Dubins–Spanier Inductive Step
The Dubins–Spanier moving-knife algorithm builds a proportional allocation on $n$ agents by induction. The key construction is an inductive proposition
$$ P(n) \;\;\text{is}\;\; \bigl(\forall\ \mu : \mathrm{Fin}\ n \to \mathrm{Measure}\ I,\ \text{prop-conditions}\bigr) \;\Rightarrow\; \exists A : \mathrm{Allocation}\ (\mathrm{Fin}\ n)\ I,\ \mathrm{IsAllocation}\ A \wedge \mathrm{IsProportional}\ n\ \mu\ A, $$
proved by induction on $n$ in
SocialChoice.FairDivision.Divisible.dubinsSpanierProportional. The
proof uses three concrete pieces:
Algorithm shape
-
Threshold step. For each remaining agent $i$, find $t_i \in I$ with $\mu_i([0, t_i]) = \mu_i(I)/n$ via the IVT lemma ([[social_choice.fair_division.divisible.cut_exists]]).
-
Argmin selection. Let $i^* = \arg\min_i t_i$ and $t^* = t_{i^*}$. Assign agent $i^*$ the piece $[0, t^*]$.
-
Recurse. Restrict the remaining agents $j \ne i^*$ to $(t^*, 1]$ and apply the algorithm recursively with $n - 1$ agents.
Why proportionality is preserved
-
The selected agent. $\mu_{i^*}([0, t^*]) = \mu_{i^*}(I) / n$ by the threshold construction. ✓
-
Each remaining agent $j$. $t^* \le t_j$ by argmin, so $\mu_j([0, t^*]) \le \mu_j(I)/n$ by monotonicity of the CDF. Hence $$ \mu_j((t^*, 1]) \;=\; \mu_j(I) - \mu_j([0, t^*]) \;\ge\; \frac{n-1}{n}\,\mu_j(I). $$ Inductively, the recursive call on $n - 1$ agents on the remaining interval yields each $j$ a piece worth at least $1/(n-1)$ of their remaining value, which (combined with the bound above) is at least $\mu_j(I)/n$. ✓
Lean implementation notes
The Lean proof uses Fin.insertNth to fold the recursively-produced
sub-allocation back into a full allocation over Fin n. Partition
validity (measurability, disjointness, cover) and proportionality are
established by ennreal_prop_step — a packaged induction-step lemma
that combines the moving-knife construction with monotonicity bounds in
ENNReal.
All sorries are closed; the resulting public theorem is [[social_choice.fair_division.divisible.dubins_spanier_proportional]].
References
- Dubins, L. E. and Spanier, E. H. (1961). "How to Cut a Cake Fairly". Amer. Math. Monthly 68: 1–17.
- [AGT Chapter 13] Nisan, Roughgarden, Tardos, and Vazirani, Algorithmic Game Theory. Moving-knife procedures.