Espaces vectoriels

Définition

Un espace vectoriel réel (abrégé EV) est un ensemble VV dont les éléments sont appelés "vecteurs", muni de deux opérations, l'addition "+" et la multiplication par des scalaires (R)(\in\R), tel que les propriétés suivantes soient satisfaites :

Exemples

  1. V=Rn,v:vecteur de Rn{+:addition vectoriellemultiplication par λR\begin{aligned}V=\R^n,&\qquad v:\text{vecteur de }\R^n\\&\begin{cases}&+:\text{addition vectorielle}\\&\text{multiplication par }\lambda\in\R\end{cases}\end{aligned}

  2. V={fonctions f:[a;b]R}V=\{\text{fonctions }f:[a;b]\rightarrow\R\}

    L5_1.png (Spécifier un $f\in V$, c'est définir $f(t)\quad\forall t\in[a;b]$ !) - **Le "+" :** si $f,g\in V,f+g\in V$ est défini par $(f\textcolor{blue}{+}g)(t)\coloneqq f(t)\textcolor{red}{+}g(t),\forall t\in[a;b] $ > **Remarque :** > > $\textcolor{blue}{+}\text{ dans V}\\\textcolor{red}{+}\text{dans }\R$ - **La multiplication par un scalaire :** si $\lambda\in\R,f\in V,\lambda f\in V$ est défini par $(\lambda f)(t)\coloneqq\lambda f(t)$ Vérifions que $V$, muni de ces deux opérations, est un EV : $$ \left.\begin{aligned} &\bullet f+g=g+f\\ &\bullet f+(g+h)=(f+g)+h\\ &\bullet \lambda (f+g)=\lambda f+\lambda g\\ &\bullet (\lambda +\mu) f=\lambda f+\mu f \\ &\bullet \lambda (\mu f)=(\lambda\mu)f=\mu(\lambda f)\\ &\bullet 1f=f\\ \end{aligned}\right\}\text{OK}

    $$

  1. V={matrices 2×2 aˋ coefficients reˊels}={(abcd)a,b,c,dR}V=\{\text{matrices }2\times2\text{ à coefficients réels}\}=\begin{Bmatrix} \left. \begin{pmatrix} a&b\\ c&d \end{pmatrix}\right\vert a,b,c,d\in\R \end{Bmatrix}

    Le "+" : (abcd)+(efgh)(a+eb+fc+gd+h)\begin{pmatrix}a&b\\c&d\end{pmatrix}+\begin{pmatrix}e&f\\g&h\end{pmatrix}\coloneqq\begin{pmatrix}a+e&b+f\\c+g&d+h\end{pmatrix}

    La multiplication : λ(abcd)(λaλbλcλd)\lambda\begin{pmatrix}a&b\\c&d\end{pmatrix}\coloneqq\begin{pmatrix}\lambda a&\lambda b\\\lambda c&\lambda d\end{pmatrix}

    Exercice : vérifier que VV est un EV.

  2. V={suite de reˊels x=(xn)n1,xnR,n1}V=\{\text{suite de réels }\bm{x}=(x_n)_{n\geqslant1},x_n\in\R,\forall n\geqslant1\}

    Le "+" : Si x=(x1,x2,x3,)y=(y1,y2,y3,)x+y(x1+y1,x2+y2,x3+y3,)\begin{aligned}\bm{x}&=(x_1,x_2,x_3,\ldots)\\\bm{y}&=(y_1,y_2,y_3,\ldots)\\\bm{x}+\bm{y}&\coloneqq(x_1\textcolor{red}{+}y_1,x_2\textcolor{red}{+}y_2,x_3\textcolor{red}{+}y_3,\ldots)\end{aligned}

    Remarque : +\textcolor{red}{+} est le "plus" dans R\R.

    La multiplication : Si λR,λx(λx1,λx2,λx3,)\lambda\in\R,\\\lambda\bm{x}\coloneqq(\lambda x_1,\lambda x_2,\lambda x_3,\ldots)

    Exercice : vérifier que VV est un EV.

Intéressant : on peut importer toutes les notions vues jusqu'à présent pour Rn\R^n dans le cadre général d'un EV abstrait. Par exemple :

  • colinéarité

  • combinaisons linéaire

  • dépendance et indépendance linéaire

  • partie engendrée

Exercices

V={fonctions f:[0;1]R}V=\{\text{fonctions }f:[0;1]\rightarrow\R\}

Soit fVf\in V, défini par f(t)tt[0;1]f(t)\coloneqq t\quad\forall t \in[0;1]

Soit gVg\in V, défini par g(t)t2t[0;1]g(t)\coloneqq t^2\quad\forall t \in[0;1]

Question : {f;g}\{f;g\} est-elle une famille libre ou liée ?

On étudie l'équation

α1f+α2g=0V,α1,α2R\alpha_1f+\alpha_2g=0_V,\quad\alpha_1,\alpha_2\in\R

qui signifie :

t[0;1],α1f(t)+α2f(t)=0V(t)=0t[0;1],α1t+α2t2p(t), polynoˆme de degreˊ 2=0\begin{array}{lcr} \forall t\in[0;1],\quad&\underbrace{\alpha_1f(t)+\alpha_2f(t)=0_V(t)}&=0\\ \forall t\in[0;1],\quad&\underbrace{\alpha_1t+\alpha_2t^2}_{p(t), \text{ polynôme de degré }2}&=0 \end{array}

Rappel : si un polynôme p(t)=α0+α1t+α2t2++αntnp(t)=\alpha_0+\alpha_1t+\alpha_2t^2+\ldots+\alpha_nt^n s'annule partout (p(t)=0t dans un intervalle)(p(t)=0\quad\forall t\text{ dans un intervalle}), alors tous ses coefficients sont nuls : α0=α1==an=0\alpha_0=\alpha_1=\ldots=a_n=0.

α1=0\longrightarrow\alpha_1=0 et α2=0\alpha_2=0.

{f;g}\longrightarrow\{f;g\} est libre.

  1. V={fonctions f:RR}V=\{\text{fonctions }f:\R\rightarrow\R\} Soit fVf\in V, défini par f(t)7tRf(t)\coloneqq 7 \forall t \in\R Soit gVg\in V, défini par g(t)cos(2t)tRg(t)\coloneqq \cos(2t) \forall t \in\R Soit hVh\in V, défini par h(t)cos2(t)tRh(t)\coloneqq \cos^2(t) \forall t \in\R Question : {f;g;h}\{f;g;h\} est-elle libre ou liée ? On étudie l'équation α1f+α2g+α3h=0\alpha_1f+\alpha_2g+\alpha_3h=0 Remarquons que
    cos2(t)=1+cos(2t)2=1147+12cos(2t)tR\begin{aligned} \cos^2(t)&=\frac{1+\cos(2t)}{2}\\ &=\frac{1}{14}7+\frac{1}{2}\cos(2t)\quad\forall t\in\R \end{aligned}
    Donc h=114f+12g    {f;g;h}h=\frac{1}{14}f+\frac{1}{2}g\iff\{f;g;h\} est liée.

Sous-espaces vectoriels

Définition

Soit VV un EV. Un sous-ensemble WVW\sub V est un sous-espace vectoriel de VV, dit "SEV" si :

Note : en faisant des combinaisons linéaires dans WW, on reste dans WW.

Exemples

  1. V=R2,W={V=\R^2,\quad W=\{vecteurs de R2\R^2 dont l'extrémité est dans le carré de côté 22 centré à l'origine }\}

    WW n'est pas un SEV, car on peut "sortir" de WW en faisant des combinaisons linéaires de vecteurs de w1,w2W\vec{w}_1,\vec{w}_2\in W.

  2. V=R2,W={V=\R^2,\quad W=\{ une droite dirigée par un v\vec{v} passant par l'origine}\}

    WW est un SEV de VV.

  3. V=R3,W={V=\R^3,\quad W=\{ un plan contenant l'origine }\}.

    WW est un SEV de VV.

  4. V={fonctions f:[a;b]R}V=\{\text{fonctions }f:[a;b]\rightarrow\R\}.

    W{f:[a;b]R telles que f(a)=f(b)}W\coloneqq\{f:[a;b]\rightarrow\R\text{ telles que }f(a)=f(b)\}.

    WW est un SEV de VV car :

    • la fonction "0V"W"0_V"\in W, puisque 0V(a)=0V(b)=00_V(a)=0_V(b)=0

    • si f,gWf,g\in W, alors f+gWf+g\in W car (f+g)(a)=f(a)+g(a)fWgW=f(b)+g(b)=(f+g)(b)\begin{aligned}(f+g)(a)&=&f(a)+&g(a)\\ &&\downarrow_{f\in W} &\downarrow_{g\in W}\\ &=&f(b)+&g(b)=(f+g)(b) \end{aligned}

    • si fW,λRf\in W,\lambda \in \R, alors λfW\lambda f\in W car (λf)(a)=λf(a)=λf(b)=(λf)(b)fW\begin{aligned}(\lambda f)(a)=\lambda\cdot f(a)&=\lambda\cdot f(b)=(\lambda f)(b)\\ &\uparrow_{f\in W} \end{aligned}

Un moyen standard de construire des SEV est de former des combinaisons linéaires d'une famille de vecteurs choisis de VV.

Définition

Soit VV un EV, v1,v2,,vpVv_1,v_2,\ldots,v_p\in V on note

Vect{v1,v2,,vp}\text{Vect}\{v_1,v_2,\ldots,v_p\}

le sous-ensemble de VV contenant toutes les combinaisons linéaires possibles des v1,v2,,vpv_1,v_2,\ldots,v_p.

Lemme

WVect{v1,v2,,vp}VW\coloneqq\text{Vect}\{v_1,v_2,\ldots,v_p\}\sub V est un SEV de VV, appelé le sous-espace engendré par v1,v2,,vpv_1,v_2,\ldots,v_p.

Preuve

WVect{v1,v2,,vp}W\coloneqq\text{Vect}\{v_1,v_2,\ldots,v_p\} est bien un SEV de VV, puisque :

Exemples VOIR GEOGEBRA

  1. V={fonctions f:RR}V=\{\text{fonctions }f:\R\rightarrow\R\}

    Soit fVf\in V, défini par f(t)ttRf(t)\coloneqq t\quad\forall t \in\R

    Soit gVg\in V, défini par g(t)t2tRg(t)\coloneqq t^2\quad\forall t \in\R

    Soit WVect{f;g}VW\coloneqq\text{Vect}\{f;g\}\sub V

    WW est l'ensemble des polynômes de degré au plus 22 qui prennent la valeur 00 en t=0t=0.

  2. V={fonctions f:RR}V=\{\text{fonctions }f:\R\rightarrow\R\}

    Soit fVf\in V, défini par f(t)sin(t)tRf(t)\coloneqq \sin(t) \forall t \in\R

    Soit gVg\in V, défini par g(t)cos(2t)tRg(t)\coloneqq \cos(2t) \forall t \in\R

    Remarque : {f;g}\{f;g\} est libre.

    Soit WVect{f;g}W\coloneqq\text{Vect}\{f;g\}

Les polynômes, exemple d'espace vectoriel

Définition

P{polynoˆmes tp(t) aˋ coefficients reˊels}\mathbb{P}\coloneqq\{\text{polynômes }t\mapsto p(t)\text{ à coefficients réels}\}

Par exemple, p(t)=1+t+2t2Pp(t)=1+t+2t^2\in\mathbb{P}

Exercice 1

Montrer que P\mathbb{P}, muni d'un "++" et d'une multiplication par des scalaires, est un EV.

Exercice 2

Si p1(t)1+2t3,p2(t)=2+t3t2,p3(t)=t+2t2t3p_1(t)\coloneqq1+2t^3,p_2(t)=2+t-3t^2,p_3(t)=-t+2t^2-t^3, est-ce que la famille {p1,p2,p3}\{p_1,p_2,p_3\} est libre ou liée ?

Définition

Pn{polynoˆmes pP de degreˊ au plusn}\mathbb{P}_n\coloneqq\{\text{polynômes }p\in\mathbb{P}\text{ de degré }\color{red}\underline{\color{black}\text{au plus}}\color{black}\,n\}

Pn\mathbb{P}_n est un SEV de P\mathbb{P}.

Applications linéaires

Soient V,VV,V' deux espaces vectoriels. On considère une application de VV dans VV', T:VVT:V\rightarrow V', c'est-à-dire une règle qui associe à chaque vVv\in V un (et un seul) vVv'\in V', noté v:T(v)v':T(v).

Définitions

Exemples

  1. V=vecteurs de RnV=\text{vecteurs de }\R^n, avec le "++",

    V=RV'=\R

    T:VVT:V\rightarrow V'

    v=(v1vn)t(v)=v1v'=\begin{pmatrix}v_1\\\vdots\\v_n\end{pmatrix}\mapsto t(\vec{v})=v_1

    Im(T)=V=R\text{Im}(T)=V'=\R, donc TT est surjective.

    TT n'est pas injective, car si v,w\vec{v},\vec{w} sont des vecteurs différents de Rn\R^n, avec v1=w1v_1=w_1. Alors T(v)=T(w)T(\vec{v})=T(\vec{w}), mais vw\vec{v}\ne\vec{w}.

  2. V=RnT:VVV=RmvT(v)Av,\begin{array}{lll} V=\R^n &T:&V\rightarrow V'\\ V'=\R^m&&\vec{v}\mapsto T(\vec{v})\coloneqq A\vec{v},\end{array}

    AA est une matrice m×nm\times n fixée.

Définition

Une application T:VVT:V\rightarrow V' est linéaire si

  1. v,wV,T(v+w)=T(v)+T(w)\forall v,w\in V,\quad T(v+w)=T(v)+T(w)

  2. vV,λR,T(λv)=λT(v)\forall v\in V,\forall\lambda\in\R,\quad T(\lambda v)=\lambda T(v)

De manière équivalente, T:VVT:V\rightarrow V' est linéaire si

T(λ1v1+λ2v2++λkvk)=λ1T(v1)+λ2T(v2)++λkT(vk)(λ1v1,λ2v2,,λkvkR,v1,v2,,vkV)\begin{aligned} &\boxed{T(\lambda_1v_1+\lambda_2v_2+\ldots+\lambda_kv_k)=\lambda_1T(v_1)+\lambda_2T(v_2)+\ldots+\lambda_kT(v_k)}\\ &(\forall\lambda_1v_1,\lambda_2v_2,\ldots,\lambda_kv_k\in\R,\forall v_1,v_2,\ldots,v_k\in V) \end{aligned}

Exemples

  1. T:RRxT(x)x2 n’est pas lineˊaire.\begin{aligned}T:\R&\rightarrow\R\\x&\mapsto T(x)\coloneqq x^2\text{ n'est pas linéaire.}\end{aligned}

    Par exemple, T(1)=1T(2)=4T(1+2)=T(3)=9\begin{aligned}&T(1)=1\\&T(2)=4\\&T(1+2)=T(3)=9\end{aligned}, donc T(1+2)T(1)+T(2)T(1+2)\ne T(1)+T(2).

  2. Le même qu'avant : T:RnRv=(v1vn)T(v)=v1\begin{aligned}T:\R^n&\rightarrow\R\\&\vec{v}=\begin{pmatrix}v_1\\\vdots\\v_n\end{pmatrix}&\mapsto T(\vec{v})=v_1\end{aligned},

    est linéaire car :

    v=(v1vn),w=(w1wn)T(v+w)=T((v1+w1vn+wn))=v1+w1=T(v)+T(w)λR,T(λv)=T((λv1λvn))=λv1=λT(v)\begin{array}{ll} &\begin{array}{lll} \bullet&\forall\vec{v}=\begin{pmatrix}v_1\\\vdots\\v_n\end{pmatrix}, \vec{w}=\begin{pmatrix}w_1\\\vdots\\w_n\end{pmatrix}\\ &\begin{aligned} T(\vec{v}+\vec{w})&=T\left(\begin{pmatrix}v_1+w_1\\\vdots\\v_n+w_n\end{pmatrix}\right)\\ &=v_1+w_1\\ &=T(\vec{v})+T(\vec{w}) \end{aligned} \end{array} &\text{}\\ &\begin{array}{ll}\bullet\quad\forall\lambda\in\R,\quad T(\lambda\vec{v})&=T\left(\begin{pmatrix}\lambda v_1\\\vdots\\\lambda v_n\end{pmatrix}\right)\\ &=\lambda v_1\\ &=\lambda T(\vec{v}) \end{array} \end{array}

  3. T:RnRmvT(v)A(v) est lineˊaire (voir plus haut).\begin{aligned}T:\R^n&\rightarrow\R^m\\\vec{v}&\mapsto T(\vec{v})\coloneqq A(\vec{v})\text{ est linéaire (voir plus haut).}\end{aligned}

  4. V={fonctions f:[0;1]R continues}V=\{\text{fonctions }f:[0;1]\rightarrow\R\text{ continues}\} (VV est un EV).

    V=RT:VVfT(f)f(0) est lineˊaire.V=R2T:VVfT(f)(01/2f(x)dx1/21f(x)dx) est lineˊaire.\begin{array}{ll}\bullet \quad V'=\R\\&\begin{aligned}T:V&\rightarrow V'\\f&\mapsto T(f) \coloneqq f(0) \text{ est linéaire}.\end{aligned}\\\bullet \quad V'=\R^2\\&\begin{aligned}T:V&\rightarrow V'\\f&\mapsto T(f) \coloneqq \begin{pmatrix}\int^{1/2}_{0}f(x)\mathrm{d}x\\\text{}\\\int^{1}_{1/2}f(x)\mathrm{d} x\end{pmatrix}\text{ est linéaire}.\end{aligned}\end{array}

Remarques

  1. Si T:VVT:V\rightarrow V' est linéaire,

    T(v+w)=T(v)+T(w)T(v\textcolor{red}{+}w)=T(v)\textcolor{blue}{+}T(w).

    Note :

    Le "+\color{red}+" dans VV !

    Le "+\color{blue}+" dans VV' !

  2. Si T:VVT:V\rightarrow V' est linéaire, alors

    T(0)=0T(\textcolor{red}{0})=\textcolor{blue}{0'}

    Note :

    Le "0\color{red}0" dans VV !

    Le "0\color{blue}0" dans VV' !

    En effet, T(0)=T(vv)ouˋ vVT lineˊaire=T(v)T(v)=0!\begin{aligned}T(0)&=T(v-v)\quad \text{où }v \in V\\ &\Big\downarrow{T\text{ linéaire}}\\&=T(v)-T(v)=0!\end{aligned}

Noyau (et image)

Définition

Soit T:VVT:V\rightarrow V' linéaire. Le noyau de TT est Ker(T){vVT(v)=0}V\mathrm{Ker}(T)\coloneqq\{v\in V\vert T(v)=0'\}\sub V.

Rappel : Ker(T)\mathrm{Ker}(T) n'est jamais vide : 0Ker(T)0\in\mathrm{Ker}(T).

Lemme

Soit T:VVT:V\rightarrow V' linéaire. Alors

  1. Ker(T)\mathrm{Ker}(T) est un SEV de VV. (Ker(T)V)\quad (\mathrm{Ker}(T)\sub V).

  2. Im(T)\mathrm{Im}(T) est un SEV de VV'. (Im(T)V)\quad (\mathrm{Im}(T)\sub V').

Preuve de 1

Soient v,wKer(T)v,w\in\mathrm{Ker}(T). Alors

v,wKer(T)T(v+w)=T(v)+T(w)=0+0=0Donc v+wKer(T).\begin{array}{lll}& &_{v,w\in\mathrm{Ker}(T)}&\\&&\curvearrowright\\\bullet &T(v+w)\quad= T(v)+T(w)&=0'+0'=0'&\\&\text{Donc }v+w\in\mathrm{Ker}(T).&&\end{array}

λR,T(λv)=λT(v)=λ0=0,Donc λvKer(T).\begin{array}{llll}\bullet&\forall\lambda\in\R,\quad T(\lambda v)=\lambda T(v)=\lambda0'=0',\\&\text{Donc }\lambda v\in\mathrm{Ker}(T).\end{array}

Comme 0Ker(T),on a bien que Ker(T) est un SEV de V.\begin{array}{lr} \text{Comme }0\in\mathrm{Ker}(T), \text{on a bien que }\mathrm{Ker}(T)\text{ est un SEV de }V.&\\ &\square \end{array}

Preuve de 2

À faire en exercice.

Exemple

T:R3Rv=(v1v2v3)T(v)v1\begin{aligned}T:\,&\R^3\rightarrow\R\\&\vec{v}=\begin{pmatrix}v_1\\v_2\\v_3\end{pmatrix}\mapsto T(\vec{v})\coloneqq v_1\end{aligned}

Calculons Ker(T)={v=(v1v2v3)R3v1=0}={(0v2v3)v2,v3 libres}={λ(010)v1+μ(001)v2λ,μ libres}=Vect{v1,v2}.\begin{aligned}\mathrm{Ker}(T)&=\left\{\vec{v}=\left.\begin{pmatrix}v_1\\v_2\\v_3\end{pmatrix}\in\R^3\right\vert v_1=0\right\}\\&=\left\{\left.\begin{pmatrix} 0\\v_2\\v_3\end{pmatrix}\right\vert v_2,v_3\text{ libres}\right\}\\ &=\left\{\lambda\underbrace{\begin{pmatrix}0\\1\\0\end{pmatrix}}_{\vec{v}_1}+\mu\left.\underbrace{\begin{pmatrix}0\\0\\1\end{pmatrix}}_{\vec{v}_2}\right\vert\lambda,\mu\text{ libres}\right\}\\ &=\mathrm{Vect}\{\vec{v}_1,\vec{v}_2\}. \end{aligned}

Description du noyau et de l'image d'applications linéaires

T:RnRmvT(v)Ax\begin{aligned}T:\,&\R^n\rightarrow\R^m\\&\vec{v}\mapsto T(\vec{v})\coloneqq A\vec{x}\end{aligned}

Ker(T)={xRnAx=0}=Ker(A)Im(T)={yRmxRn tel que Ax=y}={yRmy est combinaison lineˊaire des colonnes de A=[a1,a2,,an]}=Vect{a1,a2,,an}=Col(A) ou Im(A)\begin{aligned}\mathrm{Ker}(T)&=\{\vec{x}\in\R^n\vert A\vec{x}=\vec{0}\}=\mathrm{Ker}(A)\\ \mathrm{Im}(T)&=\{y\in\R^m\vert\exist\vec{x}\in\R^n\text{ tel que }A\vec{x}=\vec{y}\}\\ &=\{y\in\R^m\vert\vec{y}\text{ est combinaison linéaire des colonnes de }A=[\vec{a}_1,\vec{a}_2,\ldots,\vec{a}_n]\}\\ &=\mathrm{Vect}\{\vec{a}_1,\vec{a}_2,\ldots,\vec{a}_n\}\\ &=\mathrm{Col}(A)\text{ ou }\mathrm{Im}(A) \end{aligned}

Comme Ker(A),Im(A)\mathrm{Ker}(A),\mathrm{Im}(A) sont des SEV, on aimerait pouvoir les décrire comme étant engendrés par des familles finies de vecteurs particuliers.

Exemple

Calculons Ker(A)\mathrm{Ker}(A), où

A=(361+171223124584)A=\left(\begin{array}{rrrrr}-3&6&-1&\textcolor{white}{+}1&-7\\ 1&-2&2&3&-1\\ 2&-4&5&8&-4 \end{array}\right)

Ax=0matrice augmenteˊe reˊduction(12013000+1220000000)\begin{array}{rl}A\vec{x}=\vec{0}&\Longrightarrow\text{matrice augmentée}\\ &\qquad\qquad\quad\Downarrow_{\text{ réduction}}\\ &\left(\begin{array}{rrrrr:r}\textcolor{blue}{1}&-2&0&-1&3&0\\ 0&0&\textcolor{white}{+}\textcolor{blue}{1}&2&-2&0\\ 0&0&0&0&0&0 \end{array}\right) \end{array}

x2,x4,x5x_2,x_4,x_5 sont libres.

Ker(A)={x=λ(21000)v1+μ(10210)v2+ν(30201)v3λ,μ,ν libres}=Vect{v1,v2,v3}\begin{aligned}\rightarrow\mathrm{Ker}(A) &=\left\{\vec{x}=\lambda\left.\underbrace{\begin{pmatrix}2\\1\\0\\0\\0\end{pmatrix}}_{\vec{v}_1}+\mu\underbrace{\begin{pmatrix}1\\0\\-2\\1\\0\end{pmatrix}}_{\vec{v}_2}+\nu\underbrace{\begin{pmatrix}-3\\0\\2\\0\\1\end{pmatrix}}_{\vec{v}_3}\right\vert\lambda,\mu,\nu\text{ libres} \right\}\\ &=\mathrm{Vect}\{\vec{v}_1,\vec{v}_2,\vec{v}_3\} \end{aligned}

Remarques :

  • {v1,v2,v3}\{\vec{v}_1,\vec{v}_2,\vec{v}_3\} sont linéairement indépendants !

  • Le nombre de vecteurs qui sont nécessaires pour engendre le Ker(A)\mathrm{Ker}(A) est égal au nombre de variables libres dans Ax=0A\vec{x}=\vec{0}. En général, Ker(A)\mathrm{Ker}(A) pourra toujours être engendré par kk vecteurs linéairement indépendants, où kk est le nombre de variables libres dans Ax=0A\vec{x}=\vec{0}.

Bases (de SEV et d'EV)

Définition

Soit WVW\sub V un SEV de VV, et B={v1,v2,,vp}W\mathcal{B}=\{v_1,v_2,\ldots,v_p\}\sub W une famille non-vide de vecteurs de WW.

On dit que B\mathcal{B} est une base de WW si

  1. B\mathcal{B} engendre W:W=Vect{v1,v2,,vp}W:W=\mathrm{Vect}\{v_1,v_2,\ldots,v_p\}, et

  2. B\mathcal{B} est libre.

Si B={v1,v2,,vp}\mathcal{B}=\{v_1,v_2,\ldots,v_p\} est une base de WW, prenons un wWw\in W. Puisque B\mathcal{B} engendre W,α1,α2,,αpRW,\exist\alpha_1,\alpha_2,\ldots,\alpha_p\in\R tels que w=α1v1+α2v2++αpvp()\boxed{w=\alpha_1v_1+\alpha_2v_2+\ldots+\alpha_pv_p}\quad(*)

Ces α1,α2,,αp\alpha_1,\alpha_2,\ldots,\alpha_p sont les composantes de ww relativement à B\mathcal{B}.

Lemme

Les α1,α2,,αp\alpha_1,\alpha_2,\ldots,\alpha_p (associés à ww via ()(*)) sont uniques !

Preuve

Supposons que w=α1v1+α2v2++αpvp et quew=β1v1+β2v2++βpvp \begin{aligned}w=\alpha_1v_1+\alpha_2v_2+\ldots+\alpha_pv_p &\text{ et que}\\ \color{red}\underline{\color{black}\textcolor{red}{-}\quad w=\beta_1v_1+\beta_2v_2+\ldots+\beta_pv_p}\text{ }\\ \end{aligned}

 0=(α1β1)v1++(αpβp)vp\qquad\qquad\qquad\qquad\text{ }0=(\alpha_1-\beta_1)v_1+\ldots+(\alpha_p-\beta_p)v_p

Comme {v1,v2,,vp}\{v_1,v_2,\ldots,v_p\} est libre (car B\mathcal{B} est une base), on a que

α1β1=0    α1=β1 αpβp=0    αp=βp\begin{array}{lr}\begin{aligned}\alpha_1&-\beta_1=0&\implies\alpha_1=\beta_1\\ &\text{ }\vdots&\\ \alpha_p&-\beta_p=0&\implies\alpha_p=\beta_p \end{aligned}\\ &\square \end{array}

Remarque : il est important de souligner que la composante αk(k{1,2,,p})\alpha_k\quad(k\in\{1,2,\ldots,p\}) est associées à vkv_k ! Dorénavant, nous ferons donc attention à indiquer l'ordre des vecteurs vkv_k dans B\mathcal{B} :

B={v1,v2,,vp}B=(v1,v2,,vp)\cancel{\mathcal{B}=\{v_1,v_2,\ldots,v_p\}} \Longrightarrow\mathcal{B}=(v_1,v_2,\ldots,v_p)

Ce qu'on construit, en introduisant les composantes de ww dans la base B=(v1,v2,,vp)\mathcal{B}=(v_1,v_2,\ldots,v_p), c'est une fonction :

[]B:WRpw[w]B=(α1αp)\begin{aligned}[\cdot]_{\mathcal{B}}:\,&W\rightarrow\R^p\\ &\textcolor{red}{w}\mapsto[w]_{\mathcal{B}}=\color{blue}\begin{pmatrix}\alpha_1\\\vdots\\\alpha_p\end{pmatrix} \end{aligned}

Notes :

w est un vecteur "abstrait" de W.(α1αp) est la repreˊsentation "concreˋte" de w dans une base.\begin{array}{lll}&\bullet&\textcolor{red}{w} \text{ est un vecteur "abstrait" de }W.\\ &\bullet&\textcolor{blue}{\begin{pmatrix}\alpha_1\\\vdots\\\alpha_p\end{pmatrix}}\text{ est la représentation "concrète" de }w \text{ dans une base.}\end{array}

Exemple

AA, la matrice (3×5)(3\times5) du mardi. On étudie

T:R5R3xT(x)Ax\begin{aligned} T:\,&\R^5\rightarrow\R^3\\ &\vec{x}\mapsto T(\vec{x})\coloneqq A\vec{x} \end{aligned}

Ker(A)\mathrm{Ker}(A) est un SEV de R5\R^5 et on a vu que

Ker(A)=Vect{v1,v2,v3}\mathrm{Ker}(A)=\mathrm{Vect}\{\vec{v}_1,\vec{v}_2,\vec{v}_3\}, donc {v1,v2,v3}\{\vec{v}_1,\vec{v}_2,\vec{v}_3\} engendre Ker(A)\mathrm{Ker}(A). De plus, est est libre, donc B=(v1,v2,v3)\mathcal{B}=(\vec{v}_1,\vec{v}_2,\vec{v}_3) est une base de Ker(A)\mathrm{Ker}(A).

Si xKer(A)\vec{x}\in\mathrm{Ker}(A) s'écrit x=λv1+μv2+νv3\vec{x}=\lambda\vec{v}_1+\mu\vec{v}_2+\nu\vec{v}_3, on a donc

[x]B=(λμν)[\vec{x}]_{\mathcal{B}}=\begin{pmatrix}\lambda\\\mu\\\nu\end{pmatrix}

Ainsi, même si xR5\vec{x}\in\R^5, pour décrire Ker(A)\mathrm{Ker}(A) on n'a besoin que de 3 composantes !

Définition

(Cas W=VW=V) : on dit que B=(v1,,vn)V\mathcal{B}=(v_1,\ldots,v_n)\sub V est une base de VV si

  1. B\mathcal{B} engendre VV :

    V=Vect{v1,,vp}V=\mathrm{Vect}\{v_1,\ldots,v_p\}
  2. B\mathcal{B} est libre.

Exemples

  1. V=Rn:si xRnV=\R^n:\text{si }\vec{x}\in\R^n,

    x=(x1x2x3xn)=x1(1000)+x2(0100)++xn(0001)\vec{x}=\begin{pmatrix}x_1\\x_2\\x_3\\\vdots\\x_n\end{pmatrix}=x_1\begin{pmatrix}1\\0\\0\\\vdots\\0\end{pmatrix}+x_2\begin{pmatrix}0\\1\\0\\\vdots\\0\end{pmatrix}+\ldots+x_n\begin{pmatrix}0\\0\\0\\\vdots\\1\end{pmatrix}

    Donc B(e1,e2,,en)\mathcal{B}\coloneqq(\vec{e}_1,\vec{e}_2,\ldots,\vec{e}_n)

    engendre Rnest libre.}B est une base, appeleˊbase canonique de Rn\left.\begin{array}{l}\quad\bullet\quad\text{engendre }\R^n\\ \quad\bullet\quad\text{est libre.} \end{array}\right\}\mathcal{B}\text{ est une base, appelée }\textbf{base canonique }\text{de }\R^n

    Remarque : dans ce cas,

    [x]B=(x1xn)[\vec{x}]_{\mathcal{B}}=\begin{pmatrix}x_1\\\vdots\\x_n\end{pmatrix}
  2. Pn={polynoˆmes p, de degreˊ au plus eˊgal aˋ n}\mathbb{P}_n=\{\text{polynômes }p,\text{ de degré au plus égal à }n\}

    Si pP,p(t)=a0+a1t+a2t2++antnp\in\mathbb{P},p(t)=a_0+a_1t+a_2t^2+\ldots+a_nt^n

    Soit e0(t)1e0Pne1(t)te1Pnen(t)tnenPn\begin{aligned}e_0&(t)\coloneqq1&\rightarrow e_0\in\mathbb{P}_n\\ e_1&(t)\coloneqq t&\rightarrow e_1\in\mathbb{P}_n\\ &\vdots\\ e_n&(t)\coloneqq t^n&\rightarrow e_n\in\mathbb{P}_n \end{aligned}

    Donc tout pPnp\in\mathbb{P}_n peut s'écrire comme combinaison linéaire des eke_k : p=a0e0+a1e1++anenp=a_0e_0+a_1e_1+\ldots+a_ne_n.

    Donc {e0,e1,,en}\{e_0,e_1,\ldots,e_n\} engendre Pn\mathbb{P}_n.

    De plus, elle est libre. En effet, supposons que α0e0++αnen=0\alpha_0e_0+\ldots+\alpha_ne_n=\color{red}{0}

    Remarque : il s'agit du 0\color{red}{0} de Pn\mathbb{P}_n, c'est-à-dire le polynôme nul.

    signifie que ta0e0(t)+a1e1(t)++anen(t)=0\forall t\quad a_0e_0(t)+a_1e_1(t)+\ldots+a_ne_n(t)=\textcolor{blue}{0}

    Remarque : il s'agit du 0\color{blue}{0} de R\R.

    donc tα0+α1t+α2t2++αntn=0\forall t\quad\alpha_0+\alpha_1t+\alpha_2t^2+\ldots+\alpha_nt^n=0

    Par le lemme de la semaine dernière, ceci implique

    α0=α1=α2==αn=0\alpha_0=\alpha_1=\alpha_2=\ldots=\alpha_n=0

    {e0,,en}\Longrightarrow\{e_0,\ldots,e_n\} est libre.

    Donc B=(e0,e1,,en)\mathcal{B}=(e_0,e_1,\ldots,e_n) est une base de Pn\mathbb{P}_n, appelé base canonique de Pn\mathbb{P}_n.

  3. Dans R2\R^2,Bcan=((10),(01))est une base canoniqueB=((10)v1,(11)v2)est aussi une base\text{}\\\begin{aligned}&\bullet\quad\mathcal{B}_{\mathrm{can}}=\left(\begin{pmatrix}1\\0\end{pmatrix},\begin{pmatrix}0\\1\end{pmatrix}\right)\text{est une base canonique}\\ &\bullet\quad\mathcal{B}=\left(\underbrace{\begin{pmatrix}1\\0\end{pmatrix}}_{\textcolor{red}{\vec{v}_1}},\underbrace{\begin{pmatrix}1\\1\end{pmatrix}}_{\textcolor{red}{\vec{v}_2}}\right)\text{est aussi une base} \end{aligned}

    $\text{Soit }\vec{b}=\begin{pmatrix}b_1\\b_2\end{pmatrix}\text{ un vecteur quelconque de }\R^n, \text{ce qui signifie que } [\vec{b}]_{\mathcal{B}_{\mathrm{can}}}=\begin{pmatrix}b_1\\b_2\end{pmatrix}.$ 
    

    Question :

    [b]B=(α1α2)[\vec{b}]_{\mathcal{B}}=\begin{pmatrix}\alpha_1\\\alpha_2\end{pmatrix}

    On cherche α1,α2\alpha_1,\alpha_2 tels que b=α1v1+α2v2\vec{b}=\alpha_1\vec{v}_1+\alpha_2\vec{v}_2

    (b1b2)=α1(10)+α2(11)\begin{pmatrix}b_1\\b_2\end{pmatrix}=\alpha_1\begin{pmatrix}1\\0\end{pmatrix}+\alpha_2\begin{pmatrix}1\\1\end{pmatrix}

    b1=α1+α2b2=α2}{α1=b1b2α2=b2\left.\begin{aligned}&b_1=\alpha_1+\alpha_2\\ &b_2=\alpha_2 \end{aligned}\right\}\Rightarrow\begin{cases} \alpha_1=b_1-b_2\\ \alpha_2=b_2 \end{cases}

    Donc [b]B=(α1α2)\text{Donc }[\vec{b}]_{\mathcal{B}}=\begin{pmatrix}\alpha_1\\\alpha_2\end{pmatrix}

    B=((11),(33))n’est pas une baseB((11)v1,(33)v2,(23)v3) n’est pas une base,\begin{aligned}&\bullet\quad\mathcal{B}=\left(\begin{pmatrix}1\\1\end{pmatrix},\begin{pmatrix}3\\3\end{pmatrix}\right)\text{n'est pas une base} \\ &\bullet\quad\mathcal{B}\left(\underbrace{\begin{pmatrix}1\\1\end{pmatrix}}_{\vec{v}_1},\underbrace{\begin{pmatrix}3\\3\end{pmatrix}}_{\vec{v}_2},\underbrace{\begin{pmatrix}2\\3\end{pmatrix}}_{\vec{v}_3}\right) \text{ n'est pas une base,} \end{aligned}

    puisque v3=3v2v1\vec{v}_3=3\vec{v}_2-\vec{v}_1. Remarquons pourtant que si on retire v3\vec{v}_3 de B,on obtient B=((10),(11))qui est une base.\mathcal{B}, \text{on obtient }\mathcal{B}'=\left(\begin{pmatrix}1\\0\end{pmatrix},\begin{pmatrix}1\\1\end{pmatrix}\right)\text{qui est une base.}

Théorème

Soit WW un SEV d'un EV VV, et {v1,,vp}\{v_1,\ldots,v_p\} une famille qui engendre WW.

Exercice

Soit AA, la matrice (3×5)(3\times5) du mardi. Utilisons le théorème ci-dessus pour calculer Im(A)\mathrm{Im}(A) :

A=(361+171223124584)=[a1a2a3a4a5]A=\left(\begin{array}{rrrrr}-3&6&-1&\textcolor{white}{+}1&-7\\ 1&-2&2&3&-1\\ 2&-4&5&8&-4 \end{array}\right)= \left[\vec{a}_1\,\vec{a}_2\,\vec{a}_3\,\vec{a}_4\,\vec{a}_5\right]

Im(A)=Vect{a1,a2,,a5}un SEV WR3a2=2a1=Vect{a1,a3,a4,a5}a4=2a3a1=Vect{a1,a3,a5}a5=3a12a3=Vect{a1,a3}libre, engendre Im(A)\begin{array}{lll}\rightarrow\mathrm{Im}(A)&=\mathrm{Vect}\{\vec{a}_1,\vec{a}_2,\ldots,\vec{a}_5\}&\leftarrow\text{un SEV }W\sub\R^3\\ & &\downarrow\vec{a}_2=-2\vec{a}_1\\ &=\mathrm{Vect}\{\vec{a}_1,\vec{a}_3,\vec{a}_4,\vec{a}_5\}\\ & &\downarrow\vec{a}_4=2\vec{a}_3-\vec{a}_1\\ &=\mathrm{Vect}\{\vec{a}_1,\vec{a}_3,\vec{a}_5\}\\ & &\downarrow\vec{a}_5=3\vec{a}_1-2\vec{a}_3\\ &=\mathrm{Vect}\{\vec{a}_1,\vec{a}_3\}&\leftarrow\text{libre, engendre }\mathrm{Im}(A)\end{array}

Donc B=(a1,a3)\mathcal{B}=\left(\vec{a}_1,\vec{a}_3\right) est une base de Im(A)\mathrm{Im}(A).

Question : y a-t-il une méthode qui permette de trouver les dépendances linéaires entre les colonnes d'une matrice ? Oui, réponse ci-dessous.

Remarque : la réduite de AA :

(1201300+12200000)\left(\begin{array}{rrrrr} \textcolor{blue}{1}&-2&0&-1&3\\ 0&0&\textcolor{white}{+}\textcolor{blue}{1}&2&-2\\ 0&0&0&0&0 \end{array}\right)

Les colonnes 11 et 22 sont celles qui contiennent des pivots\textcolor{blue}{\text{pivots}}, et aussi celles qui forment une base de Im(A)\mathrm{Im}(A).

Théorème

Les dépendances linéaires existant entre les colonnes d'une matrice sont les même que celles existant entre les colonnes de sa réduite.

Plus précisément, si

A=[a1an]reˊduireA=[r1rn]A=\left[\vec{a}_1\cdots\,\vec{a}_n\right]\stackrel{\text{réduire}}{\rightsquigarrow}A'=\left[\vec{r}_1\cdots\,\vec{r}_n\right]

et si F{a1,,an}\mathcal{F}\sub\left\{\vec{a}_1,\ldots,\vec{a}_n\right\}, et F{r1,,rn}\mathcal{F}'\sub \left\{\vec{r}_1,\ldots,\vec{r}_n\right\} est la famille correspondante de colonnes de AA', alors

F lieˊe    F lieˊeF libre    F libre\quad\begin{array}{lll}&\mathcal{F}\text{ liée}&\iff\mathcal{F}'\text{ liée}\\ &\mathcal{F}\text{ libre}&\iff\mathcal{F}'\text{ libre} \end{array}

Preuve

Si F={ai1,,ain}\mathcal{F}=\{\vec{a}_{i_{1}},\ldots,\vec{a}_{i_{n}}\}, F={ri1,,rin}\mathcal{F}'=\{\vec{r}_{i_{1}},\cdots,\vec{r}_{i_{n}}\}, alors αi1ai1++αikaik=0αi1ri1++αikrik=0}car ces deux systeˋmes sont eˊquivalents !\begin{aligned}\left.\begin{array}{llll}&\alpha_{i_{1}}\vec{a}_{i_{1}}+&\ldots&+\alpha_{i_{k}}\vec{a}_{i_{k}}=\vec{0}\\ & &\,\,\Updownarrow\\ &\alpha_{i_{1}}\vec{r}_{i_{1}}+&\ldots&+\alpha_{i_{k}}\vec{r}_{i_{k}}=\vec{0} \end{array}\right\}\text{car ces deux systèmes sont équivalents !}\\ &\square\end{aligned}

Définition

Une colonne d'une matrice A(m×n)A\,(m\times n) est une colonne-pivot si elle devient, après réduction vers AA', une colonne qui contient un pivot.

Exemple

A=(354324618)A=(10430+10000)A=\left(\begin{array}{rrr}3&5&-4\\-3&-2&4\\6&1&-8\end{array}\right)\rightsquigarrow A'=\left(\begin{array}{rrr}\textcolor{blue}{1}&0&-\frac{4}{3}\\0&\textcolor{white}{+}\textcolor{blue}{1}&0\\0&0&0\end{array}\right)

Ici, les colonnes 11 et 22 sont des colonnes-pivot\textcolor{blue}{\text{colonnes-pivot}}.

Théorème

Les colonnes-pivot d'une matrice AA forment une base de Im(A)\mathrm{Im}(A).

Preuve

Dans le réduite de AA' :

Par le théorème précédent, ceci implique que les colonnes-pivot de AA sont indépendantes, et que les autres sont des combinaisons linéaires des colonnes-pivot, et donc {colonnes pivots}\{\text{colonnes pivots}\} est une base de Im(A).\begin{aligned}\mathrm{Im}(A).\\&\square\end{aligned}

Rappel : Si B=(v1,,vn)\mathcal{B}=(v_1,\ldots,v_n) est une base de VV, alors tout vVv\in V peut s'écrire

α1v1++αnvn,,ouˋ les α1v1++αnvnR sont uniques.&[,,]B:VRn\quad\boxed{\alpha_1v_1+\ldots+\alpha_nv_n},, \text{où les }\alpha_1v_1+\ldots+\alpha_nv_n\in\R\text{ sont uniques.} \&[,\cdot,]_\mathcal{B}:V\rightarrow\R^n

[]B:VRnv[v]B(α1α2αn)\begin{aligned}[\,\cdot\,]_\mathcal{B}:\,&V\rightarrow\R^n\\ &v\mapsto[v]_\mathcal{B}\coloneqq\begin{pmatrix}\alpha_1\\\alpha_2\\\vdots\\\alpha_n\end{pmatrix} \end{aligned}

Exercice : montrez que v[v]Bv\mapsto[v]_\mathcal{B} est linéaire.

Lemme

Si B=(v1,,vn)\mathcal{B}=(v_1,\ldots,v_n) est d'un EV VV, et si F={w1,w2,,wp}\mathcal{F}=\{w_1,w_2,\ldots,w_p\} est une famille de vecteurs de VV, avec p>np>n. Alors F\mathcal{F} est liée.

Preuve

Supposons que

λ1w1+λ2w2++λpwp=0()0[]B et lineˊariteˊλ1[w1]B++λp[wp]B=[0]B=0()1\begin{array}{ll} &\lambda_1w_1+\lambda_2w_2+\ldots+\lambda_pw_p=0\quad(*)_0\\ [\,\cdot\,]_\mathcal{B} \text{ et linéarité}\big\downarrow&\\ &\lambda_1[w_1]\mathcal{B}+\ldots+\lambda_p[w_p]\mathcal{B}=[\textcolor{blue}{0}]_\mathcal{B}=\vec{\textcolor{red}{0}}\quad(*)_1 \end{array}

Note :

"Zeˊro\textcolor{blue}{\text{Zéro}}" de VV.

"Zeˊro\textcolor{red}{\text{Zéro}}" de Rn\R^n.

On sait que dans Rn\R^n, une famille de plus de nn vecteurs est nécessairement liée. Il existe donc des coefficients λ1,,λp\lambda_1,\ldots,\lambda_p, pas tous nuls, pour lesquels ()1(*)_1 est vérifiée. Avec ces λi,()0\lambda_i,(*)_0 est aussi vérifiée, et donc F={w1,,wp} est lieˊe.\mathcal{F}=\{w_1,\ldots,w_p\}\text{ est liée.}\qquad\square

Corollaire

Toutes les bases d'un même EV contiennent le même nombre d'éléments.

Preuve

Supposons que B=(v1,,vn)\mathcal{B}=(v_1,\ldots,v_n) et B=(v1,,vm)\mathcal{B}'=(v_1',\ldots,v_m') soient deux bases d'un même EV. Comme B\mathcal{B}' est une base, {v1,,vm}\{v_1',\ldots,v_m'\} est libre, et donc par le lemme précédent, si m>nm>n, elle serait liée. Et donc mnm\leqslant n. De manière équivalente, nmn\leqslant m. Donc m=n.\begin{aligned}m=n.\\ &\qquad\qquad\qquad\qquad\qquad\qquad\qquad\qquad\qquad\qquad\qquad\qquad\qquad\qquad\quad\square\end{aligned}

Définition

Si un EV VV possède une base qui contient un nombre nn d'éléments (alors toute autre base possède aussi nn éléments par le corollaire), on appelle nn la dimension de VV. On note :

n=dim(V)\boxed{n=\mathrm{dim}(V)}

Exemples

  1. Dans R2,Bcan=((10)(01))dim(R2)=2(Plus geˊneˊralement : dim(Rn)=n)\R^2,\\\mathcal{B}_{\mathrm{can}}=\left(\begin{pmatrix}1\\0\end{pmatrix}\begin{pmatrix}0\\1\end{pmatrix}\right)\\ \text{}\\ \rightarrow\mathrm{dim}(\R^2)=2\\ \text{}\\ (\text{Plus généralement : }\mathrm{dim}(\R^n)=n)

  2. Dans P2,B=(e1,e2,e3) est une basedim(P2)=3\mathbb{P}_2,\\ \text{}\\ \mathcal{B}=(e_1,e_2,e_3)\text{ est une base}\\ \text{}\\ \mathrm{dim}(\mathbb{P}_2)=3

  3. Il existe un EV de dimension infinie :

    Par exemple, C([a,b])={f:[a,b]R continues}dim(C1([a,b]))=\mathcal{C}\left([a,b]\right)=\{f:[a,b]\rightarrow\R\text{ continues}\}\\ \text{}\\ \mathrm{dim}(\mathcal{C}^1([a,b]))=\infty

Théorème de la base complète

Dans un EV de dimension finie VV, avec un SEV WW, si F={v1,,vk}W\mathcal{F}=\{v_1,\ldots,v_k\}\sub W est libre, alors F\mathcal{F} peut être complétée :

F={v1,,vk,v~k+1,,v~p},ouˋ v~jW,\mathcal{F}'=\{v_1,\ldots,v_k,\tilde{v}_{k+1},\ldots,\tilde{v}_p\},\text{où }\tilde{v}_j\in W,

tel que F\mathcal{F}' soit une base de WW.

Exemple (pour comprendre l'idée de la preuve)

V=R3,W=un plan (qui contient l’origine),F={v1}V=\R^3,W=\text{un plan (qui contient l'origine),}\\ \mathcal{F}=\{\vec{v}_1\}

En prenant n'importe quel v~2W\vec{\tilde{v}}_2\in W qui n'est pas colinéaire à v1\vec{v}_1, on obtient F={v1,v~2}\mathcal{F}=\{\vec{v}_1,\vec{\tilde{v}}_2\} qui est libre, et qui engendre WW.

Théorème

Dans un EV VV de dimension nn, toute famille libre de nn vecteurs forme une base de VV.