aλ × Qλ . We
compute a lattice basis for the integer nullspace of Eλ and sort these vectors by increas-
ing Euclidean length. We write Rλ and Nλ for the rank and nullity of the expansion
matrix. We linearize the nonlinear identities represented by the nullspace basis vectors.
We then study the S7 -module generated by these multilinear identities, and extract a
small subset of these identities which generates the S7 -module. For each partition λ we
retain only those identities which do not belong to the submodule generated by all pre-
vious identities (for λ and all previous partitions). We write Gλ for the number of new
generators obtained for partition λ. We summarize the results of these computations:

λ 7 61 52 512 43 421 413 32 1 322 3212 314 23 1 22 13 215

Aλ 1 7 21 42 35 105 210 140 210 420 840 630 1260 2520
Qλ 0 4 16 36 28 96 204 128 196 408 840 612 1248 2520
Rλ 0 3 11 20 18 53 104 69 107 209 419 317 629 1259
Nλ 0 1 5 16 10 43 100 59 89 199 421 295 619 1261
Gλ 0 1 4 10 5 18 14 11 8 19 7 5 6 1

Altogether this gives 109 nonlinear identities whose linearizations generate the 2521-di-
mensional S7 -module of all multilinear identities satisfied by the anti-tetrad in degree 7.

Part 2. Starting at the end of the list of module generators obtained from Part 1, and
working toward the beginning of the list, we remove the current (linearized) identity
from the list and compute the dimension of the submodule generated by the remaining
identities. If the dimension reaches 2521, then the current identity is redundant, and we
remove it from the list. In this case, the current identity is a consequence of identities
which follow it in the list; by the way in which we extracted these generators, we know
that no identity is a consequence of identities which precede it in the list. The result of
this elimination process is a set of only three nonlinear identities with respectively 5, 5
and 12 terms, whose coefficients are all ±1. Even though we have three identities now
instead of the two identities of Theorem 6.2 with 6 and 21 terms, the total number of
terms is smaller. This computation establishes the next result.

Theorem 6.5. Every multilinear identity in degree 7 satisfied by the anti-tetrad is a con-
sequence of skew-symmetry in degree 4 and the linearizations of these three nonlinear
identities in degree 7, corresponding to partitions 22 13 (twice) and 215 :
       
[a, b, c, b], d, e, a − [a, b, c, b], e, d, a − [a, d, e, a], b, c, b − a, [b, c, b, d], e, a
 
+ a, [b, c, b, e], d, a ≡ 0,
       
[a, c, a, b], b, d, e − [a, c, a, e], d, b, b + [b, b, d, e], a, c, a + b, [a, c, a, b], d, e
 
− e, [a, c, a, d], b, b ≡ 0,
78 M. Bremner, S. Madariaga / Journal of Algebra 412 (2014) 51–86

       
[a, a, b, c], d, e, f + [a, a, e, f ], d, c, b − [a, a, e, b], c, d, f − [a, d, c, b], a, e, f
       
− [f, a, a, c], d, e, b + [f, d, c, b], e, a, a + [f, d, e, b], a, a, c − a, [a, b, c, d], e, f
       
− f, [a, a, d, e], b, c + f, [a, a, c, d], e, b + f, [a, b, e, d], a, c − c, [a, d, e, b], a, f ≡ 0.

7. Special identities in degree 10 for the anti-tetrad

The notions of special and exceptional for AJQS, and special identities for the anti-
tetrad, are the obvious analogues of Definition 4.1.

Open Problem 7.1. Do there exist exceptional AJQS?

Theorem 7.2. For the anti-tetrad, and for each partition λ of 10, columns “symm”,
“symmlift”, “null” and “New” in the corresponding row of Table 5 contain respectively
the multiplicity of [λ] in the modules Symm(10), Old(10), All(10) and New(10).

Open Problem 7.3. Determine a minimal set of multilinear identities in degree 10 for the
anti-tetrad which generate New(10) as an S10 -module.

In the rest of this section we present some nonlinear special identities in degree 10
for the anti-tetrad. These identities correspond to the partitions λ = st (s + t = 10) for
which column “New” in Table 5 is nonzero, namely λ = 82, 73, 64, 52 . The corresponding
Young diagrams have only two rows of lengths s and t. For these λ, the structure theory
of QS10 allows us to assume that the identity is a symmetric function of two disjoint
sets of variables, so we may reduce the size of the computations by working with non-
linear monomials which have only two variables a and b occurring respectively s and t
times.

Theorem 7.4. The multihomogeneous identities in Fig. 2 are satisfied by the anti-tetrad
in every associative algebra but are not consequences of the defining identities for AJQS.

Remark 7.5. For the anti-tetrad, we obtained better results using the LLL algorithm
with a higher value of the reduction parameter. We illustrate with λ = 64; the underlying
variables are a6 b4 which produce 210 associative monomials. The eight association types
produce respectively 82, 82, 82, 82, 32, 32, 14, 14 nonlinear quaternary monomials for a
total of 420. The 210 × 420 expansion matrix has rank 99 and nullity 321. Computing
the HNF with an initial application of LLL produces a lattice basis of the nullspace with
size ≈ 421 using Eq. (1); the coefficients belong to {0, ±1, ±2, −3}. Applying LLL with
parameter 3/4 reduces the basis size to ≈ 389 with coefficients {0, ±1, ±2}. Applying
LLL again with parameter 99/100 reduces the basis size to ≈ 337 with coefficients
{0, ±1, −2}. We obtain three special nonlinear identities for this partition.
 M. Bremner, S. Madariaga / Journal of Algebra 412 (2014) 51–86 79

Table 5
S10 -module multiplicities for the anti-tetrad in degree 10.
# λ dλ Symmetries and liftings Expansions New
rows cols sym lift rows cols rank null
1 10 1 9 8 8 8 8 1 0 8 ·
2 91 9 81 72 64 68 72 9 4 68 ·
3 82 35 315 280 222 264 280 35 15 265 1
4 812 36 324 288 218 269 288 36 19 269 ·
5 73 75 675 600 442 555 600 75 40 560 5
6 721 160 1440 1280 896 1197 1280 160 80 1200 3
7 713 84 756 672 446 632 672 84 40 632 ·
8 64 90 810 720 524 677 720 90 40 680 3
9 631 315 2835 2520 1722 2352 2520 315 160 2360 8
10 622 225 2025 1800 1200 1693 1800 225 105 1695 2
11 6212 350 3150 2800 1798 2619 2800 350 180 2620 1
12 614 126 1134 1008 612 948 1008 126 60 948 ·
13 52 42 378 336 236 309 336 42 26 310 1
14 541 288 2592 2304 1568 2155 2304 288 144 2160 5
15 532 450 4050 3600 2390 3365 3600 450 230 3370 5
16 5312 567 5103 4536 2960 4258 4536 567 276 4260 2
17 522 1 525 4725 4200 2698 3935 4200 525 265 3935 ·
18 5213 448 4032 3584 2240 3360 3584 448 224 3360 ·
19 515 126 1134 1008 604 942 1008 126 66 942 ·
20 42 2 252 2268 2016 1358 1899 2016 252 116 1900 1
21 4 2 12 300 2700 2400 1562 2239 2400 300 160 2240 1
22 432 210 1890 1680 1124 1568 1680 210 110 1570 2
23 4321 768 6912 6144 4032 5760 6144 768 384 5760 ·
24 4313 525 4725 4200 2702 3940 4200 525 260 3940 ·
25 423 300 2700 2400 1582 2260 2400 300 140 2260 ·
26 422 12 567 5103 4536 2944 4245 4536 567 291 4245 ·
27 4214 350 3150 2800 1810 2630 2800 350 170 2630 ·
28 416 84 756 672 442 627 672 84 44 628 1
29 33 1 210 1890 1680 1132 1580 1680 210 100 1580 ·
30 3 2 22 252 2268 2016 1338 1880 2016 252 136 1880 ·
31 32 212 450 4050 3600 2402 3380 3600 450 220 3380 ·
32 3 2 14 225 2025 1800 1184 1680 1800 225 120 1680 ·
33 323 1 288 2592 2304 1568 2160 2304 288 144 2160 ·
34 322 13 315 2835 2520 1726 2365 2520 315 155 2365 ·
35 3215 160 1440 1280 896 1200 1280 160 80 1200 ·
36 317 36 324 288 222 271 288 36 16 272 1
37 25 42 378 336 244 320 336 42 16 320 ·
38 2 4 12 90 810 720 516 670 720 90 50 670 ·
39 2 3 14 75 675 600 446 565 600 75 35 565 ·
40 2 2 16 35 315 280 218 260 280 35 20 260 ·
41 218 9 81 72 64 68 72 9 4 68 ·
42 10 1 9 8 8 8 8 1 0 8 ·

8. Four families of finite dimensional anti-Jordan quadruple systems; examples of

universal associative envelopes

We define four families of finite dimensional AJQS, denoted A− − − −

n , Bn , Cpqr , Dpq ; these
are subspaces of matrix algebras which are closed under the anti-tetrad.

• A−
n consists of all n × n matrices; the dimension is n .
2
−
• Bn consists of all n × n skew-symmetric matrices; the dimension is 12 n(n − 1).
80 M. Bremner, S. Madariaga / Journal of Algebra 412 (2014) 51–86

          
[a, a, b, a], a, b, a , a, a, a − a, [a, a, a, b], a, b, a , a, a − a, a, [a, a, b, a], a, a , b, a
   
+ a, a, [a, a, b, a], b, a , a, a ≡ 0,
         
[a, a, b, a], a, b, a , b, a, a + a, [a, b, a, b], a, a , b, a, a − a, [a, a, b, a], b, a, a , b, a
         
− a, [a, b, a, b], a, a, b , a, a − a, [a, b, a, b], a, b, a , a, a − [a, a, b, a], [a, b, a, b], a, a ≡ 0,
          
[a, a, b, a], b, b, a , a, a, a − a, [a, a, a, b], a, b, b , a, a + a, [a, a, b, a], a, b, b , a, a
           
− a, [a, a, b, a], b, b, a , a, a + a, [a, a, b, b], a, a, b , a, a − a, [a, a, b, b], a, b, a , a, a ≡ 0,
          
[a, a, b, a], a, b, a , a, a, b − a, [a, a, a, b], a, b, a , a, b + a, [a, a, b, a], a, a, b , a, b
           
− b, [a, a, b, a], a, b, a , a, a − a, a, [a, a, b, a], a, b , a, b + a, a, [a, a, b, a], b, a , a, b ≡ 0,
        
a, [a, a, b, a], a, a , b, b, a − a, [a, a, b, a], a, b , a, a, b + a, [a, a, b, a], a, b , b, a, a
          
+ a, [a, a, b, b], a, a , b, a, a − a, [a, a, b, a], a, b, b , a, a − a, [a, a, b, b], a, b, a , a, a
   
+ [a, a, a, b], [a, a, b, a], a, b − [a, a, a, b], b, [a, a, b, a], a ≡ 0,
        
[a, a, a, b], a, a, a , b, b, a − [a, a, a, b], a, b, b , a, a, a + a, [a, a, a, b], a, b , b, a, a
        
− a, [a, a, b, a], a, a , b, b, a + a, [a, a, b, b], a, a , a, a, b − a, [a, a, b, b], b, a , a, a, a
           
+ a, [a, a, b, a], b, b, a , a, a − a, [a, a, b, b], a, a, b , a, a + b, a, [a, a, a, b], a, b , a, a
 
− [a, a, a, b], [a, a, b, b], a, a ≡ 0,
         
b, [a, a, a, b], a, b , a, a, b + b, [a, a, b, a], a, b , a, a, b − b, [a, a, a, b], b, a, a , a, b
   
− b, [a, a, b, a], a, a, b , a, b ≡ 0,
        
[a, a, b, a], b, b, a , b, a, a − a, [a, b, a, b], a, a , b, b, a − a, [a, b, a, b], b, a , a, b, a
       
− a, [a, a, b, a], b, a, b , b, a + a, [a, b, a, b], a, a, b , b, a ≡ 0,
          
[a, a, b, a], b, b, a , b, a, a − a, [a, a, b, a], b, b, a , b, a − a, [a, b, a, b], a, b, b , a, a
       
+ a, [b, a, b, b], a, a, b , a, a − a, [b, a, b, b], a, b, a , a, a ≡ 0,
          
[a, a, b, a], b, b, a , b, a, b − a, [a, b, a, b], a, b, b , a, b + a, [b, a, b, b], a, a, b , a, b
       
− a, [b, a, b, b], a, b, a , a, b − b, [a, a, b, a], b, b, a , b, a ≡ 0.

Fig. 2. Some special identities for the anti-tetrad in degree 10.

For the remaining systems, we choose integers p, q, r ≥ 1 and consider block matrices of
size p + q + r as in Eq. (3).

−
• Cpqr (p ≥ q, r) consists of all matrices (3); the dimension is pq + qr + rp.
−
• Dpq consists of all matrices (3) with q = r where Mpq = −Mqp t
and Mqq is skew-
symmetric; the dimension is pq + 2 q(q − 1).
1

The smallest non-trivial systems in each family are A− − − −

2 , B3 , C111 , D21 with dimensions
4, 3, 3, 2 respectively. We consider these systems by increasing dimension.
 M. Bremner, S. Madariaga / Journal of Algebra 412 (2014) 51–86 81

−
Proposition 8.1. The universal associative envelope U (D21 ) is Z-graded by degree and
infinite dimensional; in degrees n ≥ 6 it is linearly isomorphic to the (commutative)
polynomial algebra F[a, b].

−
Proof. This is the basis for D21 which defines the natural representation:
⎡ ⎤ ⎡ ⎤
0 0 0 −1 0 0 0 0
⎢0 0 0 0 ⎥ ⎢0 0 0 −1 ⎥
a=⎢
⎣1
⎥, b=⎢ ⎥.
0 0 0 ⎦ ⎣0 1 0 0 ⎦
0 0 0 0 0 0 0 0

Every quadrilinear product in these generators is 0, so the original 6 ideal generators

−
have no terms of degree 1; it follows that U (D21 ) is graded by degree:

aba2 − a2 ba, ba3 − a3 b, baba − abab, b2 a2 − a2 b2 , b2 ab − bab2 , b3 a − ab3 .

These generators produce 12 distinct non-trivial compositions:

a3 ba − a4 b, a2 b2 a − a3 b2 , ba2 ba − a2 bab, ba2 b2 − a2 b3 ,

bab2 a − abab2 , bab3 − ab4 , a2 bab2 − a3 b3 , ba2 bab − abab2 a,
a4 bab − a5 b2 , a3 bab2 − a4 b3 , abab4 − a2 b5 , ba2 bab2 − a3 b4 .

After self-reduction, the 18 generators become a Gröbner basis of 13 elements:

aba2 − a2 ba, ba3 − a3 b, baba − abab, b2 a2 − a2 b2 , b2 ab − bab2 , b3 a − ab3 ,

a3 ba − a4 b, a2 b2 a − a3 b2 , ba2 ba − a2 bab, ba2 b2 − a2 b3 , bab2 a − abab2 ,
bab3 − ab4 , a2 bab2 − a3 b3 .

There is an infinite set of monomials in F a, b which are not divisible by the leading
−
monomial of any element of the Gröbner basis, and U (D21 ) is infinite dimensional. Here
−
are the basis monomials for U (D21 ) in degrees ≤ 5:

1, a, b, a2 , ab, ba, b2 , a3 , a2 b, aba, ab2 , ba2 , bab, b2 a, b3 ,

a4 , a3 b, a2 ba, a2 b2 , abab, ab2 a, ab3 , ba2 b, bab2 , b4 ,
a5 , a4 b, a3 b2 , a2 bab, a2 b3 , abab2 , ab4 , b5 .

In degree n ≥ 6 a basis consists of the monomials an−i bi for i = 0, . . . , n. To see this,

note that every element of the Gröbner basis has the form m1 − m2 , and replacing m1
by m2 moves a to the left and b to the right; now use induction on n. Thus in degrees
−
n ≥ 6, U (D21 ) is isomorphic to the (commutative) polynomial algebra Q[a, b], but there
is some noncommutativity in degrees n ≤ 5. 2
82 M. Bremner, S. Madariaga / Journal of Algebra 412 (2014) 51–86

−
Remark 8.2. We did similar computations for D31 . The original set of 36 homogeneous
generators produced 278 distinct non-trivial compositions; after self-reduction the 314
−
generators became a Gröbner basis of 94 elements. It follows that U (D31 ) is Z-graded
and infinite dimensional, and that for degree n ≥ 6, the dimension of the homogeneous
 
subspace is n+2
2 .

−
Conjecture 8.3. For all p ≥ 2, the universal associative envelope U (Dp1 ) is Z-graded and
infinite dimensional, and for degree n ≥ 6, the dimension of the homogeneous component
 
is n+p−1
p−1 , equal to that of the (commutative) polynomial algebra in p variables.

Open Problem 8.4. Study the representation theory of the anti-Jordan triple systems
− −
Dp1 and their universal associative envelopes U (Dp1 ). Comparing our results with [14]
−
suggests that U (Dp1 ) is likely to be a (generalized) down-up algebra.

−
Remark 8.5. We did similar computations for D12 . The original set of 36 non-
homogeneous generators produced 341 distinct non-trivial compositions; after self-
reduction the 377 generators became 19 elements. This set produced one more non-trivial
composition, resulting in a Gröbner basis of 20 elements:

c2 , a3 , a2 b, aba, ab2 , aca, ba2 , bab, b2 a, b3 ,

bca + acb, bcb, cac, cbc, acba + a2 cb − a, bacb + acb2 − b,
cbac − cabc − c, a2 cb2 − ab, acb2 c − bc, ca2 cb − ca.

−
From this it follows that U (D12 ) has dimension 26.

Open Problem 8.6. Study the representation theory of the anti-Jordan triple systems
− − ∼
Dpq for q ≥ 2. In particular, do we always have U (Dpq ) = Q ⊕ Mp+2q (Q)?

Proposition 8.7. We have U (C111−

) ∼
= Q ⊕ 2 M3 (Q), and hence up to isomorphism,
−
C111 has only three finite dimensional irreducible representations: 1-dimensional triv-
ial, 3-dimensional natural, and another 3-dimensional.

−
Proof. Our standard basis for C111 is the same as that for C111 ; see the proof of Propo-
sition 5.3. Only three quadrilinear products are nonzero:

[a, c, b, a] = a, [b, c, a, b] = −b, [c, b, a, c] = c.

The original set G of generators for the ideal I(G) contains 36 elements and is already
self-reduced. We obtain 333 distinct non-trivial compositions of these generators; the
resulting set of 369 generators collapses to a self-reduced set of only 13 elements, which
 M. Bremner, S. Madariaga / Journal of Algebra 412 (2014) 51–86 83

is a Gröbner basis:

a2 , b2 , c2 , aba, aca, bab, bcb, cac, cbc,

acba − abca − a, bcab − bacb + b, cbac − cabc − c, cabca + ca.

The leading monomials of these elements are the same as those of the Gröbner basis in
−
the proof of Proposition 5.3. Hence U (C111 ) has dimension 19, and a basis consists of the
−
same monomials as for U (C111 ) in the proof of Proposition 5.3. We find that U (C111 ) is
semisimple, and its center has dimension 3. The cosets of these elements form a central
basis of orthogonal primitive idempotents:

1 + abc − acb − bac + bca + cab − cba, −abc − bca − cab, acb + bac + cba.

These elements generate simple two-sided ideals of dimensions 1, 9, 9 respectively. We

omit the isomorphisms of each simple ideal with a full matrix algebra. 2

Open Problem 8.8. Study the representation theory of the anti-Jordan triple systems
−
Cpqr −
. In particular, do we always have U (Cpqr )∼
= Q ⊕ 2 Mp+q+r (Q)?
√
Proposition 8.9. We have U (B3− ) ⊗Q K ∼
= K ⊕ 3 M3 (K) where K = Q(β), β = 1 + −3.
Hence up to isomorphism, B3− has four finite dimensional irreducible representations
over K: 1-dimensional trivial, 3-dimensional natural, and two other 3-dimensional.

Proof. Our standard basis for B3− consists of these three 3 × 3 matrices defining the
natural representation:
⎡ ⎤ ⎡ ⎤ ⎡ ⎤
0 −1 0 0 0 −1 0 0 0
a = ⎣1 0 0⎦, b = ⎣0 0 0 ⎦, c = ⎣0 0 −1 ⎦ .
0 0 0 1 0 0 0 1 0

The original set G of generators for the ideal I(G) contains 36 elements and is already
self-reduced. We obtain 385 distinct non-trivial compositions of these generators; the
resulting set of 421 generators collapses to a self-reduced set of 30 elements, which is a
Gröbner basis:

aba, aca, ac2 + ab2 − a3 , bab, b2 a + ab2 − a3 , b3 − ba2 − a2 b, bca + acb,

bcb, bc2 − a2 b, ca2 − b2 c, cab + bac, cac, cba + abc, cb2 − a2 c, cbc,
c2 a − ab2 , c2 b − ba2 , c3 − b2 c − a2 c, a2 cb − a2 bc − a, ab2 c, ba3 − a3 b − c,
ba2 c, bacb + a3 c − b, a5 + cb − bc, a4 b + ac, a4 c − ab, a3 b2 + cb,
1 1 1
a3 bc + c2 − b2 + a2 , ba2 b2 − ca, ba2 bc + ba.
2 2 2
84 M. Bremner, S. Madariaga / Journal of Algebra 412 (2014) 51–86

Only 28 monomials in F a, b, c do not have a leading monomial of a Gröbner basis

element as a subword:

1, a, b, c, a2 , ab, ac, ba, b2 , bc, ca, cb, c2 , a3 , a2 b, a2 c,

ab2 , abc, acb, ba2 , bac, b2 c, a4 , a3 b, a3 c, a2 b2 , a2 bc, ba2 b.

Hence U (B3− ) has dimension 28; as before, we find that U (B3− ) is semisimple. The center
has dimension 4; finding a basis of orthogonal primitive idempotents requires a quadratic
extension of Q. The required basis is

1 + abc − acb − bac,

1 
− a2 + b2 + c2 + 2abc − 2acb − 2bac − 2a4 − 2ba2 b ,
6
1   2   
β a + b2 + c2 − 4abc + 4acb + 4bac − 2β̄ a4 + ba2 b ,
12
1   2   
β̄ a + b2 + c2 − 4abc + 4acb + 4bac − 2β a4 + ba2 b ,
12
√
where β = 1 + −3. These four elements generate simple two-sided ideals of dimensions
1, 4, 4, 4 respectively. We omit the calculation of the isomorphisms of each simple ideal
with a full matrix algebra. 2

Open Problem 8.10. Study the representation theory of the anti-Jordan triple systems
Bn− . In particular, over a quadratic extension K of Q do we always have U (Bn− ) ⊗Q K ∼
=
K ⊕ 3 Mn (K)?

∼ √
Proposition 8.11. We have U (A− 2 ) ⊗Q K = K ⊕ 6 M2 (K) where K = F (β), β = 1 + −3.
Hence up to isomorphism, A− 2 has seven finite dimensional irreducible representations
over K: 1-dimensional trivial, 2-dimensional natural, and five other 2-dimensional.

Proof. Our standard basis for A− 2 consists of the four 2 × 2 matrix units as in the proof
of Proposition 5.7. The original set G of generators for the ideal I(G) contains 120
elements and is already self-reduced. We obtain 2821 distinct non-trivial compositions
of these generators; the resulting set of 2941 generators collapses after three iterations
to a self-reduced set of only 28 elements, which is a Gröbner basis:

ad, b2 , bd − ab, c2 , cd − ac, da, db − ba, dc − ca,

d2 − cb − bc + a2 , aba, aca, acb + abc − a3 , bab, bca − abc,
1 1
bcb − ba2 − a2 b, cac, cba + abc − a3 , cbc − ca2 − a2 c, a2 bc − a4 − a,
2 2
ba3 − a3 b + b, ba2 b, ca3 − a3 c − c, ca2 b − ba2 c − d, 2
ca c,
cabc − a3 c − c, a4 b − ab, a4 c + ac, a6 − 2abc + a3 .
 M. Bremner, S. Madariaga / Journal of Algebra 412 (2014) 51–86 85

Only 25 monomials in F a, b, c, d do not have a leading monomial of a Gröbner basis

element as a subword:

1, a, b, c, d, a2 , ab, ac, ba, bc, ca, cb, a3 , a2 b,

a2 c, abc, ba2 , bac, ca2 , cab, a4 , a3 b, a3 c, ba2 c, a5 .

Hence U (A− −
2 ) has dimension 25; as before, we find that U (A2 ) is semisimple. The center
has dimension 7; splitting it requires a quadratic extension of Q. We obtain this central
basis of orthogonal primitive idempotents:

1 + a3 − 2abc + bac − cab,

1 
a + 2d + 2cb + 2abc + 2cab + a4 + 2ba2 c + 2a5 ,
6
1 
− a − 2bc + 2a3 − 2abc + 2bac − a4 − 2ba2 c + 2a5 ,
6
1  
βa − 2β̄bc − 4a3 + 4abc − 4bac − βa4 − 2βba2 c + 2β̄a5 ,
12
1  
β̄a − 2βbc − 4a3 + 4abc − 4bac − β̄a4 − 2β̄ba2 c + 2βa5 ,
12
1  
− βa + 2βd + 2β̄cb − 4abc − 4cab + βa4 + 2βba2 c + 2β̄a5 ,
12
1  
− β̄a + 2β̄d + 2βcb − 4abc − 4cab + β̄a4 + 2β̄ba2 c + 2βa5 ,
12
√
where β = 1+ −3. These seven elements generate simple two-sided ideals of dimensions
1, 4, 4, 4, 4, 4, 4 respectively. We omit the calculation of the isomorphisms of each simple
ideal with a full matrix algebra. 2

Open Problem 8.12. Study the representation theory of the anti-Jordan triple systems
A− − ∼
n . In particular, over a quadratic extension K of Q do we always have U (An ) ⊗Q K =
K ⊕ 6 Mn (K)?

References

[1] S. Bashir, Automorphisms of simple anti-Jordan pairs, PhD thesis, University of Ottawa, Canada,
2008.
[2] M.R. Bremner, How to compute the Wedderburn decomposition of a finite dimensional associative
algebra, Groups Complex. Cryptol. 3 (1) (2011) 47–66.
[3] M.R. Bremner, Lattice Basis Reduction: An Introduction to the LLL Algorithm and Its Applica-
tions, Pure Appl. Math., vol. 300, CRC Press, Boca Raton, USA, 2012.
[4] M.R. Bremner, Free associative algebras, noncommutative Gröbner bases, and universal associa-
tive envelopes for nonassociative structures, in: Proceedings of the Third Mile High Conference
on Nonassociative Algebra, Denver, Colorado, USA, August 11–17, 2013, Comment. Math. Univ.
Carolin. (2014), in press, arXiv:1303.0920.
[5] M.R. Bremner, S. Madariaga, Special identities for the pre-Jordan product in the free dendriform
algebra, Linear Algebra Appl. 439 (2) (2013) 435–454.
86 M. Bremner, S. Madariaga / Journal of Algebra 412 (2014) 51–86

[6] M.R. Bremner, L.A. Peresi, An application of lattice basis reduction to polynomial identities for
algebraic structures, Linear Algebra Appl. 430 (2009) 642–659.
[7] M.R. Bremner, L.A. Peresi, Polynomial identities for the ternary cyclic sum, Linear Multilinear
Algebra 57 (6) (2009) 595–608.
[8] M.R. Bremner, L.A. Peresi, Nonhomogeneous subalgebras of Lie and special Jordan superalgebras,
J. Algebra 322 (6) (2009) 2000–2026.
[9] M.R. Bremner, L.A. Peresi, Special identities for quasi-Jordan algebras, Comm. Algebra 39 (2011)
2313–2337.
[10] R. Carlsson, n-ary algebras, Nagoya Math. J. 78 (1980) 45–56.
[11] J. Clifton, A simplification of the computation of the natural representation of the symmetric group
Sn , Proc. Amer. Math. Soc. 83 (2) (1981) 248–250.
[12] P.M. Cohn, On homomorphic images of special Jordan algebras, Canad. J. Math. 6 (1954) 253–264.
[13] P.M. Cohn, Two embedding theorems for Jordan algebras, Proc. Lond. Math. Soc. (3) 9 (1959)
503–524.
[14] H.A. Elgendy, Universal associative envelopes of nonassociative triple systems, Comm. Algebra
42 (4) (2014) 1785–1810.
[15] H.A. Elgendy, The universal associative envelope of the anti-Jordan triple system of n × n matrices,
J. Algebra 383 (2013) 1–28.
[16] J.R. Faulkner, J.C. Ferrar, Simple anti-Jordan pairs, Comm. Algebra 8 (11) (1980) 993–1013.
[17] C.M. Glennie, Identities in Jordan algebras, in: Computational Problems in Abstract Algebra, Proc.
Conf., Oxford, 1967, Pergamon, Oxford, United Kingdom, 1970, pp. 307–313.
[18] C.M. Glennie, Some identities valid in special Jordan algebras but not valid in all Jordan algebras,
Pacific J. Math. 16 (1966) 47–59.
[19] N. Jacobson, Structure and Representations of Jordan Algebras, American Mathematical Society,
Providence, USA, 1968.
[20] O. Loos, Lectures on Jordan Triples, University of British Columbia, Vancouver, Canada, 1971.
[21] O. Loos, Representations of Jordan triples, Trans. Amer. Math. Soc. 185 (1973) 199–211.
[22] O. Loos, K. McCrimmon, Speciality of Jordan triple systems, Comm. Algebra 5 (10) (1977)
1057–1082.
[23] K. McCrimmon, Jordan triple systems: insights and ignorance, in: Proceedings of the International
Conference on Algebra, Part 2, Novosibirsk, 1989, in: Contemp. Math., vol. 131, 1992, pp. 625–637.
[24] K. McCrimmon, The role of identities in Jordan algebras, Resenhas 6 (2–3) (2004) 265–280.
[25] K. McCrimmon, A Taste of Jordan Algebras, Springer-Verlag, New York, USA, 2004.
[26] K. McCrimmon, E. Zelmanov, The structure of strongly prime quadratic Jordan algebras, Adv.
Math. 69 (2) (1988) 133–222.
[27] K. Meyberg, Lectures on Algebras and Triple Systems, The University of Virginia, Charlottesville,
USA, 1972; Available online: http://www.math.uci.edu/~brusso/Meyberg(Reduced2).pdf.
[28] E. Neher, Jordan Triple Systems by the Grid Approach, Lecture Notes in Math., vol. 1280, Springer-
Verlag, Berlin, Germany, 1987.
[29] E. Zelmanov, Prime Jordan algebras. II, Sibirsk. Mat. Zh. 24 (1) (1983) 89–104.