The New Method / Substep / Concept Idea Thread

[Thermex]

Here's my L6EP document! 86 algs, 82 disregarding EPLL with an avg of ~12 moves. Fully <RU>. Some of the cases are very difficult to recognize, especially the last set (it can be easily avoided though)

Full CLS is 104 algs, so either 190 or 186 algs. Since it only saves a few moves (possibly) you may want to stick with F2L+2GLL

Both me and @Teoidus phrased that wrong, it's not actually CLS, it's pretty much a two-gen L5C set. I'll post more about that and some of the 42 TC-conjugated CMLL set tomorrow.

 

[Neuro]

Both me and @Teoidus phrased that wrong, it's not actually CLS, it's pretty much a two-gen L5C set. I'll post more about that and some of the 42 TC-conjugated CMLL set tomorrow.

It's still 104 algs. It has identical recog to CLS (orientation only) although it is, of course, more efficient because you can cycle F2L/LL edges. Still don't think it'd be worth it, sorry

 

[Teoidus]

I think it might be lower than you'd expect, since chance of a usable pair increases a lot. I'd have to see how pseudo L6EP recognition is though compared to 2GLL

 

[Thermex]

It's still 104 algs. It has identical recog to CLS (orientation only) although it is, of course, more efficient because you can cycle F2L/LL edges. Still don't think it'd be worth it, sorry

Trust me on this one, I've genned a quarter of the algs already. If one u-layer corner is oriented in UBL, you have 18 possible orientations in the unsolved part (the unsolved pieces form a "V" shape, that's why I like to call this VLC). One of those orientations is when you have all corners oriented, which because of rotational symmetry is only one case (R U' R' U2 R U R'). For the remaining 17 orientations, the D-layer corner can be in one of 4 places relative to the oriented UBL corner, and since you already did a CP fix, you just multiply 4×17 and get 68, and then add in the one TTLL to get 69 cases. This was a bit more than I predicted (I forgot to include the TCLL and COLL cases) but still pretty learnable. Another thing to note is that a lot of the HD orientation algs are used in this set, and half of those are not even real "algorithms" (3 move inserts, well known OLLs, etc.). So that leaves like 150 algs total for the method, but like I said you could cut off about 20 cases that are just trivial. Still 45 algs more than 2GLL, but I honestly think it's worth it. You could also probably switch a couple of steps around earlier in the solve to not only boost the efficiency but maybe get rid of some of the bad L6EP cases. Here's a comparison of the two methods up until LS:

VLC+L6EP:
84 algs (VLC 69+L5EP 15=84 algs)
~20 moves (VLC 7+L5EP 11+2 AUFs=20)

2GLL:
84 algs
~20.5 moves (pair 6+alg 13+1.5 AUFs= 20.5)

Dang these are close... the main advantage of the one I came up with is that you can build onto in and add L6EP and also that when using two hands with optimal L5EP cases the movecount is closer to 17 or 18.

 

[Teoidus]

Wait, that can't be right. Inserting the last corner + orienting the remaining corners is identical to CLS in case count, so it should be 104.

EDIT: I realize my earlier post wasn't clear. The move count might be lower than youd' expect, not the case count.

 

[Thermex]

Wait, that can't be right. Inserting the last corner + orienting the remaining corners is identical to CLS in case count, so it should be 104.

EDIT: I realize my earlier post wasn't clear. The move count might be lower than youd' expect, not the case count.

Since you force one oriented corner in UBL it is actually 69 cases. Trust me, just check the math I did.

 

[Neuro]

Just realized; with 42-style recognition, you can cut down on corner cases by a LOT so that may be something to consider. I think that forcing an oriented corner in UBL *might* be counterproductive as it seems to be a waste of moves.

 

[Thermex]

Just realized; with 42-style recognition, you can cut down on corner cases by a LOT so that may be something to consider. I think that forcing an oriented corner in UBL *might* be counterproductive as it seems to be a waste of moves.

If you use 42 style solving of the corners you can signifcantly cut down on cases but not moves. You would end up having the 7 2-gen COLLs that are okay but not great. Using VLC allows for 3 move inserts among other ultra-fast cases, cutting the movecount down by 2-3 moves. Also both methods require you to orient a corner and put it in the U/B layer.

 

[Teoidus]

I agree with Neuro on this one--Pair/CLS/L6EP seems much better than Pair/setup/CLS subset/L6EP.

The CLS algs are all crazy easy 2gen stuff anyway, so why complicate things?

 

[Thermex]

I agree with Neuro on this one--Pair/CLS/L6EP seems much better than Pair/setup/CLS subset/L6EP.

The CLS algs are all crazy easy 2gen stuff anyway, so why complicate things?

Orienting one U-layer corner takes absolutely no effort to do and cuts off 35 cases. Most of the CLS cases with all corners disoriented are way worse than the ones where the corners are oriented.

Quick update on the TC stuff I've been working on: First off, I manually generated the cases for "Valiant Opposum", here are the algs: https://docs.google.com/spreadsheet...WgDXqQQOyLezr4Nscvqu3Vay-w/edit#gid=670959495. This took me a long time to do for some reason and the average for the L5C+FR edge is about 17 or 18 moves, almost exactly what CMLL is. Not really sure if I think this is even a good method or not and I don't even know if it's worth creating a wiki page/thread over it, but it's out there now. I guess it could be good for someone who uses VOP or HD on 2x2. As for the Briggs/42 TC variant @shadowlice e proposed yesterday, I tried to make to make some algs for it and it's just too darn confusing. I'm going to try to work it out today but it's really weird when you're trying to connect the FR edge to DFR corner when everything is already conjugated and moved around by the initial R move :/

 

[Pyjam]

ZZ LSLL method

1. insert edge + phase edges

4 cases including R U R', R U' R', and R U2 R', barely even a step

2. orient corners while preserving phased edges

23 cases, 6 are normal OLLs and the rest are pretty easy

3. TTLL with phased edges

one of 24 TTLL cases


Average movecounts:
step 1: 3 or 4
step 2: 9
step 3: 13 or 14

26 average movecount with less than 50 total algs, and recog is pretty easy

also you have a 1 in 11.4 chance of skipping a step, making it 2-look

I'm interested. Could you provide the algs for step 2, please?
And maybe for step 3 also.

 

[Thermex]

Okay I think I have some ideas to improve the Roux method.

1) I call this mispermuted centers. Basically the idea is that if you see a really easy FB, but the center would be wrong, just build it anyway. Before CMLL you can easily fix the centers with a couple wide u moves and an M. This adds a few moves after SB but can drop the FB move count significantly.

2) I call this liberated blocks. The idea here is that if you see a really easy SB square (or even the entire block) after you start FB, you can build that and then finish FB after. This allows some flexibility in F2B to take better advantage of lucky cases to get lower move counts.

3) This concept I call bi-color blocks. What it means is that you don't have to use the same color on bottom for both blocks. The CMLL recognition becomes slightly harder but being able to choose whichever SB is best should lower the move count a lot, and the cost is only a single R or r move after LSE.

4) S-CMLL, or Super-CMLL. This is an alternate way of doing CMLL. It requires learning only 84 algs (which is quite reasonable). The goal here is to influence EO while solving the corners to give an easier LSE case. The idea is that 42 of the algs will flip two edges while solving the corners, and 42 of them won't flip any edges while solving the corners. By adjusting the M slice to flip certain edges before performing an edge-flipping alg, or by choosing to use an EO preserving alg, you can always avoid a 6 flip and generally get better EO cases. Since the algs hardly vary in move count at all from standard CMLL (within 1 or 2 at the worst), you should get a move count reduction off of the easier LSE cases.

5) Conjugated L4EP is the idea of solving UF and UB instead of UL and UR during step 4b. Then you can offset them with a U or U' and then solve L4E with different colors (the recognition is a little harder). This should allow you to take advantage of easier 4b cases and save several moves.

6) I call this step 95, because it does step 4a and step 4b at the same time, and in hexadecimal, 4a + 4b = 95. Basically the idea is to solve EO in a specific way that puts UL and UR (or UF and UB if you combine step 95 with conjugated L4EP) into a very easily solved spot. This should shave of almost all the moves of step 4b without adding any to 4a's move count.

I think that with all these advancements, you could get sub-40 move count solves with Roux, or even sub-35, or maybe even less.

Did you get all of these tricks off Kian Mansour's above video? He goes over every single one of those there. Low 40s is probably what most people coule get after perfecting all these tricks.

 

[Teoidus]

Orienting one U-layer corner takes absolutely no effort to do and cuts off 35 cases. Most of the CLS cases with all corners disoriented are way worse than the ones where the corners are oriented.

I still don't think this is worth it. It's an extra thing to check that can be easily avoided by just learning more algs.

 

[Arc]

Did you get all of these tricks off Kian Mansour's above video? He goes over every single one of those there. Low 40s is probably what most people coule get after perfecting all these tricks.

It was satirical. Top Rouxers don't even use the first three regularly. It's totally feasible to get sub 40 move counts with vanilla Roux. If you saved an average of one move per solve with each one of those, you could get sub 35 move counts. The point is that there seems to be a huge divide between theory and practice in this thread as of late. It's great to have new ideas but people seem to be jumping on everything as if it's the best thing since sliced bread.

And there seems to be such a focus on move count versus alg count with complete disregard for anything else. One of the most interesting ideas to me in method creating is the way that CFOP keeps your wrist in a neutral position throughout all of F2L. In fact most top Rouxers solve SB as DR + 2 pairs, which also has this effect. The faster ZZ solver in the world uses EOCross, same exact effect. I think it's something to really look into.

When you gen algs, you balance move count with ergonomics to try to find the alg that can be executed the fastest, and it should be no different with method creation.

This is just my two cents.

 

[Thermex]

It was satirical. Top Rouxers don't even use the first three regularly. It's totally feasible to get sub 40 move counts with vanilla Roux. If you saved an average of one move per solve with each one of those, you could get sub 35 move counts. The point is that there seems to be a huge divide between theory and practice in this thread as of late. It's great to have new ideas but people seem to be jumping on everything as if it's the best thing since sliced bread.

And there seems to be such a focus on move count versus alg count with complete disregard for anything else. One of the most interesting ideas to me in method creating is the way that CFOP keeps your wrist in a neutral position throughout all of F2L. In fact most top Rouxers solve SB as DR + 2 pairs, which also has this effect. The faster ZZ solver in the world uses EOCross, same exact effect. I think it's something to really look into.

When you gen algs, you balance move count with ergonomics to try to find the alg that can be executed the fastest, and it should be no different with method creation.

This is just my two cents.

Agreed.

EDIT: After looking into TC variants more I really like them less and less, adding random Ms in an algorithm can be a pain, especially since Imo L7E is almost as fast as LSE. I think I'll switch over to working on L7E for a while, maybe I'll gen algs for the FR+EO (do these exist yet?)

 

[efattah]

Agreed.

EDIT: After looking into TC variants more I really like them less and less, adding random Ms in an algorithm can be a pain, especially since Imo L7E is almost as fast as LSE. I think I'll switch over to working on L7E for a while, maybe I'll gen algs for the FR+EO (do these exist yet?)

Some of the Waterman L6E algorithms can be repurposed as L7E FR+EO; in some case shorter algorithms might exist as you don't need to preserve UL and solve UR (as the Waterman algorithms do). The Waterman algorithms would only be used in the case where the FR edge is already in FR or UR slot. I'm quite curious to see the L7E FR+EO set, it could be quite useful. However with no setup moves there are at least 7 situations where FR could be (M-slice (O,D), FR(D), UR(O,D), UL(O,D)) and each of these has around 8 orientation cases so there are around 56 algorithms for the right side situation and 56 more for the left side (assuming FL+EO).

 

[Neuro]

I calculated that if you have FR in DF, you have 30 algs (15 for FR edge oriented and 15 when it's disoriented) but having it in either DF or DB gives 60 algs (this is probably the better option)

You may want to make a set for when the FR edge is in the slot but disoriented and *maybe* one for when the edge is in UL (this bumps us to 120, but you'd probably be better off learning 90, where you can have it in DF, DB, or disoriented in FR)

W/O setup, 30*7=210. I don't really think it'd be worth it to do it like this but it'd be doable.

Algs shouldn't be too hard to generate, I can probably finish it over the week if you want them.

@Arc SB can be made more efficient by using edges other than DR to build off of (similar to open-slotting in ZZ) but the rest I tend to agree with. There's a point where saving moves isn't worth the alg count (i.e. saving 3 moves by using >400 algs) I do think it's important to work on methods to get low movecounts, but if you are serious about it becoming a speedsolving method, you may need to make some sacrifices to make it approachable/feasible for people to learn.

 

[gogozerg]

One of the most interesting ideas to me in method creating is the way that CFOP keeps your wrist in a neutral position throughout all of F2L. In fact most top Rouxers solve SB as DR + 2 pairs, which also has this effect. The faster ZZ solver in the world uses EOCross, same exact effect. I think it's something to really look into.

That's one of the reasons why many people discarded the idea of starting with solving DF+DB while orienting the other edges. Long before some people presented it as the shiny new ultimate method.

 

[Abram Lookadoo]

a new method

1) solve the first two layers except the corners orientation
2) eopll 157 algorithms
3) ol8c 191 algorithms (auf+adf)

 

[Thermex]

I calculated that if you have FR in DF, you have 30 algs (15 for FR edge oriented and 15 when it's disoriented) but having it in either DF or DB gives 60 algs (this is probably the better option)

You may want to make a set for when the FR edge is in the slot but disoriented and *maybe* one for when the edge is in UL (this bumps us to 120, but you'd probably be better off learning 90, where you can have it in DF, DB, or disoriented in FR)

W/O setup, 30*7=210. I don't really think it'd be worth it to do it like this but it'd be doable.

Algs shouldn't be too hard to generate, I can probably finish it over the week if you want them.

@Arc SB can be made more efficient by using edges other than DR to build off of (similar to open-slotting in ZZ) but the rest I tend to agree with. There's a point where saving moves isn't worth the alg count (i.e. saving 3 moves by using >400 algs) I do think it's important to work on methods to get low movecounts, but if you are serious about it becoming a speedsolving method, you may need to make some sacrifices to make it approachable/feasible for people to learn.

I'll split the alg load with you, they are pretty easy to gen (I made a couple already). @crafto22 has already made many of the algs (the ones where either FR is flipped or where ULUR are oriented) so that cuts off a lot of algs that need to generated.
Calculating the number of FREO alg is quite tricky imo, but I don't think it's quite 210, especially since FR flipped is only 9 cases. anyway how about you do the DF cases and I do the DB ones? I could probably finish em by this Sunday.