Chess will never be solved, here's why

@tygxc

Positions ≠ nanoseconds.

You're basing your figures on Bob Hyatt's perft program which measures only the generation of new positions.

It is necessary to not just generate the positions but calculate leaf evaluations for the positions and manage the comparison of the evaluations and backing up to the root level. Hyatt (author of the Crafty engine) estimated at most 10% of the time was spent in generating new positions.

That should read " variant basic rules chess with no excess promotions". It seems more likely that @tygxc is proposing to consider only competition rules chess, in which case the number of positions without excess promotions would be HUGELY higher.

That's not taking account of the 3.8521... which has mysteriously disappeared altogether or the index 37 which has mysteriously shrunk to 36. Puts me in mind of one of my lecturers at university, who was said to be the only person who could make a term disappear by writing it fainter each line.

Welcome back Llama .

As for the FEN debate, it seems like Playerafar might not know how the infrastructure works. so let me make an attempt...

When you set up your position, presumably using the chess.com analysis board UI, you created a position *in that UI*.  The UI and the Stockfish engine (like almost all chess UIs and chess engines) are completely decoupled (because applications are much more modular and versatile this way).  The game you created was encoded as a FEN by the UI and sent to Stockfish for analysis.  If you cleared the board, the UI would clear the "can castle queenside" and "can castle kingside" flags (why would it not...castling is not possible?).  It is up to you while setting up the position to add these flags back in as needed.

Stockfish evaluated the FEN it was presented.

One could argue that the UI, in order to be more robust, should force each position's creator to manually set flags for castling rights and en passant, etc. at the point of submission and not allow said submission without the flags being explicitly chosen and set by the user, but...this would be a burden placed on every user that many users would find intolerable.  So, they shortcut it to the "most desired" behavior rather than setting it up so that it's is impossible to be ambiguous/uncertain of the user's intended position.

One thing the chess.com UI could do better is to present the position being submitted in a confirmation popup that clearly sets out the settings of the various flags, who is on move, etc. so that you could visually confirm quickly and easily that you are presenting for analysis what you think you are presenting for analysis .  As it sits, though, the responsibility is on you to make sure the flags are set correctly.

There *are* bugs.  This position above is one I created with castling rights both ways for white and black, and with an en passant square at g6, but the UI clears the flags and doesn't seem to retain the settings.  Nevertheless, those checkboxes are all there and that is how they are intended to be used, to allow the user to explicitly set each flag.

The above screenshot is from after hitting "clear the board" and and expanding the "Advanced" section.  Note that before clearing the board, all 4 castling rights were checked for the standard starting position.

Hey btickler. I can't quote or post diagrams yet, but I see you : )

---

Yeah, the way he described "FEN codes" it sounded like some arcane voodoo or something.

"If you used the level 9 modulated encryption frequency, sure, who knows how the program might react. Electronics in general are a magic black box that no one understand anyway, so  if you think about it, really anything is possible... which is why maybe it triggered the transverse analysis mode, which I just made up, but probably allows it to deduce whether castling is illegal. Also, shame on you for pretending to understand how FEN works when I'm obviously the one doing the heavy lifting in this discussion."

Meh, I'm a bit tired, but that's a semi decent parody I think.

Best to look at just Fool's Mate. You could probably manage that with a pencil and paper in five years.

Sounds ideal for theoretical physics.

I gather the decrement of the exponent was not just due to @tygxc's pen running a bit low on ink, but was actually explained in this post which involves a proof by conjecture.

Thanks. I'd missed that.

"As for the FEN debate, it seems like Playerafar might not know how the infrastructure works. so let me make an attempt..."
Regardless of how the 'infrastructure' works - and how the FEN works -
what the computer does is still determined by whether 'clear board' is invoked or not.
We could say -   'your modern gasoline car engine is always started by its electric starter motor' -  but you're still going to need the key.  
With the chip in it.  And you're still going to need to turn the key all the way to turn over the engine. 
Would we then hear ?
"Hey I know about electric starter engines !"  for about 100 posts ?
 
The obsession with the FEN code as substitute for whether the 'clear board' feature has been used or not -  well maybe that will last or grow.
Its very similiar to believing that a conversation about the diameter of a wavefront expanding at 2c - should instead be a conversation about 'riding the light beam'.  
That 2c is always going to be there.
As for the 'clear board' feature zapping castling rights - that might get fixed - or it might stay like that.  

@btickler probably did well here: (and in talking about 'bugs' in the system)
" It is up to you while setting up the position to add these flags back in as needed."
That's reasonable - provided its user-friendly.
And provided it doesn't display in a cumbersome or inappropriate way when the puzzle is displayed publically.  
A sudden 'castling is legal' note could or would give away the solution.

Whether programmers regard the 'system' as 'Stockfish' or as the UI plus the FEN/Pgn code plus Stockfish ....  that's semantics.
the user or customer wants to present the puzzle his/her way.
Whether accessing the feature from a number of sources -
like the analysis button at left of screen under the 5th button 'Learn'
or under the chessboard button at top left of the posting editors 
or the analysis button that might appear in posted chess diagrams - 
or when using 'Share' in the analysis feature in the 50,000+ tactics puzzles on the website. 
Are 'insider' views and lectures going to make such features more user-friendly?  No.
It doesn't take rocket science to figure that out. 
Whether the insiders are 'real insiders' like the Dev team members here on the website (and the owner of the site) - or who are 'professionals' elsewhere - but neither insiders here - nor are they professionals here on the Marketing team.  
That's right.  Chess.com has a marketing team too.  

And they should also fix the 'bug' that presents chess diagrams as 1/2-1/2 when its somebody to move and win - not draw. 

I think it's impossible to explain to people who are fundamentally chess players rather than mathematicians or computer scientists that you simply can't cut corners in proving a result (like solving chess). It's just the same as the situation with the 4-colour theorem, where there was a massive computation involved in the proof or, even more closely, the solving of 8x8 checkers.

"Interestingly at 1 nanosecond for each of 10^36 positions, it will only take you 3x10^19 years. That's 30 million million million years."
@Elroch and I in general agreement about that.
In one nanosecond the fastest computers can only do a few ops.
It might be very helpful if those well versed in programming describe their way what an 'op' is and what that translates to as far as 'solving' a position is concerned.
How many 'ops' would it take for a supercomputer to even start examining a position and determine what legal moves are available in it ?
Let alone 'solving' it many moves deep.  
I think you'd find it would take trillions of trillions of years ...
just to examine 10^36 positions.   
Let alone 'evaluate' - further let alone 'solve'.

Yes I could google it - and present internet text as to an 'op'  (single operation)  and all about 'petaflops' ...
but we've got experts here.  And professionals.  In computing.
The info could be presented objectively.  Without a 'doctrine' approach.

Extending the off topic discussion further (since we ain't going to solve chess), it is surprising how the state of the art chess engine has surprising inefficiency built in.

Here it finds the best moves to depth 20 and identifies that the two best moves for white are promoting a pawn - to either a queen or rook.  But even though the best line is to capture the promoted piece in both cases, it can't tell that the value must be the same. It evaluates them as 0.06 pawns different.

Doesn't it work by evaluating a tree of positions, but looking up new positions to see if they are in an efficient hash table, so that analysis of the position after the capture is provided to both lines?

If you are thinking it could be because the line with promoting to a queen is considered better because there are some sidelines where the queen doesn't get captured, and the evaluation is not merely from the single critical line, that sounds logical, but observing this analysis in real-time finds that promoting to a rook seems to sometimes have the better evaluation for white.

So the difference looks more like random noise in the evaluation because of the randomness in the selection of the subtree to analyse. Note that in the second example, the two critical lines already diverge at move 55, just 5 ply past the identical position, while the first example diverges at 56 black, 8 ply past the identical position.

I am not very knowledgeable about the algorithms used (I have heard the name of the relatively new search algorithm used by Stockfish but don't know what is special about it).

@MARattigan
"To be fair, it works the same in every GUI I've used as well as on the site which shall not be named, so most people will expect it to do what it does."

A professional programmer - using the 'UI' is very different from 20 million members using a 'UI'.  
Are you aware of the bug that presents the problem as '1/2-1/2' when its a win?
I could only get around that - by trial and error and forming a workaround.
Several people elsewhere agreed with me that the 'UI' or 'interface' for presenting diagrams on the site - isn't user-friendly.
But many people would disagree on what 'user friendly' means.
That would be as much of an issue for the Marketing Department as for the Dev team.
Lots of bugs on chess.com.  The alerts bugs have been going on long time now.

But this all got started here - because of the issue of supercomputers trying to examine positions to see if they could 'legally get there'.  
Then it was noted that Stockfish found castling illegal in the position I presented.
Various persons have now claimed that's 'expected'.  Apparently to finally divert away from the FEN discussion.
The reality isn't going to change that its the 'clear board' that kills the castling.  However much one wants to talk about FEN codes. 

@Elroch

I don't think your post is entirely off topic. I'll post further tomorrow.

Hey friend, @playerafar.

The "clear board" generates a FEN with no castling rights.

Then adding pieces keeps no castling rights.

Then Stockfish gets the FEN with no castling rights.

I suggest reading all about FEN - it's really not too arduous.

@playerafar

I agree the 1/2-1/2 is a pain - even when you find where to fix it. If the setup program doesn't know the result then it shouldn't display a result. That's just bad design.

Regarding post #1723 - 
with lines involving a rook or a queen with different evaluations -
I've seen Stockfish change its mind about evaluations - if you give it enough time to run.  Sometime radically change its mind.
From loss to win or vice versa.  
Could there be discrepancies regarding numerical evaluations - because the runtime on one of the lines has been different?  

There is an oddity that if there is a result in a sideline it gets displayed as the result of the game. I often come across this posting analysis diagrams for vote chess, where you discount a sideline with a different result and then have to go to the tab to remove that from being displayed as the result for the whole diagram!