Lone minor piece penalty - What did Larry mean?

[CRoberson]

In the 1990's, I implemented the lone minor piece penalty suggested by Larry Kaufman in his article published in Chess Life. I did this in NoonianChess and again in Telepath and Ares. However, I think I misinterpreted the article.

Basically, it said apply a penalty to the lone minor piece (different for bishop and knight), because it needs pieces as helpers to be effective.

What did he mean by pieces? Did he mean apply the penalty if I have a bishop and some pawns or did he mean completely lone bishop (no pieces and no pawns).

I interpreted it as lone means nothing else including pawns, but some test positions from a lose in Olivier's tournament make me think he meant apply the penalty disregarding the number of pawns.

[Ferdy]

In the 1990's, I implemented the lone minor piece penalty suggested by Larry Kaufman in his article published in Chess Life. I did this in NoonianChess and again in Telepath and Ares. However, I think I misinterpreted the article.

Basically, it said apply a penalty to the lone minor piece (different for bishop and knight), because it needs pieces as helpers to be effective.

What did he mean by pieces? Did he mean apply the penalty if I have a bishop and some pawns or did he mean completely lone bishop (no pieces and no pawns).

I interpreted it as lone means nothing else including pawns, but some test positions from a lose in Olivier's tournament make me think he meant apply the penalty disregarding the number of pawns.

Pieces would probably mean here queens, rooks, bishops, and knights.
This makes sense because with only 1 minor piece, it is difficult to create combinations and of course the mating potentials are reduced. I am talking positions with pawns .

[bob]

In the 1990's, I implemented the lone minor piece penalty suggested by Larry Kaufman in his article published in Chess Life. I did this in NoonianChess and again in Telepath and Ares. However, I think I misinterpreted the article.

Basically, it said apply a penalty to the lone minor piece (different for bishop and knight), because it needs pieces as helpers to be effective.

What did he mean by pieces? Did he mean apply the penalty if I have a bishop and some pawns or did he mean completely lone bishop (no pieces and no pawns).

I interpreted it as lone means nothing else including pawns, but some test positions from a lose in Olivier's tournament make me think he meant apply the penalty disregarding the number of pawns.

He meant single minor piece, regardless of pawns. I am not sure but many programs implement this in various (different) ways. It should be obvious that a knight vs 3 pawns is really bad if that knight is all you have. This would tend to make you not trade away your next-to-last piece. Or whatever. Crafty has it's well-known "bad trade" matrix that implements this in a way. In most cases this is probably irrelevant, because it should be symmetric, and if both sides have one piece + pawns, or two pieces + pawns, this score won't change a thing.

It really counts when one side has an extra piece, as now the other side might see that "penalty" and avoid this position (which must have been some sort of minor for pawns trade).

[CRoberson]

In the 1990's, I implemented the lone minor piece penalty suggested by Larry Kaufman in his article published in Chess Life. I did this in NoonianChess and again in Telepath and Ares. However, I think I misinterpreted the article.

Basically, it said apply a penalty to the lone minor piece (different for bishop and knight), because it needs pieces as helpers to be effective.

What did he mean by pieces? Did he mean apply the penalty if I have a bishop and some pawns or did he mean completely lone bishop (no pieces and no pawns).

I interpreted it as lone means nothing else including pawns, but some test positions from a lose in Olivier's tournament make me think he meant apply the penalty disregarding the number of pawns.

He meant single minor piece, regardless of pawns. I am not sure but many programs implement this in various (different) ways. It should be obvious that a knight vs 3 pawns is really bad if that knight is all you have. This would tend to make you not trade away your next-to-last piece. Or whatever. Crafty has it's well-known "bad trade" matrix that implements this in a way. In most cases this is probably irrelevant, because it should be symmetric, and if both sides have one piece + pawns, or two pieces + pawns, this score won't change a thing.

It really counts when one side has an extra piece, as now the other side might see that "penalty" and avoid this position (which must have been some sort of minor for pawns trade).

Thanks Bob.

Here is what I am trying to solve. It comes down to the value of a rook and a knight in a KRPP vs KNPP.

In the following position:
[D] 5rk1/p2n1pp1/1rnq3p/8/2QP4/P2BBP2/5P1P/R4RK1 b - - 0 20

Ares plays Nde5 which leads to a bad position and it does figure that out at a longer TC, but it takes 14 ply.

I've checked some other engines and the move is liked by many with 8 to 12 ply of search. I have been looking for endgame heuristic evals that would help it. It seems a reduction in the knight and/or a gain in the rook is needed. The game progresses to KRPPknpp and is lost from that point.

[D] 8/6p1/1n5p/2R5/5k2/5P2/5K1P/8 b - - 9 44

[kbhearn]

Not sure how you're getting the first position to turn into the second - after Nde5 dxe5 Nxe5 Qe4 white holds on to the extra piece which should be enough to discourage Nde5.

Regardless, if you wanted to identify the problem on the second diagram immediately it would have to be that the N is out of the play (if the N was closer to the kingside pawns a score around +1 would be entirely reasonable since sometimes a N can hold against a rook with pawns on one wing) - but the second diagram is also only a few ply away from one of the pawns falling which should escalate the score.

[CRoberson]

Not sure how you're getting the first position to turn into the second - after Nde5 dxe5 Nxe5 Qe4 white holds on to the extra piece which should be enough to discourage Nde5.

Regardless, if you wanted to identify the problem on the second diagram immediately it would have to be that the N is out of the play (if the N was closer to the kingside pawns a score around +1 would be entirely reasonable since sometimes a N can hold against a rook with pawns on one wing) - but the second diagram is also only a few ply away from one of the pawns falling which should escalate the score.

The game proceeded to that point. I think that I am over penalizing the isolated pawns which white has 4 to black's 1 after Nde5 dxe5 Nxe5 Qe4

[bob]

What am I overlooking? Nde5 throws material away? I'd play dxe5 instantly as a human.

[bob]

Not sure how you're getting the first position to turn into the second - after Nde5 dxe5 Nxe5 Qe4 white holds on to the extra piece which should be enough to discourage Nde5.

Regardless, if you wanted to identify the problem on the second diagram immediately it would have to be that the N is out of the play (if the N was closer to the kingside pawns a score around +1 would be entirely reasonable since sometimes a N can hold against a rook with pawns on one wing) - but the second diagram is also only a few ply away from one of the pawns falling which should escalate the score.

The game proceeded to that point. I think that I am over penalizing the isolated pawns which white has 4 to black's 1 after Nde5 dxe5 Nxe5 Qe4

For clarity, your program plays Nde5 on its own and throws the knight away?

I just ran this on crafty:

Code: Select all

               16     0.33     +1   2. dxe5!                         
               16     0.43   2.09   2. dxe5 Nxe5 3. Qe4 Nxd3 4. Bxb6 axb6
                                    5. Rfd1 Rd8 6. Rd2 f5 7. Qxf5 Qc5 8.
                                    Qe6+ Kf8 9. Rad1 Rd6 10. Qe4
               16->   0.51   2.09   2. dxe5 Nxe5 3. Qe4 Nxd3 4. Bxb6 axb6
                                    5. Rfd1 Rd8 6. Rd2 f5 7. Qxf5 Qc5 8.
                                    Qe6+ Kf8 9. Rad1 Rd6 10. Qe4
               17     0.69   2.29   2. dxe5 Nxe5 3. Qe4 Nxd3 4. Bxb6 axb6
                                    5. Rfd1 Rd8 6. Rd2 f5 7. Qxf5 Qc5 8.
                                    Qe6+ Kf8 9. Rad1 Rd6 10. Qe3 Rg6+ 11.
                                    Kh1
               17->   0.82   2.29   2. dxe5 Nxe5 3. Qe4 Nxd3 4. Bxb6 axb6
                                    5. Rfd1 Rd8 6. Rd2 f5 7. Qxf5 Qc5 8.
                                    Qe6+ Kf8 9. Rad1 Rd6 10. Qe3 Rg6+ 11.
                                    Kh1
               18     1.14   2.32   2. dxe5 Nxe5 3. Qe4 Nxd3 4. Bxb6 axb6
                                    5. Rfd1 Rd8 6. Rd2 f5 7. Qxf5 Qc5 8.
                                    Qe6+ Kf8 9. Rad1 Re8 10. Qg4 h5 11.
                                    Qd7
               18->   1.40   2.32   2. dxe5 Nxe5 3. Qe4 Nxd3 4. Bxb6 axb6
                                    5. Rfd1 Rd8 6. Rd2 f5 7. Qxf5 Qc5 8.
                                    Qe6+ Kf8 9. Rad1 Re8 10. Qg4 h5 11.
                                    Qd7
               19     2.11   2.66   2. dxe5 Nxe5 3. Qe4 Nxd3 4. Bxb6 axb6
                                    5. Rfd1 Rd8 6. Rd2 f5 7. Qxf5 Qc5 8.
                                    Qe6+ Kf8 9. Rad1 Re8 10. Qg4 h5 11.
                                    Qh4 Ne1 12. Qf4+ Kg8
               19->   2.63   2.66   2. dxe5 Nxe5 3. Qe4 Nxd3 4. Bxb6 axb6
                                    5. Rfd1 Rd8 6. Rd2 f5 7. Qxf5 Qc5 8.
                                    Qe6+ Kf8 9. Rad1 Re8 10. Qg4 h5 11.
                                    Qh4 Ne1 12. Qf4+ Kg8
               20     3.83   2.74   2. dxe5 Nxe5 3. Qe4 Nxd3 4. Bxb6 axb6
                                    5. Rfd1 Rd8 6. Rd2 f5 7. Qxf5 Qd5 8.
                                    Qxd5+ Rxd5 9. Rad1 Rg5+ 10. Kh1 Nf4
                                    11. Rd8+ Kh7 12. R8d6 Ra5
               20->   4.99   2.74   2. dxe5 Nxe5 3. Qe4 Nxd3 4. Bxb6 axb6
                                    5. Rfd1 Rd8 6. Rd2 f5 7. Qxf5 Qd5 8.
                                    Qxd5+ Rxd5 9. Rad1 Rg5+ 10. Kh1 Nf4
                                    11. Rd8+ Kh7 12. R8d6 Ra5
               21     7.41   2.87   2. dxe5 Nxe5 3. Qe4 Nxd3 4. Bxb6 axb6
                                    5. Rfd1 Rd8 6. Rd2 f5 7. Qxf5 Qd5 8.
                                    Qxd5+ Rxd5 9. Rad1 Rg5+ 10. Kh1 Nc5
                                    11. Rd6 b5 12. Rb1 Na4
               21->   9.80   2.87   2. dxe5 Nxe5 3. Qe4 Nxd3 4. Bxb6 axb6
                                    5. Rfd1 Rd8 6. Rd2 f5 7. Qxf5 Qd5 8.
                                    Qxd5+ Rxd5 9. Rad1 Rg5+ 10. Kh1 Nc5
                                    11. Rd6 b5 12. Rb1 Na4
               22    13.73   2.75   2. dxe5 Nxe5 3. Qe4 Nxd3 4. Bxb6 axb6
                                    5. Rfd1 Rd8 6. Rd2 f5 7. Qxf5 Qd5 8.
                                    Qxd5+ Rxd5 9. Rad1 Rg5+ 10. Kh1 Nf4
                                    11. Rd6 Ra5 12. R1d4 Ne2 13. Re4
               22->  20.94   2.75   2. dxe5 Nxe5 3. Qe4 Nxd3 4. Bxb6 axb6
                                    5. Rfd1 Rd8 6. Rd2 f5 7. Qxf5 Qd5 8.
                                    Qxd5+ Rxd5 9. Rad1 Rg5+ 10. Kh1 Nf4
                                    11. Rd6 Ra5 12. R1d4 Ne2 13. Re4 (s=2)
               23    30.99   2.80   2. dxe5 Nxe5 3. Qe4 Nxd3 4. Bxb6 axb6
                                    5. Rfd1 Rd8 6. Rd2 f5 7. Qxf5 Qd5 8.
                                    Qxd5+ Rxd5 9. Rad1 Rg5+ 10. Kh1 Nf4
                                    11. Rd8+ Kh7 12. R8d7 Re5 13. Rg1 Ne6
                                    14. Rgd1
               23->  41.79   2.80   2. dxe5 Nxe5 3. Qe4 Nxd3 4. Bxb6 axb6
                                    5. Rfd1 Rd8 6. Rd2 f5 7. Qxf5 Qd5 8.
                                    Qxd5+ Rxd5 9. Rad1 Rg5+ 10. Kh1 Nf4
                                    11. Rd8+ Kh7 12. R8d7 Re5 13. Rg1 Ne6
                                    14. Rgd1

[CRoberson]

You got it. It plays Nde5 expecting dxe5, Nxe5, Qe4, Nxd3, BxR, axb and score it as white is up by 0.82. I think I am over penalizing the white isolated pawns.

[kbhearn]

Ah it does take a while for that pinned knight (after Rd1 at the end of your stated moves) to become a dropped knight in that line - it's rather obvious that it'll happen, but no clue as to whether it's worth the effort to detect and penalise soft pins so as to discourage the line at an earlier depth.