About a month ago I was doing a post-game analysis with Crafty. When it got to the position where I executed a winning combination I was surprised that Crafty could not find the winning move. Crafty kept insisting on a move that drew instead.
Later, I saw in one of Alexandra Kostiniuk's podcasts in Chess Killer Tips where a simple mate in 2 was presented that Fritz-10 could not solve.
It would be interesting to make a list of problems that chess engines could not solve.
Fritz-10 Teaser: Mate in 2 by Almira Skripchenko
"Fritz-10 cannot see the winning move."
www.chessqueen.com
I know that engine has problems to understand positions with reversed colored bishops. Also engines cant understand positions in which weaker side can make "fortress".
Originally posted by Korch I know that engine has problems to understand positions with reversed colored bishops. Also engines cant understand positions in which weaker side can make "fortress".
That's interesting. In our study group Maxwell Smart discovered how to triangulate with his king in a way that put the weaker side in zugzwang in a queen vs bishop + knight fortress. The queen was in the corner, but the black king was only one square away from completing the fortress. I really enjoyed this discovery. I had not read about it before. Several of us have deconstructed the fortress, but it was mostly trial and error. Now, I have a pattern!
Which fortress positions are you referring to that certain engines can not evaluate?
Its really funny, Fritz 10 only finds mate in 4,
starting with Qc5+ or Qh4+, even after several minutes.
Only if you change to "searching mate"
( in my german version "Mattsuche"😉
Fritz finds the mate in two at once
Originally posted by petrovitch About a month ago I was doing a post-game analysis with Crafty. When it got to the position where I executed a winning combination I was surprised that Crafty could not find the winning move. Crafty kept insisting on a move that drew instead.
Later, I saw in one of Alexandra Kostiniuk's podcasts in Chess Killer Tips where a simple mate in 2 was prese ...[text shortened]... Mate in 2 by Almira Skripchenko
"Fritz-10 cannot see the winning move."
www.chessqueen.com
Some of my engines could find this, some couldn't. Also, if your engine allows a brute force search option, then choosing that option would also find it. And mate solver engines could also find it.
Originally posted by petrovitch About a month ago I was doing a post-game analysis with Crafty. When it got to the position where I executed a winning combination I was surprised that Crafty could not find the winning move. Crafty kept insisting on a move that drew instead.
Later, I saw in one of Alexandra Kostiniuk's podcasts in Chess Killer Tips where a simple mate in 2 was prese ...[text shortened]... Mate in 2 by Almira Skripchenko
"Fritz-10 cannot see the winning move."
www.chessqueen.com
That's got to be a programming bug if Fritz 10 can't see it.
1.Re1! [zugzwang]
1...Bg2 2.Qh4#
1...B~ 2.Qg1#
1...Kxe1 2.Qd2#
A simple brute-force search should easily find this.
Edit: I just saw MadRook's post, and I think he has the answer. Fritz on "infinite analysis" does not always find the fastest mate. He is often content just to show that there is a forced mate and leave it at that. I have Fritz 9 and it does this too.
Originally posted by SwissGambit That's got to be a programming bug if Fritz 10 can't see it.
engines don't normally look at every move. instead they 'prune' off the unlikely branches from the search tree by some heuristic (which is a fancy expression for a hardcoded rule of thumb). it's not a bug, it's a fundamental concept by which all modern engines try to make the search more efficient.
Originally posted by petrovitch That's interesting. In our study group [b]Maxwell Smart discovered how to triangulate with his king in a way that put the weaker side in zugzwang in a queen vs bishop + knight fortress. The queen was in the corner, but the black king was only one square away from completing the fortress. I really enjoyed this discovery. I had not read about it bef ...[text shortened]... attern!
Which fortress positions are you referring to that certain engines can not evaluate?[/b]
Which fortress positions are you referring to that certain engines can not evaluate?
For example final position of Game 3325196 - according to engines black has won, but in fact its pure fortress.
Originally posted by wormwood engines don't normally look at every move. instead they 'prune' off the unlikely branches from the search tree by some heuristic (which is a fancy expression for a hardcoded rule of thumb). it's not a bug, it's a fundamental concept by which all modern engines try to make the search more efficient.
in a nut shell: engines skip moves.
Well, the confusion started with the word 'can't' in "Fritz can't solve this mate in 2". If that was really true, then I would be right in calling it a bug. It is obviously false if Fritz can find the mate in 2 with brute-force search enabled.
Originally posted by SwissGambit Well, the confusion started with the word 'can't' in "Fritz can't solve this mate in 2". If that was really true, then I would be right in calling it a bug. It is obviously false if Fritz can find the mate in 2 with brute-force search enabled.
if you run it with bruteforce it'll search even less deep, evaluating positions positionally even worse than in the normal mode. the pruning is used because it improves the strength of the engine.
Originally posted by Mad Rook Some of my engines could find this, some couldn't. Also, if your engine allows a brute force search option, then choosing that option would also find it. And mate solver engines could also find it.
Please, elaborate. What was the engine that could? I think I can; I think I can. And which engines could not solve this problem or others?
Originally posted by petrovitch About a month ago I was doing a post-game analysis with Crafty. When it got to the position where I executed a winning combination I was surprised that Crafty could not find the winning move. Crafty kept insisting on a move that drew instead.
Later, I saw in one of Alexandra Kostiniuk's podcasts in Chess Killer Tips where a simple mate in 2 was prese ...[text shortened]... Mate in 2 by Almira Skripchenko
"Fritz-10 cannot see the winning move."
www.chessqueen.com
"Fritz-10 cannot see the winning move."
Clearly wrong; almost any move by White is a "winning" move.
Originally posted by wormwood if you run it with bruteforce it'll search even less deep, evaluating positions positionally even worse than in the normal mode. the pruning is used because it improves the strength of the engine.
And why does Fritz need a deep positional analysis to solve a mate in 2, again?! 🙄
I was clearly talking about the Mate in 2 problem the whole time. Thanks for the feeble attempt to lecture on the nature of chess engines, but I'd prefer that you actually read what I say and respond to it next time.
Problems that Chess Engines Can Not Solve
15 May '08 21:563 edits
Back to Top