Building a grounded power supply is the way to go. There is no way to make a ground with passive components efficiently for that much current.
----- But ----- ........
I was intrigued by the idea and thought about it a little, and came up with the circuit here as one possible basic starting point. Virtual ground is the word - the circuit is literally the buffered virtual ground of the op-amp.
Now, this is totally untested, but it seems good on paper and should provide stable regulation within mV's.
You would want R1=R2, large C, and perhaps TIP41c and TIP42c for the Q's. But for 2 amp swings the op-amp might be a little heavily loaded.
Let me know if you try something out! Good luck.
That kind of thing does work, but the ground transistors need to be every bit as beefy as the main output transistors. So you might as well build a bridged amp as this uses just as many big transistors and will avoid the need for a high current virtual ground and also load the power supply in a nicer way and if done right also cancel some distortion.
Yeah, in this case, a virtual ground is equivalent to having a bridged amplifier (H bridge) with one channel un-driven.
As with that virtual ground circuit, it looks good, except I think a pair of driver transistors are in order, -and- more powerful output transistors. Most opamps won't hold still in that circuit unless the current gain in the output stage is high. This applies equally to any audio amplifier output stage.