Bill (waross) described it as a rubber band. We can flesh out why it acts that way...
Imagine you have a permanent magnet in an external uniform magnetic field. The torque on the magnet is
T ~ |Bmag| * |Bext| * sin(delta) where delta is the angle between them
when delta is 0, the torque is zero. As you increase delta, the torque increase. But only up to a maximum where delta=pi/2... at that point the torque suddenly reverses. That's like pulling on a rubber band, the further you pull the more the tension, until suddenly the band snaps.
Above is an experiment was set up in a stationary reference frame (other than slow change in delta).
Now consider it was only stationary relative to a synchronous rotating reference frame associtaed with the external field (everything moving together at that sync speed). You can substitute power for torque since P=T*w. The external field plays the role of the stator field, the PM plays the role of the rotor field. So now we see the physical interpretation of delta as an angle between rotor field and stator field.
Let's come at it from a completely different viewpoint. Setting aside machine theory for the moment, if you study the fast decoupled load flow approximation for power transmission accross an inductive branch in a network, then you can come up with something like:
P = |V1| * |V2| *sin(delta) / |XL|
where delta is the angle between vectors V1 and V2 and XL is the inductive reactance in the series branch between nodes 1 and 2.
(actually I think the FDLF may simplify sin(delta)~delta but that simplification is not needed here).
Apply this result to a steady state 2-element model of a sync generator as an internal ideal voltage with a series synchronous reactance. Then the angle between the internal voltage source and the terminal voltage is delta which is the power angle you're talking about. The angle between the voltages is the same as the angle in the fields in the previous analogy. This may be the roundabout way of looking at it, but sometimes it helps to look from multiple angles (no pun intended).
=====================================
(2B)+(2B)' ?