The New Method / Substep / Concept Idea Thread

[efattah]

A few people asked me to post example solves and reconstructions with my new LMCF method, so I did that, this video has two example solves with reconstructions (a 10.60 second 38 move solve and a 14.96 second 48 move solve):

I'm not that experienced and turn very slowly (3 to 3.7 tps), but since this method often results in sub-40 move solutions, that's the only way I can even approach 10 seconds at the moment. My PB is 9.90 (not on video) and I have had several solves of 32-34 moves, though the average is 45.

The general procedure of the method is:
- 1 look the corners using full EG (126 algorithms)
- Solve 3 E2L pairs (edges of opposite sides) (60 algorithms)
- Solve 1 edge (if necessary; at least one edge is solved in 1/3rd of cases already)
- Orient last 5 edges (40 algorithms if you avoid rotations)
- Permute last 5 edges (3 algorithms)

The reconstructions are:
10.60 solve:
WCA scramble generated by Cube Explorer
R2 U' L2 F2 L2 U2 F2 D' L2 B' U2 L U L D B2 R' D

z x' R U' R' M2 // Build blue face, solve blue center
y // rotate
U' B U' r2 F2 U' F //EG1 Anti-Sune, diagonal antimatch case
z // rotate
B U M' U' B' // solve 1st E2L pair (one blue edge, one green edge)
R2 x L' // set up second E2L pair
U' M U // solve 2nd E2L pair (one blue edge, one green edge)
x2 // rotate
M U M' U' // solve final E2L edge piece
M2 // move M-slice to check midge orientations
M' U M' U M' U M' U // orient last 5 edges
U2 M2 U2 M' // permute last 4 edges
38 moves STM, 6 rotations

Second solve (14.96):
WCA scramble:
F D2 F2 U2 L' B D R' D F' R B2 U B' F' R' B' U'
x2
R' u U R U' z' U2 // Build blue face & center
F R U' R' F R U2 R' U F' // EG1 Pi (left diag case)
z' x
M U M' U' // 1st E2L pair
x
M2 y U M U M' U2 // 2nd E2L pair
x z
M R // Set up 3rd E2L pair
y
M' U2 M d M' U // 3rd E2L pair
y2
M' R' // Set up L5E
U' M U' M' U M' U // Orient last 5 edges
U2 M2 U2 M2 // Permute last 5 edges
48 moves STM

The second solve has way too many rotations, it isn't a very good example, but anyway this will do for now.


Eric Fattah
Vancouver, BC

 

[TDM]

A few people asked me to post example solves and reconstructions with my new LMCF method, so I did that, this video has two example solves with reconstructions (a 10.60 second 38 move solve and a 14.96 second 48 move solve):

I'm not that experienced and turn very slowly (3 to 3.7 tps), but since this method often results in sub-40 move solutions, that's the only way I can even approach 10 seconds at the moment. My PB is 9.90 (not on video) and I have had several solves of 32-34 moves, though the average is 45.

The general procedure of the method is:
- 1 look the corners using full EG (126 algorithms)
- Solve 3 E2L pairs (edges of opposite sides) (60 algorithms)
- Solve 1 edge (if necessary; at least one edge is solved in 1/3rd of cases already)
- Orient last 5 edges (40 algorithms if you avoid rotations)
- Permute last 5 edges (3 algorithms)

The reconstructions are:
10.60 solve:
WCA scramble generated by Cube Explorer
R2 U' L2 F2 L2 U2 F2 D' L2 B' U2 L U L D B2 R' D

z x' R U' R' M2 // Build blue face, solve blue center
y // rotate
U' B U' r2 F2 U' F //EG1 Anti-Sune, diagonal antimatch case
z // rotate
B U M' U' B' // solve 1st E2L pair (one blue edge, one green edge)
R2 x L' // set up second E2L pair
U' M U // solve 2nd E2L pair (one blue edge, one green edge)
x2 // rotate
M U M' U' // solve final E2L edge piece
M2 // move M-slice to check midge orientations
M' U M' U M' U M' U // orient last 5 edges
U2 M2 U2 M' // permute last 4 edges
38 moves STM, 6 rotations

Second solve (14.96):
WCA scramble:
F D2 F2 U2 L' B D R' D F' R B2 U B' F' R' B' U'
x2
R' u U R U' z' U2 // Build blue face & center
F R U' R' F R U2 R' U F' // EG1 Pi (left diag case)
z' x
M U M' U' // 1st E2L pair
x
M2 y U M U M' U2 // 2nd E2L pair
x z
M R // Set up 3rd E2L pair
y
M' U2 M d M' U // 3rd E2L pair
y2
M' R' // Set up L5E
U' M U' M' U M' U // Orient last 5 edges
U2 M2 U2 M2 // Permute last 5 edges
48 moves STM

The second solve has way too many rotations, it isn't a very good example, but anyway this will do for now.


Eric Fattah
Vancouver, BC

This is CF except you're solving edges in pairs, which is a nice idea but I can see recognition being difficult even if the solve is more efficient.
Why do you set up to L5E instead of LSE? LSE would be faster since it needs less setup, has better recognition and better lookahead.

 

[efattah]

This is CF except you're solving edges in pairs, which is a nice idea but I can see recognition being difficult even if the solve is more efficient.
Why do you set up to L5E instead of LSE? LSE would be faster since it needs less setup, has better recognition and better lookahead.

E2L pair recognition (at least as I have designed it) is slightly easier than F2L pair recognition-- whereas E2L triplet recognition is harder, except in three cases, which are the only cases where I will solve triplets in speed solves. However, E2L lookahead is just as challenging as F2L lookahead; once you ID the pair your muscles execute the 5-7 move algorithm while your eyes look for the next pair. However you have a major statistical advantage, in that once you solve the corners there is a 52% chance an edge piece is solved and about 35% chance that it is in the 8 that you are trying to solve. And in around 10% of solves you have 2 edge pieces pre-solved, and in about 10% of solves you get a triplet that solves itself while you are solving a pair. Overall I'd say recognition and lookahead are not significantly different than F2L/CFOP, except you solve the cube in 10 fewer moves on average.

Regarding L6E, you can only use Roux L6E 50% of the time because it requires the final UL and UR keyholes to be sitting on bars of equal or opposite colors, otherwise the edge orientation recognition step doesn't work. I have tried forcing this case but it breaks the flow of the solve. Either way, I find L5E faster than Roux L6E anyway (by about 15%).

I doubt anyone will be interested in using this method (especially due to the large algorithm sets and high learning curve), but I still plan on developing it and eventually competing with it.

Eric Fattah
Vancouver, BC

 

[Attila]

A few people asked me to post example solves and reconstructions with my new LMCF method, so I did that, this video has two example solves with reconstructions (a 10.60 second 38 move solve and a 14.96 second 48 move solve):

I'm not that experienced and turn very slowly (3 to 3.7 tps), but since this method often results in sub-40 move solutions, that's the only way I can even approach 10 seconds at the moment. My PB is 9.90 (not on video) and I have had several solves of 32-34 moves, though the average is 45.

The general procedure of the method is:
- 1 look the corners using full EG (126 algorithms)
- Solve 3 E2L pairs (edges of opposite sides) (60 algorithms)
- Solve 1 edge (if necessary; at least one edge is solved in 1/3rd of cases already)
- Orient last 5 edges (40 algorithms if you avoid rotations)
- Permute last 5 edges (3 algorithms)

The reconstructions are:
10.60 solve:
WCA scramble generated by Cube Explorer
R2 U' L2 F2 L2 U2 F2 D' L2 B' U2 L U L D B2 R' D

z x' R U' R' M2 // Build blue face, solve blue center
y // rotate
U' B U' r2 F2 U' F //EG1 Anti-Sune, diagonal antimatch case
z // rotate
B U M' U' B' // solve 1st E2L pair (one blue edge, one green edge)
R2 x L' // set up second E2L pair
U' M U // solve 2nd E2L pair (one blue edge, one green edge)
x2 // rotate
M U M' U' // solve final E2L edge piece
M2 // move M-slice to check midge orientations
M' U M' U M' U M' U // orient last 5 edges
U2 M2 U2 M' // permute last 4 edges
38 moves STM, 6 rotations

Second solve (14.96):
WCA scramble:
F D2 F2 U2 L' B D R' D F' R B2 U B' F' R' B' U'
x2
R' u U R U' z' U2 // Build blue face & center
F R U' R' F R U2 R' U F' // EG1 Pi (left diag case)
z' x
M U M' U' // 1st E2L pair
x
M2 y U M U M' U2 // 2nd E2L pair
x z
M R // Set up 3rd E2L pair
y
M' U2 M d M' U // 3rd E2L pair
y2
M' R' // Set up L5E
U' M U' M' U M' U // Orient last 5 edges
U2 M2 U2 M2 // Permute last 5 edges
48 moves STM

The second solve has way too many rotations, it isn't a very good example, but anyway this will do for now.


Eric Fattah
Vancouver, BC

how long you practice it?
this is your main method?

 

[efattah]

how long you practice it?
this is your main method?

I have been working on the LMCF method since I first got back into cubing in early January (so it's been about 3.5 months).
Prior to that I was cubing in the 80's. In the 80's my best ever average was 45 seconds with a method similar to Minh Thai. When I picked up a cube in January it took me 80 seconds to solve it...
Now LMCF is my main method, doing CFOP or Roux takes me about 80 seconds each (4LLL and 2 look CMLL).

The problem is I am not a top 2x2 solver (I don't do 2x2) so it takes me longer than 15 seconds to 1-look the corners, and I still don't know EG2 yet, so I can only get good times on scrambles when I can 1-look the corners. I need to be able to 1-look the corners in 15 seconds every time, to be able to compete.

The most interesting development in this method was some of the 60 E2L algorithms. Normally once the corners are solved the cube is solved with M/U moves (in classic CF solving, I mean). However as I used cube explorer to generate the E2L algorithms, a subset of the E2L cases ended up so weird (and fast) that I call them the 'black magic' set, and they defy any classic CF style moves.

One example in the black magic set is this:
U' r U' M U R' U
(puts DF edge into UL, while solving the displaced keyhole edge UR into FR; and if you happen to have the correct edge in UL, you can also get a triplet solve sometimes)
Use your right index for the second U' and watch the black magic happen!

Eric Fattah
Vancouver, BC

 

[Shiv3r]

I think that a petrus block and then insert edge and do a psuedo-L2S would actually be pretty fast. I can get sub-30 with the method i proposed earlier(to put in in perspective im sub-40 with my main) here's the method right now:
1.Create Petrus 2x2x3(during inspection)
2.insert last d-layer edge and one remaining e-slice edge, ignoring the corner under it.(you could do cross + 2 f2l2 pairs if you wanted to)
3.create the final corner-edge pair and then orient the edges with Vandenbergh-Harris last slot(~5 algs), and then without caring for the final d-layer corner, use one of the 7 MGLS CLS -I algs(7 and mirrors, there is a 2-gen alg for every single case) to orient all the final corners.
4.insert the final d-layer corner using 1 algorithm: (R' F R F')*3
5.1-Look last layer(21 cases, if you already know full PLL than you don't need to learn any new algs)

Also, if anyone has a method that starts with a 2x2x3 petrus block please tell me as that's the fastest first two steps I can do even though I know all the 4 main methods.

for this method I'm trying to come up with a way of inserting the final corner and permuting all the other corners relative to each other so the last layer is a 1-look, 4-alg last layer.

 

[SpeedCuber71]

https://www.speedsolving.com/forum/...uare-1-roux-vandenberghs-hybrid-method.60704/

 

[Shiv3r]

thats for square-1, and transferring it to 3x3 makes any square-1 algorithms hopelessly inefficient.

 

[wir3sandfir3s]

I think that a petrus block and then insert edge and do a psuedo-L2S would actually be pretty fast. I can get sub-30 with the method i proposed earlier(to put in in perspective im sub-40 with my main) here's the method right now:
1.Create Petrus 2x2x3(during inspection)
2.insert last d-layer edge and one remaining e-slice edge, ignoring the corner under it.(you could do cross + 2 f2l2 pairs if you wanted to)
3.create the final corner-edge pair and then orient the edges with Vandenbergh-Harris last slot(~5 algs), and then without caring for the final d-layer corner, use one of the 7 MGLS CLS -I algs(7 and mirrors, there is a 2-gen alg for every single case) to orient all the final corners.
4.insert the final d-layer corner using 1 algorithm: (R' F R F')*3
5.1-Look last layer(21 cases, if you already know full PLL than you don't need to learn any new algs)

Also, if anyone has a method that starts with a 2x2x3 petrus block please tell me as that's the fastest first two steps I can do even though I know all the 4 main methods.

for this method I'm trying to come up with a way of inserting the final corner and permuting all the other corners relative to each other so the last layer is a 1-look, 4-alg last layer.

Cool idea, but how efficient would it be? Let's compare this to normal L2S.
Steps 1 and 2 are the same.
For L2S, both pairs are created and put in the U layer. 1 alg is then used to orient corners, then another to permute everything. So essentially 2LLL with an XXcross 95% of the time, other 5% is better.
For your idea, EO is not needed possibly making it slightly more efficient, though I am not too sure. So, insert 1 edge, essentially doing 3 pairs just without one corner. Then 2 Algs to finish F2L and do OLL, and 1 to do PLL. Essentially a Xcross with a 3LLL 90% of the time, other 10% is better.
Obviously normal L2S is better here, but I think they can be crossed further to be made possibly better.
So, steps 1 and 2 are the same. Make sure all edges are oriented (just do EO at some point). Place the back E slice edge then do CLS, fully orienting the LL. Then, insert the last pair with the normal L2S PLL Algs. This should require more Algs than normal L2S, but you don't have to create the pairs and place them in the U layer as you normally would in L2S, finishing the pairs is made very easy here.

 

[wir3sandfir3s]

thats for square-1, and transferring it to 3x3 makes any square-1 algorithms hopelessly inefficient.

This thread isn't just for 3x3...

 

[gyroninja]

This thread isn't just for 3x3...

I think he is talking about some recent methods which suggest you use sq1 algs to permute corners while preserving co. The algs for sq1 when used on 3x3 have really high qtm.

 

[Shiv3r]

Cool idea, but how efficient would it be? Let's compare this to normal L2S.
Steps 1 and 2 are the same.
For L2S, both pairs are created and put in the U layer. 1 alg is then used to orient corners, then another to permute everything. So essentially 2LLL with an XXcross 95% of the time, other 5% is better.
For your idea, EO is not needed possibly making it slightly more efficient, though I am not too sure. So, insert 1 edge, essentially doing 3 pairs just without one corner. Then 2 Algs to finish F2L and do OLL, and 1 to do PLL. Essentially a Xcross with a 3LLL 90% of the time, other 10% is better.
Obviously normal L2S is better here, but I think they can be crossed further to be made possibly better.
So, steps 1 and 2 are the same. Make sure all edges are oriented (just do EO at some point). Place the back E slice edge then do CLS, fully orienting the LL. Then, insert the last pair with the normal L2S PLL Algs. This should require more Algs than normal L2S, but you don't have to create the pairs and place them in the U layer as you normally would in L2S, finishing the pairs is made very easy here.

I think this Idea would be probably better, but you have to remember that some people are too lazy to learn all those Algs.
I think maybe your method is faster, but I think that as an entry-level L2S idea, mine may be halfway decent. Mostly because the algorithm count is better than CFOP, and so is the movecount
anyway, I'm not saying any one is better, Im just saying that one is faster while one is easier to learn

 

[wir3sandfir3s]

I think this Idea would be probably better, but you have to remember that some people are too lazy to learn all those Algs.
I think maybe your method is faster, but I think that as an entry-level L2S idea, mine may be halfway decent. Mostly because the algorithm count is better than CFOP, and so is the movecount
anyway, I'm not saying any one is better, Im just saying that one is faster while one is easier to learn

True, but tbh I would still pick CFOP over your idea.

 

[Shiv3r]

True, but tbh I would still pick CFOP over your idea.

thats fine

 

[gyroninja]

I think that a petrus block and then insert edge and do a psuedo-L2S would actually be pretty fast. I can get sub-30 with the method i proposed earlier(to put in in perspective im sub-40 with my main) here's the method right now:
1.Create Petrus 2x2x3(during inspection)
2.insert last d-layer edge and one remaining e-slice edge, ignoring the corner under it.(you could do cross + 2 f2l2 pairs if you wanted to)
3.create the final corner-edge pair and then orient the edges with Vandenbergh-Harris last slot(~5 algs), and then without caring for the final d-layer corner, use one of the 7 MGLS CLS -I algs(7 and mirrors, there is a 2-gen alg for every single case) to orient all the final corners.
4.insert the final d-layer corner using 1 algorithm: (R' F R F')*3
5.1-Look last layer(21 cases, if you already know full PLL than you don't need to learn any new algs)

Also, if anyone has a method that starts with a 2x2x3 petrus block please tell me as that's the fastest first two steps I can do even though I know all the 4 main methods.

for this method I'm trying to come up with a way of inserting the final corner and permuting all the other corners relative to each other so the last layer is a 1-look, 4-alg last layer.

This is a less efficient version of zz-ct.

 

[Shiv3r]

This is a less efficient version of zz-ct.

what's the premise of that?
also, I think that that LCMF corners-first method may be good. i mean, cmon, he does less TPS than I do , and he can average 15 secs. With tou, I average sub-40, maybe I should try! it'll also teach me EG method too so thats a plus.
What are the algorithms effatah? for the E2L and such? please tell :3

 

[gyroninja]

what's the premise of that?
also, I think that that LCMF corners-first method may be good. i mean, cmon, he does less TPS than I do , and he can average 15 secs. With tou, I average sub-40, maybe I should try! it'll also teach me EG method too so thats a plus.
What are the algorithms effatah? for the E2L and such? please tell :3

Since it's ZZ you don't have to do vhf2l. I guess your next step is pretty much tsle. Instead of doing three sledge hammers and a pll you just do a ttll to solve the whole cube. You are doing two extra looks than zz-ct.

 

[crafto22]

Since it's ZZ you don't have to do vhf2l. I guess your next step is pretty much tsle. Instead of doing three sledge hammers and a pll you just do a ttll to solve the whole cube. You are doing two extra looks than zz-ct.

Wait this could be a really cool idea. What if we added in a bit of some other stuff too:

1. 2x2x2 in BDL (6 moves)
2. Pair in FDL + E-slice EO (6 moves)
3. 2x2x1 added on to the 2x2x2 in the back to form a 2x2x3 + 1 pair (8 moves)
4. Final E-slice edge with LL EO (6 moves)
5. COLL (10 moves)
6. L5EP (10 moves)

46 STM seems pretty good to me

 

[gyroninja]

Wait this could be a really cool idea. What if we added in a bit of some other stuff too:

1. 2x2x2 in BDL (6 moves)
2. Pair in FDL + E-slice EO (6 moves)
3. 2x2x1 added on to the 2x2x2 in the back to form a 2x2x3 + 1 pair (8 moves)
4. Final E-slice edge with LL EO (6 moves)
5. COLL (10 moves)
6. L5EP (10 moves)

46 STM seems pretty good to me

I think you replied to the wrong person.

 

[JTWong71]

Wait this could be a really cool idea. What if we added in a bit of some other stuff too:

1. 2x2x2 in BDL (6 moves)
2. Pair in FDL + E-slice EO (6 moves)
3. 2x2x1 added on to the 2x2x2 in the back to form a 2x2x3 + 1 pair (8 moves)
4. Final E-slice edge with LL EO (6 moves)
5. COLL (10 moves)
6. L5EP (10 moves)

In step 2, is the pair a 2x2x1, or a 1x1x2 Block?