You cut the top plate into a [ shape and the pole into a ] shape. The cut-out area is referred to as the "rebate" -- this rebate steers the flux towards the two separate gaps on either side of itself to extend the BL field out farther to either side than a motor without the rebate. This creates a flat BL curve and thus less BL distortion and more linear x-max.
Some people get the idea that XBL^2 actually loses BL, but it really does not, it steers it around... in some motors I've drawn up in FEA I was able to get a higher BL with XBL^2 than with other "standard" topologies. Even in motors where the "at rest" BL is a bit lower, the average BL over the useful range of the speaker will be higher.
Jacbo, You're 100% right about "it steers" it around, i think you meant to say "B" not BL,
and yes, it conserves the flux lines, but it also couples with 50% less lines in the total gaps so............in my calculations (i wrote software to do this using FEA motors i designed) i was not able to get more "BL" from any XBL^2 model i measured. in fact, i have a hypothesis that any linear motor must sacrifice BL... worded a little differently of course.
my models of XBL^2 include two solutions that are pretty convincing. I FEA'ed three motors. An overhung, and then two XBL^2 motors. One with 50% shorter and 50% lighter coil and one with 50% shorter coil with double the turns (same mass and resistance the overhung)
The results are as follows: the first XBL^2 motor has aprox an 27% BL(0) loss. Very large! the second had about a 9% BL(0) loss. Both gap widths were corrected for each coil. The first xbl^2 model used the same gap as the overhung (say # of layers on the coil) the second XBL^2 motor has a larger gap for the larger coil. the second model was much better in terms of coupling but still lower that our overhung, and the inductance of the system exceeded that of the overhung because of the extra coil in the gaps. Sure you can put a shorting ring in the notch in the gaps on either side, but you can do the same for an overhung coil with an undercut t-yoke which is basically what the XBL^2 t-yoke is. putting a shorting ring directly in the active GAP and winding it is going to reduces sensitivity the same for either XBL^2 or overhung, its a wash.
about the linear stuff, sure, the overhung has higher BL, but its still a non-linear method, the XBL^2 is a linear method, so there is the trade off. XBL^2 has lower distortion. But XBL^2 also has a difficult time correct for asymmetry about the x axis. variable coil solutions can in fact easily correct for this without having to use a very high extended poll piece. I have found that most XBL^2 motors i have modeled have higher xmax on the down-stroke. This can be corrected with a carefully designed motor im assuming.
my point is, Dan W. is misleading about his accusations according to my analysis and im out to correct the record. I have debated this with him before and i will be publishing an extensive paper on this at some point. I'm not the only one to come up with this conclusion too. Steve Moray (former Bose engineer) did similar research completely independent of me and found nearly the exact same results.
http://www.s-m-audio.com/topologies.pdf
his model included a "well hung" motor which is simialr to a DD design or a 3HP /4HP motor