Math ClubMath Club
v1 · padrão canônico

Lição 112 — Transformações lineares

Funções entre espaços vetoriais que preservam combinação linear. Representação matricial em uma base. Mudança de base. A operação fundamental que torna ML, gráficos 3D e processamento de sinais possíveis.

Used in: Leistungskurs alemão (Lineare Algebra) · Math III japonês · H2 Math singapurense · graduação engenharia 1.º semestre

T(au+bv)=aT(u)+bT(v)T(au + bv) = a\,T(u) + b\,T(v)

Uma transformação linear é uma função T:VWT: V \to W entre espaços vetoriais que preserva soma e multiplicação por escalar. Em base fixa, é representada por uma matriz: aplicar TT a um vetor vira multiplicação matricial T(v)=AvT(v) = Av.

Choose your door

Rigorous notation, full derivation, hypotheses

Definição rigorosa

Transformações lineares

"Uma transformação linear é uma função que vai de um espaço vetorial para outro e preserva as operações de espaço vetorial de adição de vetores e multiplicação por escalar." — Beezer — A First Course in Linear Algebra, §LT

"Se TT é uma transformação linear, então T(0)=0T(0) = 0." — Beezer — A First Course in Linear Algebra, Teorema LTTZZ, §LT

Representação matricial

Vbase BWbase CTcoord. x em Bcoord. [T]x em C[T] ∈ M(m×n)

Diagrama: T leva vetores de V (com base B) para W (com base C). Em coordenadas, a operação é multiplicação pela matriz [T].

Mudança de base e matrizes semelhantes

"Duas matrizes que representam a mesma transformação linear em diferentes bases são chamadas de matrizes semelhantes, e B=P1APB = P^{-1}AP para alguma matriz invertível PP." — Hefferon — Linear Algebra, cap. 3 §III.1

Composição

Exemplos resolvidos

Exercise list

44 exercises · 11 with worked solution (25%)

Application 27Understanding 4Modeling 7Challenge 2Proof 4
  1. Ex. 112.1Application

    T:R2R2T: \mathbb{R}^2 \to \mathbb{R}^2, T(x,y)=(2x,3y)T(x, y) = (2x,\, 3y). Verifique que TT é linear e encontre sua matriz.

    Show solution
    Verificamos as duas condições. Seja u=(x1,y1)u = (x_1,y_1), v=(x2,y2)v = (x_2,y_2). Soma: T(u+v)=(2(x1+x2),3(y1+y2))=T(u)+T(v)T(u+v) = (2(x_1+x_2), 3(y_1+y_2)) = T(u)+T(v). Escalar: T(au)=(2ax1,3ay1)=aT(u)T(au) = (2ax_1, 3ay_1) = aT(u). Logo TT é linear. Matriz: (2003)\begin{pmatrix}2&0\\0&3\end{pmatrix}.
    Show step-by-step (with the why)
    1. Escreva vetores genéricos u=(x1,y1)u = (x_1,y_1), v=(x2,y2)v = (x_2,y_2) e escalar aRa \in \mathbb{R}. Trabalhamos com genéricos para provar pra todo par, não só um exemplo.
    2. Condição de soma: compute T(u+v)T(u+v) e compare com T(u)+T(v)T(u)+T(v). Devem ser iguais.
    3. Condição de escalar: compute T(au)T(au) e compare com aT(u)aT(u).
    4. Para a matriz: aplique o algoritmo coluna-a-coluna. T(e1)=T(1,0)=(2,0)T(e_1) = T(1,0) = (2,0) → coluna 1. T(e2)=T(0,1)=(0,3)T(e_2) = T(0,1) = (0,3) → coluna 2.
    Macete: T(x,y)=(2x,3y)T(x,y) = (2x, 3y) é uma escala não-uniforme — estica o eixo xx por 2 e o eixo yy por 3. Transformações de escala são sempre lineares.
  2. Ex. 112.2ApplicationAnswer key

    T:R2R2T: \mathbb{R}^2 \to \mathbb{R}^2, T(x,y)=(x+1,y)T(x, y) = (x + 1,\, y). Por que TT não é transformação linear?

    Select the correct option
    Select an option first
    Show solution
    Basta testar o zero: T(0,0)=(0+1,0)=(1,0)(0,0)T(0,0) = (0+1, 0) = (1,0) \neq (0,0). Como toda transformação linear satisfaz T(0)=0T(0) = 0, esta função NÃO é linear. É uma translação, ou seja, uma transformação afim.
  3. Ex. 112.3Application

    T:RRT: \mathbb{R} \to \mathbb{R}, T(x)=x2T(x) = x^2. Dê um contraexemplo concreto para mostrar que TT não é linear.

    Show solution
    T(x)=x2T(x) = x^2 não é linear. Exemplo: T(1+1)=T(2)=4T(1+1) = T(2) = 4, mas T(1)+T(1)=1+1=24T(1)+T(1) = 1+1 = 2 \neq 4. A condição de soma falha. Adicionalmente, T(1)=11=T(1)T(-1) = 1 \neq -1 = -T(1).
  4. Ex. 112.4Application

    T:R2R2T: \mathbb{R}^2 \to \mathbb{R}^2, T(x,y)=(y,x)T(x, y) = (y,\, x). Mostre que é linear e encontre sua matriz 2×22 \times 2.

    Show solution
    T(x,y)=(y,x)T(x,y) = (y,x) é linear. Matriz: T(e1)=T(1,0)=(0,1)T(e_1) = T(1,0) = (0,1) e T(e2)=T(0,1)=(1,0)T(e_2) = T(0,1) = (1,0). Logo [T]=(0110)[T] = \begin{pmatrix}0&1\\1&0\end{pmatrix}. Esta é a matriz de permutação — ela troca as coordenadas.
  5. Ex. 112.5Application

    T:R3R2T: \mathbb{R}^3 \to \mathbb{R}^2, T(x,y,z)=(x+y,y+z)T(x, y, z) = (x + y,\, y + z). Mostre que é linear e encontre a matriz 2×32 \times 3.

    Show solution
    É linear. Matriz 2×32 \times 3: T(e1)=(1,0)T(e_1)=(1,0), T(e2)=(1,1)T(e_2)=(1,1), T(e3)=(0,1)T(e_3)=(0,1). Logo [T]=(110011)[T] = \begin{pmatrix}1&1&0\\0&1&1\end{pmatrix}.
  6. Ex. 112.6Application

    D:P3P3D: \mathcal{P}_3 \to \mathcal{P}_3, D(p)=pD(p) = p'. Encontre a matriz 4×44 \times 4 na base {1,x,x2,x3}\{1, x, x^2, x^3\}. O que é especial nessa matriz?

    Show solution
    Derivação D:P3P3D: \mathcal{P}_3 \to \mathcal{P}_3, D(p)=pD(p) = p'. Base {1,x,x2,x3}\{1, x, x^2, x^3\}: D(1)=0D(1)=0, D(x)=1D(x)=1, D(x2)=2xD(x^2)=2x, D(x3)=3x2D(x^3)=3x^2. Matriz 4×44\times4: [D]=(0100002000030000)[D] = \begin{pmatrix}0&1&0&0\\0&0&2&0\\0&0&0&3\\0&0&0&0\end{pmatrix}. É nilpotente: D4=0D^4=0.
  7. Ex. 112.7Application

    I:P2P3I: \mathcal{P}_2 \to \mathcal{P}_3, I(p)(x)=0xp(t)dtI(p)(x) = \int_0^x p(t)\,dt. Mostre que é linear e encontre a matriz 4×34 \times 3 nas bases canônicas.

    Show solution
    Integração I:P2P3I: \mathcal{P}_2 \to \mathcal{P}_3, I(p)(x)=0xp(t)dtI(p)(x) = \int_0^x p(t)\,dt. É linear pelas propriedades da integral. Imagens: I(1)=xI(1)=x, I(x)=x2/2I(x)=x^2/2, I(x2)=x3/3I(x^2)=x^3/3. Matriz 4×34\times3: colunas são (0,1,0,0)(0,1,0,0)^\top, (0,0,1/2,0)(0,0,1/2,0)^\top, (0,0,0,1/3)(0,0,0,1/3)^\top.
  8. Ex. 112.8ApplicationAnswer key

    T:M2M2T: \mathcal{M}_2 \to \mathcal{M}_2, T(A)=AT(A) = A^\top. Mostre que é linear e escreva a matriz 4×44 \times 4 na base canônica de M2\mathcal{M}_2.

    Show solution
    Transposição T(A)=ATT(A) = A^T em M2\mathcal{M}_2. Base canônica: E11,E12,E21,E22E_{11}, E_{12}, E_{21}, E_{22}. T(E11)=E11T(E_{11})=E_{11}, T(E12)=E21T(E_{12})=E_{21}, T(E21)=E12T(E_{21})=E_{12}, T(E22)=E22T(E_{22})=E_{22}. Matriz 4×44\times4: permuta as coordenadas 2 e 3. É linear, pois (A+B)T=AT+BT(A+B)^T = A^T+B^T e (aA)T=aAT(aA)^T = aA^T.
  9. Ex. 112.9Application

    Fixe BMnB \in \mathcal{M}_n. Define-se T:MnMnT: \mathcal{M}_n \to \mathcal{M}_n por T(A)=ABT(A) = AB. Mostre que TT é linear.

    Show solution
    Sim, é linear. Fixando BB, para quaisquer A1,A2A_1, A_2 e escalar aa: (A1+A2)B=A1B+A2B(A_1+A_2)B = A_1 B + A_2 B (distribuição) e (aA)B=a(AB)(aA)B = a(AB) (associatividade). Ambas as propriedades da multiplicação matricial garantem linearidade.
  10. Ex. 112.10Application

    T:M2RT: \mathcal{M}_2 \to \mathbb{R}, T(A)=detAT(A) = \det A. TT é transformação linear?

    Select the correct option
    Select an option first
    Show solution
    T(A)=detAT(A) = \det A não é linear. Para ver isso, tome A=I2A = I_2: T(2I)=det(2I)=4T(2I) = \det(2I) = 4, mas 2T(I)=21=22T(I) = 2\cdot1 = 2. Claramente 424 \neq 2. O determinante é multilinear (linear em cada coluna separadamente), mas não é linear como função de matrizes.
  11. Ex. 112.11Application

    T:P2P2T: \mathcal{P}_2 \to \mathcal{P}_2, T(p)(x)=p(2x)T(p)(x) = p(2x). Mostre que é linear e encontre a matriz diagonal na base {1,x,x2}\{1, x, x^2\}.

    Show solution
    Todo polinômio p(x)=a0+a1x+a2x2p(x) = a_0 + a_1 x + a_2 x^2 satisfaz: T(p)(x)=p(2x)=a0+2a1x+4a2x2T(p)(x) = p(2x) = a_0 + 2a_1 x + 4a_2 x^2. Na base {1,x,x2}\{1, x, x^2\}: T(1)=1T(1)=1, T(x)=2xT(x)=2x, T(x2)=4x2T(x^2)=4x^2. Matriz diagonal: diag(1,2,4)\mathrm{diag}(1,2,4).
  12. Ex. 112.12Application

    T:VWT: V \to W, T(v)=0T(v) = 0 para todo vVv \in V. Mostre que TT é linear. Qual sua matriz em qualquer base?

    Show solution
    T(v)=0T(v) = 0 para todo vv. A transformação zero é trivialmente linear: T(u+v)=0=0+0=T(u)+T(v)T(u+v) = 0 = 0+0 = T(u)+T(v). A matriz é a matriz nula 0m×n0_{m\times n}.
  13. Ex. 112.13Application

    Encontre a matriz 2×22 \times 2 da rotação de 45° no plano. Verifique que seu determinante é 1.

    Show solution
    Rotação de 45°: cos45°=sin45°=2/2\cos 45° = \sin 45° = \sqrt{2}/2. Matriz: R45=(2/22/22/22/2)R_{45} = \begin{pmatrix}\sqrt{2}/2 & -\sqrt{2}/2 \\ \sqrt{2}/2 & \sqrt{2}/2\end{pmatrix}. Verificação: det=1\det = 1 e RTR=IR^T R = I.
  14. Ex. 112.14ApplicationAnswer key

    Encontre a matriz 2×22 \times 2 da reflexão pela reta y=xy = x. Verifique que [T]2=I[T]^2 = I.

    Show solution
    Reflexão pela reta y=xy=x: T(e1)=T(1,0)=(0,1)T(e_1) = T(1,0) = (0,1) e T(e2)=T(0,1)=(1,0)T(e_2) = T(0,1) = (1,0). Logo [T]=(0110)[T] = \begin{pmatrix}0&1\\1&0\end{pmatrix}. Note [T]2=I[T]^2 = I — refletir duas vezes volta ao original.
  15. Ex. 112.15Application

    Encontre a matriz 2×22 \times 2 da projeção ortogonal no eixo yy. Verifique que P2=PP^2 = P (idempotência).

    Show solution
    Projeção ortogonal no eixo yy: T(x,y)=(0,y)T(x,y) = (0,y). Matriz: T(e1)=(0,0)T(e_1) = (0,0), T(e2)=(0,1)T(e_2) = (0,1). Logo P=(0001)P = \begin{pmatrix}0&0\\0&1\end{pmatrix}. Idempotente: P2=PP^2 = P.
  16. Ex. 112.16ApplicationAnswer key

    Encontre a matriz 2×22 \times 2 da projeção ortogonal na reta y=xy = x. Verifique idempotência.

    Show solution
    Projeção ortogonal na reta y=xy=x: a fórmula para projeção na direção de u=(1,1)/2u = (1,1)/\sqrt{2} é P=uuT=12(1111)P = uu^T = \frac{1}{2}\begin{pmatrix}1&1\\1&1\end{pmatrix}. Verificação: P2=14(2222)11=12(1111)=PP^2 = \frac{1}{4}\begin{pmatrix}2&2\\2&2\end{pmatrix} \cdot \frac{1}{1} = \frac{1}{2}\begin{pmatrix}1&1\\1&1\end{pmatrix} = P.
  17. Ex. 112.17Application

    Encontre a matriz da escala não-uniforme (x,y)(3x,2y)(x,y) \mapsto (3x,\, 2y). Qual o significado geométrico do determinante?

    Show solution
    Escala não-uniforme: T(x,y)=(3x,2y)T(x,y) = (3x, 2y). Matriz: (3002)\begin{pmatrix}3&0\\0&2\end{pmatrix}. Determinante: 66 — a transformação multiplica áreas por 6.
  18. Ex. 112.18Application

    Encontre a matriz do cisalhamento horizontal de fator 2: T(x,y)=(x+2y,y)T(x, y) = (x + 2y,\, y).

    Show solution
    Cisalhamento horizontal de fator 2: T(x,y)=(x+2y,y)T(x,y)=(x+2y, y). Matriz: (1201)\begin{pmatrix}1&2\\0&1\end{pmatrix}. Determinante: 11 — preserva áreas.
  19. Ex. 112.19Application

    Composição: rotação de 30° seguida de escala por 2. Calcule a matriz produto [E2][R30][E_2][R_{30}].

    Show solution
    Composição: rotação 30° e depois escala 2. R30=(3/21/21/23/2)R_{30} = \begin{pmatrix}\sqrt{3}/2 & -1/2 \\ 1/2 & \sqrt{3}/2\end{pmatrix}, E2=2IE_2 = 2I. Produto: E2R30=(3113)E_2 R_{30} = \begin{pmatrix}\sqrt{3} & -1 \\ 1 & \sqrt{3}\end{pmatrix}. Geometricamente: escala por 2 E rotação de 30° (nessa ordem: rotação antes, escala depois).
    Show step-by-step (with the why)
    1. Escreva a matriz de rotação de 30° usando a fórmula geral Rθ=(cosθsinθsinθcosθ)R_\theta = \begin{pmatrix}\cos\theta & -\sin\theta \\ \sin\theta & \cos\theta\end{pmatrix}. Substituindo θ=30°\theta = 30°: cos30°=3/2\cos30° = \sqrt{3}/2, sin30°=1/2\sin30° = 1/2.
    2. A escala 2 tem matriz 2I=(2002)2I = \begin{pmatrix}2&0\\0&2\end{pmatrix}. Multiplicar um vetor por esta matriz dobra cada coordenada.
    3. Composição "rotação então escala": a transformação que se aplica primeiro fica à direita no produto. Queremos escala ∘ rotação = (2I)(R30)(2I)(R_{30}).
    4. Execute o produto: multiplique linha a linha. O resultado (3113)\begin{pmatrix}\sqrt{3}&-1\\1&\sqrt{3}\end{pmatrix} tem determinante 3+1=4=223+1=4=2^2 — faz sentido, pois escalar por 2 multiplica áreas por 22=42^2 = 4.
    Macete: A transformação que se aplica primeiro fica sempre à direita no produto matricial. "Rotação depois escala" = (escala) × (rotação). Esta convenção gera um erro clássico em programação gráfica.
  20. Ex. 112.20ApplicationAnswer key

    Composição: reflexão pela reta y=xy = x e depois rotação de 90°. Calcule a matriz produto e identifique a transformação resultante.

    Show solution
    Reflexão y=xy=x seguida de rotação 90°. Reflexão: F=(0110)F = \begin{pmatrix}0&1\\1&0\end{pmatrix}. Rotação 90°: R90=(0110)R_{90} = \begin{pmatrix}0&-1\\1&0\end{pmatrix}. Produto: R90F=(0110)(0110)=(1001)R_{90}F = \begin{pmatrix}0&-1\\1&0\end{pmatrix}\begin{pmatrix}0&1\\1&0\end{pmatrix} = \begin{pmatrix}-1&0\\0&1\end{pmatrix}. Esta é a reflexão pelo eixo yy!
  21. Ex. 112.21Application

    Em R3\mathbb{R}^3, encontre a matriz 3×33 \times 3 da rotação de 90° em torno do eixo zz (o eixo zz fica fixo).

    Show solution
    Rotação de 90° no R3\mathbb{R}^3 em torno do eixo zz: eixo zz fica fixo, plano xyxy rotaciona. T(e1)=(0,1,0)T(e_1) = (0,1,0), T(e2)=(1,0,0)T(e_2) = (-1,0,0), T(e3)=(0,0,1)T(e_3) = (0,0,1). Matriz 3×33\times3: (010100001)\begin{pmatrix}0&-1&0\\1&0&0\\0&0&1\end{pmatrix}.
  22. Ex. 112.22Application

    Em R3\mathbb{R}^3, encontre a matriz 3×33 \times 3 da projeção ortogonal no plano xyxy.

    Show solution
    Projeção no plano xyxy: T(x,y,z)=(x,y,0)T(x,y,z) = (x,y,0). Coluna-a-coluna: T(e1)=(1,0,0)T(e_1)=(1,0,0), T(e2)=(0,1,0)T(e_2)=(0,1,0), T(e3)=(0,0,0)T(e_3)=(0,0,0). Matriz: (100010000)\begin{pmatrix}1&0&0\\0&1&0\\0&0&0\end{pmatrix}. Idempotente: P2=PP^2=P.
  23. Ex. 112.23Application

    T:P2P2T: \mathcal{P}_2 \to \mathcal{P}_2, T(p)=p+pT(p) = p' + p. Encontre a matriz 3×33 \times 3 na base canônica.

    Show solution
    T:P2P2T: \mathcal{P}_2 \to \mathcal{P}_2, T(p)=p+pT(p) = p' + p. Imagens da base {1,x,x2}\{1, x, x^2\}: T(1)=0+1=1T(1) = 0+1 = 1, T(x)=1+xT(x) = 1+x, T(x2)=2x+x2T(x^2) = 2x+x^2. Matriz: [T]=(110012001)[T] = \begin{pmatrix}1&1&0\\0&1&2\\0&0&1\end{pmatrix}. Triangular superior — todos os autovalores são 1.
  24. Ex. 112.24Application

    T:R3R3T: \mathbb{R}^3 \to \mathbb{R}^3, T(v)=v×(1,1,1)T(v) = v \times (1,1,1) (produto vetorial com (1,1,1)(1,1,1) fixo). Encontre a matriz 3×33 \times 3.

    Show solution
    T:R3R3T: \mathbb{R}^3 \to \mathbb{R}^3, T(v)=v×wT(v) = v \times w com w=(1,1,1)w = (1,1,1). Para v=(x,y,z)v = (x,y,z): T(v)=(yz, zx, xy)T(v) = (y-z,\ z-x,\ x-y). Matrizes coluna-a-coluna: T(e1)=(0,1,1)T(e_1)=(0,-1,1), T(e2)=(1,0,1)T(e_2)=(1,0,-1), T(e3)=(1,1,0)T(e_3)=(-1,1,0). Logo [T]=(011101110)[T] = \begin{pmatrix}0&1&-1\\-1&0&1\\1&-1&0\end{pmatrix}. Antissimétrica: [T]T=[T][T]^T = -[T].
  25. Ex. 112.25Application

    T:R2R2T: \mathbb{R}^2 \to \mathbb{R}^2, T(x,y)=(x,y)T(x,y) = (x, -y) (reflexão no eixo xx). Encontre [T][T] na base canônica e na base B={(1,1),(1,1)}\mathcal{B}' = \{(1,1),\,(1,-1)\}. Confirme que são semelhantes.

    Show solution
    Reflexão T(x,y)=(x,y)T(x,y) = (x,-y). Base canônica: [T]B=(1001)[T]_{\mathcal{B}} = \begin{pmatrix}1&0\\0&-1\end{pmatrix}. Nova base B={(1,1),(1,1)}\mathcal{B}' = \{(1,1),(1,-1)\}: T(1,1)=(1,1)=v2T(1,1)=(1,-1)=v_2, T(1,1)=(1,1)=v1T(1,-1)=(1,1)=v_1. Logo [T]B=(0110)[T]_{\mathcal{B}'} = \begin{pmatrix}0&1\\1&0\end{pmatrix}. Matrizes semelhantes, mesma transformação.
    Show step-by-step (with the why)
    1. Calcule [T]B[T]_{\mathcal{B}} na base canônica pelo método coluna-a-coluna: T(e1)=(1,0)T(e_1)=(1,0) e T(e2)=(0,1)T(e_2)=(0,-1).
    2. Para a base B={v1,v2}={(1,1),(1,1)}\mathcal{B}' = \{v_1,v_2\} = \{(1,1),(1,-1)\}, calcule T(v1)T(v_1) e T(v2)T(v_2).
    3. Expresse cada imagem em coordenadas de B\mathcal{B}': escreva T(vj)=av1+bv2T(v_j) = a\cdot v_1 + b \cdot v_2 e resolva para (a,b)(a,b). Essas coordenadas formam a coluna jj.
    4. Compare as duas matrizes. Elas são diferentes, mas representam a mesma operação geométrica.
    Observação: Em B\mathcal{B}', a reflexão troca os dois vetores de base entre si — por isso a matriz fica a permutação (01\10)\begin{pmatrix}0&1\1&0\end{pmatrix}, que é mais simples que a diagonal na base canônica!
  26. Ex. 112.26Application

    Mostre que matrizes semelhantes têm o mesmo determinante e o mesmo traço.

    Show solution
    Se ABA \sim B, existe PP invertível com B=P1APB = P^{-1}AP. Então: detB=det(P1)det(A)det(P)=detAdetPdetP=detA\det B = \det(P^{-1})\det(A)\det(P) = \frac{\det A}{\det P}\cdot\det P = \det A. E tr(B)=tr(P1AP)=tr(APP1)=tr(A)\text{tr}(B) = \text{tr}(P^{-1}AP) = \text{tr}(APP^{-1}) = \text{tr}(A) (ciclicidade do traço).
  27. Ex. 112.27ApplicationAnswer key

    Mostre que se ABA \sim B (semelhantes), então AkBkA^k \sim B^k para todo k1k \geq 1. Conclua: nilpotência é invariante por semelhança.

    Show solution
    Se ABA \sim B, existe invertível PP com B=P1APB = P^{-1}AP. Então Bk=(P1AP)k=P1AkPB^k = (P^{-1}AP)^k = P^{-1}A^kP, logo AkBkA^k \sim B^k. Se AA é nilpotente (Ak=0A^k=0), então Bk=P10P=0B^k = P^{-1}0P = 0, logo BB também é nilpotente.
  28. Ex. 112.28ModelingAnswer key

    Em computação gráfica, translação por (a,b)(a, b) em R2\mathbb{R}^2 não é transformação linear. Como coordenadas homogêneas permitem representá-la como transformação linear em R3\mathbb{R}^3? Escreva a matriz 3×33 \times 3.

    Show solution
    Coordenadas homogêneas: representar ponto (x,y)R2(x,y) \in \mathbb{R}^2 como (x,y,1)R3(x,y,1) \in \mathbb{R}^3. Translação por (a,b)(a,b): (10a01b001)(xy1)=(x+ay+b1)\begin{pmatrix}1&0&a\\0&1&b\\0&0&1\end{pmatrix}\begin{pmatrix}x\\y\\1\end{pmatrix} = \begin{pmatrix}x+a\\y+b\\1\end{pmatrix}. Translação vira multiplicação matricial — linear em R3\mathbb{R}^3 mesmo não sendo linear em R2\mathbb{R}^2.
  29. Ex. 112.29Modeling

    Portfólio com nn ativos, pesos wRnw \in \mathbb{R}^n, retornos esperados μRn\mu \in \mathbb{R}^n. O retorno esperado rp=wμr_p = w^\top \mu é transformação linear em ww? E a variância σp2=wΣw\sigma_p^2 = w^\top \Sigma w?

    Show solution
    Retorno de portfólio: rp=wμr_p = w^\top \mu. É transformação linear T:RnRT: \mathbb{R}^n \to \mathbb{R} com T(w)=μwT(w) = \mu^\top w. Linear em ww: se você dobrar todos os pesos, o retorno dobra. A variância do portfólio σp2=wΣw\sigma_p^2 = w^\top \Sigma w é forma bilinear (não linear).
  30. Ex. 112.30Modeling

    Escreva a matriz Toeplitz 3×53 \times 5 que realiza convolução linear 1D com kernel k=(1,2,1)k = (1, 2, 1) sobre um sinal de comprimento 5 (saída válida, comprimento 3).

    Show solution
    Convolução 1D com kernel de tamanho 3 sobre sinal de tamanho 5: saída de tamanho 3. Kernel k=(k0,k1,k2)k = (k_0, k_1, k_2), entrada x=(x0,,x4)x = (x_0,\ldots,x_4). Saída yi=j=02kjxi+jy_i = \sum_{j=0}^{2} k_j x_{i+j}. Matriz Toeplitz 3×53\times5: cada linha é o kernel deslocado uma posição.
  31. Ex. 112.31Modeling

    Em aprendizado de máquina, uma camada densa é y=Wx+by = Wx + b. Qual parte é transformação linear? Por que adicionar bb não torna a camada linear? De onde vem a não-linearidade de uma rede neural?

    Show solution
    Camada densa: y=Wx+by = Wx + b. A parte WxWx é transformação linear W:RninRnoutW: \mathbb{R}^{n_{\rm in}} \to \mathbb{R}^{n_{\rm out}}. O bb (bias) torna a camada afim, não linear. A não-linearidade da rede vem das funções de ativação (ReLU, sigmoid) aplicadas após cada camada afim.
  32. Ex. 112.32Modeling

    Sistema LTI: x˙=Ax\dot{x} = Ax, solução x(t)=eAtx0x(t) = e^{At}x_0. Mostre que a aplicação x0x(t)x_0 \mapsto x(t) é transformação linear. O que é a matriz eAte^{At}?

    Show solution
    Sistema LTI: x˙=Ax\dot{x} = Ax, solução x(t)=eAtx0x(t) = e^{At}x_0. A aplicação x0x(t)x_0 \mapsto x(t) é transformação linear: eAt(x0+y0)=eAtx0+eAty0e^{At}(x_0 + y_0) = e^{At}x_0 + e^{At}y_0. A matriz de transição é Φ(t)=eAt\Phi(t) = e^{At}. Em engenharia de controle, autovalores de AA determinam estabilidade do sistema.
  33. Ex. 112.33Modeling

    O operador de derivação D:PnPnD: \mathcal{P}_n \to \mathcal{P}_n tem matriz nilpotente. Explique por que Dn+1=0D^{n+1} = 0, e o que isso significa em termos de polinômios.

    Show solution
    Derivação D:PnPnD: \mathcal{P}_n \to \mathcal{P}_n. Matriz nilpotente: Dn+1=0D^{n+1} = 0 porque derivar n+1n+1 vezes qualquer polinômio de grau nn dá zero. Geometricamente: grau decresce a cada derivação, e após n+1n+1 aplicações, todo polinômio vai a zero.
  34. Ex. 112.34Modeling

    Por que dimL(V,W)=(dimV)(dimW)\dim \mathcal{L}(V, W) = (\dim V)(\dim W)? O que isso diz sobre o espaço de todas as matrizes m×nm \times n?

    Show solution
    Dimensão de L(V,W)=Hom(V,W)\mathcal{L}(V,W) = \mathrm{Hom}(V,W): todo elemento é determinado pelos valores em uma base de VV (n vetores), e cada imagem vive em WW (dimensão m). São nmn \cdot m escalares livres. Logo dimL(V,W)=mn\dim \mathcal{L}(V,W) = mn. Para V=W=RnV=W=\mathbb{R}^n: n2n^2 entradas = dimensão de Mn(R)\mathcal{M}_{n}(\mathbb{R}).
  35. Ex. 112.35UnderstandingAnswer key

    T(0)=0T(0) = 0 é condição necessária para linearidade. Dê um exemplo de TT com T(0)=0T(0) = 0 que não é linear. Por que T(0)=0T(0) = 0 não é suficiente?

    Show solution
    T(0)=0T(0)=0 é necessário mas não suficiente. Exemplo: T:RRT: \mathbb{R} \to \mathbb{R}, T(x)=x2T(x) = x^2. Temos T(0)=0T(0)=0, mas T(1+1)=41+1=T(1)+T(1)T(1+1)=4 \neq 1+1=T(1)+T(1). Logo a condição T(0)=0T(0)=0 é necessária (toda linear satisfaz) mas não suficiente (existem não-lineares que também satisfazem).
  36. Ex. 112.36Understanding

    Se A=[T]BA = [T]_{\mathcal{B}} representa TT em B\mathcal{B}, explique o significado geométrico da fórmula [T]B=P1AP[T]_{\mathcal{B}'} = P^{-1}AP. O que faz cada fator?

    Show solution
    Se AA representa TT em B\mathcal{B} e queremos [T]B[T]_{\mathcal{B}'}: seja PP a matriz cujas colunas são os vetores de B\mathcal{B}' em coordenadas de B\mathcal{B}. Então [T]B=P1AP[T]_{\mathcal{B}'} = P^{-1}AP. A fórmula lê-se: "converter de B\mathcal{B}' para B\mathcal{B} via PP, aplicar AA, converter de volta via P1P^{-1}".
  37. Ex. 112.37Understanding

    Explique, sem calcular, por que o produto matricial [T][S][T][S] é exatamente a composição TST \circ S. Qual é a relação entre a definição de produto matricial e a definição de composição?

    Show solution
    O produto matricial [T][S][T][S] realiza composição TST \circ S. Prova: para qualquer vv com coordenadas xx em B\mathcal{B}, (TS)(v)(T \circ S)(v) tem coordenadas [T]([S]x)[T]([S]x). Mas isso é exatamente ([T][S])x([T][S])x por definição de produto matricial.
  38. Ex. 112.38UnderstandingAnswer key

    Mostre que L(V,W)\mathcal{L}(V, W) (conjunto de todas as transformações lineares de VV em WW) é ele mesmo um espaço vetorial, com as operações (T1+T2)(v)=T1(v)+T2(v)(T_1 + T_2)(v) = T_1(v) + T_2(v) e (aT)(v)=aT(v)(aT)(v) = a\,T(v).

    Show solution
    Sejam T1,T2L(V,W)T_1, T_2 \in \mathcal{L}(V,W). Define-se (T1+T2)(v)=T1(v)+T2(v)(T_1+T_2)(v) = T_1(v)+T_2(v) e (aT)(v)=aT(v)(aT)(v) = a\cdot T(v). Verificação: (T1+T2)(u+v)=T1(u+v)+T2(u+v)=T1(u)+T1(v)+T2(u)+T2(v)=(T1+T2)(u)+(T1+T2)(v)(T_1+T_2)(u+v) = T_1(u+v)+T_2(u+v) = T_1(u)+T_1(v)+T_2(u)+T_2(v) = (T_1+T_2)(u)+(T_1+T_2)(v). Portanto L(V,W)\mathcal{L}(V,W) é fechado sob soma e escalar — é espaço vetorial.
  39. Ex. 112.39ChallengeAnswer key

    Encontre T:R2R2T: \mathbb{R}^2 \to \mathbb{R}^2 com T2=IT^2 = -I (o negativo da identidade). Sugere-se rotação de 90°. Qual a conexão com os números complexos?

    Show solution
    Queremos T:R2R2T: \mathbb{R}^2 \to \mathbb{R}^2 com T2=IT^2 = -I. A rotação de 90° funciona: R90=(0110)R_{90} = \begin{pmatrix}0&-1\\1&0\end{pmatrix}. Então R902=(1001)=IR_{90}^2 = \begin{pmatrix}-1&0\\0&-1\end{pmatrix} = -I. Esta propriedade é análoga a i2=1i^2 = -1 nos complexos — a rotação de 90° age como a unidade imaginária ii!
  40. Ex. 112.40Challenge

    Demonstre: toda transformação linear T:RRT: \mathbb{R} \to \mathbb{R} tem a forma T(x)=axT(x) = ax para algum aRa \in \mathbb{R}.

    Show solution
    Toda T:RRT: \mathbb{R} \to \mathbb{R} linear tem forma T(x)=T(1)xT(x) = T(1)\cdot x. Demonstração: dado qualquer xRx \in \mathbb{R}, T(x)=T(x1)=xT(1)T(x) = T(x \cdot 1) = x\,T(1) pela linearidade (preservação de escalar). Portanto a=T(1)Ra = T(1) \in \mathbb{R} determina completamente TT, e T(x)=axT(x) = ax.
  41. Ex. 112.41Proof

    Demonstração. Prove que a composição de transformações lineares é linear. Seja S:UVS: U \to V e T:VWT: V \to W ambas lineares. Mostre que TS:UWT \circ S: U \to W é linear.

    Show solution
    Sejam S:UVS: U \to V e T:VWT: V \to W lineares. Para quaisquer u1,u2Uu_1, u_2 \in U e aKa \in K: Soma: (TS)(u1+u2)=T(S(u1+u2))=T(S(u1)+S(u2))=T(S(u1))+T(S(u2))=(TS)(u1)+(TS)(u2)(T\circ S)(u_1+u_2) = T(S(u_1+u_2)) = T(S(u_1)+S(u_2)) = T(S(u_1))+T(S(u_2)) = (T\circ S)(u_1)+(T\circ S)(u_2). Escalar: análogo. Logo TST \circ S é linear. \square
  42. Ex. 112.42ProofAnswer key

    Demonstração. Prove por indução que toda transformação linear preserva combinações lineares arbitrárias: T ⁣(civi)=ciT(vi)T\!\left(\sum c_i v_i\right) = \sum c_i T(v_i).

    Show solution
    Seja T:VWT: V \to W linear e c1,,cnKc_1, \ldots, c_n \in K, v1,,vnVv_1, \ldots, v_n \in V. Então: T(i=1ncivi)=i=1nciT(vi)T\left(\sum_{i=1}^n c_i v_i\right) = \sum_{i=1}^n c_i T(v_i). Prova por indução: base n=1n=1 é a preservação de escalar; passo: T(c1v1++cnvn)=T(c1v1++cn1vn1)+T(cnvn)=i=1n1ciT(vi)+cnT(vn)T(c_1v_1 + \cdots + c_n v_n) = T(c_1v_1 + \cdots + c_{n-1}v_{n-1}) + T(c_n v_n) = \sum_{i=1}^{n-1}c_i T(v_i) + c_n T(v_n). \square
  43. Ex. 112.43Proof

    Demonstração. Prove: TT linear é injetiva     kerT={0}\iff \ker T = \{0\}.

    Show solution
    (\Rightarrow) Se TT é injetiva e T(v)=0=T(0)T(v)=0=T(0), então v=0v=0, logo kerT={0}\ker T = \{0\}. (\Leftarrow) Se kerT={0}\ker T = \{0\} e T(u)=T(v)T(u)=T(v), então T(uv)=T(u)T(v)=0T(u-v) = T(u)-T(v) = 0, logo uvkerT={0}u-v \in \ker T = \{0\}, portanto u=vu=v. \square
  44. Ex. 112.44Proof

    Demonstração. Prove o teorema de extensão linear: dados espaços vetoriais VV (dim nn) e WW, e vetores w1,,wnWw_1, \ldots, w_n \in W arbitrários, existe única transformação linear T:VWT: V \to W com T(vi)=wiT(v_i) = w_i para i=1,,ni = 1, \ldots, n.

    Show solution
    Seja B={v1,,vn}\mathcal{B} = \{v_1,\ldots,v_n\} base de VV. Todo vVv \in V se escreve unicamente como v=aiviv = \sum a_i v_i. Definindo T(v)=aiwiT(v) = \sum a_i w_i com wiWw_i \in W arbitrários: (i) TT é bem-definida (unicidade da representação); (ii) é linear (por construção); (iii) é única com esses valores na base (qualquer outra com mesmos valores coincide em toda combinação linear). Isso mostra que a base determina completamente a transformação linear. \square

Fontes

  • Beezer — A First Course in Linear Algebra — Rob Beezer · 2022 · EN · GNU FDL. §LT (Linear Transformations) e §ILT (Injective Linear Transformations). Fonte primária desta lição.
  • Hefferon — Linear Algebra — Jim Hefferon · 4.ª ed. · EN · CC-BY-SA. Cap. 3 (Maps Between Spaces): enfoque geométrico e exemplos de transformações do plano.
  • Axler — Linear Algebra Done Right — Sheldon Axler · 4.ª ed. · EN · CC-BY-NC. §3A–§3B: linear maps como objetos de primeira classe; sem determinantes como fundamento.

Updated on 2026-05-06 · Author(s): Clube da Matemática

Found an error? Open an issue on GitHub or submit a PR — open source forever.