Wednesday, October 29, 2008

Bright Ideas From Silicon


While drawing up my first list of online Jerome Gambit (1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.Bxf7+) resources (see "Jerome Gambit Scrapbook") I decided to review some games from the computer vs computer Jerome Gambit tournament that Randy Tipton of Baltimore, Maryland, ran a while back, as mentioned on his blog HANGING PAWN :: Tip's Chess Blog, subtitled A Patzer's experiments with Engines and Unorthodox Chess Openings (see "We are not alone...").

HIARCS 11.1, Deep Shredder 10 and Rybka 2.3.1 participated in the event in which White won 239 (31%), drew 76 (10%), and lost 450 (59%).

Tipton made available the 239 games won by White – won by either Hiarcs 11.1 or Deep Shredder 10, as it turns out. Hiarcs 11.1 lost 7 games; Deep Shredder 10 lost 19 games; and Rybka 2.3.1 lost 223. ("Something fishy is going on here. Very unlike Rybka, maybe its book learning was off. ")

What can we learn from these encounters between master or grandmaster level engines? A little – but, surprisingly not a whole lot.

Here is a summary of the games, and the choices the silicon giants made.

1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.Bxf7+ Kxf7

As noted, coming out of this line there were 239 White wins, played by Hiarcs 11.1 or Deep Shredder 10.

5.Nxe5+

In the games won by White, the computers showed a clear preference for the "classical" second piece sacrifice, playing it in 231 games, 97% of the time.

5.0-0 Nf6 6.c3 was seen in 4 games; 5.c3 Nf6 6.d4 exd4 7.cxd4 Bb4+ 8.Bd2 Qe7 9.e5 Ne4 was seen in 2 games; 5.d4 exd4 6.Ng5+ Ke8 7.0-0 d6 8.c3 was seen in 1 game; and 5.d3 Nf6 6.Nc3 appeared in 1 game.

5...Nxe5 6.d4

Surprisingly, this was White's choice over 6.Qh5+ by a 3 to 1 margin (in the games that the first player won).

6.Qh5+ was seen in only 52 games, which amounts to only 23% of the games with 5.Nxe5+; or 22% of the games in the whole tournament.

For the record, all of the games with 6.Qh5+ continued 6...Ke6 7.f4 d6, since they all were played by Rybka 2.3.1. This line, returning a piece to reach a more settled game, is as old as D'Aumiller - A.P., Livorno 1878 (see "My Jerome Gambit Database").

6...Qh4 7.0-0 Nf6

This move, a "TN" as far as I can tell, was the overwhelming choice in the tournament, appearing in 166 games (all played by Rybka 2.3.1.)

7...Ng4 was seen in 11 games (see "Jerome Gambit Tournament: Chapter X"); 7...Qxe4 8.dxc5 Nf6 was seen in 1 game, transposing to the main line below; and 7...d6 was seen in 1 game

8.dxc5 (144 games)

8.dxe5 was seen in 22 games

8...Qxe4 9.Nc3

Alternately, 9.Re1 was seen in 3 games

9...Qb4

9...Qc6 showed up in1 game

10.Nd5
(118 games, all Deep Shredder 10 - Rybka 2.3.1 32-bit)

10.Be3 was seen in 23 games.

10...Qxc5 11.Nxf6 Kxf6 12.Be3 Qc4 13.Bd4 c5 14.Bxe5+ Kg6

So, here we have a main line that shows up in almost half of the games in the whole tournament. After some reflection, what can we conclude?

1) 7...Nf6 is a new way to return a piece in the 5.Nxe5+ Nxe5 6.d4 Qh4 line, although it is not as strong as the traditional 7...Ng4;

2) Black still had an advantage in the main line until the error 11...Kxf6, while the alternative 11...gxf6 would have maintained that advantage (as Deep Rybka 3.0 Aquarium has confirmed);

3) The overwhelmingly chosen line of play of the tournament – the main line, above – is largely the artifact of one program's (Rybka 2.3.1) predelection for an inferior line of defense (recall Steinitz's defense vs the Evans Gambit in his games against Chigorin);

4) Whatever enlightening bits of wisdom ("new and good" as it were) that the computers have uncovered about the Jerome Gambit must be hidden in the sidelines – or in the games that White managed to lose with later moves (and which are still unavailable from HANGING PAWN).

No comments:

Post a Comment