OptML_course
OptML_course copied to clipboard
Lemma 4.3 is missing some condition
In Lemma 4.3, the easy direction (Exercise 29) can easily be proved. However, the other direction needs additional conditions, otherwise, it's not correct. For example, consider f(x) = -√x for x ≥ 0. The function and the domain are convex, but there is no subgradient at 0
True, thanks for the nice catch. We should change the lemma conditions such that the sub-gradient is non-empty in the interior points of the domain as done in (Theorem 3.1.13 in Nesterov, 2004). We will update this.