Edit page

Diferenciabilidade em Rⁿ

Diferenciabilidade em Rⁿ

Recordando a definição de diferenciabilidade em R\R, tínhamos que ff é diferenciável em aa se existe lim⁡h→0f(a+h)−f(a)h\displaystyle \lim_{h\to 0} \frac{f(a+h)-f(a)}{h}.

No entanto, este limite envolve a divisão e a divisão não está definida em dim>1dim > 1.
Deste modo, temos de reformular esta definição.

f′(a)=lim⁡h→0f(a+h)−f(a)h⇔⇔0=lim⁡h→0f(a+h)−f(a)h−f′(a)⇔⇔0=lim⁡h→0f(a+h)−f(a)−f′(a)hh⇔⇔f(a+h)−f(a)−f′(a)h=o(h), h→0\begin{darray}{l} f'( a) =\lim _{h\rightarrow 0}\frac{f( a+h) -f( a)}{h} \Leftrightarrow\\ \Leftrightarrow 0=\lim _{h\rightarrow 0}\frac{f( a+h) -f( a)}{h} -f'( a) \Leftrightarrow \\ \Leftrightarrow 0=\lim _{h\rightarrow 0}\frac{f( a+h) -f( a) -f'( a) h}{h} \Leftrightarrow \\ \Leftrightarrow f( a+h) -f( a) -f'( a) h=o( h) ,\ \quad h\rightarrow 0 \end{darray}

"o" pequeno de h

Esta nova definição contém o(h)o(h), uma nova função ao qual que chama "o pequeno de h".
Isto é simplesmente uma função de xx que dividida por xx tende para 00 quando x→0x \to 0.

Exemplos

Quando x→0x \to 0:

  1. f(x)=0f(x) = 0, f(x)x=0x=0→0\frac{f(x)}{x}=\frac{0}{x}=0\to 0

  2. f(x)=x2f(x) = x^2, f(x)x=x2x=x→0\frac{f(x)}{x}=\frac{x^2}{x}=x\to 0

  3. f(x)=xsin⁡xf(x) = x\sin x, f(x)x=xsin⁡xx=sin⁡x→0\frac{f(x)}{x}=\frac{x\sin x}{x}=\sin x \to 0

  4. g(x)=xg(x) = x, g(x)x=xx=1↛0\frac{g(x)}{x}=\frac{x}{x}=1 \not\to 0 logo, esta não é o(x)o(x)

Diferenciabilidade como Transformação Linear

Se continuarmos a desenvolver a equação obtida acima, obtemos:

f(a+h)−f(a)−f′(a)h=o(h)⇔⇔f(a+h)−f(a)=f′(a)h+o(h)f(a+h)-f(a)-f'(a)h=o(h)\Leftrightarrow \\ \Leftrightarrow f(a+h)-f(a)=f'(a)h+o(h)

No contexto deste problema, ff e aa são constantes e hh é variável.

Então, temos a função Df(a) (h)=f′(a)hDf(a)\ (h)=f'(a)h, que é uma transformação linear Df(a):R→RDf(a): \R\to \R.

Assim, chegamos à definição em dim>1dim > 1:

DEFINIÇÃO

Seja f:D⊆Rn→Rmf: D \subseteq \R^n \to \R^m é diferencial em a∈Da\in D se existir uma transformação linear Df(a):Rn→RmDf(a): \R^n \to \R^m (chamada transformação linear derivada de ff em aa), tal que

f(a+h)−f(a)=Df(a) (h)+o(h),quandoh→0f(a+h)-f(a)=Df(a)\ (h)+o(h), \quad \text{quando}\quad h \to 0

Também podemos reescrever esta definição como

lim⁡h→0⃗f(a+h)−f(a)−Df(a)(h)∣∣h∣∣=0⃗\lim_{h\to \vec 0} \frac{f(a+h)-f(a)-Df(a)(h)}{||h||}=\vec 0

Relembrando da Álgebra Linear:

  1. Uma transformação linear T:Rn→RmT:\R^n \to \R^m é univocamente representada por uma matriz (m×nm\times n) uma vez que se fixem bases em Rn\R^n e Rm\R^m.

  2. Quais são as bases preferidas?

    • A base canónica e1→=(10⋮0),e2→=(01⋮0),en→=(00⋮1)\overrightarrow{e_{1}} =\begin{pmatrix}1\\0\\\vdots \\0\end{pmatrix} ,\overrightarrow{e_{2}} =\begin{pmatrix}0\\1\\\vdots \\0\end{pmatrix} ,\overrightarrow{e_{n}} =\begin{pmatrix}0\\0\\\vdots \\1\end{pmatrix}
    • Alguma base que torne a matriz diagonal (ou "quase" diagonal)

    Por defeito, usamos a base canónica.

  3. T:R→RT:\R \to \R é representada pela matriz [c][c] onde cc é constante.

Exemplos de funções diferenciáveis
  1. Função constante

    f:Rn→Rmx⟶(22⋮2)\begin{array}{ c c } f: & \mathbb{R}^{n}\rightarrow \mathbb{R}^{m}\\ & x\longrightarrow \begin{pmatrix} 2\\ 2\\ \vdots \\ 2 \end{pmatrix} \end{array}

    Seja a∈Rna \in \R^n a=(a1a2⋮an)a=\begin{pmatrix}a_{1}\\a_{2}\\\vdots \\a_{n}\end{pmatrix}, h⃗∈Rn\vec h \in \R^n h⃗=(h1h2⋮hn)\vec h=\begin{pmatrix}h_{1}\\h_{2}\\\vdots \\h_{n}\end{pmatrix}

    f(a+h)−f(a)=(22⋮2)−(22⋮2)=(2−22−2⋮2−2)=(00⋮0)==[00…000…0⋮⋮⋱⋮00…0]⏞matriz m×n(h1h2⋮hn)⏟La(h)+(00⋮0)⏟o(h)f(a+h)-f(a)=\begin{pmatrix} 2\\ 2\\ \vdots \\ 2 \end{pmatrix} -\begin{pmatrix} 2\\ 2\\ \vdots \\ 2 \end{pmatrix} =\begin{pmatrix} 2-2\\ 2-2\\ \vdots \\ 2-2 \end{pmatrix} =\begin{pmatrix} 0\\ 0\\ \vdots \\ 0 \end{pmatrix} =\\ =\underbrace{\overbrace{\begin{bmatrix} 0 & 0 & \dotsc & 0\\ 0 & 0 & \dotsc & 0\\ \vdots & \vdots & \ddots & \vdots \\ 0 & 0 & \dotsc & 0 \end{bmatrix}}^{\text{matriz } m\times n}\begin{pmatrix} h_{1}\\ h_{2}\\ \vdots \\ h_{n} \end{pmatrix}}_{L_{a}( h)} +\underbrace{\begin{pmatrix} 0\\ 0\\ \vdots \\ 0 \end{pmatrix}}_{o( h)}

    Portanto como a∈Rna \in \R^n é genérico, então ff é diferenciável e a derivada em aa é Df(a)=0⃗Df(a) = \vec 0

  2. Função identidade

    f:Rn→Rmx⟶x\begin{array}{ c c } f: & \mathbb{R}^{n}\rightarrow \mathbb{R}^{m}\\ & x\longrightarrow x \end{array}

    Seja a∈Rna \in \R^n, h⃗∈Rn\vec h \in \R^n,

    f(a+h)−f(a)=a+h−a=h=(h1h2⋮hn)==[10…001…0⋮⋮⋱⋮00…1]⏞matriz m×n(h1h2⋮hn)⏟La(h)+(00⋮0)⏟o(h)f(a+h)-f(a)=a+h-a=h=\begin{pmatrix} h_{1}\\ h_{2}\\ \vdots \\ h_{n} \end{pmatrix} =\\ =\underbrace{\overbrace{\begin{bmatrix} 1 & 0 & \dotsc & 0\\ 0 & 1 & \dotsc & 0\\ \vdots & \vdots & \ddots & \vdots \\ 0 & 0 & \dotsc & 1 \end{bmatrix}}^{\text{matriz } m\times n}\begin{pmatrix} h_{1}\\ h_{2}\\ \vdots \\ h_{n} \end{pmatrix}}_{L_{a}( h)} +\underbrace{\begin{pmatrix} 0\\ 0\\ \vdots \\ 0 \end{pmatrix}}_{o( h)}

    Portanto ff é diferenciável em Rn\R^n e a derivada em qualquer a∈Rna \in \R^n é a identidade.

  3. Função projeção

    pi:Rn⟶Rx=(x1x2⋮xn)⟶xi\begin{array}{ c c } p_{i} : & \mathbb{R}^{n} \longrightarrow \mathbb{R}\\ & x=\begin{pmatrix} x_{1}\\ x_{2}\\ \vdots \\ x_{n} \end{pmatrix} \longrightarrow x_{i} \end{array}

    Seja a∈Rn,h∈Rna \in \R^n, h\in \R^n,

pi(a+h)−pi(a)=ai+hi−ai=hi=[0…1…0]⏞matriz 1×n(h1h2⋮hi⋮hn)⏟La(h)+(00⋮0⋮0)⏟o(h)p_{i} (a+h)-p_{i} (a)=a_{i} +h_{i} -a_{i} =h_{i}\\ =\underbrace{\overbrace{\begin{bmatrix} 0 & \dotsc & 1 & \dotsc & 0 \end{bmatrix}}^{\text{matriz } 1\times n}\begin{pmatrix} h_{1}\\ h_{2}\\ \vdots \\ h_{i}\\ \vdots \\ h_{n} \end{pmatrix}}_{L_{a}( h)} +\underbrace{\begin{pmatrix} 0\\ 0\\ \vdots \\ 0\\ \vdots \\ 0 \end{pmatrix}}_{o( h)}

Portanto as funções projeção são diferenciáveis

Derivadas Parciais

No entanto, por vezes poderá ser importante calcular derivadas segundo direções especiais - ou saber quais são essas direções especiais. Por exemplo, dada uma função ff e um ponto aa do seu domínio, ao longo de que direção e sentido me devo afastar de aa de forma a sentir a variação máxima ou mínima da função?

DEFINIÇÃO

A derivada de f:Rn→Rf: \R^n \to \R no ponto a∈Rna \in \R^n segundo o vetor v∈Rnv \in\R^n é o limite, se existir,

∂f∂v(a)=lim⁡t→0f(a+tv)−f(a)t\frac{\partial f}{\partial v}(a)= \lim_{t\to 0} \frac{f(a+tv)-f(a)}{t}

Observações:

  1. Convém, por vezes, que ∣∣v∣∣=1||v|| = 1

  2. Se v=ei→=(0⋮1⋮0)v=\overrightarrow{e_{i}} =\begin{pmatrix}0\\\vdots \\1\\\vdots \\0\end{pmatrix} então ∂f∂v(a)=∂f∂xi(a)\displaystyle \frac{\partial f}{\partial v}( a) =\frac{\partial f}{\partial x_{i}}( a), a que se chama derivada parcial em ordem a xix_i em aa.

Exemplos
f(x,y)=x2y,v=(1,2)f( x,y) =x^{2} y\quad ,\quad v=( 1,2)

Qual o valor de ∂f∂v(1,1)\frac{\partial f}{\partial v}( 1,1)?

∂f∂v(1,1)=lim⁡t→0f((1,1)+t(1,2))−f(1,1)t=lim⁡t→0f(1+t,1+2t)−12⋅1t=lim⁡t→0(1+t)2(1+2t)−1t=lim⁡t→01+2t+2t+4t2+t2+2t3−1t=lim⁡t→0t(4+5t+2t2)t=lim⁡t→0(4+5t+2t2)=4+0+0=4\begin{aligned} \frac{\partial f}{\partial v}( 1,1) & =\lim _{t\rightarrow 0}\frac{f(( 1,1) +t( 1,2)) -f( 1,1)}{t}\\ & =\lim _{t\rightarrow 0}\frac{f( 1+t,1+2t) -1^{2} \cdot 1}{t}\\ & =\lim _{t\rightarrow 0}\frac{( 1+t)^{2}( 1+2t) -1}{t}\\ & =\lim _{t\rightarrow 0}\frac{1+2t+2t+4t^{2} +t^{2} +2t^{3} -1}{t}\\ & =\lim _{t\rightarrow 0} \frac{t \left(4+5t+2t^{2}\right)}{t}\\ & =\lim _{t\rightarrow 0}\left( 4+5t+2t^{2}\right)\\ & =4+0+0\\ & =4 \end{aligned}

Propriedades de uma função diferenciável

Se ff é diferenciável em aa, isto é, se existe transformação linear Df(a):Rn→RmDf(a): \R^n\to \R^m tal que f(a+h)−f(a)=Df(a) (h)+o(h)f(a+h)-f(a)=Df(a)\ (h)+o(h), então:

  • ff é contínua em aa
  • Para todo o v∈Rn\{0⃗}v \in \R^n\backslash \{\vec 0 \} existe ∂f∂v(a)=Df(a) (v)\displaystyle \frac{\partial f}{\partial v}( a) = Df(a)\ (v) (transformação linear derivada de ff em aa calculada em vv).

Sejam f,g:D⊆Rn→Rmf, g: D \subseteq \R^n \to \R^m, λ∈R,a∈D,h∈Rn\lambda \in \R, a \in D, h \in \R^n, em que ff e gg são diferenciáveis em aa.

Então,

  • f+gf+g é diferenciável em aa
  • λf\lambda f é diferenciável em aa
  • fgfg é diferenciável em aa, sendo D(fg)(a)=g(a)Df(a)+f(a)Df(a)D( fg)( a) =g( a) Df( a) +f( a) Df( a)
  • fg\frac fg é diferenciável em aa se g(a)≠0g(a)\ne 0, sendo D(fg)(a)=g(a)Df(a)−f(a)Dg(a)g(a)2D\left(\frac{f}{g}\right)( a) =\frac{g( a) Df( a) -f( a) Dg( a)}{g( a)^{2}}

Seja f:D⊂Rn→Rmf: D \subset \R^n \to \R^m e DD conjunto aberto

  • Se ff é C1C^1 em DD então ff é diferenciável em DD.

Matriz Jacobiana

Se a função ff é diferenciável em aa então a transformação linear derivada é única.
Fixadas as bases canónicas, a matriz que a transformação linear representa é chamada jacobiana e tem por entradas as derivadas parciais em aa.

Temos então a matriz jacobiana de ff em aa:

Jaf=[∂f1∂x1(a)∂f1∂x2(a)…∂f1∂xn(a)∂f2∂x1(a)∂f2∂x2(a)…∂f2∂xn(a)⋮⋮⋱⋮∂fm∂x1(a)∂fm∂x2(a)…∂fm∂xn(a)]J^{f}_{a} =\begin{bmatrix} \frac{\partial f_{1}}{\partial x_{1}}( a) & \frac{\partial f_{1}}{\partial x_{2}}( a) & \dotsc & \frac{\partial f_{1}}{\partial x_{n}}( a)\\ \frac{\partial f_{2}}{\partial x_{1}}( a) & \frac{\partial f_{2}}{\partial x_{2}}( a) & \dotsc & \frac{\partial f_{2}}{\partial x_{n}}( a)\\ \vdots & \vdots & \ddots & \vdots \\ \frac{\partial f_{m}}{\partial x_{1}}( a) & \frac{\partial f_{m}}{\partial x_{2}}( a) & \dotsc & \frac{\partial f_{m}}{\partial x_{n}}( a) \end{bmatrix}
Exemplos de matrizes jacobianas
f:R→R,f(x)=xexf:\R \to \R\quad, \quad f(x)=xe^x
dfdx=1ex+x⋅ex=(1+x)ex\frac{df}{dx}=1e^x+x\cdot e^x=(1+x)e^x
Jxf=[(1+x)ex]J^f_x=\begin{bmatrix}(1+x) e^x\end{bmatrix}

f:R2→R,f(x,y)=x2yf: \R^2\to \R\quad,\quad f(x,y)=x^2y
∂f∂x(x,y)=2xy∂f∂y(x,y)=x2⋅1=x2\frac{\partial f}{\partial x}(x,y)=2xy\\ \frac{\partial f}{\partial y}(x,y)=x^2\cdot 1 = x^2
J(x,y)f=[2xyx2]J^f_{(x,y)}=\begin{bmatrix}2xy& x^2\end{bmatrix}

f:R→R2,f(x)=(cos⁡x,sin⁡x)f: \R\to \R^2\quad,\quad f(x)=(\cos x, \sin x)
Jxf=[−sin⁡xcos⁡x]J^f_x=\begin{bmatrix} -\sin x \\ \cos x \end{bmatrix}

f:R2→R2,f(x,y)=(x2−y2⏞f1,x2+y2⏞f2)f: \R^2\to \R^2\quad,\quad f(x,y)=(\overbrace{x^2-y^2}^{f1}, \overbrace{x^2+y^2}^{f2})
J(x,y)f=[2x−2y2x2y]J^f_{(x,y)}=\begin{bmatrix}2x & -2y\\ 2x & 2y\end{bmatrix}

warning

Pode existir a jacobiana de uma função ff em aa sem que ff seja diferenciável em aa.

Gradiente de uma função

Seja f:D⊆Rn→Rf: D \subseteq \R^n \to \R isto é, seja ff uma função escalar. Se ff é diferenciável em aa então a matriz jacobiana de ff em aa é uma matriz linha que se chama gradiente de ff em aa.

∇f(a)=(∂f∂x1(a),∂f∂x2(a),…,∂f∂xn(a))\nabla f( a) =\left(\frac{\partial f}{\partial x_{1}}( a) ,\frac{\partial f}{\partial x_{2}}( a) ,\dotsc ,\frac{\partial f}{\partial x_{n}}( a)\right)
Exemplos de diferenciabilidade

Seja f:R2→Rf: \R^2 \to \R com f(x,y)=x2yf(x,y) = x^2y

  1. Mostre que ff é diferenciável em (1,1)(1,1)

ff é diferenciável em (1,1)(1,1) se por definição:

lim⁡(h,k)→(0,0)f((1,1)+(h,k))−f(1,1)−Df(1,1)(h,k)∣∣(h,k)∣∣=0\lim_{(h,k)\to (0,0)}\frac{f((1,1)+(h,k))-f(1,1)-Df(1,1)(h,k)}{||(h,k)||}=0
J(1,1)f=(∂f∂x(1,1)∂f∂y(1,1))=(2xy∣(1,1)x2∣(1,1))=(21)J^{f}_{(1,1)} =\begin{pmatrix} \frac{\partial f}{\partial x}( 1,1) & \frac{\partial f}{\partial y}( 1,1) \end{pmatrix} =\begin{pmatrix} 2xy\Bigl|_{( 1,1)} & x^{2}\Bigl|_{( 1,1)} \end{pmatrix} =\begin{pmatrix} 2 & 1 \end{pmatrix}
lim⁡(h,k)→(0,0)f(1+h,1+k)−f(1,1)−(21)(hk)h2+k2==lim⁡(h,k)→(0,0)(1+h)2(1+k)−12⋅1−(2h+k)h2+k2=lim⁡(h,k)→(0,0)1+2h+h2+k+2kh+kh2−1−2h−kh2+k2=lim⁡(h,k)→(0,0)h2+2kh+kh2h2+k2=lim⁡(h,k)→(0,0)hh+2k+kh h2+k2=lim⁡(h,k)→(0,0)(h+2k+kh)hh2+k2=0\begin{darray}{ l } \lim\limits _{( h,k)\rightarrow ( 0,0)}\frac{f( 1+h,1+k) -f( 1,1) -\begin{pmatrix} 2 & 1 \end{pmatrix}\begin{pmatrix} h\\ k \end{pmatrix}}{\sqrt{h^{2} +k^{2}}} =\\ =\lim\limits _{( h,k)\rightarrow ( 0,0)}\frac{( 1+h)^{2}( 1+k) -1^{2} \cdot 1-( 2h+k)}{\sqrt{h^{2} +k^{2}}}\\ =\lim\limits _{( h,k)\rightarrow ( 0,0)}\frac{1+2h+h^{2} +k+2kh+kh^{2} -1-2h-k}{\sqrt{h^{2} +k^{2}}}\\ =\lim\limits _{( h,k)\rightarrow ( 0,0)}\frac{h^{2} +2kh+kh^{2}}{\sqrt{h^{2} +k^{2}}}\\ =\lim\limits _{( h,k)\rightarrow ( 0,0)} h\frac{h+2k+kh}{\ \sqrt{h^{2} +k^{2}}}\\ =\lim\limits _{( h,k)\rightarrow ( 0,0)}( h+2k+kh)\frac{h}{\sqrt{h^{2} +k^{2}}}\\ =0 \end{darray}
c.a.∣hh2+k2∣=∣h∣h2+k2=h2h2+k2⩽h2+k2h2+k2=1\text{c.a.}\\ \left| \frac{h}{\sqrt{h^{2} +k^{2}}}\right| =\frac{|h|}{\sqrt{h^{2} +k^{2}}} =\frac{\sqrt{h^{2}}}{\sqrt{h^{2} +k^{2}}} \leqslant \frac{\sqrt{h^{2} +k^{2}}}{\sqrt{h^{2} +k^{2}}} =1

Logo, ff é diferenciável em aa.

  1. Calcule ∂f∂v(1,1)\frac{\partial f}{\partial v}(1,1) para v=(1,2)v=(1,2)
∂f∂v(1,1)=Jaf(12)=(21)(12)=(2⋅1+1⋅2)=(4)\frac{\partial f}{\partial v}( 1,1) =J^{f}_{a}\begin{pmatrix} 1\\ 2 \end{pmatrix} =\begin{pmatrix} 2 & 1 \end{pmatrix}\begin{pmatrix} 1\\ 2 \end{pmatrix} =\begin{pmatrix} 2\cdot 1+1\cdot 2 \end{pmatrix} =\begin{pmatrix} 4 \end{pmatrix}

Seja f:Rn→Rmf: \R^n \to \R^m, f(x)=Axf(x)=Ax, uma transformação linear com AA a matriz que a representa, fixadas as bases canónicas.

Seja a∈Rn,h∈Rna\in\R^n, h \in \R^n,

f(a+h)−f(a)=A(a+h)−Aa=Aa+Ah−Aa=Ah+0⃗f(a+h)-f(a)=A(a+h)-Aa=Aa+Ah-Aa=Ah+\vec 0

AhAh é a transformação linear calculada em hh
0⃗\vec 0 é o o(h),h→0⃗o(h), h\to \vec 0

Portanto ff é diferenciável em qualquer a∈Rna\in\R^n e Df(a)=ADf(a)=A (AA é uma matriz)


Sendo, f:R3→R,f(x,y,z)=xyz,v=(−1,0,1)f:\R^3 \to \R\quad, \quad f(x,y,z)=xyz\quad,\quad v=(-1,0,1)

Calcular ∂f∂v(2,1,1)\frac{\partial f}{\partial v}(2,1,1):

  1. Pela definição
∂f∂v(2,1,1)=lim⁡t→0f((2,1,1)+tv)−f(2,1,1)t=lim⁡t→0f((2,1,1)+t(−1,0,1))−f(2,1,1)t=lim⁡t→0f(2−t,1,1+t)−f(2,1,1)t==lim⁡t→0(2−t)⋅1⋅(1+t)−2⋅1⋅1t==lim⁡t→02+2t−t−t2−2t==lim⁡t→0t(1−t)t=lim⁡t→0(1−t)=1\begin{aligned} \frac{\partial f}{\partial v}(2,1,1) &= \lim_{t\to 0}\frac{f((2,1,1)+tv)-f(2,1,1)}{t}\\ &= \lim_{t\to 0}\frac{f((2,1,1)+t(-1,0,1))-f(2,1,1)}{t}\\ &= \lim_{t\to 0}\frac{f(2-t,1,1+t)-f(2,1,1)}{t}=\\ &=\lim_{t \to 0}\frac{(2-t)\cdot 1 \cdot (1+t) - 2\cdot 1 \cdot 1}{t}=\\ &= \lim_{t \to 0}\frac{2+2t-t-t^2-2}{t}=\\ &=\lim_{t \to 0}\frac{t(1-t)}{t}\\ &=\lim_{t\to 0}(1-t)\\ &=1 \end{aligned}

ou seja ∂f∂v(2,1,1)=1\frac{\partial f}{\partial v}(2,1,1) = 1

  1. Através da Matriz Jacobiana

f(x,y,z)=xyz=p1(x,y,z)⋅p2(x,y,z)⋅p3(x,y,z)f(x,y,z) = xyz = p_1(x,y,z) \cdot p_2(x,y,z)\cdot p_3(x,y,z)

Como pip_i's são diferenciáveis e produto de funções diferenciáveis é diferenciável, então ff é diferenciável em qualquer ponto e existe a transformação linear derivada em qualquer ponto do domínio de ff.

J(2,1,1)f=[∂f∂x(2,1,1)∂f∂y(2,1,1)∂f∂z(2,1,1)]=[d(xyz) dx∣(2,1,1)d(xyz)dy∣(2,1,1)d(xyz)dz∣(2,1,1)]=[yzdxdx∣(2,1,1)xzdydy∣(2,1,1)xydzdz∣(2,1,1)]=[yz⋅1∣(2,1,1)xz⋅1∣(2,1,1)xy⋅1∣(2,1,1)]=[122]\begin{aligned} J^{f}_{(2,1,1)} & =\begin{bmatrix} \frac{\partial f}{\partial x} (2,1,1) & \frac{\partial f}{\partial y} (2,1,1) & \frac{\partial f}{\partial z} (2,1,1) \end{bmatrix}\\ & =\begin{bmatrix} \frac{\mathrm{d} (xyz)}{\ \mathrm{d} x}\Bigl|_{(2,1,1)} & \frac{\mathrm{d} (xyz)}{\mathrm{d} y}\Bigl|_{(2,1,1)} & \frac{\mathrm{d} (xyz)}{\mathrm{d} z}\Bigl|_{(2,1,1)} \end{bmatrix}\\ & =\begin{bmatrix} yz\frac{\mathrm{d} x}{\mathrm{d} x}\Bigl|_{(2,1,1)} & xz\frac{\mathrm{d} y}{\mathrm{d} y}\Bigl|_{(2,1,1)} & xy\frac{\mathrm{d} z}{\mathrm{d} z}\Bigl|_{(2,1,1)} \end{bmatrix}\\ & =\begin{bmatrix} yz\cdot 1\Bigl|_{(2,1,1)} & xz\cdot 1\Bigl|_{(2,1,1)} & xy\cdot 1\Bigl|_{(2,1,1)} \end{bmatrix}\\ & =\begin{bmatrix} 1 & 2 & 2 \end{bmatrix} \end{aligned}

Como ff é diferenciável, em particular em (2,1,1) tem-se

∂t∂v(2,1,1)=J(2,1,1)fv=[122][−101]==1⋅(−1)+2⋅0+1⋅2=−1+2=1\frac{\partial t}{\partial v}( 2,1,1) =J^{f}_{( 2,1,1)} v=\begin{bmatrix} 1 & 2 & 2 \end{bmatrix}\begin{bmatrix} -1\\ 0\\ 1 \end{bmatrix} =\\ =1\cdot ( -1) +2\cdot 0+1\cdot 2=-1+2=1

f(x,y)={x2yx2+y2 se (x,y)≠00se (x,y)=0f( x,y) =\begin{cases} \frac{x^{2} y}{x^{2} +y^{2}} \ & \text{se } (x,y) \neq 0\\ 0 & \text{se }( x,y) =0 \end{cases}
  1. Será que f é diferenciável em (0,0)(0,0)?

Para ff ser diferenciável em (0,0)(0,0), o seguinte tem de ser verdade.

f((a1,a2)+(h1,h2))−f(a1,a2)−Df(a1,a2)(h1,h2)∣∣(h1,h2)∣∣⟶(h1,h2)→(0,0)0\frac{f((a_1, a_2) + (h_1,h_2))-f(a_1,a_2)-Df(a_1,a_2)(h_1,h_2)}{||(h_1,h_2)||} \underset{(h_1,h_2)\to (0,0)}{\longrightarrow} 0

Se existir Df(0,0)Df(0,0) a jacobiana é J(0,0)f=[∂f∂x(0,0)∂f∂y(0,0)]J^{f}_{(0,0)} =\begin{bmatrix}\frac{\partial f}{\partial x} (0,0) & \frac{\partial f}{\partial y} (0,0)\end{bmatrix}

∂f∂x(0,0)=∂f∂e1→(0,0)=lim⁡t→0f((0,0)+t(1,0))−f(0,0)t=lim⁡t→0f(t,0)−0t=lim⁡t→0t2⋅0t2+02t=lim⁡t→00t3=0\begin{aligned} \frac{\partial f}{\partial x}( 0,0) & =\frac{\partial f}{\partial \overrightarrow{e_{1}}}( 0,0)\\ & =\lim _{t\rightarrow 0}\frac{f(( 0,0) +t( 1,0)) -f( 0,0)}{t}\\ & =\lim _{t\rightarrow 0}\frac{f( t,0) -0}{t}\\ & =\lim _{t\rightarrow 0}\frac{\frac{t^{2} \cdot 0}{t^{2} +0^{2}}}{t}\\ & =\lim _{t\rightarrow 0}\frac{0}{t^{3}}\\ & =0 \end{aligned}
∂f∂y(0,0)=∂f∂e2→(0,0)=lim⁡t→0f((0,0)+t(0,1))−f(0,0)t=lim⁡t→0f(0,t)−0t=lim⁡t→00⋅t02+t2t=lim⁡t→00t2=0\begin{aligned} \frac{\partial f}{\partial y}( 0,0) & =\frac{\partial f}{\partial \overrightarrow{e_{2}}}( 0,0)\\ & =\lim _{t\rightarrow 0}\frac{f(( 0,0) +t( 0,1)) -f( 0,0)}{t}\\ & =\lim _{t\rightarrow 0}\frac{f( 0,t) -0}{t}\\ & =\lim _{t\rightarrow 0}\frac{\frac{0\cdot t}{0^{2} +t^{2}}}{t}\\ & =\lim _{t\rightarrow 0}\frac{0}{t^{2}}\\ & =0 \end{aligned}

Portanto se ff for diferenciável em (0,0)(0,0) a Df(0,0)Df(0,0) é representada pela jacobiana J(0,0)f=[00]J^{f}_{(0,0)} =\begin{bmatrix}0 & 0\end{bmatrix}

f((0,0)+(h1,h2))−f(0,0)−Df(0,0)(h1,h2)∣∣(h1,h2)∣∣==f(h1,h2)−0−(00)(h1h2)h12+h22==h12⋅h2h12+h22−0h1−0h2h12+h22==h12⋅h2(h12+h22)32\begin{array}{ c } \frac{f(( 0,0) +( h_{1} ,h_{2})) -f( 0,0) -Df( 0,0)( h_{1} ,h_{2})}{||( h_{1} ,h_{2}) ||} =\\ =\frac{f( h_{1} ,h_{2}) -0-\begin{pmatrix} 0 & 0 \end{pmatrix}\begin{pmatrix} h_{1}\\ h_{2} \end{pmatrix}}{\sqrt{h^{2}_{1} +h^{2}_{2}}} =\\ =\frac{\frac{h^{2}_{1} \cdot h_{2}}{h^{2}_{1} +h^{2}_{2}} -0h_{1} -0h_{2}}{\sqrt{h^{2}_{1} +h^{2}_{2}}} =\\ =\frac{h^{2}_{1} \cdot h_{2}}{\left( h^{2}_{1} +h^{2}_{2}\right)^{\frac{3}{2}}} \end{array}

Não conseguimos determinar o limite, mas podemos tentar provar que não existe através dos limites direcionais h2=mh1h_2=mh_1.

lim⁡(h1,h2)→(0,0)h12⋅h2(h12+h22)32=lim⁡h1→0h12⋅mh1(h12+(mh1)2)32=lim⁡h1→0mh13h13(1+m2)32=lim⁡h1→0m(1+m)32=m(1+m)32\begin{aligned} \lim _{( h_{1} ,h_{2})\rightarrow ( 0,0)}\frac{h^{2}_{1} \cdot h_{2}}{\left( h^{2}_{1} +h^{2}_{2}\right)^{\frac{3}{2}}} & =\lim _{h_{1}\rightarrow 0}\frac{h^{2}_{1} \cdot mh_{1}}{\left( h^{2}_{1} +( mh_{1})^{2}\right)^{\frac{3}{2}}}\\ & =\lim _{h_{1}\rightarrow 0}\frac{mh^{3}_{1}}{h^{3}_{1}\left( 1+m^{2}\right)^{\frac{3}{2}}}\\ & =\lim _{h_{1}\rightarrow 0}\frac{m}{( 1+m)^{\frac{3}{2}}}\\ & =\frac{m}{( 1+m)^{\frac{3}{2}}} \end{aligned}

Logo, como o limite depende de mm, este não existe.

Como tal, a função não é diferenciável em (0,0)(0,0). No entanto, existem as derivadas parciais de ff em (0,0)(0,0).

  1. Qual o valor de ∂f∂v(0,0),v=(1,1)\frac{\partial f}{\partial v}(0,0)\quad, \quad v=(1,1)?

Como a função não é diferenciável em (0,0)(0,0), não se pode usar ∂f∂v(0,0)=Df(0,0)v\frac{\partial f}{\partial v}(0,0) = Df(0,0) v.
Tem-se então, de efetuar o cálculo pelo limite.

∂f∂v(0,0)=lim⁡t→0f((0,0)+t(1,1))−f(0,0)t=lim⁡t→0f(t,t)−0t=lim⁡t→0t2⋅tt2+t2t=lim⁡t→0t3t(2t2)=lim⁡t→0t32t3=lim⁡t→012=12\begin{aligned} \frac{\partial f}{\partial v}(0,0) &= \lim_{t\to 0}\frac{f((0,0) + t(1,1))- f(0,0)}{t}\\ & =\lim_{t\to 0}\frac{f(t,t)-0}{t}\\ & =\lim_{t\to 0} \frac{\frac{t^2\cdot t}{t^2+t^2}}{t}\\ & =\lim_{t\to 0} \frac{t^3}{t(2t^2)}\\ & =\lim_{t\to 0} \frac{t^3}{2t^3}\\ & =\lim_{t \to 0} \frac{1}{2} \\ & = \frac{1}{2} \end{aligned}

f:R2→R,f(x,y,)=∣∣(x,y)∣∣=x2+y2f: \R^2\to \R \quad,\quad f(x,y,) = ||(x,y)|| = \sqrt{x^2+y^2}

Pretende-se mostrar que ff não é diferenciável na origem ((0,0)(0,0)), ou seja, mostrar que o seguinte limite não existe ou que é diferente de zero.

lim⁡(h1,h2)→(0,0)f((0,0)+(h1,h2))−f(0,0)−Df(0,0)(h1,h2)∣∣(h1,h2)∣∣\lim_{(h_1,h_2)\to(0,0)} \frac{f((0,0) + (h_1,h_2)) - f(0,0) - Df(0,0) (h_1,h_2)}{||(h_1,h_2)||}

Se Df(0,0)Df(0,0) existisse:

∂f∂x(0,0)=lim⁡t→0f((0,0)+t(1,0))−f(0,0)t==lim⁡t→0f(t,0)−0t==lim⁡t→0t2+02t==lim⁡t→0∣t∣t(na˜o existe)\begin{aligned} \frac{\partial f}{\partial x}(0,0) &= \lim_{t\to 0} \frac{f((0,0) + t(1,0)) - f(0,0)}{t} = \\ &=\lim_{t \to 0}\frac{f(t,0)-0}{t}=\\ &=\lim_{t\to 0}\frac{\sqrt{t^2+0^2}}{t}=\\ &=\lim_{t\to 0}\frac{|t|}{t} \text{(não existe)} \end{aligned}

Como não existe ∂f∂x(0,0)\frac{\partial f}{\partial x}(0,0), então ff não é diferenciável em (0,0)(0,0).


Slides: