Teorema de Clairaut

En matemàtiques, el teorema de Clairaut (també conegut com a teorema de Schwarz o de Young) mostra la igualtat de les derivades creuades d'una funció f sempre que:

f : Ω R n R {\displaystyle f\colon \Omega \subseteq \mathbb {R} ^{n}\to \mathbb {R} }

tingui derivades parcials contínues per qualsevol punt del domini obert Ω {\displaystyle \Omega } , per exemple, prenguem el punt a = ( a 1 , a 2 , . . . , a n ) Ω {\displaystyle a=(a_{1},a_{2},...,a_{n})\in \Omega } , llavors, segons aquest teorema, per qualssevol i , j { 1 , , n } {\displaystyle i,j\in \{1,\ldots ,n\}} tenim que:

2 f x i x j ( a 1 , , a n ) = 2 f x j x i ( a 1 , , a n ) . {\displaystyle {\frac {\partial ^{2}f}{\partial x_{i}\,\partial x_{j}}}(a_{1},\dots ,a_{n})={\frac {\partial ^{2}f}{\partial x_{j}\,\partial x_{i}}}(a_{1},\dots ,a_{n}).}

Aquest teorema deu el seu nom al matemàtic i astrònom francès Alexis Clairaut.

Una conseqüència immediata d'això és que, si es compleixen les condicions del teorema de Clairaut, la matriu hessiana de la funció f serà simètrica.

Demostració

Denotarem f x i = f x i {\displaystyle f_{x_{i}}={\frac {\partial f}{\partial x_{i}}}} i f x i x j := 2 f x i x j {\displaystyle f_{x_{i}x_{j}}:={\frac {\partial ^{2}f}{\partial x_{i}\partial x_{j}}}} i demostrarem que si existeixen f x i , f x j  i  f x i x j {\displaystyle f_{x_{i}},f_{x_{j}}{\text{ i }}f_{x_{i}x_{j}}} en tot l'obert Ω {\displaystyle \Omega } i f x i x j {\displaystyle f_{x_{i}x_{j}}} és contínua en un punt a Ω {\displaystyle a\in \Omega } , aleshores f x j x i ( a ) = f x i x j ( a ) {\displaystyle \exists f_{x_{j}x_{i}}(a)=f_{x_{i}x_{j}}(a)} .

Sigui α := f x i x j ( a ) {\displaystyle \alpha :=f_{x_{i}x_{j}}(a)} . Per continuïtat de f x i x j {\displaystyle f_{x_{i}x_{j}}} en a {\displaystyle a} tenim que donat ε > 0 ,     r > 0 {\displaystyle \varepsilon >0,\ \ \exists r>0} tal que B ( a , r ) Ω {\displaystyle B(a,r)\subseteq \Omega } (per ser Ω {\displaystyle \Omega } obert) i | f x i x j ( x ) α | ε       x B ( a , r ) {\displaystyle |f_{x_{i}x_{j}}(x)-\alpha |\leq \varepsilon \ \ \ \forall x\in B(a,r)} . ( 1 ) {\displaystyle (1)}

Considerem δ := r 2 {\displaystyle \delta :={\frac {r}{\sqrt {2}}}} . Aleshores, denotant per e i {\displaystyle e_{i}} l' i {\displaystyle i} -èsim vector de la base canònica de R n {\displaystyle \mathbb {R} ^{n}} , per a tot h , k ( δ , δ ) {\displaystyle h,k\in (-\delta ,\delta )} , tenim que

| | a + h e i + k e j a | | = | | h e i + k e j | | = h 2 + k 2 < 2 δ 2 = 2 r 2 2 = | r | = r a + h e i + k e j B ( a , r ) ( 2 ) {\displaystyle ||a+he_{i}+ke_{j}-a||=||he_{i}+ke_{j}||={\sqrt {h^{2}+k^{2}}}<{\sqrt {2\delta ^{2}}}={\sqrt {2{\frac {r^{2}}{2}}}}=|r|=r\Rightarrow a+he_{i}+ke_{j}\in B(a,r)\quad \quad (2)}

En particular, com que, per ( 1 ) {\displaystyle (1)} , B ( a , r ) Ω {\displaystyle B(a,r)\subseteq \Omega } , podem definir la següent funció:

A : ( δ , δ ) 2 R {\displaystyle A:(-\delta ,\delta )^{2}\longrightarrow \mathbb {R} }

    ( h , k ) A ( h , k ) = f ( a + h e i + k e j ) f ( a + h e i ) f ( a + k e j ) + f ( a ) {\displaystyle \quad \quad \ \ (h,k)\longmapsto A(h,k)=f(a+he_{i}+ke_{j})-f(a+he_{i})-f(a+ke_{j})+f(a)}

Ara, donats h , k {\displaystyle h,k} amb 0 < | h | , | k | < δ {\displaystyle 0<|h|,|k|<\delta } , definim la funció

u : ( δ , δ ) R {\displaystyle u:(-\delta ,\delta )\longrightarrow \mathbb {R} }

    t u ( t ) = f ( a + h e i + t e j ) f ( a + t e j ) {\displaystyle \quad \quad \ \ t\longmapsto u(t)=f(a+he_{i}+te_{j})-f(a+te_{j})}

Per ( 2 ) {\displaystyle (2)} i ( 1 ) {\displaystyle (1)} , tenim que a + h e i + t e j , a + t e j B ( a , r ) Ω {\displaystyle a+he_{i}+te_{j},a+te_{j}\in B(a,r)\subseteq \Omega } , d'on, com que existeix f x j {\displaystyle f_{x_{j}}} per hipòtesi, u {\displaystyle u} és derivable i u = f x j ( a + h e i + t e j ) f x j ( a + t e j ) {\displaystyle u'=f_{x_{j}}(a+he_{i}+te_{j})-f_{x_{j}}(a+te_{j})} . Com que 0 , k ( δ , δ ) , {\displaystyle 0,k\in (-\delta ,\delta ),} podem aplicar el teorema del valor mitjà de Lagrange d'una variable a u {\displaystyle u} a l'interval amb extrems 0 {\displaystyle 0} i k {\displaystyle k} . Així,

ξ 0 , k  tal que  A ( h , k ) = u ( k ) u ( 0 ) = u ( ξ ) ( k 0 ) = [ f x j ( a + h e i + ξ e j ) f ( a + ξ e j ) ] k ( ) {\displaystyle \exists \xi \in \left\langle 0,k\right\rangle {\text{ tal que }}A(h,k)=u(k)-u(0)=u'(\xi )(k-0)=\left[f_{x_{j}}(a+he_{i}+\xi e_{j})-f(a+\xi e_{j})\right]k\quad \quad (*)}

Considerem ara

v : ( δ , δ ) R {\displaystyle v:(-\delta ,\delta )\longrightarrow \mathbb {R} }

    t v ( t ) = f x j ( a + t e i + ξ e j ) {\displaystyle \quad \quad \ \ t\longmapsto v(t)=f_{x_{j}}(a+te_{i}+\xi e_{j})}

Com que t ( δ , δ )  i  ξ 0 , k ( δ , δ ) {\displaystyle t\in (-\delta ,\delta ){\text{ i }}\xi \in \left\langle 0,k\right\rangle \subseteq (-\delta ,\delta )} , per ( 2 )  i  ( 1 ) {\displaystyle (2){\text{ i }}(1)} tenim que a + t e i + ξ e j B ( a , r ) Ω {\displaystyle a+te_{i}+\xi e_{j}\in B(a,r)\subseteq \Omega } , d'on, com que existeix f x i x j {\displaystyle f_{x_{i}x_{j}}} per hipòtesi, v {\displaystyle v} és derivable i v ( t ) = f x i x j ( a + t e i + ξ e j ) {\displaystyle v'(t)=f_{x_{i}x_{j}}(a+te_{i}+\xi e_{j})} .

Com que 0 , h ( δ , δ ) , {\displaystyle 0,h\in (-\delta ,\delta ),} podem aplicar el teorema del valor mitjà de Lagrange d'una variable a v {\displaystyle v} a l'interval amb extrems 0 {\displaystyle 0} i h {\displaystyle h} . Així,

η 0 , h  tal que  f x j ( a + h e i + ξ e j ) f x j ( a + ξ e j ) = v ( h ) v ( 0 ) = v ( η ) ( h 0 ) = f x i x j ( a + η e i + ξ e j ) h k {\displaystyle \exists \eta \in \left\langle 0,h\right\rangle {\text{ tal que }}f_{x_{j}}(a+he_{i}+\xi e_{j})-f_{x_{j}}(a+\xi e_{j})=v(h)-v(0)=v'(\eta )(h-0)=f_{x_{i}x_{j}}(a+\eta e_{i}+\xi e_{j})h\quad {\overset {\cdot k}{\Rightarrow }}}

k [ f x j ( a + h e i + ξ e j ) f x j ( a + ξ e j ) ] k = f x i x j ( a + η e i + ξ e j ) h k     ( )     A ( h , k ) = f x i x j ( a + η e i + ξ e j ) h k {\displaystyle {\overset {\cdot k}{\Rightarrow }}\left[f_{x_{j}}(a+he_{i}+\xi e_{j})-f_{x_{j}}(a+\xi e_{j})\right]k=f_{x_{i}x_{j}}(a+\eta e_{i}+\xi e_{j})hk~~{\overset {(*)}{\Rightarrow }}~~A(h,k)=f_{x_{i}x_{j}}(a+\eta e_{i}+\xi e_{j})hk} .

Definint z := a + η e i + ξ e j {\displaystyle z:=a+\eta e_{i}+\xi e_{j}} , com que h , k 0 h k 0 {\displaystyle h,k\neq 0\Rightarrow hk\neq 0} , tenim que A ( h , k ) h k = f x i x j ( z ) {\displaystyle {\frac {A(h,k)}{hk}}=f_{x_{i}x_{j}}(z)} . Observem que ( 2 ) z B ( a , r )   ( 1 )   | f x i x j α | ε {\displaystyle (2)\Rightarrow z\in B(a,r)~{\overset {(1)}{\Rightarrow }}~|f_{x_{i}x_{j}}-\alpha |\leq \varepsilon } . Així, tenim que | A ( h , k ) h k α | = | f x i x j ( z ) α | ε ( ) {\displaystyle \left|{\frac {A(h,k)}{hk}}-\alpha \right|=\left|f_{x_{i}x_{j}}(z)-\alpha \right|\leq \varepsilon \quad (**)} .

Finalment, observem que

A ( h , k ) h k = 1 k [ f ( a + h e i + k e j ) f ( a + k e i ) h f ( a + h e i ) f ( a ) h ] h 0     1 k [ f x i ( a + k e j ) f x i ( a ) ] {\displaystyle {\frac {A(h,k)}{hk}}={\frac {1}{k}}\left[{\frac {f(a+he_{i}+ke_{j})-f(a+ke_{i})}{h}}-{\frac {f(a+he_{i})-f(a)}{h}}\right]{\overset {h\rightarrow 0}{\longrightarrow }}~~{\frac {1}{k}}\left[f_{x_{i}}(a+ke_{j})-f_{x_{i}}(a)\right]\Rightarrow }

ε ( ) | A ( h , k ) h k α |     h 0     | f x i ( a + k e j ) f x i ( a ) k α | ε f x i ( a + k e j ) f x i ( a ) k     k 0     α f x j x i ( a ) = α   = def   f x i x j ( a ) {\displaystyle \Rightarrow \varepsilon {\overset {(**)}{\geq }}\left|{\frac {A(h,k)}{hk}}-\alpha \right|~~{\overset {h\rightarrow 0}{\longrightarrow }}~~\left|{\frac {f_{x_{i}}(a+ke_{j})-f_{x_{i}}(a)}{k}}-\alpha \right|\leq \varepsilon \Rightarrow {\frac {f_{x_{i}}(a+ke_{j})-f_{x_{i}}(a)}{k}}~~{\overset {k\rightarrow 0}{\longrightarrow }}~~\alpha \Rightarrow f_{x_{j}x_{i}}(a)=\alpha ~{\overset {\text{def}}{=}}~f_{x_{i}x_{j}}(a)\quad \square }

Bases d'informació
  • GEC (1)