Understanding Rubik's cube I learnt how to solve a 3x3 Rubik's cube 10 years ago. Every now and then, I picked up a cube, scrambled it, and solved it for fun. I used to work on speed-solving, and memorised lots of formulae for it. However, now I'd like to go a different way: I'd like to solve cubes using the minimal amounts of formulae.
F2L (First Two Layers)
For experienced players, it is clear that with F2L-methods one can solve the first two layers without memorizing any formulae, simply by forming bottom cross, building the 'pillars', and installing the pillars.
Top Cross
I cannot see clearly why the formulae [R'U'F'UFR] (center->bar->L->cross) work, but I have an explanation for it: There are several ways to uninstall and reinstall the pillars that we have installed in the F2L method. By observing what the reinstallations do to the top face, at the end of the day one can write down those that help form the top cross.
Top Face
The formulae [RU'L'UR'U'L] are similar to the above, but it 'reinstalls' two pillars at the same time.
Last layer
This is what I really cannot see/understand. I don't even know how people came up with these formulae..
Questions


*

*Does anyone know how people came up with the formulae for the last layer?

*How to better see how the "reinstallations" work for the top face. I know this question is very vague; I am just giving a shot to see if someone has a good way to look at them.


Thank you!
 A: 
I'd like to solve cubes using the minimal amounts of formulae.

If you don't care about speed (but do care about keeping in your head what you're doing), you can reasonably get it down to five:


*

*Something to cycle three corners, such as $[F',UBU']=  F'UBU'FUB'U'$.
This, done on different sides in different orientations, lets you get the corner pieces in the right locations one by one.

*Something to cycle three edges, such as $F^2RL'U^2R'L$.
This plus conjugations lets you put edges in the right place one by one.

*Something to twist two corners, such as $[RUR'U'RUR',D]$.

*Something to flip two edges, such as $[R'LD^2RL'FDU'R,U]$.

*Something to correct if you find you need an odd permutation of the edges or corners so (1) and (2) does not suffice ... such as $U$.

If you don't even care about whether you can keep the administration in your head, you can in principle get down to two operations, and then you don't even need to ever reorient your cube between them: see here. But that's not really realistic as a method, so it becomes more a question how far you're willing to go in the name of fewer formulas, than of approaching a mathematical limit.
A: In this video, I show how to solve the entire cube using variations of "The formulae [RU'L'UR'U'L]" you mentioned.  (Except that we include the move U to complete the sequence.)  I also explain in the video how/why that sequence works.  There are a few instances when you need to conjugate (use premoves/setup moves) before a variation of The Niklas Commutator (the algorithm above --with the added U move at the end --can be rewritten as [R, U' L' U], which is a commutator.  It is called the Niklas commutator.) is used to solve some situations.  Then those premoves/setup moves are undone afterwards.
If you are interested in learning even more about the scope of commutators, see my post here.  More specifically, we can (in theory) solve every position with a single commutator + one move such as R or R2.  So if you are actually talking about a "formula" which can represent the form of a solution to any position, then there is the formula.  That is, [X,Y] + R or R2 can solve any 3x3x3 Rubik's Cube scramble.
But I am more inclined to think that you were asking about a move sequence.  Well the video above demonstrates that direct variations (and conjugations) of one eight move commutator is sufficient to tackle the cube.

As far as answering how people came up with original sequences, this old answer of mine is in agreement with Wood.  Furthermore, in the past, I went out of my way to decompose (and/or derive) well-known last layer speedcubing algorithms.  See the links I provide in this post.  This may "simulate" the type of logic which was used to find the algorithms.

And for what it's worth (expanding on the logic of how they found the algorithms), I don't know if you solved the 4x4x4 yet, but if you have, this guide was probably published in the 80's. One of the last figures on the last page shows the single edge flip case with this caption.
I'm not sure exactly when it was (no later than 1998) that this 25 90 degree rotation algorithm was found to fix that case directly.  It was most likely found by a computer, but here is my (human) derivation of it.  (Part 1, Part 2)
Well, I took the above challenge seriously (despite that I didn't see this particular guide before I did what I am about to state) and eventually found this 18 90 degree rotation algorithm by hand (in 2016).  So in essence, I used a similar logic to what I showed above for 3x3x3 algorithms as I did for finding the above 18 move algorithm.  So if there is a human logic base, besides accidentally finding it and/or using a computer, this should put up quite a case that my own logic process is the same that theirs was (since it is very fruitful and is not dependent on computer searches).  I mean, you can (I eventually did) conduct a computer search to find the 18 move algorithm I did, but it wasn't how I originally found that 18 move algorithm, for example.
