Subprobability measures #
IsSubprobabilityMeasure μ states that μ Set.univ ≤ 1. It is the mass discipline of a
computation that may fail to return: the defect 1 - μ Set.univ is the mass that never reached a
value, so a returned result and a failure stay distinguishable without an Option wrapper.
Mathlib has no such class. IsZeroOrProbabilityMeasure is a different condition — it admits only
total mass 0 or 1, and so cannot describe a computation that fails with probability strictly
between the two.
The class has automatic instances for dirac, products, and map. A nonmeasurable pushforward of a
nonzero measure is a Dirac mass, which still has mass at most one. Bind preserves the bound
for an almost-everywhere measurable subprobability family. Exact mass preservation also
requires the relevant measurability proofs.
A measure with total mass at most one.
The total mass of a subprobability measure is at most one.
Instances
The total mass of a subprobability measure is at most one.
Every set has measure at most one under a subprobability measure.
The mass a subprobability measure does not assign to any value.
Instances For
The real-valued defect is one minus the real-valued total mass.
Independent products preserve the subprobability bound.
Pushing a subprobability measure forward along a measurable function preserves its mass bound. Stated as a theorem because the measurability hypothesis cannot be synthesised.
Pushforward preserves the subprobability bound, including the Dirac fallback for a nonmeasurable map from a nonzero measure.
Giry bind of a subprobability family against a subprobability measure stays a
subprobability measure. The measurability hypothesis is the one Measure.bind_apply needs.