necessary and sufficient conditions for a function to be DC

Hi, Does anyone know the necessary and sufficient conditions for a function to be a DC-function?

Definition: A function is a DC-function if and only if it can be written as a differnece of 2 convex functions.

-

For real functions whose domain is a real interval, it is necessary and sufficient that the second derivative is a function of bounded variation on every compact interval in the domain. Or, in terms of distributions, the second derivative must be a measure (a difference of two non-negative measures).

-
This is for real-to-real function only. –  Anton Petrunin Nov 14 '12 at 23:11
$x \: \mapsto \: x^4-0 \;\;$ seems like a counterexample to the necessity. $\;\;\;\;$ –  Ricky Demer Nov 15 '12 at 7:47
Yes, this was for functions R to R. Concerning the counterexample, I edited the statement:-) –  Alexandre Eremenko Nov 15 '12 at 12:49