techniques for analyzing a single endgame position

[nnnnnnnn]

What is the state-of-the-art in determining whether a particular endgame position is winning?

Let us say, the position has 8 non-pawn pieces and 9 pawns, so much too big for tablebases. However, I am willing to use a lot of CPU time, like weeks, to analyze it. (I already have the 7-piece tablebases).

I can let Stockfish 14 analyze it for a while. I can have engine matches. And I can just try to play out lines. But nothing is seeming really definitive. Scores hover around 1.5 for a while, basically.

[JVMerlino]

What is the state-of-the-art in determining whether a particular endgame position is winning?

Let us say, the position has 8 non-pawn pieces and 9 pawns, so much too big for tablebases. However, I am willing to use a lot of CPU time, like weeks, to analyze it. (I already have the 7-piece tablebases).

I can let Stockfish 14 analyze it for a while. I can have engine matches. And I can just try to play out lines. But nothing is seeming really definitive. Scores hover around 1.5 for a while, basically.

The only way to get a "definitive" answer, assuming that an engine hasn't shown a mate score, is to solve chess. But, more seriously, in complex positions, really all you can do is analyze with as many of the strongest engines you have, for as long as you're willing to do so, and extrapolate an "answer" from their results. Playing out lines can be helpful, but it can also lead you down an inferior branch, if the engine hasn't analyzed it for "long enough". So beware of that method unless you're willing to be extremely thorough.

[nnnnnnnn]

One feature that would be very useful would be a way to indicate that certain positions were known to be winning in the analysis of a particular position.

As an example, consider a position X one is trying to analyze. Suppose one infers that after 8 ply X leads to a position Y along a variation V, which wins for White. The analysis supporting Y as a known win might take days of computer time. Now we can backup one move along variation V from Y to get some position say Z, and analyze that, with the added information that Y is winning. This is similar to how lichess backs up in blundercheck analysis, but spread out over a much longer period of time with human intervention (and retaining only individual positions, not hashtables). Anyway, If Z is winning we can continue in this way, analyzing the positions that occur along the variation V from X to Y, and indicating to the computer that prior analysis already determined previously analyzed nodes on this variation path V are wins, until Y is reached.

[Ferdy]

What is the state-of-the-art in determining whether a particular endgame position is winning?

Let us say, the position has 8 non-pawn pieces and 9 pawns, so much too big for tablebases. However, I am willing to use a lot of CPU time, like weeks, to analyze it. (I already have the 7-piece tablebases).

I can let Stockfish 14 analyze it for a while. I can have engine matches. And I can just try to play out lines. But nothing is seeming really definitive. Scores hover around 1.5 for a while, basically.

This one is the best for analysis -- the IDea in Aquarium from chessok.

[towforce]

...The only way to get a "definitive" answer, assuming that an engine hasn't shown a mate score, is to solve chess. But, more seriously...

I seem to be the only person in the world who has looked at chess deeply and concluded that, per previous discussions in this forum, chess can be solved with today's computers.

To all of you, I look like a fantasist, but to me you all look like people saying, "to multiply something by three you have to do two additions, therefore it would be impossible to calculate something multiplied by three million by hand because it would take too long", which would actually make good sense until you discovered that there are ways to do big multiplications without doing repeated additions.

Less than four hundred years ago, some top mathematicians said that negative numbers made no sense: you cannot take more of something than there is there!

Until somebody has proven that chess cannot be solved without generating the entire game tree, it is wrong to speak as though it is proven IMO.