< Earlier Kibitzing · PAGE 3 OF 3 ·
|Jun-02-08|| ||zenpharaohs: I'm really surprised that when I say "black does not have an obviously won game" that somehow means black does not have the advantage. The point is that in the line that I propose (which I pointed out Rybka agrees with) the position ends up with a modest advantage. The material is indeed Queen to two knights, but white has big developmental advantages and black also has no attack.|
Is this line better than 10 Qa4? Hard to say. Rybka values the lines at move 10 as
23 0:00 -1.87 10.Qa4 Bxc3 11.Qxa5 Bxa5 12.Ne4 Nc6 13.g4 Bc7 14.Nd6+ Ke7 15.Bg2 h5 16.g5 b6 (114) 6
23 0:00 -1.90 10.a3 Nxb3+ 11.Nxb3 Qa6 12.axb4 Qxc4 13.Kc2 Qf4 14.f3 b6 15.e4 Ke7 16.Bb5 Rd8 (115) 6
at a depth of 23 ply. I have analyzed this position with various other engines, (Shredder, HIARCS 12 MP, etc.) which both value the position as better for black and prefer the 10 Qa4 line by almost a whole pawn. But those are weaker engines on that hardware. So I have put the problem to my strongest engine/hardware combination - Zappa Mexico II 64 bit on a fast quad core machine with a lot of memory. It likes 10 Qa4 better than 10 a3 by -1.69 to -2.27. Just a bit more than half a pawn. I will let this run on that machine for the rest of the day and see what it thinks.
The sort of values we are seeing here (-1.8 on Rybka) normally represent positions where there is an overall advantage for black, but not a clear win. I would like to see someone play black from this position and actually beat a strong engine with tournament time controls from this position.
This is not likely a mistake by Rybka.
|Jun-02-08|| ||playground player: For the second day in a row, we see White O-O-Oing into harm's way, this example being a lot more obvious than yesterday's.|
|Jun-02-08|| ||Terry McCracken: < zenpharaohs: I'm really surprised that when I say "black does not have an obviously won game" that somehow means black does not have the advantage. The point is that in the line that I propose (which I pointed out Rybka agrees with) the position ends up with a modest advantage. The material is indeed Queen to two knights, but white has big developmental advantages and black also has no attack. >|
Forget computer evals, they vary a lot. The point is Black is up a whole piece and that is enough to win outright, considering White has no longterm positional advantage.
I think people are making too much out of a rather mundane position.
|Jun-02-08|| ||Terry McCracken: <patzer2: White's blunder was 9. 0-0-0?? Instead, 9. e3>|
Correct, 9. 0-0-0?? is a blunder, 9. e3 is one of the correct moves as is 9 Rc1. IMO 9. e3 is likely best.
|Jun-02-08|| ||TheBB: <ewenardus> Yes, thanks. It would appear five years of chess hasn't made me much better :P.|
|Jun-02-08|| ||zenpharaohs: Terry McCracken: "Forget computer evals..."
No. I played for decades before computer evaluation was strong enough to be useful. Trust me, today's computer evaluations when carefully consulted, are incredibly useful. Even just on my laptop (Two cores, 2.4GHz, 4GB RAM) the stronger chess engines are really quite good. When given a long time to evaluate (Rybka 2.3.2a 64 bit has had almost two hours so far) if there was a crushing continuation for black, it would be found. It is now into the 25th ply. Back home the quad core with Zappa Mexico II 64 is probably quite a bit deeper into the analysis, I'll be able to check in about eight hours. By then the picture should be quite clear. Note that I am using two strong engines from completely different code bases - not just one. This greatly reduces the chances of a bug.
Now it's a fair question for you to answer just exactly how black converts the material into a won endgame - I am pretty sure that there is no early death for white given the depth of Rybka's analysis so far.
Now it is really clear that the material of the white clean should not be mesmerizing. If you look at the position after
9 ... Nd4
10 a3 Nxb3+
click for larger view
then it turns out that continuation
11 ... Bxc3
12 Nxa5 Bxa5
is very nearly as good as what seems to be the best line; and here the material difference (bishop to pawn) makes it pretty clear that we are talking about less than a full piece advantage.
|Jun-02-08|| ||MiCrooks: I agree Terry. Either one is an easy win with Black. Personally I would preferer to have the Q vs the 2 Knights and so would the engines mentioned but I would happily play either position against anybody. Does that mean I might not screw up still and lose - no. Does it mean I would accept a draw from a GM - no.|
Oh, BTW longer time controls against the computer favor the human :)! Most computers do not change their evaluation of the postion (ie which move is best not score) after the first few seconds. The only exception is when there is a deeply buried tactic where the payoff is deep enough off the horizon that the computer will not see the move quickly (if ever - there are positions that were found by players that computers still won't find as the payoff is beyond their ability to find).
|Jun-02-08|| ||jheiner: Saw the trapped Q instantly.--the moment I looked at the board. Didn't even have a chance to start the material analysis. Took another heartbeat to see that there was no counterplay and good game.|
Ok. Will try to post something more useful than "that was easy."
Although this felt "instant", there was actually a sequence my pattern recognition unit went through.
1. There are no checks or threats against the White K.
2. Eye is drawn to the 5x5 corner a1xe5. The Queens are the key feature there.
3. Nd4 is availalbe, it threatens the White Q and the response will be...puzzle ALMOST solved.
4. Look for counterplay against the Black K or Q. None found.
So: Checks. Captures. Threats. Responses. Counterplay.
Last thing that made this very very fast to calculate. White has no long range pieces (bishops/rooks) that are active. With the light-squared B and R hidden behind pawns, and quickly checking the f1-a6 diagonal, the only counterplay to see is against the knights. And Black's Q can escape safely to d8 in the case of something like 10.Qa3 Nb3.
|Jun-02-08|| ||jheiner: Oops. Missed the Qa4. Still Black is up a piece, but I dismissed it as a Q trap. <YouRang>...i'm keeping you company on this one.|
|Jun-02-08|| ||Terry McCracken: <zenpharaohs: Terry McCracken: "Forget computer evals..." |
I needn't be lectured on computer chess engine evals, I know how accurate and still inaccurate they can be at times for special reasons.
They're not perfect, but they are usually good, but that's not the point.
Also we're NOT looking for a "Crush" just a win, and in this case a win of a piece which is good enough.
Why you're wasting hours upon hours of computer time on a basic postion right in the opening after a serious blunder is beyond me.
The evaluations may not reflect the fact due to lack of Black's developement that White is lost. I've already explained this.
White's space and development are effecting the evaluation and giving you a lower score than a piece, but they're wrong! Once Black is in full swing the eval will keep dropping for White, to negative 3 and with decent play worse than that in less than 10 moves.
BTW Hiarcs is very strong and its eval is worse than a piece down with your a3 line. Is it better to trust it over Rybka's? In this case I'd say yes.
|Jun-02-08|| ||YouRang: <10.a3 vs. 10.Qa4> I doubt that there's any BIG difference, but it may be that the computer may rate <10.a3> higher than a human would due to differences in the way they think.|
In the 10.a3 line, we arrive pretty quickly at this position (after 10...Nxb3 11.Nxb3 Qa6 12.axb4 Qxc4 13.Kc2 Qf4):
click for larger view
About this position:
(1) It's fairly tranquil (no immediate tactics in the works)
(2) Black has a substantial material advantage, but white is somewhat better developed.
A computer will struggle with a position like this because there is no winning sequence apparent for several moves. It will give black the advantage purely from material, but not huge due to white's development.
A human will probably see the winning plan quicker:  Get our Q to a safe square where it is in checking range of white's king (probably at f5 or g6).  Castle kingside to get connected rooks on the queenside.  Play ...b6 to open up the queenside AND allow our bishop to develop at b7 or a6.  bring f8 rook to the queenside, march pawns, launch attack and win.
I bet that once we get this far, the computer evals will start rising sharply.
|Jun-02-08|| ||Terry McCracken: <MiCrooks: I agree Terry. Either one is an easy win with Black. Personally I would preferer to have the Q vs the 2 Knights and so would the engines mentioned but I would happily play either position against anybody. Does that mean I might not screw up still and lose - no. Does it mean I would accept a draw from a GM -no. >|
You see the position as I do and I agree with your assessment.
I also agree that modern computer evals although good can't be trusted 100% for the reasons you stated and for many more. They are useful of course and are usually very good, but not perfect and this appears to be the case in this position with some top engines.
So much fuss over a Monday's position seems absurd to me.
I wish they stuck to a simple forced mate:)
|Jun-02-08|| ||TheaN: 1/1
Black's somewhat attacking approach in this Nimzo-Indian pays well after 9 moves.
Simply attacking the Queen, nothing more to it.
As Clooney says: "Retreat, what else?"
Before winning this piece the situation explains the power of a Bishop that a Knight doesn't contain (and there's also a vice-versa on that): mutual defending on diagonals (B+Q). White guards the Queen with the Knight, and Black the Queen with the Bishop. However, when BxN, the Bishop STILL defends the Queen, while the Knight of White is gone: "yeah, fine, so the same would've happened if White to move."; no, because it doesn't go that way. Although 10....pass 11.Qxa4 would safe the piece it doesn't win for White.
It can't escape, and neither matter, Black just finishes off with a fatal Queen trade.
|Jun-02-08|| ||tatarch: YouRang-- good insight about the computer/human difference.|
And I like to see the arguments here over an otherwise boring puzzle-- certainly better than the usual "I got it" comments.
|Jun-02-08|| ||zenpharaohs: Terry McCracken: "Hiarcs is very strong"
Not on my hardware. I have HIARCS 12 MP and it is handily beaten by Zappa Mexico II 64 on the quad core. Take that however you like.
|Jun-02-08|| ||MostlyAverageJoe: <zenpharaohs> It might be interesting to set up an auto-play (aka shootout), using the same engine to play both sides, and see whether a3 or Qa4 brings a quicker win under otherwise identical parameter settings.|
This still would not prove which continuation is objectively stronger for humans to play, but would validate (or disprove) the engine's initial assessment of the positions. For example, if Rybka wins much faster against itself after Qa4 than after a3, it would indicate that the small evaluation difference between the two moves is incorrect.
An alternative experiment could involve reducing ELO (or time per move) for the black, and see at which setting the game can be equalized by white (playing at maximum rating). The position winnable with lower ELO/shorter time per move would be clearly easier (for the engine) to win (against engine).
|Jun-02-08|| ||zenpharaohs: YouRang: "I bet that once we get this far, the computer evals will start rising sharply."|
Not all that switfly. Once you get Zappa Mexico II 64 to this position:
click for larger view
the evaluation in favor of black has faded slightly from the -2.44 it originally had back at move 10, to -2.38 at 20 ply depth. Zappa also reports the maximum depth of line it has considered, and at this point (currently considering the 21st ply) that maximum depth is 59 ply. This means that there are some lines which Zappa has considered worth searching to 28 moves without discovering the conversion of the material advantage.
I don't see how white can obtain a draw - probably he can't. But I haven't seen anything that rules that possibility out yet.
|Jun-02-08|| ||zenpharaohs: MostlyAverageJoe: "<zenpharaohs> It might be interesting to set up an auto-play (aka shootout), using the same engine to play both sides, and see whether a3 or Qa4 brings a quicker win under otherwise identical parameter settings."|
That's an interesting idea. It might also be interesting to have different engines play the positions - sort of a round robin tournament that starts from the initial positions. That might be even more representative.
|Jun-02-08|| ||jheiner: <<MostlyAverageJoe>: The position winnable with lower ELO/shorter time per move would be clearly easier (for the engine) to win (against engine).>|
Speaking of ELO's, it's worth pointing out that the players in this case were probably class players -- or at least S Jensen, for whom this is their only game in the database. As such, a lower ELO setting might provide more realistic simulation of the games outcome.
Clearly an instructive position, and as someone who is not an Expert, I would hope they wouldn't throw in the towel so quickly.
I would be curious to see a shoot-out on this position.
|Jun-03-08|| ||MostlyAverageJoe: Okay, here's a quick Hiarcs 11.2MP shootout, tablebases disabled, 15plies/move (about 15 minutes runtime each, but this is on 8-core 3Ghz host).|
<10. Qa4> Bxc3 11. Qxa5 Bxa5 12. e3 Nc6 13. Ne4 Ke7 14. a3
Bc7 15. Nd6 b6 16. Nb5 Bb8 17. cxb6 axb6 18. f4 Bb7 19. Kb1 Rc8 20. Rc1 e5 21.
Bd3 exf4 22. exf4 Bxf4 23. Rhe1+ Kf8 24. Rcd1 Bxh2 25. Rh1 Be5 26. Be4 Nd8 27.
Bxb7 Nxb7 28. Rxh7 Na5 29. Rh3 Nxc4 30. b3 Na5 31. Rxd7 Kg8 32. Nd6 Rd8 33.
Rh8+ Kxh8 34. Rxd8+ Rxd8 35. Nxf7+ Kg8 36. Nxd8 Nxb3 37. Nc6 Bd6 38. Kc2 Na5
39. Nxa5 bxa5 40. Kb3 Kf7 41. Ka4 Ke6 42. Kxa5 Bxa3 43. Ka4 Kf5 44. Kb3 Kf4 45.
g3+ Kg4 46. Kxa3 g5 47. Kb3 Kxg3 48. Kc3 g4 49. Kd2 Kg2 50. Ke2 g3 51. Ke1 Kg1
52. Ke2 g2 53. Ke1 Kh1 54. Kf2 g1=Q+ 55. Ke2 Qd4 56. Kf3 Kh2 57. Ke2 Kg3 58.
Ke1 Kf3 59. Kf1 Qd1# 0-1
<10. a3> Nxb3+ 11. Nxb3 Qa6 12. axb4 Qxc4 13. Kc2 Qf4 14.
f3 b6 15. e4 O-O 16. Bb5 Rb8 17. Rd6 a6 18. Bxd7 a5 19. bxa5 bxc5 20. Rhd1 Qxh2
21. R1d2 Bxd7 22. Rxd7 Qg1 23. a6 c4 24. Na5 Qa1 25. Nxc4 Qxa6 26. Rc7 Rfd8 27.
Nb1 Qa4+ 28. Kc1 Qb3 29. Nba3 g5 30. Rxd8+ Rxd8 31. Nc2 Qd3 32. N4e3 Qd2+ 33.
Kb1 Rd3 34. Rc8+ Kg7 35. Ka2 h5 36. Rc3 Rxc3 37. bxc3 Qxc3 38. Kb1 Qd3 39. e5
Qb5+ 40. Kc1 Qxe5 41. Kd2 f5 42. Ne1 g4 43. Kd3 Qd6+ 44. Ke2 Qa6+ 45. Nd3 gxf3+
46. gxf3 h4 47. f4 Qa2+ 48. Kf3 Qa8+ 49. Ke2 h3 50. Nd1 h2 51. N1f2 Kg6 52.
Ne5+ Kh5 53. Nf3 h1=N 54. Nxh1 Qe4+ 55. Kf2 Kg4 56. Nh2+ Kh3 57. Nf3 Qxf4 58.
Ke2 Qe4+ 59. Kf2 Qc2+ 60. Ke3 Qc1+ 61. Ke2 Qxh1 62. Ne1 f4 63. Kd2 Qd5+ 64. Kc3
Kg3 65. Nc2 f3 66. Ne3 Qc5+ 67. Kd3 Qb5+ 68. Kd4 Kf4 69. Nc4 Qb3 70. Kc5 Qc3
71. Kd6 Qxc4 72. Ke7 Qc7+ 73. Kf6 f2 74. Kg6 f1=Q 75. Kh6 Qb1 76. Kh5 Qbh7# 0-1
Conclusion: on patzer-level engine settings, Qa4 is definitely faster to win and 10.a3 provides white with more opportunities for resistance.
With other engines and settings, the result may be different.
|Jun-03-08|| ||MostlyAverageJoe: Shootout update. I run it under time constraints (15 minutes per side + 5 sec time delay). This resulted in about four and a half hours single-processor equivalent runtime per game.|
10.a3 ... black score better than -10.00 at move 35, mate in move 63
10.Qa4 ... black score better than -10.00 at move 55, mate in move 70
Deep Shredder 11
10.a3 ... black score better than -10.00 at move 36, mate in move 55
10.Qa4 ... black score better than -10.00 at move 49, mate in move 72
Conclusion: at less-patzer-like engine settings, 10.a3 loses faster and Qa4 provides longer resistance for white.
|Jun-03-08|| ||znprdx: <MAJ ...patzer-like engine settings> I had no idea your "silicone monsters" had a kinder,gentler side. I'm still trying to understand your conclusions. Are they actually contradictory to common sense and experience? Fascinating. "opportunities for resistance" but surely doesn't that imply potential for error...|
I am astounded by the extraordinary attention paid to this position as
<Terry McCracken:> also put it. Perhaps it reveals an undelying psychological obsession with attempting to define absolute truth-which is probably why we play Chess. At least in this parody of reality we can find justice of some sort...and maybe, just maybe, sometimes the underdog can win....
|Jun-03-08|| ||zenpharaohs: znprdx: "I am astounded by the extraordinary attention paid to this position."|
It's actually the normal amount of attention. If you look at a position, you intend to settle the question, otherwise you are just fooling around. This helps reduce the false appearance of simplicity which is common in human-only analyses.
A very good example of this is the Smyslov-Liberzon game:
Smyslov vs V Liberzon, 1968
where years of commentary on how strong the game was for Smyslov have to be balanced against the fact that Liberzon actually had the advantage before blundering.
I haven't yet done the shootout but since we have already seen HIARCS 12 put in, I'll try that with a stronger engine.
|Jun-05-08|| ||MostlyAverageJoe: Shootout update. Hiarcs 11.2MP forward slide, about 24 hours total runtime per game (almost 200 hours equivalent of a single-processor machine). Depth between 20-24 plies per move.|
10.a3 ... black score better than -10.00 at move 39, mate in move 61
10.Qa4 ... black score better than -10.00 at move 43, mate in move 63
Bottom line: with gobs of silicon power thrown at the game, it matters little which losing move white choses.
|May-28-18|| ||Phony Benoni: "The Fireside Book of Chess" (Chernev & Reinfeld, 1949) dates this game as 1<1935>. Talk about your slow mail service.|
< Earlier Kibitzing · PAGE 3 OF 3 ·