LDS

The LDS algorithm (due to Ginsberg and Harvey) is a ``Limited Discrepancy Search'': it is an iterative depth first search, with the cutoff for each iteration occurring when the current path contains more than k discrepancies from the best heuristic value.

Thus, the k^th step of the iteration is roughly

    LDS(k)(var, discrepancies):
      if discrepancies > k
        return
      if var >= nvars
        (a total assignment--do whatever)
	...
	return
      Assign the value h suggested by the heuristic to var
      LDS(k)(var + 1, discrepancies)
      For each other value c that var can take on
        Assign c to var
	LDS(k)(var + 1, discrepancies + 1)

Need more information? Here are two papers by Will Harvey on LDS. The first is an IJCAI article about LDS with Matt Ginsberg. The second is Will's Ph.D. Dissertation.