Stromquist Unique-Preference Sets B(i, j)
A strengthening of the preference sets ([[social_choice.fair_division.divisible.stromquist_preference_sets]]): $B(i, j)$ collects the divisions where piece $i$ is agent $j$'s unique value-maximizer: $$ B(i, j) \;=\; \{x \in S \mid \forall k \ne i,\ v(x, j, k) < v(x, j, i)\}. $$
In Lean: strom_B i j.
Openness
strom_B_open: $B(i, j)$ is open in $\mathrm{Fin}\ n \to \mathbb{R}$.
Proof
The defining condition is a conjunction of $n - 1$ strict inequalities, each of which cuts out an open half-space because both sides are continuous ([[social_choice.fair_division.divisible.stromquist_value_continuous]]). A finite intersection of open sets is open.
Pairwise disjoint across pieces at a fixed agent
strom_B_disjoint_over_i: for fixed $j$, the family
$\{B(i, j) : i \in \mathrm{Fin}\ n\}$ is pairwise disjoint.
Proof. If $x \in B(i_1, j) \cap B(i_2, j)$ with $i_1 \ne i_2$, then agent $j$ has piece $i_1$ as their unique value-maximizer (the $B(i_1, j)$ membership) and piece $i_2$ as their unique value-maximizer (the $B(i_2, j)$ membership). Two distinct pieces cannot both be a unique maximizer, contradiction.
The intuitive content: each agent uniquely prefers at most one piece at any given division.
Role in the proof
The KKM-style argument in Stromquist's proof needs open covers (for compactness) and a way to extract a bijection between pieces and agents (for the EF assignment). The pair $(\text{open } B(i, j),\ \text{disjoint across } j)$ is exactly this data:
- Openness feeds the KKM lemma.
- Disjointness across pieces at a fixed agent makes the assignment function $\text{piece} \to \text{agent}$ injective: if two pieces had the same claimant agent, that agent would uniquely prefer both, a contradiction. Injectivity plus counting then forces the assignment to be a bijection.
The aggregated agent-preference union ([[social_choice.fair_division.divisible.stromquist_U]]) takes $U(i) = \bigcup_j B(i, j)$, which inherits openness from each $B(i, j)$.
References
- Stromquist, W. (1980). "How to Cut a Cake Fairly". Amer. Math. Monthly 87: 640–644.