What's the longest checkmate that makes sense?

[kbhearn]

KBBvKN is also a general win, and probably also falls into the understandable category while sometimes exceeding 50 moves to win the knight with perfect play - however i also suspect that the defense in that ending is even more difficult than the attack (do we even have enough occurrences of it to go over with a tablebase and see which side was more error prone?) so the theoretically cursed wins might win more often than not anyways in human play.

Note that while KNNvKP is understandable, in general the attacking side has a very hard time even doing as well as is theoretically possible - so how much leeway do you want to give the human player to demonstrate the mate? does he deserve 120 moves to prove what should be mate in 40 just because it's not the worst possible KNNvKP? What about all those games that are theoretical draws against a stubborn opponent where you're making the defender stay alive for 70 more moves (think KRBvKR that routinely the attacker will play til close to the 50 move rule because it's hard to defend even though it's drawn).

But in your most recent post you said 'significant' portion of games - and that's just not going to be the case as it's far more common that the stronger side can keep its last pawn alive and avoid any worry about the 50 move rule. It might remove some desperation escapes whereby the side that's down a piece or two knights sacced its material deficit for the last pawns that were going to promote and hoped for 50 move rule salvation but overall it'd be rare that it has any impact.

fwiw, out of my 2M game db:
72 KNNvKP - 23 were won by white anyways (and 2 won by black)
64 KPvKNN - 28 were won by black anyways
and there are theoretically drawn positions in KNNvKP so some of those draws would still be draws even with a rule change.
40 KBBvKN - 18 were won by white anyways
19 KNvKBB - 10 were won by black anyways

so of 2Mish games maybe possibly 104 were 'unfairly' drawn by the 50 move rule (probably a good number of these would've been drawn anyways) if these were the only two endings deemed worthy of 'extra' moves. Seems like an awfully small gain to make the rule more complicated than just always 50 moves.

[Lyudmil Tsvetkov]

What about the 'Troitsky line' in NN vs. P endgame.

That would be a good example, if positions of that kind can be discovered in TBs, to the point where they begin to influence the result of a good percentage of games, it doesn't make much sense to impose the 50-move rue, at least at master level.

the longest tbs mate, as posted by Greg, is some 550 moves long.

most relevant tbs mates that engines can deliver are below 200 moves, and they happen relatively frequently.

for example, RRB vs RR, or RBN vs RN; I guess most humans will be able to deliever such mates, not very difficult.

so I think it really makes sense to extend the 50-moves rule to at least 200 moves for FIDE competitions.

however, they should definitely split it into 2: the general 50-moves rule applying for any game stage, avoiding usual fortresses, not to make the game too long, and a specific subrule concerning the very late endgame with no pawn on the board, where most of those tbs mates occur, which might extend to some 200 moves or so. In this way, the games will not last too long, but still the stronger side will, justifiably so, have the opportunity to win its won game (why not?).

does anyone know what was the previous FIDE attempt at extending the rule: did they split the rule into 2, or just tried to apply it uniformly across stages? what was the reason for the mess?

this is what concerns FIDE play. what concerns computer competitions, I do not see any downsides to quickly implementing such a rule, as

1)engines use tbs
2) engines do not get tired during the game

[Lyudmil Tsvetkov]

Delivering mate with 2 bishops vs knigth is also quite easy, such endings happen quite often, are statistically relevant, so I do not see a reason to deprive the stronger player of its chance to win the game.

[duncan]

KBBvKN is also a general win, and probably also falls into the understandable category while sometimes exceeding 50 moves to win the knight with perfect play - however i also suspect that the defense in that ending is even more difficult than the attack (do we even have enough occurrences of it to go over with a tablebase and see which side was more error prone?) so the theoretically cursed wins might win more often than not anyways in human play.

Note that while KNNvKP is understandable, in general the attacking side has a very hard time even doing as well as is theoretically possible - so how much leeway do you want to give the human player to demonstrate the mate? does he deserve 120 moves to prove what should be mate in 40 just because it's not the worst possible KNNvKP? What about all those games that are theoretical draws against a stubborn opponent where you're making the defender stay alive for 70 more moves (think KRBvKR that routinely the attacker will play til close to the 50 move rule because it's hard to defend even though it's drawn).

But in your most recent post you said 'significant' portion of games - and that's just not going to be the case as it's far more common that the stronger side can keep its last pawn alive and avoid any worry about the 50 move rule. It might remove some desperation escapes whereby the side that's down a piece or two knights sacced its material deficit for the last pawns that were going to promote and hoped for 50 move rule salvation but overall it'd be rare that it has any impact.

fwiw, out of my 2M game db:
72 KNNvKP - 23 were won by white anyways (and 2 won by black)
64 KPvKNN - 28 were won by black anyways
and there are theoretically drawn positions in KNNvKP so some of those draws would still be draws even with a rule change.
40 KBBvKN - 18 were won by white anyways
19 KNvKBB - 10 were won by black anyways

so of 2Mish games maybe possibly 104 were 'unfairly' drawn by the 50 move rule (probably a good number of these would've been drawn anyways) if these were the only two endings deemed worthy of 'extra' moves. Seems like an awfully small gain to make the rule more complicated than just always 50 moves.

lyudmil says RRB vs RR, or RBN vs RN is relatively frequent. how many do you have ?

[Ozymandias]

Then it become a mess because the computers find more. Every time they change the ruler. But nobody can understand the database move.

That would be a "no", then.

[Ozymandias]

But in your most recent post you said 'significant' portion of games - and that's just not going to be the case as it's far more common that the stronger side can keep its last pawn alive and avoid any worry about the 50 move rule. It might remove some desperation escapes whereby the side that's down a piece or two knights sacred its material deficit for the last pawns that were going to promote and hoped for 50 move rule salvation but overall it'd be rare that it has any impact.

The way I see it, if we have a rule for stalemate, we should have a extended 50-move rule for the endgames that need more moves to convert a win, as long as they affect the same (or more) number of overall games.

[Ozymandias]

a specific subtle concerning the very late endgame with no pawn on the board, where most of those tbs mates occur, which might extend to some 200 moves or so.

Instead of a fixed number of moves, I would ask the player to make progress, for example, if after 50 moves they're closer to mate by say, at least 40 moves, that should be enough to allow them to continue.

[kbhearn]

lyudmil says RRB vs RR, or RBN vs RN is relatively frequent. how many do you have ?

same (mostly human otb) 2M game database
14 KRRBvKRR/KRRvKRRB - 8 won anyways
14 KRBNvKRN/KRNvKRBN - 7 won anyways

so even less common than the earlier 2 mentioned. It's possible these things are more common in correspondence though - esp if just reaching them is an instant win there where a human otb game would make reasonable effort to avoid it and keep an own pawn alive not just for 50 move rule's sake but for simplicity's sake.

[Uri Blass]

What about the 'Troitsky line' in NN vs. P endgame.

That would be a good example, if positions of that kind can be discovered in TBs, to the point where they begin to influence the result of a good percentage of games, it doesn't make much sense to impose the 50-move rue, at least at master level.

the longest tbs mate, as posted by Greg, is some 550 moves long.

most relevant tbs mates that engines can deliver are below 200 moves, and they happen relatively frequently.

for example, RRB vs RR, or RBN vs RN; I guess most humans will be able to deliever such mates, not very difficult.

I think that you overestimate humans.
Most humans are one of the following:

1)Humans who do not know the rules of chess
2)Humans with minimal knowledge about chess so they know how the piece moves but unable to win even a simple KRK endgame(usually they do not play in tournaments).


Even if we talk only about GM's then I believe that most of them have no experience in RRB vs RR or RBN vs RN and my guess is that they are going to fail to win some mate in 70 against perfect defence in a position that they did not study.

The questions should be
1)what is the tablebase mate types that humans can practically learn in the meaning that they are able to win against perfect defence in all tablebase positions?
2)What is the longest mate in these tablebases?

I do not know if KRRB vs KRR or KRBN vs KRN are possible for humans.

I am sure KQK KRK KPK KBBK KBNK are possible for humans.
Harder but still possible for humans should be KQKR

[Uri Blass]

Delivering mate with 2 bishops vs knigth is also quite easy, such endings happen quite often, are statistically relevant, so I do not see a reason to deprive the stronger player of its chance to win the game.

They do not happen quite often and delivering mate is not easy.

I am not sure if there are humans who know to do it.
I believe that there are humans who can learn how to do it but I am even not sure that the problem is not too complex for humans.