engines now good at positional chess: example position

[ozziejoe]

there was a thread marveling at the positional understanding at current programs, and I just wanted to continue that.


IN the position below, Stockfish 2.2.2. advocates a B pawn sacrifice for white, giving white a kind-of Benko Gambit in reverse. Note that this is a positional sacrifice with no clear way to recover the pawn..Stockfish assesses this position as being a +1 advantage for white, despite being a full pawn down. Stockfish sees the advantage pretty early in the search

In contrast, Deep hiarcs 14 starts out thinking black has an advantage, and after a while sees the position as equal.

What is your favorite engine's assessment?


/d r3k2r/ppq2ppp/2n2nb1/4p3/3pP1P1/B2P1N1P/2P2PB1/R2Q1RK1 w kq - 1 15 /d


[Event "?"]
[Site "?"]
[Date "????.??.??"]
[Round "?"]
[White "agressive kings indian"]
[Black "?"]
[Result "*"]
[ECO "A07"]

1.Nf3 d5 2.g3 Bg4 3.Bg2 c5 4.d3 Nf6 5.Nbd2 Nc6 6.O-O e5 7.e4 d4 8.h3 Bh5 9.g4 Bg6 10.Nc4 Bd6 11.b4 cxb4 12.a3 bxa3 13.Nxd6+ Qxd6 14.Bxa3 Qc7 *

[Michel]

[d] r3k2r/ppq2ppp/2n2nb1/4p3/3pP1P1/B2P1N1P/2P2PB1/R2Q1RK1 w kq - 1 15 [/d]

[zamar]

- White is a pawn down
+ White has a bishop pair
+ Black rook in h8 is stuck
--------------------------------
SF thinks that position is around equal just by "looking at it".

When searching deeper it becomes clear that white can improve his position with c3, Qb3, Rfc1 and it's very difficult to find any good constructive moves for black.

But in the beginning it was really just the trapped rook penalty (for h8 rook) which saved the day.

[carldaman]

Based on the experience of seeing a lot of games, and running a lot of engine analysis, I'd say all of the top 5 (or so) engines understand positional compensation very well. No surprise there. Many engines, not just the top ones, can carry out great speculative sacrifices. Those 3000+ ratings translate into some pretty amazing chess.

Regards,
CL

[zullil]

there was a thread marveling at the positional understanding of current programs, and I just wanted to continue that.


In the position below, Stockfish 2.2.2. advocates a B pawn sacrifice for white, giving white a kind-of Benko Gambit in reverse. Note that this is a positional sacrifice with no clear way to recover the pawn..Stockfish assesses this position as being a +1 advantage for white, despite being a full pawn down. Stockfish sees the advantage pretty early in the search

In contrast, Deep hiarcs 14 starts out thinking black has an advantage, and after a while sees the position as equal.

What is your favorite engine's assessment?


[d] r3k2r/ppq2ppp/2n2nb1/4p3/3pP1P1/B2P1N1P/2P2PB1/R2Q1RK1 w kq - 1 15

[pgn]
[White "Stockfish 2.2.2"]
[Black "?"]
[Result "*"]
[ECO "A07"]

1.Nf3 d5 2.g3 Bg4 3.Bg2 c5 4.d3 Nf6 5.Nbd2 Nc6 6.O-O e5 7.e4 d4 8.h3 Bh5 9.g4 Bg6 10.Nc4 Bd6 11.b4 cxb4 12.a3 bxa3 13.Nxd6+ Qxd6 14.Bxa3 Qc7 * [/pgn]

Here's the position prior to the positional sacrifice:

[D]r2qk2r/pp3ppp/2nb1nb1/2p1p3/2NpP1P1/3P1N1P/PPP2PB1/R1BQ1RK1 w kq - 3 11

[ZirconiumX]

Durandal thinks 11.Nxd6+ is good.

[pgn]
[Event "?"]
[Site "?"]
[Date "????.??.??"]
[Round "?"]
[White "agressive kings indian"]
[Black "?"]
[Result "*"]
[ECO "A07"]

1.Nf3 d5 2.g3 Bg4 3.Bg2 c5 4.d3 Nf6 5.Nbd2 Nc6 6.O-O e5 7.e4 d4 8.h3 Bh5
9.g4 Bg6 10.Nc4 Bd6 11.b4
( { [Durandal] 12:+2.24} 11.Nxd6+ Qxd6 12.Nh4 O-O 13.f3 a5 14.Kh1
Kh8 15.Qe1 Nd7 16.Nxg6+ fxg6 17.c4 Rad8 )
11...cxb4 12.a3 bxa3 13.Nxd6+ Qxd6 14.Bxa3 Qc7 *

[/pgn]

Matthew:out

[zullil]

Here's the position prior to the positional sacrifice:

[D]r2qk2r/pp3ppp/2nb1nb1/2p1p3/2NpP1P1/3P1N1P/PPP2PB1/R1BQ1RK1 w kq - 3 11

Here Stockfish considers b4 but never evaluates it highly. It then switches back to c3.

Searching: r2qk2r/pp3ppp/2nb1nb1/2p1p3/2NpP1P1/3P1N1P/PPP2PB1/R1BQ1RK1 w kq - 3 11
infinite: 1 ponder: 0 time: 0 increment: 0 moves to go: 0

24 +0.24 00:50 338727K c3 O-O cxd4 cxd4 Bd2 Bc7 Nfxe5 Nxe5 Nxe5 Bxe5 f4 Bd6
f5 Nd7 fxg6 hxg6 Rc1 Qb6 Qc2 Ne5 Bf4 Rfe8 a3 Qb5 a4
Qa6
25 +0.16 01:29 613286K b4 cxb4 a3 b3 cxb3 Bc7 Ncxe5 Nxe5 Nxe5 Bxe5 f4 Bc7
f5 Rc8 Qe1 Nd7 fxg6 hxg6 Qb4 Ne5 Bf4 Qd7 Rad1 b6
Bxe5 Bxe5
26 +0.44 01:54 787945K b4 cxb4 a3 Bc7 axb4 a6 Nfxe5 Nxe5 Nxe5 Bxe5 f4 Bc7
e5 Nd7 f5 Nxe5 fxg6 hxg6 Bxb7 Rxh3 Qe1 Rb8 Bg2 Rh4
g5 Rg4 Rxa6 Kf8 Qf2 Kg8 Bd2 Bd6 Ra5 Rc8
27 +0.20 02:57 1250M b4 cxb4 a3 Bc7 axb4 a6 Ncxe5 Nxe5 Nxe5 Bxe5 f4 Bc7
f5 O-O fxg6 hxg6 Qf3 Rc8 Qf2 Qe7 b5 Qe5 Bf3 axb5 Ra7
Rb8 Bb2 Rfd8 Rfa1 b4 R7a4 Qc5 Qd2 Qe5 Rxb4
28 +0.36 05:21 2326M c3 O-O cxd4 cxd4 b4 a6 a4 Bc7 Bd2 Nd7 b5 Ne7 b6 Bb8
Nh4 Re8 Qb3 Nc5 Qa3 Ne6 a5 Nc6 Nf5 Qd7 Rfc1 f6 Ncd6
Bxd6 Qxd6 Rad8
29 +0.40 07:45 3393M c3 dxc3 bxc3 b5 Nxd6+ Qxd6 Be3 O-O Qb3 a6 Rfd1 Rfd8
Nh4 Rac8 Rac1 Ne7 a4 b4 a5
30 +0.40 15:48 7047M c3 O-O cxd4 cxd4 b4 a6 Nxd6 Qxd6 Qb3 Nd7 Bd2 a5 a3
a4 Qa2 Ne7 Rfc1 Rac8 Bg5 Nf6 Bxf6 gxf6 Qd2 Qe6 b5
Qb3 Rcb1 Qc3 Qd1 b6 h4 h5
31 +0.53 23:21 10426M c3 dxc3 bxc3 O-O Nh4 Bc7 Bg5 Rb8 Ne3 h6 Nxg6 fxg6
Bh4 g5 Bg3 Rf7 Rb1 b5 Qc2 Qd7 Rfd1 Rb7 Nd5 a6 a3 Bd8
Qa2 Qd6 a4 b4 Ne3 Qb8
32 +0.48 31:32 14111M c3 dxc3 bxc3 O-O Be3 b5 Nxd6 Qxd6 Nh4 Rfd8 Qb3 b4
Rac1 bxc3 Qxc3 Nd7 Bxc5 Nxc5 Qxc5 Nd4 Qxd6 Rxd6 Rfe1
Ne6 Bf1 f6 Rc3 Ra6 a3 Rb8 Rec1 Nd4 Rc8+ Rxc8 Rxc8+
Kf7 Rc7+
Nodes: 14382547472
Nodes/second: 7458342
Best move: c3
Ponder move: dxc3

[Eelco de Groot]

Here's the position prior to the positional sacrifice:

[D]r2qk2r/pp3ppp/2nb1nb1/2p1p3/2NpP1P1/3P1N1P/PPP2PB1/R1BQ1RK1 w kq - 3 11

Here Stockfish considers b4 but never evaluates it highly. It then switches back to c3.

Searching: r2qk2r/pp3ppp/2nb1nb1/2p1p3/2NpP1P1/3P1N1P/PPP2PB1/R1BQ1RK1 w kq - 3 11
infinite: 1 ponder: 0 time: 0 increment: 0 moves to go: 0

24 +0.24 00:50 338727K c3 O-O cxd4 cxd4 Bd2 Bc7 Nfxe5 Nxe5 Nxe5 Bxe5 f4 Bd6
f5 Nd7 fxg6 hxg6 Rc1 Qb6 Qc2 Ne5 Bf4 Rfe8 a3 Qb5 a4
Qa6
25 +0.16 01:29 613286K b4 cxb4 a3 b3 cxb3 Bc7 Ncxe5 Nxe5 Nxe5 Bxe5 f4 Bc7
f5 Rc8 Qe1 Nd7 fxg6 hxg6 Qb4 Ne5 Bf4 Qd7 Rad1 b6
Bxe5 Bxe5
26 +0.44 01:54 787945K b4 cxb4 a3 Bc7 axb4 a6 Nfxe5 Nxe5 Nxe5 Bxe5 f4 Bc7
e5 Nd7 f5 Nxe5 fxg6 hxg6 Bxb7 Rxh3 Qe1 Rb8 Bg2 Rh4
g5 Rg4 Rxa6 Kf8 Qf2 Kg8 Bd2 Bd6 Ra5 Rc8
27 +0.20 02:57 1250M b4 cxb4 a3 Bc7 axb4 a6 Ncxe5 Nxe5 Nxe5 Bxe5 f4 Bc7
f5 O-O fxg6 hxg6 Qf3 Rc8 Qf2 Qe7 b5 Qe5 Bf3 axb5 Ra7
Rb8 Bb2 Rfd8 Rfa1 b4 R7a4 Qc5 Qd2 Qe5 Rxb4
28 +0.36 05:21 2326M c3 O-O cxd4 cxd4 b4 a6 a4 Bc7 Bd2 Nd7 b5 Ne7 b6 Bb8
Nh4 Re8 Qb3 Nc5 Qa3 Ne6 a5 Nc6 Nf5 Qd7 Rfc1 f6 Ncd6
Bxd6 Qxd6 Rad8
29 +0.40 07:45 3393M c3 dxc3 bxc3 b5 Nxd6+ Qxd6 Be3 O-O Qb3 a6 Rfd1 Rfd8
Nh4 Rac8 Rac1 Ne7 a4 b4 a5
30 +0.40 15:48 7047M c3 O-O cxd4 cxd4 b4 a6 Nxd6 Qxd6 Qb3 Nd7 Bd2 a5 a3
a4 Qa2 Ne7 Rfc1 Rac8 Bg5 Nf6 Bxf6 gxf6 Qd2 Qe6 b5
Qb3 Rcb1 Qc3 Qd1 b6 h4 h5
31 +0.53 23:21 10426M c3 dxc3 bxc3 O-O Nh4 Bc7 Bg5 Rb8 Ne3 h6 Nxg6 fxg6
Bh4 g5 Bg3 Rf7 Rb1 b5 Qc2 Qd7 Rfd1 Rb7 Nd5 a6 a3 Bd8
Qa2 Qd6 a4 b4 Ne3 Qb8
32 +0.48 31:32 14111M c3 dxc3 bxc3 O-O Be3 b5 Nxd6 Qxd6 Nh4 Rfd8 Qb3 b4
Rac1 bxc3 Qxc3 Nd7 Bxc5 Nxc5 Qxc5 Nd4 Qxd6 Rxd6 Rfe1
Ne6 Bf1 f6 Rc3 Ra6 a3 Rb8 Rec1 Nd4 Rc8+ Rxc8 Rxc8+
Kf7 Rc7+
Nodes: 14382547472
Nodes/second: 7458342
Best move: c3
Ponder move: dxc3

It does not seem Stockfish can see it on its own, Rainbow Serpent likewise would only play 11. c3 and initially does not like b4 at all, enough compensation for the pawn but it will never clear c3. The positional sacrifice does appear to be sound though; after 12.a3 this pawn is poisoned but even if you don't take White still has compensation, Multi PV 4 best moves:

[D]r2qk2r/pp3ppp/2nb1nb1/4p3/1pNpP1P1/P2P1N1P/2P2PB1/R1BQ1RK1 b kq -

Engine: Rainbow Serpent Experimental build 92 (2.3.2 + 19) Q6700, 4 threads, 512 MB
by Tord Romstad, Marco Costalba and Joona Kiiski

24 13:42 -1.13 12...a6 13.Nxd6+ Qxd6 14.axb4 O-O
15.Bd2 Rfc8 16.Qb1 Nd8 17.Qa2 Ra7
18.Rfc1 Ne6 19.c3 Raa8 20.c4 b6
21.Nh4 Ra7 22.Re1 Qc7 23.Qa4 h6
24.Nxg6 fxg6 25.Rec1 Qf7 (1.497.846.432) 1820

24 13:42 -1.17 12...Rc8 13.axb4 a6 14.Nxd6+ Qxd6
15.Bd2 O-O 16.Qb1 Ne8 17.b5 axb5
18.Qxb5 Nc7 19.Qb6 Rfd8 20.Rfb1 Ne6
21.Rb5 Nc7 22.Rb2 Ne6 23.Nh4 Qd7
24.Nxg6 hxg6 25.g5 Ra8 (1.497.846.432) 1820

24 13:42 -1.37 12...b5 13.Nxd6+ Qxd6 14.axb4 O-O
15.Bd2 Rfc8 16.Nxe5 Nxe5 17.f4 Nfd7
18.fxe5 Nxe5 19.g5 Qe6 (1.497.846.432) 1820

24 13:42 -1.41 12...O-O 13.axb4 b5 14.Nxd6 Qxd6
15.Bd2 Rfc8 16.Nxe5 Nxe5 17.f4 Nfd7
18.fxe5 Nxe5 19.g5 Qe6 20.Bf4 a6
21.Bg3 Nc6 22.Be1 Ne5 23.h4 Qg4
24.Qxg4 Nxg4 25.Bh3 Ne3 (1.497.846.432) 1820

It would seem to be opening preparation? But very nice!

Eelco

[Kyodai]

1. Nf3 d5 2. g3 Bg4 3. Bg2 c5 4. d3 Nf6 5. Nbd2 Nc6 6. O-O e5 7. e4 d4 8. h3 Bh5 9. g4 Bg6 10. Nc4 Bd6 11. b4!


Chesswise this is a very interesting pawn sacrifice from white. Also
concidering the principal question: "What should white do with his extra
tempo?" (White is playing the King's Indian)

From what I can see this b4 looks both very logical and perfectly sound.
A very good example of how white can use his extra tempo in this
opening!

On the other hand I don't like black's Bd6 at all. This simply doesn't look
good. He must be much more cautious and flexible in this position - so
the natural move for black is 10.-Nd7. Again with a wellknown position but
white a tempo ahead.

Guess the structural basics make most GM:s rather comfortable with black's
position anyhow...

[Jouni]

Another example. In latest NIC magazine Timman wonders, that in this won position Houdini only evaluates +0,86

[D]5bkr/6p1/p4pBp/8/6P1/8/PpK4P/4R3 b - - 0 34

I wonder which version of Houdini he has (illegal maybe) when H3 soon evaluates

34...h5 35.gxh5 f5 36.Rf1 a5 37.a4 f4 38.Rxf4 Bd6 39.Rh4 Kf8 40.Re4 Be7 41.Kxb2 Rh6 42.Kb3 Rxg6 43.hxg6 Bd8 44.Re5 Bc7 45.Rf5+ Ke8 46.h3 Bd8 47.Rf7 Bf6 48.Kc4 Be5 49.Ra7 Kd8 50.Rxa5 Bc7 51.Rf5 Kc8 52.a5 Kb8 53.Rf7 Be5 54.a6 Ka8 55.Rb7 Bd4
+- (5.40) Depth: 24/63