This normalization is done in order to make the operators behave more like the
original belsum and belwsum operators, which are both normalized. One advantage
of the normalization is that it allows us to describe the belief computation
of these operators in terms of various types of means (averages). For example,
belsum computes the arithmetic mean over the beliefs of the parent nodes,
whereas belwsum computes a weighted arithmetic mean. Similarly, belcombine and
belwand compute a geometric mean and weighted geometric mean, respectively.