The Gazette - August 2005
Edited by Peter Price
The views expressed in the Gazette do not necessarily reflect the policies or views of the BCA, nor those of the editor.
ANALYSING WITH THE COMPUTER - PART II
A reader’s note: You may find it handy to have 2 chess boards ready in view of the deep analysis again.
by Grand Master Rainer Knaack
Translated by Hans Cohn.
2 Interpreting the signals correctly.
In a contest between two players face-to-face it is not the better understanding of the game that counts in the final analysis, but the quality of the moves made. Usually this amounts to the same thing, but not always. We can draw a parallel with chess problems: the goal is to make the solver display a maximum of playing strength, a possibly precise indication of the assessment of the position is a secondary consideration. Therefore we have to learn to live with a few drawbacks. What matters is interpreting the messages on the screen correctly.
2.1 What do the numbers mean?
Measurements take place in pawn units, i.e. a pawn is worth one point. But since material considerations alone do not provide reasonable conclusions, positional factors must also be counted – by fractions of a point.
Bishop pair and doubled pawns are, for instance, positional factors frequently encountered. The addition of such points yields a value which may be expressed in figurine symbols. The limits are 0.20 for “White stands a little better” and 0.70 for “White stands better”. That is, of course, arbitrary and one should not take it too seriously. Above all, when a game develops a plan and carries it out purposefully, such values have little force at the outset of the plan. A recent game in the “Frankfurt Chess Classics” is a good example.
2.2 Dynamics and statics.
Diagram 6 “Fritz” – Adams; Black to move
1r3rk1; pb3pp1; 4pn2; 1p2N3; 8; 2P1P1P1; 1P4P1; 1BKR3R
Here I thought at first that something had gone radically wrong. But “Fritz” sees the position balanced which is probably correct.
24 …Rfc8 25 Rh2 g6? (after this Black has to confront several insoluble problems) 26 g4 Kg7 27 g5 Nd5 28 Rh6 Rc7 29 Bg6 fg6 30 Rf1 Black resigns
Better is 25 …b4 26 c4 after which White loses time defending the c Pawn.
A. Kf8 27 g4 g5 28 Rf1 Ke7 29 Rh6 Ng4! (the point of his 25th move) 30 Rf7 Kd6 31 Ng4 Rc4 32 Kd1 Rg4 33 Bf5 Rg2 34 Re6 with perpetual check as the logical outcome
B. g6 27 b3 Kg7 28 g4 Rc5 29 Nd7 Nd7 30 Rd7 with an unclear position.
In diagram 6 all White’s pieces displayed an impressive activity. That the assessment fluctuates round 0.0 is due to White’s poor Pawn structure. One can prove this by replacing the Pawn on g3 to f3. Analysing the position now, the computer immediately indicates values around 0.7 in White’s favour. The relationship between static material (Pawn structure) to dynamic (state of development, domination of squares) factors in diagram 6 is well represented by “Fritz”.
r1b2rk1; pppn1pbp; 5np1; 4p1B1; 2P1P3; 2N2N2; PP2BPPP; 2KR3R
This is a well known theoretical position from the exchange variation of the King’s Indian defence. White has no advantage; the static position even shows a slight plus for Black, but “Fritz” and other programs show a White plus of 0.6. The temporarily better development and the domination of the d-file are assessed as pluses for White. But practice shows that White cannot profit from this.
2.3 Indication of the relevant variation.
The humanly most plausible moves are sometimes not even indicated by the computer, e.g.
Diagram 8 Knaack – Illner, Germany 1993
5r1k; p3r3; bpR1Bqp1; 5p2; 8; 6PN; Pp1QPP2; 4R1K1
This position would have arisen if I had played 27 Rc6 instead of Rcd1. The computer in the two-variation mode shows 27 …Re6 as the only move after which White wins as I would also have seen. 28 Re6 Qe6 29 Qh6 Kg8 30 Ng5 Qe7 31 Qg6 Qg7 32 Qg7 Kg7 33 Ne6 etc.
But I feared 27 …Bb7 if then 28 Rd6 Black can get into the above variation: 28 …Re6 29 Re6 Qe6 30 Qh6 Kg8 31 Ng5 Qe7 32 Qg6 Qg7 33 Qg7 Kg7 34 Ne6 Kf7 35 Nf8 Be4! Black regains the Rook and despite a Pawn minus does not stand badly. All that I had seen, but not what awaited me on entry into the computer: after 27 …Bb7 the immediate reply is 28 Bf7!!
What do we learn from this example? As I played the game myself I was aware of 27 …Bb7 but if a third person were to analyse the game with computer he would miss the variation 27 …Bb7 28 Bf7 unless he made a point of always also analysing the humanly logical moves. I cannot stress often enough: in analysing with a computer you have to take the lead!
And another thing: moves like 28 Bf7 never get on to the screen because the programs naturally deviate beforehand. Unfortunately man is mostly all too stupid … should there be no more games of man against computer?
2.4 The horizon effect
Chess computers cannot calculate as far ahead as one would like; the limitation of the hardware is called horizon.
Diagram 9 Analytical diagram
2k4r; ppp2pp1; 3p4; 4p1Bp; 1bPP4; 4P2P; PP3P2; 3R2K1
White wins a piece with 1 a3 Ba5 2 b4 Bb6 3 c5. Programs have difficulties with such positions. They calculate 1 a3 f6 2 Bh4 g5 3 Bg3 h4 4 Bh2 and only now 4 …Ba5 already eight half-moves and the win of the black Bishop needs a few more, “pushing it beyond the horizon”.
When a program gives up material without sense, only to delay greater loss, it is called the horizon effect. If in diagram 9 the g Pawn were on g6 the program would still answer 1 a3 with 1 …f6 instead of a piece only a Pawn is lost at first. This peculiar horizon effect is not so blatant in the modern chess programs. More frequent is the practice of delaying an unmeetable threat by intervening moves.
Diagram 10 Portisch – Johannessen
r2q1rk1; 3n1p2; 4p2p; p1p1P1b1; 1p1P4; 3B1P1R; PPQ1KP2; 6R1
This position would have arisen if Black had played 20 …Bg5 instead of 20 …Ne5. After 21 Rg1 “Fritz” gives the following variations:
1. cd4 22 f4 Ne5 23 fe5 b3 24 Qd2 ba2 25 Rh6 f6 26 ef6 White stands a little better (0.22).
2. b3 22 Qc1 Kg7 23 f4 cd4 24 Rh6 Kh6 25 Rg5 Qg5 26 fg5 White stands a little better (0.47).
In variation 2. the pointless move 21 …b3 is given; after 22 ab3 Black has the same problems as before. 1. also gives …b3. Such “time moves” are a clear indication of the horizon effect. They generally indicate that the affected side is in trouble.
2.5 Indication of mate, zugzwang
What was said in the introduction goes here too: programs should play well – the exact number of moves leading to mate is not so important. Some programs merely inform that there is a forced mate in the position. “Fritz” gives the number of moves to the mate which it has just found; it does not look for another mate.
Zugzwang positions are a problem in themselves*. Programs like “Fritz” which work with the “nil move” have great problems with them. With a “nil move” the side whose turn it is to move makes one more move in the calculation of a variation (which is, of course, only possible in the calculating process); if the value of the variation does not numerically improve thereby the variation is probably not very promising. The “nil move”, above all, serves the purpose of avoiding unnecessary calculation. It makes programs more aggressive.
Diagram 11 Study by Reti Shakmaty 1928
4kr2; 5p1K; 3p1Q2; 1p4p1; 4P3; 1PP3b1; 8; 8
The start already causes the computer problems. 1 Kh6 Be5 2 Kg7 and only after 2 …Bf6 does “Fritz” see that 3 gf6 must win. But the actual “joke” comes only after 2 …Bg3. Leaving “Fritz 5.32” and “Nimzo 99” to work it out at the same time, we find that “Fritz” does not even see the draw by repetition of moves, while “Nimzo” sees the winning variation: 3 c4 bc4 4 e5!! Be5 5 bc4 and zugzwang.
In practice, zugzwang positions occur more often the less material is on the board. Correspondingly, “Fritz” then no longer uses the “nil move”, e.g. certainly not in Pawn endings. In our position, however, “Fritz” is caught on the wrong foot.
- The indicated assessment must always be questioned critically.
- You should remember that dynamic factors can quickly disappear.
- The indicated variation need not be the most logical from the human standpoint.
- The horizon effect leads to pointless move suggestions.
- When mate is shown, the indicated number of moves need not be correct.
- Zugzwang can confuse programs.
* Zugzwang is a situation in which any move by the side whose turn it is worsens its position.