Yesterday, I was going over a game with a friend of mine, and he showed me this position:
His game continued 1.Rd3?.
This could have met with a rude awakening, 1. ...Qxd3 2.Rxf8+ Kg7.
Black wins
(3.Rh8 Kxf7! not 3. ... Kxh8 4.f8=Q mate)
Anyway, that's my back story.
He set up the original diagram with his trusty Rybka going on his laptop and asked me to find the best move (I usually give him pointers, but sometimes he doesn't believe me because of the box.).
Within seconds, I lashed out the obvious 1.Rg3.
The threat of Rg8+, mating for white is super strong.
The box came up with a fascinating line. I'll only show the analysis/final diagram of what the box had to say.
1. ...h5 2.Rg8+ Kh7 3.Rexf8
Although black looks busted, white can't really get the pawn to queen!
Here, I think black holds a slight edge, according to the engine.
Now, comes the part about the horizon effect.
My friend was demonstrating the brilliance of his computer, when he showed me a way for black to go wrong.
1. ...Qxg3
2.Rxf8+ Kg7 3.Rg8+ Kxf7 4.Rxg3
He thought this was cool and began to go to the next line.
I made him hold on and asked about 4. ... b4.
This is a KILLER, that turns the box on its head.
The box's evaluation of h5 earlier was only above 2 points, but this one is over 3.
Right there, the horizon has made the box choose a second best move (h5 versus Qxg3). The problem is that here too, it is may be misevaluating the position.
I really need a stronger human, or more patient one, to check the following analysis and see if I am right. I think black only draws here !!!
5.Rg4!
5. ... c5!
6.Rc4
(Not the computer's best but quite possibly the easiest way to draw)
(A computer line ran 6.Rh4 a3 7.Rxh7+ Ke6 8.Ra7 c4 9.Ra4 b3 10.Rxa3 b2 11.Ra6+ Ke5 12.Rb6 c3 13.Kh2 c2 14.Rxb2 c1=Q 15.Rf2. White will soon play Rf3, with a fortress, similar to my main line. The computer has black with the advantage, but I am not sure that black can ever get in. There is no diagram for this because I didn't want to spoil the main line that I am posting.)
After 6.Rc4
6. ... a3 7.Rxc5 b3
8.Rb5 a2 9.Rxb3 a1=Q+ 10.Kh2
After say 10. ... Qe5+ 11.Kh1 Kg6 12.Rf3
Here, the computer has black quite won. I'm not sure that is more than a draw. White has a pretty solid fort, and I don't see a way that the black king will ever get in. I must CAUTION you that I have not studied fortresses very much, and I could be overlooking something. This really APPEARS drawn though.
Again, I need a STRONG or PATIENT player with endgame experience to evaluate the fortresses and give their opinions.
If this is correct and drawn, then the box made not one but two misevaluations in a single game (first h5 because it later thought Qxg3 was stronger/second not looking deep enough at the rook against the 3 pawns and realizing a fortress could be made).
Thanks
Interesting post.
For the first point, choosing h5 rather than Qxg3, I found that it only required a little more time to choose Qxg3. About 27 secs, though multiprocessor engines are notoriously underministic and this figure may vary significantly on different runs of the same test.
Regarding the fortress, I don't think this is so much the "horizon effect" as opposed to the difficulty of evaluating fortresses. It looks like the engine's evaluation function is poor for this position.
Originally posted by paulbuchmanfromfics
I'm neither strong nor patient, and I only skimmed through your analysis, but one thing that immediately sticks out to me is: isn't that b4 premature? in the end the 3 advanced connected pawns should be more than enough to queen one of them against the rook I think, and the only way rook gets to even try to hold them is because it has time to snatch them off. so why waste that precious tempo in a futile (assuming your draw is sound) attempt at running the pawns without support? move the black king first closer, right? the rook becomes impotent. THEN crash the pawns in. to me it looks like it just must be much stronger, and probably winning.
I didn't calculate anything. but I can't see how white stops black queen from joining the pawns after ...Ke6. he can try cutting the king off on d-file, but then you simply push the pawns again. if rook then goes to stop pawns the king catches up and protects them.
white kingside attack looks to slow, and his king can't move or black does the same things faster. I think. maybe I'm missing something silly, but to me it looks like it's a win for black.
Originally posted by wormwoodI think the point is, black can queen a pawn no problem, but can't win vs the fortress. Rybka 4 is advertised as being the first engine to be able to evaluate fortresses actually, if I'm not mistaken.
[fen]8/5k1p/2p5/1p6/p7/6RP/6P1/7K[/fen]
I'm neither strong nor patient, and I only skimmed through your analysis, but one thing that immediately sticks out to me is: isn't that b4 premature? in the end the 3 advanced connected pawns should be more than enough to queen one of them against the rook I think, and the only way rook gets to even try to hold th I think. maybe I'm missing something silly, but to me it looks like it's a win for black.
I think your evaluation is brilliant Paul.