The Bryntse Gambit

[Nordlandia]

If we use a null move, this is a possible line.

[pgn][Event "Rapid 30m+30s"] [Site "?"] [Date "????.??.??"] [Round "?"] [White "Komodo 13.01 64-bit"] [Black "Nordlandia"] [Result "*"] [PlyCount "17"] 1. f4 d5 2. Nf3 -- 3. e4 dxe4 4. Ng5 Nf6 5. Bc4 Bg4 6. Qxg4 Nxg4 7. Bxf7+ Kd7 8. Be6+ Kc6 9. Bxg4 * [/pgn]

Analysis by Komodo 13.01 64-bit:

9...Nd7 10.Ne6 Qe8
Black is better: -/+ (-0.79 ++) Depth: 36 00:05:13 4308MN, tb=598
(16.05.2019)

Analysis by Stockfish 150519 64 BMI2:

9...Nd7 10.d4 e6 11.Bxe6 b6 12.Nf7 Qe8 13.d5+ Kb7 14.Nxh8 Nc5 15.Nf7 Nxe6 16.dxe6 Qxe6 17.Ne5 g5 18.Na3 Bxa3 19.bxa3 Qd5 20.Be3 gxf4 21.Bxf4 Qa5+ 22.Ke2 Re8 23.h3 Rxe5 24.Bxe5 Qxe5 25.Ke3 Qg5+ 26.Kxe4 Qxg2+ 27.Kd3 Qg3+ 28.Kc4 Qxa3 29.c3 Qa4+ 30.Kd3 h5 31.h4 c6 32.Rae1 Qxa2 33.Re7+ Ka6 34.Rhe1 Qf2 35.Kc4 b5+ 36.Kb3 Qxh4
Black is clearly better: -/+ (-1.39 --) Depth: 33/45 00:01:01 834MN, tb=16
16.05.2019)

Or by playing the scandinavian:

[pgn][Event "Rapid 30m+30s"] [Site "?"] [Date "????.??.??"] [Round "?"] [White "Komodo 13.01 64-bit"] [Black "Nordlandia"] [Result "*"] [PlyCount "15"] [EventType "rapid"] 1. e4 d5 2. Nf3 dxe4 3. Ng5 Nf6 4. Bc4 Bg4 5. Bxf7+ Kd7 6. Qxg4+ Nxg4 7. Be6+ Kc6 8. Bxg4 * [/pgn]

[jp]

So the Bryntse Gambit is good for at least a draw. *If* white knows what he is doing, plays according to sound positional themes and understands how to take advantage of pawn and minor piece cooperation. The SF (and other AB engine) evaluations of the imbalanced lines are just mostly wrong and overoptimistic. Those lines are very very positional.

It seems premature to conclude so confidently that it's theoretically at least a draw. We'd need at least to demonstrate it against SF. Certainly Lc0 couldn't do that (but of course there's no reason to believe Lc0 as white "knows what it is doing, plays according to sound positional themes, etc.").

One more iteration. Now Bh3 is preferred:
-1.74 14. Bh3 Bxe5 ... 53. Kd2 Kxe6 (depth 61, 10:21:06)

the final position in the 14. Bh3 line is a draw. While SF says -1.74.
AB game rollouts to evaluate this gambit are not reliable.

I haven't seen it yet, but the very end of the line won't be reliable. It's way beyond 61 plies. What about at move 45 or 35?

Yes, well that's fine, but the line picked out by Stockfish, including the first move, is entirely dependent on these unreliable evaluations at the leaf nodes. That's not to say the first move is not the best move, it is to say that you can't rely on the score, and therefore can't assess the Gambit by some few monte carlo AB program rollouts.

I commented earlier that I thought there were many positions along the lines where the difference in computer evals between different moves was not much e.g. 0.2- 0.3. The Bh3/Bh5 fork was one example. (We don't see those unless we step along the lines of the computer.) That and your point suggest it's hard to pin down one PV.
Maybe we need a fairly wide early tree and then to play computer-computer games from each node in that tree.

[Dann Corbit]

I am analyzing 1000 positions with four different programs at fairly long (a few minutes per position) time control. I am using several machines so it should be ready by Monday. Caveat: I am using Arena so it might get decoupled.

[jp]

That'll be very interesting to see, Dann.

From what we've seen, there doesn't appear to be much difference between the AB & NN engines' evaluations, so if engines are misevaluating they are all doing so.

[AdminX]

UPDATE:

Development of the Bryntse Gambit
https://en.chessbase.com/post/speelman-agony-98

A few days ago I received a file in the drop box from the Polish Correspondence player, Senior IM Wladyslav Krol , with a dozen of his games in this line. So I'm continuing a look at the Bryntse this week with a selection of these with some light notes (all by me).

[jp]

Do you have a Premium account there, Ted? (Williams did an episode of his show on the article.)

[AdminX]

Do you have a Premium account there, Ted? (Williams did an episode of his show on the article.)

Yes I do, I will look for it once my current engine game is over. Do you know which episode?

Thanks for the heads up.

[jp]

I think 2019-05-06. Please let us know if he had anything interesting to add to what Speelman wrote.

[jp]

after ..15 h5, it is a very different game depending on Bf5 or Bh3
Bf5 appears the AB common move, and it looks to end up more or less the same place after a few more moves.
Bh3 is LC0 move

Though Dann's Shredder adds variation by playing Nc4 instead of 0-0 a move earlier, so maybe the place to start is one move earlier after Nc3 Bd6.

After 14. Nc3 Bd6, SF10 at low depth (depth=30) goes
(-1.76) 15. Nc4 Na6 16. Nxd6
(-1.87) 15. 0-0 h5 16. Bh3

That eval. difference may flip and get bigger at greater depths. But I think it's possible that in many positions the engines show very small differences in eval. between different moves, because they haven't got an idea for how to play either side. It's then harder to claim that there is one clear PV or even one main branching point.

In the new article linked by Ted above, in the last game, after 13.Nc4

An improvement on 13.0-0 in the previous game.

So he's endorsing what Dann's Shredder (and SF at low depth starting after ... Bd6) preferred. We should go back and analyse from that position.

[chrisw]

after ..15 h5, it is a very different game depending on Bf5 or Bh3
Bf5 appears the AB common move, and it looks to end up more or less the same place after a few more moves.
Bh3 is LC0 move

Though Dann's Shredder adds variation by playing Nc4 instead of 0-0 a move earlier, so maybe the place to start is one move earlier after Nc3 Bd6.

After 14. Nc3 Bd6, SF10 at low depth (depth=30) goes
(-1.76) 15. Nc4 Na6 16. Nxd6
(-1.87) 15. 0-0 h5 16. Bh3

That eval. difference may flip and get bigger at greater depths. But I think it's possible that in many positions the engines show very small differences in eval. between different moves, because they haven't got an idea for how to play either side. It's then harder to claim that there is one clear PV or even one main branching point.

In the new article linked by Ted above, in the last game, after 13.Nc4

An improvement on 13.0-0 in the previous game.

So he's endorsing what Dann's Shredder (and SF at low depth starting after ... Bd6) preferred. We should go back and analyse from that position.

yes, Nc4.
a good example of what I wrote a few posts ago, hang back and consolidate. in this game the two centre pawns (one passed) supported by the two minor pieces are a good match for the black queen, the black rooks are prevented from doing anything useful. Difficult to be much more than a draw for white though. The game also shows the worth of trading blacks minor pieces away, previous games the black knight was his strongest piece.
So, the heuristic rules are:
Don’t look for flashy attacks, don’t initially try advancing, consolidate pawn(s) and minor pieces.
Trade away black minor pieces, then black has no challenge available for pawn-minor piece fortress, save exchanging a rook.
Obviously, hamper black rook ability to enter the position. Lock the position as much as possible. This gambit more or less does that by itself.
Eventually use pivots built by centre pawns to advance rooks, when all pieces in the game, quite possibly some flashy tactics will emerge.