One of the side effects of writing a blog about chess is that I’ve been paying a lot more attention to ongoing events. I watched a lot of the first three rounds of Norway Chess, with the predictable result that I didn’t get my spring semester grading done quite as quickly as I’d hoped. Thursday I had to “lock in” as the kids are saying these days, so it seemed like a good opportunity to skip the games entirely and then play guess-the-move with one or two of them after the fact.
Here’s how it works: pick a game from a site where you can see the board and forward/back arrows while hiding the moves (I’ve been using theweekinchess.com). Pick a side to guess for (the winning side, generally) and play through until you’ve reached the end of the opening. Write down your candidate moves each time and a bit about each one, then click a move forward to see how you did. Repeat until you get to the end of the game. I don’t know how fun most people would find this sort of thing, but you learn a lot as you go and it’s surprisingly thrilling each time you get a move right.
As you can already tell from the headline, I picked the Caruana-Carlsen matchup from round four. Unlike the last time I did this exercise, I hadn’t seen the game before—all I knew was the black won in a queen ending—so I really had to channel my inner Magnus Carlsen, if such a thing is possible. I started after the pieces were developed and the kings castled, after white’s thirteenth move:
My first reaction was that I wished I’d picked a different game. I don’t particularly like these dry Italian Games where you maneuver for a long time hoping that the other guy will make some subtle mistake. Given how many options are available to black on each move I figured I’d get a lot wrong. But you know, learning experience and all that, plus I was curious about how Carlsen managed to beat Caruana from a position that at first glance looks so devoid of winning chances, imbalances, you name it.
I had a bunch of candidates for move 13: d5, a5, Ne7, Qe7, Qe8, Qd7. I like to start with the moves that change the position most dramatically, so first I analyzed 13… d5 14 exd5 exd5 15 e4 dxe4 16 Qb3+ Kh7 17 Nxe4 Nxe4 18 dxe4. It changes the position, but not in a way that seems great for black. I kind of liked 13… a5 to stop b4, but this makes it harder to take away c4 from white’s knight or queen. The rest of the moves all feel or less the same, but 13… Qd7 defends against Qb3, so I went with that.
Magnus played 13… Qe7. Pretty close, pretty close. After 14 Qb3 it seemed necessary to defend the b-pawn, either directly (Rab8, Na5, Nd8, b6) or indirectly (Qd7). 14… b6 15 Qc4 Na5 16 Qa2 with b4 to follow seemed annoying. I don’t know why I would follow up 13… Qe7 with 14… Qd7, so I ruled that out. 14… Rab8 made the most sense, potentially preparing b5-b4 at some point.
I was right: 14… Rab8. It’s always a relief to get on the board. Fabi responded with 15 Raf1 and I considered the following moves: Nd7, Kh7, b5. The trouble was that I didn’t have much of an idea about what black should be doing here, other than shuffling pieces around and marking time. Without much in the way of analysis I ended up choosing 15… Nd7.
Oops—15… Nd8 wasn’t even on my radar. I guess Magnus was aiming for c5, which seems pretty typical in these kinds of positions. Fabi played 16 Qd1 and I didn’t calculate much, just assumed that 16… c5 was coming next.
Wrong again: Magnus played 16… b5. Maybe he didn’t like 16… c5 17 b4. My lack of feel for these kinds of positions was continuing to show, but at least I got the next move right: 17 axb5 axb5 (automatic, but it counts) 18 Nh2.
I looked at moves that addressed white’s threat of Ng4 (Nb7, Nc6, Nd7, Rf7) and those that didn’t (b4, c5). Generally I was suspicious of lines like 18… b4 19 Ng4 Nh7 (although I decided that 20 Nxh6+ doesn’t work) and 18… Nd7 19 Rxf8+ Nxf8 20 Qh5 with Ng4 to follow. 18… Nb7 19 b4 kills the knight and 18… Nc6 feels awkward with the pawn still on c7. So I picked 18… Rf7 almost by process of elimination: it felt a little clever and a little awkward.
Once again, I was totally off track: Magnus played 18… Nf7, a move I had noticed but not seriously considered. After 19 Ng4 it’s time to think about the future of the Nf6. Exchange (Nxf6), retreat (Nd7), or ignore (Ng5)? 19… Ng5 20 Nxf6+ Rxf6 21 Rxf6 gxf6 22 Qh5 Qg7 probably isn’t losing, but I don’t know why you would allow this if you didn’t have to. 19… Nxg4 20 Qxg4 gives white a free hand on the kingside. 19… Nd7 asks what the white knight is doing on g4, and highlights one of the advantages of Carlsen’s Nf7, which overprotects h6.
Back on track: I correctly predicted the next moves for each player: 19… Nd7 20 d4. I considered various ways to play in or around the center: b4, c5, and exd4. I didn’t like the idea of giving away the c4-square, so b4 is out. 20… exd4 21 exd4 e5 gives away f5; you can imagine the Ng4 heading there via e3 in the future. So 20… c5 it is, a pretty normal and natural move in this kind of position.
Right again! I felt like I’d opened a portal into Magnus’ brain. After 20… c5 21 Qe2 black has a lot of moves to choose from:
Some previous candidates are back (exd4, b4) as well as some new arrivals (c4, cxd4, d5, h5). If we were playing mind-hand I would yell “Pawn!” at Magnus, but which pawn would he choose? I quickly dismissed c4 and h5 as a bit silly. I still didn’t like trading on d4 and giving the Ng4 access to e3. I was surprised to find that I was leaning towards d5, since it didn’t seem like there was an advantageous way for white to release the tension, and it set up c4 and b4.1
Magnus played 21… c4 (the move I dismissed as silly, oops, portal closed), and Fabi played 22 Nh2, a move that suggests that he was having trouble finding a plan. I considered a couple variations to force the issue on the queenside: 22… Nb6 23 b3 cxb3 24 Qxb5 Nd5 25 Qd3 Nf6 and 22… Ra8 23 b3 Ra2 24 bxc4 but didn’t feel that either had much promise. 22… Ra8 23 b3 d5 seemed reasonable as well as 22… d5 right away, but I felt like something was missing. It seemed like a good time to stop thinking about the queenside and switch focus to the position as a whole. The f-file lineup is tying down the Rf8, so let’s free our pieces with Nf6.
I was happy to see that I’d made the right choice, and after 22… Nf6 23 Ng4 Nd7 24 Nh2 a critical choice awaits: Magnus isn’t going to allow a repetition, but which of my previous candidate moves will he pick? I remained loyal to one of my previous moves, d5, but Magnus played 24… Ra8 and Fabi responded, as predicted with, 25 b3.
At this point my prediction game got a little shaky. I looked briefly at cxb3, b4, and d5 and decided that I still liked d5. The game continued instead 25… cxb3 26 Nxb3 and here I assumed that Rab8 is forced, but of course it’s not, and Magnus played the more forceful 26… Rac8 instead.
OK, time to refocus. After 27 Qd3 the c-pawn is defended, but the queen is no longer defending g4. I gave all my attention to Nf6, a useful idea from earlier in the game that could no longer be countered by Ng4. Magnus agreed, and after 27… Nf6 28 Nd2 I looked at a couple ways to load up on e4 (Qb7, Ng5) and somewhat lazily decided that Ng5 was better because I didn’t want to figure out if 28… Qb7 29 Rxf6 was a problem.2
Magnus played another move I hadn’t looked at, 28… Qc7, the second time I’d missed a move targeting the weak c-pawn. The game continued 29 Qxb5 Qxc3 (an easy one to predict) 30 Nhf3.
I was starting to feel more confident, as the game seems to be entering a tactical phase, and looked at Qxe3, Rb8, and exd4. I rejected the last of this trio pretty much right away, as there’s no reason to relieve white of his e3 weakness. I spend more time on 30… Rb8 31 Qe2 Rb2 32 Nb1 Rxe2 33 Nxc3 Rxe3 before deciding that actually white can play 31 Qa4 to defend along the fourth rank. It seemed like 30… Qxe3 31 dxe5 Nxe4 32 Nxe4 Qxe4 33 exd6 Nxd6 was a free pawn, so I went with it.3
30… Qxe3 31 dxe5 arrived, as predicted, but when I tried to follow up with 31… Nxe4 I discovered that Magnus had played 31… dxe5 instead. What the heck? There’s another big choice after 32 Nxe5: my candidates were Nxe4, Nxe5, Nd6, and Rc2. 32… Nxe4 33 Nxe4 Qxe4 34 Nxf7 drops a piece, so that’s out. 32… Nxe5 33 Qxe5 doesn’t seem like much at all; same with 32… Nd6 33 Qe2. I decided that I didn’t understand what Magnus was thinking but that 32… Rc2 33 Nxf7 Rxd2 or 32… Rxc2 33 Nef3 Nxe4 was the most promising.
Nope: 32… Ng5 33 Qd3:
Now my candidate moves were Qxd3, Rc3, and Qc5. The lines aren’t too difficult to calculate: 33… Qxd3 34 Nxd3 Nfxe4 35 Nxe4 Nxe4 36 Rxf8+ Rxf8 37 Re1 Ng5 38 h4 wins back the pawn on e6. 33… Rc3 34 Qxe3 Rxe3 35 Nec4 also seems quite equal. So 33… Qc5 it is, with some hope of exploiting the pin on the diagonal. It didn’t hurt that I knew that we someday had to arrive at a queen ending, which is somewhat harder to do if you allow a queen trade.
I’d correctly predicted 33… Qc5 Nef3 but picked the wrong rook: Magnus played 34… Rcd8 instead of my 34… Rfd8, which I’d picked to prevent 35 Qc44. After 35 Qb1 the question is how to increase the pressure.
Neither Nxf3+ or Rxd2 seem to do much of anything. 35… Nh5 pushed me a little past what I was comfortable calculating: 36 Nxg5 Rxd2 (36… Rxf2 37 Rxf2 Rxd2 runs into 38 Qb8+ and mate) 37 Nxe6 Qxf2+ 38 Rxf2 Rfxf2 39 Qb8+ Kf7 40 Nd8+ Ke7 (40… Kg6 41 Qb6+ and Qxf2) 41 Nc6+ Kf6 seems like the sort of thing that could be winning for black.5
Magnus made the same decision (but for the same reason?) and the game continued 35… Nh5 36 Qc1. I hadn’t even looked at white trying to bail out like this, but it’s annoying, and I suddenly felt lost. How does black keep up the pressure? Qxc1 doesn’t go anywhere. Wherever black moves on the a7-g1 diagonal can be countered by white opposing with his queen on the back rank. I guessed 36… Qe3, but Magnus played 36… Qa7, which was of course met by 37 Qa1. I then decided on 37… Qe7 38 Nxg5 Rxf2 39 Rxf2 Qxg5 40 Nf3 Qe3, restoring the long diagonal pin, but the game continued 37… Qb6 38 Qb1. Counting the moves would have been helpful here, as I didn’t realized that Magnus was probably marking time until move 40. There followed 38… Qc5 (I guessed 38… Qa7) 39 Qc1 Qe7 (I guessed 39… Qa7 again) with a better version of the plan I had seen on move 37. At this point I knew what was happening and my success rate improved: 40 Nxg5 Rxf2 41 Rxf2 Qxg5 42 Qc3:
There’s one last big decision to make. Black has pressure, but it’s not exactly clear how to cash out. I figured that the queen and rook were well placed so I should try to improve the knight. I checked 42… Nf4 first: 43 Kh2 Kh7 (43… Rd3 44 Qc8+ seems to hold) 44 Qe3 e5 45 Nf3 Qe7 46 Nxe5 and it’s white who wins the e-pawn, not black. 42… Ng3 doesn’t make an immediate threat, so I initially found it less compelling, but the threat of Rd1+ freezes the Nd2. And then I suddenly saw the tactical point: 43 Kh2 (otherwise h5-h4) 43… Rxd2! 44 Rxd2 Qxd2! 45 Qxd2 Nf1+ wins a piece.
The game unfolded as I’d predicted: 42… Ng3 43 Kh2 Rxd2 44 Qxg3 (Fabi makes the best of a bad situation) 44… Rxf2 45 Qxf2 Qe5+ 46 Kg1 Qxe4.
At last black wins a pawn. I decided to stop here, rather than double the length of this post by slogging through the queen and pawn ending. Here’s a link to the whole game. It’s worth noting that despite Carlsen’s masterful play the position is still objectively equal and would remain so until Fabi erred on move 66 and resigned on the following move. There’s a reason these two guys have been the two best players in the world for much of my adult life.
Some final thoughts:
I scored a perfect 50%: 17 moves correct, 17 incorrect. Chess moves are like DNA in that a 50% match doesn’t give you anything close to the original organism, but I felt pretty good about how I did.
I had two clear blind spots: the value of the Nd8-f7 maneuver and moves targeting the weak c-pawn. Just as in my own games, sometimes I felt lost at sea or lacked energy to figure out a position. At other times my interest spiked as the position seemed more interesting.
Almost all the variations I calculated were accurate, with the exceptions captured below in the footnotes. I would have found a way to lose to Fabi sooner or later if I had been playing the black pieces, but I’d like to think I would have made a game of it.
Normally I solve tactical problems where there is a single winning move. This exercise forced me to accept greater levels of ambiguity and frustration since there were a lot of reasonable options and none of them won by force.
I gained a newfound appreciation for how strong Magnus Carlsen is. I can’t think of many people in the history of chess who could have generated winning chances from that kind of opening position against an opponent as strong as Fabiano Caruana.
The computer assures me that I blundered a pawn here: 21… d5? 22 dxe5 Ndxe5 23 Nxe5 Nxe5 24 Rxf8+ Rxf8 25 Rxf8+ Qxf8 26 Qxb5.
Apparently it’s not good here, but 23 Rxf6 was possible instead of offering the repetition with 23 Ng4.
The problem is that even if black emerges up a pawn the game is still likely to be drawn because of how simplified it is.
You might have already noticed that 34… Rcd8 35 Qc4 loses on the spot to 35… Nxf3+!
It’s not winning for black; white has 42 Qe5+ and things fall apart. I’d like to think that once I had the position after 35… Nh5 36 Nxg5 Rxd2 37 Nxe6 on the board in front of me I would have found the correct queen sacrifice, 37… Rfxf2!
Thanks for the interesting post! Out of curiosity, did you check with an engine how much worse your 17 "misses" were than the moves that Magnus actually played? Whenever I do this exercise, I'm always surprised by how few of my guesses are significantly worse than the GM move. Yet apparently those few moves per game are worth like 700 rating points!