Local Beam Search

Local beam search tracks k states at once, keeping the best successors each step.

How does local beam search differ from running k independent hill climbers?