A Curious Position

[Mike Sherwin]

[fen]8/8/4p3/2k5/6p1/P5r1/2K5/R7 b - - 1 53 [/fen]
Blacks only winning move is Kd4. This position is from the example game from the RomiChess CPW page. The Romi version from 2006 did not find the winning move. Instead it played Kc4. The current version finds Kd4 but not till depth 25.

24 00:11 116,004k 10,185k +2.73 c5c4 a1e1 g3g2 c2d1 c4d5 e1e3 e6e5 d1e1 d5e6 e3e2 g2g3 e2a2 e5e4 a3a4 e4e3 a2h2 g3h3 h2g2 g4g3 a4a5 e6d5 a5a6 h3h1 e1e2 h1h2 g2h2
25 03:28 2,184,232k 10,464k +2.21 c5d4 a3a4 g3c3 c2d2 c3h3 a1g1 g4g3 a4a5 e6e5 d2c2 e5e4 a5a6 h3h2 c2b3 h2h6 g1d1 d4e5 d1a1

And the time goes from 00:11 to 03:28. And even SF15 does not see an overwhelming win for Kd4 right away.

So I'm wondering what is going on and how other engines handle this position? And what might be done to improve upon this?

Romi sacks a pawn on move 17 that SF15 thinks is okay and eventually gets a winning position.
[pgn][Event "ACCA 2006"]
[Site "ICC"]
[Date "2006.11.08"]
[Round "4"]
[White "Arasan"]
[Black "RomiChess"]
[Result "1/2-1/2"]

1.d4 Nf6 2.c4 e6 3.Nc3 Bb4 4.Qc2 O-O 5.a3 Bxc3+ 6.Qxc3 b6 7.Bg5 Bb7 8.e3 d6
9.Ne2 Nbd7 10.Qd3 c5 11.Nc3 Re8 12.Nb5 Qb8 13.dxc5 dxc5 14.Rd1 h6 15.Bh4 Ne5
16.Qd6 Ne4 17.Qxb8 Raxb8 18.Nxa7 Ng6 19.Bg3 Nxg3 20.hxg3 Red8 21.Rxd8+ Rxd8
22.Nb5 Ne5 23.Nc3 Kh7 24.f4 Ng4 25.e4 Kg8 26.Rg1 Nf6 27.e5 Ne4 28.Nxe4 Bxe4
29.Be2 Rd4 30.b4 Bd3 31.bxc5 bxc5 32.Kf2 Bxe2 33.Kxe2 Rxc4 34.Ra1 Ra4 35.Kd2
Kh7 36.Kc3 c4 37.Kc2 Kg6 38.g4 h5 39.Kc3 hxg4 40.g3 Kf5 41.Kd4 g5 42.fxg5 Kxg5
43.Ra2 f6 44.exf6 Kxf6 45.Ke4 Ke7 46.Kf4 Kd6 47.Ke4 Kc5 48.Ke3 Ra7 49.Ke4 c3
50.Kd3 Rf7 51.Kxc3 Rf3+ 52.Kc2 Rxg3 53.Ra1 Kc4 54.Kd2 Kd4 55.Ke2 Re3+ 56.Kf2
e5 57.a4 Rf3+ 58.Kg2 e4 59.a5 e3 60.a6 e2 61.a7 Rf8 62.Kg3 Ke3 63.Ra3+ Kd4
64.Ra4+ Ke3 65.Ra3+ Kd2 66.Ra2+ Ke3 67.Ra3+ 1/2-1/2[/pgn]

[ernest]

"And even SF15 does not see an overwhelming win for Kd4 right away."

Well, SF15 on my iPad, immediately eval -6.4 (depth 20)...

[Joost Buijs]

"And even SF15 does not see an overwhelming win for Kd4 right away."

Well, SF15 on my iPad, immediately eval -6.4 (depth 20)...

It probably is lack of an EGTB, with a 6P EGDB even the dumbest engines should find KD4 with a high score immediately.

[Mike Sherwin]

"And even SF15 does not see an overwhelming win for Kd4 right away."

Well, SF15 on my iPad, immediately eval -6.4 (depth 20)...

FEN: 8/8/4p3/2k5/6p1/P5r1/2K5/R7 b - - 1 53

Stockfish_15_x64_avx2:
NNUE evaluation using nn-6877cd24400e.nnue enabled
1/1 00:00 1k 1,205k +4.57 Kc5-c4
2/2 00:00 3k 2,569k +1.44 Rg3-e3 a3-a4
3/3 00:00 5k 4,878k +2.63 Kc5-c4 Ra1-e1 Rg3-g2+ Kc2-c1
4/4 00:00 9k 4,706k +2.12 Kc5-c4 Kc2-d2 Kc4-d4 Kd2-c2
5/5 00:00 11k 5,691k +2.87 Kc5-c4 Kc2-d2 Kc4-d4
6/7 00:00 27k 13,314k +3.29 Rg3-f3 Ra1-g1 g4-g3 Kc2-d1 Rf3xa3
7/10 00:00 52k 17,373k +1.74 Rg3-f3 a3-a4 g4-g3 Kc2-d2 Kc5-d4 a4-a5 g3-g2 Kd2-e2
8/11 00:00 132k 26,382k +1.42 Kc5-d4 a3-a4 Rg3-h3 a4-a5 g4-g3 a5-a6 Rh3-h2+ Kc2-b3

I don't use an EGTB. And even if I did it would not be the engine finding it. It takes to ply 8 for SF15 without an EGTB to determine that Kd4 is best but even then the score returned is not an overwhelming win. In fact SF15 initially prefers Kc4 over Kd4.
https://www.youtube.com/watch?v=6xijhqU8r2A

[Ajedrecista]

Hello:

I analysed the position with SF 15 (default net) with 2 cores, Multi-PV = 7 and the full set of 3, 4 and 5-man Syzygy EGTB:

Code: Select all

FEN: 8/8/4p3/2k5/6p1/P5r1/2K5/R7 b - - 1 53

Stockfish_15_x64_popcnt:
[...]
  7/8	00:01	      70.931	70.860	+0,07	Rg3e3 a3a4 Re3h3 a4a5 g4g3 a5a6 Rh3h8 a6a7
  7/9	00:01	      70.931	70.860	+0,09	Kc5b6 Kc2d2 Rg3b3 Ra1e1 Kb6a7 Kd2e2 Rb3xa3 Ke2f1
  7/9	00:01	      70.931	70.860	+0,64	Rg3h3 Ra1g1 g4g3 Kc2d3 g3g2+ Kd3e4
  7/8	00:01	      70.931	70.860	+0,65	e6e5 a3a4 Kc5c4 Kc2d2 Kc4d4 Kd2e2 Rg3h3 Ra1g1
  7/8	00:01	      70.931	70.860	+1,41	Rg3f3 a3a4 g4g3 Kc2d2 Kc5d4
  7/11	00:01	      70.931	70.860	+1,55	Kc5d4 a3a4 Rg3h3 a4a5 g4g3 a5a6 Rh3h2+ Kc2b3
  7/8	00:01	      70.931	70.860	+2,50	Kc5c4 Kc2d2 Kc4d4 Kd2c2 Rg3h3 a3a4 g4g3 a4a5
---------------------------------------------------------------------------
[...]
---------------------------------------------------------------------------
 31/28	21:15	1.571.470.345	1.231.924	 0,00	Rg3h3 Ra1g1 Kc5d4 Rg1xg4+ Kd4e3 Rg4g1 Rh3h6 Rg1e1+ Ke3d4 a3a4 e6e5 Re1d1+ Kd4e4 Rd1a1 Rh6h2+ Kc2c1 Rh2h1+ Kc1b2 Rh1xa1
 31/23	21:15	1.571.470.345	1.231.924	 0,00	Kc5d5 a3a4 Kd5e4 a4a5 Rg3g2+ Kc2c3 Rg2h2 a5a6 Rh2h8 a6a7 Rh8a8 Kc3d2 Ke4f3 Kd2e1 e6e5 Ra1a3+ Kf3g2 Ra3a2+ Kg2g3 Ra2a3+ Kg3h2 Ra3a2+
 31/27	21:15	1.571.470.345	1.231.924	 0,00	Rg3f3 a3a4 Kc5d4 a4a5 Rf3f2+ Kc2b3 g4g3 a5a6 Rf2f8 Ra1d1+ Kd4e3 Rd1e1+ Ke3f2 Re1xe6 g3g2 Re6g6 Rf8a8 Rg6f6+ Kf2e2 Rf6e6+ Ke2f3 Re6f6+ Kf3e4 Rf6g6 Ke4f3
 31/29	21:15	1.571.470.345	1.231.924	 0,00	Kc5c4 Kc2d2 e6e5 a3a4 Kc4d4 Kd2e2 Rg3h3 Ke2f2 e5e4 a4a5 e4e3+ Kf2g2 e3e2 a5a6 Rh3e3 a6a7 Re3e8 Kg2f2 g4g3+ Kf2f3 Kd4d3 Ra1a3+ Kd3c4 Ra3a1
 31/29	21:15	1.571.470.345	1.231.924	 0,00	e6e5 a3a4 Kc5c4 Kc2d2 Kc4d4 Kd2e2 Rg3h3 Ke2f2 e5e4 a4a5 e4e3+ Kf2g2 e3e2 a5a6 Rh3e3 a6a7 Re3e8 Kg2f2 g4g3+ Kf2f3 Kd4d3 Ra1a3+ Kd3c4 Ra3a1
 31/33	21:15	1.571.470.345	1.231.924	 0,00	Rg3e3 a3a4 g4g3 Kc2d2 Kc5d4 a4a5 g3g2 a5a6 Re3a3 Ra1xa3 g2g1Q Ra3a4+ Kd4c5 a6a7 Qg1g2+ Kd2e1 Qg2a8 Ke1f2 Kc5c6 Kf2e3 Kc6d6 Ke3e2 Kd6e7 Ke2f2 Ke7f7 Ra4a5 Kf7g6 Ra5a4
 31/42	21:15	1.571.470.345	1.231.924	+19,50	Kc5d4 a3a4 Rg3h3 a4a5 g4g3 a5a6 Rh3h2+ Kc2b3 Rh2h8 Ra1d1+ Kd4e4 Rd1e1+ Ke4f5 Kb3c4 Rh8a8 Kc4b5 e6e5 Kb5b6 Ra8g8 Re1g1 Kf5f4 Rg1f1+ Kf4g4 Rf1g1 Kg4f3 Rg1f1+ Kf3e2 Rf1a1 g3g2 Kb6c6 g2g1Q a6a7 Qg1xa1 Kc6d6 Qa1xa7 Kd6xe5

There were 15,064,343 TB hits up to depth 31. I want to point out that even with 5-man EGTB in a 7-man position, the drawing move 53. … ♚c4 is considered as winning and best move at depth 7 by SF 15.

Regards from Spain.

Ajedrecista.

[ernest]

...
It probably is lack of an EGTB,

Indeed no EGTB...