Proving UF5

[mark49152]

There are many challenges in the world that seem impossible (send a man to the moon, internet over the entire planet, curing cancer) but with a few years of invested time and collective effort, it becomes mainstream and widely adapted.

Such things involve judgment, planning and innovation as well as investment of effort. There are reasons we haven't built a city on the moon yet.

As I'm sure I've said before, I'm excited to see you pushing the limits and would love to see you come up with something useful. The best contribution I can make today is to point out the folly of trying to brute-force 5-style, and encourage you to approach it differently.

Personally I think the most promising way to explore 5-style would be to focus on the 4-movers plus a systematic method of setting up other cases to those, with the objective of maximising the number of cases covered. That way, you stand a better chance of achieving a reasonable move count benefit while keeping an intuitive approach to aid with learning and recall. I'm just not sure a significant enough proportion of cases could be covered with manageable effort, but it's worth a try. You could also try cancelling 4-movers into adjacent 3-cycles, e.g. solve ABCD as (ABCE)(ED).

Edit with example:
[M', U] solves IADM speffz. IADF could be solved with (IADM)(MF) where MF is [U: [M', U L' U'].

(M' U M U')(U M' U L' U' M U L U2) = M' U2 L' U' M U L U2 = [M', U2 L' U']

 

[abunickabhi]

This video is to make sure everyone is on the same page on how it is, and how this method is not for everyone.

 

[Seniorcuber]

I find the idea good, but not the approach.

Also, the number of cases should not be the main argument. Rather, is it possible to find a well structured method that enables us to tackle all cases with just a few algorithms and less or equal than three or max four setup moves.

I don't think someone learned an algorithm for each of the 440 edge 3-cycle perms. I am using a variation of the TuRBo method. 3 algorithms with their mirror and inverse (2+2+4=8), 3 edges (=6 stickers) affected, of which each can coincide with the buffer (8x6=48) cover 48 out of 440 possible permutations, and the 392 remaining can be solved with obvious setup moves (1 or 2, at max 3).
I learned 3 algorithms to solve the 440 permutations.

Nobody will learn the sheer number of 126 720 permutations. But it should be possible to reduce that number to a set of algorithms of a reasonable size, together with simple rules to determine the algorithm, its orientation and setup moves.

The algorithm MBM'B' has 8 variations (a mirror, an inverse and a double MBM'B'MBM'B'). It affects 5 edges with 10 stickers and can thus solve 80 permutations. With an AUF-move (B, B' and B2), it can solve 80x4 = 360 permutations.

There's another 'simplification' I'm thinking of : instead of considering sticker-permutations, we could consider piece-permutations, and keep track of orientation separately. The number of permutations solved with one algorithm increases : there are 2^5 = 32 sticker-perms for one piece-perm. The 8 variations cover 4 piece-perms, they affect 5 pieces and with an AUF-move, this covers 4×5×4×32 = 2560 sticker-perms.

Here is an outline of the method :

1. Solve piece-permutation, minimizing edge flips.
   • Looking at the 5 pieces involved (buffer + letter quartets), determine the setup moves needed to have four pieces on the same plane and one on the opposite side.
   • Determine the AUF-move and the variation of the algorithm required. There are two solutions for each piece-perm. Choose the one that results in less remaining flipped edges.
   • Keep track of the edges that have the wrong orientation (11 bits of information for the whole cube. Should be doable).
2. If there's a remaining 3-cycle, do it.
3. Orient flipped edges with algs of the 3EO method.
4. Handle parity, if necessary.

There are still missing pieces :

• - How to handle breaking into a new cycle
• - There are 6 permutations of 4 pieces on the same plane, given the starting point. This alg only covers 4 of them.

Nevertheless, I think this is a good starting point for tackling the problem of 5-style.
What is your opinion?

 

[Cubinwitdapizza]

I find the idea good, but not the approach.

Also, the number of cases should not be the main argument. Rather, is it possible to find a well structured method that enables us to tackle all cases with just a few algorithms and less or equal than three or max four setup moves.

I don't think someone learned an algorithm for each of the 440 edge 3-cycle perms. I am using a variation of the TuRBo method. 3 algorithms with their mirror and inverse (2+2+4=8), 3 edges (=6 stickers) affected, of which each can coincide with the buffer (8x6=48) cover 48 out of 440 possible permutations, and the 392 remaining can be solved with obvious setup moves (1 or 2,at max 3).
I learned 3 algorithms to solve the 440 permutations.

Nobody will learn the sheer number of 126 720 permutations. But it should be possible to reduce that number to a set of algorithms of a reasonable size, together with simple rules to determine the algorithm, its orientation and setup moves.

The algorithm MBM'B' has 8 variations (a mirror, an inverse and a double MBM'B'MBM'B'). It affect 5 edges with 10 stickers and can thus solve 80 permutations. With an AUF-move (B, B' and B2), it can solve 80x4 = 360 permutations.

There's another 'simplification' I'm thinking of : instead of considering sticker-permutations, we could consider piece-permutations, and keep track of orientation separately. The number of permutations solved with one algorithm increases : there are 2^5 = 32 sticker-perms for one piece-perm. The 8 variations cover 4 piece-perms, they affect 5 pieces and with an AUF-move, this covers 4×5×4×32 = 2560 sticker-perms.

Here is an outline of the method :
1. Solve piece-permutation, minimizing edge flips.
1a. Looking at the 5 pieces involved (buffer + letter quartets), determine the setup moves needed to have four pieces on the same plane and one on the opposite side.
1b. Determine the AUF-move and the variation of the algorithm required. There are two solutions for each piece-perm. Choose the one that results in less remaining flipped edges.
1c. Keep track of the edges that have the wrong orientation (11 bits of information for the whole cube. Should be doable).
2. If there's a remaining 3-cycle, do it.
3. Orient flipped edges with algs of the 3EO method.
4. Handle parity, if necessary.

There are still missing pieces :
- How to handle breaking into a new cycle
- There are 6 permutations of 4 pieces on the same plane, given the starting point. This alg only covers 4 of them.

Nevertheless, I think this is a good starting point for tackling the problem of 5-style.
What is your opinion?

I think this is a very good idea (although it got kinda deep to where I didnt understand to well but we’ll ignore that.) That would make it a lot less algs and not be as fast but still faster than 3-style if there was enough practice put into it.

 

[One Wheel]

5-style bigBLD centers seem like a much more feasible application than anything 3BLD.

 

[dudefaceguy]

I just solved edges using 5-style intuitively, which was fun. I used DF as the buffer (U in speffz) and I conjugated the other 4 of the pieces onto the U face, making the cycle I A B E (or inversions). I solved both cases with M' U M U' (or inversions). It required very long conjugates, which I remembered without writing them down surprisingly. The first alg had a 5 or 6 move conjugate, and the second had a 7 move conjugate. I performed them in a systematic way so that it was easy to reverse them (with my eyes open, anyhow).

This amounted to about 14 moves for the first cycle and 18 moves for the second, though I probably could have canceled a few. If we say 30 moves to solve 10 pieces that's 3 moves per piece - still obviously more move-efficient than 3-cycles even with my gigantic conjugates.

Since the cycles are very easy and intuitive, it might be worthwhile to learn a partial set for a few cases that can be recognized easily. I'm a super slow noob though, so I don't really know what I'm talking about. Anyhow, it was fun executing the edge solve.

Using 5 style for corners seems less useful and more difficult, since the algs are not intuitive, and there are only 8 corners so you will save 1 commutator per solve at most.

Also, how does 5 style deal with 2 cycle breaks in the same 5 piece cycle? It seems to be that you can't solve C A B A with a 4-move commutator because you need to use the same sticker twice.

5-style bigBLD centers seem like a much more feasible application than anything 3BLD.

I use 4-move commutators to solve up to 6 pieces in my big-cube method, which leaves partial centers last. My method sets up to easy 4 movers since it leaves half-centers unsolved until the last step, when they are solved with commutators. It seems like they would require extensive conjugates to get all of the pieces in the right place if all of the center pieces were unsolved.

 

[abunickabhi]

Hi dudefaceguy,

Many of the 5 cycle algs have gigantic conjugates. I agree with that. The average count right now is 10.4 moves STM after covering 60k cases (halfway there)

I am focussing a lot of fingertricks and fingertrickablity of the 10STM move alg, so that it can be faster than 2 3-style algs.

 

[abunickabhi]

I am closing onto 60k edge algs with UF buffer. This quarantine time is really helping me develop 5-style algorithm more. Can anyone help me make a JavaScript Trainer, so that people can start practising specific subsets of 5-style?

 

[TerryD]

60, 000 algs... that means you've been learning an average of more than 100 algs every day. did you learn all of them with Yo notation? is there any website where i can turn a normal alg into a Yo notation alg? I want to try learning zbll with yo notation.

 

[qwr]

Also have some critiques for the 5 year plan

Is this Soviet Russia or something? 5 year plans don't work

 

[abunickabhi]

60, 000 algs... that means you've been learning an average of more than 100 algs every day. did you learn all of them with Yo notation? is there any website where i can turn a normal alg into a Yo notation alg? I want to try learning zbll with yo notation.

ZBLL with Yo notation doesn't work that well, as it is mostly RUF or 2-gen (2GLL). Whereas in 5-cycles, the algs are generally RUFD + MES (slice moves), so the Yo notation is more varied up, and can be broken down into letter pairs/triads/quads and memorised. ZBLL with Yo notation isnt the best idea, sticking with triggers like RUR'U' and chunking them is the best way to go.

 

[abunickabhi]

The UF video playlist has almost grown to 200 videos now yay. Still a long way for the best fingertricks being shown for each alg in video form.

 

[abunickabhi]

This lockdown has proved to be fruitful for my 5-style quest. I was in two minds always, one mind thinking that method will not give any returns, and the other was that this method will just restructure the human brain, as to how we consume information and the thin line between memorisation and understanding a thing fully that we are able to reconstruct it.

This idea has been bugging me since 2014 but it has always been fuzzy in my mind, and I was not confident enough in 3-cycles and all the other tricks involved in 3BLD exec. But now atleast I have a brief idea about all the stuff like full floating, so I have a reference point to compare.

I know my approach might not be the best to tackle such a big algset, I have chosen randomized way of reassuring the idea that I have for each letter quad and each 5-cycle that I encounter, as per the memotool and the solves I do throughout the day.

I also made some roles for people to contribute to the 5-style endeavour without dealing with bare bone 10.5 STM algs, which are not fingertrick friendly at first.

The roles that I have formed currently are:
1) Fingertrick analyst
2) Communication person (discord/YT/SS Forum)
3) Mega-manual content writer
4) Solve annotator
5) Mnemonics and Letter Quads improvement engineer

Still a long way to go before this method gains traction. As Mark Rivers said, there is a reason why people have not built a colony on moon. It is all about investment of time and effort and I fully understand that.

Making cubing much more richer is one aim, because the other mind sport that I practice seriously, chess has insane theory and great personalities who have contributed. Whereas in cubing only see people get one record and quit, get fast and attend major championships and quit and so on.

Cubing is changing a lot with lot of US cubers getting sponsorships, and also the Rubik's brand kinda getting out of its slumber and using its marketing a bit (its not well directed and enough at all but that's my personal opinion).

I hope I get something out of this, and atleast people a linguist and my brain becomes a huge transfer learning framework.

Yo!

 

[abunickabhi]

Recited part 1 of 20 of all the UF5 algs. Was a fun exercise to do. My letter quads also got revised.

 

[abunickabhi]

The year 2021 has started.

My current plans is to make more UF algs, and compared more UF5 alg drill time with the corresponding 2 3-style edge algs. I need to get more comfortable in my new system of letter quads + UF5 soon, since I plan to do a lot of attempts of MBLD. I think 3BLD is a good way to notice chances where you could use a 5-cycle that is familiar, easy and you have prepared. MBLD practice also helps as well, as we can do thorough post-mortem of each of the scrambles afterwards.

I am excited to see how UF 5-cycle fares in the future. I also want to make a small subset of 5-cycles which are lesser than 10 moves STM, it will be good collection for people who don't want to learn the full set, but want fancy things in their solves. 3-style is always needed as a fall-back solution though, as a 3BLD scramble cannot be solved with pure 5-cycles, that's just not feasible.

Happy Cubing and happy new year, U' f' L' S L S' L' S L F U!

 

[abunickabhi]

Finally an update video on the 5-style blindsolving method! (Last one was 2 years back, lot of new algs and features made)

 

[abunickabhi]

Made tons of changes to the 5-style mega-manual documentation I had been writing since 2018.

It explains the back story, my motivation, example solves, feasibility and future scope.

I also talk about the 3 new concepts in more depth and clarity, in the mega-manual, Letter Quads, Yo notation and UF5 algs.When I introduced this concepts in 2018, I was not super descriptive and precise about them. I hope I am more precise in describing these ideas now.

Mega Manual

 

[abunickabhi]

https://forms.gle/pdf7nvuAZi9keaXJ7

Feedback about the 5-style method

This form is intended to get the community's opinion on alg learning and big algsets like ZBLL, 1LLL and 5-style. Do these big algsets seem worth the effort? Is there enough returns that we get, on the amout of time we invest into these methods?

 

[abunickabhi]

Just a sneak peek on how I go about generating these algorithms.

 

[abunickabhi]

I am getting close to 105k letter quads on my sheet Took 1529 days to get here. Today is the 1530th day, I am developing 5style My speed has been 68 letter quads a day 2019 was the year, I did least contribution since I was very busy.

105k LQs is really good, my 2017 version of me is super proud now. Also 20k UF5 cases are good. Its good that there are good cubers and non-blders who point out if the alg is not good, or not fingertrickable, and then I try to improve on those batch of algs.