PaperlayerView exact revision on arXiv

Source-linked AI summary

Tight p-fusion frames

Christine Bachoc, Martin Ehler

arXiv:1201.1798v2math.NAcs.IT

TL;DR

The paper sharpens fusion frames by introducing p-fusion frames and studies their structure, bounds, and constructions. Using harmonic analysis on Grassmann spaces, it characterizes tight p-fusion frames and relates them to designs and cubature formulas.

  • Problem

    The paper asks how to sharpen fusion frames through p-fusion frames and analyze the resulting objects, including bounds and structural connections.

  • Method

    The paper uses harmonic analysis on Grassmann spaces, multivariate Jacobi polynomials evaluated at principal angles, and the p-fusion frame potential to study tight p-fusion frames.

  • Results

    The paper derives a generalized simplex bound for subspaces of varying dimensions, with equality characterized by tightness and equiangularity, and relates tight p-fusion frames to cubature formulas.

  • Takeaways & Limitations

    Tight p-fusion frames provide a framework connecting sharpened fusion-frame structures with designs and cubature formulas in Grassmann spaces.

  • Takeaways & Limitations

    Existence statements are nonconstructive, so explicit constructions are needed to realize the discussed objects.

Abstract

from arXiv · show

Fusion frames enable signal decompositions into weighted linear subspace components. For positive integers p, we introduce p-fusion frames, a sharpening of the notion of fusion frames. Tight p-fusion frames are closely related to the classical notions of designs and cubature formulas in Grassmann spaces and are analyzed with methods from harmonic analysis in the Grassmannians. We define the p-fusion frame potential, derive bounds for its value, and discuss the connections to tight p-fusion frames.

1. Introduction

The paper develops p-fusion frames as a generalization of fusion frames, connecting tight versions to harmonic analysis, cubature formulas, and designs in Grassmann spaces. It also establishes generalized bounds and constructions for these objects.

  • 1. Introduction: A generalized simplex bound applies to subspaces with varying dimensions, with equality exactly when the fusion frame is tight and equiangular.The bound also generalizes Gerzon’s bound for the maximal number of equiangular lines.
  • 1. Introduction: The p-fusion frame potential extends the fusion frame potential and approximates the largest inner product among distinct subspaces.A derived bound approaches the simplex bound in the limit.
  • 1. Introduction: p-fusion frames generalize ordinary fusion frames, with tight p-fusion frames analyzed using harmonic analysis on Grassmann spaces.For equal-dimensional subspaces, the paper characterizes tightness through multivariate Jacobi polynomials evaluated at principal angles.
  • 1. Introduction: Cubatures of strength 2p are characterized as minimizers of the p-fusion frame potential, while tight p-fusion frames coincide with them only when p = 1.For p ≥2, cubatures of strength 2p are stronger than tight p-fusion frames.
  • 1. Introduction: Tight p-fusion frames exist for every integer p ≥1, and the paper gives constructions from finite orthogonal-group orbits, projective-space designs, and concatenation-like methods.The orbit construction has previously been used to derive designs in Grassmann spaces.

2. Fusion frames

Fusion frames represent signals through projections onto collections of linear subspaces. Their frame operator provides reconstruction, and tightness simplifies that operator to a scalar multiple of the identity.

  • 2. Fusion frames: The real Grassmann space Gk,d consists of all k-dimensional subspaces of R^d, while G_d is the union of these spaces for k = 1 through d−1.These spaces provide the geometric setting for the subspace collections studied later.
  • 2. Fusion frames: A fusion frame is a collection of nontrivial subspaces satisfying bounds determined by positive constants A and B.The paper assumes each subspace is neither {0} nor the whole space.
  • 2. Fusion frames: The inner product between projection operators is defined as ⟨P,Q⟩ = trace(PQ), linking subspace geometry to the fusion-frame condition.For a unit vector x, the squared projection norm can be expressed using this operator inner product.
  • 2. Fusion frames: Fusion frames project signals onto collections of linear subspaces, supporting applications such as sensor-network and computing-cluster representations.Frames can also reduce noise and support sparse representations.
  • 2. Fusion frames: The fusion frame operator is positive, self-adjoint, and invertible, yielding reconstruction through weighted projected components.For a tight fusion frame, the operator becomes S = AId, producing a simpler reconstruction formula.

3. The simplex bound and equiangular fusion frames

The section extends simplex-bound results from equidimensional to weighted, arbitrary-dimensional subspaces and characterizes equality through tightness and equiangularity. It also gives an upper bound on the number of equiangular subspaces and introduces the p-fusion frame potential as a tool for the generalized bound.

  • 3.1. The simplex bound for subspaces of equal dimension: The simplex bound lower-bounds the maximum projector inner product for equidimensional subspaces, with equality characterizing tight equidistance fusion frames.For equal-dimensional subspaces, equidistance with respect to chordal distance is equivalent to equiangularity.
  • 3.2. The simplex bound for subspaces of arbitrary dimension: The p-fusion frame potential provides an ℓp approximation to the largest pairwise projector inner product and yields a bound converging to the simplex bound.For p = 1 it extends the usual fusion frame potential, while larger p emphasize the largest pairwise inner products.
  • 3.2. The simplex bound for subspaces of arbitrary dimension: The generalized simplex bound extends the result to weighted subspaces whose dimensions may vary, with equality if and only if the fusion frame is tight and equiangular.This extension replaces equidistance with equiangularity for non-equidimensional subspaces.
  • 3.2. The simplex bound for subspaces of arbitrary dimension: For p = 1, equality in the potential bound characterizes tight fusion frames; for 1 < p < ∞, equality characterizes equiangular tight fusion frames.The equality condition follows from the corresponding norm and Cauchy–Schwarz equality conditions.
  • 3.3. The maximal number of equiangular subspaces: The section extends upper bounds on equiangular lines and equal-dimensional subspaces to equiangular subspaces of arbitrary dimensions.The proof uses linear independence of projector matrices and a Gram-matrix rank argument.

4. Tight p-fusion frames

p-fusion frames replace squared projection terms with 2p-th powers, generalizing fusion frames and reducing to tight p-frames for one-dimensional subspaces. The section establishes their defining conditions, invariances, and relations across orders.

  • Definition: p-fusion frames generalize fusion frames by replacing squared projection terms with 2p-th powers for a positive integer p.The case p = 1 recovers ordinary fusion frames.
  • Definition: Tight p-fusion frames are p-fusion frames whose lower and upper bounds coincide, while equal unit weights may be omitted from the notation.When all subspaces are one-dimensional, the construction is called a tight p-frame.
  • Frame bound: The tight-frame constant is uniquely determined by the orthogonal-group-invariant Grassmann measure and the integral of projection powers.The defining integral is independent of the selected vector or subspace because the measures are O(Rd)-invariant.
  • Relations: Reweighting a tight p-fusion frame produces tight p′-fusion frames for every 1 ≤ p′ ≤ p.The supplied passage gives the new weights as ω̃j = ωj(p − 1 + dim(Vj)/2).
  • Relations: Iteration preserves tightness across lower orders, and the union of tight p-fusion frames remains tight.The section also states that orthogonal complements of the subspaces retain tightness under the relevant conditions.

5. Equidimensional tight p-fusion frames, cubature formulas and the p-fusion frame potential

For equal-dimensional subspaces, harmonic analysis on Grassmannians characterizes tight p-fusion frames through principal angles and connects them to cubatures and the p-fusion frame potential. The section also gives bounds, existence results, and structural relations.

  • 5.2. Characterization by principal angles: Harmonic analysis on Gk,d characterizes tight p-fusion frames through principal angles between subspaces and zonal spherical polynomials.The principal-angle variables yi = cos2(θi) determine the orbit of a pair of subspaces, while the relevant polynomials are symmetric in these variables.
  • 5.1. A closed formula for the tight p-fusion frame bound: The tight-frame bound has an explicit formula obtained using Laplace-operator calculations and depends on the common subspace dimension and total weight.The section assumes k ≤ d/2 after possibly replacing subspaces by orthogonal complements.
  • 5.1. A closed formula for the tight p-fusion frame bound: Tight p-fusion frames remain tight at every lower order and after taking orthogonal complements.The lower-order statement holds for all 1 ≤ p′ ≤ p under the stated equal-dimension setting.
  • 5.2. Characterization by principal angles: The Grassmannian harmonic-analysis framework relies on irreducible decompositions of L2(Gk,d), including representations indexed by partitions and associated polynomial spaces.These decompositions support the principal-angle characterization and the positive-definiteness arguments.
  • 5.2. Characterization by principal angles: Theorem 5.3 makes the tight p-fusion-frame conditions equivalent to polynomial and principal-angle criteria.Its characteristic property involves only the principal angles of pairs (Vi, Vj).
  • 5.3. Cubature formulas and frame potential: The p-fusion frame potential admits a lower bound whose equality conditions are expressed by cubature identities and constant weighted projection-power sums.The supplied results include the constants Tk,d(p) and Al in the corresponding identities.
  • 5.3. Cubature formulas and frame potential: Tight p-fusion frames exist with cardinality bounded by the dimension of the relevant polynomial space, while strength-4p cubatures require at least dim(Pol≤2p(Gk,d)) points.A strength-2p cubature exists with n ≤ dim(Pol≤2p(Gk,d)) − 1.

6. Some constructions of tight p-fusion frames

The paper presents three constructions of tight p-fusion frames: finite-group orbits, projective-space designs, and a composition method that combines frames across dimensions. Each yields explicit structural examples under stated conditions.

  • Finite-group orbits: Finite-group orbits provide equal-weight tight p-fusion frames with strong symmetry, especially for small values of p.The construction uses orbits of finite subgroups of O(Rd).
  • Finite-group orbits: A finite subgroup G of O(Rd) yields tight p-fusion frames on every Grassmann space exactly when the equivalent conditions of Theorem 6.1 hold.One condition is that every orbit G.V is a tight p-fusion frame for all 1 ≤ k < d and V ∈ Gk,d.
  • Finite-group orbits: For p = 1, the orbit criterion reduces to irreducibility of the action on Rd; for higher p, examples include Weyl groups of specified root systems.The listed examples include A2, D4, E6, E7 for p = 2, E8 for p = 3, and H4 for p = 5.
  • Projective-space designs: The projective-design construction expresses the design condition through powers of t(P1, P2), the squared absolute Hermitian inner product between normalized representatives.This yields the tight p-fusion identity for projections onto the associated real subspaces.
  • Projective-space designs: Projective p-designs induce equal-dimension tight p-fusion frames: complex designs give dimension 2 in R2d, and quaternionic designs give dimension 4 in R4d.The correspondence follows by identifying projective spaces over C and H with real spaces and matching projection norms to projective inner products.
  • Composition construction: If F0 and F1 are tight p-fusion frames, fitting copies of F0 inside the subspaces of F1 produces another tight p-fusion frame.The construction combines frames of different dimensions through fixed isometries.
  • Composition construction: Using a tight p-frame in dimension ℓ inside the composition construction yields a tight p-frame in dimension d with nm elements.The resulting frame can be viewed either as an extension to a larger space or as a refinement by subdividing subspaces.

7. An unrestricted lower bound for the p-fusion frame potential

For subspaces of varying dimensions, the paper extends the p-fusion frame-potential analysis using positive-definite functions and the O(Rd)-decomposition of L2(Gd). This yields a lower-bound framework that accounts for representation multiplicities.

  • Unrestricted lower bound: The unrestricted lower-bound analysis removes the equal-dimension assumption and uses the O(Rd)-decomposition of L2(Gd).Nontrivial multiplicities of irreducible representations create difficulties absent from the equal-dimensional case.
  • Positive-definite decomposition: The kernel sp(V, W) = ⟨PV, PW⟩p decomposes as F0 + F1, with each component positive definite and lying in complementary tensor-product spaces.Specifically, F0 ∈ L⊗L and F1 ∈ L⊥⊗L⊥.
  • Positive-definite decomposition: The decomposition supports the lower-bound proof by separating the contribution represented in L⊗L from its orthogonal complement.The supplied proof uses positive definiteness of the two components.
  • Mixed-dimensional constants: For mixed dimensions, the constants Tl,k,d(p) are determined by the constant terms in expansions of (y1 + · · · + yk)p in zonal intertwining polynomials.These polynomials are orthogonal for the induced measures on principal-angle variables.
  • Mixed-dimensional constants: The mixed-dimensional constants satisfy Tl,k,d(1) = lk/d and are rational functions of l, k, and p.The rationality follows from the rational coefficients of the relevant polynomials.
  • Recovery of earlier bounds: When all subspaces have dimension k, the unrestricted framework recovers the earlier p-fusion frame-potential lower bound.The supplied passages state that the p = 1 case also recovers the earlier bound in Theorem 3.4.
Loading 1201.1798v2…