Why does Houdini 1.5a fail on this position?

Discussion of anything and everything relating to chess playing software and machines.

Moderators: bob, hgm, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
JohnS
Posts: 152
Joined: Sun Feb 24, 2008 1:08 am

Why does Houdini 1.5a fail on this position?

Post by JohnS » Sun Sep 18, 2011 12:20 am

[D]2b4k/8/5Pr1/5N2/8/8/8/K1B5 w - -

This is a study by Korolkov. The solution is
1. f7 Ra6+ 2. Ba3 Rxa3+ 3. Kb2 Ra2+ 4. Kc1 Ra1+ 5. Kd2 Ra2+
6. Ke3 Ra3+ 7. Kf4 Ra4+ 8. Kg5 Rg4+ 9. Kh6 Rg8 10. Ne7 Be6
11. fxg8=Q+ Bxg8 12. Ng6#

Houdini doesn't understand this and thinks it's a draw, even at the end when there is a clear mate.

25/35 0:01 0.00 1.f7 Ra6+ 2.Ba3 Rxa3+ 3.Kb2 Ra2+ 4.Kxa2 Be6+ 5.Kb2 Bxf7 (9.781.106) 5920
26/38 0:02 0.00 1.f7 Ra6+ 2.Ba3 Rxa3+ 3.Kb2 Ra2+ 4.Kxa2 Be6+ 5.Kb2 Bxf7 (14.115.604) 6086
27/40 0:03 0.00 1.f7 Ra6+ 2.Ba3 Rxa3+ 3.Kb2 Ra2+ 4.Kxa2 Be6+ 5.Kb2 Bxf7 (22.286.021) 6284
28/42 0:04 0.00 1.f7 Ra6+ 2.Ba3 Rxa3+ 3.Kb2 Ra2+ 4.Kxa2 Be6+ 5.Kb2 Bxf7 (31.581.918) 6359
29/42 0:07 0.00 1.f7 Ra6+ 2.Ba3 Rxa3+ 3.Kb2 Ra2+ 4.Kxa2 Be6+ 5.Kb2 Bxf7 (46.812.460) 6453
30/45 0:10 0.00 1.f7 Ra6+ 2.Ba3 Rxa3+ 3.Kb2 Ra2+ 4.Kxa2 Be6+ 5.Kb2 Bxf7 (68.857.680) 6515
31/45 0:16 0.00 1.f7 Ra6+ 2.Ba3 Rxa3+ 3.Kb2 Ra2+ 4.Kxa2 Be6+ 5.Kb2 Bxf7 (106.446.370) 6568
32/46 0:22 0.00 1.f7 Ra6+ 2.Ba3 Rxa3+ 3.Kb2 Ra2+ 4.Kxa2 Be6+ 5.Kb2 Bxf7 (146.973.406) 6541
33/46 0:32 0.00 1.f7 Ra6+ 2.Ba3 Rxa3+ 3.Kb2 Ra2+ 4.Kxa2 Be6+ 5.Kb2 Bxf7 (209.973.176) 6469
34/48 0:50 0.00 1.f7 Ra6+ 2.Ba3 Rxa3+ 3.Kb2 Ra2+ 4.Kxa2 Be6+ 5.Kb2 Bxf7 (315.243.095) 6273
35/48 1:15 0.00 1.f7 Ra6+ 2.Ba3 Rxa3+ 3.Kb2 Ra2+ 4.Kxa2 Be6+ 5.Kb2 Bxf7 (449.467.996) 5981
35/53 1:50 0.00 1.f7 Ra6+ 2.Ba3 Rxa3+ 3.Kb2 Ra2+ 4.Kxa2 Be6+ 5.Kb2 Bxf7 (612.062.040) 5533
best move: f6-f7 time: 1:50.609 min n/s: 5.533.000 CPU 99.5% n/s(1CPU): 5.560.804 nodes: 612.062.040
35/31 0:00 0.00 1...Ra6+ 2.Ba3 Rxa3+ 3.Kb2 Ra2+ 4.Kxa2 Be6+ 5.Kb2 Bxf7 (533.228) 2539
35/38 0:04 0.00 1...Ra6+ 2.Ba3 Rxa3+ 3.Kb2 Ra2+ 4.Kxa2 Be6+ 5.Kb2 Bxf7 (21.643.725) 4572
best move: Rg6-a6 time: 0:04.734 min n/s: 4.572.000 CPU 98.1% n/s(1CPU): 4.660.550 nodes: 21.643.725
36/34 0:00 0.00 2.Ba3 Rxa3+ 3.Kb2 Ra2+ 4.Kxa2 Be6+ 5.Kb2 Bxf7 (1.978.156) 3434
36/40 0:03 0.00 2.Ba3 Rxa3+ 3.Kb2 Ra2+ 4.Kxa2 Be6+ 5.Kb2 Bxf7 (14.427.698) 4378
best move: Bc1-a3 time: 0:03.282 min n/s: 4.378.000 CPU 95.4% n/s(1CPU): 4.589.098 nodes: 14.427.698
35/23 0:00 0.00 2...Rxa3+ 3.Kb2 Ra2+ 4.Kxa2 Be6+ 5.Kb2 Bxf7 (51.247) 483
35/50 0:02 0.00 2...Rxa3+ 3.Kb2 Ra2+ 4.Kxa2 Be6+ 5.Kb2 Bxf7 (11.980.400) 4391
best move: Ra6xa3 time: 0:02.719 min n/s: 4.391.000 CPU 99.5% n/s(1CPU): 4.413.065 nodes: 11.980.400
36/28 0:00 0.00 3.Kb2 Ra2+ 4.Kxa2 Be6+ 5.Kb2 Bxf7 (454.832) 1911
36/47 0:02 0.00 3.Kb2 Ra2+ 4.Kxa2 Be6+ 5.Kb2 Bxf7 (9.278.113) 3999
best move: Ka1-b2 time: 0:02.329 min n/s: 3.999.000 CPU 97.2% n/s(1CPU): 4.114.197 nodes: 9.278.113
35/37 0:00 0.00 3...Ra2+ 4.Kxa2 Be6+ 5.Kb2 Bxf7 (3.261.786) 3274
35/40 0:03 0.00 3...Ra2+ 4.Kxa2 Be6+ 5.Kb2 Bxf7 (11.968.791) 3715
best move: Ra3-a2 time: 0:03.218 min n/s: 3.715.000 CPU 96.9% n/s(1CPU): 3.833.849 nodes: 11.968.791
36/43 0:01 0.00 4.Kxa2 Be6+ 5.Kb2 Bxf7 (5.213.101) 3640
best move: Kb2xa2 time: 0:01.422 min n/s: 3.640.000 CPU 94.8% n/s(1CPU): 3.839.662 nodes: 5.213.101
30/33 0:01 0.00 4...Ra1+ 5.Kd2 Ra2+ 6.Kc1 (3.903.240) 2923
31/38 0:01 0.00 4...Ra1+ 5.Kd2 Ra2+ 6.Kc1 (6.140.022) 3142
32/39 0:02 0.00 4...Ra1+ 5.Kd2 Ra2+ 6.Kc1 (9.623.239) 3288
32/39 0:03 0.00 4...Ra1+ 5.Kd2 Ra2+ 6.Kc1 (12.671.271) 3349
best move: Ra2-a1 time: 0:03.782 min n/s: 3.349.000 CPU 100.2% n/s(1CPU): 3.342.315 nodes: 12.671.271
33/37 0:01 0.00 5.Kb2 Ra2+ 6.Kxa2 Be6+ 7.Ka3 Bxf7 (4.110.710) 3483
best move: Kc1-b2 time: 0:01.203 min n/s: 3.483.000 CPU 95.8% n/s(1CPU): 3.635.699 nodes: 4.110.710
33/39 0:01 0.00 5...Ra2+ 6.Kc1 Ra1+ 7.Kd2 Ra2+ (7.246.966) 3729
33/39 0:03 0.00 5...Ra2+ 6.Kc1 Ra1+ 7.Kd2 Ra2+ (12.396.636) 3853
best move: Ra1-a2 time: 0:03.234 min n/s: 3.853.000 CPU 99.9% n/s(1CPU): 3.856.856 nodes: 12.396.636
33/39 0:01 0.00 6.Kc1 Ra1+ 7.Kd2 Ra2+ 8.Kc1 (3.900.688) 3166
best move: Kd2-c1 time: 0:01.234 min n/s: 3.166.000 CPU 96.7% n/s(1CPU): 3.274.043 nodes: 3.900.688
32/38 0:01 0.00 6...Ra3+ 7.Kf4 Ra4+ 8.Kg5 Rg4+ 9.Kxg4 Bxf5+ 10.Kh4 Kg7 11.Kg5 Kxf7 (6.216.243) 3570
33/39 0:02 0.00 6...Ra3+ 7.Kf4 Ra4+ 8.Kg5 Rg4+ 9.Kxg4 Bxf5+ 10.Kh4 Kg7 11.Kg5 Kxf7 (11.124.507) 3713
33/39 0:03 0.00 6...Ra3+ 7.Kf4 Ra4+ 8.Kg5 Rg4+ 9.Kxg4 Bxf5+ 10.Kh4 Kg7 11.Kg5 Kxf7 (14.148.836) 3753
best move: Ra2-a3 time: 0:03.766 min n/s: 3.753.000 CPU 97.2% n/s(1CPU): 3.861.111 nodes: 14.148.836
33/43 0:01 0.00 7.Kf4 Ra4+ 8.Kg5 Rg4+ 9.Kxg4 Bxf5+ 10.Kh4 Kg7 11.Kg5 Kxf7 (3.793.261) 3071
best move: Ke3-f4 time: 0:01.235 min n/s: 3.071.000 CPU 93.7% n/s(1CPU): 3.277.481 nodes: 3.793.261
33/38 0:01 0.00 7...Ra4+ 8.Kg5 Rg4+ 9.Kxg4 Bxf5+ 10.Kh4 Kg7 11.Kg5 Kxf7 (4.622.820) 3642
33/38 0:01 0.00 7...Ra4+ 8.Kg5 Rg4+ 9.Kxg4 Bxf5+ 10.Kh4 Kg7 11.Kg5 Kxf7 (6.903.821) 3687
best move: Ra3-a4 time: 0:01.891 min n/s: 3.687.000 CPU 96.2% n/s(1CPU): 3.832.640 nodes: 6.903.821
34/36 0:01 0.00 8.Kg5 Rg4+ 9.Kxg4 Bxf5+ 10.Kh4 Kg7 11.Kg5 Kxf7 (6.009.091) 3837
34/38 0:02 0.00 8.Kg5 Rg4+ 9.Kxg4 Bxf5+ 10.Kh4 Kg7 11.Kg5 Kxf7 (10.701.020) 3639
best move: Kf4-g5 time: 0:02.969 min n/s: 3.639.000 CPU 99.1% n/s(1CPU): 3.672.048 nodes: 10.701.020
35/35 0:01 0.00 8...Rg4+ 9.Kxg4 Bxf5+ 10.Kh4 Kg7 11.Kg5 Kxf7 (5.894.650) 3883
36/35 0:02 0.00 8...Rg4+ 9.Kxg4 Bxf5+ 10.Kh4 Kg7 11.Kg5 Kxf7 (9.803.389) 3956
36/35 0:02 0.00 8...Rg4+ 9.Kxg4 Bxf5+ 10.Kh4 Kg7 11.Kg5 Kxf7 (9.951.982) 3939
best move: Ra4-g4 time: 0:02.547 min n/s: 3.939.000 CPU 97.0% n/s(1CPU): 4.060.824 nodes: 9.951.982
37/35 0:00 0.00 9.Kxg4 Bxf5+ 10.Kh4 Kg7 11.Kg5 Kxf7 (2.450.134) 3530
38/35 0:01 0.00 9.Kxg4 Bxf5+ 10.Kh4 Kg7 11.Kg5 Kxf7 (7.749.432) 3978
38/36 0:02 0.00 9.Kxg4 Bxf5+ 10.Kh4 Kg7 11.Kg5 Kxf7 (8.935.601) 3943
best move: Kg5xg4 time: 0:02.266 min n/s: 3.943.000 CPU 97.1% n/s(1CPU): 4.060.762 nodes: 8.935.601
38/37 0:01 0.00 9...Rg8 10.fxg8Q+ Kxg8 (7.163.152) 4028
39/37 0:03 0.00 9...Rg8 10.fxg8Q+ Kxg8 (15.015.077) 4183
40/38 0:04 0.00 9...Rg8 10.fxg8Q+ Kxg8 (20.597.746) 4207
41/40 0:06 0.00 9...Rg8 10.fxg8Q+ Kxg8 (28.833.080) 4214
42/41 0:08 0.00 9...Rg8 10.fxg8Q+ Kxg8 (36.382.691) 4196
42/41 0:09 0.00 9...Rg8 10.fxg8Q+ Kxg8 (40.461.425) 4194
best move: Rg4-g8 time: 0:09.656 min n/s: 4.194.000 CPU 98.3% n/s(1CPU): 4.266.531 nodes: 40.461.425
42/34 0:00 0.00 10.fxg8Q+ Kxg8 (2.354.618) 3256
43/41 0:06 0.00 10.fxg8Q+ Kxg8 (25.821.677) 4253
44/41 0:07 0.00 10.fxg8Q+ Kxg8 (33.238.302) 4220
45/41 0:13 0.00 10.fxg8Q+ Kxg8 (56.057.227) 4219
46/41 0:17 0.00 10.fxg8Q+ Kxg8 (74.416.278) 4186

Can anyone explain this?

Cubeman
Posts: 644
Joined: Fri Feb 02, 2007 2:11 am
Location: New Zealand

Re: Why does Houdini 1.5a fail on this position?

Post by Cubeman » Sun Sep 18, 2011 12:45 am

Some engines have in built code to recognise that K+N vs K+B is a draw, so will evaluate any such position as equal.I remember there were many engines that failed this position in the past, but the more modern ones should now know that this is a win.

EmilV
Posts: 12
Joined: Tue May 03, 2011 11:14 am
Contact:

Re: Why does Houdini 1.5a fail on this position?

Post by EmilV » Sun Sep 18, 2011 5:53 am

Here is another example
[D]5K2/2q2bpk/8/6R1/1Q6/6b1/3B4/8 w - -

Pospisil Diagrammes#6431 2009
1.Rxg7+ Kh8 2.Rg8+ Bxg8 3.Bc3+ Be5 [3...Kh7 4.Qe4+ Kh6 5.Bd2+ Kh5 6.Qf5+] 4.Qh4+ Bh7 5.Qf6+ Qg7+ 6.Qxg7+ Bxg7+ 7.Bxg7#

Houdini 1.5 cannot solve these studies, but Houdini 2.0 does.

User avatar
David Dahlem
Posts: 900
Joined: Wed Mar 08, 2006 8:06 pm

Re: Why does Houdini 1.5a fail on this position?

Post by David Dahlem » Sun Sep 18, 2011 10:42 am

JohnS wrote:[D]2b4k/8/5Pr1/5N2/8/8/8/K1B5 w - -

This is a study by Korolkov. The solution is
1. f7 Ra6+ 2. Ba3 Rxa3+ 3. Kb2 Ra2+ 4. Kc1 Ra1+ 5. Kd2 Ra2+
6. Ke3 Ra3+ 7. Kf4 Ra4+ 8. Kg5 Rg4+ 9. Kh6 Rg8 10. Ne7 Be6
11. fxg8=Q+ Bxg8 12. Ng6#

Houdini doesn't understand this and thinks it's a draw, even at the end when there is a clear mate.
Chest 5.2 doesn't find it either!!

FEN: 2b4k/8/5Pr1/5N2/8/8/8/K1B5 w - - 0 1

AutoTurbo-Search for Special-Mate [C0/R0/K4/P0/X0] in 12 ... (Hash=2044MB)
AutoTurbo-Search for Special-Mate [C0/R0/K5/P0/X0] in 12 ... (Hash=2044MB)
AutoTurbo-Search for Special-Mate [C0/R0/K6/P0/X0] in 12 ... (Hash=2044MB)
AutoTurbo-Search for Special-Mate [C0/R0/K7/P0/X0] in 12 ... (Hash=2044MB)
AutoTurbo-Search for Special-Mate [C0/R0/K8/P0/X0] in 12 ... (Hash=2044MB)
AutoTurbo-Search for Special-Mate [C0/R0/K0/P0/X0] in 12 ... (Hash=2044MB)
No Mate in 12 found ! (01:05)

Ignacio
Posts: 147
Joined: Wed Mar 08, 2006 7:15 pm

Re: Why does Houdini 1.5a fail on this position?

Post by Ignacio » Sun Sep 18, 2011 10:59 am

Hi David,
It´s mate in 12 and Chest 5.2 says:
5K2/2q2bpk/8/6R1/1Q6/6b1/3B4/8 w - - bm #12; 00:00 @ C0/R-3/K4/P3/X12;

User avatar
David Dahlem
Posts: 900
Joined: Wed Mar 08, 2006 8:06 pm

Re: Why does Houdini 1.5a fail on this position?

Post by David Dahlem » Sun Sep 18, 2011 11:20 am

Ignacio wrote:Hi David,
It´s mate in 12 and Chest 5.2 says:
5K2/2q2bpk/8/6R1/1Q6/6b1/3B4/8 w - - bm #12; 00:00 @ C0/R-3/K4/P3/X12;
That is odd. I wonder why my Chest doesn't solve it. I thought Auto Turbo is supposed to automatically find the correct parameters. :?:

Chest doesn't solve this either -

[D]5K2/2q2bpk/8/6R1/1Q6/6b1/3B4/8 w - -
Pospisil Diagrammes#6431 2009
1.Rxg7+ Kh8 2.Rg8+ Bxg8 3.Bc3+ Be5 [3...Kh7 4.Qe4+ Kh6 5.Bd2+ Kh5 6.Qf5+] 4.Qh4+ Bh7 5.Qf6+ Qg7+ 6.Qxg7+ Bxg7+ 7.Bxg7#

FEN: 5K2/2q2bpk/8/6R1/1Q6/6b1/3B4/8 w - - 0 1

Chest 5.2:

Search completed ... (Time=74.49s)
Mate in 12 found ! (1 Solution in 01:14)
12/12 01:15 54.764.704 735.165 +M12 Rg5xg7+ Kh7h8 Rg7g8+ Bf7xg8 Bd2c3+ Kh8h7 Qb4e4+ Kh7h6 Bc3d2+ Qc7f4+ Bd2xf4+ Bg3xf4 Qe4xf4+ Kh6g6 Kf8xg8 Kg6h5 Qf4g3 Kh5h6 Kg8f8 Kh6h5 Kf8f7 Kh5h6 Qg3h4+

User avatar
Ajedrecista
Posts: 1446
Joined: Wed Jul 13, 2011 7:04 pm
Location: Madrid, Spain.
Contact:

Re: Why does Houdini 1.5a fail on this position?

Post by Ajedrecista » Sun Sep 18, 2011 11:25 am

Hello:

[D]2b4k/8/5Pr1/5N2/8/8/8/K1B5 w - -

With the priceless help of online Nalimov 6-man TB, I can say that this line is not perfect. I will use Nunn's notation ('!' when a move is the only winning) and annotating between brackets imperfect play; I start with your line and when the board is with six pieces, then I use the tablebases:

1. f7 Ra6+ 2. Ba3 Rxa3+ (at this point, 3. Kb2! is claimed to be mate in 18) 3. Kb2! Ra2+ 4. Kc1! Ra1+ (lose in 14, when Rc2+ lose in 16) 5. Kd2! Ra2+ 6. Ke3 Ra3+ (lose in 11, while Ra7, Bxf5 and Kh7 are loses in 12) 7. Kf4 Ra4+ 8. Kg5 Rg4+ (lose in 5, while Bxf5 and Kh7 lose in 9) 9. Kh6! Rg8 (lose in 3, while Rg6+ lose in 4) 10. Ne7! Be6 11. fxg8=Q+ Bxg8 12. Ng6#

I think this is the reason why Chest 5.2 could not find a mate in 12; if the start of your variation is good, this will be a mate in 20. The study is wrong, but I recognize that it is extremely difficult solving this study in the correct way without computer help (and even with it...). There is a video on Youtube at this link where some variations are explained; this study was composed in 1951, so it is 60 years old! Please enjoy, although this video has the same errors.

Regards from Spain.

Ajedrecista.

User avatar
Houdini
Posts: 1471
Joined: Mon Mar 15, 2010 11:00 pm
Contact:

Re: Why does Houdini 1.5a fail on this position?

Post by Houdini » Sun Sep 18, 2011 11:56 am

John,

This is a known issue in Houdini 1.5. It has been discussed at least twice before on this forum, the Forum Search can probably help you to find the threads.
The bug has been corrected in Houdini 2.

Robert

Arpad Rusz
Posts: 270
Joined: Sat Apr 17, 2010 12:34 pm
Location: Budapest

Re: Why does Houdini 1.5a fail on this position?

Post by Arpad Rusz » Sun Sep 18, 2011 12:41 pm

Ajedrecista wrote: [D]2b4k/8/5Pr1/5N2/8/8/8/K1B5 w - -

With the priceless help of online Nalimov 6-man TB, I can say that this line is not perfect. I will use Nunn's notation ('!' when a move is the only winning) and annotating between brackets imperfect play; I start with your line and when the board is with six pieces, then I use the tablebases:

1. f7 Ra6+ 2. Ba3 Rxa3+ (at this point, 3. Kb2! is claimed to be mate in 18) 3. Kb2! Ra2+ 4. Kc1! Ra1+ (lose in 14, when Rc2+ lose in 16) 5. Kd2! Ra2+ 6. Ke3 Ra3+ (lose in 11, while Ra7, Bxf5 and Kh7 are loses in 12) 7. Kf4 Ra4+ 8. Kg5 Rg4+ (lose in 5, while Bxf5 and Kh7 lose in 9) 9. Kh6! Rg8 (lose in 3, while Rg6+ lose in 4) 10. Ne7! Be6 11. fxg8=Q+ Bxg8 12. Ng6#

I think this is the reason why Chest 5.2 could not find a mate in 12; if the start of your variation is good, this will be a mate in 20. The study is wrong...
I have to correct you. :)
This is a correct study! It's a study and not a problem so the longest line has nothing to do with its solution. White's play is dual-free except some time-wasting moves, so in Nunn's notation all White moves deserve an '!'.

User avatar
Houdini
Posts: 1471
Joined: Mon Mar 15, 2010 11:00 pm
Contact:

Re: Why does Houdini 1.5a fail on this position?

Post by Houdini » Sun Sep 18, 2011 1:06 pm

Arpad Rusz wrote:I have to correct you. :)
This is a correct study! It's a study and not a problem so the longest line has nothing to do with its solution. White's play is dual-free except some time-wasting moves, so in Nunn's notation all White moves deserve an '!'.
Agreed.
In fact Black's longest defense might be the rather uninteresting 1.f7 Rg8 2.fxg8Q Kxg8 3.Ne7+ Kf7 4.Nxc8 followed by a mate in 31.
This means the starting position appears to be mate in 35.

Robert

Post Reply