Calcul Matriciel

T:RnRm, lineˊaire. Sa matrice : A(m×n),A=[a1an]x=(x1xn)T(x)=Ax=x1a1++xnanSi A=(a11a12a1nam1am2amn), on a ak=(a1kajkamk)Donc i{1,,m}(Ax)i=l=1nailxl (ieˋme composante de AxRm)T:\R^n\rightarrow\R^m\text{, linéaire. Sa matrice : } A(m\times n),\,A=[\vec{a}_1\cdots\vec{a}_n]\\ \vec{x} = \begin{pmatrix}x_1\\\vdots\\x_n\end{pmatrix}\mapsto T(\vec{x})=A\vec{x}=x_1\vec{a}_1+\ldots+x_n\vec{a}_n\\ \text{Si } A= \begin{pmatrix}a_{11}&a_{12}&\cdots&a_{1n}\\ \vdots&\vdots& &\vdots\\a_{m1}&a_{m2}&\cdots&a_{mn}\end{pmatrix} \text{, on a }\vec{a}_k = \begin{pmatrix}a_{1k}\\\vdots\\a_{jk}\\ \vdots\\a_{mk}\end{pmatrix}\\ \text{Donc }\forall i \in \{1,\ldots,m\}\,(A\vec{x})_i = \sum^n_{l=1}a_{il}x_l \text{ ($i^{ème}$ composante de $A\vec{x}\in\R^m$)}

Opérations sur les matrices

Addition

Si A(m×n),B(m×n),A\,(m\times n),\,B\,(m\times n), alors A+B(m×n)A+B\,(m\times n) est definie ainsi :

A=(aij)i=1,,mj=1,,nB=(bij)i=1,,m,j=1,,n(A+B)ijaij+bijA =(a_{ij})_{i = 1,\ldots,m\,j = 1,\ldots,n}\\B = (b_{ij})_{i = 1,\ldots,m,\,j = 1,\ldots,n}\\ (A+B)_{ij}\coloneqq a_{ij}+b_{ij}

Multiplication par un scalaire λR\lambda\in\R

λA(m×n)\lambda A\,(m\times n) est (λA)ijλaij(\lambda A)_{ij} \coloneqq\lambda a_{ij}

Produit matriciel

Image aˋ mettre p.90\color{red}\text{Image à mettre p.90}

ST:RnRkx(ST)(x)=S(T(x))\begin{array}{rlll} S\circ T : &\R^n &\longrightarrow &\R^k\\ & \vec{x}&\longmapsto &(S\circ T)(\vec{x}) = S(T(\vec{x})) \end{array}

Exemple

BAC(320101)(123410)=(91402)2×33×22×2\begin{array}{cccc} B&A& &C\\ \begin{pmatrix}3&2&0\\-1&0&1\end{pmatrix} &\begin{pmatrix}1&2\\3&4\\1&0\end{pmatrix}&= &\begin{pmatrix}9&14\\0&-2\end{pmatrix}\\ \color{red}2\times3&\color{red}3\times2&&\color{red}2\times 2 \end{array}

Remarque : on ne peut multiplier que des matrces qui ont les bonnes dimensions ?

BAk×mm×n\begin{array}{cc} B&A\\ \uparrow&\uparrow\\ k\times m & m\times n \end{array}

Propriétés

  1. A(BC)=(AB)CA(BC) = (AB)C

  2. A(B+C)=AB+ACA(B+C)=AB+AC

  3. (A+B)C=AC+BC(A+B)C=AC+BC

  4. A(λB)=(λA)B=λ(AB)A(\lambda B)=(\lambda A)B =\lambda (AB)

  5. ABBA en geˊneˊral!\color{red}AB\not=BA\text{ en général!}

    Par exemple:

    A=(1001),B=(0110)AB=(1001)(0110)=(0110)BA=(0110)(1001)=(0110)}ABBAA=\begin{pmatrix}1&0\\0&-1\end{pmatrix},\, B=\begin{pmatrix}0&-1\\1&0\end{pmatrix}\\ \left.\begin{array}{lll}AB =\begin{pmatrix}1&0\\0&-1\end{pmatrix} \begin{pmatrix}0&-1\\1&0\end{pmatrix} = \begin{pmatrix}0&-1\\-1&0\end{pmatrix}\\ BA = \begin{pmatrix}0&-1\\1&0\end{pmatrix} \begin{pmatrix}1&0\\0&-1\end{pmatrix} = \begin{pmatrix}0&1\\1&0\end{pmatrix}\end{array}\right\}AB\not=BA

Définition

Si AB=BAAB=BA, on dit que AA et BB commutent.

Transposition

Définition

Si A(m×n)A\,(m\times n), sa transposée est une matrice n×mn\times m, notée AA^\intercal, définie ainsi :

(A)ijAjiLes colonnes de A sont les lignes de A.Les lignes de A sont les colonnes A.(A^\intercal)_{ij}\coloneqq A_{ji}\\ \color{red}\text{Les colonnes de }A^\intercal\text{ sont les lignes de }A.\\ \color{red}\text{Les lignes de }A^\intercal\text{ sont les colonnes }A.

Exemple :

A=(123456)A=(135246)(A)31=5(A)13=5A = \begin{pmatrix}1&2\\3&4\\\textcolor{red}{5}&6\end{pmatrix} \longrightarrow A^\intercal = \begin{pmatrix}1&3&\textcolor{red}{5}\\ 2&4&6\end{pmatrix}\\ \text{} \\ \textcolor{red}{(A)_{31}=5}\longrightarrow\textcolor{red} {(A^\intercal)_{13}=5}

Propriétés:

Matrice inverse

Définition

Soit T:RnRmT: \R^n \rightarrow\R^m, linéaire, bijective (c'est-à-dire injective et surjective).

    \implies \exists une application réciproque, notée

T1:RmRn,telle que T1(T(x))=xxRnT(T1(y))=yyRm()T\begin{array}{rl} & \,\, T^{-1}: \R^m\longrightarrow\R^n,\text{telle que }\\ &\boxed{\begin{aligned}T^{-1}(T(\vec{x}))=\vec{x}\quad \forall\vec{x}\in\R^n\\ T(T^{-1}(\vec{y}))=\vec{y}\quad \forall\vec{y}\in\R^m\end{aligned}}\,(*)_T \end{array}

Lemme 1

T1:RmRnT^{-1}:\R^m\longrightarrow\R^n est aussi linéaire !

Lemme 2

Si T:\R^n\longrightarrow\R^m $ est linaire et bijective, alors $n=m.

Preuve

Soit A(m×n)A\,(m\times n) la matrice qui représente TT : T(x)=Ax.T(\vec{x})=A\vec{x}.

    n=m.\implies n=m. \quad \square

Donc pour étudier les applications linéaires et bijectives, on va considérer

T:RnRnsa matrice A sera carreˊe:n×nT:\textcolor{orange}{\R^n}\longrightarrow\textcolor{blue}{\R^n}\qquad \text{sa matrice A sera} \textbf{ carrée}:n\times n

Donc, T1:RnRnT^{-1} : \textcolor{blue}{\R^n} \longrightarrow\textcolor{orange}{\R^n} est linéaire (lemme 1)

    \implies elle peut être représentée à l'aide d'une matrice, que l'on note "A1A^{-1}" :

T1(x)=A1xT^{-1}(\vec{x}) = A^{-1}\vec{x}

A1A^{-1} est une n×nn\times n appelée l'inverse de AA. Elle est caractérosée par

()T    ()A:A1(Ax)=xxRnA(A1y)=yyRn(*)_T \iff (*)_A : \boxed{\begin{aligned}A^{-1}(A\vec{x})=\vec{x} \quad \forall\,\vec{x}\in\R^n\\A(A^{-1}\vec{y})=\vec{y}\quad\forall \,\vec{y}\in\R^n\end{aligned}}

Définition

La matrice identité de dimension nn est la matrice :

In=(11001)I_n=\begin{pmatrix} 1&&&&\\&1&&\text{\huge0}&\\&&\ddots&&\\ &\text{\huge0}&&\ddots&\\&&&&1\end{pmatrix}

Propriétés

  1. Inx=xxRnI_n\vec{x}=\vec{x}\quad\forall\vec{x}\in\R^n

  2. AIn=InA=A matrices A(n×n)AI_n=I_nA=A \quad\forall\text{ matrices }A\,(n\times n )

Donc ()A(*)_A devient : A1A=InAA1=In()\boxed{\begin{aligned}A^{-1}A=I_n\\AA^{-1}=I_n\end{aligned}}\quad(*)

Image aˋ mettre p.97\color{red}\text{Image à mettre p.97}

Donc, on cherche, pour une T(x)=AxT(\vec{x})=A\vec{x} bijective, la matrice A1A^{-1}, à l'aide de ()(*), telle que T1(y)=A1yT^{-1}(\vec{y})=A^{-1}\vec{y}.

A priori, trouver A1(n×n)A^{-1}\,(n\times n) est un problème à n2n^2 inconnues !!

Définition

Pour une matrice A(n×n)A\,(n\times n) donnée, si une A1(n×n)A^{-1}\,(n\times n) existe ((satisfaisant ())(*)), AA est dite inversible, et A1A^{-1} est son inverse. Sinon, AA est singulière.

Par exemple, la matrice nulle A=(0)n×n est singulieˋre.Par exemple A=(1234) est inversible. En effet, si A1=(213212) on veˊrifie que : AA1=(1001)=I2, etA1A=(1001)=I2\begin{array}{ll} \bullet&\text{Par exemple, la matrice nulle } A= \Big(\,0\,\Big)\leftarrow n\times n \text{ est singulière.}\\ \bullet&\text{Par exemple } A=\begin{pmatrix}1&2\\3&4 \end{pmatrix}\text{ est inversible.}\\ &\text{ En effet, si }A^{-1} =\begin{pmatrix} -2&1\\\frac{3}{2} &-\frac{1}{2}\end{pmatrix}\text{ on vérifie que : }\\ & AA^{-1}=\begin{pmatrix}1&0\\0&1\end{pmatrix}=I_2\text{, et}\\ & A^{-1}A =\begin{pmatrix}1&0\\0&1\end{pmatrix}=I_2 \end{array}

Lemme 3

Si AA est inversible, son inverse est unique.

Preuve

Supposons que AB=In et BA=In, et que AC=In et CA=InAlors B=BIn=B(AC)=(BA)C=InC=C\begin{array}{ll}\text{Supposons que }AB = I_n\text{ et }BA=I_n\text{, et que }AC=I_n\text{ et }CA=I_n&\\\text{Alors }B=BI_n=B(AC)=(BA)C=I_nC=C&\\&\square\end{array}

Propriétés de la matrice inverse (lorsqu'elle existe)

Etude de l'inversibilité pour les matrices 2×22\times2

Soit A=(abcd)\text{Soit }A=\begin{pmatrix}a&b\\c&d\end{pmatrix}

  1. Comment savoir si AA est inversible ?

  2. Si AA est inversible, comment calculer A1A^{-1} ?

Pour répondre à 1 :

Lemme

On a équivalence entre

a) AA est inversible

b)(ac)et(bd)sont lineˊaires indeˊpendants.\text{b)}\begin{pmatrix}a\\c\end{pmatrix} \text{et} \begin{pmatrix}b\\d\end{pmatrix}\text{sont linéaires indépendants.}

c) adbc0ad-bc\not=0

Peuve
a)    b) : OK, puisque on sait que T(x)=Ax est bijective si et seulement si les colonnes de A sont lineˊairement indeˊpendantes (injective) et engendrent R2 (surjective).b)     c) : Supposons que (ac) et (bd) soient lineˊairement indeˊpendants (en particulier, ces deux vecteurs ne sont pas (00)), et que adbc=0. Sans perte de geˊneˊraliteˊ, supposons que a0. On a donc d=bca et(bd)=(bbca)=ba(ac),une contradiction.c)     a) : Si adbc0, supposons que(ac) et(bd)sont colineˊaires :λR tel que(ac)=λ(bd)a=λbdc=λdbadbc=λbdλdb=0,une contradiction. \begin{aligned} \text{a)$\iff$b) :}&\text{ OK, puisque on sait que }T(\vec{x})=A \vec{x}\text{ est bijective si et seulement}\\ &\text{ si les colonnes de $A$ sont linéairement indépendantes (injective)}\\ &\text{ et engendrent $\R^2$ (surjective).}\\ \text{b)$\implies$ c) :}&\text{ Supposons que }\begin{pmatrix} a\\c\end{pmatrix}\text{ et }\begin{pmatrix}b\\d\end{pmatrix} \text{ soient linéairement indépendants}\\ &\text{ (en particulier, ces deux vecteurs ne sont pas }\begin{pmatrix}0\\0\end{pmatrix} \text{), et que }\\ &\,ad-bc=0\text{. Sans perte de généralité, supposons que } a\not=0.\\ &\text{ On a donc }d=\frac{bc}{a}\text{ et}\begin{pmatrix} b\\d\end{pmatrix}=\begin{pmatrix}b\\ \frac{bc}{a}\end{pmatrix}= \frac{b}{a}\begin{pmatrix}a\\c\end{pmatrix},\text{une contradiction.}\\ \text{c)$\implies$ a) :}&\text{ Si $ad-bc\not=0$, supposons que} \begin{pmatrix}a\\c\end{pmatrix}\text{ et}\begin{pmatrix}b\\d \end{pmatrix}\text{sont colinéaires :}\\ &\begin{aligned}\exists\lambda\in\R\text{ tel que}\begin{pmatrix}a\\c \end{pmatrix}=\lambda\begin{pmatrix}b\\d\end{pmatrix}\rightarrow& \underline{\begin{array}{rr|l}&a=\lambda b&\cdot\, d\\ -&c=\lambda d&\cdot\, b\end{array}}\\ & ad-bc=\lambda bd-\lambda db = 0,\\ &\text{une contradiction. } \\& &\square \end{aligned} \end{aligned}

Définition

Si A=(abcd),det(A)adbc est le deˊtermient de A.\text{Si }A = \begin{pmatrix}a&b\\c&d\end{pmatrix}, \mathrm{det}(A)\coloneqq ad-bc \text{ est le }\textbf{détermient }\text{de }A.

L'équivalence 1)$\iff$3) peut s'exprimer ainsi :

Théorème (n=2n=2)

A(2×2)A\,(2\times 2) est inversible     \iff det(A)0\mathrm{det}(A)\not=0

Si A=(abcd)est telle que det(A)=adbc0, calculons son inverse:\text{Si }A =\begin{pmatrix}a&b\\c&d\end{pmatrix}\text{est telle que }\mathrm{det}(A) = ad-bc\not=0\text{, calculons son inverse:}

Posons : A1=(αβγδ)On veut : AA1=I2=(1001)A1A=I2Systeˋme de n2=4 inconnues :α,β,γ,δ{aα+bγ=1(1)aβ+bδ=0(2)cα+dγ=0(3)cβ+dδ=1(4)(1) et (3)    (adbc0)α=d(2) et (4)    (adbc0)β=b    A1=(dadbcbadbccadbcaadbc)=1det(A)(dbca)\begin{aligned} \text{Posons : }&A^{-1} =\begin{pmatrix}\alpha & \beta \\ \gamma & \delta \end{pmatrix}\\ \text{On veut : }&AA^{-1}=I_2=\begin{pmatrix}1&0\\0&1\end{pmatrix}\\ & A^{-1}A=I_2\\ \begin{array}{l} \color{red}\text{Système de }\\ \color{red}\text{$n^2 = 4$ inconnues :}\\ \color{red}\text{$\alpha,\beta,\gamma,\delta$} \end{array}&\left\{\begin{array}{llllll} a\alpha& &+b\gamma& &=1&(1)\\ &a\beta& &+b\delta&=0&(2)\\ c\alpha& &+d\gamma& &=0&(3)\\ &c\beta& &+d\delta&=1&(4) \end{array}\right. \end{aligned}\\ (1)\text{ et } (3) \implies (\overbrace{ad-bc}^{\not=0})\alpha = d \qquad(2)\text{ et } (4) \implies (\overbrace{ad-bc}^{\not=0})\beta = -b \\ \text{} \\ \implies\boxed{A^{-1}=\begin{pmatrix}\frac{d}{ad-bc} &\frac{-b}{ad-bc}\\ \frac{-c}{ad-bc}&\frac{a}{ad-bc}\end{pmatrix}= \frac{1}{\mathrm{det}(A)}\begin{pmatrix}d&-b\\-c&a\end{pmatrix}}
Exemple
A=(1234)det(A)=20A inversible, et A1=12(4231)=(213212)A=\begin{pmatrix}1&2\\3&4\end{pmatrix}\rightarrow\mathrm{det}(A)=-2 \not=0\rightarrow A\text{ inversible, }\\ \text{et } A^{-1}=\frac{1}{-2}\begin{pmatrix}4&-2\\-3&1\end{pmatrix} =\begin{pmatrix}-2&1\\\frac{3}{2}&-\frac{1}{2}\end{pmatrix}
Exemples (Transformations du plan)
  1. Réflexion par OxOx

    TT est injective et surjective, donc bijective. Sa réciprique est T1=TT^{-1}=T

    Par rapport à la base canonique :

    A=(1001)det(A)=10, etA1=11(1001)==A\begin{aligned} &A=\begin{pmatrix}1&0\\0&-1\end{pmatrix}\rightarrow\mathrm{det}(A)=-1\not=0 \text{, et}\\ &A^{-1}=\frac{1}{-1}\begin{pmatrix}-1&0\\0&1 \end{pmatrix}=\ldots=A \end{aligned}

    Image aˋ mettre p.102\color{red}\text{Image à mettre p.102}

  2. Projection sur OyOy

    TT n'est ni injective, ni surjective, donc par bijective. T1\rightarrow T^{-1} n'existe pas.

    Veˊrification :A=(0001)det(A)=0\text{Vérification :} A=\begin{pmatrix}0&0\\0&1\end{pmatrix}\rightarrow\mathrm{det}(A)=0

    Image aˋ mettre p.102\color{red}\text{Image à mettre p.102}

Matrice inverse et résolution d'un système n×nn\times n

Soit ()(*) un système carré (n×n)(n\times n ) : A(n×n),xRn,bRnA\,(n\times n),\,\vec{x}\in\R^{n},\,\vec{b}\in\R^n

Ax=bA\vec{x}=\vec{b}

Si AA est inversible, A1A^{-1} existe, et ()(*) est équivalent à

A1(Ax)=A1bx=A1b\begin{aligned} A^{-1}(A\vec{x})&=A^{-1}\vec{b}\\ \vec{x}&=A^{-1}\vec{b} \end{aligned}

\rightarrow en particulier, la solution de ()(*) est unique, et est donnée par x=A1b\vec{x}=A^{-1}\vec{b}

Exemple
()(5314)(x1x2)=(132)Axbdet0A inversible : A1=123(4315)x=A1b=(21)(solution unique)\begin{array}{rcccc} (*)&\begin{pmatrix}5&-3\\1&4\end{pmatrix}&\begin{pmatrix} x_1\\x_2\end{pmatrix}&=&\begin{pmatrix}13\\-2\end{pmatrix}\\ &A&\vec{x}&&\vec{b} \end{array}\\ \begin{aligned} \mathrm{det}\not=0&\rightarrow A \text{ inversible : }A^{-1} =\frac{1}{23}\begin{pmatrix}4&3\\-1&5\end{pmatrix}\\ &\rightarrow \vec{x} =A^{-1}\vec{b}=\begin{pmatrix}2\\-1\end{pmatrix} \text{(solution unique)} \end{aligned}

Iversibilité dans le cas n×nn\times n

Remarque : avant de se lancer dans le calcul de A1A^{-1}, vérifier qu'elle soit effectivement inversible !

Exemple : A=(18211174131922601102)A = \begin{pmatrix}1&8&2&-11\\1&7&4&13\\1&9&2&26\\0&11&0&-2\end{pmatrix}

Comme a3=2a1,A\vec{a}_3=2\vec{a}_1,\, A n'est pas inversible.

Méthode pour chercher A1A^{-1}

Soit A(n×n)A\,(n\times n), nous cherchons donc nue matrice A1(n×n)A^{-1}\, (n\times n), telle que

AA1=InA1A=InAA^{-1} = I_n\\ A^{-1}A=I_n

Ecrivons A1=[c1cn]ciRnA^{-1} =[\vec{c}_1\cdots\vec{c}_n]\quad\vec{c}_i\in\R^n

Comme AA1=[Ac1Ac2Acn]=[e1e2en]AA^{-1}=[A\vec{c}_1\,A\vec{c}_2\cdots A\vec{c}_n] =[\vec{e}_1\,\vec{e}_2\cdots \vec{e}_n] car AA1=InAA^{-1}=I_n

    Aci=eii=1,,nn\implies\boxed{A\vec{c}_i=\vec{e}_i\quad i=1,\ldots,n}\leftarrow n systèmes linéaire, chacun à $ n$ inconnues (les nn composantes de ci\vec{c}_i)

Idée : résoudre ces nn systèmes "en une seule fois".

Définition

Une matrice n×nn\times n est élémentaire si elle s'obtient à partir de InI_n à l'aide d'une seule opératoin élémentaire (de Type I, II ou III)

Exemple
E1=(100001010)est eˊleˊmentaire (c’est I3, surlaquelle on a fait L2L3)E2=(100020001)est eˊleˊmentaire (c’est I3, surlaquelle on a fait L22L2)(123010001)n’est pas eˊleˊmentaire\begin{array}{rll} E_1&=\begin{pmatrix}1&0&0\\0&0&1\\0&1&0\end{pmatrix}& \text{est élémentaire (c'est $I_3$, sur}\\ &&\text{laquelle on a fait $L_2\leftrightarrow L_3$)}\\ E_2&=\begin{pmatrix}1&0&0\\0&-2&0\\0&0&1\end{pmatrix}& \text{est élémentaire (c'est $I_3$, sur}\\ &&\text{laquelle on a fait $L_2\leftarrow-2L_2$)}\\ &\quad\begin{pmatrix}1&2&3\\0&1&0\\0&0&1\end{pmatrix}& \text{n'est pas élémentaire} \end{array}

Remarque :

  • Toute matrice élémentaire est inversible. (Par exemple : E11=E1E_1^{-1}=E_1)

  • On peut utiliser les matrices élémentaires pour voir les opérations élémentaires comme des produits matriciels.

    Par exemple : si A=(abcdefghi)L2L3A = \begin{pmatrix}a&b&c\\d&e&f\\g&h&i\end{pmatrix}\quad\color{red}L_2\leftrightarrow L_3

    E1A=(100001010)(abcdefghi)=(abcghidef)E_1A=\begin{pmatrix}1&0&0\\0&0&1\\0&1&0\end{pmatrix} \begin{pmatrix}a&b&c\\\color{orange}d&\color{orange}e& \color{orange}f\\\color{blue}g&\color{blue}h&\color{blue}i \end{pmatrix}=\begin{pmatrix}a&b&c\\ \color{blue}g& \color{blue}h&\color{blue}i\\\color{orange}d& \color{orange}e&\color{orange}f\end{pmatrix}

Théorème

Soit A(n×n)A\,(n\times n). Alors

AA est inversible     \iff AA est "équivalente-ligue" à InI_n (c'est-à-dire que l'on peut passer de AA à InI_n en un nombre fini d'opération élémentaire)

De plus, toute suite d'opérations élémentaires qui transforme AA en InI_n transforme aussi InI_n en A1A^{-1}.

Preuve

    \implies Supposons que AA est inversible. Donc bRn\forall\,\vec{b}\in\R^n, le système "Ax=bA\vec{x}=\vec{b}" possède une unique solution. Son "Ax=bA\vec{x}=\vec{b}" n'a pas de variables libres, donc la réduite de AA contient exactement n pivots, c'est donc InI_n.

    \impliedby Si AA est "équivalente-ligne" à InI_n, il existe des matrices élémentaires E(1),,E(N)E^{(1)},\ldots,E^{(N)} telles que

E(N)E(2)E(1)A=InE(N)1E(N1)E(1)A=E(N)1In=E(N)1E(N1)1A=E(1)1E(N)1    A est inversible, et A1=(E(1)1E(N)1)1=E(N)E(1)    A1=E(N)E(1)In\begin{array}{rlll} E^{(N)}\cdots E^{(2)}E^{(1)}A&=I_n &\\ &&\downarrow E^{(N)^{-1}}\cdot\\ E^{(N-1)}\cdots E^{(1)}A&=E^{(N)^{-1}}I_n=E^{(N)^{-1}}&\\ &&\downarrow E^{(N-1)^{-1}}\cdot\\ \vdots\\ A&=E^{(1)^{-1}}\cdots E^{(N)^{-1}} \end{array}\\\text{}\\ \begin{array}{rll} \implies A \text{ est inversible, et }A^{-1}&=(E^{(1)^{-1}} \cdots E^{(N)^{-1}})^{-1}&\\&=E^{(N)}\cdots E^{(1)}&\\ \implies A^{-1}&=E^{(N)}\cdots E^{(1)}\color{red}I_n&\\ &&\square \end{array}

Algorithme de Gauss-Jordan

Rappel :

Aci=eiE(1)Aci=E(1)eiInci=E(N)E(1)ei\begin{array}{rl}A\vec{c}_i&=\vec{e}_i\\ E^{(1)}A\vec{c}_i&=E^{(1)}\vec{e}_i\\ &\cdots\\ I_n\vec{c}_i&=E^{(N)}\cdots E^{(1)}\vec{e}_i\end{array}

Algorithme

Exemple 1
A=(1224)(AI2)=(12102401)L2L22L1=(12100021)\begin{array}{rl}A=\begin{pmatrix}1&2\\2&4\end{pmatrix} \qquad (A|I_2)& = \left(\begin{array}{rr|rr}1&2&1&0\\2&4&0&1 \end{array}\right)\\\\&\big\downarrow L_2\leftarrow L_2-2L_1\\\\&= \left(\begin{array}{rr|rr} 1&2&1&0\\0&0&-2&1 \end{array}\right) \end{array}

Comme A~I2\tilde A\not=I_2, AA singulière.

Exemple 2
A=(012103438)(AI3)=(012100103010438001)L1L2=(103010012100438001)L3L34L1=(103010012100034041)L3L3+3L2L3L32(aˋ veˊrifier !)L2L22L3L1L13L3=(1009273201024100132212)=(A~C)\begin{array}{rll} A &=\begin{pmatrix}0&1&2\\1&0&3\\4&-3&8\end{pmatrix}&\\ \text{}\\ (A|I_3)&=\left(\begin{array}{ccc|ccc}0&1&2&1&0&0\\1&0&3&0&1&0\\ 4&-3&8&0&0&1\end{array}\right)\\ &&\downarrow L_1\leftrightarrow L_2\\ &=\left(\begin{array}{ccc|ccc}1&0&3&0&1&0\\0&1&2&1&0&0\\ 4&-3&8&0&0&1\end{array}\right)&\\ &&\downarrow L_3\leftarrow L_3-4L_1\\ &=\left(\begin{array}{ccc|ccc}1&0&3&0&1&0\\0&1&2&1&0&0\\ 0&-3&-4&0&-4&1\end{array}\right)\\ &&\downarrow L_3\leftarrow L_3+3L_2\\ &&\cdots\\&&\downarrow L_3\leftarrow\frac{L_3}{2}\\ &\qquad\qquad\text{(à vérifier !)}&\cdots\\&&\downarrow L_2\leftarrow L_2-2L_3\\ &&\cdots\\&&\downarrow L_1\leftarrow L_1-3L_3\\ &=\left(\begin{array}{ccc|ccc}1&0&0&-\frac{9}{2}&7&-\frac{3}{2} \\0&1&0&-2&4&-1\\0&0&1&\frac{3}{2}&-2&\frac{1}{2}\end{array}\right) \\\\&=(\tilde A|C) \end{array}

Comme A~=In\tilde A=I_n, AA est inversible et A1=CA^{-1}=C. ((Vérifier que AC=CA=I3!)AC=CA=I_3 \,!)

Critères d'inversibilité

Théorème

Soit A(n×n)A\,(n\times n). Sont équivalents

  1. AA est inversible

  2. bRn\forall\,\vec{b}\in\R^n, le système "Ax=bA\vec{x}=\vec{b}" possède une unique solution

  3. Ax=0A\vec{x}=\vec{0} ne possède que la solution tiviale x=0\vec{x}=\vec{0}

  4. les nn colonnes de AA sont linéairement indépendantes

  5. Im(A)\mathrm{Im}(A) (ou "Col(A)\mathrm{Col}(A)") est tout Rn\R^n

  6. la réduite de AA possède un pivot sur chaque ligne

  7. AA peut se réduire à InI_n (à l'aide d'une suite finite d'opérations élémentaires)

  8. AA peut s'écrire comme produit de matrices élémentaires

Preuve
  1.     \implies 2. OK (début de la leçon)

  2.     \implies 3. OK

  3.     \implies 4. OK

  4.     \implies 5. OK

  5.     \implies 6. Si Im(A)=Rn\mathrm{Im}(A)=\R^n, tout vecteur yRn\vec{y}\in\R^n peut s'écrire comme combinaison linéaire des colonnes de AA :

    Ax=yreˊductionpas de variables libresreˊduite de A posseˋde n pivots, un dans chaque ligne\begin{aligned} A\vec{x}&=\vec{y}\\ &\downarrow\text{réduction}\\ \text{pas de }&\text{variables libres}\\ &\downarrow \end{aligned}\\ \text{réduite de $A$ possède $n$ pivots, un dans chaque ligne}
  6.     \implies 7. Donc la réduite de AA est InI_n

  7.     \implies 8. OK (voir théorème précédent)

  8.     \implies 9. Si A=E(1)E(N)A=E^{(1)}\cdots E^{(N)} où chaque E(i)E^{(i)} est élémentaire, donc inversible, alors AA est aussi inversible. \qquad\square

Conséquences
AA-1=InA1A=In}faut-il vraiment veˊrifier les deux conditions ? NON !\left.\begin{aligned}AA^\text{-1}=I_n\\A^{-1}A=I_n\end{aligned} \right\}\text{faut-il vraiment vérifier les deux conditions ? NON !}

Théorème

Si AA est (n×n)(n\times n) donnée.

  1. Si C(n×n)\exists\,C\,(n\times n) tel que CA=InCA =I_n, alors AA est inversibles, et A1=CA^{-1} = C.

  2. Si B(n×n)\exists\,B\,(n\times n) tel que AB=InAB=I_n, alors AA est inversible, et A1=BA^{-1}=B.

Preuve
  1. Supposons que CA=InCA = I_n.

    Affirmation : Ceci implique que Ax=0A\vec{x}=\vec{0} ne possède que la solution triviale. (Par le point 3 du théorème précédent, AA est donc inversible)

    En effet, si x\vec{x} est solution de Ax=0A\vec{x}=\vec{0}, alors x=Inx=(CA)x=C(Ax)=C0=0\vec{x}=I_n\vec{x}=(CA)\vec{x}=C(A\vec{x})=C\vec{0}=\vec{0}

    Donc C=CIn=A1existeC(AA1)=(CA)InA1=A1\text{Donc }C=CI_n\stackrel{A^{-1}\text{existe}}{=}C(AA^{-1})=\underbrace{(CA)}_{I_n}A^{-1}=A^{-1}

  2. Supposons que AB=InAB=I_n. Alors yRn\forall\,\vec{y}\in\R^n

    (AB)y=yA(By)=y(AB)\vec{y}=\vec{y}\\A(B\vec{y})=\vec{y}

    Donc le système "Ax=yA\vec{x}=\vec{y}" possède une solution, donnée par x=By\vec{x}=B\vec{y}.

        Im(A)=Rn.\qquad\implies\textrm{Im}(A)=\R^n.

    Par e) du théorème précédent, AA est inversible.

    Aussi, B=InB=(A1A)B=A1existeA1(AB)In=A1B=I_nB=(A^{-1}A)B\stackrel{A^{-1}\text{existe}}{=}A^{-1}\underbrace{(AB)}_{I_n}=A^{-1}\quad\square

Sur les matrices élémentaitres (et leurs inverses)

Soit B(n×n)B\,(n\times n). On étudie les matrices élémentaires associées aux opérations élémentaires de Types I, II, III, et leurs inverses.

Type I
BLiLjTijBouˋ Tij est la n×nTij=(1101101)ijComme TijTij=In,Tij est inversible, etTij1=TijB\stackrel{L_i\leftrightarrow L_j} {\longmapsto}T_{ij}B\quad\text{où $T_{ij}$ est la $n\times n$}\\\text{}\\ T_{ij}=\left(\begin{array}{cccccccc} 1&&&&&&&\\&1&&&&&&\\&&\ddots&&&&&\\&&&0&&1&&&\\&&&\downarrow& \ddots&\uparrow&&\\&&&1&&0&&&\\&&&&&&\ddots&\\&&&&&&&1 \end{array}\right)\begin{array}{l}\\ \text{} \\ \text{} \\ i \\ \text{} \\ j \\\text{}\\\text{}\end{array}\\ \text{}\\ \text{Comme } T_{ij}T_{ij}=I_n, \, T_{ij} \text{ est inversible, et} \\ \boxed{T_{ij}^{-1}=T^{ij}}
Type II
BLiλLiDi(λ)ouˋ Di(λ)B est la n×n(λ0)Di(λ)=(1101λ011)iB\stackrel{L_i\leftarrow\lambda L_i}\longmapsto D_i(\lambda) \quad \text{où }D_i(\lambda)B \text{ est la }n\times n \, (\lambda \not = 0)\\ \text{}\\ D_i(\lambda)=\left(\begin{array}{cccccccc} 1&&&&&&&\\&1&&&&&&\\&&\ddots&&&&\text{\huge0}&\\&&&1&&&&&\\&&&& \lambda&&&\\&&\text{\huge0}&&&1&&&\\&&&&&&\ddots&\\&&&&&&&1 \end{array}\right)\begin{array}{l}\\ \\ \text{} \\ \text{} \\ i \\ \text{} \\\text{}\\\text{}\end{array}\\

Comme Di(λ)Di(μ)=Di(λμ)D_i(\lambda)D_i(\mu)=D_i(\lambda\mu), et Di(1)=InD_i(1)=I_n, on peut chercher le μ\mu tel que

Di(λ)Di(μ)=In=Di(1)μ=1λD_i(\lambda)D_i(\mu) = I_n =D_i(1)\rightarrow\mu=\frac{1}{\lambda}

Donc lorsque λ0\lambda\not=0, Di(λ)D_i(\lambda) est inversible, et

Di(λ)1=Di(1λ)\boxed{D_i(\lambda)^{-1}=D_i(\frac{1}{\lambda})}
Type III
BLiLi+λLjLij(λ)Bouˋ Lij(λ) est la n×nRemarque: le 1er indice (i) est celui de la ligne qui changeTij=(11011λ101)ijB\stackrel{L_i\leftrightarrow L_i +\lambda L_j} {\longmapsto}L_{ij}(\lambda)B\quad\text{où $L{ij}(\lambda)$ est la $n\times n$}\\\textbf{Remarque: }\text{le $1^{er}$ indice $(i)$ est celui de la ligne qui change}\\\text{}\\ T_{ij}=\left(\begin{array}{ccccccccc} 1&&&&&&&&\\&1&&&&&&&\\&&\ddots&&&&&\text{\huge0}&\\&&&1&&&&&&\\&&&&1&&\lambda&& \\&&&&&\ddots&\uparrow&&\\&&&&&&1&&&\\&&\text{\huge0}&&&&&\ddots&\\&&&&&&&&1 \end{array}\right)\begin{array}{l}\\ \text{} \\ \text{} \\ \text{} \\ i \\ \text{} \\ j \\\text{}\\\text{}\end{array}\\ \text{}\\

Comme Lij(0)=InL_{ij}(0)=I_n, et Lij(μ)Lij(λ)=Lij(λ+μ)L_{ij}(\mu)L_{ij}(\lambda)=L_{ij}(\lambda+\mu), on peut chercher le λ\lambda tel que

Lij(μ)Lij(λ)=In(=Lij(0))λ=μL_{ij}(\mu)L_{ij}(\lambda)=I_n(=L_{ij}(0))\rightarrow\lambda=-\mu

Donc Lij(μ)L_{ij}(\mu) est inversibles, et

Lij(μ)1=Lij(μ)\boxed{L_{ij}(\mu)^{-1}=L_{ij}(-\mu)}

Factorisation matricielle : la décomposition "LULU"

Définition

Une C(m×n)C\,(m\times n) est

Exemples
Exercice

Démontrez qu'un produit de matrices triangulaires inférieures est triangulaire inférieur et qu'un produit de matrices triangulaires supérieures est triangulaire supérieur.

Définition

Soit AA une matrice (m×n)(m\times n). Une décomposition LULU de AA est une factorisation de la forme

A=LU ouˋ,A=LU\text{ où,}
Exemple

A=(111235468)=(100210421)L(1110+13002)UA=\left(\begin{array}{rrrr}1&1&1\\2&3&5\\4&6&8\end{array}\right)=\underbrace{\left(\begin{array}{rrrrr}\textcolor{blue}1&0&0\\2&\textcolor{blue}1&0\\4&2&\textcolor{blue}1\end{array}\right)}_L\,\underbrace{\left(\begin{array}{rrrrrr}1&1&1\\0&\textcolor{white}+1&3\\0&0&-2\end{array}\right)}_U

Pourquoi une décomposition LULU ?

Utilité : on aimerait résoudre NN systèmes linéaires

{Ax(1)=b1Ax(2)=b2Ax(N)=bN\begin{cases} &A\vec{x}^{(1)}=\vec{b}_{1}\\ &A\vec{x}^{(2)}=\vec{b}_{2}\\ &\quad\vdots\\ &A\vec{x}^{(N)}=\vec{b}_{N} \end{cases}
  1. Décomposition AA en A=LUA=LU (lorsque c'est possible)

  2. i=1,,N\forall\,i=1,\ldots,N

Reˊsoudre “Axi=bi    L(Ux(i))=bireˊsoudre Ly(i)=bireˊsoudre Ux(i)=y(i)\begin{aligned} \boxed{\text{Résoudre “$A\vec{x}^i=\vec{b}_i$" }} \iff&\boxed{L(U\vec{x}^{(i)})=\vec{b}_i}\\ &\qquad\quad\Big\Updownarrow\\ &\boxed{ \begin{array}{lllll} \bullet&\text{résoudre }L\vec{y}^{(i)}=\vec{b}_i\\ \bullet&\text{résoudre } U\vec{x}^{(i)}=\vec{y}^{(i)}\\ \end{array} } \end{aligned}

Remarque : ces systèmes étant triangulaires, ils sont rapides à résoudre !

Sur l'existence d'une décomposition LULU

Théorème

Si AA est une matrice (m×n)(m\times n), que l'on peut échelonner à l'aide d'opérations élémentaires de Type III, dans lesquelles on ne fait des additions que vers le bas, (c'est-à-dire à l'aide de matrices Lij(λ)L_{ij}(\lambda) avec i>ji>j), alors il existe une décomposition :

A=LU,ouˋA=LU, \text{où}

UU est l'échelonnée de AA.

Preuve

Supposons que l'échelonnée de AA soit

A~=ENE2E1A\tilde{A}=\color{red}\underline{\color{black}E_N\cdots E_2\,E_1}\color{black}\,A

où les E1ENE_1\cdots E_N sont de type "Lij(λ)L_{ij}(\lambda)", avec i>ji>j.

Donc, par 1., MM est inversible et

M1=E11EN11. est inversible2. est triangulaire infleˊrieure3. n’a que des “1" sur sa diagonale\begin{aligned}\qquad M^{-1}=E_1^{-1}\cdots E_N^{-1}\qquad\quad&1. \text{ est inversible}\\&2.\text{ est triangulaire inflérieure}\\&3.\text{ n'a que des “}1\text{" sur sa diagonale}\end{aligned}

A~=MA    A=M1A~\tilde{A}=MA\iff A=M^{-1}\tilde{A}

Remarque :

  • M1M^{-1} est triangle inférieure et n'a que des "11" sur sa diagonale (2. et 3.)

  • A~\tilde{A} est triangulaire supérieure

Donc on prend LM1UA~\begin{aligned}L&\coloneqq M^{-1}\\U&\coloneqq\tilde{A}\\&&\square\end{aligned}

Résumé

En partant de AA,

U=ENE2E1A121,,N produit aˋ gauche.puis :L=InE111E2121EN1N produit aˋ droite.\begin{array}{lllll}\qquad\quad&U=\color{blue}\underbrace{\color{black}E_N\color{blue}\underbrace{\color{black}\cdots\color{blue}\underbrace{\color{black}E_2\,\color{blue}\underbrace{\color{black}E_1\,A}_{{\bigcirc}\mathllap{\tiny1\,\,}}}_{{\bigcirc}\mathllap{\tiny2\,\,}}}_{\,\,\,\,\,\cdots\color{white}{\bigcirc}\mathllap{\tiny1,,}}}_{{\bigcirc}\mathllap{\tiny N\,}}\qquad&\color{red}|\text{ produit à }\textbf{gauche}.\\\text{puis :}\\&L=\color{blue}\underbrace{\color{black}\color{blue}\underbrace{\color{black}\color{blue}\underbrace{\color{black}\color{blue}\underbrace{\color{black}I_n\,E_1^{-1}}_{{\bigcirc}\mathllap{\tiny1\,\,}}\,\color{black}E_2^{-1}}_{{\bigcirc}\mathllap{\tiny2\,\,}}\color{black}\cdots}_{\,\,\,\,\,\cdots\color{white}{\bigcirc}\mathllap{\tiny1\,\,}}\color{black}\,E_N^{-1}}_{{\bigcirc}\mathllap{\tiny N\,}}\qquad&\color{red}|\text{ produit à }\textbf{droite}.\end{array}

À comprendre : que signifie "multiplier à droite pour une matrice élémentaire" ?

Type I :

Rappel :LiLjATijA\qquad\begin{array}{ccc}&\stackrel{L_i\leftrightarrow L_j}{\curvearrowright}\\A&\longmapsto &T_{ij}A\end{array}

Exemple
(abcdefghi)(010100001)=(bacedfhgi)T12\begin{aligned} \begin{pmatrix}\textcolor{blue}{a}&\textcolor{red}{b}&c\\\textcolor{blue}{d}&\textcolor{red}{e}&f\\\textcolor{blue}{g}&\textcolor{red}{h}&i\end{pmatrix} &\begin{pmatrix}0&1&0\\1&0&0\\0&0&1\end{pmatrix}= \begin{pmatrix}\textcolor{red}{b}&\textcolor{blue}{a}&c\\\textcolor{red}{e}&\textcolor{blue}{d}&f\\\textcolor{red}{h}&\textcolor{blue}{g}&i\end{pmatrix}\\&\qquad\,\, T_{12} \end{aligned}

On a échangé la première et la deuxième colonne.

Multiplier par Tij aˋ droite, c’est eˊchanger les colonnes i et j.\boxed{\text{Multiplier par }T_{ij} \text{ à }\textbf{droite}\text{, c'est échanger les colonnes }i \text{ et }j.}

Type II :

Rappel : ADi(λ)A\qquad A\mapsto D_i(\lambda) A

Exemple :
(abcdefghi)(1000λ0001)=(abλcdeλfghλi)D2(λ)\begin{aligned} \begin{pmatrix}a&b&c\\d&e&f\\g&h&i\end{pmatrix} &\begin{pmatrix}1&0&0\\0&\lambda&0\\0&0&1\end{pmatrix}= \begin{pmatrix}a&b\lambda&c\\d&e\lambda&f\\g&h\lambda&i\end{pmatrix}\\ &\quad\,\,\,\, D_2(\lambda) \end{aligned}

On a multiplié la deuxième colonne par λ\lambda.

Multiplier par Di(λ) aˋ droite, c’est multiplier la colonne i par λ.\boxed{\text{Multiplier par }D_i(\lambda) \text{ à }\textbf{droite}\text{, c'est multiplier la colonne }i \text{ par }\lambda.}

Type III :

Rappel : ALij(λ)AA\mapsto L_{ij}(\lambda)A

Exemple :
(abcdefghi)(10001λ001)=(abc+λbdef+λeghi+λh)L23(λ)\begin{aligned} \begin{pmatrix}a&b&c\\d&e&f\\g&h&i\end{pmatrix} &\begin{pmatrix}1&0&0\\0&1&\lambda\\0&0&1\end{pmatrix}= \begin{pmatrix}a&b&c+\lambda b\\d&e&f+\lambda e\\g&h&i+\lambda h\end{pmatrix}\\ &\quad\,\,\,\, L_{23}(\lambda) \end{aligned}

On a ajouté λ\lambda fois la deuxième colonne à la troisième. Multiplier par Lij(λ) aˋ droite, c’est ajouter aˋ la colonne jλ fois la ieˋme.\boxed{\text{Multiplier par }L_{ij}(\lambda) \text{ à }\textbf{droite}\text{, c'est ajouter à la colonne }j\, \lambda\text{ fois la i}^\text{ème}.}

Procédure pour la construction de LL

AIn1E1InE1112E2E1AInE11E212NENE1AUInE11EN1LN\begin{array}{rr|ll} &A&I_n&\\ \color{blue}{\bigcirc}\mathllap{\footnotesize1\,\,}&E_1&I_nE_1^{-1}&\color{blue}{\bigcirc}\mathllap{\footnotesize1\,\,}'\\ \color{blue}{\bigcirc}\mathllap{\footnotesize2\,\,}&E_2E_1A&I_nE_1^{-1}E_2^{-1}&\color{blue}{\bigcirc}\mathllap{\footnotesize2\,\,}'\\ \text{}\\ &\vdots\quad&\quad\vdots&\\ \text{}\\ \color{blue}{\bigcirc}\mathllap{\footnotesize N\,}&\underbrace{E_N\cdots E_1A}_{U}&\underbrace{I_nE_1^{-1}\cdots E_N^{-1}}_{L}&\color{blue}{\bigcirc}\mathllap{\footnotesize N\,}' \end{array}

Exemple

A=(111235468)(100010001)=In1L21(2)A=(111013468)(100210001)=InL21(2)1=InL21(2)12L31(4)L21(2)A=(111013024)(100210401)=InL21(2)L31(4)1=InL21(2)L31(4)2NL32(2)L31(4)L21(2)A=(111013002)U(100210421)=InL21(2)L31(4)L32(2)1=InL21(2)L31(4)L32(2)LN\begin{array}{rr|ll} &A=\begin{pmatrix}1&1&1\\2&3&5\\4&6&8\end{pmatrix}&\begin{pmatrix}1&0&0\\0&1&0\\0&0&1\end{pmatrix}=I_n&\\ \text{}\\ \color{blue}{\bigcirc}\mathllap{\footnotesize1\,\,}&L_{21}(-2)A=\begin{pmatrix}1&1&1\\0&1&3\\4&6&8\end{pmatrix}&\begin{pmatrix}1&0&0\\2&1&0\\0&0&1\end{pmatrix}\begin{aligned}&\text{}\\&=I_nL_{21}(-2)^{-1}\\&=I_nL_{21}(2)\\\end{aligned}&\color{blue}{\bigcirc}\mathllap{\footnotesize1\,\,}'\\ \text{}\\ \color{blue}{\bigcirc}\mathllap{\footnotesize2\,\,}&L_{31}(-4)L_{21}(-2)A=\begin{pmatrix}1&1&1\\0&1&3\\0&2&4\end{pmatrix}&\begin{pmatrix}1&0&0\\2&1&0\\4&0&1\end{pmatrix}\begin{aligned}&\text{}\\&=I_nL_{21}(2)L_{31}(-4)^{-1}\\&=I_nL_{21}(2)L_{31}(4)\\\end{aligned}&\color{blue}{\bigcirc}\mathllap{\footnotesize2\,\,}'\\ \text{}\\ &\vdots\quad&\quad\vdots&\\ \text{}\\ \color{blue}{\bigcirc}\mathllap{\footnotesize N\,}&\underbrace{L_{32}(-2)L_{31}(-4)L_{21}(-2)A=\begin{pmatrix}1&1&1\\0&1&3\\0&0&-2\end{pmatrix}\begin{aligned}&\text{}\\&\text{}\\&\text{}\\\end{aligned}}_{U}&\underbrace{\begin{pmatrix}1&0&0\\2&1&0\\4&2&1\end{pmatrix}\begin{aligned}&\text{}\\&=I_nL_{21}(2)L_{31}(4)L_{32}(-2)^{-1}\\&=I_nL_{21}(2)L_{31}(4)L_{32}(2)\end{aligned}}_{L}&\color{blue}{\bigcirc}\mathllap{\footnotesize N\,}' \end{array}

Changement de base

Soit VV un EV de dimension nn, et

B=(b1,,bn)C=(c1,,c2)}deux bases.\quad\left.\begin{aligned}&\mathcal{B}=(b_1,\ldots,b_n)\\&\mathcal{C}=(c_1,\ldots,c_2)\end{aligned}\right\}\text{deux bases.}

IMAGE P.125/126

Relatif aˋ B:[v]B=(β1βn)Relatif aˋ C:[v]C=(γ1γn)\quad\begin{aligned}&\text{Relatif à }\mathcal{B}:\\&&\left[v\right]_\mathcal{B}=\begin{pmatrix}\beta_1\\\vdots\\\beta_n\end{pmatrix}\end{aligned}\qquad\begin{aligned}&\text{Relatif à }\mathcal{C}:\\&&\left[v\right]_\mathcal{C}=\begin{pmatrix}\gamma_1\\\vdots\\\gamma_n\end{pmatrix}\end{aligned}

Y a-t-il un lien entre βn\beta_n et γn\gamma_n ?

Proposition

Il existe une matrice (m×n)(m\times n) inversible notée PCBP_{\mathcal{C}\mathcal{B}} (ou parfois PCBP_{\mathcal{C}\leftarrow\mathcal{B}}), telle que

[v]C=PCB[v]B[v]_\mathcal{C}=P_{\mathcal{C}\textcolor{red}{\mathcal{B}}}\left[v\right]_\textcolor{red}{\mathcal{B}}

De plus, PCB=[[b1]C[bn]C], etPCB1=PBC=[[c1]B[cn]B]\begin{aligned}\text{De plus, }&P_{\mathcal{C}\mathcal{B}}=\left[[b_1]_\mathcal{C}\cdots[b_n]_\mathcal{C}\right]\text{, et}\\&P_{\mathcal{C}\mathcal{B}}^{-1}=P_{\mathcal{B}\mathcal{C}}=\left[[c_1]_\mathcal{B}\cdots[c_n]_\mathcal{B}\right]\end{aligned}

(PCBP_{\mathcal{C}\mathcal{B}} et PBCP_{\mathcal{B}\mathcal{C}} sont des matrices de changement de base).

IMAGE P.126

Preuve

Soit vVv\in V, décomposé dans B\mathcal{B}:

Soit vV, deˊcomposeˊ dans B:v=β1b1++βnbn[.]C,qui est lineˊaire[v]C=β1[b1]C++βn[bn]Cil s’agit d’un produit !=[[b1]C[bn]C]matrice : PCB[v]B()1De meˆme,[v]B=[c1]B[cn]BPBC[v]C()2Il faut injecter ()2 dans ()1:[v]C=PCBPBC[v]C()v\begin{array}{llll} \text{Soit $v\in V$, décomposé dans $\mathcal{B}$:}\\ \text{}\\ \qquad v=\beta_1b_1+\ldots+\beta_nb_n&\\ &\Big\downarrow\begin{array}{llll}[\,.\,]_\mathcal{C},\\\text{qui est linéaire}\end{array}\\ \qquad\begin{aligned}[v]_\mathcal{C}&=\underbrace{\beta_1[b_1]_\mathcal{C}+\ldots+\beta_n[b_n]_\mathcal{C}}_{ \text{il s'agit d'un produit !}}\\ \text{}\\ &=\underbrace{\left[[b_1]_\mathcal{C}\cdots[b_n]_\mathcal{C}\right]}_{\text{matrice : }P_{\mathcal{C}\mathcal{B}}}[v]_\mathcal{B}\quad(*)_1 \end{aligned}\\ \text{De même,}\\ \text{}\\ \qquad[v]_\mathcal{B}=\underbrace{[c_1]_\mathcal{B}\cdots[c_n]_\mathcal{B}}_{P_{\mathcal{B}\mathcal{C}}}[v]_\mathcal{C}\quad\,\,\,(*)_2\\ \text{}\\ \text{Il faut injecter }(*)_2 \text{ dans } (*)_1 :\\ \text{}\\ \qquad[v]_\mathcal{C}=P_{\mathcal{C}\mathcal{B}}P_{\mathcal{B}\mathcal{C}}[v]_{\mathcal{C}}\quad(\forall )v \end{array}

Exercice : Si x=MxxRn, alors M=In.\vec{x}=M\vec{x}\quad\forall\vec{x}\in\R^n\text{, alors } M=I_n.

PCBPBC=In\longrightarrow P_{\mathcal{C}\mathcal{B}}P_{\mathcal{B}\mathcal{C}}=I_n

Donc PCB1=PBCP_{\mathcal{C}\mathcal{B}}^{-1}=P_{\mathcal{B}\mathcal{C}}, et PBC1=PCB\begin{aligned}P_{\mathcal{B}\mathcal{C}}^{-1}=P_{\mathcal{C}\mathcal{B}}\\&\qquad\square\end{aligned}

Exemple

IMAGE P.127

Consideˊrons v=3e1+e2    [v]Bcan=(31)\text{Considérons }\vec{v}=3\vec{e}_1+\vec{e}_2\implies[\vec{v}]_{\mathcal{B}\mathrm{can}}=\begin{pmatrix}3\\1\end{pmatrix}

But : [v]B=(αβ) ?(1.21.5) (aˋ peu preˋs, sur le dessin)\textbf{But : }[\vec{v}]_{\mathcal{B}'}=\begin{pmatrix}\alpha\\\beta\end{pmatrix}\text{ ?}\quad\color{orange}\simeq\begin{pmatrix}1.2\\-1.5\end{pmatrix}\text{ (à peu près, sur le dessin)}

Mais : [v]B=PBBcanaˋ calculer[v]Bcan(31)\text{Mais : }[\vec{v}]_{\mathcal{B}'}=\underbrace{P_{\mathcal{B}'\leftarrow\mathcal{B}_\mathrm{can}}}_{\text{à calculer}}\,\underbrace{[\vec{v}]_{\mathcal{B}\mathrm{can}}}_{\begin{pmatrix}3\\1\end{pmatrix}}

Méthode 1 pour calculer PBBcanP_{\mathcal{B}'\leftarrow\mathcal{B}_\mathrm{can}}
Par la proposition, PBBcan=[[e1]B[e2]B]Or,{b1=e1+2e2[b1]Bcan=(12)b2=e1+e2[b2]Bcan=(11){e1=13b123b2[e1]B=(1/32/3)e2=13b1+13b2[b2]B=(1/31/3)Donc PBBcan=(1/31/32/31/3)[v]B=(1/31/32/31/3)(31)=(4/35/3)\begin{array}{llll} \text{Par la proposition, }\\ \text{}\\ \qquad P_{\mathcal{B}'}\leftarrow\mathcal{B}_\mathrm{can}=\left[[\vec{e}_1]_{\mathcal{B}'}[\vec{e}_2]_{\mathcal{B}'}\right]\\ \text{Or,}\\ \qquad\begin{cases} \vec{b}_1=\vec{e}_1+2\vec{e}_2&\leftarrow[\vec{b}_1]_{\mathcal{B}\mathrm{can}}=\begin{pmatrix}1\\2\end{pmatrix}\\ \text{}\\ \vec{b}_2=-\vec{e}_1+\vec{e}_2&\leftarrow[\vec{b}_2]_{\mathcal{B}\mathrm{can}}=\begin{pmatrix}-1\\1\end{pmatrix} \end{cases}\\ \big\downarrow\\ \qquad\begin{cases} \vec{e}_1=\frac{1}{3}\vec{b}_1-\frac{2}{3}\vec{b}_2&\rightarrow[\vec{e}_1]_{\mathcal{B}'}=\begin{pmatrix}1/3\\-2/3\end{pmatrix}\\ \text{}\\ \vec{e}_2=\frac{1}{3}\vec{b}_1+\frac{1}{3}\vec{b}_2&\rightarrow[\vec{b}_2]_{\mathcal{B}'}=\begin{pmatrix}1/3\\1/3\end{pmatrix} \end{cases}\\ \text{}\\ \text{Donc }P_{\mathcal{B}'}\leftarrow\mathcal{B}_\mathrm{can}=\begin{pmatrix}1/3&1/3\\-2/3&1/3\end{pmatrix} \text{}\\ \rightarrow[\vec{v}]_{\mathcal{B}'}=\begin{pmatrix}1/3&1/3\\-2/3&1/3\end{pmatrix}\begin{pmatrix}3\\1\end{pmatrix}=\begin{pmatrix}4/3\\-5/3\end{pmatrix} \end{array}
Méthode 2 pour calculer PBBcanP_{\mathcal{B}'\leftarrow\mathcal{B}_\mathrm{can}}
PBBcan=(PBcanB)1=[[b1]Bcan[b2]Bcan]1=(1121)1=13(1121)=(1/31/32/31/3)\begin{aligned} P_{\mathcal{B}'\leftarrow\mathcal{B}_\mathrm{can}}&=\left(P_{\mathcal{B}\mathrm{can}\leftarrow\mathcal{B}'}\right)^{-1}\\ &=\left[[\vec{b}_1]_{\mathcal{B}\mathrm{can}}[\vec{b}_2]_{\mathcal{B}\mathrm{can}}\right]^{-1}\\ &=\begin{pmatrix}1&-1\\2&1\end{pmatrix}^{-1}\\ &=\frac{1}{3}\begin{pmatrix}1&1\\-2&1\end{pmatrix}=\begin{pmatrix}1/3&1/3\\-2/3&1/3\end{pmatrix} \end{aligned}

Donc pour un vecteur quelconque v,[v]Bcan=(v1v2)\text{Donc pour un vecteur quelconque }\vec{v},[\vec{v}]_{\mathcal{B}\mathrm{can}}=\begin{pmatrix}v_1\\v_2\end{pmatrix}

[v]B=(1/31/32/31/3)(v1v2)=(1/3v11/3v22/3v11/3v2)\boxed{[\vec{v}]_{\mathcal{B}'}=\begin{pmatrix}1/3&1/3\\-2/3&1/3\end{pmatrix}\begin{pmatrix}v_1\\v_2\end{pmatrix}= \begin{pmatrix}1/3v_1&1/3v_2\\-2/3v_1&1/3v_2\end{pmatrix} }