First-Order Necessary Condition (FONC)
Let $\Omega$ be a subset of $\mathbb{R}^n$ and $f\in\cal{C}^1$ a real-valued function on $\Omega$. If $\mathbf{x}^*$ is a local minimizer of $f$ over $\Omega$, then for any feasible direction $\mathbf{d}$ at $\mathbf{x}^*$, we have $$\mathbf{d}^{T}\nabla f\left(\mathbf{x}^*\right) \ge 0,$$ where $f\in\cal{C}^i$ means that a function $f$ is $i$ times continuously differentiable, a point $\mathbf{x}^*\in\Omega$ is a local minimizer of $f$ over $\Omega$ if there exists $\varepsilon\gt0$ such that $f(\mathbf{x})\ge f(\mathbf{x}^*)$ for all $\mathbf{x}\in\Omega\setminus\{\mathbf{x}^*\}$ and $\Vert\mathbf{x} - \mathbf{x}^* \Vert \lt \varepsilon$, a vector $\mathbf{d}\in\mathbb{R}^n$, $\mathbf{d}\ne\mathbf{0}$ is a feasible direction at $\mathbf{x}\in\Omega$ if there exists $\alpha_0\gt0$ such that $\mathbf{x}+\alpha\mathbf{d}\in\Omega$ for all $\alpha\in [0,\alpha_0]$, and $\mathbf{x}^T$ represents a transpose of $\mathbf{x}$ that every elements' row and column are changed from $x_{ij}$ to $x_{ji}$.
Let us consider the set-constrained problem $$\text{minimize} \quad f(\mathbf{x})$$ $$\text{subject to}\quad \mathbf{x}\in\Omega,$$ where $\Omega = \left\{ \left[x_1, x_2\right]^T : x_1^2+x_2^2=1\right\}$.
In this problem, because there are no feasible directions at any $\mathbf{x}^*$, all points in $\Omega$ satisfy the FONC for this set-constrained problem.
Why???
I do not understand this answer.