Caruana-Nepomniachtchi, Part 2
"Well, in the end, he got himself a pool - only the price turned out to be a little high."
After 25 moves, Ian Nepomniachtchi’s position against Fabiano Caruana looked like this:
Wait wait wait — somehow my classic movies file and my Caruana-Nepo file got mixed up. Let’s try it again: After 25 moves, Ian Nepomniachtchi’s position against Fabiano Caruana looked like this. He actually might have envied the guy floating facedown in the pool, given the state of things on the board:
This is a beatdown, a position that black should have no hope of coming back from. Nepo had to win, and he got himself a game, to paraphrase Sunset Boulevard — only the price turned out to be a little high. In the finest film noir tradition, let’s flash back and find out how we got here. This is where the previous post ended, after Caruana’s 14 f4:
To go from this position to the one above it in just 11 moves against one of the very strongest players in the world is a heck of a feat, and one that should have put Fabi in a playoff with Gukesh for a shot at the World Championship match. I went through three phases of thinking as I investigated the next 20+ moves of this middlegame. My first, superficial assessment, was that black should be pretty happy that he got an unclear fighting position with black. As I looked a little more deeply, I decided that black’s position was actually very difficult due to his lack of king safety and white’s potential pawn breaks in the center. I struggled to find a good plan for black and the difficulties that Nepo faced in the game felt more or less inevitable. After some more investigation and double-checking with the computer, I settled on something in-between. Black’s position is more difficult to play, but this doesn’t make it inherently unsound or mean that Nepo had to play perfectly in order to survive.
Before we get to the game, an aside about the role of computers in chess analysis. Computers are incredible at chess. They see all kinds of ideas that I would never even begin to imagine. But too much chess analysis these days is produced by a strong human chess player explaining the very nearly perfect ideas produced by a machine, which has very little to do with what the players are actually experiencing as they play. In the case of this game relying heavily on computer analysis is even worse, as both players had to play for a win. If a move makes the position worse by half a pawn it’s a mistake by the computer’s standards. But if that same move resulted in a more complicated position that increased the likelihood for a win by, say, 3%, it’s probably a pretty decent idea. So as we walk through the next phase of the game, we have to remember what’s at stake and that this game is taking place at the end of a brutally long tournament, under intense pressure. My method was to do as much analysis as I could myself in a few hours, moving pieces around on the board, and then check back and blunder-check with the machine, and I’ve tried to note when I was using the computer as guidance in the analysis that follows. Did I make a lot of mistakes as I analyzed this way? Definitely. And it all goes to show the high level of difficulty facing the players themselves.
Back to the game, at last, which continued 14… Bg4 15 Nf3 Qe7 16 e4 Bxc3 17 Qxc3 Bxf3 18 gxf3 fxe4 19 Bxe4:
Do I like this? No, I do not. I liked black’s bishop pair; now they’ve been traded for a pair of knights and white’s bishop looks like a monster. The black king was always going to be an issue, and it’s even less clear than before where it’s going to find safety. We could look at a variety of places for improvement, but I want to go back, one more time, to black’s fourteenth move and try something a little odd: 14… Nf6!?
I’ll admit that this looks dumb at first glance. White attacked our g-pawn, so we blocked its one defender. White threatened to trap our bishop with f4-f5, so instead of getting it past the f-pawn we decided to willingly drop back to d7. I spent quite a while trying to refute my own idea, first with the rather unserious 15 f5?! Bd7 16 Nf3 Qe7 17 Kb1 000 (black’s position feels much more coherent) and then with the considerably sharper 15 fxg5?! Ng4 16 g6 (not good, but we’re in refutation or bust mode) 16… Nxe3 17 gxf7+ Kf8 (taking back with either way drops the Ne3) 18 Qe2 Nxd1 19 Nxd1 Bxf7. Here’s a picture of the chaos:
The caveman in me longs to bring white’s knights to e5 and f5 and finish off the black king; the pragmatist in me is a bit concerned that black’s heavy pieces can parry the attack; the truth-seeker in me wants to know what the computer thinks, which is something along the lines of “don’t humans know about castling by hand — play 17… Kd7! instead of 17… Kf8 with a serious advantage.”
This is a long way of saying that there are a lot of options for both sides. One of the things that I like about 14… Nf6 is that it potentially lures white into a risky and potentially ineffective attack, although it’s hard to imagine Fabi going on tilt like this in such a critical game (if I were playing white, on the other hand). Nepo might have been more concerned about a positional reply like 15 Nf3, although here too there are plenty of resources and ideas.
Let’s return to move 19, as shown in the diagram below:
Nepo decided that he should grab a pawn in exchange for the suffering he’s about to endure: 19… gxf4 20 g4 000 (20… h4 21 d5 is tricky, and I almost understood it: 21… Qc5! [I tried 21… Qe5? here which my blunder-checker says loses to 22 Qa3!, not 22 dxc6 transposing] 22 dxc6 Qxc3+ 23 bxc3 bxc6 24 Bxc6 000! is the only move to hold on) 21 d5:
Nepo played 21… h4 in this position, a move that the computer thinks he should have played on move 20 instead. I say this to illustrate one of the reasons why I think black’s position is difficult: he has to decide over and over again whether or not resolve the tension in the position as the pressure increases, as demonstrated first by his choice to exchange both bishops for white’s knights, and then again with the g4 and d5 pawn breaks. I do not really want to play h5-h4 here; my amygdala is screaming at me to do something about the queenside. Here goes: 21… Nc5 22 Bf5+ Kb8 23 Rhe1 Qd6. Not so terrible, but the computer and Nepo are both better than me and I suspect that both saw 21… Nc5 22 d6! Nxe4 23 Qxh8! winning the exchange. Let’s try again with another human variation: 21… c5 22 d6 Qf6 23 Qb3 Nb6. Again, black is holding on, let’s check with the compu … what’s that? 23 Qa5! is curtains. Ugh, score two for the machine, which I guess means I should report that it feels that 21… Kb8! is more or less forced, the idea being to sidestep a future Bf5+.
After …h4 everything more or less forced for a bit now: 22 dxc6 Nc5 23 Bf5+ Kb8 24 Kb1 (it would be quite bad to allow black Qe3+ to trade the queens). I could give some variations here that are just about as a bad as black’s next move, 24… b6, and then show that computer line that maintains the position at roughly -2 instead of -3, but I don’t think it’s particularly necessary, and 24… b6 allows the very fun and aesthetic blow 25 Rd7, getting us back to the opening facedown in the pool position. I’ve been purposefully not putting evaluative marks on individual moves because black’s position didn’t go downhill because of one bad move; it went downhill because a negative trend started somewhere around move 14 or 15 and Nepo couldn’t find a way to reverse it. Now we’re talking less about how he can keep some winning chances and more about how whether he can avoid resigning before the time control at move 40:
Another forced sequence, followed by the technique of a boa constrictor: 25… Rxd7 (25… Nxd7 is much worse; it fun to work out the lines yourself so I won’t spoil them here) 26 cxd7 Rd8 27 Qd4 Nxd7 28 Rd1 Qc5 (I asked myself if black could try to unpin his pieces by playing 28… Kc7 29 Qxf4+ Kb7 30 Qd2 Kc7, but 30 Be4+! Kc8 31 Rc1+ Nc5 32 b4 is the end of the knight) 29 Qxf4+ Qc7 30 Qd2 h3 31 Be4 (not falling for 31 Bxd7? h2! regaining the piece) 31… a5 (the computer wants to give white an opportunity to fall into the previous trick, which is as good a try as any: 31… Qe5 32 Bf5 h2 33 Bxd7? Qc7!) 32 Qd5 Ka7 33 Qxf7 h2:
White has played this phase of the game incredibly well and we’re almost at the point where the win is automatic and he can sit back and relax and play more quickly and await the handshake. But we’re not quite at that point, maybe a move or two away from it, and Caruana’s excellent play has taken some time, all but seven minutes and forty seconds of his remaining time, to be exact. There’s no increment.
Fabi played 34 Qh7 which feels like a good way to avoid tricks, as it overprotects the Be4, puts another set of eyes on the h-pawn, and asks black what he’s going to do next. However, armed with the prescience of the annotator, I thought that this might be the position to go for the jugular: 34 f4! (loosening, but letting the bishop take care of h1 frees the queen and rook to make mischief) Kb8 35 Qd5! Ka7 36 g5 is curtains. I should mention that 34 f4! Ka6!? 35 g5? Nc5! 36 Qxc7 Rxd1+ 37 Kc2 Nxe4 is only a draw, but white has plenty of better 35th moves to choose from.
The game continued 34… Kb8 35 a3, with Caruana continuing to prioritize safety over putting Nepo away. It’s hard to argue with 35 g5! Nc5 36 Rxd8+! (the only winning move, but not all that hard to spot, in my opinion) 36… Qxd8 37 Qxh2+ Ka7 38 Qg1 Nxe4 39 fxe4 Qd3+ 40 Ka1 Qxe4 41 g6, but Fabi probably felt that he could keep the position as it is, get to time control, and then decide how to win the game. 35… Qe5 36 Qh6 Qc7 gets us to our next diagram, not that different from the previous diagram:
I wonder if Caruana was planning 37 Rc1 Nc5 38 b4 here. It’s hard to open up the king with so little time to consider all the possibilities. I thought I found a refutation: 38… axb4 39 axb4 h1=Q? 40 Qxh1 Qf7, getting off the c-file and threatening a perpetual on b3 and a3. The computer, bless its pedantic little heart, tells me that this loses to 41 Qh2+!, getting the queen back along the second rank with tempo, but that 39… Rd4! seems to cause some real trouble. Fabi chose more wisely for the moment: 37 g5 Rg8 38 Rh1 Nc5 and now, after all that work, after so many good moves in a row, on the verge of making it to time control, Fabi finally made a mistake that cost him everything: 39 Bh7?
He must have thought that this was the finishing touch: the black rook moves, white gobbles up the h-pawn and black resigns soon after. But by some crazy miracle, a miracle that would be repeated over and over in the next, truly exhausting phase of the game, black’s pieces turned out to be on exactly the right squares: 39… Rxg5! 40 Qxg5 Qxh7+. I’ll wrap it here for now; the next post will cover the epic battle of Q+R versus Q+N. In the meantime, please add your ideas to the comments: there’s a lot here that I didn’t have time or space to include, and probably quite a bit that I missed.
Great analysis!
When watching the third phase of the game, I kept thinking "If only Caruana had managed to eliminate that h-pawn. Surely he could have done something to eliminate that h-pawn?"
Looking back now, I guess the clearest opportunity was 35. Rd2 (instead of a3). From there Caruana shouldn't have had much trouble converting, right?
I was also wondering about 31. g5, but I guess it's not easy to get the pawn after Rh8. A cool line from the engine is Qd4 Rc8 Qd5 h2 Be4 Qc1+ Rxc1 Rxc1+ Kxc1 h1(Q)+. Seems like Caruana was right to play 31. Be4 instead of going for that.
It was interesting that in the post-game interview Caruana said that he saw 39... Rxg5 but thought it was good for him. I guess he just didn't realize how dangerous the Q+R vs. Q+N would be?