Elephantiasis: reality or myth?

[Lyudmil Tsvetkov]

Even though the side with the bishops has four bishop pairs, I think it should be won in all cases for the side with the Queens. So playing for a draw by closing the position is not really good enough Lyudmil. I made a new Scid Serpent version with increased redundancy for same squared bishops, but that does not really apply here, with four bishop pairs the redundancy factor will not be awarded unless the bishop pairs will be broken up, only redundancy of the three Queens applies. But I think the Queens still win. I started a small match of Scid Serpent agains Stockfish 4 in SCID that allows these kind of illegal positions luckily. I predict that White will win eight out of eight...

[d]bbbbkbbb/pppppppp/8/8/8/8/PPPPPPPP/Q2QK2Q w - - 0 1

It is a timecontol 1 minute no Fischer bonus match so it does not take long.

Eelco

It seems I am the only person that has actually played this particular imbalance, so I can tell you without any doubt, no predictions at all, because the position is very simple, that the bishop side will win all the games.
I would be happy if you post the results after the match is finished.

Well the results are a bit inconclusive. Maybe the opportunities for both sides are a bit more balanced than the resident elephantologists predicted

I got 3 wins with White, one draw and four wins with Black of which one was on time in an equal position, the other wins look more regular with mate. It is possible I have overlooked something as I have never played a match in SCID before and have no idea how to access the clipbase for instance. Stockfish 4 won the match so I think the Queen elephantiasis rule has to be modified a bit, it seems Stockfish plays this better without, maybe not suprising against bishops instead of knights

Eelco

Hi Eelco.
I understand it this way: the first results we have (thank you btw. for this) indicate the bishop side has the advantage. It is interesting what a larger sample of games would show.

My view is that 7 Bs are stronger than 7 Ns for the following reasons:

- bishops still defend almost the way knights do
- bishops still defend the king almost the way knights do (albeit a bit less efficiently)
- bishops additionally are able to control squares and attack from a distance
- bishops build very efficient batteries of 2 and 3 pieces along the same diagonal with a devastating effect, also additionally
- bishops additionally complement great when there are bishops on squares of different colour, which with 7 Bs is almost always the case

I think because of the above the 7Bs are much much stronger than the 7Ns. It seems also that the rule for trading Q for 2 Bs does not have a practical meaning here, as it seems impossible to trade a Q for 2 Bs, as the Bs control almost everything, difficult to attack them.

My conclusion would be that complementary control of squares matters enormously, as do bishop batteries.

Would be happy if someone provides a larger sample of games.

[hgm]

Well, so far Stockfish did not manage to get more than two draws with the Queens, against QueeNy (in case you missed the second game, look to my previous post). The latter frequently trading a Queen for two Bishops...

So what you think and what is are two entirely different things. Frequently at odds with each other...

[Lyudmil Tsvetkov]

Kanguru is another engine... I am not sure what you refer too. To me the eval graph looks pretty smooth.

Second game ended in a draw as well. Stockfish must really be extremely incompetent at this, that it cannot win from a 600 Elo weaker engine from such a won position, or the position is not nearly as won as you claim.

[pgn][Event "Computer Chess Game"]
[Site "MAKRO-PC"]
[Date "2013.10.28"]
[Round "-"]
[White "QueeNy 0.16"]
[Black "Stockfish 4 64 SSE4.2"]
[Result "1/2-1/2"]
[TimeControl "40/300"]
[FEN "bbbbkbbb/pppppppp/8/8/8/8/PPPPPPPP/Q2QK2Q w - - 0 1"]
[SetUp "1"]
1. c3 {-4.66/14} b6 {-8.82/18 10} 2. d3 {-4.78/13 4} c5 {-8.66/17 6} 3. d4
{-4.95/13 5} g6 {-8.10/18 7} 4. f3 {-5.04/13 4} cxd4 {-7.57/19 6} 5. cxd4
{-5.60/15 8} Bdc7 {-7.25/20 16} 6. Qd2 {-5.26/13 7} Bfg7 {-6.14/19 7} 7. e3
{-5.92/15 4} f5 {-5.93/21 8} 8. f4 {-5.97/15 5} Bcb7 {-5.39/21 8} 9. Qg1
{-6.13/15 5} Bxg2 {-4.78/22 24} 10. Qdxg2 {-6.18/16 8} Bxg2 {-2.84/18 1.6}
11. Qxg2 {-6.16/16 6} Bxd4 {-2.82/23 7} 12. exd4 {-5.91/15 5} Bxd4
{-2.88/24 20} 13. Qc1 {-6.21/15 6} Be6 {-2.76/24 6} 14. Qh3 {-6.47/17 6} h5
{-2.76/23 7} 15. Qd2 {-6.94/17 7} Bf6 {-2.70/21 5} 16. Qdg2 {-6.91/16 5}
Bf7 {-2.74/22 7} 17. Qb7 {-7.08/14 5} Kf8 {-2.64/22 6} 18. Qd3 {-7.17/15 9}
d5 {-2.38/19 5} 19. Qc2 {-7.02/16 6} Bxf4 {-2.40/20 6} 20. Qcc8+
{-6.85/16 5} Kg7 {-1.87/19 1.5} 21. Qbxb8 {-6.79/16 5} Bxb8 {+0.34/19 5}
22. Qxb8 {-6.66/21 7} Bd4 {+0.24/20 8} 23. Qxa7 {-6.55/21 9} e5
{+0.00/21 11} 24. Qc7 {-6.48/20 6} f4 {+0.00/24 10} 25. Ke2 {-6.18/21 6} h4
{+0.00/26 7} 26. b4 {-5.19/22 7} b5 {+0.00/26 5} 27. Qd6 {-4.89/20 5} g5
{+0.00/25 4} 28. Qd8 {-4.57/21 5} Kg6 {+0.00/26 5} 29. Qd6+ {-3.94/23 7}
Kf5 {+0.00/28 5} 30. Qf8 {-3.74/24 11} Kf6 {+0.00/28 7} 31. Qd8+
{-3.44/23 5} Kf5 {+0.00/30 5} 32. Qf8 {-0.04/28 10} Kg6 {+0.00/28 6} 33.
Qd6+ {+0.07/30 9} Kf5 {+0.00/34 10} 34. Qf8 {-0.01/26 7}
{XBoard adjudication: repetition draw} 1/2-1/2
[/pgn]

It is not about incompetence: Stockfish simply does not know how to play well with 7Ns, neither with 3Qs, and the search, althoug more efficient than Queeny search, is influenced by the eval. Stockfish is tuned for normal parameters, but not for extraordinary ones.

You claimed the 7Ns win easily, and you somewhat supported that, you claimed that the 7Bs lose easily, and you provided absolutely no evidence on this, rather refuted yourself. According to your data, which was later corrected by others and even yourself, Stockfish was losing almost all the games against the knights, now all the games are drawn, what is the conclusion?

Symptomatic is also that you suddenly stopped to post a large amount of humiliating diagrams with large swarms of knights preying on whatever other pieces they set their eyes on...

[hgm]

Well, the Bishops lost easily when QueeNy played against itself.

Stockfish seems to draw with the Bishops against QueeNy. So I would think Stockfish is stronger than QueeNy. Yet QueeNy manages to draw Stockfish rather consistently in this position. To me that seems evidence that the Queens must have the advantage. Why else would QueeNy be able to draw a stronger opponent (or beat itself) if it was not for the better start position?

I am not convinced that trading the 4th&5th Bishop for a Queen is good strategy. The Bishops seem to have great difficulty provinding King shelter at the same time as doing other useful things (like stopping passers), and against two Queens you get very easily checkmated. While the worst a single Queen can do is perpetually check you. So it could be a strategic mistake to continue Q-for-2B in the 5B-2Q situation. If Q-for-2B is not good, there would be nothing wrong with Stockfish' evaluation. In particular that it doesn't seem to be able to win with 3B vs Q and equal or even superior Pawns seems particularly disappointing: that should not be so far out of its normal parameter range.

But perhaps I indeed remember it wrong, and mixed up the 7B-vs-3Q case with the 6B-vs-3Q case. Initially I was also playing 6N-vs-3Q, and QueeNy was winning that too, both with Knights and Queens against almost all engines (including Stockfish, which then was not nearly as strong as it is now). And with Bishops it always performed much worse than with the Knights.

I don't know what your problem with the diagrams is. We are discussing Bishops now, so of course there are no diagrams with Knights. And I just learned to use the pgn forum tags, which were not working on the Linux machine I was using yesterday. (I could not see the games.) Today I am working under Windows, and there they work for me. But the final positions are still there, you just have to use the >> button to see them. So there is no need to post them separately.

At longer TC (40/20') the third game still ended in draw:
[pgn][Event "Computer Chess Game"]
[Site "MAKRO-PC"]
[Date "2013.10.28"]
[Round "-"]
[White "QueeNy 0.16"]
[Black "Stockfish 4 64 SSE4.2"]
[Result "1/2-1/2"]
[TimeControl "40/1200"]
[FEN "bbbbkbbb/pppppppp/8/8/8/8/PPPPPPPP/Q2QK2Q w - - 0 1"]
[SetUp "1"]

{--------------
b b b b k b b b
p p p p p p p p
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
P P P P P P P P
Q . . Q K . . Q
white to play
--------------}
1. g3 {-4.65/15} g6 {-9.03/21 39} 2. c3 {-5.01/16 37} f5 {-8.90/22 21} 3.
e4 {-5.11/16 30} b5 {-7.73/23 46} 4. Qf1 {-5.29/16 28} Bxe4 {-7.45/23 21}
5. Qxb5 {-5.38/16 15} Bcb7 {-7.21/24 22} 6. Qda4 {-5.14/16 16} Bec6
{-3.43/28 21} 7. Qxc6 {-5.61/19 32} Bxc6 {-3.39/28 47} 8. Qb4 {-5.54/18 44}
a5 {-3.35/28 26} 9. Qxb8 {-5.60/18 23} a4 {-3.39/28 32} 10. f4
{-5.42/17 27} e5 {-3.21/26 24} 11. fxe5 {-5.55/17 23} Bxe5 {-3.21/27 33}
12. d4 {-5.56/17 30} Bf6 {-2.66/25 1:09} 13. b3 {-5.64/16 22} Bh6
{-2.62/23 23} 14. bxa4 {-5.36/17 20} Bfg5 {+0.00/32 41} 15. Qab2
{-4.81/16 17} Bd2+ {+0.00/29 21} 16. Qxd2 {-4.59/20 19} Bxd2+ {+0.00/32 12}
17. Kxd2 {-4.72/21 19} Bxa2 {+0.00/33 42} 18. a5 {-4.59/22 27} Bc4
{+0.08/25 23} 19. Ke3 {-4.67/21 25} d6 {+0.40/27 20} 20. Qc8 {-4.64/21 19}
Bd7 {+0.62/28 50} 21. Qb8 {-4.62/21 31} Bdb5 {+0.62/32 20} 22. Qc8
{-4.76/21 23} Ba6 {+0.50/32 2:01} 23. Qb8 {-4.82/21 28} Bd5 {+0.64/28 37}
24. Qa7 {-4.73/19 28} Bf1 {+0.00/29 38} 25. Qb8 {-4.84/21 43} Kd7
{+0.70/27 18} 26. Qb2 {-4.86/20 35} Ba6 {+0.86/24 22} 27. Qb4 {-4.89/19 24}
Bdc4 {+0.74/27 39} 28. Qb8 {-5.03/20 33} Bg5+ {+0.66/26 38} 29. Kf2
{-4.79/21 19} Bd8 {+0.00/28 14} 30. Ke3 {-0.02/35 24} Bd3 {+0.64/26 15} 31.
d5 {-4.84/20 30} h5 {+0.40/26 41} 32. Qa8 {-4.65/22 45} Bab5 {+0.52/26 26}
33. h4 {-4.71/23 32} Bdc4 {+0.00/26 22} 34. Kd2 {-4.69/22 32} Bd3
{+0.00/30 11} 35. Ke3 {-0.04/28 43} g5 {+0.00/29 14} 36. hxg5 {-4.36/22 22}
Bxg5+ {+0.00/20 4} 37. Kf2 {-4.38/23 59} f4 {+0.00/29 22} 38. Qg8
{-3.87/23 37} fxg3+ {+0.00/31 15} 39. Kxg3 {-3.95/23 21} h4+ {+0.00/32 21}
40. Kg4 {-2.02/25 37} Be7 {+0.00/35 19} 41. Kf4 {-1.58/24 23} Ba4
{+0.00/37 46} 42. Ke3 {-0.09/23 26} Bab5 {+0.00/40 28} 43. Kf4
{+0.03/28 18} Bdc4 {+0.00/34 48} 44. Ke3 {-0.22/23 23} Ba6 {+0.00/41 48}
45. Qe6+ {-0.02/26 30} Ke8 {+0.00/37 33} 46. Qg6+ {-0.06/25 19} Kd7
{+0.00/51 37} 47. Qe6+ {-0.07/29 2:01} Ke8 {+0.00/56 37} 48. Qg6+
{+0.02/29 27} Kd7 {+0.00/57 55} 49. Qe6+ {-0.02/28 1:04}
{XBoard adjudication: repetition draw} 1/2-1/2
[/pgn]

[Eelco de Groot]

Hi Eelco.
I understand it this way: the first results we have (thank you btw. for this) indicate the bishop side has the advantage. It is interesting what a larger sample of games would show.

My view is that 7 Bs are stronger than 7 Ns for the following reasons:

- bishops still defend almost the way knights do
- bishops still defend the king almost the way knights do (albeit a bit less efficiently)
- bishops additionally are able to control squares and attack from a distance
- bishops build very efficient batteries of 2 and 3 pieces along the same diagonal with a devastating effect, also additionally
- bishops additionally complement great when there are bishops on squares of different colour, which with 7 Bs is almost always the case

I think because of the above the 7Bs are much much stronger than the 7Ns. It seems also that the rule for trading Q for 2 Bs does not have a practical meaning here, as it seems impossible to trade a Q for 2 Bs, as the Bs control almost everything, difficult to attack them.

My conclusion would be that complementary control of squares matters enormously, as do bishop batteries.

Would be happy if someone provides a larger sample of games.

Hi Lyudmil,

I did play a longer match, because it did not seem one of the sides is totally winning, at least not under the test circumstances and the programs not knowing much specifics. It was still an interesting fight and in the end the Bishops do seem to have the upper hand. I think it is indeed the case that having both several bishops of different colour to control more of the squares and also enough colleagues of same colour to protect each other, from a distance makes them pretty strong. More so of course if there are some breaks in the opponent pawn formation where knights would maybe be better equipped in closed positions. But I have no more than 100 games so it is not totally conclusive. Also there were quite a few timelosses playing without an increment. I preferred to have more games because the smallest increment I can give in Scid is one second. I have not counted the losses on time, I would have to do that by checking all the pgns. I hope not too many timelosses for a winning side I did not see those while watching the games, but more in drawn positions. Those distort the expected draw percentage of course and it is not certain there is an even number of White wins and Black wins that should have been drawn. Also Stockfish will play random moves if down to the last milliseconds, this favours the side of the bishop a little bit I think because it has more pieces it can blunder It also controls more squares so the side with queens in general has to play more carefully.

Scid Serpent won this time which is nothing to write home about, it should win even with an experimental material imbalance table mainly because it is based on the latest Stockfish.

Scid Serpent 51.0 / 100
Stockfish 4 49.0 / 100

+33, =26, -41

White wins 33 (The 3 Queens)
Draws 26 (underreported because of time losses)
Black wins 46 (The 7 Bishops)

Eelco

[hgm]

Well, that seems pretty close to equality. Not at all like with the Knights, which virtually always won. The special imbalance table does not seem to help much here. Which is in agreement with the fact that I noticed that Stockfish seemed to handle the Bishops better than QueeNy.

But even then, elephantiasis seems to be real enough, and it is just a matter how big the effect exactly is. Even 8 Bishops are not worth as much as 3 Queens based on standard piece values; you would need 9 to get about even. So 7 Bishops seem to perform here close to the level of 9. So the Queens seem to perform only at 78% of their nominal value.

[Lyudmil Tsvetkov]

Hi Eelco.
I understand it this way: the first results we have (thank you btw. for this) indicate the bishop side has the advantage. It is interesting what a larger sample of games would show.

My view is that 7 Bs are stronger than 7 Ns for the following reasons:

- bishops still defend almost the way knights do
- bishops still defend the king almost the way knights do (albeit a bit less efficiently)
- bishops additionally are able to control squares and attack from a distance
- bishops build very efficient batteries of 2 and 3 pieces along the same diagonal with a devastating effect, also additionally
- bishops additionally complement great when there are bishops on squares of different colour, which with 7 Bs is almost always the case

I think because of the above the 7Bs are much much stronger than the 7Ns. It seems also that the rule for trading Q for 2 Bs does not have a practical meaning here, as it seems impossible to trade a Q for 2 Bs, as the Bs control almost everything, difficult to attack them.

My conclusion would be that complementary control of squares matters enormously, as do bishop batteries.

Would be happy if someone provides a larger sample of games.

Hi Lyudmil,

I did play a longer match, because it did not seem one of the sides is totally winning, at least not under the test circumstances and the programs not knowing much specifics. It was still an interesting fight and in the end the Bishops do seem to have the upper hand. I think it is indeed the case that having both several bishops of different colour to control more of the squares and also enough colleagues of same colour to protect each other, from a distance makes them pretty strong. More so of course if there are some breaks in the opponent pawn formation where knights would maybe be better equipped in closed positions. But I have no more than 100 games so it is not totally conclusive. Also there were quite a few timelosses playing without an increment. I preferred to have more games because the smallest increment I can give in Scid is one second. I have not counted the losses on time, I would have to do that by checking all the pgns. I hope not too many timelosses for a winning side I did not see those while watching the games, but more in drawn positions. Those distort the expected draw percentage of course and it is not certain there is an even number of White wins and Black wins that should have been drawn. Also Stockfish will play random moves if down to the last milliseconds, this favours the side of the bishop a little bit I think because it has more pieces it can blunder It also controls more squares so the side with queens in general has to play more carefully.

Scid Serpent won this time which is nothing to write home about, it should win even with an experimental material imbalance table mainly because it is based on the latest Stockfish.

Scid Serpent 51.0 / 100
Stockfish 4 49.0 / 100

+33, =26, -41

White wins 33 (The 3 Queens)
Draws 26 (underreported because of time losses)
Black wins 46 (The 7 Bishops)

Eelco

Thanks Eelco!

Great contribution.
How did you ensure randomness of play?

[Lyudmil Tsvetkov]

Well, that seems pretty close to equality. Not at all like with the Knights, which virtually always won. The special imbalance table does not seem to help much here. Which is in agreement with the fact that I noticed that Stockfish seemed to handle the Bishops better than QueeNy.

But even then, elephantiasis seems to be real enough, and it is just a matter how big the effect exactly is. Even 8 Bishops are not worth as much as 3 Queens based on standard piece values; you would need 9 to get about even. So 7 Bishops seem to perform here close to the level of 9. So the Queens seem to perform only at 78% of their nominal value.

Elephantiasis=extremely good piece defence.

Let us bet that the 7Bs perform better than the 7Ns vs the 3Qs.

[Lyudmil Tsvetkov]

Well, that seems pretty close to equality. Not at all like with the Knights, which virtually always won. The special imbalance table does not seem to help much here. Which is in agreement with the fact that I noticed that Stockfish seemed to handle the Bishops better than QueeNy.

But even then, elephantiasis seems to be real enough, and it is just a matter how big the effect exactly is. Even 8 Bishops are not worth as much as 3 Queens based on standard piece values; you would need 9 to get about even. So 7 Bishops seem to perform here close to the level of 9. So the Queens seem to perform only at 78% of their nominal value.

Elephantiasis=extremely good piece defence.

Let us bet that the 7Bs perform better than the 7Ns vs the 3Qs.

[hgm]

A bet that IMO you already lost, based on the presented evidence. The 7 Knights were shown to win, even in starting positions where they did not protect each other at all, where the Pawns were fully spread out, and even when one resorted to giving them ridiculous other disadvantages, like starting their King in front of the Pawn chan, on g6.

The Bishops, however, seem to do no better than about 50%.

6 Knights also win against 3 Queens, when the side handling the Queens is slightly incompetent (e.g. does not strive for Q vs 2N, or, when it does, has less search depth than the Knights opponent that knows the danger, and thus can succesfully avoid them). 6 Bishops against 3 Queens do not stand a chance when the latter strives for Q-2B trades.

I think the most important difference is that it is very difficult for Bishops to keep them all twice protected, while for Knights this is easy. The Bishops are distributed over different colors, so with 6 you would have 3 on each color, and the only way to let them mutually protect each other twice is to put them on the same diagonal. Where they then sort of become frozen, as it would take several moves during which you are heavily exposed to switch them to another diagonal. And on the diagonal they hardly do anything useful. The opponent simply avoids that diagonal, and perpendicular to it there are only single Bishop attacts, for which single protection by a superior piece is sufficient defense.

A 2-for-1 trade-avoiding strategy for the Bishops against 3 Queens is extremely costly in terms of their usefulness (i.e. causes strong depression of their piece value), while to Knights it comes naturally. (And of course neither of them has to worry about 1-for-1 trades, which would be even more costly to avoid, basically requiring your pieces to remain under lock and key, not doing anything at all.)

The raw tactical power of a cloud of 6 Knights already inflicts a devaluation on the Queens that makes 3 Queens inferior to them. It is just that this situation can be relaxed through Q-for-2N trades that saves the day for the Queens. My prediction is therefore that 8 Knights would beat 4 Queens, despite the fact that the N-to-Q ratio is exactly 2, and Q-2N trades would be a winning path. With 8 Knights the Knights should be able to mutually protect each other well enough that no amount of Queen tactics can force a Q-2N trade. With 8 Bishops against 4 Queens, OTOH, the Bishops should be toast.

Note that mutual protection does not help you win games (although it could be great for drawing them). Essential is how much the pieces are able to do on the side, while protecting each other. The mutual protection is certainly part of the equation, as explained before: it saturates the amount of depression of the higher piece value, because no matter how dysfunctional, a piece is at least worth as much as what you are going to trade it for(*). With Queens versus Archbishops the suppression of the Queens can never get very large, as there is no way the Archbishops are going to avoid 1-on-1 trades. The mutual defense only becomes an issue when the survival of the strong-piece side hinges on 2-for-1 trades.


*) I discovered this when determining the opening value of the Camel (a (3,1) leaper), which is an absolutely worthless piece, that in the end-game is virtually always lost without any compensation at all. Yet its opening value came out very close to that of the other minors, because it has enough forking power to trade it for one when the opponent still has such minors in abundant quantity.