Chess will never be solved, here's why

@tygxc, you refer to pruning one branch based on a human evaluation by a weak chess player. This is of course of very little help to solving chess since there are quadrillions (underestimate) of positions that have never been seen before and are of a similar ilk. You wish to prune all of these.

Clearly you must be relying on a computer evaluation of these positions. It would be very bizarre if you were to suggest that you are going to treat positions with the same evaluation differently depending on other factors - I think you accept you are not as good at chess as Stockfish - so we can assume that what you are suggesting is to ignore all positions with an evaluation less than a certain level. i.e. to assume that a -3 zero ply evaluation is ALWAYS a loss.

There's the problem. That is definitely false. There are undoubtedly billions (underestimate) of positions with a zero ply evaluation like that of a blundered piece that are not lost. We very rarely see just how bad a zero play evaluation is, but everyone has seen examples of Stockfish getting the wrong evaluation of difficult positions after already searching millions of nodes to tens of ply!

So now that you know that your approach is definitely invalid, @tygxc, what are you going to do about it?

@12229

"Inductive reasoning / intuition does not eliminate 10^__ # of positions."
Perfect alpha-Beta pruning reduces the number of positions to the square root.
Imperfect alpha-beta pruning somewhat less: exponent 0.67 for checkers.
That is indeed a reduction from 10^34 to 10^17, i.e. a reduction of 10^17.
For example these are 2 legal and sensible positions, both can result from optimal play by both sides, but a weak solution to chess needs only one, not 2.

Excluding the first position also excludes all positions with a black pawn on e7.
Excluding the second position excludes all positions with a black pawn on c7.
All positions with a white pawn on e2 are no longer accessible either.

There are 10^44 legal positions, but only 10^38 without underpromotions to pieces not previously captured. That is 10^6 reduction while nobody promotes to a 3rd bishop, rook, or knight in a perfect game.

Of the 10^38 only 10^34 to 10^32 can result from optimal play by both sides.
That is a reduction of 10^4 to 10^6.
It excludes positions like this one:

It is a legal position and it does not contain underpromotions to pieces not previously captured, but cannot result from optimal play by both sides. Convince yourself and try to come up with a proof game with reasonable play. You do not even have to prove it is optimal play.

Now explain how you are going to deal with the fact that your method does not work, as explained in my last post.

@12242

"pruning one branch based on a human evaluation by a weak chess player"
++ ICCF World Championship Finalists are no weak chess players, especially if they have average 5 days per move and use computers.

"you must be relying on a computer evaluation of these positions"
++ No. The ICCF WC finalists rely on their chess culture and test ideas with their engines.

"positions with a zero ply evaluation like that of a blundered piece that are not lost"
++ There is a difference between losing and sacrificing material.
Losing material is trading material for nothing in return and loses the game.
Sacrificing material is trading material in return for some kind of positional advantage.
The sacrifice may or may not be correct and the way to decide is to calculate.
However 1 e4 e5 2 Ba6? is loss of material for no return and is losing the game, not a single ply of calculation needed and not even an ICCF (grand)master needed, nor 5 days needed.
It is obvious.

"Stockfish getting the wrong evaluation of difficult positions"
++ All Stockfish evaluations are wrong: there are only 3 possible objective, absolute evaluations: win/draw/loss. All those +0.33 etc. are approximations.
However, the objective, absolute evaluation of 1 e4 e5 2 Ba6? is a black win.

Also, 1 e4 has a higher evaluation than 1 a4, regardless of engine, engine tuning, or thinking time. However, by pure logic it is unthinkable that 1 e4 would be a draw and 1 a4 would be a win. That does not need 1 ply of calculation, only logic.

@12245

"it should take a strong player less than a day to produce an example game"
++ I have played many tournaments. Once in a while the arbiter has to intervene to supply a second queen when a player wants to queen a pawn and his queen has not yet been captured. Luxury chess sets have spare queens for that.
Sometimes underpromotions to knights happen, because of the unique properties of the knight, but when that happens, at least one knight is generally captured. I never saw an arbiter intervene to supply a 3rd knight.

Underpromotions to rook or even bishop happen but only to avoid stalemate, i.e. only by the winning side. Even in the rare cases where these happen usually at least one rook/bishop is already captured. I never saw an arbiter intervene to supply a 3rd rook or bishop.

Apart from that, the vast majority of the 10^44 legal positions has multiple underpromotions from both sides.
https://github.com/tromp/ChessPositionRanking 
First position: 7 white rooks, 3 black bishops
Second position: 4 white rooks, 3 black bishops
Third position: 4 white rooks, 3 black rooks

Those 3 randomly sampled positions thus can never result from optimal play by both sides: it makes no sense for both sides to avoid drawing by stalemate by underpromotion to R/B.

"you need to promote with check, for example" ++ The only example I know is the Lasker trap, but it results from non-optimal play by both sides.
1 d4 d5 2 c4 e5? 3 dxe5 d4 4 e3? Bb4+ 5 Bd2 exd3 6 Bxb4 exf2+ 7 Ke2 fxg1=N+

@12249

"Some 0.00 positions are decisive, and some positions with high evals are draws."
++ That is what I say the whole time. The provisional, heuristic engine evaluations like +0.33 play no role, only the final objective evaluation win/draw/loss reaching the 7-men endgame table base, or a prior 3-fold repetition, or a known drawn endgame as judged that the two ICCF World Championship Finalists after days/weeks/months of consideration.

"why engines can be trusted" ++ On the contrary: engine evaluations cannot be trusted unless they reach the 7-men endgame table base or a prior 3-fold repetition.
The way to play ICCF correspondence is exactly to aim for positions engines misjudge.

@12250

"So have I." ++ And have you ever witnessed a player requesting a 3rd rook/bishop/knight?

"this is not appropriate in a solution to chess"
++ But it is appropriate to estimate the number of positions needed to weakly solve chess.
If the Laws of Chess were changed so that you can only underpromote to a piece previously captured, then all games with optimal play from both sides would stay exactly the same.

@12252

"calculate all positions from the beginning of the game to 7-man EGTB"
++ No. Prune when possible, calculate when needed.
There is no need at all to calculate 1 e4 e5 2 Ba6? or 1 a4. That is waste of computer time.

"this would be a 32 man EGTB"
++ No, a weak solution needs only 1 black reply to the opposing white moves.
This is the major reduction to the square root of the number of positions (or power 0.67 in the case of Checkers). Schaeffer calculated to his Checkers endgame table base, but did not generate a strong solution with a 24 men table base.

we've been telling tygxc that for years bro, he just wont listen.

btw llama feel free to remind tygxc that ive brought his "arguments" to over a dozen math majors/math professors and they all found the same errors that we did.

ive gotten so used to tygxc's style of fallacies that he wont even bother to address my comments.

@12254

"You should be able to come up with examples to the contrary"
++ No. Please show me one real, serious game with 3 rooks / bishops / knights on one side.

this is a very bad suggestion for tygxc as he is notorious for completely butchering the context and meanings of quotes.

again not how proof works. its actually plenty possible to prove for the heuristic a couple of your claims, but you are so devoid of logical understanding that you cant write them.

hey tygxc remember when you claimed that a lack of counterexample constituted a formal proof?

remember when you tried to claim that "chess errors follow a poisson distribution" when chess games literally violate most of the axioms necessary?

remember how your computing time "calculation" misconstrued nodes as full positional calculations, making your end result off by a factor of over a million?

remember when you tried to claim that whilte could strategy steal any black position when it was already formally proven to be impossible? you tried to choose blacks moves when by definition such an argument must apply to any black response.

i could go on and on.

"However, the objective, absolute evaluation of 1 e4 e5 2 Ba6? is a black win."

its not an objective evaluation until you can prove that it's mate in a certain # of moves.

that's what it is by definition.

"Also, 1 e4 has a higher evaluation than 1 a4, regardless of engine, engine tuning, or thinking time""On the contrary: engine evaluations cannot be trusted"

obvious contradiction lmfao. you also have no proof that e4 is better. only conventional knowledge dictates that e4 is better.

"but a weak solution to chess needs only one, not 2"

actually you need more than that for a weak solution, as you need to prove that a draw isnt the outcome for best play on both sides. just because you have a game tree that proves that white can always achieve a draw doesnt mean there isnt a winning game tree for white.

BTW Llama man, don't let tygxc's illogic fool you into thinking that it cannot be sufficiently demonstrated that we can cut down the vast majority of underpromotions from a calculation heuristic for a weak solution. (a weak solution is just a proof of the value of the starting position along side a strategy for one of the players to guarantee that value).

bro i think llama guy got so pissed at tygxc's logic illiteracy that he cussed him out and got muted.