More generally, how do you reconcile questions that you may have about the paper's technical material?
Usually any such ex-parte communication needs to go through the relevant PC. For example I was once on a PC where a subreviewer identified a bug in a paper they were reviewing for me. They relayed it to me, I discussed it with the other PC members assigned to the paper, and then we presented our conclusions to the PC chair who initiated contact with the authors.
This is for conferences without a rebuttal process (typically true for theory confs). If the conference has a rebuttal process then there's a formal mechanism for discussing this with the authors.
Finally, if you're not sure if there's a bug, then merely raise your doubt with the PC.
Update: a secondary issue that comes to mind after seeing Jeff's comment is a matter of fairness. Suppose ex-parte communications are not officially sanctioned, but are not frowned on. Then an author who happens to know people on the PC gets a slight advantage, because PC members might be more likely to contact the author. However, an official policy on how to handle such communication that does not depend on "knowing the author" implicitly or explicitly seems fairer.