MathLabs

Worked solution: The polynomial method resolves the cap set problem (Croot-Lev-Pach, Ellenberg-Gijswijt, 2016)

Step 3 of 7: The Croot-Lev-Pach rank lemma: vanishing polynomials are rare
In plain words

Suppose a bounded-degree polynomial PP vanishes on every combination αa+βb\alpha a+\beta b of two distinct points of a set AA. Croot, Lev and Pach's trick is to build an ∣A∣×∣A∣|A| \times |A| matrix out of the values Bab:=P(αa+βb)B_{ab} := P(\alpha a+\beta b): because PP vanishes off the diagonal, this matrix is essentially diagonal, and diagonal matrices with many nonzero entries need high rank -- but the specific algebraic structure of PP forces its rank to stay small no matter how large AA is.

∣{a∈A:P(−γa)≠0}∣≤2md/2|\{a \in A : P(-\gamma a) \ne 0\}| \le 2 m_{d/2}
Detailed analysis

Ellenberg and Gijswijt (2017, Proposition 2, generalising Lemma 1 of Croot-Lev-Pach 2016) fix Fq\mathbb{F}_q elements α,β,γ\alpha,\beta,\gamma with α+β+γ=0\alpha+\beta+\gamma=0. If P∈Sn≤dP \in S_n^{\le d} satisfies P(αa+βb)=0P(\alpha a+\beta b)=0 for every a≠b∈Aa \ne b \in A, then ∣{a∈A:P(−γa)≠0}∣≤2md/2|\{a \in A : P(-\gamma a) \ne 0\}| \le 2 m_{d/2}. The proof expands P(αx+βy)P(\alpha x+\beta y): every monomial of total degree at most dd has one of its two halves (xx-part or yy-part) of degree at most d/2d/2, so P(αx+βy)=∑m∈Mnd/2m(x)Fm(y)+∑m∈Mnd/2m(y)Gm(x)P(\alpha x+\beta y) = \sum_{m \in M_n^{d/2}} m(x)F_m(y) + \sum_{m \in M_n^{d/2}} m(y)G_m(x) for some polynomials Fm,GmF_m, G_m indexed by monomials mm of degree at most d/2d/2. Evaluating at x=a,y=b∈Ax=a, y=b \in A writes the matrix BB with entries Bab:=P(αa+βb)B_{ab} := P(\alpha a+\beta b) as a sum of 2md/22m_{d/2} rank-one matrices, so rank⁡(B)≤2md/2\operatorname{rank}(B) \le 2m_{d/2}.

Terms in this step
Rank of a matrix
The minimum number of rank-one matrices (outer products of a column and a row vector) needed to write a matrix as their sum; a diagonal matrix with kk nonzero entries has rank exactly kk.
Knowledge used in this step