Documentation

MIPRE.Background.LIDT.MIPStarRE.LDT.MakingMeasurementsProjective.QXPLayerIdentities.PositiveGram.Completion

Section 5 — Positive-Gram completion #

Completion and polar-extension lemmas for the positive spectral rows of a right Gram operator.

theorem MIPStarRE.LDT.MakingMeasurementsProjective.exists_unitaryGroup_rows_extending_orthonormal {κ : Type u_1} {μ : Type u_2} [Fintype μ] [DecidableEq μ] (row : κEuclideanSpace μ) (hrow : Orthonormal row) (e : κ μ) :
∃ (U : (Matrix.unitaryGroup μ )), ∀ (i : κ) (r : μ), U (e i) r = (row i).ofLp r

A square unitary group element whose selected rows are a prescribed orthonormal family.

The row equations are stated pointwise so that later matrix calculations can rewrite entries without unfolding the chosen orthonormal basis.

theorem MIPStarRE.LDT.MakingMeasurementsProjective.exists_unitaryGroup_with_positive_gram_spectrum_rows {μ : Type u_1} {ι : Type u_2} [Fintype μ] [DecidableEq μ] [Fintype ι] [DecidableEq ι] (X : Matrix μ ι ) (Q : Matrix ι ι ) (hQ : Q.IsHermitian) (hgram : X.conjTranspose * X = Q) (e : { i : ι // 0 < hQ.eigenvalues i } μ) :
∃ (U : (Matrix.unitaryGroup μ )), ∀ (i : { i : ι // 0 < hQ.eigenvalues i }) (r : μ), U (e i) r = positiveGramSpectrumImageRows X Q hQ i r

The positive Gram image rows can be embedded into a square unitary group element.

The hypothesis e chooses distinct row positions for the strictly positive eigenvalues of the Gram operator. The theorem then completes the corresponding normalized image rows to a unitary element on the row space. This is the basis-extension ingredient needed to turn the positive spectral part into the left unitary factor of the rectangular polar construction.

theorem MIPStarRE.LDT.MakingMeasurementsProjective.exists_unitaryGroup_with_positive_gram_spectrum_rows_of_card {μ : Type u_1} {ι : Type u_2} [Fintype μ] [DecidableEq μ] [Fintype ι] [DecidableEq ι] (X : Matrix μ ι ) (Q : Matrix ι ι ) (hQ : Q.IsHermitian) (hgram : X.conjTranspose * X = Q) :
∃ (e : { i : ι // 0 < hQ.eigenvalues i } μ) (U : (Matrix.unitaryGroup μ )), ∀ (i : { i : ι // 0 < hQ.eigenvalues i }) (r : μ), U (e i) r = positiveGramSpectrumImageRows X Q hQ i r

Existential form of exists_unitaryGroup_with_positive_gram_spectrum_rows.

The normalized positive Gram image rows have cardinality at most the row dimension, so one may choose distinct row positions and then extend those rows to a square unitary group element.

theorem MIPStarRE.LDT.MakingMeasurementsProjective.exists_rectangular_coisometry_extending_orthonormal_rows {κ : Type u_1} {μ : Type u_2} {ν : Type u_3} [Fintype μ] [DecidableEq μ] [Fintype ν] (row : κEuclideanSpace ν) (hrow : Orthonormal row) (e : κ μ) (hcard : Fintype.card μ Fintype.card ν) :
∃ (W : Matrix μ ν ), W * W.conjTranspose = 1 ∀ (i : κ) (r : ν), W (e i) r = (row i).ofLp r

Extend an orthonormal row family to a rectangular coisometry.

The embedding e : κ ↪ μ specifies the rows of the rectangular matrix which must agree with the prescribed family. The dimension hypothesis Fintype.card μ ≤ Fintype.card ν supplies enough room to complete these rows to an orthonormal family indexed by all of μ.

theorem MIPStarRE.LDT.MakingMeasurementsProjective.exists_rectangular_coisometry_with_positive_gram_spectrum_right_rows {μ : Type u_1} {ι : Type u_2} [Fintype μ] [DecidableEq μ] [Fintype ι] [DecidableEq ι] (Q : Matrix ι ι ) (hQ : Q.IsHermitian) (e : { i : ι // 0 < hQ.eigenvalues i } μ) (hcard : Fintype.card μ Fintype.card ι) :
∃ (W : Matrix μ ι ), W * W.conjTranspose = 1 ∀ (i : { i : ι // 0 < hQ.eigenvalues i }) (r : ι), W (e i) r = positiveGramSpectrumRightRows Q hQ i r

Extend the positive Gram right-singular rows to a rectangular coisometry.

The selected rows are the conjugate eigenvector rows of the right Gram operator. The cardinality hypothesis is the rectangular dimension condition: there must be enough columns to complete the prescribed rows to an orthonormal row family indexed by the auxiliary row space.

Left multiplication by the transpose of a unitary group element preserves row coisometries.

This is the coisometry half of the polar-extension calculation for the candidate Xhat = Uᵀ W: the rectangular factor W has orthonormal rows, and the square factor U merely changes the orthonormal basis of the row space.

theorem MIPStarRE.LDT.MakingMeasurementsProjective.adjoint_image_eq_zero_of_unitary_positive_gram_completion_row {μ : Type u_1} {ι : Type u_2} [Fintype μ] [DecidableEq μ] [Fintype ι] [DecidableEq ι] (X : Matrix μ ι ) (Q : Matrix ι ι ) (hQ : Q.IsHermitian) (hQ_pos : Q.PosSemidef) (hgram : X.conjTranspose * X = Q) (e : { i : ι // 0 < hQ.eigenvalues i } μ) (U : Matrix μ μ ) (hU_left : U * U.conjTranspose = 1) (hU_rows : ∀ (i : { i : ι // 0 < hQ.eigenvalues i }) (r : μ), U (e i) r = positiveGramSpectrumImageRows X Q hQ i r) (a : μ) (ha : aSet.range e) :
(LinearMap.adjoint (Matrix.toEuclideanLin X)) (WithLp.toLp 2 fun (r : μ) => U a r) = 0

Completion rows of the left unitary outside the positive Gram spectrum are killed by X†.

The selected rows of U are the normalized positive images of the Gram eigenvectors. If a row index is not one of the selected positive spectral indices, the row-orthogonality of U makes that row orthogonal to all positive Gram images. The adjoint-kernel lemma then gives the stated vanishing.

theorem MIPStarRE.LDT.MakingMeasurementsProjective.positive_gram_selected_left_unitary_mixed_column {μ : Type u_1} {ι : Type u_2} [Fintype μ] [Fintype ι] [DecidableEq ι] (X : Matrix μ ι ) (Q : Matrix ι ι ) (hQ : Q.IsHermitian) (e : { i : ι // 0 < hQ.eigenvalues i } μ) (U : Matrix μ μ ) (hU_rows : ∀ (i : { i : ι // 0 < hQ.eigenvalues i }) (r : μ), U (e i) r = positiveGramSpectrumImageRows X Q hQ i r) (i : { i : ι // 0 < hQ.eigenvalues i }) (r : ι) :

The selected columns of X† Uᵀ are the positive Gram-image mixed columns.

If the selected rows of U are the normalized images of the positive Gram eigenvectors, then multiplying by X† and transposing U recovers precisely the column matrix X† Rᵀ, where R is the positive Gram-image row matrix.

theorem MIPStarRE.LDT.MakingMeasurementsProjective.positive_gram_completion_left_unitary_mixed_column_eq_zero {μ : Type u_1} {ι : Type u_2} [Fintype μ] [DecidableEq μ] [Fintype ι] [DecidableEq ι] (X : Matrix μ ι ) (Q : Matrix ι ι ) (hQ : Q.IsHermitian) (hQ_pos : Q.PosSemidef) (hgram : X.conjTranspose * X = Q) (e : { i : ι // 0 < hQ.eigenvalues i } μ) (U : Matrix μ μ ) (hU_left : U * U.conjTranspose = 1) (hU_rows : ∀ (i : { i : ι // 0 < hQ.eigenvalues i }) (r : μ), U (e i) r = positiveGramSpectrumImageRows X Q hQ i r) (a : μ) (ha : aSet.range e) (r : ι) :

The complementary columns of X† Uᵀ vanish.

This is the matrix-column form of adjoint_image_eq_zero_of_unitary_positive_gram_completion_row. It records that the arbitrary rows used to complete the positive image rows to a full unitary make no contribution after left multiplication by X†.

theorem MIPStarRE.LDT.MakingMeasurementsProjective.positive_gram_polar_extension_mixed_eq_positive_rows {μ : Type u_1} {ι : Type u_2} [Fintype μ] [DecidableEq μ] [Fintype ι] [DecidableEq ι] (X : Matrix μ ι ) (Q : Matrix ι ι ) (hQ : Q.IsHermitian) (hQ_pos : Q.PosSemidef) (hgram : X.conjTranspose * X = Q) (e : { i : ι // 0 < hQ.eigenvalues i } μ) (U : Matrix μ μ ) (W : Matrix μ ι ) (hU_left : U * U.conjTranspose = 1) (hU_rows : ∀ (i : { i : ι // 0 < hQ.eigenvalues i }) (r : μ), U (e i) r = positiveGramSpectrumImageRows X Q hQ i r) (hW_rows : ∀ (i : { i : ι // 0 < hQ.eigenvalues i }) (r : ι), W (e i) r = positiveGramSpectrumRightRows Q hQ i r) :

The polar-extension mixed product is the positive-spectrum mixed product.

The candidate Xhat = Uᵀ W has selected rows matching the positive left and right singular rows. After expanding the middle index, the selected part is exactly the positive-spectrum square-root product, while every complementary row is killed by X†.

theorem MIPStarRE.LDT.MakingMeasurementsProjective.positive_gram_polar_extension_mixed_eq_sqrt_unitaryGroup {μ : Type u_1} {ι : Type u_2} [Fintype μ] [DecidableEq μ] [Fintype ι] [DecidableEq ι] (X : Matrix μ ι ) (Q : Matrix ι ι ) (hQ : Q.IsHermitian) (hQ_pos : Q.PosSemidef) (hgram : X.conjTranspose * X = Q) (e : { i : ι // 0 < hQ.eigenvalues i } μ) (U : (Matrix.unitaryGroup μ )) (W : Matrix μ ι ) (hU_rows : ∀ (i : { i : ι // 0 < hQ.eigenvalues i }) (r : μ), U (e i) r = positiveGramSpectrumImageRows X Q hQ i r) (hW_rows : ∀ (i : { i : ι // 0 < hQ.eigenvalues i }) (r : ι), W (e i) r = positiveGramSpectrumRightRows Q hQ i r) :

The polar-extension mixed product is the square root of the Gram operator.

This combines the finite-dimensional completion calculation for Xhat = Uᵀ W with the positive-spectrum square-root identity.