In my new book “The Silicon Road to Chess Improvement” I spend a whole chapter on engine sacrifices. One of the many sacrifices that engines play nowadays are sacrifices to clear lines for their major pieces. We saw many such long-term sacrifices from AlphaZero and modern engines are just as ready – indeed eager – to do so! One great example was from the game Combusken-rofChade in round 3 of the ongoing TCEC Swiss 2 (https://tcec-chess.com/#div=sw2&game=102&season=21)
Engines can teach us how to play like Tal or Shirov at their best! Understanding this theme is a key building block in creating promising attacking positions.
This position arose after 11 moves of the sharp 7.g4 Semi-Slav popularized by Alexei Shirov. The g4-pawn is hanging but somewhat poisoned as after …Nxg4, Rg1 White wins back the pawn on g7.
Many moves come to mind in this position: 12.g5, 12.Rg1, maybe 12.Bf4 or 12.Bd2. Not however the move Combusken played!
12.Be3!? Nxg4
This pawn grab is possible now as 13.Rg1 is met by 13…Nxe3! What is Combusken’s idea?
13.Bd2!
Position after 13.Bd2! (game)
That was it! Combusken was simply looking to create an open file for its rook! So it tempted Black to take on g4 by dangling its bishop on e3! The open g-file proves to be very awkward for Black, and the extra pawn for the moment plays no part.
13…Ngf6 14.Rg1
14…g5
It’s already a bit awkward for Black to defend the g7-pawn. 14…Kf8 is possible, but it’s not particularly attractive to block in the rook on h8. 14…g6 15.e5 Nd5 16.Rxg6 is completely disastrous.
15.h4 gxh4 16.Bf4 Bb7 17.0–0–0 Qa5 18.Nd2 Ba6 19.Nc4 Bxc4 20.Bxc4
White is two pawns down, but somehow it doesn’t look like it! White’s space advantage is substantial and very solid: it’s hard to see how Black can safely challenge it, particularly as Black lacks a safe spot for the king. As you can see from the rest of the game, Combusken gradually encroached on rofChade’s position, rofChade seeming powerless to strike back at the white position.
20…Rg8 21.Rxg8+ Nxg8 22.Kb1 Rc8 23.Be2 Ngf6 24.b3 h5 25.Nb2 Nb6 26.Bf1 Nh7 27.f3 Nf8 28.Qg2 Ng6 29.Be3 Nd7 30.Bc4 Ngf8 31.Rg1 Nf6 32.Bf2 a6 33.Rc1 Ng6 34.Be3 Nf8 35.Nd3 N6d7 36.f4 c5 37.d5 Qb6 38.dxe6 fxe6 39.f5 e5 40.Rd1 Qf6 41.Qh2 Qc6 42.Bd5 Qf6 43.Bc1 Kd8 44.Bb2 Bd6 45.Bb7 Rb8 46.Bxa6 Qe7 47.Bc1 Kc7 48.Nb2 Nb6 49.Rg1 Qh7 50.Nc4 Nxc4 51.Bxc4 Nd7 52.Rg6 Rb6 53.Qxh4 Bf8 54.Be6 Rc6 55.Bg5 Rd6 56.Bg8 Qh8 57.Bc4 Qh7 58.Be6 Kb8 59.Kc2 c4 60.Bg8 Qh8 61.Bxc4 Qh7 62.Bg8 Qh8 63.Be6 Qh7 64.Qe1 Rc6+ 65.Kb1 h4 66.Qxh4 Qxh4 67.Bxh4 Kc7 68.Rg8 Rd6 69.Bxd7 Bh6 70.Be6 Rd1+ 71.Kc2 Rd2+ 72.Kc1 Rg2+ 73.Kb1 Rh2 74.Bg5 Bxg5 75.Rxg5 Rf2 76.Rg6 Rf4 77.Bd5 Kd7 78.f6 Rf3 79.f7 Rf1+ 80.Kc2 Rf2+ 81.Kd3 Ke7 82.Kc4 Rc2+ 83.Kxb4 Rf2 84.Rg8 Rxf7 85.Bxf7 Kxf7 86.Rg2 Kf6 87.Kc5 Ke7 88.Kc6 Ke6 89.Rf2 Ke7 90.Kc7 Ke6 91.a3 Ke7 92.b4 Ke6 93.Kd8 Kd6 94.Rf6#
1–0
Deepmind just published [this article](https://arxiv.org/pdf/2111.09259.pdf) recently about the evolution of (multiple instances of) AlphaZero’s opening theory and opening preferences as it trains and becomes stronger, and how it compares to human opening theory throughout history. I thought it would be interesting for you.
Thanks Madeleine! That’s really fascinating – thanks for pointing it out! Best Wishes, Matthew