1.d4 d5 2.c4 e5 3.dxe5 d4 4.Nf3 Nc6 5.a3 g6 6.Bg5 Qd7 7.Nbd2 h6 8.Bh4 g5 9.Bg3 g4 10.Nh4 Qd8 11.Qb1 Be7 12.Nf5 Bg5 13.f4 gxf3 14.Nxf3 Bxf5 15.Qxf5 Nge7 16.Qe4 Qd7 17.Bf2 Qf5 18.Qxf5 Nxf5 19.g3 Ne3 20.h4 Nc2+ 21.Kd1 Ne3 22. Kc1 Be7 <23. Bxe3> (I forced shreeder to play this) 23...dxe3

click for larger view

sliding forward: 24. Kc2 Rd8 25. Bh3 Nd4+ 26. Nxd4 Rxd4 27. Rhd1 Rxc4+ 28. Kd3 (picking up Pe3?!) 28...Rc5 29. Rac1 Rg8 30. Rxc5 Bxc5 31. Rc1 Bb6 32. Bc8 Rxg3 33. Bxb7 Rg4 34. Bc6+ Ke7

click for larger view

Analysis by Stockfish 2.2.2 JA (36-ply, 13 hours): 1. ± (0.72): 19.Ne4 Qg6 20.Bd3 Nce7 21.Rg1 0-0-0 22.Bh4 Bxe4 23.Rxg6 Rxd3 24.Rg7 Rd2+>

25.Ke1 Bxb1 26.Kxd2 Nf5 27.Rxf7 Ngh6 28.Rf6 Be4 29.Bg5 Rd8+ 30.Ke2 Ng8 31.Rxf5 Bxf5 32.Bxd8 Kxd8

click for larger view

Everyone should watch this video from D. Monokroussos: http://www.chessvideos.tv/forum/vie...

There have been several methods translating endgame positions into tablebases. The most state-of-the-art method has been invented by Eugene Nalimov. Modern engines almost exclusively use tablebases of the Nalimov type. It must be noted that endgame tablebases (often referred to as EGTB's) are not a wonder weapon. This is because the size of EGTB's increases in an extreme manner with the pieces on the chessboard. Even endgames with only 4 pieces (including kings) occupy 30 MB space, and 5-man EGTB's require more than 7 gigabytes! To generate 6-man EGTB's would require several month's time (and you'll probably will not have the required disk space) even on most up-to-date computers. <To create 7-man tablebases is currently simply fantasy.> Since endgames often include more pieces than that, you can imagine that the use of EGTB's is rather limited.

Because of the size of EGTB's, it is common practice to use only subsets of, say, 5-man tablebases. You can install e.g. only EGTB's including rooks, since these are the most important ones. However, this can create problems for certain engines. Why? Consider that the engine has reached a position with 2 kings, one rook for each side, and a pawn. You expect that it is about to promote that pawn into queen. But it doesn't. Why? EGTB's tell him that the position is mate in, say, 14 moves. But after queening that pawn, he does not find the corresponding EGTB. Its limited-depth search returns a value of 10 or so (it can't see tha mate yet), which is worse than mate. So it decides it will be better not to promote the pawn. It makes useless moves with its rook and king, and the game is drawn by the 50-move or repetition rule. Some engines can overcome this problem, some can't. You should always check what the case is (for example, by setting up an appropriate position and see if the engine can promote the pawn). If your engine can't work properly with incomplete tablebases, you'd better set it to use only EGTB's with smaller number of pieces. Also, note that for an engine to use EGTB's, additional operative RAM space is required, in which it processes the EGTB's. Additional RAM may be reserved for storing parts of the EGTB's in use (this will increase search speed, because fewer disk accesses are required).

I've found a free Tablebase generator :)

Please see the links below.

http://kirill-kryukov.com/chess/dis...

http://www.mtu-media.com/finalgen/h...

Do you know this FinalGen software ?

I'm not generating any Rybka-based analysis. As you may know, I'm deliberately staying away from engines that others' use; Rybka, Stockfish, Houdini since I believe that using different engines gives us an opportunity to get different ideas and different lines, as well as eliminating any biases that a particular engine may have. Or at least identify them.

The only other person that I know is generating Rybka based analyses is <WinKing> who is using Rybka 3. And, believe me, I would know. For what it's worth, the latest CCRL 40/40 tournament results show Rybka 4.1's ELO rating (4 CPU) to be 3161 and ranked #2-3 with Stockfish 2.2.2 and Rybka 3's ELO rating (also 4 CPU) to be 3134 and ranked #4 tied with several other engines such as Critter 1.4, Houdini 1.5a/2.0c, and Vitruvius 1.11C (which I had never heard of before). So, if relative ELO ratings are meaningful as far as analyses are concerned, there isn't that much difference in playing strength between and the quality of their analyses should be comparable at comparable search depths.

In case anyone is interested, Vitruvius looks like a relatively new chess program from Italy and it's a Ribbolit/IvanHoe derivative. It's style of play is claimed to be "highly speculative" and "human like" with an ELO rating above 3100. So they seem to have a marketing team behind them. They claim that Vitruvius beat Rybka 4 in a 15-game, 15 minutes per game match by a score of 7 ½ - 4 ½ (+3 -0 =0) as well as "crushing" Fritz 13 in a 50-game match (no details provided – yet). The Rybka games are provided on their web page.

I have an an opinion, which I cannot prove, that Rybka uses a different set of heuristics which are more effective at very long time controls, and which are not as effective at the shorter time controls used for the tournaments that are performed to generate the ratings.

[+0.19] d=27 18.g3 Ng6 19.Bg2 Ngxe5 20.Nxe5 NXe5 21.Qxd4 Rd8 22.Qxa7 Bd2+ 23.Kf1 0-0 24.Qxb7 c6 25.Qb3 Nxc4 26.Qa4 Be3 27.Bf3 Nd2+ 28.Kg2 Nxf3

13.e4 Bxf5 14.exf5 Qe7 15.Qc2 (the mainline as given by RV and others. At this point most analyses are settled on 15...Nxe5, but here is my proposition:

15...h5!?

I am not sure if this has been discussed already, so forgive me if I am duplicating, but here follows the "forced line" I mentioned above:

16.h4 Bf6 17.e6 fxe6 18.0-0-0:

click for larger view

initially, I looked at 18...e5!? here, creating a great pawn centre with the follow-up plan of attacking the f5 pawn (an effective strategy in all these lines, N.b. we have already gained back the gambit pawn) however, on deeper analysis, it appears to be suboptimal and simply 0-0-0 is better. One of the reasons for this is so that we can use the e5 square for pieces. I also looked at 0-0 at this moment and subsequent moves, but it's unwise and for fairly obvious reasons. After giving up our light squared bishop, there is a keen sense of weakness on the light squares (especially the e4 square, which will become a good outpost) and white, indeed, has a decent edge in resultant positions - but we have play too. As mentioned, we can start targeting the f5 pawn quickly, and use the e5 square as a base for operations, following up with Qg7 and Be5/Ne5. We can also target the h4 pawn.

One continuation goes: 18...0-0-0 19.Bd3 Be5!? (Nh6 is tempting and the engines jump for it immediately but it's strategically incorrect) 20.Rde1 Qg7 bolstering e5 and prophylaxis against f3/f4 ideas) 21.fxe6 Nge7:

click for larger view

we can round up the e6 pawn before long, create pressure down the f file and make e5 our positional stronghold. White maintains an edge, as I said, but our now passed d pawn is a big asset and prevents white from simplifying to an endgame too quickly. We will have to take extra care regarding our h5 and g4 pawns, and we will be relieving white of the bishop pair via Bxg3 shortly, creating further targets on the kingside. It looks very interesting anyway, and we always have the option of playing e5 instead of pursuing this line, which is more solid but less dynamic. To give everyone some idea visually, we can expect something like this:

click for larger view

In these lines there is the added benefit that it will take white 3 moves to challenge our pawn centre properly thanks to our g4 pawn. There are many possibilities and ideas for both sides, so it will remain juicy and complex - looks like fun!

Guess it wasn't a quick note after all.

Cheers,

Only one slight problem: 15.f4! is the surprise white move that tanks the position and causes us grief. If it were not for this surprise, everything you say would be at the top of our interest list.

The problem as I see it is that while in game-like conditions we have a clear criteria as to which engine is better in head-to-head competition (i.e. which engine wins the majority of their games), we don't have an equally clear criteria for analysis-like conditions. The best criteria would seem to be which engine provides the most accurate evaluation of the position. But how do we go about figuring that out? None of us have GM playing strength or positional judgment to properly evaluate the engine's evaluation (if we did, what would we be doing playing in this game?). I suppose that we could get one (or more, why not?) grandmasters to comment on the engines' evaluations but the last time I checked Anand was busy preparing for a trivial event (the WC Championship challenge) and so far Carlsen has not returned my calls. :-) And even if we could get several grandmasters to pass judgment on various engines' evaluations the results would still have to be probabilistic; i.e. which engines perform the best evaluations of various positions most of the time so that we could assign a confidence factor to an engine's evaluation of a certain position. And, of course, some engines may be more accurate at evaluating closed positions, some may be more accurate at evaluating open positions, etc., further compounding the problem.

Rankings per CEGT 40/20 games, May-27-12, single core computers:

Rank Engine Name Elo
1. Houdini 1.5a x64 1CPU 3015
2. Houdini 2.0c x64 1CPU 3006
3. Komodo 4.0 x64 1CPU 2979
4. Critter 1.4 x64 1CPU 2978
5. Critter 1.2 x64 1CPU 2975
6. Stockfish 2.2.2 x64 1CPU 2973
7. Houdini 1.5a w32 1CPU 2972
8. Critter 1.4 w32 1CPU 2965
9. Deep Rybka 4.1 x64 1CPU 2962

Rankings per CCRL 40/40 games, May-28-12, single core 32-bit computers:

Rank Engine Name Elo
1. Houdini 2.0c 32-bit 3163
2. Komodo 4 64-bit 3152
3. Critter 1.4 64-bit 3139
4. Rybka 4.1 64-bit 3138
5. Stockfish 2.2.2 64-bit 3127

Rankings per CCRL 40/40 games, May-28-12, 4-core 64-bit computers:

Rank Engine Name Elo
1. Houdini 2.0c 64-bit 4CPU 3212
2 3 Rybka 4 64-bit 4CPU 3161
2 3 Stockfish 2.2.2 64-bit 4CPU 3161
4. Critter 1.4 64-bit 4CPU 3159
5. IvanHoe 9.46h 64-bit 4CPU 3128
6. Komodo 4 64-bit 3119

Rankings per CEGT 40/120 games, May-27-12, single core computers:

Rank Engine Name Elo
1. Houdini 1.5 x64 2962
2. Stockfish 2.2.2 x64 2956
3. Komodo 4.0 x64 2950
4. Critter 1.4 x64 2943
5. Rybka 4.0 x64 2940

So at the fastest time control (40/20) Rybka was rated "only" 9th while at the slowest time control (40/120) it was rated 5th. Of course, we all know the dangers and/or silliness of trying to wildly extrapolate the results from tournaments based on an average 30 seconds/move (40/20) and 3 minutes/move (4/120) to analyses averaging, what, 10 or more hours per move (without even considering your 19-day analysis run!). So again, as usual, I have no answers.

The latest AT spreadsheet has 5 analyses with 13.e4 Bxf5 14.exf5 Qe7 15.Qc2 h5 with an average eval (3 different engines, discounting different versions of the same engine) of [+0.47]. It also has 18 analyses with 13.e4 Bxf5 14.exf5 Qe7 15.f4 with an average eval (4 different engines, again discounting different versions of the same engine) of [+0.65]. So 15.f4 definitely presents Black with more problems than 15.Qc2. And, of course, after 15.f4 Black's bishop is attacked so we don't have time for 15...h5. Then, one of the analyses stops after 15...gxf3, two continue 15...gxf3 16.Nxf3 0-0-0 17.Be2 Be3, and the rest continue 15...gxf3 16.Nxf3 Nf6 17.Bd3 0-0-0.

I agree <RV>