engines & positional play

[smirobth]

I am sure if you do a database search over CCRL or CEGT games, you will find MANY successfull sacrifices of a FULL exchange. Of course, you would need to review it from the human perspective, if they are to be considered tactically or positionally.

I would start with Quad-Rybka games and/or with the 40/120 downloads from CEGT, for this search.

Another engine which is capable of BIG longe-range sacrifices, is Junior (maybe older versions even more than the latest).

I don't deny that Junior and other programs will sac a full exchange. But it is my contention that they rarely (if ever) do so if a king attack is not involved. If it is so easy to find an example why hasn't anyone already done so in this thread? You say I could find one if I looked. Perhaps; but I am the one claiming they aren't so easy to find, so it seems to me the burden of proof for a counter example to my claim is on others.

[Graham Banks]

I am sure if you do a database search over CCRL or CEGT games, you will find MANY successfull sacrifices of a FULL exchange. Of course, you would need to review it from the human perspective, if they are to be considered tactically or positionally.

I would start with Quad-Rybka games and/or with the 40/120 downloads from CEGT, for this search.

Another engine which is capable of BIG longe-range sacrifices, is Junior (maybe older versions even more than the latest).

Yes, Robin could use the following tool to search databases specifically for full exchange sacs:
http://www.rbnn.com/cql/

Regards, Graham.

Thanks Graham. I haven't used this program before. I have chessbase, but it is not a simple task to find exchange sacs with chessbase. Can you give me an example of how to define a CQL query for an exchange sac? Their sac example shows only a queen sac, but an exchange sac is a lot harder to define since it involves a rook on one side and either a bishop or knight on the other (not just one piece).

I've not had a chance to take a look, but I'm told that is very specific, thorough and is easy to use.

Regards, Graham.

[Mike S.]

it is not a simple task to find exchange sacs with chessbase.

That query language thing doesn't look easy to me either. - With ChessBase, try the "difference" option in the material search. One side is up one minor piece, the other side has one rook more, over a reasonable number of moves (to make sure it isn't short-term tactics). That should work.

[Dann Corbit]

I looked for examples four or so years ago and couldn't find any.

I asked for some help on the Rybka forum and Vas replied:

Graham,

I definitely agree with your comment. For some ancient examples, take a look here:

http://www.rybkachess.com/index.php?aus ... hange+sacs
http://www.rybkachess.com/index.php?aus ... sacrifices

Newer Rybka versions are even much better at this.

Vas

Thanks Graham. However in the first link you provided every single exchange sac involved not only positional compensation, but also at least one pawn for additional compensation and sometimes two pawns (which is no sac at all). And the second link shows only pawn sacs, not exchange sacs. So these links are making my case for me .... programs don't seem to sac a full exchange for purely positional compensation.

I have seen some versions of Rybka make the right move here:
[d]r1bqk1nr/pppp1ppp/2n5/b7/2BpP3/2P2N2/P4PPP/RNBQ1RK1 b kq - bm Nge7;

Though most engines will toil for hours, scheming about capturing that lovely pawn.

This Evan's gambit position is an example of avoiding a pawn sac, not offering an exchange sac. Also in this position d6 is just as good as Nge7, so perhaps this would be better as an avoid move position. It is an nice position to test comps with though. Taking the pawn is very bad but not obvious.

Some very strong engines will grind on it for hours and hours and never see the problem.

A long time ago {perhaps 5-8 years ago}, I ran crafty against this problem for 50 hours (950 MHz computer) and it never found the solution. However, after you make the pawn capture, it will eventually see the problem with that. Also, after you make the move Nge7 it will eventually see that the position is good.

I have not tested modern crafty against this with modern hardware.

[Vladimir Xern]

Not quite sure if it suits your qualifications, but what about Junior's ...Rxc4! against Brutus?

[d] r1r5/3nbpk1/4qnp1/4p1Bp/1pN1P3/1PP2Q2/4N1PP/3R1R1K b - -

[Graham Banks]

I am the one claiming they aren't so easy to find, so it seems to me the burden of proof for a counter example to my claim is on others.

A good example has been posted in this thread on the Rybka forum:
http://rybkaforum.net/cgi-bin/rybkaforu ... id=2977#fp

[smirobth]

I am the one claiming they aren't so easy to find, so it seems to me the burden of proof for a counter example to my claim is on others.

A good example has been posted in this thread on the Rybka forum:
http://rybkaforum.net/cgi-bin/rybkaforu ... id=2977#fp

Hi Graham,
I went to the Rybka site and saw the same positions from Vas again (no full exchange sac), plus this single additional one:
[d]rq4kr/1bn2pp1/4p3/p1R1N2p/1p2n3/7P/PP1B1PP1/3QRBK1 w - -
with the continuation 26.Rxe4 Bxe4 27.Nxf7 Kxf7 leading to this position:
[d]rq5r/2n2kp1/4p3/p1R4p/1p2b3/7P/PP1B1PP1/3Q1BK1 w - -
Which is pretty obviously a king attack. I am talking about 100% positional (quiet) compensation. Like a destroyed pawn structure combined with a forever knight outpost on a dominating square from which it cannot ever be dislodged. Or sacs to allow a side in trouble to make a fortress draw. These types of purely positional sacs do occur in GM games. They don't seem to occur in computer games; unless you count exposed kings or pawns nearing promotion as positional factors (which could be argued).

So I am still standing by my claim.

[smirobth]

I have found some positions that I consider to be purely positional full exchange sacs, and that some programs have no trouble seeing, thus I stand corrected:

Van Wely,Loek - Topalov,Veselin
1.d4 d5 2.c4 c6 3.Nf3 Nf6 4.Nc3 a6 5.a4 e6 6.Bg5 h6 7.Bh4 dxc4 8.e3
[d]rnbqkb1r/1p3pp1/p1p1pn1p/8/P1pP3B/2N1PN2/1P3PPP/R2QKB1R b KQkq - 0 8
8...b5! 9.axb5 cxb5 10.Bxf6 gxf6 11.Nxb5 axb5 12.Rxa8 Bb4+ 13.Ke2 Bb7 14.Ra1 f5 15.Ne5 Rg8 16.f4 Nc6 17.Nf3 Na5 18.Kf2 Nb3 19.Ra7 Be4 20.Ra2 e5 21.fxe5 f4 22.Be2 fxe3+ 23.Kxe3 Qd5 24.g3 Nxd4 25.Nxd4 Bxh1 26.Bf3 Qxe5+ 27.Kf2 Bc5 28.Bxh1 Bxd4+ 29.Kf1 Rg5 30.Bf3 Kf8 31.Kg2 Qe3 32.Kh3 Kg7 33.b3 cxb3 34.Ra3 b4 35.Rxb3 Bc3 36.Qe2 Qc5 37.Qd3 Qc8+ 38.Kg2 Ra5 39.Qc2 Qe6 40.Qb1 Ra1 41.Qc2 Bd4 42.Bd1 Qe1 43.Bf3 Qf1# 0–1

Nikolic,Predrag - Mamedyarov,Shakhriyar
1.d4 Nf6 2.Nf3 g6 3.g3 Bg7 4.Bg2 0–0 5.0–0 d5 6.Nbd2 b6 7.c4 e6 8.cxd5 exd5 9.Nb3 Re8 10.Bf4 Ba6 11.Re1 Nh5 12.Bg5 Qd6 13.Rc1 h6 14.Be3 Nd7 15.Rc2
[d]r3r1k1/p1pn1pb1/bp1q2pp/3p3n/3P4/1N2BNP1/PPR1PPBP/3QR1K1 b - - 0 15
15...Rxe3!? 16.fxe3 Re8 17.Qc1 c5 18.dxc5 bxc5 19.Rd1 c4 20.Nbd4 Nhf6 21.Nd2 h5 22.Nf1 Bh6 23.Rc3 h4 24.gxh4 Ne4 25.Ra3 Ndf6 26.Nf3 c3 27.Qc2 cxb2 28.Rb3 Ng4 29.h3 Ngf2 30.Rd4 Bc8 31.Qxb2 Nxh3+ 32.Bxh3 Bxh3 33.Ng5 Bxg5 34.hxg5 Qe5 0–1

Ivanchuk,Vassily - Kramnik,Vladimir
1.e4 c5 2.Nf3 Nc6 3.d4 cxd4 4.Nxd4 Nf6 5.Nc3 d6 6.Bg5 e6 7.Qd2 a6 8.0–0–0 h6 9.Be3 Be7 10.f4 Nxd4 11.Bxd4 b5 12.Qe3 Qc7 13.e5 dxe5 14.Bxe5
[d]r1b1k2r/2q1bpp1/p3pn1p/1p2B3/5P2/2N1Q3/PPP3PP/2KR1B1R b kq - 0 14
14...Ng4! 15.Qf3 Nxe5 16.Qxa8 Nd7 17.g3 Nb6 18.Qf3 Bb7 19.Ne4 f5 20.Qh5+ Kf8 21.Nf2 Bf6 22.Bd3 Na4 23.Rhe1 Bxb2+ 24.Kb1 Bd5 25.Bxb5 Bxa2+ 26.Kxa2 axb5 27.Kb1 Qa5 28.Nd3 Ba3 29.Ka2 Nc3+ 30.Kb3 Nd5 31.Ka2 Bb4+ 32.Kb1 Bc3 0–1

[Dann Corbit]

A little over ten years ago, another great chess player got a surprise:
[d]r1r1q1k1/6p1/p2b1p1p/1p1PpP2/PPp5/2P4P/R1B2QP1/R5K1 w - - bm axb5; am Qb6;

The computer played axb5. Back in those days, none of the PC programs could find it. I bet that lots of them can find it now.

[smirobth]

A little over ten years ago, another great chess player got a surprise:
[d]r1r1q1k1/6p1/p2b1p1p/1p1PpP2/PPp5/2P4P/R1B2QP1/R5K1 w - - bm axb5; am Qb6;

The computer played axb5. Back in those days, none of the PC programs could find it. I bet that lots of them can find it now.

Hi Dann,
This famous position has been discussed a lot, and I think the consensus these days is that Qb6 is stronger than axb5. axb5 doesn't win against best defense (Kasparov resigned in a drawn position!), while Qb6 very likely might win by force.