I have an ellipse $\mathcal{E} = \{x^TAx = 1\}$, and I have a connected subset of an ellipse $U\subset \mathcal{E}$
For a given $\theta$ let $x_U^*(\theta) = \arg \sup\{\langle x,\theta\rangle, x\in U\} $ and $x_\mathcal{E}^*(\theta) = \arg \sup\{\langle x,\theta\rangle, x\in \mathcal{E}\}$. Now in general $x_U^*(\theta) \neq x_\mathcal{E}^*(\theta)$.
But if $\theta$ and $U$ is such that $x_U^*(\theta)$ is an interior point of $U$ then it seems that both points are the same. This can be seen easily in the case of a sphere, though I cannot prove it. Can anyone help?
EDIT it's like suprema of a linear function over a convex surface occurs at boundary but if it is an interior point then it must be a global optima