这是一种迭代算法,为了求解多变量方程f⃗(x⃗)=0⃗\vec{f}(\vec{x}) = \vec{0}f(x)=0, 其中x⃗,f⃗(x⃗)∈Rn\vec{x}, \vec{f}(\vec{x}) \in \R^nx,f(x)∈Rn,则该方程的近似最优解可由下式给出
x⃗k+1=x⃗k−J−1(f⃗(x⃗))f⃗(x⃗)\vec{x}_{k+1} = \vec{x}_k - \mathbf{J}^{-1}(\vec{f}(\vec{x})) \vec{f}(\vec{x}) xk+1=xk−J−1(f(x))f(x)
其中
J(f⃗(x⃗))=[∂fi(x⃗)∂xj]i,ji,j=1,2,...,n\mathbf{J}^{}(\vec{f}(\vec{x})) = \begin{bmatrix} \frac{\partial f_i(\vec{x})}{\partial x_j} \end{bmatrix}_{i,j} \quad\quad i,j = 1,2,...,n J(f(x))=[∂xj∂fi(x)]i,ji,j=1,2,...,n
当
∣∣x⃗k+1−x⃗k∣∣≤ϵ1或者∣∣f⃗(x⃗k)∣∣≤ϵ2||\vec{x}_{k+1} - \vec{x}_k|| \leq \epsilon_1 或者||\vec{f}(\vec{x}_k)|| \leq \epsilon_2 ∣∣xk+1−xk∣∣≤ϵ1或者∣∣f(xk)∣∣≤ϵ2
时,算法停止