Pipeline-generated

How statement and proof provenance work

The first chip identifies the source of the statement or construction; the second identifies the source of its local proof or verification.

Literature-sourced: the exact statement appears in a cited source; only wording and notation differ.
AI-adapted: a semantically identical restatement of literature-sourced material, modulo indexing, notation, and boundary cases adopted by the library.
AI-generated: a genuinely novel statement formulated by AI, with no source for the claim itself.

These labels describe origin, not correctness: citations and verification chips remain separate evidence.

✓ 8 results · all verified · 2 also independently AI-judged

Every result on this page is machine-checked by a proof checker and read in full and owner-audited; the judge is an additional, independent cross-model AI review of the proofs. The 6 not AI-judged were verified by owner audit (typically over a confirmed judge false positive), not failures.

Measurable Hilbert Fields and Direct-Integral Operators

1 · Prerequisites

2 · Summary

This page builds the measurable bookkeeping that lets a family of Hilbert spaces be integrated along a measure space, together with the operators that respect the family. A measurable complex Hilbert field over a sigma-finite standard-Borel measure space $(X, B, μ)$ consists of separable complex fibres $H_{x}$ and a countable fundamental sequence $e_{n} (x)$ whose Gram coefficients $x \mapsto ⟨ e_{n} (x), e_{m} (x) ⟩$ are all Borel, with ${e_{n} (x) : n \in N}$ complex-linearly dense in every fibre. A section $ξ$ is measurable exactly when every coefficient $x \mapsto ⟨ ξ (x), e_{n} (x) ⟩$ is Borel. Inner products are linear in the first variable, zero fibres are allowed, the measure is not assumed complete, and sections are identified only when they agree off a Borel null set; nothing in the definition presumes a measurable raw field or a completed base.

The first structural result is Measurable sections have measurable pointwise inner products: the coefficient test is equivalent to testing against every measurable section, the pointwise norm $x \mapsto ∥ ξ (x) ∥$ and every pairing $x \mapsto ⟨ ξ (x), η (x) ⟩$ of measurable sections are measurable, and measurable sections are closed under measurable scalar combinations and pointwise norm limits. The proof codes every finite rational-complex combination of the fundamental family by one fixed finite-sequence code, uses rational density to obtain a countable pointwise-dense family, recovers the fibre norm as a countable supremum, and replaces arbitrary witnesses by least-index approximants; the argument is choice-free, and the zero fibre and the dependent fundamental vectors are handled without division by zero.

The direct integral $\int_{X}^{\oplus} H_{x} d μ (x)$ is then defined as the sections of finite integrated squared norm modulo agreement off a measurable null set, with $⟨ [ξ], [η] ⟩ = \int_{X} ⟨ ξ (x), η (x) ⟩ d μ (x)$ . The definition Direct integral of a measurable Hilbert field proves that square-integrable sections form a vector space, that the quotient operations respect almost-everywhere classes, that the pairing is finite by fibrewise Cauchy--Schwarz followed by the scalar $L^{2}$ Cauchy--Schwarz inequality, and that the result is a genuine inner-product space: representative independence uses the almost-everywhere invariance of the integral, linearity uses linearity of the $L^{1}$ integral, and positive definiteness uses the zero-integral criterion. Completeness is deliberately not part of the definition.

Direct integrals of measurable Hilbert fields are Hilbert spaces supplies it. A Cauchy sequence is thinned to a summable one, representatives are chosen for the countably many classes, and the resulting series of fibre increments converges outside the measurable set where its norm sum is infinite; dominated convergence identifies the limit and completes the Cauchy sequence. Separability is obtained from the AC-qualified countable Borel generating algebra, a finite-measure exhaustion and a local pi--lambda approximation for scalar complex $L^{2}$ , a measurable fibrewise Gram--Schmidt frame whose zero remainders are assigned the zero vector, and Parseval together with dominated convergence for the coordinate truncations; the final density estimate is taken over the measurable sets where the frame vector is nonzero. The theorem states its exact use of the axiom of choice: choosing the countable family of representatives, the standard-Borel coding and generating algebra, and the countable-choice hypothesis of the published Parseval theorem.

Operator fields are handled next. Measurable and decomposable operator fields defines a field $(T_{x})$ with $T_{x} \in B (H_{x})$ to be weakly measurable when every fundamental matrix coefficient $x \mapsto ⟨ T_{x} e_{n} (x), e_{m} (x) ⟩$ is measurable, shows this is equivalent to measurability of $x \mapsto ⟨ T_{x} ξ (x), η (x) ⟩$ for all measurable sections, and proves the fibre norm $x \mapsto ∥ T_{x} ∥$ is measurable, so essential boundedness is meaningful. A bounded operator is decomposable when it is induced by such a field. Measurable essentially bounded operator fields act decomposably proves that every weakly measurable essentially bounded field acts on the direct integral by $[ξ] \mapsto [x \mapsto T_{x} ξ (x)]$ , with $∥ T ∥ = {ess sup}_{x} ∥ T_{x} ∥$ exactly. The upper bound follows from the pointwise operator inequality; the lower bound localizes a superlevel set to a finite positive-measure set, tests with an indicator section, and lets a rational level approach the essential supremum. The adjoint field $(T_{x}^{*})$ and the product field $(T_{x} S_{x})$ are again weakly measurable and essentially bounded, and they induce $T^{*}$ and $T S$ ; the adjoint claim uses the Hilbert adjoint theorem with its countable-choice hypothesis supplied by AC.

The diagonal algebra of a direct integral is $D = {M_{f} : f \in L^{\infty} (X, μ)}$ , with $M_{f}$ acting by scalar multiplication. Decomposable operators are the commutant of diagonal multiplication identifies its commutant exactly with the decomposable operators. The inclusion that decomposable operators commute with every $M_{f}$ is pointwise; the converse reconstructs an arbitrary $S \in D^{'}$ from its images of normalized fundamental sections localized to the members of a finite-measure exhaustion, using commutation with the corresponding characteristic multipliers to glue the local representatives into a measurable field, establishing the pointwise bound on a dense rational-complex test family on one common conull set, extending fibrewise by density, and finally recovering $S$ from the induced operator on a dense span of localizations. The theorem also proves that a weakly measurable essentially bounded field inducing a fixed operator is unique up to a null set. Diagonal multipliers form a von Neumann algebra then shows $D$ is a unital abelian $*$ -subalgebra with $D^{''} = D$ , hence a concrete von Neumann algebra, even when zero fibres make $f \mapsto M_{f}$ noninjective: countably many measurable rank-one fields $R_{n, m} (x) ξ = ⟨ ξ, u_{n} (x) ⟩ u_{m} (x)$ lie in $D^{'}$ , and fibrewise commutation with them forces every $T \in D^{''}$ to be scalar on each fibre, with the scalar recovered measurably on the least-index partition of the nonzero fibers.

The spectral model closes the page. A first tool is A separably acting abelian von Neumann algebra has a self-adjoint generator: on a nonzero separable complex Hilbert space, every abelian concrete von Neumann algebra is $W^{*} (S)$ for one bounded self-adjoint $S$ . The argument makes the weak operator topology on the unit ball of $A$ countably based, builds a weakly dense sequence by AC, splits it into self-adjoint contractions, enumerates their dyadic threshold projections, and encodes the whole enumeration in one norm-convergent ternary series whose spectral projections are decoded by explicit continuous digit functions. Spectral multiplicity model for separably acting abelian von Neumann algebras then decomposes $H$ into cyclic reducing summands, forms the weighted common measure $μ = \sum_{j} a_{j} μ_{j}$ with Radon--Nikodym densities $h_{j}$ , builds the measurable field $H_{t} = C^{m (t)}$ with multiplicity $m (t) = \sum_{j} 1_{{h_{j} > 0}} (t)$ and a unitary $U$ with $U S U^{- 1} = M_{t}$ and $U A U^{- 1} = {M_{f} : f \in L^{\infty} (K, μ)}$ , and proves that for a fixed generator the measure class and the almost-everywhere multiplicity are unique: measure classes are detected by the common spectral scalar measure, equivalent measures are bridged by the Radon--Nikodym square-root unitary, and an off-diagonal block of the direct sum of two models is decomposable, so its fibres are unitaries and the dimensions agree. Changing the generator is not covered by the uniqueness clause.

Alongside these items the page fixes the surrounding vocabulary it uses: Von Neumann algebras and commutants states the concrete unital weak-operator-closed convention, the commutant and double commutant, and the generated algebra $W^{*} (S)$ , with no bicommutant theorem assumed, and the adjoint convention on $B (H)$ is invoked under the exact countable-choice hypothesis of the published adjoint theorem. The axiom of choice is assumed exactly where the items state it; the field, section, direct-integral and operator-field definitions are choice-free.

3 · Logical flowchart

4 · Definitions, theorems and proofs

DefinitionDefinition: AI-adaptedProof: Not applicablejudge pass (gpt-6-sol)audited 2026-09-30Open item page →

Von Neumann algebras and commutants

Definition

Assume AC. Let $H$ be a complex Hilbert space and let $B (H)$ denote its bounded linear operators. Applying AC to a family indexed by the natural numbers supplies Countable Choice, the hypothesis of Hilbert-adjoint identities, which provides adjoints on $B (H)$ . A concrete von Neumann algebra on $H$ is a unital $*$ -subalgebra $A \subseteq B (H)$ that is closed in the weak operator topology (WOT). Here unital means $I_{H} \in A$ ; the zero Hilbert space is allowed, with $I_{H} = 0$ and sole unital algebra ${0}$ . This is the only use of AC in this definition.

For any set $S \subseteq B (H)$ , its commutant and double commutant are $S^{'} : = {T \in B (H) : T S = S T for every S \in S}, S^{''} : = (S^{'})^{'} .$ Commutants are taken inside $B (H)$ . The von Neumann algebra generated by $S$ is $W^{*} (S) : = {\overline{{Alg}^{*} (S \cup {I_{H}})}}^{W O T},$ the WOT closure in $B (H)$ of the unital $*$ -algebra generated by $S$ . No bicommutant theorem is part of these definitions.

Two elementary properties will be used. For fixed $S \in B (H)$ , $⟨ T S ξ, η ⟩ = ⟨ T (S ξ), η ⟩, ⟨ S T ξ, η ⟩ = ⟨ T ξ, S^{*} η ⟩$ are WOT-continuous matrix coefficients as functions of $T$ . Thus each equation $T S = S T$ defines a WOT-closed set, and so $S^{'}$ is WOT-closed. If $S$ is self-adjoint, then $S^{'}$ is a unital $*$ -subalgebra: products preserve commutation with every $S$ , and taking adjoints of $T S = S T$ gives $S T^{*} = T^{*} S$ . These facts do not identify $W^{*} (S)$ with $S^{''}$ .

DefinitionDefinition: AI-adaptedProof: Not applicablejudge pass (gpt-6-sol)audited 2026-09-30Open item page →

Measurable Hilbert field from a countable fundamental family

Definition

Let $(X, B, μ)$ be a standard-Borel space with a sigma-finite measure on its Borel sigma-algebra; the measure is not assumed complete. A measurable complex Hilbert field with countable fundamental family consists of a separable complex Hilbert space $H_{x}$ for each $x \in X$ and a sequence of vectors $e_{n} (x) \in H_{x}$ ( $n \in N$ ) such that $x ⟼ ⟨ e_{n} (x), e_{m} (x) ⟩_{H_{x}}$ is Borel measurable for every $n, m$ , and the complex-linear span of ${e_{n} (x) : n \in N}$ is dense in $H_{x}$ for every $x$ . The fibres may be zero-dimensional; no positive lower bound on their dimensions is imposed. The inner product is linear in its first variable, as in the library's complex inner-product convention.

A section is a choice of a vector $ξ (x) \in H_{x}$ for each $x$ . It is a measurable section when all its fundamental coefficients $x ⟼ ⟨ ξ (x), e_{n} (x) ⟩_{H_{x}}, n \in N,$ are Borel measurable complex functions. Each $e_{n}$ is itself measurable by the Gram-coefficient condition. Pointwise sums and multiplication by a measurable complex scalar function are taken in the corresponding fibres.

Two measurable sections are identified almost everywhere when there is a Borel null set $N$ such that they agree at every $x \notin N$ . This formulation works on the given, possibly noncomplete, measure space and does not silently adjoin arbitrary subsets of null sets to $B$ .

LemmaStatement: AI-adaptedProof: AI-adaptedprecheck passjudge pass (gpt-6-sol)audited 2026-09-30Open item page →

Measurable sections have measurable pointwise inner products

Statement

Let $(H_{x}, e_{n} (x))_{x \in X}$ be a measurable complex Hilbert field with countable fundamental family, with the inner product linear in its first variable. For a section $ξ$ , the following are equivalent:

every coefficient $x \mapsto ⟨ ξ (x), e_{n} (x) ⟩$ is measurable;
$x \mapsto ⟨ ξ (x), η (x) ⟩$ is measurable for every measurable section $η$ .

For such sections, $x \mapsto ∥ ξ (x) ∥$ and $x \mapsto ⟨ ξ (x), η (x) ⟩$ are measurable. Measurable sections are closed under measurable scalar combinations and under pointwise norm limits.

Facts & Assumptions

[F1]

Each fibre is separable, the fundamental Gram coefficients are measurable, and the countable fundamental family has dense complex-linear span in every fibre (Measurable Hilbert field from a countable fundamental family).

[F2]

The complex inner product is linear in its first variable and conjugate-linear in its second (Real and complex inner-product spaces and their induced length).

[F3]

Every inner-product pairing satisfies $∣ ⟨ u, v ⟩ ∣ \leq ∥ u ∥ ∥ v ∥$ (Cauchy–Schwarz: $∣ ⟨ x, y ⟩ ∣ \leq ∥ x ∥ ∥ y ∥$ , with equality exactly for dependent pairs).

[F4]

Sequential suprema of measurable real functions and pointwise limits of measurable real-valued functions are measurable (Sequential suprema, infima, limsup, liminf, and pointwise limits of measurable functions are measurable).

[F5]

A function between measurable spaces is measurable exactly when inverse images of measurable target sets are measurable (A measurable function between measurable spaces).

[F6]

Let $β : N^{2} \to N$ be the bijection in [F7]. Define $c_{0} (()) = 0$ and $c_{k + 1} (a_{0}, \dots, a_{k}) = β (a_{0}, c_{k} (a_{1}, \dots, a_{k}))$ . Then $c (a_{0}, \dots, a_{k - 1}) = β (k, c_{k} (a_{0}, \dots, a_{k - 1}))$ is injective on all finite sequences: inverse pairing first recovers the length and then each entry.

[F7]

There is a specified bijection between $N$ and $N \times N$ ( $N \times N \approx N$ ).

[F8]

There is a bijection $ρ : Q \to N$ ( $Q$ is countably infinite).

[F9]

The rational embedding is dense in $R$ ; every real is approximated within any positive rational tolerance, and a rational lies strictly between any two distinct reals (The rationals embed densely in the reals).

[F10]

Complex modulus is subadditive and multiplicative (Conjugation is an involutive real-field automorphism, $z \overline{z} = ∣ z ∣^{2}$ , and modulus is definite, multiplicative, and subadditive).

[F11]

$d_{C} (z, w) = ∣ z - w ∣$ is the Euclidean metric on $R^{2}$ (The Euclidean metric, convergence, Cauchy sequences, and continuity on the complex plane).

[F12]

Rational open boxes give a countable basis in every finite-dimensional real coordinate space ( $Q^{n}$ is a countable dense subset of $R^{n}$ , and rational open boxes form a countable basis).

[F13]

Continuous maps have Borel preimages (A continuous map has Borel preimages of Borel sets).

[F14]

Composition with a Borel map preserves measurability (Composition with a Borel measurable outer map preserves measurability).

Proof

technique · direct, using an explicitly coded countable dense family

Given: A countable fundamental family and a section whose fundamental coefficients are measurable.

1.1F1F2F5F6F7F8F9F10F11F12F13F14construct

Fix a bijection $ρ : Q \to N$ from [F8] and a bijection $β : N^{2} \to N$ from [F7]. Encode a term $(n, a, b)$ , representing $(a + i b) e_{n}$ , by $κ (n, a, b) = β (n, β (ρ (a), ρ (b)))$ , and encode a finite list of such term-codes by the finite-sequence code [F6]. To decode $j$ , first write $β^{- 1} (j) = (k, m)$ and recursively apply $β^{- 1}$ to recover a length- $k$ list from $m$ ; accept it only if the final remainder is $0$ , and otherwise use the empty list. The recursion takes exactly $k$ steps, so this defines $q_{j}$ for every $j$ , and every rational-complex finite combination occurs since each finite list has its code. Only the two single bijections [F7,F8] are fixed. The scalar set $Q + i Q$ is dense in $C$ : for $z = a + i b$ and $ε > 0$ , choose rationals $p, q$ strictly between $a - ε / 3, a + ε / 3$ and $b - ε / 3, b + ε / 3$ using [F9], and [F10] gives $∣ z - (p + i q) ∣ \leq ∣ a - p ∣ + ∣ b - q ∣ < ε$ . Given $v \in H_{x}$ , first approximate it within $ε / 2$ by a finite complex combination $u = \sum_{r = 1}^{N} z_{r} e_{n_{r}} (x)$ using [F1]. With $C = \sum_{r} ∥ e_{n_{r}} (x) ∥$ , choose rational-complex $c_{r}$ with $∣ z_{r} - c_{r} ∣ < ε / (2 (1 + C))$ ; the fibre norm triangle inequality and homogeneity give $∥ u - \sum_{r} c_{r} e_{n_{r}} (x) ∥ \leq \sum_{r} ∣ z_{r} - c_{r} ∣ ∥ e_{n_{r}} (x) ∥ < ε / 2$ . Thus $(q_{j} (x))_{j}$ is dense in every fibre. These finitely many existential instantiations are not an axiom-of-choice use. For decoded $q_{j} = \sum_{r = 1}^{N} c_{r} e_{n_{r}}$ , the coefficient $⟨ q_{j} (x), e_{m} (x) ⟩ = \sum_{r} c_{r} ⟨ e_{n_{r}} (x), e_{m} (x) ⟩$ is a finite Gram sum. The real and imaginary coordinate maps are continuous for [F11]'s metric and Borel by [F13]; finite tuples are measurable in $R^{2 N}$ by the countable rational-box basis [F12]; and the finite-sum map is continuous and Borel [F13], so composition [F14] makes each $q_{j}$ measurable. The same argument makes $∥ q_{j} (x) ∥^{2}$ measurable from its finite Gram expansion. The square-root map is continuous on $[0, \infty)$ because for $r \geq s \geq 0$ , $(\sqrt{r} - \sqrt{s})^{2} \leq r - s$ ; hence it is Borel [F13], and [F14] makes $∥ q_{j} (x) ∥$ measurable.

1.2F2F5F11F12F13F14algebra

If $a, b$ are measurable complex scalar functions and $ξ, η$ have measurable coefficients, then $⟨ a ξ + b η, e_{n} ⟩ = a ⟨ ξ, e_{n} ⟩ + b ⟨ η, e_{n} ⟩$ by [F2]. The finite tuple of inputs is measurable in $C^{4} ≅ R^{8}$ because its real coordinates are Borel [F11,F13,F14] and rational boxes [F12] form a countable basis. The map $(u, z, v, w) \mapsto u z + v w$ is continuous and Borel [F13], so composition [F14] makes each displayed coefficient measurable. Thus measurable sections are closed under measurable scalar combinations.

1.3F2F3F4F5F11F13F14

Suppose $ξ_{k} (x) \to ξ (x)$ in norm pointwise and each $ξ_{k}$ is measurable. For every $n$ , Cauchy--Schwarz [F3] gives $∣ ⟨ ξ_{k} (x) - ξ (x), e_{n} (x) ⟩ ∣ \leq ∥ ξ_{k} (x) - ξ (x) ∥ ∥ e_{n} (x) ∥ \to 0$ . The complex coefficients converge pointwise; their real and imaginary coordinates are Borel by [F11,F13] and composition [F14] preserves measurability. The pointwise-limit theorem [F4] makes both real limits measurable, so every fundamental coefficient of $ξ$ is measurable.

2.1F1F2F3F4F5F10F11F12F13F14step 1.1algebra

For a coefficient-measurable $ξ$ , $⟨ ξ, q_{j} ⟩$ is a finite linear combination of its fundamental coefficients by [F2] and step 1.1. Put $b_{j} (x) = ∣ ⟨ ξ (x), q_{j} (x) ⟩ ∣ / ∥ q_{j} (x) ∥$ when $∥ q_{j} (x) ∥ > 0$ , and $b_{j} (x) = 0$ otherwise. Modulus is continuous by the reverse triangle inequality from [F10,F11]; division is continuous where the denominator is positive, and extending by zero on the Borel set where it vanishes gives a Borel map. The input tuple is measurable by the rational-box basis [F12], so composition [F13,F14] makes $b_{j}$ measurable. Cauchy--Schwarz [F3] gives $b_{j} (x) \leq ∥ ξ (x) ∥$ . On a nonzero fibre, the normalized nonzero $q_{j} (x)$ are dense in the unit sphere: for any unit $u$ , take the least index $j_{m}$ with $∥ q_{j_{m}} (x) - u ∥ < 1 / m$ ; then normalization converges to $u$ . If $ξ (x) \neq 0$ , apply this to $u = ξ (x) / ∥ ξ (x) ∥$ and use [F3] to obtain $\sup_{j} b_{j} (x) = ∥ ξ (x) ∥$ ; for $ξ (x) = 0$ both sides vanish. On a zero fibre every test and the norm are zero. Thus $∥ ξ (x) ∥ = \sup_{j} b_{j} (x)$ in every fibre, and the countable-supremum clause [F4] makes the norm measurable.

3.1F1F2F5step 1.1step 1.2step 2.1construct

Fix a measurable section $η$ and $k \geq 1$ . By step 1.2, $η - q_{j}$ is measurable; step 2.1 then makes $E_{j, k} = {x : ∥ η (x) - q_{j} (x) ∥ < 1 / k}$ measurable. Density from step 1.1 gives $⋃_{j} E_{j, k} = X$ . At each $x$ , take the least $j_{k} (x)$ with $x \in E_{j_{k} (x), k}$ . The sets $E_{j, k} ∖ ⋃_{i < j} E_{i, k}$ form a measurable partition, so $η_{k} (x) = q_{j_{k} (x)} (x)$ is measurable coefficient by coefficient and $∥ η_{k} (x) - η (x) ∥ < 1 / k$ . Least-index selection is canonical and uses no choice.

4.1F2F3F4F5F11F13F14step 1.1step 3.1

For each fixed $j$ , $⟨ ξ, q_{j} ⟩$ is measurable by the finite-combination argument of step 1.1, so $⟨ ξ, η_{k} ⟩$ is measurable on the partition of step 3.1. Cauchy--Schwarz [F3] gives $∣ ⟨ ξ (x), η_{k} (x) - η (x) ⟩ ∣ \leq ∥ ξ (x) ∥ / k \to 0$ , so the pairings converge pointwise to $⟨ ξ, η ⟩$ . The real and imaginary coordinate maps are continuous for [F11]'s metric and Borel [F13]; composition [F14] and pointwise-limit measurability [F4] show the limit is measurable. This proves coefficient measurability implies measurability of pairing with every measurable section.

5.1F1step 4.1∎

Conversely, if pairing with every measurable section is measurable, each $e_{n}$ is a measurable section by its Gram coefficients [F1]. Testing against $e_{n}$ gives every fundamental coefficient measurable, which is the coefficient criterion. Together with step 4.1 this proves both directions of the equivalence.

LemmaStatement: AI-adaptedProof: AI-adaptedprecheck passjudge pass (gpt-6-sol)audited 2026-09-30Open item page →

A separably acting abelian von Neumann algebra has a self-adjoint generator

Statement

Assume AC. Let $H$ be a nonzero separable complex Hilbert space and let $A \subseteq B (H)$ be an abelian von Neumann algebra (so its elements commute pairwise). Then there is a bounded self-adjoint $S \in A$ such that $A = W^{*} (S) .$

Facts & Assumptions

[F1]

Separability means that some at most countable subset is dense. In the library, “at most countable” means finite or countably infinite. Since $H$ is nonzero, a dense subset is nonempty; its finite or countable listing, with one point repeated in the finite case, gives a dense sequence (Separability: the existence of an at most countable dense subset, Finite, countably infinite, countable, uncountable).

[F2]

Weak operator tests are the scalar functionals $T \mapsto f (T x)$ (Strong and weak operator topologies). Under the first-variable-linear inner-product convention, Riesz representation under Countable Choice writes each bounded linear $f$ as $f (v) = ⟨ v, y ⟩$ for some $y \in H$ (Real and complex inner-product spaces and their induced length, Riesz representation for Hilbert spaces, The Axiom of Countable Choice ( ${A C}_{ω}$ )). Thus WOT on $B (H)$ is generated by the matrix coefficients $T \mapsto ⟨ T x, y ⟩$ .

[F3]

The topology of $C$ is the $d_{2}$ topology on $R^{2}$ ; the product topology there is the $d_{2}$ topology, and rational open boxes give a countable basis. Hence $C$ is second countable (The Euclidean metric, convergence, Cauchy sequences, and continuity on the complex plane, For $n \geq 1$ the product topology on $n$ copies of the usual topology of $R$ is the metric topology of $d_{\infty}$ on $R^{n}$ , and hence also of $d_{1}$ and $d_{2}$ , so $R^{n}$ as a product and $R^{n}$ as a metric space are one space, $Q^{n}$ is a countable dense subset of $R^{n}$ , and rational open boxes form a countable basis, Second countability: an at most countable basis for the topology).

[F4]

There is a bijection $N \to N \times N$ ( $N \times N \approx N$ ). Under Countable Choice, a countable product of second-countable spaces is second countable (The product set $\prod_{i \in I} X_{i}$ of functions choosing a point in each factor, the projections, the box topology, and the product topology as the initial topology of the projections; the empty product is a one-point space, Assuming countable choice, a countable product of second countable spaces is second countable), and every subspace of a second-countable space is second countable (Second countability is hereditary). AC supplies the Countable Choice hypothesis by The Axiom of Choice and The Axiom of Countable Choice ( ${A C}_{ω}$ ).

[F5]

$A$ is a unital star-subalgebra of $B (H)$ closed in WOT; “abelian” means that its elements commute pairwise. Also $W^{*} (S)$ is the WOT closure of the unital star-algebra generated by $S$ (Von Neumann algebras and commutants).

[F6]

The operator norm is subadditive and absolutely homogeneous, as follows from its unit-ball supremum definition (The operator norm as the least bound and as the unit-sphere or unit-ball supremum).

[F7]

A Hilbert space is Banach (Hilbert space), so $B (H)$ is Banach by If (Y) is Banach then (\mathcal B(X,Y)) is Banach. Under composition it is a nonzero unital complex Banach algebra: composition is associative and submultiplicative (Composition satisfies |ST|\le|S|,|T|), the bounded-operator space has pointwise linear operations (The spaces (\mathcal B(X,Y)) and (\mathcal B(X)) of bounded linear operators), $I_{H}$ is its unit and has norm $1$ by the operator-norm definition on nonzero $H$ (The operator norm as the least bound and as the unit-sphere or unit-ball supremum), and $I_{H} \neq 0$ . The operator spectrum is the spectrum in $B (H)$ : by the two definitions, $λ I - T$ has a bounded everywhere-defined inverse exactly when it is invertible as an element of $B (H)$ (Spectrum and resolvent of a bounded operator, Spectrum and resolvent set in a Banach algebra). Thus if $T$ is self-adjoint, $σ (T) \subseteq R$ (Spectrum of a self adjoint operator is real) and $σ (T)$ lies in the closed disk of radius $∥ T ∥$ by the Banach-algebra spectral bound (Unital Banach algebra, Spectrum is nonempty compact and norm bounded).

[F8]

The continuous self-adjoint calculus is isometric, sends the coordinate function to $T$ , and has range $C^{*} (I, T)$ (Continuous functional calculus for bounded self adjoint operators). For normal $T$ , $C^{*} (I, T)$ is the norm closure of the unital star-algebra generated by $T$ (C star algebra generated by a normal operator).

[F9]

For bounded normal $T$ , the Borel calculus extends the continuous calculus, is a unital star-homomorphism, sends $1_{B}$ to the spectral projection $E (B)$ , has $∥ f (T) ∥ \leq ∥ f ∥_{\infty}$ , and uniformly bounded pointwise spectral-a.e. convergent Borel functions converge in SOT (Borel functional calculus for bounded normal operators). A spectral PVM takes values in orthogonal projections, which are contractive (Projection valued measure). SOT is stronger than WOT, and norm convergence implies WOT convergence, by the operator-topology definitions and boundedness of each WOT functional (Strong and weak operator topologies).

[F10]

Real polynomials uniformly approximate every continuous real function on a closed interval (Polynomials are uniformly dense in $C ([a, b], R)$ for every closed interval).

[F11]

AC is the choice-function axiom and in particular applies to any family of nonempty basic open sets (The Axiom of Choice).

[F12]

Hilbert adjoints obey $(R + T)^{*} = R^{*} + T^{*}$ and $(a R)^{*} = \overline{a} R^{*}$ , with $∥ R^{*} ∥ = ∥ R ∥$ (Hilbert-adjoint identities).

[F13]

Composition of bounded operators satisfies $∥ R T ∥ \leq ∥ R ∥ ∥ T ∥$ (Composition satisfies |ST|\le|S|,|T|).

[F14]

Every self-adjoint bounded operator is normal (Self-adjoint, positive, unitary and normal operators).

Proof

technique · direct

Given: AC, a nonzero separable complex Hilbert space $H$ , and an abelian von Neumann algebra $A \subseteq B (H)$ .

1.1F2F6

By [F1], choose a dense sequence $(x_{n})_{n \in N}$ in $H$ . On the operator unit ball $B_{1} (H)$ consider the coefficient map $Φ (T) = (⟨ T x_{n}, x_{m} ⟩)_{(n, m) \in N^{2}} .$ Its coordinates are WOT-continuous by [F2]. Conversely, they induce the WOT on this ball: if $T, U \in B_{1} (H)$ , then $∥ U - T ∥ \leq 2$ , and for dense sequence approximants $x_{i} \to x$ , $x_{j} \to y$ , $∣ ⟨ (U - T) x, y ⟩ - ⟨ (U - T) x_{i}, x_{j} ⟩ ∣ \leq 2 ∥ x - x_{i} ∥ ∥ y ∥ + 2 ∥ x_{i} ∥ ∥ y - x_{j} ∥ .$ Thus every finite matrix-coefficient neighborhood can be refined by one involving only the countable family of coordinates of $Φ$ .

1.2construct

Fix $k \geq 1$ . For a digit string $η = (η_{1}, \dots, η_{k}) \in {0, 1}^{k}$ , set $s_{η} : = \sum_{j = 1}^{k} 3^{- j} η_{j}, I_{η} : = [s_{η}, s_{η} + \frac{1}{2} 3^{- k}] .$ These finitely many intervals lie in $[0, 1 / 2]$ and are pairwise separated. Indeed, if two strings first differ at index $ℓ \leq k$ , the difference of their left endpoints has absolute value at least $3^{- ℓ} - \sum_{j = ℓ + 1}^{k} 3^{- j} = \frac{1}{2} 3^{- ℓ} + \frac{1}{2} 3^{- k};$ subtracting their common interval length $\frac{1}{2} 3^{- k}$ leaves a gap at least $\frac{1}{2} 3^{- ℓ} > 0$ . Assign value $η_{k}$ on $I_{η}$ ; define $f_{k}$ linearly across each gap between consecutive intervals, with endpoint values inherited from those intervals, and constantly on each exterior interval with the value at the adjacent endpoint. This defines a continuous real piecewise-linear function on $[- 1 / 2, 1 / 2]$ . Its values on $[0, 1 / 2]$ are thus $0$ or $1$ on the cylinder intervals according to the $k$ -th digit.

2.1F3F4step 1.1

By [F3, F4], the product $C^{N^{2}}$ is second countable. Pulling its countable basis back by $Φ$ gives a countable basis for $B_{1} (H)$ , because step 1.1 shows that its initial topology is the relative WOT. Its subspace $A_{1} : = {T \in A : ∥ T ∥ \leq 1}$ is second countable by [F4]. Fix a countable basis of $A_{1}$ .

3.1F5F11step 2.1

For each nonempty member of this basis, AC [F11] selects one operator in that member; for an empty member put $0$ . Enumerating the countable basis, and repeating an entry if it is finite, gives a sequence $(T_{j})_{j \in N}$ WOT-dense in $A_{1}$ . Indeed every nonempty relatively open subset contains a basis member and hence a selected point.

4.1F5F6F12step 3.1

For $j \in N$ define $X_{j, 0} : = \frac{T_{j} + T_{j}^{*}}{2}, X_{j, 1} : = \frac{T_{j} - T_{j}^{*}}{2 i} .$ Because $A$ is a star-subalgebra, both lie in $A$ ; [F6, F12] show they are self-adjoint contractions, and $T_{j} = X_{j, 0} + i X_{j, 1}$ . All these operators commute because $A$ is abelian.

5.1F5F7F8F9step 4.1

For each $X = X_{j, τ}$ and real threshold $q$ , let $P_{X} (q) : = E_{X} ((q, \infty)) .$ This projection belongs to $A$ . To see this directly, on $σ (X) \subseteq [- 1, 1]$ set $g_{m} (t) : = \min {1, m \max {t - q, 0}} .$ Each $g_{m}$ is continuous, lies between $0$ and $1$ , and converges pointwise to $1_{(q, \infty)}$ , including value $0$ at $t = q$ . By [F8], $g_{m} (X)$ lies in the norm closure of the unital star-algebra generated by $X$ , hence in $A$ : norm convergence implies WOT convergence and $A$ is WOT-closed by [F5], [F9]. By [F9], $g_{m} (X)$ converges strongly, therefore weakly, to $P_{X} (q)$ ; WOT-closedness gives $P_{X} (q) \in A$ . The spectral bound $σ (X) \subseteq [- 1, 1]$ is [F7].

6.1F4F5step 5.1

For $n \in N$ and $0 \leq ℓ \leq 2^{n}$ put $r_{n, ℓ} : = - 2 + 4 ℓ 2^{- n} .$ The projections $P_{X_{j, τ}} (r_{n, ℓ})$ , indexed by $(j, τ, n, ℓ)$ , form an explicitly countable family: repeated pairing from [F4] enumerates $N^{4}$ , and invalid $τ$ or $ℓ$ indices may be assigned the zero projection. List the family, allowing repetitions, as $(P_{k})_{k \geq 1}$ . Every $P_{k}$ is in $A$ by step 5.1, so these projections commute pairwise.

7.1F7F9step 4.1step 5.1step 6.1

Fix $X = X_{j, τ}$ . For each $n$ , the intervals $(r_{n, ℓ - 1}, r_{n, ℓ}]$ , $1 \leq ℓ \leq 2^{n}$ , partition $(- 2, 2]$ and cover $σ (X)$ . Their spectral projections are $Q_{n, ℓ} = P_{X} (r_{n, ℓ - 1}) - P_{X} (r_{n, ℓ}) .$ The step function taking value $r_{n, ℓ - 1}$ on this interval differs from the identity function by at most $4 \cdot 2^{- n}$ on $σ (X)$ . The norm formula for the Borel calculus in [F9] therefore gives $∥ X - \sum_{ℓ = 1}^{2^{n}} r_{n, ℓ - 1} Q_{n, ℓ} ∥ \leq 4 \cdot 2^{- n} .$ Each approximant is a finite linear combination of the listed projections, so every $X_{j, τ}$ is a norm, hence WOT, limit of such combinations.

7.2F5F6F9F12step 6.1algebra

Define $S_{N} : = \sum_{k = 1}^{N} 3^{- k} P_{k}, S : = \sum_{k = 1}^{\infty} 3^{- k} P_{k} .$ By [F9], $∥ P_{k} ∥ \leq 1$ . Since $\sum_{k \geq 1} 3^{- k} = 1 / 2$ , the series converges in operator norm. Each $S_{N}$ is self-adjoint and belongs to $A$ ; its norm limit $S$ is self-adjoint and belongs to $A$ by WOT-closedness. Also $∥ S_{N} ∥, ∥ S ∥ \leq 1 / 2$ .

8.1F5step 3.1step 4.1step 7.1

Let $C$ be the unital star-algebra generated by $(P_{k})$ . It is contained in $A$ , and its WOT closure is contained in $A$ by [F5]. The WOT closure of a linear subspace is linear because addition and scalar multiplication are WOT-continuous. Step 7.1 puts every $X_{j, τ}$ in that closure, hence every $T_{j} = X_{j, 0} + i X_{j, 1}$ there. Since $(T_{j})$ is WOT-dense in $A_{1}$ by step 3.1, the closure contains $A_{1}$ . Every $T \in A$ is a scalar multiple of an element of $A_{1}$ (with $T = 0$ immediate), so ${\overline{C}}^{W O T} = A$ .

8.2F5F8F10step 6.1step 7.2step 1.2

For $N \geq k$ and $ϵ \in {0, 1}^{N}$ , put $R_{j, 1} = P_{j}, R_{j, 0} = I - P_{j}, Q_{ϵ} : = \prod_{j = 1}^{N} R_{j, ϵ_{j}} .$ The $P_{j}$ commute, so the $Q_{ϵ}$ are pairwise orthogonal projections whose sum is $I$ , and $S_{N} = \sum_{ϵ \in {0, 1}^{N}} (\sum_{j = 1}^{N} 3^{- j} ϵ_{j}) Q_{ϵ} .$ For every polynomial $p$ , the finite orthogonal resolution gives $p (S_{N}) = \sum_{ϵ} p (\sum_{j = 1}^{N} 3^{- j} ϵ_{j}) Q_{ϵ}$ . Uniform polynomial approximation and the isometric calculus [F8], [F10] pass this identity to $f_{k}$ , so $f_{k} (S_{N}) = \sum_{ϵ \in {0, 1}^{N}} f_{k} ⁣ (\sum_{j = 1}^{N} 3^{- j} ϵ_{j}) Q_{ϵ} = P_{k},$ because the scalar sum belongs to $I_{(ϵ_{1}, \dots, ϵ_{k})}$ and the function has value $ϵ_{k}$ there. The last equality uses $\sum_{ϵ : ϵ_{k} = 1} Q_{ϵ} = P_{k}$ .

8.3F6F7F8F10F13step 7.2step 1.2

The continuous calculus values converge in norm: $f_{k} (S_{N}) ⟶ f_{k} (S) .$ For any $δ > 0$ , [F10] gives a real polynomial $p$ with $\sup_{[- 1 / 2, 1 / 2]} ∣ f_{k} - p ∣ < δ$ . By [F7, F8], the spectra of $S_{N}$ and $S$ lie in this interval and the calculus is isometric, so each of $∥ f_{k} (S_{N}) - p (S_{N}) ∥$ and $∥ f_{k} (S) - p (S) ∥$ is below $δ$ . Since $S_{N} \to S$ in norm and multiplication is norm-continuous by [F13], $p (S_{N}) \to p (S)$ ; addition and scalar multiplication are norm-continuous by [F6]. Letting $δ ↓ 0$ proves the convergence.

9.1F5F8F14step 8.2step 8.3

Steps 8.2 and 8.3 show $P_{k} = f_{k} (S)$ for every $k$ . The operator $S$ is self-adjoint and hence normal by [F14]; by [F8], $f_{k} (S) \in C^{*} (I, S)$ , which is contained in the norm closure of the unital star-algebra generated by $S$ and therefore in $W^{*} (S)$ . Hence $P_{k} \in W^{*} (S)$ .

10.1F2F5F12step 8.1step 7.2step 9.1∎

By [F5], $W^{*} (S)$ is the WOT closure of the unital star-algebra $D$ generated by $S$ . This closure is itself a unital star-algebra: adjoint and each fixed left or right multiplication are WOT-continuous, as their matrix coefficients are WOT tests from [F2] and the adjoint identity [F12]. If $x, y \in {\overline{D}}^{W O T}$ , first approximate $y$ by elements of $D$ with left multiplication fixed at any $d \in D$ to get $d y \in {\overline{D}}^{W O T}$ ; then approximate $x$ with right multiplication fixed at $y$ to get $x y \in {\overline{D}}^{W O T}$ . The adjoint map preserves the closure by the same continuity, and $I \in D$ . Hence $W^{*} (S)$ is a WOT-closed unital star-algebra. It contains every $P_{k}$ by step 9.1, so it contains $C$ and its WOT closure $A$ by step 8.1. Conversely, $S \in A$ by step 7.2 and $A$ is a WOT-closed unital star-algebra, so the WOT closure of the unital star-algebra generated by $S$ is contained in $A$ . Thus $A = W^{*} (S)$ , with $S$ bounded and self-adjoint as required.

Remarks

The nonzero hypothesis is required by the continuous and Borel spectral calculus suppliers. One-dimensional $H$ and the scalar algebra $A = C I$ are included: the threshold projections may be only $0$ and $I$ , and the same argument still produces a generator.
The grid $[- 2, 2]$ places the full spectrum inside the half-open cells $(r_{n, ℓ - 1}, r_{n, ℓ}]$ , so spectral atoms at either spectral endpoint are included. Values at a threshold itself lie in the interval to its left, consistent with $P_{X} (q) = E_{X} ((q, \infty))$ .
AC is used only at the declared WOT-base selection and through the exact Countable Choice qualifications recorded above; the threshold enumeration and ternary coding are explicit.

DefinitionDefinition: AI-adaptedProof: AI-generatedaudited 2026-09-30Open item page →

Direct integral of a measurable Hilbert field

Definition

Let $(X, B, μ)$ carry a measurable complex Hilbert field $(H_{x}, e_{n} (x))_{x \in X}$ as in Measurable Hilbert field from a countable fundamental family, with inner products linear in the first variable. Write $M (H)$ for its measurable sections. Define the square-integrable section space $M_{2} (H) : = {ξ \in M (H) : \int_{X} ∥ ξ (x) ∥_{H_{x}}^{2} d μ (x) < \infty} .$ The integrand is a nonnegative measurable function: the section lemma Measurable sections have measurable pointwise inner products makes $x \mapsto ∥ ξ (x) ∥$ measurable, and composition with $t \mapsto t^{2}$ preserves measurability by Composition with a Borel measurable outer map preserves measurability, whose measurability convention is inverse-image measurability A measurable function between measurable spaces. Its integral is the nonnegative Lebesgue integral The nonnegative Lebesgue integral.

For $ξ, η \in M (H)$ , put $ξ \sim_{μ} η$ when there is a Borel $μ$ -null set $N$ such that $ξ (x) = η (x)$ for every $x \notin N$ , using the noncomplete-base convention of the field definition. The direct integral is the quotient set

\int_{X}^{\oplus} H_{x} d μ (x) : = M_{2} (H) / \sim_{μ} .

Write $[ξ]$ for the class of $ξ$ . Addition and scalar multiplication are induced by pointwise fibre operations, and the inner product is $⟨ [ξ], [η] ⟩ : = \int_{X} ⟨ ξ (x), η (x) ⟩_{H_{x}} d μ (x) .$

The admissible sections form a vector space. Measurability of pointwise linear combinations follows from the section lemma. If $a = ∥ ξ (x) ∥$ and $b = ∥ η (x) ∥$ , fibrewise expansion gives $∥ ξ (x) + η (x) ∥^{2} = a^{2} + b^{2} + 2 Re ⟨ ξ (x), η (x) ⟩ \leq a^{2} + b^{2} + 2 a b \leq 2 (a^{2} + b^{2}) .$ by Cauchy–Schwarz, $Re z \leq ∣ z ∣$ for complex $z$ , and $2 a b \leq a^{2} + b^{2}$ (since $(a - b)^{2} \geq 0$ ). The real-part bound follows from $∣ u + i v ∣ = \sqrt{u^{2} + v^{2}}$ and $u^{2} + v^{2} \geq u^{2}$ when $u > 0$ ; it is immediate when $u \leq 0$ , and taking nonnegative square roots gives $u \leq ∣ u + i v ∣$ in the positive case. The modulus and conjugation laws are recorded in Conjugation is an involutive real-field automorphism, $z \overline{z} = ∣ z ∣^{2}$ , and modulus is definite, multiplicative, and subadditive. By monotonicity, positive homogeneity, and additivity of the nonnegative integral (Monotonicity and nonnegative homogeneity of the nonnegative integral, Additivity of the nonnegative Lebesgue integral), the integral of the squared norm of a sum is finite whenever those of $ξ$ and $η$ are finite; the comparison function $a^{2} + b^{2}$ is measurable by Borel composition Composition with a Borel measurable outer map preserves measurability and measurable arithmetic Arithmetic and lattice operations preserve measurability whenever they are defined. For $c \neq 0$ , $∥ c ξ (x) ∥^{2} = ∣ c ∣^{2} ∥ ξ (x) ∥^{2}$ and positive homogeneity keeps its integral finite; for $c = 0$ the zero function has integral zero. Hence $M_{2} (H)$ is a complex vector space. The pointwise bound uses Cauchy–Schwarz: $∣ ⟨ x, y ⟩ ∣ \leq ∥ x ∥ ∥ y ∥$ , with equality exactly for dependent pairs and the norm and scalar conventions of Real and complex inner-product spaces and their induced length.

The quotient and its operations are well-defined. The empty set is a Borel null set by the $m = 0$ clause of Finite and countable subadditivity of measures, so $\sim_{μ}$ is reflexive; it is symmetric by equality, and it is transitive because two Borel null witnesses have a null union by the same finite-subadditivity theorem. This uses the definition of almost-everywhere equality Measure-null sets and almost-everywhere statements relative to a measure. If $ξ \sim_{μ} ξ^{'}$ and $η \sim_{μ} η^{'}$ , then outside the union of their two witnesses the pointwise sums agree, and so do the pointwise scalar multiples. The same finite union argument shows that these operations do not depend on representatives.

The displayed pairing is finite. By the section lemma, $h (x) = ⟨ ξ (x), η (x) ⟩$ is measurable. The modulus $∣ h ∣$ is measurable by Borel composition with the complex modulus. The measurable real-valued functions $f (x) = ∥ ξ (x) ∥$ and $g (x) = ∥ η (x) ∥$ belong to $L^{2} (μ)$ by the definition of that space The function space $L^{p} (μ)$ for $0 < p < \infty$ . Fibrewise Cauchy–Schwarz and then the scalar $L^{2}$ Cauchy–Schwarz inequality Cauchy-Schwarz inequality for $L^{2}$ give $\int_{X} ∣ h (x) ∣ d μ (x) \leq \int_{X} f (x) g (x) d μ (x) \leq ∥ f ∥_{2} ∥ g ∥_{2} < \infty .$ Since $f, g \geq 0$ , the scalar theorem's $∣ f g ∣$ is $f g$ . Here $f g$ is measurable by Arithmetic and lattice operations preserve measurability whenever they are defined, and the first inequality uses monotonicity of the nonnegative integral Monotonicity and nonnegative homogeneity of the nonnegative integral. Thus $h$ is an integrable complex function by Integrable real and complex functions, and their integrals. More explicitly, write $h = u + i v$ . The real coordinate projections are Borel, so $u$ and $v$ are measurable, and $∣ u ∣, ∣ v ∣ \leq ∣ h ∣$ makes both real functions integrable. Their classes, rather than a class of the complex function $h$ , belong to the real space $L^{1} (μ)$ of The space $L^{p} (μ)$ as the quotient by null functions. The complex integral here means precisely $\int h = \int u + i \int v$ .

The pairing does not depend on representatives. If $ξ \sim_{μ} ξ^{'}$ and $η \sim_{μ} η^{'}$ , their pairings agree outside the union of the two null witnesses. Both pairings are integrable by the preceding estimate. Their real parts agree almost everywhere, as do their imaginary parts. Applying Two integrable functions are equal almost everywhere exactly when all of their indefinite integrals agree separately to these real $L^{1}$ classes, with $A = X$ , gives equality of both component integrals and hence of the complex integrals.

The quotient is an inner-product space. Fibrewise linearity in the first variable passes through the integral by real linearity in The Lebesgue integral is linear on $L^{1} (μ)$ , applied to the integrable real and imaginary parts. Indeed, if $h = u + i v$ and $c = a + i b$ , then $c h = (a u - b v) + i (a v + b u)$ , so real linearity gives $\int c h = c \int h$ . For two integrable pairings $h = u + i v$ and $k = s + i t$ , $h + k = (u + s) + i (v + t)$ gives $\int (h + k) = \int h + \int k$ in the same way; all these real components are integrable by real linearity. Their complex combinations are integrable since $∣ c h ∣ = ∣ c ∣ ∣ h ∣$ and $∣ h + k ∣ \leq ∣ h ∣ + ∣ k ∣$ , using nonnegative integral monotonicity and additivity. Fibrewise conjugate symmetry passes through it as well: for integrable $h = u + i v$ , the definition gives $\int \overline{h} = \int u - i \int v = \overline{\int h}$ ; integrability of $\overline{h}$ follows from $∣ \overline{h} ∣ = ∣ h ∣$ Real and imaginary parts, complex conjugation, and modulus and Conjugation is an involutive real-field automorphism, $z \overline{z} = ∣ z ∣^{2}$ , and modulus is definite, multiplicative, and subadditive. Conjugate symmetry and first-variable linearity give conjugate-linearity in the second variable. For every class, $⟨ [ξ], [ξ] ⟩ = \int_{X} ∥ ξ (x) ∥^{2} d μ (x) \geq 0.$ This value is zero exactly when $∥ ξ (x) ∥^{2} = 0$ almost everywhere, by A nonnegative measurable function has integral $0$ exactly when it vanishes almost everywhere. Fibrewise positive definiteness says this is exactly when $ξ (x) = 0$ off a Borel null set, that is, when $[ξ] = [0]$ . Therefore the pairing is positive definite. These arguments use the ordinary complex inner-product axioms Real and complex inner-product spaces and their induced length and the integral convention above; they do not use any form of the axiom of choice.

No completion is part of this definition. The next theorem proves that this inner-product space is complete under its induced norm.

Boundary cases. If $X = \emptyset$ , or if every fibre is zero-dimensional, there is only the zero section and the quotient is the zero inner-product space. If $μ (X) = 0$ , then $X$ itself is a Borel null set, so every two square-integrable sections are equivalent and the quotient is again zero. For a one-point base $X = {x_{0}}$ with $μ ({x_{0}}) = m > 0$ and $H_{x_{0}} = C$ , take $e_{1} (x_{0}) = 1$ and $e_{n} (x_{0}) = 0$ for $n \geq 2$ . Every section has the form $ξ_{z} (x_{0}) = z$ and $⟨ [ξ_{z}], [ξ_{w}] ⟩ = m z \overline{w} .$ The direct integral is one-dimensional, with norm $\sqrt{m} ∣ z ∣$ . If instead $m = 0$ , the preceding zero-measure calculation applies.

TheoremStatement: AI-adaptedProof: AI-generatedprecheck passaudited 2026-09-30Open item page →

Direct integrals of measurable Hilbert fields are Hilbert spaces

Statement

Assume AC. Let $(X, B, μ)$ be a sigma-finite standard-Borel measure space, and let $(H_{x}, e_{n} (x))_{x \in X}$ be a measurable complex Hilbert field with a countable fundamental family. The direct integral $H = \int_{X}^{\oplus} H_{x} d μ (x)$ of Direct integral of a measurable Hilbert field is complete and separable. Hence, with its already-defined inner product, it is a separable Hilbert space. Inner products are linear in their first variable.

Facts & Assumptions

[F1]

The direct integral is the quotient of square-integrable measurable sections by equality off a measurable null set, and its inner product is the integral of the fibre inner products (Direct integral of a measurable Hilbert field).

[F2]

Each fibre is a complete Hilbert space, and the specified fundamental family has dense complex-linear span in that fibre (Hilbert space, Measurable Hilbert field from a countable fundamental family).

[F3]

Measurable sections have measurable pointwise norms and pairings, are closed under measurable scalar combinations, and are closed under pointwise norm limits (Measurable sections have measurable pointwise inner products).

[F4]

The fibre inner products are linear in the first variable, and the induced inner-product norm is absolutely homogeneous and satisfies the triangle inequality (Real and complex inner-product spaces and their induced length, The inner-product norm is definite, homogeneous, and satisfies the triangle inequality).

[F5]

Real Minkowski bounds the scalar $L^{2}$ norm of a finite sum, and complex $L^{2} (μ; C)$ has its quotient norm and satisfies Minkowski's inequality (Minkowski's inequality for integrals, including $p = \infty$ , The function space $L^{p} (μ)$ for $0 < p < \infty$ , Complex Lp classes and Euclidean test-function conventions, Complex Holder, Minkowski, and the quotient norm).

[F6]

Increasing nonnegative measurable functions satisfy monotone convergence (Monotone convergence for the integral).

[F7]

Pointwise almost-everywhere convergence under one integrable majorant implies convergence of the integrals (Dominated convergence).

[F8]

A nonnegative measurable function with finite integral is finite almost everywhere (A nonnegative measurable function with finite integral is finite almost everywhere).

[F9]

A sigma-finite measure has a countable cover by measurable finite-measure sets (Finite, sigma-finite, and semifinite measures).

[F10]

Finite and countable unions obey measure subadditivity (Finite and countable subadditivity of measures).

[F11]

Measures are continuous from below on increasing measurable sets (Continuity from below for measures), and when the smaller set has finite measure, a set difference has the corresponding difference of measures (Measure of a set difference when the smaller set has finite measure).

[F12]

A standard-Borel space has a measurable structure presented by a Polish space (Standard Borel spaces). Under AC it has a countable algebra generating that structure (Standard borel spaces have countable generating and measure determining algebras). The corollary's construction uses a bimeasurable coding into a Borel subset of $[0, 1]$ (Standard borel spaces admit bimeasurable real codings).

[F13]

A pi-system contained in a lambda-system has its generated sigma-algebra contained in that lambda-system (Dynkin's pi-lambda theorem).

[F14]

Complex finite simple functions whose nonzero sets have finite measure are dense in complex $L^{p}$ for finite $p$ , in particular in complex $L^{2}$ (Complex finite-simple and smooth compact-support density for finite p, Complex Lp classes and Euclidean test-function conventions).

[F15]

For a complete orthonormal family, Parseval's equality holds; the published result assumes Countable Choice (Parseval equivalences for an orthonormal family, Orthonormal families, complete orthonormal systems and Hilbert bases, The Axiom of Countable Choice ( ${A C}_{ω}$ )).

[F16]

AC gives a choice function on every family of nonempty sets, and hence in particular supplies Countable Choice (The Axiom of Choice, The Axiom of Countable Choice ( ${A C}_{ω}$ )).

[F17]

There is a bijection between $N^{2}$ and $N$ , a bijection between $Q$ and $N$ ( $N \times N \approx N$ , $Q$ is countably infinite). From a fixed bijection $β : N^{2} \to N$ , define $c_{0} (()) = 0$ and $c_{k + 1} (a_{0}, \dots, a_{k}) = β (a_{0}, c_{k} (a_{1}, \dots, a_{k}))$ ; then $c (a_{0}, \dots, a_{k - 1}) = β (k, c_{k} (a_{0}, \dots, a_{k - 1}))$ is an injective code for finite sequences. Every nonempty countable set can be enumerated by a surjection from $N$ (A nonempty set is at most countable iff it is a surjective image of $N$ ).

[F18]

Rational numbers are dense in $R$ , and complex modulus obeys the triangle inequality; complex numbers have real and imaginary coordinates (The rationals embed densely in the reals, Real and imaginary parts, complex conjugation, and modulus, Conjugation is an involutive real-field automorphism, $z \overline{z} = ∣ z ∣^{2}$ , and modulus is definite, multiplicative, and subadditive).

[F19]

Measurable scalar functions are closed under finite arithmetic and pointwise limits (Arithmetic and lattice operations preserve measurability whenever they are defined, Sequential suprema, infima, limsup, liminf, and pointwise limits of measurable functions are measurable).

[F20]

Composition with a Borel map preserves measurability; continuous maps have Borel preimages (Composition with a Borel measurable outer map preserves measurability, A continuous map has Borel preimages of Borel sets, A measurable function between measurable spaces).

[F21]

Inner-product Cauchy--Schwarz bounds a coefficient by the product of the two vector norms (Cauchy–Schwarz: $∣ ⟨ x, y ⟩ ∣ \leq ∥ x ∥ ∥ y ∥$ , with equality exactly for dependent pairs).

[F22]

The nonnegative integral is monotone and positively homogeneous (Monotonicity and nonnegative homogeneity of the nonnegative integral).

Proof

Proof technique: direct summable-subsequence construction for completeness; countable scalar simple functions and a measurable fibrewise orthonormal family for separability.

Given: AC, the sigma-finite standard-Borel measure space, the measurable Hilbert field and its countable fundamental family, and the direct-integral inner-product space.

1.1F1F3F16construct

Given a Cauchy sequence $(z_{n})$ in $H$ , for each $k \geq 1$ let $N_{k}$ be the least index after which every pair of terms is less than $2^{- k}$ apart. [given, F1, F3, F16, construct] Set $n_{1} = N_{1}$ and $n_{k + 1} = \max (N_{k + 1}, n_{k} + 1)$ . Then $(n_{k})$ is strictly increasing and $∥ z_{n_{k + 1}} - z_{n_{k}} ∥_{H} < 2^{- k}$ . Each class $z_{n_{k}}$ has a nonempty set of square-integrable measurable representatives; AC [F16] selects one representative $ξ_{k}$ for each $k$ . Put $d_{k} = ξ_{k + 1} - ξ_{k}$ and $h_{k} (x) = ∥ d_{k} (x) ∥$ . By [F1, F3, F16, construct], $d_{k}$ is a square-integrable measurable section, $h_{k}$ is measurable, and $∥ h_{k} ∥_{2} = ∥ [d_{k}] ∥_{H} < 2^{- k} .$

1.2F9F10F12

By [F12, F16], fix a countable algebra $A$ generating $B$ . [given, F9, F10, F12, F16] The cited real coding identifies the standard-Borel structure with a Borel subset of $[0, 1]$ , and the corollary takes the pullback algebra generated by rational cuts. By sigma-finiteness [F9], choose a sequence $(E_{j})$ of finite-measure measurable sets covering $X$ and set $X_{m} = ⋃_{j \leq m} E_{j}$ . Subadditivity [F10] makes each $X_{m}$ finite measure, and the sequence increases to $X$ .

1.3F2F3F4F20algebra

Recursively construct measurable fibrewise Gram--Schmidt sections. [given, F2, F3, F4, F20, construct] Put $v_{0} = e_{0}$ and, for $n > 0$ , $v_{n} (x) = e_{n} (x) - \sum_{j < n} ⟨ e_{n} (x), u_{j} (x) ⟩ u_{j} (x) .$ Set $u_{n} (x) = r (∥ v_{n} (x) ∥) v_{n} (x)$ , where $r (0) = 0$ and $r (t) = 1 / t$ for $t > 0$ . By [F3] each $v_{n}$ is measurable and its norm is measurable. The scalar map $r$ is Borel, being continuous on $(0, \infty)$ and defined separately on the Borel singleton ${0}$ ; [F20] makes $u_{n}$ measurable. At each fibre the nonzero $u_{n} (x)$ are orthonormal: for $ℓ < n$ with $u_{ℓ} (x) \neq 0$ , the induction hypothesis gives $⟨ v_{n} (x), u_{ℓ} (x) ⟩ = ⟨ e_{n} (x), u_{ℓ} (x) ⟩ - \sum_{j < n} ⟨ e_{n} (x), u_{j} (x) ⟩ ⟨ u_{j} (x), u_{ℓ} (x) ⟩ = 0;$ when $u_{ℓ} (x) = 0$ the pairing vanishes directly. Normalizing a nonzero $v_{n}$ preserves these orthogonality relations. Inductively, each $e_{n} (x)$ lies in the span of $u_{0} (x), \dots, u_{n} (x)$ : when $v_{n} (x) = 0$ it lies in the previous span, and otherwise $v_{n} (x) = ∥ v_{n} (x) ∥ u_{n} (x)$ . Their nonzero subfamily therefore has dense span in $H_{x}$ by [F2]. This handles dependent vectors without ever dividing by zero.

2.1step 1.1F5F6F19algebra

For $N \geq 1$ put $g_{N} = \sum_{k = 1}^{N} h_{k}$ . [step 1.1, F5] These are nonnegative measurable functions increasing in $N$ . Minkowski gives $∥ g_{N} ∥_{2} \leq \sum_{k = 1}^{N} ∥ h_{k} ∥_{2} < \sum_{k = 1}^{N} 2^{- k} < 1.$ Let $g = \lim_{N} g_{N} = \sup_{N} g_{N}$ . Scalar pointwise-limit measurability [F19] makes $g$ measurable. Monotone convergence [F6] applied to $g_{N}^{2}$ gives $\int_{X} g^{2} d μ = \lim_{N} \int_{X} g_{N}^{2} d μ \leq 1.$ Thus $g$ is finite outside the measurable null set $E = {x : g (x) = + \infty}$ by [F8].

2.2step 1.2F10F11F13algebra

Fix $m$ . [step 1.2, F13] $A_{m} = {A \cap X_{m} : A \in A}$ is a pi-system generating the trace sigma-algebra on $X_{m}$ . Let $D_{m}$ consist of measurable $B \subseteq X_{m}$ such that for every $ε > 0$ some $A_{m} \in A_{m}$ satisfies $μ (B △ A_{m}) < ε$ . It contains $A_{m}$ . It is a lambda-system: relative complements preserve symmetric-difference measure; for pairwise disjoint $B_{j} \in D_{m}$ , continuity from below and finite measure of $X_{m}$ give $μ ((⋃_{j} B_{j}) ∖ (⋃_{j \leq J} B_{j})) < ε / 2$ for some $J \geq 1$ . Approximate each of the first $J$ sets within $ε / (2 J)$ and take their finite union in $A_{m}$ ; finite subadditivity [F10] bounds the resulting symmetric difference by $ε$ . Dynkin's theorem [F13] now gives that every measurable subset of $X_{m}$ belongs to $D_{m}$ . If $B$ has finite measure in $X$ , then $B \cap X_{m} ↑ B$ , so [F11] gives $μ (B ∖ X_{m}) \to 0$ . It follows that every finite-measure measurable $B$ is approximable in measure by some $A \cap X_{m}$ with $A \in A$ : given $ε > 0$ , choose $m$ so that $μ (B ∖ X_{m}) < ε / 2$ , then apply $B \cap X_{m} \in D_{m}$ to choose $A_{m} = A \cap X_{m}$ with $μ ((B \cap X_{m}) △ A_{m}) < ε / 2$ . The inequality $μ (B △ A_{m}) \leq μ (B ∖ X_{m}) + μ ((B \cap X_{m}) △ A_{m})$ proves the required approximation.

2.3step 1.3F1F3F7F15F16F21F22algebra

At each $x$ , let $I_{x} = {n : u_{n} (x) \neq 0}$ . [step 1.3, F15, F16] It is an orthonormal family with dense span in $H_{x}$ by step 1.3, so Parseval gives, for each $w \in H_{x}$ , $∥ w ∥^{2} = \sum_{n \in I_{x}} ∣ ⟨ w, u_{n} (x) ⟩ ∣^{2},$ where the sum is the increasing limit of finite partial sums. In particular, for a measurable square-integrable section $ξ$ , the functions $a_{n} (x) = ⟨ ξ (x), u_{n} (x) ⟩$ are measurable by [F3] and lie in complex scalar $L^{2}$ by Cauchy--Schwarz [F21] and monotonicity [F22]. The finite coordinate sections $ξ_{N} = \sum_{n < N} a_{n} u_{n}$ are measurable. Finite orthogonality gives $0 \leq ∥ ξ (x) - ξ_{N} (x) ∥^{2} = ∥ ξ (x) ∥^{2} - \sum_{n < N} ∣ a_{n} (x) ∣^{2} \leq ∥ ξ (x) ∥^{2} .$ The initial finite subsets $I_{x} \cap {0, \dots, N - 1}$ exhaust the finite subsets of $I_{x}$ , so Parseval makes this residual tend pointwise to zero. Also $∥ ξ_{N} (x) ∥^{2} = \sum_{n < N} ∣ a_{n} (x) ∣^{2} \leq ∥ ξ (x) ∥^{2}$ , so each $ξ_{N}$ is square-integrable. Dominated convergence [F7] proves $∥ [ξ] - [ξ_{N}] ∥_{H} \to 0$ .

3.1step 2.1F1F2F3F20algebra

For $x \notin E$ , $\sum_{k} ∥ d_{k} (x) ∥ = g (x) < \infty$ , so completeness of $H_{x}$ gives a limit of $ξ_{1} (x) + \sum_{k \geq 1} d_{k} (x)$ . [step 2.1, F2] Define $ξ (x)$ to be that limit off $E$ and $0$ on $E$ . For each $N$ the section $ζ_{N} = 1_{X ∖ E} (ξ_{1} + \sum_{k = 1}^{N} d_{k})$ is measurable by [F3, F20]. The sequence $ζ_{N} (x)$ converges in norm to $ξ (x)$ for every $x$ , so $ξ$ is measurable by [F3]. Off $E$ , $∥ ξ (x) ∥ \leq ∥ ξ_{1} (x) ∥ + g (x)$ , while on $E$ it is zero. Therefore $∥ ξ (x) ∥^{2} \leq 2 ∥ ξ_{1} (x) ∥^{2} + 2 g (x)^{2}$ everywhere, and the right side has finite integral. Thus $ξ$ is square integrable and $[ξ] \in H$ .

3.2F5F14F17F18step 2.2algebra

The set $T = {A \cap X_{m} : A \in A, m \in N}$ is countable and consists of finite-measure sets. [step 2.2, F17] To see countability, enumerate the nonempty countable algebra and pair its indices with $m$ using [F17]. Every $T \in T$ has finite measure. Let $S$ be the scalar functions that are finite sums $\sum_{r < N} q_{r} 1_{T_{r}}$ with $q_{r} \in Q + i Q$ and $T_{r} \in T$ , including the empty sum. This family is countable by pairing the natural indices for $T_{r}$ and the rational real and imaginary parts, then using the finite-sequence code [F17]. Each member is measurable and lies in $L^{2}$ , since its support is a finite union of finite-measure sets. To prove density, fix $f \in L^{2} (μ; C)$ and $ε > 0$ . By [F14], choose a finite-measure-support simple function $s = \sum_{r < N} c_{r} 1_{B_{r}}$ with $μ (B_{r}) < \infty$ and $∥ f - s ∥_{2} < ε / 3$ . If $N = 0$ , the empty sum belongs to $S$ and already approximates $f$ within $ε$ . Suppose $N \geq 1$ and put $δ = ε / (3 N)$ . For each $r$ , write $c_{r} = a_{r} + i b_{r}$ . If $μ (B_{r}) = 0$ , take $q_{r}^{'} = 0$ . Otherwise rational density [F18] gives $p_{r}, q_{r} \in Q$ such that, with $q_{r}^{'} = p_{r} + i q_{r}$ , $∣ c_{r} - q_{r}^{'} ∣ \sqrt{μ (B_{r})} < δ / 2$ ; here $∣ c_{r} - q_{r}^{'} ∣ \leq ∣ a_{r} - p_{r} ∣ + ∣ b_{r} - q_{r} ∣$ by [F18]. For these fixed $q_{r}^{'}$ , approximate $B_{r}$ in measure by $T_{r} \in T$ using step 2.2, choosing $μ (B_{r} △ T_{r}) < (δ / (2 (1 + ∣ q_{r}^{'} ∣)))^{2}$ . Minkowski [F5] gives $∥ c_{r} 1_{B_{r}} - q_{r}^{'} 1_{T_{r}} ∥_{2} \leq ∣ c_{r} - q_{r}^{'} ∣ \sqrt{μ (B_{r})} + ∣ q_{r}^{'} ∣ \sqrt{μ (B_{r} △ T_{r})} < δ .$ Thus $t = \sum_{r < N} q_{r}^{'} 1_{T_{r}} \in S$ , and Minkowski gives $∥ s - t ∥_{2} \leq \sum_{r < N} ∥ c_{r} 1_{B_{r}} - q_{r}^{'} 1_{T_{r}} ∥_{2} < N δ = ε / 3$ . Hence $∥ f - t ∥_{2} < 2 ε / 3 < ε$ , so $S$ is dense in scalar $L^{2} (μ; C)$ .

4.1F1F4F7step 1.1step 3.1

For $x \notin E$ , [step 1.1, step 3.1, F1, F4, F7] $∥ ξ (x) - ξ_{k} (x) ∥ \leq \sum_{j \geq k} h_{j} (x) \leq g (x),$ and the left side tends to zero as $k \to \infty$ . Its square is measurable, converges to zero almost everywhere, and is dominated by $g^{2} \in L^{1} (μ)$ . Dominated convergence [F7] yields $∥ [ξ] - z_{n_{k}} ∥_{H}^{2} = \int_{X} ∥ ξ (x) - ξ_{k} (x) ∥^{2} d μ (x) ⟶ 0.$ Since $(z_{n})$ is Cauchy, for $ε > 0$ choose $K$ so that $∥ z_{n} - z_{m} ∥ < ε / 2$ for $n, m \geq K$ , then choose $k$ with $n_{k} \geq K$ and $∥ z_{n_{k}} - [ξ] ∥ < ε / 2$ . The triangle inequality [F4] gives $∥ z_{n} - [ξ] ∥ < ε$ for every $n \geq K$ . Thus the entire sequence converges, proving completeness.

4.2F1F3F4F5F17step 2.3step 3.2algebra

Define the countable candidate family $Q = {\sum_{n < N} s_{n} u_{n} : N \in N, s_{n} \in S}$ . [step 2.3, step 3.2, F17] Its elements are coded by finite sequences from the countable set of pairs $(n, s)$ , using [F17]. Every member is a square-integrable measurable section: pointwise triangle inequality and complex $L^{2}$ Minkowski give $∥ \sum_{n < N} s_{n} u_{n} ∥_{H} \leq \sum_{n < N} ∥ s_{n} ∥_{2} < \infty$ . Given $[ξ] \in H$ and $ε > 0$ , choose $N \geq 1$ so that $∥ [ξ] - [ξ_{N}] ∥ < ε / 2$ by step 2.3. For each $n < N$ , density of $S$ [step 3.2] supplies $s_{n}$ with $∥ [a_{n}] - [s_{n}] ∥_{L^{2} (μ; C)} < ε / (2 \sqrt{N})$ . Put $Y_{n} = {x : u_{n} (x) \neq 0}$ ; it is measurable because $∥ u_{n} (\cdot) ∥$ is measurable. Since $a_{n} = 0$ off $Y_{n}$ and $u_{n}$ has norm one on $Y_{n}$ , pointwise orthogonality gives ${∥ [ξ_{N}] - [\sum_{n < N} s_{n} u_{n}] ∥}_{H}^{2} = \sum_{n < N} \int_{Y_{n}} ∣ a_{n} - s_{n} ∣^{2} d μ \leq \sum_{n < N} ∥ [a_{n}] - [s_{n}] ∥_{L^{2}}^{2} < ε^{2} / 4.$ The triangle inequality proves that $Q$ is dense in $H$ . The only selections here are finitely many scalar approximants for a fixed $ξ, N$ ; no choice principle is used in constructing the countable set $Q$ .

5.1step 4.1step 4.2F1∎

Step 4.1 proves completeness, and step 4.2 provides a countable dense family. Together with the inner-product structure of [F1], this proves that $H$ is a separable Hilbert space.

Boundary cases

If $X = \emptyset$ , or all fibres are zero, the direct integral is the zero Hilbert space and its singleton is countable and dense. On a null base every square-integrable section represents zero, and the estimates above still apply. For a one-dimensional fibre, Gram--Schmidt yields at most one nonzero frame vector and Parseval is the one-coordinate identity. Dependent or zero fundamental vectors give $v_{n} = 0$ and are assigned $u_{n} = 0$ ; finite-measure exhaustions that stabilize and zero-measure pieces are included in the finite-measure and null-set arguments. There is no interval endpoint parameter. AC is used exactly as stated in the axiom_use field; least-index subsequence selection and Gram--Schmidt are canonical. The theorem has no iff assertion.

Source qualifications

Bekka--de la Harpe, Chapter 1 §1.G, printed pp. 59–60, define a countable fundamental family, measurable sections, the almost-everywhere quotient, and the integrated inner product, then state that the resulting space is Hilbert without proving completeness in that passage. Bruhat, Part III Chapter 10 §1.3, printed p. 95, explicitly says completeness follows by imitating the Riesz--Fischer proof but leaves the argument to the reader; §1.5, printed p. 96, gives fibrewise orthogonalization and zeroes a vector when its orthogonal remainder vanishes. Bruhat's framework is a locally compact topological/Lusin field, not this standard-Borel measurable convention. The summable-subsequence proof, null-set modification, finite-measure $π$ -- $λ$ approximation, and measurable Gram--Schmidt construction above supply the details in the present setting; no unstated Bruhat hypothesis is used.

DefinitionDefinition: AI-adaptedProof: AI-generatedprecheck passaudited 2026-09-30Open item page →

Measurable and decomposable operator fields

Definition

Let $(X, B, μ; H_{x}, e_{n} (x))$ be the measurable Hilbert field of Measurable Hilbert field from a countable fundamental family, and let $H = \int_{X}^{\oplus} H_{x} d μ (x)$ be the direct-integral inner-product space of Direct integral of a measurable Hilbert field. The inner products are linear in their first variables. An operator field is a family $(T_{x})_{x \in X}$ with $T_{x} \in B (H_{x})$ for every $x$ .

The field is weakly measurable when every fundamental matrix coefficient $x ⟼ ⟨ T_{x} e_{n} (x), e_{m} (x) ⟩$ is measurable. As proved below, this is equivalent to measurability of $x \mapsto ⟨ T_{x} ξ (x), η (x) ⟩$ for every pair of measurable sections $ξ, η$ . The proof obtains this equivalence from pointwise boundedness of each $T_{x}$ , so the equivalence also holds when the field is essentially bounded.

For a weakly measurable operator field, the operator-norm function $x \mapsto ∥ T_{x} ∥$ is measurable. Such a field is essentially bounded when ${ess sup}_{x \in X} ∥ T_{x} ∥ < \infty$ , with essential supremum taken with respect to $μ$ .

A bounded operator $S \in B (H)$ is decomposable if there is a weakly measurable, essentially bounded operator field $(T_{x})$ such that for every square-integrable measurable section $ξ$ , the pointwise section $(T ξ) (x) : = T_{x} ξ (x)$ is again square-integrable and $S [ξ] = [T ξ]$ . The square-integrability and class equality are part of this definition; the next theorem proves that every weakly measurable essentially bounded field indeed has this action. Operator fields are identified when equal off a measurable $μ$ -null set.

Facts & Assumptions

[F1]

Each fundamental vector $e_{n}$ is a measurable section because its Gram coefficients are measurable (Measurable Hilbert field from a countable fundamental family).

[F2]

The section lemma proves the coefficient test for measurability and its equivalence with testing all measurable-section pairings (Measurable sections have measurable pointwise inner products).

[F3]

Let $β : N^{2} \to N$ be the bijection in [F8]. Define $c_{0} (()) = 0$ and $c_{k + 1} (a_{0}, \dots, a_{k}) = β (a_{0}, c_{k} (a_{1}, \dots, a_{k}))$ . Then $c (a_{0}, \dots, a_{k - 1}) = β (k, c_{k} (a_{0}, \dots, a_{k - 1}))$ is injective on all finite sequences: the outer pairing recovers the length, and repeated inverse pairing recovers each entry. This is the finite-sequence code used below.

[F4]

The direct integral is the quotient of square-integrable measurable sections by almost-everywhere equality (Direct integral of a measurable Hilbert field).

[F5]

$B (X, Y)$ consists of bounded linear operators between normed spaces (The spaces (\mathcal B(X,Y)) and (\mathcal B(X)) of bounded linear operators).

[F6]

The operator norm is the supremum over the unit ball; rescaling gives $∥ T u ∥ \leq ∥ T ∥ ∥ u ∥$ (The operator norm as the least bound and as the unit-sphere or unit-ball supremum).

[F7]

Cauchy--Schwarz gives $∣ ⟨ u, v ⟩ ∣ \leq ∥ u ∥ ∥ v ∥$ (Cauchy–Schwarz: $∣ ⟨ x, y ⟩ ∣ \leq ∥ x ∥ ∥ y ∥$ , with equality exactly for dependent pairs).

[F8]

There is a bijection between $N^{2}$ and $N$ ( $N \times N \approx N$ ).

[F9]

A countable supremum of measurable extended-real functions is measurable (Sequential suprema, infima, limsup, liminf, and pointwise limits of measurable functions are measurable).

[F10]

The essential supremum of a measurable real function is the infimum of its almost-everywhere bounds; finite essential supremum defines essential boundedness (The essential supremum of a measurable function with respect to a measure).

[F11]

Measurability is inverse-image measurability (A measurable function between measurable spaces).

[F12]

Composition with a Borel map preserves measurability (Composition with a Borel measurable outer map preserves measurability).

[F13]

Continuous maps have Borel preimages (A continuous map has Borel preimages of Borel sets).

[F14]

Complex modulus is subadditive and multiplicative (Conjugation is an involutive real-field automorphism, $z \overline{z} = ∣ z ∣^{2}$ , and modulus is definite, multiplicative, and subadditive).

[F15]

$d_{C} (z, w) = ∣ z - w ∣$ is the Euclidean metric on $R^{2}$ (The Euclidean metric, convergence, Cauchy sequences, and continuity on the complex plane).

[F16]

Rational boxes form a countable basis in each finite-dimensional real coordinate space ( $Q^{n}$ is a countable dense subset of $R^{n}$ , and rational open boxes form a countable basis).

[F17]

The inner product is linear in its first variable and conjugate-linear in its second (Real and complex inner-product spaces and their induced length).

[F18]

There is a bijection between $Q$ and $N$ ( $Q$ is countably infinite).

[F19]

Rational numbers approximate every real and lie strictly between any two distinct reals (The rationals embed densely in the reals).

[F20]

The complex-linear span of the fundamental family is dense in every fibre (Measurable Hilbert field from a countable fundamental family).

[F21]

The fibre norm is absolutely homogeneous and satisfies the triangle inequality (The inner-product norm is definite, homogeneous, and satisfies the triangle inequality).

[F22]

On a nonzero domain the operator norm is the supremum over the unit sphere (The operator norm as the least bound and as the unit-sphere or unit-ball supremum).

[F23]

Measurable sections have measurable pointwise norms and pairings, and are closed under measurable scalar combinations and pointwise norm limits (Measurable sections have measurable pointwise inner products).

Proof

technique · direct

Given: The measurable field, its countable fundamental family, and a pointwise field $T_{x} \in B (H_{x})$ whose matrix coefficients are measurable.

1.1F1F3F8F14F18F19F20F21F23construct

Fix a bijection $ρ : Q \to N$ from [F18] and a bijection $β : N^{2} \to N$ from [F8]. Encode a term $(n, a, b)$ , representing $(a + i b) e_{n}$ , by $κ (n, a, b) = β (n, β (ρ (a), ρ (b)))$ . Use the injective finite-sequence code [F3] on each finite list of term codes; if a natural $j$ is not in the code's range, decode it as the empty list, and otherwise use its unique decoded list. Define $q_{j} (x) = \sum_{r} (a_{r} + i b_{r}) e_{n_{r}} (x)$ over that list, with the empty sum zero. Every rational-complex finite combination occurs, and each $q_{j}$ is a measurable section by [F1,F23]. These values are dense in every fibre: given $v \in H_{x}$ and $ε > 0$ , choose a finite complex combination $u = \sum_{r = 1}^{N} z_{r} e_{n_{r}} (x)$ with $∥ v - u ∥ < ε / 2$ by [F20]. Put $C = \sum_{r} ∥ e_{n_{r}} (x) ∥$ and use [F19] to choose a rational $η$ with $0 < η < ε / (4 (1 + C))$ ; approximate each real and imaginary coordinate of $z_{r}$ within $η$ by rationals $a_{r}, b_{r}$ . Absolute homogeneity and the triangle inequality [F21], together with the complex modulus triangle inequality [F14], give $∥ u - \sum_{r} (a_{r} + i b_{r}) e_{n_{r}} (x) ∥ \leq 2 η C < ε / 2$ . Hence $q_{j} (x)$ is dense. This construction fixes only individual bijections and makes no arbitrary selections, so it uses no AC.

1.2F2F11F12F13F14F15F16F17construct

Fix $j, m$ and write $q_{j} = \sum_{r = 1}^{N} c_{r} e_{n_{r}}$ with its finite rational-complex coefficients. First-variable linearity [F17] gives $⟨ T_{x} q_{j} (x), e_{m} (x) ⟩ = \sum_{r} c_{r} ⟨ T_{x} e_{n_{r}} (x), e_{m} (x) ⟩$ . If this combination is empty, then $T_{x} q_{j} (x) = 0$ and is measurable. Otherwise the finitely many complex coefficient functions form a measurable map into their real-coordinate space because rational boxes give a countable basis [F15,F16]; for input tuples $z, w$ , [F14] gives $∣ \sum_{r} c_{r} z_{r} - \sum_{r} c_{r} w_{r} ∣ \leq \sum_{r} ∣ c_{r} ∣ ∣ z_{r} - w_{r} ∣$ , so the finite-sum map is continuous in these finitely many coordinates and hence Borel [F13]; composition [F12] makes this coefficient measurable. The coefficient criterion [F2] therefore makes $x \mapsto T_{x} q_{j} (x)$ a measurable section.

2.1F11F12F13F21F23step 1.1step 1.2construct

Put $r_{j} (x) = 1 / ∥ q_{j} (x) ∥$ when $∥ q_{j} (x) ∥ > 0$ and $r_{j} (x) = 0$ otherwise, and set $u_{j} (x) = r_{j} (x) q_{j} (x)$ . Since $∥ q_{j} ∥$ is measurable [F23], define $g : [0, \infty) \to R$ by $g (0) = 0$ and $g (t) = 1 / t$ for $t > 0$ . For every Borel $A \subseteq R$ , $g^{- 1} (A)$ is the union of ${0}$ when $0 \in A$ and $(0, \infty) \cap (t \mapsto 1 / t)^{- 1} (A \cap (0, \infty))$ , a Borel set by continuity of reciprocal on $(0, \infty)$ and [F13]. Thus $g$ is Borel and $r_{j} = g \circ ∥ q_{j} ∥$ is measurable by [F11,F12]; hence $u_{j}$ is a measurable section by [F23]. It has norm one when $q_{j} (x) \neq 0$ and is zero otherwise. On a nonzero fibre, the nonzero normalized $q_{j} (x)$ are dense in the unit sphere: for each unit vector $u$ and each $k$ , take the least $j$ with $∥ q_{j} (x) - u ∥ < 1 / k$ ; these approximants are eventually nonzero and their normalizations converge to $u$ , since $∥ q_{j} / ∥ q_{j} ∥ - u ∥ \leq 2 ∥ q_{j} - u ∥$ when $q_{j} \neq 0$ . This is a least-index construction, not a choice of arbitrary witnesses. By step 1.2, each $T_{x} q_{j}$ is measurable; measurable scalar closure [F23] makes $T_{x} u_{j}$ measurable.

2.2F1F5F23step 1.1step 1.2

Suppose first that the fundamental matrix coefficients are measurable. For any measurable section $ξ$ and $k \geq 1$ , the sets $E_{j, k} = {x : ∥ ξ (x) - q_{j} (x) ∥ < 1 / k}$ are measurable by [F23] and cover $X$ by step 1.1. At each $x$ take the least qualifying $j_{k} (x)$ ; the sets $E_{j, k} ∖ ⋃_{i < j} E_{i, k}$ form a measurable partition; call its pieces $P_{j, k}$ . For each $m$ and Borel $B \subseteq C$ , the inverse image of $B$ under the $m$ th coefficient of $ξ_{k} (x) = q_{j_{k} (x)} (x)$ is $⋃_{j} (P_{j, k} \cap {x : ⟨ q_{j} (x), e_{m} (x) ⟩ \in B})$ , which is measurable, so $ξ_{k}$ is a measurable section and $∥ ξ_{k} (x) - ξ (x) ∥ < 1 / k$ . The inverse image under the $m$ th coefficient of $T ξ_{k}$ is $⋃_{j} (P_{j, k} \cap {x : ⟨ T_{x} q_{j} (x), e_{m} (x) ⟩ \in B})$ , which is also measurable; thus $T ξ_{k}$ is measurable. Pointwise boundedness of $T_{x}$ [F5] gives $∥ T_{x} ξ_{k} (x) - T_{x} ξ (x) ∥ \leq ∥ T_{x} ∥ / k \to 0$ for every $x$ , so [F23] makes $T ξ$ measurable and its pairing with every measurable $η$ measurable. Conversely, if all section pairings are measurable, test on $ξ = e_{n}$ and $η = e_{m}$ , both measurable by [F1], to recover every fundamental coefficient. This proves both directions of the equivalence; no essential bound is needed because each individual $T_{x}$ is bounded.

3.1F6F7F8F9F12F13F14F15F17F22F23step 2.1step 1.2algebra

For each $j, k$ , the pairing $x \mapsto ⟨ T_{x} u_{j} (x), u_{k} (x) ⟩$ is measurable by [F23]. Modulus is continuous by the reverse triangle inequality from [F14,F15], so [F13] makes these moduli Borel and [F12] makes them measurable. For every nonzero fibre, $∥ T_{x} ∥ = \sup_{j, k \in N} ∣ ⟨ T_{x} u_{j} (x), u_{k} (x) ⟩ ∣$ . The upper bound follows from [F6,F7], since $∥ u_{j} ∥, ∥ u_{k} ∥ \leq 1$ . For the reverse bound, [F22] gives $∥ T_{x} ∥ = \sup_{∥ u ∥ = 1} ∥ T_{x} u ∥$ , and Cauchy--Schwarz with $v = T_{x} u / ∥ T_{x} u ∥$ when $T_{x} u \neq 0$ gives $∥ T_{x} u ∥ = \sup_{∥ v ∥ = 1} ∣ ⟨ T_{x} u, v ⟩ ∣$ . The normalized $q_{j}$ are dense in the unit sphere by step 2.1. First-variable linearity [F17], the modulus triangle inequality [F14], and [F6,F7] give $∣ ⟨ T_{x} u, v ⟩ - ⟨ T_{x} u_{j}, u_{k} ⟩ ∣ \leq ∥ T_{x} ∥ (∥ u - u_{j} ∥ + ∥ v - u_{k} ∥)$ , proving continuity in both unit vectors. On a zero fibre all $u_{j}$ and the operator norm are zero, so the same supremum formula holds. Use [F8] to enumerate pairs $(j, k)$ by one natural index; [F9] then makes $x \mapsto ∥ T_{x} ∥$ measurable, including on zero fibres.

4.1F4F5F10step 3.1construct∎

Since step 3.1 proves $∥ T_{x} ∥$ measurable and nonnegative, [F10] defines ${ess sup}_{x} ∥ T_{x} ∥$ and the field is essentially bounded exactly when that value is finite. By [F4], $S \in B (H)$ is decomposable when there is such a field for which each pointwise action $T ξ$ belongs to the square-integrable section space and $S [ξ] = [T ξ]$ for every class. This states the well-definedness required by the formula; it does not presume the next theorem's existence result. If two fields agree off a measurable null set, their pointwise actions agree there, so they give the same direct-integral class. The definition and measurable-field arguments use no axiom of choice.

Boundary cases

If $X = \emptyset$ , its unique operator field has no coefficients to test; measurability is vacuous, and the essential supremum is $0$ because every $M \geq 0$ is an almost-everywhere bound. The direct integral is the zero space, so its only bounded operator is decomposable. If every fibre is zero, then $T_{x} = 0$ and every tested coefficient and operator norm is zero; the direct integral is again the zero space. The normalization in step 2.1 sends every zero test vector to zero without division by zero. On a one-point base of finite positive mass with $H_{x_{0}} = C$ , $e_{1} (x_{0}) = 1$ , and $e_{n} (x_{0}) = 0$ for every $n \in N ∖ {1}$ , the field $T_{x_{0}} z = a z$ has matrix coefficient $a$ at $(n, m) = (1, 1)$ and zero coefficients otherwise, and its norm is $∣ a ∣$ . The field acts by scalar multiplication on the one-dimensional direct integral. If the whole base is null, every measurable norm is zero almost everywhere, so the essential bound and direct integral are zero. There is no endpoint parameter. The construction uses the previously coded countable family and least qualifying indices, with no axiom of choice.

Source qualifications

Bekka--de la Harpe define a measurable operator field by testing all measurable sections and state that an essentially bounded field induces the pointwise operator, with operator norm equal to the essential supremum; their displayed passage cites Dixmier--von Neumann for that norm assertion. That passage does not prove the countable coefficient criterion or measurability of the norm, which are established above. Bruhat's Proposition 6 uses a locally compact, Lusin/topological field, local boundedness, and continuity off sets of small measure. His §1.8 also states the action result in that setting. These are contextual comparisons only; neither is used to import hypotheses or proof steps into the standard-Borel measurable-field definition here.

TheoremStatement: AI-adaptedProof: AI-adaptedprecheck passaudited 2026-09-30Open item page →

Measurable essentially bounded operator fields act decomposably

Statement

Assume AC. Let $(X, B, μ)$ and $(H_{x}, e_{n} (x))_{x \in X}$ be a measurable complex Hilbert field with countable fundamental family, over the sigma-finite standard-Borel base of Measurable Hilbert field from a countable fundamental family. Put $H = \int_{X}^{\oplus} H_{x} d μ (x),$ the Hilbert space of Direct integrals of measurable Hilbert fields are Hilbert spaces. Every weakly measurable essentially bounded field $(T_{x})_{x \in X}$ from Measurable and decomposable operator fields induces a well-defined bounded operator $T = \int_{X}^{\oplus} T_{x} d μ (x) \in B (H),$ given on classes by $T [ξ] = [x \mapsto T_{x} ξ (x)]$ , and $∥ T ∥ = {ess sup}_{x \in X} ∥ T_{x} ∥ .$ For two such fields $T$ and $S$ , the adjoint field $(T_{x}^{*})$ and product field $(T_{x} S_{x})$ are weakly measurable and essentially bounded. Their induced operators are respectively $T^{*}$ and $T S$ . All inner products are linear in their first variable.

Facts & Assumptions

[F1]

The fundamental sequence has dense complex span in each fibre, and each fundamental vector is a measurable section (Measurable Hilbert field from a countable fundamental family).

[F2]

The direct integral is the quotient of square-integrable measurable sections by almost-everywhere equality, and its inner product is the integral of the fibre inner products (Direct integral of a measurable Hilbert field).

[F3]

Under AC, this direct integral is a complete Hilbert space (Direct integrals of measurable Hilbert fields are Hilbert spaces).

[F4]

Weak measurability means that every fundamental matrix coefficient of the operator field is measurable (Measurable and decomposable operator fields).

[F5]

For a weakly measurable field, pairings against every pair of measurable sections are measurable (Measurable and decomposable operator fields).

[F6]

Measurable sections have measurable pointwise norms and pairings and are closed under measurable scalar combinations (Measurable sections have measurable pointwise inner products).

[F7]

The pointwise operator-norm function of a weakly measurable field is measurable (Measurable and decomposable operator fields).

[F8]

A finite essential supremum is an almost-everywhere bound and is the least such bound (The essential supremum of a measurable function with respect to a measure, The essential supremum is attained as the least essential bound).

[F9]

The nonnegative integral is monotone, homogeneous, and additive (Monotonicity and nonnegative homogeneity of the nonnegative integral, Additivity of the nonnegative Lebesgue integral).

[F10]

A nonnegative function has zero integral over a measurable null set, and integration over a measurable set is multiplication by its indicator (A nonnegative integral over a null set vanishes, Integral over a measurable subset).

[F11]

For an indicator function, the nonnegative integral equals the measure of its set (The nonnegative integral agrees with the simple integral on simple functions, The integral of a nonnegative simple function).

[F12]

A sigma-finite measure space has a countable cover by measurable sets of finite measure (Finite, sigma-finite, and semifinite measures).

[F13]

$B (X, Y)$ is the set of bounded linear operators from $X$ to $Y$ (The spaces (\mathcal B(X,Y)) and (\mathcal B(X)) of bounded linear operators).

[F14]

The operator norm satisfies $∥ A u ∥ \leq ∥ A ∥ ∥ u ∥$ and on a nonzero domain is the supremum over the unit sphere (The operator norm as the least bound and as the unit-sphere or unit-ball supremum).

[F15]

Composition of bounded operators satisfies $∥ A B ∥ \leq ∥ A ∥ ∥ B ∥$ (Composition satisfies |ST|\le|S|,|T|).

[F16]

AC implies Countable Choice; under Countable Choice bounded operators between Hilbert spaces have unique adjoints satisfying $⟨ A u, v ⟩ = ⟨ u, A^{*} v ⟩$ , $∥ A^{*} ∥ = ∥ A ∥$ (The Axiom of Choice, The Axiom of Countable Choice ( ${A C}_{ω}$ ), Hilbert-adjoint identities).

[F17]

In real coordinates complex conjugation is the reflection $(a, b) \mapsto (a, - b)$ , which preserves the Euclidean complex metric; every continuous map has Borel preimages (Real and imaginary parts, complex conjugation, and modulus, The Euclidean metric, convergence, Cauchy sequences, and continuity on the complex plane, A continuous map has Borel preimages of Borel sets).

[F18]

Composing a measurable map with a Borel map preserves measurability, with measurability understood through inverse images (Composition with a Borel measurable outer map preserves measurability, A measurable function between measurable spaces).

[F19]

The complex modulus is $1$ -Lipschitz for the Euclidean metric, since modulus subadditivity gives $∣ ∣ z ∣ - ∣ w ∣ ∣ \leq ∣ z - w ∣$ (The Euclidean metric, convergence, Cauchy sequences, and continuity on the complex plane, Conjugation is an involutive real-field automorphism, $z \overline{z} = ∣ z ∣^{2}$ , and modulus is definite, multiplicative, and subadditive).

[F20]

Cauchy--Schwarz holds in each fibre: $∣ ⟨ u, v ⟩ ∣ \leq ∥ u ∥ ∥ v ∥$ (Cauchy–Schwarz: $∣ ⟨ x, y ⟩ ∣ \leq ∥ x ∥ ∥ y ∥$ , with equality exactly for dependent pairs).

[F21]

$Q$ is in bijection with $N$ and $N^{2}$ is in bijection with $N$ ( $Q$ is countably infinite, $N \times N \approx N$ ). Given a fixed bijection $β : N^{2} \to N$ , define $c_{0} (()) = 0$ and $c_{k + 1} (a_{0}, \dots, a_{k}) = β (a_{0}, c_{k} (a_{1}, \dots, a_{k}))$ ; the code $c (a_{0}, \dots, a_{k - 1}) = β (k, c_{k} (a_{0}, \dots, a_{k - 1}))$ injects all finite sequences of naturals into $N$ .

[F22]

Rational numbers are dense in the real numbers (The rationals embed densely in the reals).

[F23]

Fibre norms are absolutely homogeneous and satisfy the triangle inequality (The inner-product norm is definite, homogeneous, and satisfies the triangle inequality).

[F24]

Complex modulus is subadditive (Conjugation is an involutive real-field automorphism, $z \overline{z} = ∣ z ∣^{2}$ , and modulus is definite, multiplicative, and subadditive).

[F25]

Inner products are linear in their first variable and conjugate-linear in the second (Real and complex inner-product spaces and their induced length).

[F26]

Finite and countable unions satisfy measure subadditivity (Finite and countable subadditivity of measures).

[F27]

Each fundamental vector $e_{n}$ is a measurable section (Measurable Hilbert field from a countable fundamental family).

Proof

technique · direct, using pointwise localization and a countable dense family in each fibre

Given: AC, the measurable Hilbert field, and one or two weakly measurable essentially bounded operator fields on it.

1.1F5F27

Let $ξ$ be a measurable section. For each fundamental vector $e_{m}$ , the function $x \mapsto ⟨ T_{x} ξ (x), e_{m} (x) ⟩$ is measurable by the all-sections coefficient criterion in [F5]. Thus all fundamental coefficients of $x \mapsto T_{x} ξ (x)$ are measurable, so this pointwise image is a measurable section.

1.2F1F6F21F22F23F24F27construct

Enumerate all finite rational-complex linear combinations of the fundamental sections as $(q_{j})_{j \in N}$ . Such an enumeration exists by [F21]; each $q_{j}$ is a measurable section by [F6] and [F27]. The family is pointwise dense: first approximate a given vector by a finite complex combination using [F1], then approximate each of its finitely many complex coefficients by an element of $Q + i Q$ using [F22]. If $u = \sum_{r = 1}^{N} z_{r} e_{n_{r}} (x)$ is such a combination and $a_{r} + i b_{r}$ are the rational approximants, [F23] and [F24] bound the replacement error by $\sum_{r = 1}^{N} ∣ z_{r} - (a_{r} + i b_{r}) ∣ ∥ e_{n_{r}} (x) ∥$ , which can be made arbitrarily small.

1.3F4F16F17F18F25algebra

For each $x$ , the fibre adjoint $T_{x}^{*}$ exists by [F16]. With inner products linear in the first variable [F25], its matrix coefficients satisfy $⟨ T_{x}^{*} e_{n} (x), e_{m} (x) ⟩ = \overline{⟨ T_{x} e_{m} (x), e_{n} (x) ⟩}$ . Conjugation is an isometry because if $z = a + b i$ and $w = c + d i$ then $d_{C} (\overline{z}, \overline{w})^{2} = (a - c)^{2} + (b - d)^{2} = d_{C} (z, w)^{2}$ . Thus it is continuous and Borel by [F17]; [F4] and [F18] then make the right side measurable. Hence $T_{x}^{*}$ is weakly measurable. Also $∥ T_{x}^{*} ∥ = ∥ T_{x} ∥$ by [F16], so it is essentially bounded with the same essential bound.

2.1F2F6F7F8F9F10F13F14step 1.1

Put $s = {ess sup}_{x} ∥ T_{x} ∥ < \infty$ , and choose a measurable null set $N$ outside which $∥ T_{x} ∥ \leq s$ by [F8]. For $f (x) = ∥ T_{x} ξ (x) ∥^{2}$ , step 1.1 and [F6] give nonnegative measurability. The pointwise operator bound in [F14] gives $f 1_{X ∖ N} \leq s^{2} ∥ ξ ∥^{2}$ . By [F9] and [F10], $\int_{X} f d μ = \int_{X} f 1_{X ∖ N} d μ + \int_{X} f 1_{N} d μ \leq s^{2} \int_{X} ∥ ξ ∥^{2} d μ < \infty$ , where the second integral is zero because $N$ is null. Hence the pointwise image is square-integrable. If $ξ = ξ^{'}$ outside a measurable null set, then $T_{x} ξ (x) = T_{x} ξ^{'} (x)$ there. If two field representatives agree outside a measurable null set, their pointwise images also agree there. Pointwise linearity and the same estimate show that $[ξ] \mapsto [T ξ]$ is a well-defined bounded linear operator on $H$ , independent of both section and field representatives, with $∥ T ∥ \leq s$ .

2.2F5F6F14F17F18F19F20F21F23step 1.2construct

Define $u_{j} (x) = q_{j} (x) / ∥ q_{j} (x) ∥$ when $q_{j} (x) \neq 0$ and $u_{j} (x) = 0$ otherwise. The norm is measurable by [F6]. The function $g (0) = 0$ and $g (t) = 1 / t$ for $t > 0$ is Borel, by splitting each preimage into its possible zero part and the preimage under the continuous reciprocal on $(0, \infty)$ using [F17]. Thus $u_{j} = (g \circ ∥ q_{j} ∥) q_{j}$ is measurable by [F6] and [F18]. It has norm at most one, and the nonzero $u_{j} (x)$ are dense in the unit sphere of each nonzero fibre: given a unit $v$ and $ε > 0$ , choose $q_{j} (x)$ with $∥ q_{j} (x) - v ∥ < \min (ε / 2, 1 / 2)$ by step 1.2. Then $q_{j} (x) \neq 0$ and $∥ q_{j} (x) / ∥ q_{j} (x) ∥ - v ∥ \leq 2 ∥ q_{j} (x) - v ∥ < ε$ . Consequently, for every fibre, $∥ T_{x} ∥ = \sup_{j, k \in N} ∣ ⟨ T_{x} u_{j} (x), u_{k} (x) ⟩ ∣$ . On a nonzero fibre, the operator norm is the supremum of $∥ T_{x} u ∥$ over unit $u$ by [F14], and $∥ T_{x} u ∥ = \sup_{∥ v ∥ = 1} ∣ ⟨ T_{x} u, v ⟩ ∣$ by [F20] and the choice $v = T_{x} u / ∥ T_{x} u ∥$ when $T_{x} u \neq 0$ . The pairing is continuous in both unit vectors: its change is at most $∥ T_{x} ∥ (∥ u - u^{'} ∥ + ∥ v - v^{'} ∥)$ by [F14] and [F20]. Density of both normalized families therefore gives the displayed supremum. On a zero fibre every term and $∥ T_{x} ∥$ are zero. Each displayed coefficient is measurable by [F5] and [F18], and its modulus is measurable because [F19] makes the modulus continuous and [F18] preserves measurability under composition. Thus the norm is a countable measurable supremum.

2.3F5F8F15F26step 1.1

Let $S$ be another weakly measurable essentially bounded field. By step 1.1, $x \mapsto S_{x} e_{n} (x)$ is a measurable section. The all-sections criterion [F5], applied to $T$ and that section, shows that every fundamental coefficient $⟨ T_{x} S_{x} e_{n} (x), e_{m} (x) ⟩$ is measurable. Hence $(T_{x} S_{x})$ is weakly measurable. The pointwise composition bound [F15] and the two essential bounds [F8] show that $∥ T_{x} S_{x} ∥$ is bounded almost everywhere by the product of the two essential suprema: remove the union of their two null exceptional sets, which is null by [F26]. Thus the product field is essentially bounded.

3.1F2F6F8F9F10F11F12F14F20F22F26step 2.1step 2.2algebra

If $s = 0$ , step 2.1 gives $T = 0$ and hence $∥ T ∥ = s$ . Suppose $s > 0$ , and fix $0 < c < s$ . The set $A_{c} = {x : ∥ T_{x} ∥ > c}$ has positive measure, since otherwise $c$ would be an almost-everywhere bound contradicting the leastness of $s$ in [F8]. By the countable supremum in step 2.2, $A_{c} = ⋃_{j, k \in N, n \in N_{> 0}} {x : ∣ ⟨ T_{x} u_{j} (x), u_{k} (x) ⟩ ∣ > c + 1 / n}$ . Countable subadditivity [F26] gives one such set positive measure. Intersect it with a set of a sigma-finite cover from [F12] so the resulting measurable set $E$ has finite positive measure. On $E$ both test vectors have norm one; by Cauchy--Schwarz in [F20], $∥ T_{x} u_{j} (x) ∥ > c + 1 / n$ . For $ζ = 1_{E} u_{j}$ , the section lemma [F6] gives measurability, and it is square-integrable because $∥ ζ (x) ∥^{2} = 1_{E} (x)$ . Hence $∥ ζ ∥^{2} = μ (E) > 0$ and $∥ T ζ ∥^{2} = \int_{E} ∥ T_{x} u_{j} (x) ∥^{2} d μ (x) \geq (c + 1 / n)^{2} μ (E)$ . The indicator integral is [F11], [F10] identifies it as an integral over $E$ , and [F9] gives the comparison. Since $μ (E) > 0$ , its defining unit-vector supremum [F14] gives $∥ T ∥ \geq ∥ T ζ ∥ / ∥ ζ ∥ \geq c + 1 / n > c$ . If $∥ T ∥ < s$ , rational density [F22] gives a $c$ strictly between $∥ T ∥$ and $s$ , contradicting the established inequality; hence $∥ T ∥ \geq s$ , and step 2.1 proves equality.

4.1F2F3F16step 1.1step 1.3step 2.1step 2.3∎

Let $\tilde{T} = \int^{\oplus} T_{x}^{*} d μ$ be the induced operator from step 2.1 applied to the adjoint field. For direct-integral vectors $ξ, η$ , the fibre adjoint identity gives $⟨ T ξ, η ⟩_{H} = \int_{X} ⟨ T_{x} ξ (x), η (x) ⟩ d μ (x) = \int_{X} ⟨ ξ (x), T_{x}^{*} η (x) ⟩ d μ (x) = ⟨ ξ, \tilde{T} η ⟩_{H}$ . Uniqueness of Hilbert adjoints in [F16] therefore gives $\tilde{T} = T^{*}$ . For the product field, step 2.1 and the definition of induced action yield, for every $ξ \in H$ , $(\int^{\oplus} T_{x} S_{x} d μ) ξ = [x \mapsto T_{x} (S_{x} ξ (x))] = T (S ξ)$ , so its induced operator is $T S$ . This proves adjoint and multiplication compatibility.

Boundary cases

If $X = \emptyset$ , the direct integral and its only induced operator are zero, and the essential supremum of the empty norm field is zero. If every fibre is zero, all fields and induced operators are zero. For a one-point base of mass $m > 0$ with fibre $C$ and $T_{x_{0}} z = a z$ , the induced operator is the same scalar map and has norm $∣ a ∣ = ∥ T_{x_{0}} ∥$ . A zero field on a nonzero direct integral has $s = 0$ and is covered by step 3.1. Zero test vectors are assigned the zero normalized section, so no division by zero occurs. The exact-norm argument has no interval endpoint parameter. AC is used through the Hilbert-space and adjoint suppliers in [F3] and [F16]; the countable test family and countable-union localization use no further choice. The theorem has no iff assertion, so the two iff boundary axes do not apply.

Source qualifications

Bekka--de la Harpe, §1.G.3, printed p. 60, states the pointwise field action and essential-supremum norm formula, but cites Dixmier--von Neumann for the norm equality; §1.H.1 states the general commutant theorem and refers its proof to Dixmier. Neither cited passage supplies the varying-fibre proof written here. Bruhat, Part III Chapter 10 §§1.7–1.8, printed pp. 99–101, works with a locally bounded operator field and a Lusin/topological measurability convention. It gives the measurable action and upper norm bound, and states an exact norm formula in Theorem 2; its printed argument does not supply the countable localization used here for the lower bound. Those conventions and source statements are context only; the proof above derives the result for the standard-Borel countable-fundamental-family convention of this pair.

TheoremStatement: AI-adaptedProof: AI-adaptedprecheck passaudited 2026-09-30Open item page →

Decomposable operators are the commutant of diagonal multiplication

Statement

Assume AC. Let $(X, B, μ)$ be a sigma-finite standard-Borel measure space, let $(H_{x}, e_{n} (x))_{x \in X}$ be a measurable complex Hilbert field with a countable fundamental family, and set $H = \int_{X}^{\oplus} H_{x} d μ (x),$ with the direct-integral Hilbert-space structure. For each $f \in L^{\infty} (X, μ)$ let $M_{f} \in B (H)$ act by scalar multiplication, and put $D = {M_{f} : f \in L^{\infty} (X, μ)}$ . Then $D^{'} = {S \in B (H) : S is decomposable},$ where decomposable has the meaning in Measurable and decomposable operator fields. For a fixed $S$ , any weakly measurable essentially bounded field inducing $S$ is unique up to a $μ$ -null set.

Facts & Assumptions

Given: AC; a sigma-finite standard-Borel measure space; a measurable complex Hilbert field with countable fundamental family; its direct-integral Hilbert space; the scalar-multiplication algebra $D$ ; and, for the converse direction, a bounded operator $S \in D^{'}$ .

[F1]

The fibres are separable complex Hilbert spaces; measurable sections are characterized by Borel fundamental coefficients, and each fundamental vector is measurable (Measurable Hilbert field from a countable fundamental family).

[F2]

The direct integral is the quotient of square-integrable measurable sections by equality off a Borel null set, with norm squared the integral of the fibre norm squared (Direct integral of a measurable Hilbert field).

[F3]

Under AC, the direct integral of this field over a sigma-finite standard-Borel base is a separable Hilbert space (Direct integrals of measurable Hilbert fields are Hilbert spaces).

[F4]

The commutant is taken inside the bounded operators on the Hilbert space; $S \in D^{'}$ means $S M_{f} = M_{f} S$ for every $f \in L^{\infty}$ (Von Neumann algebras and commutants).

[F5]

A decomposable operator is induced by a weakly measurable essentially bounded operator field. For any such field, its action is bounded and its operator norm equals the essential supremum of the fibre norms (Measurable and decomposable operator fields, Measurable essentially bounded operator fields act decomposably).

[F6]

The complex-linear span of the fundamental family is dense in each fibre (Measurable Hilbert field from a countable fundamental family).

[F7]

Measurable sections have measurable pointwise norms and pairings, and measurable scalar multiples remain measurable (Measurable sections have measurable pointwise inner products).

[F8]

Sigma-finiteness gives a countable measurable cover by finite-measure sets; taking finite unions makes it increasing (Finite, sigma-finite, and semifinite measures).

[F9]

The nonnegative integral is monotone and positively homogeneous, additive on finite sums, and agrees with the simple integral on indicators (Monotonicity and nonnegative homogeneity of the nonnegative integral, Additivity of the nonnegative Lebesgue integral, The nonnegative integral agrees with the simple integral on simple functions).

[F10]

Dominated convergence applies to measurable functions bounded in modulus by one integrable function (Dominated convergence).

[F11]

Countable unions of Borel null sets are Borel null sets (Finite and countable subadditivity of measures).

[F12]

The essential supremum is the infimum of the almost-everywhere bounds (The essential supremum of a measurable function with respect to a measure).

[F13]

The complex rationals are countable and dense ( $Q$ is countably infinite, The rationals embed densely in the reals), and $N^{2}$ is in bijection with $N$ ( $N \times N \approx N$ ). From a fixed bijection $β : N^{2} \to N$ , the recursive code $c_{0} (()) = 0$ , $c_{k + 1} (a_{0}, \dots, a_{k}) = β (a_{0}, c_{k} (a_{1}, \dots, a_{k}))$ , and $c (a_{0}, \dots, a_{k - 1}) = β (k, c_{k} (a_{0}, \dots, a_{k - 1}))$ injects the set of finite sequences into $N$ . Pairing these codes with the rational-complex coefficients shows that the finite rational-complex combinations of a countable family form a countable test family.

[F14]

AC supplies a choice function for every family of nonempty sets, in particular a countable family of nonempty sets (The Axiom of Choice).

[F15]

Complex $L^{\infty}$ consists of measurable functions with finite essential bound, modulo almost-everywhere equality (Complex Lp classes and Euclidean test-function conventions).

[F16]

The nonnegative integral of a nonnegative measurable function is defined as the supremum of the simple integrals of its nonnegative simple minorants (The nonnegative Lebesgue integral).

[F17]

Composition of a measurable map with a Borel measurable outer map is measurable (Composition with a Borel measurable outer map preserves measurability).

[F18]

A continuous map between topological spaces has Borel preimages of Borel sets (A continuous map has Borel preimages of Borel sets).

[F19]

Complex $L^{\infty}$ classes have measurable representatives and identify representatives that agree almost everywhere (Complex Lp classes and Euclidean test-function conventions).

[F20]

Arithmetic operations on measurable extended-real functions, when defined, preserve measurability (Arithmetic and lattice operations preserve measurability whenever they are defined).

[F21]

Superlevel sets of a measurable real-valued function are measurable (Threshold characterisations of real-valued and extended-real-valued measurability).

[F22]

For a nonnegative measurable function, integration over a measurable set is integration after multiplication by its indicator (Integral over a measurable subset).

[F23]

The operator norm is a bound: $∥ S ξ ∥ \leq ∥ S ∥ ∥ ξ ∥$ for every vector in its domain (The operator norm as the least bound and as the unit-sphere or unit-ball supremum).

[F24]

A section belongs to the direct integral's prequotient space exactly when its pointwise squared norm has finite nonnegative integral (Direct integral of a measurable Hilbert field).

Proof

technique · localize, glue, and extend by density

1.1F1F2F3F4F5F7F12F15F19givenalgebra

Fix $f \in L^{\infty}$ and take a Borel representative, available from its measurable-function quotient [F19]. The scalar field $R_{x} = f (x) I_{H_{x}}$ is weakly measurable: $f e_{n}$ is a measurable section by [F1, F7], so each matrix coefficient $⟨ f (x) e_{n} (x), e_{m} (x) ⟩$ is measurable. Also $∥ R_{x} ∥ \leq ∣ f (x) ∣$ , so [F12, F15] make it essentially bounded. The action theorem [F5] therefore defines a bounded multiplication operator $M_{f}$ ; changing the representative on a null set does not change its action on direct-integral classes by [F2, F19]. Now let $T$ be induced by any weakly measurable essentially bounded field $(T_{x})$ . Pointwise linearity gives $T_{x} (f (x) ξ (x)) = f (x) T_{x} ξ (x)$ for every measurable section $ξ$ , and the action theorem makes both operators bounded. Hence $T M_{f} = M_{f} T$ on every class, proving the decomposable inclusion.

1.2F1F2F4F6F7F8F9F13F14F16F17F18F22F24givenconstruct

For the converse, fix $S \in D^{'}$ and write $C = ∥ S ∥$ . Enumerate by the finite-sequence coding and the countability of $Q + i Q$ all finite rational-complex combinations $q_{n}$ of the fundamental sections $e_{j}$ . They are measurable by [F1, F7] and pointwise dense in each fibre by [F6, F13]. Define $b (0) = 0$ and $b (t) = 1 / t$ for $t > 0$ . The reciprocal map on $(0, \infty)$ is continuous, so its Borel preimages are Borel by [F18]; separating the Borel singleton ${0}$ shows that $b$ is Borel. Set $u_{n} (x) = b (∥ q_{n} (x) ∥) q_{n} (x)$ . By [F7, F17], each $u_{n}$ is a measurable section; it has norm one where $q_{n} (x) \neq 0$ and is zero otherwise. Its nonzero values are dense in the unit sphere of every nonzero fibre: if $z$ is a unit vector, choose rational-complex combinations $q_{n}$ arbitrarily close to $z$ using [F6, F13]; eventually they are nonzero, and $∥ q_{n} / ∥ q_{n} ∥ - z ∥ \leq 2 ∥ q_{n} - z ∥$ . Thus $∥ u_{n} (x) ∥ \leq 1$ and their complex span is dense in every fibre. Choose an increasing Borel exhaustion $E_{k} ↑ X$ with $μ (E_{k}) < \infty$ by [F8]. By [F9, F16, F22], $\int_{X} ∥ 1_{E_{k}} u_{n} ∥^{2} d μ \leq μ (E_{k}) < \infty$ , so $[1_{E_{k}} u_{n}] \in H$ by [F24]. For each pair $(k, n)$ choose a measurable square-integrable representative $η_{k, n}$ of $S [1_{E_{k}} u_{n}]$ by the defining prequotient space [F24]. This is a countable choice, supplied by AC.

2.1F1F2F4F7F11step 1.1step 1.2construct

If $ℓ \geq k$ , commutation with $M_{1_{E_{k}}}$ , which exists by step 1.1, gives $[1_{E_{k}} η_{ℓ, n}] = [η_{k, n}]$ because $S \in D^{'}$ by step 1.2 and [F4]. The set where these two sections differ is Borel by [F7] and null by [F2]. Take the countable union over $k, n, ℓ$ ; it is Borel null by [F11]. Off that set the representatives agree on every overlap. Define $v_{n}$ on the disjoint layers $E_{k} ∖ E_{k - 1}$ by $η_{k, n}$ , and set $v_{n} = 0$ on the exceptional null set (with $E_{0} = \emptyset$ ). For every fundamental index $m$ , each coefficient $⟨ v_{n}, e_{m} ⟩$ is Borel on every layer by [F1]; the preimage of a Borel set is the countable union of its Borel preimages intersected with the layers, together with its preimage on the Borel exceptional set. Thus all fundamental coefficients of $v_{n}$ are Borel, so $v_{n}$ is a measurable section by [F1]. On every $E_{k}$ it represents the local action of $S$ on $1_{E_{k}} u_{n}$ .

3.1F2F4F9F13F16F20F22F23F24step 1.1step 1.2step 2.1algebra

Let $q = \sum_{j = 1}^{m} c_{j} u_{n_{j}}$ , where the coefficients are in $Q + i Q$ ; these tests form a countable family by [F13]. The section $q$ is bounded by the construction in step 1.2. Its squared norm and the squared norm of $v_{q}$ are measurable by [F20], so the displayed integrals are defined. For every Borel $A \subseteq E_{k}$ , step 2.1 and commutation with $M_{1_{A}}$ , which exists by step 1.1, show that $S [1_{A} q] = [1_{A} v_{q}]$ : first apply the local representative identity to each $u_{n_{j}}$ on $E_{k}$ , then use commutation and linearity. Thus the operator norm bound [F23] gives $\int_{A} ∥ v_{q} (x) ∥^{2} d μ (x) = ∥ S [1_{A} q] ∥^{2} \leq C^{2} ∥ 1_{A} q ∥^{2} = C^{2} \int_{A} ∥ q (x) ∥^{2} d μ (x) .$ Both integrands are integrable on $E_{k}$ : $q$ is bounded on a finite-measure set by [F9, F22], and $v_{q}$ represents the image under the bounded operator $S$ of $[1_{E_{k}} q]$ , so belongs to the prequotient space [F24].

4.1F9F11F13F16F20F21F22step 3.1algebra

Put $g = ∥ v_{q} ∥^{2}$ and $h = C^{2} ∥ q ∥^{2}$ on $E_{k}$ . Their measurability and that of $g - h$ follow from [F20], so $A_{ϵ} = {x \in E_{k} : g (x) - h (x) \geq ϵ}$ is Borel by [F21] for each positive rational $ϵ$ . The localized inequality, monotonicity, additivity, and the simple integral of an indicator give $\int_{A_{ϵ}} h d μ + ϵ μ (A_{ϵ}) \leq \int_{A_{ϵ}} g d μ \leq \int_{A_{ϵ}} h d μ .$ Here [F22] identifies each integral over $A_{ϵ}$ with the integral of the corresponding indicator product. Thus $μ (A_{ϵ}) = 0$ . Density of the rationals [F13] gives ${g > h} = ⋃_{ϵ \in Q, ϵ > 0} A_{ϵ}$ , so $g \leq h$ almost everywhere on $E_{k}$ . There are only countably many $q, k, ϵ$ , by [F13]; one Borel null set makes all inequalities hold simultaneously. Enlarge the gluing exceptional set by this null set and replace every $v_{n}$ by zero on it; this preserves measurability and each local equivalence class. Consequently, outside one Borel null set $∥ \sum_{j = 1}^{m} c_{j} v_{n_{j}} (x) ∥ \leq C ∥ \sum_{j = 1}^{m} c_{j} u_{n_{j}} (x) ∥$ for every rational-complex finite combination.

5.1F1F5F6F7F12F13F23step 1.2step 4.1construct

Off the common exceptional null set, the rational-complex span of the $u_{n} (x)$ is dense: their complex span is dense by step 1.2 and [F6], and rational-complex coefficients approximate every complex coefficient by [F13]. Define $T_{x} (\sum_{j} c_{j} u_{n_{j}} (x)) = \sum_{j} c_{j} v_{n_{j}} (x)$ for $c_{j} \in Q + i Q$ . The pointwise estimate in step 4.1 makes this assignment well-defined and bounded by $C$ ; it is $Q + i Q$ -linear. Its unique continuous extension to $H_{x}$ , which exists by fibre completeness, is complex-linear because $Q + i Q$ is dense in $C$ . Thus it is a bounded operator $T_{x} \in B (H_{x})$ with $∥ T_{x} ∥ \leq C$ by [F23]. Set $T_{x} = 0$ on the exceptional set and on zero fibres. For each fundamental vector $e_{n}$ , the fixed finite-sequence code has an index $j (n)$ for the one-term combination $e_{n}$ ; hence $T_{x} e_{n} (x) = ∥ e_{n} (x) ∥ v_{j (n)} (x)$ , with both sides zero where $e_{n} (x) = 0$ . By [F7], these are measurable sections and their pairings with every $e_{m}$ are measurable. Thus $(T_{x})$ is weakly measurable by [F5], and $∥ T_{x} ∥ \leq C$ everywhere, so it is essentially bounded by [F12].

6.1F2F3F4F5step 1.1step 2.1step 5.1

The action theorem induces $\hat{S} = \int_{X}^{\oplus} T_{x} d μ (x)$ and gives $∥ \hat{S} ∥ = {ess sup}_{x} ∥ T_{x} ∥ \leq C$ . For every $k, n$ , the representatives agree outside a null set, so $\hat{S} [1_{E_{k}} u_{n}] = S [1_{E_{k}} u_{n}]$ . Both operators commute with $D$ by step 1.1 and the assumption on $S$ , so they agree on all bounded scalar localizations of these sections.

6.2F2F5F6F11givenstep 1.2step 5.1algebra

If two weakly measurable essentially bounded fields induce the same operator, [F5] makes their actions agree on every $[1_{E_{k}} u_{n}]$ . The quotient definition supplies a Borel null set for each such equality; the countable union is null by [F11]. Off it the fields agree on every $u_{n} (x)$ , hence on their dense span by step 1.2 and [F6], and then on $H_{x}$ by boundedness. On zero fibres both fields are the zero operator. This proves uniqueness up to a null set.

7.1F2F5F6F7F10F15F17F19step 1.1step 1.2step 6.1construct

These localizations have dense linear span in $H$ . Indeed, for $[ξ] \in H$ and $ε > 0$ , [F10] lets us first restrict to some finite-measure $E_{k}$ and then to $F = E_{k} \cap {∥ ξ ∥ \leq R}$ with arbitrarily small $L^{2}$ error. On $F \cap {ξ \neq 0}$ , the normalized section is $b (∥ ξ ∥) ξ$ , where the Borel reciprocal function $b$ was constructed in step 1.2; it is measurable by [F7, F17]. For any $δ > 0$ , density of the $u_{n} (x)$ in the unit sphere, proved in step 1.2, gives a countable measurable cover by sets ${x : ∥ b (∥ ξ (x) ∥) ξ (x) - u_{n} (x) ∥ < δ}$ . Assign each point the least qualifying $n$ ; this is a measurable partition. Its first $N$ pieces exhaust all but a set on which the $L^{2}$ norm of $ξ$ tends to zero, by [F10]. For each retained partition set $P_{n}$ , the section $1_{P_{n}} ∥ ξ ∥ u_{n}$ equals $M_{ϕ_{n}} [1_{E_{k}} u_{n}]$ for $ϕ_{n} = 1_{P_{n}} ∥ ξ ∥$ ; this is a member of $L^{\infty}$ by [F15, F19], since $P_{n} \subseteq F \subseteq E_{k}$ has finite measure and $∥ ξ ∥ \leq R$ there. Thus the finite sum over the first $N$ pieces lies in the span on which $S$ and $\hat{S}$ agree by steps 1.1 and 6.1. On retained pieces its pointwise error is below $δ ∥ ξ ∥$ ; on the discarded tail its error is $∥ ξ ∥$ . Taking $δ$ , the tail, and the two initial truncation errors small proves density. Since $S$ and $\hat{S}$ are bounded by [F5] and agree on this dense span, $S = \hat{S}$ . Thus every member of $D^{'}$ is decomposable.

8.1

If $X = \emptyset$ or every fibre is zero, then $H = {0}$ , $D^{'} = {0}$ , and the zero field is the unique field. If every fibre is one-dimensional, each constructed fibre operator is scalar, while the localized density and uniqueness arguments remain valid. Finite-dimensional and varying-dimension fibres require no separate choice: the dense normalized sections and the pointwise inequality include zero and dependent fundamental vectors. Null exceptional sets are combined by a countable union, and the proof uses an arbitrary measure space rather than an interval, so there are no endpoint cases. Steps 1.1–7.1 prove both commutant inclusions, and step 6.2 proves uniqueness. AC is used for the countable representative choice in step 1.2 and to meet the commutant, Hilbert-space, and action supplier hypotheses [F3–F5]; the proof's indexing, partitions and extensions are explicit. [F3, F4, F5, F6, F11, F14, step 1.1, step 1.2, step 5.1, step 7.1, step 6.2, algebra] $□$

Source qualifications

Bruhat, Part III Chapter 10 §1.8, Theorem 2, printed pp. 100–101, gives the diagonal-commutant characterization and starts the converse by localizing fundamental vectors on compact sets and gluing their images. That argument uses a continuous/Lusin-field convention and does not provide the common countable rational-span estimate in the standard-Borel convention used here; the proof above supplies the finite-measure exhaustion, measurable gluing, pointwise bound, and density steps directly. Bekka–de la Harpe, Chapter 1 §1.H, Theorem 1.H.1, printed p. 65, states the general-field result and refers its proof to Dixmier; their Theorem 1.H.4, printed pp. 67–68, proves the converse for constant separable fibres after reducing a sigma-finite measure to an equivalent probability measure. Neither reduction is imported here.

Boundary cases

Empty: $X = \emptyset$ gives $H = 0$ and both sides consist of the zero operator, as checked in step 8.1.
Zero: All zero fibres give the zero direct integral and unique zero field; mixed zero fibres are assigned zero in step 5.1 and are covered by uniqueness in step 6.2.
One: Every bounded operator on a one-dimensional fibre is scalar, so the constructed fibre maps have the asserted pointwise form; step 8.1 checks that no density or uniqueness argument changes.
Degenerate: Zero, dependent, and varying finite-dimensional fundamental vectors are included by the normalized family and dense-span extension in steps 1.2 and 5.1.
Endpoints: Not applicable; the base is an arbitrary sigma-finite standard-Borel space with no interval parameter.
Nonempty choice: AC selects the countably many local representatives in step 1.2 and meets the assumptions of [F3–F5]; all subsequent partitions and limits are explicit.
Iff forward: Every decomposable operator commutes with all diagonal multipliers in step 1.1.
Iff reverse: Every operator in the commutant is induced by the constructed field in steps 1.2–7.1.

LemmaStatement: AI-adaptedProof: AI-adaptedprecheck passaudited 2026-09-30Open item page →

Diagonal multipliers form a von Neumann algebra

Statement

Assume AC. Let $(X, B, μ)$ be a sigma-finite standard-Borel measure space, let $(H_{x}, e_{n} (x))_{x \in X}$ be a measurable complex Hilbert field with a countable fundamental family from Measurable Hilbert field from a countable fundamental family, and put $H = \int_{X}^{\oplus} H_{x} d μ (x),$ the direct integral of Direct integral of a measurable Hilbert field. For $f \in L^{\infty} (X, μ)$ , let $M_{f}$ be scalar multiplication on $H$ , and set $D = {M_{f} : f \in L^{\infty} (X, μ)} \subseteq B (H) .$ Then $D$ is a unital abelian $*$ -subalgebra, $D^{''} = D,$ and hence $D$ is a concrete von Neumann algebra. This remains true when zero fibres make $f \mapsto M_{f}$ noninjective. On the zero Hilbert space, $D = {0}$ and its sole element is the identity operator. Inner products are linear in their first variable.

Facts & Assumptions

Given: AC; the preceding direct integral and its scalar-multiplication operators; a measurable Hilbert field with its countable fundamental family; and the action, commutant, and WOT conventions in the cited items.

[F1]

Each fundamental vector $e_{n}$ is a measurable section, their complex span is dense in every fibre, and zero-dimensional fibres are allowed (Measurable Hilbert field from a countable fundamental family).

[F2]

Measurable sections have measurable pointwise norms and pairings, and multiplication by a measurable scalar function preserves measurability (Measurable sections have measurable pointwise inner products).

[F3]

Measurability is defined by inverse-image measurability (A measurable function between measurable spaces).

[F4]

A bounded operator field is weakly measurable when its fundamental matrix coefficients are measurable, and is decomposable when weakly measurable and essentially bounded (Measurable and decomposable operator fields).

[F5]

Weakly measurable essentially bounded fields act on the direct integral; the induced norm is the essential supremum of the fibre norms, and products and adjoints of fields induce the corresponding operator products and adjoints (Measurable essentially bounded operator fields act decomposably).

[F6]

The commutant of all scalar multipliers consists exactly of the decomposable operators, and two fields inducing the same operator agree almost everywhere (Decomposable operators are the commutant of diagonal multiplication).

[F7]

For any operator set, a commutant is WOT closed; if a set is self-adjoint, its commutant is a unital $*$ -subalgebra. Commutants are taken inside $B (H)$ (Von Neumann algebras and commutants).

[F8]

A countable union of Borel null sets is a Borel null set (Finite and countable subadditivity of measures).

[F9]

The operator norm bounds the image norm (The operator norm as the least bound and as the unit-sphere or unit-ball supremum).

[F10]

An essentially bounded measurable scalar or norm function has finite essential supremum (The essential supremum of a measurable function with respect to a measure).

[F11]

The inner product is linear in its first variable and conjugate-linear in its second (Real and complex inner-product spaces and their induced length).

[F12]

AC means every family of nonempty sets has a choice function (The Axiom of Choice); here it is assumed only to meet the hypotheses of the preceding Hilbert-field, action, commutant, and adjoint results.

[F13]

The direct integral is the quotient of square-integrable measurable sections modulo Borel null sets, and scalar and fibrewise operations act on classes pointwise (Direct integral of a measurable Hilbert field).

[F14]

$L^{\infty} (X, μ; C)$ consists of a.e.-equivalence classes of measurable complex functions with finite essential-supremum modulus (Complex Lp classes and Euclidean test-function conventions).

[F15]

Arithmetic operations on measurable real functions preserve measurability; complex addition, multiplication, and conjugation are measurable by applying these rules to real and imaginary parts (Arithmetic and lattice operations preserve measurability whenever they are defined, Complex Lp classes and Euclidean test-function conventions).

[F16]

Complex modulus is multiplicative and subadditive, and conjugation preserves modulus (Conjugation is an involutive real-field automorphism, $z \overline{z} = ∣ z ∣^{2}$ , and modulus is definite, multiplicative, and subadditive).

[F17]

A finite essential supremum of a real measurable function is an almost-everywhere bound (The essential supremum is attained as the least essential bound).

[F18]

Cauchy–Schwarz bounds the modulus of an inner product by the product of the vector norms (Cauchy–Schwarz: $∣ ⟨ x, y ⟩ ∣ \leq ∥ x ∥ ∥ y ∥$ , with equality exactly for dependent pairs).

[F19]

A commutant is WOT closed because each commutation equation is a WOT-closed condition (Von Neumann algebras and commutants).

[F20]

A Borel function composed with a measurable map is measurable (Composition with a Borel measurable outer map preserves measurability).

[F21]

A continuous map has Borel inverse images of Borel sets (A continuous map has Borel preimages of Borel sets).

Proof

technique · direct, using countably many measurable rank-one fields

Given: AC and the measurable direct-integral field above.

1.1F1F2F3F20F21construct

For each $n$ , define $u_{n} (x) = {\begin{cases} e_{n} (x) / ∥ e_{n} (x) ∥, & ∥ e_{n} (x) ∥ > 0, \\ 0, & ∥ e_{n} (x) ∥ = 0. \end{cases}$ The norm is measurable by [F2]. The scalar function $r (0) = 0$ and $r (t) = 1 / t$ for $t > 0$ is Borel: its inverse image of any Borel set is the possible singleton contribution at $0$ together with the inverse image under the continuous reciprocal map on $(0, \infty)$ , using [F21]. Thus $u_{n} = (r \circ ∥ e_{n} ∥) e_{n}$ is a measurable section by [F1, F2, F3, F20]. Each $u_{n}$ is either zero or a unit vector. Its pointwise linear span is dense in $H_{x}$ , since each $e_{n} (x)$ is a scalar multiple of $u_{n} (x)$ and [F1] gives dense span. In particular, if every $u_{n} (x)$ is zero then $H_{x} = {0}$ .

1.2F2F4F5F6F9F11F12F18construct

For $n, m \in N$ , define the rank-one field $R_{n, m} (x) ξ = ⟨ ξ, u_{n} (x) ⟩ u_{m} (x), ξ \in H_{x} .$ By the first-variable linearity in [F11], this is a bounded linear operator of norm at most one, including on zero fibres, by [F9, F18]. For each pair of fundamental sections $e_{i}, e_{j}$ , the section $R_{n, m} e_{i} = ⟨ e_{i}, u_{n} ⟩ u_{m}$ is measurable by [F2]; pairing it with $e_{j}$ is measurable by [F2] again. Hence every fundamental matrix coefficient of $R_{n, m}$ is measurable, so [F4] makes it a weakly measurable, essentially bounded field. By [F5] it induces a decomposable operator $Q_{n, m} = \int_{X}^{\oplus} R_{n, m} (x) d μ (x)$ , which belongs to $D^{'}$ by [F6].

1.3F5F7F8F11F12F13F14F15F16F17givenalgebra

Let $f, g \in L^{\infty} (X, μ; C)$ and choose measurable representatives. Their moduli have finite essential suprema by [F14], so [F17] gives finite bounds outside null sets; [F8] combines the two exceptional sets. The inequalities in [F16] show that $f + g$ , $f g$ , $a f$ , and $\overset{ˉ}{f}$ are essentially bounded for every $a \in C$ . Their measurability follows from [F14,F15], so all belong to $L^{\infty} (X, μ; C)$ . Pointwise action on direct-integral classes [F13] and product/adjoint compatibility [F5,F11] give $M_{f} M_{g} = M_{f g}$ , $M_{f} + M_{g} = M_{f + g}$ , $a M_{f} = M_{a f}$ , and $M_{f}^{*} = M_{\overset{ˉ}{f}}$ . Also $M_{1} = I_{H}$ , including when $H = {0}$ . Thus $D$ is a unital abelian $*$ -subalgebra. Consequently $D \subseteq D^{'}$ and $D \subseteq D^{''}$ by [F7].

2.1F4F5F6F7F8F12step 1.2step 1.3

Let $T \in D^{''}$ . Since $D \subseteq D^{'}$ , $T$ commutes with every element of $D$ , so $T \in D^{'}$ . By [F6], there is a weakly measurable essentially bounded field $(T_{x})$ inducing $T$ . For each $n, m$ , $T$ commutes with $Q_{n, m}$ because $Q_{n, m} \in D^{'}$ . By the product clause in [F5], the fields $(T_{x} R_{n, m} (x))$ and $(R_{n, m} (x) T_{x})$ are weakly measurable and essentially bounded and induce $T Q_{n, m}$ and $Q_{n, m} T$ , respectively. These induced operators are equal; the uniqueness clause in [F6] therefore makes the two fields equal outside a Borel null set. There are only countably many pairs $(n, m)$ , so [F8] gives one Borel null set $N$ outside which all fibrewise commutation identities hold simultaneously.

3.1F1F9F11step 1.1step 1.2step 2.1

Fix $x \notin N$ . If $H_{x} = {0}$ , then $T_{x} = 0$ . Otherwise some $u_{n} (x)$ is nonzero by step 1.1. For every nonzero $u_{n} (x)$ , $R_{n, n} (x)$ is the orthogonal projection onto its one-dimensional span. Commutation with this projection shows that $T_{x} u_{n} (x) = λ_{n} u_{n} (x)$ , where $λ_{n} = ⟨ T_{x} u_{n} (x), u_{n} (x) ⟩$ . If $u_{n} (x)$ and $u_{m} (x)$ are both nonzero, apply $T_{x} R_{n, m} (x) = R_{n, m} (x) T_{x}$ to $u_{n} (x)$ : the left side is $λ_{m} u_{m} (x)$ and the right side is $λ_{n} u_{m} (x)$ , so $λ_{m} = λ_{n}$ . Thus a single scalar $λ_{x}$ satisfies $T_{x} u_{n} (x) = λ_{x} u_{n} (x)$ for every $n$ (also when $u_{n} (x) = 0$ ). Their span is dense by step 1.1, and boundedness of $T_{x}$ extends this identity to all of $H_{x}$ . Hence $T_{x} = λ_{x} I_{H_{x}}$ .

4.1F1F2F3F4F5F9F10F12F13F14F17F18F19step 1.3step 2.1step 3.1algebra∎

Let $Z = {x : H_{x} = {0}}$ , which is Borel because $Z = ⋂_{n} {x : ∥ e_{n} (x) ∥ = 0}$ by [F1,F2,F3]. On $X ∖ (N \cup Z)$ , partition into the Borel sets $B_{n}$ on which $n$ is the least index with $u_{n} (x) \neq 0$ . Define $λ (x) = {\begin{cases} ⟨ T_{x} u_{n} (x), u_{n} (x) ⟩, & x \in B_{n}, \\ 0, & x \in N \cup Z . \end{cases}$ For each $n$ , the displayed pairing is measurable by the all-section coefficient criterion in [F4] and the measurable-section pairing result [F2]. The countable Borel partition therefore makes $λ$ Borel. Cauchy–Schwarz and the operator-norm bound [F9, F18] give $∣ λ (x) ∣ \leq ∥ T_{x} ∥$ on the $B_{n}$ ; since the field $(T_{x})$ is essentially bounded by [F4,F10], [F17] gives a finite a.e. bound for its norm, so $λ$ represents an element of $L^{\infty} (X, μ; C)$ by [F14]. Step 3.1 gives $T_{x} = λ (x) I_{H_{x}}$ outside $N$ , including zero fibres. The action definition then implies $T = M_{λ}$ on direct-integral classes by [F5,F13]. This proves $D^{''} \subseteq D$ , and step 1.3 gives the reverse inclusion. Finally, [F19] says $D^{''}$ , being a commutant, is WOT closed. Thus $D = D^{''}$ is a concrete von Neumann algebra.

Boundary cases

Empty: If $X = \emptyset$ , then $H = {0}$ , $D = {0}$ , and $D^{''} = D$ ; its sole operator is the identity on the zero space.
Zero: On all-zero fibres the same zero-space calculation applies. On a mixed field, the proof defines $λ = 0$ on zero fibres; multiplier values there may lie in the kernel of $f \mapsto M_{f}$ , which does not affect the operator equality.
One: On a one-dimensional nonzero fibre, every bounded fibre operator is scalar and the rank-one commutation argument gives exactly that scalar.
Degenerate: Zero or dependent fundamental vectors normalize to zero or repeat directions; their total span remains dense, and the countable null-set union handles all pairs simultaneously. Fibre dimensions may vary.
Endpoints: Not applicable; the base has no interval parameter.
Nonempty choice: AC from [F12] is assumed to invoke the cited direct-integral, commutant, and adjoint conventions. The countable family, null-set union, and least-index partition use no additional choice.
Iff forward: The inclusion $D \subseteq D^{''}$ follows from the commutant definition in step 1.3.
Iff reverse: Steps 2.1–4.1 show each $T \in D^{''}$ is a scalar multiplier, including on zero fibres.

Source qualifications

Bruhat, Part III Chapter 10 §1.8, Theorem 3, printed pp. 101–102, argues that the scalar diagonal algebra is weakly closed by countably many fibrewise rank-one tests in his continuous-sum setting. His field convention is based on locally compact spaces and Lusin-type measurability; it is not silently identified with the standard-Borel measurable-field convention here. Bekka and de la Harpe, Chapter 1 §1.H, Proposition 1.H.2, printed pp. 65–66, prove WOT closure of diagonal multipliers for a constant Hilbert fibre by a weak-star compactness argument. Neither passage proves this varying-fibre, possibly nonfaithful statement in the present convention; the rank-one field, common-null-set, scalar-recovery, and zero-fibre arguments above are local.

TheoremStatement: AI-adaptedProof: AI-adaptedprecheck passaudited 2026-09-30Open item page →

Spectral multiplicity model for separably acting abelian von Neumann algebras

Statement

Assume AC. Let $A \subseteq B (H)$ be an abelian concrete von Neumann algebra on a nonzero separable complex Hilbert space $H$ . There exist a bounded self-adjoint operator $S \in A$ with $A = W^{*} (S)$ , a nonempty compact set $K = σ (S) \subseteq R$ , a nonzero finite regular Borel measure $μ$ on $K$ , and a Borel function $m : K ⟶ {1, 2, 3, \dots} \cup {\infty}$ (whose values are immaterial on a $μ$ -null set) such that, for the measurable field $H_{t} = C^{m (t)}$ when $m (t) < \infty$ and $H_{t} = ℓ^{2} (N)$ when $m (t) = \infty$ , there is a unitary $U : H ⟶ \int_{K}^{\oplus} H_{t} d μ (t)$ with $U S U^{- 1} = M_{t}, U A U^{- 1} = {M_{f} : f \in L^{\infty} (K, μ)} .$ For this fixed generator $S$ , every such spectral model has the same measure class on $K$ and the same multiplicity function $μ$ -almost everywhere. Changing $S$ can change the spectral coordinate and is not part of the uniqueness assertion. Inner products are linear in their first variable.

Facts & Assumptions

Given: AC; the abelian concrete von Neumann algebra $A$ on the nonzero separable complex Hilbert space $H$ ; and the measurable-field, direct-integral, spectral-calculus, and measure-theoretic conventions named below.

[F1]

Every such $A$ has a bounded self-adjoint generator $S \in A$ with $A = W^{*} (S)$ (A separably acting abelian von Neumann algebra has a self-adjoint generator).

[F2]

The spectrum of a bounded operator is nonempty compact and norm bounded; the spectrum of a self-adjoint operator is real. Thus $K = σ (S)$ is a nonempty compact subset of $R$ (Spectrum is nonempty compact and norm bounded, Spectrum of a self adjoint operator is real).

[F3]

A bounded normal operator on a nonzero separable Hilbert space has a finite or countable orthogonal decomposition into nonzero cyclic reducing subspaces; their closed Hilbert sum is $H$ (Maximal orthogonal family of cyclic reducing subspaces).

[F4]

For each cyclic vector $x_{j}$ in this decomposition, its scalar spectral measure $μ_{j} = E_{x_{j}}$ is a nonzero finite regular Borel measure on $K$ , and the cyclic unitary $V_{j} : L^{2} (K, μ_{j}) \to H_{j}$ intertwines multiplication by every bounded Borel function with the corresponding Borel function of $S$ (Cyclic spectral representation).

[F5]

Positive countable weighted sums of measures are measures; an absolutely continuous sigma-finite signed measure has a measurable density unique almost everywhere, with the set-integral formula (Nonnegative scalar multiples and countable weighted sums of measures are measures, A sigma-finite signed measure that is absolutely continuous with respect to a sigma-finite positive measure has a unique almost-everywhere density).

[F6]

If $ν (B) = \int_{B} h d μ$ for a nonnegative measurable density $h$ , then for every nonnegative measurable $g$ , $\int g d ν = \int g h d μ .$ Here this identity is proved locally from the set formula: it holds for nonnegative simple $g$ by the simple-integral definition; increasing simple approximation and monotone convergence give it for general $g$ (The integral of a nonnegative simple function, The nonnegative Lebesgue integral, Every nonnegative measurable function is the increasing limit of simple measurable functions, Monotone convergence for the integral).

[F7]

A countable fundamental family defines a measurable Hilbert field; the direct integral is the quotient of square-integrable measurable sections, and under AC it is a separable Hilbert space (Measurable Hilbert field from a countable fundamental family, Direct integral of a measurable Hilbert field, Direct integrals of measurable Hilbert fields are Hilbert spaces).

[F8]

On a compact metric space every open set is sigma-compact. Under countable choice, every locally finite Borel measure is regular in the stronger Radon convention when its open sets are sigma-compact; a finite regular Borel measure on an LCH space has the real-valued $C_{c}$ dense in real $L^{p}$ for finite $p$ under DC. The library's $C_{c} (K)$ convention is real-valued until a complex convention is stated explicitly (Locally compact topological space: every point has a compact neighbourhood; and what this says in a metric space, Radon measure on an LCH space, Compact support, $C_{c} (X)$ , and $C_{0} (X)$ , Sigma-compact open sets make locally finite Borel measures regular, C_c(X) is dense in L^p(mu) for a Radon measure).

[F9]

Under AC, $D = {M_{f} : f \in L^{\infty}}$ on a measurable Hilbert field's direct integral is a concrete von Neumann algebra, and its commutant is exactly the decomposable operators; bounded measurable operator fields act with norm the essential supremum and products and adjoints act fibrewise (Diagonal multipliers form a von Neumann algebra, Decomposable operators are the commutant of diagonal multiplication, Measurable essentially bounded operator fields act decomposably, Measurable and decomposable operator fields).

[F10]

The bounded Borel functional calculus sends $1_{B} (S)$ to the spectral projection $E (B)$ ; for $M_{t}$ the projection is multiplication by $1_{B}$ . A PVM defines the finite positive scalar measure $B \mapsto ⟨ E (B) x, x ⟩ = ∥ E (B) x ∥^{2}$ (Borel functional calculus for bounded normal operators, Scalar and complex measures from a pvm).

[F11]

For bounded self-adjoint $M_{t}$ , the bounded Borel calculus acts by $φ (M_{t}) = M_{φ}$ ; for continuous $φ$ this lies in the norm-closed unital $*$ -algebra generated by $M_{t}$ , which is contained in $W^{*} (M_{t})$ . WOT-closed sets are norm closed (Borel functional calculus for bounded normal operators, Continuous functional calculus for bounded self adjoint operators, C star algebra generated by a normal operator, Strong and weak operator topologies, Von Neumann algebras and commutants).

[F12]

Complex $L^{\infty}$ consists of measurable equivalence classes with finite essential bound. Since the base $K$ carries its Borel sigma-algebra, its measurable representatives are Borel. A finite essential supremum is an almost-everywhere bound (Complex Lp classes and Euclidean test-function conventions, The essential supremum of a measurable function with respect to a measure, The essential supremum is attained as the least essential bound, Standard Borel spaces).

[F13]

A finite measure on $K$ is sigma-finite; nonnegative density integrals define measures; countable weighted sums and countable unions of null sets are valid (Finite, sigma-finite, and semifinite measures, The indefinite integral of a nonnegative measurable function is a measure, Additivity of the nonnegative Lebesgue integral, Finite and countable subadditivity of measures).

[F14]

The usual real line is Polish: it is complete and the countable dense rationals witness separability. Its Borel space is standard Borel, and the Borel subset $K$ is standard Borel under AC. Separability supplies a countable dense sequence in each Hilbert space at issue ( $R$ and $R^{n}$ for $n \geq 1$ with the Euclidean metric are complete, componentwise from the Cauchy criterion in $R$ , The rationals embed densely in the reals, $Q$ is countably infinite, Polish spaces are separable completely metrizable spaces, Standard Borel spaces, Borel subspaces admit polish presentations, Separability: the existence of an at most countable dense subset).

[F15]

Measurable functions remain measurable under Borel composition and the listed arithmetic operations, including the real-imaginary formulas for complex operations; continuous maps have Borel preimages. Complex conjugation and modulus obey their Euclidean algebraic laws (A measurable function between measurable spaces, Composition with a Borel measurable outer map preserves measurability, Arithmetic and lattice operations preserve measurability whenever they are defined, A continuous map has Borel preimages of Borel sets, Complex Lp classes and Euclidean test-function conventions, Real and imaginary parts, complex conjugation, and modulus, Conjugation is an involutive real-field automorphism, $z \overline{z} = ∣ z ∣^{2}$ , and modulus is definite, multiplicative, and subadditive).

[F16]

AC implies DC and countable choice, which are the hypotheses of the regularity and $C_{c}$ -density suppliers in [F8] (The Axiom of Choice, AC implies DC implies countable choice).

[F17]

The complex inner product is linear in its first variable. For an orthogonal projection $P$ , $⟨ P x, x ⟩ = ⟨ P x, P x ⟩ = ∥ P x ∥^{2}$ because $P = P^{*} = P^{2}$ ; spectral projections have this property (Real and complex inner-product spaces and their induced length, Hilbert space, Scalar and complex measures from a pvm).

Proof

technique · direct construction from cyclic spectral measures, followed by a local decomposable-intertwiner argument for uniqueness

Given: AC and $A \subseteq B (H)$ as in the statement.

1.1F1F2

By [F1], choose the bounded self-adjoint generator $S$ with $A = W^{*} (S)$ . By [F2], $K = σ (S)$ is nonempty compact and contained in $R$ .

1.2F3F4

Regard $S$ as a bounded normal operator. Apply [F3] and enumerate the nonzero cyclic reducing summands as $H_{j}$ , where the index set is either ${1, \dots, N}$ or $N$ . Choose a cyclic vector $x_{j} \neq 0$ in each summand, and apply [F4] to obtain finite nonzero regular Borel measures $μ_{j}$ on $K$ and cyclic unitaries $V_{j} : L^{2} (K, μ_{j}) \to H_{j}$ . Their direct sum is a unitary from $⨁_{j} L^{2} (K, μ_{j})$ onto $H$ , and it intertwines every bounded Borel function of $S$ componentwise. AC supplies the cyclic decomposition and the cyclic vectors; no zero summand is used.

1.3F5F8F16algebra

Put $a_{j} = \frac{2^{- j}}{1 + μ_{j} (K)}, μ = \sum_{j} a_{j} μ_{j} .$ By [F5] this is a Borel measure, and $0 < μ (K) = \sum_{j} \frac{2^{- j} μ_{j} (K)}{1 + μ_{j} (K)} \leq \sum_{j} 2^{- j} < \infty .$ It dominates each $μ_{j}$ because $μ (B) \geq a_{j} μ_{j} (B)$ for every Borel $B$ . It is regular by [F8]: it is finite and hence locally finite on compact $K$ , and every open subset of the compact metric space $K$ is sigma-compact. For an open $O \neq K$ , compact sets $C_{n} = {t \in K : d (t, K ∖ O) \geq 1 / n}$ increase to $O$ ; the cases $O = \emptyset$ and $O = K$ are immediate.

1.4F5F13F15

By [F5], take Borel Radon--Nikodym densities $h_{j} = d μ_{j} / d μ$ . Each $h_{j}$ is nonnegative almost everywhere: if $E_{n} = {h_{j} \leq - 1 / n}$ , then $0 \leq μ_{j} (E_{n}) = \int_{E_{n}} h_{j} d μ \leq - μ (E_{n}) / n,$ so $μ (E_{n}) = 0$ ; the negative set is their countable union. Replace $h_{j}$ by zero there. Since $μ_{j} (K) < \infty$ , the set formula also implies that $h_{j}$ is finite almost everywhere. Replace it by zero on any Borel null set where it is nonfinite, so each chosen $h_{j}$ is Borel, finite and nonnegative. Let $F_{j} = {t : h_{j} (t) > 0}, K_{0} = ⋃_{j} F_{j} .$ The set-integral formula in [F5] gives $μ_{j} (K ∖ F_{j}) = 0$ and then $μ_{j} (K ∖ K_{0}) = 0$ for every $j$ . The weighted-sum formula yields $μ (K ∖ K_{0}) = 0$ .

1.5F7F14F15

Define $m_{0} (t) = \sum_{j} 1_{F_{j}} (t)$ on $K_{0}$ , and set $m (t) = m_{0} (t)$ there and $m (t) = 1$ on $K ∖ K_{0}$ . Countable sums of Borel indicators make $m$ Borel; it takes values in ${1, 2, \dots} \cup {\infty}$ and is the number of active cyclic coordinates for $μ$ -almost every $t$ . For $A_{r} = {t : m (t) \geq r}$ , the $r$ -th active index $k_{r} (t) = \min {j : t \in F_{j}, 1 + # {i < j : t \in F_{i}} = r}$ is Borel on $A_{r} \cap K_{0}$ : its level set at $j$ is $F_{j} \cap {\sum_{i < j} 1_{F_{i}} = r - 1}$ . The rank $r_{j} (t) = 1 + \sum_{i < j} 1_{F_{i}} (t)$ is Borel on $F_{j}$ . These least-index definitions introduce no choice.

1.6F6F7F13F14F15

Give $H_{t}$ the span of the first $m (t)$ standard vectors in $ℓ^{2}$ ; when $m (t) = \infty$ this is all of $ℓ^{2}$ . The sections $e_{r} (t) = 1_{A_{r}} (t) e_{r}$ form a countable fundamental family because their Gram coefficients are $δ_{r s} 1_{A_{r} \cap A_{s}}$ and their span is dense in every fibre. Thus this is a measurable Hilbert field over the standard-Borel finite-measure space $(K, μ)$ . Its direct integral identifies with $⨁_{r \geq 1} L^{2} (A_{r}, μ)$ : a section has measurable coordinates $f_{r}$ supported on $A_{r}$ , its squared fibre norm is $\sum_{r} ∣ f_{r} ∣^{2}$ , and monotone convergence of the finite coordinate sums gives $\int_{K} \sum_{r} ∣ f_{r} ∣^{2} d μ = \sum_{r} \int_{A_{r}} ∣ f_{r} ∣^{2} d μ .$ This also proves the identification in both directions, including the countably infinite fibre.

1.7F6F7F15

On $K_{0}$ , define a map from the cyclic direct sum by $f_{r} (t) = \sqrt{h_{k_{r} (t)} (t)} g_{k_{r} (t)} (t) (t \in A_{r} \cap K_{0}),$ and put $f_{r} = 0$ on the remaining null set. The rank partitions and [F15] make every coordinate Borel. The identity [F6] gives $\sum_{r} \int_{A_{r}} ∣ f_{r} ∣^{2} d μ = \sum_{j} \int_{F_{j}} h_{j} ∣ g_{j} ∣^{2} d μ = \sum_{j} \int_{K} ∣ g_{j} ∣^{2} d μ_{j} .$ For each finite partial sum these equalities follow by applying [F6] on each rank piece; monotone convergence passes to the countable sums, and the active rank enumeration is a bijection from the active $j$ coordinates to the $r = 1, \dots, m (t)$ coordinates at every $t \in K_{0}$ . Hence the map is well-defined and isometric. It is onto: for a field $(f_{r})$ , set $g_{j} (t) = {\begin{cases} f_{r_{j} (t)} (t) / \sqrt{h_{j} (t)}, & t \in F_{j}, \\ 0, & t \notin F_{j} . \end{cases}$ The Borel rank partition makes $g_{j}$ measurable, and the same integral identity shows that $(g_{j})$ lies in the cyclic Hilbert sum and maps back to $(f_{r})$ almost everywhere. Thus the constructed map $W$ is unitary and commutes with all bounded Borel scalar multipliers.

1.8F9F11algebra

Let $D = {M_{f} : f \in L^{\infty} (K, μ)}$ on this direct integral. By [F9], $D$ is a WOT-closed unital $*$ -algebra and contains $M_{t}$ because $K$ is compact. Therefore $W^{*} (M_{t}) \subseteq D .$

1.9F5F7F11F13algebra

For the reverse inclusion, fix $f \in L^{\infty} (K, μ)$ and let $M = ∥ f ∥_{\infty}$ . By [F12] choose a Borel representative and change it on a Borel $μ$ -null set so that $∣ f (t) ∣ \leq M$ everywhere. If $M = 0$ , $M_{f} = 0$ . Assume $M > 0$ . Choose a countable dense sequence $η_{j}$ in the nonzero direct-integral Hilbert space and measurable representatives; [F7] makes $t \mapsto ∥ η_{j} (t) ∥^{2}$ Borel. Define $ν_{j} (B) = \int_{B} ∥ η_{j} (t) ∥^{2} d μ (t), b_{j} = \frac{2^{- j}}{1 + ∥ η_{j} ∥^{2}}, ν = \sum_{j} b_{j} ν_{j} .$ Each $ν_{j} ≪ μ$ , so changing $f$ on a Borel $μ$ -null set does not alter any $L^{2} (ν_{j})$ class or the $L^{2} (ν)$ class. By [F13] each $ν_{j}$ and $ν$ is a Borel measure, and $ν (K) \leq 2$ since $ν_{j} (K) = ∥ η_{j} ∥^{2}$ and $\sum_{j \in N} 2^{- j} = 2$ . It is nonzero because a dense sequence in a nonzero Hilbert space cannot consist entirely of zero vectors. By [F8] $ν$ is regular, and [F16] supplies DC for [F8]'s hypotheses and the $p = 2$ case of the $C_{c} (K)$ -density theorem. Write $f = u + i v$ with real-valued $u, v$ . Both belong to real $L^{2} (ν)$ because $∣ u ∣, ∣ v ∣ \leq ∣ f ∣$ . Since the library's $C_{c} (K)$ is real-valued and $K$ is compact, $C_{c} (K) = C (K; R)$ ; apply the density theorem separately to choose $u_{n}, v_{n} \in C (K; R)$ converging to $u, v$ in real $L^{2} (ν)$ . Then $ψ_{n} = u_{n} + i v_{n} \in C (K; C)$ and $∥ ψ_{n} - f ∥_{L^{2} (ν)} \to 0$ . Radially clip each $ψ_{n}$ to the disc of radius $M$ : since $∣ f ∣ \leq M$ , the clipped functions $ψ_{n}^{(c)}$ satisfy $∣ ψ_{n}^{(c)} (t) - f (t) ∣ \leq ∣ ψ_{n} (t) - f (t) ∣$ and $∥ ψ_{n}^{(c)} ∥_{\infty} \leq M$ , and they still converge to $f$ in $L^{2} (ν)$ . Fix $ξ$ in the direct integral and a member $η_{j}$ of the dense sequence. Cauchy--Schwarz in $L^{2} (μ)$ and the pointwise bound $∣ ⟨ ξ (t), η_{j} (t) ⟩ ∣ \leq ∥ ξ (t) ∥ ∥ η_{j} (t) ∥$ give $∣ ⟨ (M_{ψ_{n}^{(c)}} - M_{f}) ξ, η_{j} ⟩ ∣ \leq ∥ ξ ∥_{H} (\int_{K} ∣ ψ_{n}^{(c)} - f ∣^{2} ∥ η_{j} ∥^{2} d μ)^{1 / 2}$ , and the integral equals $∥ ψ_{n}^{(c)} - f ∥_{L^{2} (ν_{j})}^{2} \leq b_{j}^{- 1} ∥ ψ_{n}^{(c)} - f ∥_{L^{2} (ν)}^{2}$ because $ν = \sum_{k} b_{k} ν_{k}$ with $b_{j} > 0$ . Hence $⟨ (M_{ψ_{n}^{(c)}} - M_{f}) ξ, η_{j} ⟩ \to 0$ for every $j$ . The clipped approximants are uniformly bounded by $M$ , so $∣ ⟨ (M_{ψ_{n}^{(c)}} - M_{f}) ξ, η - η_{j} ⟩ ∣ \leq 2 M ∥ ξ ∥_{H} ∥ η - η_{j} ∥_{H}$ extends this convergence from the dense family to every $η$ ; thus $M_{ψ_{n}^{(c)}} \to M_{f}$ in the weak operator topology. Each $M_{ψ_{n}^{(c)}} = ψ_{n}^{(c)} (M_{t})$ lies in $C^{*} (I, M_{t}) \subseteq W^{*} (M_{t})$ by the continuous functional calculus, and $W^{*} (M_{t})$ is WOT closed, so $M_{f} \in W^{*} (M_{t})$ .

1.10F5F7F10F14F17

Fix $S$ and compare any two models $(μ, m, U_{1})$ and $(ν, n, U_{2})$ on $K$ . Choose a countable dense sequence $(z_{j})$ in $H$ and define $ρ (B) = \sum_{j} \frac{2^{- j}}{1 + ∥ z_{j} ∥^{2}} ⟨ E (B) z_{j}, z_{j} ⟩ .$ By [F5, F10, F17] this is a finite nonzero Borel measure. For a Borel set $B$ , $ρ (B) = 0$ exactly when $E (B) z_{j} = 0$ for every $j$ , which by density and boundedness of the projection $E (B)$ is equivalent to $E (B) = 0$ . In either model, [F10] identifies $U_{i} E (B) U_{i}^{- 1}$ with $M_{1_{B}}$ , and this operator is zero exactly when its model measure of $B$ is zero: if the measure is positive, the section $1_{B} e_{1}$ is a nonzero square-integrable section because the fibre dimension is at least one almost everywhere. Therefore $μ (B) = 0 ⟺ E (B) = 0 ⟺ ν (B) = 0,$ so the two measures have the same measure class.

2.1F4F7step 1.2step 1.7

Compose $W$ with the inverse of the direct sum of the cyclic unitaries $V_{j}$ . This gives a unitary $U : H ⟶ \int_{K}^{\oplus} H_{t} d μ (t)$ that intertwines every bounded Borel function of $S$ with the same scalar multiplier; in particular $U S U^{- 1} = M_{t}$ .

2.2F5F6F15step 1.10

Let $g = d ν / d μ$ and $q = d μ / d ν$ . Since both measures are finite, their Radon--Nikodym set formulas make these densities finite almost everywhere. As in Step 1.4, the threshold-set argument gives nonnegative Borel versions; replace them by zero on the Borel null sets where they are negative or nonfinite. If $g = 0$ on a set of positive $μ$ -measure, the set formula gives zero $ν$ -measure there, contradicting $μ ≪ ν$ ; similarly $q > 0$ $ν$ -almost everywhere. Applying [F6] to the set formula for $g$ and the nonnegative function $q$ gives $\int_{B} q g d μ = \int_{B} q d ν = μ (B) (B \subseteq K Borel) .$ Uniqueness in [F5] therefore gives $q g = 1$ almost everywhere. The map $J : L^{2} (K, ν; C^{n (t)}) ⟶ L^{2} (K, μ; C^{n (t)}), J ξ = \sqrt{g} ξ,$ is an isometry by [F6]; multiplication by $\sqrt{q}$ is its inverse, so it is unitary. It commutes with every bounded Borel scalar multiplier. Hence $R = J U_{2} U_{1}^{- 1}$ is a unitary from the first model to the second model written over $μ$ , and it intertwines every scalar multiplier.

3.1F1F9step 2.1step 1.8step 1.9

Since $A = W^{*} (S)$ , unitary conjugation and steps 1.8--1.9 yield $U A U^{- 1} = W^{*} (U S U^{- 1}) = W^{*} (M_{t}) = D .$ These are both inclusions: WOT closedness gives $W^{*} (M_{t}) \subseteq D$ , and the bounded-continuous approximation gives the reverse inclusion.

3.2F7F9F13step 2.2

Over $(K, μ)$ form the measurable direct-sum field $Z_{t} = C^{m (t)} \oplus C^{n (t)}$ using the interleaved fundamental families: their Gram coefficients are the two Borel Gram matrices in diagonal blocks, so the field is measurable and has dense fundamental span. Its direct integral identifies with the Hilbert sum of the two model spaces by pairing section coordinates and adding their squared norms. In that sum, define the off-diagonal operator $\hat{R} (x, y) = (0, R x) .$ It commutes with every scalar multiplier. The commutant theorem [F9] makes it decomposable, say $\hat{R} = \int^{\oplus} T_{t} d μ (t)$ on $Z_{t}$ . Let $P_{F}, P_{G}$ be the global projections onto the first and second summands, induced by the pointwise projections $Q_{F} (t), Q_{G} (t)$ . Globally, $\hat{R} = P_{G} \hat{R} P_{F}, {\hat{R}}^{*} \hat{R} = P_{F}, \hat{R} {\hat{R}}^{*} = P_{G} .$ The adjoint/product and exact-norm clauses in [F9] imply, outside a common Borel null set, $T_{t} = Q_{G} (t) T_{t} Q_{F} (t), T_{t}^{*} T_{t} = Q_{F} (t), T_{t} T_{t}^{*} = Q_{G} (t) .$ Indeed each difference field induces the zero operator, so its essential supremum norm is zero; the finitely many exceptional null sets can be united. On this common conull set, the off-diagonal block of $T_{t}$ is a unitary map $C^{m (t)} \to C^{n (t)}$ . The dimensions of unitarily isomorphic finite or countably infinite Hilbert spaces agree, so $m (t) = n (t)$ almost everywhere. This argument includes both different finite ranks and the finite/infinite and infinite/infinite cases; it uses no published intertwiner-dimension lemma.

4.1

Steps 1.1--1.9, 2.1, and 3.1 prove existence of the generator, finite regular measure, measurable multiplicity field, unitary spectral model and algebra identity. Steps 1.10, 2.2, and 3.2 prove fixed-generator measure-class and almost-everywhere multiplicity uniqueness. No uniqueness is asserted across different generators. [step 1.1, step 1.2, step 1.7, step 2.1, step 3.1, step 1.10, step 2.2, step 3.2] $□$

Boundary cases

Empty: Inapplicable because $H \neq 0$ and the spectrum of a bounded operator on a nonzero complex Hilbert space is nonempty; the cyclic family is therefore nonempty.
Zero: The zero operator is allowed. Then $K = {0}$ ; the nonzero cyclic measure and the construction remain valid. The zero multiplier $f = 0$ is handled separately in Step 1.9 when its essential bound is zero.
One-dimensional: A one-dimensional $H$ is cyclic, so there is one summand and $m = 1$ almost everywhere; the same formulas give the scalar representation.
Degenerate: Each chosen cyclic vector and measure is nonzero; finite cyclic decompositions, overlaps of the active sets, null exceptional sets, and infinite multiplicity are handled explicitly. Values on $K ∖ K_{0}$ are assigned arbitrarily because that set is $μ$ -null.
Endpoints: No interval endpoints are removed. The compact spectral set may be a singleton or may contain its minimum and maximum; all Borel sets, including endpoint singletons, are included in the spectral-projection and measure-class arguments. Radial clipping includes the boundary $∣ z ∣ = M$ .
Choice: AC is stated and is a direct dependency. It supplies the generator, countable cyclic decomposition, Radon--Nikodym densities, direct-integral and operator-field hypotheses, and implies DC and countable choice for regularity and $C_{c}$ density. Countable representatives of dense vectors are selected under AC. Weighting and active-index enumeration are explicit.
Iff cases: The algebra identity has both inclusions proved separately in Steps 1.8 and 1.9. The fibrewise multiplicity conclusion in Step 3.2 follows in both directions from $T_{t}^{*} T_{t} = Q_{F} (t)$ and $T_{t} T_{t}^{*} = Q_{G} (t)$ on one common conull set.

Source qualifications

Anantharaman--Popa, Chapter 8 §8.1, Theorem 8.1.1, printed pp. 122--123, starts with a separable module over a standard probability-space model. Its proof passes from cyclic modules to a multiplicity partition and says that the partition details are an exercise; its uniqueness argument uses commutants. Remark 8.1.2 relates the result to the spectral multiplicity theorem for one self-adjoint operator. The present proof does not import the omitted partition details or its measure normalization.

Bekka--de la Harpe, Chapter 1 §1.G, Proposition 1.G.2, printed p. 60, states the measurable dimension-strata reduction and refers its proof elsewhere. Their Theorem 1.H.1, printed p. 65, states the intertwiner/decomposability criterion for varying fields and refers its general-field proof to Dixmier--von Neumann. Here the active-rank map is constructed directly, and the fixed-generator uniqueness proof uses the locally proved commutant and operator-action results of this pair. No citation is treated as a proof of these local steps.

5 · Examples, counterexamples and false statements

None yet.

Measurable Hilbert Fields and Direct-Integral Operators

1 · Prerequisites

2 · Summary

3 · Logical flowchart

4 · Definitions, theorems and proofs

Von Neumann algebras and commutants

Definition

Measurable Hilbert field from a countable fundamental family

Definition

Measurable sections have measurable pointwise inner products

Statement

Facts & Assumptions

Proof

A separably acting abelian von Neumann algebra has a self-adjoint generator

Statement

Facts & Assumptions

Proof

Remarks

Direct integral of a measurable Hilbert field

Definition

Direct integrals of measurable Hilbert fields are Hilbert spaces

Statement

Facts & Assumptions

Proof

Boundary cases

Source qualifications

Measurable and decomposable operator fields

Definition

Facts & Assumptions

Proof

Boundary cases

Source qualifications

Measurable essentially bounded operator fields act decomposably

Statement

Facts & Assumptions

Proof

Boundary cases

Source qualifications

Decomposable operators are the commutant of diagonal multiplication

Statement

Facts & Assumptions

Proof

Source qualifications

Boundary cases

Diagonal multipliers form a von Neumann algebra

Statement

Facts & Assumptions

Proof

Boundary cases

Source qualifications

Spectral multiplicity model for separably acting abelian von Neumann algebras

Statement

Facts & Assumptions

Proof

Boundary cases

Source qualifications

5 · Examples, counterexamples and false statements

Sources