Tuesday, September 30, 2014

The horizon

If there is a red line in my chess then it must be that I try to avoid chance as much as possible. This can be seen e.g. in a reaction of Kara in which he expresses his surprise about my depth of preparation. In my article which games to analyze I explain in detail how I try to extract lessons from the analyses. Again I try to arm myself against haphazard repetitions. In my OTB-games you can clearly see an allergy for risks to avoid that the result depends too much on luck. For this I already once received right or wrong critique see Lintons reaction on the article Tactic.

However assuming that I always avoid risks is nonsense as I am e.g. no pragmatic player, see chessintuition part 2 or somebody never daring to play a gambit. Now I do admit that the balance clearly leans to prudence and especially playing economically. Playing economically was already once touched in a reaction of my article my most beautiful move. If I can choose between sacrificing material of which the complications are obscure and between (preferably without spending much reflection-time) a quiet continuation which still permits to maintain a position with some prospects then I choose invariably for the second option.

So it happened in round 5 against the British player Andrew Stone that I after a long reflection anyway didn't sacrifice my knight but preferred to retract it to f6. I imagine MNb will probably be shocked again if he sees that I once more chose for the retracting move but sacrificing somebody else's pieces is always easier.
[Event "Open Gent 5de ronde"] [Date "2014"] [White "Stone, A."] [Black "Brabo"] [Result "*"] [ECO "A10"] [WhiteElo "2200"] [BlackElo "2333"] [SetUp "1"] [FEN "r1b2rk1/1p3qpp/3b4/p1p1p3/2PpPpn1/3P2PP/PPNB1PB1/1R1Q1RK1 b - - 0 19"] [PlyCount "13"] 19... f3 $1 {(I looked at this move at least 10 minutes during the game but I did not succeed to calculate all the consequences. Eventually I chose for Nf6 which is risk-free and still sufficient for a small advantage. By the way also Stockfish recommends Nf6.)} 20. hxg4 {(The refutation according to the engines but Bxf3 is better.)} (20. Bh1 Qh5 $1 {(The engines prefer Nh6 but black has of course to select the same road as in the mainline.)} 21. h4 { (Hxg4 transposes to the mainline but h4 is an important alternative. So f3 hangs and I was not sure in the game of the consequences.)} Be7 {(Black wants to demolish the kingside with a sacrifice on h4. The engines have no difficulty to proof that whites position is desperate.)} 22. Ne1 (22. Bxf3 Bxh4 23. Qe2 Bg5 24. Bxg4 Bxg4 25. f3 Bxf3 26. Qh2 Qxh2 27. Kxh2 Bxd2 $19) 22... Bxh4 23. Nxf3 Bxg3 {(The fastest and the most spectacular win.)} 24. fxg3 Rxf3 25. Rxf3 Qh2 26. Kf1 Qxh1 27. Ke2 Qg2 28. Ke1 Nh2 29. Rf2 {(White had to give up the rook as now he is mated immediately.)} Qg1 30. Ke2 Bg4 31. Rf3 Bxf3#) (20. Bxf3 $1 Qxf3 21. Qxf3 Rxf3 22. hxg4 (22. Ne1 Rxf2 23. Rxf2 Nxf2 24. Kxf2 Bxh3 {(Till here I had calculated in the game and I judged this as very good for me. My engines confirm and even show it is winning provided that some accurate moves are played.)} 25. b3 (25. Kg1 a4 $19 {(An important move to keep the queen-side flexible.)}) 25... Rf8 26. Kg1 Rf1 27. Kh2 Rf2 28. Kxh3 Rxd2 $19) 22... Rxd3 23. Rfd1 Rf3 $1 {(During the game I looked at Bxg4 with a clear advantage for black but I have to admit that the subtle Rf3 shown by the engines is still stronger.)} 24. Ne1 Rf7 25. f3 Be6 26. b3 a4 27. Ra1 b5 $19 {(Black started with an attack on the king-side but now breaks through winning on the queen-side despite the material-equilibrium.)}) 20... Bxg4 $1 {(Fxg2 is the move which the engines prefer even after minutes calculating but that does not give a winning advantage. However once the piece-sacrifice Bxg4 is executed on the board then it does not take long before they adapt drastically their evaluation.)} 21. Bh1 Qh5 { (At the board I mainly looked at the plan of putting a rook on h6. With the bishop on c1 this is not simple but by playing Ra6-Be7-Rf6 it becomes possible. The question of course is if black has sufficient time. I was pretty confident about it in the game.)} 22. Bc1 $5 {(A profound move which indicates big problems for white. The purpose is to play Qd2 so the queen can be sacrificed on h6 if black wants to maneuver both rooks to h6.)} Rf6 23. Ne1 $5 Raf8 {(The plan with Ra6 which I discovered during the game, wins too but Raf8 is still stronger.)}24. a3 Be7 25. b4 g5 $19 {(Rb6 is maybe a bit stronger but this idea with g5 which I missed in the game, is sufficient proof of whites hopeless situation.)} *
After the game it took me a lot of effort to verify the piece-sacrifice but now I dare to state that it is fully correct. However I would not mention this if there was nothing special about. When I let the engines Houdini 2 and Stockfish 4 calculate on the critical position then none of them found the key-move even on my fastest PC. Something like that I hadn't encountered before with those programs. Was it still possible as human to find a tactical idea at the board which was beyond the horizon of the best engines?

Via the wikispace of testpositions for chess-engines I tried to find recent examples from the tournament-chess. However I didn't have much luck as everything which I checked was pretty quickly solved by my top-engines. E.g. also the testposition 201 out of  the standard arasan testsuite. This is an extract from the game Hikaru Nakamura - Anish Giri played in the 2012 Fide Grandprix at Londen.
[Event "1st FIDE GP London 2012"] [Site "London ENG"] [Date "2012.10.02"] [Round "10"] [White "Nakamura, Hi"] [Black "Giri, A."] [Result "1-0"] [ECO "C42"] [WhiteElo "2783"] [BlackElo "2730"] [SetUp "1"] [FEN "r3k3/1p4p1/1Bb1Bp1p/P1p1bP1P/2Pp2P1/3P4/5K2/4R3 w - - 0 47"] [PlyCount "37"] 47. g5 {(A fabulous move. It takes Houdini about 50 seconds to find it. Stockfish needs a bit more with 80 seconds.)} hxg5 48. h6 gxh6 49. Rxe5 fxe5 50. f6 Bd7 51. f7 Ke7 52. Bxd7 Kxd7 53. Bxc5 h5 54. f8=Q Rxf8 55. Bxf8 h4 56. Bh6 g4 57. Bg5 h3 58. Bh4 Kd6 59. Bg3 Ke6 60. Ke2 Kd6 61. Kd2 Kc5 62. Bxe5 Kb4 63. Kc2 Kxa5 64. Kb3 Kb6 65. Bxd4 1-0
Therefore I also looked at some positions from older games which some testers use. One of them was a critical position of the famous game David Bronstein - Ljubomir Ljubjevic. I recently bumped by accident on this game when reading My great predecessors part 2.
[Event "Petropolis Interzonal"] [Site "Petropolis"] [Date "1973.08.07"] [Round "11"] [White "Bronstein, David I"] [Black "Ljubojevic, Ljubomir"] [Result "1-0"] [ECO "B03"] [WhiteElo "2585"] [BlackElo "2570"] [SetUp "1"] [FEN "r4rk1/pp1n1p1p/1nqP2p1/2b1P1B1/4NQ2/1B3P2/PP2K2P/2R5 w - - 0 22"] [PlyCount "39"] [EventDate "1973.07.23"] [EventRounds "17"] [EventCountry "BRA"] [EventCategory "12"] 22. Rxc5 {(My 2 top engines react very differently on this position. Houdini shows Rxc5 immediately but Stockfish even after an hour is still stuck by a4 which nevertheless is also evaluated as winning. However once Rxc5 is executed on the board then Stockfish almost immediately admits that it wins much quicker than a4.)} Nxc5 23. Nf6 Kh8 24. Qh4 Qb5 25. Ke3 h5 26. Nxh5 Qxb3 27. axb3 Nd5 28. Kd4 Ne6 29. Kxd5 Nxg5 30. Nf6 Kg7 31. Qxg5 Rfd8 32. e6 fxe6 33. Kxe6 Rf8 34. d7 a5 35. Ng4 Ra6 36. Ke5 Rf5 37. Qxf5 gxf5 38. d8=Q fxg4 39. Qd7 Kh6 40. Qxb7 Rg6 41. f4 1-0
It is naturally not because I can't find immediately examples from the tournament practice which engines can't solve that they don't exist. However from an older blogarticle Shirovs brilliant Bh3 we can deduct that the examples are not widely spread anymore. I am curious if there are readers knowing such specific positions from practice or maybe encountered them when analyzing their own games. At chess problems composers often work several days which permits sometimes still to fool the best engines. An example of such puzzle I found on a forum in which white gives mate in 60 moves !
[Event "Mate in 60 moves"] [Date "1949"] [White "Karel"] [Black "Fabel"] [Result "1-0"] [SetUp "1"] [FEN "8/4K3/4NN2/p3p3/rnp1p3/1pk5/bp1n4/qrb1N3 w - - 0 1"] [PlyCount "119"] 1. Ke8 Ra3 2. Kd7 Ra4 3. Kc8 Ra3 4. Kb7 Ra4 5. Kb6 Ra3 6. Kb5 e3 7. Kb6 Ra4 8. Kb7 Ra3 9. Kc8 Ra4 10. Kd7 Ra3 11. Ke8 Ra4 12. Kf8 Ra3 13. Kf7 Ra4 14. Ke8 Ra3 15. Kd7 Ra4 16. Kc8 Ra3 17. Kb7 Ra4 18. Kb6 Ra3 19. Kb5 e4 20. Kb6 Ra4 21. Kb7 Ra3 22. Kc8 Ra4 23. Kd7 Ra3 24. Ke8 Ra4 25. Kf8 Ra3 26. Kf7 Ra4 27. Ke8 Ra3 28. Kd7 Ra4 29. Kc8 Ra3 30. Kb7 Ra4 31. Kb6 Ra3 32. Kb5 e2 33. Kb6 Ra4 34. Kb7 Ra3 35. Kc8 Ra4 36. Kd7 Ra3 37. Ke8 Ra4 38. Kf8 Ra3 39. Kf7 Ra4 40. Ke8 Ra3 41. Kd7 Ra4 42. Kc8 Ra3 43. Kb7 Ra4 44. Kb6 Ra3 45. Kb5 e3 46. Kb6 Ra4 47. Kb7 Ra3 48. Kc8 Ra4 49. Kd7 Ra3 50. Ke8 Ra4 51. Kf8 Ra3 52. Kf7 Ra4 53. Ke8 Ra3 54. Kd7 Ra4 55. Kc8 Ra3 56. Kb7 Ra4 57. Kb6 Ra3 58. Kb5 a4 59. Kb6 {(Black finally run out of moves.)} Nf1 (59... Nd3 60. Nd5#) 60. Ne4# 1-0
Of course this is not a normal position anymore but it does show that the human player isn't fully defeated by the engines. In the category of exceptional positions certainly belongs also the position of my game. The temporarily locked bishop on h1 and the preliminary control of the critical square h6 are a funny concurrence which engines today can't handle. To be more precise the HW and SW which I use today can't. Some readers certainly possess stronger equipment which maybe can sufficiently shift the horizon so a different image is created.



  1. As a correspondence chess player I could probably speak for hours about the chess engines and their evaluations of different positions. The overall opinion of most of the OTB players is that engines are so strong, that no human can "argue" with them. And I can only smile when even strong OTB players ask me how can I play correspondence chess in these modern times where everybody uses strong chess engines and fast computers. As I am too tired to explain through and through in which positions the computers are weaker than humans and why, I won't do it here, too. I won't speak. Instead, I prefer to show you a game played in late year 2012, a correspondence chess game between two of my countrymen-colleagues who both played it with engines' assistance. The first player is a FIDE IM, while the second one....well, just an engine operator, as it seems: https://www.iccf.com/game?id=357514 The first game shown here on this page, somehow reminded me of the game I just posted. You can copy and paste it in your chess software, run your engines and see their evaluations! And you'll understand what I mean!

  2. My title seems to be misleading if I interpret properly your reaction. Yes long term advantages like in the game you show are clearly beyond the horizon of an engine but that is not the idea of the article.
    With the article I am trying to discover if there exist still positions in which a human can outsmart an engine in pure tactics so without getting any external help to calculate. The example that you give doesn't contain tactics like I show in my article.

    I am interested in OTB games in which this tactical horizon is demonstrated. Also I don't want to consider correspondence games as there is no clear line between human input and engine output.

  3. First there were the Nolot positions, of which the Bronstein - Ljubjevic position is a member, which were famously difficult for computers to solve. Even Deep Thought, the predecessor to Deep Blue, analyzed them. Nowadays, however, they've all been solved, some of them trivially, or been proven incorrect. The toughest was probably 13...axb5 from Malaniuk - Ivanchuk, 1988, but it has now been solved in the last few years. Shirov's ...Bh3 was another long-time hurdle, but has been found by engines too as you know.

    One more famous position that has been resistant to computers has been Gusev - Averbakh, 1946. 4q1kr/p6p/1prQPppB/4n3/4P3/2P5/PP2B2P/R5K1 w - - bm Qxe5

    Nevertheless, a combination of Houdini, big hardware, and probably a little bit of luck managed to find that one as well after searching 33 billion positions. As you say, it's getting tougher to find a deep tactic that can stump engines for long.

  4. Engines' parameters can be modified to exclusively find tactics in practically every position. So, they are able to find every tactic blow, in fact....but only if a human knows there is one in a specific position and let them find it. Strong correspondence chess players know this and when they "smell something", they let them loose. By the way, Houdini 4 in tactical regime is not the strongest one in this specific area. There are several others that can do better job. Anyway, the problem is that one can't always analyse with a modified engine, because when modified, such an engine is slow in calculating. So, being a strong OTB player (who can "smell something" in a concrete position) is obligatory to become a strong correspondence chess player.
    As for the tactics themselves, we all know that tactics are not only sparkling blows and sacrifices. Long time strategy tactics are what the engines can't solve, for now. For example, Topalov's favourite quality sacrifices which only change the balance and can be crucial (or not) in the long run, sometimes 20-30 moves ahead. Such tactical strikes are difficult to be properly evaluated by the today's engines, because they prune variations much sooner, if not "seeing" fast results. One specific case is when, as a result of a little pawn sacrifice or alike, an opponent's piece is isolated for a long time. That's why I showed the above game, probably inspired by the game brabo analysed here.

  5. "Engines' parameters can be modified to exclusively find tactics in practically every position. "
    I agree but again this article is about using an engine without modifying anything at all. I am comparing pure human power with pure engine power. B.t.w. the analysis published in the article is clearly a cooperation between human and engine. I didn't play the move in OTB although I did see a major part of the variations without any help.

    "Topalov's favourite quality sacrifices which only change the balance and can be crucial (or not) in the long run, sometimes 20-30 moves ahead."
    For me these are positional sacrifices and not tactical ones. In my game it is a tactical sacrifice as whites counterplay on the queen-side can easily unlock the bishop of h1 if blacks king-side attack is not fast enough. The mainlines are only 10 moves deep often ending with mate so pretty amazing that today's top-engines have so much troubles finding it.

  6. If you want to read the original version of the chessbase article then you have to use a different link: http://chessbookreviews.wordpress.com/2014/09/11/choosing-a-chess-engine/


Note: Only a member of this blog may post a comment.