Workshop: The Engines and Caruana's Queen Sacrifice

By Arno Nickel

Some remarks about the game Caruana-Nakamura from the perspective of a correspondence player with a certain amount of experience of analysing with engines.

In the Najdorf Sicilian new attempts in an early stage of the game are made again and again but this basically only proves the viability of the system.

Fabiano Caruana and Hikaru Nakamura at the London Chess Classic (Photo: Lennart Ootes)

If the whole line with 13...g5 should turn out to be dubious, many players will probably try 13...Bb7 which does not abandon the idea g7-g5 but postpones it, e.g. 14.Bg2 Rc8 15. Kb1 g5 16. Qh3 Nh7. The online database only shows four correspondence games with this line but all these games are relatively recent and they all ended with a draw.

But I think Black also has good chances to equalize in the line 13...g5 14.h4 gxf4 15.Be2 and now (instead of Nakamura's 15...b4?! and instead of Vachier-Lagrave's 15...Rg8) 15...Ne5, if you follow the variation indicated by Vachier-Lagrave (quoted according to Michal Krasenkow on ChessBase) and correct it slightly on move 21 - again according to correspondence material provided by the Live-Database:

16.Qxf4 Nexg4 17.Bxg4 e5 18.Nd5 Nxd5 19.Qf3 Bxg4 20.Qxg4 Nf6 21.Qf3 and now (instead of 21...exd4?) 21...Rc8.

The resulting position with 21...Rc8 might not be to everyone's taste with Black but the engines agree - even at greater search depths (e.g. Stockfish 8 at depth 35) that the position offers equal chances. I don't know whether anyone analyzed this deeply but apart from two correspondence games from 2013 and 2014, which both ended in a draw there are no games in the online database.

However, wenn comparing tournament chess and correspondence chess one has to be aware of some fundamental differences in the analytical approach:

1.) The tournament player neither can nor wants to memorize lengthy lines but usually tries to find a repertoire that avoids such lines and that first of all suits his understanding, his preferences and his playing style. The correspondence player does not need to remember lengthy and complicated lines but he has to know what he is doing when allowing or heading for openings with a lot of lines. He also would like to rely on his positional understanding but he always has to remain very self-critical and he always has to be ready to change plans if sober analysis requires this.

2.) The tournament player has good reasons to play for a win with Black - of course, depending on the opponent and the concrete situation. In fact, I suppose that Nakamura had hoped for or thought he might get winning chances after 15...b4. But this is a luxury the correspondence chess player can only very rarely not afford in the age of engines and computers. He principally has to assume that the opponent does not make mistakes. White might allow himself slight inaccuracies but before Black gets real winning chances White must make serious errors. However, one unexpected blunder is enough - after that it's only a matter of execution even if this lasts for 30 or 50 moves.

3.) In critical positions or in crucial situations the correspondence player has to squeeze out as much as possible from his computer. This is less letting the computer stupidly calculate ever more deeply (people like to say "over night) but working through variation trees. Because he cannot analyze each and every line he has to - a bit similar to scientific research - devise appropriate work plans for critical positions, which will later allow him to make the right decisions. A lot of moves might require only little time for analysis but individual moves (key moves) often require weeks.

As a result the correspondence chess player today uses and maintains the computer - if I may dare this comparison - like a Formula 1 racing car whereas the tournament player most of the time uses it only like a Volkswagen Rabbit to just get from A to B. Therefore one has to take computer-based statements of tournament players about their games with a certain care.

The assault gave Hikaru Nakamura something to think about. Particularly after 21.Nf5 - (Photo: Lennart Ootes)

I would like to give an example from the commentary by grandmaster Michal Krasenkow (which you can play through at the end of the article), whom I, despite my peripheral criticism, value highly, and I do indeed remember well that I once lost a drawish rook endgame in a freestyle game against the Rybka team which he had guided because I succumbed in time-trouble to the insinuations of the machine (they still have problems in rook endings; though less than humans).

Now Krasenkow writes about Caruana's wonderful move 21.Nf5 as point of the queen sacrifice that went before: "I must say that such 'real' queen sacrifices are an Achilles' heel of analytic engines. They almost always evaluate positions with, say, two minor pieces for a queen in favour of the strongest piece, while in reality, in human play, there may be a more than sufficient compensation for the material."

A couple of years ago I might have subscribed to this judgement but as far as the top engines of today are concerned it is no longer valid and has to be seen as dated. First of all, all engines rather evaluate the position when the queen sacrifice was played as slightly better for White. And as far as the real point - 21. Nf5 - is concerned it is not really necessary that the engines immediately realize that this is a winning move.

21. Nf5 - as played in Caruana-Nakamura  - in the Live-book since April 2016

Even the engines which at first prefer 21.Nc6 (with a slight plus for White) do not at all dismiss 21.Nf5 as much worse. This becomes apparent if you enter the move or if you let the computer indicate several main lines. This clearly shows that it only takes some computing time until all engines see this as a winning line; that is the engine are indeed able - and well able - to see if one side has positional compensation for a piece. Some see sooner some see later that it is more than just compensation. I don't know if Caruana found the move over the board. It is possible because the situation looks rather gloomy after this cheeky knight move which – in regard to the square f5 - often appears in the Najdorf. However, it is also quite possible that he had come across this idea (positional queen sacrifice on f6 with continuous pressure) at home, maybe in a slighly different position without b5-b4.

20.Nf5 in the Live-Book (without b5-b4): known since April 2016

However, after Caruana's queen sacrifice the question arises if it also works in the line with 15...Ne5 and now 16.Qxf4 Nexg4 17.Nxg4 e5 - instead of 18.Nd5 (Krasenkow/Vachier-Lagrave) - 18. Dxf6(!?) and after 18...Bxf6 19. Nd5 Qd8 again 20. Nf5! (in the Live-Book since April 2016) and now the only but maybe crucial difference to the position in the game is that Black has not played ...b4 yet and that his pawn structure on the queenside is therefore still fairly intact. Someone will definitely try this soon (may the “force” - incorporated by the engines - be with him or her), at the latest in a correspondence game and this will open another page in the endless chapter about the Najdorf.

Apart from the engines, for me as correspondence player the online-database is the most important tool to get an orientation in the opening, much more so because I tend to be somewhat lazy here. But if you read and interpret the move statistics (left) in the online database correctly you can quickly get fantastic insights into past and current opening trends and their chances and risks. You create your own individual tree by merging the games you consider important though I here prefer to keep a certain clarity. If this clarity is no longer guaranteed I split the material into a number of files or trees. But it is important to evaluate the games that make up the tree and to understand why they ended the way they did.

Let's Check (available online if the engine is activated) gives you a good idea of current evaluation trends because it bundles important information:

The evaluations of three independent engines or at least analyses (sometimes an engine appears twice). They often indicate very deep calculations which save the user calculation time if he is only interested in a first check. Because only three half-moves are shown this is indeed only a quick check. The date indicates topicality. Of course, in a lot of cases you have to keep in mind that a different move than the one favored by the engines can be "good" or "better", particularly in positions in which nothing seems to happen or that are still undeveloped. And searching with the Let’s Check method is also influenced by certain trends.

I only rarely use the Live-Book because it hardly ever gives new useful hints apart from the material that has already been established. The prompt that a move appears there which does not appear in the online database and in Let's Check, will make me wonder whether the move is any good or useful. The evaluations in the Live-Book are often inconsistent if you go deeper into the lines. It is also often hard to comprehend why one move should be better than another. The games on which these evaluations are based are not given. Therefore you never know whether they are any good. Finally, if you really want to know you simply have to analyse the move yourself. 

That is how I see it. The Live-Book is rather useful if you see its function together with Let’s Check. The Live-Book gives you a broader data basis about the way the engines treat the position in question.

Translation: Johannes Fischer

Michal Krasenkow's analysis of the game