bm Kd4: engine solutions?

[Howard E]

[d] 4B3/8/8/2p1k1p1/5p2/2pb1PP1/7P/3K4 b - - bm Kd4;

[Eelco de Groot]

[d] 4B3/8/8/2p1k1p1/5p2/2pb1PP1/7P/3K4 b - - bm Kd4;

Hello Howard,

Blueberry does not score very well in the EET test from Walter eigenmann but seems to get this one right for a change:

[d]4B3/8/8/2p1k1p1/5p2/2pb1PP1/7P/3K4 b - -

Engine: Blueberry Beta 4 DM70 Build 421 (Athlon 2009 MHz, 256 MB)
by F. Letouzey, T. Gaksch, E. de Groot

1/08 0:00 +2.94 1...fxg3 2.hxg3 (422)

1/16 0:00 +3.07 1...Bf5 (12.660)

1/16 0:00 +3.10 1...Kd4 (12.696)

2/16 0:00 +2.28 1...Kd4 2.gxf4 gxf4 3.Ba4 (13.613)

3/19 0:00 +2.28 1...Kd4 2.gxf4 gxf4 3.Ba4 (142.621)

4/35 0:00 +2.79 1...Kd4 2.gxf4 gxf4 3.h4 Ke3 4.Bd7 Kxf3
5.Ba4 (314.286)

5/35 0:00 +2.87 1...Kd4 2.gxf4 gxf4 3.h4 Ke3 4.Bc6 Kd4 (551.324)

6/39 0:01 +2.32 1...Kd4 2.gxf4 gxf4 3.h4 Ke3 4.Bc6 Bg6
5.h5 c2+ 6.Kc1 Bxh5 7.Kxc2 Bxf3 (1.184.949) 714

7/41 0:05 +2.60 1...Kd4 2.gxf4 gxf4 3.h4 Ke3 4.Bc6 Bg6
5.h5 c2+ 6.Kc1 Bxh5 7.Be4 (4.126.462) 746

8/45 0:13 +2.34 1...Kd4 2.gxf4 gxf4 3.h4 Ke3 4.Bc6 Bg6
5.Kc1 c4 6.Bd5 Kd4 7.h5 Bxh5 8.Be4 Bg4 (11.463.000) 831

9/46 0:24 +2.45 1...Kd4 2.gxf4 gxf4 3.h4 Ke3 4.Bc6 Bg6
5.Kc1 c4 6.Ba4 Kxf3 7.h5 Bxh5 8.Bd1+ Ke3
9.Bxh5 f3 10.Bxf3 Kxf3 (20.194.602) 812

10/46 0:32 +2.46 1...Kd4 2.gxf4 gxf4 3.h4 Ke3 4.Bc6 Bg6
5.Kc1 c4 6.Bd5 Kd4 7.h5 Bxh5 8.Be4 Ke3
9.Kc2 Bxf3 10.Bxf3 Kxf3 11.Kxc3 (25.800.524) 805

11/47 0:51 +2.46 1...Kd4 2.gxf4 gxf4 3.h4 Ke3 4.Bc6 Bg6
5.Kc1 c4 6.Bd5 Kd4 7.h5 Bxh5 8.Be4 Ke3
9.Kc2 Bxf3 10.Bxf3 Kxf3 11.Kxc3 (40.766.666) 797

12/50 1:28 +2.46 1...Kd4 2.gxf4 gxf4 3.h4 Ke3 4.Bc6 Bg6
5.Kc1 c4 6.Bd5 Kd4 7.h5 Bxh5 8.Be4 Ke3
9.Kc2 Bxf3 10.Bxf3 Kxf3 11.Kxc3 (70.292.999) 797

13/56 2:32 +2.44 1...Kd4 2.gxf4 gxf4 3.h4 Ke3 4.Bc6 Bg6
5.Kc1 c2 6.Bd5 Kf2 7.Kd2 Kg3 8.Bc6 Kxh4
9.Kc1 c4 10.Ba4 Kg3 11.Bxc2 Bxc2
12.Kxc2 Kxf3 (121.121.160) 795

14/62 8:52 +2.47 1...Kd4 2.gxf4 gxf4 3.Bc6 Bf5 4.h4 Ke5
5.Ba4 Kf6 6.Bc6 Kg6 7.Be8+ Kh6 8.h5 c4
9.Kc1 Kg5 10.Ba4 Kxh5 11.Kd1 (431.347.350) 810

15/62 8:53 +2.46 1...Kd4 2.gxf4 gxf4 3.h4 Ke3 4.Bc6 c2+
5.Kc1 Bg6 6.Bd5 Kf2 7.Be6 Kxf3 8.Bh3 Bd3
9.Bd7 (432.045.572) 810

16/66 10:30 +3.68 1...Kd4 2.gxf4 gxf4 3.h4 Ke3 4.Bc6 Bg6
5.Kc1 Kf2 6.Bd5 Kg3 7.Bc6 Kxh4 8.Bd7 Bd3
9.Ba4 c4 10.Bc6 Kg3 11.Ba4 Kxf3 (511.501.943) 811

17/66 11:26 +4.14 1...Kd4 2.gxf4 gxf4 3.h4 Ke3 4.Bc6 Bg6
5.Kc1 Kf2 6.Ba4 Kg3 7.Bd1 Kxh4 8.Bb3 Kg3
9.Bd5 Bh5 10.Kc2 Bxf3 11.Bc4 Bb7
12.Bf1 f3 13.Kxc3 Bd5 (560.559.872) 816

18/66 11:33 +4.13 1...Kd4 2.gxf4 gxf4 3.h4 Ke3 4.Bc6 Bg6
5.Kc1 Kf2 6.Bd5 Kg2 7.Bc6 Kh3 8.Kd1 Kxh4
9.Ke1 Bf5 10.Ke2 Kg5 11.Kd1 Kf6
12.Be8 (567.121.736) 817

19/66 11:45 +4.13 1...Kd4 2.gxf4 gxf4 3.h4 Ke3 4.Bc6 Bg6
5.Kc1 Kf2 6.Bd5 Kg2 7.Bc6 Kh3 8.Kd1 Kxh4
9.Ke1 c4 10.Kd1 Bf5 11.Bd5 Bg6 (576.263.231) 817

20/78 18:01 +5.64 1...Kd4 2.gxf4 gxf4 3.h4 Ke3 4.Bc6 Bg6
5.Kc1 Kf2 6.Bd5 Kg3 7.Kd1 Kxh4 8.Bc6 Bf5 (910.826.304) 841

21/78 20:39 +5.64 1...Kd4 2.gxf4 gxf4 3.h4 Ke3 4.Bc6 Bg6
5.Kc1 Kf2 6.Bb7 Kg3 7.Kd1 Kxh4 8.Ke2 Kh3
9.Ba8 Bb1 10.Kd1 Kg3 11.Kc1 Bg6
12.Bc6 Bh5 (1.054.073.269) 850

best move: Ke5-d4 time: 32:58.610 min n/s: 869.771 CPU 100.0% n/s(1CPU): 869.771 nodes: 1.720.720.000

Eelco

[ernest]

32-bit, Core 2 Duo @3GHz (512MB hash, 3-4-5 TBs)
Analysis by Rybka 3 32-bit:
1...fxg3 2.hxg3[] Kd4 3.Kc1 Ke3 4.f4 gxf4 5.gxf4 Kxf4 6.Bh5 Ke3 7.Bg4 Ke4 8.Bh5 Ke3 9.Bg4 Ke4 10.Bh5 Ke3 11.Bg4 Ke4 12.Bh5 Ke3 13.Bg4 Ke4 14.Bh5 Ke3 15.Bg4 Ke4 16.Bh5 Ke3
-+ (-2.76) Depth: 26 00:00:36 3542kN, tb=2975
1...Kd4
-+ (-2.96 !) Depth: 26 00:01:05 7691kN, tb=4562

Analysis by Zappa Mexico II:
1...Kd4 2.gxf4 gxf4 3.h4 Ke3 4.Bc6 Bg6 5.Bd5 Kd3 6.Be6 Bh5
-+ (-2.89) Depth: 12/40 00:00:00 386kN, tb=2

[Howard E]

Thanks, looks like all the fruit/toga variants find this quickly.
Are your changes to toga more tactical overall?

[Howard E]

Thanks, Zappa Mexico is fast here.
The first line of analysis by Rybka3-32bit shows that no progress can be made from gxf3. Curious that my 32bit single cpu Rybka 3
finds Kd4 on the 28th ply, with 512 hash under Arena.

[Howard E]

Bright-0.4a: 8 threads, i7-920,512 hash

1.
29/67 04:28 3.975.203,023 14.818 +3.15 fxg3 hxg3 Kd4 Bc6 Ke3 f4 gxf4 gxf4 Kxf4 Bd5 Ke3 Kc1 Bf5 Bc6 Kd4 Bb5 Bd3 Bc6 Ba6 Bd7 Bb7 Be8 Be4 Ba4 Bh7 Bd1 Bg8 Ba4 Be6 Bc6 Bd5 Bd7
29/67 06:56 2.011.837.135 4.826 +3.79 Kd4 gxf4 gxf4 Bc6 Ke3 h4 Bg6 Bb7 Kf2 Bd5 Kg3 Bc6 Kxh4 Ke2 Kg5 Bd7 Kf6 Bb5 Ke5 Ba4 Kd4 Bb5 Bc2 Ba6 Bb3 Bb7 Bd5 Bc8 Bc4+ Kd1 Bd3

[Howard E]

i7-920, 512 hash

Thinker-Inert-MPx64: infinite mode stopped after 10 minutes.

32 10:03 2.328.727,217 3.861 -2.44 fxg3 hxg3

Any insights on what makes Zappa and toga so quick to solution time?

[Eelco de Groot]

Thanks, looks like all the fruit/toga variants find this quickly.
Are your changes to toga more tactical overall?

Hello Howard,

Yes, probably the overall playing strength is still bad, but trying to improve the King attacks a little, so that is what I would call tactics although that is not a factor in this position. For instance 2 checks in q-search should help a little with King attacks, and I attempted to detect X-ray attacks on the King.

But the King attack is not really the strongest point of Fruit, so it is not spectacular in Blueberry at all but that also means there is some room for improvement there. Hiarcs and Glaurung, Rebel I think all are much better in King attacks ! Also, because it is based on Toga II 1.4 beta5c, tactics is not really quite as good as in the older Toga 1.2.1a, pruning rules were introduced for this version that help positionally and in getting greater general depth, but hurt a little in tactics sometimes. So that is again an area where some improvement should be possible but as always it is not easy not to let that hurt overall playing strength. I hardly have tested with games yet, I think Blueberry is far from ready for that If I knew Rybka's secret to get its much better depth and tactical strength, it would be a different matter Chess programming I feel is a little like alchemy, we have not yet discovered the periodic table and are sometimes trying to make gold from lead We need a fusion reactor for that, and I believe lead is exactly the worst possible candidate element they could think of for fusion reactions!

Eelco

[ernest]

Thinker-Inert-MPx64: infinite mode stopped after 10 minutes.

What kind of infinite mode, with Thinker?

[Howard E]

In Arena when one puts the time to move on infinite mode the engine searches until the user stops the search. Thinker will then give you its choice of move and a 2 ply output.