The Superfinal of the 14th edition of the TCEC is becoming an incredible show. The traditional competition between engines, which has always lately been dominated by consecutive versions of Stockfish, has encountered a new spectacular finalist: Leela, an engine that, despite being only 10 months old, already proved superiority in the last round to well-established engines such as Komodo and Houdini. In this article we will go through some of the most memorable moments of the Superfinal, which has been ongoing since the 4th of February.
In the last few years, public interest in engines has dramatically increased. Nowadays, not only are we concerned about using them as an essential tool for analysis, but we also want to know which one is the best!
After the astonishing appearance of AlphaZero in December 2017, many things have changed. Stockfish has constantly been improving after the release of each new version, and many people believe it might already have reached the level of being able to beat older versions of AlphaZero. On the other hand, Leela, an open source engine, is based on DeepMind’s publication about AlphaZero. Leela improves only by self-training, as she started out only knowing the basic rules of chess.
At the time this article is being written, the match couldn’t be more equal; it’s a tie at 22 points each! As the match is set to last 100 games, the result will most likely be uncertain until the very last minute.
As expected, both engines have left us with memorable high quality games. We will now show some of the most brilliant moments that we have encountered in the match thus far:
1. d4 ♘f6 2. c4 e6 3. g3 c5 4. d5 exd5 5. cxd5 b5 Until here it was all theory for both engines. In the TCEC they choose "fun" positions that both engines have to play with White and Black. Stockfish finds what is considered to be the "refutation" of this line.
19. ♕g6! Initiating a piece and pawn sacrifice for a non-forced attack. This whole next sequence was believed by the engine to be forced and was calculated by Stockfish already by move 16, otherwise it would have evaluated that the sacrifice didn't work!
23... ♗d4 A good defensive try!
26. ♖e1! Taking away the escape squares for Black's king.
29. g4! Another piece into the final attack! The white pawn becomes decisive.
29... ♘c7 30. g5 The game is now decided. Stockfish had to have seen this variation already by move 16 or even 14 to enter it, otherwise its evaluation would have been different and it would have played a calmer move instead.
30... ♘e8 31. ♖g1 ♔e6 32. gxf6 ♘xf6 33. ♕g6 ♔e7 34. ♖e1+ ♔d8 35. ♖xf6 gxf6 36. ♕xg8+ ♔c7 37. ♕g7 ♔b7 38. ♔g3 ♔a6 39. h4 b4 40. ♗f4 c3 41. bxc3 ♕xc3+ 42. ♖e3 ♕a1 43. ♕xd7 ♖g8+ 44. ♗g5 ♕g1+ 45. ♔h3 ♕f1+ 46. ♔g4 ♕g1+ 47. ♖g3 ♕b6 48. ♖d3 fxg5 49. ♖d6 gxh4+ 50. ♔xh4 ♖h8+ 51. ♔g5 ♖g8+ 52. ♔f6 ♖f8+
There is no doubt that Leela has been pressuring Stockfish, but how do you decide the game? Stockfish removed the rook allowing g4 to be played, giving White an "easy" win. Both engines saw it immediately after it was played. The question is, how could White win if Black would have played Ke6 instead? If you have time and feel like improving your endgame technique, try to come up with some ideas!
55... ♔e6 To win this endgame, White needs to force zugzwang. 56. h4 This advance, even though it doesn't look very natural, will be the key. The idea will be to sacrifice on g4, so that if its taken with the h-pawn, h5 can be played and White can create a passed pawn. 56... ♔d6 Not wanting to remove the rook on f7 to "avoid" g4. (56... ♖c7 57. ♗b1 ♖f7 gets to the same position.) 57. ♗b1! A precise move that you only understand several moves later. The black king will be worse on e6 than on d6 because it will limit the movement of the bishop on b7. 57... ♔e6 58. g4! A beautiful and instructive pawn sacrifice to open up the kingside. 58... fxg4+ (58... hxg4 59. h5 gxh5 60. g6 ♖c7 61. ♔g5! and the passed g-pawn is worth more than the 3 black pawns on the kingside. When there are passed pawns it's important to remember that many times the quality of the pawn is more important than the quantity.) 59. ♔g3 ♗c8 (59... ♖f3+ 60. ♔g2 ♗c8 61. ♗xg6 ♖h3 would be very close to draw with the king on d6, but... 62. ♗f7+! ♔f5 63. ♖c7 and the bishop is trapped!) 60. ♖a3! The pawn on g6 will fall and after that the whole chain of pawns on the kingside will follow. 60... ♖g7 61. ♖a8 ♖c7 (61... ♗d7 62. ♖b8 and the pawn on b6 falls! Just like the last one, this variation would not be possible if the black king was on d6... ; 61... ♖g8 62. ♗xg6 ) 62. ♗xg6 and White's g-pawn will decide the game.
60. ♗f5! ♗xf5 61. ♔xf5 ♔f8 and the game ended in Leela's favour. It is not precisely easy for a human to see everything but I will leave it for you as a problem that you can solve in the comment section.
The games between engines are very complex, profound and instructive if worked with properly. In addition, given the equality between these engines, we will experience a lot of excitement in the TCEC superfinal. Who will win? LCZero or Stockfish? In a few weeks we’ll know!
We respect your privacy and data protection guidelines. Some components of our site require cookies or local storage that handles personal information.