1.Rb1! (the only winning move)

[biscuit1953]

D. Gurgenidze L. Mitrofanov (First Prize, Molodoi Leninets 1982)

Only a couple of the top engines I tested could find the winning move in less than 5 minutes.
Houdini 1.5 (10 seconds)
Critter 1.4 (12 seconds)
8/8/8/2pK3p/8/5N2/1P4pq/5Rbk w - - 0 1

[Harvey Williamson]

Hiarcs 14 goes for Rb1 instantly

[biscuit1953]

Does that include the winning score or did it take longer for that?

[Harvey Williamson]

Does that include the winning score or did it take longer for that?

Winning score took a little longer

[syzygy]

I was wondering why my engine finds Rb1 immediately with a score of +3.6 after 0.08s and a score of +6.1 after 3.2s, until I realised I had disabled null move for testing on some other position.

With null move enabled, it happens to play Rb1 as well, but the score quickly goes to +0.0.

[biscuit1953]

I was wondering why my engine finds Rb1 immediately with a score of +3.6 after 0.08s and a score of +6.1 after 3.2s, until I realised I had disabled null move for testing on some other position.

With null move enabled, it happens to play Rb1 as well, but the score quickly goes to +0.0.

Actually Graham Banks explains this on page 414 of "The Mammouth Book of Chess." He says,

"Unfortunately, the standard algorithms to prune out the variation tree are very likely to prune out the winning line, as the moves only make sense when you have seen the end of the line. Black's plan is to play ...c4, ...h4, ...h3, give up his c-pawn, sacrifice his queen (by checking the white king) and then to play ...h2, when White would have no good way to lift the stalemate."

"The older engines didn't get very far with this study, either considering the position drawn, or claiming that a number of rook moves win for White (whereas only one of them does). Only Fritz 6 with selectivity set to zero (i.e. less pruning) found the solution, but I only knew to activate this option because I already knew what the position involved - and the normal reason for using an analysis engine is to tell you something you don't already know!"

[biscuit1953]

I was wondering why my engine finds Rb1 immediately with a score of +3.6 after 0.08s and a score of +6.1 after 3.2s, until I realised I had disabled null move for testing on some other position.

With null move enabled, it happens to play Rb1 as well, but the score quickly goes to +0.0.

Actually Graham Burgess explains this on page 414 of "The Mammouth Book of Chess." He says,

"Unfortunately, the standard algorithms to prune out the variation tree are very likely to prune out the winning line, as the moves only make sense when you have seen the end of the line. Black's plan is to play ...c4, ...h4, ...h3, give up his c-pawn, sacrifice his queen (by checking the white king) and then to play ...h2, when White would have no good way to lift the stalemate."

"The older engines didn't get very far with this study, either considering the position drawn, or claiming that a number of rook moves win for White (whereas only one of them does). Only Fritz 6 with selectivity set to zero (i.e. less pruning) found the solution, but I only knew to activate this option because I already knew what the position involved - and the normal reason for using an analysis engine is to tell you something you don't already know!"

I meant to say Graham Burgess.

[MrEdCollins]

Interestingly, Spike 1.4 also finds Rb1, almost instantly. It takes about three seconds on my machine, however, before this engine confirms the move is best and gives a large score for White.

1 [-2.41] 1.Nxh2 Kxh2 (0.00)

3 [-2.14] 1.Nxh2 h4 2.Rxg1+ Kxg1 3.Nf3+ Kf2 (0.00)

4 [-2.47] 1.Nxh2 h4 2.Re1 Kxh2 (0.00)

5 [-2.70] 1.Nxh2 Kxh2 2.Rf5 h4 3.Ke4 (0.00)

5 [-2.29] 1.Re1 h4 2.Nxh2 Kxh2 3.Ke4 (0.00)

5 [-2.20] 1.Rd1 h4 2.Nxh2 Kxh2 3.Ke4 (0.00)

6 [-3.06] 1.Rd1 h4 2.Nxh2 Kxh2 3.Ke4 h3 4.Kf3 (0.00)

7 [-2.98] 1.Rd1 h4 2.Ke4 h3 3.b3 (0.01)

7 [-1.88] 1.Rd1 h4 2.Ke4 c4 3.Nxh2 Kxh2 4.Kf3 (0.01)

8 [-3.24] 1.Rd1 h4 2.Ke4 c4 3.Rc1 h3 4.Nxh2 Kxh2 5.Kf3 Bd4 (0.01)

9 [-2.87] 1.Rd1 h4 2.b3 h3 3.Ke4 (0.01)

9 [-3.49] 1.Rd1 c4 2.Ke4 h4 3.Rc1 h3 4.Nxh2 Kxh2 5.Kf3 Bd4 (0.01)

9 [-3.16] 1.Rb1 c4 2.b3 c3 3.Rc1 h4 4.Nxh2 Kxh2 5.Ke4 Be3 (0.03)

10 [-3.00] 1.Rb1 c4 2.b3 c3 3.Rc1 h4 4.b4 (0.03)

10 [-2.59] 1.Rb1 c4 2.b3 c3 3.Rc1 h4 4.b4 h3 5.b5 (0.03)

10 [-1.18] 1.Rb1 c4 2.b3 cxb3 3.Ke4 h4 4.Kd5 h3 5.Nxh2 Kxh2 6.Rxb3 Bf2 (0.03)

11 [-2.35] 1.Rb1 c4 2.b3 cxb3 3.Ke4 h4 4.Kd5 b2 5.Nxh2 Kxh2 6.Rxb2 Bf2 7.Rxf2 h3 (0.06)

12 [-1.30] 1.Rb1 c4 2.Rc1 h4 3.Ke4 h3 4.Kd5 c3 5.bxc3 (0.08)

12 [+0.00] 1.Rb1 c4 2.Rc1 h4 3.Ke4 h3 4.Kd5 c3 5.bxc3 Qd6+ 6.Kxd6 h2 7.Nxh2 Kxh2 8.Ke5 Bf2 9.c4 g1=Q 10.Rxg1 Kxg1 (0.09)

14 [+0.00] 1.Rb1 c4 2.Rc1 h4 3.Ke4 h3 4.Kd5 c3 5.bxc3 Qd6+ 6.Kxd6 h2 7.Nxh2 Kxh2 8.Ke5 Bf2 9.Ke4 g1=Q 10.Rxg1 Kxg1 (0.10)

15 [+0.00] 1.Rb1 c4 2.Rc1 h4 3.Ke4 h3 4.Kd5 c3 5.bxc3 Qd6+ 6.Kxd6 h2 7.Nxh2 Kxh2 8.Ke5 Bf2 9.c4 g1=Q 10.Rxg1 Kxg1 (0.12)

17 [+0.00] 1.Rb1 c4 2.Rc1 h4 3.Ke4 h3 4.Kd5 c3 5.bxc3 Qd6+ 6.Kxd6 h2 7.Nxh2 Kxh2 8.Ke5 Bf2 9.c4 g1=Q 10.Rxg1 Kxg1 11.Kd6 (0.15)

18 [+0.00] 1.Rb1 c4 2.Rc1 h4 3.Ke4 h3 4.Kd5 c3 5.bxc3 Qd6+ 6.Kxd6 h2 7.Nxh2 Kxh2 8.Ke5 Bf2 9.c4 g1=Q 10.Rxg1 Kxg1 11.Kd6 Kh2 (0.17)

19 [+0.00] 1.Rb1 h4 2.b3 h3 3.Re1 c4 4.bxc4 Qd6+ 5.Kxd6 h2 6.Nxh2 Kxh2 7.c5 Bf2 8.Rd1 g1=Q 9.Rxg1 Bxc5+ 10.Kxc5 Kxg1 (0.31)

20 [+0.09] 1.Rb1 h4 2.b3 h3 3.Ke4 c4 4.bxc4 (0.40)

20 [+0.82] 1.Rb1 h4 2.b3 h3 3.Ke4 c4 4.bxc4 Qf4+ 5.Kxf4 h2 6.Nxh2 Kxh2 7.Kf3 Bd4 8.Rb7 g1=N+ 9.Ke4 Bf2 10.Rg7 Kh1 11.Rg6 Nh3 12.Rg8 Kh2 13.Rh8 Kg3 14.Rg8+ Kh4 15.Kf3 (0.78)

21 [+0.85] 1.Rb1 c4 2.Rc1 h4 3.Ke4 h3 4.Rb1 c3 5.bxc3 Qf4+ 6.Kxf4 h2 7.Nxh2 Kxh2 8.Kf3 Bc5 9.Rb2 Bf2 10.Rb7 Be1 11.c4 g1=N+ 12.Ke3 Kg2 13.Rg7+ Kf1 (1.06)

22 [+1.01] 1.Rb1 c4 2.Ke4 c3 3.bxc3 h4 4.c4 (1.33)

22 [+1.33] 1.Rb1 c4 2.Ke4 c3 3.bxc3 h4 4.c4 h3 5.Re1 (1.37)

22 [+1.97] 1.Rb1 c4 2.Ke4 c3 3.bxc3 h4 4.c4 h3 5.Re1 Qf4+ 6.Kxf4 h2 7.Nxh2 Kxh2 8.Kf3 Bb6 9.Re2 Bf2 10.Re7 Bd4 11.Rh7+ Kg1 12.Rh4 Bf2 13.Rg4 (1.45)

22 [+5.97] 1.Rb1 c4 2.Ke4 h4 3.Rc1 h3 4.Rb1 c3 5.bxc3 Qf4+ 6.Kxf4 h2 7.Nxh2 Kxh2 8.Kf3 Bc5 9.Rb2 Kh3 10.Rxg2 Kh4 11.c4 Bd6 12.Rd2 Bb8 13.Rd5 Bc7 14.c5 Ba5 15.c6 Bc7 16.Rb5 Bd6 17.Rd5 Bc7 (2.12)

23 [+5.22] 1.Rb1 c4 2.Ke4 h4 3.Rc1 h3 4.Rb1 c3 5.bxc3 Qf4+ 6.Kxf4 h2 7.Nxh2 Kxh2 8.Kf3 Bc5 9.Rb2 Kh3 10.Rxg2 Kh4 11.c4 Bd6 12.Rd2 Bb8 13.Rd5 Bc7 14.c5 Ba5 15.c6 Bc7 16.Rd7 (2.26)

24 [+6.43] 1.Rb1 c4 2.Ke4 h4 3.Rc1 h3 4.Rb1 c3 5.bxc3 Qf4+ 6.Kxf4 h2 7.Nxh2 Kxh2 8.Kf3 Bc5 9.Rb2 Kh3 10.Rxg2 Kh4 11.Rd2 Ba3 12.Rd5 Be7 13.c4 Bf6 14.c5 Be7 15.c6 (2.55)

24 [+6.65] 1.Rb1 c4 2.Ke4 h4 3.Rc1 h3 4.Rb1 c3 5.bxc3 Qf4+ 6.Kxf4 h2 7.Nxh2 Kxh2 8.Kf3 Ba7 9.Rb2 Kh3 10.Rxg2 Kh4 11.c4 Kh5 12.Ke4 Bb6 13.Kd5 Kh6 14.c5 Ba5 15.c6 Bc7 16.Ke6 Bf4 17.Kd7 Kh5 (2.77)

25 [+6.81] 1.Rb1 c4 2.Ke4 h4 3.Rc1 h3 4.Rb1 c3 5.bxc3 Qf4+ 6.Kxf4 h2 7.Nxh2 Kxh2 8.Kf3 Ba7 9.Rb2 Kh3 10.Rxg2 Kh4 11.c4 Kh5 12.Ke4 Bb6 13.Kd5 Bc7 14.c5 Kh6 15.c6 Kh7 16.Ke6 Kh6 17.Kd7 Be5 18.c7 (3.12)

[MrEdCollins]

Strelka 5.5 realizes the move is best in 6.64 seconds, on my machine:

22 [+14.03] 1.Rb1 h4 2.Kc6 h3 3.Kb7 c4 4.Ka8 c3 5.bxc3 Qb8+ 6.Rxb8 Bf2 7.Rh8 g1=N 8.Nxg1 Kxg1 9.Rxh3 Kg2 10.Rh6 Kg3 11.Rg6+ Kh4 12.c4 Kh5 13.Rg8 Kh6 14.Kb7 Bc5 15.Rc8 Bd4 16.c5 Kg6 17.c6 Be5 18.c7 Bxc7 19.Kxc7 Kf5 20.Re8 Kf6 21.Kd6 Kf5 22.Kd5 Kf4 23.Re4+ Kf5 24.Rc4 Kf6 25.Rf4+ Ke7 (6.64)