Difficult position to Critter

[Jouni]

This position is well-known from many testsuites. Without tablebases Critter
sees fast, that e4 is winning, but with tablebases it goes to Kg3? Why

[D]8/1p4kP/5pP1/3p4/8/4P3/7K/8 w - - 0 1

Analysis by Critter 1.0 64-bit:

...
1.e4 d4 2.Kg3 Kh8 3.Kf4 d3 4.Ke3 Kg7 5.Kxd3 b5 6.Kc3 Kh8 7.Kd4 b4 8.Kc4 Kg7 9.Kxb4 Kxg6 10.h8Q Kf7 11.Kc5 f5 12.Qh7+ Kf8 13.exf5 Ke8 14.f6 Kd8 15.f7 Ke7 16.f8Q+ Kxf8
+- (#31) Depth: 24/62 00:01:41 445mN

but with tablebases

...
1.Kg3 b5 2.Kf3 b4 3.Ke2 f5 4.Kd3 b3 5.Kc3 d4+ 6.exd4 f4 7.d5 f3 8.d6
+- (3.19) Depth: 28/75 00:01:32 250mN, tb=1193451

Jouni

[Arpad Rusz]

That position is a study by N.D.Grigoriev (Shakhmaty v SSSR, 1932)

[D]8/1p4kP/5pP1/3p4/8/4P3/7K/8 w - - 0 1
White wins

1. e4!! (1. Kg3? b5 2. Kf3 (2. e4 b4! 3. exd5 b3 =; 2. Kf4 b4 3. Kf5 b3 4. Ke6 b2 5. h8=Q+ Kxh8 6. Kf7 b1=Q 7. g7+ Kh7 8. g8=Q+ Kh6 =) 2... b4 3. Ke2 f5 4.Kd3 b3 5. Kc3 d4+! 6. exd4 f4 =) 1... dxe4 (1... d4 2. Kg3 b5 3. Kf4 3... b4 4. Kf5 b3 5. Ke6 b2 6. h8=Q+ Kxh8 7. Kf7 b1=Q 8. g7+ Kh7 9. g8=Q+ Kh6 10. Qg6#) 2. Kg3 b5 3. Kf4 e3! (3... b4 4. Kf5 b3 5. Ke6 b2 6. h8=Q+ Kxh8 7. Kf7 b1=Q 8. g7+ Kh7 9. g8=Q+ Kh6 10. Qg6#) 4. Kxe3 f5 5. Kf4! (5. Kd4? f4! =) 5... b4 6. Ke5! b3 7. Ke6 b2 8. h8=Q+ Kxh8 9. Kf7 b1=Q 10. g7+ Kh7 11. g8=Q+ Kh6 12. Qg6#

[PauloSoare]

Without tablebases-445 Mn
With tablebases-250 Mn

Is this the reason?

[Jouni]

After more testing this position seems to belong "very difficult pawn endings"! Even if engines finds e4 fast it can later come back to about 0,00
many times. But only Critter behaves differently with bases. BTW seems to be mate in 24 as three engines think.

Jouni

[Mike S.]

Didn't you set the tablebases usage to "everywhere"? Considering that I have only singlecore P4 and 32-bit only, Critter 1.0 switches to 1.e4 quite soon here:

Critter 1.0 [P4/2.4 GHz 128 MB hash]:

2/4 00:00 112 0 +2,14 1.Kg3 b5 2.Kf4 b4
3/4 00:00 168 0 +2,14 1.Kg3 b5 2.Kf4 b4
(...)
14/40 00:01 140.677 257.179 +3,24 1.Kg3 b5 2.Kf3 b4 3.Ke2 f5 4.Kd3 b3 5.Kc3 d4+ 6.exd4 f4 7.d5 f3 8.d6 f2 9.d7 f1Q 10.d8Q Qf3+ 11.Kb2 Qe2+ 12.Kxb3 Qf3+ 13.Kc4 Qe2+ 14.Kc5 Qc2+ 15.Kb5 Qe2+ 16.Kc6 Qe6+ 17.Qd6 Qe4+ 18.Kd7
(...)
23/59 00:22 11.747.822 527.257 +3,24 1.Kg3 b5 2.Kf3 b4 3.Ke2 f5 4.Kd3 b3 5.Kc3 d4+ 6.exd4 f4 7.d5 f3 8.d6 f2 9.d7 f1Q 10.d8Q Qf3+ 11.Kb2 Qe2+ 12.Kxb3 Qe6+ 13.Kc3 Qe3+ 14.Kb4 Qe4+ 15.Ka3 Qf3+ 16.Kb2 Qe2+ 17.Kb1 Qb5+ 18.Kc2 Qa4+ 19.Kc3 Qa1+ 20.Kb4 Qb1+ 21.Ka3 Qc1+ 22.Kb3 Qe3+ 23.Ka4 Qe4+
23/59+ 00:27 14.169.358 523.588 +3,44 1.e4
23/59+ 00:28 14.476.093 524.610 +3,64 1.e4
(...)
23/60 01:46 50.734.061 479.048 +12,69 1.e4 dxe4 2.Kg3 e3 3.Kf3 e2 4.Kxe2 b5 5.Ke3 b4 6.Kd4 Kh8 7.Kc4 Kg7 8.Kxb4 Kh8 9.Kc4 Kg7 10.Kd4 Kh8 11.Kd5 Kg7 12.Kc4 Kh8 13.Kb3 Kg7 14.Kc3 Kh8 15.Kd3 Kg7 16.Kc2 Kh8 17.Kb2 Kg7 18.Kb3 Kh8 19.Kc2 Kg7 20.Kd1 Kh8 21.Ke2 Kg7 22.Ke3 f5 23.Kd3 Kh8 24.Kc3 Kg7 25.Kd4 f4 26.Kd3 Kh8 27.Kd2 Kg7 28.Ke2 Kh8 29.Kd3 Kg7 30.Kc3 Kh8
(...)
24/60- 02:15 66.984.870 497.799 +5,47 1.e4 dxe4 2.Kg3 e3 3.Kf3 e2 4.Kxe2 b5 5.Ke3 b4 6.Kd4 Kh8 7.Kc4 Kg7 8.Kxb4 Kh8 9.Kc4 Kg7 10.Kd4 Kh8 11.Kd5 Kg7 12.Kc4 Kh8 13.Kb3 Kg7 14.Kc3 Kh8 15.Kd3 Kg7 16.Kc2 Kh8 17.Kb2 Kg7 18.Kb3 Kh8 19.Kc2 Kg7 20.Kd1 Kh8 21.Ke2 Kg7 22.Ke3 f5 23.Kd3 Kh8 24.Kc3 Kg7 25.Kd4 f4 26.Kd3 Kh8 27.Kd2 Kg7 28.Ke2 Kh8 29.Kd3 Kg7 30.Kc3 Kh8
Best move: e4, Value: +5,47, Depth: 24/60, Time: 02:18,839, 68.780.032 Nodes, 499.822 N/sec.

(~277,000 TBHits; 3p+4p Gaviotas only)

For comparison: Stockfish 201+GTB

Stockfish 2.01-GTB def:
1/1 00:00 31 219 +1,54 1.Kg3
2/2 00:00 68 435 +1,19 1.Kg3 b5
(...)
22/39 00:39 11.808.078 299.408 +0,92 1.Kg3 b5 2.Kf4 b4 3.Kf5 b3 4.Ke6 b2 5.h8Q+ Kxh8 6.Kf7 b1Q 7.g7+ Kh7 8.g8Q+ Kh6 9.Qg4 Qb7+ 10.Kxf6 Qb2+ 11.Qd4 Qf2+ 12.Ke5 Kg5 13.Kxd5 Qf7+ 14.Kd6 Qf8+ 15.Kc6 Qe8+ 16.Kc5 Qe7+ 17.Kb5 Qb7+ 18.Kc4 Qa6+ 19.Kb4 Qb7+ 20.Kc4
22/19+ 00:41 12.301.837 295.646 +1,04 1.e4 dxe4 2.Kg3 b5 3.Kf4 e3 4.Kxe3 b4 5.Kd4 f5 6.Ke5 b3 7.Ke6 b2 8.h8Q+ Kxh8 9.Kf7 b1Q 10.g7+
(...)
28/35+ 01:39 25.457.740 256.379 +M45 1.e4 d4 2.Kg3 b5 3.Kf4 d3 4.Ke3 d2 5.Kxd2 Kh8 6.Kc2 b4 7.Kb3 Kg7 8.Kc4 b3 9.Kc3 Kh8 10.Kxb3 Kg7 11.Kc3 Kh8 12.Kd4 Kg7 13.e5 f5 14.e6 f4 15.e7 Kh6 16.e8Q Kg5 17.g7 Kh4 18.g8Q
Best move: e4, Value: +Mate in 45, Depth: 28/35, Time: 02:24,980, 36.831.373 Nodes, 254.091 N/sec.

(~1,487,000 TBHits)

I have the GTBs on a Corsair USB memory stick and have set 8 MB tbs. cache. But no 5p Gaviota tables. With 5p tables, maybe a slowdown effect is happening here?

[PauloSoare]

After more testing this position seems to belong "very difficult pawn endings"! Even if engines finds e4 fast it can later come back to about 0,00
many times. But only Critter behaves differently with bases. BTW seems to be mate in 24 as three engines think.

Jouni

Maybe a mate in 23?

[Event "?"]
[Site "?"]
[Date "2011.03.24"]
[Round "?"]
[White "Endgame- Jouni"]
[Black "?"]
[Result "1-0"]
[Annotator "Soares,Paulo"]
[SetUp "1"]
[FEN "8/1p4kP/5pP1/3p4/8/4P3/7K/8 w - - 0 1"]
[PlyCount "45"]

1. e4 dxe4 2. Kg3 b5 (2... e3 3. Kf3 e2 4. Kxe2 b5 5. Ke3 b4 6. Kd4 b3 7. Kc3
f5 8. Kxb3 f4 9. Kc3 f3 10. Kd2 f2 11. Ke2 f1=Q+ 12. Kxf1 Kh8 13. Kf2 Kg7 14.
Kf3 Kh8 15. Kf4 Kg7 16. Kf5 Kh8 17. Ke6 Kg7 18. h8=Q+ Kxh8 19. Kf6 Kg8 20. g7
Kh7 21. Kf7 Kh6 22. g8=Q Kh5 23. Qg3 Kh6 24. Qg6#) 3. Kf4 e3 4. Kxe3 b4 5. Kd4
b3 6. Kc3 f5 7. Kxb3 f4 8. Kc3 f3 9. Kd3 f2 10. Ke2 f1=Q+ 11. Kxf1 Kh8 12. Kf2
Kg7 13. Kf3 Kh8 14. Kf4 Kg7 15. Kf5 Kh8 16. Ke6 Kg7 17. h8=Q+ Kxh8 18. Kf6 Kg8
19. g7 Kh7 20. Kf7 Kh6 21. g8=Q Kh5 22. Qg3 Kh6 23. Qh4# 1-0

[PauloSoare]

Chest is not good to find the mate? I tried, but I do not know how to use correctly, was spending too much time .

[Jouni]

Wait longer:

23/59+ 00:11 21,458,778 1,897,327 +7.55 1.e3-e4
23/59+ 00:15 30,447,265 1,934,387 +9.80 1.e3-e4
23/59+ 00:24 49,244,697 2,047,170 +13.17 1.e3-e4
23/60 00:24 50,734,061 2,042,845 +12.69 1.e3-e4 d5xe4 2.Kh2-g3 e4-e3 3.Kg3-f3 e3-e2 4.Kf3xe2 b7-b5 5.Ke2-e3 b5-b4 6.Ke3-d4 Kg7-h8 7.Kd4-c4 Kh8-g7 8.Kc4xb4 Kg7-h8 9.Kb4-c4 Kh8-g7 10.Kc4-d4 Kg7-h8 11.Kd4-d5 Kh8-g7 12.Kd5-c4 Kg7-h8 13.Kc4-b3 Kh8-g7 14.Kb3-c3 Kg7-h8 15.Kc3-d3 Kh8-g7 16.Kd3-c2 Kg7-h8 17.Kc2-b2 Kh8-g7 18.Kb2-b3 Kg7-h8 19.Kb3-c2 Kh8-g7 20.Kc2-d1 Kg7-h8 21.Kd1-e2 Kh8-g7 22.Ke2-e3 f6-f5 23.Ke3-d3 Kg7-h8 24.Kd3-c3 Kh8-g7 25.Kc3-d4 f5-f4 26.Kd4-d3 Kg7-h8 27.Kd3-d2 Kh8-g7 28.Kd2-e2 Kg7-h8 29.Ke2-d3 Kh8-g7 30.Kd3-c3 Kg7-h8
24/60- 00:27 57,258,573 2,073,684 +11.90 1.e3-e4 d5xe4 2.Kh2-g3 e4-e3 3.Kg3-f3 e3-e2 4.Kf3xe2 b7-b5 5.Ke2-e3 b5-b4 6.Ke3-d4 Kg7-h8 7.Kd4-c4 Kh8-g7 8.Kc4xb4 Kg7-h8 9.Kb4-c4 Kh8-g7 10.Kc4-d4 Kg7-h8 11.Kd4-d5 Kh8-g7 12.Kd5-c4 Kg7-h8 13.Kc4-b3 Kh8-g7 14.Kb3-c3 Kg7-h8 15.Kc3-d3 Kh8-g7 16.Kd3-c2 Kg7-h8 17.Kc2-b2 Kh8-g7 18.Kb2-b3 Kg7-h8 19.Kb3-c2 Kh8-g7 20.Kc2-d1 Kg7-h8 21.Kd1-e2 Kh8-g7 22.Ke2-e3 f6-f5 23.Ke3-d3 Kg7-h8 24.Kd3-c3 Kh8-g7 25.Kc3-d4 f5-f4 26.Kd4-d3 Kg7-h8 27.Kd3-d2 Kh8-g7 28.Kd2-e2 Kg7-h8 29.Ke2-d3 Kh8-g7 30.Kd3-c3 Kg7-h8
24/60- 00:28 59,299,978 2,088,615 +11.11 1.e3-e4 d5xe4 2.Kh2-g3 e4-e3 3.Kg3-f3 e3-e2 4.Kf3xe2 b7-b5 5.Ke2-e3 b5-b4 6.Ke3-d4 Kg7-h8 7.Kd4-c4 Kh8-g7 8.Kc4xb4 Kg7-h8 9.Kb4-c4 Kh8-g7 10.Kc4-d4 Kg7-h8 11.Kd4-d5 Kh8-g7 12.Kd5-c4 Kg7-h8 13.Kc4-b3 Kh8-g7 14.Kb3-c3 Kg7-h8 15.Kc3-d3 Kh8-g7 16.Kd3-c2 Kg7-h8 17.Kc2-b2 Kh8-g7 18.Kb2-b3 Kg7-h8 19.Kb3-c2 Kh8-g7 20.Kc2-d1 Kg7-h8 21.Kd1-e2 Kh8-g7 22.Ke2-e3 f6-f5 23.Ke3-d3 Kg7-h8 24.Kd3-c3 Kh8-g7 25.Kc3-d4 f5-f4 26.Kd4-d3 Kg7-h8 27.Kd3-d2 Kh8-g7 28.Kd2-e2 Kg7-h8 29.Ke2-d3 Kh8-g7 30.Kd3-c3 Kg7-h8
24/60- 00:29 60,732,218 2,098,701 +9.92 1.e3-e4 d5xe4 2.Kh2-g3 e4-e3 3.Kg3-f3 e3-e2 4.Kf3xe2 b7-b5 5.Ke2-e3 b5-b4 6.Ke3-d4 Kg7-h8 7.Kd4-c4 Kh8-g7 8.Kc4xb4 Kg7-h8 9.Kb4-c4 Kh8-g7 10.Kc4-d4 Kg7-h8 11.Kd4-d5 Kh8-g7 12.Kd5-c4 Kg7-h8 13.Kc4-b3 Kh8-g7 14.Kb3-c3 Kg7-h8 15.Kc3-d3 Kh8-g7 16.Kd3-c2 Kg7-h8 17.Kc2-b2 Kh8-g7 18.Kb2-b3 Kg7-h8 19.Kb3-c2 Kh8-g7 20.Kc2-d1 Kg7-h8 21.Kd1-e2 Kh8-g7 22.Ke2-e3 f6-f5 23.Ke3-d3 Kg7-h8 24.Kd3-c3 Kh8-g7 25.Kc3-d4 f5-f4 26.Kd4-d3 Kg7-h8 27.Kd3-d2 Kh8-g7 28.Kd2-e2 Kg7-h8 29.Ke2-d3 Kh8-g7 30.Kd3-c3 Kg7-h8
24/60- 00:31 64,960,002 2,095,686 +8.14 1.e3-e4 d5xe4 2.Kh2-g3 e4-e3 3.Kg3-f3 e3-e2 4.Kf3xe2 b7-b5 5.Ke2-e3 b5-b4 6.Ke3-d4 Kg7-h8 7.Kd4-c4 Kh8-g7 8.Kc4xb4 Kg7-h8 9.Kb4-c4 Kh8-g7 10.Kc4-d4 Kg7-h8 11.Kd4-d5 Kh8-g7 12.Kd5-c4 Kg7-h8 13.Kc4-b3 Kh8-g7 14.Kb3-c3 Kg7-h8 15.Kc3-d3 Kh8-g7 16.Kd3-c2 Kg7-h8 17.Kc2-b2 Kh8-g7 18.Kb2-b3 Kg7-h8 19.Kb3-c2 Kh8-g7 20.Kc2-d1 Kg7-h8 21.Kd1-e2 Kh8-g7 22.Ke2-e3 f6-f5 23.Ke3-d3 Kg7-h8 24.Kd3-c3 Kh8-g7 25.Kc3-d4 f5-f4 26.Kd4-d3 Kg7-h8 27.Kd3-d2 Kh8-g7 28.Kd2-e2 Kg7-h8 29.Ke2-d3 Kh8-g7 30.Kd3-c3 Kg7-h8
24/60- 00:31 66,984,870 2,103,862 +5.47 1.e3-e4 d5xe4 2.Kh2-g3 e4-e3 3.Kg3-f3 e3-e2 4.Kf3xe2 b7-b5 5.Ke2-e3 b5-b4 6.Ke3-d4 Kg7-h8 7.Kd4-c4 Kh8-g7 8.Kc4xb4 Kg7-h8 9.Kb4-c4 Kh8-g7 10.Kc4-d4 Kg7-h8 11.Kd4-d5 Kh8-g7 12.Kd5-c4 Kg7-h8 13.Kc4-b3 Kh8-g7 14.Kb3-c3 Kg7-h8 15.Kc3-d3 Kh8-g7 16.Kd3-c2 Kg7-h8 17.Kc2-b2 Kh8-g7 18.Kb2-b3 Kg7-h8 19.Kb3-c2 Kh8-g7 20.Kc2-d1 Kg7-h8 21.Kd1-e2 Kh8-g7 22.Ke2-e3 f6-f5 23.Ke3-d3 Kg7-h8 24.Kd3-c3 Kh8-g7 25.Kc3-d4 f5-f4 26.Kd4-d3 Kg7-h8 27.Kd3-d2 Kh8-g7 28.Kd2-e2 Kg7-h8 29.Ke2-d3 Kh8-g7 30.Kd3-c3 Kg7-h8
24/72 00:40 86,557,832 2,127,512 +3.24 1.Kh2-g3 b7-b5 2.Kg3-f3 b5-b4 3.Kf3-e2 f6-f5 4.Ke2-d3 b4-b3 5.Kd3-c3 d5-d4+ 6.e3xd4 f5-f4 7.d4-d5 f4-f3 8.d5-d6 f3-f2 9.d6-d7 f2-f1Q 10.d7-d8Q Qf1-f3+ 11.Kc3-b2 Qf3-e2+ 12.Kb2xb3 Qe2-e6+ 13.Kb3-c3 Qe6-e3+ 14.Kc3-b4 Qe3-e4+ 15.Kb4-a3 Qe4-f3+ 16.Ka3-a2 Qf3-e2+ 17.Ka2-b1 Qe2-b5+ 18.Kb1-c1 Qb5-c5+ 19.Kc1-d1 Qc5-g1+ 20.Kd1-d2 Qg1-f2+ 21.Kd2-d3 Qf2-f3+ 22.Kd3-c2 Qf3-e2+ 23.Kc2-c1 Qe2-e1+ 24.Kc1-c2 Qe1-f2+
25/72 01:11 154,066,204 2,164,854 +3.16 1.Kh2-g3 b7-b5 2.Kg3-f3 b5-b4 3.Kf3-e2 f6-f5 4.Ke2-d3 b4-b3 5.Kd3-c3 d5-d4+ 6.e3xd4

Jouni

[ernest]

With 5p tables, maybe a slowdown effect is happening here?

Indeed, with 5p tables (of course "everywhere"), no solution (stopped after 10 minutes, depth=29), But nothing to do with "slowdown" (still more than 700kN/s).

But no problem with Houdini, also using Gaviota TBs (finds e4 immediately)