Stockfish Syzygy Endgame

[Andre]

[d]8/1k6/8/Q7/7p/6p1/6pr/6Kb w[/d]

Strangely in the above diagram, Stockfish Syzygy cannot found the winning move Qc5 (only considers the draw with Qb6+) when the regular Stockfish finds the good move in a few seconds.
Any reason for this behavior?

[Damir]

there is no winning move. The position is draw. You are just dreaming.

[jd1]

there is no winning move. The position is draw. You are just dreaming.

I'm not 3000+ like Stockfish , but isn't this a winning strategy:
- Force Black to move something other than the king (this is done by cornering the king, e.g. 1. Qc5 Kb8 2. Qc6 Ka7 3. Qb5 Ka8 4. Qb6).
- Then Black has only two legal moves:
- if Rh3 then the queen checks until it can take the rook, e.g. 5. Qc6+ Kb8 6. Qd6+ Ka8 7. Qd8+, 8. Qd7+ and 9. Qxh3. This is forced for all Black responses.
- if Black plays h3, the White queen manoveuers until it checks the Black King and takes the g3 pawn, trapping the h2 rook.

Jerry

[jpqy]

Qc5! is winning move..indeed Stockfish with RTB gives draw ,Qc7?? Kxc7 draw..but when you go one move back ,SF RTB see directly Mat!

1.Qc5! Ka6 2.Qb4 Rh3 [2...Ka7 3.Qb5 Ka8 4.Qb6 Rh3 5.Qd8+ Ka7 6.Qd7+ Kb6 7.Qxh3 ]

3.Qd6+ Ka5 4.Qd2+ Ka4 [4...Ka6 5.Qd3+ Ka5 6.Qf5+ Kb4 7.Qxh3 Kc3 8.Qxh4 Kb2 9.Qxg3 Kc2 10.Qe3 Kb2 11.Kf2 Kc2 12.Ke2 g1Q 13.Qxg1 Bb7 14.Qc5+ Kb3 15.Kd3 Ba6+ 16.Kd4 Bf1 17.Qc3+ Ka2 18.Qc2+ Ka3 19.Kc3 Bc4 20.Qb2+ Ka4 21.Qb4# ]

5.Qd7+ Kb4 6.Qxh3 Kb5 7.Qxh4 Ka6 8.Qxg3 * same leads to Mat.

JP.

[Damir]

The problem is Black can only shufle the king. All the other moves are losing.

[jpqy]

Yes..but black has to move his rook when king can't move anymore..and then after a few checks more queen can take rook on h3! game over..

[Uri Blass]

[d]8/1k6/8/Q7/7p/6p1/6pr/6Kb w[/d]

Strangely in the above diagram, Stockfish Syzygy cannot found the winning move Qc5 (only considers the draw with Qb6+) when the regular Stockfish finds the good move in a few seconds.
Any reason for this behavior?

Maybe a bug in implementing tablebase code.
I do not have syzygy but my stockfish has no problem to find Qc5 in a few seconds using 3 cores.


[D]8/1k6/8/Q7/7p/6p1/6pr/6Kb w - - 0 1

Stockfish_13110108_x64_modern_sse42:
1/1 00:00 116 116,000 0.00 Qa5-b6+ Kb7xb6
2/2 00:00 396 198,000 0.00 Qa5-b6+ Kb7xb6
3/3 00:00 964 482,000 0.00 Qa5-b6+ Kb7xb6
4/3 00:00 1,029 514,500 0.00 Qa5-b6+ Kb7xb6
5/3 00:00 1,172 586,000 0.00 Qa5-b6+ Kb7xb6
6/3 00:00 1,647 823,500 0.00 Qa5-b6+ Kb7xb6
7/3 00:00 2,474 1,237,000 0.00 Qa5-b6+ Kb7xb6
8/3 00:00 3,697 1,232,333 0.00 Qa5-b6+ Kb7xb6
9/3 00:00 5,334 1,778,000 0.00 Qa5-b6+ Kb7xb6
10/3 00:00 9,191 1,838,200 0.00 Qa5-b6+ Kb7xb6
11/3 00:00 12,446 2,074,333 0.00 Qa5-b6+ Kb7xb6
12/3 00:00 15,050 2,150,000 0.00 Qa5-b6+ Kb7xb6
13/3 00:00 18,727 1,560,583 0.00 Qa5-b6+ Kb7xb6
14/3 00:00 23,999 1,714,214 0.00 Qa5-b6+ Kb7xb6
15/3 00:00 30,416 1,901,000 0.00 Qa5-b6+ Kb7xb6
16/3 00:00 38,571 2,268,882 0.00 Qa5-b6+ Kb7xb6
17/3 00:00 40,866 2,270,333 0.00 Qa5-b6+ Kb7xb6
18/3 00:00 48,850 2,442,500 0.00 Qa5-b6+ Kb7xb6
19/3 00:00 56,475 2,567,045 0.00 Qa5-b6+ Kb7xb6
20/3 00:00 66,347 2,764,458 0.00 Qa5-b6+ Kb7xb6
21/3 00:00 77,979 2,888,111 0.00 Qa5-b6+ Kb7xb6
22/3 00:00 91,797 2,961,193 0.00 Qa5-b6+ Kb7xb6
23/3 00:00 108,900 3,111,428 0.00 Qa5-b6+ Kb7xb6
24/3 00:00 124,760 3,119,000 0.00 Qa5-b6+ Kb7xb6
25/3 00:00 160,015 2,963,240 0.00 Qa5-b6+ Kb7xb6
26/3 00:00 181,354 2,973,016 0.00 Qa5-b6+ Kb7xb6
27/3 00:00 208,942 2,862,219 0.00 Qa5-b6+ Kb7xb6
28/3 00:00 220,598 2,828,179 0.00 Qa5-b6+ Kb7xb6
29/3 00:00 256,138 2,613,653 0.00 Qa5-b6+ Kb7xb6
30/3 00:00 307,862 2,587,075 0.00 Qa5-b6+ Kb7xb6
31/3 00:00 335,183 2,618,617 0.00 Qa5-b6+ Kb7xb6
32/3 00:00 363,369 2,732,097 0.00 Qa5-b6+ Kb7xb6
33/3 00:00 402,130 2,773,310 0.00 Qa5-b6+ Kb7xb6
34/3 00:00 434,533 2,785,467 0.00 Qa5-b6+ Kb7xb6
35/3 00:00 510,368 2,883,435 0.00 Qa5-b6+ Kb7xb6
36/3 00:00 552,547 2,776,618 0.00 Qa5-b6+ Kb7xb6
37/3 00:00 589,536 2,834,307 0.00 Qa5-b6+ Kb7xb6
38/3 00:00 638,811 2,575,850 0.00 Qa5-b6+ Kb7xb6
39/3 00:00 662,012 2,488,766 0.00 Qa5-b6+ Kb7xb6
40/3 00:00 725,888 2,452,324 0.00 Qa5-b6+ Kb7xb6
41/3 00:00 810,658 2,557,280 0.00 Qa5-b6+ Kb7xb6
42/63+ 00:03 14,333,835 4,672,045 +9.35 Qa5-c5 Kb7-a6 Qc5-b4 Rh2-h3 Qb4-d6+ Ka6-a5 Qd6-d2+ Ka5-a6 Qd2-h6+ Ka6-a7 Qh6-g7+ Ka7-a8 Qg7-e5 Ka8-b7 Qe5-e7+ Kb7-b6 Qe7-e6+ Kb6-c5 Qe6-f5+ Kc5-d4 Qf5xh3 Kd4-e3 Qh3xh4 Ke3-f3 Qh4-d4 Kf3-e2 Qd4-c3 Ke2-d1 Qc3-d3+ Kd1-c1 Qd3-d7 Kc1-b2 Qd7-d6 Kb2-c2 Qd6-d4 Kc2-c1 Qd4-e3+ Kc1-b2 Qe3-g5 Kb2-b3 Qg5-f5 Kb3-b2 Qf5-f6+ Kb2-b1 Qf6-d6 Kb1-c2

I can also get this move with 1 core at smaller depth and I suspect stockfish has bugs in using many cores efficiently and for some reason with many cores it can see less at the same depth.

[D]8/1k6/8/Q7/7p/6p1/6pr/6Kb w - - 0 1

Stockfish_13110108_x64_modern_sse42:
1/1 00:00 116 58,000 0.00 Qa5-b6+ Kb7xb6
2/2 00:00 396 198,000 0.00 Qa5-b6+ Kb7xb6
3/3 00:00 964 321,333 0.00 Qa5-b6+ Kb7xb6
4/3 00:00 1,029 343,000 0.00 Qa5-b6+ Kb7xb6
5/3 00:00 1,172 390,666 0.00 Qa5-b6+ Kb7xb6
6/3 00:00 1,647 411,750 0.00 Qa5-b6+ Kb7xb6
7/3 00:00 2,474 618,500 0.00 Qa5-b6+ Kb7xb6
8/3 00:00 3,671 734,200 0.00 Qa5-b6+ Kb7xb6
9/3 00:00 5,321 886,833 0.00 Qa5-b6+ Kb7xb6
10/3 00:00 9,182 1,020,222 0.00 Qa5-b6+ Kb7xb6
11/3 00:00 11,911 1,082,818 0.00 Qa5-b6+ Kb7xb6
12/3 00:00 15,122 1,163,230 0.00 Qa5-b6+ Kb7xb6
13/3 00:00 18,742 1,249,466 0.00 Qa5-b6+ Kb7xb6
14/3 00:00 23,938 1,329,888 0.00 Qa5-b6+ Kb7xb6
15/3 00:00 30,214 1,162,076 0.00 Qa5-b6+ Kb7xb6
16/3 00:00 38,081 1,190,031 0.00 Qa5-b6+ Kb7xb6
17/3 00:00 41,387 1,254,151 0.00 Qa5-b6+ Kb7xb6
18/3 00:00 47,499 1,249,973 0.00 Qa5-b6+ Kb7xb6
19/3 00:00 53,280 1,268,571 0.00 Qa5-b6+ Kb7xb6
20/3 00:00 61,968 1,347,130 0.00 Qa5-b6+ Kb7xb6
21/3 00:00 72,835 1,400,673 0.00 Qa5-b6+ Kb7xb6
22/3 00:00 85,058 1,394,393 0.00 Qa5-b6+ Kb7xb6
23/3 00:00 100,523 1,377,027 0.00 Qa5-b6+ Kb7xb6
24/3 00:00 123,998 1,291,645 0.00 Qa5-b6+ Kb7xb6
25/3 00:00 160,926 1,319,065 0.00 Qa5-b6+ Kb7xb6
26/3 00:00 177,039 1,311,400 0.00 Qa5-b6+ Kb7xb6
27/3 00:00 202,227 1,256,068 0.00 Qa5-b6+ Kb7xb6
28/3 00:00 221,244 1,257,068 0.00 Qa5-b6+ Kb7xb6
29/3 00:00 254,410 1,235,000 0.00 Qa5-b6+ Kb7xb6
30/3 00:00 277,852 1,229,433 0.00 Qa5-b6+ Kb7xb6
31/3 00:00 314,816 1,215,505 0.00 Qa5-b6+ Kb7xb6
32/3 00:00 387,645 1,160,613 0.00 Qa5-b6+ Kb7xb6
33/3 00:00 451,702 1,164,180 0.00 Qa5-b6+ Kb7xb6
34/3 00:00 484,317 1,169,847 0.00 Qa5-b6+ Kb7xb6
35/3 00:00 567,554 1,220,546 0.00 Qa5-b6+ Kb7xb6
36/3 00:00 617,543 1,208,499 0.00 Qa5-b6+ Kb7xb6
37/3 00:01 660,596 1,181,745 0.00 Qa5-b6+ Kb7xb6
38/53 00:13 36,449,199 2,888,895 +9.71 Qa5-c5 Kb7-b8 Qc5-c6 Kb8-a7 Qc6-b5 Ka7-a8 Qb5-b6 h4-h3 Qb6-a6+ Ka8-b8 Qa6-d6+ Kb8-b7 Qd6xg3 Kb7-c6 Qg3xh2 Kc6-c5 Qh2xh3 Kc5-d4 Qh3-e6 Kd4-d3 Qe6-e5 Kd3-d2 Kg1-f2 Kd2-d3 Qe5-d5+ Kd3-c2 Kf2-g1 Kc2-b2 Qd5-c6 Kb2-b1 Qc6-f6 Kb1-c2 Qf6-b6 Kc2-c3 Qb6-d6 Kc3-b3 Qd6-d2 Kb3-c4 Qd2-b2 Kc4-d5 Qb2-g7 Kd5-e4 Qg7-h7+ Ke4-f4 Qh7-c7+ Kf4-f5 Qc7-c4 Kf5-f6 Qc4-d5 Kf6-e7 Qd5-a8 Ke7-d7 Qa8-g8 Kd7-c7 Qg8-g6 Kc7-b7 Qg6-e6 Kb7-c7 Qe6-e7+ Kc7-c8 Qe7-h4 Kc8-b7 Qh4-d4 Kb7-a6 Qd4-f4 Ka6-b6 Qf4-c4 Kb6-b7 Qc4-b5+ Kb7-c8 Qb5-c6+ Kc8-d8 Qc6-d6+ Kd8-c8 Qd6-e7 Kc8-b8 Qe7-d7 Kb8-a8 Qd7-b5 Ka8-a7 Qb5-c6 Ka7-b8 Qc6-d7
39/53 00:31 89,131,879 2,853,407 +9.71 Qa5-c5 Kb7-b8 Qc5-c6 Kb8-a7 Qc6-b5 Ka7-a8 Qb5-b6 h4-h3 Qb6-a6+ Ka8-b8 Qa6-d6+ Kb8-b7 Qd6xg3 Kb7-c6 Qg3xh2 Kc6-c5 Qh2xh3 Kc5-d4 Qh3-e6 Kd4-d3 Qe6-e5 Kd3-d2 Kg1-f2 Kd2-d3 Qe5-d5+ Kd3-c2 Kf2-g1 Kc2-b1 Qd5-b3+ Kb1-c1 Qb3-d3 Kc1-b2 Qd3-b5+ Kb2-c3 Qb5-e2 Kc3-b4 Qe2-d3 Kb4-a5 Qd3-a3+ Ka5-b6 Qa3-e7 Kb6-a5 Qe7-f7 Ka5-a4 Qf7-d5 Ka4-a3 Qd5-d3+ Ka3-a2 Qd3-c3 Ka2-b1 Qc3-b3+

[syzygy]

I see no evidence of a bug in the tablebase code.

Regular Stockfish does not easily find the win. At the moment I am unable to test with TB probes, but TB cutoffs do affect the search tree and it is very well possible that it causes Stockfish to have even more problems with this particular position.

[Laskos]

I can also get this move with 1 core at smaller depth and I suspect stockfish has bugs in using many cores efficiently and for some reason with many cores it can see less at the same depth.

This is probably wrong Uri, I just tested the latest Stockfish on Arasan 16 250 positions suite to depth 20 on 1 thread and 4 threads, the results were 91 solved vs 90 solved, well within statistical noise. They are the same strength to the same depth.

[BBauer]

I had a similar position (zug).
Stockfish-syzygy with 6 pieces stoped after 7 plies.
Stockfish-syzygy with 5 pieces stoped after 11 plies.
Stockfish-syzygy with 0 pieces stoped run up to some 40 plies and crashed.
Stockfish-syzygy was versin of last thursday.

My impression is: Stockfish-syzygy has problems with zug-positions.

Kind regards
Bernhard