Computer Chess Club Archives




Subject: Re: About False Fail Highs, professionals, and MTD searches

Author: Ulrich Tuerke

Date: 01:50:40 04/13/02

Go up one level in this thread

>Why don't MTD(n,f) searchers have trouble with this? If the
>PVS can have a fail high/fail low sequence, couldn't this
>happen with MTD(n,f) as well? As far as I see, the result
>would be that the MTD quits (it has converged, after all)
>and ends up with the move that corresponds to the false-fail
>high one in PVS. These are often blunders. Why doesn't this

as far as I understand (not too much), these search anomalies can't even
theoretically be a problem in MTD.
IMHO, the idea is: you start the search at some initial score with a zero
window. In case of a fail high, you repeat the search shifting the zero window
to larger values until a fail low happens. That's by definition the true score
and you may start the next iteration.
So, in my understanding, fail low in a fail high research (and vice versa) is a
very common thing in MTDF. It's just an indicator that the true score had been



This page took 0.23 seconds to execute

Last modified: Thu, 07 Jul 11 08:48:38 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.