Quantitative edges of a positive and negative block spike
Abstract
A rank-one diagonal spike in a Hermitian block matrix admits two-sided spectral-edge estimates. For a unit v, Hermitian S with Sv = 0, L >= 1 and L >= ‖K X S‖, every t >= 96L has an error at most 891L^4/t^3 at each edge, and 1782L^4/t^3 in their sum.
Definition 1.1 (Top spectral edge).
Formalization. D5/S3/Quantum/BlockNorm/SpikeEdgeEstimate.edgeMax (✓ std3).
Source. Repository-derived.
Commentary.
The real supremum of the real spectrum. Hermitian matrices on nonempty finite types have a nonempty compact real spectrum.
Definition 1.2 (Bottom spectral edge).
Formalization. D5/S3/Quantum/BlockNorm/SpikeEdgeEstimate.edgeMin (✓ std3).
Source. Repository-derived.
Commentary.
The real infimum of the real spectrum, using the same real-spectrum convention as the top edge.
Definition 1.3 (Hermitian block pencil).
Formalization. D5/S3/Quantum/BlockNorm/SpikeEdgeEstimate.K (✓ std3).
Source. Repository-derived.
Commentary.
K X D is Matrix.fromBlocks D X (Matrix.conjTranspose X) (-D); it is Hermitian when D is Hermitian.
Lemma 1.4 (Upper spectral shift).
Proof. Machine-checked in Lean as D5/S3/Quantum/BlockNorm/SpikeEdgeEstimate.upper_shift_iff (✓ std3). ∎
Source. Repository-derived.
Commentary.
A scalar upper spectral bound is exactly positivity of the scalar shift minus the Hermitian matrix.
Lemma 1.5 (Action of a rank-one matrix).
Proof. Machine-checked in Lean as D5/S3/Quantum/BlockNorm/SpikeEdgeEstimate.outer_action (✓ std3). ∎
Source. Repository-derived.
Commentary.
Matrix.vecMulVec and WithLp.ofLp are the actual matrix and vector carriers. The scalar is the complex inner product.
Lemma 1.6 (Hermitian rank-one matrix).
Proof. Machine-checked in Lean as D5/S3/Quantum/BlockNorm/SpikeEdgeEstimate.outer_hermitian (✓ std3). ∎
Source. Repository-derived.
Commentary.
The conjugate transpose of the rank-one matrix equals the matrix itself.
Definition 1.7 (First edge-sum coefficient).
Formalization. D5/S3/Quantum/BlockNorm/SpikeEdgeEstimate.defect1 (✓ std3).
Source. Repository-derived.
Commentary.
The first coefficient measures the squared norm difference between the adjoint action and the original action.
Definition 1.8 (Second edge-sum coefficient).
Formalization. D5/S3/Quantum/BlockNorm/SpikeEdgeEstimate.defect2 (✓ std3).
Source. Repository-derived.
Commentary.
The second coefficient is the difference of the real S-pairings, with the original action first and the adjoint action second.
Theorem 1.9 (Two-sided positive-edge enclosure).
Proof. Machine-checked in Lean as D5/S3/Quantum/BlockNorm/SpikeEdgeEstimate.positive_spike_enclosure (✓ std3). ∎
Source. Repository-derived.
Commentary.
The normalized test vector is formed from e + εr₁ + ε²r₂ at ε = 1/t. Its Rayleigh quotient supplies the lower edge bound; the complement gap and residual-energy estimate supply the upper bound. Both estimates hold at every finite t >= 96L.
Theorem 1.10 (Two-sided edge-sum enclosure).
Proof. Machine-checked in Lean as D5/S3/Quantum/BlockNorm/SpikeEdgeEstimate.edge_sum_bound (✓ std3). ∎
Source. Repository-derived.
Commentary.
Unitary block reflection transfers the positive-edge estimate to the negative edge. Their leading ±t terms cancel. The sum has the corrected coefficient ‖X* v‖² - ‖X v‖² at order 1/t and the difference of the S-pairings at order 1/t².
References
- Truth anchor:
D5/S3/Quantum/BlockNorm/SpikeEdgeEstimate.K - Truth anchor:
D5/S3/Quantum/BlockNorm/SpikeEdgeEstimate.defect1 - Truth anchor:
D5/S3/Quantum/BlockNorm/SpikeEdgeEstimate.defect2 - Truth anchor:
D5/S3/Quantum/BlockNorm/SpikeEdgeEstimate.edgeMax - Truth anchor:
D5/S3/Quantum/BlockNorm/SpikeEdgeEstimate.edgeMin - Truth anchor:
D5/S3/Quantum/BlockNorm/SpikeEdgeEstimate.edge_sum_bound - Truth anchor:
D5/S3/Quantum/BlockNorm/SpikeEdgeEstimate.outer_action - Truth anchor:
D5/S3/Quantum/BlockNorm/SpikeEdgeEstimate.outer_hermitian - Truth anchor:
D5/S3/Quantum/BlockNorm/SpikeEdgeEstimate.positive_spike_enclosure - Truth anchor:
D5/S3/Quantum/BlockNorm/SpikeEdgeEstimate.upper_shift_iff - Dependency: D5/S3/Weil/GroundMode/ResidualDrivenProjectiveEnergy