I am working on a problem to prove $$ f(x,y) = |xy| + a(x^2 + y^2) $$ is convex for $a \ge 1/2$.

My approach is to show that the Hessian is PSD for the cases where $x \not= 0, y \not= 0$.

However, this approach breaks down for cases when $x$ or $y$ are zero since the function is not differentiable at these points.

Any suggestions on other approaches that get around this?


$\begin{array}\\ f(x,y) &= |xy| + a(x^2 + y^2)\\ &= |xy| + (a-\tfrac{1}{2})(x^2+y^2)+\tfrac{1}{2}|x|^2+\tfrac{1}{2}|y|^2\\ &= (a-\tfrac{1}{2})(x^2+y^2) + \tfrac{1}{2}(|x|+|y|)^2\\ \end{array} $

The first term, $(a-\tfrac{1}{2})(x^2+y^2)$, is clearly convex.
The second term is the composition of the (outer) convex increasing nonnegative function $\xi\mapsto \tfrac{1}{2}\max\{0,\xi^2\}$ and the inner function $(x,y)\mapsto |x|+|y|$.

  • $\begingroup$ Hi Max, forgot to ask. What was the intuition that guided you to deriving this expression for $f(x,y)$? $\endgroup$ – ted Mar 16 '18 at 18:20
  • $\begingroup$ I knew that $xy$ is not convex, so completing the squares seems natural. Plus you should have to use that $a\geq \tfrac{1}{2}$. $\endgroup$ – max_zorn Mar 16 '18 at 22:22

Possible hint:

$\begin{array}\\ f(x,y) &= |xy| + a(x^2 + y^2)\\ &= |xy| + a(x^2+2|xy| + y^2)-2a|xy|\\ &= (1-2a)|xy| + a(|x|+|y|)^2\\ \end{array} $

  • $\begingroup$ When $a=1$, $(1-2a)|xy|$ is not convex. $\endgroup$ – max_zorn Mar 4 '18 at 6:34

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.