14 June 2021

CCC No-castling Events

For my previous off-week engine post, CCC PGN IV (May 2021), I set myself a task:-

To locate the PGN for the five [CCC] events. [...] Locating the PGN turned out to be fairly trivial.

Now what? I decided to look at a no-castling event, where the engines are prohibited from castling to either side. This is an idea that GM Kramnik and others have floated to counter the growing corpus of opening theory. The supporters of the no-castling idea are generally anti-chess960 and since I'm a keen proponent of chess960, it might be worthwhile to check out the competition.

I located a post from early last year, TCEC S17 L1 Underway; CCC12 Bonus Series (February 2020), where I wrote,

After the event ended, the CCC ran a series of bonus matches including a round robin using 'No-Castling' rules. The current tournament, a six-engine affair titled 'A February Event', ends in a few days.

While scrolling back through the CCC archive to locate the PGN, I found three other 'no castling' events. Here is a summary of the four events in chronological order, where the date is for the first game in the event:-

  • 2020-01-25: 'CCC12 Bonus: No-Castling'
  • 2020-09-18: 'No-Castle II'
  • 2021-04-03: 'To Castle Or Not To Castle'
  • 2021-04-04: 'To Castle Or Not To Castle II'

All four events use the same technique to force a no-castling position. For each Rook-Knight pair on the back rank of the traditional start position they develop the Knight to its third rank, then move the Rook to the square vacated by the Knight, then return the Rook and Knight to their original start squares. This means that after eight move pairs (16-ply), the traditional start position is reached for the second time. Because the Rooks have moved, castling is no longer possible.

That sequence complicates the initial analysis, where I use SCID. A search on the repeated position (White's 9th move) returns the initial position (White's 1st move) where White develops a Knight to its third rank. Only after White's first real move (1.e4, 1.d4, etc.) does the search return real results. It turned out that not being able to calculate the frequency of the first real move was not important. The operators of the events had forced White's first real move, meaning that the engines were initially on their own for Black's first real move (1...e5, 1...d5, etc.)

For the 'No-Castle II' event, I encountered a number of irregularities. The first game in the PGN file is a standard Nimzo-Indian, where Black castles on the fourth move (1.d4 Nf6 2.c4 e6 3.Nc3 Bb4 4.e3 O-O). Only after that game does the event switch to no-castling. Later in the file, a monster game with '[PlyCount "489"]' makes SCID skid to a halt.

As for the last two events, 'To Castle Or Not To Castle', these force a different no-castling strategy. White moves the Knights back and forth for the first few moves, leaving the castling options intact, while Black loses the option(s) as described above. For the first few games on the file, Black loses only the O-O option, leaving O-O-O intact. I didn't check whether the same no-castling strategy applied throughout the two matches, or whether the 'O-O / O-O-O' options are turned off using other sequences.

That's not much of an analysis for this post, but I ran out of time to do more. I might come back to the subject another time. At least I verified that the CCC PGN files are accessible for further investigation and work as advertised.

For more about the subject, see my previous post The Engines' Value of Castling (May 2015). Using a different opening trick, I worked out that the value of castling is roughly equivalent to a Pawn. I should repeat the experiment with the same trick used in the two 'To Castle Or Not To Castle' events.

No comments: