Small representative sets for Boolean sumsets #
For arbitrary subsets A and B of the Boolean cube, choose A₀ ⊆ A and B₀ ⊆ B such that
A + B = (A₀ + B) ∪ (A + B₀) and |A₀| + |B₀| ≤ 3 ∑_{j ≤ n/3} binom(n,j).
Here addition of bit strings is coordinatewise XOR. There is no degree or circuit-size
assumption on A and B.
This is the case q = 2 (the field 𝔽₂) of [Ellenberg2017Sumsets, Theorem 1, pp. 1–2]. The proof
uses the general image-sparsification theorem with restricted low-degree evaluation spaces.
Its input is the common slice-space containment of the polynomial pullbacks; we do not use or
formalize the theorem of Meshulam used in the published proof. All choices are nonuniform.
Main results #
Cslib.BitString.exists_add_cover_of_degree: an asymmetric, degree-parameterized bound.Cslib.BitString.exists_add_cover: the3times low-degree monomial-count bound.
References #
- J. S. Ellenberg, Sumsets as unions of sumsets of subsets, arXiv:1612.01929v2, Theorem 1 and its proof, pp. 1–3.
Restricting the degree-p and degree-q evaluation spaces to the two summands yields an
exact sumset cover with budget D(p) + D(q) + 2^n - D(p+q+1). Parentheses around the
codimension keep the subtraction in natural numbers local to that term.
Ellenberg's sumset cover on the Boolean cube. At most three times the number of
degree-⌊n/3⌋ monomials suffice to generate the whole sumset using translates of the original
summands. This includes empty summands and the zero-dimensional cube.