My take. (Subject to correction by anyone who knows the subject better - like having DONE it. B-) )
First approximation:
- Hole is the size of the magnet.
- Side of the coil is half the width of the hole.
- Space between magnets is half the width of the magnet.
- Coils butt up against each other.
In this ideal, where the stator is half-paved with copper and the field goes straight across, this would cause each phase to have a trapezoidal waveform:
- 1/3 cycle ramp between + and -,
- 1/6 cycle at -,
- 1/3 cycle ramp between - and +,
- 1/6 cycle at +
In Y the phases would be ramping so that as one goes down another in series with it would be going up (as the magnets transition gradually from one coil to the next, keeping an equal number of turns under the poles), resulting in a nearly ripple-free voltage. (Of course IR drop fouls that a tad.)
Note that is one of the non-sinusoidal waveforms where a delta connection sums to zero and (if we could keep to this ideal waveform) a delta would also have negligible circulating current and a ripple-less rectified output.
In the real world the field spreads out from the poles a tad. So make the magnet to a tad narrower than the ideal to compensate. Because the magnet size is fixed that's done by moving 'em out a tad so the spacing is a little more than half the pole width. Scale the coils up, keeping the leg width = half hole width relation.
Also in the real world the coils are radial, the magnets are rectangles rather than trapezoids / pie segments, and the assembly is not so far from the center that this doesn't cause issues. So let the hole be wider at the outside than the inside. Magnet over both sides of the coil cancels out voltage, ending up with turns contributing resistance but no voltage. Thus err on the side of widening the hole so it's about the size of the "slightly spread pole" near the inside end and has extra gap near the outside end.
"Wasting magnets" means losing potential power generation - and can be compensate by making the magnets stronger, larger, or tightenin the gap. "Wasting turns" means consuming mechanical power making extra resistive heating without making extra output, which means compensation would require making the swept area larger. Better to "waste magnets" - and buy strong ones. The radial flux with the single layer of coils ALREADY "wastes magnets" big time by having the magnets over a hole rather than a coil for about half their travel (and doubling up on the number of super-strong magnets). A little extra wedge of "wasted magnet" from widening the outer part of the coil hole is almost lost in the noise.
