Charge of the Light Brigade

[Karlo Bala]

Here, the advantage of having queens is that they could trade whenever they want for two pieces, but the engine choose not to do so.
Miguel

Yes. For example, possibility of good exchange is why two bishops are often stronger then two knights or knight + bishop.

[Dann Corbit]

I have started a live demo match at http://80.100.28.169/gothic/QN.html , starting from the position:

[d]2n1kn2/3ppp2/1nn2nn1/8/8/8/3PPP2/1Q1QK1Q1 w - -

Currently I am playing Crafty against an engine tuned for this. Seems Crafty gets whipped, as it often loses with the Queens.

I am thinking of holding a tourney with this start position (which is reachable from the FIDE initial position, and thus normal Chess). If people want to subnmit a specially tuned engine, they are welcome!

Black has a bit more material, but also multiple smaller pieces of equal value are usually better than one large piece with the sum value of the smaller pieces.

I guess that in general the knights will win. Is there an engine that can hold its own with the queens?

Never mind the "more material" bit.
A queen should be 3x knight, so white appears ahead on wood. I guess that it is one of those "Material Imbalance" problems.

I suspect that black is favored here, despite the queen heavy opposition

It is great example of "redundancy of heavy pieces". Side with queens should win easily. It just need to exchange queen for two knight twice, but neither program except QueeNy doesn't know that.

Exactly. A piece of cake for a human, but it is too late for a computer when they realize that.

This is similar to have a huge bonus for passed pawns. Why a passed pawn is good? because it could promote, but if the bonus is bigger than the value of the piece it could promote, it will reach the 7th rank and it will stay there forever. Here, the advantage of having queens is that they could trade whenever they want for two pieces, but the engine choose not to do so.

Miguel

I think that for a chess program, the idea is very easily explained by the material imbalance table, then.

nnnnnn verse QQQ BAD for side with queens
nn verse Q Good for side with queen

With this information, chess programs will automatically trade down.

[hgm]

It is great example of "redundancy of heavy pieces". Side with queens should win easily. It just need to exchange queen for two knight twice, but neither program except QueeNy doesn't know that.

Wait a minute! It is not that easy. Against the 'naive' engines, that think Q >> 2N it is, because they handle the Knights careless. (Crafty even allows a Q vs 3N trade, which is one of the reasons it is especially bad at this.) But When 2N is set > Q, like in QueeNy, the program starts to avoid such trades, by keeping all its Knights doubly defended, or defended by a Pawn.

When QueeNy plays itself, it really has a very hard time to force the first Q vs 2N trade. Often it only happens because the N side voluntarily trades 2N for Q+P. I am not sure yet if this is good, or not. In any case it could be easily prevented by increasing the Knight value still further. Yet it seems that with only 6 Knights, you cannot avoid Q vs 2N trades forever, against an equally strong opponent. But the search depth is really important then. If the Knights side out-searches the Queens side, it will likely be able to avoid trading down. So although the Queens probably have a won position against 6 Knights, it is by no means a 'trivial win', which could already be won with a low search depth against perfect play. There is always very deep tactics in these positions. Often the engines start seeing the possibility for the forced trade only at d=15 (with R=2 NMP and all-non-capture LMR).

Of course, like Dan points out, the evaluation should really be handled by a material table; only the 5th and 6th Knight are worth that much (or alternatively and equivalently, the 3rd Queen that little). The trade 2nd Q vs 3rd and 4th Knight is probably already bad. QueeNy sometimes fails to win with the Queens because it to early engages in such a trade, and when the opponent still has 3 Pawns in the initial position it can then build a fortress. (This happened against Stockfish, and in a few other cases it could only crack the fortress with large difficulty.) It would be better to wait with that trade as the Queen side until you have gained one or more Pawns, or until the Pawn position has opened up. So when Q is kept fixed at 950, the 3rd and 4th Knight should add up to 900. rather than the 1000 for the 5th and 6th Knight. Then the Queens would only go for Q vs NNP trades.

Another flaw of QueeNy is that it does knot know that KNNK is draw; it therefore soemtimes offers an easy escape from KNNNNKQ or KNNNPKQ. This can of course also be handled with a material table.

If black gets an extra Knight, (on d8), the position is really heavily won for the Knights. It has become much easier for black to keep all its Knights doubly defended, and so avoid Q vs 2N trades. In addition, Q vs 2N trades won't save you anymore. You will end up in a Pawn ending with one Knight down, or, if you stop early enough, with one Queen against 3 Knights, which in a position with close-together Pawns and no passers seems badly lost for the Queen.