ON THE CONJECTURE OF SHANG ABOUT FREE ALTERNATIVE ALGEBRAS

arXiv:2503.16074v1 [[Link]] 20 Mar 2025

VLADIMIR DOTSENKO

A BSTRACT. Kashuba and Mathieu [12] proposed a conjecture on vanishing

of some components of the homology of certain Lie algebras, implying a de-
scription of the GL d -module structure of the free d-generated Jordan algebra.
Their conjecture relies on a functorial version of the Tits–Kantor–Koecher con-
struction that builds Lie algebras out of Jordan algebras. Recently, Shang [17]
used a functorial construction of Allison, Benkart and Gao that builds Lie alge-
bras out of alternative algebras to propose another conjecture on vanishing of
some components of the homology of certain Lie algebras, implying a descrip-
tion of the GL d -module structure of the free d-generated alternative algebra.
In this note, we explain why the conjecture of Shang is not true.

I NTRODUCTION
An alternative algebra is a nonassociative algebra in which the associator
(a, b, c) = (ab)c − a(bc)
is a skew-symmetric function of its arguments a, b, c. The most famous instance
of octonions is given by octonions of Cayley [3]; a systematic theory of alterna-
tive algebras was developed by Zorn [22].
A lot of structural results on alternative algebras are known [21], yet free al-
ternative algebras remain a mysterious object. They have many unexpected
properties: for instance, free alternative algebras on sufficiently many genera-
tors contain central and nilpotent elements [18]. Recently, Shang [17], inspired
by a conjecture of Kashuba and Mathieu on free Jordan algebras [12], proposed
a conjecture that would shed light on free alternative algebras. Namely, he con-
jectured that the homology of Lie algebras obtained from free alternative alge-
bras by a construction of Allison, Benkart, and Gao [1, Th. 4.13], if viewed as an
sl3 -module, contains no trivial or adjoint isotypic components in homological
degrees greater than one. This conjecture would imply a description of the GL d -
module structure of the free d -generated alternative algebra for each d ∈ N.
We show in this note that the conjecture of Shang is not true. It is worth men-
tioning that, as indicated to the author by Olivier Mathieu when discussing the
results of this note, one heuristic reason that the conjecture of Shang is “too good
to be true” is that the Allison–Benkart–Gao Lie algebra has three different sl3 -
module types of nontrivial relations, so it is plausible that interaction of those
relations creates trivial and adjoint components in higher homological degrees.
In fact, we give three different obstructions to the conjecture. Two of them come
from the concrete situations where free alternative algebras are well understood:
1
2 VLADIMIR DOTSENKO

we show that conjecture fails for the free alternative algebra on three generators
(described explicitly by Iltyakov [9]) and for the free alternative superalgebra on
one odd generator (described explicitly by Shestakov and Zhukavets [19]). One
other obstruction is of more fundamental nature: it turns out that the conjecture
predicts a virtual non-effective S 10 -module structure on the component Alt(10)
of the alternative operad. It is worth noting that in parallel to this computa-
tion, we computed the prediction of the Kashuba–Mathieu conjecture for the
S n -module structure on the component Jord(n) of the Jordan operad, and that
prediction gives an effective module for n ≤ 32, which makes us cautiously opti-
mistic about the latter conjecture.
This paper is organized as follows. In Section 1, we recall the necessary defi-
nitions and the main conjecture. In Section 2, we discuss an operadic viewpoint
of the Allison–Benkart–Gao construction and of the conjecture of Shang and its
superalgebra version. In Section 3, we explain why the conjecture is false for
most free alternative algebras and superalgebras. Finally, in Appendix A, we give
the SageMath [20] code that computes the predictions of the conjecture that we
show to be problematic.

Acknowledgements. We thank Iryna Kashuba, Olivier Mathieu, and Ivan Shes-

takov for many useful discussions. Work of V.D. was supported by the French
national research agency (project ANR-20-CE40-0016) and by Institut Universi-
taire de France. Some of the computations presented here were done during the
visit of V.D. to the Shenzhen International Center for Mathematics; he thanks
this institution for hospitality and excellent working conditions.

1. T HE A LLISON –B ENKART–G AO FUNCTOR AND THE CONJECTURE OF S HANG

Throughout this paper, we use the following notation in any alternative alge-
bra A: f ·g = 12 ( f g +g f ), L g ( f ) = g · f , R g ( f ) = f ·g . Recall that for any alternative
algebra A and any a, b ∈ A, the operator D a,b ∈ End (A) defined by the formula

D a,b := [L a , L b ] + [R a , R b ] + [L a , R, b]

is a derivation of A. These operators satisfy the following identities [15, Sec. III.8]:

(1) D a,b + D b,a = 0,

(2) D ab,c + D bc,a + D ca,b = 0,
(3) [D, D a,b ] = D D(a),b + D a,D(b) for all D ∈ Der(A).

Derivations of this form are called inner; they arise as a particular case of the
general theory of inner derivations of nonassociative algebras [15, Sec. II.3]. The
last property shows that inner derivations Inner(A) form an ideal of the Lie alge-
bra Der(A) of all derivations of A. It is known [1, Sec. 2.35] that for an alternative
 ON THE CONJECTURE OF SHANG ABOUT FREE ALTERNATIVE ALGEBRAS 3

algebra A, the vector space sl3 ⊗ A ⊕Inner(A) has a Lie algebra structure given by
2 1 1
µ ¶
[x ⊗ a, y ⊗ b] = [x, y] ⊗ a · b + x y + y x − tr(x y)I 3 ⊗ [a, b] + tr(x y)D a,b ,
3 2 3
[D a,b , x ⊗ c] = x ⊗ D a,b (c),
[D a,b , D c,d ] = D D a,b (c),d + D c,D a,b (d) .

However, inner derivations are not functorial, and hence this construction does
not give a functor from the category of alternative algebras to the category of
Lie algebras. However, this is remedied if one passes to the universal central
extension of this latter algebra, described as follows [1, Th. 4.13]. Guided by
Equations (1) and (2), one associates to the given alternative algebra A the vector
space
B (A) := Λ2 (A)/(ab ∧ c + bc ∧ a + c a ∧ b : a, b, c ∈ A)
which one can think of as a functorial version of the space of Inner(A). Further-
more, one defines on the vector space

ABG(A) := sl3 ⊗ A ⊕ B (A)

a skew-symmetric bracket by
2 1 1
µ ¶
(4) [x ⊗ a, y ⊗ b] = [x, y] ⊗ a · b + x y + y x − tr(x y)I 3 ⊗ [a, b] + tr(x y)a ∧ b,
3 2 3
(5) [a ∧ b, x ⊗ c] = x ⊗ D a,b (c),
(6) [a ∧ b, c ∧ d ] = D a,b (c) ∧ d + c ∧ D a,b (d ).

One can show that these formulas make ABG(A) into a Lie algebra that one calls
the Allison–Benkart–Gao Lie algebra associated to A. Clearly, the construction
ABG(A) is functorial, and produces a Lie algebra in the category of sl3 -modules
having only trivial and adjoint components. Shang proves in [17, Th. 2.2] that
for each such Lie algebra g, the multiplicity of the adjoint component has an
alternative algebra structure, which he calls the Berman–Moody functor. Fur-
thermore, he shows in [17, Th. 2.3] that ABG is the left adjoint of the Berman–
Moody functor. Thus, this situation is parallel to that discussed in [12], where the
functorial Tits–Allison–Gao Lie algebra TAG(J ) built out of a Jordan algebra J is
studied; it is a Lie algebra in the category of sl2 -modules having only trivial and
adjoint components. The multiplicity of the adjoint module in such a Lie alge-
bra has a natural Jordan algebra structure; this defines the Tits functor from the
category of such Lie algebras to the category of Jordan algebras, and TAG is the
left adjoint of the Tits functor, so that in particular TAG(Jord(V )) is the free Lie
algebra the category of sl2 -modules having only trivial and adjoint components.
The conjecture proposed in [12] asserts that the homology H• (TAG(Jord(V )), k)
has no trivial or adjoint components in homological degrees greater than one; if
it is true, it would give character formulas for free Jordan algebras. Shang pro-
posed the following analogue of this conjecture for free alternative algebras.
4 VLADIMIR DOTSENKO

Conjecture 1.1 ([17, Conjecture 1]). Let Alt(V ) denote the free alternative algebra
generated by a finite-dimensional vector space V . The sl3 -module
Hk (ABG(Alt(V )), k)
has no trivial or adjoint component for k > 1.
It is shown in [17, Lemma 4.1] that there exist unique elements a(V ) and b(V )
in the augmentation ideal of the Grothendieck ring of GL(V ) for which we have
the following equalities in the Grothendieck ring of GL(V ) × P SL 3 :
[λ(a(V )[L(α1 + α2 )] + b(V )[L(0)]) : [L(0)]] = [k]
[λ(a(V )[L(α1 + α2 )] + b(V )[L(0)]) : [L(α1 + α2 )]] = [V ].
Here λ is the homological λ-operation satisfying λ(x + y) = λ(x)λ(y) and given
on each effective class x = [U ] by
(−1)k [Λk (U )].
X
λ(x) =
k≥0
The proof of [17, Lemma 4.1] gives an explicit recursive procedure for computing
a(V ) and b(V ); in Appendix A, we give the SageMath code implementing that
procedure.
According to [17, Th. 4.1] that Conjecture 1.1 implies the following conjec-
ture on character formulas for Alt(V ) and B (Alt(V )) in the Grothendieck ring of
GL(V ).
Conjecture 1.2 ([17, Conjecture 2]). Let Alt(V ) denote the free alternative algebra
generated by a finite-dimensional vector space V . In the Grothendieck ring of
GL(V ), we have
[Alt(V )] = a(V ), [B (Alt(V )) = b(V )].

2. O PERADS , ALGEBRAS , AND SUPERALGEBRAS

Let us offer some basic insight to the operad theory for a ring theorist reader,
referring to [6, Sec. 2.3] for further details and to the monograph [13] for system-
atic information on operads. Recall that, to a sequence {K (n)}n≥0 , where each
K (n) is a right module over the symmetric group S n , one can associate a functor
K : Vect → Vect given by the formula
K (n) ⊗kS n V ⊗n .
M
K(V ) :=
n≥0
Functors like that are called analytic functors [11]: their value on V is a “cate-
gorified Taylor series” with the “iterated derivatives” K (n) (one can view the ten-
soring over kS n as the categorical division by n!). An important class of analytic
functors come from free algebras of various kinds. For instance, if we denote
by Alt(n) the subspace of the free alternative algebra Alt(x 1 , . . . , x n ) consisting of
all elements of degree exactly one in each generator x 1 , . . . , x n , this vector space
has a natural right S n -action (by permutations of the generators x 1 , . . . , x n ), and
hence the collection of all these spaces
Alt := {Alt(n)}n≥1
 ON THE CONJECTURE OF SHANG ABOUT FREE ALTERNATIVE ALGEBRAS 5

gives rise to an analytic functor. It is easy to see that Alt(V ) naturally identified
with the free alternative algebra generated by V : indeed, we take all possible
multilinear operations that can be defined on alternative algebras, and “substi-
tute” elements of V by considering Alt(n) ⊗kS n V ⊗n . Furthermore, the natural
map
Alt(Alt(V )) → Alt(V )
which simply forgets the layered structure of alternative operations on the left,
considered together with the obvious inclusion V ,→ Alt(V ) makes the analytic
functor Alt into a monad; monads like that are called operads.
For our purposes, it will be important that analytic functors form a symmetric
monoidal category with respect to the so called Cauchy product; if F = {F (n)}n≥0
and G = {G(n)}n≥0 are two analytic functors, we may define a new analytic func-
tor F ⊗ G whose n-th component is given by
S
M
IndS nk ×S l (F (k) ⊗G(l )).
k+l =n

It is easy to check that this product is associative and admits symmetry isomor-
phisms F ⊗ G → G ⊗ F satisfying all necessary axioms of a symmetric monoidal
category. In particular, one can talk about Lie algebras in this category, which
historically are called twisted Lie algebras [2]. Concretely, a twisted Lie algebra g
may be viewed as a Lie algebra of the form
M
g= g(n)
n≥0

where each g(n) is a right kS n -module, and the Lie bracket maps g(n) ⊗ g(m) to
g(n + m) and is S n × S m -equivariant. Clearly, if g is a twisted Lie algebra and V is
a vector space, then we may view g as an analytic functor and obtain the vector
space g(V ); the twisted Lie algebra structure of g induces an honest Lie algebra
structure on g(V ).
The following result lifts the Allison–Benkart–Gao construction to the level
of twisted Lie algebras, and sheds new light on the functorial properties of that
construction. To state it, recall that every vector space U gives rise to the con-
stant Schur functor 1U with
(
U , n = 0,
1U (n) =
0, n > 0.

Moreover, giving a U a structure of an algebra over an operad O is the same as

giving the Schur functor 1U a structure of a left O -module [7, Sec. 3.2.9]. Recall
also that if R is a right O -module and L is a left O -module, we can form the
relative composition product R ◦O L, see [7, Sec. 5.1.5].

Proposition 2.1.
(1) There exists a twisted Lie algebra ABG such that we have a Lie algebra
isomorphism
ABG(Alt(V )) ∼
= ABG (V ).
6 VLADIMIR DOTSENKO

(2) The twisted Lie algebra ABG is a right module over the alternative op-
erad. Moreover, for every alternative algebra A, we have a natural iso-
morphism of left Lie-modules
1ABG(A) ∼
= ABG ◦Alt 1 A .

Proof. Let V be a finite-dimensional vector space. The free alternative algebra

Alt(V ) can be described as the value of the Schur functor of the alternative op-
erad on the vector space V [13]:
Alt(V ) ∼ Alt(n) ⊗kS n V ⊗n .
M
=
n≥1

Moreover, the functor

= Λ2 (Alt(V ))/(ab ∧ c + bc ∧ a + c a ∧ b : a, b, c ∈ Alt(V ))

B (Alt(V )) ∼
is also analytic. Indeed, we may take the obvious functorial version of the defi-
nition of B , looking, for each n, at the elements of B (Alt(x 1 , . . . , x n )) consisting of
all elements of degree exactly one in each generator x 1 , . . . , x n . Equivalently, one
may define
B := B (Alt) = Λ2 (Alt)/(ab ∧ c + bc ∧ a + c a ∧ b : a, b, c ∈ Alt),
where the exterior power is now taken with respect to the Cauchy product. We
can now define a twisted Lie algebra structure on the analytic functor
ABG = sl3 ⊗ Alt ⊕B (Alt)
by the same formulas (4)–(6). We conclude that
ABG(Alt(V )) = sl3 ⊗ Alt(V ) ⊕ B (Alt(V )) ∼
= (sl3 ⊗ Alt ⊕B (Alt))(V )
is an analytic functor and a twisted Lie algebra (one can even view it as the
Allison–Benkart–Gao construction applied to the twisted alternative algebra Alt),
and that ABG(Alt(V )) ∼ = ABG (V ).
Furthermore, the operad Alt is tautologically a right Alt-module, and B (Alt)
is a right Alt-module since B (Alt(V ))) is manifestly stable under all endomor-
phisms of Alt(V ), so the analytic functor ABG is a right Alt-module. Moreover, a
direct computation shows that the twisted Lie algebra structure on ABG com-
mutes with the right Alt-module action. It remains to notice that
Alt ◦Alt 1 A ∼
= 1A ,
B (Alt) ◦Alt 1 A ∼
=1 B (A) ,

where the first isomorphism is obvious, and the second follows from the fact
that B (Alt) is defined by the same formula as B (A), and the relative composition
product “computes” all the products in A after the evaluation of the analytic
functor B (Alt)(A) by taking the corresponding coequalizer. Since the twisted
Lie algebra structure on ABG commutes with the right Alt-module action, the
induced Lie algebra structure on the relative composite product matches the
Allison–Benkart–Gao Lie algebra structure on the vector space ABG(A). 
 ON THE CONJECTURE OF SHANG ABOUT FREE ALTERNATIVE ALGEBRAS 7

Let us also outline a functorial viewpoint on Lie algebra homology; for sim-
plicity, we shall focus on the homology with trivial coefficients. In classical text-
books, one would often find the either the definition of the homology as a de-
rived functor
(L)
H• (L, k) := TorU
• (k, k),
or a definition via the Chevalley–Eilenberg complex
H• (L, k) := H• (Λ(L), d ),
where in Λ(L) we place Λk (L) in the homological degree k, and the differential d
is given by the formula
(−1)i + j −1 [x i , x j ] ∧ x 1 ∧ · · · ∧ x̂ i · · · ∧ x̂ j ∧ · · · ∧ x k ,
X
d (x 1 ∧ · · · ∧ x k ) =
1≤i < j ≤k

where the notation x̂ i means that this factor must be omitted. While these def-
initions are completely unambiguous, they disguise one conceptual aspect of
the definition. The operad of Lie algebras is Koszul, and its Koszul dual is the
operad of commutative associative algebras. This allows us to interpret the ho-
mology of a Lie algebra L as the homology of its bar construction [13]. In our
particular case, the space of exterior forms Λ(L) is really a disguise of S c (sL),
the cofree conilpotent cocommutative coassociative coalgebra on sL (the vec-
tor space L homologically shifted by one using an odd element s), and d is the
unique coderivation of S c (sL) extending the map
S c (sL) ։ S 2 (sL) → sL
made of the projection onto S c (sL) ։ S 2 (sL) and the map S 2(sL) → sL corre-
sponding to the Lie bracket (a skew-symmetric bilinear map V × V → V is the
same as symmetric bilinear map sV × sV → sV ); note that, similarly how deriva-
tions of free algebras are determined by restriction to generators, coderivations
of cofree conilpotent coalgebras are determined by corestriction to cogenera-
tors. With that in mind, we obtain the following result.
Proposition 2.2. We have
H• (ABG(V ), k) ∼
= H• (ABG , k)(V ).
Proof. The underlying vector space of the Chevalley–Eilenberg complex of the
Lie algebra ABG(V ) is, as we already indicated,
(S c (s ABG(V )), d ),
= s Lie¡ (V ), where Lie¡ is the Koszul dual cooperad
and, if we note that S c (sV ) ∼
of the Lie operad [13, Sec. 7.2], the differential d comes from the Koszul twist-
ing cochain κ : Lie¡ → Lie. Thus, computing the differential (and its homology)
commutes with the evaluation of analytic functors on V . 
Corollary 2.3. Conjecture 1.1 holds for all choices of the vector space of generators
V if and only if the sl3 -module
Hk (ABG , k)
has no trivial or adjoint component for k > 1.
8 VLADIMIR DOTSENKO

Proof. Proposition 2.2 implies the “if” part of the statement. The “only if” part,
that is the assertion that vanishing of the trivial and the adjoint component
of Hk (ABG(V ), k) for all V implies the same for Hk (ABG , k), follows from the
fact that the “Taylor coefficients” of an analytic functor can be uniquely recon-
structed from its values using the Schur–Weyl duality [8]. 
Everything discussed above admits an analogue for free alternative superal-
gebras: indeed, one may generalize the Allison–Benkart–Gao construction to al-
ternative superalgebras and then simply replace the Lie algebra homology by the
Lie superalgebra homology in the statement of Conjecture 1.1. In fact, such gen-
eralizations cost nothing: over the years, the author of this note has been adver-
tising the obvious viewpoint that superalgebras over a given operad are merely
algebras over the same operad in a larger symmetric monoidal category [4], and
hence various results that can be stated and proved in terms of the correspond-
ing operad (e.g., the Poincaré–Birkhoff–Witt type theorems [5] or the Nielsen–
Schreier property [6]) are automatically true for superalgebras if already proved
for algebras. To give a yet another example of this approach, let us record the
following result.

Proposition 2.4. Suppose that Conjecture 1.1 is true for all free alternative alge-
bras. Then it is true for all free alternative superalgebras.

Proof. The free alternative superalgebra Alt(V0 | V1 ) generated by the superspace

V0 | V1 can also be written as a value of the same Schur functor:
Alt(V0 | V1 ) ∼
= Alt(n) ⊗kS n (V0 | V1 )⊗n ,
where the latter formula uses the symmetric monoidal structure on the category
of Z/2Z-graded vector spaces given by the Koszul sign rule
σ(u ⊗ v) = (−1)|u||v| v ⊗ u.
The Lie superalgebra homology also admits a description as the homology of
(S c (sL), d ), where S c (sL) now denotes the cofree conilpotent cocommutative
coassociative co-superalgebra generated by sL (which is nothing but the cofree
conilpotent cocommutative coassociative coalgebra in the symmetric monoidal
category of Z/2Z-graded vector spaces), and d is the unique coderivation of
S c (sL) extending the map
S c (sL) ։ S 2 (sL) → sL
made of the projection onto S c (sL) ։ S 2 (sL) and the map S 2(sL) → sL corre-
sponding to the Lie superalgebra structure. The obvious analogue of Proposi-
tion 2.2 holds, so
H• (ABG(Alt(V0 | V1 )), k) ∼
= H• (ABG , k)(V0 | V1 ),
and the claim follows from Corollary 2.3. 
Remark 2.5. The same line of reasoning applies to the Kashuba–Mathieu con-
jecture on free Jordan algebras [12]: the same argument as above shows that
 ON THE CONJECTURE OF SHANG ABOUT FREE ALTERNATIVE ALGEBRAS 9

validity of that conjecture for all Jordan algebras automatically implies the su-
peralgebra version of the conjecture proposed in [16].

3. C OUNTEREXAMPLES TO THE CONJECTURE

3.1. The conjecture fails for most free alternative superalgebras. Our precise
statement justifying the failure of the conjecture of Shang is the following result.
Theorem 3.1. Conjecture 1.1 fails for the free alternative superalgebra
Alt(x 1 , . . . , x m | y 1 , . . . , y n )
whenever m > 2 or n > 0.
Proof. The free alternative superalgebra Alt(x 1 , . . . , x m | y 1 , . . . , y n ) is Nm+n -graded,
and this grading induces a grading on the Lie superalgebra
ABG(Alt(x 1 , . . . , x m | y 1 , . . . , y n )),
as well as on its homology. Considering the graded components supported on
N J ⊂ Nm+n for various subsets J of the set of variables, we see that our assertion
would follow if we prove it for (m | n) = (3 | 0) and for (m | n) = (0 | 1). The first of
these results is proved in Proposition 3.4 below, and the second in Proposition
3.7 below. 
Remark 3.2. Conjecture 1.1 is true for the free alternative algebra Alt(x) [17,
Th. 5.1], so our result prompts a natural question if it is true for the only remain-
ing free alternative superalgebra: the free alternative algebra on two generators.
According to a theorem of Artin [22], this algebra coincides with free associative
algebra on two generators. We checked that the prediction of Conjecture 1.2 for
this algebra in degrees up to 15, so one may be cautiously optimistic.
To explain the proof in the case (m | n) = (3 | 0), let us briefly recall the results
of Iltyakov [9] who described a basis of Alt(a, b, c). For brevity, we denote below
L f ,g (h) = R g R f − R f ·g .
Theorem 3.3 ([9]). Consider the following elements in Alt(a, b, c):
n n n n n n n
W0 := {L a,b
1 2
L a,c 3
L c,a 4
L a,a 5
L b,b L c,c6 L a,[b,c]
7
(a, b, c) : n 1 , . . . , n 7 ≥ 0},
W1 := {[w, a], [w, b], [w, c] : w ∈ W0 },
W2 := {(w, a, b), (w, b, c), (w, a, c) : w ∈ W0 },
W := W0 ⊔ W1 ⊔ W2 ,
W′ := {w · (a, b, c) : w ∈ W0 },
n n n
B := {R c c R b b R a a w : n a , n b , n c ≥ 0, w ∈ W ⊔ W ′ }
Then we have a vector space isomorphism
Alt(a, b, c) ∼
= Ass(a, b, c) ⊕ kB.
This result allows us to disprove Conjecture 1.1 in the case (m | n) = (3 | 0).
Proposition 3.4. Conjecture 1.1 fails for the algebra Alt(a, b, c).
10 VLADIMIR DOTSENKO

Proof. The vector space isomorphism

Alt(a, b, c) ∼
= Ass(a, b, c) ⊕ kB
of Theorem 3.3 can be used to prove the following formula for the Hilbert series
of Alt(a, b, c):
1 t3
+ .
1 − 3t (1 − t )6 (1 − t 2 )3 (1 − t 3 )
Indeed, we first note that the vector space kW0 looks like the module with one
generator of weight 3 over the algebra of polynomials in six generators of weight
2 and one generator of weight 3, so its Hilbert series is
t3
.
(1 − t 2 )6 (1 − t 3 )
The vector space kW1 is the direct sum of three copies of kW0 with weight shifted
by one, so its Hilbert series is
3t 4
.
(1 − t 2 )6 (1 − t 3 )
The vector space kW2 is the direct sum of three copies of kW0 with weight shifted
by two, so its Hilbert series is
3t 5
.
(1 − t 2 )6 (1 − t 3 )
The vector space kW ′ is one copy of kW0 with weight shifted by three, so its
Hilbert series is
t6
.
(1 − t 2 )6 (1 − t 3 )
Therefore, the vector space kW ⊕ kW′ = kW0 ⊕ kW1 ⊕ kW2 ⊕ kW′ has the Hilbert
series
t 3 (1 + 3t + 3t 2 + t 3 ) t 3 (1 + t )3 t3
= = .
(1 − t 2 )6 (1 − t 3 ) (1 − t 2 )6 (1 − t 3 ) (1 − t )3 (1 − t 2 )3 (1 − t 3 )
Finally, the vector space kB looks like the module generated by the vector space
kW ⊕ kW′ over the algebra of polynomials in three variables of weight 1, so its
Hilbert series is
t3
.
(1 − t )6 (1 − t 2 )3 (1 − t 3 )
1
It remains to recall that the Hilbert series of Ass(a, b, c) is 1−3t . Expanding the
result to order 7, we obtain
1 + 3t + 9t 2 + 28t 3 + 87t 4 + 267t 5 + 804t 6 + 2388t 7 + O(t 8 ).
At the same time, computing the prediction of Conjecture 1.2 in SageMath [20]
gives us
1 + 3t + 9t 2 + 28t 3 + 87t 4 + 267t 5 + 804t 6 + 2373t 7 + O(t 8 ).
The resulting mismatch disproves Conjecture 1.2, and, due to [17, Th. 4.1], Con-
jecture 1.1 as well. 
 ON THE CONJECTURE OF SHANG ABOUT FREE ALTERNATIVE ALGEBRAS 11

Remark 3.5. In fact, using the albert program [10], we can get more precise in-
formation in the case of Alt(a, b, c). Using that program, we find that the dimen-
sion of the multihomogeneous component of multidegree (3, 3, 1) in Alt(a, b, c)
is equal to 154, as opposed to the prediction 152 of the conjecture of [17], and
that the dimension of the multihomogeneous component of multidegree (3, 2, 2)
in Alt(a, b, c) is equal to 236, as opposed to the prediction 233 of the conjecture
of [17]. Altogether these mismatches, multiplied by the cardinalities of S 3 -orbits
of the respective multidegrees, add up to 15, the total dimension mismatch.

To explain the proof in the case (m | n) = (0 | 1), let us briefly recall the results
of Shestakov and Zhukavets [19] who described a basis of Alt(∅ | x).

Theorem 3.6 ([19, Th. 4.2, Cor. 5.2]). The dimension d k of the weight k compo-
nent of Alt(∅ | x) is as follows:
1
d 1 = 1, d 2 = 1, d 3 = 2, d k = 2(k − 3) + (1 + (−1)k(k+1)/2 ) for k > 3.
2
More precisely, if we denote

t := x 2 ,
x (1) := x, x (k+1) := x (k) x − (−1)k xx (k) for k > 0,
z (k) := x (k) t − t x (k) ,
u (k) := x (k) x (3) + (−1)k x (3) x (k) ,

the following elements form a basis of Alt(∅ | x):

t m x σ, m ≥ 0, σ ∈ {0, 1}, m + σ ≥ 1,
m (k+2) σ
t (x x ), m ≥ 0, k > 0, σ ∈ {0, 1},
m (4k+ǫ) σ
t (u x ), m ≥ 0, k > 0, ǫ, σ ∈ {0, 1},
t m (z (4k+ǫ) x σ ), m ≥ 0, k > 0, ǫ, σ ∈ {0, 1}.

This result allows us to disprove Conjecture 1.1 in the case (m | n) = (0 | 1).

Proposition 3.7. Conjecture 1.1 fails for the superalgebra Alt(∅ | x).

Proof. Using the formula for the numbers d k from Theorem 3.6, we obtain the
sequence
1, 1, 2, 3, 4, 6, 9, 11, 12, 14, . . .
At the same time, computing the prediction of Conjecture 1.2 in SageMath [20]
gives us
1, 1, 2, 3, 4, 6, 9, 11, 12, 12, . . .
The resulting mismatch disproves Conjecture 1.2, and, due to [17, Th. 4.1], Con-
jecture 1.1 as well. 
12 VLADIMIR DOTSENKO

3.2. The conjecture and the Schur positivity property. One may view the cal-
culation of Proposition 3.7 as that of the multiplicity of the sign S 10 -module in
the component Alt(10) of the alternative operad. A yet another problem emerges
if one uses the recursive procedure of [17, Lemma 4.1] to compute multiplicities
of all irreducible S 10-modules in that component (because of the Schur–Weyl
duality, this essentially means working in the Grothendieck ring of GL(V ) for an
unspecified V ).
Proposition 3.8. Conjecture 1.2 predicts, for the S 10 -action on the component
Alt(10) of the alternative operad, a virtual module that is not effective (that is,
includes some irreducible S 10-modules with negative coefficients).
Proof. Computing the prediction of Conjecture 1.2 in SageMath [20] as a com-
bination of Schur functions (which correspond to characters of irreducible S n -
modules [14]), we get the following result:

12s 1,1,1,1,1,1,1,1,1,1 + 16s 2,1,1,1,1,1,1,1,1 + (−12)s 2,2,1,1,1,1,1,1 + (−32)s 2,2,2,1,1,1,1

+ (−27)s 2,2,2,2,1,1 + (−11)s 2,2,2,2,2 + (−2)s 3,1,1,1,1,1,1,1 + (−56)s 3,2,1,1,1,1,1
+ (−41)s 3,2,2,1,1,1 + 16s 3,2,2,2,1 + (−16)s 3,3,1,1,1,1 + 69s 3,3,2,1,1 + 67s 3,3,2,2
+ 68s 3,3,3,1 + 26s 4,1,1,1,1,1,1 + 95s 4,2,1,1,1,1 + 226s 4,2,2,1,1 + 145s 4,2,2,2
+ 251s 4,3,1,1,1 + 412s 4,3,2,1 + 124s 4,3,3 + 217s 4,4,1,1 + 179s 4,4,2 + 109s 5,1,1,1,1,1
+ 356s 5,2,1,1,1 + 415s 5,2,2,1 + 472s 5,3,1,1 + 361s 5,3,2 + 268s 5,4,1 + 42s 5,5
+ 151s 6,1,1,1,1 + 365s 6,2,1,1 + 218s 6,2,2 + 307s 6,3,1 + 90s 6,4 + 110s 7,1,1,1
+ 172s 7,2,1 + 75s 7,3 + 44s 8,1,1 + 35s 8,2 + 9s 9,1 + s 10 .
We note that some of the coefficients are negative: those of
s 2,2,1,1,1,1,1,1, s 2,2,2,1,1,1,1, s 2,2,2,2,1,1, s 2,2,2,2,2,
s 3,1,1,1,1,1,1,1, s 3,2,1,1,1,1,1, s 3,2,2,1,1,1, s 3,3,1,1,1,1.
As a consequence, we see that the predicted answer is a virtual character that is
not effective. 
Remark 3.9. The Schur function among the ones listed in the proof above that
corresponds to a partition with the smallest possible number of rows is s 2,2,2,2,2.
Therefore starting from Alt(a 1 , a 2 , a 3 , a 4 , a 5 ), there are certainly unexpected ho-
mology classes in the trivial and/or adjoint component that prevent the conjec-
tural answer even from being Schur-positive (let alone correct).

A PPENDIX A. T HE SageMath CODE

The following code computes the prediction of Conjecture 1.2 for the sym-
metric group actions on the components of the operad of alternative algebras.
It literally follows the recursive procedure described in [17, Lemma 4.1]. Note
that in the language of symmetric functions [14], the operation
(−1)k [Λk (U )]
X
λ(x) =
k≥0
 REFERENCES 13

used to define the elements a(V ) and b(V ) corresponds to the plethysm with the
element k≥0 (−1)k e k . To extract the trivial and adjoint multiplicities we use the
P

Weyl character formula for sl3 , see, e.g. [8].

R.<x,y> = LaurentPolynomialRing(QQ,2)
Sym = SymmetricFunctions(R)
s = Sym.s()
e = Sym.e()

def computealtlazy(dd):
[Link].display_length(dd)
L = LazySymmetricFunctions(e)
la = L(lambda n: (-1)^n*e[n])
a = [0 for t in range (1,dd+2)]
b = [0 for t in range (1,dd+2)]
echar = 1
a[1] = s[1]
ad = x^2*y+x/y+y/x+x*y^2+1/(x^2*y)+1/(x*y^2)+2
print(a[1],"\n")
for i in range (1,dd) :
echar=echar*la(a[i]*ad+b[i])
new = (1-x*y^-1)*(1-x*y^2)*(1-x^2*y)*echar
newp = [Link](i+1)
a[i+1] = newp.map_coefficients(lambda cf:cf[-2,-1])
b[i+1] = newp.map_coefficients(lambda cf:cf[0,0])
print(a[i+1],"\n")

R EFERENCES
[1] Bruce Allison, Georgia Benkart, and Yun Gao. “Central extensions of Lie al-
gebras graded by finite root systems”. In: Math. Ann. 316.3 (2000), pp. 499–
527.
[2] M. G. Barratt. “Twisted Lie algebras”. In: Geometric applications of homo-
topy theory (Proc. Conf., Evanston, Ill., 1977), II. Vol. 658. Lecture Notes in
Math. Springer, Berlin-New York, 1978, pp. 9–15.
[3] Arthur Cayley. “XXVIII. On Jacobi’s Elliptic functions, in reply to the Rev.
Brice Bronwin; and on Quaternions”. In: The London, Edinburgh, and Dublin
Philosophical Magazine and Journal of Science 26.172 (1845), pp. 208–211.
[4] V. Dotsenko. “Review of the article “Superalgebras and operads. I.” by S. N.
Tronin”. In: Mathematical Reviews 2555888 (2009).
[5] Vladimir Dotsenko and Pedro Tamaroff. “Endofunctors and Poincaré–Birk-
hoff–Witt theorems”. In: Int. Math. Res. Not. IMRN 16 (2021), pp. 12670–
12690.
[6] Vladimir Dotsenko and Ualbai Umirbaev. “An effective criterion for Nielsen–
Schreier varieties”. In: Int. Math. Res. Not. IMRN 23 (2023), pp. 20385–
20432.
14 REFERENCES

[7] Benoit Fresse. Modules over operads and functors. Vol. 1967. Lecture Notes
in Mathematics. Springer-Verlag, Berlin, 2009, pp. x+308.
[8] William Fulton and Joe Harris. Representation theory. A first course. Vol. 129.
Graduate Texts in Mathematics. Springer-Verlag, New York, 1991, pp. xvi+551.
[9] A. V. Iltyakov. “Free alternative algebras of rank 3”. In: Algebra i Logika 23.2
(1984), pp. 136–158, 240.
[10] David P. Jacobs. “The Albert Nonassociative Algebra System: A Progress
Report”. In: Proceedings of the International Symposium on Symbolic and
Algebraic Computation. ISSAC ’94. Oxford, United Kingdom: Association
for Computing Machinery, 1994, 41–44.
[11] André Joyal. “Une théorie combinatoire des séries formelles”. In: Adv. in
Math. 42.1 (1981), pp. 1–82.
[12] Iryna Kashuba and Olivier Mathieu. “On the free Jordan algebras”. In: Adv.
Math. 383 (2021), Paper No. 107690, 35.
[13] Jean-Louis Loday and Bruno Vallette. Algebraic operads. Vol. 346. Grund-
lehren der mathematischen Wissenschaften [Fundamental Principles of
Mathematical Sciences]. Springer, Heidelberg, 2012, pp. xxiv+634.
[14] I. G. Macdonald. Symmetric functions and Hall polynomials. Second. Ox-
ford Classic Texts in the Physical Sciences. The Clarendon Press, Oxford
University Press, New York, 2015, pp. xii+475.
[15] Richard D. Schafer. An introduction to nonassociative algebras. Pure and
Applied Mathematics, Vol. 22. Academic Press, New York-London, 1966,
pp. x+166.
[16] Shikui Shang. “The Z2 -graded dimensions of the free Jordan superalgebra
J (D 1 | D2)”. In: (2022). arXiv: 2211.09393 [[Link]].
[17] Shikui Shang. “Allison-Benkart-Gao functor and the cyclicity of free alter-
native functors”. In: (2025). arXiv: 2312.16369 [[Link]].
[18] I. P. Shestakov. “Free alternative algebras”. In: Mat. Zametki 25.5 (1979),
pp. 775–783, 800.
[19] Ivan Shestakov and Natalia Zhukavets. “The free alternative superalge-
bra on one odd generator”. In: Internat. J. Algebra Comput. 17.5-6 (2007),
pp. 1215–1247.
[20] The Sage Developers. SageMath, the Sage Mathematics Software System
(Version 9.5). [Link] 2022.
[21] K. A. Zhevlakov, A. M. Slin’ko, I. P. Shestakov, and A. I. Shirshov. Rings that
are nearly associative. Vol. 104. Pure and Applied Mathematics. Translated
from the Russian by Harry F. Smith. Academic Press, Inc. [Harcourt Brace
Jovanovich, Publishers], New York-London, 1982, pp. xi+371.
[22] Max Zorn. “Theorie der alternativen ringe”. In: Abhandlungen aus dem
Mathematischen Seminar der Universität Hamburg 8.1 (1931), pp. 123–
147.

I NSTITUT DE R ECHERCHE M ATHÉMATIQUE AVANCÉE , UMR 7501, U NIVERSITÉ DE S TRASBOURG

ET CNRS, 7 RUE R ENÉ -D ESCARTES , 67000 S TRASBOURG , F RANCE
Email address: vdotsenko@[Link]