For any author it’s a wonderful moment to see a new book published, and that happened to me a week ago with the launch of “The Silicon Road to Chess Improvement” (New in Chess). The book explains many new and original ways to use engines for training, and distills opening and middlegame strategies from the greatest and most spectacular engine games played in recent years.
I saw an interview just today with Daniil Dubov in which he said that engines have improved as much in the past 3 years as they did in the past 15 years. That may be overstating it (Stoofvlees’ creator Gian-Carlo Pascutto shared the following interesting link https://rebel13.nl/misc/stats.html which details engine improvement objectively over the years) but certainly the current level of engines is astonishing and the beauty of the chess they play is also staggering… which is why I love it so much!
The current top dog is Stockfish which defeated its eternal adversary Leela Zero 2.5-1.5 in the final of the TCEC Cup. The game Stockfish won featured an extraordinary moment which I couldn’t explain at the time so I thought it would be fun to take a closer look!
The opening – a g3 King’s Indian – was fairly biased with Stockfish evaluating White’s chances as +1.22 and Leela as +0.46. Note that the scale of Leela’s evaluations is considerably depressed in comparison to Stockfish’s. These evaluations are roughly similar: a clear White advantage, but still some way from winning.
Stockfish had held the draw as Black in the first game so now it was Leela’s turn! At this point Leela’s evaluation was just 0.53 for White but Stockfish was already at a massive 3.24 which rose to 3.55 after the next move: a huge difference in insight between the 2 engines! For myself, I understood at a glance that White is clearly better but to say that we are in guaranteed win territory would never occur to me!
I devoted a chapter in my latest book “The Silicon Road to Chess Improvement” to the early big evaluations displayed by Stockfish (you ‘ll be shocked to hear that they were all completely correct!) so I thought it would be interesting to examine this position in depth and work out what Stockfish had understood and everyone else hadn’t!
Black’s d-pawn is the most obvious weakness, disconnected from the rest of its structure and vulnerable to attack by all of White’s pieces. On the plus side however, the advance of the d-pawn has freed the d5-outpost for the black knight, allowing Black to combine knight and rook against White’s most obvious weakness: the f4-pawn.
Note however that the d5-outpost is unstable due to the weakness of the e6-pawn: pressure against the d5-knight from White’s queen and rook can easily be increased by Nxe6, knocking away the black knight’s support.
However, the black d4-pawn is not White’s greatest source of advantage in this position: it is the weakness of Black’s king (pinned to the corner due to the open g-file with just the h7-pawn in front of it) combined with White’s magisterial knight on g5 that really causes the damage.
Although it doesn’t look like it at first glance, the knight is more or less untouchable on g5 as long as White retains the option to counter …h6 with Qh5. From g5, the knight joins in to create mating threats against h7 while the pressure against e6 is tactically catastrophic for Black in many scenarios. Just imagine a Black king on g7 and a white rook on c1 capturing the knight on c7 and you’ll see what I mean! On top of that, the knight on g5 plugs the g-file, preventing any Black attempts at counterplay against the white king.
How does this add up to +3.54 though? Very simply, the tactical possibilities in the position created by the combination of the knight on g5 and vulnerable black king are so powerful that Black is essentially paralyzed. Paralyzed you say, how? Well try and find a move for Black after Stockfish’s next move!
35.Rd2!!
This caused some commotion in the TCEC chat! A waiting move and Stockfish’s evaluation rises to 3.54 while Leela’s also shoots above +1 (a sign of a losing position for Leela). I was asked to provide some expert commentary but wasn’t able to be very specific! I guessed that Leela had prepared some fraught lines in the case of immediate action but that Stockfish had seen that there was no rush and that a strengthening move was possible after which Leela’s line no longer worked. It’s kind of true, but it’s actually more than that: Black has no good moves available! Moreover, that d1 square for the queen comes in very handy in a couple of variations!
Leela played a waiting move 35…b5 , hoping to push the moment of truth further back in time! Let’s look first at the active, “human” attempts.
i) 35…h6
is the most critical try for Black on each move. Driving away the knight would be a huge achievement for Black.
36.Qh5 Qg7
ia) 37.Rc2 was my idea which doesn’t quite make it. 37…Kg8 The move I’d underestimated. It looks awfully risky to move the king to the g-file, but it hangs together due to a downside of 35.Rd2! 38.Nxe6 Nxe6 39.Rg2 winning the queen? Well… 39…Qxg2+ 40.Kxg2 Nxf4+ Oops!
ib) 37.Rg2 Despite about half an hour’s analysis at the board, I didn’t manage to find Stockfish’s precise tactical line.
37…Rf5 The defence I couldn’t break while analysing by myself. 38.Rc2
Now! The feint to the g-file has drawn the black rook off the back rank after which the threat of Rxc7 is considerably more destructive! (38.Nf7+ Qxf7 39.Qxh6+ Qh7) 38…Qf8 Stockfish’s best defense. It looks perpetual-scary, but an engine has no fear: only the power of its calculation! (38…Nd5 39.Rc8+ Rf8 40.Nxe6 Nxf4 41.Nxg7 Nxh5 42.Rxf8+ wins; 38…Kg8 39.Rxc7 Qxc7 40.Qe8+ Rf8 41.Qxf8+ Kxf8 42.Nxe6+ Kg8 43.Nxc7 d3 44.Nd5 d2 45.Nc3 Just in time!) 39.Qg6 Rxg5 40.fxg5 Qf4+ 41.Kg2 Qg4+ 42.Kf1 Qh3+ 43.Ke1 Qe3+ 44.Re2 Qc1+ 45.Kf2 Qf4+ 46.Kg1 Qg3+ 47.Kf1 Qh3+ 48.Ke1 Qg3+ 49.Kd1 Qxb3+ 50.Qc2 was one of Stockfish’s lines.;
ii) 35…Na6
So… if 35…h6 doesn’t work, let’s try and activate other stuff! 35…Na6 is obvious, ready to bring the knight to c5 supporting …d4–d3 and attacking b3 while still defending the e6–pawn. 36.Qh5 Rxf4 37.Qh6
was my line… and it works! The black rook can no longer get back to the 8th rank and it is surprisingly short of stable squares on the f-file. 37…Nc5 The logical follow-up allows a stunning riposte (37…Rf5 38.Rxd4; 37…Qg7 38.Qxg7+ Kxg7 39.Nxe6+ That fork again!; 37…d3 38.Nxh7; 37…Qe7 38.Kg3 No good squares for the rook! If the rook moves from f4, then d4 falls giving White an invasion channel on the d-file as well as the opportunity to switch to the g-file via g4. 38…Rf5 39.Rxd4 threatening Rd8+ 39…Kg8 40.Nxh7) 38.Rxd4
The one we all wanted to see! 38…Rxd4 (38…Qxd4 39.Qxh7#) 39.Qf8#;
iii) 35…Rf5
This was my favourite attempt after seeing that 35…h6 didn’t work. The rook keeps attacking f4, while the knight on g5 is pinned to the queen after White attacks with Qh5. However, danger now strikes from the other wing! 36.Qb7
White strikes on the queenside. I call the adeptness of modern engines to combine threats on both wings “whole board play”, another topic to which I devote a chapter in my new book “The Silicon Road to Chess Improvement” Qb7 threatens Rc2 winning the paralyzed c7–knight. 36…Rxf4 (36…d3 37.Qb8+ Ne8 (37…Kg7 38.Qxc7 is that trick again!) 38.Rxd3 exploits the weakness of the back rank) 37.Rc2 Rxh4+ 38.Kg3 Rh5 39.Rxc7 Rxg5+ 40.Kh4 and Black loses the whole house!
iv) 35…Rg8
My last attempt to find something constructive and non-destructive for Black. Black keeps the 8th rank covered, contest the g-file and eyes the knight on g5 for possible …Rxg5 desperadoes. 36.Qh5 (36.Qe4 greedily going after the d4–pawn gives Black a great chance for counterplay 36…h6 37.Rxd4 Qe7 Stockfish’s move (37…Qg7 38.Rd3 hxg5 39.hxg5 Rf8 40.Rh3+ Kg8 was my line, but Stockfish sees a draw by repetition with 41.Rh6 Rf5 42.Qd3 Qe7 43.Qh3 Qg7 44.Qd3) 38.Rd3 (38.Nf3 Qg7 is rather embarrassing for White!) 38…hxg5 39.hxg5 Rg7 with an extra piece for Black) 36…Ne8 (36…Rg7 37.Qd1 making very nice use of 35.Rd2 for a mini-Alekhine’s Gun! The queen helps to attack d4 while – in contrast to Qe4 – keeping in touch with h5 to dissuade …h6 from Black. 37…Nd5 38.Rxd4 Qe7 39.f5 wins. Black cannot hold the knight on d5.; 36…Rf8 37.Qh6 Qg7 38.Qxg7+ Kxg7 39.Rc2 when the fork on e6 will again prove terminal) 37.Rxd4 Rxg5 38.Qd1
A very neat idea that I’d missed in my calculations. The rook on d4 is protected and White will capture the rook on g5 on the next move.
Back to the game now where Leela played 35…b5 36.a5 a6
…and now…
37.Rd1!!
Another waiting move from Stockfish accompanied by a 4.71 evaluation! The tactics are slightly different with the rook on d1, but no less potent!
37…Qe7
Leela gives in and jettisons the d4-pawn. Let’s see what happened against other tries!
i) 37…h6
Again the most critical continuation 38.Qh5 Qg7 39.Rc1 immediately now works – in contrast to positions with the rook on the second rank – as after 39…Kg8 40.Nxe6 Nxe6 41.Rg1 there is no …Qxg2+ and …Nxf4+!
ii) 37…Rg8 38.Rc1 with Rc6–d6 to follow was Stockfish’s line. 38…Nd5 seemed like the best shot since White is no longer attacking the d4–pawn (38…h6 39.Qh5) 39.f5 exf5 40.Qh5 Rf8 41.Qh6 Qe7 42.Qd6 (42.Nxh7 Qxe5+; 42.Kh1 threatening Nxh7 42…Ra8 43.Rg1 with e6 to follow is the perhaps more human way to finish!) 42…Qxd6 43.exd6 Nf6 44.Ne6 Rg8 45.Nxd4 is +5 for White according to Stockfish! If it’s not worried about checks on g4, I guess I’m not either!
After 37…Qe7…
…Stockfish had little trouble converting.
38.Rxd4 Nd5 39.Kg2 b4 40.Nxe6 Qxe6 41.Rxd5
41…Qg6+ 42.Kh1 Qf5 43.Rd4 Qg6 44.h5 Qf5 45.Kh2 Qc2+ 46.Qg2 Qf5 47.Qe2 Qe6 48.Qf3 Qf5 49.Rxb4 Qc2+ 50.Kh3 Qc8+ 51.Kh4 Qc2 52.Rc4 Qh2+ 53.Kg4 Rg8+ 54.Kf5 Rf8+ 55.Ke6 Qh4 56.Rc7 Re8+ 57.Kf7 Qd8 58.Qb7 Rf8+ 59.Ke6 Qd3 60.Rd7 Qf5+ 61.Kd6 Re8 62.b4 h6 63.Kc7 Qxf4 64.Qd5 Qxb4 65.Rh7+ Kxh7 66.Qf7+ Kh8 67.Qxe8+ Kg7 68.Qg6+ Kh8 69.Qd6 Qc3+ 70.Kd7 Qh3+ 71.Ke7 Qc3 72.Qxh6+ Kg8 73.Qg5+ Kh7 74.Kf6 Qc6+ 75.e6 Qc3+ 76.Qe5 Qd2 77.Qe4+ Kh8 78.Kg6 Qg5+ 79.Kxg5 Kg7 80.e7 Kf7 81.e8R Kg7 82.Qg6# 1–0
Just to verify the power of Stockfish’s play, I checked what would have happened after 2 other tempting “human” moves instead of the amazing 35.Rd2!!!
i) 35.Qe4
This greedy attempt to win the d4-pawn immediately gives Black a welcome chance to chase away the knight from g5.
35…h6 36.Rxd4 Qg7 37.Nh3 (37.Nf3 Rg8 is really strong) 37…Rg8 38.Rd3 (38.Qf3 Nd5 was my line which looked decent for Black to me.) 38…Nd5 39.f5 and now a lovely move…! 39…Nf4
Aaah! 40.Nxf4 Qg1+ 41.Kh3 Qg4+ 42.Kh2 Qg1+ with a draw by perpetual
ii) 35.Qh5
was the obvious move for White. I wondered whether it offered enough to win, though perhaps less optimally than via Stockfish’s 35.Rd2. 35…Rxf4 36.Qh6 This move again, x-ray targeting the rook on f4. 36…Qg7 Black has this resource now as the knight on c7 defends e6 against a knight fork 37.Qxg7+ Kxg7 38.Rc1
A cruel idea, as the knight cannot move and cannot be defended! 38…Kg6
A lovely move I was very pleased to find without engine assistance! (38…Rxh4+ 39.Kg3 Rh5 40.Rxc7+ Kg6 41.Nf3 Keeps control for White) 39.Rxc7 (39.Kg3 Nd5) 39…Rxh4+ 40.Nh3 (40.Kg3 Kxg5 41.Rg7+ Kh6) 40…Re4 or Stockfish’s 40…Kf5 are both good enough for Black. This defensive technique of transposing to a pawns vs knight ending (with a rook or queen) is a typical one, and one that I highlight in my latest book “The Silicon Road to Chess Improvement” in the chapter on engine defense.
Thanks for sharing. I just ordered the book 🙂
The position looks like a zugswang endgame study with lots of pieces on the board.
Is Leela able to see the position is lost given more time or is it beyond reach?
Yeaaah! Thanks! I haven’t tested this one specifically… The TCEC box built up 6.4M nodes on this position which it would take my machine some time to emulate! I think if it was going to spot the problems, it would have done so much earlier! Best Wishes, Matthew
Amazing analysis. I’ve ordered your book and can’t wait to read it altho it’s not on Amazon until late November.
(Macbeth59)
Thanks so much Carlton! Yes indeed, looks like it’s going to take a while on Amazon. I’ve ordered one too 🙂 🙂 Best Wishes Matthew
What a beauty! Thanks for sharing!
Thanks Cesar! Glad you enjoyed it!