Stockfish, the Strong

Which engine is the strongest at chess960? According to Chessdom.com, it's Stockfish, as in Stockfish is TCEC double champion.

The TCEC Grand champion Stockfish, modification 260614, won another prestigious computer chess competition – the Fischer Random Chess (FRC) tournament, organized as a TCEC Season 6 Special Event. [...] The reigning champion Stockfish was convincing with 25/28 points, leaving the runner-up Houdini 4 full 3 points behind. Critter 1.6a took the bronze, collecting 17,5 points.

TCEC is, of course, the Thoresen Chess Engines Competition. In May, Chessdom posted an Interview with Martin Thoresen – organizer and director of TCEC, where we learn that

Some people refer to TCEC as the "unofficial Computer World Championship". The organization ICGA is hosting the official World Championship, but they have lost a lot of interest over the years – in particular because none of the strongest engines are participating there and they play very few games.

and that

I [Thoresen] was a part of the CCRL rating list prior to starting TCEC. I ran a lot of test games for them, but it really didn’t interest me as much after a while – I was more interested in organizing tournaments and broadcasting it for others to watch

I downloaded the PGN file containing the games from the event and will look at it in another post. These games will make a valuable addition to my own copy of the Chess Jungle's database; see Chess960 Database, Part II for more info.

Clinging to the Past

In my previous post, It's Not Unusual, I brought up GM Soltis's 'Chess to Enjoy'column in the current issue of Chess Life. The focus in that post was on combinations, but Soltis has more to say that is relevant to chess960. After discussing combinations, he moves to the value of studying the games of top players.

But the problem is that games played by today's top grandmasters have less to teach the aspiring student than do older games. Why? Because a typical game by elite grandmasters these days begins with 15-plus moves of computer-checked home analysis. What happens in those moves is often impenetrable if you're not already an expert in that opening.

He repeats the idea a few paragraphs later.

Something similar happens in grandmaster games in which the opening is highly analyzed, such as the Sveshnikov Variation of the Sicilian Defense, the Marshall Gambit in the Ruy Lopez, the main line of the Classical Variation of the King's Indian Defense, or one of several others. In each case, the game doesn't really start until move 20 or so. How much can you possibly learn from them?

And finally concludes that there is not much to learn from these games.

And you learn more when you see how a great player punishes mistakes. But today, when a Nakamura or a Carlsen wins, it is usually because their opponent made errors that are subtle. The mistakes are so below-the-surface that it's hard -- if not impossible -- to learn from them. In short, we have more and more games to study and yet fewer and fewer teachable moments. A well-known American grandmaster, who has given thousands of lessons, put it well. "Today's games," he told me, "are bad for your chess."

The same criticism, 'bad for your chess', is sometimes aimed at chess960. If studying modern games that use the traditional start position is considered bad for your chess, does the argument have any real weight against chess960?

Why doesn't GM Soltis just accept the obvious fact that computers have changed chess forever and adopt chess960? A look at Bookfinder.com reveals that he has published many dozens of chess books, the majority of them on the opening. The popularity of chess960 would severely diminish interest in those books. Is it any wonder that Soltis, and the rest of the chess publishing sector, will cling to traditional chess for as long as they can?

It's Not Unusual

It's hardly news when a GM starts complaining about the advanced state of opening theory, but how often do you hear the same lament about the middlegame? In the July 2014 issue of Chess Life, GM Andy Soltis does exactly that in his 'Chess to Enjoy'column, titled 'Too Much of a Good Thing'. The reason for his grievance is a set of five combinations, all featuring combinations that start with a Queen for Rook pseudosacrifice on d7. He says,

It's not unusual that all five games have remarkably similar features. Chess combinations have a way of being repeated. It is very rare to come upon a truly original combination. This is why studying great tactical battles of the past is so useful.

It's not unusual? Why should it be otherwise? Since every game in every tournament -- chess960 tournaments excepted, of course -- has the same pieces starting on the same squares, and since every first move of those pieces goes to the same small set of target squares, what do you expect? Combinations showing the Bishop sitting on b1 or the Knight on d6 don't occur very often.

The following diagram shows the six positions that GM Soltis chose for his monthly puzzle set, a regular feature of his column. The positions are all from John Grefe games of the 1970s.

Take a good look at the positions. Three games have a Bishop sitting on the c/f start square, and four have the piece on the long diagonal. Five games have a Knight sitting on a square that can be reached on a single move move from its traditional start square.

In chess960, where the pieces often occupy squares that are unusual in traditional chess, original combinations occur in the majority of games. When you limit the number of start squares for the pieces, you limit the number of positions that arise from sequences of logical moves.

A Better Pawn Method

A short formality at the beginning of every chess960 game is to choose the start position for that game. If you're playing in a tournament, the position is always given, but if you're playing for fun, how do you select a random position? Here is a new method that was flagged to me by its creators: The Pawn Method: An easy way to set up Fischer Random Chess positions.

The Pawn Method is a simpler, faster way of setting up a starting position for Fischer Random Chess (aka Chess960). Unlike other methods, it doesn't require dice, coins, or a complicated process. It has been proven to select one of the 960 possible starting positions with perfect randomness. Created by Robby Walker and Tim Suzman.

Since web pages come and go, I'll quote the gist of the method here.

On the bottom of each Pawn, write the following: [White Pawns] Q B B N N... [Black Pawns] L1 L2 L3 L4 R1 R2 R3 R4

That only needs to be done once. For each game,

Begin by randomly shuffling the White Pawns in their starting row. Look under each Pawn. If the Pawn says Q, B, or N, place a Queen, Bishop, or Knight in the corresponding starting square.

Did the two Bishops land on the same color square? Randomly select one Black Pawn and look at the bottom. The number (1-4) tells you which number square of the opposite color to move the bishop to. Move the Bishop to that new square, swapping it with the piece already there if there is one.

Fill the three empty squares with Rook - King - Rook, in that order.

A few years ago I mentioned a similar, simpler method in Chess960 Waits for No One (see the last quote in the post). Ichabod, a professional statistician, commented on the post to point out that it didn't produce an even distribution over the 960 start positions. How does the Walker / Suzman method fare? I asked Ichabod for his opinion and he replied,

There are 1680 the letters on the bottom of the White Pawns can be arranged. 960 of those will be valid positions, leaving 720 invalid positions. The Black Pawns can rearrange each invalid position into 8 other positions, or 5,760 ways total. 5,760 / 960 = 6. That means that if the 5,760 ways are evenly distributed among the 960 legal positions, then the method works. My attempts to write a Python program to confirm that it does this have failed (they're giving invalid results meaning there is an error in the program), so I can't confirm it. My lunch hour is up, so I don't have time to fix the program.

A day later he confirmed that the method worked.

I redid the Python program. The system works. Each possible 960 position has the same probability of being chosen.

While that's a good start for the 'The Pawn Method', what about the statement that 'it doesn't require dice, coins, or a complicated process'. While it doesn't require dice or coins, it does require a special chess set. Lacking that set, how do you generate a new start position? You will still need dice, coins, or yarrow sticks. As for whether or not it is a 'complicated process', that requires a practical field test where real people judge its simplicity.

Chess960 certainly needs a simple, effective method to generate a new start position. Should chess sets be distributed with a mechanism to do that? Perhaps the time has already arrived.