Let $\{ f_1,f_2,\ldots,f_m \}$ be a set of convex functions, where $f_i : C \subset \mathbb{R}^n \to \mathbb{R}$ and with $C$ a convex set. Then,

$$F(x) := \max \{ f_1(x),f_2(x), \ldots, f_m(x) \}$$ is a convex function. What happens if each $f_i$ is strictly convex? Is $F$ also strictly convex?

I believe that it is not true. If it is the case, a counterexample would be great.

  • $\begingroup$ It is true when your family of convex functions is finite, but when taking the (pointwise) supremum of an infinite family, it may not be. For example, if you consider $f_n(x) = \frac{x^2 - 1}{n}$ over the interval $C = [-1, 1]$, the pointwise supremum is constant. $\endgroup$ – Theo Bendit May 24 at 2:16

Consider $C = \mathbf{R}^n$ for simplicity; let $k = \arg\max_{i \in [m]} f_i(\lambda x + (1 - \lambda) y)$. Then

$$ \begin{align} \max_{i=1, \dots, m} f_i(\lambda x + (1 -\lambda)y) &= f_k(\lambda x + (1 - \lambda)y) \overset{(*)}{<} \lambda f_k(x) + (1- \lambda) f_k(y) \\ &\leq \lambda \max_{i} f_i(x) + (1-\lambda) \max_{i} f_i(y), \end{align} $$ where $(*)$ used the fact that all the $f_i$'s are strictly convex, so $F$ is in fact strictly convex.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.