Is Black's compensation after 16. g3 really sufficient?

[Ovyron]

The recent flood of 3300+ engines who all play like Stockfish (and have ppl running off to do sim tests!) gets a bit boring imo .

Did I miss this recent flood? Or are you talking about direct Stockfish derivatives?

[mmt]

[d]rnb1r1k1/pp3ppp/2p5/3p4/3P1Pnq/P1NBPK2/1PQBN1P1/R4R2 w - - 0 16

The position is from the 'Man vs Machine 5th Game' played in 2003 Kasparov vs Deep Junior.


Direct Link: https://i.ibb.co/64hfWYw/2019-09-24-4-23-46.jpg

Calculating 1 billion+ nodes each after g3, LC0 likes Nh2+ and Qh2 about evenly and confirms that the sacrifice wasn't sound:

-1.28 Nh2+ Kf2 Ng4+ Ke1 Qh3 f5 Nd7 e4 Nh2 Rf4 Qg2 Qb1 Nf6 Kd1 Nh5 Kc2 Nf3 Rxf3 Qxf3 Qh1 dxe4 Bxe4 Qxh1 Rxh1 Nf6 Bd3 g6 Nf4 Bxf5 Bxf5 gxf5 Ng2 Rad8 Bg5 Kg7 Nh4 Re6 Nxf5+ Kg6 Bxf6 Kxf6 Rf1 h5 d5 cxd5 Nd4+ Kg6 Nxe6 fxe6 Kd3 e5 b4 Rd7 Rf8 Kg5 Rg8+ Kf6 a4 b6 Rh8 Kg5 Nd1 Rg7 Ne3 d4 Nc4 Kf6 Rh6+ Kg5 Re6 Kg4 Nd6 Rc7 Ne4 Kf5 Rh6 Kg4 Rg6+ Kf5 Rg5+ Ke6 Rxh5 Rc1 Rh6+ Kf5 Rf6+ Kg4 Rf7 Re1 Nd6 Re3+ Kc4 a5 b5 Ra3 Kd5 Rxa4 (N: 1201M, P: 9.93%, V: -0.273, Q: -0.437)
-1.29 Qh2 Rae1 g6 e4 Nd7 e5 Nb6 f5 Rxe5 Nf4 Nc4 Bc1 Bxf5 Qxh2 Nxh2+ Kg2 Nxf1 dxe5 Bxd3 Nxd3 Nfd2 Ne2 Re8 Nd4 Ne4 Nf3 h5 a4 b6 Bf4 c5 b4 cxb4 Nxb4 Nc3 Rc1 Ne2 Rb1 d4 Nd5 Kg7 Bg5 Nc3 Rc1 Nxd5 Rxc4 Ne3+ Bxe3 dxe3 Kf1 Re7 Ke2 g5 Re4 Kg6 Kxe3 Rc7 Kd3 Rd7+ Rd4 Rxd4+ Nxd4 h4 g4 h3 Nf3 a6 Ke4 b5 axb5 axb5 Kd5 b4 Kc4 f5 exf6 Kxf6 Kxb4 h2 Nxh2 Ke5 Kc5 Kf4 Kd4 Kg3 Ke5 Kxh2 Kf5 Kg2 Kxg5 Kg3 Kh5 Kf4 g5 Ke5 Kh6 Kf5 g6 Kf6 Kh7 Kg5 g7 Kh5 g8=Q Kh4 Kg6 Kg4 Qd8 Kf3 Kf5 Ke3 (N: 1166M, P: 22.91%, V: -0.469, Q: -0.439)

[jp]

Stockfish with a small change predicts 19.Nd1 and then 19...Nh2 even from move 11 after 2 minutes correctly predicting the lines after the Bishop sac as played in the match but deviating with 16.g3. So not so much 19.f5 from the diagram.

But the scores are not really going up enough to say that White would be winning.

In the position after 16. g3 Nh2+ 17. Kf2 Ng4+ 18. Ke1 Qh3,

[d]rnb1r1k1/pp3ppp/2p5/3p4/3P1Pn1/P1NBP1Pq/1PQBN3/R3KR2 w - -

(an older version of) SF sticks with 19. Nd1 and the eval isn't really increasing with depth:
+1.31 (depth 47), +1.37, +1.15, +1.17, +1.26 (depth 51).

Its PV at depth 51 is this:

[pgn][FEN "rnb1r1k1/pp3ppp/2p5/3p4/3P1Pn1/P1NBP1Pq/1PQBN3/R3KR2 w - - 5 9"] 9.Nd1 Nh2 10.Bxh7+ Kh8 11.Rf2 Qh5 12.Bd3 Nf3+ 13.Rxf3 Qxf3 14.Bb4 Nd7 15.Nec3 Qxg3+ 16.Qf2 Qxf2+ 17.Nxf2 a5 18.Bd6 Rxe3+ 19.Kd2 Re6 20.Bc7 b5 21.Ng4 Ra7 22.Rh1+ Kg8 23.Bh7+ Kf8 24.f5 Rxc7 25.fxe6 Nb6 26.Ne5 Bxe6 27.Bd3 Ke7 28.b3 Rc8 29.Ne2 Nd7 30.Nxd7 Bxd7 31.Rh7 Rg8 32.Ke3 Re8 33.Nf4 Bg4 34.Kf2 Kf8 35.Nh5 Bxh5 36.Rxh5 g6 37.Rh7 Kg8[/pgn]

In this line, Black has an endgame with three pawns for a minor piece.

[Nordlandia]

In that line black got three pawns for the piece, but more often than not; minor piece is better than three pawns. all other things being equal.

[jp]

Yes, after all the engines are giving an advantage of over 1.00 to White. The question is whether Black can draw. If Black is worse after 16. g3 but not losing, one can argue about whether the word "unsound" really applies to Black's play.

[Eelco de Groot]

Yes, after all the engines are giving an advantage of over 1.00 to White. The question is whether Black can draw. If Black is worse after 16. g3 but not losing, one can argue about whether the word "unsound" really applies to Black's play.

Hello Jean Paul, sorry, I missed your reply here! No sure still, but also have not changed mind.

If 16. g3 is winning then Kasparov missed it. I think he would have mentioned it if he found a win later, some players would rather forget it or at least not publish it but I think Kasparov would appreciate the truth more than feeling annoyed with himself having missed a win.

I think I let Bluefish analyze this too, at least I think this was Bluefish. The scores are after a long while positive for White but for me I am not convinced this advantage means a win, but I have not looked at the given lines or anything, I just forgot to stop the analysis

Position after 16. g3, after a few days:

[d]rnb1r1k1/pp3ppp/2p5/3p4/3P1Pnq/P1NBPKP1/1PQBN3/R4R2 b - -

Engine: SF20190925 MOD NoContempt (512 MB)
(Bluefish I think, one thread, Eval for White)
by T. Romstad, M. Costalba, J. Kiiski, G. Linscott

58 6830:12+1.56 16...Qh2 17.Rae1 g6 18.e4 Nd7 19.e5 Nb6
20.f5 Rxe5 21.Nf4 Nc4 22.Bc1 Bxf5
23.Qxh2 Nxh2+ 24.Kg2 Nxf1 25.dxe5 Bxd3
26.Nxd3 Nfd2 27.Nd1 b6 28.N1f2 h5
29.g4 Kg7 (753.843.242.628) 1839

58 6830:12+1.92 16...Nh2+ 17.Kf2 Ng4+ 18.Ke1 Qh3
19.f5 Nd7 20.e4 dxe4 21.Nxe4 Nb6
22.Bf4 Bxf5 23.Nd6 Bxd3 24.Qxd3 Re6
25.Kd2 Qh5 26.Rh1 Qg6 27.Qxg6 fxg6
28.Nxb7 h6 29.Nc5 Re7 (753.843.242.628) 1839

58 6830:12+5.42 16...Qh5 17.Rh1 Nxe3+ 18.Rxh5 Bg4+
19.Kf2 Nxc2 20.Rah1 Bxh5 21.Rxh5 g6
22.Rh1 Nxd4 23.Nxd4 Nd7 24.g4 Rac8
25.g5 f6 26.gxf6 Nxf6 27.f5 Re7
28.fxg6 hxg6 29.Bxg6 Rf8 (753.843.242.628) 1839

58 6830:12+7.88 16...Qh3 17.Rh1 Rxe3+ 18.Bxe3 Nh2+
19.Rxh2 Bg4+ 20.Kf2 Qxh2+ 21.Ke1 Nd7
22.Kd2 Re8 23.Qb1 Nb6 24.b3 Nc8
25.Qh1 Qxh1 26.Rxh1 h5 27.f5 f6
28.Nd1 Nd6 29.Nf2 Bxf5 (753.843.242.628) 1839

58 6830:12+9.24 16...Rxe3+ 17.Bxe3 Nh2+ 18.Kf2 Qh5
19.Ng1 Nd7 20.Qe2 Qxe2+ 21.Ngxe2 Nxf1
22.Kxf1 Nf6 23.f5 Bd7 24.Kf2 h6
25.Kf3 Nh7 26.Rh1 Ng5+ 27.Bxg5 hxg5
28.g4 Rd8 29.b4 Be8 (753.843.242.628) 1839

58 6830:12+9.82-- 16...Nxe3 17.gxh4 (753.843.242.628) 1839

best move: Qh4-h2 time: 6830:12.718 min n/s: 1.839.482 nodes: 753.843.242.628