Home Page Chess Life Online 2015 March Chess Life Bonus: More Games From Thoresen Competition
|Chess Life Bonus: More Games From Thoresen Competition|
|By IM Erik Kislik|
|April 1, 2015|
the April Chess Life, available
now to USCF Members by clicking here, IM Erik Kislik writes
about the Thoresen Chess Engines Competition (TCEC) which features perhaps the
highest-level chess being played these days. Kislik provided wonderfully
detailed annotations, but it all couldn't fit within the article. Some of the
references in the below article are more fully developed within the print story,
such as the opening book discussion. We hope you enjoy this bonus material.
Before I get into a more nuanced discussion of computer chess and how we can learn from it, I would like to show a game from Season One between Rybka 4 and Houdini 1.5a that was very eye-opening. This was played in game one of the "Elite Match" (now known as the SuperFinal). Rybka was currently the reigning champion and it was unclear how the untested Houdini would fare against it. This game demonstrated a number of interesting peculiarities in computer chess: namely how engines handle material imbalances, the bishop pair, and a weak king position.
A lot of room to go wrong
TCEC Season One Elite Match (1), 2/2011
This position is very interesting for a number of reasons. First, Black has sacrificed a pawn for activity and is willing to give up a second pawn on d3 for maximum activity and play against the white king. Second, White lacks a clear plan and an obvious way to complete development, so there is a lot of room to go wrong here due to the lack of straightforwardness. Rybka started with the riskiest move, immediately grabbing the second pawn and forcing its own king out into the open.
I don't like this move at all, because Black's play now becomes so obvious that I would expect many strong players to find all of Black's next few moves without much effort. After the king takes back on d3 and Black plays Na4, the plan of following up with ... b5-b4 is not easy to stop. By not being so materialistic, White could have considered improving each of his pieces individually in a much calmer fashion. Most importantly, his a3-knight would have made more sense if he hadn't taken on d3, as at least it would restrict Black's b6-knight in that case.
It is difficult to try to criticize engine games with simple human logic but I will try to do so here. By playing the move 17. h4 White benefits in numerous ways: He advances his flank passed pawn, threatens to take on d3 under better circumstances, keeps the knight on g5 to hinder 0-0-0 (in view of the weakness of f7), and in some cases the move Rh2 turns out to be surprisingly effective in defending White's kingside structure. Black has few obvious moves besides 17. ... Bc6 here. One additional advantage of not taking on d3 is shown: Now the b6-knight is restricted and has difficulty moving in view of its need to defend the c4-pawn. 17. h4! Bc6 (White is winning after 17. ... 0-0-0? 18. Be3) 18. Nf3! Now that g2 is defended, White can consider taking on d3 or playing Rb1 followed by Be3, which would help White complete his development. 18. ... 0-0-0 19. Rb1 White now prepares the obvious Be3, when Black fails to prove any serious compensation. 19. ... Bf6! Black frees up the g-file and puts pressure on the h4-pawn. (19. ... Nxc1+ 20. Rbxc1 has the nice idea of Ng5 followed by Be4, which would most likely get rid of Black's bishop pair.) 20. Rh2!? This move appeals to me a lot because it plans to simply play Be3 or g3, restricting Black's knights and moving closer to the completion of his development. (20. Be3? is the most obvious try, but chasing material is problematic in view of White's weak kingside: 20. ... Nd5 21. Nxc4?! N5f4+! 22. Kf1 Nxg2! is bad for White; 20. g3 Rh5 threatens the very dangerous ... Rf5 and gives Black an unpleasant initiative.) 20. ... Nd5! This threatens to play ... b5, locking out the a3-knight. (White is slightly better after 20. ... Kb8 21. g3) 21. Kf1 Nxc1! (21. ... b5 22. Bxd3 cxd3 23. Bd2 prepares to play the calm Re1! and forces Black to now seek out a way to equalize.) 22. Rxc1 Nf4 23. Nxc4 Nxg2! 24. Nfe5!? leads to wild and intricate play, in which White is not worse. In summary, White could have chosen a very calm defensive plan and held the balance. This is an odd situation where I would expect human logic in many cases to trump low depth computer analysis. A human will try to think about how to get all of the pieces working together effectively and how to solve all of his problems, namely his king position. With a number of precise moves, White puts the pressure on Black to prove he has enough compensation. In the game continuation on the other hand, the burden was squarely on White's shoulders to demonstrate he was objectively fine; 17.Nf3 One of the most obvious moves is this one, intending to try to consolidate by bringing White's pieces back to safe squares, yet the d3-knight is quite a thorn for White to deal with in this case. 17. ... 0-0-0 18. Rd1 e5 fails to drive the knight out of d3 how White would like to; White has the upper hand after 17. Rb1 Bc6.
17. ... cxd3+ 18. Kxd3
Playing through the game now with the data available on the TCEC website, I can see that Rybka evaluates its advantage as +.38 at depth 20 after playing 18. Kxd3. This is a very materialistic evaluation and one which has no particular meaning unless White can concretely complete its development. The evaluation looks simply incorrect to me though, and goes against human intuition here: It is hard to imagine in what circumstances Black can ever be worse here, or even how exactly White can get his pieces out without losing a lot of material. The white king is weak, Black's two bishops are unhindered, and White has trouble dealing with Black's most obvious threats.
18. ... Na4!
Black entertains many aggressive ideas now, including even the extravagant ... Rh6-b6!?
This is a solid move, intending to play h4 and g4 in the near future, when the pawns may begin to counter for something.
19. Nf3 Rh5 20. Ke2 would have attempted to consolidate White's position a little bit. Nevertheless, the threat of ... b4 is still difficult to deal with after 20. ... b5 21. Nc2 is the most obvious move to restrict ... b4, yet 21. ... b4! is a nifty sacrifice aimed at the fact that White's king has no safe square to go to. 22. Nxb4 Bb5+ 23. Ke1 Nxc3 and White's king comes under very unpleasant fire.
19. ... a5! 20. Ne4?
20. Ne4 fails completely at the goal of consolidating and must be regarded as a major error, since it wastes a lot of time in a concrete position. Moving the other knight was far more logical.
20. Nc2! b5 21. a3! would have been the only way to stop ... b4 and try to hold on tight. Black still retains slightly better chances after 21. ... Nc5+ 22. Ke2 Bf5 23. Ne4! White uses a common defensive tactic when up material: by giving back a pawn, he gets rid of Black's strong c5-knight and is close to consolidating his position. 23. ... Nxe4 24. fxe4 Bxe4 25. Ne1 Be5 26. h4 leads to a position that is still fairly difficult for White, but only slightly worse. Black has ideas like ... Rg8 and ... Ra6, although it is not clear if he has any more than full compensation for the pawn and a slight initiative.
20. ... f5!
This is the obvious move that must be played sooner or later: Black drives the knight away from e4 and increases his initiative.
Rybka still sensed no danger here, giving a nebulous and meaningless 0.00 score. Granted, it was a low depth (19), yet it is more evidence that one needs to be cautious about trusting numbers that don't immediately make sense to the human mind.
21. Ng3! would have given White chances to defend by sticking the knight on e2 and hoping for the best. At the very least this would have been much better than the game continuation, in which the knight had nothing to do on f2 at all. 21. ... Rc8 22. Ne2 b5 23. b3! helps get the c1-bishop out and exchange off some pieces, to ease the defense. 23. ... Nxc3 24. Bb2 Nxe2 25. Bxg7 Nf4+ 26. Kd2 Rh7 27. g3 Rxg7 28. gxf4 Rg2+ 29. Kd3 b4 30. Nc4 Be6 is definitely better for Black, but there is no win in sight.
21. ... b5 22. Nc2
22. b3 is a desperate engine try, although Black is probably still winning after 22. ... Nxc3 23. Bd2 Nd5 24. Rae1 a4.
22. ... b4!
This crucial move opens up everything and exposes how weak White's queenside and how unstable his pieces are, especially his king.
23. c4 f4 24. Bxf4 Nxb2+ 25. Ke2 Nxc4 26. Nd3 Rd8 is also completely winning.
23. ... Kf7
Black threatens ... Ra6 or ... Rhc8 with decisive effect. 23. ... Rc8 is also strong, but there is a certain beauty to the slowness of the move 23. ... Kf7!?
24. Ke3 Ra6 and White's king will not survive without major loss of material. (Also possible is 24. ... Rhc8!?)
24. ... Rxa5 25. Kd2 Rd8
Black is now completely winning, and with a couple of accurate moves, won material:
26. Nb4 Re5 27. Nfd3 Bb5 28. Re1 Nc5 29. Rxe5 Bxe5 30. f4
30. a4 Nxd3 31. axb5 Nxb4+ 32. Ke2 Nc2 33. Ra4 Nd4+ 34. Kf2 Rc8 35. Bd2 Rc2 also wins for Black.
30. ... Bf6
Black won a piece and kept the bishop pair, leaving White completely paralyzed. I'll include the rest of the moves with no comment just for clarity since it was a very clean conversion:
31. Ke1 Nxd3+ 32. Nxd3 Bxd3 33. a4 Rc8 34. a5 Rc2 35. Bd2 Rxb2 36. a6 Be4 37. Ra3 Bxg2 38. a7 Rb1+ 39. Ke2 Ba8 40. Be1 Bd4 41. Ra2 Rb3 42. Bg3 Ke6 43. Kf1 Bc5 44. Ke2 Kd7 45. Kf1 Rb4 46. Ke1 Bd6 47. Kf2 Bxf4 48. h4 Bh6 49. Kf1 Rb1+ 50. Be1 e5 51. h5 f4 52. Rd2+ Kc7 53. Rc2+ Kb6, White resigned.
The previous game was fascinating because numerous mistakes were made by Rybka that were not so difficult to point out. We were also able to see generally what flaws a lot of engines of the time had. Season One's elite match ended in a crushing victory for Houdini over Rybka by the score of 23.5-16.5 and effectively put both Houdini and TCEC on the map.
Season Seven featured one of the most amazing games I have ever seen before, which quite deservedly won "Craziest Game of the Year" on chess.com. Let's try to make sense of some of the moves in this incredible encounter. The game was played in stage two of Season Seven, which was played with no opening book. All other stages of this season of the tournament had an opening book that started the game on White's ninth move.
Slav Defense (D11)
TCEC Season Seven (stage two), 2014
1. d4 d5 2. c4 dxc4
It is rare to see an engine enter the Queen's Gambit Accepted. Most engines will grant White a comfortable advantage in view of the fact that White's bishop comes out to c4 quickly and he has additional central control after e3 and Bxc4.
3. e3 Be6?!
This is a very rare move that optically looks very risky in view of the f8-bishop's inability to find a decent square now. Watch how it gets crushed. 3. ... e6 4. Bxc4 Nf6 5. Nf3 a6 6. 0-0 c5 is the main line, in which Black intends to play ... b5 and ... Bb7 to develop comfortably if White doesn't hinder him.
4. Nf3 c6 5. Nbd2 b5
Trying to hold onto the pawn is consistent, so it's not too surprising to see it show up on the board. Unless Black is going to fight hard to hold onto his material, he should not play 3. ... Be6. 5. ... Nf6 6. Nxc4 is clearly not what Black wants. 6. ... Nbd7 7. b3 Bf5 8. Ba3 e6 9. Bxf8 Nxf8 10. Bd3 makes it difficult for Black to fully equalize in view of his passive pawn structure.
6. a4 Nf6 7. axb5 cxb5 8. b3!
... and now White gets his pawn back by force. In the complexity that follows, Black is merely hoping to cling to his c-pawn for as long as possible.
8. ... c3 9. Bxb5+ Bd7 10. Bxd7+ Nbxd7 11. Nb1!
This allows White to try to slowly surround the pawn after Qc2.
11. ... Rc8 12. Qc2
White threatens the natural Nxc3.
12. ... Nd5
Black threatens ... Nb4 so 13. Ba3 is not only forced, but what White wants to play anyway.
13. ... e5!
This is Black's best try to simplify matters, but it does not seem to solve all of his problems in the concrete lines that follow.
14. Bxf8 Kxf8 15. 0-0
White has various threats, including dxe5, Rc1, and Rxa7.
15. ... Nb4 16. Qe4 c2 17. Na3! Nf6 18. Qb7!
This cumbersome move turns out to be clearly best, in view of the fact that 18. Qxe5 Nd3 would have won material for Black.
18. ... Nfd5 19. Nxe5
White finally gets a clear threat: mate on f7. White's queen play that follows is very curious and difficult to plan from afar as a human player.
19. ... Qc7 20. Qb5!
White decides to keep the queens on the board so as to maximize his potential to create threats, starting with e3-e4 on deck now.
20. ... a6 21. Qa4
e4 is still a serious threat.
21. ... a5
This defends the b4-knight so that e4 doesn't win immediately. Now ... Nc3 is a real threat.
22. Nb5 c1=Q! 23. Raxc1 Qxc1
This is a really shocking move in view of its slowness. I am amazed just looking at it even now after seeing it 10 times. White threatens to simply take the queen. Qc2 is forced and White has the very nasty 25. Nd6, which eyes key squares such as c8, e8, and f7.
24. ... Qc2 25. Nd6
25. ... Rb8?
This is a key mistake, showing that Jonny was much too concerned with material instead of his king position. Better was 25. ... f6!. Incidentally, this is a rare case in a computer game where a human may stumble upon the best defense leading to a draw (based on human judgment of both White's attack and the coming rook ending), while a computer probably would not. 26. e4 Qc7! (26. ... fxe5 27. exd5 threatens the poisonous Qd7.) This would have been a fresh defense, challenging both knights directly, enabling Black to reach a drawn endgame after 27. Nxc8 Nc3 28. Qd7! Qxd7 29. Nxd7+ Kf7 30. Ncb6 Ne2+ 31. Kh2 Nxd4 when it looks like Black can forcibly take the b3-pawn and take play into a four versus three rook ending that is theoretically drawn: 32. Nc5 Nbc6 33. Nc4 Rb8 34. Ra1 Nxb3 35. Nxb3 Rxb3 36. Nxa5 Nxa5 37. Rxa5 Rb2 and a few more moves may demonstrate a typical drawn position in which White is not able to make any progress: 38. Ra7+ Kg6 39. Kg3 h5 40. h4 Rb3+ 41. f3 Rb2.
26. e4! Nb6 27. Qb5 Qc7
27. ... Kg8!? 28. Nexf7 Qc7 29. Qe5 Rf8 30. Qe6! snags an Exchange and leads to a position in which White has four pawns for the piece. White has good winning chances in the resulting position, but this was definitely Black's most worthwhile defensive try. 30. ... Rxf7 31. Nxf7 Qxf7 32. Qxb6 h6 33. Qxa5 Nd3 34. Qc3 Nf4 35. Qe3.
28. Ndxf7! Rg8
28. ... Kg8! 29. Nxh8 Na8 was the comical defense that Stockfish showed in its "display" mode during the game. It turns out that there is no clear win here and Black still retains some chances of drawing. 30. Qd7!? Kxh8 31. Qf5 White can play either Ra1-xa5 (since the queen would defend the rook from the side after a discovered check) or d5-d6, when Black faces a very difficult defense.
29. Qxa5 N6d5! 30. Qa1!
White has four pawns for the rook and a deadly attack. In view of the uselessness of Black's rooks, Black will not be able to extricate himself from this mess anytime soon.
30. ... Nf4 31. d5
White threatens both d6 and Ng5. There is no way to defend against all of White's threats.
31. ... Nbd3 32. Qa3+ Qe7 33. d6! Qf6 34. d7+ Qe7 35. Qa5!
Now simply queening on d8 is on the menu.
35. ... Nxe5 36. Nxe5
White threatens Rd1.
36. ... Qd6
36. ... g6 37. Nc6 wins the b8-rook at least.
This is a very brave move, walking into the line of sight of the queen. Nevertheless, in the game 38. f4 closed the window.
37. ... Ng6 38. f4
This is the only move, but fully sufficient.
38. ... Ke7
With the queen forced away, f5 is coming up.
39. ... Qd4
39. ... Qxd7 40. f5 Nh8 41. Qe5+ Kf7 42. Nd6+ turns out the lights on the black king.
40. f5 Nf8
40. ... Nh4! looked more stubborn from a human perspective: 41. Qc7 Rbd8 42. Qg3 g5!? 43. e5 Rxd7 44. e6 is given as above +4 by the engine, but a human would need to calculate out the entire line to have confidence in it. 48. Qb4 is a very hard move to see, if I do say so myself. 44. ... Ra7 45. f6+ Kxe6 46. f7 Rf8 47. Qe1+ Kd5 48. Qb4! Rfxf7 49. Qd6+ Ke4 50. Re1+ Kd3 51. Re3+ and White snags the queen. Perhaps if White sees 48. Qb4! he can convince himself the black king is just too exposed and must end up captured.
This is a phenomenal move, threatening Rd2 and the queening of the d-pawn. The rook must be captured. 41. d8=Q+ Rxd8 42. Qb4+ Kf7 43. e5 would have been a less extravagant winning line, when Black's pieces are totally and completely paralyzed and White threatens Qb7+.
41. ... Qxf2 42. Qe5+ Kxd7 43. Qxb8
White threatens Ne5+ or Nb6+ with a direct mate.
43. ... Ke7 44. Qe5+
Now White grabs the g8-rook and slowly wraps things up.
44. ... Kd8 45. Qd5+ Ke7 46. Qxg8 Qf4+ 47. Kg1
White blocks checks from the black queen with the instructive 50. Ne3!. This defensive motif was described by Mark Dvoretsky in one of his books, based on a lesson he learned from a game against GM Tigran Petrosian. The pattern certainly sticks in my mind.
47. ... Qc1+ 48. Kf2 Qf4+ 49. Ke2 Qxe4+ 50. Ne3! Qe5 51. Kf3
The king is totally safe. Being two pawns up, the win should be easy now.
51. ... h6 52. b4 Ke8
White also wins after 52. ... h5 53. Qd5.
53. Qd5 Qf6 54. Qe4+ Kf7 55. b5
The pawn can roll comfortably.
55. ... Nd7 56. Qc4+ Ke8 57. Qc8+ Qd8 58. Qxd8+ Kxd8 59. Ke4 Nf8 60. Kd5 Kc8, Black resigned.
... and Black finally gave up.
The next game I will show was also quite unusual. Black is to move in the following position (game 40 in the most recent SuperFinal).
After a Grunfeld
TCEC Season Seven SuperFinal (40), 12/23/2014
Black to move
This position came from a very complicated g3 Grunfeld variation, played often by Garry Kasparov with the black pieces in his heyday. Black must play with extreme dynamism to justify his play. I am often shocked, dazed, or confused when I see certain moves in chess games. This one is a perfect example. I do not understand the point of the move 28. ... h6 at all. I don't see any variation where it gives the black king useful luft or is connected to an effective plan in any way. 28. ... Be6! is much more critical and comes with a real threat.
28. ... h6?
White now has time to improve his pieces. 28. ... Be6! By blocking the e-file, Black threatens ... Rxe7 because the queen on d2 is undefended. White's most likely reply is 29. Rd1, but then the c-pawn starts up in motion: 29. Rd1 (29. Bh4 Bg4 is even.) 29. ... Bxc3! 30. Rxc3 Na4 and the pawn is awkward for White to deal with. 31. Rcc1 c3 32. Qc2 Nb6 gives Black excellent counterplay and is at least equal.
In the game, White was able to use the move h4 to provoke the undesirable ... h5, when Black has very unpleasant weaknesses.
29. ... Be6 30. Qd1!
Can you guess the point of this strange-looking move? White would like to play 31. Rxe6! fxe6 32. Bh3 with the queen defended on d1 at the end of the variation so that ... Rxe7 cannot be played.
30. ... Bg4 31. Qc1!
Offers up an Exchange that cannot be taken.
31. ... Bd4 32. Qd2! Bg7
Black admits that he can't take it. 32. ... Bxe3+ 33. Rxe3 plans Ne4, when Black's king position is atrocious.
Again White toys with the incredible threat of 34. a4!? Nxa4 35. Nxa4 Qxa4 36. f5!? Bxf5 37. Qd5, which has the unbelievable idea of Qxf7+!!.
33. ... Be6?!
33. ... Bh3 was a reasonable try to relieve some of the pressure, yet White's d-pawn is still very powerful. White is doing very well after 34. Bxh3 (34. Bh1!? intending Kh2 as in the game may also be strong.) 34. ... Qxh3 35. Rd1 Bxc3 (35. ... Nd7 36. Qe2 and White threatens the crushing g4!) 36. Rxc3)
34. Qc1 Bh3 35. Bh1 Bg4
Black can find absolutely nothing worthwhile to do. The e7-bishop is a very painful thorn in its side. Clearly Black wishes it could have played 28. ... Be6 and received some of his own counterplay.
White intends the consolidating Bg2 followed by b5 and a4, exactly as played in the game.
36. ... h5 37. Bg2
Black is paralyzed, so White even has ideas like Re5!?-a5, since it is unlikely that Black can get away with taking on e5.
37. ... Bf5 38. b5 Rc5 39. Bf6!
This exploits the fact that the rook on c5 is undefended (39. ... Rxe3 40. Qxe3).
39. ... Rcc8 40. Be5! f6
40. ... Red8 41. Bxg7 Kxg7 42. Rd1 Re8 This is forced since Re7 was the threat. 43. a4-a5 is too hard to stop. 43. ... Rxe3 44. Qxe3 Rd8 45. a5 Nc8 46. Qd4+ Kh7 47. Qxc4 Nxd6 48. Qd4 was also lost for Black. a7 is undefended and Nd5 is an important possibility as well.
Despite equal material, Black is unlikely to succeed in defending this.
41. ... Red8
41. ... Rxe3 42. Qxe3 Qxd6 43. a4! Rd8 44. Rd1 Qe6 45. a5 Nc8 46. Bxb7 is most likely a win for White.
42. Bxb6 axb6 43. Nd5
In view of both Nxb6 and Ne7+, White wins the Exchange and slowly converts his advantage into a win.
43. ... Qxd6 44. Ne7+ Kh7 45. Nxc8 Rxc8 46. Re7 Rc7 47. Rxc7 Qxc7 48. Bf1 Qc5 49. Qe3 Qxb5 50. Qd4
And White won on move 89. This game was an excellent example of the theme of paralysis, but more importantly we saw in a crystal-clear form the value of a tempo. When one wastes a full tempo in an extremely critical position, it is unlikely to end well. In this case, White was able to fully consolidate after Black wasted time and then it was only a question of whether Black would make a miracle escape or get a lucky save.
Unfortunately this day was not so fortuitous. This example shows that human intuition still sometimes trumps bizarre miscalculations by the computer, even at high depths on strong hardware. I greatly enjoyed following this game, so I do not mind seeing dubious play get refuted convincingly. After all, completely perfect play would not be all that much fun if every game was drawn and the superior engine was not able to demonstrate its superiority over the board.
Find the Chess Life Magazine story here.