A very hard endgame study

[zullil]

Thank you, Louis!

The last output- line of your engine deviates not until move nr. 32, ignoring transpositions, from the author's solution.

I always started backward solving too soon, thinking I might have missed something at the beginning.

Great study

But I'm surprised by SF's next iteration:

Code: Select all

info depth 46 seldepth 100 multipv 1 score cp 77 nodes 542694557106 nps 34912809 hashfull 999 tbhits 0 time 15544282 pv e8d7 f3d3 d7e6 d3e3 e6f7 e3f3 f7g8 f3e3 g8f8 e3f3 f8e8 f3e3 g2g1 g5g4 g1g2 e3f3 e8d7 f3d3 d7e6 d3e3 e6f7 e3f3 f7e8 f3e3 g2g1 g3g2 g1g2 g4g3 g2g1 e3d3 e8f7 d3f3 f7e6 f3e3 e6d7 e3d3 d7e8 d3e3 g1c1 c4d3 e8d7 d3d2 c1f1 d2e2 f1f4 g3g2 f4g4 e2f1 e7e8q e3e8 d7e8 g2g1q g4g1 f1g1 e8f7 g1g2 f7g7 g2f3 g7h7

Is there a flaw in the study? Was going to stop, but this odd output has me curious.

[peter]

Code: Select all

info depth 46 seldepth 100 multipv 1 score cp 77 nodes 542694557106 nps 34912809 hashfull 999 tbhits 0 time 15544282 pv e8d7 f3d3 d7e6 d3e3 e6f7 e3f3 f7g8 f3e3 g8f8 e3f3 f8e8 f3e3 g2g1 g5g4 g1g2 e3f3 e8d7 f3d3 d7e6 d3e3 e6f7 e3f3 f7e8 f3e3 g2g1 g3g2 g1g2 g4g3 g2g1 e3d3 e8f7 d3f3 f7e6 f3e3 e6d7 e3d3 d7e8 d3e3 g1c1 c4d3 e8d7 d3d2 c1f1 d2e2 f1f4 g3g2 f4g4 e2f1 e7e8q e3e8 d7e8 g2g1q g4g1 f1g1 e8f7 g1g2 f7g7 g2f3 g7h7

Is there a flaw in the study? Was going to stop, but this odd output has me curious.

10.(4.)Kg8? instead of Ke8! simply loses the "plan" again, Louis. Replacing that in the line played out should be enough to bring it back into hash again.
It's a typical kind of overwriting good hash by bad hash in stand- alone computing, and the eval drops.
Probably it will rise again with ongoing computing, but if the bad hash is getting more and more, of course it can go on dropping further on to 0.00 also.
That's why I was in doubt for a while too, but I don't think anymore the study to suck , having seen the way of winning, I'm rather sure, it's just the engine that sucks dealing with it.

[Sven]

Code: Select all

info depth 46 seldepth 100 multipv 1 score cp 77 nodes 542694557106 nps 34912809 hashfull 999 tbhits 0 time 15544282 pv e8d7 f3d3 d7e6 d3e3 e6f7 e3f3 f7g8 f3e3 g8f8 e3f3 f8e8 f3e3 g2g1 g5g4 g1g2 e3f3 e8d7 f3d3 d7e6 d3e3 e6f7 e3f3 f7e8 f3e3 g2g1 g3g2 g1g2 g4g3 g2g1 e3d3 e8f7 d3f3 f7e6 f3e3 e6d7 e3d3 d7e8 d3e3 g1c1 c4d3 e8d7 d3d2 c1f1 d2e2 f1f4 g3g2 f4g4 e2f1 e7e8q e3e8 d7e8 g2g1q g4g1 f1g1 e8f7 g1g2 f7g7 g2f3 g7h7

Is there a flaw in the study? Was going to stop, but this odd output has me curious.

10.(4.)Kg8? instead of Ke8! simply loses the "plan" again, Louis. Replacing that in the line played out should be enough to bring it back into hash again.
It's a typical kind of overwriting good hash by bad hash in stand- alone computing, and the eval drops.
Probably it will rise again with ongoing computing, but if the bad hash is getting more and more, of course it can go on dropping further on to 0.00 also.
That's why I was in doubt for a while too, but I don't think anymore the study to suck , having seen the way of winning, I'm rather sure, it's just the engine that sucks dealing with it.

10.Kg8 is only a delay of four plies, after 12.Ke8 the PV follows the original solution of the author again. 26...Ke2 (28...Ke2 with the delay) deviates from it (original Rd3+) but the PV above is still correct even there, its final position is clearly won for white (trivial pawn ending), only the score does not reflect it, probably since there are no passers on the board. Well, that's really hard, we are around ply 70 from the root already ...

[peter]

[10.Kg8 is only a delay of four plies, after 12.Ke8 the PV follows the original solution of the author again. 26...Ke2 (28...Ke2 with the delay) deviates from it (original Rd3+) but the PV above is still correct even there, its final position is clearly won for white (trivial pawn ending), only the score does not reflect it, probably since there are no passers on the board. Well, that's really hard, we are around ply 70 from the root already ...

[lech]

I gave one hour for my newest (non-published) version of Sting,
My old dual core returns:
[d]8/p6p/4P2p/p3K1pP/pPk5/p2r2p1/P5R1/8 w - - bm b5;

Code: Select all

info depth 29 multipv 1 score cp 0 nodes 860381906 nps 872197 time 986453 pv b4a5 d3e3 e5d6 e3d3 d6e5
info depth 30
info nodes 860381906 nps 872197 time 986453
info currmove b4a5 currmovenumber 1
info nodes 1041401252 nps 867834 time 1200000
info currmove b4b5 currmovenumber 2
info nodes 1193789818 nps 867974 time 1375375
info currmove e6e7 currmovenumber 3
info nodes 1206888595 nps 868119 time 1390234
info currmove g2e2 currmovenumber 4
info nodes 1496332690 nps 854589 time 1750938
info currmove g2b2 currmovenumber 5
info currmove g2g1 currmovenumber 6
info nodes 1499442239 nps 854603 time 1754547
info currmove e5e4 currmovenumber 7
info nodes 1522415475 nps 854457 time 1781734
info currmove g2d2 currmovenumber 8
info currmove g2c2 currmovenumber 9
info nodes 1620269386 nps 855206 time 1894594
info currmove g2h2 currmovenumber 10
info currmove e5f6 currmovenumber 11
info nodes 1630238892 nps 855242 time 1906172
info currmove e5f5 currmovenumber 12
info nodes 1631003136 nps 855229 time 1907094
info currmove g2f2 currmovenumber 13
info currmove g2g3 currmovenumber 14
info depth 30 multipv 1 score cp 0 nodes 1631420366 nps 855245 time 1907547 pv b4a5 d3e3 e5d6 e3d3 d6e5
info depth 31
info nodes 1631420366 nps 855238 time 1907563
info currmove b4a5 currmovenumber 1
info nodes 1705816365 nps 856279 time 1992125
info currmove b4b5 currmovenumber 2
info depth 31 multipv 1 score cp 8 lowerbound nodes 2541069848 nps 862475 time 2946250 pv b4b5
info nodes 2541069848 nps 862475 time 2946250
info currmove b4b5 currmovenumber 1
info depth 31 multipv 1 score cp 8 nodes 3115713157 nps 865423 time 3600219 pv b4b5 g5g4 e5e4 d3d4 e4f5 d4d3 e6e7 d3e3
info nodes 3115713157 nps 865423 time 3600219
bestmove b4b5 ponder g5g4

I think the future version of Sting will get it, but a stronger computer will be welcome.

BTW. Some words about my work on the next version of Sting. I found many new bugs in the code of based Stockfish2.1.1 .
It was easy to correct it.

I found a good code for the walking (maybe rather LAZY) King. and Sting hold on Stockfish's 2.1.1 strength (++).
They are the good news.

But there is some other strange bug in the code of Stockfish 2.1.1 and I can't get it.
I tried the newest version of Stockfish 6 and it looks that the same bug is continued there too.