Approximability of convex programming (convex optimization)

Question

Convex optimization is defined here: https://cstheory.stackexchange.com/questions/22314/definition-of-convex-optimization-problem-by-stephen-boyd-and-lieven-vandenbergh

The problem is NP-hard: https://mathoverflow.net/questions/92939/is-that-true-all-the-convex-optimization-problems-can-be-solved-in-polynomial-ti

But is anything known about the approximation complexity of the problem? Does it have a PTAS (poly time approximation scheme)? Or is there a proof that a PTAS is impossible unless P=NP? Are there any known results on upper/lower bounds on its approximability?

Any information will be much appreciated.

Answer 1

You haven't defined convex optimization in general, and indeed it is not clear how one would encode a general convex optimization problem. However, to answer your question it suffices to consider the special case of copositive programming. It is known that Independent Set can be expressed as a copositive program, and this problem is NP-hard to approximate up to a factor of $n^{1-\epsilon}$ for every $\epsilon > 0$. In particular, there is no PTAS.

• Sameer Gupta: Thanks.. Let me check to make sure that the reduction from Ind.Set to Copositive Prog is an approximation preserving reduction.