Possible search improvment

[bob]

If you have good move ordering, wouldn't the tactics all be near the top end of the search anyway?

No. Tactics are distributed throughout the tree, not just near the root.

[MattieShoes]

The problem is that it hardly takes any time to search moves that really are bad. And sometimes a move that looks really bad at 12 ply turns out to be a brilliant sacrifice at 13 ply...

Yes, I could see that happening. What if we optimized the depth so that this couldn't happen? Is it possible to do such a thing?

You can generalize what he's saying by saying a move that appears bad at ply N may prove to be good at ply N+1.

[MattieShoes]

If you have good move ordering, wouldn't the tactics all be near the top end of the search anyway?

[d]1r1r2k1/2p1qp1p/6p1/ppQB1b2/5Pn1/2R1P1P1/PP5P/R1B3K1 b

Best move here is Qe4. Figuring that out takes a search. If you've culled it, you'll never find it, but your opponent will. This happens to be the crux position, but a search 2 moves ago would have been hitting this same position, and the results of the Q sac could affect the score several moves ago. Tactics happen everywhere in the tree.

This is an extreme example, but there's lots of situations where, say, sacrificing an edge pawn results in eventual domination of the center and pressure on the king. Losing the pawn would be enough to push it out of your top 10. Also, if your opponent makes a move that you've not bothered to search previously, your subsequent search will be much slower because you don't have hash entries helping you order moves effeciently.