The idea is to represent the values of a boolean function ƒ in a table of 2k rows, representing the possible values of the k first variables x1, ..., ,xk, and 2n−k columns representing the values of the other variables.
Let A1, ..., Ap be a partition of the rows of this table such that for i < p, |Ai| = s and | A p | = s ′ ≤ s {\displaystyle |A_{p}|=s'\leq s} . Let ƒi(x) = ƒ(x) iff x ∈ Ai.
Moreover, let B i , w {\displaystyle B_{i,w}} be the set of the columns whose intersection with A i {\displaystyle A_{i}} is w {\displaystyle w} .