convexity about a function of many variable

68 Views Asked by At

assume $\mathcal X$ is a convex set and $\mathit f(x,a)$ is a function that $$f:\mathcal X \times \Bbb R \to \Bbb R$$

$f(x,a)$ is convex with respect to $x$ for every $a$ and $f(x,a)$ is convex with respect to $a$ for every $x$

Is that function $f(x,a)$ is convex with respect to $(x,a)$??

if not, under what condition that can provide convexity to guarantee function $f(x,a)$ is convex with respect to $(x,a)$?

I know it goes to prove $f((1-\lambda )(x,a)+\lambda(x',a')) \le (1-\lambda)f(x,a)+\lambda f(x',a') $

but I didn't know how to go on when I take this step

$f((1-\lambda )(x,a)+\lambda(x',a'))$ = $f((1-\lambda)x+\lambda x' ,(1-\lambda)a+\lambda a')$

1

There are 1 best solutions below

0
On BEST ANSWER

No, in order to guarantee concavity in more than one dimension, you need the Hessian Matrix to be positive semi definite over the entire domain. For an example of a function that is convex in every independent variable, but not in the combined domain look at

$$ xy - x - y $$

on $\mathbb{R}x\mathbb{R}$. This function is linear with respect to each independent variable, however it is not convex as a whole. There is even a saddle point at $(1,1)$