Alphabeta Math
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.

✓ 11 results · all verified · 5 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.

Complete Reducibility for Compact Groups

1 · Prerequisites

2 · Summary

On a compact Hausdorff group the normalized Haar probability of haar-measure-existence-and-uniqueness can be used to average intrinsic data of a representation. Averaging a Hermitian inner product produces a positive definite invariant form, so a closed invariant subspace of a finite-dimensional unitary representation has an invariant orthogonal complement; induction on the dimension then shows that every finite-dimensional complex representation of a compact group is a direct sum of irreducible ones. The averaging operator on homomorphism spaces is defined as a weak operator integral against Haar measure, and its scalar pairing formula reduces all of its properties to the invariance of the measure; in particular it is a contraction projecting onto the space of intertwining operators.

The theory of compact operators supplies the infinite-dimensional half. Weak operator integration of the conjugation orbit of a positive rank-one operator produces a nonzero compact self-adjoint intertwiner, which Schur's lemma turns into a nonzero scalar; a compact scalar identity forces finite dimension. Convolution on L2(K) is treated separately through its Hilbert--Schmidt kernel, and the norm continuity of conjugation orbits is proved only for finite-rank operators, which is exactly what the averaging argument needs. The consequences are collected in the finite-dimensionality of continuous irreducible unitary representations of a compact group, using the Bochner integration and compact-operator machinery of banach-valued-integration-and-the-radon-nikodym-property, compact-operators-and-riesz-schauder-theory and compact-self-adjoint-hilbert-schmidt-and-trace-class-operators.

Schur orthogonality for general compact groups is proved by averaging rank-one maps with normalized Haar measure: matrix coefficients of inequivalent irreducible representations are orthogonal in L2(K), and a single irreducible satisfies the 1/d normalization in the first-variable-linear convention. These identities define the isotypic projection Pσ=dσ∫Kχσ(k)‾ π(k) dμ(k) attached to an irreducible σ and a strongly continuous unitary representation π. Its images are sums of finitely many σ-copies, and the averaged operator is shown to be a bounded self-adjoint idempotent commuting with π(K) whose range is exactly the σ-isotypic subspace, with orthogonal ranges for inequivalent types.

No completeness statement is made here. The page deliberately does not assert that the isotypic subspaces span the whole representation space, does not form a sum over the unitary dual and does not prove Peter--Weyl density; those belong to the later theory of general compact groups. The Axiom of Choice is consumed through the existence and uniqueness of Haar measure and through the Bochner and Hilbert-space suppliers, while the finite-dimensional averaging arguments themselves are choice free apart from their cited inputs.

3 · Logical flowchart

4 · Definitions, theorems and proofs

DefinitionDefinition: Literature-sourcedProof: Not applicablejudge pass (gpt-6.1-sol)audited 2026-10-02Open item page →

Averaged Hermitian form for a compact group

Definition

Assume the Axiom of Choice (The Axiom of Choice). Let K be a compact Hausdorff group and let μ be its normalized Haar probability measure (Normalized Haar probability on a compact group); this is the only place the Axiom of Choice is consumed by the definition.

Let V be a finite-dimensional complex vector space and let ρ:K→GL⁡(V) be a continuous finite-dimensional complex representation: a homomorphism of groups such that ρ is continuous when GL⁡(V)⊆End⁡(V) carries the topology induced by a norm on End⁡(V). In finite dimension any two norms on End⁡(V) induce the same topology (A chosen algebraic basis identifies a finite-dimensional normed space with a coordinate space), so the continuity requirement does not depend on the norm chosen.

Let h0 be a Hermitian inner product on V that is linear in the first variable and conjugate-linear in the second (Real and complex inner-product spaces and their induced length). The averaged Hermitian form of the pair (ρ,h0) is the map h:V×V→C defined by

h(v,w):=∫Kh0(ρ(k)v,ρ(k)w) dμ(k).

Well-definedness and conventions. The form h0 is linear in the first and conjugate-linear in the second variable, and so is h: for each fixed k the integrand is linear in v and conjugate-linear in w, and these properties pass through the integral. Hermitian symmetry likewise passes to the limit because the integrand of h(w,v) is the complex conjugate of the integrand of h(v,w) for every k. For fixed v,w the integrand k↦h0(ρ(k)v,ρ(k)w) is continuous: ρ is continuous, evaluation g↦ρ(g)v is therefore continuous, and h0 is continuous on the finite-dimensional space V×V (The inner product is jointly continuous). A continuous complex function on the compact space K is bounded and integrable against the Borel probability measure μ, so the displayed integral is a finite complex number and h is a sesquilinear form, linear in its first variable and conjugate-linear in its second. Whether h is positive definite and ρ(K)-invariant is a theorem, not a convention: those two properties are proved in Averaging a Hermitian form unitarizes a finite-dimensional compact-group representation.

LemmaStatement: Literature-sourcedProof: Literature-sourcedprecheck passjudge pass (gpt-6.1-sol)audited 2026-10-02Open item page →

Invariant orthogonal complements in unitary representations

Statement

Let G be a topological group, let π:G→U(H) be a strongly continuous unitary representation of G on a complex Hilbert space H (Strongly continuous unitary representations, invariant linear subspaces and intertwiners, Hilbert space), and let M⊆H be a closed invariant subspace. Then the orthogonal complement M⊥ (Orthogonality and the orthogonal complement) is a closed invariant subspace of H. This assertion is choice free.

Facts & Assumptions

Given: a topological group G, a strongly continuous unitary representation π on a complex Hilbert space H, and a closed invariant subspace M⊆H.

[F1]

Each π(g) is a bijective isometry with π(g)−1=π(g)∗, so ⟨π(g)x,y⟩=⟨x,π(g)−1y⟩ for all x,y∈H; and M is invariant, meaning π(g)M=M for every g∈G. (Strongly continuous unitary representations, invariant linear subspaces and intertwiners)

[F2]

For a subset S of an inner-product space, S⊥={v:⟨v,s⟩=0 for every s∈S} is a linear subspace, and orthogonality is symmetric. (Orthogonality and the orthogonal complement, Real and complex inner-product spaces and their induced length)

[F3]

The inner product is jointly continuous, so for each fixed y the map x↦⟨x,y⟩ is continuous. (The inner product is jointly continuous)

[F4]

In a metric space open balls are open, arbitrary unions of open sets are open, and a set is closed exactly when its complement is open; consequently {0} is closed in C, since its complement is the union of the open balls B(z,∣z∣) over z≠0. (Arbitrary unions and finite intersections of open sets are open, open balls are open and closed balls are closed, The metric topology: a set is open when every one of its points has a ball around it inside the set; closed means open complement)

[F5]

A closed set is the complement of an open set, and arbitrary intersections of closed sets are closed because arbitrary unions of open sets are open. (Topology on a set, open and closed sets, clopen sets, the closed-set axiomatisation, and the coarser/finer comparison)

Proof

technique · direct
1.1F1F2

Let x∈M⊥, y∈M and g∈G. Then ⟨π(g)x,y⟩=⟨x,π(g)−1y⟩, and π(g)−1y∈M because π(g)−1M=M; hence ⟨π(g)x,y⟩=0 and π(g)x∈M⊥, so π(g)M⊥⊆M⊥. Replacing g by g−1 gives M⊥⊆π(g)M⊥ as well, so π(g)M⊥=M⊥: the complement is invariant.

1.2F3F4F5

The complement of M⊥ in H is the union over y∈M of the sets {x:⟨x,y⟩≠0}, each of which is the preimage under the continuous map x↦⟨x,y⟩ of the open set C∖{0}; a union of open sets is open, so the complement of M⊥ is open and M⊥ is closed.

2.1F2step 1.1step 1.2∎

Together with the fact that M⊥ is a linear subspace, steps 1.1 and 1.2 show that M⊥ is a closed invariant subspace of H, with no use of any choice principle.

LemmaStatement: Literature-sourcedProof: Literature-sourcedprecheck passjudge pass (gpt-6.1-sol)audited 2026-10-02Open item page →

Averaging a Hermitian form unitarizes a finite-dimensional compact-group representation

Statement

Assume the Axiom of Choice (The Axiom of Choice). Let K be a compact Hausdorff topological group (Topological group: multiplication and inversion are continuous, Open cover, subcover, and compact topological space; a compact subset is a subspace that is compact in its own right, Hausdorff space: distinct points have disjoint open neighbourhoods; every metrizable space is Hausdorff and the indiscrete topology on two points is not) with normalized Haar probability measure μ (Normalized Haar probability on a compact group). Let V be a finite-dimensional complex vector space, let ρ:K→GL⁡(V) be a continuous finite-dimensional complex representation of K (A finite-dimensional representation ρ:G→GL⁡(V) over a field, and its degree), let h0 be a Hermitian inner product on V that is linear in the first variable and conjugate-linear in the second (Real and complex inner-product spaces and their induced length, Real and complex inner product spaces, with the inner product linear in the first argument), and let h(v,w):=∫Kh0(ρ(k)v,ρ(k)w) dμ(k) be the averaged form of the pair (ρ,h0) (Averaged Hermitian form for a compact group). Then

  1. h(v,v)>0 for every v∈V with v≠0; that is, h is positive definite, and
  2. h(ρ(g)v,ρ(g)w)=h(v,w) for every g∈K and all v,w∈V; that is, h is K-invariant.

Consequently h is an inner product on V, and every ρ(g) is a unitary operator of the finite-dimensional inner product space (V,h) (Linear isometries, and orthogonal or unitary operators on finite-dimensional inner product spaces), so the representation ρ is unitary for the averaged form h.

Facts & Assumptions

Given: AC, a compact Hausdorff group K with normalized Haar probability μ, a continuous finite-dimensional complex representation ρ:K→GL⁡(V), a Hermitian inner product h0 on V linear in the first variable, and the averaged form h.

[F1]

The averaged form is well defined: h is a sesquilinear form on V, linear in the first variable and conjugate-linear in the second, it is Hermitian in the sense h(v,w)=h(w,v)‾, the integrand k↦h0(ρ(k)v,ρ(k)w) is continuous on K for all v,w∈V, and a continuous complex function on the compact space K is bounded and integrable against the Borel probability measure μ, so the defining integral is a finite complex number (Averaged Hermitian form for a compact group).

[F2]

Normalized Haar: μ is a Borel probability measure with μ(K)=1 that is left and right invariant, μ(aE)=μ(E) and μ(Ea)=μ(E) for every Borel set E⊆K and every a∈K, and μ(U)>0 for every nonempty open U⊆K (Normalized Haar probability on a compact group, Haar measure is positive on nonempty open sets and finite on compact sets, Measure spaces).

[F3]

The form h0 is an inner product: it is linear in the first variable, conjugate-linear in the second, Hermitian, and positive definite, h0(v,v)≥0 with h0(v,v)=0 exactly for v=0; its induced length is ∥v∥h:=h(v,v) for any inner product h (Real and complex inner-product spaces and their induced length, Real and complex inner product spaces, with the inner product linear in the first argument).

[F4]

ρ is a group homomorphism with ρ(e)=IV and ρ(kg)=ρ(k)ρ(g) for all k,g∈K, and each ρ(g) is an invertible linear map of V; the map ρ:K→GL⁡(V) is continuous (A finite-dimensional representation ρ:G→GL⁡(V) over a field, and its degree).

[F5]

A continuous self-map T of the measure space (K,B(K),μ) is Borel measurable, and it is measure preserving when μ(T−1E)=μ(E) for every Borel E; in that case ∫Kf∘T dμ=∫Kf dμ for every integrable f (Measure-preserving transformations and systems, Integral invariance under measure-preserving maps). Right translation Tg(k):=kg is a homeomorphism because multiplication in a topological group is continuous (Topological group: multiplication and inversion are continuous), and by [F2] it is measure preserving: Tg−1E=Eg−1 has μ(Eg−1)=μ(E).

[F6]

Nonnegative measurable real functions have an extended integral that is monotone and positively homogeneous, and for nonnegative simple functions it agrees with the simple integral ∫∑jcjχEj dμ=∑jcjμ(Ej); in particular ∫c 1U dμ=c μ(U) for c≥0. For a real measurable f≥0 the Lebesgue integral of f equals this nonnegative integral, since the negative part vanishes (Monotonicity and nonnegative homogeneity of the nonnegative integral, The nonnegative integral agrees with the simple integral on simple functions, The integral of a nonnegative simple function, Integrable real and complex functions, and their integrals).

[F7]

A linear map T:V→W between inner product spaces is a linear isometry if ∥Tv∥=∥v∥ for every v, and an invertible linear isometry from a finite-dimensional complex inner product space to itself is a unitary operator (Linear isometries, and orthogonal or unitary operators on finite-dimensional inner product spaces).

Proof

technique · direct
1.1F1F3F4

Fix v,w∈V and put fv,w(k):=h0(ρ(k)v,ρ(k)w) for k∈K. By [F1] the function fv,w is continuous on K, hence its integral against μ is a finite complex number, and by [F4] ρ(e)=IV, so fv,w(e)=h0(v,w). If v≠0, then fv,v is real-valued with fv,v(e)=h0(v,v)>0 and fv,v≥0 pointwise, by [F3].

2.1F1F4F5step 1.1

Let g∈K and v,w∈V. Substituting the pair (ρ(g)v,ρ(g)w) into the defining integral of [F1], using the homomorphism property ρ(k)ρ(g)=ρ(kg) of [F4] and the notation of step 1.1, gives h(ρ(g)v,ρ(g)w)=∫Kh0(ρ(kg)v,ρ(kg)w) dμ(k)=∫Kfv,w(kg) dμ(k). The right translation Tg(k)=kg is a measure-preserving homeomorphism by [F5], and fv,w is continuous hence integrable, so the integral invariance theorem of [F5] gives ∫Kfv,w(kg) dμ(k)=∫Kfv,w(k) dμ(k)=h(v,w). Hence h(ρ(g)v,ρ(g)w)=h(v,w) for all g∈K and v,w∈V.

2.2F1F2F6F8step 1.1

Let v∈V with v≠0, put f:=fv,v and ε:=h0(v,v)/2>0, and let U:=f−1[(ε,∞)]. Since f is continuous by [F1] and (ε,∞) is open in R, [F8] shows that U is open in K; and e∈U because f(e)=h0(v,v)>ε by step 1.1, so U is nonempty. By step 1.1, f≥0 everywhere and f>ε on U, so f≥ε1U pointwise. Monotonicity and the indicator computation of [F6] applied to the real nonnegative function f give h(v,v)=∫Kf dμ≥∫Kε1U dμ=ε μ(U)>0, the final inequality by positivity of μ on the nonempty open set U in [F2]. Hence h is positive definite.

3.1F1F3F4F7step 2.1step 2.2∎

By [F1] the form h is sesquilinear and Hermitian; step 2.2 makes it positive definite, so h is an inner product on V, and step 2.1 makes h invariant under every ρ(g). Hence for every g∈K and v∈V one has h(ρ(g)v,ρ(g)v)=h(v,v), so the induced lengths of [F3] satisfy ∥ρ(g)v∥h=∥v∥h: each ρ(g) is a linear isometry of the finite-dimensional inner product space (V,h). Each ρ(g) is invertible by [F4], so [F7] makes every ρ(g) a unitary operator for h. Thus h is a positive-definite K-invariant Hermitian form on V, and the representation ρ is unitary for the averaged form h.

TheoremStatement: Literature-sourcedProof: Literature-sourcedprecheck passaudited 2026-10-02Open item page →

Complete reducibility of finite-dimensional compact-group representations

Statement

Assume the Axiom of Choice (The Axiom of Choice). Let K be a compact Hausdorff topological group (Topological group: multiplication and inversion are continuous, Open cover, subcover, and compact topological space; a compact subset is a subspace that is compact in its own right, Hausdorff space: distinct points have disjoint open neighbourhoods; every metrizable space is Hausdorff and the indiscrete topology on two points is not), let V be a finite-dimensional complex vector space, and let ρ:K→GL⁡(V) be a continuous finite-dimensional complex representation (A finite-dimensional representation ρ:G→GL⁡(V) over a field, and its degree, Finite-dimensional vector space, and its dimension dim⁡FV; infinite-dimensional means having no finite basis). Then ρ is completely reducible (A completely reducible representation as a finite direct sum of irreducible subrepresentations): there are finitely many irreducible subrepresentations V1,…,Vr⊆V (Subrepresentations, direct sums of representations, and irreducibility) with V=V1⊕⋯⊕Vr, the empty direct sum being allowed, so the zero representation is completely reducible.

Facts & Assumptions

Given: AC, a compact Hausdorff group K, a finite-dimensional complex vector space V, and a continuous finite-dimensional complex representation ρ:K→GL⁡(V).

[F1]

Averaging unitarizes: with a normalized Haar probability measure μ on K, for every Hermitian inner product h0 on V linear in the first variable, the averaged form h(v,w)=∫Kh0(ρ(k)v,ρ(k)w) dμ(k) is a positive-definite K-invariant Hermitian form, and every ρ(g) is a unitary operator for h (Averaging a Hermitian form unitarizes a finite-dimensional compact-group representation, Real and complex inner-product spaces and their induced length).

[F2]

The orthogonal complement of a closed invariant subspace of a strongly continuous unitary representation on a complex Hilbert space is again a closed invariant subspace (Invariant orthogonal complements in unitary representations, Hilbert space).

[F3]

If W is a subspace of a finite-dimensional real or complex inner product space V, then V=W⊕W⊥ (For a subspace W of a finite-dimensional inner product space, V=W⊕W⊥, Linear subspace of a vector space).

[F4]

A finite-dimensional subspace of a normed space is closed, in ZF (A finite-dimensional normed subspace is closed).

[F6]

Strong induction: if a property of naturals holds at n whenever it holds at every m<n, then it holds at every n (Strong (complete) induction).

[F7]

Definitions: a subrepresentation of ρ is a ρ-invariant linear subspace; ρ is irreducible when V≠0 and its only subrepresentations are 0 and V; ρ is completely reducible when V=V1⊕⋯⊕Vr with Vi irreducible subrepresentations, the empty sum allowed (Subrepresentations, direct sums of representations, and irreducibility, A completely reducible representation as a finite direct sum of irreducible subrepresentations, A finite-dimensional representation ρ:G→GL⁡(V) over a field, and its degree).

[F8]

A finite-dimensional normed space is a Banach space (Every finite-dimensional normed space is Banach, Banach space), and a complex inner-product space whose induced-length metric is complete is a complex Hilbert space (Hilbert space).

[F9]

Any two norms on a finite-dimensional complex vector space are equivalent (All norms on a finite-dimensional complex normed space are equivalent, Equivalent norms, and the dictionary with equivalent metrics); the operator norm satisfies ∥Bv∥≤∥B∥ ∥v∥ for bounded operators, which are continuous (The operator norm as the least bound and as the unit-sphere or unit-ball supremum, A bounded linear operator between normed spaces).

[F11]

Under AC every compact Hausdorff group has a normalized Haar probability measure. (Normalized Haar probability on a compact group)

Proof

technique · induction
1.1F5F7base

Base case. If dim⁡V=0, then V={0} by [F5], and V is the empty direct sum of irreducible subrepresentations, which [F7] allows; hence the zero representation is completely reducible.

1.2F5F7ih

Induction step setup. Fix a natural number n≥1 and assume the induction hypothesis: every continuous finite-dimensional complex representation of K on a complex vector space of dimension m<n is completely reducible. Let ρ:K→GL⁡(V) be a continuous finite-dimensional complex representation with dim⁡V=n; then V≠{0} by [F5].

1.3F1F5F8F11

By [F11], fix a normalized Haar probability measure μ on K. By [F5] fix an ordered basis (b1,…,bn) of V and let h0 be the Hermitian inner product in these coordinates, h0(∑iaibi,∑icibi):=∑iaici‾. By [F1] the averaged form h is a positive-definite K-invariant Hermitian form on V, so h is an inner product on V and every ρ(g) is a unitary operator for h. Since V has the ordered basis (b1,…,bn) of finite length, [F8] makes (V,∥⋅∥h) a Banach space for the norm induced by h, so (V,h) is a complex Hilbert space.

2.1F1F9step 1.3

The representation ρ is strongly continuous for the norm ∥⋅∥h: for fixed v∈V and g0∈K, the operator norm inequality of [F9] gives ∥ρ(g)v−ρ(g0)v∥h≤∥ρ(g)−ρ(g0)∥h∥v∥h, and g↦ρ(g) is continuous at g0 for the operator norm because it is continuous for the topology of some norm on the finite-dimensional complex space End⁡(V) and all such norms are equivalent by [F9]. Hence g↦ρ(g)v is continuous, and ρ:K→U(V,h) is a strongly continuous unitary representation of K on the complex Hilbert space (V,h).

2.2F7step 1.2

Irreducible case. If ρ is irreducible, then by [F7] its only subrepresentations are 0 and V; since V≠{0} by step 1.2, the space V is itself an irreducible subrepresentation and V=V is a direct sum with the single summand V, so ρ is completely reducible.

2.3F7step 1.2

Non-irreducible case. If ρ is not irreducible, then, since V≠{0}, [F7] provides a subrepresentation M⊆V with M≠{0} and M≠V.

3.1F2F3F4F5F10step 2.1step 2.3

In the situation of step 2.3, M is a finite-dimensional subspace of V, hence closed in (V,∥⋅∥h) by [F4], and it is invariant by definition; applying the complement lemma [F2] to the strongly continuous unitary representation ρ on the Hilbert space (V,h) from step 2.1, the orthogonal complement M⊥ is a closed invariant subspace. By the orthogonal decomposition [F3], V=M⊕M⊥; by the dimension formula [F10] and dim⁡M≥1, one has dim⁡M<n by [F5], and dim⁡M⊥=dim⁡V−dim⁡M<n.

4.1F7step 1.2step 3.1

Apply the induction hypothesis of step 1.2 to the restrictions ρM(g):=ρ(g)∣M and ρM⊥(g):=ρ(g)∣M⊥. These are continuous finite-dimensional complex representations: invariance makes ρ(g)∣M a linear self-map of M and injectivity of ρ(g) makes it invertible, the homomorphism property is inherited, and continuity follows from ∥ρ(g)∣M−ρ(g0)∣M∥≤∥ρ(g)−ρ(g0)∥, with the same argument for M⊥. Since dim⁡M<n and dim⁡M⊥<n by step 3.1, the induction hypothesis gives irreducible subrepresentations with M=M1⊕⋯⊕Mr and M⊥=Mr+1⊕⋯⊕Ms; these Mi are also irreducible subrepresentations of ρ, and concatenating with V=M⊕M⊥ gives V=M1⊕⋯⊕Ms, so ρ is completely reducible in this case as well.

5.1F6step 1.1step 2.2step 4.1discharge-induction∎

Discharge. Every continuous finite-dimensional complex representation of K of dimension n≥1 is completely reducible, by the two cases of steps 2.2 and 4.1 together with the induction hypothesis of step 1.2, and the case n=0 is step 1.1; strong induction [F6] therefore proves that every continuous finite-dimensional complex representation of K is completely reducible.

DefinitionDefinition: Literature-sourcedProof: Not applicableaudited 2026-10-02Open item page →

Haar averaging of bounded operators as a weak operator integral

Definition

Assume the Axiom of Choice (The Axiom of Choice). Let K be a compact Hausdorff group with normalized Haar probability measure μ (Normalized Haar probability on a compact group). Under the assumed Axiom of Choice the Axiom of Countable Choice holds as well (AC supplies the countable and dependent choices used in Banach integration).

Let π:K→U(H) and σ:K→U(J) be strongly continuous unitary representations of K on complex Hilbert spaces H and J (Strongly continuous unitary representations, invariant linear subspaces and intertwiners, Hilbert space), and let T∈B(H,J) be a bounded linear operator (A bounded linear operator between normed spaces). The pairing below is linear in the first variable and conjugate-linear in the second (Real and complex inner-product spaces and their induced length).

The scalar integrals. Fix v∈H and w∈J and put

bv(w):=∫K⟨σ(k)Tπ(k)−1v, w⟩ dμ(k).

The integrand is continuous on K. Indeed, k↦Tπ(k)−1v is continuous because π is strongly continuous and T is bounded, and for a continuous map ψ into J the map k↦σ(k)ψ(k) is continuous because every σ(k) is an isometry: for fixed k0,

∥σ(k)ψ(k)−σ(k0)ψ(k0)∥≤∥ψ(k)−ψ(k0)∥+∥(σ(k)−σ(k0))ψ(k0)∥,

and both terms tend to zero as k→k0. A continuous complex function on the compact space K is bounded and Borel, so the displayed integral is a finite complex number; and ∣⟨σ(k)Tπ(k)−1v,w⟩∣≤∥T∥ ∥v∥ ∥w∥ for every k by the Cauchy–Schwarz inequality (Cauchy–Schwarz: ∣⟨x,y⟩∣≤∥x∥ ∥y∥, with equality exactly for dependent pairs), so ∣bv(w)∣≤∥T∥ ∥v∥ ∥w∥. For fixed v the assignment w↦bv(w) is conjugate-linear and ∥w∥↦∥bv(w)∥ is bounded by ∥T∥ ∥v∥; hence

φv(w):=bv(w)‾=∫K⟨w,σ(k)Tπ(k)−1v⟩ dμ(k)

is a bounded linear functional on J of norm at most ∥T∥ ∥v∥.

The definition. Under Countable Choice the Hilbert space J has the Riesz representation property (Riesz representation for Hilbert spaces): there is a unique vector A(T)v∈J with φv(w)=⟨w,A(T)v⟩ for every w∈J, equivalently

⟨A(T)v,w⟩=bv(w)=∫K⟨σ(k)Tπ(k)−1v, w⟩ dμ(k)(v∈H, w∈J).

This determines A(T) as a map H→J, the Haar average of T with respect to the pair (σ,π).

Well-definedness. For each v the vector A(T)v is unique, so A(T) is a well-defined function. It is linear: if v=av1+bv2 then, by the conjugate-linearity in f of the representing vector recorded in Riesz representation for Hilbert spaces, the vector representing φav1+bv2=a‾ φv1+b‾ φv2 is aA(T)v1+bA(T)v2; and ∥A(T)v∥=∥φv∥≤∥T∥ ∥v∥ because the Riesz representation is isometric, so A(T) is a bounded linear operator with ∥A(T)∥≤∥T∥. The assignment A is itself linear in T: for fixed v,w the integrand is linear in T, so the scalar integral is, and equality of the weak matrix elements together with uniqueness of the Riesz vector gives A(aT+bT′)=aA(T)+bA(T′). Thus A:B(H,J)→B(H,J) is a well-defined map.

This is a weak operator integral. Only scalar functions are integrated in the definition: for fixed v and w, the continuous function k↦⟨σ(k)Tπ(k)−1v,w⟩ is integrated against the scalar measure μ, and the vector A(T)v is then produced by the Riesz representation theorem. No operator-norm continuity of the orbit k↦σ(k)Tπ(k)−1 of an arbitrary bounded operator T is assumed or asserted. For finite-rank T:H→J, it follows from Finite-rank conjugation orbits are operator-norm continuous as follows. Equip H⊕J with the sum inner product; it is a Hilbert space because Cauchy sequences converge in each coordinate. The representation τ(k)(v,w)=(π(k)v,σ(k)w) is unitary and strongly continuous. The bounded finite-rank operator S(v,w)=(0,Tv) satisfies τ(k)Sτ(k)−1(v,w)=(0,σ(k)Tπ(k)−1v). The cited lemma makes this conjugation orbit norm continuous, and restriction to H⊕{0} followed by projection onto J does not increase operator norm, proving the claimed continuity for T. Only the Axiom of Choice is used, through normalized Haar measure and Countable Choice. That A is an idempotent contraction onto the bounded intertwiners, with operator norm one exactly when Hom⁡K(H,J)≠{0}, is the content of Haar averaging projects contractively onto the bounded intertwiners ↗, which justifies the present definition.

LemmaStatement: Literature-sourcedProof: AI-adaptedprecheck passaudited 2026-10-02Open item page →

Haar averaging projects contractively onto the bounded intertwiners

Statement

Assume the Axiom of Choice (The Axiom of Choice). Let K be a compact Hausdorff group with normalized Haar probability measure μ (Normalized Haar probability on a compact group), and let π:K→U(H) and σ:K→U(J) be strongly continuous unitary representations of K on complex Hilbert spaces H and J (Strongly continuous unitary representations, invariant linear subspaces and intertwiners, Hilbert space). Write Hom⁡K(H,J):={T∈B(H,J):Tπ(g)=σ(g)T for every g∈K} for the space of bounded intertwiners and let A:B(H,J)→B(H,J) be the Haar averaging operator of Haar averaging of bounded operators as a weak operator integral, so that ⟨A(T)v,w⟩=∫K⟨σ(k)Tπ(k)−1v,w⟩ dμ(k) for all T∈B(H,J), v∈H, w∈J. Then:

  1. A is idempotent, A∘A=A, and its range is exactly Hom⁡K(H,J): A(T) intertwines for every bounded T, and A(T)=T for every bounded intertwiner T;
  2. A is a contraction, ∥A(T)∥≤∥T∥ for every T∈B(H,J), hence ∥A∥≤1;
  3. ∥A∥=1 if Hom⁡K(H,J)≠{0}, and ∥A∥=0 if Hom⁡K(H,J)={0}.

Facts & Assumptions

Given: AC, a compact Hausdorff group K with normalized Haar probability μ, strongly continuous unitary representations π on H and σ on J, the averaging map A of the definition item, and a bounded linear operator T∈B(H,J).

[F1]

The operator A(T) is well defined by the weak operator integral: for all v∈H and w∈J the displayed pairing formula holds; the integrand is continuous on K, hence bounded and integrable against μ; A is linear in T; and ∥A(T)∥≤∥T∥ (Haar averaging of bounded operators as a weak operator integral, A bounded linear operator between normed spaces). The definition uses Countable Choice, which follows from AC, through the Riesz representation theorem for J (Riesz representation for Hilbert spaces, Real and complex inner-product spaces and their induced length, AC supplies the countable and dependent choices used in Banach integration).

[F2]

The normalized Haar probability is left invariant and μ(K)=1; for h∈K the left translation Th(k):=hk is a measurable self-map with μ(Th−1E)=μ(h−1E)=μ(E), hence measure preserving, so ∫Kf(hk) dμ(k)=∫Kf(k) dμ(k) for every integrable f (Normalized Haar probability on a compact group, Measure-preserving transformations and systems, Integral invariance under measure-preserving maps, Measure spaces).

[F3]

π and σ are group homomorphisms into the unitary groups, so π(e)=idH, π(h−1k)−1=π(k)−1π(h), σ(hk)=σ(h)σ(k), and σ(h)−1=σ(h−1); each σ(h) is unitary with ⟨σ(h)x,y⟩=⟨x,σ(h)−1y⟩. A bounded operator T:H→J is an intertwiner, written T∈Hom⁡K(H,J), exactly when Tπ(g)=σ(g)T for every g∈K (Strongly continuous unitary representations, invariant linear subspaces and intertwiners).

[F4]

For bounded operators ∥Bv∥≤∥B∥ ∥v∥ for every vector v, the operator norm is the supremum of ∥Bv∥ over the closed unit ball (also when the domain is zero), and ∥B∥=0 exactly when B=0 (The operator norm as the least bound and as the unit-sphere or unit-ball supremum, A bounded linear operator between normed spaces).

Proof

technique · direct
1.1F1F2F3

Let T∈B(H,J), h∈K, v∈H and w∈J. Using unitarity of σ(h) and the pairing formula of [F1], ⟨σ(h)A(T)v,w⟩=⟨A(T)v,σ(h)−1w⟩=∫K⟨σ(k)Tπ(k)−1v,σ(h)−1w⟩ dμ(k)=∫K⟨σ(h)σ(k)Tπ(k)−1v,w⟩ dμ(k)=∫K⟨σ(hk)Tπ(k)−1v,w⟩ dμ(k) by the homomorphism property of [F3]. Substituting k=h−1k′ and using left invariance of μ as recorded in [F2], this equals ∫K⟨σ(k′)Tπ(h−1k′)−1v,w⟩ dμ(k′)=∫K⟨σ(k′)Tπ(k′)−1π(h)v,w⟩ dμ(k′)=⟨A(T)π(h)v,w⟩, by [F3] and the pairing formula of [F1] again. As v,w are arbitrary, σ(h)A(T)=A(T)π(h), and as h is arbitrary, A(T) is an intertwiner.

1.2F1F3

Let T∈Hom⁡K(H,J) and k∈K. Then σ(k)Tπ(k)−1=σ(k)σ(k)−1T=T by the intertwining relation and the homomorphism properties of [F3]. Hence the integrand of the pairing formula is the constant k↦⟨Tv,w⟩, whose integral against the probability measure μ is again ⟨Tv,w⟩; therefore ⟨A(T)v,w⟩=⟨Tv,w⟩ for all v,w and A(T)=T.

1.3F1F4

For every T∈B(H,J) the definition of A gives ∥A(T)∥≤∥T∥ by [F1], so the operator norm of the linear map A satisfies ∥A∥≤1 by the supremum description in [F4].

2.1F1step 1.1step 1.2

Let T∈B(H,J). By step 1.1 the operator A(T) lies in Hom⁡K(H,J), and step 1.2 applied to that intertwiner gives A(A(T))=A(T). Hence A∘A=A, so A is idempotent, and range⁡(A)⊆Hom⁡K(H,J); conversely every S∈Hom⁡K(H,J) satisfies S=A(S) by step 1.2, so Hom⁡K(H,J)⊆range⁡(A). Thus the range of A is exactly Hom⁡K(H,J).

3.1F4step 1.2step 1.3step 2.1∎

If Hom⁡K(H,J)≠{0}, choose a nonzero intertwiner S. Then S=A(S) by step 1.2, so ∥S∥=∥A(S)∥≤∥A∥ ∥S∥ by [F4], and since ∥S∥≠0 this gives ∥A∥≥1; with step 1.3, ∥A∥=1. If instead Hom⁡K(H,J)={0}, then range⁡(A)={0} by step 2.1, so A(T)=0 for every T and ∥A∥=0 by [F4]. Together with the contraction bound this proves the precise norm statement.

LemmaStatement: Literature-sourcedProof: AI-adaptedprecheck passaudited 2026-10-02Open item page →

L² convolution on a compact group is Hilbert–Schmidt

Statement

Assume the Axiom of Choice. Let K be a compact Hausdorff group with normalized Haar probability measure μ (Normalized Haar probability on a compact group) and let f∈L2(K,μ;C) be a class (Complex Haar L^p spaces and compactly supported functions). Then the left convolution Cf on L2(K,μ;C),

(Cfh)(x)=∫Kf(xy−1)h(y) dμ(y),

is a well-defined bounded linear operator whose definition does not depend on the chosen measurable representative of f and which is Hilbert–Schmidt with ∥Cf∥HS=∥f∥2; in particular Cf is compact. The operator acts on the regular representation only; no assertion is made about integrating a convolution kernel on an arbitrary irreducible representation.

Facts & Assumptions

Given: a compact Hausdorff group K with normalized Haar probability μ, a class f∈L2(K,μ;C), and Countable Choice as a consequence of AC.

[A1]

The Axiom of Choice is assumed. (The Axiom of Choice)

[F1]

Assume AC. For a Radon measure μ on an LCH space X the complex spaces L1(X,μ;C) and L2(X,μ;C) are complete, and Cc(X;C) is dense in both. (Completeness of the complex Haar L1 and L2 spaces and density of Cc)

[F2]

Under AC a class of the completed product of two sigma-finite measure spaces defines a bounded kernel operator, with ∑e∥Tke∥2=∥k∥22 for every Hilbert basis, so Tk is Hilbert–Schmidt with ∥Tk∥HS=∥k∥2; the operator is independent of the chosen representative of k. (L two kernels give Hilbert–Schmidt operators)

[F3]

Under Countable Choice every Hilbert–Schmidt operator is compact. (Hilbert–Schmidt operators are compact)

[F4]

For sigma-finite measure spaces and a product-measurable nonnegative function the iterated integrals exist and agree with the product integral. (Tonelli's theorem for nonnegative measurable functions on a sigma-finite product)

[F5]

The normalized Haar probability is a left Haar measure, is inversion invariant, and left translations and inversion preserve it; integrals of nonnegative measurable functions and of integrable real or complex functions are invariant under measure-preserving maps. (Normalized Haar probability on a compact group, Left Haar integral and left Haar measure, Integral invariance under measure-preserving maps)

[F6]

AC implies Countable Choice, the hypothesis needed for the Hilbert-space and completed-product L2 results [F9, F10] and the compactness conclusion [F3]. (AC supplies the countable and dependent choices used in Banach integration, The Axiom of Countable Choice (ACω))

[F8]

The product sigma-algebra is the sigma-algebra generated by measurable rectangles; pointwise limits of measurable functions are measurable. (The product sigma-algebra and its finite iterates, Sequential suprema, infima, limsup, liminf, and pointwise limits of measurable functions are measurable)

[F9]

Under Countable Choice L2 of a measure space is a Hilbert space for the integral pairing, so Cauchy–Schwarz holds for it (L2 with the integral pairing is a Hilbert space, Cauchy–Schwarz: ∣⟨x,y⟩∣≤∥x∥ ∥y∥, with equality exactly for dependent pairs).

[F10]

The completed product measure is the completion of the product measure on the product sigma-algebra. (The completed product measure)

Proof

technique · direct
1.1F7

Let u∈C(K) and put qu(x,y):=u(xy−1). Inversion and multiplication are continuous, so qu is continuous on the compact space K×K.

2.1F7F8step 1.1

For each ϵ>0, continuity gives an open-rectangle cover of K×K on each member of which qu varies by less than ϵ; compactness gives a finite subcover. Disjointify its rectangles by successive differences and on each nonempty piece use a value of qu from its rectangle. These pieces are measurable in the product sigma-algebra, so this gives a measurable simple function within ϵ of qu everywhere. Taking ϵ=1/n and applying pointwise-limit measurability to the real and imaginary parts shows that qu is product-measurable.

3.1F4F5step 2.1

For fixed x, the map y↦xy−1 preserves μ by [F5]. Tonelli applied to the measurable function ∣qu∣2 therefore gives ∥qu∥L2(μ×μ)2=∫K∫K∣u(xy−1)∣2 dμ(y)dμ(x)=∥u∥22.

4.1A1F1F9F10step 3.1

The map u↦qu is an isometry from C(K) into L2 of the completed product measure by step 3.1. Since C(K) is dense in L2(K) and the completed-product L2 space is complete, it extends to an isometry U:L2(K)→L2(μ×μ‾). Choose un∈C(K) with ∥f−un∥2→0 and put κ:=Uf=lim⁡nqun; then ∥κ∥2=∥f∥2, and the class κ is independent of the approximating sequence.

5.1F2F5F7F9step 1.1step 4.1

Choose measurable representatives of f and h∈L2(K) and define Cfh(x):=∫Kf(xy−1)h(y) dμ(y). For each x, the change of variables y↦xy−1 preserves μ, so Cauchy–Schwarz makes this integral finite and gives ∣Cfh(x)−Cunh(x)∣≤∥f−un∥2∥h∥2 for every continuous approximant un chosen in step 4.1, uniformly in x. Each Cunh is continuous: joint continuity of qun from step 1.1 and compactness of K give sup⁡y∣qun(x,y)−qun(x0,y)∣→0 as x→x0, and h∈L1(K) since μ(K)=1. Thus Cfh is a measurable uniform limit. By [F2], Tqun=Cun and ∥Tqun−Tκ∥≤∥qun−κ∥2→0, so the uniform limit also gives Cfh=Tκh as an L2 class.

6.1F2step 4.1step 5.1

The kernel theorem [F2] makes Tκ Hilbert–Schmidt with ∥Tκ∥HS=∥κ∥2=∥f∥2. By step 5.1 this operator is exactly Cf, so Cf is bounded and has the asserted Hilbert–Schmidt norm.

7.1F3F5F6step 6.1∎

If f is changed on a null set N, then for each x the convolution integrand changes only for y∈N−1x, since xy−1∈N iff y∈N−1x; this set is null by inversion and right invariance of μ. Changing the representative of the input h also leaves every section integral unchanged. Thus Cf is representative-independent; since Countable Choice holds, [F3] makes this Hilbert–Schmidt operator compact.

LemmaStatement: Literature-sourcedProof: AI-adaptedprecheck passjudge pass (gpt-6.1-sol)audited 2026-10-02Open item page →

Finite-rank conjugation orbits are operator-norm continuous

Statement

Assume the Axiom of Choice. Let G be a topological group and let π:G→U(H) be a strongly continuous unitary representation on a complex Hilbert space H (Strongly continuous unitary representations, invariant linear subspaces and intertwiners, Hilbert space). Let T∈B(H) be a bounded linear operator (A bounded linear operator between normed spaces) whose range T(H) is finite dimensional (that is, T is a finite-rank operator). Then the conjugation orbit

g⟼π(g)Tπ(g)−1

is continuous from G to B(H) for the operator norm (The operator norm as the least bound and as the unit-sphere or unit-ball supremum).

Facts & Assumptions

Given: a topological group G, a strongly continuous unitary representation π on a complex Hilbert space H, and a bounded finite-rank operator T.

[A1]

The Axiom of Choice is assumed. (The Axiom of Choice)

[F1]

For every v∈H the orbit map g↦π(g)v is norm continuous, and each π(g) is a bijective isometry, so π(g)−1=π(g)∗ and ∥π(g)v∥=∥v∥. (Strongly continuous unitary representations, invariant linear subspaces and intertwiners)

[F2]

Countable Choice holds under AC, and it is the hypothesis of the Riesz representation theorem. (AC supplies the countable and dependent choices used in Banach integration, Riesz representation for Hilbert spaces)

[F3]

A finite-dimensional inner product space has an orthonormal basis, and for an orthonormal basis e0,…,er−1 of a finite-dimensional subspace every vector v of that subspace satisfies v=∑i<r⟨v,ei⟩ei. (Every finite-dimensional real or complex inner product space has an orthonormal basis, Bessel's inequality for a finite orthonormal list and Parseval's identity for an orthonormal basis)

[F4]

Cauchy–Schwarz: ∣⟨x,y⟩∣≤∥x∥ ∥y∥. (Cauchy–Schwarz: ∣⟨x,y⟩∣≤∥x∥ ∥y∥, with equality exactly for dependent pairs)

[F5]

The operator norm is a bound and a least bound: ∥Sx∥≤∥S∥ ∥x∥, and ∥S∥ is the least such constant. (The operator norm as the least bound and as the unit-sphere or unit-ball supremum, A bounded linear operator between normed spaces)

Proof

technique · direct
1.1A1F2

The assumed Axiom of Choice supplies Countable Choice, so the Riesz representation theorem is available for bounded linear functionals on H.

1.2F3

Since T(H) is finite dimensional, it has an orthonormal basis a0,…,ar−1, and for every x∈H the vector Tx lies in T(H), so Tx=∑i<r⟨Tx,ai⟩ai.

1.3F1F4F5

For fixed a,b∈H the rank-one operator Ra,bx:=⟨x,b⟩a is bounded with ∥Ra,bx∥≤∥a∥ ∥b∥ ∥x∥ by Cauchy–Schwarz, so ∥Ra,b∥≤∥a∥ ∥b∥; and for g∈G one has π(g)Ra,bπ(g)−1=Rπ(g)a,π(g)b, because ⟨π(g)−1x,b⟩=⟨x,π(g)b⟩ and π(g) is linear.

2.1F2F4step 1.1step 1.2

Each functional x↦⟨Tx,ai⟩ is bounded, since ∣⟨Tx,ai⟩∣≤∥T∥ ∥x∥; by Riesz representation there is for each i a unique vector bi∈H with ⟨Tx,ai⟩=⟨x,bi⟩ for all x, so T=∑i<r⟨ ⋅ ,bi⟩ai=∑i<rRai,bi.

3.1step 1.3step 2.1

Conjugating the finite sum of step 2.1 and using step 1.3 termwise gives π(g)Tπ(g)−1=∑i<rπ(g)Rai,biπ(g)−1=∑i<rRπ(g)ai,π(g)bi for every g∈G.

4.1F1F4F5step 3.1

For g,h∈G and vectors a,b,a′,b′, Ra,b−Ra′,b′=Ra−a′,b+Ra′,b−b′, so ∥Ra,b−Ra′,b′∥≤∥a−a′∥ ∥b∥+∥a′∥ ∥b−b′∥; applying this with a=π(g)ai, b=π(g)bi, a′=π(h)ai, b′=π(h)bi and using unitarity bounds ∥π(g)Tπ(g)−1−π(h)Tπ(h)−1∥ by ∑i<r(∥bi∥ ∥π(g)ai−π(h)ai∥+∥ai∥ ∥π(g)bi−π(h)bi∥).

5.1F1step 4.1∎

The finitely many orbit maps g↦π(g)ai and g↦π(g)bi are norm continuous at h by strong continuity, so the bound of step 4.1 tends to zero as g→h; hence g↦π(g)Tπ(g)−1 is continuous in operator norm at every h∈G.

LemmaStatement: Literature-sourcedProof: AI-adaptedprecheck passaudited 2026-10-02Open item page →

A positive rank-one Haar average is a nonzero compact intertwiner

Statement

Assume AC. Let K be a compact Hausdorff group with normalized Haar probability μ (Normalized Haar probability on a compact group, Topological group: multiplication and inversion are continuous, Open cover, subcover, and compact topological space; a compact subset is a subspace that is compact in its own right, Hausdorff space: distinct points have disjoint open neighbourhoods; every metrizable space is Hausdorff and the indiscrete topology on two points is not), let π:K→U(H) be a strongly continuous unitary representation on a nonzero complex Hilbert space H (Strongly continuous unitary representations, invariant linear subspaces and intertwiners, Hilbert space), and let ξ∈H with ξ≠0. Write Rξx:=⟨x,ξ⟩ξ and φ(k):=π(k)Rξπ(k)−1 (Real and complex inner-product spaces and their induced length). Then φ is continuous in operator norm and Bochner integrable, and

Qξ:=∫Kφ(k) dμ(k)

is a bounded operator that is self-adjoint with ⟨Qξx,x⟩≥0 for every x∈H, satisfies Qξ≠0, is compact (Compact linear operator), and commutes with π(K): π(g)Qξ=Qξπ(g) for every g∈K. The Axiom of Choice is consumed through Haar measure, the Bochner framework and the compact-operator norm limit.

Facts & Assumptions

Given: AC, a compact Hausdorff group K with normalized Haar probability μ, a strongly continuous unitary representation π on a complex Hilbert space H, and ξ∈H, ξ≠0. Under AC Countable Choice holds (AC supplies the countable and dependent choices used in Banach integration).

[A1]

Normalized Haar: μ is a left-invariant and right-invariant probability Borel measure, and μ(U)>0 for every nonempty open U⊆K (Normalized Haar probability on a compact group, Haar measure is positive on nonempty open sets and finite on compact sets, Measure spaces).

[A2]

Each π(k) is unitary with π(k)−1=π(k−1) and ⟨π(k)x,y⟩=⟨x,π(k)−1y⟩, π is a homomorphism, and k↦π(k)x is continuous for every x (Strongly continuous unitary representations, invariant linear subspaces and intertwiners).

[A3]

The inner product is linear in the first variable and conjugate-linear in the second, ⟨x,y⟩=⟨y,x⟩‾, ∥x∥2=⟨x,x⟩≥0 with equality only for x=0, and ∣⟨x,y⟩∣≤∥x∥ ∥y∥ (Real and complex inner-product spaces and their induced length, Cauchy–Schwarz: ∣⟨x,y⟩∣≤∥x∥ ∥y∥, with equality exactly for dependent pairs).

[A5]

Conjugation orbits of a bounded finite-rank operator are continuous in operator norm (Finite-rank conjugation orbits are operator-norm continuous).

[A6]

Bochner framework: a strongly measurable function into a Banach space is Bochner integrable exactly when the integral of its norm is finite, the integral is the norm limit of the integrals of L1-approximating simple functions, and ∥∫Ef dμ∥≤∫E∥f∥ dμ (Strongly measurable Banach-valued function, Banach-valued simple function and integral, Bochner-integrable function, Bochner integrability criterion, Bochner integral norm inequality, Banach space, If (Y) is Banach then (\mathcal B(X,Y)) is Banach). A bounded linear map commutes with the Bochner integral (Bounded linear maps commute with Bochner integration), so for fixed v,w∈H the bounded functional S↦⟨Sv,w⟩ on B(H) gives ⟨(∫Kf dμ)v,w⟩=∫K⟨f(k)v,w⟩dμ(k) for every Bochner integrable f:K→B(H).

[A8]

Finite linear combinations of compact operators are compact, and a norm limit of compact operators is compact under Countable Choice (Linear combinations of compact operators are compact, Norm limit of compact operators is compact).

[A9]

Left translation is a measurable measure-preserving self-map of K by [A1], so for every integrable f:K→C and g∈K one has ∫Kf(gk) dμ(k)=∫Kf(k) dμ(k) (Integral invariance under measure-preserving maps).

[A10]

Continuity of maps into K, balls and neighbourhoods are as in Continuity of a map of topological spaces at a point and globally.

Proof

technique · direct
1.1A3A4

The map Rξx=⟨x,ξ⟩ξ is linear and bounded with ∥Rξx∥≤∥x∥ ∥ξ∥2 by Cauchy--Schwarz, and ∥Rξξ∥=∥ξ∥3, so ∥Rξ∥=∥ξ∥2; its range is contained in the span Cξ of the one-term list ξ, which admits the ordered basis of length one given by ξ because ξ≠0, so Rξ is compact by [A4]. Moreover ⟨Rξx,y⟩=⟨x,ξ⟩⟨ξ,y⟩=⟨x,⟨y,ξ⟩ξ⟩=⟨x,Rξy⟩, so Rξ is self-adjoint, ⟨Rξx,x⟩=∣⟨x,ξ⟩∣2≥0, and ⟨Rξξ,ξ⟩=∥ξ∥4>0, so Rξ≠0.

2.1A2A4A5step 1.1

For every k∈K the operator φ(k)=π(k)Rξπ(k)−1 is bounded; it is compact by [A4], because φ(k)x=⟨x,π(k)ξ⟩π(k)ξ has range in the one-dimensional span of the nonzero vector π(k)ξ; it is self-adjoint and non-negative because ⟨φ(k)x,y⟩=⟨Rξπ(k)−1x,π(k)−1y⟩=⟨π(k)−1x,Rξπ(k)−1y⟩=⟨x,φ(k)y⟩ and ⟨φ(k)x,x⟩=⟨Rξπ(k)−1x,π(k)−1x⟩≥0 by step 1.1; and unitary conjugation preserves norms, so ∥φ(k)∥=∥Rξ∥=∥ξ∥2. The orbit map k↦φ(k) is continuous in operator norm by [A5], since Rξ is bounded of finite rank.

3.1A6A7step 2.1

The orbit φ:K→B(H) is continuous by step 2.1 into the Banach space B(H), so its image φ[K] is a compact subset of the metric space B(H) and hence totally bounded: for each n≥1 there is a finite Fn⊆φ[K] with φ[K]⊆⋃y∈FnB(y,1/n); choosing these nets for all n, and enumerating each finite net, is licensed by Countable Choice and finite choice. List Fn={y0,…,ym} and put Aj:=φ−1[B(yj,1/n)]∖⋃i<jAi: each Aj is Borel, being a difference of Borel sets, the Aj are pairwise disjoint, and they cover K; hence tn:=∑jyj1Aj is a B(H)-valued measurable simple function, and ∥tn(k)−φ(k)∥<1/n for every k∈K, because k lies in the first Aj whose net point is within 1/n of φ(k). Thus the tn converge to φ pointwise in norm, so φ is strongly measurable; and since ∥φ(k)∥=∥ξ∥2 for every k and μ(K)=1, one has ∫K∥φ∥ dμ=∥ξ∥2<∞ and ∫K∥φ−tn∥ dμ≤1/n→0, so φ is Bochner integrable and Qξ:=∫Kφ dμ∈B(H) is defined with ∥Qξ∥≤∫K∥φ∥ dμ=∥ξ∥2.

4.1A6step 3.1

For all v,w∈H the pairing formula holds: ⟨Qξv,w⟩=∫K⟨φ(k)v,w⟩dμ(k), by applying the commuting theorem of [A6] to the bounded linear functional S↦⟨Sv,w⟩ on B(H) and the integrable function φ.

4.2A8step 2.1step 3.1

The operator Qξ is compact. By the norm inequality of [A6], ∥Qξ−∫Ktn dμ∥≤∫K∥φ−tn∥ dμ≤1/n, so Qξ is the norm limit of the integrals ∫Ktn dμ. Each of those is ∑jμ(Aj)yj, a finite linear combination of net points yj∈φ[K], and each such point equals φ(kj) for some kj∈K and is therefore compact by step 2.1; so each ∫Ktn dμ is compact by [A8], and the norm limit Qξ is compact by [A8] under Countable Choice.

5.1A3step 2.1step 4.1

The operator Qξ is self-adjoint with ⟨Qξx,x⟩≥0 for every x: by step 4.1 and the pointwise properties of step 2.1, ⟨Qξx,x⟩=∫K⟨φ(k)x,x⟩dμ(k)≥0, and ⟨Qξx,y⟩=∫K⟨φ(k)x,y⟩dμ(k)=∫K⟨x,φ(k)y⟩dμ(k)=∫K⟨φ(k)y,x⟩dμ(k)‾=⟨Qξy,x⟩‾=⟨x,Qξy⟩.

5.2A1A2A3A10step 2.1step 4.1

Qξ≠0: by steps 2.1 and 4.1, ⟨Qξξ,ξ⟩=∫K∣⟨π(k)−1ξ,ξ⟩∣2dμ(k). The integrand is continuous, non-negative, and equals ∥ξ∥4>0 at k=e, so by continuity there is an open neighbourhood U of e on which it exceeds 12∥ξ∥4; then the integral is at least 12∥ξ∥4μ(U)>0 by positivity of μ on nonempty open sets, since U is nonempty. Hence ⟨Qξξ,ξ⟩>0 and in particular Qξ≠0.

5.3A1A2A9step 2.1step 4.1

Qξ commutes with π(K): for g∈K and v,w∈H, the function k↦⟨φ(k)π(g)v,w⟩ is continuous and bounded on compact K by step 2.1, hence integrable against the probability μ. Using unitarity, the relation π(g)φ(k)=φ(gk)π(g), and the translation invariance of the scalar integral, ⟨π(g)Qξv,w⟩=⟨Qξv,π(g)−1w⟩=∫K⟨φ(k)v,π(g)−1w⟩dμ(k)=∫K⟨π(g)φ(k)v,w⟩dμ(k)=∫K⟨φ(gk)π(g)v,w⟩dμ(k)=∫K⟨φ(k)π(g)v,w⟩dμ(k)=⟨Qξπ(g)v,w⟩; since this holds for all v,w, the operators agree.

6.1A1A6A8step 4.2step 5.1step 5.2step 5.3∎

Collecting: φ is norm continuous and Bochner integrable and Qξ=∫Kφ dμ is a bounded self-adjoint operator with ⟨Qξx,x⟩≥0 for all x (step 5.1), nonzero (step 5.2), compact (step 4.2), and commuting with every π(g) (step 5.3). The Axiom of Choice entered only through the normalized Haar measure of [A1], the countable selections in the Bochner and net constructions of step 3.1, and the countable-choice compact-operator limit of [A8].

LemmaStatement: Literature-sourcedProof: Literature-sourcedprecheck passaudited 2026-10-02Open item page →

A nonzero compact scalar identity forces finite dimension

Statement

Let H be a real or complex Hilbert space (Hilbert space) and let c be a nonzero scalar. If the scalar operator cIH is a compact operator (Compact linear operator), then H is finite dimensional: it admits an ordered basis of finite length. This implication is choice free.

Facts & Assumptions

Given: a real or complex Hilbert space H, a nonzero scalar c, and the assumption that cIH is compact.

[F1]

A linear operator T is compact exactly when the closure of T(B‾) is compact, where B‾={x:∥x∥≤1} is the closed unit ball. (Compact linear operator)

[F2]

If T is compact and A is bounded, then the composite TA is compact. (Compositions with a compact operator are compact)

[F3]

A normed space X has compact closed unit ball if and only if X admits an ordered basis of finite length. (The closed unit ball is compact if and only if the normed space is finite-dimensional)

[F4]

For all vectors x,y of a normed space, ∣ ∥x∥−∥y∥ ∣≤∥x−y∥. (The reverse triangle inequality in a normed space)

[F5]

The scalar operator c−1IH is bounded, with ∥c−1IH∥≤∣c∣−1, including H={0}. (A bounded linear operator between normed spaces, The operator norm as the least bound and as the unit-sphere or unit-ball supremum)

Proof

technique · direct
1.1F5algebra

The scalar operator c−1IH satisfies ∥c−1IHx∥=∣c∣−1∥x∥ for every x∈H, so it is bounded with bound ∣c∣−1 so its operator norm is at most ∣c∣−1, including when H={0}.

1.2F4

The closed unit ball B‾ is closed in H: if ∥x∥>1, put r=∥x∥−1>0. Whenever ∥y−x∥<r, [F4] gives ∥y∥≥∥x∥−∥y−x∥>1, so the open ball of radius r about x is disjoint from B‾. Its complement is therefore open.

2.1F2step 1.1

The composite (cIH)∘(c−1IH) is compact by [F2], applied with the compact operator cIH and the bounded operator c−1IH; this composite is the identity IH.

3.1F1step 1.2step 2.1

By [F1] applied to IH, the closure of IH(B‾)=B‾ in H is compact. By step 1.2 the ball B‾ is closed, so this closure is B‾ itself; hence B‾ is a compact subset of H.

4.1F3step 3.1∎

By [F3] applied to the normed space H, compactness of B‾ means that H admits an ordered basis of finite length, that is, H is finite dimensional.

TheoremStatement: Literature-sourcedProof: AI-adaptedprecheck passjudge pass (gpt-6.1-sol)audited 2026-10-02Open item page →

Irreducible unitary representations of compact groups are finite dimensional

Statement

Facts & Assumptions

Given: AC, a compact Hausdorff group K, and an irreducible strongly continuous unitary representation π of K on a nonzero complex Hilbert space H.

[F1]

Under AC there is a normalized Haar probability measure on K, and for every ξ∈H with ξ≠0 the rank-one average Qξ:=∫Kπ(k)Rξπ(k)−1 dμ(k), where Rξx=⟨x,ξ⟩ξ, is a well-defined bounded operator on H that is self-adjoint, nonzero, compact (Compact linear operator), and satisfies π(g)Qξ=Qξπ(g) for every g∈K (A positive rank-one Haar average is a nonzero compact intertwiner, Normalized Haar probability on a compact group, A bounded linear operator between normed spaces).

[F2]

Schur's lemma: for an irreducible strongly continuous unitary representation of a topological group on a nonzero complex Hilbert space, every bounded operator commuting with the representation is a scalar multiple of the identity (Schur lemma for complex unitary representations).

[F3]

If a nonzero scalar multiple cIH of the identity of a real or complex Hilbert space is a compact operator, then H admits an ordered basis of finite length; this implication is choice free (A nonzero compact scalar identity forces finite dimension).

[F4]

π is a group homomorphism, and IH≠0 because H≠{0} (Strongly continuous unitary representations, invariant linear subspaces and intertwiners).

Proof

technique · direct
1.1F1

Since H≠{0}, choose an element ξ∈H with ξ≠0, which is exactly the hypothesis under which [F1] applies. The operator Qξ of [F1] is therefore well defined, bounded, self-adjoint, nonzero, compact, and commutes with every operator π(g), that is, Qξπ(g)=π(g)Qξ for all g∈K.

2.1F2F4step 1.1

The operator Qξ is a bounded operator commuting with the irreducible representation π, so Schur's lemma [F2] provides a scalar c with Qξ=cIH. Since Qξ≠0 by step 1.1 while IH≠0 by [F4], the scalar satisfies c≠0.

3.1F3step 2.1∎

Now cIH=Qξ is a nonzero compact scalar identity with c≠0, so [F3] applied to the Hilbert space H shows that H admits an ordered basis of finite length. Hence every irreducible strongly continuous unitary representation of K is finite dimensional.

TheoremStatement: Literature-sourcedProof: Literature-sourcedprecheck passaudited 2026-10-02Open item page →

Schur orthogonality for general compact groups

Statement

Assume the Axiom of Choice (The Axiom of Choice). Let K be a compact Hausdorff topological group (Topological group: multiplication and inversion are continuous, Open cover, subcover, and compact topological space; a compact subset is a subspace that is compact in its own right, Hausdorff space: distinct points have disjoint open neighbourhoods; every metrizable space is Hausdorff and the indiscrete topology on two points is not) with normalized Haar probability measure μ (Normalized Haar probability on a compact group), and let π:K→U(V) and ρ:K→U(W) be irreducible strongly continuous unitary representations of K on nonzero complex Hilbert spaces V and W (Strongly continuous unitary representations, invariant linear subspaces and intertwiners, Hilbert space). Let d:=dim⁡CV be the degree of π, so that d<∞ and d≥1. Matrix coefficients are written k↦⟨π(k)v,w⟩ and k↦⟨ρ(k)v′,w′⟩ with the pairing linear in the first variable (Matrix coefficient of a unitary representation, Real and complex inner-product spaces and their induced length). Then:

  1. (Inequivalent pair.) If π and ρ are not unitarily equivalent (Linear isometries, and orthogonal or unitary operators on finite-dimensional inner product spaces), then ∫K⟨π(k)v,w⟩ ⟨ρ(k)v′,w′⟩‾ dμ(k)=0 for all v,w∈V and all v′,w′∈W: the matrix coefficients of inequivalent irreducibles are orthogonal in L2(K).
  2. (A single irreducible.) For all v,w,v′,w′∈V, ∫K⟨π(k)v,w⟩ ⟨π(k)v′,w′⟩‾ dμ(k)=1d ⟨v,v′⟩ ⟨w,w′⟩‾. The case d=1 is included.
  3. (Equivalent models.) If U:V→W is a unitary intertwiner, that is Uπ(k)=ρ(k)U for every k∈K, then ∫K⟨π(k)v,w⟩ ⟨ρ(k)Uv′,Uw′⟩‾ dμ(k)=1d ⟨v,v′⟩ ⟨w,w′⟩‾ for all v,w,v′,w′∈V.

Facts & Assumptions

Given: AC; a compact Hausdorff group K with normalized Haar probability μ; irreducible strongly continuous unitary representations π on the nonzero complex Hilbert space V and ρ on the nonzero complex Hilbert space W; vectors v,w∈V and v′,w′∈W.

[F1]

Finite dimensionality: the representation spaces of irreducible strongly continuous unitary representations of K are finite dimensional, so d:=dim⁡V<∞ and dim⁡W<∞, and d≥1 because V≠{0} (Irreducible unitary representations of compact groups are finite dimensional, Strongly continuous unitary representations, invariant linear subspaces and intertwiners).

[F2]

Schur's lemma: every bounded self-intertwiner of an irreducible strongly continuous unitary representation is a scalar multiple of the identity, and a nonzero bounded intertwiner between two such representations forces them to be unitarily equivalent; hence inequivalent irreducibles admit no nonzero bounded intertwiner (Schur lemma for complex unitary representations).

[F3]

Haar averaging: the weak operator integral ⟨A(T)x,z⟩=∫K⟨σ(k)Tπ0(k)−1x,z⟩ dμ(k) defines a linear contraction A on the bounded operators between the carrier spaces of strongly continuous unitary representations π0 and σ, and its range is exactly the space of bounded intertwiners (Haar averaging of bounded operators as a weak operator integral, Haar averaging projects contractively onto the bounded intertwiners); here A is linear and ∥A(T)∥≤∥T∥ (A bounded linear operator between normed spaces, The operator norm as the least bound and as the unit-sphere or unit-ball supremum).

[F4]

Trace: for a finite-dimensional complex vector space with an orthonormal basis e1,…,ed, the trace of an endomorphism T satisfies tr⁡(T)=∑i=1d⟨Tei,ei⟩ and tr⁡(cI)=c d; similar endomorphisms have equal trace, so tr⁡(π(k)Tπ(k)−1)=tr⁡(T) for every k (The basis-independent trace of an endomorphism of a finite-dimensional vector space, Similar matrices have the same trace, Every finite-dimensional real or complex inner product space has an orthonormal basis).

[F5]

Orthonormal expansions: for an orthonormal basis e1,…,ed of a finite-dimensional inner product space and any vector y, one has y=∑i=1d⟨y,ei⟩ei; the pairing is linear in the first variable and conjugate-linear in the second, so a finite sum pulls out of the first variable, ⟨∑i=1dciei,z⟩=∑i=1dci⟨ei,z⟩ (Bessel's inequality for a finite orthonormal list and Parseval's identity for an orthonormal basis, Real and complex inner-product spaces and their induced length).

[F6]

Scalar integral: for a probability measure, ∫K1 dμ=1, finite sums and scalar multiples of integrable functions integrate termwise (The Lebesgue integral is linear on L1(μ), Integrable real and complex functions, and their integrals).

[F7]

A rank-one operator Tx=⟨x,v′⟩v is linear (Linear map between vector spaces over the same field) and bounded with ∥Tx∥≤∥v′∥ ∥x∥ ∥v∥ by the Cauchy--Schwarz inequality (Cauchy–Schwarz: ∣⟨x,y⟩∣≤∥x∥ ∥y∥, with equality exactly for dependent pairs).

[F8]

A unitary intertwiner U:V→W is a bijective linear isometry satisfying Uπ(k)=ρ(k)U; hence ⟨ρ(k)Uv′,Uw′⟩=⟨Uπ(k)v′,Uw′⟩=⟨π(k)v′,w′⟩ (Linear isometries, and orthogonal or unitary operators on finite-dimensional inner product spaces, Real and complex inner-product spaces and their induced length).

Proof

technique · direct
1.1F1

Fix v,w∈V and v′,w′∈W. By [F1] the dimensions d=dim⁡V and dim⁡W are finite and d≥1.

1.2F2F3F7

Inequivalent case. Assume first that π and ρ are not unitarily equivalent, and define T:W→V by Tx:=⟨x,v′⟩v. By [F7] the operator T is bounded of rank one. Applying the Haar averaging operator A of [F3] to the pair (ρ,π) of representations, the operator A(T):W→V is a bounded intertwiner, so A(T)ρ(k)=π(k)A(T) for every k. Since π and ρ are inequivalent irreducible representations, Schur's lemma [F2] forces A(T)=0.

1.3F2F3F7

Same-representation case. Now take ρ=π on W=V and T:V→V, Tx=⟨x,v′⟩v as before, a bounded finite-rank operator. The average A(T):V→V is a bounded self-intertwiner of the irreducible representation π, so Schur's lemma [F2] provides a scalar c with A(T)=cI.

1.4F4F5

The rank-one trace is tr⁡(T)=⟨v,v′⟩: in the orthonormal basis e1,…,ed, the matrix of T has entries ⟨Tej,ei⟩=⟨ej,v′⟩⟨v,ei⟩, so tr⁡(T)=∑i=1d⟨ei,v′⟩⟨v,ei⟩=∑i=1d⟨v,ei⟩⟨ei,v′⟩=⟨∑i=1d⟨v,ei⟩ei,v′⟩=⟨v,v′⟩, pulling the finite sum out of the first variable by [F5] and using the orthonormal expansion v=∑i=1d⟨v,ei⟩ei of [F5].

2.1F3F7step 1.2

Evaluating at w′ and w, the weak pairing formula of [F3] for the pair (ρ,π) and the definition of T give ⟨A(T)w′,w⟩=∫K⟨π(k)Tρ(k)−1w′,w⟩ dμ(k)=∫K⟨ρ(k)−1w′,v′⟩ ⟨π(k)v,w⟩ dμ(k). Unitarity of ρ(k) gives ⟨ρ(k)−1w′,v′⟩=⟨w′,ρ(k)v′⟩=⟨ρ(k)v′,w′⟩‾, so the integral equals ∫K⟨π(k)v,w⟩ ⟨ρ(k)v′,w′⟩‾ dμ(k). Since A(T)=0 by step 1.2, this integral is 0.

2.2F3F4F6step 1.3

The trace of A(T) computes c and the trace of T: by [F4] applied to an orthonormal basis e1,…,ed of V, tr⁡(A(T))=∑i=1d⟨A(T)ei,ei⟩=∑i=1d∫K⟨π(k)Tπ(k)−1ei,ei⟩ dμ(k)=∫Ktr⁡(π(k)Tπ(k)−1)dμ(k)=∫Ktr⁡(T) dμ(k)=tr⁡(T), using the weak pairing formula of [F3] for ⟨A(T)ei,ei⟩, termwise integration by [F6], the trace of the conjugated endomorphism in [F4], and μ(K)=1 in [F6]. On the other hand tr⁡(A(T))=tr⁡(cI)=cd by [F4].

3.1F3step 2.2step 1.4step 2.1algebra

Combining steps 1.3, 2.2 and 1.4 gives cd=tr⁡(A(T))=tr⁡(T)=⟨v,v′⟩, hence c=d−1⟨v,v′⟩. Evaluating the weak pairing formula of [F3] at w′ and w exactly as in step 2.1 then gives ∫K⟨π(k)v,w⟩ ⟨π(k)v′,w′⟩‾ dμ(k)=⟨A(T)w′,w⟩=c ⟨w′,w⟩=1d⟨v,v′⟩⟨w,w′⟩‾, where the last equality uses ⟨w′,w⟩=⟨w,w′⟩‾.

4.1F8step 2.1step 3.1∎

Equivalent models. If U:V→W is a unitary intertwiner, then by [F8] the coefficient ⟨ρ(k)Uv′,Uw′⟩ equals ⟨π(k)v′,w′⟩, so the integral in claim 3 reduces to the integral of claim 2 and equals d−1⟨v,v′⟩⟨w,w′⟩‾. Together with steps 2.1 and 3.1 this proves the orthogonality of matrix coefficients for inequivalent irreducibles, the normalized d−1 formula for a single irreducible including d=1, and the equivalent-model form.

DefinitionDefinition: Literature-sourcedProof: Not applicablejudge pass (gpt-6.1-sol)audited 2026-10-02Open item page →

Compact-group isotypic projection

Definition

Assume the Axiom of Choice (The Axiom of Choice). Let K be a compact Hausdorff topological group (Topological group: multiplication and inversion are continuous, Open cover, subcover, and compact topological space; a compact subset is a subspace that is compact in its own right, Hausdorff space: distinct points have disjoint open neighbourhoods; every metrizable space is Hausdorff and the indiscrete topology on two points is not) with normalized Haar probability measure μ (Normalized Haar probability on a compact group); by AC the Axiom of Countable Choice holds (AC supplies the countable and dependent choices used in Banach integration).

Let σ:K→U(Vσ) be an irreducible strongly continuous unitary representation of K on a nonzero complex Hilbert space Vσ (Strongly continuous unitary representations, invariant linear subspaces and intertwiners). By Irreducible unitary representations of compact groups are finite dimensional the space Vσ is finite dimensional; write dσ:=dim⁡CVσ<∞,dσ≥1, the degree of σ (Finite-dimensional vector space, and its dimension dim⁡FV; infinite-dimensional means having no finite basis). Fix an orthonormal basis e1,…,edσ of Vσ (Bessel's inequality for a finite orthonormal list and Parseval's identity for an orthonormal basis). The character of σ is χσ(k):=tr⁡σ(k)(k∈K), the trace of the endomorphism σ(k) of Vσ (The basis-independent trace of an endomorphism of a finite-dimensional vector space). It is a class function, χσ(hkh−1)=χσ(k) for all h,k∈K, because conjugation by σ(h) is a similarity (Similar matrices have the same trace), and it is continuous: for a finite-dimensional space the strong continuity of σ makes every matrix coefficient k↦⟨σ(k)ei,ej⟩ continuous, and the trace is the finite sum ∑i=1dσ⟨σ(k)ei,ei⟩ of these.

Let π:K→U(H) be a strongly continuous unitary representation of K on a complex Hilbert space H, and fix v∈H. The integrand is the function fv:K→H,fv(k):=χσ(k)‾ π(k)v.

Well-definedness of the vector-valued integral. The function fv is continuous: k↦π(k)v is norm-continuous by strong continuity of π (Strongly continuous unitary representations, invariant linear subspaces and intertwiners), k↦χσ(k)‾ is continuous, and products of a continuous scalar function with a continuous vector-valued function are continuous. Its image fv[K] is therefore a compact subset of the metric space H (A continuous image of a compact space is compact; a continuous real-valued map on a nonempty compact space attains a maximum and a minimum; and a continuous bijection from a compact space to a Hausdorff space is a homeomorphism, For a metric space with its metric topology, compactness in the topological sense is compactness in the metric sense, and the two notions of compact subset coincide), hence totally bounded (A compact metric space is complete and totally bounded, and neither implication uses any choice principle, Finite ε-net and totally bounded metric space, Open ball, closed ball and sphere in a metric space, Metric space: d(x,y)=0 iff x=y, symmetry, and the triangle inequality; pseudometric and ultrametric): for every n≥1 there is a finite Fn⊆fv[K] with fv[K]⊆⋃y∈FnB(y,1/n). Choosing one such finite net for each n, and enumerating each finite net, uses Countable Choice and finite choice only (The Axiom of Countable Choice (ACω), Every natural-number-indexed list of nonempty sets has a choice function on its family of values). Writing Aj:=fv−1[B(yj,1/n)]∖⋃i<jAi for an enumeration Fn={y0,…,ym} produces pairwise disjoint Borel sets covering K (A continuous map has Borel preimages of Borel sets, The Borel sigma-algebra of a topological space) and hence a measurable H-valued simple function tn=∑jyj1Aj with ∥tn(k)−fv(k)∥<1/n for every k; thus fv is the pointwise norm limit of simple functions, that is, strongly measurable (Strongly measurable Banach-valued function, Banach-valued simple function and integral). Moreover ∥χσ(k)‾π(k)v∥=∣χσ(k)∣ ∥v∥≤Mv for all k, where Mv:=(max⁡k∈K∣χσ(k)∣)∥v∥<∞: the continuous function ∣χσ∣ on the compact space K is bounded (A continuous image of a compact space is compact; a continuous real-valued map on a nonempty compact space attains a maximum and a minimum; and a continuous bijection from a compact space to a Hausdorff space is a homeomorphism), and μ is a probability. Hence ∫K∥fv∥ dμ≤Mv<∞, so fv is Bochner integrable by the Bochner integrability criterion (Bochner integrability criterion, Bochner-integrable function, Measure spaces).

The definition. With the Bochner integral just justified, put Pσv:=dσ∫Kfv(k) dμ(k)=dσ∫Kχσ(k)‾ π(k)v dμ(k)(v∈H). This defines a map Pσ:H→H, the σ-isotypic projection attached to the irreducible σ and the representation π; the norm inequality for Bochner integrals (Bochner integral norm inequality) gives ∥Pσv∥≤dσ∫K∣χσ∣ dμ ∥v∥ for every v. Only scalar functions are integrated directly in the construction: for each k the integrand is the vector χσ(k)‾π(k)v, and the choice of nets above is the only place Countable Choice is consumed.

The σ-isotypic subspace. A closed linear subspace M⊆H is σ-isotypic of type σ, or simply a σ-copy, when π(k)M=M for every k and there is a unitary intertwiner U:Vσ→M with Uσ(k)=π(k)U for every k; that is, when π∣M is unitarily equivalent to σ (Strongly continuous unitary representations, invariant linear subspaces and intertwiners, Linear subspace of a vector space). The σ-isotypic subspace of H is the closed linear span Hσ:=span⁡‾{ M⊆H:M is a σ-copy }, the smallest closed invariant subspace of H containing every σ-copy.

What is not asserted here. No sum over the unitary dual of K is formed, and no equality ∑σPσ=IH is claimed: completeness of the family of isotypic subspaces is the content of Peter--Weyl theory, which is not available at this point. All that is claimed about Pσ below is proved in Isotypic projections are mutually orthogonal equivariant projections ↗ without any density or dual-sum assertion: Pσ is a bounded self-adjoint idempotent commuting with π(K), its range is exactly Hσ, and distinct inequivalent types give orthogonal ranges. The boundedness, linearity and self-adjointness of Pσ are not part of the definition; they are theorems.

TheoremStatement: Literature-sourcedProof: AI-adaptedprecheck passjudge pass (gpt-6.1-sol)audited 2026-10-02Open item page →

Isotypic projections are mutually orthogonal equivariant projections

Statement

Assume the Axiom of Choice (The Axiom of Choice). Let K be a compact Hausdorff topological group (Topological group: multiplication and inversion are continuous, Open cover, subcover, and compact topological space; a compact subset is a subspace that is compact in its own right, Hausdorff space: distinct points have disjoint open neighbourhoods; every metrizable space is Hausdorff and the indiscrete topology on two points is not) with normalized Haar probability measure μ (Normalized Haar probability on a compact group). Let σ be an irreducible strongly continuous unitary representation of K of degree dσ, let π:K→U(H) be a strongly continuous unitary representation of K on a complex Hilbert space H (Strongly continuous unitary representations, invariant linear subspaces and intertwiners, Hilbert space), let Pσ:H→H be the σ-isotypic projection and Hσ the σ-isotypic subspace of Compact-group isotypic projection. Then:

  1. Pσ is a bounded linear operator, with ∥Pσv∥≤dσ(∫K∣χσ∣ dμ)∥v∥ for every v∈H;
  2. Pσ is self-adjoint: ⟨Pσv,w⟩=⟨v,Pσw⟩ for all v,w∈H;
  3. Pσ commutes with π(K): Pσπ(g)=π(g)Pσ for every g∈K;
  4. Pσ fixes every σ-copy: if M⊆H is a σ-copy, then Pσx=x for every x∈M;
  5. Pσ is idempotent, Pσ2=Pσ, and its range is exactly the σ-isotypic subspace: range⁡(Pσ)=Hσ;
  6. if τ is an irreducible strongly continuous unitary representation of K inequivalent to σ, with isotypic projection Pτ and isotypic subspace Hτ, then PσPτ=0=PτPσ and ⟨Pσv,Pτw⟩=0 for all v,w∈H.

No assertion is made that the sum of the operators Pσ over the unitary dual is IH; no completeness, density or Peter--Weyl statement is used.

Facts & Assumptions

Given: AC; a compact Hausdorff group K with normalized Haar probability μ; an irreducible strongly continuous unitary representation σ of K of degree dσ; a strongly continuous unitary representation π of K on a complex Hilbert space H; the isotypic projection Pσ, the character χσ, the σ-copies and the isotypic subspace Hσ of Compact-group isotypic projection.

[F1]

Well-definedness and norm bound: for each v the integrand fv(k)=χσ(k)‾π(k)v is continuous and Bochner integrable, Pσv=dσ∫Kfv dμ, and ∥Pσv∥≤dσ∫K∣χσ∣ dμ ∥v∥ (Compact-group isotypic projection, Bochner integral norm inequality, Hilbert space).

[F2]

Weak pairing formula: for all v,y∈H, ⟨Pσv,y⟩=dσ∫Kχσ(k)‾⟨π(k)v,y⟩ dμ(k), obtained by applying the theorem that bounded linear maps commute with Bochner integrals to the bounded functional x↦⟨x,y⟩ (Bounded linear maps commute with Bochner integration, Real and complex inner-product spaces and their induced length).

[F3]

The character is a continuous class function with χσ(k−1)=χσ(k)‾: the trace of a unitary endomorphism is the sum of its eigenvalues on the unit circle, and χσ(hkh−1)=χσ(k) (Compact-group isotypic projection, The basis-independent trace of an endomorphism of a finite-dimensional vector space, Similar matrices have the same trace); in particular ∣χσ∣ is bounded on the compact space K.

[F4]

Haar invariance: μ(K)=1; the maps k↦hk, k↦kh and k↦k−1 are measure preserving, so integrals of integrable functions are unchanged under these substitutions (Normalized Haar probability on a compact group, Measure-preserving transformations and systems, Integral invariance under measure-preserving maps, Measure spaces).

[F5]

The scalar Lebesgue integral is linear, so finite sums and scalar multiples integrate termwise and ∫f‾ dμ=∫f dμ‾ by the definition of the complex integral (The Lebesgue integral is linear on L1(μ), Integrable real and complex functions, and their integrals).

[F6]

Schur orthogonality (Schur orthogonality for general compact groups): for an orthonormal basis e1,…,edσ of the carrier of σ one has χσ(k)‾=∑i⟨σ(k)ei,ei⟩‾, (i)∫K⟨π(k)v,w⟩ ⟨ρ(k)v′,w′⟩‾ dμ(k)=0 for irreducible π,ρ that are not unitarily equivalent, and (ii)∫K⟨σ(k)a,b⟩ ⟨σ(k)ei,ei⟩‾ dμ(k)=dσ−1⟨a,ei⟩⟨b,ei⟩‾ for all a,b in the carrier of σ (Every finite-dimensional real or complex inner product space has an orthonormal basis, Bessel's inequality for a finite orthonormal list and Parseval's identity for an orthonormal basis).

[F7]

Schur's lemma: every bounded self-intertwiner of an irreducible strongly continuous unitary representation is a scalar multiple of the identity, and a nonzero bounded intertwiner between irreducible such representations makes them unitarily equivalent (Schur lemma for complex unitary representations, Linear isometries, and orthogonal or unitary operators on finite-dimensional inner product spaces).

[F8]

Bochner framework for the auxiliary averages: a continuous map g:K→H into the Banach space H has compact image and is attained as a pointwise norm limit of H-valued measurable simple functions built from finite nets, the selection of nets costing only Countable Choice, which the standing AC supplies; if ∥g(k)∥≤M for all k then ∫K∥g∥ dμ≤M<∞ and g is Bochner integrable with ∥∫Kg dμ∥≤M; every bounded linear map commutes with the Bochner integral, so for the bounded linear functional x↦⟨x,y⟩ one has ⟨∫Kg dμ,y⟩=∫K⟨g(k),y⟩ dμ(k) (Compact-group isotypic projection, Strongly measurable Banach-valued function, Banach-valued simple function and integral, Bochner-integrable function, Bochner integrability criterion, Bochner integral norm inequality, Bounded linear maps commute with Bochner integration, The Axiom of Countable Choice (ACω), AC supplies the countable and dependent choices used in Banach integration).

[F9]

Orthogonal projections: for a closed subspace M⊆H the orthogonal projection pM is linear and self-adjoint with pMx=x for x∈M; if M is invariant under a unitary representation, then so is M⊥ (The Hilbert orthogonal projection onto a closed subspace, Invariant orthogonal complements in unitary representations).

Proof

technique · direct
1.1F1given

If H={0} then Pσ=0, Hσ={0} and every assertion is immediate, so assume H≠{0} throughout; nothing below uses more than H≠{0} where a vector is exhibited.

1.2F1F2F5F10

Linearity and boundedness. Let v,v′∈H, a,b∈C and y∈H. By [F2], linearity of each π(k) and termwise integration [F5], ⟨Pσ(av+bv′),y⟩=dσ∫Kχσ(k)‾⟨π(k)(av+bv′),y⟩ dμ(k)=a⟨Pσv,y⟩+b⟨Pσv′,y⟩; a vector is determined by its pairings, so Pσ(av+bv′)=aPσv+bPσv′. Moreover ∣⟨Pσv,y⟩∣≤dσ∫K∣χσ(k)∣ ∣⟨π(k)v,y⟩∣ dμ(k)≤dσ(∫K∣χσ∣ dμ)∥v∥ ∥y∥ by Cauchy--Schwarz [F10]; applying this with y=Pσv when Pσv≠0 gives ∥Pσv∥≤dσ(∫K∣χσ∣ dμ)∥v∥. Thus Pσ is a bounded linear operator.

1.3F2F3F4F5

Self-adjointness. By [F2] and unitarity of π(k), ⟨Pσv,y⟩=dσ∫Kχσ(k)‾⟨v,π(k)−1y⟩ dμ(k); substituting k↦k−1, which preserves μ by [F4], and using χσ(k−1)‾=χσ(k) from [F3] turns this into dσ∫Kχσ(k)⟨v,π(k)y⟩ dμ(k)=dσ∫Kχσ(k)‾⟨π(k)y,v⟩ dμ(k)‾=⟨Pσy,v⟩‾=⟨v,Pσy⟩, the conjugation inside the scalar integral being justified by [F5]. Hence Pσ is self-adjoint.

1.4F2F3F4

Equivariance. Let g∈K. For v,y∈H, [F2] and the homomorphism property give ⟨Pσπ(g)v,y⟩=dσ∫Kχσ(k)‾⟨π(kg)v,y⟩ dμ(k); right translation u↦ug−1 is measure preserving by [F4], so substituting k=ug−1 this equals dσ∫Kχσ(ug−1)‾⟨π(u)v,y⟩ dμ(u), and since χσ is a class function the identity χσ(ug−1)=χσ(g−1u) of [F3] gives dσ∫Kχσ(g−1u)‾⟨π(u)v,y⟩ dμ(u); left translation u↦gu is measure preserving by [F4], so substituting u=gk gives dσ∫Kχσ(k)‾⟨π(gk)v,y⟩ dμ(k)=dσ∫Kχσ(k)‾⟨π(k)v,π(g)−1y⟩ dμ(k)=⟨Pσv,π(g)−1y⟩=⟨π(g)Pσv,y⟩, the second-to-last equality by unitarity of π(g). As y is arbitrary, Pσπ(g)=π(g)Pσ.

1.5F2F5F6F9

Fixing a single σ-copy. Let M⊆H be a σ-copy with unitary intertwiner U:Vσ→M, and let x∈M, y∈H. Writing pM for the orthogonal projection onto the closed subspace M [F9], the invariance of M gives π(k)x∈M, hence ⟨π(k)x,y⟩=⟨π(k)x,pMy⟩. For x=Ua and pMy=Ub with a,b∈Vσ, intertwining and unitarity of U give ⟨π(k)Ua,Ub⟩=⟨σ(k)a,b⟩, and the character expansion of [F6] together with Schur orthogonality (ii) yields ⟨Pσx,y⟩=dσ∑i∫K⟨σ(k)a,b⟩⟨σ(k)ei,ei⟩‾ dμ(k)=dσ∑idσ−1⟨a,ei⟩⟨b,ei⟩‾=∑i⟨a,ei⟩⟨b,ei⟩‾=⟨a,b⟩=⟨Ua,Ub⟩=⟨x,pMy⟩=⟨x,y⟩. Hence Pσx=x for every x in any σ-copy.

1.6F2F6F9

Killing inequivalent copies. Let τ be an irreducible strongly continuous unitary representation of K inequivalent to σ, let M⊆H be a τ-copy with unitary intertwiner U:Vτ→M, and let x=Ua∈M, y∈H with pMy=Ub. Then π(k)x∈M, so ⟨π(k)x,y⟩=⟨τ(k)a,b⟩, and [F2], [F6] with Schur orthogonality (i) applied to the inequivalent irreducibles τ and σ give ⟨Pσx,y⟩=dσ∑i∫K⟨τ(k)a,b⟩⟨σ(k)ei,ei⟩‾ dμ(k)=0. Hence Pσx=0 for every x in any τ-copy.

1.7F2F3F4F5F7F8F10

The range lies in the isotypic subspace. Fix v∈H and, for each i, define Ai(w):=∫K⟨w,σ(k)ei⟩ π(k)v dμ(k) for w∈Vσ. The integrand gi,w(k):=⟨w,σ(k)ei⟩π(k)v is continuous and satisfies ∥gi,w(k)∥≤∥w∥ ∥v∥ for every k, because σ(k)ei has norm 1 and π(k)v has norm ∥v∥, so gi,w is Bochner integrable and ∥Ai(w)∥≤∥w∥ ∥v∥ by [F8]; moreover, since the bounded linear functional x↦⟨x,y⟩ commutes with the Bochner integral, ⟨Ai(w),y⟩=∫K⟨w,σ(k)ei⟩⟨π(k)v,y⟩ dμ(k) for every y∈H. The map Ai is linear: the scalar integrand in this pairing is linear in w and the scalar integral is linear [F5], so ⟨Ai(aw+bw′),y⟩=a⟨Ai(w),y⟩+b⟨Ai(w′),y⟩ for all y, and a vector is determined by its pairings. For h∈K and y∈H, unitarity of π(h) together with the pairing formula gives ⟨π(h)Ai(w),y⟩=⟨Ai(w),π(h)−1y⟩=∫K⟨w,σ(k)ei⟩⟨π(hk)v,y⟩ dμ(k), and the substitution k↦h−1k, measure preserving by [F4], turns this into ∫K⟨w,σ(h−1k)ei⟩⟨π(k)v,y⟩ dμ(k)=∫K⟨σ(h)w,σ(k)ei⟩⟨π(k)v,y⟩ dμ(k)=⟨Ai(σ(h)w),y⟩; hence π(h)Ai=Aiσ(h) and each Ai is a bounded intertwiner. If Ai≠0 then ker⁡Ai is a σ-invariant subspace, because Aiσ(h)=π(h)Ai for all h, and ker⁡Ai≠Vσ; by irreducibility ker⁡Ai={0}, so Ai is injective, and its image Mi=Ai(Vσ) is finite dimensional (hence closed [F10]) and π(K)-invariant, because π(h)Mi=Aiσ(h)Vσ⊆Mi; a closed invariant subspace N⊆Mi pulls back under the injective intertwiner Ai to the σ-invariant subspace Ai−1(N) of the irreducible Vσ, which is {0} or Vσ, so π∣Mi is irreducible and the nonzero bounded intertwiner Ai makes π∣Mi unitarily equivalent to σ by Schur's lemma [F7]; thus Mi is a σ-copy and Ai(ei)∈Hσ, while if Ai=0 then Ai(ei)=0∈Hσ. Finally, for every y, summing the pairing formula over i and using linearity of the scalar integral [F5] and the trace formula χσ(k)=∑i⟨σ(k)ei,ei⟩ of [F3] together with conjugate symmetry ⟨ei,σ(k)ei⟩=⟨σ(k)ei,ei⟩‾ gives ⟨dσ∑iAi(ei),y⟩=dσ∫Kχσ(k)‾⟨π(k)v,y⟩ dμ(k)=⟨Pσv,y⟩ by [F2]; hence Pσv=dσ∑iAi(ei) lies in Hσ.

2.1F1F10step 1.2step 1.5step 1.7

Fixing the isotypic subspace and identifying the range. By step 1.5, Pσ fixes every vector lying in a σ-copy; by linearity (step 1.2) it fixes the linear span of all σ-copies, and since it is bounded, hence continuous, it fixes the closure of that span: Pσx=x for every x∈Hσ. Combined with step 1.7, for every v∈H the vector Pσv lies in Hσ and is therefore fixed, so Pσ2=Pσ. Hence range⁡(Pσ)⊆Hσ by step 1.7 and Hσ⊆range⁡(Pσ) because x=Pσx for x∈Hσ; the range is exactly Hσ.

3.1F6step 1.3step 1.6step 2.1

Distinct inequivalent types. Let τ be irreducible and inequivalent to σ with isotypic projection Pτ and isotypic subspace Hτ. Repeating steps 1.6, 1.7 and 2.1 with τ in place of σ shows Pσ vanishes on every τ-copy, hence by linearity and continuity on Hτ, and that range⁡(Pτ)=Hτ; therefore PσPτ=0. Exchanging the roles of σ and τ gives PτPσ=0. Finally, for v,w∈H, self-adjointness (step 1.3) and idempotence (step 2.1) give ⟨Pσv,Pτw⟩=⟨Pσ2v,Pτw⟩=⟨Pσv,PσPτw⟩=0, so the ranges of Pσ and Pτ are orthogonal.

4.1step 1.2step 1.3step 1.4step 1.5step 2.1step 3.1∎

Collecting steps 1.2, 1.3, 1.4, 1.5, 2.1 and 3.1: Pσ is a bounded self-adjoint idempotent commuting with π(K), fixes every σ-copy, has range exactly the σ-isotypic subspace Hσ, and the projections of inequivalent irreducibles multiply to zero with orthogonal ranges. At no point is any sum over the unitary dual asserted, and no density or completeness statement is used.

5 · Examples, counterexamples and false statements

None yet.

Sources