Alphabeta Math
TheoremStatement: Literature-sourcedProof: AI-adaptedPipeline-generatedjudge pass (gpt-6-luna)
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.

Conforming Galerkin invariant subspaces converge across an isolated cluster

Statement

Assume Countable Choice and the compact coercive form setting of Discrete spectrum and min–max principle for a compact coercive form. Suppose a continuous eigenvalue λ occupies the isolated cluster of indices r,…,s, so λr=⋯=λs=λ and λs+1>λ, with λr−1<λ when r>1. Let E:=span⁡{er,…,es}, Es:=span⁡{e1,…,es}, and let Vh⊆V be conforming with dim⁡Vh≥s. Put δs:=δa(Es,Vh) using Best-approximation gap of a finite-dimensional eigenspace. If δs is sufficiently small that the eigenvalue theorem Conforming finite-element eigenvalues have a quadratic best-approximation error applies at index s and, when r>1, at index r−1, and Csδs2<12(λs+1−λ) and, when r>1, Cr−1δs2<12(λ−λr−1), then the discrete Ritz eigenspace Eh associated with indices r,…,s has dimension s−r+1 and δa(E,Eh)≤Cr,s,λ δs. This compares the invariant subspaces across the whole repeated eigenvalue; it does not select individual eigenvectors inside the cluster.

Facts & Assumptions

Given: Countable Choice, the compact coercive form pair (V,H,a,J), the isolated repeated eigenvalue λ, the indices r≤s, the conforming finite-dimensional subspace Vh, and the small best-approximation gap δs.

[F1]

The continuous eigenvectors form an H-orthonormal basis and their energy-normalized versions form an a-orthonormal basis; the Ritz quotient has the continuous min–max principle (Discrete spectrum and min–max principle for a compact coercive form).

[F2]

Discrete Ritz eigenvalues satisfy min–max, and the conforming eigenvalue theorem gives λj≤λh,j≤λj+Cjδj2 when δj is sufficiently small (Finite-dimensional conforming Ritz values satisfy min–max, Conforming finite-element eigenvalues have a quadratic best-approximation error).

[F3]

The energy projection Ph is best approximation and is orthogonal (The conforming energy projection is the best approximation).

[F4]

If a compact self-adjoint operator has a spectral cluster separated from its remaining spectrum by Δ>0, an ε residual on an invariant subspace bounds the directed subspace gap by ε/Δ (A residual and a spectral separation bound an invariant-subspace gap).

[F5]

On a Hilbert space, every bounded linear functional has a unique Riesz representative and the representative has the same norm as the functional (Riesz representation for Hilbert spaces, The operator norm as the least bound and as the unit-sphere or unit-ball supremum); Cauchy--Schwarz holds for every real or complex inner product (Cauchy–Schwarz: ∣⟨x,y⟩∣≤∥x∥ ∥y∥, with equality exactly for dependent pairs).

[F6]

Orthogonality of the energy projection gives ∥Phv∥a2+∥v−Phv∥a2=∥v∥a2, so ∥Ph∥≤1. A bounded operator with range in the finite-dimensional space Vh has a compact closure of its unit-ball image: choose a finite orthonormal basis of Vh by finite Gram--Schmidt, use its real (Rd or R2d) coordinates, and apply Heine--Borel to the closed bounded coordinate closure. The inclusion of that compact set into V is isometric; its ambient closure is a closed subset of this compact set and is compact. The compact-operator definition then gives compactness (The conforming energy projection is the best approximation, Compact linear operator, Heine-Borel in Rn: with the Euclidean metric a subset of Rn is compact if and only if it is closed and bounded, and the proof by bisection uses no choice principle; the same holds on the real line, A subset of a metric space is open in the subspace metric exactly when it is the trace of an open set of the ambient space, and it is compact as a metric space in its own right exactly when every family of open subsets of the ambient space covering it, indexed or not, has finitely many members covering it, A compact subset of a metric space is closed and bounded, A closed subset of a compact metric space is compact).

Proof

technique · direct

Given: The continuous eigenpairs (λj,ej), the isolated cluster λr=⋯=λs=λ, Vh, and δs.

1.1F2givenchoose

Since Er−1⊆Es when r>1, its directed best-approximation gap is at most δs. Choose δs below the smallness thresholds for the upper eigenvalue theorem at s and, if r>1, at r−1. At index s, [F2] gives λh,s≤λ+Csδs2. For r≤j≤s, Ritz ordering gives λh,j≤λh,s, while the conforming lower bound in [F2] gives λh,j≥λj=λ. Hence λ≤λh,j≤λ+Csδs2. When r>1, Ritz ordering and the upper bound at r−1 give λh,j≤λr−1+Cr−1δs2<λ−12(λ−λr−1) for j<r. For each discrete index j>s that exists, [F2] gives λh,j≥λj≥λs+1>λ. Thus exactly the s−r+1 indexed Ritz eigenvalues in the cluster are separated from all other discrete Ritz eigenvalues.

1.2F1F5givenconstruct

Boundedness and coercivity give α∥v∥V2≤a(v,v)≤M∥v∥V2, so the energy norm is equivalent to the Hilbert norm of V and V is complete in the energy norm. Thus (V,a) is a Hilbert space. For fixed u, the map v↦(Ju,Jv)H‾=(Jv,Ju)H is a bounded linear functional on (V,a). By [F5] there is a unique Ku∈V with a(v,Ku)=(Jv,Ju)H for every v; conjugating gives a(Ku,v)=(Ju,Jv)H. Uniqueness in this identity makes K linear. If CJ=∥J∥V→H, then coercivity and Cauchy–Schwarz give ∣a(Ku,v)∣=∣(Ju,Jv)H∣≤(CJ2/α)∥u∥a∥v∥a. Taking the supremum over ∥v∥a=1 and using Riesz norm equality yields ∥Ku∥a≤(CJ2/α)∥u∥a. Hermitian symmetry shows that K is self-adjoint. The continuous eigenrelation in [F1] gives Kej=λj−1ej and K(ej/λj)=λj−1(ej/λj). In particular, on E the operator K acts as μI for μ:=1/λ.

2.1F2F3F5F6step 1.1givenalgebra

Let Ph be the energy projection from [F3] and set Kh:=PhKPh:V→V. It is self-adjoint because Ph is an orthogonal projection and K is self-adjoint. It is bounded as a composition of bounded operators, and its range lies in the finite-dimensional space Vh. To check compactness under the actual compact-closure definition, the image Kh(B‾V) is bounded in Vh. Choose a finite a-orthonormal basis of Vh by finite Gram--Schmidt and identify its real coordinates with Rd or R2d. The closure of this image in Vh is closed and bounded in those coordinates, hence compact by Heine--Borel. Its inclusion into V is isometric, so it is compact and closed in V; therefore the ambient closure of Kh(B‾V) is a closed subset of this compact set and is compact. The compact-operator definition now makes Kh compact [F6]. It is positive because a(Khu,u)=a(KPhu,Phu)=∥JPhu∥H2≥0. For uh,vh∈Vh, a(Khuh,vh)=a(Kuh,vh)=(Juh,Jvh)H. If Khuh=0, taking vh=uh gives ∥Juh∥H2=0, so injectivity of J gives uh=0. Thus Kh∣Vh is positive definite. Moreover, Khuh=μhuh is equivalent by the displayed identity to b(uh,vh)=μha(uh,vh) for every vh∈Vh, the Ritz equation with λh,j=1/μh. Conversely, the same identity and nondegeneracy of a on Vh show that every Ritz eigenvector is an eigenvector of Kh. Its positive eigenspaces are therefore exactly the conforming Ritz eigenspaces. Put μ:=1/λ. If r>1, set γ−:=λ−λr−1; set Δ+:=1/λ−1/λs+1>0. When r>1, put Δ−:=1/(λ−γ−/2)−1/λ>0. Step 1.1 and smallness give a positive distance from μ to all eigenvalues outside the r,…,s Ritz cluster: lower-index reciprocal eigenvalues are at least 1/(λ−γ−/2) when r>1, higher-index reciprocals are at most 1/λs+1 when present, and zero is separated by μ. Define Δr,s,λ:=min⁡{μ,Δ+} if r=1 and Δr,s,λ:=min⁡{μ,Δ+,Δ−} if r>1. The cluster subspace Eh of Kh has dimension s−r+1 and this positive spectral separation from its complement.

3.1F1F3step 2.1givenalgebra

For u∈E, the inclusion E⊆Es and [F3] give ∥Phu−u∥a≤δs∥u∥a. Since Ku=μu, Khu−μu=PhK(Phu−u)+μ(Phu−u). The orthogonal projection Ph has norm at most one, so ∥(Kh−μI)u∥a≤(∥K∥+μ)δs∥u∥a. The constant Δr,s,λ from step 2.1 is therefore independent of δs under the chosen smallness bound.

4.1F4step 2.1step 3.1givenalgebra∎

Apply the residual subspace estimate [F4] to Kh, its separated cluster Eh, and the residual bound from step 3.1. It yields δa(E,Eh)≤∥K∥+μΔr,s,λ δs, where Δr,s,λ>0 is the uniform spectral separation obtained in step 2.1. This is the claimed invariant-subspace estimate, including the case of multiplicity greater than one.

Depends on

Used by

Dependency tree · two levels

80 results within two dependency steps of this one, each drawn at its shortest distance from it. An arrow runs from a result to what uses it, so the chart reads left to right and ends at this result, which carries a heavier outline. Every node is a link to that result. Click elsewhere on the chart to enlarge it.

Sources