variations upto positions without any (more) captures

[tissatussa]

are those ever considered by coders ? then which engines are using such positions ? i imagine (only) those positions can truely be judged by their material and piece positions, otherwise material imbalance on a certain depth / ply can be temporary due to a unfinished capture sequence ?

i might not be using the right terms here, but i hope my question clear ..

[AAce3]

are those ever considered by coders ? then which engines are using such positions ? i imagine (only) those positions can truely be judged by their material and piece positions, otherwise material imbalance on a certain depth / ply can be temporary due to a unfinished capture sequence ?

i might not be using the right terms here, but i hope my question clear ..

https://www.chessprogramming.org/Quiescence_Search

[okidoki]

In addition to what has been share by AAce3, please do consider Quiescence-Search(QSearch) for the engine. It does make a lot of difference even if the depth searched is less and more number of nodes are searched as compared to one without QSearch.

IMO one typical flaw is to check for captures only sequences in QSearch. Although unfinished capture sequence(horizon effect) is bad one should consider all the moves which would lead to a non-quiet positions which are not captures. That being said "Quiet" is something which is not a standard and YMMV.

As an example if the search terminates by playing pawn on 7th rank, and there are no captures, then the win or loss will be dependent if that pawn can be either stopped from queening or be captured.

Thus one might also consider passed pawns moves at 7th-rank (again YMMV) as not quiet and consider them in QSearch. The downside of this is more moves would be explored in such cases.

Entire point I am trying to make here is horizon effect is not only due to unfinished capture sequences.

[hgm]

Searching Pawn pushes in QS is still safe, as these are also irreversible moves, and you will eventually run out of those, just as you will run out of material to capture. Whether it pays to do so depends on whether the moves really cause such a large score swing that the threat of them deserves to be called non-quiet. One could argue that creating a 5th-rank protected passer by pushing a Pawn forward already is valuable enough.

As to pushes to 7th rank... The problem there is that the opponent has to move first, before you get the chance to promote. And he might have no moves that are suitable for QS. You can of course already give a hefty bonus for being on 7th rank in the PST. (Micro-Max values a 7th-rank Pawn as 250 cP.) That ensures at least that the opponent cannot discard the line as 'uninteresting' (and reduce it by LMR to push a subsequent promotion over the horizon). But it might be too optimistic in case the Pawn can be easily blocked, and later gobbled up. If you don't give a hefty bonus, the opponent might simply stand pat in the move after you pushed to 7th rank, and you still would not see the promotion.

A better way would be to treat a Pawn on 7th the same way as delivering a check: search it even in QS, with a 1-ply extensions that forces the opponent to search all reply moves, so that you are sure to get a chance for trying out whether the actual promotion move (also always searched in QS) is any good.

[Mike Sherwin]

HGM, I wish to know what you think of my proposal that I have posted about a few times.

In my new engine if I ever get that far the plan is to call Qsearch(alpha, beta, 1, 1);

Each side gets a one time gen of non capture moves after the capture moves are tried.

I think it may solve many problems?