Tuesday, March 03, 2009

Gelfand-Markowski Revisited, or: Who Do You Trust?

A month ago I posted this position from Gelfand-Markowski, 1998 over the label "What's the Best Move?":

White to Move

The reason I was interested in the position was that I thought it was a fine example of the kind of play that one can probably never learn from tactics practice of the "Seven Circles" variety.

Two intrepid readers ventured lines.

Anonymous offered: "...what about 1.e4-e5. If the rook retreats, then White wins d6. If 1...d6xe5, then White has gained e4 for his knight (while denying e5 to the opposing knight), and he could follow up with c4-c5, Bc4(+) and Nd6, which would be more than enough for a pawn. This relinquishes f5 to the enemy bishop, but that seems livable. That leaves 1...Nxe5 2.Ne4 R moves 3.Nxd6. Again, White might continue with c4-c5 and Bc4(+), which adds up to a clear plus, perhaps winning."

Grandmaster of Blogging Liquid Egg Product wrote "...I'd have played c5, but seems inferior to the above commenter. Envisioning something like:
1. c5 dxc5
2. Bc4+ K-moves
3. e5

Gallagher gives (from the above diagram), "22. e5!! dxe5 (Or 22... Nxe5 23. Ne4 Rf8 24. Nxd6 with a clear advantage to White) 23. Ne4 Rf7 24. c5 Kg7 25. Bc4 Re7 26. a4

26... Re8? (b6 offered more resistance)," and Black resigned after 27. Rf1 Re7 28. Rd2 b5 29. axb5 cxb5 30. Bd5 1-0. I wouldn't have resigned just yet, but okay...

All fine, right? "Gelfand's move opened up the game for all of White's forces to zoom into action and one slip by Black and it was over" is the story line. Anonymous seems to have gotten it right, while Donnie wanted to play the second best move.

And yet...just for fun I set a couple of engines (Gambit Fruit and Toga II) that came with ChessDB (free download) to work and from the second position above, after a good half-hour both engines give 26...a5(!) 27. b5 cxb5 28. cxb5 Nf6 29. Nxf6 Kxf6

as best, with a tiny edge for Black! One of the engines even thinks Donnie's 22. c5 is (slightly) best in the first game position!

So who do you trust, GM Gallagher, a computer, or none of the above? Was 22. e5 (!!) or (!?).

I almost always think that the questions are more interesting than the answers, and so it is here. What do you think?


Rocky said...

These kinds of positions reveal how unrefined my chess sense is.

I probably would have considered e5, but not c5.

I keep seeing Black's d6 pawn defended only once and I feel compelled to double up the rooks and take that lonely d6 pawn. I also see White's g3 & b4 pawns are undefended and that worries me. But I also take comfort that Black still has to develop his bishop and a8 rook and so I tend to think White has a slight advantage and perhaps by gobbling up a pawn and exchanging pieces with Black, White will eventually promote a pawn.

Do you have Fritz? I tend to trust Fritz with stuff like this.

chesstiger said...

I looked at both 1. c5 and 1. e5 and after some consideration my choice was 1. e5 because 1. c5 seems to just throw away a pawn after 1. c5 dxc5 2. bxc5 Nxc5 while 1. e5 might do the same but the pluspion will be a loose pawn that will have black's pieces occupied to defend it.

Anonymous said...

After 12+ minutes and 20 ply of brute force analysis, Rybka gives

1.e5 +0.12
1.c5 -0.05
1.Rd2 -0.08

Considering that engines tend to overweigh material, we should feel confident that White has ample compensation after 1.e5.

True, in the second diagram, Rybka prefers Black by -0.37. However, she considers most of White's earlier moves in that line to be sub-optimal as well.

But this is to put too much stock in computer evaluations. Ultimately I think Gallagher is right: Gelfand played 1.e5 and beat a grandmaster; after 1.Rd2, he'd be slogging out a draw.

If you go through Kasparov's wins, you'll see that he routinely goes down 0.50 pawns in order to create complications and generate practical winning chances (against humans).

For more proof of the fallibility of computers, ask your computer to evaluate a tabiya position from the Sicilian Najdorf or King's Indian Samisch.

tanc (happyhippo) said...


I would have played 1.e5 because it seems to make the most sense.

I guess I was looking for ways to make White's pieces active and sac'ing the pawn does just that as White can easily hop the Knight onto e4 if dxe5 and the attack down the d-file is incredibly strong.

my logic is that by tying Black down to the defence of the Knight and thus the defence of the Bishop (as White's Knight can easily hop onto d6 at the crucial moment), Black's a-Rook will never see the light of day. And Black cannot play a5 either to liberate that Rook.

Effectively, Black is playing with a whole Rook down.

Derek Slater said...

Great post - this is part of what draws us back to chess again and again.