If you promote to Q, he would play Rf7+, and have a perpetual on f7/f8 unless you abandon the Q. If you do the latter he would have either his two Rooks attacking the singly defended Q, or retaliate against Rb7 when you capture with the Rook. So that would just lose a Pawn.
By promoting to N he can interpose the Knight on e7 after Rf7+. In the short run that gains him a Knight, which might be valued more than a 7th-rank Pawn..
rvida wrote:Critter played c8=N here and eventually won, but the move choice seems odd. It sees a better move only after half minute.
c8=N is not odd at all. It is the only promotion out of four possible ones that does not lead to a draw immediately. The reason is that all other promotions are answered by Rf7+, you can easily see what happens then. (In the meantime while I wrote this answer HGM has already explained it! )
To find out that other moves, like Rb4 (or Rb1, or Re3) are even better than the winning move c8=N seems to require some deeper search, so it does not surprise me at all that Rb4 is not displayed as best move immediately. I analyzed the position a little bit with the help of free Rybka 2.2n2 32bit (I don't own Rybka 3) and found that after a move like Rb1, it takes about 18 plies until White can finally promote and win a Rook. That c8=N gives an advantage of N vs P can be seen much faster.
hgm wrote:If you promote to Q, he would play Rf7+, and have a perpetual on f7/f8 unless you abandon the Q. If you do the latter he would have either his two Rooks attacking the singly defended Q, or retaliate against Rb7 when you capture with the Rook. So that would just lose a Pawn.
Thanks, now I understand.
hgm wrote:
In the short run that gains him a Knight, which might be valued more than a 7th-rank Pawn..
Indeed, this is the cause. The passed pawn is valued only 3.26. After promotion it is: -1.00 for loosing a pawn, +3.25 for gaining a knight, +0.87 bonus for being a minor piece up for a total of 3.12. After the knight gains some mobility and threatens black king, its value outweights the value of the passer.
After tweaking/increasing passed pawn evaluation now it switches to a different move much faster:
hgm wrote:
In the short run that gains him a Knight, which might be valued more than a 7th-rank Pawn..
Indeed, this is the cause. The passed pawn is valued only 3.26. After promotion it is: -1.00 for loosing a pawn, +3.25 for gaining a knight, +0.87 bonus for being a minor piece up for a total of 3.12. After the knight gains some mobility and threatens black king, its value outweights the value of the passer.
After tweaking/increasing passed pawn evaluation now it switches to a different move much faster:
You can easily and quickly go into a wrong direction by
a) tuning evaluation based on one single position,
b) trying to increase passed pawn eval such that the engine prefers "infinitely" having a 7th rank passer on the board over actually promoting it.
It may be right that underpromotion to a knight is not the major goal of having a passer since usually it will be promoted to a queen. But I doubt that a very high static bonus for 7th rank passers is always adequate, since you really don't know whether the pawn can promote and win a huge amount of material or not. To decide about this is more tactics than positional evaluation IMO, so the search should find that out.
)