Not logged inRybka Chess Community Forum
Up Topic The Rybka Lounge / Test Positions / Even harder Morozevich v Vachier-Lagrave position
- - By CSullivan (**) [us] Date 2010-06-17 17:32
Don't try this unless you have some time on your hands (position from analysis of Morozevich vs. Vachier-Lagrave, Biel 2009):
r6r/4k3/p2bP2p/2n2PP1/2BQn2P/1pP1p3/1q1BK3/3R3R b - - 0 29

Black has a material advantage, but White is threatening to equalize (at least!) with 30.Qg7+.

Despite what your engine will tell you for at least a few minutes, the solution does NOT begin with 29...exd2.  Rybka3(4-cpu) took between 15 and 20 minutes (Q6600@3GHz/1GB hash) before discovering that 30.Rh3!! saves the game: 30...Rhf8 (30...Bg3 31.f6+ Ke8 32.Bd5! leads to a very complicated perpetual-check draw) 31.f6+ Nxf6 32.gxf6+ Rxf6 33.Rxd2 Rf2+ 34.Kxf2 Qxd2+ 35.Qxd2 Ne4+ 36.Ke1 Nxd2 37.Kxd2 (approximately -1.00, according to both Rybka and Zappa)
r7/4k3/p2bP2p/8/2B4P/1pP4R/3K4/8 b - - 0 37

and neither Rybka nor Zappa found a win in this position of equal material and opposite-colored bishops.

Stockfish 1.7.1 (64-bit) seems relatively clueless about the correct solution: using 1 core, it still shows 29...exd2 (-0.92, full 27 plies) as best after 38 hours; using 4 cores, it starts to see the correct solution (-0.92, 28th ply) at 13 hours!  At one point, Stockfish scores 29...exd2 as high as -4.00!

It takes a while before Rybka starts to fail high with the correct 29th move, and it can take a very long time before reaching a winning score (approximately -2.00):
Rybka3 (1-cpu, 256MB hash) reaches -1.99 on the 18th ply and 18 hours.
Rybka3 (1-cpu, 1GB hash) reaches -1.77 on the 17th ply and 7 hours.
Rybka3 (4-cpu, 1GB hash) reaches -2.01 on the 17th ply and 2 hours or 2.5 hours [I ran this twice].
Zappa Mexico II (64-bit) takes about twice as long as Rybka3 to reach -2.00 with the correct variation.

I will post the solution in a day or two, depending on interest (or lack thereof!).
Parent - - By Felix Kling (Gold) [de] Date 2010-06-17 17:51
Qc2... needed to help Rybka a bit :) I guess the cluster would solve this one rather quickly.
Parent - By CSullivan (**) [us] Date 2010-06-20 16:29
Felix,
I suspect that a GM might find the solution to this faster than engines -- at least, engines running on "normal" hardware.   It appears that your assistance was right on the money!
Charles
Parent - - By yanquis1972 (****) [us] Date 2010-06-17 20:05
rybka 4 appears to blow R3 out of the water on finding the first key move --

Analysis by Deep Rybka 4 x64 AN:

1...Raf8 2.f6+ Nxf6 3.gxf6+ Rxf6 4.Rhg1 Rh7 5.Qxe3 Qc2 6.Rc1 Qf5 7.Rcf1 Qh5+ 8.Ke1
  =/+  (-0.65)   Depth: 6   00:00:00  113kN
1...Raf8
  -/+  (-0.77 ++)   Depth: 7   00:00:00  124kN
1...Raf8
  -/+  (-0.92 ++)   Depth: 7   00:00:00  133kN
1...Raf8 2.f6+ Nxf6 3.gxf6+ Rxf6 4.Rhg1 Rh7 5.Qxe3 Qc2 6.Rde1 b2 7.Ba2 Qf5 8.Rb1 Bf4
  -/+  (-0.77)   Depth: 7   00:00:00  133kN
1...Raf8 2.f6+ Nxf6 3.gxf6+ Rxf6
  -/+  (-0.89 ++)   Depth: 8   00:00:00  174kN
1...exd2 2.Qg7+ Kd8 3.Qxh8+ Kc7 4.Qxa8 Nxc3+ 5.Kf3 Qc2 6.Qa7+ Kc6 7.Qa8+ Kb6
  -/+  (-1.04 ++)   Depth: 8   00:00:00  284kN
1...exd2 2.Qg7+ Kd8[] 3.Qxh8+ Kc7 4.Qxa8 Qxc3 5.Bxa6 Ng3+ 6.Kf2 Qd4+ 7.Kg2 Nxh1 8.Qa7+ Kc6 9.Bb7+ Kb5 10.Ba6+ Kb4 11.Qb6+ Kc3 12.Qa5+ Kc2
  -/+  (-0.91)   Depth: 8   00:00:01  325kN
1...exd2 2.Qg7+ Kd8[] 3.Qxh8+ Kc7 4.Qxa8 Qxc3 5.Bxa6 Ng3+ 6.Kf2 Qd4+ 7.Kg2 Nxh1 8.Qa7+ Kc6 9.Bb7+ Kb5 10.Ba6+ Kb4 11.Qb6+ Kc3 12.Qa5+ Kc2
  -/+  (-1.03 ++)   Depth: 9   00:00:01  416kN
1...exd2 2.Qg7+ Kd8[] 3.Qxh8+ Kc7 4.Qxa8 Qxc3 5.Bxa6 Ng3+ 6.Kf2 Qd4+ 7.Kg2 Nxh1 8.Qa7+ Kc6 9.Bb7+ Kb5 10.Ba6+ Kb4 11.Qb6+ Kc3 12.Qa5+ Kc2
  -/+  (-1.18 ++)   Depth: 9   00:00:01  472kN
1...exd2 2.Qg7+ Kd8[] 3.Qxh8+ Kc7 4.Qxa8 Qxc3 5.Bxa6 Ng3+ 6.Kf2 Qd4+ 7.Kg2 Nxh1 8.Qa7+ Kc6 9.Bb7+ Kb5 10.Ba6+ Kb4 11.Qb6+ Kc3 12.Qa5+ Kc2
  -+  (-1.58 ++)   Depth: 9   00:00:02  775kN
1...exd2 2.Qg7+ Kd8[] 3.Qxh8+ Kc7 4.Qg7+ Kb6 5.Qd4 hxg5 6.hxg5 Rd8 7.Rh3 Be7 8.Bd5 Nxg5 9.Rxd2 Qb1
  -/+  (-1.00)   Depth: 9   00:00:02  973kN
1...exd2 2.Qg7+ Kd8[] 3.Qxh8+ Kc7 4.Qg7+ Kb6 5.Qd4 hxg5 6.hxg5 Rd8 7.Rh3 Be7 8.Bd5 Rxd5 9.Qxd5 Bxg5 10.Qc4 Bf6
  -/+  (-1.12 ++)   Depth: 10   00:00:03  1092kN
1...exd2 2.Qg7+ Kd8[] 3.Qxh8+ Kc7 4.Qg7+ Kb6 5.Qd4 hxg5 6.hxg5 Rd8
  -/+  (-1.27 ++)   Depth: 10   00:00:03  1147kN
1...exd2 2.Qg7+ Kd8[] 3.Qxh8+ Kc7[] 4.Qg7+ Kb6[] 5.Qd4 hxg5 6.hxg5 Rd8 7.Rh3 Be7 8.Bd5 Rxd5 9.Qxd5 Bxg5 10.Qc4 Bf6 11.Qb4+ Kc6 12.Kf1 Nxc3 13.e7
  -+  (-1.67 ++)   Depth: 10   00:00:03  1278kN
1...exd2 2.Qg7+ Kd8[] 3.Qxh8+ Kc7[] 4.Qg7+ Kb6[] 5.Qd4 hxg5 6.hxg5 Rd8 7.Rh3 Be7 8.Qg7 Bxg5 9.e7 Rc8 10.Bd5 Qc2 11.Qg6+ Ka7 12.Qf7 Qxd1+ 13.Kxd1 Nf2+ 14.Ke2 d1Q+ 15.Kxf2 Qd2+ 16.Kf1 Kb6
  -+  (-1.88)   Depth: 10   00:00:03  1456kN
1...exd2 2.Qg7+ Kd8[] 3.Qxh8+ Kc7[] 4.Qg7+ Kb6 5.Qd4 hxg5 6.hxg5 Rd8 7.Rh3 Be7 8.Qg7 Bxg5 9.e7 Rc8 10.Bd5 Qc2 11.Qg6+ Ka7 12.Qf7 Qxd1+ 13.Kxd1 Nf2+ 14.Ke2 d1Q+ 15.Kxf2 Qd2+ 16.Kf1 Kb6
  -+  (-1.88)   Depth: 11   00:00:04  1659kN
1...exd2 2.Qg7+[] Kd8[] 3.Qxh8+ Kc7[] 4.Qg7+ Kb6 5.Qd4 hxg5 6.hxg5 Rd8 7.Rh3 Be7 8.Qg7 Bxg5 9.e7 Rc8 10.Bd5 Qc2 11.Qg6+ Ka7
  -+  (-1.88)   Depth: 12   00:00:07  2833kN
1...exd2 2.Rh3 Rhf8 3.f6+ Nxf6 4.gxf6+ Rxf6 5.Rxd2 Rf2+[] 6.Kxf2 Qxd2+
  -/+  (-0.98)   Depth: 13   00:00:29  10472kN
1...exd2 2.Rh3 Rhf8 3.f6+ Nxf6 4.gxf6+ Rxf6 5.Rxd2 Rf2+ 6.Kxf2[] Qxd2+ 7.Qxd2[] Ne4+[]
  -/+  (-0.93)   Depth: 14   00:00:41  14386kN
1...exd2 2.Rh3 Rhf8 3.f6+ Nxf6 4.gxf6+ Rxf6 5.Rxd2[] Rf2+ 6.Kxf2[] Qxd2+ 7.Qxd2 Ne4+[] 8.Ke2 Nxd2[] 9.Kxd2 Bf4+ 10.Kd3 Rb8 11.Rf3 Bd6 12.Rf1 a5 13.Bd5 Rb6 14.Rf7+ Ke8 15.Rb7 Rxb7 16.Bxb7 Ke7
  -/+  (-0.93)   Depth: 15   00:01:08  22917kN

i'll let it run for a bit to see if it finds the answer to it.
Parent - - By yanquis1972 (****) [us] Date 2010-06-17 20:26
hung (not in the sense R4 users are used to :mad:) on Qc2 for some minutes before it finally officially popped up in the pv --

  -/+  (-0.93)   Depth: 16   00:02:24  49047kN
1...exd2 2.Rh3[] Rhf8 3.f6+ Nxf6 4.gxf6+ Rxf6 5.Rxd2 Rf2+ 6.Kxf2[] Qxd2+ 7.Qxd2
  -/+  (-0.92)   Depth: 17   00:05:55  125mN, tb=23
1...Qc2 2.Qg7+ Kd8 3.Qxh8+ Kc7
  -/+  (-1.05 ++)   Depth: 17   00:17:25
407mN, tb=28
1...Qc2 2.Qg7+ Kd8[]
  -/+  (-1.20 ++)   Depth: 17   00:19:32  462mN, tb=29
1...Qc2 2.Qg7+ Kd8[] 3.Qxh8+ Kc7 4.Qxa8 exd2[]
  -+  (-1.60 ++)   Depth: 17   00:23:32  564mN, tb=40

(also 4x3ghz, but 2GB hash)
Parent - - By CSullivan (**) [us] Date 2010-06-20 17:08 Edited 2010-06-20 17:10

> 1...Qc2 2.Qg7+ Kd8[] 3.Qxh8+ Kc7 4.Qxa8 exd2[]
>    -+  (-1.60 ++)   Depth: 17   00:23:32  564mN, tb=40


Very interesting....
After 23 minutes, Rybka4 still has not quite hit upon the winning variation, although
r6r/4k3/p2bP2p/2n2PP1/2BQn2P/1pP1p3/1q1BK3/3R3R b - - 0 29

29...Qc2!! is the right first move.
(A) If White accepts the double-rook sacrifice, then the best line is 30.Qg7+ Kd8 31.Qxh8 Kc7 32.Qxa8

Now the best move is NOT Rybka4's 32...exd2 [how can this be "forced"?], when finding a win for Black is very difficult; for example: 33.Qa7+ Nb7 34.Qd4 Nxc3+ 35.Kf3 Nxd1 36.Bd3 Qc1 37.e7 Bxe7 38.Qg7 Qc6+ 39.Ke2 Qg2+ 40.Kxd1 Qxh1+ 41.Kxd2 Qh2+ 42.Ke3 Kd7 43.f6
8/1n1kb1Q1/p4P1p/6P1/7P/1p1BK3/7q/8 b - - 0 43

and this position is probably a draw.

Rather, the winning continuation is 32...Nxd2! 33.Qa7+ Kc6 34.Qa8+ Kb6 35.Qd8+ Ka7 36.Qxd6 Nde4+ 37.Rd2 exd2 38.Qc7+ Nb7 39.Ke3 Nxc3! 40.e7 d1=Q 41.Rxd1 Qe4+ 42.Kf2 Nxd1+.

(B) The other main variation (which also requires precise calculation) is 30.Kxe3 Raf8 31.Rdf1 Kd8!! 32.Bd5 Nxd2 33.Qxd2 Qxd2+ 34.Kxd2 hxg5 35.hxg5 Rxh1 36.Bxh1 Nxe6 and Black has a winning endgame.
Parent - By yanquis1972 (****) [us] Date 2010-06-20 17:47
thanks for the update c-- very interesting indeed! i decided to run the position fresh to see how long it would take rybka to get the full variation -- what follows is a great example of SMP randomness...(the only change i've made to the engine is setting tablebase usage to 'normally' (which i doubt had any real effect in this position) - but you'd think its a completely different engine altogether!) --

rybka gets the correct rook sac variation before deciding white's better off declining:

r6r/4k3/p2bP2p/2n2PP1/2BQn2P/1pP1p3/1q1BK3/3R3R b - - 0 1


Analysis by Deep Rybka 4 x64 AN:

1...exd2 2.Qg7+ Kd8 3.Qxh8+ Kc7 4.Qxa8 Nxc3+ 5.Kf3 Qc2 6.Qa7+ Kc6 7.Qa8+ Kb6 8.Qd8+ Bc7 9.Qd4 Nxd1 10.Bd3 Qc1 11.Qb4+ Ka7 12.Bc4 Nxe6 13.fxe6 Nc3 14.Qc5+ Kb8 15.Qf8+ Kb7 16.Qb4+ Kc6
  -/+  (-1.01)   Depth: 6   00:00:00  0kN
1...exd2 2.Qg7+ Kd8 3.Qxh8+ Kc7 4.Qxa8 Nxc3+ 5.Kf3 Qc2 6.Qa7+ Kc6 7.Qa8+ Kb6 8.Qd8+ Bc7 9.Qd4 Nxd1 10.Bd3 Qc1 11.Qb4+ Ka7 12.Bc4 Nxe6 13.fxe6 Nc3 14.Qc5+ Kb8 15.Qf8+ Kb7 16.Qb4+ Kc6
  -/+  (-1.01)   Depth: 7   00:00:00  0kN
1...exd2 2.Qg7+ Kd8[] 3.Qxh8+ Kc7[] 4.Qxa8 Nxc3+ 5.Kf3 Qc2 6.Qa7+ Kc6 7.Qa8+ Kb6 8.Qd8+ Bc7 9.Qd4 Nxd1 10.Bd3 Qc1 11.Qb4+ Ka7 12.Bc4 Nxe6 13.fxe6 Nc3 14.Qc5+ Kb8 15.Qf8+ Kb7 16.Qb4+ Kc6
  -/+  (-1.01)   Depth: 8   00:00:00  0kN
1...exd2 2.Qg7+ Kd8[] 3.Qxh8+ Kc7[] 4.Qxa8 Nxc3+ 5.Kf3 Qc2 6.Qa7+ Kc6 7.Qa8+ Kb6 8.Qd8+ Bc7 9.Qd4 Nxd1 10.Bd3 Qc1 11.Qb4+ Ka7 12.Bc4 Nxe6 13.fxe6 Nc3 14.Qc5+ Kb8 15.Qf8+ Kb7 16.Qb4+ Kc6
  -/+  (-1.01)   Depth: 9   00:00:00  0kN
1...exd2 2.Qg7+ Kd8[] 3.Qxh8+ Kc7[] 4.Qxa8 Nxc3+ 5.Kf3 Qc2 6.Qa7+ Kc6 7.Qa8+ Kb6 8.Qd8+ Bc7 9.Qd4 Nxd1 10.Bd3 Qc1 11.Qb4+ Ka7 12.Bc4 Nxe6 13.fxe6 Nc3 14.Qc5+ Kb8 15.Qf8+ Kb7 16.Qb4+ Kc6
  -/+  (-1.13 ++)   Depth: 10   00:00:00  71kN
1...exd2 2.Qg7+ Kd8[] 3.Qxh8+ Kc7[] 4.Qxa8 Nxc3+ 5.Kf3 Qc2 6.Qa7+ Kc6 7.Qa8+ Kb6 8.Qd8+ Bc7 9.Qd4 Nxd1 10.Bd3 Qc1 11.Qb4+ Ka7 12.Bc4 Nxe6 13.fxe6 Nc3 14.Qc5+ Kb8 15.Qf8+ Kb7 16.Qb4+ Kc6
  -/+  (-1.28 ++)   Depth: 10   00:00:00  111kN
1...exd2 2.Qg7+ Kd8[] 3.Qxh8+ Kc7[] 4.Qxa8 Nxc3+ 5.Kf3 Qc2 6.Qa7+ Kc6 7.Qa8+ Kb6 8.Qd8+ Bc7 9.Qd4 Nxd1 10.Bd3 Qc1 11.Qb4+ Ka7 12.Bc4 Nxe6 13.fxe6 Nc3 14.Qc5+ Kb8 15.Qf8+ Kb7 16.Qb4+ Kc6
  -+  (-1.63)   Depth: 10   00:00:01  572kN
1...exd2 2.Qg7+ Kd8[] 3.Qxh8+ Kc7[] 4.Qg7+ Kb6[] 5.Qd4 hxg5 6.hxg5 Be7 7.Rh3 Rd8 8.Bd5 Rxd5 9.Qxd5 Bxg5
  -+  (-1.66)   Depth: 11   00:00:02  1116kN
1...exd2 2.Qg7+ Kd8[] 3.Qxh8+ Kc7[] 4.Qg7+ Kb6[] 5.Qd4 hxg5[] 6.hxg5 Be7 7.Rh3 Rd8[] 8.Qg7 Bxg5 9.e7 Rc8 10.Bd5 Qc2 11.Qg6+ Nf6 12.Qxg5 Nxd5 13.Rxd2 Qe4+ 14.Re3 Nxe3 15.Qxe3 Qg4+ 16.Ke1 Qh4+
  -+  (-1.78 ++)   Depth: 12   00:00:10  3932kN
1...exd2 2.Rh3[] Rhf8 3.f6+ Nxf6 4.gxf6+ Rxf6 5.Rxd2 Rf2+ 6.Kxf2[] Qxd2+ 7.Qxd2 Ne4+ 8.Ke2 Nxd2 9.Kxd2 Bf4+ 10.Kd3 Rd8+ 11.Ke4 b2 12.Ba2 Bc7 13.Rf3 a5 14.Rf7+ Kd6 15.e7 Rb8 16.Kd3 a4
  -/+  (-0.74)   Depth: 12   00:00:31  10758kN
1...exd2 2.Rh3[] Rhf8 3.f6+ Nxf6 4.gxf6+ Rxf6 5.Rxd2 Rf2+ 6.Kxf2[] Qxd2+ 7.Qxd2 Ne4+ 8.Ke2 Nxd2[] 9.Kxd2 Bf4+ 10.Kd3 Rd8+ 11.Ke4 b2 12.Ba2 Bc7 13.Rf3 a5 14.Rf7+ Kd6 15.e7 Rb8 16.Kd3 a4
  -/+  (-0.74)   Depth: 13   00:00:34  11900kN
1...exd2 2.Rh3[] Rhf8 3.f6+ Nxf6 4.gxf6+ Rxf6 5.Rxd2 Rf2+ 6.Kxf2[] Qxd2+ 7.Qxd2 Ne4+ 8.Ke2 Nxd2[] 9.Kxd2 Bf4+ 10.Kd3 Rd8+ 11.Ke4 b2 12.Ba2 Bc7 13.Rf3 a5 14.Rf7+ Kd6 15.e7 Rb8 16.Kd3 a4
  -/+  (-0.74)   Depth: 14   00:00:50  16712kN
1...exd2 2.Rh3[] Rhf8 3.f6+ Nxf6 4.gxf6+ Rxf6 5.Rxd2[] Rf2+ 6.Kxf2[] Qxd2+ 7.Qxd2 Ne4+[] 8.Ke2 Nxd2[] 9.Kxd2 Bf4+ 10.Kd3 Rd8+ 11.Ke4 b2 12.Ba2 Bc7 13.Rf3 a5 14.Rf7+ Kd6 15.e7 Rb8 16.Kd3 a4
  -/+  (-0.74)   Depth: 15   00:01:11  23627kN
1...exd2 2.Rh3[] Rhf8 3.f6+ Nxf6 4.gxf6+ Rxf6 5.Rxd2[] Rf2+ 6.Kxf2[] Qxd2+ 7.Qxd2 Ne4+[] 8.Ke2 Nxd2[] 9.Kxd2 Bf4+ 10.Kd3 Rd8+ 11.Ke4[] b2 12.Ba2[] Bc7 13.Rd3 a5 14.Rxd8 Kxd8 15.Kd4 Bd6 16.c4 Ba3
  -/+  (-0.77)   Depth: 16   00:02:23  45646kN, tb=17
1...Qc2 2.Qg7+ Kd8[] 3.Qxh8+ Kc7 4.Qxa8 Nxd2 5.Qa7+ Kc6 6.Qa8+ Kb6 7.Qd8+ Ka7 8.Qxd6 Nde4+ 9.Rd2 exd2
  -/+  (-0.86 ++)   Depth: 16   00:06:01  127mN, tb=20

1...Qc2 2.Qg7+ Kd8[] 3.Qxh8+ Kc7 4.Qxa8 Nxd2 5.Qa7+ Kc6 6.Qa8+ Kb6
  -/+  (-1.01 ++)   Depth: 16   00:08:28  185mN, tb=104
1...Qc2 2.Qg7+ Kd8[] 3.Qxh8+ Kc7 4.Qxa8 Nxd2 5.Qa7+ Kc6 6.Qa8+ Kb6
  -+  (-1.41 ++)   Depth: 16   00:13:00  289mN, tb=259
1...Qc2 2.Kxe3 Raf8[] 3.Rdf1 Qxd2+ 4.Qxd2 Nxd2 5.Kxd2 hxg5 6.hxg5 Rxh1 7.f6+ Ke8 8.Rxh1[] Bf4+ 9.Kd1[] Bxg5 10.Rh5[] b2 11.f7+ Ke7[] 12.Kc2[] Rb8[] 13.Kb1 Be3 14.Rh7 Kd6 15.Rg7 Ne4 16.f8Q+ Rxf8
  -+  (-1.54)   Depth: 16   00:19:52  430mN, tb=474
1...Qc2 2.Kxe3 Raf8[] 3.Rdf1 Qxd2+ 4.Qxd2 Nxd2 5.Kxd2 hxg5 6.hxg5 Rxh1 7.f6+ Ke8 8.Rxh1 Bf4+ 9.Kd1 Bxg5 10.Rh5 Be3 11.Re5 Rxf6 12.Rxe3[] Rf4 13.Be2 Ke7 14.Re5 Na4 15.Bxa6 Rf2 16.Ke1 Rg2
  -+  (-1.47)   Depth: 17   00:24:05  520mN, tb=532


really insane position.
Up Topic The Rybka Lounge / Test Positions / Even harder Morozevich v Vachier-Lagrave position

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill