[Event "Wch 2018"]
[Site "London"]
[Date "2018.11.09"]
[Round "1"]
[White "Caruana, Fabiano"]
[Black "Carlsen, Magnus"]
[Result "1/2-1/2"]
[ECO "B31"]
[Annotator "AlphaZero - 1 minute / move"]
[PlyCount "229"]
[SourceVersionDate "2018.11.09"]
1. e4 c5 2. Nf3 Nc6 3. Bb5 g6 4. Bxc6 dxc6 5. d3 Bg7 6. h3 Nf6 7. Nc3 Nd7 8.
Be3 e5 9. O-O b6 10. Nh2 Nf8 11. f4 exf4 12. Rxf4 Be6 13. Rf2 h6 {[#] #1 The
first really critical moment in the game. Caruana has implemented the f4-break
in a direct manner. This wasn't AlphaZero's first choice - for example,
AlphaZero liked 10.a3 supporting a queenside break with b2-b4 - but AlphaZero
was still broadly upbeat about White's chances (more so than other engines).
However, White now has to make some crucial decisions about the disposition of
his forces leading into the middlegame. This position is strategically
complex and extremely interesting. Black has the 2 bishops, but a damaged
queenside pawn structure with doubled c-pawns. However. On the plus side,
those c-pawns control the central squares d5 and d4. Black has lost his
central presence after f4, …exf4 which gives White the idea of establishing
a double pawn centre in the future, and the removal of the e5-pawn exposes the
possible vulnerability of the dark squares in Black's position: d6 (no longer
protected by a pawn on c7), e5, f6, g7,h6. Black’s development is also
temporarily awkward due to the “Chess960” knight on f8 which interrupts
the communication between Black’s kingside and queenside. From move 6
onwards, AlphaZero was heavily advocating the move a4 (it became a running
theme of our analysis during the game). Clearly AlphaZero felt that the
inclusion of the moves a4 ...a5 (the natural Black reaction, blocking the
further advance of the a-pawn) benefited White in most positions. So it was no
surprise that AlphaZero also recommended 14.a4 in this position. However, we
only understood the subtlety of AlphaZero's 14.a4 when we saw its
recommendations against Caruana's choice in the game continuation - which were
identical to Carlsen's choices. AlphaZero was very wary of Black castling
queenside in this position - which would free Black to target the white king
with pawns and pieces - and wanted to force Black's king to the kingside after
which any advance of Black's kingside pawns also involves danger for Black's
own king.} 14. Qd2 (14. a4 a5 {is a key response. ...a5 makes Black's future
king's position on the queenside more "airy" but it is safer than allowing
White to set up a pressure point against Black's queenside pawn structure with
a5. It also asks White to demonstrate a follow-up plan now the obvious
continuation a4-a5 has been blocked. Moreover, on the plus side, if queenside
castling is unattractive Black also gains the idea ...Ra7-d7 to prevent d3-d4.
Although ...Rd7 takes away the d7-square from Black's knight on f8, this
knight may be developed via ...g5 and ...Ng6 or even via ...Nh7(-f6).} (14...
g5 {as in the game} 15. a5 {and if} Ng6 16. Qh5 {was AlphaZero's reason for
delaying the placement of its queen in comparison with the game. With its last
three moves, AlphaZero has made the black king uncomfortable on both wings: 16.
..Qc7 (to castle queenside) would allow 17.axb6 while 16...0-0 17.Nf3, eyeing
potential sacrifices on g5 while preparing to build up further with Raf1 and
Ne2-g3 is troubling for Black.} O-O {Not the best move, but just to illustrate
White's play.} 17. Nf3 {Threatening Bxg5} Re8 18. Raf1 Qc7 19. Ne2 {when
AlphaZero gives itself a massive 85% expected score!}) (14... Nd7 {AlphaZero's
main line.} 15. a5 O-O 16. Qd2 g5 17. Raf1 {Switching back to Caruana's plan
now that Black's king is committed to the kingside.} Qe7 18. Nf3 {and now
AlphaZero looked at two lines: 18...Rae8 and 18...f5. A sample line goes:} Rae8
19. Ne2 Qd6 20. b3 f6 (20... f5 21. Nxg5 hxg5 22. exf5 Bf7 (22... Bd5 23. c4)
23. Bxg5 {is worth considering: Black's minor pieces are limited in scope: for
example, the light-squared bishop bites against White's queenside pawn
structure which AlphaZero has placed on light squares. If White can exchange
off the dark-squared bishops then Black will face big problems defending the
king, either against a direct assault by the pieces or against a pawn storm.})
21. Kh1 {with a 60% expected score.}) 15. Nf3 {was AlphaZero's #1 plan, to
break in the centre with d3-d4 before Black completes its development. It is
risky from the pawn structure point of view: the black doubled c-pawn is
liquidated and White accepts an isolated pawn on e4. However, looking at the
disposition of Black's pieces, it makes perfect sense: Black has placed the
bishop rather than the knight on e6 and has foregone the opportunity to
prevent d4 permanently. Moreover, at the moment, the in-transit position of
the knight on f8, blocking in the rook on h8, cuts the Black position in two
by interrupting the coordination of the h8-rook with the a8-rook. It is worth
trying to exploit this as there may be some unexpected tactical opportunities
based on Black's weak back rank.} Qc7 16. d4 O-O-O (16... Rd8 {leaves Black's
king uncomfortable in the centre due to the "Chess960 knight" on f8 preventing
castling.} 17. Rd2 cxd4 18. Bxd4 Bxd4+ {and now for example} 19. Nxd4 Bc8 20.
e5 Ne6 21. Ne4 {with a 66% expected score for White.}) 17. Qd3 {I like this
move of AlphaZero's very much: the queen eyes the a6 square, again emphasising
the weakness of the Black queenside after a4, ...a5.} (17. Rd2 cxd4 18. Bxd4
Bxd4+ 19. Rxd4 Nd7 {is much more pleasant for Black than after 16...Rd8.
Black's king is safer on c8 than on e8 despite the draughty a5-b6-c6 pawn
structure.}) 17... cxd4 {This is the main line, but AlphaZero considers a few
other moves.} 18. Bxd4 {AlphaZero's preference was for this move, which I
think is related to its love of creating and exploiting weak square complexes
within the opponent's position. The plan of exchanging dark-squared bishops
and then the surprising idea of following up with a quick e4-e5 fits this
imperative very well. It was less keen on the idea that attracted me the most
- 18.Nxd4 - but did its best to keep me happy!} (18. Nxd4 {The pin on the
d4-knight is quite scary, but by keeping the dark-squared bishop on the board,
pointing at the b6-pawn, White emphasises the weakening of Black's queenside
resulting from White throwing in a4 ...a5. Moreover, Black's pawn on a5 is
now within reach of the b2-b4 break from White which opens up some tempting
sacrificial possibilities. The following analysis is not definitive, but gives
an idea of the danger that Black faces in this position as well as his active
resources.} Kb7 {I felt that this was probably Black's best move, despite the
tactical danger of 18...Qb7. One regrouping that appealed to me was ...Bc8
followed by ...Ne6, though it seems difficult to realise due to the weakness
of the f7-pawn.} (18... Qb7 {is a scary move, threatening to win a piece with .
..c5 or ...Bxd4 followed by ...c5 by sidestepping the attack on the queen
after Nxe6. After ...c5, I would love to get something on to the c-file, even
at the cost of a piece but all my attempts based around Ra3-c3 came to naught.
Some AlphaZero help however revealed the following attacking lines:} 19. Rd2 {
was AlphaZero's main line and if Black takes White head on and grabs material,
then White has a great time attacking!} (19. Nce2 {was another move AlphaZero
looked at - talk about ignoring Black's threat!} c5 20. c3 cxd4 21. cxd4 {
White's central control, the vulnerability of Black's queenside (watch out for
b4!) and Black's interrupted communication lines due to the knight on f8 give
White good compensation for the piece. AlphaZero's initial main lines from
move 19 continued:}) 19... c5 20. Ncb5 cxd4 21. c3 Bd5 {was AlphaZero's
defence.} (21... Kb8 22. cxd4 {was AlphaZero's choice with a 75% expected
score.} Bc8 23. b4 {with a huge attack}) 22. cxd4 Bxe4 23. Rc1+ Kb8 24. Bf4+
Ka8 25. Nc7+ Ka7 26. Qe3 {with an 85% expected score.}) 19. Rb1 {intending b4.}
(19. b4 {appealed to me of course, but I couldn't quite make it work. In a lot
of lines, I wish I had played 13.Rf1 instead of 13.Rf2 to be able to swing my
king's rook over to the queenside in one go!} axb4 20. a5 bxc3 21. axb6 (21.
Qa6+ {Every chessplayer will know that this *has* to work... and every
chessplayer will know that it doesn't!} Kxa6 22. axb6+ Kb7 23. Ra7+ Kxb6 24.
Nxe6+ Bd4 {wins}) (21. Rb1 Bxd4 22. Bxd4 c5 {is unfortunately (for White) a
safe way for Black to neutralise the attack.} 23. Rxb6+ Qxb6 24. axb6 Rxd4 25.
Qxc3 Nd7) 21... Qxb6 22. Nxe6 Rxd3 23. Rxf7+ Rd7 24. Rxd7+ Nxd7 25. Bxb6 Kxb6 {
is pleasant for Black}) (19. Rd1 {Protecting the knight on d4 to allow the
queen to retreat to f1 targeting the f7-pawn was my default option. AlphaZero
is lukewarm.} Rg8 20. Qf1 Bxd4 21. Bxd4 g5 {and all of a sudden, AlphaZero is
coming in with ...Ng6 and ...g4-g3!} 22. Rfd2 Ng6 23. Qf6 g4 24. h4 g3 {
and now AlphaZero thinks that White should force perpetual with} 25. Be5 Nxe5
26. Rxd8 Ng4 27. Qd4 Rxd8 28. Qxd8 Qf4 29. Rd7+) 19... Bxd4 20. Bxd4 Rg8 (20...
c5 21. Nd5 Bxd5 22. Bf6 {wins}) 21. Qe3 {and AlphaZero contents itself with a
modest 54% expected score.}) 18... Bxd4 19. Nxd4 Nd7 20. Nxe6 fxe6 21. e5 g5 (
21... Nc5 22. Qxg6 Qxe5 {was part of AlphaZero's main line which it assessed
between 65-70% expected score.}) (21... Nxe5 22. Qe3 {gives Black a few
worries about holding the e6-pawn and the kingside pawns, while Black's
draughty second rank may prove awkward if White's major pieces get active.})
22. Re1 Nc5 23. Qe3 Qe7 24. Qf3 Kb7 25. Ref1 Qg7 26. Qe2 Rh7 27. Rf6 Qd7 {
was AlphaZero's line that ends up around 56% expected score. White has a small
edge, but nothing major. However, the plan of 14.a4 seems like a good way for
White to put a few challenges in Black's path.}) 14... g5 15. Raf1 Qd6 16. Ng4
O-O-O 17. Nf6 {[#] #2 When we placed this move on the board, AlphaZero burst
into life screaming 25% expected score for White (down from around 50% the
move before). When analysing with AlphaZero, it strikes me how quick AlphaZero
is to evaluate the long-term prospects of both sides. We'll see the reason
AlphaZero thinks White's long-term prospects are poor after a few more moves
from both sides.} Nd7 18. Nh5 Be5 19. g4 f6 {[#]In this position, Black's king
is completely safe. Not only has White not yet started any operations against
Black's king, all of White's pieces are on the kingside, far away from Black's
king. The integrity of White's kingside cover on the other hand depends
entirely on the knight on h5. If this knight moves, then Black will play ...h5
and break open the kingside; if White leaves the knight on h5, then Black can
capture it with ...Bf7xh5 and then open up the g-file with ...g4 or even
target the h5-pawn with ....Qe6-e8 and ...Nf8-e6-g7/f4. It takes Black some
time, but since White will be even slower creating any counterplay against
Black's king, the position is desperate for White in the long term.} 20. b3 (
20. Nd1 {I mis-clicked and entered this move instead of 20.b3. AlphaZero
surprised me with the lovely idea} f5 21. exf5 Bf7 {intending ...Bxh5 and ...
Nf6 followed by doubling on the e-file and then rounding up the h-pawn with
the black queen! A great idea!}) 20... Bf7 21. Nd1 Nf8 {[#] #3 A remarkable
concept from Magnus and yet it didn't take us completely by surprise. This was
not AlphaZero's preferred move, but while taking a look at the line 21...Kb7,
its main variation 22.Qe2 Nf8 23.Nxf6 Bg6 24.Nh5 Ne6 popped up with just a 35%
expected score for White. We were looking at this, feeling that it looked good,
but was it really necessary?... when Magnus' move was played! Great minds
think alike! The strategy is really good. Black's knight on d7 is passive, but
the light-squared bishop has cleared a magnificent outpost on e6 for the
knight. Moving it there would cost a pawn...but how much can White do with
that pawn? The f-file is opened for White's doubled rooks, but White's knight
on f6 is in the way and it doesn't have any dangerous squares to go to from
there: Black's king is safe! The only path is back to h5 and there it will be
taken exposing the white king further. Furthermore, all White's entry squares
on the f-file are covered, and once the knight comes to e6, then it can block
the f-file by moving to f4 as occurred in the game. It's really advanced
strategy and it was interesting to see it also feature in some of AlphaZero's
analysis. Magnus played really beautifully in this part of the game. In
particular, the idea of simply capturing the h5-pawn in order to be able to
lift a blockade on g4 with ...h5 featured early in AlphaZero's variations and
turned out to be really strong. But didn't Fabiano defend well!} 22. Nxf6 Ne6
23. Nh5 Bxh5 24. gxh5 Nf4 25. Bxf4 gxf4 26. Rg2 Rhg8 27. Qe2 Rxg2+ 28. Qxg2 Qe6
29. Nf2 Rg8 30. Ng4 Qe8 31. Qf3 Qxh5 32. Kf2 Bc7 33. Ke2 Qg5 34. Nh2 h5 35. Rf2
Qg1 36. Nf1 h4 37. Kd2 Kb7 38. c3 Be5 39. Kc2 Qg7 40. Nh2 Bxc3 41. Qxf4 Bd4 42.
Qf7+ Ka6 43. Qxg7 Rxg7 44. Re2 Rg3 45. Ng4 Rxh3 46. e5 Rf3 47. e6 Rf8 48. e7
Re8 49. Nh6 h3 50. Nf5 Bf6 51. a3 b5 52. b4 cxb4 53. axb4 Bxe7 54. Nxe7 h2 55.
Rxh2 Rxe7 56. Rh6 Kb6 57. Kc3 Rd7 58. Rg6 Kc7 59. Rh6 Rd6 60. Rh8 Rg6 61. Ra8
Kb7 62. Rh8 Rg5 63. Rh7+ Kb6 64. Rh6 Rg1 65. Kc2 Rf1 66. Rg6 Rh1 67. Rf6 Rh8
68. Kc3 Ra8 69. d4 Rd8 70. Rh6 Rd7 71. Rg6 Kc7 72. Rg5 Rd6 73. Rg8 Rh6 74. Ra8
Rh3+ 75. Kc2 Ra3 76. Kb2 Ra4 77. Kc3 a6 78. Rh8 Ra3+ 79. Kb2 Rg3 80. Kc2 Rg5
81. Rh6 Rd5 82. Kc3 Rd6 83. Rh8 Rg6 84. Kc2 Kb7 85. Kc3 Rg3+ 86. Kc2 Rg1 87.
Rh5 Rg2+ 88. Kc3 Rg3+ 89. Kc2 Rg4 90. Kc3 Kb6 91. Rh6 Rg5 92. Rf6 Rh5 93. Rg6
Rh3+ 94. Kc2 Rh5 95. Kc3 Rd5 96. Rh6 Kc7 97. Rh7+ Rd7 98. Rh5 Rd6 99. Rh8 Rg6
100. Rf8 Rg3+ 101. Kc2 Ra3 102. Rf7+ Kd6 103. Ra7 Kd5 104. Kb2 Rd3 105. Rxa6
Rxd4 106. Kb3 Re4 107. Kc3 Rc4+ 108. Kb3 Kd4 109. Rb6 Kd3 110. Ra6 Rc2 111. Rb6
Rc3+ 112. Kb2 Rc4 113. Kb3 Kd4 114. Ra6 Kd5 115. Ra8 1/2-1/2
[Event "Wch 2018"]
[Site "London"]
[Date "2018.11.10"]
[Round "2"]
[White "Carlsen, Magnus"]
[Black "Caruana, Fabiano"]
[Result "1/2-1/2"]
[ECO "D37"]
[Annotator "AlphaZero - 1 min / move"]
[PlyCount "97"]
[SourceVersionDate "2018.11.09"]
1. d4 Nf6 2. Nf3 d5 3. c4 e6 4. Nc3 Be7 5. Bf4 O-O 6. e3 c5 7. dxc5 Bxc5 8. Qc2
Nc6 9. a3 Qa5 10. Rd1 Rd8 {After this unusual move from Caruana, AlphaZero
joined the ranks of every commentator - human and engine - in recommending 11.
Nd2! AlphaZero was pretty upbeat about White's chances giving between 65% and
70% expected score.} 11. Be2 Ne4 12. O-O Nxc3 13. bxc3 h6 14. a4 Ne7 {A great
concept from Caruana and also AlphaZero's instant reaction! With 13...h6
(protecting the kingside against Ng5) and 14...Ne7 (making cxd5 unattractive
due to Nxd5 hitting the bishop on f4 and the c3-pawn), Black has neutralised
much of White's potential for causing trouble.} 15. Ne5 Bd6 (15... f6 {was
AlphaZero's preferred idea, when it considered Black would have just more than
50% expected score after} 16. Ng6 e5 17. Bg3 Be6 18. cxd5 Nxd5 19. Bc4 Qb6 20.
Ba2 Qc6 21. Bb1 Qe8 22. a5 Rac8 23. h3 e4 {Caruana's choice is super-solid and
I can imagine it appeals more in a practical game than weakening the kingside
light squares with ...f6.}) 16. cxd5 Nxd5 17. Bf3 (17. Nxf7 {[#] #1 Again
AlphaZero's instant choice, and the best way to avoid getting into trouble
with White.} Kxf7 18. Bh5+ Kg8 19. Bxd6 Rxd6 20. c4 {was AlphaZero's
preference above 20.e4.} Qc7 {was AlphaZero's main line and unfortunately
easily enough to contain White's initiative.} 21. Qd3 Bd7 22. cxd5 Bxa4 23. Rc1
Qd7 24. dxe6 Qxe6 25. Bg4 Rxd3 26. Bxe6+ Kf8 27. Rc7 Rd6 28. Rf7+ Kg8 29. Ba2
Rd2 30. Be6 {was the source of AlphaZero's 50% evaluation, leading to a draw
by repetition.}) 17... Nxf4 18. exf4 Bxe5 19. Rxd8+ Qxd8 20. fxe5 Qc7 21. Rb1
Rb8 22. Qd3 Bd7 23. a5 Bc6 24. Qd6 Qxd6 25. exd6 Bxf3 26. gxf3 Kf8 27. c4 Ke8
28. a6 b6 29. c5 Kd7 30. cxb6 axb6 31. a7 Ra8 32. Rxb6 Rxa7 33. Kg2 e5 34. Rb4
f5 35. Rb6 Ke6 36. d7+ Kxd7 37. Rb5 Ke6 38. Rb6+ Kf7 39. Rb5 Kf6 40. Rb6+ Kg5
41. Rb5 Kf4 42. Rb4+ e4 43. fxe4 fxe4 44. h3 Ra5 45. Rb7 Rg5+ 46. Kf1 Rg6 47.
Rb4 Rg5 48. Rb7 Rg6 49. Rb4 1/2-1/2
[Event "Wch 2018"]
[Site "London"]
[Date "2018.11.12"]
[Round "3"]
[White "Caruana, Fabiano"]
[Black "Carlsen, Magnus"]
[Result "1/2-1/2"]
[ECO "B31"]
[Annotator "AlphaZero - 1 minute / move"]
[PlyCount "97"]
[SourceVersionDate "2018.11.09"]
1. e4 c5 2. Nf3 Nc6 3. Bb5 g6 4. Bxc6 dxc6 5. d3 Bg7 6. O-O (6. h3 Nf6 7. Nc3
b6 8. Be3 e5 9. O-O O-O 10. a3 Qe7 11. Qb1 Nh5 12. b4 f5 13. bxc5 f4 14. Bd2
bxc5 15. Qb3+ Be6 16. Qa4 Rac8 17. Qa5 g5 18. Na4 g4 19. hxg4 Bxg4 20. Qxc5 Qf6
21. Nh2 f3 22. Nxg4 Qg6 23. Qe7 fxg2 24. Rfb1 Qxg4 25. Qg5 Qe2 26. Qe3 Qg4 27.
Qg5 Qxg5 28. Bxg5 Nf4 29. Bxf4 exf4 30. Kxg2 f3+ 31. Kf1 Rf4 32. c3 Rd8 33. d4
Bh6 34. Ke1 Rxe4+ 35. Kd1 c5 36. Kc2 cxd4 37. Kd3 Re2 38. c4 Rxf2 39. Rd1 Re2 {
0-1 (39) Caruana,F (2820)-Carlsen,M (2862) Wijk aan Zee 2015}) 6... Qc7 7. Re1
{Also AlphaZero's reaction.} e5 8. a3 {AlphaZero was recommending 8.a4 with a
66% expected score. 8.a3 brought things back to a 63% expected score.} Nf6 9.
b4 O-O 10. Nbd2 (10. bxc5 {was AlphaZero's choice when I wasn't at all sure
how Black was intending to continue.} Nd7 11. Be3 f5 {is the obvious idea,
aiming to drive away the bishop from e3 with ...f4 followed by recapturing on
c5. It leads to a scenario reminiscent of these players' game at Wijk Aan Zee
in 2015: White has a queenside advantage but Black is coming for White's king.
That may have been a reason for Magnus to offer it and for Fabiano to choose a
quieter path. This turned out to be an excellent piece of judgment from
Fabiano.} 12. a4 f4 13. Bd2 {is AlphaZero's subtle improvement over 13.Bc1
with the following ideas:} (13. Bc1 Nxc5 14. Ba3 b6 15. Nbd2 Rd8 16. Bxc5 bxc5
17. Rb1 Be6 18. Ng5 Bc8 19. Nc4 Qe7 20. Nf3 g5 21. h3 h5 22. Nh2 {is
AlphaZero's idea with a 75% expected score, although to human eyes, Black's
kingside advance looks quite dangerous.}) 13... a5 (13... Nxc5 14. Bb4 b6 (
14... Qb6 15. Ba3 {is very awkward for Black.} Rd8 16. a5 Qxa5 17. Nbd2 Be6 18.
Ng5 Re8 19. Qb1 {with an 85% expected score for White.}) 15. a5 Be6 16. Ng5 Bd7
17. Nd2 Bf6 18. Ngf3 Rfd8 19. Ra3 Ne6 20. Bc3 b5 21. Qa1 {and the e5-pawn
falls with an 80% expected score for White.}) (13... Qd8 {was another
AlphaZero suggestion, leaving the pawn on c5 in order to push through the
kingside pawns. AlphaZero strongly prefers White, but the variations felt
fraught for a human player.}) 14. Bc1 Nxc5 15. Ba3 b6 16. Nbd2 Rd8 17. Bxc5
bxc5 18. Rb1 {The inclusion of the moves Bd2 and ...a5 has loosened Black's
cover of his queenside by unprotecting the b6-square, which makes generating
White counterplay on that wing much easier.} Qe7 19. Nc4 Be6 20. Nfd2 {is an
80% expected score for White.}) 10... Bg4 11. h3 Bxf3 12. Nxf3 cxb4 13. axb4 a5
(13... Nd7 14. Be3 a6 {was AlphaZero's preferred continuation albeit with a
67% expected score for White.}) 14. bxa5 Rxa5 15. Bd2 (15. Rxa5 Qxa5 16. Bd2
Qc7 17. Qa1 Nd7 18. Qa7 Rb8 19. Ra1 Bf8 20. Be3 {and AlphaZero gave White a
massive 80% expected score. Unfortunately for Fabiano, his move order gives
Black a chance to hold on to the a-file after which White's advantage
gradually dissipates.}) 15... Raa8 16. Qb1 Nd7 17. Qb4 Rfe8 18. Bc3 b5 19. Rxa8
Rxa8 20. Ra1 Rxa1+ 21. Bxa1 Qa7 22. Bc3 Qa2 23. Qb2 Qxb2 24. Bxb2 f6 25. Kf1
Kf7 26. Ke2 Nc5 27. Bc3 Ne6 28. g3 Bf8 29. Nd2 Ng5 30. h4 Ne6 31. Nb3 h5 32.
Bd2 Bd6 33. c3 c5 34. Be3 Ke7 35. Kd1 Kd7 36. Kc2 f5 37. Kd1 fxe4 38. dxe4 c4
39. Nd2 Nc5 40. Bxc5 Bxc5 41. Ke2 Kc6 42. Nf1 b4 43. cxb4 Bxb4 44. Ne3 Kc5 45.
f4 exf4 46. gxf4 Ba5 47. f5 gxf5 48. Nxc4 Kxc4 49. exf5 1/2-1/2
[Event "Wch 2018"]
[Site "London"]
[Date "2018.11.13"]
[Round "4"]
[White "Carlsen, Magnus"]
[Black "Caruana, Fabiano"]
[Result "1/2-1/2"]
[ECO "A29"]
[Annotator "AlphaZero - 1 min / move"]
[PlyCount "67"]
[SourceVersionDate "2018.11.09"]
1. c4 e5 2. Nc3 Nf6 3. Nf3 Nc6 4. g3 d5 5. cxd5 Nxd5 6. Bg2 Bc5 {AlphaZero's
favourite move after an initial preference for 6...Nb6.} 7. O-O O-O 8. d3 Re8 {
Also AlphaZero's first choice.} 9. Bd2 {AlphaZero's main move.} Nxc3 (9... Nb6
{followed by ...Bf8 (transposing into a 6...Nb6 line in which White has not
played the most critical moves) was AlphaZero's intention. It seems very
sensible, but has never been played before.} 10. Rc1 Bf8) 10. Bxc3 Nd4 11. b4 {
A novelty, varying from a previous game of Caruana's.} (11. Rc1 Bb6 12. Re1 Bg4
13. Bxd4 exd4 14. a3 c6 15. Nd2 a5 16. Bf3 Be6 17. Qa4 Bc7 18. Ne4 Bd5 19. Nc5
Rb8 20. Bxd5 Qxd5 21. Qc4 Red8 22. b4 axb4 23. axb4 Qh5 24. b5 Ba5 25. Rf1 cxb5
26. Qxb5 b6 27. Qa4 Qxe2 28. Na6 Ra8 29. Nc7 Rac8 30. Qb5 Bc3 31. Nd5 Qe6 32.
Nf4 Qd6 33. Rb1 Rb8 34. Rfc1 h6 35. Rc2 Rdc8 36. Re2 Qc6 37. Nd5 Kh7 38. Qb3
Re8 39. Rc2 Re5 40. Nxc3 Rc5 41. Qxf7 dxc3 42. d4 Rc4 43. d5 Qg6 44. Qxg6+ Kxg6
45. Rb3 b5 46. Rcxc3 Rxc3 47. Rxc3 b4 48. Rb3 Kf5 49. Kf1 Ke5 50. Ke2 Kxd5 51.
Kd3 Kc5 52. Kc2 Rf8 53. f4 Re8 54. Rd3 Kc4 55. Rd7 Re2+ 56. Kb1 Kb3 57. Rd3+
Kc4 58. Rd7 Rxh2 59. Rxg7 Kb3 60. Kc1 Rh1+ 61. Kd2 Kb2 62. f5 Rf1 63. g4 b3 64.
Ke2 Rf4 65. Ke3 Rf1 66. Ke2 Rf4 67. Ke3 {1/2-1/2 (67) So,W (2778)-Caruana,F
(2816) Paris 2018}) 11... Bd6 12. Rb1 {These moves also formed AlphaZero's
chosen plan. AlphaZero was lukewarm about White's chances.} Nxf3+ 13. Bxf3 a6
14. a4 c6 {White has to decide whether to play b5 now, which creates a
weakness in Black's position but simplifies the position, or to keep
complexity in the position by delaying b5. Keeping complexity means not
carrying out the plan you really want to play as Black can hold back b4-b5
securely with ...Bd7. In principle, such positions without knights feel quite
"flat" - it's very hard for either side to create severe problems for the
opponent.} 15. Re1 {Presumably to sidestep a future ...Bh3 with tempo? It's a
brave decision not to execute your desired goal in the opening to keep the
position tense. Unfortunately, after 15...Bd7, it's not easy to find a
convincing plan for White.} (15. Qc2 Bd7 16. Bd2 {AlphaZero simply wanted to
play 15.b5 with a small edge but when I pressed it to find something else, it
came up with this idea: moving the bishop round to e3 and then to c5 to exert
a little pressure on Black's position. It's not much, but it's a plan!} Qe7 {
Preventing Be3} 17. Rfc1 h6 18. Qb2 {looking for Be3-c5. AlphaZero then
expects Black to prepare ...Be6-d5 to neutralise White's light-squared bishop.}
) 15... Bd7 16. e3 Qf6 (16... Qe7 {was AlphaZero's favourite idea, with an
unusual and strong queenside plan in mind:} 17. Bg2 Rac8 18. Qb3 h6 19. Qb2 b5
20. Ra1 {was my attempt to maintain tension in the position and keep control...
it failed!} (20. axb5 cxb5 21. d4 (21. Ra1 Rxc3) 21... e4 22. d5 f6 {when
AlphaZero evaluated the position as a 55% expected score for Black.}) 20... a5
21. bxa5 b4 22. Bd2 c5 23. a6 Bc6 (23... e4 24. Bxe4 c4 {was also assessed as
a 75% expected score. It's definitely worth seeing, but I think that 23...Bc6
is safe and strong.}) 24. Bxc6 Rxc6 {was also assessed around a 75% expected
score by AlphaZero. Black will gang up on the a4-pawn.}) 17. Be4 Bf5 {
AlphaZero didn't like this move particularly, feeling that it moved the
assessment from slightly better for Black to slightly better for White. Once
again, AlphaZero was recommending 17...Rac8 followed by ...b5.} 18. Qf3 Bxe4
19. Qxf6 gxf6 20. dxe4 b5 {AlphaZero gave White an expected score of 53% here
- just a very small advantage. It's very important to stop a5: after a move
like 20...Red8 21.a5, AlphaZero gave White an expected score of 62%.} 21. Red1
Bf8 22. axb5 axb5 23. Kg2 Red8 24. Rdc1 Ra4 25. Be1 Rc8 26. Rc2 Kg7 27. Kf3 h5
28. Ke2 Kg6 29. h3 f5 30. exf5+ Kxf5 31. f3 Be7 32. e4+ Ke6 33. Bd2 Bd6 34.
Rbc1 1/2-1/2
[Event "Wch 2018"]
[Site "London"]
[Date "2018.11.15"]
[Round "5"]
[White "Caruana, Fabiano"]
[Black "Carlsen, Magnus"]
[Result "1/2-1/2"]
[ECO "B31"]
[Annotator "AlphaZero 1 minute / move"]
[PlyCount "67"]
[SourceVersionDate "2018.11.09"]
1. e4 c5 2. Nf3 Nc6 3. Bb5 g6 4. O-O Bg7 5. Re1 e5 6. b4 (6. Bxc6 {was
AlphaZero's preferred choice, heading for the structure we saw in the first
and third games but 6.b4 was its second choice. The b4-break is very thematic
- we saw Fabiano playing it in the third game (after preparation with a2-a3) -
but feels more like an Evans Gambit when played immediately!}) 6... Nxb4 (6...
cxb4 {is also reasonable according to AlphaZero. It came up with the following
plan:} 7. a3 b3 8. cxb3 Nge7 9. Bc4 {A new move: 9.Bb2 is White's normal
approach, but AlphaZero wants to put the bishop on another diagonal.} O-O 10.
Nc3 d6 11. a4 a6 12. h3 Be6 13. Ba3 {with a 62% expected score.}) 7. Bb2 a6 8.
a3 axb5 9. axb4 Rxa1 10. Bxa1 d6 11. bxc5 Ne7 12. Qe2 {AlphaZero's second move}
(12. cxd6 Qxd6 13. d4 exd4 14. Bxd4 Bxd4 15. Nxd4 b4 16. Nb5 Qxd1 17. Rxd1 O-O
18. c3 {was AlphaZero's line with a 60% expected score which looks a little
optimistic to me!}) 12... b4 13. Qc4 {AlphaZero's second move} (13. c3 {
was AlphaZero's main line and pretty interesting.} dxc5 (13... bxc3 14. Qb5+
Nc6 15. Nxc3 O-O 16. cxd6 Qxd6 17. Nd5 {with a small edge for White.}) 14. Qb5+
Nc6 15. Qxc5 b3 16. c4 Qa5 17. Qxa5 Nxa5 18. d3 O-O 19. Bc3 b6 20. Kf1 f5 21.
Nbd2 fxe4 22. Nxe4 Bf5 23. Nfd2 Bh6 24. Bxa5 Bxe4 25. dxe4 bxa5 26. Nxb3 a4 {
was AlphaZero's main line at move 13 with a 55% expected score.}) 13... Qa5 14.
cxd6 Be6 15. Qc7 Qxc7 16. dxc7 Nc6 17. c3 Kd7 18. cxb4 Ra8 19. Bc3 Kxc7 20. d3
Kb6 21. Bd2 Rd8 22. Be3+ Kb5 23. Nc3+ Kxb4 24. Nd5+ Bxd5 25. exd5 Rxd5 26. Rb1+
Kc3 27. Rxb7 Nd8 28. Rc7+ Kxd3 29. Kf1 h5 30. h3 Ke4 31. Ng5+ Kf5 32. Nxf7 Nxf7
33. Rxf7+ Bf6 34. g4+ 1/2-1/2
[Event "Wch 2018"]
[Site "London"]
[Date "2018.11.16"]
[Round "6"]
[White "Carlsen, Magnus"]
[Black "Caruana, Fabiano"]
[Result "1/2-1/2"]
[ECO "C42"]
[Annotator "AlphaZero 1 minute / move"]
[PlyCount "160"]
[SourceVersionDate "2018.11.09"]
1. e4 e5 2. Nf3 Nf6 3. Nxe5 d6 4. Nd3 Nxe4 5. Qe2 Qe7 6. Nf4 Nc6 7. Nd5 Nd4 8.
Nxe7 Nxe2 9. Nd5 Nd4 10. Na3 Ne6 11. f3 N4c5 12. d4 Nd7 13. c3 c6 14. Nf4 Nb6
15. Bd3 d5 16. Nc2 Bd6 17. Nxe6 Bxe6 18. Kf2 h5 {A move AlphaZero had been
advocating for quite some time. Caruana's next sequence of moves followed
AlphaZero's recommendations.} 19. h4 Nc8 20. Ne3 Ne7 21. g3 c5 {I wondered
whether Caruana would play this move of AlphaZero's. It shows an excellent
objective appraisal of Black's active chances in this position, and nerves of
steel when there seems little wrong with just maintaining the status quo as
Black.} 22. Bc2 {The coming passage of play is a good example of why I love
analysing with AlphaZero. AlphaZero is extremely sensitive to the optimal
placement of pieces. As soon as Carlsen played the bishop to c2, AlphaZero
started screaming a negative expected score for White and insisted the bishop
should be returned to d3 on the next move. It also started to find some
excellent deployments of the Black pieces.} (22. dxc5 Bxc5 23. Rd1 O-O {
and ...d5-d4 cannot be stopped. AlphaZero's main line ended up with a small
edge for Black.}) 22... O-O 23. Rd1 Rfd8 (23... a5 {was AlphaZero's favourite
move, when after} 24. a4 cxd4 25. cxd4 Nc6 26. Bb1 Rfe8 27. Rd3 Ra6 28. Bd2 Rb6
{it saw a 66% expected score for Black.}) 24. Ng2 cxd4 25. cxd4 Rac8 (25...
Rdc8 {was AlphaZero's instant choice again, keeping the a8-rook to support an
advance of the queenside pawns.} 26. Ne3 (26. Bd3 Bf5 27. Ne3 Bxd3 28. Rxd3 b5
29. Rc3 f5 30. f4 Kf7 {with a 59% expected score for Black.}) (26. Bb3 a5 27.
a4 Ra6 {The reason AlphaZero wants its rook on a8.} 28. Bf4 Kf8 29. Bxd6 Rxd6
30. Rd2 Rb6 {66% expected score for Black.}) 26... a5 27. Bd2 Nc6 28. Be1 a4
29. a3 b5 {followed by ...Rab8 with a pleasant initiative for Black.}) 26. Bb3
Nc6 27. Bf4 Na5 28. Rdc1 Bb4 29. Bd1 Nc4 30. b3 Na3 {AlphaZero was at a 60%
expected score for Black here which rose very slightly over the next few moves.
} (30... Nd6 31. Rxc8 Rxc8 32. Rc1 Rxc1 33. Bxc1 Nb5 34. Nf4 Nc3 35. Bc2 Nxa2
36. Bb2 Bd2 37. Nxh5 Bd7 38. g4 Bb5 39. Bb1 Nc1 40. Bc2 b6 {60% expected score
for Black.}) 31. Rxc8 Rxc8 32. Rc1 Nb5 33. Rxc8+ Bxc8 34. Ne3 Nc3 35. Bc2 Ba3
36. Bb8 a6 37. f4 Bd7 38. f5 Bc6 39. Bd1 Bb2 40. Bxh5 Ne4+ (40... Nxa2 {
was AlphaZero's preferred choice with an expected score of 65% after} 41. Ba7
Nc3 42. Be2 Ne4+ 43. Kg2 Bc1 44. Nf1 Nd6 45. g4 Nc8 46. Bb8 Bb2 47. Nh2 Bxd4
48. Nf3 Bf6 49. Kg3 Be8 50. Kf4 a5) 41. Kg2 Bxd4 42. Bf4 Bc5 43. Bf3 Nd2 44.
Bxd5 Bxe3 45. Bxc6 Bxf4 46. Bxb7 Bd6 47. Bxa6 Ne4 48. g4 Ba3 {AlphaZero was
now at an 82% expected score.} 49. Bc4 Nc3 50. b4 Bxb4 51. Kf3 Kf8 52. g5 Na4
53. Bb5 {Looking through the lines and trying out a few lines, AlphaZero
clearly had the following schematic plan in mind if left to its own devices:
1. Place the knight on d6 2. Place the pawn on f6 3. Place the bishop on d2 4.
Place the knight on e5 5. Bring the king in. In other words, optimise the
position of the knight, then fix the kingside pawn structure, then reoptimise
the knight and then activate the king. It's a pretty good plan, but
unfortunately White can achieve the fortress-type position that Carlsen
achieved in the game.} Nc5 (53... Nb6 54. a4 f6 {was AlphaZero's main move.} (
54... Be1 55. h5 (55. Kg4 {This attempt of mine best reflects the plan
AlphaZero was aiming for} Nd5 56. Bc6 Ne3+ 57. Kf4 Nc4 58. Bb5 Nd6 59. Kg4 Ke7
60. Bd3 f6 61. Kh5 Nf7 62. Bc4 Ne5 63. Bd5 Kd6 64. Bb7 Kc7 65. Ba6 Kb6 66. Be2
Ka5 67. Bd1 Kb4 68. Bc2 Nf3) 55... Bd2 56. f6 gxf6 57. gxf6 Nd5 58. Ke2 Ba5 59.
h6 Nxf6 60. Bc4 {AlphaZero's lines here were ending up with White bringing the
king to the queenside and winning the bishop for the pawn and then drawing the
bishop vs knight and pawn endgame.}) 55. Kg4 Bd2 56. Kh5 Nd5 57. Bc4 Ne7 {
A great position for the knight, attacking f5 and ready to transfer to c6.
However, White still has his trump in the a-pawn.} 58. Be6 Ke8 59. gxf6 gxf6
60. a5 Bxa5 61. Kh6 Kf8 62. Kh7 Be1 63. h5 {which should hold just as in the
game. I did get tricked a few times by AlphaZero when defending it, but I
don't think I was finding best play for White.}) 54. a4 f6 55. Kg4 Ne4 56. Kh5
Be1 57. Bd3 Nd6 58. a5 Bxa5 59. gxf6 gxf6 60. Kg6 Bd8 61. Kh7 Nf7 62. Bc4 Ne5
63. Bd5 Ba5 64. h5 Bd2 65. Ba2 Nf3 66. Bd5 Nd4 67. Kg6 Bg5 68. Bc4 Nf3 69. Kh7
Ne5 70. Bb3 Ng4 71. Bc4 Ne3 72. Bd3 Ng4 73. Bc4 Nh6 74. Kg6 Ke7 75. Bb3 Kd6 76.
Bc2 Ke5 77. Bd3 Kf4 78. Bc2 Ng4 79. Bb3 Ne3 80. h6 Bxh6 1/2-1/2
[Event "Wch 2018"]
[Site "London"]
[Date "2018.11.18"]
[Round "7"]
[White "Carlsen, Magnus"]
[Black "Caruana, Fabiano"]
[Result "1/2-1/2"]
[ECO "D37"]
[Annotator "AlphaZero 1 minute / move"]
[PlyCount "79"]
[SourceVersionDate "2018.11.09"]
1. d4 Nf6 2. Nf3 d5 3. c4 e6 4. Nc3 Be7 5. Bf4 O-O 6. e3 c5 7. dxc5 Bxc5 8. Qc2
Nc6 9. a3 Qa5 {The second outing for the 5.Bf4 Queen's Gambit Declined, and
Carlsen varies from 10.Rd1 with the third most popular move in this position.}
10. Nd2 Qd8 {Caruana shows the depth of his preparation by choosing a very
rare continuation. AlphaZero liked it although it wasn't AlphaZero's first
choice.} (10... Bb4 {AlphaZero's main move. This line was played
(unsuccessfully) as Black by Kasimdzhanov (Caruana's second) while Simen
Agdestein (Carlsen's former trainer) played a fairly important game as White.}
11. cxd5 exd5 12. Nb3 Bxc3+ 13. bxc3 Qd8 {AlphaZero prefers this modest
retreat after first looking at 13...Qa4.} 14. Be2 Bd7 (14... Ne4 {was the only
other high-level outing in recent years} 15. f3 Nf6 16. g4 Be6 17. O-O Rc8 18.
Qd2 {was balanced and complicated in Topalov,V -Nakamura,H Paris 2017}) 15. O-O
Rc8 16. Rfd1 Na5 {was a 56% expected score for White: nothing major, but I
could imagine that the imbalance of 2 bishops vs bishop and knight might have
attracted Carlsen to the position.}) 11. Nb3 (11. Rd1 e5 12. Bg5 d4 13. Nb3 Be7
{transposes back into a former main line of the 5.Bf4 QGD, normally reached
after 10.Rd1 (instead of Nd2) Be7 11.Nd2 e5 12.Bg5 d4 13.Nb3 Qd8. This is
interesting because Fabiano played 10...Rd8 instead of 10...Be7 in the second
game. I guess we can deduce from this that Magnus would have met 10.Rd1 Be7
with 11.Be2 as he does not choose to exploit this transposition.} 14. Be2 a5
15. O-O a4 16. Nc1 g6 {and AlphaZero considers that Black has few problems
(51% expected score for White.)} 17. e4 Qa5 18. Bxf6 Bxf6 19. Nd5 Bg7 20. c5
Be6 21. Nb6 {51% expected score.}) (11. Nf3 {AlphaZero and Anish Giri were of
the same opinion! This was AlphaZero's main line though after 11...Qa5, it
preferred 12.Rd1!} Qa5) 11... Bb6 {A very sensible novelty and again
AlphaZero's recommendation.} 12. Be2 (12. O-O-O {was the move I wanted to see!}
Bd7 {Threatening to open the c-file with ...dxc4 if White starts any kingside
action such as g4.} 13. Kb1 Rc8 14. Bg5 {Here AlphaZero had 2 lines, both of
which were very close to each other in terms of evaluation:} h6 (14... dxc4 15.
Nd2 Ne5 16. Nde4 {As we shall see, this theme of playing Bg5 followed by
Nd2-e4 after Black captures on c4 also could have occurred in the game.} Qe7
17. f4 Nd3 18. Bxd3 cxd3 19. Qxd3 Bc6 {with a 55% expected score for White.})
15. Bh4 Na5 16. Nd2 Qe7 17. e4 Bd4 {was a very fraught-looking line that
AlphaZero was calculating on the 12th move.} 18. f4 dxe4 19. Ncxe4 Bc6 20. Nf3
Bxe4 21. Qxe4 Qc7 {It hangs together tactically by a miracle it seems!} 22. Qc2
Be3 23. Bxf6 gxf6 24. f5 e5 25. Qe4 Qb6 26. Be2 Rfd8 {which AlphaZero assessed
as a 53% expected score for White. To be honest, looking at these lines, I
would be fairly happy to try them as White as they seem more risky for Black
than for White!}) (12. Rd1 Qe7 13. Bd3 (13. cxd5 Nxd5 14. Nxd5 exd5 15. Rxd5
Be6 16. Bd6 Qf6 17. Rd3 Rfd8 {jumped to a 90% expected score for Black, so I
guess we can call that a hot pawn!}) 13... h6 14. O-O dxc4 15. Bxc4 e5 16. Bg3
Be6 17. Nd2 Rfd8 {was a sample line of AlphaZero's with just a 53% expected
score for White.}) 12... Qe7 (12... dxc4 {was AlphaZero's choice and the
super-safe option.} 13. Bxc4 Bd7 14. O-O (14. Rd1 Rc8 15. O-O Ne7 16. Qe2 a6 {
was an alternative plan of AlphaZero's.}) 14... Bc7 15. Bxc7 Qxc7 16. Be2 Rac8
17. Rac1 Ne5 {with a 54% expected score for White. Caruana's choice is
slightly more risky.}) 13. Bg5 (13. Bg3 {was AlphaZero's main idea with an
interesting tactical idea.} Bd7 (13... dxc4 14. Nd2 {The point of 13.Bg3 is
that the Black reply ...Nd5 no longer hits the bishop on f4.}) 14. c5 {and now
AlphaZero's main line when analysing at move 13 was} Bxc5 15. Nxd5 Nxd5 16.
Qxc5 Qf6 17. Rb1 Rac8 18. O-O a6 19. Nd2 h5 {AlphaZero likes this move!} 20.
Ne4 Qh6 21. Nd6 h4 22. Bf4 Nxf4 23. exf4 Rb8 {with a 57% expected score for
White.}) 13... dxc4 14. Nd2 Ne5 15. O-O (15. Nce4 {was AlphaZero's main line
and lead to some unbalanced play after} Bd7 16. Qc3 Nxe4 17. Nxe4 f6 18. Qxe5
Bc6 (18... fxg5 19. Bxc4 Rac8 20. b3 h6 21. Rd1 {was a 61% expected score for
White.}) 19. Bxc4 Rae8 20. O-O Bc7 21. Bxf6 gxf6 22. Qd4 Be5 23. Qd3 {with a
56% expected score for White. I would have enjoyed seeing AlphaZero on the
Black side - it plays such positions with active counterplay very well - but
it wasn't to be as Carlsen went for the super-solid continuation. The rest of
the game stayed balanced until the end.}) 15... Bd7 16. Bf4 Ng6 17. Bg3 Bc6 18.
Nxc4 Bc7 19. Rfd1 Rfd8 20. Rxd8+ Rxd8 21. Rd1 Rxd1+ 22. Qxd1 Nd5 23. Qd4 Nxc3
24. Qxc3 Bxg3 25. hxg3 Qd7 26. Bd3 b6 27. f3 Bb7 28. Bxg6 hxg6 29. e4 Qc7 30.
e5 Qc5+ 31. Kh2 Ba6 32. Nd6 Qxc3 33. bxc3 f6 34. f4 Kf8 35. Kg1 Ke7 36. Kf2 Kd7
37. Ke3 Bf1 38. Kf2 Ba6 39. Ke3 Bf1 40. Kf2 1/2-1/2
[Event "Wch 2018"]
[Site "London"]
[Date "2018.11.19"]
[Round "8"]
[White "Caruana, Fabiano"]
[Black "Carlsen, Magnus"]
[Result "1/2-1/2"]
[ECO "B33"]
[Annotator "AlphaZero 1 minute / move"]
[PlyCount "75"]
[SourceVersionDate "2018.11.09"]
1. e4 c5 2. Nf3 Nc6 3. d4 cxd4 4. Nxd4 Nf6 5. Nc3 e5 6. Ndb5 d6 7. Nd5 Nxd5 8.
exd5 Nb8 9. a4 (9. c4 {was AlphaZero's preference with a 62% expected score.})
9... Be7 10. Be2 O-O 11. O-O Nd7 12. Bd2 f5 13. a5 a6 14. Na3 e4 (14... f4 {
was AlphaZero's move for Black with a very aggressive plan of transferring the
rook to the kingside while protecting the d6-pawn laterally. The following
line was its main line while considering 14...f4 for a minute.} 15. Bb4 Rf6 16.
Re1 Rh6 17. Nc4 Bf8 18. Ra3 Qg5 19. g3 e4 (19... Nf6 {was the move order that
AlphaZero went for when I left AlphaZero to run for a little longer:} 20. Bxd6
Bxd6 21. Nxd6 e4 22. Qd4 e3 23. Nxc8 Rxc8 24. d6 Qf5 25. Qxf4 Qxf4 26. gxf4 Ne4
27. Rxe3 (27. fxe3 Rg6+ 28. Kf1 Nd2+ 29. Kf2 Ne4+ {is a great draw by
repetition.}) 27... Rg6+ 28. Rg3 Nxg3 29. hxg3 Rxd6 {with a 60% expected score
for White. Great line!}) 20. Qd4 Nf6 21. Bxd6 Bxd6 22. Nxd6 e3 23. Qxf4 Qxf4
24. gxf4 Nxd5 25. Ne4 exf2+ 26. Nxf2 Nxf4 {with a 58% expected score for White.
}) 15. Nc4 Ne5 16. Nb6 Rb8 17. f4 exf3 18. Bxf3 g5 (18... Bf6 {was AlphaZero's
main move when it considered 2 ideas:} 19. c3 (19. Bc3 {Another move AlphaZero
was spending time on, introducing the idea of Bxe5 followed by Bg4 whenever
Black plays ...f4.} f4 20. Ra4 g5 21. Bxe5 Bxe5 22. Bg4 {was the main line
with a 56% expected score.}) 19... Nd7 20. Be3 Be5 21. Qd2 Nxb6 22. axb6 Re8
23. Bd4 Bd7 {with a 58% expected score for White.}) 19. c4 f4 20. Bc3 (20. Be4
{AlphaZero's second best move. It starts by considering this move and then
switches rapidly to the powerful 20.Bc3.} Bf5 21. Bxf5 Rxf5 22. Qc2 Rf7 23.
Rae1 Qc7 24. b4 Rbf8 25. Kh1 f3 26. g3 h5 27. Be3 h4 {with a 65% expected
score but some danger for the White king! It worried me anyway!}) 20... Bf5 {
AlphaZero did not like this move at all!} (20... Qc7 {was AlphaZero's main
move for Black, preventing c4-c5 for now,} 21. Rc1 (21. b4 Bf6 22. Rc1 Bf5 23.
Bxe5 Bxe5 24. Bg4 {was AlphaZero's initial line.}) 21... Bf5 (21... Bf6 22.
Bxe5 Bxe5 23. c5 Bxb2 24. Rc4 Bf5 25. c6 Rbe8 26. Bg4 Bg7 {and now either 27.
h4 or 27.Kh1 gave a 72% expected score for White.}) 22. b4 Bf6 23. Bxe5 Bxe5
24. Bg4 Qg7 25. Kh1 (25. c5 dxc5 26. bxc5 f3 27. Qxf3 Bd7 28. Qe4 Bb5 29. Rf5
Rbe8 30. Rxf8+ Rxf8 31. d6 Kh8 32. d7 {gave a 62% expected score for White.})
25... Bxg4 26. Qxg4 h5 27. Qe6+ (27. Qxh5 g4 {followed by ...Rf6-h6.}) (27. Qd1
g4 28. c5 Rbe8 29. Rc2 Rf7 30. c6 {was a 79% expected score for White.}) 27...
Qf7 (27... Kh8 28. Nd7 Rbe8 29. Qh3 {76% expected score.} Rf7 30. c5 Qg6 31.
Nxe5 dxe5 32. d6 g4 33. Qc3 Kh7 34. b5 axb5 35. a6 bxa6 36. Rcd1 Rd7 37. Qb3
Kg7 38. Qa2 {79% expected score.}) 28. Qxf7+ Rxf7 29. c5 Rd8 30. Rc4 Rc7 31.
Re4 {with a 65% expected score for White.}) (20... Bf6 21. c5 dxc5 22. d6 Nxf3+
(22... Be6 23. Be4) 23. Rxf3 Bg4 24. Qd5+ Kh8 25. Rf2 {gave an 82% expected
score for White.}) (20... g4 21. Be4) 21. c5 {The only move! 84% expected
score for White.} Nxf3+ (21... Bf6 22. c6 Qe7 23. h3 Rbe8 24. Bxe5 Bxe5 25. Bh5
Bg6 26. Bg4 f3 27. Bxf3 Rf4 28. Nd7 Bxb2 29. Ra4 {with an 84% expected score
for White.}) (21... g4 22. Be2 (22. Bxe5 dxe5 23. Bxg4 Bxc5+) 22... f3 23. gxf3
gxf3 24. Bxf3 {was something I thought fitted Magnus' mood that day, but I was
wrong this time!} Bg5 25. Bg2 Be3+ 26. Kh1 Qh4 (26... Bxc5 27. Bxe5 dxe5 28.
Rxf5 Rxf5 29. Qc2 Qf8 30. Nd7 {wins}) 27. Ra4 Qg5 (27... Ng4 28. h3 Nf2+ 29.
Rxf2 Qxf2 30. Rg4+ Bg6 31. cxd6 {with an 83% expected score for White}) 28. h4
Qg6 29. Rxf5 Qxf5 30. cxd6 Nf3 31. d7 Bxb6 32. axb6 Rf7 33. Rc4 {and I guess
that I should stop this madness somewhere: this is a very long way away from
move 21! AlphaZero gives an 85% expected score for White.}) 22. Qxf3 dxc5 23.
Rad1 (23. Rae1 {also reached a high evaluation but AlphaZero plumped very
quickly for 23.Rad1.} Bf6 24. h4 gxh4 (24... Bd4+ 25. Bxd4 cxd4 26. hxg5 Qxg5
27. Qxf4 Qxf4 28. Rxf4 d3 29. d6 Rbe8 30. Rxe8 Rxe8 31. Kh2 {77% expected
score for White}) 25. Qxf4 Bxc3 26. bxc3 Bg6 27. Qh2 Qg5 28. Nd7 Rxf1+ 29. Rxf1
Rc8 30. Qh3 Kg7 31. c4 {77% expected score for White.}) 23... Bd6 24. h3 (24.
Qh5 {was the move AlphaZero wanted (just like the rest of the world!)} Qe8 (
24... Bg6 25. Qh3 Bf5 26. Qh6 Rf7 27. Rfe1 Bf8 28. Qh5 Bc2 29. Rd2 Bg6 30. Qg4
Bf5 31. Qf3 {is an 87% expected score for White..}) 25. Qxg5+ Qg6 26. Qxg6+
hxg6 27. Nc4 Rbd8 (27... Be7 28. d6 Bg5 29. Rd5 {85% expected score for White.}
) 28. Nxd6 Rxd6 29. Be5 Rd7 30. Bxf4 {and AlphaZero considered this to be
completely winning. For example:} c4 31. Rd4 Bd3 32. Re1 Rf5 33. d6 Rxa5 34. h4
Rf5 (34... Rb5 35. Re7 Bf5 36. Bg5 Rxb2 37. Rxc4 Rxe7 (37... Rxd6 38. Rcc7) 38.
dxe7 Re2 39. Rd4) (34... Kf7 35. Bg5 Bf5 36. g4 Bd3 37. Bd2 {followed by mate
on the kingside.}) 35. g4 Rf8 36. Kf2 b5 37. Re7 Rf7 38. Re5 {93% expected
score.}) 24... Qe8 25. Nc4 Qg6 26. Nxd6 Qxd6 27. h4 gxh4 28. Qxf4 Qxf4 29. Rxf4
h5 30. Re1 Bg4 31. Rf6 Rxf6 32. Bxf6 Kf7 33. Bxh4 Re8 {56% expected score.} 34.
Rf1+ Kg8 35. Rf6 Re2 36. Rg6+ Kf8 37. d6 Rd2 38. Rg5 1/2-1/2