Once upon a time there was a man named Emanuel Lasker who happened to be quite remarkable in a number of fields, one of them being chess, where he reigned as world champion for an absurdly long time. Fast forward one hundred years, and Lasker is largely overlooked in favor of his successors (Capablanca, Alekhine) who possessed styles of play far easier to appreciate. He’s also the only world champion who I remember best for a game he lost:
In this mundane and equal position Lasker, playing black, decided to fight for a win and walked his king to c7, starting with 27… Ke8!? He baited his opponent, Karl Schlechter, into a queenside attack, gained a significant positional advantage, won a couple pawns, then lost control of the position and resigned after some resourceful play by white. Check it out; it’s a cool game that just happens to also have been the fifth game of the 1910 world championship match.
I’m telling you all this because Schelchter-Lasker and Caruana-Nepomniachtchi have one important thing in common: each game was decided during the fourth phase. Normally we think of phases of a chess game in the same way the my elementary school taught me about states of matter: three phases, opening, middlegame, and ending (or solid, liquid, and gas), but it’s a lie, there are actually four. In physics we’re talking about plasma. In chess the fourth phase is defined by each player having a queen and one other piece, often a rook. In some ways it’s like a middlegame (mating attacks are quite common, king safety is important), in other ways it’s like an ending (passed pawns can be quite valuable, the board geometry and coordination of pieces is more endgame-like).
Humans struggle to play the fourth phase accurately, partly due to its rarity and partly due to the fact that it requires a tremendous amount of precise calculation. Computers can find hidden geometrical patterns that lead to mate and perpetual check far more quickly and easily than even the strongest human players and so the gap between what the computer suggests and the play of human beings can be shockingly wide. I have an as yet unconfirmed theory that games that remain in the fourth phase for more than a few moves tend to be epic in scope, with crazy ideas and tactical reversals the norm rather than the exception.
I worked on these annotations first without the help of the computer, but I’d already had my ideas influenced by watching the broadcast: I knew which positions were the important turning points and the some key ideas for both sides. Even so, I struggled to get my bearings repeatedly. Let’s start with the position after Nepo’s 40th move:
A surface-level appraisal: White is up an exchange. Black’s pawn on h2 feels very dangerous. Both kings are exposed. White’s in check, so black is going to have some kind of initiative to start; whether it’s lasting or not is another story. White’s pieces (unprotected rook h1, pawn f3) are a little loose. Let’s call it “unclear.”
My more detailed evaluation, after moving the pieces around and blundering things for a while, more or less in the order that I realized them to be true:
Without some crazy resource, a queen trade will result in a white victory.
Black’s exposed king gives the white queen tactical access to certain key squares that can’t be reached in a single move (for example Qd8+ K-any Qd1 allows the queen to move from g5 to d1 with tempo).
Despite the weakness of white’s king, only two perpetual check patterns came up in my analysis: one with blacks’ queen on c2 and the knight hopping between b3 and c1, and the other involving the sacrifice of the knight to draw the king out into the open.
The evaluation of many positions hinges on the black knight. It’s often best placed on c5, where it defends the black king and can advance to b3 or d3 to attack. In some cases white wants to lure the knight forward and out of position.
Somewhat surprisingly to me, white wants to trade the f-pawn for the h-pawn to stabilize the position and gain more space for the rook. Trading these pawns, using mating threats to force off the queens, and slowly advancing with king and rook is the winning strategy if black puts up maximum resistance.
Conversely, black wants to keep the kingside pawns in order to split white’s attention between the h-pawn and mating/perpetual threats on the queenside.
I doubt that the players were thinking about the position in this way at all; they were likely spending their time more concretely, calculating as much as they could. Caruana thought for a few minutes and played 41 Ka1, a move that sent the eval bar tumbling downward. I don’t think this is a typical case of a player relaxing and losing focus on the first move after reaching time control; the choice between the move Fabi played and the winning 41 Ka2! Qf7+ 42 Ka1 Nb3+ 43 Kb1 Qh7+ 44 Ka2 (safe at last!) 44… Qf7 45 Qg3+ Ka7 46 Qxh2 Nd2+ 47 Ka1 Nb3+ 48 Kb1 Qxf3 is an extremely difficult one. You might see that 41 Ka2 wins and still play 41 Ka1 because you’ll only realize that it doesn’t work if you find black’s brilliant idea on move 44, a resource difficult enough that Nepo didn’t find it over the board.
The game continued 41… Qc2 42 Qg8+ Ka7 43 Ka2 (black was trying to execute the first of the two perpetual ideas in #3 above: 43 Qh8? Nb3+ 44 Ka2 Nc1+ 45 Rxc1 Qxc1 46 Qxh2 Qc4+ 47 b3 Qf7! threatens both … Qxf3 and … a4) 43… a4. The battle over the a2-g8 diagonal has given white no time to round up the h2-pawn; Robert Hess on the broadcast quipped that white really wishes he could play his queen to h9 in order to take care of both at once. Fabi played a great move, 44 f4, really putting Nepo to the test:
I already knew from the broadcast that there was only one move to draw in this position: 44… Ka6! I can explain the logic behind it: black’s queen and knight are ideally placed on c2 and c5. The only weakness is the black king, since white can play Qg7+ followed by Qc3 to defend against the Nb3-c1 idea, so by process of elimination we should move the king. I can demonstrate how it works with the following variation: 44… Ka6! 45 Qa8+ Kb5 46 Qd5 Ka6! (can’t allow the pin on the knight) 47 f5 Nd3! (the computer showed me that 47… Nb3? fails to cover e1 and thus loses to 48 Qa8+ Kb5 49 Qe8+! Ka5 50 Qe1+! Ka6 51 f6) 48 Qa8+ Kb5 49 Qe8+ Ka5 50 Qh8 Nc1+ 51 Rxc1 Qxc1 52 Qxh2 Qc4+ with a draw. But just because I can figure all this out once I know about the idea doesn’t mean there’s much of a chance that I would play this over the board. Nepo didn't. He instead picked the wrong moment to bring in the knight, 44… Nb3 (see note #4), and now white is back in the driver’s seat.
Caruana played the next dozen moves very well, and I’ll provide only light commentary: 45 Qg7+ Ka6 46 Qc3 (putting an end black’s Nc1-b3 idea) 46… Qg2 (switching back to h-pawn, but white untangles nicely) 47 Qc4+ Kb7 48 Re1 (generating some mating threats to stop 48… h1=Q? 49 Re7+ Kb8 50 Qc7+ Ka8 51 Qa7#) 48… Nc5 49 Qf1 Qd5+ 50 Kb1 (not falling for 50 Ka1? Nb3+, winning the white queen next move) Qf5+ 51 Ka1 Qc2 52 f5:
It’s worth comparing this diagram to the one above it. Black’s only “accomplishment” is that his king is on b7 rather than a7, where it’s actually more exposed. His pawns, knight, and queen are exactly where we found them on eight moves earlier. White, on the other hand, has reactivated his rook, advanced his passed pawn, and stationed his queen where it supports the passer, overprotects the h1 queening square, defends against Nb3-c1, and covers c4. I honestly expected black to resign at any moment, but he played on and got to unleash one more trick: 52… Nd3 53 Rb1 Nc5 54 f6 Nb3+ 55 Ka2 Nd2 56 Qh1+ Ka7 57 Rc1 Qb3+ 58 Ka1 Qe6:
I don’t know if you have a favorite move in this wild and crazy game, but I definitely do, and that move is 58… Qe6! It takes care of some important stuff, like threatening … Nb3+ and covering the advance of the f-pawn, but it’s what it doesn’t do that makes it really special. Most people, including myself, would decide, quite prudently, to stop Rc7+ by playing 58… Qf7. Nepo must have realized that the rook check is actually a mistake and an extremely enticing one at that. This is high level swindling in action, and it tricked Caruana from making further progress with the calm 59 Rd1! Nb3+ 60 Kb1 Qxf6 61 Qxh2, a line we would have surely seen after 58… Qf7. Instead he played for a quicker kill: 59 Rc7+ Ka6 60 f7 (there’s no longer time for 60 Qxh2? Qe1+ 61 Ka2 Qb1#) 60… Nb3+ 61 Kb1:
We’re truly on the knife’s edge now. Only 61… Nd2+ forces a draw, since after 62 Kc2 Qf5+! 63 Kxd2 Qf4+ white cannot run back to the c-file without losing to rook with check, and the queen can never block lest the h-pawn promotes. From an aesthetic point of view, this would have been a lovely finish to the game, but from a sporting point of view it’s utterly unthinkable: the players cannot afford a draw. I wonder if Nepo saw it and rejected it, or simply didn’t have lines that led to perpetuals on his radar. After all, if he plays 61… Nd2+ a draw is the only conceivable result. After the move in the game, 61… Qf5+, a mistake in any other context, Nepo is technically losing but there remains a chance that Caruana could blunder.
Again Fabi made progress toward the elusive win: 62 Ka2 Nc5 63 Qa8+ (not 63 Qxh2? Qd5+ 64 Kb1 Qd1+ with a simple draw) 63… Kb5 64 Qc6+ (64 f8=Q? Qe6+ 65 Kb1 Qe1+ is another perpetual) Ka6 65 Qa8+ Kb5:
One last piece of geometry is required: the white queen needs to move from a8 to e2 with tempo. It’s the fourth phase, so just about anything is possible: 66 Qe8+! Ka5 67 Qe1+ Ka6 68 Qe2+ Qd3 (68… Nd3 69 Qxh2 and there’s no perpetual now that the knight is in the way on d3) 69 Qxd3+ Nxd3 70 f8=Q h1=Q 71 Qc8+! Black loses his knight or his king. In any normal situation you’d expect Caruana to find this 99 times out of a hundred, maybe 999 times out of a thousand. But with seconds left, he made a final mistake: 66 Qc6+ Ka6 and now 67 Qa8+ would be an immediate draw by repetition. At last this dramatic phase of the game came to an end, but not without one more clever tactic: 67 Re7 Qf1 68 Qa8+ Kb5 69 Qe8+ (tragically too late) 69… Ka6 70 Qa8+ Kb5 71 Qe8+ Ka6 72 Re4. It’s worth noting that Fabi’s not trying to torture Nepo by moving back and forth but is simply trying to gain some time back on the clock before making a decision about how to get the maximum from the position. The problem left is that there’s nothing left to extract, so he sets one final trap:
Nepo did not fall for the lovely 72… h1=Q? 73 Qa8+ Kb5 74 Rb4#, instead simplifying to a dead drawn queen ending: 72… Nxe4 73 Qxa4+ Kb7 74 Qxe4+ Ka7 75 Qa4+ Kb7 76 Qd7+ Ka6 77 Qc8+ Ka6 78 f8=Q Qxf8 79 Qxf8 h1=Q:
Whew. That was some wild chess. I hope you enjoyed these annotations of the fourth phase; I certainly learned a lot as I worked on them. Despite there being 30 more moves, there’s not much chess content left. I’ll wrap up with one more post in the next few days about the queen ending and some final thoughts on the entire game.