In this position (Paul Keres v. Laszlo Szabo, USSR v. Hungary, Budapest 1955)
John Nunn says [The World's Greatest Chess Games, p. 206], "There is no defence."
However, 19...Rb4!!! 20.Qg5 (nothing else is any better) g6 21.Qf6 Rxb2!!!
and the game will end in perpetual check.
But the real point of this post concerns a variation that Nunn discusses: (continuing from the first diagram) 19...Qb4
(A) After 20.Qg5 (given by Nunn), the best I can find is 20...g6 21.h5 Qe7 22.Qf4 Qb4 23.Qf6 Qe7 24.Ne4 Rb4 25.c4 Qxf6 26.Nxf6+ Kg7 27.Nxd7 and White has reached a narrowly winning endgame with a bishop & knight for a rook.
(B) Also leading to a win is 20.Qh6 g6 21.h5 Qf8 22.Qd2; for example: 22...Rb4 23.Ne4 Rxe4 24.Bxe4 Re7 25.Rd3 Be8 26.Rd8 Rc7 27.Ra8 c5 28.Qd8 Re7 29.Bc6, etc.
(C) It takes Rybka 3 (1 core) about 71 minutes to (unexpectedly) fail high with 20.Bxh7+!!!
on the 21st iteration; Rybka 3 shows +4.13 after 7 hours (21st iteration). Moves like 20.Bxh7+ are the reason that I like chess. After finding that Rybka 3 with 2 cores finds 20.Bxh7+ fairly quickly (approximately 8 minutes, as I recall), I wondered how long it would take Houdini to find the move.
(D) Houdini 1.5 (2 cores) thinks for 28 hours and a full 29 iterations giving (+3.01) 20.b3 Qf8 21.Ne4 Kh8 22.Ng5 f5 23.exf6 Qxf6 24.Rf3 Qa1+ 25.Kd2 Rf8 26.Qa7 Qh1 27.Rxf8 Rxf8 28.Qxd7 Qxg2+ 29.Kc3 Qf2 30.Nxe6, and White has reached another (winning) bishop & knight vs. rook ending. However, when I force the continuation 20.Bxh7+, Houdini very quickly becomes enthusiastic and after 20...Kxh7 21.Qg5 Houdini gives (+4.63 after 31 iterations and 5 hours) 21...g6 22.h5 Rg8 23.Rh3 Rg7 24.hxg6+ Kg8 25.Qh6 fxg6 26.Qh8+ Kf7 27.Rf3+ Ke7 28.Qxg7+ Kd8 29.b3 Rb7 30.Rd3 and this position is rated +5.90 (after 30 iterations and 4 hours).
I appreciate strong chess-playing programs, but it saddens me when a truly strong program, given plenty of time (28 hours!), fails to find the most crushing move and settles for a mere "technical" win.
1r2r1k1/3b1ppp/p1p1p3/q3P3/7P/2NBQ1R1/PPP3P1/2K5 b - - 0 19
John Nunn says [The World's Greatest Chess Games, p. 206], "There is no defence."
However, 19...Rb4!!! 20.Qg5 (nothing else is any better) g6 21.Qf6 Rxb2!!!
4r1k1/3b1p1p/p1p1pQp1/q3P3/7P/2NB2R1/PrP3P1/2K5 w - - 0 22
and the game will end in perpetual check.
But the real point of this post concerns a variation that Nunn discusses: (continuing from the first diagram) 19...Qb4
1r2r1k1/3b1ppp/p1p1p3/4P3/1q5P/2NBQ1R1/PPP3P1/2K5 w - - 0 20
(A) After 20.Qg5 (given by Nunn), the best I can find is 20...g6 21.h5 Qe7 22.Qf4 Qb4 23.Qf6 Qe7 24.Ne4 Rb4 25.c4 Qxf6 26.Nxf6+ Kg7 27.Nxd7 and White has reached a narrowly winning endgame with a bishop & knight for a rook.
(B) Also leading to a win is 20.Qh6 g6 21.h5 Qf8 22.Qd2; for example: 22...Rb4 23.Ne4 Rxe4 24.Bxe4 Re7 25.Rd3 Be8 26.Rd8 Rc7 27.Ra8 c5 28.Qd8 Re7 29.Bc6, etc.
(C) It takes Rybka 3 (1 core) about 71 minutes to (unexpectedly) fail high with 20.Bxh7+!!!
on the 21st iteration; Rybka 3 shows +4.13 after 7 hours (21st iteration). Moves like 20.Bxh7+ are the reason that I like chess. After finding that Rybka 3 with 2 cores finds 20.Bxh7+ fairly quickly (approximately 8 minutes, as I recall), I wondered how long it would take Houdini to find the move.(D) Houdini 1.5 (2 cores) thinks for 28 hours and a full 29 iterations giving (+3.01) 20.b3
Qf8 21.Ne4 Kh8 22.Ng5 f5 23.exf6 Qxf6 24.Rf3 Qa1+ 25.Kd2 Rf8 26.Qa7 Qh1 27.Rxf8 Rxf8 28.Qxd7 Qxg2+ 29.Kc3 Qf2 30.Nxe6, and White has reached another (winning) bishop & knight vs. rook ending. However, when I force the continuation 20.Bxh7+, Houdini very quickly becomes enthusiastic and after 20...Kxh7 21.Qg5 Houdini gives (+4.63 after 31 iterations and 5 hours) 21...g6 22.h5 Rg8 23.Rh3 Rg7 24.hxg6+ Kg8 25.Qh6 fxg6 26.Qh8+ Kf7 27.Rf3+ Ke7 28.Qxg7+ Kd8 29.b3 Rb7 30.Rd3 and this position is rated +5.90 (after 30 iterations and 4 hours).I appreciate strong chess-playing programs, but it saddens me when a truly strong program, given plenty of time (28 hours!), fails to find the most crushing move and settles for a mere "technical" win.
New game
Analysis by Spark-1.0-win64-mp-corei:
20.Bxh7+ Kxh7 21.Qg5 Qf8 22.Ne4 g6 23.Nf6+ Kg7 24.Ng4 Qh8 25.h5 Qxh5 26.Qf6+ Kf8 27.Rh3 Qxh3 28.gxh3 Red8 29.Qh8+ Ke7 30.Qh4+ Kf8 31.Nf6 Bc8 32.Qh8+ Ke7 33.Ng8+ Kd7 34.Qf6 Kc7 35.Qxf7+ Rd7 36.Qxe6
+/- (0.75 --) Depth: 23/62 00:05:49 4463mN
(Prakash, Bhubaneswar 24.08.2012)
1r2r1k1/3b1ppp/p1p1p3/4P3/1q5P/2NBQ1R1/PPP3P1/2K5 w - - 0 1
Analysis by Spark-1.0-win64-mp-corei:
20.Bxh7+ Kxh7 21.Qg5 Qf8 22.Ne4 g6 23.Nf6+ Kg7 24.Ng4 Qh8 25.h5 Qxh5 26.Qf6+ Kf8 27.Rh3 Qxh3 28.gxh3 Red8 29.Qh8+ Ke7 30.Qh4+ Kf8 31.Nf6 Bc8 32.Qh8+ Ke7 33.Ng8+ Kd7 34.Qf6 Kc7 35.Qxf7+ Rd7 36.Qxe6
+/- (0.75 --) Depth: 23/62 00:05:49 4463mN
(Prakash, Bhubaneswar 24.08.2012)
Thanks for the feedback!
it's a real shame this engine seems to no longer be in development. amazing tactician.
Rybka 3 says Qh6!!
New game
Analysis by Rybka 3:
20.Qh6 g6[] 21.h5[] Qf8 22.Qg5 Qe7 23.Qe3[] Kh8 24.Ne4 Rb4 25.Rh3 Rxe4 26.Qxe4 Qg5+ 27.Kb1 Kg7 28.hxg6 hxg6[] 29.Bxa6 Ra8[] 30.Bc4 Ra5[] 31.Re3 c5[] 32.Rd3 Bb5[]
+/- (1.23) Depth: 19 00:05:56 89205kN
(Prakash, Bhubaneswar 24.08.2012)
New game
1r2r1k1/3b1ppp/p1p1p3/4P3/1q5P/2NBQ1R1/PPP3P1/2K5 w - - 0 1
Analysis by Rybka 3:
20.Qh6 g6[] 21.h5[] Qf8 22.Qg5 Qe7 23.Qe3[] Kh8 24.Ne4 Rb4 25.Rh3 Rxe4 26.Qxe4 Qg5+ 27.Kb1 Kg7 28.hxg6 hxg6[] 29.Bxa6 Ra8[] 30.Bc4 Ra5[] 31.Re3 c5[] 32.Rd3 Bb5[]
+/- (1.23) Depth: 19 00:05:56 89205kN
(Prakash, Bhubaneswar 24.08.2012)
True, and given enough time, it will probably change to 20.Bxh7+.
The Rybka Cluster 64 on PV=2 gets Bh7+ as its top choice in 50 seconds. (see screenshot).
It shows a massive eval for Bh7+ after 2 mins 30 sec.
It shows a massive eval for Bh7+ after 2 mins 30 sec.
Damn!!! the kn/s
Om,is 64 cores
I can see that.
apart from that,i think the cluster isnt obfuscated the kn/s,because if you get the ration kn/ by cores,you get higher kn/s than my the "home" engine
maybe to "impress" the people renting the cluster thinkng something like : "h my god,to much higher ratio than the uci engine"...
people usually doesnt think more than what fulfill their thoughts
maybe to "impress" the people renting the cluster thinkng something like : "h my god,to much higher ratio than the uci engine"...
people usually doesnt think more than what fulfill their thoughts
On the cluster, the kn/s are de-obfuscated (or obfuscated another way)
83865 kn/s divided by 64 cores = 1310 kn/s/core ...looks like a "normal" value
83865 kn/s divided by 64 cores = 1310 kn/s/core ...looks like a "normal" value
agree
read my post above
and that confirms that rybka doesnt "count" the plies in a different way,that is true,but is obfuscated to make belive its competitors that the strength is based on knoledge rather than about its speed calculating
read my post above
and that confirms that rybka doesnt "count" the plies in a different way,that is true,but is obfuscated to make belive its competitors that the strength is based on knoledge rather than about its speed calculating
> I appreciate strong chess-playing programs, but it saddens me when a truly strong program, given plenty of time (28 hours!), fails to find the most crushing move and settles for a mere "technical" win.
This is part of the search strategy of most modern chess engines: if a winning move is found (b3! in this example), it is more important for the engine to validate this winning line than to explore other lines (Bxh7!!, Qh6!, h5!) that might or might not be even better. It's actually a strategy often used by human players as well.
The engine allows you to circumvent this "make sure to win" strategy by setting the multi-PV parameter, basically you're telling to engine to look for more than 1 good move.
Below the output after 20 minutes (not hours!) of analysis with Houdini using MPV=4. The very nice Bxh7!! tops the charts from search dept 24 on.
24 19:47 +3.37 1.Bxh7+ Kxh7 2.Qg5 Qf8 3.Ne4 g6 4.Rf3 Kg7 5.h5 Qe7 6.Nf6 Qb4 7.b3 Rh8 8.Nxd7 Rxh5 9.Qf6+ Kh6 10.Nxb8 Qxb8 11.Qf4+ Kg7 12.Qxf7+ Kh6 13.Qf4+ Kh7 14.g4 (6.166.354.137) 5192 TB:46 24 19:47 +1.83 1.b3 Kh8 2.Ne4 Qf8 3.Kb1 Red8 4.Ng5 Be8 5.Nxh7 Rxd3 6.Qxd3 Qb4 7.Re3 Kg8 8.Re4 Qc5 9.Nf6+ Kf8 10.Qg3 Rb4 11.Re1 Qa5 12.Rd1 Rb5 13.Nxe8 Kxe8 14.Qxg7 (6.166.354.137) 5192 TB:46 24 19:47 +1.66 1.Qh6 g6 2.h5 Qf8 3.Qg5 Qe7 4.Qe3 Kh8 5.Ne4 Rg8 6.Nf6 Rg7 7.hxg6 fxg6 8.Bxg6 Rf8 9.Bd3 Rxf6 10.exf6 Qxf6 11.Rf3 Qe7 12.Qe5 Kg8 13.Qb8+ Be8 14.Be4 (6.166.354.137) 5192 TB:46 24 19:47 +1.52 1.h5 Qh4 2.Ne4 Kh8 3.Nf6 Rxb2 4.Rg4 Rb1+ 5.Kxb1 Qh1+ 6.Kb2 Rb8+ 7.Kc3 gxf6 8.exf6 Qxh5 9.Qg5 Qxg5 10.Rxg5 h6 11.Rh5 Ra8 12.Rxh6+ Kg8 13.Rh7 Be8 14.Rg7+ (6.166.354.137) 5192 TB:46
Regarding Houdini 2.0c , I need few tips.
1. Is running Houdini with MPV better than SPV in IA ? If yes How much numbers ? 3 ? 4? 5? 10? Does it Make the engine slow ?
2. If I run in MPV what hash is good ? I most of time use 1024 MB hash and I have 8 GB RAM.
3. Is it good to have learning on while playing engine matches in Playchess Server ?
1. Is running Houdini with MPV better than SPV in IA ? If yes How much numbers ? 3 ? 4? 5? 10? Does it Make the engine slow ?
2. If I run in MPV what hash is good ? I most of time use 1024 MB hash and I have 8 GB RAM.
3. Is it good to have learning on while playing engine matches in Playchess Server ?
1GB hash is fine, 2GB is probably better if you're going to let it run for awhile, but i'm not sure it matters much at all. i believe uly here uses very small amounts of hash & he seems to have a very good grasp of what he's doing. if you're running rybka with large pages, more hash is better as it should increase speed. houdini pro also supports large pages.
EVERYTHING i've read on this board says MPV is far less efficient overall than SPV. and it slows down the engine considerably, just think of it as 2pv, 50% speed, 4pv, 25% speed, etc. what i understand the power users do is simply tell the engine to look at the 'next best' move, this way you see alternatives while still running at 100% capacity -- afaik all GUIs support this feature, although i've been out of the loop long enough now i can't remember how to do it in aquarium...help anyone?
EVERYTHING i've read on this board says MPV is far less efficient overall than SPV. and it slows down the engine considerably, just think of it as 2pv, 50% speed, 4pv, 25% speed, etc. what i understand the power users do is simply tell the engine to look at the 'next best' move, this way you see alternatives while still running at 100% capacity -- afaik all GUIs support this feature, although i've been out of the loop long enough now i can't remember how to do it in aquarium...help anyone?
Ya. I gotcha. You have to do ctrl + move.
thanks-- knew it was something like that but i'd forgotten over my lazy summer.
And here the output after some 6 hours, Bxh7 is very good but b3 is not a lot worse:
29 374:48 +4.26 1.Bxh7+ Kxh7 2.Qg5 Qf8 3.Ne4 g6 4.Rf3 Kg7 5.h5 Qe7 6.Nf6 Rh8 7.h6+ Rxh6 8.Nh5+ Rxh5 9.Qxe7 Be8 10.Rf1 Rh8 11.b3 Ra8 12.Qc7 c5 13.Qxc5 Rd8 14.c4 (128.856.589.044) 5729 TB:467 29 374:48 +3.34 1.b3 g6 2.Ne4 Kg7 3.Nf6 Qe7 4.h5 Rh8 5.Kb1 Be8 6.Rh3 h6 7.hxg6 fxg6 8.Nxe8+ Qxe8 9.Bxg6 Qxg6 10.Qa7+ Kg8 11.Qxb8+ Kh7 12.Qc7+ Qg7 13.Qxc6 Qxe5 14.Qd7+ (128.856.589.044) 5729 TB:467 29 374:48 +2.01 1.Qh6 g6 2.h5 Qf8 3.Qg5 Qe7 4.Qe3 Kg7 5.Ne4 Rb4 6.c4 Rh8 7.Nf6 Be8 8.b3 Rb7 9.Kb2 a5 10.Rh3 h6 11.hxg6 fxg6 12.Nxe8+ Qxe8 13.Bxg6 Kxg6 14.Rg3+ (128.856.589.044) 5729 TB:467 29 374:48 +1.19 1.h5 Kh8 2.Rxg7 Kxg7 3.Qg5+ Kh8 4.Qf6+ Kg8 5.Qh6 f5 6.exf6 Re7 7.Qg5+ Kh8 8.fxe7 Qd4 9.Qh6 Qg7 10.Qe3 Rg8 11.h6 Qf6 12.Ne4 Qxe7 13.Nc5 Rxg2 14.Be4 (128.856.589.044) 5729 TB:467
> it is more important for the engine to validate this winning line than to explore other lines (Bxh7!!, Qh6!, h5!) that might or might not be even better. It's actually a strategy often used by human players as well.
That's right, I've just finished reading "your best move" by Per Osman which is the best book I've read on the subject of how humans think OTB. It's a significant step up on Kotov's work, IMO.
> This is part of the search strategy of most modern chess engines: if a winning move is found (b3! in this example), it is more important for the engine to validate this winning line than to explore other lines (Bxh7!!, Qh6!, h5!) that might or might not be even better. It's actually a strategy often used by human players as well.
As a small addendum, Houdini 3 will have a "Tactical Mode" in which tactical solutions are preferred (aka as "beautiful" solutions).
In Tactical Mode Houdini 3 produces the following result (in normal analysis without multi-PV):
1r2r1k1/3b1ppp/p1p1p3/4P3/1q5P/2NBQ1R1/PPP3P1/2K5 w - -
Engine: Houdini 3 (2048 MB)
by Robert Houdart
21/70 1:55 +1.35 1.b3 Kh8 2.Qg5 g6 3.Qf6+ Kg8 4.h5 Qa3+
5.Kb1 Qe7 6.hxg6 hxg6 7.Ne4 Rbd8
8.Bxa6 Qxf6 9.Nxf6+ Kf8 10.Kb2 Ke7
11.Nxe8 Bxe8 12.Rd3 Ra8 13.Bc4 Ra5
14.Re3 (623.607.346) 5406
21/70 3:17 +1.50++ 1.Bxh7+ (1.089.693.273) 5504
21/70 3:22 +1.72++ 1.Bxh7+ (1.115.306.227) 5502
18/70 3:24 +2.21++ 1.Bxh7+ (1.125.715.659) 5501
18/70 3:26 +2.57 1.Bxh7+ Kxh7 2.Qg5 Qf8 3.Ne4 g6
4.Rf3 Qe7 5.Nf6+ Kg7 6.h5 Qb4 7.b3 Rh8
8.Nxd7 Rxh5 9.Qf6+ Kh6 10.Nxb8 Qxb8
11.g4 Rh1+ 12.Kb2 Qg8 13.g5+ Kh7
14.Qxf7+ (1.137.705.493) 5499
...
22/75 4:25 +3.08 1.Bxh7+ Kxh7 2.Qg5 Qf8 3.Ne4 g6
4.Rf3 Kg7 5.h5 Qe7 6.Nf6 Rh8 7.h6+ Rxh6
8.Nh5+ Rxh5 9.Qxe7 Rh1+ 10.Kd2 Be8
11.Kc3 Rh2 12.Rh3 Rxh3+ 13.gxh3 c5
14.Qxc5 (1.465.518.203) 5522
> As a small addendum, Houdini 3 will have a "Tactical Mode" in which tactical solutions are preferred (aka as "beautiful" solutions).
> In Tactical Mode Houdini 3 produces the following result (in normal analysis without multi-PV):
> Engine: Houdini 3 (2048 MB)
by Robert Houdart
21/70 1:55 +1.35 1.b3 Kh8 2.Qg5 g6 3.Qf6+ Kg8 4.h5 Qa3+
5.Kb1 Qe7 6.hxg6 hxg6 7.Ne4 Rbd8
8.Bxa6 Qxf6 9.Nxf6+ Kf8 10.Kb2 Ke7
11.Nxe8 Bxe8 12.Rd3 Ra8 13.Bc4 Ra5
14.Re3 (623.607.346) 5406
21/70 3:17 +1.50++ 1.Bxh7+ (1.089.693.273) 5504
21/70 3:22 +1.72++ 1.Bxh7+ (1.115.306.227) 5502
18/70 3:24 +2.21++ 1.Bxh7+ (1.125.715.659) 5501
Very impressive. Was that on 3 cores?
Correct, 3 cores of a standard Core i5-750, 2048 MB hash.
I've seen you post a lot of 3 core stuff. How come you do it this way instead of using all 4 cores if you have them to use?
I usually leave 1 core for other activities (e-mail, internet, streaming, ...).
It's a great way to put a multi-core CPU to good use.
It's a great way to put a multi-core CPU to good use.
do you find this to be more efficient than simply all 4 cores, low priority? i used to do the 3 core thing till i switched to this (since aquarium is my primary GUI, i'm not sure why i was using 3 cores in the first place).
Leaving 1 core for non-chess stuff improves the consistency of the results - the output is not a lot influenced by my other activities on the computer.
If I would use all 4 cores for the chess analysis, any non-chess related activity would immediately impact the analysis (lower the nps).
Either way it's not a big deal.
If I would use all 4 cores for the chess analysis, any non-chess related activity would immediately impact the analysis (lower the nps).
Either way it's not a big deal.
Good teaser!
A glimpse on release date?
A glimpse on release date?
hmm, yes, i notice it doesn't say 'dev' anymore..
> A glimpse on release date?
Early October is the most likely period, I'll release more details on the Houdini Facebook page when available.
> As a small addendum, Houdini 3 will have a "Tactical Mode"
Will there also be an auto-tune feature (as with Houdini 2) or any other kind of optimiser? I don't want to test your program when it comes out without making sure it's running optimally first
> This is part of the search strategy of most modern chess engines: if a winning move is found (b3! in this example), it is more important for the engine to validate this winning line than to explore other lines (Bxh7!!, Qh6!, h5!) that might or might not be even better. It's actually a strategy often used by human players as well
Does/should this logic reverse when an engine suddenly realizes(as it searches deeper) that its current best move is losing? It would seem then it's not as important to validate this move is losing as it is to find another move that is not losing. In case of H3 I wonder if it would be helped by auto switching into its tactical mode when this happens to try and save the game? Humans use this too. You gamble more when you are losing and less when you are winning.
> Does/should this logic reverse when an engine suddenly realizes(as it searches deeper) that its current best move is losing? It would seem then it's not as important to validate this move is losing as it is to find another move that is not losing.
Engines already do this, when a move fail lows they search for alternatives even without knowing the exact score of the previously best move.
> In case of H3 I wonder if it would be helped by auto switching into its tactical mode when this happens to try and save the game? Humans use this too. You gamble more when you are losing and less when you are winning.
Interesting idea. Note that the Tactical Mode wastes a lot of time looking for tactics that usually don't work - this may or may not be the best choice in a fail-low situation, hard to tell.
> Engines already do this, when a move fail lows they search for alternatives even without knowing the exact score of the previously best move.
This is not good for analysis. This strategy has advantages in games because there it's paramount that a better move is found and played within the time limit if there's one.
For analysis, what is paramount is knowing the score and refuting line of the move that failed low, specially if the moves from opponent that kill the move also work for the other moves but user interaction is needed to check this. It also is faster in the cases where no better alternative moves are there and the engine will have to go back to solve exact score of the move anyway.
So, I hope Houdini has some kind of "solve exact score of failing move" option.
what is better,go for a ''beauty'' win,or for the sure way and a ''technical'' win?
Dvoretsky said that play for the beuty moves is an error,and even more if we have a technical win
Dvoretsky said that play for the beuty moves is an error,and even more if we have a technical win
Even more, it is very difficult at the present time to assign a numerical value to "beauty", and thus even more difficult to to code it into a working program.
right,is difficult
but we can let the GM of the past help a bit with that;'beauty' is with sacrifices or a strong and very tactical atack
so if an engines has 2 ways to win,one long term and 'slowly',and another short term and sacrificing pawns/pieces,the engine can choose play 'beauty'
of course,that will be an error,because the engie must not take risks if there is a sure win without that risks,just to play beuty
but we can let the GM of the past help a bit with that;'beauty' is with sacrifices or a strong and very tactical atack
so if an engines has 2 ways to win,one long term and 'slowly',and another short term and sacrificing pawns/pieces,the engine can choose play 'beauty'
of course,that will be an error,because the engie must not take risks if there is a sure win without that risks,just to play beuty
Hi CSullivan,
Very interesting position from a game with historical interest!
I guess Keres' mistake was a half-move earlier from your first diagram:
Here White should not play 19.Bd3?! which breaks the communication between the Rg3 rook and the c3 knight (i.e. it unprotects it). Better in my opinion is to "slow built" with 19.Qg5! g6 20. Rd3! which impedes with black strong defensive manoeuvre Qd8-Rb4 followed by the landing of a black heavy piece on h4. The sequence you shown beginning with 19.Bd3 Rb4!!! is truly remarkable in my opinion.
Thank you for this,
[Event "Budapest"]
[Site "Budapest"]
[Date "1955.??.??"]
[EventDate "?"]
[Round "?"]
[Result "1-0"]
[White "Paul Keres"]
[Black "Laszlo Szabo"]
[ECO "B64"]
[WhiteElo "?"]
[BlackElo "?"]
[PlyCount "45"]
1.e4 c5 2.Nf3 Nc6 3.d4 cxd4 4.Nxd4 Nf6 5.Nc3 d6 6.Bg5 e6 7.Qd2
Be7 8.O-O-O O-O 9.f4 a6 10.e5 dxe5 11.Nxc6 bxc6 12.fxe5 Nd7
13.h4 Rb8 14.Qe3 Re8 15.Rh3 Qa5 16.Bxe7 Rxe7 17.Rg3 Re8
18.Rxd7 Bxd7 19.Bd3 h6 20.Qf4 Kf8 21.Rxg7 Kxg7 22.Qf6+ Kf8
23.Bg6 1-0
Very interesting position from a game with historical interest!
I guess Keres' mistake was a half-move earlier from your first diagram:
1r2r1k1/3b1ppp/p1p1p3/q3P3/7P/2N1Q1R1/PPP3P1/2K2B2 w - - 0 19
Here White should not play 19.Bd3?! which breaks the communication between the Rg3 rook and the c3 knight (i.e. it unprotects it). Better in my opinion is to "slow built" with 19.Qg5! g6 20. Rd3! which impedes with black strong defensive manoeuvre Qd8-Rb4 followed by the landing of a black heavy piece on h4. The sequence you shown beginning with 19.Bd3 Rb4!!! is truly remarkable in my opinion.
Thank you for this,
[Event "Budapest"]
[Site "Budapest"]
[Date "1955.??.??"]
[EventDate "?"]
[Round "?"]
[Result "1-0"]
[White "Paul Keres"]
[Black "Laszlo Szabo"]
[ECO "B64"]
[WhiteElo "?"]
[BlackElo "?"]
[PlyCount "45"]
1.e4 c5 2.Nf3 Nc6 3.d4 cxd4 4.Nxd4 Nf6 5.Nc3 d6 6.Bg5 e6 7.Qd2
Be7 8.O-O-O O-O 9.f4 a6 10.e5 dxe5 11.Nxc6 bxc6 12.fxe5 Nd7
13.h4 Rb8 14.Qe3 Re8 15.Rh3 Qa5 16.Bxe7 Rxe7 17.Rg3 Re8
18.Rxd7 Bxd7 19.Bd3 h6 20.Qf4 Kf8 21.Rxg7 Kxg7 22.Qf6+ Kf8
23.Bg6 1-0
Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill