Physlib

PhyslibAlpha.AlgebraicFramework.Representation.Covariance.Finite

Covariance for finite-outcome measurements

`Covariance.lean` sets up covariance in full measure-theoretic generality: a measurable action of `G` on an outcome space `Ω`, transported to a physical system via a homomorphism into `Symmetry E`. For a *finite* outcome type `ι`, this specializes drastically: a group acting on a finite label set needs no measurability machinery at all, since every subset of a finite type is automatically "measurable" in the relevant sense — a permutation of a finite set permutes it, full stop. This file builds the induced symmetry action `G →* Symmetry (ι → ℝ)` on the classical `ι`-outcome system (`ClassicalSystem.lean`, `FiniteOutcome.lean`), and uses it to specialize two of `Covariance.lean`'s general structural theorems — covariance preserved under composition, and under postprocessing — to finite-outcome measurements presented as channels (`Measurement.channelEquiv`).

The induced action is the standard "functions on a `G`-set" representation: `(σ • f) i = f (σ⁻¹ • i)`. This is a genuine *left* action — `σ • (τ • f) = (σ * τ) • f` — precisely because of the inverse: composing `f ↦ f ∘ (σ⁻¹ • ·)` then `f ↦ f ∘ (τ⁻¹ • ·)` composes the permutations as `(στ)⁻¹ = τ⁻¹σ⁻¹` in the matching order (`inducedAction`'s `map_mul'` spells this out) — using `σ` without inversion would instead give an *anti*-homomorphism. This matches the standard convention for the classical system, `(β_g f)(x) = f(g⁻¹x)`.

Since a finite-outcome measurement `M : (ι → ℝ) →ₚ₁[ℝ] E` is already a channel (`FiniteOutcome.lean`), its covariance under this induced action, matched to a target symmetry `ρ : G →* Symmetry E`, is *literally* `Covariance.lean`'s general channel notion, `M.IsCovariant inducedAction ρ` — no new predicate is needed. The two theorems proved here, `postprocess_isCovariant` and `marginal_isCovariant`, both fall out of `UnitalPositiveLinearMap.IsCovariant.comp` plus the fact that postprocessing and marginalizing are literal channel composition (`Postprocessing.lean`).

Main definitions

- `inducedLinearMap`, `inducedChannel`, `inducedSymmetry`, `inducedAction` : the homomorphism `G →* Symmetry (ι → ℝ)` induced by a `MulAction G ι` on a finite outcome-label type. - `Measurement.postprocess_isCovariant` : post-processing by an equivariant classical channel preserves covariance. - `Measurement.classicalPullback_isCovariant`, `Measurement.marginal_isCovariant` : marginals of a covariant joint measurement, for a diagonal product action on the joint outcome type, are covariant.

Post-processing by an equivariant classical channel preserves covariance

A finite-outcome measurement `M : (ι → ℝ) →ₚ₁[ℝ] E` is covariant under a `G`-action on its outcome labels `ι`, matched to a target symmetry `ρ : G →* Symmetry E`, precisely when `M.IsCovariant inducedAction ρ` — `Covariance.lean`'s general notion, instantiated with the induced action from `inducedAction` above. No new predicate is needed: this *is* that notion.

Marginals of a covariant joint measurement are covariant

12 declarations

definition

Linear map on ι→R\iota \to \mathbb{R} induced by σ∈G\sigma \in G

For a group GG acting on a finite set ι\iota of outcome labels, this definition is the linear map from the space of functions ι→R\iota \to \mathbb{R} to itself induced by an element σ∈G\sigma \in G. It maps each function ff to the function i↦f(σ−1⋅i)i \mapsto f(\sigma^{-1} \cdot i), representing the standard action of a group on the space of functions over a GG-set by pulling back along the inverse action.

theorem

Evaluation of the induced linear map: (σ⋅f)(i)=f(σ−1⋅i)(\sigma \cdot f)(i) = f(\sigma^{-1} \cdot i)

For a group GG acting on a finite set ι\iota of outcome labels, let f:ι→Rf: \iota \to \mathbb{R} be a function and σ∈G\sigma \in G. The linear map induced by σ\sigma, denoted as inducedLinearMap(σ)\text{inducedLinearMap}(\sigma), maps ff to a new function whose value at an index i∈ιi \in \iota is given by f(σ−1⋅i)f(\sigma^{-1} \cdot i).

definition

Channel on ι→R\iota \to \mathbb{R} induced by σ∈G\sigma \in G

For an element σ\sigma of a group GG acting on a finite set ι\iota of outcome labels, this definition is the unital positive linear map (channel) from the space of functions ι→R\iota \to \mathbb{R} to itself induced by σ\sigma. It maps each function ff to the function i↦f(σ−1⋅i)i \mapsto f(\sigma^{-1} \cdot i). The map is positive because permuting coordinates preserves non-negativity, and it is unital because it preserves the constant function 1\mathbf{1}.

theorem

Evaluation of the Induced Channel: (inducedChannel σ)f(i)=f(σ−1⋅i)(\text{inducedChannel } \sigma) f (i) = f(\sigma^{-1} \cdot i)

Let GG be a group acting on a finite set ι\iota. For any element σ∈G\sigma \in G, function f:ι→Rf : \iota \to \mathbb{R}, and index i∈ιi \in \iota, the value of the induced channel (inducedChannel σ)(\text{inducedChannel } \sigma) applied to ff and evaluated at ii is given by f(σ−1⋅i)f(\sigma^{-1} \cdot i).

theorem

The induced channel MσM_\sigma is an order-automorphism for any σ∈G\sigma \in G

Let GG be a group acting on a finite set ι\iota of outcome labels. For any element σ∈G\sigma \in G, the induced channel Mσ:(ι→R)→(ι→R)M_\sigma: (\iota \to \mathbb{R}) \to (\iota \to \mathbb{R}), defined as the unital positive linear map such that (Mσf)(i)=f(σ−1⋅i)(M_\sigma f)(i) = f(\sigma^{-1} \cdot i), is an order-automorphism. This means that MσM_\sigma is a bijection and preserves the pointwise ordering of functions (i.e., f≤gf \le g if and only if Mσf≤MσgM_\sigma f \le M_\sigma g).

definition

Symmetry on ι→R\iota \to \mathbb{R} induced by σ∈G\sigma \in G

For an element σ\sigma of a group GG acting on a finite set ι\iota of outcome labels, the induced symmetry is the order-automorphism of the space of functions ι→R\iota \to \mathbb{R}. It bundles the induced channel MσM_\sigma—which maps a function ff to the function i↦f(σ−1⋅i)i \mapsto f(\sigma^{-1} \cdot i)—with the proof that this map is an order-automorphism (i.e., a bijective map that preserves the pointwise ordering of functions).

theorem

The underlying map of inducedSymmetry(σ)\text{inducedSymmetry}(\sigma) is inducedChannel(σ)\text{inducedChannel}(\sigma)

Let GG be a group acting on a finite set ι\iota of outcome labels. For any element σ∈G\sigma \in G, the underlying unital positive linear map (channel) of the symmetry inducedSymmetry(σ)\text{inducedSymmetry}(\sigma) on the space of functions ι→R\iota \to \mathbb{R} is equal to the induced channel inducedChannel(σ)\text{inducedChannel}(\sigma).

definition

Induced group action G→Symmetry(ι→R)G \to \text{Symmetry}(\iota \to \mathbb{R})

The induced action is a group homomorphism G→Symmetry(ι→R)G \to \text{Symmetry}(\iota \to \mathbb{R}) derived from the action of a group GG on a finite set of outcome labels ι\iota. For each element σ∈G\sigma \in G, the homomorphism provides a symmetry (specifically, an order-preserving automorphism) of the space of real-valued functions ι→R\iota \to \mathbb{R}. This symmetry transforms a function ff into a new function i↦f(σ−1⋅i)i \mapsto f(\sigma^{-1} \cdot i), which ensures that the mapping behaves as a left group action.

theorem

The underlying map of inducedAction(σ)\text{inducedAction}(\sigma) is inducedChannel(σ)\text{inducedChannel}(\sigma)

For any element σ\sigma of a group GG acting on a finite set ι\iota of outcome labels, the underlying map of the symmetry inducedAction(σ)\text{inducedAction}(\sigma) on the space of functions ι→R\iota \to \mathbb{R} is equal to the induced channel inducedChannel(σ)\text{inducedChannel}(\sigma).

theorem

Post-processing a Covariant Measurement by an Equivariant Channel is Covariant

Let GG be a group acting on finite sets of outcome labels ι\iota and κ\kappa. For any such set XX, let the induced action of GG on the space of functions X→RX \to \mathbb{R} be defined by (σ⋅f)(x)=f(σ−1⋅x)(\sigma \cdot f)(x) = f(\sigma^{-1} \cdot x) for σ∈G\sigma \in G. Let EE be a physical system (an ordered R\mathbb{R}-module) and ρ:G→Symmetry(E)\rho : G \to \text{Symmetry}(E) be a group homomorphism representing the symmetry on EE. Consider a finite-outcome measurement M:(ι→R)→EM : (\iota \to \mathbb{R}) \to E and a classical post-processing channel K:(κ→R)→(ι→R)K : (\kappa \to \mathbb{R}) \to (\iota \to \mathbb{R}). If MM is covariant with respect to the induced action on ι→R\iota \to \mathbb{R} and ρ\rho, and KK is equivariant (covariant with respect to the induced actions on κ→R\kappa \to \mathbb{R} and ι→R\iota \to \mathbb{R}), then the post-processed measurement M∘KM \circ K is covariant with respect to the induced action on κ→R\kappa \to \mathbb{R} and ρ\rho.

theorem

Equivariant Map ff Induces Covariant Classical Pullback Channel

Let GG be a group, and let ι\iota and κ\kappa be types (representing outcome label sets) equipped with GG-actions. Suppose f:κ→ιf: \kappa \to \iota is a GG-equivariant map, such that for all g∈Gg \in G and k∈κk \in \kappa, the condition f(g⋅k)=g⋅f(k)f(g \cdot k) = g \cdot f(k) holds. Then the classical pullback channel (classicalPullback f):(ι→R)→(κ→R)(\text{classicalPullback } f) : (\iota \to \mathbb{R}) \to (\kappa \to \mathbb{R}), defined by precomposing a function with ff, is covariant with respect to the induced group actions on the function spaces. The induced action of g∈Gg \in G on a function pp is given by (g⋅p)(x)=p(g−1⋅x)(g \cdot p)(x) = p(g^{-1} \cdot x).

theorem

Marginals of a Covariant Joint Measurement are Covariant

Let GG be a group acting on finite sets of outcome labels ι\iota and κ\kappa. Let EE be a physical system represented by an ordered real vector space with a unit 1∈E1 \in E, and let ρ:G→Symmetry(E)\rho: G \to \text{Symmetry}(E) be a group homomorphism representing the symmetry action of GG on the system. Suppose J:(ι×κ→R)→EJ : (\iota \times \kappa \to \mathbb{R}) \to E is a joint measurement (a unital positive linear map). If JJ is covariant with respect to the diagonal action of GG on ι×κ\iota \times \kappa and the symmetry ρ\rho on EE (i.e., J(σ⋅f)=ρ(σ)(J(f))J(\sigma \cdot f) = \rho(\sigma)(J(f)) for all σ∈G\sigma \in G, where the induced action on functions is (σ⋅f)(x)=f(σ−1⋅x)(\sigma \cdot f)(x) = f(\sigma^{-1} \cdot x)), then the marginal measurement on ι\iota is also covariant. The marginal measurement on ι\iota is defined as the post-processing of JJ by the classical channel mapping f∈(ι→R)f \in (\iota \to \mathbb{R}) to f∘pr1∈(ι×κ→R)f \circ \text{pr}_1 \in (\iota \times \kappa \to \mathbb{R}), where pr1:ι×κ→ι\text{pr}_1: \iota \times \kappa \to \iota is the projection onto the first factor. This marginal is covariant with respect to the action of GG on ι\iota and the symmetry ρ\rho.