Want to take part in these discussions? Sign in if you have an account, or apply for one below
Vanilla 1.1.10 is a product of Lussumo. More Information: Documentation, Community Support.
Created:
Maharam’s theorem states a complete classification of isomorphism classes of the appropriate category of measurable spaces.
In the σ-finite case, the theorem classifies measure spaces up to an isomorphism. Here an isomorphism is an equivalence class of measurable bijections $f$ with measurable inverse such that $f$ and $f^{-1}$ preserve measure 0 sets.
As explained in the article categories of measure theory, for a truly general, unrestricted statement for non-σ-finite spaces there are additional subtleties to consider: equality almost everywhere must be refined to weak equality almost everywhere, and σ-finiteness should be relaxed to a combination of Marczewski-compactness and strict localizibility.
In this unrestricted form, by the Gelfand-type duality for commutative von Neumann algebras, Maharam’s theorem also classifies isomorphism classes of localizable Boolean algebras, abelian von Neumann algebras, and hyperstonean spaces (or hyperstonean locales).
Every object in one of the above equivalent categories canonically decomposes as a coproduct (disjoint union) of ergodic objects. Here an object $X$ is ergodic if the only subobjects of $X$ invariant under all automorphisms of $X$ are $\emptyset$ and $X$ itself.
Furthermore, an ergodic object $X$ is (noncanically, using the axiom of choice) isomorphic to $\mathfrak{c}\times 2^\kappa$, where $\kappa$ is 0 or infinite, and $\mathfrak{c}$ is infinite if $\kappa$ is infinite. Here the cardinal $\mathfrak{c}$ is known as the cellularity of $X$ and $\kappa$ is its Maharam type.
In particular, if $\kappa=0$, we get a classification of isomorphism classes of atomic measure spaces: they are classified by the cardinality $\mathfrak{c}$ of their set of atoms.
Otherwise, $\kappa$ is infinite, and we get a classification of isomorphism classes of ergodic atomless (or diffuse) measure spaces: such spaces are isomorphic to $\mathfrak{c}\times 2^\kappa$, where $\mathfrak{c}$ and $\kappa$ are infinite cardinals.
Thus, a completely general object $X$ has the form
$\coprod_\kappa \mathfrak{c}_\kappa\times 2^\kappa,$where $\kappa$ runs over 0 and all infinite cardinals, $\mathfrak{c}_\kappa$ is a cardinal that is infinite or 0 if $\kappa\ne0$, and $\mathfrak{c}_\kappa\ne0$ only for a set of $\kappa$.
The original reference is
A modern exposition can be found in Chapter 33 (Volume 3, Part I) of
Added a relative version:
There is also a relative version of Maharam’s theorem, which classifies morphisms in any of the equivalent categories considered above.
Observe that morphisms $2^\kappa\to2^\lambda$ exist if and only if $\kappa\ge\lambda$. For example, there are no morphisms from the terminal space $2^0$ (i.e., a singleton) to the real line $\mathbf{R}\cong 2^{\aleph_0}$, since the image of such a point is a measure 0 subset, whose preimage therefore cannot have measure 0. In the language of commutative von Neumann algebras, this translates to saying that there are no normal *-homomorphisms $L^\infty(\mathbf{R})\to\mathbf{C}$.
Observe also that we have a natural notion of locality for a measure space: a covering family is given by a family of measurable subsets whose essential supremum equals the entire space. (This is more than just an analogy to open covers in topological spaces: when translated to the language of locales, the two notions become identical.)
With these two observations in mind, we can succinctly formulate the relative Maharam theorem as follows: every morphism $f\colon X\to Y$ locally in $X$ and $Y$ is isomorphic to a morphism of the form $2^\kappa\to2^\lambda$, where $\kappa\ge\lambda$ and the map is given by projecting to the first $\lambda$ coordinates.
1 to 2 of 2