1: \begin{proof} To show convexity, we need to argue that the function $\fp\w)$ is a convex function of $\w=\x-\x_0$.\\

2: Observe that $g(\w)=f(\x_0+\w)-f(\x_0)$ is a convex function of $\w$ and behaves same as the directional derivative $\fp\w)$ for sufficiently small $\w$. More rigorously, from \eqref{dirder3}, for any $\w_1,\w_2\in\R^n$ and $ \delta>0$ there exists $\eps>0$ such that, we have:

3: \beq

4: g(\eps\w_1)\leq \fp\eps\w_1)+\delta\eps, ~g(\eps\w_2)\leq \fp\eps\w_2)+\delta\eps\nn

5: \eeq

6: Hence, for any $0\leq c\leq 1$:

7: \bea

8: \fp\eps(c\w_1+(1-c)\w_2))&\leq g(\eps(c\w_1+(1-c)\w_2))\nn\\

9: &\leq cg(\eps\w_1)+(1-c)g(\eps\w_2)\nn\\

10: &\leq c\fp\eps\w_1)+(1-c)\fp\eps\w_2)+\eps \delta\nn

11: \end{align}

12: Making use of the fact that $\fp\eps\s)=\eps \fp\s)$ for any direction $\s$, we obtain:

13: \beq

14: \fp c\w_1+(1-c)\w_2)\leq c\fp\w_1)+(1-c)\fp\w_2)+ \delta\nn

15: \eeq

16: Letting $\delta\rightarrow 0$, we may conclude with the convexity of $\fp\cdot)$ and problem (\ref{first or}).

17: \end{proof}

18: