2

In an engine match Leela zero vs Komodo 14 Dragon at hyper-long time 6 hours each engine) in an Alekhine Defense game, Modern variation, the following rook endgame appeared on the board.

[FEN "4r1k1/pp3p2/6p1/2B3K1/2P4P/8/PP3P2/6R1 b - - 0 1"]

BLACK to move. According to KOMODO 14 Dragon internal analysis the position is + 1,99 at White at depth = 49 63.2M of tablebases. Meaning that we do not know if the above position is a win or a draw. Please, someone with a software powerful enough can go deeper? Is the position above, critical for the evaluation of the Alekhine opening, a win for White or a draw?

7
  • The position is : WHITE King on g5 Rook on g1 Bishop on c5 pawns on a2 b2 c4 f2 h4. BLACK King on g8 Rook on e8 pawns on a7 b7 f7 g6 BLACK to move. Stefano.
    – Stefano
    Aug 19 at 0:16
  • OK. You just had a K where there should have been an R in the FEN. I corrected it and made it into a diagram. Voting to reopen since it's now a legal position, although I'm not sure if we'll be able to do what Komodo couldn't in a 6 hour per side game.
    – D M
    Aug 19 at 2:56
  • 9
    This is not the first time an endgame position is posted with the mention "it is critical for the evaluation of the opening XXX". This is a misunderstanding of how openings (and chess) work : there are many ways to play the Alekhine defense without reaching this endgame, and even if the two very same supercomputer played an Alekhine against each other, they certainly wouldn't reach it.
    – Evargalo
    Aug 19 at 8:42
  • 1
    We cannot write answers right now. Rook endgames are tricky and often concidered drawish in spite of extra material. Here, the split white pawns on the kingside also favor the defense. The queenside majority seems to be strong enough for White to be winning but it is hard to make progress after the natural line 1...Re5 2.Kf4 Rc5 3.b3 b5 4.Rc1 Kf8 5.Ke4 bc 6.bc Ke7 7.Kd4 Kd6 and it seems tough to crack.
    – Evargalo
    Aug 19 at 9:03
  • @SmallChess isn't it only a bishop and a pawn, with the bishop about to fall to Re5?
    – Cleveland
    Aug 19 at 12:24
4

The Position: Win or Draw

It is as Draw! Which I am genuinely surprised about to be honest. My initial review of the board would say that with the following; White should be winning (which they are in the starting position);

  • 5 v 5 Pawns
  • 1 v 1 Rook
  • White has a Bishop
  • Whites King has been engaged in the offence.

However as black has the move initiative ... they are able to slowly equalize the position.

On the Topic of Table Bases

End Game TableBases now aren't necessarily as relevant as they use to be. Which is probably a good thing. Reasonable TableBases are rare, and more importantly they are impractical. If you have a table base of 5 pieces, if I have a table base of 6 piece end games...well, then there is no competition. Here is where it gets tricky. How do you reasonably provide that to engine,

The Syzygy 7 Piece TableBase is 18tb , containing 4.23e14 positions. A Syzygy 8 piece TableBase (which doesn't even exist and is what is technically required for this position) is something like 3.8e16 positions and would require probably 4.5PB of storage if not more.

Interesting Take Aways

  • Stockfish sees the position as a draw and equal (0.00) at move 20.

  • Stockfish more efficient at analyzing the position and reaching the position at a depth of of 35.

  • Looks like there is a new build of stockfish planned that appears to be slightly more efficient at reviewing the position.

  • List item

New Stockfish Version Confirmed???

The Results

[Event "StockFish 14 Analysis of Position Depth 52 @923753 kN/s"]
[SetUp "1"]
[FEN "4r1k1/pp3p2/6p1/2B3K1/2P4P/8/PP3P2/6R1 b - - 0 1"]
[PlyCount "66"]

1... Re5+ 2. Kf4 Rxc5 3. b3 Rf5+ 4. Ke3 Re5+ 5. Kf3 Rf5+ 6. Ke2 Re5+ 7. Kf1 Kg7 8. Rg3 a6 9. Re3 Rh5 10. Ke2 Rxh4 11. Re7 b5 12. Kd3 Kf6 13. Re1 Rh5 14. Kd4 Rh4+ 15. Kd5 bxc4 16. bxc4 Rh2 17. c5 Rxf2 18. a4 Rd2+ 19. Kc6 Ra2 20. Kb6 Rxa4 $11 21. c6 Rc4 22. c7 Rc2 23. Re8 g5 24. c8=R Rxc8 25. Rxc8 g4 26. Rf8 Kf5 27. Kxa6 g3 28. Rxf7+ Kg4 29. Kb5 g2 30. Rg7+ Kf3 31. Kc5 Kf2 32. Kc4 Kf1 33. Kc5 g1=Q+ 34. Rxg1+ Kxg1

Method

Hijacked the GPU Cluster from Work (again...see the footer as to why). Running StockFish14 as its a "neutral" engine in regards to the analysis / outcome of the game...in that it wasn't involved. It is also faster and more accurate at pruning positions that Komodo.

Also decided to run this through ChessBase to see what other folks had found.

I originally did the analysis on this but stuffed up and had as White to Move. Which was White had a forced mate sequence (Mate in 20 though)

2
  • On move 17 (White), try Re2 instead of c5. It seems White wins after both ... Rh1 and ...g5. The evaluation is at more than 6 ( 6.36) at d= 35/88 after a brief analysis ( 21 minutes and 32 seconds). It seems White wins.
    – user27314
    Sep 2 at 0:42
  • Honestly at the beginning I was surprised too.
    – user27314
    Sep 2 at 1:00

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.