DAUT

[Rein Halbersma]

In his book "Secrets of Practical Chess", John Nunn discusses the idea of Don't Analyse Unnecessary Tactics" (DAUT, p21). His idea is that humans shouldn't analyze a root node too deeply if they have a) a sound advantage, b) at least one promising move that already shows up after a shallow calculation, so that they c) can in all likelihood afford to skip possibly equivalent -but much deeper- tactical lines.

One could alternatively call this idea Multiple Reductions as opposed to Singular Extensions. It is somewhat reminiscent of Multi-Cut but then reversed in the side to move. The idea fits well into Nunn's overall philosophy of avoiding time trouble by playing a series of "good enough" moves until a critical decision really has to be taken.

I wonder if it would be feasible to test this idea within a chess program such as Stockfish that counts the tree sizes of root moves for move ordering. One could e.g. test after shallow searches if the top (say) 3 advantageous moves score within some margin of each other (this would require to do multi-pv at low search depths), with at least one root move having a significantly smaller tree size than the others.

Any comments?

Rein

[mcostalba]

Any comments?

Actually this is what PV-centric engines (like Stockfish) already do.

PV is searched much deeper and at a much greater detail then siblings move and is not easy for an engine to switch to another best move, normally this occurs not when engine finds a better alternative (as people could think) but infact when engine finds some unexpected problem with the supposedly best move (the PV fails low).

So I would say that this Nunn principle is already there in any PV-centric engine, like are almost all the top ones for which we have the sources and probably also for the others...