Documentation

ToCslib.Probability.Kernel.Subprobability

Subprobability kernels #

IsSubprobabilityKernel κ states that every measure produced by κ has total mass at most one. It is the kernel-level mass discipline for a parameterized computation that may fail or diverge.

Mathlib's IsZeroOrMarkovKernel is different: it allows only kernels that are identically zero or have mass exactly one at every input. A subprobability kernel may lose a different, nontrivial amount of mass at each input.

The class is closed under the standard kernel operations used by probabilistic semantics. In particular, a subprobability kernel is finite and hence s-finite, so Mathlib's composition and product APIs apply without additional finiteness assumptions.

A kernel whose value at every input is a subprobability measure.

  • measure_univ_le' (a : α) : (κ a) Set.univ ≤ 1

    Every kernel value has total mass at most one.

Instances
    theorem ProbabilityTheory.Kernel.measure_univ_le {α : Type u} {β : Type v} [MeasurableSpace α] [MeasurableSpace β] (κ : Kernel α β) [IsSubprobabilityKernel κ] (a : α) :
    (κ a) Set.univ ≤ 1

    Every value of a subprobability kernel has total mass at most one.

    theorem ProbabilityTheory.Kernel.measure_le_one {α : Type u} {β : Type v} [MeasurableSpace α] [MeasurableSpace β] (κ : Kernel α β) [IsSubprobabilityKernel κ] (a : α) (s : Set β) :
    (κ a) s ≤ 1

    Every measurable event has mass at most one under a subprobability kernel.

    Applying a subprobability kernel produces a subprobability measure.

    @[instance 100]

    A subprobability kernel is uniformly finite, with bound one.

    @[instance 100]

    Every Markov kernel is a subprobability kernel.