When to play the Rxc3 exchange sacrifice in the Najdorf: A data-driven approach

A little while ago I decided to refresh my opening repertoire and play more popular, mainstream openings. As black against the open Sicilian, I decided, I would play the Najdorf variation. I read a number of things about the ideas for both sides in that variation, and found one theme particularly interesting: the positional exchange sacrifice black can often make by using their rook to capture white's knight on c3.

I was excited to try it out. In a rapid game on Lichess, I got a chance. The opportunity presented itself, and I played Rxc3. But it backfired, and I lost. Analyzing the game with Stockfish afterwards, the engine hated the move. And that got me thinking: when is Rxc3 a good move, and when is it not?

Feeding the data into the computer

I downloaded the December 2021 PGN file from the Lichess database and fed the first gigabyte (about 430,000 chess games) into my computer. I programmed the computer to find positions from those games that met some criteria:

1. The opening of the game was the Najdorf variation
2. It's in the first 20 moves of the game
3. It's black to move
4. There's a knight on c3
5. Rxc3 is a legal move
6. White will not be able to recapture on c3 with a piece

(I added that last one after one full iteration of this - Stockfish never liked Rxc3 if it didn't mess up white's pawn structure.)

This resulted in 284 positions, from 116 games.

I had Stockfish analyze all of these positions. First, I had it filter out the games where the eval was under -2 or over +2. I don't need to look at games where one side is totally winning. This left 164 positions. Then I asked Stockfish the big question: Is Rxc3 a good move?

It didn't have to be the best move, but if the eval of Rxc3 was within 0.5 points of the best move, I considered that a yes.

I manually went though a bunch of these positions and found more filtering I had to do - some of these positions had some unrelated threat black had to respond to, or white playing bxc3 actually undoubled their pawns instead of doubling them. I ultimately made a collection of 30 "Yes" positions (where Rxc3 is a good move) and 30 "No" positions (where it is not). All 60 of these positions are at the bottom of this post.

Results

White's castling status had a big effect on whether Rxc3 was a good move.

• In positions where white's king had not castled, Rxc3 was a good move 100% of the time (although only five of the 60 positions met this criteria).
• In positions where white had castled queenside, Rxc3 was a good move 87% of the time (13/15).
• In positions where white had castled kingside, Rxc3 was a good move just 30% of the time (12/40).

These numbers make some intuitive sense. If white is uncastled, or castled queenside, then Rxc3 is both an attack on the king and a positional sacrifice. If white is castled kingside, Rxc3 is only a positional sacrifice, and you'd expect it to work less often.

Sometimes, there's a pawn on e4, and the knight on c3 is the only defender of it, and playing Rxc3 allows black to win this pawn immediately. This is also a good predictor of whether Rxc3 is a good move.

• In positions where Rxc3 wins the e4 pawn, it was a good move 77% of the time (10/13)
• In other positions, Rxc3 was a good move 43% of the time (20/47)

One thing that surprised me was that having a bishop on g7 (the "Dragondorf" arrangement) did not have a positive effect whether Rxc3 was a good move. I looked at positions where black had either already played Bg7 or had played g6 and looked like they were about to play Bg7.

• In Dragondorf positions, Rxc3 was a good move 38% of the time (3/8)
• In non-Dragondorf positions, Rxc3 was a good move 52% of the time (27/52)

What did I miss? I would love to hear suggestions of other positional factors to analyze, especially from stronger players with experience on either side of this opening. Leave a comment or contact me on Twitter at boldmovebydan.

The positions

All these positions are from games played on Lichess in December of 2021. Click on a link and it will open inline without popping up a new tab.

Positions where Rxc3 is a bad move

Positions where Rxc3 is a good move