A relaxation term is used to govern the speen of adadptation of many kinds of iterative algorithm. These algorithms are often have the form:
new_value = old_value + λ change_direction.
The term λ is the relaxation term. If this is too large the algorithm is unstable swinging wildly between different values, if it is too small the algorithm takes a very ling time to converge. In some algoruthms, notably simulated annealing the relaxation terms starts relatively large (but not so large to cause instability) to allow rapid movement towards the general site of a solutuon, and then gradually reduced to allow fine tuning.
Used on page 116