I have a function $f:\mathbb{R}^p\rightarrow\mathbb{R}$. For the purposes of illustration let's say that $p=2$. I want to prove that $f$ is convex. Suppose I can show that both

$f(\alpha x_1 + (1-\alpha)x_2,y)\leq\alpha f(x_1,y)+(1-\alpha)f(x_2,y) \quad \forall y \in \mathbb{R}$


$f(x,\alpha y_1 + (1-\alpha)y_2)\leq\alpha f(x,y_1)+(1-\alpha)f(x,y_2) \quad \forall x \in \mathbb{R}$

is true, i.e. I can show that the resulting univariate functions obtained by holding all but one of the arguments fixed is convex.

Is this sufficient to prove that $f$ is convex in $x$ and $y$? Namely

$f(\alpha x_1 + (1-\alpha)x_2,\alpha y_1 + (1-\alpha)y_2)\leq \alpha f(x_1,y_1) + (1-\alpha)f(x_2,y_2)$.

I am looking for a proof or counter example.

Merely applying each conditional inequality sequentially gives $f(\alpha x_1 + (1-\alpha)x_2,\alpha y_1 + (1-\alpha)y_2)\leq \alpha^2 f(x_1,y_1) + \alpha (1-\alpha) f(x_1,y_2) + \alpha (1-\alpha) f(x_2,y_1) + (1-\alpha)^2 f(x_2,y_2)$

If $f$ is a convex function, this implies the first two univariate inequalities of this post. This post is asking if the reverse direction holds.


NO, Take $f(x,y) =xy$ then $f$ is linear in each arguments, but not convex. Look at its hessian.


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.