What does "fail low at the root" mean?

[watersky33]

I often see quotes such as "fail low at the root" is problematic, but what does that mean really? Why would I get into a situation that none of the root moves is good? I mean, one of the root moves must be the actual move. Thanks for some nice explanation.

[hyatt]

Most programs use an aspiration window at the root. If you complete iteration N with a score of something like +0.16, then when you start the next iteration, the search is more efficient if you start with alpha = 0.0, beta = 0.32 or something similar. Now your search can prune branches with scores < 0.0 or > 0.32, which speeds the search up. But on occasion that window is not wide enough and the score returns 0.0, which is not the real score, but the original lower bound you gave the search. You know the score is at least dropping to 0.0 and it might be dropping further, so you have to reduce the alpha bound and search again. Getting this 0.0 lower bound on the search is called "failing low at the root".

It is, in general, a bad thing to happen, because it means the score is going to be less than you expected.