MPSI · Représentation matricielle

Matrice d'une application linéaire

Matrice d'une application linéaire

Question

Une application linéaire uL(E,F)u\in\mathcal{L}(E,F) est entièrement déterminée par les images u(e1),,u(en)u(e_1),\dots,u(e_n) des vecteurs d'une base de EE : c'est le théorème de détermination d'une application linéaire. Or chacun de ces nn vecteurs de FF se code, dans une base de FF, par une colonne de pp scalaires. Que reste-t-il de uu si l'on range ces nn colonnes côte à côte ? Toute l'information, ou seulement une partie ?

Définition 1 : Matrice d'une application linéaire

Soient BE=(e1,,en)\mathcal{B}_E=(e_1,\dots,e_n) une base de EE, BF\mathcal{B}_F une base de FF et uL(E,F)u\in\mathcal{L}(E,F).

On appelle matrice de uu dans les bases BE\mathcal{B}_E et BF\mathcal{B}_F la matrice

MatBE,BF(u)=MatBF(u(e1),,u(en)).\operatorname{Mat}_{\mathcal{B}_E,\mathcal{B}_F}(u)=\operatorname{Mat}_{\mathcal{B}_F}\big(u(e_1),\dots,u(e_n)\big).

Remarque :

On pose BF=(f1,,fp)\mathcal{B}_F=(f_1,\dots,f_p). Pour tout j1,nj\in\llbracket 1,n\rrbracket, on pose u(ej)=i=1pai,jfi\displaystyle u(e_j)=\sum_{i=1}^{p} a_{i,j}\, f_i. Ainsi

MatBE,BF(u)=(a1,1a1,nap,1ap,n)Mp,n(K).\operatorname{Mat}_{\mathcal{B}_E,\mathcal{B}_F}(u)= \begin{pmatrix} a_{1,1} & \cdots & a_{1,n}\\ \vdots & & \vdots\\ a_{p,1} & \cdots & a_{p,n} \end{pmatrix}\in\mathcal{M}_{p,n}(\mathbb{K}).

Remarque :

Retenir la règle de lecture : une colonne par vecteur de la base de départ, une ligne par vecteur de la base d'arrivée. La matrice a donc p=dimFp=\dim F lignes et n=dimEn=\dim E colonnes — l'ordre des dimensions est inversé par rapport à l'écriture L(E,F)\mathcal{L}(E,F).

Exemple :

Pour u=idEu=\operatorname{id}_E et une base B\mathcal{B} de EE, on a MatB(idE)=In\operatorname{Mat}_{\mathcal{B}}(\operatorname{id}_E)=I_n, puisque idE(ej)=ej\operatorname{id}_E(e_j)=e_j pour tout jj. En revanche, si l'on prend deux bases différentes B\mathcal{B} et B\mathcal{B}' de EE, la matrice MatB,B(idE)\operatorname{Mat}_{\mathcal{B},\mathcal{B}'}(\operatorname{id}_E) n'est plus InI_n du tout. Que vaut-elle alors ?

Test 1 : Lecture des colonnes

La jj-ième colonne de MatBE,BF(u)\operatorname{Mat}_{\mathcal{B}_E,\mathcal{B}_F}(u) est formée des coordonnées de u(ej)u(e_j) dans la base BF\mathcal{B}_F.

Exercice 1

Déterminer les matrices, dans les bases canoniques, des applications linéaires suivantes :

  1. f:R3R2(x,y,z)(2xy+z,  xz)f : \begin{array}{ccl} \mathbb{R}^3 & \longrightarrow & \mathbb{R}^2\\ (x,y,z) & \longmapsto & (2x-y+z,\; x-z) \end{array}

  2. g:K3[X]K3[X]PP+XPg : \begin{array}{ccl} \mathbb{K}_3[X] & \longrightarrow & \mathbb{K}_3[X]\\ P & \longmapsto & P+XP' \end{array}

Solution :(cliquer pour afficher)
  1. On a MatB3,B2(f)=(211101)\operatorname{Mat}_{\mathcal{B}_3,\mathcal{B}_2}(f)=\begin{pmatrix} 2 & -1 & 1\\ 1 & 0 & -1 \end{pmatrix}. On remarque que :
MatB2(f(x,y,z))=(2xy+zxz)=(211101)(xyz)=MatB3,B2(f)(xyz).\begin{aligned} \operatorname{Mat}_{\mathcal{B}_2}\big(f(x,y,z)\big) &=\begin{pmatrix} 2x-y+z\\ x-z \end{pmatrix} =\begin{pmatrix} 2 & -1 & 1\\ 1 & 0 & -1 \end{pmatrix} \begin{pmatrix} x\\ y\\ z \end{pmatrix}\\[1mm] &=\operatorname{Mat}_{\mathcal{B}_3,\mathcal{B}_2}(f)\cdot\begin{pmatrix} x\\ y\\ z \end{pmatrix}. \end{aligned}
  1. On a g(1)=1g(1)=1, g(X)=2Xg(X)=2X, g(X2)=3X2g(X^2)=3X^2 et g(X3)=4X3g(X^3)=4X^3, donc
MatB,B(g)=(1000020000300004).\operatorname{Mat}_{\mathcal{B},\mathcal{B}}(g)= \begin{pmatrix} 1 & 0 & 0 & 0\\ 0 & 2 & 0 & 0\\ 0 & 0 & 3 & 0\\ 0 & 0 & 0 & 4 \end{pmatrix}.

Matrice d'un endomorphisme

Définition 2 : Matrice d'un endomorphisme

Soient B\mathcal{B} une base de EE et uL(E)u\in\mathcal{L}(E).

On appelle matrice de uu dans la base B\mathcal{B} la matrice de uu dans les bases B\mathcal{B} et B\mathcal{B}. On note

MatB(u)=MatB,B(u).\operatorname{Mat}_{\mathcal{B}}(u)=\operatorname{Mat}_{\mathcal{B},\mathcal{B}}(u).

Remarque :

  1. La matrice d'un endomorphisme dans une base est une matrice carrée.
  2. Si A=MatB(u)A=\operatorname{Mat}_{\mathcal{B}}(u), on dit que la matrice AA représente uu dans la base B\mathcal{B}.

Exercice 2

Soient (a,b)R2(a,b)\in\mathbb{R}^2 et φ:CCz(a+ib)z\varphi : \begin{array}{ccl} \mathbb{C} & \longrightarrow & \mathbb{C}\\ z & \longmapsto & (a+ib)\,z \end{array}.

Déterminer la matrice de φ\varphi dans la base canonique B=(1,i)\mathcal{B}=(1,i).

Solution :(cliquer pour afficher)

On a φ(1)=a+ib\varphi(1)=a+ib et φ(i)=b+ia\varphi(i)=-b+ia, ainsi

MatB(φ)=(abba).\operatorname{Mat}_{\mathcal{B}}(\varphi)=\begin{pmatrix} a & -b\\ b & a \end{pmatrix}.

Un dictionnaire entre applications linéaires et matrices

Question

On sait maintenant fabriquer une matrice à partir d'une application linéaire. Deux questions se posent aussitôt. Cette correspondance perd-elle de l'information : deux applications distinctes peuvent-elles avoir la même matrice ? Et toute matrice est-elle atteinte ? Autrement dit : le dictionnaire uAu\leftrightarrow A est-il un vrai dictionnaire, et respecte-t-il les opérations ?

Proposition 1 : Isomorphisme entre ℒ(E,F) et ℳₚ,ₙ(𝕂)

Soient BE\mathcal{B}_E une base de EE et BF\mathcal{B}_F une base de FF. Alors l'application

φ:L(E,F)Mp,n(K)uMatBE,BF(u)\varphi : \begin{array}{ccl} \mathcal{L}(E,F) & \longrightarrow & \mathcal{M}_{p,n}(\mathbb{K})\\ u & \longmapsto & \operatorname{Mat}_{\mathcal{B}_E,\mathcal{B}_F}(u) \end{array}

est un isomorphisme.

Démonstration :

  • Soient u,vL(E,F)u,v\in\mathcal{L}(E,F) et λK\lambda\in\mathbb{K}. On note BE=(e1,,en)\mathcal{B}_E=(e_1,\dots,e_n) et BF=(f1,,fp)\mathcal{B}_F=(f_1,\dots,f_p), et on pose
MatBE,BF(u)=(ai,j)1ip, 1jnetMatBE,BF(v)=(bi,j)1ip, 1jn.\operatorname{Mat}_{\mathcal{B}_E,\mathcal{B}_F}(u)=(a_{i,j})_{1\leqslant i\leqslant p,\ 1\leqslant j\leqslant n} \quad\text{et}\quad \operatorname{Mat}_{\mathcal{B}_E,\mathcal{B}_F}(v)=(b_{i,j})_{1\leqslant i\leqslant p,\ 1\leqslant j\leqslant n}.

Soit j1,nj\in\llbracket 1,n\rrbracket. On a

(u+λv)(ej)=u(ej)+λv(ej)=i=1pai,jfi+λi=1pbi,jfi=i=1p(ai,j+λbi,j)fi.(u+\lambda v)(e_j)=u(e_j)+\lambda\, v(e_j) =\sum_{i=1}^{p} a_{i,j}\, f_i+\lambda\sum_{i=1}^{p} b_{i,j}\, f_i =\sum_{i=1}^{p}\big(a_{i,j}+\lambda\, b_{i,j}\big)\, f_i.

Ainsi

MatBE,BF(u+λv)=(ai,j+λbi,j)=(ai,j)+λ(bi,j)=MatBE,BF(u)+λMatBE,BF(v),\operatorname{Mat}_{\mathcal{B}_E,\mathcal{B}_F}(u+\lambda v) =\big(a_{i,j}+\lambda\, b_{i,j}\big) =\big(a_{i,j}\big)+\lambda\,\big(b_{i,j}\big) =\operatorname{Mat}_{\mathcal{B}_E,\mathcal{B}_F}(u)+\lambda\,\operatorname{Mat}_{\mathcal{B}_E,\mathcal{B}_F}(v),

donc φ(u+λv)=φ(u)+λφ(v)\varphi(u+\lambda v)=\varphi(u)+\lambda\,\varphi(v). Alors φ\varphi est une application linéaire.

  • Soit uL(E,F)u\in\mathcal{L}(E,F) tel que φ(u)=0\varphi(u)=0. Alors u(ej)=0u(e_j)=0 pour tout j1,nj\in\llbracket 1,n\rrbracket, donc u=0u=0. Donc Kerφ={0L(E,F)}\operatorname{Ker}\varphi=\left\{0_{\mathcal{L}(E,F)}\right\}, d'où φ\varphi est injective.

  • Comme dimL(E,F)=np=dimMp,n(K)\dim\mathcal{L}(E,F)=np=\dim\mathcal{M}_{p,n}(\mathbb{K}), φ\varphi est un isomorphisme.

Remarque :

L'application φ:L(E,F)Mp,n(K)uMatBE,BF(u)\varphi : \begin{array}{ccl} \mathcal{L}(E,F) & \longrightarrow & \mathcal{M}_{p,n}(\mathbb{K})\\ u & \longmapsto & \operatorname{Mat}_{\mathcal{B}_E,\mathcal{B}_F}(u) \end{array} est un isomorphisme. En particulier :

  1. Pour tout MMp,n(K)M\in\mathcal{M}_{p,n}(\mathbb{K}), il existe une unique application vL(E,F)v\in\mathcal{L}(E,F) telle que MatBE,BF(v)=M\operatorname{Mat}_{\mathcal{B}_E,\mathcal{B}_F}(v)=M.
  2. Pour tous u,vL(E,F)u,v\in\mathcal{L}(E,F), si MatBE,BF(u)=MatBE,BF(v)\operatorname{Mat}_{\mathcal{B}_E,\mathcal{B}_F}(u)=\operatorname{Mat}_{\mathcal{B}_E,\mathcal{B}_F}(v), alors u=vu=v.

Test 2 : Même matrice, même application ?

Si u,vL(E,F)u,v\in\mathcal{L}(E,F) vérifient MatBE,BF(u)=MatBE,BF(v)\operatorname{Mat}_{\mathcal{B}_E,\mathcal{B}_F}(u)=\operatorname{Mat}_{\mathcal{B}'_E,\mathcal{B}'_F}(v) pour certaines bases BE,BE\mathcal{B}_E,\mathcal{B}'_E de EE et BF,BF\mathcal{B}_F,\mathcal{B}'_F de FF, alors u=vu=v.

Question

Le dictionnaire est établi, mais il serait sans intérêt s'il ne servait qu'à ranger des nombres. Le vrai enjeu est calculatoire : si AA code uu et si XX code xx, l'objet AXAX code-t-il quelque chose de reconnaissable ?

Proposition 2 : Traduction matricielle de l'égalité y = u(x)

Soient BE\mathcal{B}_E une base de EE, BF\mathcal{B}_F une base de FF et uL(E,F)u\in\mathcal{L}(E,F). Soit (x,y)E×F(x,y)\in E\times F.

On pose A=MatBE,BF(u)A=\operatorname{Mat}_{\mathcal{B}_E,\mathcal{B}_F}(u), X=MatBE(x)X=\operatorname{Mat}_{\mathcal{B}_E}(x) et Y=MatBF(y)Y=\operatorname{Mat}_{\mathcal{B}_F}(y). Alors :

Y=AX    y=u(x).Y=AX \iff y=u(x).

En particulier :

MatBF(u(x))=MatBE,BF(u)MatBE(x).\operatorname{Mat}_{\mathcal{B}_F}\big(u(x)\big)=\operatorname{Mat}_{\mathcal{B}_E,\mathcal{B}_F}(u)\cdot\operatorname{Mat}_{\mathcal{B}_E}(x).

Démonstration :

On note BE=(e1,,en)\mathcal{B}_E=(e_1,\dots,e_n) et BF=(f1,,fp)\mathcal{B}_F=(f_1,\dots,f_p). On pose

X=(x1xn),Y=(y1yp)etA=(ai,j)1ip, 1jn.X=\begin{pmatrix} x_1\\ \vdots\\ x_n \end{pmatrix},\qquad Y=\begin{pmatrix} y_1\\ \vdots\\ y_p \end{pmatrix} \qquad\text{et}\qquad A=(a_{i,j})_{1\leqslant i\leqslant p,\ 1\leqslant j\leqslant n}.

On a

Y=AX    i1,p,    yi=j=1nai,jxj    i=1pyifi=i=1p(j=1nai,jxj)fi    y=j=1n(i=1pai,jfi)xj    y=j=1nu(ej)xj    y=u(j=1nxjej)    y=u(x).\begin{aligned} Y=AX &\iff \forall i\in\llbracket 1,p\rrbracket,\;\; y_i=\sum_{j=1}^{n} a_{i,j}\, x_j\\ &\iff \sum_{i=1}^{p} y_i\, f_i=\sum_{i=1}^{p}\left(\sum_{j=1}^{n} a_{i,j}\, x_j\right) f_i\\ &\iff y=\sum_{j=1}^{n}\left(\sum_{i=1}^{p} a_{i,j}\, f_i\right) x_j\\ &\iff y=\sum_{j=1}^{n} u(e_j)\, x_j\\ &\iff y=u\left(\sum_{j=1}^{n} x_j\, e_j\right)\\ &\iff y=u(x). \end{aligned}

Rédaction — Calculer u(x) matriciellement :

Pour calculer l'image d'un vecteur xx par uu à l'aide d'une matrice :

Étape 1. Écrire A=MatBE,BF(u)A=\operatorname{Mat}_{\mathcal{B}_E,\mathcal{B}_F}(u) (une colonne par u(ej)u(e_j)).

Étape 2. Écrire X=MatBE(x)X=\operatorname{Mat}_{\mathcal{B}_E}(x) — attention, les coordonnées de xx dans BE\mathcal{B}_E, pas dans une autre base.

Étape 3. Calculer le produit Y=AXY=AX.

Étape 4. Relire YY comme une colonne de coordonnées dans BF\mathcal{B}_F : le vecteur u(x)u(x) est i=1pyifi\displaystyle\sum_{i=1}^{p} y_i f_i.

Les étapes 2 et 4 sont celles que l'on oublie quand les bases ne sont pas canoniques.

Exercice 3

Soit fL(R2)f\in\mathcal{L}(\mathbb{R}^2) dont la matrice dans les bases B=((1,1),(1,1))\mathcal{B}=\big((1,1),(1,-1)\big) au départ et C=(e1,e2)\mathcal{C}=(e_1,e_2) (base canonique) à l'arrivée est

A=(1203).A=\begin{pmatrix} 1 & 2\\ 0 & 3 \end{pmatrix}.

Calculer f(3,1)f(3,1).

Solution :(cliquer pour afficher)

Étape 2. On décompose x=(3,1)x=(3,1) dans B\mathcal{B}. On cherche (α,β)(\alpha,\beta) tel que α(1,1)+β(1,1)=(3,1)\alpha(1,1)+\beta(1,-1)=(3,1), c'est-à-dire

{α+β=3αβ=1    {α=2β=1\begin{cases} \alpha+\beta=3\\ \alpha-\beta=1 \end{cases} \iff \begin{cases} \alpha=2\\ \beta=1 \end{cases}

donc X=(21)X=\begin{pmatrix} 2\\ 1\end{pmatrix}.

Étape 3. Y=AX=(1203)(21)=(43)Y=AX=\begin{pmatrix} 1 & 2\\ 0 & 3\end{pmatrix}\begin{pmatrix} 2\\ 1\end{pmatrix}=\begin{pmatrix} 4\\ 3\end{pmatrix}.

Étape 4. La base d'arrivée est la base canonique, donc f(3,1)=(4,3)f(3,1)=(4,3).

Remarquons que le calcul aurait été faux si l'on avait posé X=(31)X=\begin{pmatrix} 3\\ 1\end{pmatrix} : la colonne d'entrée se lit toujours dans la base de départ de la matrice.

Matrice d'une composée

Question

L'ensemble L(E)\mathcal{L}(E) n'est pas seulement un espace vectoriel : c'est un anneau, dont la multiplication est la composition. Le dictionnaire précédent traduit l'addition et la multiplication externe. Traduit-il aussi la composition — et si oui, en quelle opération sur les matrices ?

Proposition 3 : Matrice d'une composée

Soient BE\mathcal{B}_E une base de EE, BF\mathcal{B}_F une base de FF, GG un K\mathbb{K}-espace vectoriel de dimension finie et BG\mathcal{B}_G une base de GG.

Soient uL(E,F)u\in\mathcal{L}(E,F) et vL(F,G)v\in\mathcal{L}(F,G). Alors :

MatBE,BG(vu)=MatBF,BG(v)MatBE,BF(u).\operatorname{Mat}_{\mathcal{B}_E,\mathcal{B}_G}(v\circ u)=\operatorname{Mat}_{\mathcal{B}_F,\mathcal{B}_G}(v)\cdot\operatorname{Mat}_{\mathcal{B}_E,\mathcal{B}_F}(u).

Démonstration :

On pose BE=(e1,,en)\mathcal{B}_E=(e_1,\dots,e_n), BF=(f1,,fp)\mathcal{B}_F=(f_1,\dots,f_p) et BG=(g1,,gq)\mathcal{B}_G=(g_1,\dots,g_q). Soit j1,nj\in\llbracket 1,n\rrbracket. On a

MatBG((vu)(ej))=MatBG(v(u(ej)))=MatBF,BG(v)MatBF(u(ej))=MatBF,BG(v)MatBE,BF(u)MatBE(ej).\begin{aligned} \operatorname{Mat}_{\mathcal{B}_G}\big((v\circ u)(e_j)\big) &=\operatorname{Mat}_{\mathcal{B}_G}\big(v(u(e_j))\big)\\ &=\operatorname{Mat}_{\mathcal{B}_F,\mathcal{B}_G}(v)\cdot\operatorname{Mat}_{\mathcal{B}_F}\big(u(e_j)\big)\\ &=\operatorname{Mat}_{\mathcal{B}_F,\mathcal{B}_G}(v)\cdot\operatorname{Mat}_{\mathcal{B}_E,\mathcal{B}_F}(u)\cdot\operatorname{Mat}_{\mathcal{B}_E}(e_j). \end{aligned}

Or MatBG((vu)(ej))=MatBE,BG(vu)MatBE(ej)\operatorname{Mat}_{\mathcal{B}_G}\big((v\circ u)(e_j)\big)=\operatorname{Mat}_{\mathcal{B}_E,\mathcal{B}_G}(v\circ u)\cdot\operatorname{Mat}_{\mathcal{B}_E}(e_j).

Ainsi la jj-ième colonne de la matrice MatBE,BG(vu)\operatorname{Mat}_{\mathcal{B}_E,\mathcal{B}_G}(v\circ u) est égale à la jj-ième colonne de la matrice MatBF,BG(v)MatBE,BF(u)\operatorname{Mat}_{\mathcal{B}_F,\mathcal{B}_G}(v)\cdot\operatorname{Mat}_{\mathcal{B}_E,\mathcal{B}_F}(u). D'où

MatBE,BG(vu)=MatBF,BG(v)MatBE,BF(u).\operatorname{Mat}_{\mathcal{B}_E,\mathcal{B}_G}(v\circ u)=\operatorname{Mat}_{\mathcal{B}_F,\mathcal{B}_G}(v)\cdot\operatorname{Mat}_{\mathcal{B}_E,\mathcal{B}_F}(u).

Remarque :

  1. La matrice MatBE(ej)\operatorname{Mat}_{\mathcal{B}_E}(e_j) est la jj-ième colonne de la matrice InI_n.
  2. La matrice AMatBE(ej)A\cdot\operatorname{Mat}_{\mathcal{B}_E}(e_j) est la jj-ième colonne de la matrice AA.

Test 3 : Ordre des facteurs

Pour uL(E,F)u\in\mathcal{L}(E,F) et vL(F,G)v\in\mathcal{L}(F,G), on a MatBE,BG(vu)=MatBE,BF(u)MatBF,BG(v)\operatorname{Mat}_{\mathcal{B}_E,\mathcal{B}_G}(v\circ u)=\operatorname{Mat}_{\mathcal{B}_E,\mathcal{B}_F}(u)\cdot\operatorname{Mat}_{\mathcal{B}_F,\mathcal{B}_G}(v).

Corollaire 1

Soit B\mathcal{B} une base de EE. Pour tous u,vL(E)u,v\in\mathcal{L}(E) :

MatB(vu)=MatB(v)MatB(u).\operatorname{Mat}_{\mathcal{B}}(v\circ u)=\operatorname{Mat}_{\mathcal{B}}(v)\cdot\operatorname{Mat}_{\mathcal{B}}(u).

Démonstration :

Il suffit d'appliquer la proposition précédente avec F=G=EF=G=E et BE=BF=BG=B\mathcal{B}_E=\mathcal{B}_F=\mathcal{B}_G=\mathcal{B} : elle donne

MatB,B(vu)=MatB,B(v)MatB,B(u),\operatorname{Mat}_{\mathcal{B},\mathcal{B}}(v\circ u)=\operatorname{Mat}_{\mathcal{B},\mathcal{B}}(v)\cdot\operatorname{Mat}_{\mathcal{B},\mathcal{B}}(u),

c'est-à-dire, avec la notation des endomorphismes, MatB(vu)=MatB(v)MatB(u)\operatorname{Mat}_{\mathcal{B}}(v\circ u)=\operatorname{Mat}_{\mathcal{B}}(v)\cdot\operatorname{Mat}_{\mathcal{B}}(u).

Remarque :

  1. L'application φ:L(E)Mn(K)uMatB(u)\varphi : \begin{array}{ccl} \mathcal{L}(E) & \longrightarrow & \mathcal{M}_{n}(\mathbb{K})\\ u & \longmapsto & \operatorname{Mat}_{\mathcal{B}}(u) \end{array} est un morphisme d'anneaux.
  2. MatB(idE)=In\operatorname{Mat}_{\mathcal{B}}(\operatorname{id}_E)=I_n.

Exercice 4

On considère

f:R3R2(x,y,z)(x+y,  yz)etg:R2R2(a,b)(2ab,  a+b)f : \begin{array}{ccl} \mathbb{R}^3 & \longrightarrow & \mathbb{R}^2\\ (x,y,z) & \longmapsto & (x+y,\; y-z) \end{array} \qquad\text{et}\qquad g : \begin{array}{ccl} \mathbb{R}^2 & \longrightarrow & \mathbb{R}^2\\ (a,b) & \longmapsto & (2a-b,\; a+b) \end{array}

Déterminer la matrice de gfg\circ f dans les bases canoniques, de deux façons : par calcul direct de gfg\circ f, puis par la formule de la composée.

Solution :(cliquer pour afficher)

Calcul direct. Pour tout (x,y,z)R3(x,y,z)\in\mathbb{R}^3 :

(gf)(x,y,z)=g(x+y,  yz)=(2(x+y)(yz),  (x+y)+(yz))=(2x+y+z,  x+2yz),(g\circ f)(x,y,z)=g(x+y,\;y-z)=\big(2(x+y)-(y-z),\;(x+y)+(y-z)\big)=(2x+y+z,\;x+2y-z),

d'où

MatB3,B2(gf)=(211121).\operatorname{Mat}_{\mathcal{B}_3,\mathcal{B}_2}(g\circ f)=\begin{pmatrix} 2 & 1 & 1\\ 1 & 2 & -1 \end{pmatrix}.

Par la formule. On a A=MatB3,B2(f)=(110011)A=\operatorname{Mat}_{\mathcal{B}_3,\mathcal{B}_2}(f)=\begin{pmatrix} 1 & 1 & 0\\ 0 & 1 & -1\end{pmatrix} et B=MatB2(g)=(2111)B=\operatorname{Mat}_{\mathcal{B}_2}(g)=\begin{pmatrix} 2 & -1\\ 1 & 1\end{pmatrix}. Alors

BA=(2111)(110011)=(211121).BA=\begin{pmatrix} 2 & -1\\ 1 & 1\end{pmatrix}\begin{pmatrix} 1 & 1 & 0\\ 0 & 1 & -1\end{pmatrix} =\begin{pmatrix} 2 & 1 & 1\\ 1 & 2 & -1 \end{pmatrix}.

Les deux méthodes coïncident. Noter que le produit ABAB, lui, n'est même pas défini.

Caractérisations matricielles

Question

La bijectivité de uu est une propriété abstraite : elle parle d'antécédents. L'inversibilité de AA est une propriété calculatoire : elle se teste par pivot de Gauss. Le dictionnaire échange-t-il ces deux notions ? Et si oui, sous quelle hypothèse sur les dimensions — car une matrice non carrée n'a aucune chance d'être inversible.

Proposition 4 : Caractérisation matricielle de la bijectivité

Supposons que EE et FF ont la même dimension. Soient BE\mathcal{B}_E une base de EE et BF\mathcal{B}_F une base de FF.

Une application uL(E,F)u\in\mathcal{L}(E,F) est bijective si et seulement si MatBE,BF(u)\operatorname{Mat}_{\mathcal{B}_E,\mathcal{B}_F}(u) est inversible. Dans ce cas :

(MatBE,BF(u))1=MatBF,BE(u1).\big(\operatorname{Mat}_{\mathcal{B}_E,\mathcal{B}_F}(u)\big)^{-1}=\operatorname{Mat}_{\mathcal{B}_F,\mathcal{B}_E}\big(u^{-1}\big).

Démonstration :

  • On suppose que uu est bijective. On a
MatBE,BF(u)MatBF,BE(u1)=MatBF,BF(uu1)=MatBF(idF)=I.\operatorname{Mat}_{\mathcal{B}_E,\mathcal{B}_F}(u)\cdot\operatorname{Mat}_{\mathcal{B}_F,\mathcal{B}_E}\big(u^{-1}\big) =\operatorname{Mat}_{\mathcal{B}_F,\mathcal{B}_F}\big(u\circ u^{-1}\big) =\operatorname{Mat}_{\mathcal{B}_F}(\operatorname{id}_F)=I.

De même, on a MatBF,BE(u1)MatBE,BF(u)=I\operatorname{Mat}_{\mathcal{B}_F,\mathcal{B}_E}\big(u^{-1}\big)\cdot\operatorname{Mat}_{\mathcal{B}_E,\mathcal{B}_F}(u)=I. Donc MatBE,BF(u)\operatorname{Mat}_{\mathcal{B}_E,\mathcal{B}_F}(u) est inversible et (MatBE,BF(u))1=MatBF,BE(u1)\big(\operatorname{Mat}_{\mathcal{B}_E,\mathcal{B}_F}(u)\big)^{-1}=\operatorname{Mat}_{\mathcal{B}_F,\mathcal{B}_E}\big(u^{-1}\big).

  • Réciproquement, on suppose que MatBE,BF(u)\operatorname{Mat}_{\mathcal{B}_E,\mathcal{B}_F}(u) est inversible. Soit vL(F,E)v\in\mathcal{L}(F,E) tel que MatBF,BE(v)=(MatBE,BF(u))1\operatorname{Mat}_{\mathcal{B}_F,\mathcal{B}_E}(v)=\big(\operatorname{Mat}_{\mathcal{B}_E,\mathcal{B}_F}(u)\big)^{-1}. On a MatBE,BF(u)MatBF,BE(v)=I\operatorname{Mat}_{\mathcal{B}_E,\mathcal{B}_F}(u)\cdot\operatorname{Mat}_{\mathcal{B}_F,\mathcal{B}_E}(v)=I, donc
MatBF,BF(uv)=MatBF(idF).\operatorname{Mat}_{\mathcal{B}_F,\mathcal{B}_F}(u\circ v)=\operatorname{Mat}_{\mathcal{B}_F}(\operatorname{id}_F).

Alors uv=idFu\circ v=\operatorname{id}_F. Or dimE=dimF\dim E=\dim F, donc uu est bijective et u1=vu^{-1}=v.

Test 4 : Bijectivité sans hypothèse sur les dimensions

Soient EE et FF deux espaces de dimensions finies quelconques, BE\mathcal{B}_E et BF\mathcal{B}_F des bases de EE et FF, et uL(E,F)u\in\mathcal{L}(E,F). Alors uu est bijective si et seulement si MatBE,BF(u)\operatorname{Mat}_{\mathcal{B}_E,\mathcal{B}_F}(u) est inversible.

Corollaire 2

Soient B\mathcal{B} une base de EE et uL(E)u\in\mathcal{L}(E). Alors uu est un isomorphisme si et seulement si MatB(u)\operatorname{Mat}_{\mathcal{B}}(u) est inversible, et on a :

MatB(u1)=(MatB(u))1.\operatorname{Mat}_{\mathcal{B}}\big(u^{-1}\big)=\big(\operatorname{Mat}_{\mathcal{B}}(u)\big)^{-1}.

Démonstration :

On applique la proposition précédente avec F=EF=E et BE=BF=B\mathcal{B}_E=\mathcal{B}_F=\mathcal{B} : l'hypothèse dimE=dimF\dim E=\dim F est trivialement vérifiée. On obtient que uu est bijective — c'est-à-dire, uu étant un endomorphisme, que uu est un isomorphisme de EE — si et seulement si MatB,B(u)=MatB(u)\operatorname{Mat}_{\mathcal{B},\mathcal{B}}(u)=\operatorname{Mat}_{\mathcal{B}}(u) est inversible, et dans ce cas

(MatB(u))1=MatB,B(u1)=MatB(u1).\big(\operatorname{Mat}_{\mathcal{B}}(u)\big)^{-1}=\operatorname{Mat}_{\mathcal{B},\mathcal{B}}\big(u^{-1}\big)=\operatorname{Mat}_{\mathcal{B}}\big(u^{-1}\big).

Exercice 5

Soit uL(R2)u\in\mathcal{L}(\mathbb{R}^2) défini par u(x,y)=(x+2y,  x+3y)u(x,y)=(x+2y,\;x+3y).

Montrer que uu est un automorphisme de R2\mathbb{R}^2 et déterminer u1u^{-1}.

Solution :(cliquer pour afficher)

Dans la base canonique B\mathcal{B}, on a u(e1)=(1,1)u(e_1)=(1,1) et u(e2)=(2,3)u(e_2)=(2,3), donc

A=MatB(u)=(1213).A=\operatorname{Mat}_{\mathcal{B}}(u)=\begin{pmatrix} 1 & 2\\ 1 & 3 \end{pmatrix}.

Cherchons l'inverse de AA par la méthode du pivot. Le système {x+2y=ax+3y=b\begin{cases} x+2y=a\\ x+3y=b\end{cases} donne, par L2L2L1L_2\leftarrow L_2-L_1, y=bay=b-a puis x=a2(ba)=3a2bx=a-2(b-a)=3a-2b. Ainsi AA est inversible et

A1=(3211).A^{-1}=\begin{pmatrix} 3 & -2\\ -1 & 1 \end{pmatrix}.

D'après le corollaire, uu est un automorphisme de R2\mathbb{R}^2 et MatB(u1)=A1\operatorname{Mat}_{\mathcal{B}}(u^{-1})=A^{-1}, c'est-à-dire

u1(x,y)=(3x2y,  x+y).u^{-1}(x,y)=(3x-2y,\;-x+y).

Vérification : u1(u(x,y))=(3(x+2y)2(x+3y),  (x+2y)+(x+3y))=(x,y)u^{-1}\big(u(x,y)\big)=\big(3(x+2y)-2(x+3y),\;-(x+2y)+(x+3y)\big)=(x,y).

Question

Décider qu'une famille de nn vecteurs est une base demande en principe deux vérifications : la liberté et le caractère générateur. En dimension finie, on sait déjà qu'une seule des deux suffit quand le cardinal est bon. Peut-on aller plus loin et remplacer ce travail par un simple calcul sur une matrice ?

Proposition 5 : Caractérisation matricielle des bases

Soient B\mathcal{B} une base de EE et F=(x1,,xn)\mathcal{F}=(x_1,\dots,x_n) une famille de vecteurs de EE de cardinal nn.

Alors F\mathcal{F} est une base de EE si et seulement si la matrice MatB(x1,,xn)\operatorname{Mat}_{\mathcal{B}}(x_1,\dots,x_n) est inversible.

Démonstration :

On pose B=(e1,,en)\mathcal{B}=(e_1,\dots,e_n). Soit uL(E)u\in\mathcal{L}(E) tel que u(ei)=xiu(e_i)=x_i pour tout i1,ni\in\llbracket 1,n\rrbracket. On a

MatB(u)=MatB(u(e1),,u(en))=MatB(x1,,xn).\operatorname{Mat}_{\mathcal{B}}(u)=\operatorname{Mat}_{\mathcal{B}}\big(u(e_1),\dots,u(e_n)\big)=\operatorname{Mat}_{\mathcal{B}}(x_1,\dots,x_n).

F\mathcal{F} est une base de EE si et seulement si uu est bijective, ce qui est équivalent à MatB(u)\operatorname{Mat}_{\mathcal{B}}(u) est inversible, c'est-à-dire MatB(x1,,xn)\operatorname{Mat}_{\mathcal{B}}(x_1,\dots,x_n) est inversible.

Test 5 : Famille libre et matrice inversible

Soient EE de dimension nn, B\mathcal{B} une base de EE et F\mathcal{F} une famille libre de vecteurs de EE. Alors MatB(F)\operatorname{Mat}_{\mathcal{B}}(\mathcal{F}) est inversible.

Exercice 6

Dans K2[X]\mathbb{K}_2[X], on pose P1=1+XP_1=1+X, P2=X+X2P_2=X+X^2 et P3=1+X2P_3=1+X^2.

Montrer que F=(P1,P2,P3)\mathcal{F}=(P_1,P_2,P_3) est une base de K2[X]\mathbb{K}_2[X].

Solution :(cliquer pour afficher)

La famille F\mathcal{F} est de cardinal 3=dimK2[X]3=\dim\mathbb{K}_2[X]. Sa matrice dans la base canonique B=(1,X,X2)\mathcal{B}=(1,X,X^2) est

M=MatB(F)=(101110011).M=\operatorname{Mat}_{\mathcal{B}}(\mathcal{F})=\begin{pmatrix} 1 & 0 & 1\\ 1 & 1 & 0\\ 0 & 1 & 1 \end{pmatrix}.

Échelonnons MM :

(101110011)L2L2L1(101011011)L3L3L2(101011002).\begin{pmatrix} 1 & 0 & 1\\ 1 & 1 & 0\\ 0 & 1 & 1 \end{pmatrix} \underset{L_2\leftarrow L_2-L_1}{\longrightarrow} \begin{pmatrix} 1 & 0 & 1\\ 0 & 1 & -1\\ 0 & 1 & 1 \end{pmatrix} \underset{L_3\leftarrow L_3-L_2}{\longrightarrow} \begin{pmatrix} 1 & 0 & 1\\ 0 & 1 & -1\\ 0 & 0 & 2 \end{pmatrix}.

La matrice échelonnée obtenue est triangulaire supérieure à coefficients diagonaux tous non nuls, donc MM est inversible. D'après la caractérisation matricielle des bases, F\mathcal{F} est une base de K2[X]\mathbb{K}_2[X].