Evolution algebras, automorphisms, and graphs
Alberto Elduque*⋆*
Departamento de Matemáticas e Instituto Universitario de Matemáticas y Aplicaciones,
Universidad de Zaragoza, 50009 Zaragoza, Spain
[email protected]
and
Alicia Labra*⋆⋆*
Departamento de Matemáticas,
Facultad de Ciencias, Universidad de Chile. Casilla 653, Santiago, Chile
[email protected]
Abstract.
The affine group scheme of automorphisms of an evolution algebra E with E2=E is shown to lie
in an exact sequence 1→D→Aut(E)→S, where D, diagonalizable,
and S, constant, depend solely on the directed graph associated to E.
As a consequence, the Lie algebra of derivations Der(E) (with E2=E) is shown to be trivial if the characteristic of the ground field
is [math] or 2, and to be abelian, with a precise description, otherwise.
⋆ Supported by grants MTM2017-83506-C2-1-P (AEI/FEDER, UE) and E22_17R (Diputación
General de Aragón).
Part of this research was done while this author was visiting the Departamento de Matemáticas, Facultad de Ciencias, Universidad de
Chile, supported by FONDECYT grant 1170547.
⋆⋆ Supported by FONDECYT 1170547.
1. Introduction
Evolution algebras were introduced in 2006 by Tian and Vojtechovsky (see [10]) and present many connections with other fields like graph theory, group theory or Markov chains,
to mention a few (see Tian’s monograph [9]). They have received considerable attention in the last years (see [1] and the references therein).
In this paper, all the algebras considered will be defined over a ground field F, of arbitrary characteristic, and their dimension will be finite. An algebra is just a vector space A endowed with a bilinear map (the multiplication) A×A→A,(x,y)↦xy.
Definition 1.1**.**
An evolution algebra is an algebra E endowed with a basis B={v1,v2,…,vn}, called a natural basis, such that vivj=0 for any 1≤i=j≤n.
Given any evolution algebra E with natural basis B={v1,v2,…,vn}, and multiplication determined by
vi2=∑i=1nαijvj (αij∈F), an associated (directed) graph Γ=Γ(E,B) is defined in
[6]. The set of vertices V of Γ is just the natural basis, and the set of edges E⊆V×V consists of those pairs (vi,vj)
with αij=0, that is, (vi,vj)∈E if vj appears in vi2 with nonzero coefficient.
The graph Γ=Γ(E,B) is used in [6, 7] to get new results on these algebras and to provide new natural proofs of some known results.
In particular, it is shown in [6] that the group of automorphisms Aut(E) is finite if E2=E (or equivalently the matrix
(αij) is regular). Over an infinite field F, the regular matrices form a Zariski open, and hence dense, set in Matn(F). So, in a way, we have that
Aut(E) is finite for “almost all” evolution algebras.
Over fields of positive characteristic, or over nonalgebraically closed fields of characteristic [math], the affine group scheme of automorphisms Aut(E) contains much more information than Aut(E) including, in particular, the information on the Lie algebra of derivations Der(E).
Here we follow the functorial approach to affine group schemes (see for instance [11]). An affine group scheme is a representable group-valued functor defined on the category AlgF of unital commutative, associative algebras. Thus, given an evolution algebra E, Aut(E) is the functor AlgF⟶Grp that takes any object R in AlgF to the group Aut(ER) of automorphisms, as an R-algebra, of ER:=E⊗FR. The action on morphisms is the natural one.
The Lie algebra Lie(Aut(E)) is canonically isomorphic to the Lie algebra of derivations Der(E)={δ∈EndF(E)∣δ(xy)=δ(x)y+xδ(y) for any x,y∈E} (see [5, Example A.43]).
Now, the fact that Aut(E) is finite if
E2=E [6, Theorem 4.3] shows, in particular, that Aut(EFalg) is finite, where Falg is an algebraic closure of F, and hence the affine group scheme Aut(E) is finite, that is, the Hopf algebra that represents it is finite dimensional.
If the characteristic of the ground field F is [math], then any finite affine group scheme is étale, and hence the Lie algebra is trivial. Therefore
[6, Theorem 4.8] implies Der(E)=0 if E2=E and char(F)=0.
(This result over C has been proven in [2, Theorem 2.6] in a different way).
However, as some examples in [3] show, this is no longer true if char(F)>0.
The goal of the present paper is to show that given any evolution algebra E with E2=E, there is an exact sequence (7)
[TABLE]
where Aut(Γ) is the constant group scheme attached to the group of automorphism of the graph associated to E in [6], while
Diag(Γ) is a finite diagonalizable group scheme defined in terms solely of Γ. That is the elements in the exact sequence, except
Aut(E) itself, depend only on Γ (!!).
An affine group scheme is diagonalizable if it is a subscheme of a torus [11, §2.2] or, equivalently, if the representing Hopf algebra is the gruop
algebra of a finitely generated abelian group. In our situation, Diag(Γ) turns out to be a product of schemes of roots of unity μN (N∈N),
where μN(R)={r∈R∣rN=1} for any R in AlgF, which is represented by the quotient F[x]/(xN−1), that is, the group algebra of the cyclic group of order N.
On the other hand, given a finite group G, the associated constant group scheme G is the group scheme represented by FG:=Maps(G,F)=⨁g∈GFϵg, where
[TABLE]
(see [11, §2.4]). For any R in AlgF without proper idempotents, G(R) is (isomorphic to) the group G.
Note that FG≃F×F×⋯×F is the cartesian product of ∣G∣ copies of F.
In particular, FG is a separable algebra and hence G is étale.
The paper is structured as follows. Section 2 will be devoted to define and study the diagonalizable affine group scheme Diag(Γ) associated to any graph. For connected Γ, Diag(Γ) is either trivial or isomorphic to μN for some natural number N, given by the so called balance of Γ. Section 3 will deal with the group of automorphisms of a graph. Its main result: Theorem 3.2, gives the exact sequence
(7) mentioned above. This exact sequence induces a short exact sequence (8) which does not split in general.
Finally Section
4 is devoted to describe the Lie algebra of derivations of any evolution algebra E with E2=E. The description is a direct consequence of our results on the affine group scheme Aut(E). It turns out that Der(E) depends only on the graph.
2. The diagonal group of a graph
All the graphs considered in this paper are directed graphs. These are pairs Γ=(V,E), consisting of a finite set of vertices V and a set of edges (or arrows)
E⊆V×V.
Given such a graph, we need some definitions
A path is a sequence γ=(v0,e1,v1,…,vn−1,en,vn) where n≥0, v0,…,vn∈V, e1,…,en∈E, and for each i=1,…,n, either ei=(vi−1,vi) or ei=(vi,vi−1).
We define the balance of the path γ as the integer
[TABLE]
that is, b(γ) is obtained by adding 1 if the edge ei goes in the “right” direction (from v0 to vn) and −1 if the edge ei goes in the “wrong” direction, and summing over i.
The balance of Γ is defined as the greatest common divisor of the absolute values of the balances of the cycles in Γ:
[TABLE]
A cycle is a path γ=(v0,e1,v1,…,vn−1,en,vn) with v0=vn.
The indegree of a vertex v is the natural number (or [math])
[TABLE]
while the outdegree is
[TABLE]
The vertex v is said to be a source if deg−(v)=0, and a sink if deg+(v)=0.
Γ is said to be connected if the underlying undirected graph is connected, that is, if for every v,w∈V there exists a path
[TABLE]
with v0=v and vn=w. Any graph Γ is the “disjoint union” of its
connected components
Definition 2.1**.**
The diagonal group of a graph Γ=(V,E) is the (diagonalizable) affine group scheme Diag(Γ) given by
[TABLE]
with the natural morphisms.
Note that Diag(Γ) is a subgroup scheme of the torus (Gm)∣V∣
Let us see a few examples.
Example 2.2**.**
[TABLE]
then
[TABLE]
Example 2.3**.**
[TABLE]
Γ has no sinks.
If φ∈Diag(Γ)(R) and φ(a)=μ (∈R×), then φ(b)=μ2, φ(c)=μ4, and φ(b)=φ(c)2, that is μ2=μ8, so μ6=1. Hence Diag(Γ)≃μ6.
Example 2.4**.**
[TABLE]
Γ has no sources.
Again, if φ∈Diag(Γ)(R) and φ(c)=μ, then φ(b)=μ2, φ(a)=μ4, and φ(c)=φ(b)2, that is, μ=μ4, so μ3=1. Hence Diag(Γ)≃μ3.
From the definitions, we get at once the next result:
Proposition 2.5**.**
Let Γ=(V,E) be a graph with connected components Γi=(Vi,Ei), i=1,…,n (so that V=V1∪˙⋯∪˙Vn). Then
[TABLE]
If m=2s+1 is an odd natural number the square map
[TABLE]
is a group automorphism for any R in AlgF, with inverse r⟶r21:=rs+1.
Therefore, expressions like r2−3 make sense: r2−3=((r21)21)21.
Lemma 2.6**.**
Let Γ=(V,E) be a graph, γ=(v0,e1,v1,…,vn−1,en,vn) be a path in Γ. Let φ∈Diag(Γ)(R) for R in AlgF,
such that φ(vi)∈μmi(R) with mi odd for any i=0,…,n. Then φ(vn)=φ(v0)2b(γ).
Proof.
Imagine that γ=(v0,e1,v1,e2,v2,e3,v3) with e1=(v1,v0), e2=(v1,v2), and e3=(v3,v2), so b(γ)=−1.
[TABLE]
Then
As e1=(v1,v0)∈E, φ(v0)=φ(v1)2, so φ(v1)=φ(v0)21=φ(v0)2−1.
As e2=(v1,v2)∈E, φ(v2)=φ(v1)2, so φ(v2)=(φ(v0)2−1)2=φ(v0).
As e3=(v3,v2)∈E, φ(v2)=φ(v3)2, so φ(v3)=φ(v2)−1=φ(v1)2−1=φ(v0)2b(γ).
The general argument follows the same lines.
∎
Our next result determines the diagonal group of connected graphs without sources. Note that the graphs attached to evolution algebras E
with E2=E have no sources.
Theorem 2.7**.**
Let Γ=(V,E) be a connected graph with no sources. Then Diag(Γ)≃μN where N=2b(Γ)−1.
Proof.
First, the arguments in the proof of [6, Theorem 4.8] show that for any R in AlgF, any φ∈Diag(Γ)(R), and any vector v∈V, φ(v)∈μ2s−1(R) for some natural number s.
Fix a vertex a∈V, and consider the restriction homomorphism
[TABLE]
We will follow several steps:
Φa is one-to-one.
Actually, for R in AlgF, and φ∈Diag(Γ)(R), with φ(a)=1, by connectedness for any vertex v∈V there is a path
γ=(v0,e1,v1,…,vn−1,en,vn) with v0=a and vn=v. By Lemma 2.6, φ(v)=φ(a)2b(γ)=12b(γ)=1.
For any R in AlgF, and φ∈Diag(Γ)(R), φ(a)∈μN(R).
Indeed, by the previous argument, for any v∈V, φ(v)=φ(a)2b(γ) for any path γ connecting a and v. As the order of φ(a) is odd, φ(a) and φ(v) generate the same subgroup of R×.
In particular φ(a) and φ(v) have the same order.
Given any cycle γ=(v0,e1,v1,…,vn−1,en,vn) in Γ (vn=v0), we get φ(v0)=φ(v0)2b(γ), or
φ(v0)2b(γ)−1=1. Thus the order of φ(a) divides 2∣b(γ)∣−1 for any cycle γ. Using that 2gcd(m1,m2)−1=gcd(2m1−1,2m2−1), our result follows.
The image of Φa is exactly μN.
For any R in AlgF, and any μ∈μN(R), define φ:V⟶R× as follows: For any v∈V, select a path connecting a and v:
γ=(v0,e1,v1,…,vn−1,en,vn) with v0=a and vn=v, and define φ(v)=μ2b(γ). This is well defined, because for any other path
γ^=(v^0,e^1,v^1,…,v^n−1,e^n,v^n)
connecting a=v^0 and v=v^n, then
[TABLE]
is a cycle with balance b(γγ^−1)=b(γ)−b(γ^) and, therefore,
μ=μ2b(γ)−b(γ^). Hence
[TABLE]
Finally, φ∈Diag(Γ)(R), because for any e=(v,w)∈E, if γ=(v0,e1,v1,…,vn−1,en,vn) is a path connecting a=v0 and v=vn, then
γ′=(v0,e1,v1,…,vn−1,en,vn,e,w) is a path connecting a and w with b(γ′)=b(γ)+1. Hence,
φ(w)=μ2b(γ′)=(μ2b(γ))2=φ(v)2.
∎
Corollary 2.8**.**
Let Γ=(V,E) be a connected graph with no sources and with a loop e=(v,v). Then Diag(Γ)=1.
Let E be an evolution algebra with natural basis B={v1,…,vn} and let Γ=Γ(E,B)=(V,E) be the attached graph (V=B). For any
R in AlgF, and any φ∈Diag(Γ)(R), φ induces the linear (diagonal) isomorphism
[TABLE]
Let vi2=∑j=1nαijvj for i=1,…,n, with αij∈F, then
[TABLE]
and φ^(vi)2=φ(vi)2∑j=1nαijvj.
But if αij=0, then (vi,vj)∈E, so φ(vj)=φ(vi)2. Hence φ^∈Aut(ER) and we
obtain the following result:
Theorem 2.9**.**
Let E be an evolution algebra with natural basis B and let Γ=Γ(E,B) be the attached graph. Then there is an injective homomorphism
ι:Diag(Γ)⟶Aut(E) such that for any R in AlgF, and any R-point φ∈Diag(Γ)(R), ι(φ)=φ^
(as in (1)).
3. Graph Automorphisms
The goal of this section is, given an evolution algebra E with E2=E with attached graph Γ(E,B) (which is independent, up to isomorphism, of the natural
basis B chosen [6, Corollary 4.7]), to show the existence of a natural homomorphism
[TABLE]
where Aut(Γ) is the constant group scheme attached to the group of automorphisms of Γ, denoted by Aut(Γ). If B={v1,…,vn} is a natural basis we may identify Aut(Γ) with a subgroup of the symmetric group Sn of degree n:
[TABLE]
If we just look at the rational points in Aut(E)=Aut(E)(F), any φ∈Aut(E) has an attached permutation σ∈Aut(Γ) such that
φ(vi)∈F×vσ(i) for any i=1,…,n ([6, Theorem 4.4]). Thus the coordinate matrix of φ relative to B is a monomial matrix (i.e., it
has exactly one nonzero entry in each row and column).
In order to deal with the group scheme Aut(E), some extra care must be taken. Let R be in AlgF, and let φ∈Aut(E)(R)=Aut(ER),
with φ(vi)=∑j=inrijvj for any i=1,…,n. Then r=\det\bigl{(}r_{ij}\bigr{)}\in R^{\times}:
[TABLE]
For any i=j we have 0=φ(vivj)=φ(vi)φ(vj)=∑k=1nrikrjkvk2.
Because E2=E, {v12,…,vn2} form a basis of E and hence
[TABLE]
Therefore, for any σ=τ in Sn,
(rσ(1)1⋯rσ(n)n)(rτ(1)1⋯rτ(n)n)=0.
For any σ∈Sn, consider the element
[TABLE]
Then 1=∑σ∈Sneσφ, and eσφeτφ=0 for σ=τ in Sn. Therefore, the eσφ’s are orthogonal idempotent elements, and
R=⨁σ∈SnReσφ. Moreover, (3) implies
[TABLE]
and the coordinate matrix \bigl{(}r_{ij}\bigr{)} of φ splits into a sum of monomial matrices over the orthogonal ideals Reσφ. Thus, for instance, with n=3 we have:
[TABLE]
and A=\bigl{(}r_{ij}\bigr{)}=\sum_{\sigma\in S_{3}}A_{\sigma}, with Aσ=eσφA∈Mat3(Reσφ) a monomial matrix thanks to (4):
[TABLE]
Moreover, if σ∈Sn and eσφ=0, then the monomial matrix
[TABLE]
where Eij denotes the matrix with 1 in the (ij) slot and [math]’s elsewhere, correspond to an automorphism of EReσφ. This forces
σ∈Aut(Γ). Therefore,
[TABLE]
Recall that the coordinate Hopf algebra of the constant group scheme Aut(Γ) is \mathbb{F}^{\operatorname{\mathrm{Aut}}(\Gamma)}=\mathrm{Maps}\bigl{(}\operatorname{\mathrm{Aut}}(\Gamma),\mathbb{F}\bigr{)}, which has a natural basis {ϵσ∣σ∈Aut(Γ)}, with
[TABLE]
Then Aut(Γ)(R) is identified with HomAlgF(FAut(Γ),R).
We are ready to define the homomorphism ρ in (2). For R in AlgF and φ∈Aut(E)(R)=Aut(ER), the image of φ under ρ is defined as the element ρ(φ)∈HomAlgF(FAut(Γ),R) given by
[TABLE]
It is trivially checked that this gives a homomorphism ρ:Aut(E)→Aut(Γ).
Remark 3.1*.*
Exactly as over F, if R in AlgF has no proper idempotents, then 1=eσφ for a unique σ∈Aut(Γ) and the matrix of φ is a monomial matrix attached to σ. In this case Aut(Γ)(R)≃Aut(Γ) and ρ(φ) is just σ under this identification.
The main result of this section is the following:
Theorem 3.2**.**
Let E be an evolution algebra with E2=E and natural basis B={v1,…,vn}. Let Γ=Γ(E,B) be its associated graph. Then the sequence
[TABLE]
is exact.
Proof.
ker(ρ)(R) consists of the automorphisms φ∈Aut(E)(R)=Aut(ER) such that eσφ=0 for any 1=σ∈Aut(Γ). Hence 1=e1φ and φ is diagonal, that is, the elements of B are eigenvectors for φ. These automorphisms are precisely the elements in the image of ι.
∎
Example 3.3**.**
The homomorphism ρ is not surjective in general. Take, for instance, the evolution algebra E=Fv1⊕Fv2, with natural basis B={v1,v2}, and multiplication given by v12=v1+αv2, v22=βv1+v2, with 0=α,β∈F, α=β, αβ=1. Then the associated graph Γ(E,B) is the complete graph
[TABLE]
While Aut(Γ)=C2, let us check that Aut(E)=1. To do that, it is enough to prove that Aut(ER)=1 for R in AlgF without proper idempotents.
The arguments above show that the coordinate matrix relative to B={v1,v2} of any φ∈Aut(ER) is either
[TABLE]
with r1,r2∈R×.
In the first case φ(v12)=r1v1+αr2v2, while φ(v1)2=r12(v1+αv2), so r12=r1=r2, and hence, due to the absence of
proper idempotents, φ=id.
In the second case φ(v12)=r1v2+αr2v1, while φ(v1)2=r12v22=r12(βv1+v2), so r12=r1 and αr2=βr12. Hence, r1=1, r2=βα−1=1. But φ(v22)=φ(v2)2 forces r2=1, a contradiction.
Any subgroup scheme of a constant group scheme is itself a constant group scheme. Hence we have the next consequence:
Corollary 3.4**.**
Let E be an evolution algebra with E2=E and natural basis B={v1,…,vn}. Let Γ=Γ(E,B) be its associated graph. Then there is a subgroup
H of Aut(Γ) and a short exact sequence
[TABLE]
where H is the constant group scheme associated to H.
Example 3.5**.**
The short exact sequence in Corollary 3.4 does not split in general. Take, for instance the evolution algebra E=Fv1⊕Fv2 with
v12=v2, v22=αv1, with 0=α∈F.
The associated graph is
[TABLE]
Then Diag(Γ)=μ3 (Theorem 2.7) and ρ:Aut(E)⟶Aut(Γ)≃C2 is surjective, as it is so over an algebraic closure Falg. Indeed, over Falg the assignment
[TABLE]
gives an automorphism φ with ρ(φ) being the generator of Aut(Γ). Moreover, φ2=id and this proves that (8) splits over Falg.
Let us check that the short exact sequence
[TABLE]
splits if and only if there is μ∈F such that α=μ3.
Actually, if α=μ3 the assignment (9) makes sense over F, so the sequence splits.
Conversely, if (10) splits, there is an automorphism φ∈Aut(E) with φ2=id, such that φ(v1)∈F×v2, φ(v2)∈F×v1.
With φ(v1)=νv2, φ(v2)=μv1, we get ν=μ−1, as φ2=id, and
[TABLE]
so that α=μ3. ∎
4. Derivations
The results of the previous sections allow us to compute easily the Lie algebra of derivations of any evolution algebra
E, with E2=E. This Lie algebra depends only on the associated graph!
Theorem 4.1**.**
Let E be an evolution algebra with E2=E. Let B be a natural basis and let Γ=Γ(E,B) be the attached graph. Then:
- (1)
If the characteristic of F is [math] or 2, then Der(E)=0.
2. (2)
If the characteristic of F is p=0,2, then Der(E) is an abelian Lie algebra whose dimension is the number of connected components
Γi of Γ such that the order of 2 in Z/pZ divides the balance b(Γi).
Proof.
The exact sequence (7) induces an exact sequence (see eq. [Milne, 10d]):
[TABLE]
But Lie(Aut(Γ))=0, as Aut(Γ) is a constant group scheme, and hence étale. On the other hand, Lie(Aut(E))=Der(E)
(see [5, Example A.43]), so that Der(E) is isomorphic to Lie(Diag(Γ)) through the differential of ι.
However, Lie(μm) is either [math] if char(F)∤m, or it has dimension 1 if char(F)∣m (see [5, Example A42]). Hence
Theorem 2.7 gives the results.
∎
Remark 4.2*.*
As mentioned in the Introduction, the fact that Der(E) is [math] for any evolution algebra E with E2=E over C has already been proved in
[2, Theorem 2.1].
Consider the algebra of dual numbers F[ϵ]=F1⊕Fϵ,
with ϵ2=0, and the natural homomorphism π:F[ϵ]⟶F in AlgF (π(1)=1, π(ϵ)=0).
Given a graph Γ=(V,E), Lie(Diag(Γ)) is the kernel of the induced group homomorphism π∗:Diag(Γ)(F[ϵ])⟶Diag(Γ)(F). The elements of kerπ∗ are the maps
[TABLE]
for a linear map δ:V⟶F, such that, for any (v,w)∈E, φ(w)=φ(v)2, which is equivalent to δ(w)=2δ(v).
Therefore we obtain the following straightforward consequence of Theorems 4.1, 2.7 and 2.9.
Corollary 4.3**.**
Let E be an evolution algebra with E2=E over a field F of characteristic p=0,2. Let B be a natural basis and let
Γ=Γ(E,B) be the associated graph. Let Γi=(Vi,Ei) (Vi⊆B), i=1,⋯,r, be the connected components
of Γ such that p∣2b(Γi)−1. For any i=1,…,r, fix an element vi∈Vi. Then a basis of Lie(Diag(Γ)) is given by
δ^1,⋯,δ^r, where
δ^i(v)=0* if v∈/Vi,*
δ^i(vi)=vi,
δ^i(w)=2b(γ)w* if w∈Vi and γ=(w0,e1,w1,…,en,wn) is a path connecting w0=vi and wn=w.*
Example 4.4**.**
The evolution algebra E in Example 3.3 has trivial group scheme of automorphisms, so Der(E)=0 for any ground field F.
However, for the evolution algebra E in Example 3.5, we have the short exact sequence in (10), and Diag(Γ)≃ μ3. Hence Der(E)=0 unless char(F)=3. In the later case, Der(E) is spanned by the map d:v1↦v1, v2↦2v2=−v2.
Remark 4.5*.*
It must be remarked that for α=1, the evolution algebra E in Example 3.5 is the two-dimensional split para-Hurwitz algebra, and hence, for arbitrary α (=0), E is a symmetric composition algebra (see [4] and references therein).
As shown in Example 3.5, the short exact sequence
[TABLE]
splits if and only if α∈F3, that is, if and only if E is, up to isomorphism, the split two-dimensional para-Hurwitz algebra.