Endgame best move

[swami]

FEN: 4r3/7R/3K1k2/3RpP2/4P3/6r1/8/8 w - - 0 1

Bright-0.3a:
18/46 00:07 88.140.442 11.488 +2.92 Rh6+ Kf7 Re6 Rxe6+ fxe6+ Ke8 Rc5 Rd3+ Kxe5 Rh3 Rc8+ Ke7 Rc7+ Ke8 Rg7 Rh2 Rb7 Rh1 Rd7 Rh3 Ra7 Rh2 Rf7 Rb2 Rg7
19/48 00:17 213.110.154 12.387 +2.92 Rh6+ Kf7 Re6 Rxe6+ fxe6+ Ke8 Rc5 Rd3+ Kxe5 Rh3 Rc8+ Ke7 Rc7+ Ke8 Rg7 Rh2 Rb7 Rh1 Rd7 Rh3 Ra7 Rh1 Rb7 Re1 Kd6 Rd1+ Kc5
20/52 00:48 621.145.661 12.827 +2.92 Rh6+ Kf7 Re6 Rxe6+ fxe6+ Ke8 Rc5 Rd3+ Kxe5 Rh3 Rc8+ Ke7 Rc7+ Ke8 Rg7 Rh2 Rb7 Rh1 Rd7 Rf1 Rg7 Rb1 Rh7 Rc1 Ra7 Re1 Kd6 Rd1+ Kc5
21/52 01:26 1.126.341.993 12.955 +2.92 Rh6+ Kf7 Re6 Rxe6+ fxe6+ Ke8 Rc5 Rd3+ Kxe5 Rh3 Rc8+ Ke7 Rc7+ Ke8 Rg7 Rh2 Rb7 Rh1 Rd7 Rf1 Rg7 Rb1 Rf7 Rc1 Ra7 Re1 Kd6 Rd1+ Kc5
22/54 04:37 3.604.213,823 12.973 +2.92 Rh6+ Kf7 Re6 Rxe6+ fxe6+ Ke8 Rc5 Rd3+ Kxe5 Rh3 Rc8+ Ke7 Rc7+ Ke8 Rg7 Rh2 Rb7 Rh1 Rd7 Rf1 Rg7 Rb1 Rf7 Rc1 Ra7 Rh1 Rb7 Re1 Kd6 Rd1+ Kc5
23/62 10:04 3.547.233,483 5.870 +2.92 Rh6+ Kf7 Re6 Rxe6+ fxe6+ Ke8 Rc5 Rd3+ Kxe5 Rh3 Rc8+ Ke7 Rc7+ Ke8 Rg7 Rh2 Rb7 Rh1 Rd7 Rf1 Rg7 Rb1 Rf7 Rc1 Rh7 Re1 Ra7 Re2 Kd6 Rd2+ Kc5
23/62 15:34 3.596.190,950 3.847 +2.92 Rh6+ Kf7 Re6 Rxe6+ fxe6+ Ke8 Rc5 Rd3+ Kxe5 Rh3 Rc8+ Ke7 Rc7+ Ke8 Rg7 Rh2 Rb7 Rh1 Rd7 Rf1 Rg7 Rb1 Rf7 Rc1 Rh7 Re1 Ra7 Re2 Kd6 Rd2+ Kc5

No luck with Bright 0.3a
8 threads
core i7 - 920
512 hash
bitbases on (5 piece)

do these times seem to be consistent for 8 threads? (8 threads for a quad with hyper threading turned on)

Maybe the older version of Bright was missing something that prevents itself from finding the move?

CCT version of Bright is about 75 elo better than 0.3a or may even be more. Allard told me at ICC that he had improved some endgame knowledge with the last version, so 75 improvement is attributable to those endgame knowledge additions...

Thanks for running the test with Bright 0.3a btw...

[ernest]

Hi Howard,

The final word is: Rh6+ wins, Kd7 only draws

At least that comes out of the German CSS-Forum, where some experts studied the position in detail.

http://forum.computerschach.de/cgi-bin/ ... l?tid=1013

[Howard E]

Thanks for the correction and link.
I did not save the Rybka analysis but will let it go longer next time.
Also, I think I will use this position to have engines play it out in a game setting, using a long time control.
Anyway to translate the findings from German to English?
Or maybe you could post the main lines here.

[Howard E]

1.Th6+ Kf7 2.Kd7 Tb8 3.Tdd6 Tb7+ 4.Kc8 Tb4 5.Td7+ Kg8 6.f6 Tc3+ 7.Kd8 Tb8+ 8.Ke7 Tc1 9.Ke6 Tb6+ 10.Kxe5 Tb5+ 11.Ke6 Tb6+ 12.Ke7 Tb8 13.e5 Te1 14.Td8+ Txd8 15.Tg6+ Kh7 16.Tg7+ Kh6
+- (#40) Tiefe: 24/67 04:54:59 26169mN, tb=12490991

I just cut and pasted this from Zappa Mexico's analysis, an engine I do not have.

I was totally blind to white's 2nd move, Kd7. So many engines prefer Re6 here, followed by the pawn capture. Among them Rybka3 - 32bit and Toga 2 1.4se.

I appreciate the lengthy analysis from the testers on that site.
It is a fun position that came from a home engine vs engine game.

[ernest]

1.Th6+ Kf7 2.Kd7 Tb8 3.Tdd6 Tb7+ 4.Kc8 Tb4 5.Td7+ Kg8 6.f6 Tc3+ 7.Kd8 Tb8+ 8.Ke7 Tc1 9.Ke6 Tb6+ 10.Kxe5 Tb5+ 11.Ke6 Tb6+ 12.Ke7 Tb8 13.e5 Te1 14.Td8+ Txd8 15.Tg6+ Kh7 16.Tg7+ Kh6
+- (#40) Tiefe: 24/67 04:54:59 26169mN, tb=12490991

I just cut and pasted this from Zappa Mexico's analysis, an engine I do not have.

Yes, and you can add the refutation of 1.Kd7:
(also, don't forget, all this analysis was done with the full 6-piece Nalimov...)

Code: Select all

4r3/3K3R/5k2/3RpP2/4P3/6r1/8/8 b - - 0 1
Analysis by Zappa Mexico II:
1...Tgg8 2.Td6+ Kg5 3.Kc6 Ta8 4.Kd5 Ta5+ 5.Ke6 Tf8 6.Ke7 Tg8 7.Td7 Taa8 8.Kd6 Kf4 9.Kd5 Tge8 10.Th4+ Kg3 11.Th1 Ta5+
  +/=  (0.68)   Tiefe: 24/62   01:33:27  16139mN, tb=4434305
1...Tgg8 2.Td6+ Kg5 3.Kc6 Ta8 4.Kd5 Ta5+ 5.Ke6 Tf8 6.Ke7 Tg8 7.Td7 Taa8 8.Kd6 Kf4 9.Kd5 Tge8 10.Th4+ Kg3 11.Th1 Ta5+ 12.Kc6 Kf4 13.Th4+ Kg5 14.Th2 Kf4 15.Tf2+ Kxe4 16.Te2+ Kxf5 17.Tf2+ 
  +/=  (0.68)   Tiefe: 25/63   01:44:38  18540mN, tb=6491361
and even on longer time basis:
1...Tgg8 2.Td6+ Kg5 3.Tg6+ Txg6 4.Kxe8 Tb6 5.Th1 Tb7 6.Tc1 Kf4 7.Tc4 Kg5 8.Ta4 Th7 9.Ta1 Tb7 10.Ta5 Kf4 11.Kd8 
  +/=  (0.69)   Tiefe: 26/67   14:41:36  55844mN, tb=21731242
1...Tgg8 2.Td6+ Kg5 3.Tg6+ Txg6 4.Kxe8 Tb6 5.Th1 Tb7 6.Tc1 Kf4 7.Tc4 Kg5 8.Ta4 Th7 9.Ta1 Tb7 
  +/=  (0.69)   Tiefe: 26/67   15:43:49  58680mN, tb=22893820

As you know, you can get some translation of the German comments with Google tools...

I appreciate the lengthy analysis from the testers on that site.

I think, the CSS-Forum "testers" also did appreciate that very interesting position. So thanks to you too!