From:

~~~ ~~~ Subject:

As I mentioned in my last message, I used symmetries to reduce the

number of candidate sequences for the superflip. Here's how:

Suppose we have a sequence for the superflip that has at least 4 syllables.

(Here, a syllable is a maximal sequence of commuting face turns, i.e., a

maximal sequence of face turns on the same axis.) The sequence of axes

in these syllables must look like

(1) Z X ... X Y, (2) Z Y ... X Y, (3) X Z ... X Y, or (4) X Y ... X Y,

for some distinct axes X, Y, and Z. Remember that the superflip is

central, so we can cyclically permute the sequence of syllables. If

doing this always results in pattern (4), we only use two axes, but

this can't flip any edges; hence, we can get (1), (2) or (3). By inverting

the (order 2) superflip we can change (2) to (3). Then we have (1)

or (3). By applying a symmetry of the cube, we can let X, Y and Z be

the FB, UD, and LR axes, respectively.

We still have some of the symmetry group to work with, namely the set of

the eight symmetries of the cube that fix all cube axes. If we need to,

we can apply a 180-degree rotation of the cube about the UD or LR axes,

which lets us restrict the first FB syllable to 9 of the 15 possibilities;

then, rotating about the FB axis, we can do the same for the last UD syllable.

Finally, we can reflect the cube through the plane between R and L; this lets

us restrict the first LR syllable to 9 possibilities, although it expands the

number of possibilities for the last UD and first FB syllables to 10 each.

Some more estimated runtimes for my Shamir implementation: 20 CPU hr for a 20 qtw superflip; 190 CPU hr for a 22 qtw superflip. -- David Moews dmoews@xraysgi.ims.uconn.edu