Anti-Engine Puzzle

[Paloma]

[d]8/8/2k5/3N4/8/4N1p1/4Kprp/B5bn w - - 0 1

[Ferdy]

[d]8/8/2k5/3N4/8/4N1p1/4Kprp/B5bn w - - 0 1

So what is the solution?

Code: Select all

                                  Name  Hash  Threads  bm1  ce1   bm2  ce2  acd  Timems
0                          CDrill 1800  None        1  Kf1 -199  Nb4+ -343    9   60000
1                          Crystal 3.1   256        1  Kf1   15   Nf1    8   25   60000
2              Deuterium v2021.1.38.29   256        1  Kf1  164  Ne7+  -49   27   60000
3         Deuterium v2021.1.38.29 NNUE   256        1  Kf1   17  Ne7+   -4   19   60000
4  Lc0 v0.27.0 w67900 30x384 gtx 1650s  None        1  Kf1   25  Nb4+  -18    9   60000
5                         Stockfish 13   256        1  Kf1   12  Nb4+   11   32   60000

[Milos]

[d]8/8/2k5/3N4/8/4N1p1/4Kprp/B5bn w - - 0 1

So what is the solution?

Code: Select all

                                  Name  Hash  Threads  bm1  ce1   bm2  ce2  acd  Timems
0                          CDrill 1800  None        1  Kf1 -199  Nb4+ -343    9   60000
1                          Crystal 3.1   256        1  Kf1   15   Nf1    8   25   60000
2              Deuterium v2021.1.38.29   256        1  Kf1  164  Ne7+  -49   27   60000
3         Deuterium v2021.1.38.29 NNUE   256        1  Kf1   17  Ne7+   -4   19   60000
4  Lc0 v0.27.0 w67900 30x384 gtx 1650s  None        1  Kf1   25  Nb4+  -18    9   60000
5                         Stockfish 13   256        1  Kf1   12  Nb4+   11   32   60000

There is no solution. Two knights and a bishop can not mate without help of the king but this hasn't been analytically proven. So the "puzzle" is 99.999999% draw. And virtually any engine sees that instantly.

[mvanthoor]

[d]8/8/2k5/3N4/8/4N1p1/4Kprp/B5bn w - - 0 1

So what is the solution?

AFAIK, if white lets one or more of the black pieces escape, he's going to lose. So the solution / trick is to play Kf1, so black has no moves with any piece except for the king. I don't know if it's possible to mate with 2 knights and 1 bishop without the help of the king.

[Uri Blass]

If the target is draw there are many solutions.

Mating with 2 knights and bishop seems impossible and the evidence is that chest did not find a forced mate for white even when I put the black king in one of the corners.

[Uri Blass]

If white can push the king to a1 maybe white can force mate but I am not sure if it is possible.

The following is mate in 9 for white.
[d]8/8/5B2/4N3/8/6p1/3N1prp/k4Kbn w - - 11 1

[Nay Lin Tun]

Some people feel proud that they found the puzzles that engines are hard to solve.

If the puzzles are " too hard to solve" by engines, then those positions are unlikely to occour in real chess or rubbish positions.

In fact, Stockfish get bettter and better by selectively searching and excluding searching rubbish positions (average branching factor of Stockfish search is 2 out of 30), and human will get better too by excluding those rubbish in search.

[Dann Corbit]

There is a post on that position in this russian forum from 2017:
https://forchess.ru/showthread.php?p=146383
I did not see any resolution there either.

The position is not in YACPDB or the PDB server.

[Dann Corbit]

I think if I were trying to win this position, I would bring all my pieces to bear on the cluster of black chessmen in the corner.
It seems easily possible for the king to take the rook and knight.
If white can take the rest of the chessmen and only give up one knight, he could win the game with KNB v K
I cannot say that I have a clear plan on how to proceed, but it does seem at least remotely possible.

[peter]

[d]8/8/2k5/3N4/8/4N1p1/4Kprp/B5bn w - - 0 1

[pgn][Event "Outskirts?"]
[Site "?"]
[Date "2020.09.14"]
[Round "?"]
[White "White, to Move"]
[Black "and Win"]
[Result "1-0"]
[SetUp "1"]
[FEN "8/8/2k5/3N4/8/4N1p1/4Kprp/B5bn w - - 0 1"]
[PlyCount "95"]

1. Kf1 Kc5 2. Nf6 Kd6 3. Nfd5 Kc5 4. Nf4 Kb4 5. Bb2 Kb5 6. Ba3 Kc6 7. Ne6 Kd7
8. Nd4 Ke8 9. Bd6 Kd7 10. Nc4 Ke8 11. Ne5 Kd8 12. Ne6+ Ke8 13. Ng7+ Kd8 14.
Ne6+ Kc8 15. Nc5 Kd8 16. Nf7+ Ke8 17. Ne5 Kd8 18. Nc6+ Kc8 19. Ne7+ Kd8 20.
Nc6+ Ke8 21. Bb8 Kf7 22. Be5 Kg6 23. Ne7+ Kf7 24. Nc8 Kg6 25. Nd6 Kg5 26. Ne6+
Kh5 27. Nf5 Kg6 28. Nd6 Kh7 29. Nf8+ Kg8 30. Ng6 Kh7 31. Ne7 Kh6 32. Nf7+ Kh7
33. Ng5+ Kh6 34. Bf6 Kh5 35. Nf7 Kg4 36. Bg5 Kf3 37. Nd6 Kg4 38. Bc1 Kh5 39.
Nc4 Kg4 40. Ne5+ Kh4 41. Bd2 Kh5 42. Nd5 Kh4 43. Nf6 Kh3 44. Nc6 Kh4 45. Nd4
Kh3 46. Ne6 Kh4 47. Bg5+ Kh3 48. Nf4# 1-0


[/pgn]

DTM I'm not sure about, got it just by playing out and it's too long to keep it in hash easily at Backward, but it should work within 50 moves boundary, which is the main problem as for engines' output, I guess, getting once and again 50 moves- draw into search- horizon before mating score arises.
If it's a study, which I don't know, neither the author, if it is one, 50 moves wouldn't count, studies don't have such a boundary. Found it once at Outskirts- Forum from a user I don't know neither and without solution too.
Maybe engines with 50 moves rule to be switched off would get it soon, not as for DTM but as for White wins- eval.

Edit: For those registered at Outskirts, here's what Peter Grayson wrote about the problem there then:

http://outskirts.altervista.org/forum/v ... ost#p19077