Isométries vectoriels

Isométrie vectoriel en dimension 3

($\star$) Correction

Déterminer la nature de l'endomorphisme $f$ de $\R^3$, dont la matrice dans la base canonique de $\R^3$ est donnée ci-après, et préciser les éléments caractéristique de $f$ $$\mathbf{a)\,}\,\,\,\dfrac{-1}{27}\begin{pmatrix} 2&-26&7\\-23&2&14\\14&7&22 \end{pmatrix},\quad\mathbf{b)\,}\,\,\,\dfrac{1}{9}\begin{pmatrix} -7&4&-4\\ 4&-1&-8\\ -4&-8&-1 \end{pmatrix},\quad$$ $$\mathbf{c)\,}\,\,\,\dfrac{1}{9}\begin{pmatrix} 7&4&4\\ 4&1&-8\\ 4&-8&1 \end{pmatrix},\quad\mathbf{d)\,}\,\,\,\dfrac{1}{9}\begin{pmatrix} 7&-4&4\\ 4&8&1\\ 4&-1&-8 \end{pmatrix},$$

Correction

  1. $\mathbf{a)\,}\,$ $f$ est une rotation d'axe $u=(7,7,-3)$ et d'angle $\theta=\Arccos \left(\frac{-53}{54}\right)[2\pi]$.
  2. $\mathbf{b)\,}\,$ $f$ est une symétrie orthogonale par rapport à l'axe $u= (1,2,-2)$.
  3. $\mathbf{c)\,}\,$ $f$ est la réflexion par rapport au plan d'équation $x-2y-2z=0$.
  4. $\mathbf{d)\,}\,$ $f$ est la composé d'une rotation d'axe $u=(1,0,4)$ et d'angle $-\Arccos \left(\frac{8}{9}\right)$ et de la réflexion par rapport au plan d'équation $x-4z=0$.

($\star$) Correction

Déterminer la matrice $A$ dans une b.o.n.d $(i,j,k)$ de $\R^3$, de la rotation d'axe dirigé orienté par $i+j+k$ et d'angle $\dfrac{\pi}{3}[2\pi]$.

Correction

Notons $u$ la rotation d'axe dirigé orienté par $i+j+k$ et d'angle $\dfrac{\pi}{3}[2\pi]$.

On pose $I=\dfrac{i+j+k}{\sqrt{3}},\,J=\dfrac{i-j}{\sqrt{3}}$ et $K=I\wedge J=\dfrac{i+j-2k}{\sqrt{6}}$ de sorte que $J,\,K\in \Vect(I)^\perp$ et $\BB'=(I,J,K)$ soit une b.o.n.d de $\R^3$.

La matrice de $u$ dans $\BB'$ est $\begin{pmatrix} 1&0&0\\ 0&\frac{1}{2}&\frac{-\sqrt{3}}{2}\\ 0&\frac{\sqrt{3}}{2}&\frac{1}{2} \end{pmatrix}$. En utilisant la formule de changement de base, on trouve: $$ \begin{array}{lcl} A&=&M_\BB(u)=M_\BB^{\BB'}(Id) M_{\BB'}(u)M_{\BB'}^\BB(Id)\\ &&\\ &=&\begin{pmatrix} \frac{1}{\sqrt{3}}&\frac{1}{\sqrt{2}}&\frac{1}{\sqrt{6}}\\ \frac{1}{\sqrt{3}}&\frac{-1}{\sqrt{2}}&\frac{1}{\sqrt{6}}\\ \frac{1}{\sqrt{3}}&0&\frac{-2}{\sqrt{6}} \end{pmatrix} \begin{pmatrix} 1&0&0\\ 0&\frac{1}{2}&\frac{-\sqrt{3}}{2}\\ 0&\frac{\sqrt{3}}{2}&\frac{1}{2} \end{pmatrix} \begin{pmatrix} \frac{1}{\sqrt{3}}&\frac{1}{\sqrt{3}}&\frac{1}{\sqrt{3}}\\ \frac{1}{\sqrt{2}}&\frac{-1}{\sqrt{2}}&0\\ \frac{1}{\sqrt{6}}&\frac{1}{\sqrt{6}}&\frac{-2}{\sqrt{6}} \end{pmatrix} \\ &&\\ &=&\dfrac{1}{3}\begin{pmatrix} 2&-1&2\\ 2&2&-1\\ -1&2&2 \end{pmatrix} . \end{array} $$

(DS 2022) Correction

Déterminer la matrice $A$ dans une b.o.n.d $(i,j,k)$ de $\R^3$, de la rotation d'axe dirigé orienté par $i-k$ et d'angle $\dfrac{\pi}{4}[2\pi]$.

Correction

Notons $u$ la rotation d'axe dirigé orienté par $i-k$ et d'angle $\dfrac{\pi}{4}[2\pi]$.

On pose $I=\dfrac{i-k}{\sqrt{2}},\,J=\dfrac{i+k}{\sqrt{2}}$ et $K=I\wedge J=-j$ de sorte que $J,\,K\in \Vect(I)^\perp$ et $\BB'=(I,J,K)$ soit une b.o.n.d de $\R^3$.

La matrice de $u$ dans $\BB'$ est $\begin{pmatrix} 1&0&0\\ 0&\frac{1}{\sqrt{2}}&\frac{-1}{\sqrt{2}}\\ 0&\frac{1}{\sqrt{2}}&\frac{1}{\sqrt{2}} \end{pmatrix}$. En utilisant la formule de changement de base, on trouve: $$ \begin{array}{lcl} A&=&M_\BB(u)=M_\BB^{\BB'}(Id)M_{\BB'}(u)M_{\BB'}^\BB(Id) \\ &&\\ &=&\begin{pmatrix} \frac{1}{\sqrt{2}} &\frac{1}{\sqrt{2}}&0\\ 0 &0 &-1\\ \frac{-1}{\sqrt{2}}&\frac{1}{\sqrt{2}}&0 \end{pmatrix} \begin{pmatrix} 1&0&0\\ 0&\frac{1}{\sqrt{2}}&\frac{-1}{\sqrt{2}}\\ 0&\frac{1}{\sqrt{2}}&\frac{1}{\sqrt{2}} \end{pmatrix} \begin{pmatrix} \frac{1}{\sqrt{2}}&0&\frac{-1}{\sqrt{2}}\\ \frac{1}{\sqrt{2}}&0&\frac{ 1}{\sqrt{2}}\\ 0&-1&0 \end{pmatrix}\\ &&\\ & =&\dfrac{1}{2}\begin{pmatrix} 1+\dfrac{1}{\sqrt{2}}&1&-1+\dfrac{1}{\sqrt{2}}\\ -1&\sqrt{2}&-1\\ -1+\dfrac{1}{\sqrt{2}}&1&1+\dfrac{1}{\sqrt{2}} \end{pmatrix} . \end{array}$$

($\star\star$) Correction

Soit $E$ un espace euclidien de dimension 3. Soit $f\in \mathcal{S}\mathcal{O}(E)$ une rotation d'axe dirigé par le vecteur unitaire $e$ et d'angle $\theta$. Montrer la relation suivante: $$\forall x\in E,\quad f(x)=\cos(\theta)x+\sin(\theta) \,e\wedge x+(1-\cos(\theta))\scal{x}{e}e.$$ On suppose qu'il existe $x\in E$ non nul tel que $\scal{x}{f(x)}=0$, caractériser dans ce cas $f$.

Correction

Notons $H=\Vect(e)^\perp$, soit $(u,v=e\wedge u)$ une b.o.n de $H$, ainsi $(e,u,v)$ une b.o.n.d de $E$. Déterminons la matrice de $f$ dans cette base. $$\begin{array}{lcl} f(e)&=&\cos(\theta)e+\sin(\theta) \,e\wedge e+(1-\cos(\theta))\scal{e}{e}e=\cos(\theta)e+0+(1-\cos(\theta))e\\ &=&e\\ f(u)&=&\cos(\theta)u+\sin(\theta) \,e\wedge u+(1-\cos(\theta))\scal{u}{e}e= \cos(\theta)u+\sin(\theta) v+(1-\cos(\theta))0\\ &=&\cos(\theta)u+\sin(\theta) v\\ f(v)&=&\cos(\theta)v+\sin(\theta) \,e\wedge v+(1-\cos(\theta))\scal{v}{e}e= \cos(\theta)v-\sin(\theta) u+(1-\cos(\theta))0\\ &=&\cos(\theta)v-\sin(\theta) v \end{array}$$ Ainsi, la matrice de $f$ dans cette base est $\begin{pmatrix} 1&0&0\\ 0&\cos(\theta)&-\sin(\theta)\\ 0&\sin(\theta)&\cos(\theta) \end{pmatrix}$, donc $f$ est une rotation d'axe dirigé par $e$ et d'angle $\theta$.
Pour tout $x\in E\setminus\{0\}$, $$\scal{x}{f(x)}=\scal{x}{\cos(\theta)x+\sin(\theta) \,e\wedge x+(1-\cos(\theta))\scal{x}{e}e}=\cos(\theta)\norme{x}^2+(1-\cos(\theta))\scal{x}{e}^2.$$ Ainsi, si $\cos(\theta)>0$ alors $\scal{x}{f(x)}>0$. Donc on ne peut pas trouver dans ce $x$ non nul tel que $x$ soit orthogonale à $f(x)$.
Supposons alors que $\cos(\theta)\leq 0$, et posons pour $t\in \R$, $$\varphi(t)= tu +(1-t) e,\, \text{ et } \psi (t) = \scal{f(\varphi (t))}{\varphi(t)}.$$ $\psi\in \CC(\R,\R)$ comme composition des fonctions continues, de plus, $$\psi(0)=\scal{f(\varphi(0))}{\varphi(0)}=\scal{f(e)}{e}=1,\quad \psi(1)=\scal{f(\varphi(1))}{\varphi(1)}=\scal{f(u)}{u}=\cos(\theta)\leq 0.$$ Donc d'après le TVI, il existe $t_0\in ]0,1]$ tel que $\psi(t_0)=0$, i.e. il existe $x\in \R^3$ tel que $x\perp f(x)$.

($\star$) Correction

Soit $A = \begin{pmatrix} a &-b &-c &-d \cr b & a & d &-c \cr c &-d & a & b \cr d & c &-b & a \cr \end{pmatrix} \in \MM_4(\R)$, avec $a,b,c,d$ non tous nuls. Calculer $A\,{\,}^t A$,en déduire que $A$ est inversible. Quelles conditions sur $a,b,c,d$ pour que $A\in \mathcal{O}(4)$?

Correction

On calcul, $$\begin{array}{lcl} A{\,}^t A&=&\begin{pmatrix} a &-b &-c &-d \cr b & a & d &-c \cr c &-d & a & b \cr d & c &-b & a \cr \end{pmatrix}\begin{pmatrix} a &b &c &d \cr -b & a &- d &c \cr -c &d & a & -b \cr -d & -c &b & a \cr \end{pmatrix}\\ &&\\ &=&(a^2+b^2+c^2+d^2)\begin{pmatrix} 1&0&0&0\\ 0&1&0&0\\ 0&0&1&0\\ 0&0&0&1 \end{pmatrix} \end{array}$$ Comme $a,b,c,d$ non tous nuls alors $a^2+b^2+c^2+d^2\neq 0$ donc $A$ est inversible et $A^{-1}=\dfrac{1}{a^2+b^2+c^2+d^2}{\,}^t A$.
$A\in \mathcal{O}(4)$ ssi $A{\,}^t A=I_4$ donc pour que $A$ soit dans $\mathcal{O}(4)$ il faut et il suffit que $a^2+b^2+c^2+d^2=1$.

($\star$) Correction

Soit $u\in \R^3$ tel que $\norme{u}=1$. On considère $\fonct{f}{\R^3}{\R^3}{x}{x\wedge u}$. Calculer $f^3$ en fonction de $f$.

Correction

On pose $I=u$ et soit $J\in \Vect(I)^\perp$ tel que $\norme{J}=1$. On pose ensuite $K=I\wedge J$ de sorte que $\BB=(I,J,K)$ soit une base orthonormé direct de $E$.
On a, $$\left\{\begin{array}{l} f(I)=0,\\ f(J)=J \wedge I=-K,\\ f(K)=K \wedge I=-J \end{array} \right.\Longrightarrow M_\BB(f)=\begin{pmatrix} 0&0&0\\ 0&0&1\\ 0&-1&0 \end{pmatrix}. $$ Donc $$M_\BB(f^3)=(M_\BB(f))^3=\begin{pmatrix} 0&0&0\\ 0&0&-1\\ 0&1&0 \end{pmatrix}=-M_\BB(f)$$ on en déduit alors que $f^3=-f$.

($\star$) Correction

Soit $E$ un espace vectoriel euclidien orienté de dimension $3$. Soit $u\in E$. On définit $\fonct{f}{E}{E}{x}{u\wedge(u\wedge x)}$. Montrer que $f$ est symétrique. Quelles sont les valeurs propres de $f$ ?

Correction

On complète $(u/\norme{u})$ en une base orthonormée directe.
On pose $I=\dfrac{u}{\norme{u}}$ et soit $J\in \Vect(I)^\perp$ tel que $\norme{J}=1$. On pose ensuite $K=I\wedge J$ de sorte que $\BB=(I,J,K)$ soit une base orthonormé direct de $E$.
On a, $$\left\{\begin{array}{l} f(I)=0,\\ f(J)=u \wedge (\norme{u} (I\wedge J))=-\norme{u}^2J,\\ f(K)=u \wedge (\norme{u} (I\wedge K))=-\norme{u}^2K \end{array} \right.\Longrightarrow M_\BB(f)=\begin{pmatrix} 0&0&0\\ 0&-\norme{u}^2&0\\ 0&0&-\norme{u}^2 \end{pmatrix}. $$ On obtient une matrice symétrique. Les valeurs propres sont $0$ (multiplicité $1$) et $-\norme{u}^2$ (multiplicité $2$).

($\star\star$) Correction

Soit $E$ un e-v euclidien orienté de dimension 3, et $(a,b,c)$ une base de $E$. Déterminer $(x,y,z) \in E^3$ tels que : $ x \wedge y = c \quad , \quad y \wedge z = a \quad , \quad z \wedge x = b $.
On pourra commencer par établir les formules : $(w \wedge u) \wedge (u \wedge v) = [u,v,w]u,\quad [v \wedge w,w \wedge u , u \wedge v] = [u,v,w]^2 $.

Correction

Supposons que une telle solution existe, alors $x\in \Vect(c)^\perp$ et également $x\in \Vect (b)^\perp$, on conclut alors qu'il existe $\alpha\in \R^*$ tel que $x=\alpha \,(c\wedge b)$. Le même raisonnement montre qu'il existe $\beta,\gamma\in \R^*$ tels que $y= \beta (a\wedge c),\, z=\gamma (a\wedge b)$. (comme la famille $(a,b,c)$ est libre alors les réels $\alpha,\,\beta$ et $\gamma$ ne peuvent pas être nuls).
D'autre part, on a, $$c=x\wedge y =\alpha\beta\left( (c\wedge b)(a\wedge c)\right)=\alpha\beta\left( (b\wedge c)\wedge (c\wedge a)\right)=\alpha\beta [c,a,b]c=\alpha\beta [a,b,c]c.$$ Le même raisonnement pour $a$ et $b$ nous donne $$1=\alpha\beta [a,b,c]=\alpha\gamma [a,b,c]=\beta\gamma [a,b,c]\Longrightarrow (\alpha\beta\gamma)^2[a,b,c]^3=1$$ On en déduit alors,

  1. Si $[a,b,c]< 0$ alors le système n'admet pas de solutions.
  2. Si $[a,b,c]>0$ alors le système admet des solutions, et l'ensemble de solutions est $$\left\{\left(\frac{1}{\sqrt{[a,b,c]}}(b\wedge c),\frac{1}{\sqrt{[a,b,c]}}(c\wedge a),\frac{1}{\sqrt{[a,b,c]}}(a\wedge b)\right);\left(\frac{-1}{\sqrt{[a,b,c]}}(b\wedge c),\frac{-1}{\sqrt{[a,b,c]}}(c\wedge a),\frac{-1}{\sqrt{[a,b,c]}}(a\wedge b)\right)\right\}.$$

($\star$) Correction

Soit $u$ un vecteur unitaire de l'espace euclidien $\R^3$. On définit l'application $f$ par: $$\fonct{f}{\R^3}{\R^3}{x}{\dfrac{1}{2}\left(x+\scal{x}{u}\,u+\sqrt{3}\,u\wedge\,x\right)}.$$

  1. Montrer que $f\in \LL(\R^3)$.
  2. Soit $\BB=(u,\,v,\,w)$ une base orthonormale directe de $\R^3$. Écrire la matrice de $f$ dans la base $\BB$ puis reconnaitre $f$.

Correction

  1. Soient $x,y\in \R^3$ et $\lambda\in \R$, on a $$\begin{array}{lcl} f(\lambda x+y)&=&\dfrac{1}{2}\left((\lambda x+y)+\scal{\lambda x+y}{u}\,u+\sqrt{3}\,u\wedge\,(\lambda x+y)\right)\\ &=&\dfrac{1}{2}\left(\lambda x+y+\lambda\scal{ x}{u}\,u+\scal{y}{u}\,u+\sqrt{3}\,\lambda \,u\wedge\,x+\sqrt{3}\,u\wedge y\right)\\ &=&\dfrac{\lambda}{2}\left(x+\scal{x}{u}\,u+\sqrt{3}\,u\wedge\,x\right)+\dfrac{1}{2}\left(y+\scal{y}{u}\,u+\sqrt{3}\,u\wedge\,y\right)\\ &=&\lambda f(x)+f(y) \end{array}$$ On en déduit que $f\in \LL(\R^3)$.
  2. On a $$\left\{\begin{array}{lcl} f(u)&=&u\\ f(v)&=&\dfrac{1}{2}\left(v+\sqrt{3}\,w\right)\\ f(w)&=&\dfrac{1}{2}\left(w-\sqrt{3}\,v\right) \end{array} \right.\Longrightarrow M_\BB(f)=\begin{pmatrix} 1&0&0\\ 0&\dfrac{1}{2}&\dfrac{-\sqrt{3}}{2}\\ 0&\dfrac{\sqrt{3}}{2}&\dfrac{1}{2} \end{pmatrix}.$$ Il s'agit alors d'une rotation d'axe $D=\Vect (u)$ et d'ange $\theta=\dfrac{\pi}{3}[2\pi]$.

Endomorphisme orthogonale

($\star\star$) Correction

Soit $A\in \mathcal{O}_n(\R)$. Montrer les relations suivantes: $$\mathbf{a)\,}\, \, \dsum_{1\leq i,j\leq n}a_{i\,j}^2=n,\quad \mathbf{b)\,}\,\, \abs{\dsum_{1\leq i,j\leq n}a_{i\,j}}\leq n,\quad\mathbf{c)\,}\,\, n\leq \dsum_{1\leq i,j\leq n}\abs{a_{i\,j}}\leq n\sqrt{n}. $$

Correction

  1. $\mathbf{a)\,}\,$ On a $\dsum_{1\leq i,j\leq n}a_{i\,j}^2=\tr (A{\,}^t A)=\tr (I_n)=n$ car $A\in \mathcal{O}_n(\R)$.
  2. $\mathbf{b)\,}\,$ Posons $X=\begin{pmatrix} 1\\\vdots\\1 \end{pmatrix}$, alors $\scal{X}{AX}=\dsum_{1\leq i\,j\leq n}a_{i\,j}$, en utilisant l'inégalité de Cauchy-Schwarz, on obtient: $$\abs{\dsum_{1\leq i\,j\leq n}a_{i\,j}}=\abs{\scal{X}{AX}}\leq \norme{AX}\norme{X}=\norme{X}^2=n.$$
  3. $\mathbf{c)\,}\,$ Comme $A\in \mathcal{O}_n(\R)$ alors on a, pour tout $(i,j)\in \inter{1,n}^2,\,\,a_{i\,j}\in [-1,1]$ donc $a_{i\,j}^2\leq \abs{a_{i\,j}}$. Ce qui donne $$ n= \dsum_{1\leq i,j\leq n}a_{i\,j}^2\leq \dsum_{1\leq i\,j\leq n}\abs{a_{i\,j}}.$$ D'autre part, on a $$\abs{\dsum_{1\leq i\,j\leq n}\abs{a_{i\,j}}}\leq \sqrt{\dsum_{1\leq i\,j\leq n}\abs{a_{i\,j}}^2}\sqrt{\dsum_{1\leq i\,j\leq n}1^2}=n\sqrt{\dsum_{1\leq i\,j\leq n}a_{i\,j}^2}=n\sqrt{n}.$$

($\star$) Correction

Soit $A\in \MM_n(\R)$ antisymétrique. Montrer que $I_n+A$ est inversible. On pose $B=(I_n+A)^{-1}(I_n-A)$, montrer que $B\in \mathcal{O}_n(\R)$ puis calculer $\det(B)$.

Correction

Soit $X\in \R^n$ tel que $(I_n+A)X=0$ ce qui donne $AX=-X$ puis ${\,}^t (AX)=-{\,}^t X$ or ${\,}^t (AX)= {\,}^t X{\,}^t A=-{\,}^t X A$. On a donc $$AX=-X\Longrightarrow {\,}^t X A={\,}^t X\Longrightarrow {\,}^t X A\,X={\,}^t X\,X\Longrightarrow -{\,}^t X \,X={\,}^t X\,X=\norme{X}^2,$$ on en déduit que $X=0$ d'où $I_n+A$ est inversible.
Montrons que $B\in \mathcal{O}_n(\R)$, $$\begin{array}{lcl} B B^t&=&(I_n+A)^{-1}(I_n-A) \left((I_n+A)^{-1}(I_n-A)\right)^t \\ &=&(I_n+A)^{-1}(I_n-A) (I_n-A)^t( (I_n+A)^{-1})^t\\ &=&(I_n+A)^{-1}(I_n-A)(I_n+A_n)(I_n-A)^{-1} \end{array}$$ les matrices $(I_n+A)$ et $(I_n-A)$ commutent entre elles puisqu'elles appartiennent à $\R[A]$, ce qui donne $$B\,{\,}^t B= (I_n+A)^{-1}(I_n+A)(I_n-A)(I_n-A)^{-1}=I_n\Longrightarrow B\in \mathcal{O}_n(\R).$$ D'autre part, $\det (B)=\dfrac{\det(I_n-A)}{\det( I_n+A)}=\dfrac{\det({\,}^t (I_n-A))}{\det(I_n+A)}=\dfrac{\det(I_n+A)}{\det(I_n+A)}=1.$

($\star$) Correction

Soient $E$ un espace euclidien et $f\in \mathcal{O}(E)$. Montrer que $\Ker (f-\mathrm{Id}e)=\im (f-\mathrm{Id}e)^\perp$.

Correction

Soient $x\in \Ker (f-\mathrm{Id}e)$ et $y\in \im (f-\mathrm{Id}e)$, il existe $z\in E$ tel que $y=f(z)-z$. On a alors $$\scal{x}{y}=\scal{x}{f(z)-z}=\scal{f(x)}{f(z)-z}=\scal{f(x)}{f(z)}-\scal{x}{z}=0.$$ On en déduit que $x\in \im(f-\mathrm{Id}e)^\perp$ et $y\in \Ker(f-\mathrm{Id}e)^\perp$, donc $\Ker(f-\mathrm{Id}e)\subset\im(f-\mathrm{Id}e)^\perp$, mais comme $\dim(\Ker(f-\mathrm{Id}e))=\dim(E)-\dim(\im (f-\mathrm{Id}e))=\dim(\im(f-\mathrm{Id}e)^\perp)$ on conclut alors que $\Ker (f-\mathrm{Id}e)=\im (f-\mathrm{Id}e)^\perp$.

($\star$) Correction

Pour $\theta,\varphi\in \R$, on note $R_\theta=\begin{pmatrix} \cos(\theta)&-\sin(\theta)\\\sin(\theta)&\cos(\theta) \end{pmatrix}$, et $S_\varphi=\begin{pmatrix} \cos(\varphi)&\sin(\varphi)\\\sin(\varphi)&-\cos(\varphi) \end{pmatrix}$.
Calculer les produit $R_\theta R_{\theta'},\,R_\theta S_\varphi,\,S_\varphi R_\theta, \,S_\varphi S_{\varphi'}$ ($\theta,\,\theta',\,\varphi,\,\varphi'\in \R$).

Correction

Application direct du cours.

($\star\star$) Correction

Soit $E$ en espace euclidien, soient $u\in E$ non nul et $g\in \mathcal{O}(E)$. On note $\sigma$ la symétrie orthogonale par rapport à $(\R u)^{\perp}$. Décrire l'endomorphisme $g\circ \sigma\circ g^{-1}$.

Correction

On a $\sigma(u)=-u$, notons $h=g\circ \sigma \circ g^{-1}$ et $v=g(u)$ alors : $$h(v)=g(\sigma (g^{-1}(g(u))))=g(\sigma (u))=-g(u)=-v.$$ Notons $H=\Vect (v)^\perp$, on a pour tout $x\in H,\, 0=\scal{x}{g(u)}=\scal{g^{-1}(x)}{u}$ donc $g^{-1}(x)\in \Vect(u)^\perp$ d'où $\sigma (g^{-1}(x))=g^{-1}(x)$, soit finalement $h(x)=x$.
On en déduit que $h$ est la symétrie orthogonale par rapport à $H=\Vect (g(u))^\perp$.

($\star\star$) Correction

Soient $E$ espace euclidien et $u\in \mathcal{O}(E)$. Pour $k\in \N^*$, on pose $v_k=\dfrac{1}{k}\dsum_{j=0}^{k-1}u^j$. Montrer que la suite $(v_k)_{k\geq 1}$ converge vers le projecteur orthogonale sur $\Ker (\mathrm{Id}e -u)$.

Correction

On a $\Ker (\mathrm{Id}e -u)=\im (\mathrm{Id}e -u)^\perp$ (exercice précédent).

  1. Soit $x\in \Ker (\mathrm{Id}e -u)$, i.e. $u(x)=x$ alors pour tout $j\in \N$, $u^j(x)=x$ donc $v_k(x)=x$ (pour tout $k\in \N$), ainsi $v_k(x)\tendvers{k}{\infty}x$.
  2. Soit $x\in \im (\mathrm{Id}e-u)$ donc il existe $y\in E$ tel que $x=y-u(y)$ alors pour tout $j\in \N$, $u^j(x)=u^j(y)-u^{j+1}(y)$, d'où $$\forall k\in \N^*, v_k(x)=\dfrac{1}{k}\dsum_{j=0}^{k-1}(u^j(y)-u^{j+1}(y))=\dfrac{y-u^k(y)}{k}\Longrightarrow\norme{v_k(x)}\leq\dfrac{2\norme{y}}{k}\tendvers{k}{\infty}\,0.$$

On en déduit que $(v_k)$ converge vers le projeté orthogonal sur $\Ker(u-\mathrm{Id}e )$.

Oraux

($\star\star$) Correction

Soient $a,b\in \R^3$, résoudre le système $a\wedge x=b$ d'inconnue $x\in \R^3$.

Correction

On distingue deux cas:

  1. Si $\scal{a}{b}\neq 0$ alors il n'y a pas de solutions.
  2. Si $\scal{a}{b}=0$, alors on distingue également 3 cas:
    1. Si $a=0$ et $b\neq 0$ alors il n'y a pas de solutions.
    2. Si $a=b=0$ alors l'ensemble de solutions est $\R^3$.
    3. Si $a\neq 0$. On pose $u=\dfrac{a}{\norme{a}}$, $v\in \Vect(u)^\perp$ unitaire et $w=u\wedge v$ (de sorte que $(u,v,w)$ soit une b.o.n.d de $\R^3$).
      Soit $x=\alpha u+\beta v+\gamma w\in \R^3$, on a $a\wedge x= b$ implique $$\norme{a}\beta w-\norme{a}\gamma v =\scal{b}{w}w+\scal{b}{v}v\Leftrightarrow \left\{\begin{array}{l} \alpha\in \R\\ \beta =\dfrac{\scal{b}{w}}{\norme{a}}\\ \gamma= \dfrac{-\scal{b}{v}}{\norme{a}} \end{array} \right.\Leftrightarrow x=\alpha u +\dfrac{-1}{\norme{a}}(u\wedge b)$$ On en déduit alors que l'ensemble de solutions est $\left\{ \alpha a-\dfrac{1}{\norme{a}^2}(a\wedge b),\,\alpha\in \R\right\}$.
(CCP 2018) Correction

Soit $E$ un espace euclidien et $u$ un endomorphisme orthogonal. On pose $v=u-\mathrm{Id}e$.

  1. Montrer que $E=\operatorname{Im}v\,\overset{\perp }{\oplus}\operatorname{Ker}v$.
  2. On pose $f_n=\displaystyle\frac 1n \sum_{k=0}^{n-1}u^k$ pour tout $n\in\mathbb N^*$.
    1. Soit $x\in E$. Montrer qu'il existe $(y,z)\in \operatorname{Ker}v\times \operatorname{Im}v$ tel que $f_n(x)=y+\displaystyle\frac 1n \sum_{k=0}^{n-1}u^k(z)$ pour tout entier $n\geqslant 1$.
    2. En déduire que $(f_n)_{n\in\mathbb N^*}$ converge vers le projecteur orthogonal sur $\operatorname{Ker}v$.

Correction

  1. Soient $y\in \im (v),\, x\in \Ker (v)$, il existe $z\in E,\, y=v(z)=u(z)-z$, $$\begin{array}{lcl} \scal{y}{x}&=&\scal{v(z)}{x}=\scal{u(z)-z}{x}\\ &=&\scal{u(z)}{x}-\scal{z}{x}\\ &=&\scal{u(z)}{u(x) }-\scal{z}{x} \quad\quad \quad (\text{ car }u(x)=x)\\ &=&0 \end{array}$$ Donc $\im (v)=\Ker (v)^\perp$.
    1. Soit $x\in E$, il existe un unique couple $(y,z)\in \Ker (v)\times \im (v)$ tel que $x=y+z$, donc $$f_n(x)=f_n(y+z)=f_n(y)+f_n(z)=\frac{1}{n}\dsum_{k=1}^{n-1}u^k(y)+ \frac{1}{n}\dsum_{k=1}^{n-1}u^k(z)$$ comme $y\in \Ker (v)$ alors $v(y)=0$ soit $y=u(y)$, ainsi $$f_n(x)=\frac{1}{n}\dsum_{k=1}^{n-1}y+ \frac{1}{n}\dsum_{k=1}^{n-1}u^k(z)=y+\frac{1}{n}\dsum_{k=1}^{n-1}u^k(z)$$
    2. D'après la question précédente, pour tout $x\in E$, il existe $y\in \Ker (v)$ et $z\in \im (v)$ tel que $f_n(x)=y+f_n(z)$, or puisque $z\in \im (v)$, on peut écrire $z=v(z_1)=u(z_1)-z_1$ ($z_1\in E$), ce qui donne $$f_n(x)-y=\dfrac{1}{n}\dsum_{k=1}^n u^k (u(z_1)-z_1)=\dfrac{1}{n}(u^n(z_1)-u(z_1)).$$ Ainsi, en tenant compte que $\Norme{u}\leq 1$, on obtient $$\norme{f_n(x)-y}=\dfrac{1}{n}\norme{u^n(z_1)-u(z_1)}\leq \dfrac{2\norme{z_1}}{n}\tendversN\,0.$$ Ce qui prouve que $(f_n)_{n\in\mathbb N^*}$ converge vers le projecteur orthogonal sur $\operatorname{Ker}v$.

(CCP MP 2017) Correction

Soit $n\in\mathbb{N}^*$ ; on munit $\mathbb{R}^n$ de sa structure euclidienne usuelle dont on note$ \langle \,, \, \rangle$ le produit scalaire. Soit $u\in\mathbb{R}^n$ un vecteur unitaire et $f$ l'endomorphisme de $\mathbb{R}^n$ défini par : $\forall x\in\mathbb{R}^n,\, f(x)=x-2\langle{u,x\rangle} u$.

  1. Montrer que $f$ est orthogonal. Préciser sa nature.
  2. On prend $n=3$.
    1. Donner la matrice, dans la base b canonique de $\mathbb{R}^3$, de la symétrie orthogonale par rapport au plan d'équation $x+y+z=0$.
    2. Donner la matrice, dans la base $b$, de la symétrie orthogonale par rapport à la droite $\text{Vect}((1,-1,0))$.
    3. Soit $A=\frac 13\left(\begin{array} {rrr}1 & \ \ 2 & -2\\ 2 & 1 & 2\\ -2 & 2 & 1\end{array}\right)$. Déterminer les caractéristiques de l'endomorphisme associé à $A$.

Correction

  1. Soit $x\in \R^n$, on a $$\norme{f(x)}^2=\norme{x-2\scal{u}{x}u}^2=\norme{x}^2-2\scal{x}{2\scal{u}{x}u}+4\scal{u}{x}^2\norme{u}^2=\norme{x}^2 $$ ce qui prouve que $f\in \mathcal{O}(\R^n)$.
    On note $F=\Vect (u)$, alors, $$\forall y\in F^\perp,\,\,f(y)=y-2\scal{y}{u}u=y$$ d'autre part, $f(u)= u-2\norme{u}^2u=-u$, on en déduit alors que $f$ est la symétrie orthogonale par rapport à $F^\perp$.
    1. On $F=\Vect(u)$ avec $u=\dfrac{1}{\sqrt{3}}(1,1,1)$ de sorte que $F^\perp=\{(x,y,z)\in \R^3,\,x+y+z=0\}$. Alors d'après la question précédente, (en notant $f$ la symétrie orthogonale par rapport à $F^\perp$), $$\begin{array}{lcl} f(i)&=&i-2\scal{u}{i}u=(1,0,0)-\frac{2}{3}(1,1,1)=\frac{1}{3}(1,-2,-2)\\ f(j)&=&j-2\scal{u}{j}u=(0,1,0)-\frac{2}{3}(1,1,1)=\frac{1}{3}(-2,1,-2)\\ f(k)&=&k-2\scal{u}{k}u=(0,0,1)-\frac{2}{3}(1,1,1)=\frac{1}{3}(-2,-2,1) \end{array}$$ Donc la matrice de $f$ dans la base canonique est $\dfrac{1}{3}\begin{pmatrix} 1&-2&-2\\ -2&1&-2\\ -2&-2&1 \end{pmatrix}$.
    2. On note $u=\dfrac{1}{\sqrt{2}}(1,-1,0)$ et $F=\Vect(u)$ donc $F^\perp =\{(x,y,z)\in \R^3,\,\, x-y=0\}$. Pour tout $X=(x,y,z)\in \R^3$, on écrit, $$X=\underset{\in F}{\underbrace{((x-y)/2,(y-x)/2,0)}}+\underset{\in F^\perp}{\underbrace{((x+y)/2,(x+y)/2,z)}}$$ donc (en notant $f$ la symétrie orthogonale par rapport à $F$), $$\forall X=(x,y,z)\in \R^3,\,f(X)= ((x-y)/2,(y-x)/2,0)-((x+y)/2,(x+y)/2,z)=(-y,-x,-z)$$ $$ \left. \begin{array}{lcl} f(i)&=&(0,-1,0)\\ f(j)&=&(-1,0,0)\\ f(k)&=&(0,0,-1) \end{array}\right\}\Longrightarrow M(f)=\begin{pmatrix} 0&-1&0\\ -1&0&0\\ 0&0&-1 \end{pmatrix}$$
    3. Un calcul simple montre que $A{\,}^t A=I_3$, puis $$AX=X\Longleftrightarrow X\in \{(x,y,z)\in \R^3,\, x-y+z=0\},\, AX=-X\Longleftrightarrow X\in \Vect ((1,-1,1)).$$ Donc $A$ est la matrice de la symétrie orthogonale par rapport au plan d'équation $x-y+z=0$.

(Mines 2106) Correction

Soit $f$ un endomorphisme bijectif de l'espace euclidien $E$, vérifiant : $\forall x,y\in E,\ \langle f(x),y \rangle=-\langle x,f(y) \rangle$.

  1. Montrer que $s=f\circ f$ est symétrique.
  2. Soit $a$ une valeur propre de $s$ et $V_a$ l'espace propre associé.
    1. Soit $ x\in V_a\setminus \{0_E\}$, montrer que $\langle s(x),x\rangle=a\|x\|^2=-\|f(x)\|^2$.
    2. En déduire que $a< 0$.
    3. Soit $F=\operatorname{Vect}(x,f(x))$. Montrer que $F$ et $F^\perp$ sont stables par $f$.
    4. Montrer que $\dim F=2$.

Correction

  1. Soit $x,y\in E$, on a $$\begin{array}{lcl} \scal{s(x)}{y}&=&\scal{f\circ f(x)}{y}=-\scal{f(x)}{f(y)}=\scal{x}{f\circ f(y)}\\ &=&\scal{x}{s(y)} \end{array}$$ ce qui montre que $s$ est symétrique.
  2. Soit $a\in \Sp (s)$, $x\in V_a$ non nul, on a $$ \scal{s(x)}{x}=\scal{a x}{x}=a\norme{x}^2,\, \scal{s(x)}{x}=\scal{f(f(x))}{x}=-\scal{f(x)}{f(x)}=-\norme{f(x)}^2.$$ Ceci montre que $a=-\dfrac{\norme{f(x)}^2}{\norme{x}^2}\leq 0$, et comme $f$ est bijective ($0\not\in \Sp (s)$), alors $a< 0$.
    On a évidement $f(x)\in F$ et $f (f(x))=s(x)=ax\in F$ ceci prouve que $F$ est stable par $f$.
    Soit $y\in F^\perp$, alors $$\scal{f(y)}{x}=-\scal{y}{f(x)}=0, \quad \scal{f(y)}{f(x)}=-\scal{y}{f\circ f(x)}=-a\scal{y}{x}=0$$ donc $f(y)\in F^\perp$, ce qui prouve aussi que $f^\perp $ est stable par $f$.
    Supposons que $(x,f(x))$ est liée, alors il existe $\beta \in \R$ tel que $f(x)=\beta x$ donc $s(x)=f(f(x))=\beta^2 x$, autrement dit, $\beta=a< 0$ ce qui est impossible.
    Donc la famille $(x,f(x))$ est libre, soit $\dim (F)=2$.

(Arts et Métiers PSI 2017) Correction

  1. Montrer que, $(P,Q)\mapsto \langle P,Q\rangle= P(0)Q(0) + P'(0)Q'(0) + P''(0)Q''(0)$ est un produit scalaire sur $\mathbb R_2[X]$.
  2. Montrer que la famille $(1, X, X^2)$ est orthogonale.
  3. Orthonormaliser la famille $(1, X, X^2)$.
  4. Soit $f$ l'application définie pour tout $P\in\mathbb R_2[X]$ par $f(P) = P(1-X)$. Montrer que $f$ est un automorphisme et expliciter $f^{-1}$. L'application $f$ est-elle une isométrie ?

Correction

  1. On a $$\begin{array}{lcl} \scal{1}{X}&=&1\times 0+ 0\times 1+0\times 0=0\\ \scal{1}{X^2}&=&1\times 0+ 0\times 0+ 0\times 2 =0\\ \scal{X}{X^2}&=&0\times 0 +1\times 0+ 0\times 2=0 \end{array}$$ donc la famille $(1,X,X^2)$ est une famille orthogonale.
  2. Comme la famille $(1,X,X^2)$ est orthogonale, donc pour orthonormaliser il suffit de diviser chaque vecteur par sa norme, $$\begin{array}{lcl} \scal{1}{1}&=&1\times 1+ 0\times 1+0\times 0=1\\ \scal{X}{X}&=&0\times 0+ 1\times 1+ 0\times 0 =1\\ \scal{X^2}{X^2}&=&0\times 0 +0\times 0+ 2\times 2=4 \end{array}$$ On obtient alors $(1,X,\frac{1}{2}X^2)$.
  3. Il est clair que $f$ est linéaire, de plus $f$ est bijective et $f^{-1}(P)=P(-X+1)=f(P)$, i.e. $f^{-1}=f$.
    Soit $P=\frac{X^2}{2}$ alors $f(P)= (1/2)-X+\frac{X^2}{2}$, d'où $$\norme{P}^2=1\neq \norme{f(P)}^2=\dfrac{1}{4}+1+1=\dfrac{9}{4}.$$ Donc $f$ n'est pas une isométrie vectoriel.

(C.C.E. Mines MP 2015) Correction

  1. Montrer que $(M,N)\mapsto \mathrm{Tr}({}^t MN)$ permet de définir un produit scalaire sur $\MM_n(\mathbb R)$.
  2. Soit $A\in \MM_n(\mathbb R)$ et $f:M\mapsto A\times M$. Montrer que $f$ est une isométrie vectorielle de $\MM_n(\mathbb R)$ ssi $A\in \mathcal{O}(n)$.

Correction

  1. Résultat du cours.
  2. $$\begin{array}{lcl} f\in \mathcal{O}(\MM_n(\R)&\Longleftrightarrow& \forall N,\,M\in\MM_n(\R),\,\,scal{f(M)}{f(N)}=\scal{M}{N}\\ &\Longleftrightarrow& \forall N,\,M\in\MM_n(\R),\, \tr ({\,}^t (AM) (AN))=\tr({\,}^t MN)\\ &\Longleftrightarrow& \forall N,\,M\in\MM_n(\R),\, \,\tr ({\,}^t M ({\,}^t A A) N))=\tr({\,}^t MN)\\ &\Longleftrightarrow& {\,}^t A A=I_n\Longleftrightarrow A\in \mathcal{O}(n). \end{array}$$

(E3A 2012) Correction

  1. Dans $\R^3$ euclidien orienté usuel, on note $u$ l'endomorphisme dont la matrice canoniquement associée est $B=\begin{pmatrix}0&0&1\\0&-1&0\\1&0&0\end{pmatrix}$. Donner la nature géométrique de $u$ et ses éléments caractéristiques.
  2. Soit $d\in \R$ et $P(X)=X^3+X^2+d$. Déterminer les valeurs de $d$ telles que le polynôme $P$ soit scindé sur $\R$.
  3. Soit $Q(X)=X^3+aX^2+bX+c\in \R_3[X]$ dont on note $\alpha,\beta$ et $\gamma$ les racines dans $\C$.
    1. Exprimer les coefficients $a$, $b$ et $c$ à l'aide des racines $\alpha,\beta$ et $\gamma$.
    2. Déterminer tous les triplets $(a,b,c)\in \R^3$ tels que la matrice $A=\begin{pmatrix}\alpha&\beta&\gamma\\\beta&\gamma&\alpha\\\gamma&\alpha&\beta\end{pmatrix}$ soit la matrice d'une rotation de $\R^3$ euclidien orienté usuel.

Correction

  1. On vérifie facilement que $B{\,}^t B=B^2=I_3$, puis $\det(B)=1$. Il s'agit alors d'une rotation. Pour déterminer l'axe de la rotation, on résout $u(x)=x$, $$u(x)=x\Longleftrightarrow B\begin{pmatrix} a\\b\\c \end{pmatrix}=\begin{pmatrix} a\\b\\c \end{pmatrix}\Longleftrightarrow\left\{\begin{array}{lcl} c&=&a\\ b&=&-b\\ a&=&c \end{array}\right.\Longleftrightarrow X\in \Vect(\begin{pmatrix} 1\\0\\1 \end{pmatrix}).$$ Donc l'axe de la rotation est la droite $D=\Vect (\dfrac{i+k}{\sqrt{2}})$. Il faut ensuite déterminer l'angle $\theta$, d'après le cours, on a $$1+2\cos(\theta)=\tr (u)=-1\Longrightarrow \cos(\theta)=-1\Longrightarrow \theta=\pi [2\pi].$$
  2. Un étude rapide de la fonction $f(x)=x^3+x^2+d$, donne le tableau de variation suivant Donc pour avoir trois racines réelles il faut que $d\in \left[\frac{-4}{27},0\right]$.

  3. On écrit, $$(X-\alpha)(X-\beta)(X-\gamma)=(X^2-(\alpha+\beta)X+\alpha\beta)(X-\gamma)=X^3-(\alpha+\beta+\gamma)X^2+(\alpha\beta+\alpha\gamma+\beta\gamma)X-\alpha\beta\gamma $$ Par identification, on trouve: $$a=-(\alpha+\beta+\gamma),\quad b=\alpha\beta+\alpha\gamma+\beta\gamma,\,\quad c=-\alpha\beta\gamma.$$ Puis, on calcul, $$\begin{array}{lcl} A{\,}^t A&=&\begin{pmatrix} \alpha^2+\beta^2+\gamma^2&\alpha\beta+\beta\gamma+\gamma\beta&\alpha\gamma+\beta\alpha+\gamma\beta\\ \alpha\beta+\beta\gamma+\gamma\beta&\alpha^2+\beta^2+\gamma^2 &\alpha\beta+\beta\gamma+\gamma\beta\\ \alpha\beta+\beta\gamma+\gamma\beta&\alpha\beta+\beta\gamma+\gamma\beta&\alpha^2+\beta^2+\gamma^2 \end{pmatrix}\\ &&\\ &=&\begin{pmatrix} a^2-2b&b&b\\ b&a^2-2b&b\\ b&b&a^2-2b \end{pmatrix} \end{array}$$ Ainsi, $A$ est une matrice d'une rotation ssi $$\alpha,\,\beta,\,\gamma\in \R,\quad a^2-b=1,\quad b=0,\quad \det(A)=1\Longleftrightarrow \alpha,\,\beta,\,\gamma\in \R,\quad a=\mp 1,\, b=0,\, \det(A)=1$$ Le calcul de $\det (A)$ donne (en utilisant $b=0$), $$\begin{array}{lcl} \det(A)&=&\alpha(\gamma\beta-\alpha^2)-\beta (\beta^2-\alpha\gamma)+\gamma (\beta\alpha-\gamma^2)\\ &=&-(\alpha+\beta+\gamma)(\alpha^2 +\beta^2+\gamma^2)= -(\alpha+\beta+\gamma)\left( (\alpha+\beta+\gamma)^2-2(\alpha\beta+\alpha\gamma+\gamma\beta)\right)\\ &=&a ((-a)^2-2b)=a \end{array}$$ donc $A$ est une matrice d'une rotation ssi $a=1,\,b=0$ et $\alpha,\,\beta,\,\gamma\in \R$, ce qui implique en utilisant la question précédente, $c\in \left[\frac{-4}{27},0\right]$.

(Centrale 2018) Correction

On se donne un $n$- de réels $(x_{1}, x_{2}, \dots, x_{n})$. On pose : $$s=\sum_{i=1}^{n}{x_{i}},\ \alpha=\sum_{1 \leqslant i < j \leqslant n}{x_{i}x_{j}}, \, X=(x_{1}\ \ x_{2}\ \ \cdots\ \ x_{n})^{\mathrm{T}},\ U=(s-x_{1}\ \ s-x_{2}\ \ \cdots\ \ s- x_{n})^{\mathrm{T}},$$ $A=UX^{\mathrm{T}} \text{ et } M=A-\alpha I_{n}.$

  1. Déterminez les éléments propres de $A$.
  2. A quelle condition sur $\alpha $ la matrice $A$ est-elle diagonalisable ?
  3. Calculez le déterminant de $M$.
  4. On suppose dans cette question que $\alpha \neq 0$. Interprétez géométriquement l'endomorphisme de $\mathbb{R}^{n}$ canoniquement associé à $\frac{1}{a}M$.
  5. A quelle condition portant sur les réels $(x_{1}, x_{2},\dots, x_{n})$ l'endomorphisme canoniquement associé à $M$ est une symétrie orthogonale ?

Correction

On suppose que $X\neq 0$ (le cas $X=0$ ne représente aucun intérêt), ce qui implique $U\neq 0$, en effet si on suppose que $U=0$, on aurait $$\forall i\in \inter{1,n},\,\, x_i=s\Longrightarrow s=\dsum_{i=1}^n x_i=\dsum_{i=1}^n s=ns \Longrightarrow s=0.$$ Donc $X=0$ ce qui est impossible .
Notons également $F=\Vect (X)$ et $G=F^\perp$

  1. La matrice $A$ est le produit d'une matrice colonne par une matrice ligne, donc $\rg (A)= 1$.
    Pour tout $Y\in G$, on a $$AY=UX^TY =U (X^TY) =\scal{X}{T}U=0 \Longrightarrow Y\in \Ker (A),$$ Ainsi $0$ est une valeur propre d'ordre au moins $n-1$ (puisque $\dim (G)=n-1$), et vu que $A\neq 0$ donc $0$ est une v.p. d'ordre $n-1$ et $E_0(A)=G$.
    Pour trouver la deuxième valeur propre $\lambda$, on a $$\lambda=\dsum_{\beta\in \Sp (A)}\beta =\tr (A)= \dsum_{j=1}^n (s-x_j)x_j=\dsum_{i=1}^n sx_j-\dsum_{j=1}^nx_j^2=\left(\dsum_{i=1}^nx_i\right)^2-\norme{X}^2=\dsum_{i\neq j}x_ix_j=2\alpha.$$
  2. La matrice $A$ est diagonalisable ssi $\tr (A)\neq 0$, i.e. $\alpha\neq 0$.
  3. Supposons que $\alpha\neq 0$, alors $\Sp (A)=\{0,2\alpha\}$. On a $$\forall Y\in G=E_0(A),\,\quad MY=(A-\alpha I_n)Y=AY-\alpha Y=-\alpha Y$$ donc $-\alpha$ est une valeur propre de $M$ d'ordre $n-1$.
    Soit $Z$ un vecteur propre de $A$ associé à la v.p. $2\alpha$, alors $MZ=AZ-\alpha Z=\alpha Z$, donc $\alpha$ est une valeur propre de $A$ d'ordre $1$.
    On en déduit que $\det (M)= (-1)^{n-1}\alpha^n$.
    Si $\alpha =0$ alors $\det (M)=0$ puisque $0\in \Sp (M)$.
  4. D'après la question précédente, $$\forall Y\in E_0(A),\,\,\quad \dfrac{1}{\alpha} MY=\dfrac{1}{\alpha}(A-\alpha I_n)Y=-Y,$$ et $$\forall Y\in E_{2\alpha}(A),\,\, \dfrac{1}{\alpha} MY=\dfrac{1}{\alpha}(A-\alpha I_n)Y=Y.$$ Donc $\frac{1}{\alpha}M$ est la matrice d'une symétrie par rapport à $E_{2\alpha}(A)$ parallèlement à $E_0(A)$.
  5. La première condition est $\alpha\neq 0$, dans ce cas $M$ est inversible.
    Dans ce cas, $\Sp(M)=\{-\alpha,\alpha\}$, ainsi pour que $M$ soit la matrice d'une symétrie il faut que $\alpha\in \{-1,1\}$. Les espaces propres de $M$ sont $G$ et $E_\alpha(M)$, il faut que $E_\alpha (M)$ soit orthogonale à $G$, i.e. $E_\alpha(M)=G^\perp=F$. Donc il faut que $X$ soit un vecteur propre associé à la v.p. $\alpha$.
    $X$ vecteur propre de $M$, $$MX = \alpha X\Longleftrightarrow AX=2\alpha X\Longleftrightarrow \norme{X}^2 U=2\alpha X.$$ Donc $U$ doit être colinéaire à $X$, ce qui se traduit par, $$\exists \gamma\in \R^*,\quad \forall i \in \inter{1,n},\quad s-x_i= \gamma x_i\Longleftrightarrow \exists x\in \R^*,\quad \forall i\in \inter{1,n},\, x_i=x.$$ Conclusion: $M$ est la matrice d'une symétrie orthogonale ssi il existe $x\in \R^*$ tel que $X = (x\,x\,\cdots\,x)^T$.

(Mines-Télécom PSI 2019) Correction

On munit $\mathbb{R}^3$ de sa structure euclidienne canonique, on travaille relativement à la base canonique. Déterminer la matrice de la symétrie orthogonale par rapport au plan d'équation $y =x\tan \theta$, avec $\theta \in [0,\frac{\pi}{2}[$.

Correction

On note $P=\{(x,y,z)\in \R^3,\,\, y=x\tan (\theta)\}$. On pose $e_1=(\cos (\theta),\sin(\theta),0)$, $e_2=(0,0,1)$. La famille $(e_1,e_2)$ est une b.o.n du plan $P$, on définit ensuite $e_3=e_1\wedge e_2 =(\sin(\theta),-\cos (\theta),0)$ de sorte que $(e_1,e_2,e_3)$ est une b.o.n directe de $\R^3$.
La matrice de la symétrie orthogonale par rapport à $P$ dans la base $(e_1,e_2,e_3)$ est $\begin{pmatrix} 1&0&0\\ 0&1&0\\ 0&0&-1 \end{pmatrix}$. On en déduit alors que la matrice de cette symétrie dans la base canonique est $$A =\begin{pmatrix} \cos(\theta)&0&\sin(\theta)\\ \sin(\theta)&0&-\cos(\theta)\\ 0&1&0 \end{pmatrix}\begin{pmatrix} 1&0&0\\ 0&1&0\\ 0&0&-1 \end{pmatrix} \begin{pmatrix} \cos(\theta)&\sin(\theta)&0\\ 0&0&1\\ \sin(\theta)&-\cos(\theta)&0 \end{pmatrix}=\begin{pmatrix} \cos(\theta)^2-\sin(\theta)^2&2\cos(\theta)\sin(\theta)&0\\ 2\cos(\theta)\sin(\theta)&\sin(\theta)^2-\cos(\theta)^2&0\\ 0&0&1 \end{pmatrix}.$$

Endomorphisme symétrique

($\star$) Correction

Soit $A=(a_{ij})$ une matrice réelle symétrique d'ordre $n$. On note $\lambda_1,\ldots,\lambda_n$ ses valeurs propres (distinctes ou non). Prouver que : $\dsp \sum_{i=1}^n{\lambda_i^2} = \dsp \sum_{i=1}^n \dsp \sum_{j=1}^n {a_{ij}^2}$.

Correction

D'après le cours, il existe une b.o.n $\BB'=(e_1,\cdots,e_n)$ de $\R^n$ formée des vecteurs propres de $A$ (i.e. $Ae_i=\lambda_i\,e_i$). De plus, en notant $P$ la matrice de passage de la base canonique $\BB$ à $\BB'$, $P=M_\BB^{\BB'}(Id)$. Alors on a $P\in \mathcal{O}_n(\R)$ et $A=PD{\,}^t P$ avec $D=\mathrm{diag} (\lambda_1,\cdots,\lambda_n)$. On en déduit alors $${\,}^t A\,A={\,}^t (PD{\,}^t P)(PD{\,}^t P)=PD({\,}^t P \,P)D{\,}^t P=P(D^2){\,}^t P \Longrightarrow \tr ({\,}^t A A)=\tr(PD^2{\,}^t P)=\tr (D^2).$$ Pour conclure, il suffit de remarque que $\tr ({\,}^t A A)=\dsp\sum_{i=1}^n \dsp \sum_{j=1}^n {a_{ij}^2}$ et $\tr(D^2)=\dsp \sum_{i=1}^n{\lambda_i^2}$.

($\star\star$) Correction

On note $\scal{.}{.}$ le produit scalaire canonique sur $\R^n$ et $\norme{.}$ la norme associée. Pour toute famille $u=(u_1,\cdots,u_p)$ de $(\R^n)^p$, on note $G_u$ la matrice carrée d'ordre $p$ dont le terme d'indice $(i,j)\in \inter{1,p}^2$ est $\scal{u_i}{u_j}$.

  1. Montrer que $G_u$ est inversible ssi la famille $(u_1,\cdots,u_p)$ est libre.
  2. Démontrer que $\det G_u \geq 0$ et que $\det G_u \leq \dsp \prod_{i=1}^p \norme{u_i}^2$.

Correction

  1. Soit $X={\,}^t (\alpha_1,\cdots,\alpha_p)\in \R^p$, on pose $u(X)=\dsum_{i=1}^p\alpha_iu_i\in \Vect((u_1,\cdots,u_p))\subset \R^n$, on a $$\scal{X}{G_uX}=\sum_{i=1}^p\sum_{j=1}^p\alpha_i\alpha_j\scal{u_i}{u_j}=\sum_{i=1}^p\sum_{j=1}^p \scal{\alpha_iu_i}{\alpha_j u_j}=\scal{\sum_{i=1}^p\alpha_iu_i}{\sum_{j=1}^p\alpha_j u_j}=\norme{u(X)}^2.$$ Notons $\lambda_1,\cdots,\lambda_p$ les valeurs propres de $G_u$, la relation précédente montre que $\Sp (G_u)\subset \R_+$. En effet, si $X_i$ est un vecteur propre associé à $\lambda_i$ alors on a $\lambda_i\norme{X_i}^2=\norme{u(X_i)}^2$ donc $\lambda_i\geq 0$.
    Supposons que $(u_1,\cdots,u_p)$ est une famille libre, alors pour tout $X\in \R^p$ non nuls on a $u(X)\neq 0$ donc $\scal{X}{G_u X}>0$, donc les valeurs propres de $G_u$ sont strictement positives, i.e. $0\not \in \Sp (G_u)$, d'où $G_u$ est inversible.
    Inversement, si $G_u$ est inversible alors $0\not \in \Sp(G_u)$ donc $$\forall X\in \R^p\setminus \{0\},\,\,\scal{X}{G_uX}>0$$ Si $\dsum_{k=1}^p\alpha_ku_k=0$ ceci implique $\scal{X}{G_uX}=0$ donc $X=0$ i.e. $\alpha_1=\cdots=\alpha_p=0$ donc la famille la famille $(u_1,\cdots,u_p)$ est libre.
  2. Pour montrer que $\det G_u\geq 0$ il suffit de remarquer que $\Sp (G_u)\subset \R_+$ et comme $\dsp\det(G_u)=\prod_{\lambda\in \Sp (G_u)}\lambda$ on trouve le résultat.
    Une autre méthode: Si la famille $(u_1,\cdots,u_p)$ est liée alors $\det(G_u)=0$, sinon on note $(e_1,\cdots , e_p)$ une b.o.n de $\Vect ((u_1,\cdots,u_p))$, alors $$\forall i,j\in \inter{1,p},\,u_i=\sum_{k=1}^p\scal{u_i}{e_k}e_k,\,\,\scal{u_i}{u_j}=\dsum_{k=1}^p\scal{u_i}{e_k}\scal{u_j}{e_k}.$$ Alors en notant $A=(\scal{u_i}{e_j})_{0\leq i,j\leq p}\in \MM_p(\R)$, on a $G_u=A{\,}^t A$ donc $\det (G_u)=\det(A)^2>0$.
    Montrons que $\det G_u \leq \dsp \prod_{i=1}^p \norme{u_i}^2$ par récurrence sur $p\in \N^*$.
    Pour $p=1$ la relation est évidente. Supposons que $p\geq 1$ et que la relation est vrai au rang $p$. Soit $(u_1,\cdots,u_p,u_{p+1})$ une famille de $(p+1)$ éléments dans $\R^n$. Notons $G_u=(\scal{u_i}{u_j})_{1\leq i,j\leq p+1}\in \MM_{p+1}(\R)$ et $G=(\scal{u_i}{u_j})_{1\leq i,j\leq p}\in \MM_{p}(\R)$. Soit $y$ la projection orthogonale de $u_{p+1}$ sur $\Vect ((u_1,\cdots,u_p))$ alors $$\forall j\in \inter{1,p},\,\,\scal{u_{p+1}}{u_j}=\scal{y}{u_j},\text{ et la famille }(u_1,\cdots,u_p,y)\text{ est liée}.$$ Ce qui donne $$\begin{array}{lcl} \det(G_u)&=&\begin{vmatrix} \scal{u_1}{u_1}&\cdots&\scal{u_1}{u_p}&\scal{u_1}{u_{p+1}}\\ \vdots& & & \vdots\\ \scal{u_1}{u_p}&\cdots&\scal{u_p}{u_p}&\scal{u_p}{u_{p+1}}\\ \scal{u_1}{u_{p+1}}&\cdots&\scal{u_p}{u_{p+1}}&\scal{u_{p+1}}{u_{p+1}} \end{vmatrix}\\ &&\\ &=&\begin{vmatrix} \scal{u_1}{u_1}&\cdots&\scal{u_1}{u_p}&\scal{u_1}{y}\\ \vdots& & & \vdots\\ \scal{u_1}{u_p}&\cdots&\scal{u_p}{u_p}&\scal{u_p}{y}\\ \scal{u_1}{y}&\cdots&\scal{u_p}{y}&\scal{u_{p+1}-y}{u_{p+1}-y}+\scal{y}{y} \end{vmatrix}\\ &&\\ &=& \begin{vmatrix} \scal{u_1}{u_1}&\cdots&\scal{u_1}{u_p}&\scal{u_1}{y}\\ \vdots& & & \vdots\\ \scal{u_1}{u_p}&\cdots&\scal{u_p}{u_p}&\scal{u_p}{y}\\ \scal{u_1}{y}&\cdots&\scal{u_p}{y}&\scal{y}{y} \end{vmatrix}\\ && + \begin{vmatrix} \scal{u_1}{u_1}&\cdots&\scal{u_1}{u_p}&\scal{u_1}{y}\\ \vdots& & & \vdots\\ \scal{u_1}{u_p}&\cdots&\scal{u_p}{u_p}&\scal{u_p}{y}\\ \scal{u_1}{y}&\cdots&\scal{u_p}{y}&\scal{u_{p+1}-y}{u_{p+1}-y} \end{vmatrix}\\ &=&\begin{vmatrix} \scal{u_1}{u_1}&\cdots&\scal{u_1}{u_p}&\scal{u_1}{y}\\ \vdots& & & \vdots\\ \scal{u_1}{u_p}&\cdots&\scal{u_p}{u_p}&\scal{u_p}{y}\\ \scal{u_1}{y}&\cdots&\scal{u_p}{y}&\scal{u_{p+1}-y}{u_{p+1}-y} \end{vmatrix}\\ &&\\ &=&\norme{u_{p+1}-y}^2 \begin{vmatrix} \scal{u_1}{u_1}&\cdots&\scal{u_1}{u_p}\\ \vdots& & \\ \scal{u_1}{u_p}&\cdots&\scal{u_p}{u_p} \end{vmatrix}\\ &\leq &\norme{u_{p+1}}^2\det(G)\leq \norme{u_{p+1}}^2\prod_{k=1}^p\norme{u_k}^2 \end{array}$$ Ceci prouve que la relation est vrai au rang $p+1$.

($\star\star$) Correction

Soit $E=\R_n[X]$ ($n\in \N^*$) muni du produit scalaire $\scal{P}{Q}=\dsp\int_0^1P(t)Q(t)\ud t$. On définit $u:E\longmapsto E$ par $$\forall P\in E,\quad u(P)=\int_0^1(X+t)^nP(t)\ud t.$$

  1. Montrer que $u$ est un endomorphisme symétrique de $E$, en déduire qu'il existe une base orthonormale $(P_0,\cdots , P_n)$ de $E$ formée de vecteurs propres de $E$. On note $\lambda_0,\cdots,\lambda_n$ les valeurs propres associées.
  2. Montrer que : $\forall x,y\in \R,\quad (x+y)^n= \dsum_{k=0}^n\lambda_kP_k(x)P_k(y)$. En déduire $\tr (u)$.

Correction

  1. Soit $P=\dsum_{\ell =1}^na_\ell X^\ell\in E$, on a $$\int_0^1(X+t)^nP(t)\ud t=\int_0^1\dsum_{\ell=1}^n{n \choose \ell}t^{n-\ell}X^\ell P(t)\ud t= \dsum_{k=1}^n{n \choose \ell}\left(\int_0^1t^{n-k}P(t)\ud t\right)X^k\in E\Longrightarrow u(P)\in E .$$ Soient $P_1,P_2\in E$ et $\lambda\in \R$, on a $$\begin{array}{lcl} u(\lambda P_1+P_2)&=& \dsp\int_0^1(X+t)^n (\lambda P_1+P_2)(t)\ud t\\ &&\\ &=&\dsp \int_0^1 (X+t)^n\left(\lambda P_1(t)+P_2(t)\right)\ud t\\ &&\\ &=&\dsp \lambda \int_0^1(X+t)^nP_1(t)\ud t+\int_0^1 (X+t)^nP_2(t)\ud t\\ &&\\ &=&\lambda u(P_1)+u(P_2) \end{array}$$ On en déduit $u\in \LL(E)$. Soient $P,Q\in E$, on a $$\begin{array}{lcl} \scal{u(P)}{Q}&=&\dsp \int_0^1\left(\int_0^1(t+u)^nP(u)\ud u\right)Q(t)\ud t\\ &&\\ &=&\dsp \int_0^1\int_0^1(t+u)^nP(u)Q(t)\ud u\ud t\\ &&\\ &=&\dsp \int_0^1\left(\int_0^1(t+u)^nQ(t)\ud t\right)P(u)\ud u=\scal{P}{u(Q)} \end{array}$$ Donc $u$ est un endomorphisme symétrique. On en déduit que $u$ est diagonalisable dans une b.o.n $(P_0,\cdots, P_n)$ de $E$ formée des vecteurs propres de $u$.
  2. Soit $y\in \R$, on considère le polynôme $P=(X+y)^n\in E$, d'après la question précédente, on a $$P=(X+y)^n=\dsum_{k=0}^n\scal{P}{P_k}P_k(X),\text{ avec }\scal{P}{P_k}=\int_0^1(t+y)^nP_k(t)\ud t= u(P_k)(y)=\lambda_kP_k(y),$$ d'où $$\forall y\in \R,\quad (X+y)^n=\dsum_{k=0}^n\lambda_kP_k(y)P_k(X)\Longrightarrow \forall (x,y)\in \R^2,\quad (x+y)^n=\dsum_{k=0}^n\lambda_kP_k(y)P_k(x).$$ En prenant $y=x$ dans la relation précédent, puis en intégrant entre $0$ et $1$, on trouve, $$\int_0^1(2x)^n\ud x= \int_0^1\dsum_{k=0}^n\lambda_kP_k(x)^2\ud x= \dsum_{k=0}^n\lambda_k\int_0^1P_k(x)^2\ud x=\dsum_{k=0}^n\lambda_k\norme{P_k}^2=\dsum_{k=0}^n\lambda_k$$ Or $\dsp \int_0^1(2x)^n\ud x=\dfrac{2^n}{n+1}$, et $\dsum_{k=0}^n\lambda_k=\tr (u)$.

($\star\star\star$) Correction

Soient $f_1, \dots ,f_n \in \CC([0,1],\R)$. Pour $(i,j) \in \inter{1,n}^2$, on pose : $$ a_{ii} = \dsp \sum_{\underset{j \neq i}{j=1}}^n \dsp \int_0^1f_j^2(t) \ud t \quad \textrm{et, pour }i\neq j, a_{ij}=- \dsp \int_0^1 f_i(t)f_j(t) \ud t $$ Soit enfin $A$ la matrice $A=(a_{ij}) \in \MM_n(\R)$.

  1. Montrer que $A$ est symétrique positive.
  2. A quelle condition $A$ est-elle définie positive ?

Correction

On définit sur $E=\CC([0,1],\R)$ le produit scalaire $\scal{f}{g}=\dsp\int_0^1f(t)g(t)\ud t$ et on note $\norme{\cdot}_E$ la norme associée à ce p.s.
On note $B=(\scal{f_i}{f_j})_{1\leq i,j\leq n}\in \MM_n(\R)$, de sorte que $A=\dsp\left(\dsum_{k=1}^n\norme{f_k}_E^2\right)I_n-B$. Il est clair que $A$ est symétrique.
Soit $X={\,}^t (x_1,\cdots,x_n)\in \R^n$, on a $$\begin{array}{lcl} \scal{X}{AX}_{\R^n}&=& \scal{X}{\left(\dsum_{k=1}^n\norme{f_k}_E^2\right)X}_{\R^n}-\scal{X}{BX}_{\R^n}=\left(\dsum_{k=1}^n\norme{f_k}_E^2\right)\norme{X}_{\R^n}^2-\dsum_{1\leq i,j\leq n}x_i\scal{f_i}{f_j}x_j\\ &=&\left(\dsum_{k=1}^n\norme{f_k}_E^2\right)\norme{X}_{\R^n}^2-\norme{\dsum_{k=1}^n x_kf_k}_E^2. \end{array}$$ D'autre part, on a: $$\norme{\dsum_{k=1}^n x_kf_k}_E\leq \dsum_{k=1}^n\abs{x_k}\norme{f_k}_E\leq \sqrt{\dsum_{k=1}^nx_k^2}\sqrt{\dsum_{k=1}^n\norme{f_k}_E^2}=\norme{X}_{\R^n}\sqrt{\dsum_{k=1}^n\norme{f_k}_E^2}.$$ On obtient finalement, $$\scal{X}{AX}_{\R^n}=\left(\dsum_{k=1}^n\norme{f_k}_E^2\right)\norme{X}_{\R^n}^2-\norme{\dsum_{k=1}^n x_kf_k}_E^2\geq \left(\dsum_{k=1}^n\norme{f_k}_E^2\right)\norme{X}_{\R^n}^2-\norme{X}_{\R^n}^2\left(\dsum_{k=1}^n\norme{f_k}_E^2\right)=0.$$ Ce qui implique que $A$ est symétrique positive.
On peut également écrire: $$\forall X\in \R^n,\,\quad \scal{X}{AX}_{\R^n}=\dsum_{i=1}^{n-1}\left(\dsum_{j=i+1}^n\norme{x_jf_i-x_if_j}_E^2\right).$$ Avec cette écriture, on voit que la matrice $A$ n'est pas définie positive ssi $\rg (f_1,\cdots,f_n)\leq 1$.

($\star\star$) Correction

Soient $u$ et $v$ deux endomorphismes symétriques d'un e-v euclidien $E$. Démontrer que les quatre propositions suivantes sont équivalentes :

  1. $u \circ v$ est symétrique.
  2. $u \circ v = v \circ u$.
  3. il existe une base orthonormale de $E$ formée de vecteurs propres communs à $u$ et a $v$.
  4. Il existe une base orthonormale $\BB$ de $E$ telle que les matrices de $u$ et $v$ dans $\BB$ soient diagonales.

Correction

$\mathbf{a)} \Longrightarrow \mathbf{b)}$ Supposons que $u\circ v $ est symétrique. Soit $x\in E$, pour tout $y\in E$, on a $$\begin{array}{lcl} \scal{u\circ v (x)}{y}&=&\scal{x}{u\circ v(y)}\\ &=&\scal{u (x)}{v(y)}\quad\quad (\text{car } u \text{ est symétrique})\\ &=&\scal{v\circ u (x)}{y}\quad\quad (\text{car } v \text{ est symétrique}) \end{array}$$ Donc $$\forall y\in E,\, \scal{u\circ v (x)}{y}=\scal{v\circ u(x)}{y}$$ Ce qui implique que $u\circ v(x)=v\circ u(x)$.
$\mathbf{b)} \Longrightarrow \mathbf{c)}$ Supposons que $u\circ v=v\circ u$, puisque $u$ est symétrique il existe une b.o.n de $\BB=(b_1,\cdots,b_n)$ formée des vecteurs propres de $u$. Soit $k\in \inter{1,n}$, on a $$u(v(b_k))=v(u(b_k))=v(\lambda_k b_k)=\lambda_k v(b_k)$$ donc $v(b_k)$ est un vecteur propre de $u$ associé à $\lambda_k$, il existe alors $\alpha_k\in \R$ tel que $v(b_k)=\alpha_k b_k$, ceci prouve que $b_k$ est aussi un vecteur propre de $v$.

($\star\,--\,\star\star$) Correction

Diagonaliser les matrices : $$ A=\left( \begin{array}{ccccc} 1 & 1 & \ldots & 1 & 1 \\ 1 & 0 & \ldots & 0 & 1 \\ \vdots & \vdots & \ldots & \vdots & \vdots \\ 1 & 0 & \ldots & 0 & 1\\ 1 & 1 & \ldots & 1 & 1 \end{array} \right) \quad ,\quad\quad \quad B=\left( \begin{array}{cccccc} 2 & 1 & 0 & \ldots & \ldots & 0 \\ 1 & 2 & 1 & 0 & \ldots &0 \\ 0 & 1 & 2 & 1 & \ldots & 0 \\ 0 & 0 & \ddots & \ddots & \ddots & 0 \\ 0 & 0 & \ldots & 1 & 2 & 1 \\ 0 & 0 & \ldots & \ldots & 1 & 2 \end{array} \right).$$ $$\quad$$

Correction


  1. Si on note $(C_1,\cdots,C_n)$ les colonnes de $A$, on a , $$C_1-C_n=0,\,\,C_3-C_2=C_4-C_2\cdots=C_{n-1}-C_2=0\Longrightarrow 0\in \Sp(A) \text{ et } \dim (\Ker(A))\geq n-2.$$ Soit $\lambda\in \Sp(A),\,\lambda \neq 0$, et $X={\,}^t (x_1,\cdots,x_n)\in \R^n,$ non nul, un vecteur propre associé à $\lambda$, alors: $$AX=\lambda X\Leftrightarrow\left\{\begin{array}{lcl} \dsum_{k=1}^nx_k&=&\lambda x_1\\ x_1+x_n&=&\lambda x_2\\ \vdots&& \\ x_1+x_n&=&\lambda x_{n-1}\\ \dsum_{k=1}^nx_k&=&x_n \end{array} \right. \Leftrightarrow \left\{\begin{array}{l} x_1=x_n\\ x_2=\cdots=x_{n-1}=\dfrac{x_1+x_n}{\lambda}\\ x_1+x_n+(n-2)\dfrac{x_1+x_n}{\lambda}=\lambda x_1=\lambda x_n \end{array} \right. \Leftrightarrow \left\{\begin{array}{l} x_1=x\in \R^*\\ x_n=x\\ x_j=\dfrac{2x}{\lambda},\,\,j\in \inter{2,n-1}\\ \lambda^2 -2\lambda-2(n-2)=0 \end{array} \right. $$ La dernière équation nous donne que les deux v.p restant de $A$ sont $1\mp \sqrt{1+2(n-2)}$. Donc $\Sp (A)=\{1-\sqrt{1+2(n-2)},0,1+\sqrt{1+2(n-2)}\}$.
  2. On écrit $B$ sous la forme: $$B=2I_n + C\text{ avec } C = \left( { \begin{array}{cccc} 0 & 1 & {} & 0 \\ 1 & \ddots & \ddots & {} \\ {} & \ddots & \ddots & 1 \\ 0 & {} & 1 & 0 \\ \end{array} } \right) \in \MM_n (\mathbb{C})$$ Ainsi si $\lambda\in \Sp (B)$ alors il existe $X\in \R^n$ non nul tel que $$BX=\lambda X \Longleftrightarrow 2X+CX=\lambda X \Longleftrightarrow CX =(\lambda -2)X \Longleftrightarrow 2-\lambda \in \Sp(C).$$ Or $\Sp(C)=\Big\{-2\cos\left(\dfrac{k\pi}{n+1}\right),\,k\in \inter{1,n} \Big\}$, (voir exercice réduction) donc $$\Sp (B)=\Big\{2-2\cos\left(\dfrac{k\pi}{n+1}\right),\,k\in \inter{1,n} \Big\} $$
($\star\star$) Correction

Soient $A=(a_{ij})$ et $B=(b_{ij})$ $\in \MM_n(\R)$, symétriques, positives. Soit $C=(c_{ij})$ la matrice définie par : $ \forall (i,j) \in \inter{1,n}^2 , \ c_{ij}=a_{ij}b_{ij}$. Montrer que $C$ est symétrique positive.
Montrer que, si on suppose de plus $A$ et $B$ définies positives, alors $C$ est définie positive.

Correction

Le faite que $C$ est symétrique est évident!
Comme $A$ est symétrique alors $A$ est diagonalisable, de plus il existe $P=(p_{ij})_{1\leq i,j\leq n}\in \mathcal{O}_n(\R)$ et $D$ une matrice diagonale tel que $A=PD{\,}^t P$ (les éléments diagonaux de $D$ sont les valeurs propres de $A$, $\lambda_1,\cdots,\lambda_n$).
On obtient alors, pour tout $(i,j)\in \inter{1,n}^2$, $a_{ij}=\dsp\dsum_{k=1}^n\lambda_kp_{ik}p_{jk}$.
Soit $X={\,}^t (x_1,\cdots,x_n)\in \R^n$, on a $$\scal{X}{CX}=\dsum_{1\leq i,j\leq n}x_ic_{ij}x_j=\dsum_{1\leq i,j\leq n}x_ia_{ij}b_{ij}x_j= \dsum_{1\leq i,j\leq n}x_i\left(\dsum_{k=1}^n\lambda_kp_{ik}p_{jk}\right) b_{ij}x_j=\dsum_{k=1}^n\lambda_k\left(\dsum_{1\leq i,j\leq n} (p_{ik}x_i)b_{ij}(p_{jk}x_k)\right).$$ On note alors pour $k\in \inter{1,n}$, $X_k={\,}^t (p_{1k}x_1,\cdots,p_{nk}x_n)\in \R^n$. On obtient alors, $\dsp \scal{X}{CX}=\dsum_{k=1}^n\lambda_k\scal{X_k}{BX_k}$.

  1. Si $A,B$ sont positives, alors, pour tout $ k\in \inter{1,n},\,\lambda_k\geq 0$ et pour tout $Y\in \R^n,\,\,\scal{Y}{BY}\geq 0$, ce qui donne, $$\forall X\in \R^n,\,\, \scal{X}{CX}=\dsum_{k=1}^n\lambda_k\scal{X_k}{BX_k}\geq 0\Longrightarrow C \text{ est sysmétrique positive}.$$
  2. Si $A,B$ sont définies positives , alors, pour tout $ k\in \inter{1,n},\,\lambda_k>0 $ et pour tout $Y\in \R^n\setminus\{0\},\,\,\scal{Y}{BY}> 0$, ce qui donne, $$\forall X\in \R^n\setminus\{0\},\,\neq 0,\, \scal{X}{CX}=\dsum_{k=1}^n\lambda_k\scal{X_k}{BX_k}> 0\Longrightarrow C \text{ est sysmétrique définie positive}.$$ (Remarque: si $X$ est non nul alors il existe au moins $k\in \inter{1,n}$ tel que $X_k\neq 0$ car $P$ est inversible)

($\star$) Correction

Soient $E$ un espace vectoriel euclidien, $a$ et $b$ deux endomorphismes symétriques définis positifs tels que \begin{equation*} \forall x\in E\setminus\{0\} \qquad \scal{a(x)}{x}< \scal{b(x)}{x}. \end{equation*} Soit $\lambda\in\Sp(b^{-1}a)$. Montrer que $\sum\limits_{n\geq 1}\dfrac{\lambda^n}{n}$ converge.

Correction

Soit $\lambda\in\Sp(b^{-1}a)$. Alors il existe $x\neq 0$ tel que $b^{-1}\big(a(x)\big)=\lambda x$, donc $a(x)=\lambda b(x)$ donc \begin{equation*} \scal{a(x)}{x} = \lambda \scal{b(x)}{x} < \scal{b(x)}{x}. \end{equation*} Or, $a$ et $b$ étant définis positifs, on en en déduit que $\lambda>0$ d'une part et que $\lambda< 1$ d'autre part. Et notamment, la série de l'énoncé converge.

(CCP PC 2003) Correction

On pose $E=\R_n[X]$ que l'on munit du produit scalaire $\scal PQ=\dsp\int_{-1}^1 P(t)\,Q(t)\,\ud t$. On considère l'application $$\fonct{u}{E}{\R[X]}{P}{2XP'+(X^2-1)P''}.$$

  1. Montrer que $u\in\LL(E)$.
  2. Montrer que $u$ est diagonalisable et que, si $P_k$ et $P_\ell$ sont des vecteurs propres de valeurs propres distinctes, alors $\scal{P_k}{P_\ell}=0$.
  3. Trouver les éléments propres de~$u$ pour $n=3$.

Correction

  1. immédiat
  2. $u$ est symétrique pour le produit scalaire considéré. En effet, soient $P,Q\in E$, on a $$\begin{array}{lcl} \scal{u(P)}{Q}&=&\dsp \int_{-1}^1\left(2tP'(t)+(t^2-1)p''(t)\right)Q(t)\ud t=\int_{-1}^1\left((t^2-1)P'(t)\right)'Q(t)\ud t\\ &=&\dsp \left[(t^2-1)P'(t)Q(t)\right]_{-1}^1-\int_{-1}^1(t^2-1)P'(t)Q'(t)\ud t=-\int_{-1}^1(t^2-1)P'(t)Q'(t)\ud t\\ &=&\dsp -\left[(t^2-1)Q'(t)P(t)\right]_{-1}^1+\int_{-1}^1(2tQ'(t)+(t^2-1)Q''(t))P(t)\ud t=\scal{P}{u(Q)} \end{array}$$ On en déduit que $u$ est symétrique.
  3. On suppose dans cette question que $E=\R_3[X]$, on a $$ \left\{\begin{array}{lcl} u(1)&=&0\\ u(X)&=&2X\\ u(X^2)&=&6X^2-2\\ u(X^3)&=&12X^3-6X \end{array} \right. \Longrightarrow M(u)=\begin{pmatrix} 0&0&-2&0\\ 0&2&0&-6\\ 0&0&6&0\\ 0&0&0&12 \end{pmatrix}\Longrightarrow \Sp(u)=\{0,2,6,12\}.$$ Après calculs, on trouve: $$P_0=1,\quad P_1=X,\quad P_2=3X^2-1,\quad P_{3}=5X^3-3X.$$

(X 2015) Correction

Soit $A\in \MM_n(\R)$ inversible. Montrer qu'il existe une base orthonormale $(X_1,\cdots,X_n)$ de $\R^n$ telle que $(AX_1,\cdots,AX_n)$ soit aussi une base orthogonale de $\R^n$.

Correction

On pose $B=A^t\,A$, $B$ est symétrique réelle donc diagonalisable dans une b.o $(X_1,\cdots,X_n)$, avec $BX_i=\lambda_i X_i$ ($\lambda_i\in \Sp (B)$).
Comme la matrice $A$ est inversible alors $(AX_1,\cdots,AX_n)$ est une base de $\R^n$.

Soit $i,\,j\in \inter{1,n}$ tels que $i\neq j$, on a $$\scal{AX_i}{AX_j}= (AX_i)^t\,(AX_j) =X_i^tA^t\,AX_j =X_i^t BX_j =\lambda_j \scal{X_i}{X_j}=0$$ Ceci prouve que la famille $(AX_1,\cdots,AX_n)$ est orthogonale.