If \( f(\mathbf{x}) \) has a local extremum at \( \mathbf{x}^* \) and is differentiable, then \( \nabla f(\mathbf{x}^*) = \mathbf{0} \).
For a function \( f(\mathbf{x}) \) about point \( \mathbf{x}^* \):
\[ f(\mathbf{x}) \approx f(\mathbf{x}^*) + \nabla f(\mathbf{x}^*)^T (\mathbf{x} - \mathbf{x}^*) + \frac{1}{2} (\mathbf{x} - \mathbf{x}^*)^T H(\mathbf{x}^*) (\mathbf{x} - \mathbf{x}^*) \]
Let \( \nabla f(\mathbf{x}^*) = \mathbf{0} \). Then:
Example: For \( f(x, y) = x^2 - y^2 \):
Saddle point at \( (0,0) \): First partial derivatives are zero, Hessian is indefinite.
Example: Consider the matrix:
\[ A = \begin{pmatrix} 4 & 1 & 2 \\ 1 & 3 & 1 \\ 2 & 1 & 2 \end{pmatrix} \]
Find and classify critical points of \( f(x, y) = x^2 + 2y^2 - 2xy - 2y \).