27 September 2010

Some Truths Cannot Be Proven?

Continuing with Fischer - Geller, Bled 1961, the most interesting position for further analysis is shown in the following diagram. White's last move attacked the d-Pawn for a second time.

Bled 1961

(After 18.Bc1-f4)

Geller played 18...d5. Fischer gave the move a '?' and commented

Loses outright. In the post-mortem Tal tried to hold the game with 18...Rd8 19.Qe2 hxg4; but after 20.hxg4 Black is in virtual zugzwang (*). If 20...Qh7? 21.Bxd6+ wins.

Spending a half page on the same position, Kasparov also gave Geller's move a '?', but for different reasons. After 18...Rd8, he first dismissed Fischer's suggestion, 19.Qe2 hxg4 20.hxg4, with the move 20...f5, eventually arriving at a Queen endgame on move 40. He concluded that 'the win for White is problematic'. He then gave two 'more interesting continuations', 19.Qb3 and 19.g5. On the first move he concluded, 'Black retains drawing chances', and on the second, 'White should win'.

Kasparov's analysis in the Predecessors series, while always impressive, has been dogged by the criticism that it depends too much on the work of computers. I was reminded of this in a comment to my previous post in this series, Fischer Overestimates His Position.

Fischer saw the truth in this game. Dmitry Plisetsky and Kasparov's super computers only show that some truths cannot be proven.

These days you don't need a supercomputer to get an amazingly accurate tactical analysis. I submitted the position after 18.Bf4 Rd8 to two different engines. After Fischer's suggestion 19.Qe2, both found 20...f5 within a few seconds. Likewise, both engines quickly found Kasparov's suggestions 19.Qb3 and 19.g5, and ranked them well ahead of their third choice, 19.Qe2.

The problem with this sort of analysis is that it risks becoming obsolete as computers get faster and as chess playing software becomes stronger. In Rybka 1 - Fischer / Huebner / Kasparov 0, I already discussed one well known position where Kasparov missed the strongest move.

In the position under examination in this post, the engines improve on Kasparov's 19.Qb3 Qf6! 20.Qb6 hxg4 21.Rad1! (leading to 'Black retains drawing chances') with the zwischenzug 20.Qb4, when 20...c5 looks forced, but further weakens the d-Pawn. Only now does White play Kasparov's suggestion 21.Qb6 hxg4. In addition to 22.Rad1, White has 22.h4, stopping all counterplay on the h-file and leading to the win of the exchange. This looks stronger than the positions after 19.g5; we go from 'White should win' to White wins.

Refinements like these don't mean we should just dismiss the analysis in Predecessors. It means we should take the analysis with a grain of salt (like all chess analysis) and use Kasparov's investigations as a map for discovering interesting positions and as a starting point for further analyses. That's exactly what I intend to do with future posts in this series.


(*) I'm not sure what Fischer meant by 'virtual zugzwang'. Is it a position that is almost a zugzwang, but not quite?

No comments: