You should learn commutators.
A B A' B'
Example: r U r' u' r U' r' u
Last edited by Unnoticed; 07-02-2012 at 08:39 AM.
To do this intuitively, use the same technique for solving a big cube. (1) do the faces, then fix the edges (if necessary), and (2) solve the 3x3. The only thing is that in solving the 3x3, you want to make sure you don't change the orientation of your faces.
(1) To get the faces, follow this guy's recipe for last two faces. Note that since what you're doing is actually "breaking" the edges, you have to do it a little differently: you want to make sure that the cubie that you grab has the color that you need. Also, his technique assumes that your last two faces are adjacent. Since the color scheme here mixes opposite faces you could either (a) move some center cubies around so that they're adjacent to where you want them, or (b) what I did, which is to replace all the 90 degree slice moves with 180 degree slice moves. Either way, here's the video:
"The Borrowing Method"
In solving the 7x7, I use the above method, modified slightly. It turns out that the last two faces is the only part where my brain is faster than my fingers. My modification is that if I can move two or more cubies at the same time I'll do that. The change to the above algorithm is to move more than one slice at a time. You still are messing around with two "slices", but now one (or both) of your slices is actually more than one slice (which need not be adjacent). Eventually I'll have to trade this algorithm for better, but right now my lousy times are dominated by (1) my slow finger speeds, and (2) edge pairing which I expect to slowly improve. I just got a metronome, maybe it will help.
(2) To solve the 3x3x3 without messing up the faces, use a 3x3 solving technique which is pure commutator. That is, use RUR'U' and RU'R'U repeatedly and nothing but this. Each of these algorithms has just as many R as R' (and just as many U as U') and so there is no net rotation on a face. Another example of a commutator algorithm is F (RUR'U') F', which also does not change the orientation of a face. (By the way, when solving a cube with images on the sides, it's useful to have 3x3x3 solving algorithms that do rotate a face.)
This is actually how I used to solve cubes (for 30 years) until I decided to speed cube (cheaper than golf) a few months ago. I get all the edges into place first (ignoring the corners), and then solve the corners by using (RUR'U')(RUR'U')(RUR'U') which swaps corners but doesn't move any edges.
Last edited by CarlBrannen; 07-02-2012 at 10:39 PM.