Documentation

Oseledets.Lyapunov.ExteriorNorm.Basic

Operator norm of the exterior power: trivializations and the compound matrix #

For finite-dimensional real inner product spaces E, F, this module builds the operator-norm theory of the k-th exterior power ⋀^k f of a linear map f : E →ₗ[ℝ] F: the submultiplicativity engine, the Hodge and orthonormal-basis trivializations, and the compound matrix with its Cauchy–Binet multiplicativity. The Plücker eigenvalue ceilings and the Weyl perturbation theory that build on this layer live in Oseledets.Lyapunov.ExteriorNorm.Plucker and Oseledets.Lyapunov.ExteriorNorm.Weyl.

Main definitions #

Main results #

Implementation notes #

The type ⋀[ℝ]^k E is definitionally ↥(Submodule …) and already carries an AddCommGroup instance coming from the ambient submodule. Asserting or installing a fresh NormedAddCommGroup (⋀[ℝ]^k E) would create an AddCommGroup/topology diamond that breaks even IsTopologicalAddGroup synthesis on ⋀^k E.

To stay diamond-free we never put a normed structure on ⋀^k E. Instead we carry an explicit linear trivialization ε : ⋀^k E ≃ₗ[ℝ] EuclideanSpace ℝ (Fin n) as data and measure the operator norm of the conjugated map in the genuine Euclidean target. The canonical such trivialization (exteriorTrivialization) exists because ⋀^k E is a finite free -module.

A small set of trivialization/ordering helper lemmas in this file are not private because the Plucker layer downstream reuses them; they are internal infrastructure, not public API.

The submultiplicativity engine #

We carry explicit linear trivializations ε : ⋀^k · ≃ₗ EuclideanSpace ℝ (Fin n) as data and take the operator norm of the conjugated exterior map in the genuine Euclidean target.

The k-th exterior map ⋀^k f, conjugated through trivializations of source and target exterior powers into genuine Euclidean spaces.

Equations
Instances For
    noncomputable def Oseledets.ExteriorNorm.exteriorOpNorm {E : Type u_1} {F : Type u_2} [NormedAddCommGroup E] [NormedSpace E] [NormedAddCommGroup F] [NormedSpace F] {nE nF : } (k : ) (εE : (⋀[]^k E) ≃ₗ[] EuclideanSpace (Fin nE)) (εF : (⋀[]^k F) ≃ₗ[] EuclideanSpace (Fin nF)) (f : E →ₗ[] F) :

    The exterior-power operator norm of f, measured through the trivializations εE, εF. When εE, εF are the orthonormal-wedge isometries for the Hodge inner product, this is the genuine ‖⋀^k f‖.

    Equations
    Instances For
      @[simp]
      theorem Oseledets.ExteriorNorm.exteriorOpNorm_comp_le {E : Type u_1} {F : Type u_2} {G : Type u_3} [NormedAddCommGroup E] [NormedSpace E] [NormedAddCommGroup F] [NormedSpace F] [NormedAddCommGroup G] [NormedSpace G] {nE nF nG : } (k : ) (εE : (⋀[]^k E) ≃ₗ[] EuclideanSpace (Fin nE)) (εF : (⋀[]^k F) ≃ₗ[] EuclideanSpace (Fin nF)) (εG : (⋀[]^k G) ≃ₗ[] EuclideanSpace (Fin nG)) (f : E →ₗ[] F) (g : F →ₗ[] G) :
      exteriorOpNorm k εE εG (g ∘ₗ f) exteriorOpNorm k εF εG g * exteriorOpNorm k εE εF f

      Submultiplicativity of the exterior-power operator norm. Pure functoriality (exteriorPower.map_comp, with the middle trivialization telescoping) together with the submultiplicativity of the continuous-linear-map operator norm (opNorm_comp_le).

      Existence of trivializations #

      Every ⋀^k E (a finite free -module) admits a linear equiv to a Euclidean space, via its finrank basis.

      A canonical linear trivialization of ⋀^k E into a Euclidean space, via the finrank basis.

      Equations
      Instances For

        The Hodge trivialization #

        For an inner product space E, the Hodge trivialization of ⋀^k E is the linear equiv to EuclideanSpace that sends the orthonormal wedge basis — the k-fold wedges e_{i₁} ∧ ⋯ ∧ e_{i_k} of the standard orthonormal basis {eᵢ} of E — to the standard Euclidean basis. It is a concrete piece of data (no inner product is installed on ⋀^k E, avoiding the AddCommGroup/topology diamond). Measuring the exterior operator norm through this trivialization gives the genuine ‖⋀^k f‖ for the Hodge inner product.

        The wedge basis of ⋀^k E induced by the standard orthonormal basis of E: its elements are the k-fold wedge products of distinct standard basis vectors. As a Basis it is data, and under the Hodge inner product it is orthonormal.

        Equations
        Instances For

          The reindexing equiv powersetCard (Fin (finrank E)) k ≃ Fin (finrank (⋀^k E)) witnessing that both index sets have the same cardinality (finrank E).choose k.

          Equations
          Instances For

            The Hodge trivialization of ⋀^k E: the linear equiv to a Euclidean space sending the orthonormal wedge basis to the standard Euclidean basis.

            Equations
            • One or more equations did not get rendered due to their size.
            Instances For

              The bridge to singular values #

              ‖⋀^k f‖ = ∏_{i<k} σᵢ(f), measured through the Hodge trivializations of source and target. Mathematically: an SVD of f diagonalizes ⋀^k f on the orthonormal bases of k-fold wedges of singular vectors; the operator norm is attained on the top wedge u₀ ∧ ⋯ ∧ u_{k-1}, whose image has norm ∏_{i<k} σᵢ(f) (the largest wedge product, since σ is antitone). This requires the SVD-decomposition packaging, the orthonormality of the wedge basis for the Hodge inner product, and a diagonal-operator-norm computation, none of which are currently in Mathlib.

              The det-Gram (Hodge) bilinear form and orthonormal-wedge trivializations #

              We carry the Hodge inner product on ⋀^k E as a plain bilinear form hodgeForm (never as a typeclass instance, to avoid the AddCommGroup/topology diamond). It is defined by reusing the exterior-power dual pairing composed with the inner-product-to-dual map innerₗ, so on ιMulti families it is the determinant of the Gram matrix det ⟪vᵢ, wⱼ⟫.

              On ιMulti families, the Hodge form is the determinant of the Gram matrix ⟪vⱼ, wᵢ⟫.

              Reindexing powersetCard ι k ≃ Fin (finrank (⋀^k E)) for an o.n. basis b indexed by ι.

              Equations
              Instances For

                The wedge trivialization attached to an arbitrary orthonormal basis b of E.

                Equations
                • One or more equations did not get rendered due to their size.
                Instances For

                  The trivialization is a Hodge isometry: the L2 inner product of trivialized vectors equals the Hodge form.

                  Change of coordinates between two o.n.-basis wedge trivializations of the same space is an L2 isometry (the compound ⋀^k Q of the orthogonal change of basis). The two bases may be indexed differently; only the space E (hence finrank (⋀^k E)) matters.

                  Equations
                  Instances For
                    theorem Oseledets.ExteriorNorm.onbChange_apply {E : Type u_1} [NormedAddCommGroup E] [InnerProductSpace E] [FiniteDimensional E] {ιE : Type u_3} {ιE' : Type u_4} [Fintype ιE] [LinearOrder ιE] [Fintype ιE'] [LinearOrder ιE'] (b : OrthonormalBasis ιE E) (b' : OrthonormalBasis ιE' E) (k : ) (p : EuclideanSpace (Fin (Module.finrank (⋀[]^k E)))) :
                    (onbChange b b' k) p = (onbTriv b' k) ((onbTriv b k).symm p)

                    Operator norm of a map with orthogonal images of an orthonormal basis #

                    The Gram determinant and the top-k product maximization #

                    theorem Oseledets.ExteriorNorm.prod_le_prod_top {n k : } (σ : Fin n) ( : Antitone σ) (hpos : ∀ (i : Fin n), 0 σ i) (S : (Set.powersetCard (Fin n) k)) (top : Fin kFin n) (htop : ∀ (i : Fin k), (top i) = i) :
                    aS, σ a i : Fin k, σ (top i)

                    Top-k prefix maximizes the product of antitone nonnegative weights. For antitone nonnegative σ : Fin n → ℝ, the product over any k-subset is at most the product over the top-k prefix indices (any top with (top i : ℕ) = i).

                    theorem Oseledets.ExteriorNorm.top_elem_ge {n k : } (hk1 : 1 k) (hkn : k n) (e : Fin k ↪o Fin n) (htop : Finset.image (fun (j : Fin k) => e j) Finset.univ Finset.image (fun (i : Fin k) => i, ) Finset.univ) :
                    k (e k - 1, )

                    The top element of a non-top k-subset is ≥ k. If the ordered enumeration e of a k-subset S of Fin n does not enumerate the top prefix {0,…,k-1}, then its largest element e ⟨k-1⟩ has value ≥ k. (Otherwise all of S lies in {0,…,k-1}, forcing S = top prefix.)

                    theorem Oseledets.ExteriorNorm.prod_le_second_aux {n : } (m : ) (lam : ) (hanti : Antitone lam) (hpos : ∀ (i : ), 0 lam i) (e : Fin (m + 1) ↪o Fin n) (htopge : m + 1 (e m, )) :
                    j : Fin (m + 1), lam (e j) (∏ iFinset.range m, lam i) * lam (m + 1)

                    The second-largest k-subset product bound. For antitone nonnegative lam, the product of lam over the ordered enumeration e of a k-subset whose top element is ≥ k (i.e. a non-top subset) is at most the second-largest product (∏_{i<k-1} lam i)·lam k. The top factor drops from lam (k-1) to lam k; the remaining k-1 factors are bounded by the prefix {0,…,k-2}.

                    The bridge. Through the Hodge trivializations of source and target, the exterior operator norm equals the product of the top-k singular values: exteriorOpNorm k (hodgeTrivialization k) (hodgeTrivialization k) f = ∏_{i<k} σᵢ(f).

                    The Plücker bridge: eigen-diagonalization of the compound #

                    For a symmetric map f with orthonormal eigenbasis u and eigenvalues lam, the compound ⋀^k f is diagonal in the wedge basis of u: it scales the wedge u_S by ∏_{a ∈ S} lam a. This is the abstract spectral core behind the Plücker top eigenpair and the second-eigenvalue ceiling established below.

                    theorem Oseledets.ExteriorNorm.prod_ofFinEmbEquiv_symm {ι : Type u_3} [LinearOrder ι] {k : } (lam : ι) (S : (Set.powersetCard ι k)) :
                    i : Fin k, lam ((Set.powersetCard.ofFinEmbEquiv.symm S) i) = aS, lam a

                    Product reindexing. A product of lam over the ordered enumeration of a k-subset S equals the product of lam over S.

                    theorem Oseledets.ExteriorNorm.map_exteriorPower_wedgeBasis_eq {E : Type u_1} [NormedAddCommGroup E] [InnerProductSpace E] {ι : Type u_3} [Fintype ι] [LinearOrder ι] (f : E →ₗ[] E) (u : OrthonormalBasis ι E) (lam : ι) (hf : ∀ (i : ι), f (u i) = lam i u i) (k : ) (S : (Set.powersetCard ι k)) :

                    Eigen-diagonalization of the compound (abstract). For a linear map f with an orthonormal eigenbasis u (f (u i) = lam i • u i), the compound ⋀^k f scales each wedge basis vector u_S by the subset product ∏_{a ∈ S} lam a.

                    The compound matrix and the operator-norm/compound bridge #

                    For a square matrix M, the conjugated exterior map ⋀^k (toEuclideanLin M) through the orthonormal-wedge trivialization of the standard basis EuclideanSpace.basisFun is itself toEuclideanLin of an explicit compound matrix compoundMatrix k M, whose entries are the k × k minors of M. This converts the (analytically delicate) exterior operator norm into the L2 operator norm of a concrete matrix of determinants — the form needed for measurability.

                    onbTriv sends the i-th wedge basis vector (under (wIndexEquiv b k).symm) to the i-th standard Euclidean basis vector.

                    The compound matrix C_k(M): its (t, s) entry is the k × k minor of M obtained by selecting the rows enumerated by the t-th k-subset and the columns enumerated by the s-th k-subset (under the standard orthonormal-wedge index equivalence).

                    Equations
                    • One or more equations did not get rendered due to their size.
                    Instances For

                      The compound bridge. Through the orthonormal-wedge trivializations of the standard basis, ⋀^k (toEuclideanLin M) is toEuclideanLin of the compound matrix C_k(M).

                      Matrix-level Cauchy–Binet. The k-th compound of a matrix product is the product of the compounds: C_k(B * M) = C_k(B) * C_k(M). This is the multiplicativity of the compound matrix, proved via the functoriality ⋀^k(B ∘ M) = ⋀^k B ∘ ⋀^k M (exteriorPower.map_comp) transported through the standard orthonormal-wedge trivialization by the compound bridge conjExteriorMap_eq_toEuclideanLin_compound.

                      Compound of a transpose. The k-th compound matrix of Mᵀ is the transpose of the k-th compound of M: C_k(Mᵀ) = C_k(M)ᵀ. Entrywise this is det(Mᵀ.minor) = det(M.minorᵀ) (Matrix.det_transpose), since the row/column enumerators at index t, s coincide.

                      The k-th compound of the Gram matrix Mᵀ M is (C_k M)ᵀ (C_k M), i.e. the Gram matrix of the compound. Combines compoundMatrix_mul with compoundMatrix_transpose.

                      Cauchy–Binet, linear-map form. toEuclideanLin of the k-th compound of a product is the composition of the compounds. This is the form consumed by the rank-1 exterior Rayleigh-deficit chain below, where the right-hand factor is post-composed with the inverse compound.

                      Top singular value vs. the sum of squared singular values (Frobenius) #

                      The Frobenius back-transport below needs ‖M‖_op ≤ ‖M‖_F. Stated through the singular-value bridge (toEuclideanLin) to avoid the L2-operator vs. Frobenius NormedAddCommGroup-instance diamond on Matrix. The core inequality is that the top squared singular value is at most the sum of all squared singular values; the sum equals tr(MᵀM) = ‖M‖_F².

                      The top squared singular value of toEuclideanLin M is at most the sum of all squared singular values. The sum over Fin d equals tr(MᵀM) = ‖M‖_F² (the Hilbert–Schmidt norm squared); combined with ‖M‖_op = σ₀ this yields ‖M‖_op ≤ ‖M‖_F.

                      The top singular value of toEuclideanLin M is at most the Frobenius norm √(∑ σᵢ²). Immediate from singularValues_zero_sq_le_sum and Real.sqrt.

                      The L2 operator-norm/Frobenius bridge. The squared L2 operator norm ‖M‖² of a matrix is at most the sum of the squared singular values of toEuclideanLin M (the squared Frobenius norm). The L2 matrix norm ‖M‖ is by definition the operator norm of toEuclideanLin M on EuclideanSpace; expanding any vector in the singular-value eigenbasis u of adjoint f ∘ₗ f (with f = toEuclideanLin M), the images f uⱼ are pairwise orthogonal with ‖f uⱼ‖ = σⱼ, so ‖f v‖² = ∑ⱼ ⟪uⱼ, v⟫² σⱼ² ≤ (∑ σᵢ²) ‖v‖².

                      The sum of the squared singular values of toEuclideanLin N equals the trace of the Gram matrix tr(Nᵀ N) (the squared Frobenius norm). Both sides are the trace of the self-adjoint operator adjoint f ∘ₗ f = toEuclideanLin (Nᵀ N): in the singular-value eigenbasis its diagonal entries are the eigenvalues σᵢ², while as a matrix its trace is tr(Nᵀ N).

                      A matrix U with orthonormal columns (Uᵀ U = 1) is an isometry, so its L2 operator norm is at most 1.

                      theorem Oseledets.ExteriorNorm.gram_eigenvalues_le_opNorm_sq {k : } (W : Matrix (Fin k) (Fin k) ) (hGherm : (W.transpose * W).IsHermitian) (i : Fin k) :
                      hGherm.eigenvalues i W ^ 2

                      The eigenvalues of the Gram matrix Wᵀ W are bounded by the squared L2 operator norm ‖W‖². Each eigenvalue μᵢ of the Hermitian matrix G = Wᵀ W, with unit eigenvector bᵢ, equals ‖toEuclideanLin W (bᵢ)‖² ≤ ‖W‖².

                      theorem Oseledets.ExteriorNorm.norm_proj_sub_le_wedge {d k : } (U V : Matrix (Fin d) (Fin k) ) (hU : U.transpose * U = 1) (hV : V.transpose * V = 1) :
                      U * U.transpose - V * V.transpose ^ 2 2 * k * (1 - (U.transpose * V).det ^ 2)

                      The Frobenius back-transport. For matrices U, V with orthonormal columns (Uᵀ U = 1, Vᵀ V = 1), the squared L2 operator norm of the difference of the orthogonal projectors U Uᵀ and V Vᵀ is bounded by 2 k (1 - det(Uᵀ V)²). Chain: self-adjoint idempotents of trace k; ‖P − P'‖²_op ≤ ∑σᵢ² = tr((P−P')²) = 2k − 2 tr(P P'); tr(P P') = ‖Uᵀ V‖_F² = tr(G) for the Gram G = (Uᵀ V)ᵀ (Uᵀ V); then the elementary AM-GM k ∏ tᵢ ≤ ∑ tᵢ over the eigenvalues tᵢ ∈ [0, 1] of G, with ∏ tᵢ = det G = det(Uᵀ V)².

                      The product of singular values is the L2 operator norm of the compound matrix. Combining the singular-value bridge with the compound identity: ∏_{i<k} σᵢ(toEuclideanLin M) = ‖C_k(M)‖.