Brause and the DDG

Introduction

Brause is a crafty chess program at Free Internet Chess Servers (FICS). Steffen A. Jakob added the DDG in Brause's opening book in September 1996, so it started to play 3.c4 against the French Defence, and has created hundreds of games.

Brause started to play at Internet Chess Club (ICC) in 1997. First running on 200 MHz Pentium 6, it soon reached the rating over 2600.

Results

The strength of crafty programs depends on the CPU power, and can be anything between 2200 and 2800. In September 1996, Brause had got 83 % wins, 5 % draws, and 12 % losses in the last 4000 matches. After 400 DDG games, Brause has scored 75 % wins, 10 % draws, and 15 % losses. The QGA games, where Black plays dxc4, are not counted in because they are not DDG-like.

During the first month, Brause lost three times to a program named Fuzzy (gnuchess 4.0pl77 on PowerPC 604 @ 150), whose rating was 2417. Another program, jos (parrot 1.0 on 170 MHz Sun Ultra 1), with rating 2373, has also won Brause more than once. The average rating of other winners and drawers has been 2000-2100. The weakest winner was rated 1940, and the weakest drawer 1920.

In any case, it is not easy to beat Brause. During the first month, it was beaten a few times in less than 40 moves; during the second month, more than 50 moves very usually required; and during the third month, it took over 60 moves by default. By far the shortest loss so far was against Bjerre (rating 2117), played in October 1996:

1.e4 e6 2.d4 d5 3.c4 Bb4+ 4.Nc3 dxe4 5.f3 Nf6 6.a3 Bxc3+ 7.bxc3 c5 8.Bg5 Qa5 9.Bd2 Nbd7 10.Qe2 b6 11.fxe4 Bb7 12.d5 O-O 13.Qd3 exd5 14.exd5 Rfe8+ 15.Ne2 Ne5 16.Qh3 Nxc4 17.Qd3 0-1

Fortunately, Brause does not play 6.a3?! anymore.

To counterbalance, here is the shortest and very unique win, against mega (rating 2321), played in January 1997:

1.e4 e6 2.d4 d5 3.c4 dxe4 4.Nc3 Bb4 5.Be3 Nf6 6.Qb3 Qe7 7. a3 Ba5?? 8.Qa4+ Nc6 9.b4 Bb6 10.c5 1-0

Opening book

Here are listed how often different variations were played by Brause's opponents. Note that after 4.Nc3 Nf6 5.f3, almost every other player chose 5...exf3. However, it should be noted that they most probably knew they were playing against a computer.

3.c4 (#473) 3...dxe4 (84 %) 3...dxc4 (5 %) 3...misc (11 %) 3.c4 dxe4 4.Nc3 (#378) 4...Nf6 (67 %) 4...misc (33 %) 3.c4 dxe4 4.Nc3 Nf6 5.f3 (#177) 5...exf3 (46 %) 5...Bb4 (34 %) 5...c5 (12 %) 5...misc (7 %)

Yobes has written almost completely own opening book for Brause, which contains a lot of really strange openings, some of them invented by himself. The DDG seems to fit in well!

All the programs that have played the DDG seem to prefer Nge2 to f3, so one of my first suggestions was that at least some f3 variations should be added in the opening book.

I had always thought that programs prefer middlegame tactics, but Brause seemed to be happy to exchange queens and go to endgame -- and usually win in spite of being a pawn or two down! Especially, when Black played c5, Brause played Be3 by default, which often led to exchange of queens. The "normal" continuation is d5, which often transposes to the QGD, Tarrasch Defence, Marshall Gambit.

Apparently most opponents are surprised by the opening, thus playing some inferior lines. For example, exf3 has been played too often voluntarily, without making White's life difficult with some waiting moves.

Learning mode

In November 1996, the new book learning feature was added to Brause. Therefore, it sometimes played bad book lines until it learned not to do that again.

It will be interesting to see how Brause will learn. There are some unique problems related to the DDG (and offbeat gambits in general) and ICS:

One cannot expect any "standard" moves, let alone lines, from the opponent simply because there are no such things in the DDG; in practice, everyone has to invent new moves and new plans.
After 3.c4, the position is no more balanced, so an objective plan might be to look for a way to equalize. I believe, however, that the right plan is to seek time and space advantage instead of material balance.
Players at ICS have wide range of levels. Playing some line against weak players most of the time, and only seldom against a strong one, does not give a desired result. Even if the total result of the games were 90 %, it could only mean that 90 % of the opponents were novices who blundered in a good position.

Improvements

Even though Brause keeps getting excellent results, it plays a couple of critical lines in which it is quite easy to bust. The main problem is, of course, that there are too many critical lines in the DDG: White cannot avoid them all if Black plays the best moves. Against experienced DDG opponents, the best try is to vary between variations, like sometimes defer the f3-move, and even re-establish 4.Nc3 Nf6 5.Nge2. I do not like to play it, but it seems to have worked fine in e-mail games.

Here are a few suggestions for improvements that could make it score even better.

3...dxe4 4.Nc3 Nc6 5.d5?! exd5 6.cxd5 has resulted some draws. I think that White should not play d5 unless Black plays e5 or c5. Therefore, I recommend 5.Be3.

4...Nf6 5.f3

A) 5...Bb4 is critical.

6.a3?! is simply bad in view of 6...Bxc3+ 7.bxc3 c5 (or 7...b6), giving Black very good game. I therefore recommend the following moves.
6.Qa4+ Nc6 7.a3 may lead to difficult game. 7...Bxc3+ 8.bxc3 is the standard line, where White's d4 is supported. Despite the half-open b-file, White often castles queenside! Not necessarily better, but worth trying is 7.Be3.
6.Be3 c5! is yet another problem. Unfortunately, I cannot think any better 6th moves for White. Other 5th moves could be tried every now and then, hoping for transposition to some better line. The main alternatives are 5.Be3 and 5.Nge2.

B) 5...c5 6.Be3?! cxd4 7.Qxd4 has been used several times to beat Brause. Exchanging queens just cannot be a good idea. I recommend 6.d5, which usually leads to the Keres Variation 6...exd5 7.cxd5 exf3 8.Nxf3 Bd6, and now Brause often plays 9.Bb5+ Bd7 10.Qe2+!?, happily exchanging queens. At least for humans, who are not so superior in endgames, 10.O-O looks stronger, giving some attacking chances.

C) 5...exf3 6.Nxf3 Bb4 7.Qa4+ is a strange line; e.g., 7.Bg5 looks better.

Moreover, there are some "flaws" that are much more hard to prepare. Brause often plays very anti-positional moves, like c5, leaving Black's knight a nice place on d5. Similarly, it may move the rook from the half-open f-file to "attack" b7, or to defend d4-pawn. Hardly any human player would play such moves.

In February 1997, Brause's opening looks pretty good; if there are strategical errors, they are too subtle for me to recognize. However, there is one strange idea Brause keeps repeating, namely playing d5 even though the d4-pawn was not threatened: 4.Nc3 Nf6 5.Be3 (5.Nge2 b6 6.Bg5 Bb7 7.d5) Nc6 (or 5...b6) 6.d5.

To me, d5 is suspicious in all of those positions. In particular, the very idea of Be3 is to support d4, so playing d5 does not look a good idea. Unfortunately, I do not know what to recommend, except that after 5.Be3 Nc6, White could easily play 6.f3, and if 6...Bb4, then 7.Qc2.

Too many games are drawn by move repetition. Maybe Brause objectively thought that its position was not better, but at the same forgot that it should continue because of its brilliant endgame technique. Exactly for the same reason, its strategy to simplify has been so successful: being a pawn down, it keeps exchanging pieces and even queens, and very often wins the endgame. However, Yobes noted that Brause's endgame is rather bad on 90 MHz Pentium, but 200 MHz makes it a monster.