Tuesday, January 5, 2016

Lomonosov 7 men tablebases

Thanks to Vladimir Makhnychev and Victor Zakharov there are since july 2012 the 7 men tablebases (tbs) on a computer of the state-university of Moscow. This university was founded by Mikhail Lomonosov, the founder of science in Russia and a scientist of many domains. Approximately 140 TB of data (500.000 billion unique positions - without mirroring or rotations) are necessary to house this, such capacity hosting is today only possible for the big boys.

Fortunately access is online available for the customers of Convekta (people having bought Houdini 5 Aquarium (Pro) or Chess Assistant 14 Pro). For non-customers it is sufficient to pay 20 USD. The reason is that this is the only place in the world where the tbs are stored. If 2 million players want to check an endgame, or worse, an engine is calculating with the tbs then the internet-traffic is unsustainable. So that explains the threshold.

But chessok claims that there is also free access for people with an Android-gsm. That is true, although I got occasionally "connection failed" messages. In this article I give a few examples of my own experiences with this app (on the Internet you find some more; if you click on above link on "shared" then you see what VIP members have shared online; this section also already contains some interesting positions).

Picture 1
In the first example (see picture 1) you see how 4 pawns can win from a rook (mate in 45). White starts with Kd3 (normal as pawn e3 is attacked). The menu-buttons "White" till "Play" can be swapped by the optimal route for the win, or by a summary of the moves with their distance till mate (or a conversion to an endgame with less pieces which also wins).

But one of the best features is the button below the black cross. You can use this to move the pieces on the board, but if you select the button and stay on the piece by keeping your finger on the touchscreen, then it shows all the other squares where that piece can stand with the evaluation of the position. In the picture 1 I have clicked on the pawn f5. You see that if the pawn is on f5 then white mates in 45 ("45" on field f5). If the pawn in on f4 then it is a draw "=", if it is on b5 then mate in 23 and on g7 it is mate in 1 (because white has the move and will play g8Q#).

Picture 2
You can do this for each piece, so let us test what will be the result for pawn e3 - see picture 2. Here we notice the value of the pawnchain: white can still easily lose if the e-pawn becomes a double f-pawn (except f6 as then there follows fxe7). Or he loses or he only makes a draw if it is an another file on the second or third row. Even on a6 or h6 it still is only a draw. Besides this endgame pops up more frequently than you would think. Kasparov once got it on the board against Ye see Olympiade of 2002; Jakovenko, Bareev, Karjakin, Naiditsch and Leko also encountered it at least once; Megabase 2016 gives more than 500 games of 4 pawns or more against a rook.

OK - another example which better illustrates the power of the tablebases. Honestly who is interested in K+R+R versus K+B+B+N? In the Megadatabase 2016 there is only 1 game with this material imbalance see Furlan-Sale (1996), in which quickly exchanges happened to draw. So the study of that endgame is not really practical. But before you can create the endgame of K+3p vs K+2p, you first need to create the tablebases of KRR-KBBN. The cute pawnendgames of K+3p vs K+2p are now finally available. I toyed a bit with a pawnendgame, put the kings on not so standard positions to get a long mate and the result can be seen in the pictures 3 and 4. It is mate in 24 and white wins the quickest with 1.h5. Now if you let Stockfish and Komodo analyze this position then they also solve this fast (Stockfish initially shows mate in 42 with 1.Kd5; even after 10 minutes it sticks with Kd5 (and mate in 29). But that is not the point, with this app you can see what happens if you put the piece on another square.

Pictures 3 and 4
Further you also have some "fun" buttons like the 4 buttons on the top-right (see below the arrow to the right and the arrow to the left). With those buttons can move all pieces to the right, left, up or down (if there is sufficient margin of course). That is not something which goes quickly in the Fritz or Chessbase interface, no?

The second picture above (picture 4) I already checked where the pawn c7 can stand, to hold the position for black. Rather astonishing is that black can keep the position with the pawn on d6, e7 and e6 (so you give white the most distanced pawn!). That is not something you would expect immediately... As you see from the picture I had some problems with the connection at that time but as I stored the position as a favorite on my gsm, it still remembered all my earlier actions. Nice feature.

Also something nice, which maybe can be expanded in the future (to 4-men tbs): all the 3-piece tbs are immediately stored on your gsm. So you can always check how it exactly was with K vs K+p or make a bet with your friends in how many moves it is mate with K vs K+Q or R. Another extra is the endgame K+Q vs K+R which is useful for training.

But probably the most useful part of the Lomonsov tablebases are likely the endgames with rook or other light piece and 2 versus 1 pawn. The richness of those endgames is enormous. Let us have a look at below position which does not pop up in the Megadatabase 2016 (I didn't check the database of Harold van der Heijden). I could represent this as my own study but I don't want to push it. In the world of endgame-compositions this is a sensitive topic: using such databases would not only be anymore about searching beauty but a composer would have enough with an introduction, to end the study with a tablebase which wins. A better use of the tablebases is to control studies with maximum 7 pieces as John Nunn already did 3 times: to write a book about such positions which explain the process in human language.

Picture 5
The position: white wins in 95 moves with Rf8-a8+ (the numbers give the win/ loss moves if the white king is on different squares.) Again we see here classic logic: white wins if he is in front of the black pawn and loses if he is behind the pawn. It becomes a draw if he is too far away (right side of the board). Personally I find the square f5 interesting. Why is there still a win on f5 but not on the squares e6, f6, g6 or g5? Why is there so big difference between d1 and e1? Is it because white can exchange the rooks and he is in the quadrant with his king? Here it is obvious that it matters who is moving first. If white is in check (by the rook or b-file) then the win is vanished and black holds a draw. All this can be learned without moving a piece in the position.

Pictures 6 and 7
Another example. Can a knight and bishop win against 3 pawns. In the position on the left we see that it's a draw with white to move as the king stands on a8. Only if the king is on a7 it is also a draw. As long the king doesn't block the bishop (pawn a2 is promoting so white must start with Bh2-e5) and it is not check with a pawn (that would be an illegal position but this knowledge is not in the tbs) then white wins. Also Kg5 is forbidden as after 1.Be5 there is 1....f6+ followed up with 2...fxe5. I didn't check why exactly only a8 and a7 are draws but I assume it is because otherwise the king is too late to neutralize the pawns on the other wing.

Still a quick check where the pawn g7 can stand and how this influences the result: again no surprise. The more the pawn advanced on the king-side, the better for black, the closer to the white king, the better for white.

Picture 8
We are almost done with our examples: if KNN wins against King and pawn, how about KNN vs Kppp? An example which I found very surprising is that the place of the knight on h1 matters a lot. Only a few squares give a win. It also shows how small the path is to a win. In practice I see white trying to win this endgame, but most likely pull the emergency break (sacrificing the knights) when the black pawns can't be stopped anymore.

Pictures 9 and 10

But the app offers also other advantages than just checking positions. First you can also have classes (see pictures on the right). They are arranged via theme and contain some nice exercises. It is nice also to have an indication of the estimated elo-level each time so you know how hard the problem is. This indicator is often reliable but sometimes they contain some small errors. The position White: Kb8,b5; black Ka8, Bb1, Bh6, Nc2, Ng5 - white plays and wins, is definitely easier than 2500 elo.

You can also store positions in your favorite folder (in pgn or epd) or even share, store in dropbox or send via mail. There is even a possibility to save the testposition in  the ChessKing training-program so you can still work with it later on your PC.

Conclusion: a very handy app (if you have access to the server), which surely improves the knowledge of endgames. Besides there is even already a manual for the app see Manual for "7-piece Chess Endgame Training". Personally I don't understand why the free consultation of the 7-men tbs is allowed via the app, but not via the website. Market logic I assume...

People already satisfied with the 6-men tbs, can consult online them on the site of Shredder or the one of ChessOK itself or just download them (like at kirill-kryukov).


No comments:

Post a Comment