White plays 1.e4 c5 2.a3 with the intention of playing 3.b4, diverting Black's c-pawn from the centre and allowing White to build a strong central position with his e and d pawns. Depending on Black's 2nd move this may or may not involve sacrificing a pawn. Of course White can play the Wing Gambit (1.e4 c5 2.b4) or the Wing Gambit Delayed (1.e4 c5 2.Nf3 d6 3.b4) without the fanfare of 2.a3.
I have tried 2.a3 a few times in online blitz games with pretty good results. I have not used it with the intention of playing b4 though because I’ve also tried the Wing Gambit and WG Delayed and have not gotten very good positions.
The following online game was quite interesting. We reached what appears to be a standard Sicilian position and after a flurry of exchanges beginning at move 9 we reached a position where I had 2R’s, N and 5P’s against my opponent’s R, 2B’s and 6P’s. These positions are the kind that, in my opinion, engines don’t evaluate very well, so going over it with Houdini 2 doesn’t answer a lot of questions. In fact, while analyzing it with a couple of different engines, there were different suggestions for both of us at almost every move, but most of them seem to lead to nothing significantly better than what we actually played. Still, it seems to me that all the winning chances were Black’s but when he missed his best chance at move 37 the game was a draw. Black kept playing though hoping for a mistake which is exactly what happened. Unfortunately for him, it was his mistake and it cost him the game.