Questions by Christopher Conkie and Michael Diosi .

Thank you for agreeing to give us this interview.


Question 1.

Hello Eiko. About one year ago you officially released your program Freezer. Can you sum up what Freezer is about ?

Freezer is a tool that allows you to analyze certain types of endgames without introducing errors. It lets you build complete endgame databases that are similar to the Nalimov tablebases.

Question 2.

Tablebases is a good point. On one hand there are the normal chess playing programs like Fritz and Shredder, and on the otherhand there are these databases. How does Freezer come into play?

The ordinary analysis in chess programs is based on evaluations that of course include approximations and errors. Endgame databases include the pure truth; the Nalimov tablebases for example tell you the exact number of moves to the mate position. As nice as the tablebases are, they are packed with one huge disadvantage: they tend to be huge in size. Although the 3-men tablebases do only take up about 30 MB, all 5-men tablebases sum up to approx. 7 GB, also being not a real problem anymore. But all the 6-men tablebases will take up about 1.5 TB which is just a bit too much for almost all chess players. This is where Freezer comes into play. Instead of generating databases that are huge and contain many positions that are of no interest for a given analysis scenario, it is now possible to build just parts of it. This makes it possible to solve endgames that were out of reach before.

Question 3.

So how is it possible to leave out some positions ?

Consider a classic endgame: King, rook, and pawn vs. king, knight, and pawn. The pawns shall be blocked for now, ie. on d4 and d5. The classical databases would consider all posisitions with these pieces, and they would contain those positions that cannot be reached from the special given position. We are talking about all placements of the two pawn, giving a number of 1560 possibilites. So what Freezer generates will be only a small fraction of a classical database. In this case you can get the same relevant information but only look at a tiny part of the whole thing.

Question 4.

And those restrictions are the heart of Freezer ?

Exactly. By restricting pieces to certain squares big problems can be cooked into small ones. Of course, you cannot consider arbitrary large endgame problems (chess would be boring then), but with certain types of positions you can get well ahead of the tablebases.

Question 5.

How many pieces are suitable for Freezer ? How long does one have to wait for the result ?

This is difficult to answer in general. The more blocked a position is, the more likely you can solve the position. The main problem is how to cope with the many positions that appear after capture moves or pawn promotions. You have to tell Freezer how to evaluate these positions (for example you could tell Freezer to look up the correct result in Nalimov tableses). We could solve many 6-men and 7-men endings, and in extreme cases we could handle more than 15 pieces, but those were more or less special chess problem positions. The main focus is on positions with six to eight pieces. But even if you don't manage to get an exact result, it may be possible to get good ideas on what aspects in what direction to go during normal analysis.

Question 6.

Well, chess is played forwards. From the beginning to the middle game, up to the endgame and checkmate or stalemate. Chess programs also search forward. Why does your program work backward ?

Indeed chess is being played forward, and in theory you could solve the game forward, like special mate solving programs do. But for the general endgame construction, the retrograd algorithms that work from back to front seem to work very well. Simply put, this works as follows. First, take all mate positions, than takeback one ply, giving you all the win-in-1 positions. Takeback another ply, and you get the candidates for the next loss or win that is one move deeper etc. You proceed in doing so until you cannot mark any more new positions as win or loss, and you are left with the draw positions. The problem with doing forward search is the unknown width and depth of the search tree; retrograde algorithms solve this issue.

Question 7.

The whole concept of Freezer is quite new. How was the idea born ?

The idea was developed together with Prof. Ingo Althöfer. After my study of computer science I looked for possibilites to write my Ph.D. thesis on a computer chess related topic. Searching for computer chess and science will lead you to Ingo Althöfer rather quickly, and after some talks we decided on the main direction of research. And one of the fruits of these thoughts is the program Freezer.

Question 8.

Now a question that probably many people have asked: Why is the program called "Freezer"?

The principle of restricting pieces is like freezing those pieces to the chessboard. The program icon also contains an ice block.

Question 9.

Can you show us a position that illustrates what Freezer is about?

I would like to show one type of position that is especially well suited for Freezer, the fortress. This example has been shown before, but it demonstrates the fundamentals of Freezer very well. We are talking about the following position:

 

8/6k1/6p1/8/q7/4R3/5PK1/8 b - - 0 1