07.07.07 · representation-theory / compact-lie

Mackey theory of induced representations and systems of imprimitivity

shipped3 tiersLean: none

Anchor (Master): Mackey, *Imprimitivity for representations of locally compact groups I*, Proc. Natl. Acad. Sci. USA 35 (1949) 537–545; Mackey, *Induced representations of locally compact groups I*, Ann. Math. 55 (1952) 101–139; Mackey, *Induced Representations of Groups and Quantum Mechanics* (Benjamin 1968); Folland *A Course in Abstract Harmonic Analysis* (CRC 1995) Ch. 6

Intuition Beginner

Suppose you understand a small group well — its symmetries, its representations, how it acts on states. The small group sits inside a larger one. Mackey's question is: can you build representations of the big group cheaply, out of what you already know about the small one? The answer is yes, and the recipe is called induction. You spread the small-group representation out over all the ways the big group can move the small one around, and the big group acts by shuffling those copies. The Poincaré-group story of the previous unit is one case of this single recipe.

Why bother packaging it as a general machine? Because the same recipe keeps reappearing. A free non-relativistic particle, a relativistic particle, a particle in a periodic crystal, an atom in a magnetic field — each is a representation of some symmetry group, and in each case the group is built by gluing a simple piece (translations, say) to a piece that rotates it. Once you see the pattern, you stop solving each problem from scratch. You read off the answer from two ingredients: which orbit the momentum sits on, and what the leftover symmetry does to it.

The everyday picture is a carousel of identical seats. Each seat carries the same little gadget — that is your small-group representation. The carousel turns the seats around a circle; that turning is the big group permuting copies. To know the whole system you only need one seat's gadget plus the rule for how the carousel cycles the seats. Mackey's theorem says every representation built this way is captured by exactly that data, and a clean test tells you when a given representation is one of them. This idea builds toward the entire particle table.

Visual Beginner

A horizontal circle stands for the coset space $G / H$ , the "seats on the carousel." Above each point floats a small box, all boxes identical, each holding one copy of the subgroup representation. The large group acts in two coordinated ways: it slides the points around the circle (permuting which box is where), and inside each box it twists the contents by an element of the small group. A second panel shows the same data as a measuring device: a projection asking "which region of the circle is the state supported on?", drawn as a shaded arc, with an arrow showing that moving the state by a group element shifts the arc accordingly.

The picture captures the two halves of the imprimitivity theorem: an induced representation always comes with a "where is it supported" measurement that moves correctly under the group, and conversely any representation carrying such a measurement is induced.

Worked example Beginner

Build a representation of the Euclidean group of the plane and watch the recipe produce a familiar object: a wave of fixed wavelength. Use the group of rigid motions of the plane, made of translations and rotations. We start from a single translation character — a plane wave of one fixed momentum — and induce.

Step 1. Fix a momentum vector of length $5$ , say pointing along the first axis: $k = (5, 0)$ . The translation by a vector $a = (a_{1}, a_{2})$ multiplies the wave by the phase $cos (5 a_{1}) +$ a matching sine part; the size is always $1$ , so this is a one-number "character" of the translations.

Step 2. Rotations move the momentum around a circle. Rotate $k = (5, 0)$ by a quarter turn and you get $(0, 5)$ ; by a half turn, $(- 5, 0)$ . Every rotation keeps the length at $5$ . So the set of momenta the recipe visits is the full circle of radius $5$ — this circle is one orbit.

Step 3. Ask which rotations leave $k = (5, 0)$ exactly fixed. Only the identity rotation does, since any nonzero rotation moves the arrow off itself. So the leftover symmetry group here is the smallest possible: just the identity. That leftover group is the little group.

Step 4. Assemble. The induced representation lives on functions on the circle of radius $5$ . A translation multiplies by the position-dependent phase; a rotation slides the function around the circle. Pick the function concentrated near $(5, 0)$ and rotate by a quarter turn: it moves to sit near $(0, 5)$ . The whole representation is the bundle of plane waves of wavelength fixed by the radius $5$ .

What this tells us: starting from one plane wave and a circle of radius $5$ , the recipe rebuilds the space of all states of a free particle with that fixed energy. The orbit (the circle) sets the energy; the little group (here only the identity) tells you there is no extra internal label. Change the radius and you change the energy; the machine is the same.

Check your understanding Beginner

Exercise (easy, multiple choice).

In Mackey's recipe, the "little group" attached to a chosen momentum is:

A. The whole big group
B. The subgroup of symmetries that leave that chosen momentum exactly fixed
C. The set of all momenta of the same length
D. The space of cosets

Hint

It is the leftover symmetry once you have used up the freedom to move the momentum to a standard position — the symmetries that do nothing to the chosen momentum.

Answer

B. The subgroup of symmetries that leave that chosen momentum exactly fixed. The little group is the stabiliser of the standard momentum. The set of all momenta of the same length (option C) is the orbit, a different object; the cosets (option D) parametrise the orbit. The orbit and the little group together are the two ingredients of the classification.

Formal definition Intermediate+

Let $G$ be a second-countable locally compact group and $H \leq G$ a closed subgroup. The homogeneous space $G / H$ carries a $G$ -quasi-invariant Borel measure $μ$ : for each $g$ the translate $g_{*} μ$ has the same null sets as $μ$ , with a positive Radon–Nikodym cocycle $$ \lambda(g, \dot x) = \frac{d(g_*\mu)}{d\mu}(\dot x), \qquad \lambda(g_1 g_2, \dot x) = \lambda(g_1, g_2 \dot x),\lambda(g_2, \dot x). $$ Given a unitary representation $σ : H \to U (V_{σ})$ , the induced representation $Ind_{H}^{G} (σ)$ acts on the Hilbert space $H_{σ}$ of (equivalence classes of) measurable functions $f : G \to V_{σ}$ satisfying the covariance relation $f (g h) = σ (h)^{- 1} f (g)$ for $h \in H$ and $\int_{G / H} ∥ f (g) ∥^{2} d μ (\overset{g}{˙}) < \infty$ . The action is $$ \bigl(\mathrm{Ind}H^G(\sigma)(g_0),f\bigr)(g) = \lambda(g_0^{-1}, \dot g)^{1/2}, f(g_0^{-1} g), $$ the square root of the cocycle restoring unitarity. Equivalently, after choosing a Borel section $s : G / H \to G$ , $\mathcal{H}\sigma \cong L^2(G/H, \mu; V_\sigma) $an d t h e a c t i o n c a r r i es a$ \sigma $- v a l u e d m u l t i pl i er b u i l t f r o m$ s$.

A system of imprimitivity for $G$ based on $G / H$ is a pair $(U, P)$ , where $U : G \to U (H)$ is a unitary representation and $P$ is a projection-valued measure on the Borel sets of $G / H$ valued in projections on $H$ , satisfying the covariance relation $$ U(g), P(E), U(g)^{-1} = P(g \cdot E) $$ for every $g \in G$ and every Borel set $E \subseteq G / H$ . The projection $P (E)$ answers the measurement "is the state supported over the region $E$ of $G / H$ ?", and covariance says moving the state by $g$ moves the region by $g$ . Two systems of imprimitivity are equivalent when a single unitary intertwines both the representations and the projection-valued measures.

The induced representation is canonically a system of imprimitivity: on $L^{2} (G / H, μ; V_{σ})$ the projection-valued measure $P_{0} (E) =$ multiplication by the indicator $1_{E}$ satisfies the covariance relation with $U = Ind_{H}^{G} (σ)$ by inspection. The content of Mackey's theorem is the converse.

For the semidirect-product specialisation take $G = A ⋊ H$ with $A$ locally compact abelian and normal. The Pontryagin dual $A$ carries the dual action $(h \cdot χ) (a) = χ (h^{- 1} a)$ . The action is regular when the orbits of $H$ on $A$ are locally closed and a Borel cross-section for the orbit space exists. For a character $χ \in A$ , the little group is the stabiliser $H_{χ} = {h \in H : h \cdot χ = χ}$ , and the isotropy subgroup is $A ⋊ H_{χ}$ . The character $χ$ extends to $A ⋊ H_{χ}$ by $χ (a, h) = χ (a)$ on the abelian factor.

Counterexamples to common slips

A system of imprimitivity is more than a representation plus a commuting algebra of multiplications: the covariance relation $U (g) P (E) U (g)^{- 1} = P (g E)$ ties the two together. Dropping covariance leaves a meaningless pairing; the entire force of the theorem lives in that compatibility.
The base space must be the homogeneous space $G / H$ , not an arbitrary $G$ -space. The imprimitivity theorem characterises induction from $H$ precisely because $G / H$ is transitive; over a non-transitive $G$ -space one obtains a direct integral of induced pieces, one per orbit.
Regularity is a genuine hypothesis, not bookkeeping. For non-regular (non-type-I) groups the orbit space is not a standard Borel space, distinct orbits cannot be separated by invariant Borel sets, and the parametrisation of irreducibles by (orbit, little-group irrep) fails. The Mackey machine is a theorem about regular semidirect products.
The little group is the stabiliser of a character of $A$ , an element of $A$ , not of an element of $A$ itself. Confusing $A$ with $A$ inverts the orbit geometry: for $A = R^{n}$ the two are isomorphic, but the dual action carries the inverse transpose, and momentum lives in $A$ .

Key theorem with proof Intermediate+

Theorem (Mackey's imprimitivity theorem; Mackey 1949, 1952). Let $G$ be a second-countable locally compact group, $H \leq G$ a closed subgroup, and $G / H$ the homogeneous space with its quasi-invariant measure class. A unitary representation $U$ of $G$ on a separable Hilbert space $H$ admits a system of imprimitivity $(U, P)$ based on $G / H$ if and only if $U$ is unitarily equivalent to an induced representation $Ind_{H}^{G} (σ)$ for some unitary representation $σ$ of $H$ . The representation $σ$ is determined up to unitary equivalence by $(U, P)$ , and the equivalence is functorial: equivalent systems of imprimitivity correspond to equivalent inducing data. ^{[Mackey 1949]}

Proof. The forward direction was recorded above: $Ind_{H}^{G} (σ)$ carries the canonical projection-valued measure $P_{0} (E) = M_{1_{E}}$ by multiplication, and the covariance relation is the identity $g \cdot 1_{E} = 1_{g E}$ transported through the unitary action. The substance is the converse.

Let $(U, P)$ be a system of imprimitivity based on $G / H$ . The projection-valued measure $P$ on the standard Borel space $G / H$ diagonalises $H$ as a direct integral $$ \mathcal{H} \cong \int^\oplus_{G/H} \mathcal{H}_{\dot x}, d\mu(\dot x), $$ a measurable field of Hilbert spaces, with $P (E)$ acting as multiplication by $1_{E}$ . The commutative von Neumann algebra generated by ${P (E)}$ is maximal abelian among the algebras it is required to commute with, so this direct-integral decomposition is the spectral resolution of $P$ , unique up to the measure class.

Covariance forces $U (g)$ to map the fibre over $\overset{x}{˙}$ to the fibre over $g \overset{x}{˙}$ : for $f$ supported in $E$ (so $P (E) f = f$ ), the relation $P (g E) U (g) = U (g) P (E)$ gives $P (g E) (U (g) f) = U (g) f$ , hence $U (g) f$ is supported in $g E$ . Thus $U (g)$ is a measurable field of unitary maps $u (g) : H_{g^{- 1} \overset{x}{˙}} \to H_{\overset{x}{˙}}$ , weighted by the cocycle $λ (g, \cdot)^{1/2}$ for unitarity against the quasi-invariant $μ$ .

Flatten the field over a Borel section $s : G / H \to G$ , identifying each fibre $H_{\overset{x}{˙}}$ with a fixed model space $V$ via $s (\overset{x}{˙})$ . Transport of $u (g)$ to the model space produces, for each pair $(g, \overset{x}{˙})$ , the element $$ \sigma_0(g, \dot x) = s(\dot x)^{-1}, g, s(g^{-1} \dot x) \in H, $$ which lies in $H$ because it fixes the basepoint coset: $s (\overset{x}{˙})^{- 1} g s (g^{- 1} \overset{x}{˙})$ sends $H \in G / H$ to itself. The representation property $U (g_{1}) U (g_{2}) = U (g_{1} g_{2})$ becomes the cocycle identity $σ_{0} (g_{1} g_{2}, \overset{x}{˙}) = σ_{0} (g_{1}, \overset{x}{˙}) σ_{0} (g_{2}, g_{1}^{- 1} \overset{x}{˙})$ , so the fibrewise unitaries assemble into a single homomorphism $σ : H \to U (V)$ once evaluated at the basepoint, independent of the section up to equivalence.

The resulting data — the field $V$ over $G / H$ , the cocycle $σ_{0}$ , the cocycle weight $λ^{1/2}$ — is by construction the induced representation $Ind_{H}^{G} (σ)$ written in the section flattening, with its canonical projection-valued measure recovered as the multiplication algebra. Uniqueness of $σ$ follows because the fibre $V$ with its $H$ -action is the restriction of $U$ to the isotropy data at the basepoint, an invariant of $(U, P)$ . A different section changes $σ_{0}$ by a coboundary and leaves $σ$ in the same equivalence class. $□$

Bridge. The imprimitivity theorem builds toward every orbit-and-little-group classification, and the foundational reason is that it converts an external datum — a representation of a possibly enormous group — into the internal datum of a single representation of a subgroup plus a geometric measurement on $G / H$ . This is exactly the move that powers the Wigner classification of 07.07.06: there the projection-valued measure is the joint spectral measure of the momentum operators, the base $G / H$ is a mass shell, and the recovered $σ$ is the spin or helicity representation of the little group. The semidirect-product specialisation that follows generalises that single case: it identifies the dual of the abelian factor with momentum space and identifies $G / H$ with a Lorentz orbit, so the bridge is the statement that diagonalising the abelian factor is choosing a system of imprimitivity. The central insight is that induction and imprimitivity are two faces of one adjunction — induction is left to restriction by Frobenius reciprocity 07.01.08 in the finite case, and the projection-valued measure is the locally compact replacement for the coset-permutation bookkeeping. Putting these together, the non-relativistic free particle of 09.03.03 and the relativistic particle of 07.07.06 are the same theorem read on two different abelian normal factors, and the little-group method appears again in the crystallographic and magnetic-translation settings.

Exercises Intermediate+

Exercise 3 (medium, symbolic).

Verify the Radon–Nikodym cocycle identity $λ (g_{1} g_{2}, \overset{x}{˙}) = λ (g_{1}, g_{2} \overset{x}{˙}) λ (g_{2}, \overset{x}{˙})$ from $λ (g, \overset{x}{˙}) = \frac{d ( g _{*} μ )}{d μ} (\overset{x}{˙})$ , where $(g_{*} μ) (E) = μ (g^{- 1} E)$ .

Hint

Use the chain rule for Radon–Nikodym derivatives, $\frac{d ( ν _{1} )}{d μ} = \frac{d ν _{1}}{d ν _{2}} \frac{d ν _{2}}{d μ}$ , applied to the chain $μ \to (g_{2})_{*} μ \to (g_{1} g_{2})_{*} μ$ .

Answer

Write the pushforward composition $(g_{1} g_{2})_{*} μ = (g_{1})_{*} ((g_{2})_{*} μ)$ . The Radon–Nikodym derivative factors by the chain rule: $$ \lambda(g_1 g_2, \dot x) = \frac{d((g_1 g_2)* \mu)}{d\mu}(\dot x) = \frac{d((g_1)((g_2)_\mu))}{d((g_2)*\mu)}(\dot x)\cdot \frac{d((g_2)\mu)}{d\mu}(\dot x). $$ The second factor is $λ (g_{2}, \overset{x}{˙})$ . The first factor is $λ (g_{1}, \cdot)$ evaluated after the change of base point produced by $g_{2}$ : since $\frac{d((g_1)_\nu)}{d\nu}(\dot x) $i s$ \frac{d((g_1)_*\mu)}{d\mu} $t r an s p or t e d, i t e q u a l s$ \lambda(g_1, g_2^{-1}\cdot) $in o n eco n v e n t i o n; w i t h t h e a c t i o n co n v e n t i o n u se d h er e i t r e a d s$ \lambda(g_1, g_2 \dot x) $. H e n ce$ \lambda(g_1 g_2, \dot x) = \lambda(g_1, g_2 \dot x),\lambda(g_2, \dot x)$. Rubric: full credit for the chain-rule factorisation and identification of the two factors.

Exercise 4 (medium, symbolic).

For $G = A ⋊ H$ with $A$ abelian, show that the dual action $(h \cdot χ) (a) = χ (h^{- 1} a)$ is a genuine left action of $H$ on $A$ , i.e. $(h_{1} h_{2}) \cdot χ = h_{1} \cdot (h_{2} \cdot χ)$ .

Hint

Apply both sides to an arbitrary $a \in A$ and use that $H$ acts on $A$ by automorphisms.

Answer

Apply both sides to $a \in A$ . The right side: $(h_{1} \cdot (h_{2} \cdot χ)) (a) = (h_{2} \cdot χ) (h_{1}^{- 1} a) = χ (h_{2}^{- 1} (h_{1}^{- 1} a)) = χ ((h_{2}^{- 1} h_{1}^{- 1}) a)$ , using that $H$ acts on $A$ by automorphisms so the actions compose. The left side: $((h_{1} h_{2}) \cdot χ) (a) = χ ((h_{1} h_{2})^{- 1} a) = χ ((h_{2}^{- 1} h_{1}^{- 1}) a)$ . The two agree for all $a$ , so $(h_{1} h_{2}) \cdot χ = h_{1} \cdot (h_{2} \cdot χ)$ . The inverse in the definition is exactly what makes this a left rather than a right action. Rubric: full credit for evaluating on $a$ and matching the composed automorphism.

Exercise 5 (medium, short-answer).

Explain why the regularity hypothesis (orbits locally closed, Borel cross-section exists) cannot be dropped, by describing what goes wrong with the parametrisation of irreducibles when the orbit space is not a standard Borel space.

Hint

The classification says irreducibles correspond to pairs (orbit, little-group irrep). What must be true of the orbit space for "choose an orbit" to be a measurable, separating operation?

Answer

The Mackey machine asserts a bijection between irreducible unitary representations of $G = A ⋊ H$ and pairs consisting of an $H$ -orbit $O \subseteq A$ and an irreducible representation of the little group of a standard point. For this to be a clean parametrisation, the orbit space $A / H$ must be a standard Borel space, so that distinct orbits are separated by invariant Borel sets and a measurable choice of standard point exists. When the action is non-regular — orbits not locally closed, the orbit space not countably separated — the quotient Borel structure collapses: invariant Borel sets cannot separate orbits, the direct-integral decomposition of a representation need not concentrate on a single orbit, and there exist irreducibles whose spectral measure is supported on a whole non-closed orbit class with no standard point to induce from. The group is then non-type-I and admits irreducibles not of the form $Ind (χ \otimes σ)$ . Rubric: full credit for naming the standard-Borel requirement, orbit separation, and the failure of single-orbit concentration / type-I.

Exercise 7 (hard, short-answer).

State the Mackey machine for a regular semidirect product $G = A ⋊ H$ ( $A$ abelian) and prove that the induced representation $Ind_{A ⋊ H_{χ}}^{G} (χ \otimes σ)$ restricts on $A$ to a representation whose spectral measure is supported on the orbit $O = H \cdot χ$ .

Hint

On the model space $L^{2} (O, μ; V_{σ})$ , identify how $A$ acts by characters, and read off the support of the resulting projection-valued measure on $A$ .

Answer

Mackey machine: for $A$ abelian and the dual action regular, every irreducible unitary representation of $G = A ⋊ H$ is equivalent to $U^{(O, σ)} = Ind_{A ⋊ H_{χ}}^{G} (χ \otimes σ)$ for an orbit $O = H \cdot χ$ and an irreducible $σ$ of the little group $H_{χ}$ ; inequivalent pairs give inequivalent irreducibles. Support claim: the induced space is realised as $L^{2} (O, μ; V_{σ})$ , with $O ≅ H / H_{χ}$ identified with the orbit through the boost section. For $a \in A$ and $ψ \in L^{2} (O, μ; V_{σ})$ , $$ \bigl(U^{(\mathcal{O},\sigma)}(a)\psi\bigr)(\chi') = \chi'(a),\psi(\chi'), \qquad \chi' \in \mathcal{O}, $$ because the extension $χ \otimes σ$ acts on the $A$ -factor by the character of the point and induction transports this fibrewise across the orbit. Thus $U ∣_{A} = \int_{O}^{\oplus} χ^{'} d μ (χ^{'})$ , a multiplication representation, whose projection-valued measure on $A$ is $P (E) = M_{1_{E \cap O}}$ . Its support is $O$ . This is the system of imprimitivity that the imprimitivity theorem recognises, with base $A \supseteq O ≅ H / H_{χ}$ . Rubric: full credit for the statement plus the fibrewise character action establishing support on $O$ .

Exercise 8 (hard, short-answer).

Prove the induction-in-stages identity $Ind_{K}^{G} (σ) ≅ Ind_{H}^{G} (Ind_{K}^{H} (σ))$ for closed subgroups $K \leq H \leq G$ at the level of the defining covariant-function model, and explain why it is the locally compact form of transitivity of induction.

Hint

Both sides are spaces of functions on $G$ with covariance conditions; exhibit a measure-preserving bijection between the function spaces that intertwines the left-translation actions.

Answer

The left side $Ind_{K}^{G} (σ)$ is functions $f : G \to V_{σ}$ with $f (g k) = σ (k)^{- 1} f (g)$ for $k \in K$ , square-integrable over $G / K$ . The right side is functions $F : G \to H_{Ind_{K}^{H} σ}$ with $F (g h) = (Ind_{K}^{H} (σ) (h))^{- 1} F (g)$ for $h \in H$ , square-integrable over $G / H$ ; here $H_{Ind_{K}^{H} σ}$ is itself functions $φ : H \to V_{σ}$ with $φ (hk) = σ (k)^{- 1} φ (h)$ . Define $Φ (f) (g) (h) = f (g h)$ for $g \in G, h \in H$ . Check the inner covariance: $Φ (f) (g) (hk) = f (g hk) = σ (k)^{- 1} f (g h) = σ (k)^{- 1} Φ (f) (g) (h)$ , so $Φ (f) (g) \in H_{Ind_{K}^{H} σ}$ . Check the outer covariance: $Φ (f) (g h^{'}) (h) = f (g h^{'} h) = Φ (f) (g) (h^{'} h) = (Ind_{K}^{H} (σ) (h^{'})^{- 1} Φ (f) (g)) (h)$ , matching the required transformation. The Fubini-type factorisation $G / K \to G / H \to H / K$ of the quasi-invariant measures makes $Φ$ unitary, and it manifestly intertwines left translation, so it is the desired equivalence. This is transitivity of induction: inducing through the intermediate group $H$ in two steps agrees with inducing directly from $K$ , the locally compact analogue of the finite-index identity $Ind_{K}^{G} = Ind_{H}^{G} \circ Ind_{K}^{H}$ . Rubric: full credit for the map $Φ$ , both covariance checks, and the measure factorisation.

Lean formalization Intermediate+

Mathlib has Haar measure, $L^{p}$ spaces, and the semidirect product, but no induced representations of locally compact groups and no imprimitivity theorem. The intended formalisation reads schematically:

import Mathlib.MeasureTheory.Group.Measure
import Mathlib.MeasureTheory.Function.LpSpace
import Mathlib.GroupTheory.SemidirectProduct

variable {G H : Type*} [Group G] [TopologicalSpace G] [LocallyCompactSpace G]
variable [Group H] [TopologicalSpace H]

/-- A system of imprimitivity for `U : G → U(ℋ)` based on `G/H`:
    a projection-valued measure `P` on `G/H` with the covariance
    `U g • P E • (U g)⁻¹ = P (g • E)`. -/
structure SystemOfImprimitivity
    (U : G → (ℋ →L[ℂ] ℋ)) (P : Set (G ⧸ H) → (ℋ →L[ℂ] ℋ)) : Prop where
  isProjectionValued : sorry
  covariance : ∀ (g : G) (E : Set (G ⧸ H)), U g * P E * (U g)⁻¹ = P (g • E)

/-- The unitary induced representation `Ind_H^G σ` on the L² sections
    of the associated bundle over `G/H` with quasi-invariant measure. -/
noncomputable def inducedRep (σ : H → (V →L[ℂ] V)) :
    G → ((G ⧸ H → V) →L[ℂ] (G ⧸ H → V)) :=
  sorry  -- covariant functions with Radon–Nikodym cocycle weight

/-- Mackey's imprimitivity theorem: `U` carries a system of imprimitivity
    based on `G/H` iff it is induced from a representation of `H`. -/
theorem mackey_imprimitivity (U : G → (ℋ →L[ℂ] ℋ)) :
    (∃ P, SystemOfImprimitivity U P) ↔
      ∃ (σ : H → (V →L[ℂ] V)), U ≃ᵤ inducedRep σ :=
  sorry  -- the converse direction is the content

The proof gap is large. Mathlib needs the quasi-invariant measure on $G / H$ with its Radon–Nikodym cocycle, the unitary structure on the space of $L^{2}$ covariant sections, projection-valued measures over a standard Borel space, the direct-integral decomposition a projection-valued measure induces, and the regularity hypothesis for the semidirect-product specialisation. Each is reachable from the extant Haar and $L^{p}$ infrastructure together with a Pontryagin-duality completion, but the imprimitivity theorem itself has no Mathlib analogue.

Advanced results Master

Theorem (Mackey machine for regular semidirect products; Mackey 1949–58). Let $G = A ⋊ H$ with $A$ locally compact abelian and the dual action of $H$ on $A$ regular. For an orbit $O = H \cdot χ \subseteq A$ with little group $L = H_{χ}$ and an irreducible unitary representation $σ$ of $L$ , the representation $U^{(O, σ)} = Ind_{A ⋊ L}^{G} (χ \otimes σ)$ is irreducible. Every irreducible unitary representation of $G$ is equivalent to some $U^{(O, σ)}$ , and $U^{(O, σ)} ≅ U^{(O^{'}, σ^{'})}$ if and only if $O = O^{'}$ and $σ ≅ σ^{'}$ (after conjugating standard points). ^{[Mackey 1952]}

The proof is the imprimitivity theorem applied twice. Restricting an irreducible $U$ to the abelian $A$ produces, by the SNAG theorem, a projection-valued measure on $A$ ; the semidirect relation makes it covariant for the $H$ -action, hence a system of imprimitivity based on $A$ . Irreducibility forces the spectral measure onto a single orbit $O ≅ H / L$ , reducing the base to a homogeneous space, and the imprimitivity theorem then realises $U$ as induced from the isotropy subgroup $A ⋊ L$ . On $A ⋊ L$ the character $χ$ is $L$ -fixed, so the inducing representation factors as $χ \otimes σ$ , and irreducibility of $U$ is equivalent to irreducibility of $σ$ by a commutant computation. Folland Ch. 6 §6.6 gives the complete account. ^{[Folland 1995]}

Theorem (Mackey's irreducibility and intertwining criteria; Mackey 1952). For a closed subgroup $H \leq G$ and representations $σ, τ$ of $H$ , the intertwining space of $Ind_{H}^{G} (σ)$ and $Ind_{H}^{G} (τ)$ is computed from the double-coset space $H \ G / H$ : each double coset $H g H$ contributes intertwiners built from the $H \cap g H g^{- 1}$ -intertwiners of $σ$ and the $g$ -conjugate of $τ$ . In particular $Ind_{H}^{G} (σ)$ is irreducible iff $σ$ is irreducible and, for every $g \in / H$ , the representations $σ$ and $σ^{g}$ of $H \cap g H g^{- 1}$ are disjoint.

This is the Mackey subgroup theorem and the Mackey irreducibility criterion. For the semidirect product it specialises: the double cosets reduce to the orbit structure, the disjointness condition is automatic away from the little group, and the criterion collapses to "irreducible little-group representation gives irreducible induced representation." The finite-group ancestor is the Mackey formula for the restriction-of-induction $Res_{K} Ind_{H}^{G} (σ) ≅ ⨁_{K g H} Ind_{K \cap g H g^{- 1}}^{K} (σ^{g})$ , itself a refinement of Frobenius reciprocity 07.01.08.

Theorem (the $E (2)$ and $E (3)$ free particles via the machine). Let $G = R^{n} ⋊ SO (n)$ be the Euclidean group. The orbits of $SO (n)$ on $R^{n} ≅ R^{n}$ are the spheres of radius $r \geq 0$ . For $r > 0$ the little group is $SO (n - 1)$ ; the irreducible representations $U^{(r, σ)}$ , with $σ$ an irrep of $SO (n - 1)$ , are the states of a free particle of fixed energy $\propto r^{2}$ and internal symmetry type $σ$ . The orbit $r = 0$ has little group $SO (n)$ itself, giving the zero-momentum representations.

For $n = 3$ this is the content of 09.03.03: the spheres are the energy shells, the little group $SO (2)$ supplies the helicity-type label, and the induced representation acts on $L^{2}$ of the sphere. The Euclidean and Poincaré cases are the same theorem on two different normal factors — $R^{3}$ with rotational orbits, $R^{1, 3}$ with Lorentzian orbits — which is why the relativistic classification was obtainable before the general machine existed.

Theorem (cocycle / projective refinement). When the inducing data is twisted by a Borel $2$ -cocycle $ω \in Z^{2} (H, T)$ , the imprimitivity theorem holds verbatim for $ω$ -representations: $ω$ -systems of imprimitivity based on $G / H$ correspond to $ω$ -induced representations. For $G = A ⋊ H$ this produces the projective irreducibles of $G$ , which for $A = R^{2 n}$ and $H$ a symplectic group recover the metaplectic representation.

The cocycle refinement is what makes the imprimitivity theorem the correct framework for quantum mechanics, where Wigner's theorem already forces projective representations. Mackey's Induced Representations of Groups and Quantum Mechanics develops exactly this reading: the canonical commutation relations are a system of imprimitivity for $R^{n} ⋊ R^{n}$ based on configuration space, and the Stone–von Neumann uniqueness theorem is the imprimitivity theorem for that pair. ^{[Mackey 1968]}

Synthesis. The imprimitivity theorem is the foundational reason that every orbit-and-little-group classification works, and the central insight is that diagonalising an abelian normal factor is equipping a representation with a system of imprimitivity over the dual group. Putting these together, the Mackey machine reads off the irreducibles of a regular semidirect product from two geometric data: the orbit, which fixes the spectral support, and the little-group representation, which fixes the internal multiplicity; this is exactly the structure that the Wigner classification of 07.07.06 exhibits with mass shells and spin, that 09.03.03 exhibits with energy spheres and helicity, and that the Stone–von Neumann theorem exhibits with configuration space and the canonical commutation relations. The finite-group theory generalises through Frobenius reciprocity 07.01.08 and the Mackey subgroup formula into the locally compact theory, and the projective refinement is dual to it in the sense that a $2$ -cocycle on the little group threads through induction unchanged, so the metaplectic and spin representations sit inside the same machine. The whole arc — induction, imprimitivity, the dual action, the regularity hypothesis, the orbit decomposition — identifies the representation theory of $A ⋊ H$ with the equivariant geometry of $A$ , and it builds toward the orbit method for nilpotent and solvable groups, where coadjoint orbits replace the dual of the abelian factor.

Full proof set Master

Proposition (the induced representation is a system of imprimitivity). On the model space $L^{2} (G / H, μ; V_{σ})$ , the pair $(Ind_{H}^{G} (σ), P_{0})$ with $P_{0} (E) = M_{1_{E}}$ (multiplication by the indicator of $E \subseteq G / H$ ) is a system of imprimitivity based on $G / H$ .

Proof. Each $P_{0} (E)$ is a self-adjoint idempotent, so a projection, and $E \mapsto P_{0} (E)$ is countably additive in the strong operator topology because indicators of disjoint sets multiply to zero and sum to the indicator of the union; hence $P_{0}$ is a projection-valued measure. For covariance, compute on $ψ \in L^{2} (G / H, μ; V_{σ})$ with $U = Ind_{H}^{G} (σ)$ : $$ \bigl(U(g) P_0(E) U(g)^{-1} \psi\bigr)(\dot x) = \lambda(g^{-1}, \dot x)^{1/2}, \bigl(P_0(E) U(g)^{-1}\psi\bigr)(g^{-1}\dot x). $$ The inner factor is $1_{E} (g^{- 1} \overset{x}{˙}) (U (g)^{- 1} ψ) (g^{- 1} \overset{x}{˙})$ , and $1_{E} (g^{- 1} \overset{x}{˙}) = 1_{g E} (\overset{x}{˙})$ . Carrying the indicator out front and recombining the two cocycle factors $λ (g^{- 1}, \overset{x}{˙})^{1/2}$ and the one inside $U (g)^{- 1}$ (which multiply to $1$ by the cocycle identity applied to $g g^{- 1} = e$ ) returns $1_{g E} (\overset{x}{˙}) ψ (\overset{x}{˙}) = (P_{0} (g E) ψ) (\overset{x}{˙})$ . Hence $U (g) P_{0} (E) U (g)^{- 1} = P_{0} (g E)$ . $□$

Proposition (Stone–von Neumann as imprimitivity). Let $G = R^{n} ⋊ R^{n}$ with the second factor acting as the identity at the group level but pairing with the first by the canonical symplectic form, modelled by position multiplications $Q_{j}$ and momentum translations $P_{j}$ . A system of imprimitivity for the translation group $R^{n}$ (momenta) based on configuration space $R^{n} = G / H$ with $H = {0} ⋊ R^{n}$ is unique up to multiplicity, and the irreducible one is the Schrödinger representation on $L^{2} (R^{n})$ .

Proof. The projection-valued measure $P$ on configuration space $R^{n}$ together with the momentum representation $U$ satisfying $U (a) P (E) U (a)^{- 1} = P (E + a)$ is, by definition, a system of imprimitivity based on $R^{n} = G / H$ with $H$ the momentum subgroup. The imprimitivity theorem identifies it with $Ind_{H}^{G} (σ)$ for a representation $σ$ of $H ≅ R^{n}$ . The irreducible characters of $R^{n}$ are one-dimensional, so the irreducible inducing $σ$ are characters; each yields the same induced representation up to equivalence because the orbit is a single point translated to all of configuration space, giving $L^{2} (R^{n})$ with $P (E) = M_{1_{E}}$ (position) and $U (a)$ translation (momentum). This is the Schrödinger representation, and uniqueness up to multiplicity is the imprimitivity theorem's uniqueness clause. The canonical commutation relations $[Q_{j}, P_{k}] = i δ_{j k}$ are the infinitesimal form of the covariance $U (a) P (E) U (a)^{- 1} = P (E + a)$ . $□$

Proposition (irreducibility from an irreducible little group). For a regular semidirect product $G = A ⋊ H$ , orbit $O = H \cdot χ$ , and irreducible $σ$ of $L = H_{χ}$ , the induced representation $U^{(O, σ)}$ is irreducible.

Proof. Let $T$ be a bounded operator commuting with $U^{(O, σ)} (g)$ for all $g$ . Commuting with $A$ , whose restriction acts on $L^{2} (O, μ; V_{σ})$ by multiplication by the characters $χ^{'}$ ranging over the orbit, forces $T$ to be decomposable: $T = \int_{O}^{\oplus} T (χ^{'}) d μ (χ^{'})$ with $T (χ^{'}) \in End (V_{σ})$ , since the multiplication algebra by orbit characters is maximal abelian (the orbit characters separate points of $O$ ). Commuting with the $H$ -part then forces the equivariance $T (h \cdot χ^{'}) = σ (W (h, χ^{'})) T (χ^{'}) σ (W (h, χ^{'}))^{- 1}$ , where $W (h, χ^{'})$ is the little-group cocycle from the section. The orbit is a single $H$ -orbit, so $T$ is determined by its value $T (χ)$ at the standard point, transported by transitivity. Finally $T (χ)$ commutes with $σ (L)$ for every little-group element; since $σ$ is irreducible, Schur's lemma gives $T (χ) = c I$ . Hence $T = c I$ and $U^{(O, σ)}$ is irreducible. $□$

Proposition (inequivalence across orbits). If $O \neq = O^{'}$ then $U^{(O, σ)} \neq ≅ U^{(O^{'}, σ^{'})}$ for any $σ, σ^{'}$ .

Proof. Restrict each representation to the abelian factor $A$ . By the proposition computing the spectral support, $U^{(O, σ)} ∣_{A}$ has projection-valued measure supported exactly on $O$ , and $U^{(O^{'}, σ^{'})} ∣_{A}$ on $O^{'}$ . A unitary equivalence of representations of $G$ restricts to a unitary equivalence of the representations of $A$ , which must carry one spectral measure to the other and hence preserve the support. Since $O \neq = O^{'}$ are disjoint $H$ -orbits in $A$ , no such equivalence exists. Given $O = O^{'}$ , equivalence further requires the little-group representations to match: restricting to the isotropy subgroup $A ⋊ L$ recovers $χ \otimes σ$ , so $σ ≅ σ^{'}$ . $□$

Proposition (transitivity of induction). For closed subgroups $K \leq H \leq G$ and a unitary representation $σ$ of $K$ , $Ind_{K}^{G} (σ) ≅ Ind_{H}^{G} (Ind_{K}^{H} (σ))$ .

Proof. Realise $Ind_{K}^{G} (σ)$ on covariant functions $f : G \to V_{σ}$ with $f (g k) = σ (k)^{- 1} f (g)$ , square-integrable over $G / K$ . Define $Φ (f) (g) (h) = f (g h)$ for $g \in G$ , $h \in H$ . The inner relation $Φ (f) (g) (hk) = f (g hk) = σ (k)^{- 1} f (g h)$ shows $Φ (f) (g) \in H_{Ind_{K}^{H} σ}$ ; the outer relation $Φ (f) (g h^{'}) (h) = f (g h^{'} h) = Φ (f) (g) (h^{'} h)$ shows $Φ (f)$ transforms by $Ind_{K}^{H} (σ) (h^{'})^{- 1}$ under right multiplication by $h^{'} \in H$ , so $Φ (f) \in H_{Ind_{H}^{G} (Ind_{K}^{H} σ)}$ . The quasi-invariant measures factor through the tower $G / K \to G / H$ with fibre $H / K$ by the cocycle factorisation $λ^{G / K} = λ^{G / H} \cdot λ^{H / K}$ , making $Φ$ an isometry onto its image; surjectivity follows by reversing the construction with $f (g) = Φ (f) (g) (e)$ . The map intertwines left translation by $G$ by inspection, so it is a unitary equivalence. $□$

Connections Master

Induced representation 07.01.07. Mackey's $Ind_{H}^{G}$ for a closed subgroup of a locally compact group is the analytic completion of the finite-group induction defined there: cosets become a measure space $G / H$ , the permutation action becomes a quasi-invariant measure with a Radon–Nikodym cocycle, and the finite direct sum over cosets becomes an $L^{2}$ direct integral. Every theorem in this unit reduces, when $G$ is finite, to a statement already available from the finite construction, which is the correct way to read the abstractions back into computable form.
Frobenius reciprocity 07.01.08. The imprimitivity theorem is the geometric face of the same induction–restriction adjunction. Frobenius reciprocity computes $Hom_{G} (Ind_{H}^{G} σ, π) ≅ Hom_{H} (σ, Res_{H} π)$ ; the Mackey subgroup theorem refines it by computing $Res_{K} Ind_{H}^{G}$ over double cosets, and the irreducibility criterion for induced representations falls out of that refinement. The adjunction is what guarantees the inducing datum $σ$ recovered by the imprimitivity theorem is unique.
Wigner's classification of the Poincaré group 07.07.06. That unit is the single most important application of this machine: the abelian factor is the spacetime translations $R^{1, 3}$ , the dual is momentum space, the orbits are the mass shells and light cone, and the little groups $SU (2)$ , $ISO (2)$ , $SU (1, 1)$ supply spin and helicity. Wigner obtained the result by hand in 1939, before Mackey's general theorem; reading the two together shows precisely which steps were special to the Lorentzian orbit geometry and which are the universal imprimitivity argument.
Quantum free particle as a representation of $E (3)$ 09.03.03. The non-relativistic free particle is the Mackey machine for $E (3) = R^{3} ⋊ SO (3)$ , with spherical orbits in momentum space and little group $SO (2)$ . It is the same theorem as the Poincaré case on a different abelian normal factor, and it is the cleanest finite-dimensional-little-group instance in which to watch the orbit-and-stabiliser data assemble into an irreducible representation.
Casimir element 07.06.10. On a representation produced by the machine, the Casimir invariants of the enveloping algebra act as the scalars that label the orbit (the value of the abelian invariant, e.g. mass-squared) and the little-group representation (the value of the little-group Casimir, e.g. spin). The general Schur-lemma principle that central elements act by scalars on irreducibles is what makes these orbit and little-group labels well-defined invariants of the equivalence class.

Historical & philosophical context Master

George Mackey developed the theory in a sequence of papers beginning with Imprimitivity for representations of locally compact groups I, Proceedings of the National Academy of Sciences 35 (1949) 537–545 ^{[Mackey 1949]}, which proved the imprimitivity theorem characterising induced representations by a system of imprimitivity over $G / H$ , and continuing through Induced representations of locally compact groups I, Annals of Mathematics 55 (1952) 101–139 ^{[Mackey 1952]}, which constructed $Ind_{H}^{G}$ in full generality with quasi-invariant measures and established the subgroup and irreducibility theorems. The construction generalised Frobenius's finite-group induction from Über Relationen zwischen den Charakteren einer Gruppe und denen ihrer Untergruppen, Sitzungsberichte der Preußischen Akademie der Wissenschaften (1898) 501–515 ^{[Frobenius 1898]}, replacing the sum over cosets by an integral over a quasi-invariant measure and the coset-permutation action by a Radon–Nikodym cocycle.

Mackey recognised that his machine subsumed Wigner's 1939 classification of the Poincaré group and the Stone–von Neumann uniqueness theorem for the canonical commutation relations as instances of a single statement, a reading he set out in the monograph Induced Representations of Groups and Quantum Mechanics (Benjamin 1968) ^{[Mackey 1968]}. Sternberg's Group Theory and Physics (CUP 1994) ^{[Sternberg 1994]} presents the imprimitivity theorem and the little-group method as the apparatus that makes Wigner's classification a theorem rather than a case computation, and Folland's A Course in Abstract Harmonic Analysis (CRC 1995) ^{[Folland 1995]} gives the modern measure-theoretic treatment, including the regularity hypothesis under which the orbit-and-little-group parametrisation of $A ⋊ H$ is exact. The non-regular case, where the orbit space is not countably separated and the group is not type I, was identified by Mackey and analysed further by Glimm and by Effros, who characterised type-I groups by the topological regularity of the orbit space.

Bibliography Master

@article{Mackey1949,
  author  = {Mackey, George W.},
  title   = {Imprimitivity for representations of locally compact groups {I}},
  journal = {Proceedings of the National Academy of Sciences},
  volume  = {35},
  number  = {9},
  year    = {1949},
  pages   = {537--545}
}

@article{Mackey1952,
  author  = {Mackey, George W.},
  title   = {Induced representations of locally compact groups {I}},
  journal = {Annals of Mathematics},
  volume  = {55},
  number  = {1},
  year    = {1952},
  pages   = {101--139}
}

@article{Mackey1953,
  author  = {Mackey, George W.},
  title   = {Induced representations of locally compact groups {II}: The {F}robenius reciprocity theorem},
  journal = {Annals of Mathematics},
  volume  = {58},
  number  = {2},
  year    = {1953},
  pages   = {193--221}
}

@book{Mackey1968,
  author    = {Mackey, George W.},
  title     = {Induced Representations of Groups and Quantum Mechanics},
  publisher = {W. A. Benjamin},
  year      = {1968}
}

@article{Frobenius1898,
  author  = {Frobenius, Ferdinand Georg},
  title   = {{\"U}ber Relationen zwischen den Charakteren einer Gruppe und denen ihrer Untergruppen},
  journal = {Sitzungsberichte der Preu{\ss}ischen Akademie der Wissenschaften zu Berlin},
  year    = {1898},
  pages   = {501--515}
}

@article{Wigner1939,
  author  = {Wigner, Eugene P.},
  title   = {On unitary representations of the inhomogeneous {L}orentz group},
  journal = {Annals of Mathematics},
  volume  = {40},
  number  = {1},
  year    = {1939},
  pages   = {149--204}
}

@book{Sternberg1994,
  author    = {Sternberg, Shlomo},
  title     = {Group Theory and Physics},
  publisher = {Cambridge University Press},
  year      = {1994}
}

@book{Folland1995,
  author    = {Folland, Gerald B.},
  title     = {A Course in Abstract Harmonic Analysis},
  publisher = {CRC Press},
  year      = {1995}
}

@book{Hall2013,
  author    = {Hall, Brian C.},
  title     = {Quantum Theory for Mathematicians},
  publisher = {Springer},
  series    = {Graduate Texts in Mathematics},
  volume    = {267},
  year      = {2013}
}

Prerequisites

07.01.07
07.01.08
07.06.01
07.07.03
07.07.06
09.03.03

Tier anchors

beginner: Sternberg, *Group Theory and Physics* (CUP 1994), Ch. 9 informal opening on building big-group representations from a subgroup; Woit, *Quantum Theory, Groups and Representations* (Springer 2017), Ch. 42
intermediate: Sternberg *Group Theory and Physics* Ch. 9 (induced representations and the Mackey machine); Hall *Quantum Theory for Mathematicians* (GTM 267, 2013) Ch. 17 §17.2–17.4; Folland *A Course in Abstract Harmonic Analysis* (CRC 1995) Ch. 6 §6.4–6.6
master: Mackey, *Imprimitivity for representations of locally compact groups I*, Proc. Natl. Acad. Sci. USA 35 (1949) 537–545; Mackey, *Induced representations of locally compact groups I*, Ann. Math. 55 (1952) 101–139; Mackey, *Induced Representations of Groups and Quantum Mechanics* (Benjamin 1968); Folland *A Course in Abstract Harmonic Analysis* (CRC 1995) Ch. 6

References

Mackey — Imprimitivity for representations of locally compact groups I · Proc. Natl. Acad. Sci. USA 35 (1949) 537–545; the imprimitivity theorem characterising induced representations by a system of imprimitivity over $G/H$
Mackey — Induced representations of locally compact groups I · Ann. Math. 55 (1952) 101–139; the construction of $\mathrm{Ind}_H^G$ for a closed subgroup of a locally compact group, with quasi-invariant measures and the cocycle
Mackey — Induced Representations of Groups and Quantum Mechanics · Benjamin 1968, Chs. 1–3; the imprimitivity theorem, the Mackey machine for regular semidirect products, and the quantum-mechanical reading
Frobenius — Über Relationen zwischen den Charakteren einer Gruppe und denen ihrer Untergruppen · Sitzungsberichte Akad. Wiss. Berlin (1898) 501–515; the original finite-group induction that Mackey generalised
Sternberg — Group Theory and Physics · CUP 1994, Ch. 9; induced representations, the Mackey machine, and the little-group method as the engine behind Wigner's classification
Folland — A Course in Abstract Harmonic Analysis · CRC 1995, Ch. 6 §6.4–6.6; the imprimitivity theorem and the Mackey-machine classification for regular semidirect products
Hall — Quantum Theory for Mathematicians · Springer GTM 267, 2013, Ch. 17 §17.2–17.4; the Mackey machine for $A \rtimes H$ with $A$ abelian

Estimated time

beginner: 20m
intermediate: 45m
master: 85m