betaFeedback

WorksMathematics

Decomposition of trilinear forme

Abstract

In this work, we are interested in the decomposition of trilinear forms. We associate this decomposition with the decomposition into irreducible subspaces of the group algebra associated with the symmetric group. We have a correspondence between a ternary form and a symmetric trilinear form, in the same way that we have the correspondence between a quadratic form and a symmetric quadratic form. We are then interested in the decomposition of ternary forms into a sum of cubes of linear forms and as an application we find the Tartaglia formulae on the expression of the roots of equations of degree 3.

Introduction

The decomposition of a bilinear form into a sum of a symmetric form and a skew-symmetric form is actually based on the decomposition into irreducible subspaces of the associative group algebra defined by the symmetric group Σ2\Sigma_2 of degree 2. Starting from this remark, we propose a decomposition of trilinear forms by describing a decomposition of the group algebra associated with the symmetric group Σ3\Sigma_3 of degree 3. Recall that if GG is a finite group, the group algebra A(G)A(G) defined by GG is the finite dimensional vector space generated by GG and this space is naturally endowed with an associative algebra structure and a GG-module structure.

We extend this result to trilinear applications with values in a vector space. This is the case, for example, of the associator of a multiplication of a non-associative algebra. This makes it possible to highlight particular classes of such algebras.

To complete this work, we are interested in ternary forms (also called cubic forms) associated with symmetric trilinear forms. We show under which conditions these ternary forms on a two-dimensional K\mathbb{K}-vector space EE can be decomposed into a sum of two cubes of linear forms. As a direct consequence, we find again the Tartaglia formulae giving the expressions of roots of real equations of degree 3.

We briefly touch upon finding solutions to degree 5 equations using this method.

The trilinear forms and more generally the multilinear forms occur in different mathematical areas and subject such as cryptography (see [2]) and also in the study of properties of algebra laws (see [1]).

Definition

Trilinear forms

Let EE be a K\mathbb{K}-vector space where K=R\mathbb{K} = \mathbb{R} or C\mathbb{C}. A map

f:E3→Kf : E^3 \to\mathbb{K}

is a trilinear form if ff is a linear form with respect to each of its arguments. If dim⁡E=n\dim E = n and if {e1,⋯ ,en}\{e_1,\cdots,e_n\} is a basis of EE, then the writing of ff in this basis is an homogeneous algebraic expression of degree 3. The constants

aijk=f(ei,ej,ek)a_{ijk}=f(e_i,e_j,e_k)

are the structure constants (or coefficients) of ff with respect to this basis. The formulae of change of basis are the classical formulae corresponding to f∈(E⊗E⊗E)∗f\in(E\otimes E\otimes E)^*.

Definition 1. Let ff be a trilinear form on EE. Then

(1) ff is symmetric if for every x1,x2,x3∈Ex_1,x_2,x_3\in E,

∀σ∈Σ3,f(xσ(1),xσ(2),xσ(3))=f(x1,x2,x3),\forall\sigma\in\Sigma_3,\quad f(x_{\sigma(1)},x_{\sigma(2)},x_{\sigma(3)})=f(x_1,x_2,x_3),

(2) ff is alternating if it vanishes when two inputs are the same.

A trilinear form on EE is antisymmetric (skew-symmetric) if

f(xσ(1),xσ(2),xσ(3))=−f(x1,x2,x3)f(x_{\sigma(1)},x_{\sigma(2)},x_{\sigma(3)})=-f(x_1,x_2,x_3)

for every transposition σ∈Σ3\sigma\in\Sigma_3 and for every x1,x2,x3∈Ex_1,x_2,x_3\in E. But, since we assume that the characteristic of K\mathbb{K} is 0, the notions of antisymmetric and alternating trilinear forms coincide.

A pp-linear form f:Ep→Kf:E^p\to\mathbb{K} is defined in the same way as a linear form with respect to each of its pp arguments.

Symmetrization-alternatization

Let ff be a trilinear form on EE. The trilinear form fsf_s defined by

fs(x1,x2,x3)=∑σ∈Σ3f(xσ(1),xσ(2),xσ(3))f_s(x_1,x_2,x_3)=\sum_{\sigma\in\Sigma_3}f(x_{\sigma(1)},x_{\sigma(2)},x_{\sigma(3)})

is symmetric. It is called the symmetrization of ff.

The trilinear form faf_a defined by

fa(x1,x2,x3)=∑σ∈Σ3sgn⁡(σ)f(xσ(1),xσ(2),xσ(3))f_a(x_1,x_2,x_3)=\sum_{\sigma\in\Sigma_3}\operatorname{sgn}(\sigma)f(x_{\sigma(1)},x_{\sigma(2)},x_{\sigma(3)})

is alternating. It is called the alternatization of ff.

Remark : case p=2p=2. It is well known, that for bilinear forms we have a converse to this construction. Given a symmetric bilinear form f1f_1 and an alternating bilinear form f2f_2, there is a unique bilinear form ff such that

fs=f1,fa=f2.f_s=f_1,\quad f_a=f_2.

This form is given by

f=fs+fa2f=\frac{f_s+f_a}{2}

(recall that we assume that Char⁡(K)=0\operatorname{Char}(\mathbb{K})=0).

In the next section, we will give an interpretation of this decomposition in terms of representations of Σ3\Sigma_3 to extend it to the trilinear case.

The group algebra K[Σp]K[\Sigma_p]

Generalities

If Σp\Sigma_p is the symmetric group of degree pp, then the group algebra K[Σp]K[\Sigma_p] is the associative algebra over KK whose underlying KK-module is the free module over KK on the underlying set of Σp\Sigma_p and whose multiplication is given on basis elements by the group operation. An element of K[Σp]K[\Sigma_p] is a formal combination

v=∑σ∈Σpaσσv = \sum_{\sigma\in\Sigma_p} a_\sigma\sigma

The action of Σp\Sigma_p on K[Σp]K[\Sigma_p] is defined by

(σ,v=∑iaiσi)⟶σv=∑iaiσ−1∘σi.(\sigma,v=\sum_i a_i\sigma_i)\longrightarrow\sigma v=\sum_i a_i\sigma^{-1}\circ\sigma_i.

For each v∈K[Σp]v\in K[\Sigma_p], we denote by O(v)\mathcal{O}(v) the corresponding orbit and by Fv=K(O(v))F_v=K(\mathcal{O}(v)) the linear subspace of K[Σp]K[\Sigma_p] generated by O(v)\mathcal{O}(v). A subspace FF is called invariant if for every σ∈Σp\sigma\in\Sigma_p, we have σF⊂F\sigma F\subset F. In particular the subspaces FvF_v are invariant.

Examples. We denote by τi,j\tau_{i,j} the transposition between ii and jj.

(1) p=2p=2. We have

Σ2={Id,τ1,2}\Sigma_2=\{Id,\tau_{1,2}\}

and

K[Σ2]={a1Id+a2τ1,2, ai∈K}.K[\Sigma_2]=\{a_1Id+a_2\tau_{1,2},\ a_i\in K\}.

If v=a1Id+a2τ1,2v=a_1Id+a_2\tau_{1,2}, then O(v)={v,a2Id+a1τ1,2}\mathcal{O}(v)=\{v,a_2Id+a_1\tau_{1,2}\}. We deduce that dim⁡Fv=2\dim F_v=2 except if a1−a2=0a_1-a_2=0.

Proposition 2. Assume that v≠0v\ne0. Then dim⁡Fv=2\dim F_v=2 except if v=a1(Id−τ1,2)v=a_1(Id-\tau_{1,2}) or v=a1(Id+τ1,2)v=a_1(Id+\tau_{1,2}). In these last cases, if a1≠0a_1\ne0, dim⁡Fv=1\dim F_v=1

If we denote by

V2=Id−τ1,2,W2=Id+τ1,2V_2=Id-\tau_{1,2},\qquad W_2=Id+\tau_{1,2}

then FV2F_{V_2} and FW2F_{W_2} are the only one-dimensional invariant subspaces of K[Σ2]K[\Sigma_2] and we have the classical decomposition

K[Σ2]=FV2⊕FW2.K[\Sigma_2]=F_{V_2}\oplus F_{W_2}.

(2) p=3p=3. We have

Σ3={Id,τ1,2,τ1,3,τ2,3,c,c2}\Sigma_3=\{Id,\tau_{1,2},\tau_{1,3},\tau_{2,3},c,c^2\}

where cc is the cycle (2,3,1)(2,3,1) and c2=c∘cc^2=c\circ c. If we denote by V3V_3 and W3W_3 the vectors

V3=Id−τ1,2−τ1,3−τ2,3+c+c2,W3=Id+τ1,2+τ1,3+τ2,3+c+c2,V_3=Id-\tau_{1,2}-\tau_{1,3}-\tau_{2,3}+c+c^2,\qquad W_3=Id+\tau_{1,2}+\tau_{1,3}+\tau_{2,3}+c+c^2,

then dim⁡FV3=dim⁡FW3=1\dim F_{V_3}=\dim F_{W_3}=1 and these spaces are the only one-dimensional invariant subspaces of K[Σ3]K[\Sigma_3].

Irreducible subspaces of dimension 2 for p=3p = 3

An invariant subspace FF of K[Σ3]\mathbb{K}[\Sigma_{3}] is called irreducible if any invariant subspace of FF is trivial. The only one-dimensional irreducible subspaces are FV3F_{V_{3}} and FW3F_{W_{3}}.

Assume now that FF is a 2-dimensional irreducible subspace of K[Σ3]\mathbb{K}[\Sigma_{3}]. For any v≠0∈Fv \ne0 \in F, {σv,σ∈Σ3}\{\sigma v, \sigma\in\Sigma_{3}\} is a generating family of FF and there exists σ∈Σ3\sigma\in\Sigma_{3}, such as {v,σv}\{v,\sigma v\} is a basis of FF.

Let v=a1Id+a2τ12+a3τ13+a4τ23+a5c1+a6c2v = a_{1}Id + a_{2}\tau_{12} + a_{3}\tau_{13} + a_{4}\tau_{23} + a_{5}c_{1} + a_{6}c_{2} be in K[Σ3]\mathbb{K}[\Sigma_{3}].

The family {v,τ1,2v}\{v,\tau_{1,2}v\} is linearly dependent

The hypothesis implies τ1,2v=λv\tau_{1,2}v = \lambda v. Solving this linear system, we obtain λ=±1\lambda= \pm1 and v=(a1,a1,a3,a4,a4,a3)v = (a_{1},a_{1},a_{3},a_{4},a_{4},a_{3}) or (a1,−a1,a3,a4,−a4,−a3)(a_{1},-a_{1},a_{3},a_{4},-a_{4},-a_{3}). Since dim⁡Fv=2\dim F_{v} = 2, then there exists σ∈Σ3\sigma\in\Sigma_{3} such that {v,σv}\{v,\sigma v\} is a basis of FvF_{v}. If we also have that τ1,3v=λv\tau_{1,3}v = \lambda v, then v=V3v = V_{3} or W3W_{3} and FvF_{v} is not irreducible. Then we can assume that {v,τ1,3v}\{v,\tau_{1,3}v\} is a basis of FvF_{v}.

In the first case, that is v=(a1,a1,a3,a4,a4,a3)v = (a_{1},a_{1},a_{3},a_{4},a_{4},a_{3}), we have

v+τ1,3v+τ2,3v=(a1+a3+a4)W3.v + \tau_{1,3}v + \tau_{2,3}v = (a_{1} + a_{3} + a_{4})W_{3}.

If a1+a3+a4≠0a_{1} + a_{3} + a_{4} \ne0, W3∈FvW_{3} \in F_{v} and this subspace is not irreducible. Then a1+a3+a4=0a_{1} + a_{3} + a_{4} = 0 and

v=(a1,a1,a3,−a1−a3,−a1−a3,a3).v = (a_{1},a_{1},a_{3},-a_{1}-a_{3},-a_{1}-a_{3},a_{3}).

In the second case, that is v=(a1,−a1,a3,a4,−a4,−a3)v = (a_{1},-a_{1},a_{3},a_{4},-a_{4},-a_{3}), we have

v−τ1,3v−τ2,3v=(a1−a3−a4)V3v - \tau_{1,3}v - \tau_{2,3}v = (a_{1} - a_{3} - a_{4})V_{3}

This implies a4=a1−a3a_{4} = a_{1} - a_{3} and

v=(a1,−a1,a3,a1−a3,−a1+a3,−a3).v = (a_{1},-a_{1},a_{3},a_{1}-a_{3},-a_{1}+a_{3},-a_{3}).

The family {v,τ1,2v}\{v,\tau_{1,2}v\} is a basis of FvF_{v}

In this case we consider the vector u=v+τ1,2vu = v + \tau_{1,2}v. This vector is in FvF_{v} then FuF_{u} is a subspace of FvF_{v}. If dim⁡Fu=1\dim F_{u} = 1, then Fu=FV3F_{u} = F_{V_{3}} or FW3F_{W_{3}} and FvF_{v} is not irreducible. Then Fu=FvF_{u} = F_{v}. But {u,τ1,2u}\{u,\tau_{1,2}u\} is linearly dependent. So we return to the previous case.

Proposition 3. Let us consider the vectors

u1=(1,1,0,−1,−1,0),u2=(1,1,−1,0,0,−1).u_{1} = (1,1,0,-1,-1,0), \quad u_{2} = (1,1,-1,0,0,-1).

and

u3=(1,−1,0,1,−1,0),u4=(1,−1,1,0,0,−1).u_{3} = (1,-1,0,1,-1,0), \quad u_{4} = (1,-1,1,0,0,-1).

Then the subspaces FuiF_{u_{i}} are 2-dimensional irreducible subspaces.

Theorem 4. We have the following direct decomposition of K[Σ3]\mathbb{K}[\Sigma_{3}] in irreducible subspaces

K[Σ3]=FV3⊕FW3⊕Fui⊕Fuj,i≠j.\mathbb{K}[\Sigma_{3}] = F_{V_{3}} \oplus F_{W_{3}} \oplus F_{u_{i}} \oplus F_{u_{j}}, \quad i \ne j.

In fact the vectors V3,W3,ui,τ13ui,uj,τ13uj, (i≠j)V_{3}, W_{3}, u_{i}, \tau_{13}u_{i}, u_{j}, \tau_{13}u_{j}, \ (i \ne j) are linearly independent.

Decomposition of a trilinear form

Let φ\varphi be a pp-linear form on EE and v=∑aiσi∈K[Σp]v=\sum a_i\sigma_i\in\mathbb{K}[\Sigma_p]. We define the pp-linear form φv\varphi_v by

φv(x1,…,xp)=∑aiφ(xσi(1),…,xσi(p)).\varphi_v(x_1,\ldots,x_p)=\sum a_i\varphi(x_{\sigma_i(1)},\ldots,x_{\sigma_i(p)}).

For example, the symmetric and alternating part of φ\varphi correspond to

φs=φVp,φa=φWp.\varphi_s=\varphi_{V_p},\quad\varphi_a=\varphi_{W_p}.

In particular, if φ\varphi is a bilinear form, the decomposition in symmetric and alternating part corresponds to

φ=12(φV2+φW2)\varphi=\frac{1}{2}(\varphi_{V_2}+\varphi_{W_2})

and it is associated with the decomposition

R[Σ2]=FV2⊕FW2\mathbb{R}[\Sigma_2]=F_{V_2}\oplus F_{W_2}

in irreducible subspaces. We generalize this remark to the trilinear case.

Decomposition of trilinear maps

We consider the following basis of R[Σ3]\mathbb{R}[\Sigma_3]

B={V3,W3,u1,u2,u3,u4}.\mathcal{B}=\{V_3,W_3,u_1,u_2,u_3,u_4\}.

Lemma 5. We have

6Id=V3+W3+u1+u2+u3+u4.6Id=V_3+W_3+u_1+u_2+u_3+u_4.

Proof. In fact

V3=Id−τ1,2−τ1,3−τ2,3+c+c2=(1,−1,−1,−1,1,1),W3=Id+τ1,2+τ1,3+τ2,3+c+c2=(1,1,1,1,1,1),u1=Id+τ1,2−τ2,3−c=(1,1,0,−1,−1,0),u2=Id+τ1,2−τ1,3−c2=(1,1,−1,0,0,−1),u3=Id−τ1,2+τ2,3−c=(1,−1,0,1,−1,0),u4=Id−τ1,2+τ1,3−c2=(1,−1,1,0,0,−1)\begin{aligned} V_3&=Id-\tau_{1,2}-\tau_{1,3}-\tau_{2,3}+c+c^2=(1,-1,-1,-1,1,1),\\ W_3&=Id+\tau_{1,2}+\tau_{1,3}+\tau_{2,3}+c+c^2=(1,1,1,1,1,1),\\ u_1&=Id+\tau_{1,2}-\tau_{2,3}-c=(1,1,0,-1,-1,0),\\ u_2&=Id+\tau_{1,2}-\tau_{1,3}-c^2=(1,1,-1,0,0,-1),\\ u_3&=Id-\tau_{1,2}+\tau_{2,3}-c=(1,-1,0,1,-1,0),\\ u_4&=Id-\tau_{1,2}+\tau_{1,3}-c^2=(1,-1,1,0,0,-1) \end{aligned}

We deduce the identity.

We define the polarisation of a trilinear form

Definition 6. Let φ\varphi be a real trilinear form. The polarization of φ\varphi is the decomposition

φ=16(φa+φs+φu1+φu2+φu3+φu4)\varphi=\frac{1}{6}(\varphi_a+\varphi_s+\varphi_{u_1}+\varphi_{u_2}+\varphi_{u_3}+\varphi_{u_4})

So we have

φa(x,y,z)=φ(x,y,z)−φ(y,x,z)−φ(z,y,x)−φ(x,z,y)+φ(y,z,x)+φ(z,x,y),φs(x,y,z)=φ(x,y,z)+φ(y,x,z)+φ(z,y,x)+φ(x,z,y)+φ(y,z,x)+φ(z,x,y),φu1(x,y,z)=φ(x,y,z)+φ(y,x,z)−φ(x,z,y)−φ(y,z,x),φu2(x,y,z)=φ(x,y,z)+φ(y,x,z)−φ(z,y,x)−φ(z,x,y),φu3(x,y,z)=φ(x,y,z)−φ(y,x,z)+φ(x,z,y)−φ(y,z,x),φu4(x,y,z)=φ(x,y,z)−φ(y,x,z)+φ(z,y,x)−φ(z,x,y).\begin{aligned} \varphi_a(x,y,z)&=\varphi(x,y,z)-\varphi(y,x,z)-\varphi(z,y,x)-\varphi(x,z,y)+\varphi(y,z,x)+\varphi(z,x,y),\\ \varphi_s(x,y,z)&=\varphi(x,y,z)+\varphi(y,x,z)+\varphi(z,y,x)+\varphi(x,z,y)+\varphi(y,z,x)+\varphi(z,x,y),\\ \varphi_{u_1}(x,y,z)&=\varphi(x,y,z)+\varphi(y,x,z)-\varphi(x,z,y)-\varphi(y,z,x),\\ \varphi_{u_2}(x,y,z)&=\varphi(x,y,z)+\varphi(y,x,z)-\varphi(z,y,x)-\varphi(z,x,y),\\ \varphi_{u_3}(x,y,z)&=\varphi(x,y,z)-\varphi(y,x,z)+\varphi(x,z,y)-\varphi(y,z,x),\\ \varphi_{u_4}(x,y,z)&=\varphi(x,y,z)-\varphi(y,x,z)+\varphi(z,y,x)-\varphi(z,x,y). \end{aligned}

Proposition 7. The trilinear forms φu1\varphi_{u_1} and φu2\varphi_{u_2} satisfy the following properties:

φui(x,y,z)−φui(y,x,z)=0,φui(x,y,z)+φui(z,y,x)+φui(x,z,y)=0.\varphi_{u_i}(x,y,z)-\varphi_{u_i}(y,x,z)=0,\qquad\varphi_{u_i}(x,y,z)+\varphi_{u_i}(z,y,x)+\varphi_{u_i}(x,z,y)=0.

The trilinear forms φu3\varphi_{u_3} and φu4\varphi_{u_4} satisfy the following properties:

φui(x,y,z)+φui(y,x,z)=0,φui(x,y,z)−φui(z,y,x)−φui(x,z,y)=0.\varphi_{u_i}(x,y,z)+\varphi_{u_i}(y,x,z)=0,\qquad\varphi_{u_i}(x,y,z)-\varphi_{u_i}(z,y,x)-\varphi_{u_i}(x,z,y)=0.

Polynomial expression

We assume E=KnE = \mathbb{K}^n. Let φ\varphi be a trilinear form on EE. If

x1=(x11,x12,⋯ ,x1n), x2=(x21,x22,⋯ ,x2n), x3=(x31,x32,⋯ ,x3n)x_1 = (x_1^1,x_1^2,\cdots,x_1^n),\ x_2 = (x_2^1,x_2^2,\cdots,x_2^n),\ x_3 = (x_3^1,x_3^2,\cdots,x_3^n)

are three vectors of EE, then

φ(x1,x2,x3)=∑ai,j,kx1ix2jx3k\varphi(x_1,x_2,x_3) = \sum a_{i,j,k}x_1^i x_2^j x_3^k

is identified with an homogeneous polynomial expression PφP_\varphi of degree 33 in the variables xjix_j^i.

  1. If φ=φs\varphi= \varphi_s is symmetric, then PφP_\varphi is a symmetric polynomial. But any symmetric polynomial in n3n^3 variables x11,x12,⋯ ,x1n,x21,x22,⋯ ,x2n,x31,x32,⋯ ,x3nx_1^1,x_1^2,\cdots,x_1^n,x_2^1,x_2^2,\cdots,x_2^n,x_3^1,x_3^2,\cdots,x_3^n is representable in a unique way as a polynomial in the elementary symmetric polynomials. But PφP_\varphi is homogeneous of degree 33. The previous decomposition is quite trivial but we can refine this decomposition.

Examples

i) n=2n = 2 We have the decomposition

φ(x1,x2,x3)=a1x11x21x31+a2x12x22x32+a3(x11x21x32+x11x22x31+x12x21x31)+a4(x11x22x32+x12x22x31+x12x21x32).\varphi(x_1,x_2,x_3) = a_1x_1^1x_2^1x_3^1+a_2x_1^2x_2^2x_3^2+a_3(x_1^1x_2^1x_3^2+x_1^1x_2^2x_3^1+x_1^2x_2^1x_3^1)+a_4(x_1^1x_2^2x_3^2+x_1^2x_2^2x_3^1+x_1^2x_2^1x_3^2).

ii) n=3n = 3, we have

Proposition 8. Any symmetric trilinear form on EE is a linear combination of the following symmetric forms:

(1) φs,1,2,3(x1,x2,x3)=∑i≠j≠kx1ix2jx3k\varphi_{s,1,2,3}(x_1,x_2,x_3)=\sum_{i\ne j\ne k}x_1^i x_2^j x_3^k

(2) φs,i,j(x1,x2,x3)=x1ix2ix3j+x2ix3ix1j+x1ix3ix2j,i≠j\varphi_{s,i,j}(x_1,x_2,x_3)=x_1^i x_2^i x_3^j+x_2^i x_3^i x_1^j+x_1^i x_3^i x_2^j,\quad i\ne j

(3) φs,i(x1,x2,x3)=x1ix2ix3i,i=1,2,3\varphi_{s,i}(x_1,x_2,x_3)=x_1^i x_2^i x_3^i,\quad i=1,2,3.

  1. If φ\varphi is a trilinear alternating form, we have the decomposition

φ(x1,x2,x3)=∑1≤i<j<k≤naijk∑σ∈Σ3ϵ(σ)x1σ(i)x2σ(j)x3σ(k)\varphi(x_1,x_2,x_3)=\sum_{1\le i<j<k\le n}a_{ijk}\sum_{\sigma\in\Sigma_3}\epsilon(\sigma)x_1^{\sigma(i)}x_2^{\sigma(j)}x_3^{\sigma(k)}

For example, if n=3n=3 then φ\varphi is colinear to the trilinear form corresponding to the determinant, that is

φa(x1,x2,x3)=det⁡(x1,x2,x3)=∑σ∈Σ3ϵ(σ)x1σ(1)x2σ(2)x3σ(3).\varphi_a(x_1,x_2,x_3)=\det(x_1,x_2,x_3)=\sum_{\sigma\in\Sigma_3}\epsilon(\sigma)x_1^{\sigma(1)}x_2^{\sigma(2)}x_3^{\sigma(3)}.

From now, we assume that n=3n=3. In this case we have φ(x1,x2,x3)=Pφ\varphi(x_1,x_2,x_3)=P_\varphi is a polynomial in variables x11,x21,x31,x12,x22,x32,x13,x23,x33x_1^1,x_2^1,x_3^1,x_1^2,x_2^2,x_3^2,x_1^3,x_2^3,x_3^3.

  1. We saw that the form φu1\varphi_{u_1} satisfies the following symmetry relations:

φu1(x1,x2,x3)−φu1(x2,x1,x3)=0,\varphi_{u_1}(x_1,x_2,x_3)-\varphi_{u_1}(x_2,x_1,x_3)=0,
φu1(x1,x2,x3)+φu1(x3,x2,x1)+φu1(x1,x3,x2)=0.\varphi_{u_1}(x_1,x_2,x_3)+\varphi_{u_1}(x_3,x_2,x_1)+\varphi_{u_1}(x_1,x_3,x_2)=0.

We have

Proposition 9. The polynomial associated with φu1\varphi_{u_1} or φu2\varphi_{u_2} is a linear combination of the following polynomials

{2x1ix2ix3j−x1ix2jx3i−x1jx2ix3i,i≠j,x11x22x33+x12x23x31−x13x21x32−x11x23x32,x12x23x31+x13x21x32−x13x22x31−x11x22x33.\left\{ \begin{aligned} &2x_1^i x_2^i x_3^j-x_1^i x_2^j x_3^i-x_1^j x_2^i x_3^i,\quad i\ne j,\\ &x_1^1x_2^2x_3^3+x_1^2x_2^3x_3^1-x_1^3x_2^1x_3^2-x_1^1x_2^3x_3^2,\\ &x_1^2x_2^3x_3^1+x_1^3x_2^1x_3^2-x_1^3x_2^2x_3^1-x_1^1x_2^2x_3^3. \end{aligned} \right.
  1. The form φu3\varphi_{u3} (and φu4\varphi_{u4}) satisfies the following relations:

φu3(x1,x2,x3)+φu3(x2,x1,x3)=0,\varphi_{u3}(x_1,x_2,x_3)+\varphi_{u3}(x_2,x_1,x_3)=0,
φu3(x1,x2,x3)−φu3(x3,x2,x1)−φu3(x1,x3,x2)=0.\varphi_{u3}(x_1,x_2,x_3)-\varphi_{u3}(x_3,x_2,x_1)-\varphi_{u3}(x_1,x_3,x_2)=0.

We have

Proposition 10. The polynomial associated with φu3\varphi_{u3} or φu4\varphi_{u4} is a linear combination of the following polynomials

{x1ix2jx3i−x1jx2ix3ii≠j,x11x22x33−x12x21x33+x11x23x32−x13x21x32,x12x23x31−x13x22x31+x11x23x32−x13x21x32.\begin{cases} x_1^i x_2^j x_3^i-x_1^j x_2^i x_3^i & i\ne j,\\ x_1^1x_2^2x_3^3-x_1^2x_2^1x_3^3+x_1^1x_2^3x_3^2-x_1^3x_2^1x_3^2,\\ x_1^2x_2^3x_3^1-x_1^3x_2^2x_3^1+x_1^1x_2^3x_3^2-x_1^3x_2^1x_3^2. \end{cases}

Nonassociative algebras. A decomposition of the associator

The multiplication μ\mu of an algebra AA over K\mathbb{K} is given by a bilinear map

μ:A×A→A.\mu:A\times A\to A.

Its associator Aμ\mathcal{A}_{\mu} is the trilinear map given by Aμ(x,y,z)=μ(μ(x,y),z)−μ(x,μ(y,z))\mathcal{A}_{\mu}(x,y,z)=\mu(\mu(x,y),z)-\mu(x,\mu(y,z)). For simplification, we will denote xyxy instead of μ(x,y)\mu(x,y) and refer to the associator as A\mathcal{A}. The algebra (A,μ)(A,\mu) is associative when A=0\mathcal{A}=0. It is nonassociative if A\mathcal{A} is null or not null. As with trilinear forms, for any vector v=∑aiσi∈K[Σ3]v=\sum a_i\sigma_i\in\mathbb{K}[\Sigma_3], we define the trilinear map Av\mathcal{A}_v by

Av(x1,x2,x3)=∑aiA(xσi(1),xσi(2),xσi(3)).\mathcal{A}_v(x_1,x_2,x_3)=\sum a_i\mathcal{A}(x_{\sigma_i(1)},x_{\sigma_i(2)},x_{\sigma_i(3)}).

According to the previous section, we have

A=16(AV3+AW3+Au1+Au2+Au3+Au4).\mathcal{A}=\frac{1}{6}\left(\mathcal{A}_{V_3}+\mathcal{A}_{W_3}+\mathcal{A}_{u1}+\mathcal{A}_{u2}+\mathcal{A}_{u3}+\mathcal{A}_{u4}\right).

This decomposition highlights the following classes of nonassociative algebras:

  1. Totally symmetric algebras. The associator AV3\mathcal{A}_{V_3} satisfies the relations

AV3(x,y,z)=AV3(y,x,z)=AV3(z,y,x).\mathcal{A}_{V_3}(x,y,z)=\mathcal{A}_{V_3}(y,x,z)=\mathcal{A}_{V_3}(z,y,x).

If (A,μ)(A,\mu) is a nonassociative algebra whose associator corresponds to AV3\mathcal{A}_{V_3}, it satisfies the previous symmetric relations and it is a nonassociative totally symmetric algebra.

Remark: If we consider a nonassociative algebra whose associator A\mathcal{A} satisfies

AV3=0,\mathcal{A}_{V_3}=0,

that is,

A(x1,x2,x3)+A(x2,x1,x3)+A(x3,x2,x1)+A(x1,x3,x2)+A(x2,x3,x1)+A(x3,x1,x2)=0\mathcal{A}(x_1,x_2,x_3)+\mathcal{A}(x_2,x_1,x_3)+\mathcal{A}(x_3,x_2,x_1)+\mathcal{A}(x_1,x_3,x_2)+\mathcal{A}(x_2,x_3,x_1)+\mathcal{A}(x_3,x_1,x_2)=0

it is a 3-power associative algebras.

  1. Alternative algebras. The associator AW3\mathcal{A}_{W_3} satisfies the relations

AW3(x,y,z)=−AW3(y,x,z)=−AW3(z,y,x).\mathcal{A}_{W_3}(x,y,z)=-\mathcal{A}_{W_3}(y,x,z)=-\mathcal{A}_{W_3}(z,y,x).

If (A,μ)(A,\mu) is a nonassociative algebra whose associator corresponds to AW3\mathcal{A}_{W_3}, it satisfies the previous symmetric relations and it is an alternative algebra.

Remark: If we consider a nonassociative algebra whose associator A\mathcal{A} satisfies

AW3=0\mathcal{A}_{W_3}=0

that is

A(x1,x2,x3)−A(x2,x1,x3)−A(x3,x2,x1)−A(x1,x3,x2)+A(x2,x3,x1)+A(x3,x1,x2)=0\mathcal{A}(x_1,x_2,x_3)-\mathcal{A}(x_2,x_1,x_3)-\mathcal{A}(x_3,x_2,x_1)-\mathcal{A}(x_1,x_3,x_2)+\mathcal{A}(x_2,x_3,x_1)+\mathcal{A}(x_3,x_1,x_2)=0

it is a Lie-admissible algebra.

Numerous studies have focused on these two classes. See for example [4].

Left-symmetric 3-power associative algebras

Since

Au1(x1,x2,x3)=A(x1,x2,x3)+A(x2,x1,x3)−A(x1,x3,x2)−A(x2,x3,x1)\mathcal{A}_{u_1}(x_1,x_2,x_3)=\mathcal{A}(x_1,x_2,x_3)+\mathcal{A}(x_2,x_1,x_3)-\mathcal{A}(x_1,x_3,x_2)-\mathcal{A}(x_2,x_3,x_1)

this trilinear map satisfies

{Au1(x1,x2,x3)−Au1(x2,x1,x3)=0,Au1(x1,x2,x3)+Au1(x3,x2,x1)+Au1(x1,x3,x2)=0\begin{cases} \mathcal{A}_{u_1}(x_1,x_2,x_3)-\mathcal{A}_{u_1}(x_2,x_1,x_3)=0,\\ \mathcal{A}_{u_1}(x_1,x_2,x_3)+\mathcal{A}_{u_1}(x_3,x_2,x_1)+\mathcal{A}_{u_1}(x_1,x_3,x_2)=0 \end{cases}

An algebra A whose associator A\mathcal{A} satisfies the first relation is left symmetric which implies that it is also Lie-admissible and it is 3-power associative from the second relation.

Remark. Let us consider the class of nonassociative algebras whose associator A\mathcal{A} satisfies

Au1=0\mathcal{A}_{u_1}=0

that is

A(x1,x2,x3)+A(x2,x1,x3)−A(x1,x3,x2)−A(x2,x3,x1)=0.\mathcal{A}(x_1,x_2,x_3)+\mathcal{A}(x_2,x_1,x_3)-\mathcal{A}(x_1,x_3,x_2)-\mathcal{A}(x_2,x_3,x_1)=0.

This class does not seem to have been the subject of particular studies. However, let us note that this class contains the class of right-symmetric algebras.

Left-alternative Lie-admissible algebra

We have

Au3(x1,x2,x3)=A(x1,x2,x3)−A(x2,x1,x3)+A(x1,x3,x2)−A(x2,x3,x1)\mathcal{A}_{u_3}(x_1,x_2,x_3)=\mathcal{A}(x_1,x_2,x_3)-\mathcal{A}(x_2,x_1,x_3)+\mathcal{A}(x_1,x_3,x_2)-\mathcal{A}(x_2,x_3,x_1)

these trilinear map satisfies

{Au3(x1,x2,x3)+Au3(x2,x1,x3)=0,Au3(x1,x2,x3)−Au3(x3,x2,x1)−Au3(x1,x3,x2)=0\begin{cases} \mathcal{A}_{u_3}(x_1,x_2,x_3)+\mathcal{A}_{u_3}(x_2,x_1,x_3)=0,\\ \mathcal{A}_{u_3}(x_1,x_2,x_3)-\mathcal{A}_{u_3}(x_3,x_2,x_1)-\mathcal{A}_{u_3}(x_1,x_3,x_2)=0 \end{cases}

The first relation shows that the algebra is left-alternative which implies that it is 3-power-associative. The second identity implies that this algebra is Lie-admissible.

Remark. If we consider a nonassociave algebra whose associator A\mathcal{A} satisfies

Au3=0,\mathcal{A}_{u_3}=0,

that is,

A(x1,x2,x3)−A(x2,x1,x3)+A(x1,x3,x2)−A(x2,x3,x1)=0.\mathcal{A}(x_1,x_2,x_3)-\mathcal{A}(x_2,x_1,x_3)+\mathcal{A}(x_1,x_3,x_2)-\mathcal{A}(x_2,x_3,x_1)=0.

This class of nonassociative algebras contains the class of right-alternative algebras that is satisfying

A(x1,x2,x3)+A(x1,x3,x2).\mathcal{A}(x_1,x_2,x_3)+\mathcal{A}(x_1,x_3,x_2).

On the reduction of cubic forms

Symmetric trilinear forms and cubic forms

The decomposition of a trilinear form focuses on the study of antisymmetric, symmetric forms and those defined by vectors u1u_1 and u2u_2. Let EE be a finite-dimensional K\mathbb{K}-vector space. The classification of alternating trilinear forms is already presented in [3]. The study of bilinear symmetric forms is associated with the study of quadratic forms.

Recall that a quadratic form is a map q:E→Kq:E\to\mathbb{K} such that q(av)=a2q(v)q(av)=a^2q(v) for all a∈Ka\in\mathbb{K}, v∈Ev\in E and the function f:E×E→Kf:E\times E\to\mathbb{K} defined by f(u,v)=12(q(u+v)−q(u)−q(v))f(u,v)=\frac{1}{2}(q(u+v)-q(u)-q(v)) is a bilinear form on EE. Such a map can be identified to a homogeneous polynomial of degree 2 in nn variables with coefficients in K\mathbb{K}. A major result specifies that any quadratic form can be written as a linear combination of squares of linearly independent linear forms. What about the cubic forms? We assume now that K=C\mathbb{K}=\mathbb{C}. The problem that naturally arises is that of the decomposition of cubic forms.

A cubic form is a homogeneous polynomial of degree 3 in nn variables: For example

  • binary cubic form: q(x,y)=ax3+bx2y+cxy2+dy3q(x,y)=ax^3+bx^2y+cxy^2+dy^3

  • ternary cubic form: q(x,y,z)=ax3+bx2y+cx2z+dxy2+exz2+fxyz+gy3+hy2z+kyz2+lz3q(x,y,z)=ax^3+bx^2y+cx^2z+dxy^2+exz^2+fxyz+gy^3+hy^2z+kyz^2+lz^3

with a,b,⋯ ,l∈Ka,b,\cdots,l\in\mathbb{K}.

If φ\varphi is a trilinear form on the K\mathbb{K} vector space EE, the associated cubic form is

qφ(u)=φ(u,u,u).q_\varphi(u)=\varphi(u,u,u).

It is clear that if φ\varphi is skewsymmetric then qφ=0q_\varphi=0. But from Proposition 7, it is the same if φ=φui\varphi=\varphi_{u_i} for i∈{1,2,3,4}i\in\{1,2,3,4\}. We deduce that the cubic forms are in a one-to-one correspondence with the symmetric trilinear forms.

The next section concerns the reduction of binary cubic forms, that is when EE is 2-dimensional.

Decomposition of a binary cubic form

In this paragraph, we are interested in the decomposition of a binary cubic form into a sum of cubes of linear forms.

Proposition 11. There are binary cubic forms that cannot be written as a sum of cubes of independent linear forms.

Proof. Let us consider, for example, the form

q(x,y)=x3+x2y.q(x,y)=x^3+x^2y.

Assume that

q(x,y)=(α1x+β1y)3+(α2x+β2y)3.q(x,y)=(\alpha_1x+\beta_1y)^3+(\alpha_2x+\beta_2y)^3.

This is equivalent to the algebraic system

{α13+α23=13α12β1+3α22β2=13α1β12+3α2β22=0β13+β23=0\begin{cases} \alpha_1^3+\alpha_2^3=1\\ 3\alpha_1^2\beta_1+3\alpha_2^2\beta_2=1\\ 3\alpha_1\beta_1^2+3\alpha_2\beta_2^2=0\\ \beta_1^3+\beta_2^3=0 \end{cases}

Let us note that β2=0\beta_2=0 implies β1=0\beta_1=0 and α12β1+α22β2=0\alpha_1^2\beta_1+\alpha_2^2\beta_2=0 which is impossible. Then β2≠0\beta_2\ne0 (and also α1,α2,β1\alpha_1,\alpha_2,\beta_1). We deduce

α2=−α1β12β22\alpha_2=-\frac{\alpha_1\beta_1^2}{\beta_2^2}

and

3(α12β1+α22β2)=3(α12β1+α12β14β24β2)=3α12β1(1+β13β23)=3α12β1(β23+β13β23)=0≠13(\alpha_1^2\beta_1+\alpha_2^2\beta_2)=3\left(\alpha_1^2\beta_1+\frac{\alpha_1^2\beta_1^4}{\beta_2^4}\beta_2\right)=3\alpha_1^2\beta_1\left(1+\frac{\beta_1^3}{\beta_2^3}\right)=3\alpha_1^2\beta_1\left(\frac{\beta_2^3+\beta_1^3}{\beta_2^3}\right)=0\ne1

so this algebraic system has no solution.

□\square

Since there are binary cubic forms that are the sum of cubes of linear forms, we can wonder about the structure of such forms that are written in this way. Let q(x,y)=ax3+bx2y+cxy2+dy3q(x,y)=ax^3+bx^2y+cxy^2+dy^3 a binary cubic form. When a≠0a\ne0, a classical change of variables x=X−b3ayx=X-\frac{b}{3a}y leads to aX3+pXy2+qy3aX^3+pXy^2+qy^3 with p=c−b23ap=c-\frac{b^2}{3a}, q=d−bc3a+2b327a2q=d-\frac{bc}{3a}+\frac{2b^3}{27a^2} so allows to reduce the problem for a≠0a\ne0 to write as the sum of cubes of linear forms the binary cubic forms

q(x,y)=ax3+cxy2+dy3 with a≠0q(x,y)=ax^3+cxy^2+dy^3\text{ with }a\ne0

Assume that

q(x,y)=(α1x+β1y)3+(α2x+β2y)3.q(x,y)=(\alpha_1x+\beta_1y)^3+(\alpha_2x+\beta_2y)^3.

Then we have

{α13+α23=a3α12β1+3α22β2=03α1β12+3α2β22=cβ13+β23=d\begin{cases} \alpha_1^3+\alpha_2^3=a\\ 3\alpha_1^2\beta_1+3\alpha_2^2\beta_2=0\\ 3\alpha_1\beta_1^2+3\alpha_2\beta_2^2=c\\ \beta_1^3+\beta_2^3=d \end{cases}

If c=0c=0 then qq is trivially a sum of linear forms. If one of the coefficients αi,βi\alpha_i,\beta_i is zero, then c=0c=0 and qq is indeed a sum of cubes of independent linear forms (two cubes if d≠0d\ne0). Then we assume that α1α2β1β2≠0\alpha_1\alpha_2\beta_1\beta_2\ne0. Then

α1=cα233aβ12,β2=−α12α22β1,β1=−c2α249a2β13\alpha_1=\frac{c\alpha_2^3}{3a\beta_1^2},\qquad\beta_2=-\frac{\alpha_1^2}{\alpha_2^2}\beta_1,\qquad\beta_1=-\frac{c^2\alpha_2^4}{9a^2\beta_1^3}

We put u=α2u=\alpha_2, v=α2β1v=\frac{\alpha_2}{\beta_1}. Then

α1=c3auv2,β2=−c29a2uv3,β1=uv.\alpha_1=\frac{c}{3a}uv^2,\qquad\beta_2=-\frac{c^2}{9a^2}uv^3,\qquad\beta_1=\frac{u}{v}.

The equations α13+α23=a\alpha_1^3+\alpha_2^3=a, β13+β23=d\beta_1^3+\beta_2^3=d are equivalent to

u3(1+c327a3v6)=a,u^3\left(1+\frac{c^3}{27a^3}v^6\right)=a,
u3(1−c693a6v12)=dv3.u^3\left(1-\frac{c^6}{9^3a^6}v^{12}\right)=dv^3.

We deduce

1−c327a3v6=dav3.1-\frac{c^3}{27a^3}v^6=\frac{d}{a}v^3.

We put V=v3V=v^3. Then VV is a root of

c327a3V2+daV−1=0.\frac{c^3}{27a^3}V^2+\frac{d}{a}V-1=0.

The discriminant is Δ=27ad2+4c327a3\Delta= \frac{27ad^{2}+4c^{3}}{27a^{3}}. We deduce

V=v3=27a32c3(−da±Δ)V=v^{3}=\frac{27a^{3}}{2c^{3}}\left(-\frac{d}{a}\pm\sqrt{\Delta}\right)

and

u3(2−daV)=a.u^{3}\left(2-\frac{d}{a}V\right)=a.

Since we have assumed that a≠0a\ne0, the system has a solution if and only if

4c3+27ad2≠04c^{3}+27ad^{2}\ne0

and this condition is equivalent to Δ≠0\Delta\ne0.

We can now calculate the coefficients αi\alpha_i and βi\beta_i. We have

v3=27a32c3(−da+εΔ),u3=a22a−dv3v^{3}=\frac{27a^{3}}{2c^{3}}\left(-\frac{d}{a}+\varepsilon\sqrt{\Delta}\right),\qquad u^{3}=\frac{a^{2}}{2a-dv^{3}}

We deduce

u3=2c327a(aΔ−εdΔ)u^{3}=\frac{2c^{3}}{27a\left(a\Delta-\varepsilon d\sqrt{\Delta}\right)}

with ε2=1\varepsilon^{2}=1, and

u=c3(2a2(aΔ−εdΔ))13,v=3ac(εaΔ−d2a)13.u=\frac{c}{3}\left(\frac{2}{a^{2}\left(a\Delta-\varepsilon d\sqrt{\Delta}\right)}\right)^{\frac{1}{3}},\qquad v=\frac{3a}{c}\left(\frac{\varepsilon a\sqrt{\Delta}-d}{2a}\right)^{\frac{1}{3}}.

We check that the equations α13+α23=a\alpha_{1}^{3}+\alpha_{2}^{3}=a and β13+β23=d\beta_{1}^{3}+\beta_{2}^{3}=d are indeed satisfied. Let’s look at the second equation. The first one is identical. We have

α2=u,β1=uv,α1=c3auv2,β2=−c29a2uv3.\alpha_{2}=u,\quad\beta_{1}=\frac{u}{v},\quad\alpha_{1}=\frac{c}{3a}uv^{2},\quad\beta_{2}=-\frac{c^{2}}{9a^{2}}uv^{3}.

We have

β13+β23=(uv)3[1−c693a6v12].\beta_{1}^{3}+\beta_{2}^{3}=\left(\frac{u}{v}\right)^{3}\left[1-\frac{c^{6}}{9^{3}a^{6}}v^{12}\right].

We have seen that

c327a3v6+dav3−1=0.\frac{c^{3}}{27a^{3}}v^{6}+\frac{d}{a}v^{3}-1=0.

We deduce

v6=27a3c3−27a2dc3v3.v^{6}=\frac{27a^{3}}{c^{3}}-\frac{27a^{2}d}{c^{3}}v^{3}.

This implies

v12=272a6ac6[ac3+27a2d2−(27ad3+2dc3)v3].v^{12}=\frac{27^{2}a^{6}}{ac^{6}}\left[ac^{3}+27a^{2}d^{2}-(27ad^{3}+2dc^{3})v^{3}\right].

We put

ρ=ac3+27a2d2−(27ad3+2dc3)v3\rho=ac^{3}+27a^{2}d^{2}-(27ad^{3}+2dc^{3})v^{3}

and β13+β23=d\beta_{1}^{3}+\beta_{2}^{3}=d is equivalent to

u3(1−ρac3)=dv3.u^{3}\left(1-\frac{\rho}{ac^{3}}\right)=dv^{3}.

But u3=a22a−dv3u^3=\frac{a^2}{2a-dv^3}. Then we have to verify that

a2(1−ρac3)=2adv3−d2v6=2adv3−d2(27a3c3−27a2dc3v3).a^2\left(1-\frac{\rho}{ac^3}\right)=2adv^3-d^2v^6=2adv^3-d^2\left(\frac{27a^3}{c^3}-\frac{27a^2d}{c^3}v^3\right).

This is equivalent to

ac3−ρ=(2dc3+27ad3)v3−27a2d2.ac^3-\rho=(2dc^3+27ad^3)v^3-27a^2d^2.

And we find again the expression of ρ\rho and the equation β13+β23=d\beta_1^3+\beta_2^3=d is satisfied.

Theorem 12. Let q(x,y)=ax3+cxy2+dy3q(x,y)=ax^3+cxy^2+dy^3 be a ternary form with a≠0a\ne0.

If c=0c=0, the binary cubic form q(x,y)=ax3+dy3q(x,y)=ax^3+dy^3 writes trivially as a sum of two linear forms.

If c≠0c\ne0 we can write qq as a sum of the cube of two linear forms if and only if

δ=4c3+27ad2≠0\delta=4c^3+27ad^2\ne0

and we have

q(x,y)=(α1x+β1y)3+(α2x+β2y)3q(x,y)=(\alpha_1x+\beta_1y)^3+(\alpha_2x+\beta_2y)^3

with

α2=u,β1=uv,α1=c3auv2,β2=−c29a2uv3\alpha_2=u,\quad\beta_1=\frac{u}{v},\quad\alpha_1=\frac{c}{3a}uv^2,\quad\beta_2=-\frac{c^2}{9a^2}uv^3

and

u=c3(2a2(aΔ−dΔ))13,v=3ac(−d2a+Δ2)13u=\frac{c}{3}\left(\frac{2}{a^2(a\Delta-d\sqrt{\Delta})}\right)^{\frac{1}{3}},\quad v=\frac{3a}{c}\left(-\frac{d}{2a}+\frac{\sqrt{\Delta}}{2}\right)^{\frac{1}{3}}

with Δ=δ27a3\Delta=\frac{\delta}{27a^3}.

It remains to examine the case q(x,y)=bx2y+cxy2+dy3q(x,y)=bx^2y+cxy^2+dy^3, that is a=0a=0 in the initial case. If d≠0d\ne0 the study is similar to the previous case exchanging the role of xx and yy. We can therefore only study the particular case

q(x,y)=bx2y+cxy2.q(x,y)=bx^2y+cxy^2.

If bc=0bc=0, then qq cannot be written as a sum of the cube of linear forms. Assume bc≠0bc\ne0. Since α13+α23=β13+β23=0\alpha_1^3+\alpha_2^3=\beta_1^3+\beta_2^3=0, we can choose α1=k1α2, β1=k2β2\alpha_1=k_1\alpha_2,\ \beta_1=k_2\beta_2 with k13=k23=−1k_1^3=k_2^3=-1 and 1+k12k2≠01+k_1^2k_2\ne0 and 1+k1k22≠01+k_1k_2^2\ne0. Let’s take, for example, k1=−1,k2=−jk_1=-1,k_2=-j. In this case 1+k12k2=1−j1+k_1^2k_2=1-j and 1+k1k22=2+j1+k_1k_2^2=2+j. We deduce

α23=b2(1+k1k22)3c(1+k12k2)2=b2(1+2j)9c,β23=−(1+2j)c23b.\alpha_2^3=\frac{b^2(1+k_1k_2^2)}{3c(1+k_1^2k_2)^2}=\frac{b^2(1+2j)}{9c},\quad\beta_2^3=-\frac{(1+2j)c^2}{3b}.

Proposition 13. If bc≠0bc\ne0 then

q(x,y)=bx2y+cxy2q(x,y)=bx^2y+cxy^2
=[−(b2(1+2j)9c)13x+j2((1+2j)c23b)13y]3+[(b2(1+2j)9c)13x−j((1+2j)c23b)13y]3.=\left[-\left(\frac{b^2(1+2j)}{9c}\right)^{\frac{1}{3}}x+j^2\left(\frac{(1+2j)c^2}{3b}\right)^{\frac{1}{3}}y\right]^3+\left[\left(\frac{b^2(1+2j)}{9c}\right)^{\frac{1}{3}}x-j\left(\frac{(1+2j)c^2}{3b}\right)^{\frac{1}{3}}y\right]^3.

Example. We saw that q(x,y)=x3+x2yq(x,y)=x^3+x^2y could not decompose like this. The change of coordinates x1=x−13y,y=yx_1=x-\frac{1}{3}y,y=y allows to write in canonical form

q(x1,y)=x13−13x1y2+227y3q(x_1,y)=x_1^3-\frac{1}{3}x_1y^2+\frac{2}{27}y^3

and Δ=0\Delta= 0. But in this case, although qq does not decompose as a sum of cubes, we have the decomposition

q(x1,y)=(x1−13y)3+(x1−13y)2y.q(x_1,y)=\left(x_1-\frac{1}{3}y\right)^3+\left(x_1-\frac{1}{3}y\right)^2y.

We will see that when δ=27ad2+4c3=0\delta=27ad^2+4c^3=0, qq will decompose in this form, that is q(x,y)=(α1x+β1y)3+(α1x+β1y)2(α2x+β2y)q(x,y)=(\alpha_1x+\beta_1y)^3+(\alpha_1x+\beta_1y)^2(\alpha_2x+\beta_2y).

Proposition 14. Let be q(x,y)=ax3+cxy2+dy3q(x,y)=ax^3+cxy^2+dy^3 with a≠0a\ne0 and δ=27ad2+4c3=0\delta=27ad^2+4c^3=0. Then we have the following decomposition

q(x,y)=(α1x+β1y)3+(α1x+β1y)2(α2x+β2y).q(x,y)=(\alpha_1x+\beta_1y)^3+(\alpha_1x+\beta_1y)^2(\alpha_2x+\beta_2y).

Proof. The algebraic system associated with this decomposition is

{α13+α12α2=a,3α12β1+α12β2+2α1α2β1=0,3α1β12+2α1β1β2+α2β12=c,β13+β12β2=d.\begin{cases} \alpha_1^3+\alpha_1^2\alpha_2=a,\\ 3\alpha_1^2\beta_1+\alpha_1^2\beta_2+2\alpha_1\alpha_2\beta_1=0,\\ 3\alpha_1\beta_1^2+2\alpha_1\beta_1\beta_2+\alpha_2\beta_1^2=c,\\ \beta_1^3+\beta_1^2\beta_2=d. \end{cases}

Let us note that if c=0c=0, d=0d=0 and qq is already decomposed. Then c≠0c\ne0. In this case the coefficients α1,β1\alpha_1,\beta_1 are non zero. We deduce

α2=a−α13α12,β1=3d2cα1,β2=4c29dα12−3d2cα1.\alpha_2=\frac{a-\alpha_1^3}{\alpha_1^2},\qquad\beta_1=\frac{3d}{2c}\alpha_1,\qquad\beta_2=\frac{4c^2}{9d\alpha_1^2}-\frac{3d}{2c}\alpha_1.

In particular, if α1=−a13\alpha_1=-a^{\frac{1}{3}}, then α2=2a13\alpha_2=2a^{\frac{1}{3}} and β1=−3d2ca13=β2\beta_1=-\frac{3d}{2c}a^{\frac{1}{3}}=\beta_2.

Corollary 15. Consider q(x,y)=ax3+cxy2+dy3q(x,y)=ax^3+cxy^2+dy^3 with a≠0a\ne0 and 27ad2+4c3=027ad^2+4c^3=0. Then we have the following decomposition

q(x,y)=(−a13x−3d2ca13y)3+(−a13x−3d2ca13y)2(2a13x−3d2ca13y).q(x,y)=\left(-a^{\frac{1}{3}}x-\frac{3d}{2c}a^{\frac{1}{3}}y\right)^3+\left(-a^{\frac{1}{3}}x-\frac{3d}{2c}a^{\frac{1}{3}}y\right)^2\left(2a^{\frac{1}{3}}x-\frac{3d}{2c}a^{\frac{1}{3}}y\right).

This is equivalent to

q(x,y)=−a[(x−3d2cy)3−(x−3d2cy)2(2x−3d2ca13y)].q(x,y)=-a\left[\left(x-\frac{3d}{2c}y\right)^3-\left(x-\frac{3d}{2c}y\right)^2\left(2x-\frac{3d}{2c}a^{\frac{1}{3}}y\right)\right].

Roots of the equation ax3+bx2+cx+d=0ax^3+bx^2+cx+d=0

To find the roots of an equation of degree 33, we have only to consider the case a≠0a\ne0. Using the previous decompositions, we can also consider b=0b=0, that is, ax3+cx+d=0ax^3+cx+d=0. If c(4c3+27ad2)=0c(4c^3+27ad^2)=0, the roots are easy to find. If c(4c3+27ad2)≠0c(4c^3+27ad^2)\ne0, we have

q(x,1)=ax3+cx+d=(α1x+β1)3+(α2x+β2)3q(x,1)=ax^3+cx+d=(\alpha_1x+\beta_1)^3+(\alpha_2x+\beta_2)^3

with

α2=u,β1=uv,α1=c3auv2,β2=−c29a2uv3.\alpha_2=u,\qquad\beta_1=\frac{u}{v},\qquad\alpha_1=\frac{c}{3a}uv^2,\qquad\beta_2=-\frac{c^2}{9a^2}uv^3.

and

u=c3(2a2(aΔ−dΔ))13,v=3ac(−d2a+Δ2)13u=\frac{c}{3}\left(\frac{2}{a^{2}(a\Delta-d\sqrt{\Delta})}\right)^{\frac{1}{3}},\quad v=\frac{3a}{c}\left(-\frac{d}{2a}+\frac{\sqrt{\Delta}}{2}\right)^{\frac{1}{3}}

We deduce that

x1=−β1+β2α1+α2x_{1}=-\frac{\beta_{1}+\beta_{2}}{\alpha_{1}+\alpha_{2}}

is a root of the equation ax3+cx+d=0ax^{3}+cx+d=0.

Theorem 16. (Tartaglia’s roots) If c(4c3+27ad2)≠0c(4c^{3}+27ad^{2})\ne0 and a≠0a\ne0 then

x1=−1v+c3avx_{1}=-\frac{1}{v}+\frac{c}{3a}v

with

v=3ac(−d+Δ2a)13v=\frac{3a}{c}\left(\frac{-d+\sqrt{\Delta}}{2a}\right)^{\frac{1}{3}}

is a root of ax3+cx+d=0ax^{3}+cx+d=0.

Proof.

q(x,y)=(α1x+β1)3+(α2x+β2)3q(x,y)=(\alpha_{1}x+\beta_{1})^{3}+(\alpha_{2}x+\beta_{2})^{3}

with α2=u\alpha_{2}=u, β1=uv\beta_{1}=\frac{u}{v}, α1=c3auv2\alpha_{1}=\frac{c}{3a}uv^{2}, β2=−c29a2uv3\beta_{2}=-\frac{c^{2}}{9a^{2}}uv^{3}. We have a root

x1=−β1+β2α1+α2=−1v+c3av.x_{1}=-\frac{\beta_{1}+\beta_{2}}{\alpha_{1}+\alpha_{2}}=-\frac{1}{v}+\frac{c}{3a}v.

▢

Remarks. If Δ>0\Delta>0, then v∈Rv\in\mathbb{R}. We have seen that

v3=27a32c3(−da±Δ)v^{3}=\frac{27a^{3}}{2c^{3}}\left(-\frac{d}{a}\pm\sqrt{\Delta}\right)

To find a root of ax3+cx+d=0ax^{3}+cx+d=0 we have considered that ±=+\pm=+. If v1v_{1} corresponds to: ±=+\pm=+ and v2v_{2} to ±=−\pm=-, then we have c3av2=−1v1\frac{c}{3a}v_{2}=-\frac{1}{v_{1}} and we find the same root.

If Δ<0\Delta<0 and x1∈Rx_{1}\in\mathbb{R}, then we find again the previous conditions. If x1∈C−Rx_{1}\in\mathbb{C}-\mathbb{R}, then we find the conjugate root.

On the equations of degree 5

In the same vein, we conclude this work by examining what this degree 3 method can induce for degree 5 equations.

Let us consider a binary pentatic form

f(x,y)=ax5+bx4y+cx3y2+dx2y3+exy4+fy5.f(x,y)=ax^{5}+bx^{4}y+cx^{3}y^{2}+dx^{2}y^{3}+exy^{4}+fy^{5}.

As for the binary cubic case, if a≠0a\ne0, we can always consider that b=0b=0. We moreover assume that e=0e=0, that is

f(x,y)=ax5+cx3y2+dx2y3+fy5.f(x,y)=ax^{5}+cx^{3}y^{2}+dx^{2}y^{3}+fy^{5}.

If ff is the sum of two 5-power linear equations:

f(x,y)=(α1x+β1y)5+(α2x+β2y)5,f(x,y)=(\alpha_{1}x+\beta_{1}y)^{5}+(\alpha_{2}x+\beta_{2}y)^{5},

we have the following algebraic system

{α15+α25=a,α14β1+α24β2=0,α13β12+α23β22=c10,α12β13+α22β23=d10,α1β14+α2β24=0,β15+β25=f.\left\{ \begin{aligned} \alpha_1^5+\alpha_2^5&=a,\\ \alpha_1^4\beta_1+\alpha_2^4\beta_2&=0,\\ \alpha_1^3\beta_1^2+\alpha_2^3\beta_2^2&=\frac{c}{10},\\ \alpha_1^2\beta_1^3+\alpha_2^2\beta_2^3&=\frac{d}{10},\\ \alpha_1\beta_1^4+\alpha_2\beta_2^4&=0,\\ \beta_1^5+\beta_2^5&=f. \end{aligned} \right.

If (c,d)=(0,0)(c,d)=(0,0) it is trivial that ff is the sum of two 5-power linear equations. If (c,d)≠(0,0)(c,d)\ne(0,0), we have α1α2β1β2≠0\alpha_1\alpha_2\beta_1\beta_2\ne0. The second equation is equivalent to

β2=−β1(α1α2)4.\beta_2=-\beta_1\left(\frac{\alpha_1}{\alpha_2}\right)^4.

This implies

{α13β12+α23β22=α13β12aα15=c10,α12β13+α22β23=α12β13a(α25−α15)α210=d10,α115+α215=0,α15+α25=a,β15+β25=f.\left\{ \begin{aligned} \alpha_1^3\beta_1^2+\alpha_2^3\beta_2^2&=\alpha_1^3\beta_1^2\frac{a}{\alpha_1^5}=\frac{c}{10},\\ \alpha_1^2\beta_1^3+\alpha_2^2\beta_2^3&=\alpha_1^2\beta_1^3\frac{a(\alpha_2^5-\alpha_1^5)}{\alpha_2^{10}}=\frac{d}{10},\\ \alpha_1^{15}+\alpha_2^{15}&=0,\\ \alpha_1^5+\alpha_2^5&=a,\\ \beta_1^5+\beta_2^5&=f. \end{aligned} \right.

But α115+α215=(α15+α25)(α110−α15α25+α210)\alpha_1^{15}+\alpha_2^{15}=(\alpha_1^5+\alpha_2^5)(\alpha_1^{10}-\alpha_1^5\alpha_2^5+\alpha_2^{10}) implying α110−α15α25+α210=0\alpha_1^{10}-\alpha_1^5\alpha_2^5+\alpha_2^{10}=0 because a≠0a\ne0. In this case

(α1α2)5=−j or −j2.\left(\frac{\alpha_1}{\alpha_2}\right)^5=-j\ \text{or}\ -j^2.

When (α1α2)5=−j2\left(\frac{\alpha_1}{\alpha_2}\right)^5=-j^2 we have

α15=−j2a1−j2,α25=a1−j2\alpha_1^5=\frac{-j^2a}{1-j^2},\qquad\alpha_2^5=\frac{a}{1-j^2}

But (α1α2)5=−j\left(\frac{\alpha_1}{\alpha_2}\right)^5=-j implies

α25=a1−j=−j2a1−j2,α15=a1−j2\alpha_2^5=\frac{a}{1-j}=\frac{-j^2a}{1-j^2},\qquad\alpha_1^5=\frac{a}{1-j^2}

and all the equations are symmetric in α1,α2\alpha_1,\alpha_2 so we can consider (α1α2)5=−j2\left(\frac{\alpha_1}{\alpha_2}\right)^5=-j^2 and

α15=−j2a1−j2,α25=a1−j2,β15=f1−j2,β25=−j2f1−j2.\alpha_1^5=\frac{-j^2a}{1-j^2},\qquad\alpha_2^5=\frac{a}{1-j^2},\qquad\beta_1^5=\frac{f}{1-j^2},\qquad\beta_2^5=\frac{-j^2f}{1-j^2}.

We have

α1α2=−jξ,\frac{\alpha_1}{\alpha_2}=-j\xi,

where ξ\xi is a fifth root of the unity. This implies

α22β13=c10ξ3(j2−1),α22β13=d10ξ2(j2−1).\alpha_2^2\beta_1^3=\frac{c}{10\xi^3(j^2-1)},\qquad\alpha_2^2\beta_1^3=\frac{d}{10\xi^2(j^2-1)}.

We deduce

102af=cd10^2af=cd

and

c5=−105a3f2.c^5=-10^5a^3f^2.

Theorem 17. Let f(x,y)=ax5+cx3y2+dx2y3+fy5f(x,y)=ax^{5}+cx^{3}y^{2}+dx^{2}y^{3}+fy^{5} a penlinear form with

102af=cd, c5=−105a3f2.10^{2}af=cd,\ c^{5}=-10^{5}a^{3}f^{2}.

Then this form can be written as follows

f(x,y)=(α1x+β1y)5+(α2x+β2y)5f(x,y)=(\alpha_{1}x+\beta_{1}y)^{5}+(\alpha_{2}x+\beta_{2}y)^{5}

with

α1=(−j2a1−j2)15,α2=(a1−j2)15,β1=(f1−j2)15,β2=(−j2f1−j2)15.\alpha_{1}=\left(\frac{-j^{2}a}{1-j^{2}}\right)^{\frac{1}{5}},\quad\alpha_{2}=\left(\frac{a}{1-j^{2}}\right)^{\frac{1}{5}},\quad\beta_{1}=\left(\frac{f}{1-j^{2}}\right)^{\frac{1}{5}},\quad\beta_{2}=\left(\frac{-j^{2}f}{1-j^{2}}\right)^{\frac{1}{5}}.

Corollary 18. The equation of degree 5

ax5+cx3+dx2+f=0ax^{5}+cx^{3}+dx^{2}+f=0

with

102af=cd, c5=−105a3f2.10^{2}af=cd,\ c^{5}=-10^{5}a^{3}f^{2}.

admits the following root

x=−β1−β2α1+α2.x=\frac{-\beta_{1}-\beta_{2}}{\alpha_{1}+\alpha_{2}}.

This work was presented at the University of Cagliari in honor of the retirement of Professor Paola Piu.

Elisabeth Remm Université de Haute-Alsace, IRIMAS UR 7499, F-68100 Mulhouse, France.

Email address: [email protected]

Email address: [email protected]

References

  1. [1]Barreiro E., Benayadi S., Rizzo C. Nearly associative algebras. Journal of Algebra 658 (2024) 821–868.DOI
  2. [2]Boneh D., Silverberg A. Applications of multilinear forms to cryptography. Contemporary Mathematics 324, AMS (2003) 71–90.
  3. [3]Capdevielle B. Classification des formes trilinéaires alternées en dimension 6. L’Enseignement Mathéma-tique 18, 1972-1973DOI
  4. [4]Goze M., Remm E. A Class of Nonassociative Algebras. Algebra Colloquium 14 : 2 (2007) 313–326.

Paper details

vlnc:W_6pve5k3e
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents