Measure Disintegration & Applications

• Disintegration of measures is the process of decomposing a measure into a family of conditional measures concentrated on the fibers of a measurable map.
• Its formulation supports applications in ergodic decomposition, Bayesian inversion, and optimal transport through the use of measurable kernels.
• The theory rigorously links measure theory with geometry and dynamics, offering concrete insights into conditional probabilities and system classifications.

Searching arXiv for recent and foundational papers on disintegration of measures, conditional measures, and related geometric/dynamical formulations. Disintegration of measures is the decomposition of a measure $\mu$ along the fibers of a measurable map $\pi:X\to Y$ into a measurable family of conditional measures $\{\mu_y\}_{y\in Y}$ such that $\mu_y$ is concentrated on $\pi^{-1}(y)$ for $\pi_\#\mu$-almost every $y$ and

$$\int_X f(x)\,d\mu(x)=\int_Y\left(\int_X f(x)\,d\mu_y(x)\right)\,d(\pi_\#\mu)(y)$$

for every integrable $f$. In standard Borel settings this is the measure-theoretic content of regular conditional probability, and in modern work it also appears as a structural tool in ergodic decomposition, optimal transport, foliation theory, Bayesian inversion, Gaussian conditioning, and dynamical systems (Cho et al., 2017, Possobon et al., 2022).

1. Classical formulation and regular conditional probabilities

In its classical form, disintegration starts with a measurable map $T:X\to Y$ between standard Borel spaces and a probability measure $\pi:X\to Y$0 on $\pi:X\to Y$1. Writing $\pi:X\to Y$2, Rokhlin’s theorem yields a measurable family $\pi:X\to Y$3 such that $\pi:X\to Y$4 is supported on $\pi:X\to Y$5 for $\pi:X\to Y$6-almost every $\pi:X\to Y$7, the map $\pi:X\to Y$8 is measurable for every measurable $\pi:X\to Y$9, and

$\{\mu_y\}_{y\in Y}$0

Equivalently, for integrable $\{\mu_y\}_{y\in Y}$1,

$\{\mu_y\}_{y\in Y}$2

This is the precise sense in which disintegration identifies conditional laws on fibers (Possobon et al., 2022).

The same construction is often expressed in the language of kernels. For a joint law $\{\mu_y\}_{y\in Y}$3 on $\{\mu_y\}_{y\in Y}$4 with $\{\mu_y\}_{y\in Y}$5-marginal $\{\mu_y\}_{y\in Y}$6, a disintegration with respect to $\{\mu_y\}_{y\in Y}$7 is a kernel $\{\mu_y\}_{y\in Y}$8 such that

$\{\mu_y\}_{y\in Y}$9

This kernel is a regular conditional probability of $\mu_y$0 given $\mu_y$1, unique $\mu_y$2-almost everywhere. In Bayesian language, if a prior $\mu_y$3 on $\mu_y$4 and a channel $\mu_y$5 generate a joint law

$\mu_y$6

then a Bayesian inversion $\mu_y$7 is exactly a disintegration of $\mu_y$8 in the reverse direction (Cho et al., 2017).

Several later works emphasize that existence and uniqueness beyond probability measures require explicit hypotheses. For $\mu_y$9-finite Radon measures, existence is obtained under assumptions such as metrizability of the source, countable generation of the target $\pi^{-1}(y)$0-algebra, and $\pi^{-1}(y)$1-finiteness of the pushforward; uniqueness remains only almost-everywhere (Costa et al., 1 Aug 2025). Outside standard Borel or comparable Radon settings, disintegrations may fail to exist, may fail to be measurable, or may cease to be unique in any pointwise sense (Cho et al., 2017).

2. Structural variants: ergodic decomposition, systems of measures, and canonical models

A major structural use of disintegration is ergodic decomposition. For a Polish locally compact group $\pi^{-1}(y)$2 acting on a Polish space $\pi^{-1}(y)$3 with an invariant probability $\pi^{-1}(y)$4, there exists a Borel map $\pi^{-1}(y)$5 such that $\pi^{-1}(y)$6 is an ergodic invariant measure, $\pi^{-1}(y)$7, and

$\pi^{-1}(y)$8

in the weak-* sense, where $\pi^{-1}(y)$9. This barycentric identity lifts from measures to $\pi_\#\mu$0-spaces:

$\pi_\#\mu$1

and the Koopman representation disintegrates as an $\pi_\#\mu$2-direct integral of order indecomposable representations (Jeu et al., 2015). In this form, disintegration is not merely a conditional probability device; it is a representation-theoretic decomposition principle.

A second structural framework is the theory of systems of measures on a map $\pi_\#\mu$3. A system $\pi_\#\mu$4 is a family of measures concentrated on fibers $\pi_\#\mu$5. The distinction between Borel systems of measures and continuous systems of measures is central: in the former, $\pi_\#\mu$6 is Borel for every Borel set $\pi_\#\mu$7; in the latter, $\pi_\#\mu$8 is continuous for every $\pi_\#\mu$9. This framework is stable under composition, lifting, and fibred products, and was developed partly to support categorical constructions for topological groupoids and Haar systems (Censor et al., 2010).

A third extension replaces standard Borel spaces by canonical compact Hausdorff models of probability algebras. In that setting, a morphism of probability algebras $y$0 induces a continuous map $y$1, and there exists a unique family of Radon probability measures $y$2 such that $y$3 is continuous for every $y$4, $y$5 is supported on the fiber $y$6, and the disintegration identity holds pointwise rather than merely almost everywhere (Jamneshan et al., 2020). This replaces the usual measurable-selection viewpoint by a functorial compact-Hausdorff model.

3. Transport, Wasserstein geometry, and disintegration maps

In optimal transport, disintegration of a plan $y$7 with respect to its first marginal takes the form

$y$8

where $y$9 is the disintegration map. This makes every transport plan a measurable field of conditional probabilities on the target. The paper on transport classes defines an equivalence relation on plans by

$$\int_X f(x)\,d\mu(x)=\int_Y\left(\int_X f(x)\,d\mu_y(x)\right)\,d(\pi_\#\mu)(y)$$0

where $$\int_X f(x)\,d\mu(x)=\int_Y\left(\int_X f(x)\,d\mu_y(x)\right)\,d(\pi_\#\mu)(y)$$1 and $$\int_X f(x)\,d\mu(x)=\int_Y\left(\int_X f(x)\,d\mu_y(x)\right)\,d(\pi_\#\mu)(y)$$2. The pushforward $$\int_X f(x)\,d\mu(x)=\int_Y\left(\int_X f(x)\,d\mu_y(x)\right)\,d(\pi_\#\mu)(y)$$3 records the law of conditional measures and determines the second marginal through the barycentric constraint

$$\int_X f(x)\,d\mu(x)=\int_Y\left(\int_X f(x)\,d\mu_y(x)\right)\,d(\pi_\#\mu)(y)$$4

Fixing such a transport class turns the constrained Kantorovich problem into an abstract Monge problem on $$\int_X f(x)\,d\mu(x)=\int_Y\left(\int_X f(x)\,d\mu_y(x)\right)\,d(\pi_\#\mu)(y)$$5 with lifted cost

$$\int_X f(x)\,d\mu(x)=\int_Y\left(\int_X f(x)\,d\mu_y(x)\right)\,d(\pi_\#\mu)(y)$$6

so that the Monge case corresponds to the special class where $$\int_X f(x)\,d\mu(x)=\int_Y\left(\int_X f(x)\,d\mu_y(x)\right)\,d(\pi_\#\mu)(y)$$7 (Granieri et al., 2012).

A different geometric viewpoint studies disintegration maps themselves as maps into Wasserstein space. For a Borel map $$\int_X f(x)\,d\mu(x)=\int_Y\left(\int_X f(x)\,d\mu_y(x)\right)\,d(\pi_\#\mu)(y)$$8 and disintegration $$\int_X f(x)\,d\mu(x)=\int_Y\left(\int_X f(x)\,d\mu_y(x)\right)\,d(\pi_\#\mu)(y)$$9, one defines

$f$0

Under local compactness and separability hypotheses, this map is Borel, and if $f$1 it is nearly weakly continuous in the sense of Lusin. Under stronger geometric hypotheses, such as bijective fibers or metric measure foliations, $f$2 becomes weakly continuous or even an isometry (Possobon et al., 2022). The same framework defines transport classes through the pushforward $f$3 and reformulates constrained transport as a Monge problem from $f$4 to $f$5.

Recent work sharpens this geometry by comparing how supports of conditional measures move in $f$6 to how the measures move in Wasserstein space. For a disintegration map $f$7, the derivative-like quantity

$f$8

leads to an energy $f$9. Under full-support assumptions on the conditionals, the condition $T:X\to Y$0 is equivalent to the fibers forming a $T:X\to Y$1-metric measure foliation, that is,

$T:X\to Y$2

for all $T:X\to Y$3 (Münch et al., 17 Sep 2025). This suggests that disintegration maps can classify when a family of conditional measures is geometrically induced by a foliation.

A separate generalization replaces measurable functions by measurable multifunctions $T:X\to Y$4. In that setting, fibers $T:X\to Y$5 may overlap, and disintegration exists precisely when the closed graph of $T:X\to Y$6 satisfies the mass inequality

$T:X\to Y$7

This coupling-based criterion yields a disintegration theorem for multifunctions and leads to the notion of asymptotic disintegrability of metric Polish probability spaces (Fill et al., 1 Jul 2025).

4. Dynamical disintegration: regularity, atomicity, entropy, and dimension

In dynamical systems, disintegration is frequently taken along invariant foliations or factor maps. For hyperbolic skew products

$T:X\to Y$8

with expanding base and contracting fiber, disintegration of the SRB measure along stable fibers gives conditional measures $T:X\to Y$9 and a quotient observable

$\pi:X\to Y$00

Under bounded distortion and contraction assumptions, this operator preserves Hölder regularity and, in a stronger setting, $\pi:X\to Y$01 regularity:

$\pi:X\to Y$02

and for $\pi:X\to Y$03 one has $\pi:X\to Y$04 with an explicit derivative bound (Butterley et al., 2015). Here disintegration functions as a regular quotienting mechanism rather than a singular decomposition.

The opposite regime is atomic disintegration. For locally constant skew-products over the full shift with circle fiber maps,

$\pi:X\to Y$05

a measure with periodic repetitive pattern is obtained as a weak limit of equidistributed periodic measures whose words satisfy a summable noise condition

$\pi:X\to Y$06

The main theorem proves that such a limit measure $\pi:X\to Y$07 has atoms in its disintegration along central fibers, and if $\pi:X\to Y$08 is ergodic then the disintegration is atomic almost everywhere (Santiago et al., 2020). The proof combines Feldman–Katok control of the projected base dynamics, zero entropy of the base limit, and a contraction-box argument that traps a positive proportion of mass on a countable fiber set.

A related but sharper phenomenon occurs for accessible derived-from-Anosov diffeomorphisms on $\pi:X\to Y$09. If Lebesgue measure has atomic disintegration along the center foliation, then the atomicity is mono-atomic: there is exactly one atom per leaf. The paper further gives an open nonempty class where the center exponent is negative while the linearization has positive center exponent, and in that class the center disintegration of Lebesgue measure is atomic, hence mono-atomic (Ponce et al., 2013). This shows that central atomicity can be robust in partially hyperbolic settings without being reducible to skew-product structure.

Disintegration also supports finer geometric invariants. For $\pi:X\to Y$10-stationary measures on flag manifolds, disintegration along the one-dimensional foliation $\pi:X\to Y$11 yields conditional measures $\pi:X\to Y$12 on circle fibers. Under ergodicity, a moment condition, and a uniqueness assumption on stationary lifts, these conditionals are exact dimensional and satisfy

$\pi:X\to Y$13

where $\pi:X\to Y$14 is a foliation entropy and $\pi:X\to Y$15 is the adjacent Lyapunov gap (Lessa, 2019). In a different direction, entropy itself satisfies a chain rule under disintegration:

$\pi:X\to Y$16

providing a general conditional-entropy identity and an asymptotic interpretation through fiberwise typical sets (Vigneaux, 2021).

5. Bayesian inversion, Gaussian conditioning, and constructive formulas

Disintegration is the formal content of conditioning in Bayesian statistics. In categorical language, a prior $\pi:X\to Y$17 on $\pi:X\to Y$18 and a channel $\pi:X\to Y$19 determine a joint law

$\pi:X\to Y$20

and a Bayesian inversion $\pi:X\to Y$21 is any kernel satisfying

$\pi:X\to Y$22

With densities, this is the familiar Bayes formula

$\pi:X\to Y$23

for $\pi:X\to Y$24-almost every $\pi:X\to Y$25 (Cho et al., 2017).

For Gaussian measures on separable Banach spaces, disintegration with respect to bounded linear operator data recovers infinite-dimensional Gaussian conditioning. If $\pi:X\to Y$26 is Gaussian on $\pi:X\to Y$27 with mean $\pi:X\to Y$28 and covariance $\pi:X\to Y$29, and $\pi:X\to Y$30 is bounded linear, then the posterior along noiseless operator data has Gaussian disintegrations; in the Hilbert or finite-dimensional form, and with additive Gaussian noise covariance $\pi:X\to Y$31, the posterior mean and covariance are

$\pi:X\to Y$32

$\pi:X\to Y$33

The same paper proves that sequential assimilation yields the same final posterior as batch assimilation, giving infinite-dimensional analogues of Gaussian vector update identities (2207.13581).

A more delicate issue is constructive conditioning on submanifolds. If $\pi:X\to Y$34 is $\pi:X\to Y$35 on a manifold $\pi:X\to Y$36 and $\pi:X\to Y$37, then for regular values $\pi:X\to Y$38 the disintegration density on the fiber $\pi:X\to Y$39 is not generally the restricted ambient density. Instead,

$\pi:X\to Y$40

and in Euclidean space this becomes $\pi:X\to Y$41 with $\pi:X\to Y$42 (Costa et al., 1 Aug 2025). The paper shows that the often-used “restricted density” on the constraint surface generally disagrees with the disintegration density, and that conditional modes derived from these two constructions need not coincide.

6. Foundational, computability, and set-theoretic frontiers

Disintegration has nontrivial computational content. For complete separable metric spaces, the disintegration operator along a projection, restricted to measures admitting a unique continuous disintegration, is strongly Weihrauch equivalent to the limit operator:

$\pi:X\to Y$43

When uniqueness fails, a disintegration can still be extracted from a computable basis of continuity sets with the Vitali covering property, and the resulting operator remains strongly Weihrauch reducible to $\pi:X\to Y$44 (Ackerman et al., 2015). This identifies conditioning at points as a limit-complete operation rather than a computable one.

At the opposite foundational extreme, set-theoretic pathologies appear already for $\pi:X\to Y$45-finite disintegrations. Maharam asked whether every disintegration of a $\pi:X\to Y$46-finite measure into $\pi:X\to Y$47-finite measures is uniformly $\pi:X\to Y$48-finite. Under the Continuum Hypothesis, and also under $\pi:X\to Y$49, there exist counterexamples: one can construct a $\pi:X\to Y$50-finite disintegration that is not uniformly $\pi:X\to Y$51-finite, and under $\pi:X\to Y$52 this can be done using special $\pi:X\to Y$53 sets (Backs et al., 2013). This shows that even basic regularity questions about disintegration are sensitive to additional axioms of set theory.

The contemporary picture is therefore highly stratified. On standard Borel and Radon spaces, disintegration is the backbone of regular conditional probability. In canonical compact-Hausdorff models it becomes pointwise unique and weak-* continuous (Jamneshan et al., 2020). In transport and geometry it organizes conditional measures as Wasserstein-valued fields (Possobon et al., 2022). In dynamics it can be regular, exact dimensional, atomic, or mono-atomic depending on the mechanism generating the invariant measure (Butterley et al., 2015, Santiago et al., 2020). In constructive Bayesian settings it controls the difference between conditioning and naive restriction (Costa et al., 1 Aug 2025). A plausible implication is that “disintegration of measures” is no longer a single theorem but a family of theories: measure-theoretic, geometric, dynamical, algorithmic, and foundational, linked by the same fiberwise integral identity but distinguished by radically different regularity, rigidity, and computability phenomena.