Weighted finite collections and certificate mass decomposition #
Weighted collections and the positive/negative mass decomposition of a dual certificate for the low-frequency construction and recombination pipeline.
Weighted finite collections #
A weighted finite collection of subsets of U.
The weights are nonneg reals with positive total weight, so the collection
represents a (possibly unnormalized) distribution over finsets.
This avoids the need for literal multisets (which would require rational certificates and denominator clearing).
- J : Type v
Finite index type.
Fintype instance for J.
- decJ : DecidableEq self.J
DecidableEq instance for J.
The indexed family of sets.
Weight of each index.
Weights are nonnegative.
Weights sum to a positive value (total mass).
Instances For
Total weight of a weighted collection.
Equations
- C.totalWeight = ∑ j : C.J, C.weight j
Instances For
Item frequency: the weighted fraction of sets containing item i.
Instances For
Weighted average deficit.
Equations
- C.avgDeficit f M = (∑ j : C.J, C.weight j * KaltonRoberts.deficit f M (C.sets j)) / C.totalWeight
Instances For
Weighted average surplus.
Equations
- C.avgSurplus f M = (∑ j : C.J, C.weight j * KaltonRoberts.surplus f M (C.sets j)) / C.totalWeight
Instances For
Item frequency is nonneg.
Item frequency is at most 1.
Certificate mass decomposition #
Positive mass of a dual certificate: p = ∑_S max(λ(S), 0).
Instances For
Positive mass is nonneg.
Negative mass is nonneg.
p ≤ 1
q ≤ 1
Positive mass is positive when there is a positively-weighted set.
Negative mass is positive when there is a negatively-weighted set.
Zero-marginal frequency identity #
Item frequency bound: for each item i, the sum of positive weights
over sets containing i is at most the negative mass q.
Symmetric bound: for each item i, the sum of negative weights
over sets containing i is at most the positive mass p.
Positive and negative weighted collections from a certificate #
The positive weighted collection from a dual certificate.
Equations
- One or more equations did not get rendered due to their size.
Instances For
The negative weighted collection from a dual certificate.
Equations
- One or more equations did not get rendered due to their size.
Instances For
The total weight of the positive collection is p.
The total weight of the negative collection is q.
Frequency and deficit bounds for certificate collections #
Swapping lemma: ensure q ≤ 1/2 #
By possibly replacing f by −f, we may arrange q ≤ 1/2 ≤ p.
Augmented collections (Lemma 2.3 of the paper) #
The augmented positive collection A consists of:
- Positive active sets P (with weight λ⁺(P))
- Complements Nᶜ of negative active sets N (with weight λ⁻(N))
Total weight = p + q = 1. Every item has frequency exactly q (by the zero-marginal property). Average deficit ≤ q(1 − u).
Symmetrically for the augmented negative collection B.
The augmented positive collection from a dual certificate (Lemma 2.3). Includes positive active sets and complements of negative active sets.
Equations
- One or more equations did not get rendered due to their size.
Instances For
The augmented negative collection from a dual certificate (Lemma 2.3). Includes negative active sets and complements of positive active sets.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Total weight of the augmented positive collection is 1.
Total weight of the augmented negative collection is 1.
Uniform weighted collection from a finite family #
Build a WeightedCollection with uniform weight 1 from a finite family
of subsets indexed by a type J with positive cardinality.
Equations
- KaltonRoberts.WeightedCollection.uniformOfFamily sets hJ = { J := J, finJ := inst✝¹, decJ := inst✝, sets := sets, weight := fun (x : J) => 1, weight_nonneg := ⋯, total_pos := ⋯ }
Instances For
Total weight of a uniform collection equals the cardinality.
Item frequency in a uniform collection equals
card {j | i ∈ sets j} / card J.
Average deficit of a uniform collection equals
(∑ j, deficit f M (sets j)) / card J.
Average surplus of a uniform collection equals
(∑ j, surplus f M (sets j)) / card J.