EconCSLib.SocialChoice.FairDivision.Divisible.Basic #
Foundational definitions for divisible goods (cake-cutting) allocation.
Main definitions #
Allocation N Ω— a function assigning each agent a subset of the cakeΩIsAllocation— predicate that an allocation is a measurable partition ofΩCakeValuation N Ω V— abstract valuation mapping cake-subsets to values inVIsNormalized— predicate that each agent values the whole cake at1MeasureValuation— standard valuation where each agent's value is a measureIsContiguousAllocation— each agent's piece is a (convex) interval inℝIsEnvyFree,IsProportional,IsEquitable— fairness predicates specialized to divisible allocations
Design #
The "cake" is an abstract type Ω with a MeasurableSpace Ω. This mirrors the indivisible
track (where G is an abstract type of goods), with Set Ω playing the role of Finset G.
Allocation is a plain function type alias; the partition predicate
IsAllocation is kept separate so that algorithms can manipulate raw allocations
before proving correctness. This follows the same separation principle as Allocation /
IsAllocation in SocialChoice.FairDivision.Indivisible.Basic.
CakeValuation N Ω V stays value-polymorphic because measure-based valuations naturally
take values in ENNReal. The bundled real-valued divisible cardinal interface lives in
Divisible.Instance.
MeasureValuation is the standard concrete case where each agent's value for a piece is
given by their personal (possibly non-atomic) measure. This corresponds to AdditiveValuation
in the indivisible track.
The fairness predicates IsEnvyFree, IsProportional, and IsEquitable are thin
specializations of the shared SocialChoice.FairDivision fairness layer to the share
type Set Ω.
References #
- Nisan et al., Algorithmic Game Theory, Chapter 13
- Procaccia, "Cake Cutting: Not Just Child's Play" (2013)
- Robertson–Webb, Cake-Cutting Algorithms (1998)
Fairness predicates #
Envy-free (EF): every agent weakly prefers their own piece over any other agent's piece.
∀ i j, μ_i(A_j) ≤ μ_i(A_i).
For divisible goods with non-atomic measures, EF allocations always exist. This is the key difference from the indivisible setting where EF need not exist.
Equations
- SocialChoice.FairDivision.Divisible.IsEnvyFree μ A = ∀ (i j : N), μ.val i (A j) ≤ μ.val i (A i)
Instances For
Proportional (PROP): each agent values their piece at least 1 / n of the whole cake.
Stated as μ_i(Ω) ≤ n * μ_i(A_i) to avoid division.
Equations
Instances For
Equitable: every agent assigns exactly the same value to their own piece.
Equations
- SocialChoice.FairDivision.Divisible.IsEquitable cv A = ∀ (i j : N), cv.val i (A i) = cv.val j (A j)
Instances For
EF implies proportional #
EF implies proportional for MeasureValuation partitions.