betaFeedback

WorksMathematics

The algebraic classification of three-dimensional nilpotent left-symmetric dialgebras

Abstract

Left-symmetric dialgebras (LSDA) are to left-symmetric algebras what Leibniz algebras are to Lie algebras. We classify the three-dimensional nilpotent LSDA over the complex numbers. To do this we adapt the annihilator-extension (Skjelbred-Sund) method to dialgebras, using the annihilator with respect to both products on both sides, and we compute the second cohomology of all two-dimensional nilpotent LSDA. Up to isomorphism, besides the null and split algebras there are nine non-split algebras and families; the 2-step nilpotent families depend on four and three parameters. We give isomorphism criteria and separating invariants, identify the left-symmetric and diassociative algebras in the list, and complete the two-dimensional classification of an earlier paper.

Introduction

An algebra (A,⋅)(A,\cdot) is left-symmetric if the associator (x,y,z)=(xy)z−x(yz)(x,y,z)=(xy)z-x(yz) is symmetric in xx and yy. These algebras go back to Cayley’s work on rooted trees [5] and arose independently in Vinberg’s theory of homogeneous convex cones [24] and in the study of affine structures on Lie groups. They are also called pre-Lie, Vinberg or Koszul–Vinberg algebras. The commutator of a left-symmetric product is a Lie bracket, and left-symmetric algebras appear in operad theory [6], deformation theory, renormalisation and Rota–Baxter theory [16]. Burde’s survey [4] collects their main applications in geometry and physics. Classifications are known in low dimensions: Kim [14] described the complete left-symmetric structures on nilpotent Lie algebras of small dimension, and Bai [2] classified the three-dimensional left-symmetric algebras over C\mathbb{C}.

Loday introduced dialgebras, vector spaces with two products ⊣\dashv and ⊢\vdash, as the natural setting for Leibniz algebras [18, 19]: an associative dialgebra (diassociative algebra) gives a Leibniz algebra in the same way an associative algebra gives a Lie algebra. Diassociative algebras have since been studied structurally [10, 17], and the nilpotent ones were classified in dimension three [21] and four [22]. Felipe [8] defined left-symmetric dialgebras (LSDA) as the dialgebra analogue of left-symmetric algebras: they are to left-symmetric algebras what Leibniz algebras are to Lie algebras. When ⊣=⊢\dashv=\vdash the axioms reduce to the left-symmetric identity, and every diassociative algebra is an LSDA. The author began the structure theory of LSDA in [20], introducing annihilators and nilpotency and classifying the two-dimensional LSDA over a field of characteristic zero.

The standard tool for nilpotent algebras of small dimension is the method of annihilator (central) extensions of Skjelbred and Sund [23], first used for nilpotent Lie algebras [7, 9]. Every nilpotent algebra is an annihilator extension of a smaller one, and isomorphism classes of extensions correspond to orbits of the automorphism group on subspaces of a second cohomology space. Hegazi, Abdelwahab and Calderón Martín adapted it to Malcev and other non-associative algebras [11], and it has since been applied to many varieties (Leibniz [1], Jordan, Novikov, bicommutative, Zinbiel and others); see the survey by Kaygorodov [13]. Boza, Fedriani, Núñez and Tenorio [3] review the history of Lie algebra classifications.

For LSDA no classification beyond dimension two is known, and the two-dimensional list of [20 Theorem 2.6] needs to be completed (Remark 3.2). The cohomology needed for annihilator extensions has not been set up for LSDA either. A difficulty specific to dialgebras is the choice of annihilator: the extension method needs the subspace killed on both sides by both products, which is smaller than the ideal used in [20 Definition 2.3].

In this paper we classify all three-dimensional nilpotent left-symmetric dialgebras over C\mathbb{C} (Theorem 4.1), using annihilator extensions adapted to dialgebras. Besides the null and split algebras, there are nine non-split algebras and families, two of which depend on four and three parameters.

Preliminaries

All vector spaces are finite-dimensional over C\mathbb{C}. For subspaces UU, WW of an algebra and a product ∘\circ, U∘WU \circ W is the span of all u∘wu \circ w.

Definition 2.1 ([8, 20]). A left-symmetric dialgebra (LSDA) is a vector space SS with two bilinear products ⊣\dashv and ⊢\vdash such that, for all x,y,z∈Sx,y,z \in S,

x⊣(y⊣z)=x⊣(y⊢z)(1)x \dashv(y \dashv z) = x \dashv(y \vdash z) \tag*{(1)}
(x⊢y)⊢z=(x⊣y)⊢z(2)(x \vdash y) \vdash z = (x \dashv y) \vdash z \tag*{(2)}
x⊣(y⊣z)−(x⊣y)⊣z=y⊢(x⊣z)−(y⊢x)⊣z(3)x \dashv(y \dashv z) - (x \dashv y) \dashv z = y \vdash(x \dashv z) - (y \vdash x) \dashv z \tag*{(3)}
x⊢(y⊢z)−(x⊢y)⊢z=y⊢(x⊢z)−(y⊢x)⊢z.(4)x \vdash(y \vdash z) - (x \vdash y) \vdash z = y \vdash(x \vdash z) - (y \vdash x) \vdash z. \tag*{(4)}

We keep the numbering of [20]. A homomorphism of LSDA is a linear map preserving both products.

Remark 2.2. If ⊣=⊢\dashv=\vdash, then (1.1) and (1.2) are trivial and (1.3), (1.4) both become the left-symmetric identity, so an LSDA with equal products is a left-symmetric algebra. Every diassociative algebra [19] is an LSDA: its axioms make both sides of (1.3) and of (1.4) zero and give (1.1), (1.2) directly. Moreover, the commutator [x,y]=x⊣y−y⊢x[x,y]=x \dashv y-y \vdash x of an LSDA satisfies the (right) Leibniz identity [[x,y],z]=[[x,z],y]+[x,[y,z]][[x,y],z]=[[x,z],y]+[x,[y,z]] [8], [20, Theorem 1.13]; for diassociative algebras this is Loday’s functor Dias →\to Leib, and for ⊣=⊢\dashv=\vdash it is the commutator LSA →\to Lie. These relations form the extended Loday diagram of [20] (Figure 1), in which hooked arrows are inclusions of subcategories and the horizontal arrows on the right are the commutator functors.

Figure 1. The extended Loday diagram: associative algebras (As), diassociative algebras (Dias), left-symmetric algebras (LSA), left-symmetric dialgebras (LSDA), Lie and Leibniz (Leib) algebras.

Definition 2.3. Let SS be an LSDA.

  1. The annihilator of SS is Ann⁡(S)={x∈S:x⊣S=S⊣x=x⊢S=S⊢x=0}\operatorname{Ann}(S)=\{x\in S:x\dashv S=S\dashv x=x\vdash S=S\vdash x=0\}. It is an ideal of $S.

(2) Put S1=SS^{1}=S and Sk=∑i+j=k(Si⊣Sj+Si⊢Sj)S^{k}=\sum_{i+j=k}(S^{i}\dashv S^{j}+S^{i}\vdash S^{j}) for k≥2k\ge2. Then Sk+1⊆SkS^{k+1}\subseteq S^{k} for all k≥1k\ge1: this is clear for k=1k=1, and for k≥2k\ge2 every summand Si∘SjS^{i}\circ S^{j} (∘∈{⊣,⊢}\circ\in\{\dashv,\vdash\}) of Sk+1S^{k+1} has i≥2i\ge2 or j≥2j\ge2, so by induction it lies in Si−1∘SjS^{i-1}\circ S^{j} or Si∘Sj−1S^{i}\circ S^{j-1}, which are summands of SkS^{k}. The algebra SS is nilpotent if St=0S^{t}=0 for some tt; the least such tt is the nilpotency index.

(3) SS is split if S=S′⊕CeS=S'\oplus\mathbb{C}e, a direct sum of ideals with e∈Ann⁡(S)e\in\operatorname{Ann}(S), and non-split otherwise.

The annihilator in [20 Definition 2.3] is the intersection of the right ⊣\dashv-annihilator {x:S⊣x=0}\{x:S\dashv x=0\} and the left ⊢\vdash-annihilator {x:x⊢S=0}\{x:x\vdash S=0\}. It contains Ann⁡(S)\operatorname{Ann}(S) and may be strictly larger. Annihilator extensions need the subspace killed by both products on both sides, which is Ann⁡(S)\operatorname{Ann}(S) above.

Lemma 2.4. Every non-zero nilpotent LSDA SS has Ann⁡(S)≠0\operatorname{Ann}(S)\ne0. More precisely, if tt is the nilpotency index of SS, then 0≠St−1⊆Ann⁡(S)0\ne S^{t-1}\subseteq\operatorname{Ann}(S).

Proof. Since S≠0S\ne0 we have t≥2t\ge2, and St−1≠0S^{t-1}\ne0 by minimality of tt. The terms with (i,j)=(1,t−1)(i,j)=(1,t-1) and (t−1,1)(t-1,1) in the definition of StS^{t} give S⊣St−1S\dashv S^{t-1}, S⊢St−1S\vdash S^{t-1}, St−1⊣SS^{t-1}\dashv S, St−1⊢S⊆St=0S^{t-1}\vdash S\subseteq S^{t}=0. Hence St−1⊆Ann⁡(S)S^{t-1}\subseteq\operatorname{Ann}(S). □

Since Ann⁡(S)\operatorname{Ann}(S) is an ideal, S/Ann⁡(S)S/\operatorname{Ann}(S) is again a nilpotent LSDA, of smaller dimension by Lemma 2.4. This is the basis of the inductive method below.

Definition 2.5. Let AA be an LSDA and VV a vector space. A pair θ=(θ⊣,θ⊢)\theta=(\theta_{\dashv},\theta_{\vdash}) of bilinear maps A×A→VA\times A\to V is a 2-cocycle if, for all x,y,z∈Ax,y,z\in A,

θ⊣(x,y⊣z)=θ⊣(x,y⊢z),θ⊢(x⊢y,z)=θ⊢(x⊣y,z),θ⊣(x,y⊣z)−θ⊣(x⊣y,z)=θ⊢(y,x⊣z)−θ⊣(y⊢x,z),θ⊢(x,y⊢z)−θ⊢(x⊢y,z)=θ⊢(y,x⊢z)−θ⊢(y⊢x,z).\begin{aligned} \theta_{\dashv}(x,y\dashv z)&=\theta_{\dashv}(x,y\vdash z),& \theta_{\vdash}(x\vdash y,z)&=\theta_{\vdash}(x\dashv y,z),\\ \theta_{\dashv}(x,y\dashv z)-\theta_{\dashv}(x\dashv y,z)&=\theta_{\vdash}(y,x\dashv z)-\theta_{\dashv}(y\vdash x,z),\\ \theta_{\vdash}(x,y\vdash z)-\theta_{\vdash}(x\vdash y,z)&=\theta_{\vdash}(y,x\vdash z)-\theta_{\vdash}(y\vdash x,z). \end{aligned}

The cocycles form a space Z2(A,V)Z^{2}(A,V). For a linear map f:A→Vf:A\to V, the pair δf=(f(x⊣y),f(x⊢y))\delta f=(f(x\dashv y),f(x\vdash y)) is a coboundary; the coboundaries form B2(A,V)B^{2}(A,V), and H2(A,V)=Z2(A,V)/B2(A,V)H^{2}(A,V)=Z^{2}(A,V)/B^{2}(A,V). We write Ann⁡(θ)={x∈A:θ∘(x,A)=θ∘(A,x)=0 for ∘∈{⊣,⊢}}\operatorname{Ann}(\theta)=\{x\in A:\theta_{\circ}(x,A)=\theta_{\circ}(A,x)=0\text{ for }\circ\in\{\dashv,\vdash\}\}. For θ∈Z2(A,V)\theta\in Z^{2}(A,V) the annihilator extension Aθ=A⊕VA_{\theta}=A\oplus V has the products

(x+u)⊣(y+w)=x⊣Ay+θ⊣(x,y),(x+u)⊢(y+w)=x⊢Ay+θ⊢(x,y).(x+u)\dashv(y+w)=x\dashv_{A}y+\theta_{\dashv}(x,y),\qquad (x+u)\vdash(y+w)=x\vdash_{A}y+\theta_{\vdash}(x,y).

Lemma 2.6. Let θ\theta be a pair of bilinear maps A×A→VA\times A\to V. Then AθA_{\theta} is an LSDA if and only if θ∈Z2(A,V)\theta\in Z^{2}(A,V). Moreover B2(A,V)⊆Z2(A,V)B^{2}(A,V)\subseteq Z^{2}(A,V).

Proof. Products involving VV vanish, so it suffices to check (1.1)–(1.4) on x,y,z∈Ax,y,z\in A. For example x⊣(y⊣z)=x⊣A(y⊣Az)+θ⊣(x,y⊣Az)x\dashv(y\dashv z)=x\dashv_{A}(y\dashv_{A}z)+\theta_{\dashv}(x,y\dashv_{A}z) in AθA_{\theta}. Each identity in AθA_{\theta} therefore splits into its AA-component, which is the same identity in AA, and its VV-component, which is the corresponding condition of Definition 2.5. For δf\delta f, each condition is ff applied to an identity of AA, so it holds. □

Lemma 2.7. Ann⁡(Aθ)=(Ann⁡(A)∩Ann⁡(θ))⊕V\operatorname{Ann}(A_{\theta})=(\operatorname{Ann}(A)\cap\operatorname{Ann}(\theta))\oplus V.

Proof. V⊆Ann⁡(Aθ)V\subseteq\operatorname{Ann}(A_{\theta}). For x,y∈Ax,y\in A we have x⊣y=x⊣Ay+θ⊣(x,y)x\dashv y=x\dashv_{A}y+\theta_{\dashv}(x,y), with the two terms in AA and VV. So x⊣Aθ=0x\dashv A_{\theta}=0 iff x⊣AA=0x\dashv_{A}A=0 and θ⊣(x,A)=0\theta_{\dashv}(x,A)=0; the same holds for the other three conditions. □

Lemma 2.8. Let SS be an LSDA with V=Ann⁡(S)≠0V=\operatorname{Ann}(S)\ne0 and A=S/VA=S/V. Then S≅AθS\cong A_{\theta} for some θ∈Z2(A,V)\theta\in Z^{2}(A,V) with Ann⁡(A)∩Ann⁡(θ)=0\operatorname{Ann}(A)\cap\operatorname{Ann}(\theta)=0.

Proof. Choose a complement S=A′⊕VS=A'\oplus V and let π\pi be the projection onto A′A' along VV. Identify A′A' with A=S/VA=S/V; since VV is an ideal, the products of AA are x∘Ay=π(x∘y)x\circ_{A}y=\pi(x\circ y). Put θ∘(x,y)=x∘y−π(x∘y)∈V\theta_{\circ}(x,y)=x\circ y-\pi(x\circ y)\in V for ∘∈{⊣,⊢}\circ\in\{\dashv,\vdash\}. Then S=AθS=A_{\theta}, so θ\theta is a cocycle by Lemma 2.6, and Lemma 2.7 gives V=Ann⁡(S)=(Ann⁡(A)∩Ann⁡(θ))⊕VV=\operatorname{Ann}(S)=(\operatorname{Ann}(A)\cap\operatorname{Ann}(\theta))\oplus V. □

Fix a basis e1,…,ese_1,\ldots,e_s of VV and write θ=∑θiei\theta=\sum\theta_i e_i with θi∈Z2(A,C)\theta_i\in Z^2(A,\mathbb{C}). For φ∈Aut⁡(A)\varphi\in\operatorname{Aut}(A) put (φθ)(x,y)=θ(φx,φy)(\varphi\theta)(x,y)=\theta(\varphi x,\varphi y). Since Ann⁡(A)∩Ann⁡(θ+δf)=Ann⁡(A)∩Ann⁡(θ)\operatorname{Ann}(A)\cap\operatorname{Ann}(\theta+\delta f)=\operatorname{Ann}(A)\cap\operatorname{Ann}(\theta), the condition Ann⁡(A)∩Ann⁡(θ)=0\operatorname{Ann}(A)\cap\operatorname{Ann}(\theta)=0 depends only on the subspace ⟨[θ1],…,[θs]⟩⊆H2(A,C)\langle[\theta_1],\ldots,[\theta_s]\rangle\subseteq H^2(A,\mathbb{C}).

Lemma 2.9. Let θ,η∈Z2(A,V)\theta,\eta\in Z^2(A,V) with Ann⁡(A)∩Ann⁡(θ)=Ann⁡(A)∩Ann⁡(η)=0\operatorname{Ann}(A)\cap\operatorname{Ann}(\theta)=\operatorname{Ann}(A)\cap\operatorname{Ann}(\eta)=0.

  1. AθA_\theta is split if and only if [θ1],…,[θs][\theta_1],\ldots,[\theta_s] are linearly dependent in H2(A,C)H^2(A,\mathbb{C}).

  2. If AθA_\theta and AηA_\eta are non-split, then Aθ≅AηA_\theta\cong A_\eta if and only if there is φ∈Aut⁡(A)\varphi\in\operatorname{Aut}(A) with ⟨[φθ1],…,[φθs]⟩=⟨[η1],…,[ηs]⟩\langle[\varphi\theta_1],\ldots,[\varphi\theta_s]\rangle=\langle[\eta_1],\ldots,[\eta_s]\rangle.

Proof. (1) If the classes are dependent, change the basis of VV so that θs=δf\theta_s=\delta f with f:A→Cf:A\to\mathbb{C}. Replacing AA by its graph {x−f(x)es}\{x-f(x)e_s\} changes θ\theta by a coboundary and makes θs=0\theta_s=0; then Ces\mathbb{C}e_s is an ideal with complement A⊕⟨e1,…,es−1⟩A\oplus\langle e_1,\ldots,e_{s-1}\rangle, so AθA_\theta is split. Conversely, let Aθ=S′⊕CeA_\theta=S'\oplus\mathbb{C}e be a direct sum of ideals with e∈Ann⁡(Aθ)=Ve\in\operatorname{Ann}(A_\theta)=V. Then V=(V∩S′)⊕CeV=(V\cap S')\oplus\mathbb{C}e. Choose a complement S′=A′′⊕(V∩S′)S'=A''\oplus(V\cap S'); all products lie in S′S', so with respect to Aθ=A′′⊕VA_\theta=A''\oplus V and a basis of VV ending with ee, the last component of the cocycle vanishes. A change of complement and of basis of VV replaces the θi\theta_i by invertible combinations plus coboundaries, so the classes [θi][\theta_i] are dependent.

(2) Let ψ:Aθ→Aη\psi:A_\theta\to A_\eta be an isomorphism. By Lemma 2.7 both annihilators equal VV, so ψ(V)=V\psi(V)=V. Hence ψ(x+v)=φˉ(x)+g(x)+Mv\psi(x+v)=\bar{\varphi}(x)+g(x)+Mv with φˉ∈Aut⁡(A)\bar{\varphi}\in\operatorname{Aut}(A), g:A→Vg:A\to V linear and M∈GL⁡(V)M\in\operatorname{GL}(V). Comparing the VV-components of ψ(x∘y)\psi(x\circ y) and ψ(x)∘ψ(y)\psi(x)\circ\psi(y) for ∘∈{⊣,⊢}\circ\in\{\dashv,\vdash\} gives

Mθ+δg=φˉη.M\theta+\delta g=\bar{\varphi}\eta.

Conversely, if φˉ\bar{\varphi}, MM and gg satisfy this equation, the same computation shows that ψ(x+v)=φˉ(x)+g(x)+Mv\psi(x+v)=\bar{\varphi}(x)+g(x)+Mv is an isomorphism. Applying φˉ−1\bar{\varphi}^{-1} gives M(φθ)+δ(gφˉ−1)=ηM(\varphi\theta)+\delta(g\bar{\varphi}^{-1})=\eta with φ=φˉ−1\varphi=\bar{\varphi}^{-1}. Thus Aθ≅AηA_\theta\cong A_\eta iff [η1],…,[ηs][\eta_1],\ldots,[\eta_s] are the images of [φθ1],…,[φθs][\varphi\theta_1],\ldots,[\varphi\theta_s] under some invertible matrix MM. For two independent families this is exactly equality of their spans. □

Corollary 2.10. The non-split nn-dimensional nilpotent LSDA with ss-dimensional annihilator correspond bijectively to the pairs (A,W)(A,W), where AA runs over the (n−s)(n-s)-dimensional nilpotent LSDA up to isomorphism and WW over the Aut⁡(A)\operatorname{Aut}(A)-orbits of ss-dimensional subspaces W⊆H2(A,C)W\subseteq H^2(A,\mathbb{C}) with Ann⁡(A)∩Ann⁡(W)=0\operatorname{Ann}(A)\cap\operatorname{Ann}(W)=0. Every split algebra is B⊕CB\oplus\mathbb{C} with BB nilpotent of dimension n−1n-1.

Proof. Lemmas 2.4, 2.8 and 2.9. □

Two-dimensional nilpotent LSDA and their cohomology

Notation. In all multiplication tables only the non-zero products of basis vectors are listed. For an algebra with basis e1,…,ene_1,\ldots,e_n we write Δij⊣\Delta^{\dashv}_{ij} (resp. Δij⊢\Delta^{\vdash}_{ij}) for the cocycle with θ⊣(ei,ej)=1\theta_{\dashv}(e_i,e_j)=1 (resp. θ⊢(ei,ej)=1\theta_{\vdash}(e_i,e_j)=1) and all other values 00. For θ∈Z2(A,C)\theta\in Z^2(A,\mathbb{C}) we put λij=θ⊣(ei,ej)\lambda_{ij}=\theta_{\dashv}(e_i,e_j) and ρij=θ⊢(ei,ej)\rho_{ij}=\theta_{\vdash}(e_i,e_j). N1\mathcal{N}_1 and N20\mathcal{N}_2^0 denote the null algebras of dimensions 1 and 2.

Proposition 3.1. Every two-dimensional nilpotent LSDA over C\mathbb{C} is isomorphic to the null algebra N20\mathcal{N}_2^0 or to exactly one algebra

N2[α:β]:e1⊣e1=αe2,e1⊢e1=βe2,[α:β]∈P1.\mathcal{N}_2[\alpha:\beta]:\quad e_1\dashv e_1=\alpha e_2,\qquad e_1\vdash e_1=\beta e_2,\qquad[\alpha:\beta]\in\mathbb{P}^1.

Always Ann⁡(N2[α:β])=⟨e2⟩\operatorname{Ann}(\mathcal{N}_2[\alpha:\beta])=\langle e_2\rangle, and Aut⁡(N2[α:β])\operatorname{Aut}(\mathcal{N}_2[\alpha:\beta]) consists of the maps φ:e1↦pe1+qe2\varphi:e_1\mapsto pe_1+qe_2, e2↦p2e2e_2\mapsto p^2e_2 with p≠0p\ne0.

Proof. Let SS be non-null. By Lemma 2.4, dim⁡Ann⁡(S)=1\dim\operatorname{Ann}(S)=1, and A=S/Ann⁡(S)A=S/\operatorname{Ann}(S) is one-dimensional nilpotent, hence A=N1A=\mathcal{N}_1. Since N1\mathcal{N}_1 has zero products, every pair θ=(αΔ11⊣,βΔ11⊢)\theta=(\alpha\Delta^{\dashv}_{11},\beta\Delta^{\vdash}_{11}) is a cocycle and B2(N1)=0B^2(\mathcal{N}_1)=0, so H2(N1,C)=C2H^2(\mathcal{N}_1,\mathbb{C})=\mathbb{C}^2. The condition Ann⁡(N1)∩Ann⁡(θ)=0\operatorname{Ann}(\mathcal{N}_1)\cap\operatorname{Ann}(\theta)=0 means θ≠0\theta\ne0. Aut⁡(N1)=C∗\operatorname{Aut}(\mathcal{N}_1)=\mathbb{C}^* acts by p2p^2 on both coordinates, so the one-dimensional subspaces ⟨θ⟩\langle\theta\rangle, that is the points [α:β]∈P1[\alpha:\beta]\in\mathbb{P}^1, are pairwise inequivalent. By Corollary 2.10 this gives the list. An automorphism preserves S2=⟨e2⟩S^2=\langle e_2\rangle, so φ(e2)=re2\varphi(e_2)=re_2 and φ(e1)=pe1+qe2\varphi(e_1)=pe_1+qe_2; comparing φ(e1)∘φ(e1)\varphi(e_1)\circ\varphi(e_1) with φ(e1∘e1)\varphi(e_1\circ e_1) gives r=p2r=p^2. □

:::

Remark 3.2 (Corrections to [20]). In Case 2 of the proof of [20 Theorem 2.6] (α3=0\alpha_3=0) the identities give α1=0\alpha_1=0 and α2β1=0\alpha_2\beta_1=0, but only the subcase α2≠0\alpha_2\ne0, β1=0\beta_1=0, β2≠0\beta_2\ne0 was treated. It yields the nilpotent algebra e2⊣e2=ae1e_2\mathbin{\dashv}e_2=ae_1, e2⊢e2=e1e_2\mathbin{\vdash}e_2=e_1 (a≠0a\ne0), which is N2[a:1]\mathcal{N}_2[a:1] after interchanging e1e_1 and e2e_2; this algebra does not appear in the list L21\mathcal{L}_2^1–L26\mathcal{L}_2^6 of the statement. The remaining subcases give three further algebras: e2⊣e2=e1e_2\mathbin{\dashv}e_2=e_1 (that is, N2[1:0]\mathcal{N}_2[1:0]), e2⊢e2=e1e_2\mathbin{\vdash}e_2=e_1 (that is, N2[0:1]\mathcal{N}_2[0:1]), and the non-nilpotent algebra e2⊢e1=e1e_2\mathbin{\vdash}e_1=e_1, obtained after the change e2↦e2−be1e_2\mapsto e_2-be_1 removes a product e2⊢e2=be1e_2\mathbin{\vdash}e_2=be_1. The last algebra is not isomorphic to any of L21\mathcal{L}_2^1–L26\mathcal{L}_2^6, because its product ⊣\dashv is zero, while each of L21\mathcal{L}_2^1–L26\mathcal{L}_2^6 has ⊣≠0\dashv\ne0. Hence [20 Theorem 2.6] should be supplemented by the nilpotent algebras N2[α:β]\mathcal{N}_2[\alpha:\beta] of Proposition 3.1 and by the algebra e2⊢e1=e1e_2\mathbin{\vdash}e_1=e_1; all of them satisfy (1.1)–(1.4).

In what follows we normalise α=1\alpha=1 when α≠0\alpha\ne0, so the algebras are N2[1:β]\mathcal{N}_2[1:\beta] (β∈C\beta\in\mathbb{C}) and N2[0:1]\mathcal{N}_2[0:1].

Proposition 3.3. Let A=N2[α:β]A=\mathcal{N}_2[\alpha:\beta]. A pair θ\theta is a cocycle if and only if

λ22=ρ22=0,(α−β)λ12=(α−β)ρ21=0,αρ12=αλ12+(β−α)λ21,\lambda_{22}=\rho_{22}=0,\qquad(\alpha-\beta)\lambda_{12}=(\alpha-\beta)\rho_{21}=0,\qquad\alpha\rho_{12}=\alpha\lambda_{12}+(\beta-\alpha)\lambda_{21},

and B2(A,C)=⟨αΔ11⊣+βΔ11⊢⟩B^2(A,\mathbb{C})=\langle\alpha\Delta_{11}^{\dashv}+\beta\Delta_{11}^{\vdash}\rangle. Consequently:

AAConditions on θ\thetaB2B^2dim⁡H2\dim H^2Basis of H2H^2
N10\mathcal{N}_1^0none02Δ11⊣, Δ11⊢\Delta_{11}^{\dashv},\ \Delta_{11}^{\vdash}
N20\mathcal{N}_2^0none08all Δij⊣, Δij⊢\Delta_{ij}^{\dashv},\ \Delta_{ij}^{\vdash}
N2[1:β]\mathcal{N}_2[1:\beta], β≠1\beta\ne1λ12=λ22=ρ21=ρ22=0\lambda_{12}=\lambda_{22}=\rho_{21}=\rho_{22}=0, ρ12=(β−1)λ21\rho_{12}=(\beta-1)\lambda_{21}⟨Δ11⊣+βΔ11⊢⟩\langle\Delta_{11}^{\dashv}+\beta\Delta_{11}^{\vdash}\rangle2[Δ11⊣][\Delta_{11}^{\dashv}], [Δ21⊣+(β−1)Δ12⊢][\Delta_{21}^{\dashv}+(\beta-1)\Delta_{12}^{\vdash}]
N2[0:1]\mathcal{N}_2[0:1]λ12=λ21=λ22=ρ21=ρ22=0\lambda_{12}=\lambda_{21}=\lambda_{22}=\rho_{21}=\rho_{22}=0⟨Δ11⊢⟩\langle\Delta_{11}^{\vdash}\rangle2[Δ11⊣][\Delta_{11}^{\dashv}], [Δ12⊢][\Delta_{12}^{\vdash}]
N2[1:1]\mathcal{N}_2[1:1]λ22=ρ22=0\lambda_{22}=\rho_{22}=0, λ12=ρ12\lambda_{12}=\rho_{12}⟨Δ11⊣+Δ11⊢⟩\langle\Delta_{11}^{\dashv}+\Delta_{11}^{\vdash}\rangle4[Δ11⊣][\Delta_{11}^{\dashv}], [Δ21⊣][\Delta_{21}^{\dashv}], [Δ21⊢][\Delta_{21}^{\vdash}], [Δ12⊣+Δ12⊢][\Delta_{12}^{\dashv}+\Delta_{12}^{\vdash}]

Table 3.

Here dim⁡Z2\dim Z^2 equals 2, 8, 3, 3, 5 in the five rows.

Proof. The only non-zero products of AA are e1⊣e1=αe2e_1\mathbin{\dashv}e_1=\alpha e_2 and e1⊢e1=βe2e_1\mathbin{\vdash}e_1=\beta e_2, so a term θ(u,v∘w)\theta(u,v\circ w) or θ(u∘v,w)\theta(u\circ v,w) of Definition 2.5 is non-zero only when the inner pair is (e1,e1)(e_1,e_1). Evaluating the four conditions on all basis triples gives:

  • first condition, (ei,e1,e1)(e_i,e_1,e_1): (α−β)λi2=0(\alpha-\beta)\lambda_{i2}=0 for i=1,2i=1,2;

  • second condition, (e1,e1,ej)(e_1,e_1,e_j): (α−β)ρ2j=0(\alpha-\beta)\rho_{2j}=0 for j=1,2j=1,2;

  • third condition, triples (e1,e1,e1)(e_1,e_1,e_1), (e1,e1,e2)(e_1,e_1,e_2), (e2,e1,e1)(e_2,e_1,e_1), (e1,e2,e1)(e_1,e_2,e_1): α(λ12−λ21)=αρ12−βλ21\alpha(\lambda_{12}-\lambda_{21})=\alpha\rho_{12}-\beta\lambda_{21}, (α−β)λ22=0(\alpha-\beta)\lambda_{22}=0, αλ22=0\alpha\lambda_{22}=0 and αρ22=0\alpha\rho_{22}=0;

  • fourth condition, triples (e2,e1,e1)(e_2,e_1,e_1), (e1,e2,e1)(e_1,e_2,e_1): βρ22=0\beta\rho_{22}=0.

All remaining triples give 0=00=0. Since (α,β)≠(0,0)(\alpha,\beta)\ne(0,0), these equations are equivalent to the stated ones. For f=μe1∗+νe2∗f=\mu e_1^*+\nu e_2^* we get δf=ν(αΔ11⊣+βΔ11⊢)\delta f=\nu(\alpha\Delta_{11}^{\dashv}+\beta\Delta_{11}^{\vdash}). Specialising to α=1,β≠1\alpha=1,\beta\ne1, to [0:1][0:1] and to [1:1][1:1] gives the rows of the table; for N10\mathcal{N}_1^0 and N20\mathcal{N}_2^0 every pair is a cocycle and B2=0B^2=0. □

Lemma 3.4. Let A=N2[α:β]A=\mathcal{N}_2[\alpha:\beta], φ∈Aut⁡(A)\varphi\in\operatorname{Aut}(A) as in Proposition 3.1, and k∈C∗k\in\mathbb{C}^* a rescaling of V=CV=\mathbb{C}. Then kφθk\varphi\theta has

λ12′=kp3λ12,λ21′=kp3λ21,ρ12′=kp3ρ12,ρ21′=kp3ρ21,\lambda'_{12}=kp^3\lambda_{12},\qquad\lambda'_{21}=kp^3\lambda_{21},\qquad\rho'_{12}=kp^3\rho_{12},\qquad\rho'_{21}=kp^3\rho_{21},
λ11′=k(p2λ11+pq(λ12+λ21)),ρ11′=k(p2ρ11+pq(ρ12+ρ21)).\lambda'_{11}=k(p^2\lambda_{11}+pq(\lambda_{12}+\lambda_{21})),\qquad\rho'_{11}=k(p^2\rho_{11}+pq(\rho_{12}+\rho_{21})).

Proof. Expand θ(φei,φej)\theta(\varphi e_i,\varphi e_j) with φe1=pe1+qe2\varphi e_1=pe_1+qe_2, φe2=p2e2\varphi e_2=p^2e_2 and use λ22=ρ22=0\lambda_{22}=\rho_{22}=0. □

The main theorem

Theorem 4.1 (Main Theorem). Every three-dimensional nilpotent left-symmetric dialgebra over C\mathbb{C} is isomorphic to exactly one of the following algebras: the null algebra N3\mathcal{N}_3, a split algebra N2[α:β]⊕C\mathcal{N}_2[\alpha:\beta]\oplus\mathbb{C} with [α:β]∈P1[\alpha:\beta]\in\mathbb{P}^1, or one of the nine non-split algebras and families F1\mathcal{F}_1–F7\mathcal{F}_7, DD and $T below. Only non-zero products are listed; the basis is chosen so that e3e_3 spans Ann⁡(S)\operatorname{Ann}(S) (for F7F_7, e2e_2 and e3e_3 span it), hence every product with e3e_3 as a factor is zero.

Algebras with S/Ann⁡(S)S/\operatorname{Ann}(S) two-dimensional and non-null (nilpotency index 4):

  • F1(β)F_1(\beta), β∈C∖{1}\beta\in\mathbb{C} \setminus\{1\}: e1⊣e1=e2e_1 \dashv e_1=e_2, e1⊢e1=βe2e_1 \vdash e_1=\beta e_2, e2⊣e1=e3e_2 \dashv e_1=e_3, e1⊢e2=(β−1)e3e_1 \vdash e_2=(\beta-1)e_3;

  • F2F_2: e1⊢e1=e2e_1 \vdash e_1=e_2, e1⊢e2=e3e_1 \vdash e_2=e_3;

  • F3F_3: e1⊣e1=e3e_1 \dashv e_1=e_3, e1⊢e1=e2e_1 \vdash e_1=e_2, e1⊢e2=e3e_1 \vdash e_2=e_3;

  • F4(λ,m)F_4(\lambda,m), (λ,m)∈C2(\lambda,m)\in\mathbb{C}^2: e1⊣e1=e1⊢e1=e2e_1 \dashv e_1=e_1 \vdash e_1=e_2, e2⊣e1=λe3e_2 \dashv e_1=\lambda e_3, e2⊢e1=(λ−1)e3e_2 \vdash e_1=(\lambda-1)e_3, e1⊣e2=e1⊢e2=me3e_1 \dashv e_2=e_1 \vdash e_2=me_3;

  • F5[λ:m]F_5[\lambda:m], [λ:m]∈P1[\lambda:m]\in\mathbb{P}^1: e1⊣e1=e1⊢e1=e2e_1 \dashv e_1=e_1 \vdash e_1=e_2, e2⊣e1=e2⊢e1=λe3e_2 \dashv e_1=e_2 \vdash e_1=\lambda e_3, e1⊣e2=e1⊢e2=me3e_1 \dashv e_2=e_1 \vdash e_2=me_3;

  • F6[λ:m]F_6[\lambda:m], [λ:m]∈P1[\lambda:m]\in\mathbb{P}^1: as F5[λ:m]F_5[\lambda:m] but with e1⊣e1=e2+e3e_1 \dashv e_1=e_2+e_3.

Algebra with two-dimensional annihilator (nilpotency index 3):

  • F7F_7: e1⊣e1=e2e_1 \dashv e_1=e_2, e1⊢e1=e3e_1 \vdash e_1=e_3.

2-step nilpotent algebras with S/Ann⁡(S)≅N20S/\operatorname{Ann}(S)\cong\mathcal{N}_2^0 (nilpotency index 3):

  • D(α,δ,γ)D(\alpha,\delta,\gamma), with α=(α1,α2)\alpha=(\alpha_1,\alpha_2), δ=(δ1,δ2)\delta=(\delta_1,\delta_2), γ=(γ1,γ2)∈C2\gamma=(\gamma_1,\gamma_2)\in\mathbb{C}^2 such that γ≠0\gamma\ne0, or γ=0\gamma=0 and α≠0≠δ\alpha\ne0\ne\delta:

    e1⊣e1=α1e3e_1 \dashv e_1=\alpha_1e_3, e2⊣e2=δ1e3e_2 \dashv e_2=\delta_1e_3, e1⊣e2=−e2⊣e1=γ1e3e_1 \dashv e_2=-e_2 \dashv e_1=\gamma_1e_3,

    e1⊢e1=α2e3e_1 \vdash e_1=\alpha_2e_3, e2⊢e2=δ2e3e_2 \vdash e_2=\delta_2e_3, e1⊢e2=−e2⊢e1=γ2e3e_1 \vdash e_2=-e_2 \vdash e_1=\gamma_2e_3;

  • T(x,y)T(x,y), with x=(x1,x2)x=(x_1,x_2), y=(y1,y2)∈C2y=(y_1,y_2)\in\mathbb{C}^2 such that x+y≠0x+y\ne0: e1⊣e2=x1e3e_1 \dashv e_2=x_1e_3, e2⊣e1=y1e3e_2 \dashv e_1=y_1e_3, e1⊢e2=x2e3e_1 \vdash e_2=x_2e_3, e2⊢e1=y2e3e_2 \vdash e_1=y_2e_3, and moreover e1⊣e1=e3e_1 \dashv e_1=e_3 if x2+y2≠0x_2+y_2\ne0, e1⊢e1=e3e_1 \vdash e_1=e_3 if x2+y2=0x_2+y_2=0.

Algebras with different names are not isomorphic. Within the families:

(1) F1(β)F_1(\beta), F4(λ,m)F_4(\lambda,m), F5[λ:m]F_5[\lambda:m], F6[λ:m]F_6[\lambda:m] and N2[α:β]⊕CN_2[\alpha:\beta]\oplus\mathbb{C} are isomorphic only for equal parameters (equal points of P1\mathbb{P}^1 for the projective ones);

(2) D(α,δ,γ)≅D(α′,δ′,γ′)D(\alpha,\delta,\gamma)\cong D(\alpha',\delta',\gamma') if and only if there are λ,μ,ν∈C∗\lambda,\mu,\nu\in\mathbb{C}^* with ν2=λμ\nu^2=\lambda\mu such that (α′,δ′,γ′)=(λα,μδ,νγ)(\alpha',\delta',\gamma')=(\lambda\alpha,\mu\delta,\nu\gamma) or (α′,δ′,γ′)=(λδ,μα,νγ)(\alpha',\delta',\gamma')=(\lambda\delta,\mu\alpha,\nu\gamma);

(3) T(x,y)≅T(x′,y′)T(x,y)\cong T(x',y') if and only if (x′,y′)=ν(x,y)(x',y')=\nu(x,y) for some ν∈C∗\nu\in\mathbb{C}^*; that is, [x1:x2:y1:y2]∈P3[x_1:x_2:y_1:y_2]\in\mathbb{P}^3 is a complete invariant.

Thus DD has four essential parameters (two when γ=0\gamma=0) and TT has three. Normal forms for DD: if γ≠0\gamma\ne0 one may take γ=(1,γ2)\gamma=(1,\gamma_2) or (0,1)(0,1), and then (α,δ)(\alpha,\delta) is determined up to (tα,t−1δ)(t\alpha,t^{-1}\delta) and the swap (α,δ)↦(δ,α)(\alpha,\delta)\mapsto(\delta,\alpha); if γ=0\gamma=0, the algebra depends only on the unordered pair of points [α1:α2][\alpha_1:\alpha_2], [δ1:δ2]∈P1[\delta_1:\delta_2]\in\mathbb{P}^1.

Proof of the main theorem

Let SS be a three-dimensional nilpotent LSDA. By Lemma 2.4, dim⁡Ann⁡(S)≥1\dim\operatorname{Ann}(S)\ge1, and dim⁡Ann⁡(S)=3\dim\operatorname{Ann}(S)=3 means S=N3S=\mathcal{N}_3. If SS is split, then S=B⊕CS=B\oplus\mathbb{C} with BB a two-dimensional nilpotent LSDA, so SS is N3\mathcal{N}_3 or N2[α:β]⊕C\mathcal{N}_2[\alpha:\beta]\oplus\mathbb{C} by Proposition 3.1. In the latter case every x∉Ann⁡(S)x\notin\operatorname{Ann}(S) satisfies x⊣x=cαe2x\dashv x=c\alpha e_2 and x⊢x=cβe2x\vdash x=c\beta e_2 with c≠0c\ne0, so [α:β][\alpha:\beta] is an isomorphism invariant. From now on SS is non-split, and by Corollary 2.10 it corresponds to a pair (A,W)(A,W) with dim⁡A=3−dim⁡Ann⁡(S)\dim A=3-\dim\operatorname{Ann}(S).

Two-dimensional annihilator

If dim⁡Ann⁡(S)=2\dim\operatorname{Ann}(S)=2, then A=N1A=\mathcal{N}_1 and WW is a two-dimensional subspace of H2(N1,C)=C2H^2(\mathcal{N}_1,\mathbb{C})=\mathbb{C}^2, so W=⟨[Δ11⊣],[Δ11⊢]⟩W=\langle[\Delta_{11}^{\dashv}],[\Delta_{11}^{\vdash}]\rangle is the whole space. The condition Ann⁡(N1)∩Ann⁡(W)=0\operatorname{Ann}(\mathcal{N}_1)\cap\operatorname{Ann}(W)=0 holds. Taking θ=Δ11⊣e2+Δ11⊢e3\theta=\Delta_{11}^{\dashv}e_2+\Delta_{11}^{\vdash}e_3 gives F7F_7.

One-dimensional annihilator, A=N2[α:β]A=\mathcal{N}_2[\alpha:\beta]

Now dim⁡Ann⁡(S)=1\dim\operatorname{Ann}(S)=1, W=⟨[θ]⟩W=\langle[\theta]\rangle and V=Ce3V=\mathbb{C}e_3. By Lemma 2.9 the relevant group is Aut⁡(A)×C∗\operatorname{Aut}(A)\times\mathbb{C}^{*} acting as in Lemma 3.4, on classes modulo B2B^2. Since Ann⁡(A)=⟨e2⟩\operatorname{Ann}(A)=\langle e_2\rangle and λ22=ρ22=0\lambda_{22}=\rho_{22}=0, the condition Ann⁡(A)∩Ann⁡(θ)=0\operatorname{Ann}(A)\cap\operatorname{Ann}(\theta)=0 means (λ12,λ21,ρ12,ρ21)≠0(\lambda_{12},\lambda_{21},\rho_{12},\rho_{21})\ne0. In each case AθA_\theta has the products of AA plus the e3e_3-components given by θ\theta.

Case I: A=N2[1:β]A=\mathcal{N}_2[1:\beta], β≠1\beta\ne1. By Proposition 3.3, θ=λ11Δ11⊣+ρ11Δ11⊢+s(Δ21⊣+(β−1)Δ12⊢)\theta=\lambda_{11}\Delta^{\dashv}_{11}+\rho_{11}\Delta^{\vdash}_{11}+s(\Delta^{\dashv}_{21}+(\beta-1)\Delta^{\vdash}_{12}) with s≠0s\ne0. By Lemma 3.4, s↦kp3ss\mapsto kp^3s and (λ11,ρ11)↦k(p2(λ11,ρ11)+pqs(1,β−1))(\lambda_{11},\rho_{11})\mapsto k(p^2(\lambda_{11},\rho_{11})+pqs(1,\beta-1)). Modulo B2=⟨(1,β)⟩B^2=\langle(1,\beta)\rangle the pair (λ11,ρ11)(\lambda_{11},\rho_{11}) is recorded by the number βλ11−ρ11\beta\lambda_{11}-\rho_{11}, and the shift (1,β−1)(1,\beta-1) changes it by kpqs≠0kpqs\ne0 per unit of qq. Choosing qq makes this class zero, and k=(p3s)−1k=(p^3s)^{-1} gives s=1s=1. Hence there is a single orbit, represented by θ=Δ21⊣+(β−1)Δ12⊢\theta=\Delta^{\dashv}_{21}+(\beta-1)\Delta^{\vdash}_{12}, which gives F1(β)F_1(\beta).

Case II: A=N2[0:1]A=\mathcal{N}_2[0:1]. Here θ=cΔ11⊣+sΔ12⊢\theta=c\Delta^{\dashv}_{11}+s\Delta^{\vdash}_{12} modulo B2=⟨Δ11⊢⟩B^2=\langle\Delta^{\vdash}_{11}\rangle, with s≠0s\ne0. Since λ12=λ21=0\lambda_{12}=\lambda_{21}=0, Lemma 3.4 gives (c,s)↦(kp2c,kp3s)(c,s)\mapsto(kp^2c,kp^3s). The orbits are c=0c=0, giving F2F_2, and c≠0c\ne0 (take p=c/sp=c/s and k=(p3s)−1k=(p^3s)^{-1}, so c=s=1c=s=1), giving F3F_3. Whether c=0c=0 is invariant, so F2≇F3F_2\not\cong F_3.

Case III: A=N2[1:1]A=\mathcal{N}_2[1:1]. Take the representative with ρ11=0\rho_{11}=0 and write θ=dΔ11⊣+λΔ21⊣+ρΔ21⊢+m(Δ12⊣+Δ12⊢)\theta=d\Delta^{\dashv}_{11}+\lambda\Delta^{\dashv}_{21}+\rho\Delta^{\vdash}_{21}+m(\Delta^{\dashv}_{12}+\Delta^{\vdash}_{12}), with w=(λ,ρ,m)≠0w=(\lambda,\rho,m)\ne0. The class of d=λ11−ρ11d=\lambda_{11}-\rho_{11} modulo B2B^2 is well defined, and Lemma 3.4 gives

w↦kp3w,d↦k(p2d+pq(λ−ρ)).w\mapsto kp^3w,\qquad d\mapsto k(p^2d+pq(\lambda-\rho)).

If λ≠ρ\lambda\ne\rho, choose qq with d=0d=0 and kk with kp3(λ−ρ)=1kp^3(\lambda-\rho)=1, so λ−ρ=1\lambda-\rho=1. The transformations preserving this normal form satisfy kp3=1kp^3=1 and q=0q=0, so they fix (λ,m)(\lambda,m). This gives F4(λ,m)F_4(\lambda,m), with pairwise non-isomorphic members. If λ=ρ\lambda=\rho, then d↦kp2dd\mapsto kp^2d and w=(λ,λ,m)↦kp3ww=(\lambda,\lambda,m)\mapsto kp^3w, with (λ,m)≠0(\lambda,m)\ne0. For d=0d=0 only the point [λ:m]∈P1[\lambda:m]\in\mathbb{P}^1 survives, giving F5[λ:m]F_5[\lambda:m]. For d≠0d\ne0 put k=(p2d)−1k=(p^2d)^{-1}, so d=1d=1 and w↦(p/d)ww\mapsto(p/d)w; again only [λ:m][\lambda:m] survives, giving F6[λ:m]F_6[\lambda:m]. The quantities λ−ρ\lambda-\rho (zero or not) and dd (zero or not when λ=ρ\lambda=\rho) are invariant, so F4F_4, F5F_5 and F6F_6 are mutually non-isomorphic.

One-dimensional annihilator, A=N20A=\mathcal{N}_2^0 (the 2-step case)

Now every pair is a cocycle and B2=0B^2=0, so θ\theta is a pair of 2×22\times2 matrices B1=(λij)B_1=(\lambda_{ij}), B2=(ρij)B_2=(\rho_{ij}). Since Aut⁡(N20)=GL⁡2\operatorname{Aut}(\mathcal{N}_2^0)=\operatorname{GL}_2, Lemma 2.9 says that (B1,B2)(B_1,B_2) and (kPTB1P,kPTB2P)(kP^{T}B_1P,kP^{T}B_2P) give isomorphic algebras for P∈GL⁡2P\in\operatorname{GL}_2, k∈C∗k\in\mathbb{C}^{*}, and only these. The algebra is non-split if and only if there is no x≠0x\ne0 with Bix=BiTx=0B_ix=B_i^{T}x=0 for i=1,2i=1,2.

Step 1 (symmetric and skew parts). Write Bi=Si+γiJB_i=S_i+\gamma_iJ with SiS_i symmetric and J=(01−10)J=\begin{pmatrix}0&1\\-1&0\end{pmatrix}. Since PTJP=det⁡(P)JP^{T}JP=\det(P)J, the group acts by Si↦kPTSiPS_i\mapsto kP^{T}S_iP and γ↦kdet⁡(P)γ\gamma\mapsto k\det(P)\gamma. Because Bix=BiTx=0B_ix=B_i^{T}x=0 iff Six=0S_ix=0 and γiJx=0\gamma_iJx=0, the algebra is non-split iff γ≠0\gamma\ne0, or γ=0\gamma=0 and S1,S2S_1,S_2 have no common kernel vector.

Step 2 (orthogonal geometry). Let QQ be the space of symmetric 2×22\times2 matrices with the quadratic form q(S)=det⁡Sq(S)=\det S and its polar form ⟨ , ⟩\langle\ ,\ \rangle. The map ρ(P)S=PTSP/det⁡P\rho(P)S=P^{T}SP/\det P preserves qq, its kernel is the group of scalar matrices, and its image is a connected three-dimensional subgroup of O(Q,q)O(Q,q), hence equal to SO(Q,q)SO(Q,q). With c=kdet⁡Pc=k\det P we have kPTSP=cρ(P)SkP^{T}SP=c\rho(P)S, so the action on (S1,S2,γ)(S_1,S_2,\gamma) is (cρ(P)S1,cρ(P)S2,cγ)(c\rho(P)S_1,c\rho(P)S_2,c\gamma) with c∈C∗c\in\mathbb{C}^{*} and ρ(P)∈SO(Q,q)\rho(P)\in SO(Q,q) independent. The non-zero isotropic vectors of QQ are the symmetric matrices of rank one.

Step 3 (two types). Let U=⟨S1,S2⟩U=\langle S_1,S_2\rangle. Either UU lies in a non-degenerate plane of QQ, or UU is a degenerate plane; in the second case U=ℓ⊥U=\ell^\perp for the isotropic line ℓ=rad⁡U\ell=\operatorname{rad}U, because QQ contains no totally isotropic plane. Every vector, and every degenerate plane, lies in a non-degenerate plane, and by Witt’s theorem [15] all non-degenerate planes are conjugate under SO(Q,q)SO(Q,q) to the plane Δ0\Delta_0 of diagonal matrices. So we get:

  • type D: after a congruence S1S_1 and S2S_2 are both diagonal, i.e. they are simultaneously diagonalisable by congruence;

  • type T: S1,S2S_1,S_2 are linearly independent and the binary form det⁡(sS1+tS2)\det(sS_1+tS_2) has a double root.

After moving ℓ\ell to ⟨E11⟩\langle E_{11}\rangle, type TT has both SiS_i in ℓ⊥={(abb0)}\ell^\perp=\left\{\begin{pmatrix}a&b\\b&0\end{pmatrix}\right\}. The type is invariant under the group, so no algebra of type DD is isomorphic to one of type TT.

Step 4 (type D). Write Bi=(αiγi−γiδi)B_i=\begin{pmatrix}\alpha_i & \gamma_i \\ -\gamma_i & \delta_i\end{pmatrix}. For P=diag⁡(p,r)P=\operatorname{diag}(p,r) the action is

(α,δ,γ)↦(kp2α,kr2δ,kprγ).(\alpha,\delta,\gamma)\mapsto(kp^2\alpha,kr^2\delta,kpr\gamma).

which realises exactly the triples (λ,μ,ν)(\lambda,\mu,\nu) with ν2=λμ\nu^2=\lambda\mu; for P=JP=J and k=1k=1 it swaps α\alpha and δ\delta and fixes γ\gamma. Conversely, let (P,k)(P,k) map one algebra of type DD in diagonal form to another, with c=kdet⁡Pc=k\det P. Then ρ(P)\rho(P) restricts to an isometry from ⟨S1,S2⟩\langle S_1,S_2\rangle onto ⟨c−1S1′,c−1S2′⟩\langle c^{-1}S'_1,c^{-1}S'_2\rangle, both inside Δ0\Delta_0. By Witt’s theorem in Δ0\Delta_0 it extends to h∈O(Δ0)h\in O(\Delta_0), and h⊕det⁡(h)h\oplus\det(h) on Δ0⊥\Delta_0^\perp equals ρ(P0)\rho(P_0) with P0P_0 diagonal or antidiagonal, since those are exactly the PP for which PT(1001)PP^{\mathsf T}\begin{pmatrix}1&0\\0&1\end{pmatrix}P is proportional to (1001)\begin{pmatrix}1&0\\0&1\end{pmatrix}. Hence the equivalence is realised by the maps above, which proves the isomorphism criterion for DD. If γ=0\gamma=0, a common kernel vector exists iff α=0\alpha=0 (then e1e_1) or δ=0\delta=0 (then e2e_2).

Step 5 (type T). Since ℓ=rad⁡U\ell=\operatorname{rad}U is determined by the algebra, equivalences between normal forms fix ⟨E11⟩\langle E_{11}\rangle, so PTE11PP^{\mathsf T}E_{11}P is a multiple of E11E_{11} and P=(p0sr)P=\begin{pmatrix}p&0\\s&r\end{pmatrix} is lower triangular. Writing Bi=(aixiyi0)B_i=\begin{pmatrix}a_i&x_i\\y_i&0\end{pmatrix}, a direct computation gives

(x,y)↦kpr(x,y),a↦k(p2a+ps(x+y)).(x,y)\mapsto kpr(x,y),\qquad a\mapsto k(p^2a+ps(x+y)).

Type TT means that a=(a1,a2)a=(a_1,a_2) and x+yx+y are linearly independent. Keeping kprkpr fixed, aa can be moved to any vector independent of x+yx+y: we take a=(1,0)a=(1,0) if x2+y2≠0x_2+y_2\ne0 and a=(0,1)a=(0,1) otherwise. The maps preserving this aa are those with kp2=1kp^2=1 and s=0s=0, and they act by (x,y)↦ν(x,y)(x,y)\mapsto\nu(x,y) with ν=r/p\nu=r/p arbitrary. Hence [x1:x2:y1:y2]∈P3[x_1:x_2:y_1:y_2]\in\mathbb{P}^3 is a complete invariant. Type TT is always non-split: if S1v=S2v=0S_1v=S_2v=0 with Si=(aibibi0)S_i=\begin{pmatrix}a_i&b_i\\b_i&0\end{pmatrix} and b≠0b\ne0, then biv1=0b_iv_1=0 gives v1=0v_1=0 and then biv2=0b_iv_2=0 gives v2=0v_2=0.

This completes the proof of Theorem 4.1, except for the non-isomorphism of algebras with different names, which follows from Proposition 6.1.

Invariants, non-isomorphism and special subclasses

Table 1 lists invariants of all algebras of Theorem 4.1. Here Der⁡(S)\operatorname{Der}(S) is the Lie algebra of derivations, i.e. linear maps DD with D(x∘y)=Dx∘y+x∘DyD(x\mathbin{\circ}y)=Dx\mathbin{\circ}y+x\mathbin{\circ}Dy for ∘∈{⊣,⊢}\mathbin{\circ}\in\{\dashv,\vdash\}; its dimension equals dim⁡Aut⁡(S)\dim\operatorname{Aut}(S). For the families, dim⁡Der⁡\dim\operatorname{Der} is given at generic parameters.

Algebradim Ann S/Ann⁡(S)S/\operatorname{Ann}(S)nil. indexdim S2S^2dim Der⊣=⊢\dashv=\vdashdiassociative
F1(β)F_1(\beta)1N2[1:β]\mathcal{N}_2[1:\beta]422nono
F2F_21N2[0:1]\mathcal{N}_2[0:1]423nono
F3F_31N2[0:1]\mathcal{N}_2[0:1]422nono
F4(λ,m)F_4(\lambda,m)1N2[1:1]\mathcal{N}_2[1:1]422nono
F5[λ:m]F_5[\lambda:m]1N2[1:1]\mathcal{N}_2[1:1]423yesiff λ=m\lambda=m
F6[λ:m]F_6[\lambda:m]1N2[1:1]\mathcal{N}_2[1:1]422noiff λ=m\lambda=m
F7F_72N1\mathcal{N}_1323noyes
D(α,δ,γ)D(\alpha,\delta,\gamma)1N20\mathcal{N}_2^0313iff α1=α2,δ1=δ2,γ1=γ2\alpha_1=\alpha_2,\delta_1=\delta_2,\gamma_1=\gamma_2yes
T(x,y)T(x,y)1N20\mathcal{N}_2^0313noyes
N2[α:β]⊕C\mathcal{N}_2[\alpha:\beta]\oplus\mathbb{C}2–315iff α=β\alpha=\betayes
N3\mathcal{N}_33–209yes…

Table 1. Invariants of the three-dimensional nilpotent LSDA.

Proposition 6.1. The algebras of Theorem 4.1 with different names are pairwise non-isomorphic.

Proof. dim⁡Ann⁡(S)\dim\operatorname{Ann}(S) and the isomorphism class of S/Ann⁡(S)S/\operatorname{Ann}(S) are invariants. By Proposition 3.1 they separate F1(β)F_1(\beta) (for each β\beta), the group {F2,F3}\{F_2,F_3\}, the group {F4,F5,F6}\{F_4,F_5,F_6\}, the group {D,T}\{D,T\}, F7F_7, the split algebras and N3\mathcal{N}_3. Inside each group the algebras correspond to different orbits by Lemma 2.9 and § 5: c=0c=0 or not for F2,F3F_2,F_3; λ−ρ=0\lambda-\rho=0 or not and d=0d=0 or not for F4,F5,F6F_4,F_5,F_6; simultaneous diagonalisability for DD and TT. The column dim⁡Der⁡\dim\operatorname{Der} gives an independent check for F2≇F3F_{2} \not\cong F_{3} and for F5≇F4,F6F_{5} \not\cong F_{4}, F_{6}. □

Corollary 6.2 (Left-symmetric algebras). Up to isomorphism, the three-dimensional nilpotent complex left-symmetric algebras are N3\mathcal{N}_{3}, N2[1:1]⊕C\mathcal{N}_{2}[1:1]\oplus\mathbb{C} (e1e1=e2e_{1}e_{1}=e_{2}), F5[λ:m]F_{5}[\lambda:m] (e1e1=e2e_{1}e_{1}=e_{2}, e2e1=λe3e_{2}e_{1}=\lambda e_{3}, e1e2=me3e_{1}e_{2}=me_{3}) and the 2-step algebras eiej=Bije3e_{i}e_{j}=B_{ij}e_{3} with

B=I,B=J,B=J+E11,B=J+diag⁡(1,μ)  (μ≠0).B=I,\qquad B=J,\qquad B=J+E_{11},\qquad B=J+\operatorname{diag}(1,\mu)\;(\mu\ne0).

Proof. These are the algebras of Theorem 4.1 with ⊣=⊢\dashv=\vdash. For DD this means one form B=(ag−gd)B=\begin{pmatrix}a&g\\-g&d\end{pmatrix}. If g=0g=0 then a,d≠0a,d\ne0 and B∼IB\sim I. If g≠0g\ne0 we normalise g=1g=1; then (a,d)∼(λa,λ−1d)(a,d)\sim(\lambda a,\lambda^{-1}d) and the swap, so adad is the only invariant, with the two orbits (0,0)(0,0) and (1,0)(1,0) when ad=0ad=0. In TT the condition a∥x+ya\parallel x+y would be needed, which is excluded. □

Corollary 6.3 (Diassociative algebras). The three-dimensional nilpotent complex LSDA that are diassociative are F5[1:1]F_{5}[1:1], F6[1:1]F_{6}[1:1], F7F_{7}, all D(α,δ,γ)D(\alpha,\delta,\gamma), all T(x,y)T(x,y), N2[α:β]⊕C\mathcal{N}_{2}[\alpha:\beta]\oplus\mathbb{C} and N3\mathcal{N}_{3}.

Proof. Algebras with S3=0S^{3}=0 satisfy every identity of degree three, in particular the five diassociative axioms. For F1F_{1}–F6F_{6}, the axioms on basis triples hold only for F5[λ:m]F_{5}[\lambda:m] and F6[λ:m]F_{6}[\lambda:m] with λ=m\lambda=m; for instance (e1⊣e1)⊣e1=λe3(e_{1}\dashv e_{1})\dashv e_{1}=\lambda e_{3} and e1⊣(e1⊣e1)=me3e_{1}\dashv(e_{1}\dashv e_{1})=me_{3} in F5F_{5} and F6F_{6}. □

Remark 6.4. For a generic algebra of type DD the stabiliser of (B1,B2)(B_{1},B_{2}) in GL2×C∗\mathrm{GL}_{2}\times\mathbb{C}^{*} is one-dimensional, and the maps ei↦ei+cie3e_{i}\mapsto e_{i}+c_{i}e_{3} (i=1,2i=1,2) give two more dimensions; this agrees with dim⁡Der⁡=3\dim\operatorname{Der}=3 in Table 1.

Concluding remarks

The three-dimensional nilpotent LSDA form nine non-split classes, with a four-parameter family already in the 2-step case. Natural next steps are the geometric classification (degenerations and irreducible components of the variety of three-dimensional nilpotent LSDA), the classification of four-dimensional nilpotent LSDA by the same method, and the classification of all three-dimensional LSDA, extending Theorem 2.6 of [20].

References

  1. [1]S. Albeverio, B. A. Omirov, I. S. Rakhimov, Classification of 4-dimensional nilpotent complex Leibniz algebras, Extracta Math. 21 (2006), no. 3, 197–210.DOI
  2. [2]C. Bai, Bijective 1-cocycles and classification of 3-dimensional left-symmetric algebras, Comm. Algebra 37 (2009), no. 3, 1016–1057.DOI
  3. [3]L. Boza, E. M. Fedriani, J. Núñez, Á. F. Tenorio, A historical review of the classifications of Lie algebras, Rev. Un. Mat. Argentina 54 (2013), no. 2, 75–99.
  4. [4]D. Burde, Left-symmetric algebras, or pre-Lie algebras in geometry and physics, Cent. Eur. J. Math. 4 (2006), no. 3, 323–357.DOI
  5. [5]A. Cayley, On the theory of analytic forms called trees, in: Collected Mathematical Papers, Vol. 3, Cambridge Univ. Press, 1890, 242–246.
  6. [6]F. Chapoton, M. Livernet, Pre-Lie algebras and the rooted trees operad, Internat. Math. Res. Notices 2001, no. 8, 395–408.DOI
  7. [7]W. A. de Graaf, Classification of 6-dimensional nilpotent Lie algebras over fields of characteristic not 2, J. Algebra 309 (2007), no. 2, 640–653.DOI
  8. [8]R. Felipe, A brief foundation of the left-symmetric dialgebras, Comunicaciones del CIMAT, No. I-11-02, 2011.
  9. [9]M. Goze, Y. Khakimdjanov, Nilpotent Lie Algebras, Mathematics and its Applications 361, Kluwer, Dordrecht, 1996.DOI
  10. [10]C. M. González, Associative algebras from a structural viewpoint, Comm. Algebra 41 (2013), no. 5, 1903–1912.DOI
  11. [11]A. S. Hegazi, H. Abdelwahab, A. J. Calderón Martín, The classification of n-dimensional non-Lie Malcev algebras with (n − 4)-dimensional annihilator, Linear Algebra Appl. 505 (2016), 32–56.
  12. [13]I. Kaygorodov, Non-associative algebraic structures: classification and structure, Commun. Math. 32 (2024), no. 3, 1–62.DOI
  13. [14]H. Kim, Complete left-invariant affine structures on nilpotent Lie groups, J. Differential Geom. 24 (1986), no. 3, 373–394.DOI
  14. [15]T. Y. Lam, Introduction to Quadratic Forms over Fields, Graduate Studies in Mathematics 67, Amer. Math. Soc., Providence, 2005.DOI
  15. [16]X. Li, D. Hou, C. Bai, Rota–Baxter operators on pre-Lie algebras, J. Nonlinear Math. Phys. 14 (2007), no. 2, 269–289.DOI
  16. [17]L. Lin, Y. Zhang, F[x, y] as a dialgebra and a Leibniz algebra, Comm. Algebra 38 (2010), no. 9, 3417–3447.
  17. [18]J.-L. Loday, Une version non commutative des algèbres de Lie: les algèbres de Leibniz, Enseign. Math. (2) 39 (1993), no. 3–4, 269–293.DOI
  18. [19]J.-L. Loday, A. Frabetti, F. Chapoton, F. Goichot, Dialgebras and Related Operads, Lecture Notes in Mathematics 1763, Springer, Berlin, 2001.DOI
  19. [20]I. M. Rikhsioboev, On the classification of left-symmetric dialgebras, Contemp. Math. 672, Amer. Math. Soc., Providence, 2016. DOI 10.1090/conm/672/13464.
  20. [21]I. M. Rikhsioboev, I. S. Rakhimov, W. Basri, Classification of 3-dimensional complex diassociative algebras, Malaysian J. Math. Sci. 4 (2010), no. 2, 241–254.
  21. [22]I. M. Rikhsioboev, I. S. Rakhimov, W. Basri, On four dimensional nilpotent diassociative algebras, J. Gen. Lie Theory Appl. 9 (2015), no. 1, 7 pp.
  22. [23]T. Skjelbred, T. Sund, Sur la classification des algèbres de Lie nilpotentes, C. R. Acad. Sci. Paris Sér. A-B 286 (1978), no. 5, A241–A242.
  23. [24]È. B. Vinberg, The theory of homogeneous convex cones, Trudy Moskov. Mat. Obšč. 12 (1963), 303–358.

Paper details

vlnc:W_562fsm39
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY-SA 4.0
  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents