Category of Markov kernels¶

A category whose objects are measurable spaces and whose morphisms are Markov kernels, composed by integrating one conditional probability kernel through another.

Version: v2 · 2026-09-19 · History
Domain-specific #: 3615
Domain group: Formal Sciences
Origin domain: Mathematics
Subdomain: Categorical Probability → Mathematics
Aliases: Stoch

Core Idea¶

Variants use probability subprobability or s-finite kernels and may quotient by almost-sure equality, composition requires measurability and Tonelli-style conditions, and deterministic functions embed only through Dirac kernels. A kernel maps each input point to a probability measure on the output space measurably; composing kernels marginalizes the intermediate state, while identity arrows are Dirac measures. The abstraction is therefore identified by a declared carrier, a transformation or constraint over that carrier, and an invariant that tells an analyst whether the named structure is genuinely present.

How would you explain it like I'm…

Spinner Machines

Imagine a spinner for every square on a game board: from each square, your spinner tells how likely you are to land on each next square. Using one spinner and then another is like one bigger spinner that adds up every in-between square you might have passed through. The collection of all these spinners, and the rule for chaining them, is the category of Markov kernels.

Chaining Chance Machines

A Markov kernel is like a random machine: for each input, it gives chances for each possible output. You can connect two random machines, and to find the chances of the final output you add up over every possible middle result. The machine that always returns its input unchanged works as the "do nothing" machine. Ordinary, non-random rules fit in too, as machines that give one output with 100% chance. The category of Markov kernels is the collection of all these spaces and random machines with these connecting rules.

Probability Transition Arrows

In the category of Markov kernels, the objects are measurable spaces (sets where you can assign probabilities to suitable subsets), and an arrow from X to Y is a Markov kernel: a rule that assigns to each point of X a probability distribution on Y, in a measurable way. Composing two kernels means going from X to Y at random and then from Y to Z at random, and integrating (averaging) over all the intermediate points in Y. The identity arrow sends each point to a Dirac measure — the distribution that is certain to stay at that point. An ordinary deterministic function becomes a kernel by sending each point to the Dirac measure at its image. There are variants: some allow subprobability or s-finite kernels instead of probability kernels, and some treat kernels as equal if they agree almost surely; you must state which you're using.

The category of Markov kernels takes measurable spaces as objects and Markov kernels as morphisms. A kernel k: X → Y is a function k(x, B) from X times the σ-algebra of Y to the nonnegative reals such that for each x, k(x, ·) is a probability measure, and for each measurable B, x ↦ k(x, B) is measurable. Composition marginalizes the intermediate state: (l ∘ k)(x, C) = ∫_Y l(y, C) k(x, dy), and verifying that this is well defined and associative uses measurability and Tonelli/Fubini-style arguments. Identities are Dirac kernels, δ_x(B) = 1 if x ∈ B and 0 otherwise, and they satisfy the unit laws. Deterministic measurable functions f embed only through their Dirac kernels x ↦ δ_{f(x)}. Variants replace probability kernels with subprobability or s-finite kernels, may quotient by almost-sure equality, and may add a monoidal product (product of kernels) when needed, so the exact variant must be made explicit.

Scope of Application¶

Category of Markov kernels belongs to categorical probability and is useful where the analyst can specify the typed categorical probability carrier, including objects, relations, parameters, conventions, evidence, boundaries, and comparison targets, then evaluate the objects as measurable spaces, morphism kernel k from X times sigma-algebra of Y to nonnegative reals, probability normalization and measurability conditions, Dirac identity kernel, composition integral over the intermediate space, associativity and unit laws, deterministic measurable-function embedding, monoidal product when used and probability subprobability or s-finite variants are explicit.

Clarity¶

The abstraction clarifies a crowded vocabulary by making the objects as measurable spaces, morphism kernel k from X times sigma-algebra of Y to nonnegative reals, probability normalization and measurability conditions, Dirac identity kernel, composition integral over the intermediate space, associativity and unit laws, deterministic measurable-function embedding, monoidal product when used and probability subprobability or s-finite variants are explicit the center of the account. A claim should name the carrier, the governing operation or relation, the applicable assumptions, and the recognition test.

Manages Complexity¶

Without the abstraction, an analyst must reason directly over many local details: the carrier roles, admissibility assumptions, competing conventions, derived invariants, boundary cases, and proof or validation obligations specific to Category of Markov kernels. Category of Markov kernels compresses them into the roles in the structural signature. That compression permits comparison across instances without erasing the variables that determine validity. It also exposes which details may be varied safely and which are constitutive.

Abstract Reasoning¶

Identify the carrier. State what the elements, states, objects, or observations are: the typed categorical probability carrier, including objects, relations, parameters, conventions, evidence, boundaries, and comparison targets. Reject examples whose alleged carrier belongs to a different problem.
Lock the constitutive rule. Express the objects as measurable spaces, morphism kernel k from X times sigma-algebra of Y to nonnegative reals, probability normalization and measurability conditions, Dirac identity kernel, composition integral over the intermediate space, associativity and unit laws, deterministic measurable-function embedding, monoidal product when used and probability subprobability or s-finite variants are explicit independently of one notation or implementation.

Knowledge Transfer¶

Knowledge transfers strongly among subfields of categorical probability because they reuse the typed categorical probability carrier, including objects, relations, parameters, conventions, evidence, boundaries, and comparison targets, A kernel maps each input point to a probability measure on the output space measurably; composing kernels marginalizes the intermediate state, while identity arrows are Dirac measures., and type the carrier, state every parameter and convention in the definition, test that the objects as measurable spaces, morphism kernel k from X times sigma-algebra of Y to nonnegative reals, probability normalization and measurability conditions, Dirac identity kernel, composition integral over the intermediate space, associativity and unit laws, deterministic measurable-function embedding, monoidal product when used and probability subprobability or s-finite variants are explicit, compare the nearest accepted identity, and report counterexamples, uncertainty, and limiting cases.

Relationships to Other Abstractions¶

Current abstraction Category of Markov kernels Domain-specific

Parents (1) — outgoing typed relations

Category of Markov kernels is a kind of Relation Prime

It remains its own entry because its identity is fixed by the objects as measurable spaces, morphism kernel k from X times sigma-algebra of Y to nonnegative reals, probability normalization and measurability conditions, Dirac identity kernel, composition integral over the intermediate space, associativity and unit laws, deterministic measurable-function embedding, monoidal product when used and probability subprobability or s-finite variants.

Hierarchy path (1) — routes to 1 parentless root

Category of Markov kernels → is a kind of Relation

Neighborhood in Abstraction Space¶

Category of Markov kernels sits in a crowded region of the domain-specific corpus (15^th percentile for distinctiveness): several abstractions share nearly its structure, so a description that fits it tends to fit its neighbors too.

Family — Markov Processes & Random Generation (8 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-10-25