Chess will never be solved, here's why

Regarding 'empiricism' and @tygxc with very fancy footwork to dance around it ...
He can always just quote 'experts' ...
"The van den Herik definition is generally accepted."
Chess might never be solved ...  correct.
But Dancing is never going to be a 'solved' game ...
that seems 'even more certain'?  Is that grammatically wrong?
Its not just chess that might never be 'solved'.
And would we want to 'solve' games of verbal footwork ?
Lawyers try to do that all the time.
And they often succeed. 
The juries or the judge do what the lawyer wants.

I wonder how @tygxc justifies this to himself ...
what are the internal sentences ?
"I know more mathematics than they do."
Or:
"I am quoting Dr. Herik and peer-reviewed articles".

Could the critical point come if he is asked to qualify or acknowledge as to when any old empirical data cannot justify a mathematical conclusion ?
Even saying 'solved' can be quite a claim.
This is more of a claim - its 'will/can be solved - just a matter of money'
but ...  becomes even more of a stretch ...
'will/can be solved because of this empirical data - plus Dr. Herik and other peer-review' 
(Reminds me of a name.  Mercola.  Dr. Mercola.  I don't know who Dr. Herik is and I don't want to know.   Mercola is arguably the biggest quack in all of history.)
But if Herik is a legitimate Doc - that still doesn't justify a claim of 'will be solved' based on empiricism and 'peer review'.  

@playerafar

Professor van den Herik in fact. Recently retired (see this).

Definitely not a quack.

Okay Martin. Another good post by you.
I'll change my mind and click on the Dr. Van den Herik link. 
Okay did that.  He definitely is 'Recognized' and would be 'the real thing'.
But does that mean that because @tygxc quotes him - then quotes from Dr. Van den Herik then become the be-all and the end-all of the discussion?
I think its ... No.
But its a well-known tactic to quote from authorities - similiar to the 'peer-review' tactic.  Even Mercola got 'peer review'.
Issue - will @tygxc be using quotes from  Dr. Van den Herik to try to obfuscate any empiricism @tygxc is using ? ...  

#2691
"You could even post a pseudocode, or even a code in a simple language (like Python, JavaScript) with some comments, that simulates what your algorithm is supposed to"
++ I do not intend to write any software, I intend to use already written software like Stockfish running on already existing cloud engines to weakly solve chess.

"Especially, it would be nice if, running it, we could see how many nodes are expected to be in the search tree, on average, every 17 s."
++ One cloud engine in 17 s adds 4 nodes to the solution tree and 17 billion nodes to the search tree. Thus in 5 years 3 cloud engines add 4*10^17 nodes to the search tree and
3 engines * 5 a * 365.25 d/a * 24 h/d * 3600 s/h * 4 nodes / 17 s = 10^8 nodes to the solution tree.
This 10^17 nodes in the search tree and 10^8 nodes in the solution tree is commensurate with the 10^14 nodes in the search tree and 10^7 nodes in the solution tree for checkers.

#2652

""any opposition" means against any possible move by the opponent (not only the supposedly best ones), so in a game-theoretic sense you cannot skip 1. e4 e5 2. Ba6"
++ Per Merriam-Webster Dictionary: "oppose = to place opposite or against something (so as to provide resistance or contrast) OR to strive against : RESIST"
Neither 1 a4 nor 1 e4 e5 2 Ba6 oppose i.e. resist against the draw.

"Next to brute-force methods it is often beneficial to incorporate knowledge-based
methods in game-solving programs. Their main advantage is providing an appropriate
move ordering or selection in the search trees." - Prof. em. J. van den Herik
So it is allowed to incorporate knowledge about chess for move ordering and selection.

For that reason 1 a4 and 1 e4 e5 2 Ba6 do not belong to "any opposition" and can be skipped i.e. not selected because ordered lower.

"more about how to create in 5 years a supposedly unbeatable chess engine, rather than providing a proof tree of a weak solution."
++ No, I do not intend to create any engine, I intend to use cloud engines, i.e. existing software like Stockfish running on existing hardware, to weakly solve chess.

great!that is good to know!

One cloud engine in any amount of time gives  you 4 candidate moves which could possibly be added to the solution. That is assuming you are using an engine with MultiPV=4 (which would reduce its depth of search).

Three cloud engines will give you 12. Of those 12 candidates at most three would be added to the solution (normally 0).

The solution tree for checkers contains 10¹⁴ nodes from the 24 man starting position to the 10 man tablebases according to Schaeffer in the paper you keep quoting, not 10⁷. (You have a wonderful facility for jettisoning 99.99999% or more of your problem overboard as and when required.)

You also assume that:  (a) chess is a draw from the starting position, and (b) a solution of chess from the 32 man starting position to the 7 man tablebases will contain no more positions than a solution of checkers from the 24 man starting position to the 10 man tablebases.

If (a) is true then (b) is, to say the least, questionable.

Nevertheless it means that if (a) is true, then at 1s per move there is a lower bound of 10¹⁴ /3 secs = 1,056,993 years. on the time you will take to complete your solution. That is, of course, an absurdly conservative lower bound. (At your 17s/move, revised down from 60h/move, you could postulate instead an equally absurdly low lower bound of around 18 million years.)

The story is altogether different if you drop (a). 

If, for example you could prove that the mate 1.f3 e5 2.g4 Qh4 was a shortest possible forced mate from the starting position, you would need only about 600 nodes in your solution.

You would probably struggle to prove that the mate I gave is forced, but you may feel a different mate in say 16 moves might be forcible. If so finding that in 5 years could be feasible - but only if it's really there.

Please, could you stop repeating the same unsolicited things? It's just spam. Cannot you use links, if anything?
Did you understand that we never said that heuristics are not allowed for move ordering and selection, in the search for a solution to chess? Quite the opposite actually. The issue is about the other special way you think you can use them, but I will not talk about that now.

What? I thought you had conceded that in fact the size of the proof tree in checkers is not 10⁷. The size of the stored proof tree is 10⁷. The upper bound for the proof tree is 10¹⁴, the number of positions searched. Where in the paper do you read that the stored tree is in fact the entire proof tree (see here, after the yellow line)? With your criterion, since you say that the optimal
moves are among those 4, the proof tree in chess would have an upper bound of 10⁸ nodes. Do you understand that it is implausible, to use an euphemism, that the entire proof tree in chekers is bigger than the entire proof tree in chess, don't you?

#2703

"One cloud engine in any amount of time gives  you 4 candidate moves which could possibly be added to the solution." ++ The point of the 17 s/move (revised from 60 h/move) is to ascertain that the optimal move is within the top 4 engine moves with 1 miss per 10^20 positions.

"assuming you are using an engine with MultiPV=4 (which would reduce its depth of search)."
++ No, the depth of search is crucial. The heuristic is best first. The 4 candidates are generated simultaneously. First the trunk is takes e.g. from an ICCF WC draw game. Then in a first verification pass all white moves starting from the last and working back to the first after a predecessor are replaced by the 2nd choice. Then a 2nd verification with 3rd choice. Then a 3rd verification pass with the 4th choice.

"Three cloud engines will give you 12." ++ No, the 3 cloud engines are meant to work in parallel. One for 1 e4, one for 1 d4, one for 1 c4 / 1 Nf3.

"The solution tree for checkers contains 10¹⁴ nodes from the 24 man starting position to the 10 man tablebases according to Schaeffer in the paper you keep quoting, not 10⁷."
++ I quote from the paper
"there are 10^7 positions in the stored proof tree, each representing a search of 10^7 positions"
https://www.researchgate.net/publication/231216842_Checkers_Is_Solved

"You also assume that:  (a) chess is a draw from the starting position"
++ Yes, that is right.

"(b) a solution of chess from the 32 man starting position to the 7 man tablebases will contain no more positions than a solution of checkers from the 24 man starting position to the 7 man tablebases." ++ I would start from 26-men tabiya and checkers ends at a 10-men database. The checkers search tree had 10^14 nodes, for chess I reckon with 10^17 nodes.

"If (a) is true then (b) is, to say the least, questionable."
++ Yes, you can question the number of legal, sensible, reachable, and relevant positions. This is the key question. Legal we know 10^44 from Tromp. Sensible 10^38 - 10^32 from Gourion with Tromp's conjecture. Reachable estimated 10^18. Relevant estimated 10^17. We will only know for sure after it is done if even for 1 tabiya of 1 ECO code e.g. C67. Before he had done it, Schaeffer calculated with 10^9 positions for checkers, see his earlier paper. So he did not know it before either.

"at 1s per move there is a lower bound of 10¹⁴ /3 secs = 1,056,993 years on the time you will take to complete your solution."
++ In 1,056,993 years 3 cloud engines of 10^9 nodes/s consider
3 engines * 1056993 a * 365.25 d/a * 24 h/d * 3600 s/h * 10^9 nodes/s/engine = 10^23 nodes.
I cannot rule out there might be 10^23 legal, sensible, reachable, and relevant positions.

"The story is altogether different if you drop (a)."
++ I consider the game-theoretic value of chess a draw.
It is the general consensus of experts: Steinitz, Lasker, Capablanca, Spassky, Fischer, Adorjan.
AlphaZero autoplay converges to all draws with increasing time, even if stalemate is a win.
TCEC had to impose slightly imbalanced openings to avoid all draws.
ICCF WC converges to all draws.

"you may feel a different mate in say 16 moves might be forcible.
If so finding that in 5 years could be feasible."
++ Proving a draw is harder than proving a win.
Checkers is an easier game than Chess or Losing Chess, but was harder to solve than Losing Chess because Checkers is a draw and losing Chess a win.

No, your stated plans do not involve "weakly solving" chess (PROVING its value). They involve increasing your confidence in what the value is, and then confusing increased confidence with certainty.

Your plans do not even involve generating a strategy to achieve the optimal result (whatever it might be) against 1. a4 (which may be an optimal first move for white. Or may not). They don't even involve any improvement in confidence of the result with optimal play after 1. a4: they rely on the reasoning "Imperfect players believe the often played moves are at least as good as the very rarely played and little analysed move 1. a4, so they are".

#2704

"could you stop repeating the same unsolicited things?"
++ It is relevant to my point that 1 a4 and 1 e4 e5 2 Ba6 may be skipped

"Cannot you use links, if anything?"
++ Some people are reticent or too lazy to open links. I quote for clarity.

"we never said that heuristics are not allowed for move ordering and selection, in the search for a solution to chess? Quite the opposite actually."
++ Some people keep insisting that I am not allowed to select and must try all moves.

"I thought you had conceded that in fact the size of the proof tree in checkers is not 10⁷."
"there are 10^7 positions in the stored proof tree, each representing a search of 10^7 positions"
https://www.researchgate.net/publication/231216842_Checkers_Is_Solved 
So for chechers a proof tree on 10^7 and a search tree of 10^14.
For chess I project a proof tree of 10^8 and a search tree of 10^17.
"the entire proof tree in chekers is bigger than the entire proof tree in chess"
Checkers 10^7 proof tree < chess 10^8 proof tree,
checkers 10^14 search tree < chess 10^17 search tree

#2706

"No, your stated plans do not involve "weakly solving" chess (PROVING its value)."
++ Yes, my plans involve weakly solving chess i.e. proving,
just like done for Checkers and Losing Chess.

"1. a4 (which may be an optimal first move for white. Or may not)."
++ I use knowledge about chess to select the moves 1 e4, 1 d4, 1 c4, 1 Nf3 to be superior compared to 1 a4, and thus to select the moves accordingly. 1 e4, 1 d4, 1 c4, 1 Nf3 oppose to the draw, 1 a4 does not.

"they rely on the reasoning "Imperfect players believe the often played moves are at least as good as the very rarely played and little analysed move 1. a4, so they are"."
++ No, it is knowledge about the game.
"Next to brute-force methods it is often beneficial to incorporate knowledge-based
methods" - Prof. em. van den Herik.
I am allowed to incorporate chess knowledge.

Lie #7. The paper clearly states that the stored proof tree is not the entire proof tree, and you deliberately ignore that.

#2709
Quoted from the paper:

"there are 10^7 positions in the stored proof tree, each representing a search of 10^7 positions"
"Hence, 10^14 is a good ballpark estimate of the forward search effort"
Thus search tree for checkers = 10^14.
Search tree for chess projected = 10^17.
I quote from the paper.
Can you quote from the paper anything contradictory?

I have already done it here. Follow the link in that post. The paper contradicts that the entire proof tree is 10⁷ nodes, it's down black and white.

Obviously ignoring definitions in common dictionaries, that do not nor can include all the specific and technical uses of some words, I think that the following is a very good example of how "any opposition" is considered in our context. In fact, the authors use just van den Herik's (cited in the references) definition of weakly solved, explaining later its meaning:

  "Weakly solved means a strategy [ . . . ] has been determined for the initial position
to achieve the game-theoretic value against any opposition. [ . . . ]
   In terms of prunings in branch and bound tree search, a sound pruning condition when solving a weaker solution concept may not hold in stronger ones. This is caused by the removal of the assumption of optimal/perfect plays when dealing with stronger solution concepts. For example in alpha-beta prunings, when the min player obtains an A-costs which is lower than the lb (i.e. max player's last found best), we cannot immediately backtrack if we want to tackle weakly solved solutions, where we assume the max player is the adversary. The reason behind is that we cannot assume the max player must play a perfect move. We have to consider all moves for the max player. The situation is similar if we assume the min player is the adversary." (emphasis mine)

https://www.researchgate.net/publication/262296151_Consistencies_for_Ultra-Weak_Solutions_in_Minimax_Weighted_CSPs_Using_the_Duality_Principle

@MARattigan, I can understand your point about using that definition (e. g. a perfect strategy to gain... a loss?). There is no problem if we consider that if the game is, let's say, a win for White, it has to be proven that White can actually win (the strategy must not work only for the losing side). But indeed, other ways to state the concept may be clearer.