设矩阵A,向量 x ⃗ \vec{x} x 则有
∂ A x ⃗ ∂ x ⃗ = A T ∂ A x ⃗ ∂ x ⃗ T = A ∂ ( x ⃗ T A ) ∂ x ⃗ = A ∂ ( x ⃗ T A x ⃗ ) ∂ x ⃗ = ( A T + A ) x ⃗ \frac{\partial A\vec{x}}{\partial \vec{x}} = A ^ {T} \\ \frac{\partial A\vec{x}}{\partial \vec{x} ^ {T}} = A \\ \frac{\partial (\vec{x} ^ {T}A)}{\partial \vec{x}} = A \\ \frac{\partial (\vec{x} ^ {T}A\vec{x})}{\partial \vec{x}} = (A ^ {T} + A)\vec{x} \\ ∂x ∂Ax =AT∂x T∂Ax =A∂x ∂(x TA)=A∂x ∂(x TAx )=(AT+A)x 特别的,如果 A = A T A = A ^ {T} A=AT(A为对称矩阵),则: ∂ ( x ⃗ T A x ⃗ ) ∂ x ⃗ = 2 A x ⃗ \frac{\partial (\vec{x} ^ {T}A\vec{x})}{\partial \vec{x}} = 2A\vec{x} ∂x ∂(x TAx )=2Ax