Thanks, Teemuk and Darwin. Well, it MAY work, and could be built as a curio, but if simplicity is the name of the game, the bias can be simplified (and bettered) by pulling the zener (which does NOT track the output stage temperature) and recalculating the bias resistors , replacing the unnecessary FET "current source" by a 2k7 resistor, and some other minor changes. The "power stage" has only a gain of 6 (100K/15K),and isn´t properly stabilized. An LM3886 is as simple and much better.