To find some fixes to the initial core growth problem we presented last time, let’s refine the split lemma as our first step. We show:
Lemma 1 (\(r\)-Split Lemma): let \( r \in [m] \) divide both \( m \in [n] \) and \( n=|X| \). For any \( {\mathcal F} \subset {X \choose m} \), there exists an \( r \)-split \( {\boldsymbol X}=(X_1, X_2, \ldots, X_r) \) and subfamily \( {\mathcal F}’ \) of cardinality at least \( {n/r \choose m/r}^r |{\mathcal F}| \Big/ {n \choose m} \) such that \[ |U \cap X_j| = \frac{m}{r}, \quad \textrm{for all}~ U \in {\mathcal F}’ ~\textrm{and}~ X_j. \]
Here an \( r \)-split is a natural generalization of an \( m \)-split, so \( X_j \) are mutually disjoint \( n/r \)-sets as defined last time. The claim generalizes the original split lemma that finds \( |{\mathcal F}’| \ge \left( \frac{n}{m} \right)^m |{\mathcal F}| \Big/ {n \choose m} > |{\mathcal F}| e^{-m}, \) which is given on p.2 of Paper 1 ver. 1 on the left. Let’s say such an \( {\mathcal F}’ \) is on the \(r\)-split \({\boldsymbol X} \) as well.
Corollary 3.2 on p. 11 of Paper 3 further generalizes this to a primitively weighted \( X \). There, the primitive norm \( \| \cdot \| \) is defined just as the 1/15/24 post by a one-parameter weight function \( w_* \). The whole Section 3 of the paper is about splitting the \(X \) according to a given \( {\mathcal F} \). This could be a tool set for finding some topological properties of \(2^X \).
Our proof of the lemma uses the same technique as the original split lemma, described on the extra page of the 3/3/23 post. The claim given there finds the next \( X_j \) such that the incident subfamily \( {\mathcal F}_{\boldsymbol X} \subset {\mathcal F} \) is large enough meeting \( |U \cap X_i| = 1 \) for all \( i \le j \). For Lemma 1, we change its RHS into \( m/r \).
Let’s see how Paper 3 finds the first \( n/r \)-set \( X_1 \) in the last paragraph of p. 10, where \[ q= \frac{m}{r}, \quad d= \frac{n}{r}, \quad j=0, \quad {\boldsymbol X} = ( \emptyset ), \quad {\mathcal F}_{\boldsymbol X} = {\mathcal F}, \quad \textrm{and} \quad \gamma_{\boldsymbol X} = e^{-\kappa({\mathcal F})}. \]
It is rephrased as follows:
The factor \( \prod_{i=0}^{r-1} {n-di \choose d } \) is the number of \(r \)-splits \({\boldsymbol X} \), so there must be a subfamily \( {\mathcal F}’ \subset {\mathcal F} \) on some \({\boldsymbol X} \) such that \( |{\mathcal F}’ | \ge {d \choose q}^r e^{-\kappa({\mathcal F})} = {d \choose q}^r |{\mathcal F}| \Big/ {n \choose m}. \)
We’ll detail the induction step for any \( j \) to finish proving Lemma 1 in the next post.
Links for mobile devices