A balanced approach to imbalances

[petero2]

I think you left out one of the most fundamental aspect of imbalances, which is Reinhard Scharnagle's 'elephantiasis effect'. The factors you take into account would have a very hard time explaining why 7 Knights are so much stronger than 3 Queens, while any reasonable set of piece values would predict exactly the opposite.

Do you know approximately how many extra pawns the queen side need to make this even? I tried giving the queen side two extra pawns, which didn't seem to help much:
[pgn]
[Event "Computer Chess Game"]
[Site "acer.localdomain"]
[Date "2013.10.26"]
[Round "-"]
[White "Komodo 5.1r2 64-bit"]
[Black "TQueeny 64-bit"]
[Result "0-1"]
[TimeControl "60/300"]
[FEN "nnnnknnn/1pppppp1/8/8/8/8/PPPPPPPP/Q2QK2Q w - - 0 1"]
[SetUp "1"]

{--------------
n n n n k n n n
. p p p p p p .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
P P P P P P P P
Q . . Q K . . Q
white to play
--------------}
1. c4 {+8.22/16 71577:57} c6 {+0.08/17 9} 2. b4 {+8.42/18 11} f6
{+0.14/19 0.1} 3. d4 {+8.32/18 15} Nhf7 {+0.33/18 8} 4. Qc3 {+8.29/17 6} e5
{+0.48/17 2.2} 5. Qg3 {+8.03/16 28} g6 {+0.36/20 0.1} 6. Kf1 {+8.12/15 18}
Nge7 {+0.72/16 8} 7. dxe5 {+7.77/16 42} fxe5 {+0.65/17 9} 8. h4
{+7.76/16 7} d6 {+0.84/18 0.7} 9. a4 {+7.99/16 10} Nbd7 {+1.04/17 8} 10.
Qgb3 {+7.78/15 14} Nc7 {+1.15/18 7} 11. h5 {+7.64/16 14} gxh5 {+1.18/18 7}
12. Qxh5 {+7.54/16 7} d5 {+1.22/18 0.1} 13. c5 {+7.32/16 8} d4
{+1.48/18 0.1} 14. g3 {+6.92/15 9} Ned5 {+1.88/18 7} 15. g4 {+5.37/14 9}
Nde6 {+2.51/17 7} 16. Qc1 {+5.30/15 6} Nce7 {+2.74/17 7} 17. a5
{+4.65/14 6} e4 {+3.20/17 25} 18. Kg1 {+4.67/14 10} Ne5 {+3.38/17 6} 19. a6
{+3.31/18 21} Nxa6 {+3.94/17 6} 20. Qg3 {+3.08/14 6} N5g6 {+4.14/17 18} 21.
Qh1 {+3.29/16 5} Nc3 {+4.33/17 8} 22. Kf1 {+2.59/21 3} Ned5 {+3.97/17 2.1}
23. e3 {+2.60/14 1.9} d3 {+4.74/18 13} 24. Qb2 {+1.02/15 3} Nge7
{+4.80/17 1.4} 25. Kg2 {+3.26/15 5} Nb5 {+4.87/18 16} 26. Qh5
{+3.56/12 1.3} Ng5 {+4.31/17 4} 27. Qge5 {+2.75/13 1.2} Nfe6 {+4.90/18 3}
28. Qc1 {+1.65/17 1.6} Nbc7 {+5.20/18 15} 29. Qeb2 {+2.58/12 1.3} Naxb4
{+5.65/18 4} 30. Qba1 {+1.77/15 5} Nba6 {+6.27/19 4} 31. Kh1 {+1.08/17 0.1}
Kd7 {+6.85/18 4} 32. Qab2 {+0.90/18 0.1} Naxc5 {+7.20/19 4} 33. Qd1
{+0.33/17 0.1} Nd6 {+8.33/19 4} 34. Qhh8 {-0.55/15 2.4} Nf3 {+9.19/18 4}
35. Qh7 {-1.18/16 0.1} b5 {+9.33/19 4} 36. Qh3 {-1.75/14 1.1} b4
{+9.93/19 2.7} 37. g5 {-2.51/15 1.5} Nc4 {+10.01/17 6} 38. Qba1
{-3.26/15 0.7} Nc3 {+11.57/18 4} 39. Qdf1 {-3.90/19 0.1} Ncd2 {+11.45/18 3}
40. Qfc1 {-3.84/16 0.6} N7b5 {+12.36/19 3} 41. Qcb2 {-4.17/14 0.6} Ndb3
{+12.21/19 3} 42. Qaf1 {-5.37/17 0.1} d2 {+12.60/18 3} 43. Qh7
{-5.76/19 0.1} Nexg5 {+13.59/18 6} 44. Qg7 {-5.94/19 0.1} Nge6
{+13.99/18 3} 45. Qh7 {-6.09/15 0.6} Kd6 {+14.74/18 2.5} 46. Qc2
{-5.61/13 0.3} Na3 {+14.22/18 7} 47. Qb2 {-5.99/21 0.1} d1=Q
{+14.40/17 2.5} 48. Qxd1+ {-6.44/15 1.0} Nxd1 {+14.71/19 1.5} 49. Qe2
{-6.87/15 0.4} Nc3 {+16.44/18 2.5} 50. Qf1 {-8.77/19 0.1} Nbd2
{+16.90/17 2.3} 51. Qfh3 {-9.52/18 0.1} b3 {+19.70/19 2.4} 52. Qg3+
{-7.63/14 0.5} Kd5 {+22.38/20 2.3} 53. Qh5+ {-10.55/15 1.7} Kc4
{+24.16/21 0.5} 54. Kg2 {-11.58/17 0.3} b2 {+23.88/18 2.1} 55. Qb8
{-13.64/17 0.1} Nab5 {+26.83/18 2.4} 56. Qbh8 {-15.80/17 0.1} Ng6
{+31.32/17 2.2} 57. Qh1 {-18.71/13 0.3} Nxh8 {+41.87/17 1.8} 58. Kg3
{-99.78/18 0.1} Ne2+ {+99.73/15 4} 59. Kg4 {-99.80/12 0.1} b1=Q
{+99.75/7 0.1} 60. Qg2 {-99.88/14 0.2} Nf7 {+99.85/9 0.1} 61. Kh5
{-99.90/16 1.2} Qb2 {+99.91/10 8} 62. Qg3 {-99.92/15 1.2} Nxg3+
{+99.93/8 0.6} 63. Kg4 {-99.94/17 2.6} Qh8 {+99.95/6 0.1} 64. fxg3
{-99.96/16 2.4} Qh6 {+99.97/4 0.1} 65. Kf5 {-99.98/17 2.5} Qg5#
{+99.99/2 0.1}
{Xboard adjudication: Checkmate} 0-1
[/pgn]
TQueeny is my own engine texel, modified to understand this material imbalance, based on what you wrote about how QueeNy works.

I also let TQueeny play both the queen and the knight side, but the queen side lost anyway:
[pgn]
[Event "Computer Chess Game"]
[Site "acer.localdomain"]
[Date "2013.10.26"]
[Round "-"]
[White "TQueeny 64-bit"]
[Black "TQueeny 64-bit"]
[Result "0-1"]
[TimeControl "60/300"]
[FEN "nnnnknnn/1pppppp1/8/8/8/8/PPPPPPPP/Q2QK2Q w - - 0 1"]
[SetUp "1"]

{--------------
n n n n k n n n
. p p p p p p .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
P P P P P P P P
Q . . Q K . . Q
white to play
--------------}
1. b4 {-0.09/18 71577:56} f6 {+0.14/18 9} 2. d4 {-0.31/17 1.3} Nhf7
{+0.14/18 8} 3. c4 {-0.37/17 8} e5 {+0.30/17 9} 4. Qc3 {-0.43/17 0.3} c6
{+0.45/17 8} 5. Qg3 {-0.43/17 0.5} g6 {+0.43/18 8} 6. dxe5 {-0.45/17 0.4}
fxe5 {+0.51/18 7} 7. h4 {-0.63/17 0.7} Nf6 {+0.59/18 7} 8. f4
{-0.59/17 1.5} d6 {+0.62/19 11} 9. fxe5 {-0.54/19 0.1} dxe5 {+0.68/20 7}
10. Qf3 {-0.72/19 9} Nbd7 {+0.78/18 7} 11. g4 {-0.66/19 1.2} Nc7
{+0.78/17 5} 12. Qhf1 {-0.67/17 2.6} Ng8 {+0.92/17 4} 13. e3 {-0.79/16 9}
c5 {+1.01/16 8} 14. bxc5 {-0.93/17 7} Nce6 {+1.17/18 6} 15. c6
{-1.24/17 1.3} bxc6 {+1.07/19 4} 16. Q3g2 {-1.12/17 8} Nec5 {+1.41/16 6}
17. a4 {-1.28/17 2.3} e4 {+1.51/17 3} 18. a5 {-1.38/16 7} Nfe5 {+1.47/16 6}
19. h5 {-1.55/15 8} g5 {+1.89/17 0.1} 20. a6 {-1.89/16 7} Nf3+ {+1.90/16 4}
21. Ke2 {-2.33/20 18} Nxa6 {+2.03/20 6} 22. Qa1 {-2.28/19 0.9} Nac5
{+2.07/19 5} 23. Qg3 {-2.35/16 5} Ngf6 {+2.25/17 7} 24. Qfb1 {-2.38/17 0.1}
Ne7 {+2.45/17 5} 25. Qbb2 {-2.71/17 10} Kf7 {+2.66/17 5} 26. h6
{-2.74/16 4} Nd3 {+3.07/16 6} 27. Qc3 {-2.80/17 7} Nd3e5 {+3.10/18 0.1} 28.
Qcxe5 {-3.01/17 13} Ndxe5 {+4.28/20 5} 29. Qaxe5 {-3.82/20 1.0} Nxe5
{+4.58/23 5} 30. Qxe5 {-4.72/23 1.1} Nde6 {+4.90/21 4} 31. Kf1
{-4.97/20 2.5} c5 {+5.66/23 6} 32. Ke2 {-5.78/22 6} Neg8 {+6.61/22 5} 33.
h7 {-6.19/26 1.3} N8xh7 {+7.03/25 4} 34. Qh2 {-7.25/25 5} Nhf8
{+7.89/25 0.4} 35. Kd1 {-7.81/25 9} N8d7 {+8.35/24 6} 36. Qd2
{-8.69/24 1.0} Ng7 {+9.03/24 5} 37. Ke1 {-9.73/26 8} Nh6 {+9.70/24 5} 38.
Kf1 {-10.33/25 2.6} Nhxg4 {+11.18/25 2.6} 39. Kg1 {-11.40/21 9} Nf5
{+12.63/23 11} 40. Qc3 {-14.52/23 17} Kg6 {+10.90/21 0.5} 41. Kh1
{-16.49/21 10} Nde5 {+19.86/20 5} 42. Qa1 {-21.81/20 2.2} Nf2+
{+27.71/20 2.7} 43. Kg1 {-26.30/20 7} Nfd3 {+30.49/20 0.1} 44. Kf1
{-30.01/19 20} Nfg4 {+31.42/19 0.9} 45. Qa6+ {-30.80/18 3} Kh5
{+38.84/19 2.1} 46. Qa1 {-40.21/18 5} Ngxe3+ {+99.63/18 0.6} 47. Ke2
{-39.92/18 4} Nf3 {+99.65/14 17} 48. Qh8+ {-99.82/16 16} Kg4 {+99.83/16 5}
49. Qh3+ {-99.84/16 5} Kf4 {+99.85/16 0.1} 50. Qxf5+ {-99.86/15 1.4} Nxf5
{+99.87/14 0.1} 51. Kd1 {-99.88/13 0.1} e3 {+99.89/12 0.1} 52. Ke2
{-99.90/11 0.2} Nf2 {+99.91/10 0.1} 53. Kf1 {-99.92/9 0.1} Ng3+
{+99.93/8 0.1} 54. Kg2 {-99.94/7 0.1} e2 {+99.95/6 0.1} 55. Kxf2
{-99.96/5 0.1} e1=Q+ {+99.97/4 0.1} 56. Kg2 {-99.98/3 0.1} Qf1#
{+99.99/2 0.1}
{Xboard adjudication: Checkmate} 0-1
[/pgn]

[Lyudmil Tsvetkov]

Another game against Rybka, and another draw. I think white has the advantage, but short on time difficult to seek winning continuations, so settling for the draw.

[pgn][Event "Computer chess game"]
[Site "OWNER-PC"]
[Date "2013.10.26"]
[Round "?"]
[White "owner"]
[Black "Rybka 2.2 64 bit"]
[Result "1/2-1/2"]
[BlackElo "2000"]
[Time "16:39:04"]
[WhiteElo "2400"]
[TimeControl "300"]
[SetUp "1"]
[FEN "1nn1k1n1/pppppppp/8/8/8/8/PPPPPPPP/3QK3 w - - 0 1"]
[Termination "normal"]
[PlyCount "44"]
[WhiteType "human"]
[BlackType "program"]

1. d4 Nf6 {(Ng8-f6 Qd1-d3 Nb8-c6 c2-c3 d7-d6 g2-g4 h7-h6 h2-h3 Nc8-b6 e2-e4
Ke8-d7 Ke1-e2) -1.25/15 17} 2. c4 Nc6 {(Nb8-c6 Qd1-d3 e7-e6 Qd3-g3 Nc8-d6
c4-c5 Nd6-f5 Qg3xc7 Nf5xd4 Qc7xb7 Nf6-e4 e2-e3 Nd4-c2+ Ke1-e2) -1.01/16 12}
3. d5 Nd8 {(Nc6-d8 Qd1-d3 Nc8-b6 b2-b3 d7-d6 e2-e4 c7-c6 Ke1-e2 g7-g6 f2-f3
c6xd5 c4xd5) -0.47/15 7} 4. Qd4 d6 {(d7-d6 e2-e4 e7-e6 Ke1-e2 c7-c6 g2-g4
e6-e5 Qd4-d3 h7-h6 h2-h3 c6xd5) -0.42/14 5} 5. e4 e6 {(e7-e6 Ke1-e2 c7-c6
f2-f3 e6xd5 c4xd5 c6xd5 e4xd5 a7-a6 g2-g4 Nc8-e7 Qd4-a4+ Ke8-f8 Qa4-a3)
-0.41/14 5} 6. f3 exd5 {(e6xd5 e4xd5 h7-h6 Qd4-e3+ Ke8-f8 Ke1-f2 c7-c6
Qe3-e1 c6xd5 Qe1-a5 Nc8-b6 c4xd5 Kf8-e7) -0.43/14 18} 7. cxd5 h6 {(h7-h6
g2-g4 a7-a6 h2-h4 c7-c5 Qd4-c3 Ke8-d7 g4-g5 h6xg5 h4xg5 Nf6-h5 Qc3-d2
Nc8-e7) -0.51/15 10} 8. g4 a6 {(a7-a6 h2-h4 c7-c5 Qd4-c3 Nc8-b6 g4-g5 h6xg5
h4xg5 Nf6-h5 Ke1-f2 Nb6-d7 Kf2-e3 Nd7-e5 f3-f4) -0.49/16 6} 9. h3 Ne7
{(Nc8-e7 Ke1-d2 Ke8-d7 f3-f4 Nf6-e8 a2-a4 c7-c6 a4-a5 c6xd5 e4xd5 Ne8-f6
g4-g5) -0.44/15 5} 10. f4 Ng6 {(Ne7-g6 e4-e5 d6xe5 f4xe5 Nf6-d7 e5-e6
Nd7-e5 Qd4-a4+ c7-c6 e6xf7+ Ne5xf7 d5xc6 Nd8xc6 Qa4-b3) -0.34/16 4} 11. e5
dxe5 {(d6xe5 f4xe5 Nf6-d7 e5-e6 Nd7-e5 Qd4-c3 f7xe6 Qc3xc7 e6xd5 Qc7xg7
Nd8-f7 Qg7-f6) -0.23/17 4} 12. fxe5 Nd7 {(Nf6-d7 e5-e6 Nd7-e5 Qd4-c3 f7xe6
Qc3xc7 e6xd5 Qc7xg7 Nd8-f7 Qg7-f6 Ke8-d7 Ke1-d1 Ng6-e7 h3-h4) -0.29/18 4}
13. e6 Nde5 {(Nd7-e5 Qd4-c3 f7xe6 Qc3xc7 e6xd5 Qc7xg7 Nd8-f7 Qg7-f6 Ke8-d7
Ke1-d1 Ng6-e7 h3-h4 Ne7-g6 g4-g5) -0.25/14 1} 14. exf7+ Ndxf7 {(Nd8xf7
Qd4-a7 Nf7-d8 Qa7-e3 Nd8-f7 Ke1-d1 b7-b6 Kd1-c2 Ke8-d7 Kc2-c3 Nf7-g5 h3-h4
Ng5-f3) -0.01/15 6} 15. Qa7 Nd6 {(Nf7-d6 Qa7-b8+ Ke8-d7 Qb8-g8 Nd6-e8
Ke1-d2 Kd7-d8 Qg8-e6 Ne5-c4+ Kd2-c3 Ng6-f4 Qe6-f7 Nf4-e2+ Kc3-c2) 0.00/17
6} 16. Qb8+ Kd7 {(Ke8-d7 Qb8-g8 Nd6-e8 Ke1-d2 Kd7-d8 Qg8-e6 Ne5-c4+ Kd2-c3
Ng6-f4 Qe6-f7 Nf4-e2+ Kc3-c2 Ne2-d4+ Kc2-c3) 0.00/18 2} 17. Qg8 Ne8
{(Nd6-e8 Ke1-d2 Kd7-d8 Qg8-e6 Ne5-c4+ Kd2-c3 Ng6-f4 Qe6-f7 Nf4-e2+ Kc3-c2
Ne2-d4+ Kc2-c3 Nd4-b5+ Kc3-b3) 0.00/18 3} 18. Kd1 Kd8 {(Kd7-d8 Qg8-e6
Ne8-d6 Qe6-g8+ Nd6-e8 Qg8-e6 Ne8-d6 Qe6-g8+ Nd6-e8 Qg8-e6 Ne8-d6 Qe6-g8+
Nd6-e8 Qg8-e6) 0.00/19 4} 19. Qe6 Nd6 {(Ne8-d6 Qe6-g8+ Nd6-e8 Qg8-e6 Ne8-d6
Qe6-g8+ Nd6-e8 Qg8-e6 Ne8-d6 Qe6-g8+ Nd6-e8 Qg8-e6 Ne8-d6 Qe6-g8+) 0.00/19
1} 20. Qg8+ Ne8 {(Nd6-e8 Qg8-e6) 0.00/25 2} 21. Qe6 Nd6 {(Ne8-d6 Qe6-g8+)
0.00/22 3} 22. Qg8+ Ne8 {(Nd6-e8) 0.00/27 3 3-fold repetition} 1/2-1/2
[/pgn]

[d]3kn1Q1/1pp3p1/p5np/3Pn3/6P1/7P/PP6/3K4 w - - 16 23

For the time being, I do not seem very much influenced by elephantiasis. (Do not get me wrong, Harm, I like your theory, but I will beat Queeny)

[Lyudmil Tsvetkov]

First game against Queeny - I lost on time in a completely won position. 5 minutes each side is not much for a human in an unfamiliar environment. Extremely funny - Queeny is very weak indeed. It beats top engines with its score rising continuously, but against me its score was dropping all the time: from +4.40 to +1.50. I do not believe how many weak moves that engine made - Rybka definitely played much stronger. And at the end of the game, when Arena announced forfeit on time, Queeny checked me, did not allow me to respond, then grabbed my king (which Arena allowed), and finally proudly crashed

This is definitely the funniest experience I have had against a chess engine. Thank you, Harm, for the opportunity.

[pgn][Event "Computer chess game"]
[Site "OWNER-PC"]
[Date "2013.10.26"]
[Round "?"]
[White "QN"]
[Black "owner"]
[Result "0-1"]
[BlackElo "2400"]
[Time "17:07:38"]
[WhiteElo "2000"]
[TimeControl "300"]
[SetUp "1"]
[FEN "1nn1k1n1/pppppppp/8/8/8/8/PPPPPPPP/3QK3 w - - 0 1"]
[Termination "time forfeit"]
[PlyCount "56"]
[WhiteType "human"]
[BlackType "human"]

1. d4 Nf6 {(g8f6 e2e4 f6e4 d1g4 e4f6 g4g7 d7d5 g2g4 b8c6 g4g5 f6e4 g7g8
e8d7 g8f7) +4.14/15 7} 2. c4 Nc6 {(b8c6 d4d5 c6e5 d1b3 b7b6 b3c3 e5g6 f2f3
e7e5 g2g4 h7h6 e2e4 c8d6 g4h5 f6h5) +4.44/16 6} 3. d5 Ne5 {(c6e5 d1b3 b7b6
b3g3 d7d6 b2b3 g7g6 f2f4 e5d7 g3f3 d7c5 b3b4 c5e4 b4b5) +4.35/17 18} 4. Qd4
d6 {(d7d6 c4c5 e7e6 f2f4 e5d7 c5c6 b7c6 d5c6 d7b6 e2e3 c8e7 d4c3 b6d5 c3c1)
+4.64/16 5} 5. f4 Ned7 {(e5d7 g2g4 f6g4 d4g7 d7f6 h2h3 g4e3 b2b3 c8b6 e1d2
e3f5 g7h8 e8d7 d2d3 f5g3 e2e3) +4.59/17 9} 6. e4 e5 {(e7e5 d5e6 f7e6 d4e3
e6e5 g2g3 c8b6 b2b3 e8f7 e1e2 a7a6 e3f3 d7c5 f4e5 d6e5 b3b4) +5.14/18 6} 7.
dxe6 fxe6 {(f7e6 e4e5 c7c5 d4e3 d6e5 f4e5 f6g4 e3g5 g4e5 g5g7 e5c4 b2b3
c4e3 e1e2 e3d5 g7h7) +4.95/17 5} 8. e5 c5 {(c7c5 d4e3 d6e5 f4e5 f6g4 e3g5
g4e5 g5g7 e5c4 e1e2 c8d6 b2b3 c4e5 g7h8 e8e7 h8h7 e7f6 h2h3) +4.74/19 5} 9.
Qe3 dxe5 {(d6e5 f4e5 f6g4 e3g5 g4e5 g5g7 e5c4 g7h7 c8b6 e1e2 c4e5 h2h3 c5c4
h7e4 b6d5 e4d4 a7a6 g2g4) +4.54/20 7} 10. fxe5 Ng4 {(f6g4 e3g5 g4e5 g5g7
e5c4 g7h7 c8b6 h2h3 c4e5 e1e2 c5c4 h7h6 e8e7) +4.61/19 6} 11. Qg5 Ngxe5
{(g4e5 g5g7 e5c4 g7h7 c8b6 h2h3 c4e5 e1e2 c5c4 h7h6 e8e7 h6h4 e7d6 h4d4
b6d5 d4a7 d5f4 e2d2 d7c5) +4.60/19 6} 12. Qxg7 Nxc4 {(e5c4 g7h7 c8b6 h2h3
c4e5 b2b3 c5c4 b3c4 b6c4 h7h8 e8e7 h8a8 e5d3 e1f1 d7c5 a8a7 c4e3 f1g1)
+4.56/18 7} 13. Qxh7 Nc8b6 {(c8b6 h7e4 e6e5 e4b7 c4b2 b7a7 b2d3 e1d2 e5e4
h2h3 b6c4 d2c3 d7e5 a2a3) +4.36/17 7} 14. h4 Nce5 {(c4e5 h4h5 c5c4 h5h6
c4c3 b2c3 b6c4 h7b1 c4b2 h6h7 b2d3 e1e2 d3f4) +3.39/16 4} 15. h5 Nf8 {(d7f8
h7b7 f8d7 b7e4 b6d5 h5h6 e8f7 e4h4 d5e3 h4e4 e3c4 h6h7 f7g7 e4h4) +2.88/17
4} 16. Qxb7 Nfd7 {(f8d7 b7e4 e8f8 h5h6 f8g8 e1e2 b6c4 h6h7 g8h8 e4c2 h8g7
g2g3 c4b6 e2e3 b6d5) +2.54/17 3} 17. Qxa7 Kf7 {(e8f7 a7a3 c5c4 a3g3 b6a4
e1f1 f7e8 b2b4 c4b3 a2b3 a4c5 f1e2) +2.38/17 8} 18. a4 Nc8 {(b6c8 a7a6 c8e7
h5h6 d7f6 a6d6 e7g6 d6c7 f6d7 a4a5 e5d3 e1e2 g6f4 e2e3) +2.32/17 5} 19. Qb7
Nd6 {(c8d6 b7a6 d6c4 b2b3 e5d3 e1e2 c4e5 a6d3 e5d3) +2.50/17 4} 20. Qc7 Ne8
{(d6e8 c7b7 f7f6 a4a5 e8d6 b7c7 d6b5 c7b7 b5d4 a5a6 c5c4 a6a7 d7c5 b7c8
c5d3) +2.21/17 3} 21. Qd8 c4 {(c5c4 h5h6 d7f6 d8d4 e5g4 h6h7 f7g7 d4c4 g7h7
c4e6 e8g7 e6c8 h7g6 b2b3 g7f5) +1.08/17 4} 22. a5 Nc5 {(d7c5 e1f1 e8f6 a5a6
c5a6 d8a5 e5g4 a5a6 f6h5 f1g1 h5f6 a6c4 g4e3 c4f4 e3d5) -0.53/18 2} 23. h6
Nf6 {(e8f6 e1f1 f7g6 a5a6 c5a6 d8d6 e5g4 d6a6 e6e5 f1e2 g4h6 a6c4 h6f5 b2b4
f5d4) -0.84/17 4} 24. Qc7+ Nfd7 {(f6d7 c7c8 f7g6 c8h8 c4c3 b2c3 c5d3 e1f1
g6f5 h6h7 e5c4 h8d4 c4d2 f1e2) +0.36/17 3} 25. Qd6 Ncd3+ {(c5d3 e1e2 d3b2
a5a6 b2a4 d6a3 a4b6 a3g3 e5g6 h6h7 e6e5 e2e1 c4c3 g3g6 f7g6 h7h8) +1.28/17
2} 26. Kd2 Nxb2 {(d3b2 a5a6 b2a4 a6a7 a4b6 d6b8 c4c3 d2c2 f7g6 b8h8 e5c4
h8g7 g6h5 a7a8 b6a8) +0.09/18 4} 27. a6 c3+ {(c4c3 d2c3 b2c4 d6d4 c4b6 a6a7
f7g6 d4d6 g6h6 d6e6 h6g7 e6e8 g7f6 a7a8 b6a8 e8a8 f6f5 g2g3 d7c5) -0.95/19
4} 28. ... {0-1 White forfeits on time} cxd2 29. ... {White forfeits on
time} 0-1
[/pgn]

[d]8/2Q2k2/4pn1P/P1n1n3/2p5/8/1P4P1/4K3 b - - 2 24

Well, if this is not won, I do not know. Now I will win a couple of games in a row, but I think even this one should be considered for a white win, as Harm allowed me to use as much time as I want.

[hgm]

I unearthed what I got on my disk, and winboard from the year 2000 was the last version used by me. Why should I use winboard now when I have Fritz with a wonderful interface and usually sufficient features?

You answered that yourself already, right?

Still, as the positions interest me, I downloaded Arena and played some games in it. The interface is not extremely convenient, but still better than winboard.

Well, if you want to limp along with Arena when you can't limp along with Fritz, that's fine with me. Just stop whining to us about the interface if you choose an inferior one by your own ignorance...

Just started playing Rybka 2.2, which is the strongest engine coming with Arena. Below a typical game. 5 minutes each side.

Soon I will download Queeny and start beating the .... out of it. One step at a time.

No win, hey?

Of course you should win this position easily against QueeNy. It doesn't have at least5 Knights, so QueeNy is completely mis-tuned for it. But if you want to waste your time, go ahead.

[hgm]

Do you know approximately how many extra pawns the queen side need to make this even? I tried giving the queen side two extra pawns, which didn't seem to help much:

I never tried that. All I know is that with 6 Knights vs 3 Queens (equal Pawns) it is badly lost for the Knights, with 7 Knights it is easily won. The extra Knight devaluates all three Queens, however, so its marginal value could be close to 6 Pawns. So it could be that with 7 Knights it is effectively +3, and with 6 Knights -3. That doesn't necessarily mean 3 extra Pawns would be enough; Pawns usually devaluate for the side that has a numeric minority in pieces (e.g. R vs 2 minors), because they become a liability, and it get get easily more attacked then you can protect them. The 'Knight cloud' gobbles up Pawns quite easily. The most helpful would probably be to use the Pawns to build a strong King fortress, perhaps with some doubled Pawns to keep the Knights away.

[hgm]

And at the end of the game, when Arena announced forfeit on time, Queeny checked me, did not allow me to respond, then grabbed my king (which Arena allowed), and finally proudly crashed .

I warned you! Don't come complaining to us if you insist 0n using inferior interfaces!

If you are done fooling around, you can start to test elephantiasis, by playing 3Q vs 7N.

[Lyudmil Tsvetkov]

And at the end of the game, when Arena announced forfeit on time, Queeny checked me, did not allow me to respond, then grabbed my king (which Arena allowed), and finally proudly crashed .

I warned you! Don't come complaining to us if you insist 0n using inferior interfaces!

If you are done fooling around, you can start to test elephantiasis, by playing 3Q vs 7N.

I will do that, but why do not you acknowledge that you were wrong: a queen is not about equal, but actually much stronger than 3Ns? We were talking science, did not we?

[hgm]

The only reason it can beat the top engines so badly is that it depreciates the Queens (or, equivalently, upgrades the Knights) according to the elephantiasis theory.

Is this theory written down somewhere?

It is clear that the effect exists, which means that it makes sense to incorporate it in the material balance evaluation (as that can usually be done for free), but I wonder how it could best be expressed.

There is no formal write-up of the theory. It was mentioned a few times before, but it isn't really much of a theory. It can be worded in one sentence:

The evaluation contains terms proportional to the products of the number of white pieces of type A and the number of black pieces of type B.

It could be added that that the sign of such terms is likely to favor the owner of the weakest pieces in the product, as weak pieces hinder strong pieces more than the other way around, and the product just gives the number of interactions of this type.

The rest is just conjecture. (In particular that the coefficients c_ij are proportional to the value of the weak piece, and to the square root of the value of the strong piece.) At the moment it would be best to consider the c_ij simply fit parameters.

Note that Larry Kaufman claimed there is a measurable 'Knight-pair penalty', but considering the empirical results on 7N vs 3Q it seems that this isn't something that extrapolates to larger numbers of Knights. It is probably there in the GM data because having only two Knights makes you vulnerable for sacrifices based on the fact that KNNK is draw. So I expect it to be a 'Knights-only' penalty, that does not grow with the number of Knights.

As I explained earlier in this thread, Bishops do not work well at all in large numbers against Queens. I guess their c_ij are much smaller than those of the Knights. Bishops on the same color tend to be in each other's way, strongly reducing the amount of board control they provide (and thus the elephantiasis effect). But for Knights the effect seems large (c_NQ ~ 75 cP). You would have to renormalize the base values for introduction of the quadratic term, though, so that the basic trades in the full setup retain their normal value (e.g. Q vs 3 minors ~ 0)

[Eelco de Groot]

The only reason it can beat the top engines so badly is that it depreciates the Queens (or, equivalently, upgrades the Knights) according to the elephantiasis theory.

Is this theory written down somewhere?

It is clear that the effect exists, which means that it makes sense to incorporate it in the material balance evaluation (as that can usually be done for free), but I wonder how it could best be expressed.

It is not particularly relevant for games where you don't underpromote to knight. Rainbow Serpent already has a term for this in the imbalance code from Tord that I think is maybe two years old, certainly more than one.

Code: Select all

  // Polynomial material balance parameters
  const Value RedundantQueenPenalty   = Value(320);
  const Value RedundantRookPenalty    = Value(550);
  const Value RedundantKnightPenalty  = Value(-50); //[Based on Harm-Geert Muller's experiments with more than two knights] 
  const Value RedundantBishopPenalty  = Value( 20); //[Two same color bishops are probably slightly redundant, but untuned. EdG]
  const Value DelayMinortradeBonus    = Value(125); //[For Kingattacks you like to have a full complement of knights and bishops,
//[three knights might be powerful too, but the second degree bonus is only
//in combination with the bishop pair]

Code: Select all

  // Redundancy of major pieces, formula based on Kaufman's paper
  // "The Evaluation of Material Imbalances in Chess"
  if (pieceCount[Us][ROOK] > 0)
      value -=  RedundantRookPenalty * (pieceCount[Us][ROOK] - 1)
              + RedundantQueenPenalty * pieceCount[Us][QUEEN];
  else if (pieceCount[Us][QUEEN] > 1)
		value -= (pieceCount[Us][QUEEN] - 1) * Value(100) + (pieceCount[Us][QUEEN] - 2) * Value(150);
  if (pieceCount[Us][KNIGHT] >= 3)
	  value -= (pieceCount[Us][KNIGHT] + 1) * (pieceCount[Us][KNIGHT]) * (pieceCount[Us][KNIGHT] - 1) * (pieceCount[Us][KNIGHT] - 2) * RedundantKnightPenalty;
  if (pieceCount[Us][BISHOP] == 2 && !pieceCount[Us][NO_PIECE_TYPE])
	  value -= (pieceCount[Us][BISHOP]) * RedundantBishopPenalty;

It is just a crude attempt however I am almost sure that Rainbow Serpent handles this better than Stockfish. But I don't want to spoil the party here Please take no notice.

Eelco

[Eelco de Groot]

Regarding Peter Österlund's question, I don't have WinBoard installed on this computer yet and Shredder will not accept the 'Charge of the Light Brigade' position. I don't know if SCID, for which the engine was really intended can handle it. I should check that. But for now, just in a command window. The RedundantKnightPenalty computation is not very accurate yet I think, but this position is probably still won for the Knights. Just a short search with the original 'Scid Serpent':

Rainbow Serpent Build 025 130624 by Tord Romstad, Marco Costalba and Joona Kiiski
position fen [d]nnnnknnn/2pppp2/8/8/8/8/PPPPPPPP/Q2QK2Q w - - 0 1
go infinite
info depth 1 seldepth 2 score cp -654 nodes 51 nps 17000 time 3 multipv 1 pv b2b
4
info depth 2 seldepth 2 score cp -872 nodes 229 nps 38166 time 6 multipv 1 pv b2
b4 h8g6
info depth 3 seldepth 4 score cp -525 nodes 404 nps 40400 time 10 multipv 1 pv b
2b4 h8g6 g2g4
info depth 4 seldepth 4 score cp -747 nodes 1194 nps 79600 time 15 multipv 1 pv
b2b4 h8g6 g2g4 a8b6
info depth 5 seldepth 6 score cp -501 nodes 1466 nps 69809 time 21 multipv 1 pv
b2b4 h8g6 g2g4 c8b6 h1a8 b6a8
info depth 6 seldepth 6 score cp -501 nodes 1870 nps 69259 time 27 multipv 1 pv
b2b4 h8g6 g2g4 c8b6 h1a8 b6a8
info depth 7 seldepth 10 score cp -488 nodes 2698 nps 79352 time 34 multipv 1 pv

.
.
.

info depth 28 seldepth 50 score cp -125 nodes 1382923347 nps 3009999 time 459443
multipv 1 pv b2b3 f7f6 g2g4 c7c6 f2f4 d7d6 d2d4 h8f7 c2c4 a8c7 e2e4 e7e5 f4f5 e
5d4 d1d4 c8e7 h2h4 c6c5 h4h5 c5d4
info depth 29 currmove b2b3 currmovenumber 1
stop
bestmove b2b3 ponder f7f6

Take the centipawn scores with a few grains of salt

Eelco