Documentation

LeanPool.ViscositySolutionTheory.ViscositySolns.Comparison.MatrixInequalities.BlockStructure

Ported for Lean Pool: imports relocated and code adapted to Lean/Mathlib v4.34.0.

Matrix inequalities for comparison (BlockStructure) #

Part of the block-matrix development used in the doubling-of-variables proof of comparison. Split from MatrixInequalities.lean; see the umbrella module docstring.

@[reducible, inline]

Hessian matrices on the product coordinate space R^n × R^n.

Equations
Instances For
    @[reducible, inline]

    Coordinate vectors on the product space R^n × R^n.

    Equations
    Instances For

      Convert a pair of R^n points into one point of the product coordinate space.

      Equations
      Instances For

        Identify the product coordinate space R^n × R^n, indexed by Fin n ⊕ Fin n, with the coordinate space indexed by Fin (n + n).

        Equations
        Instances For
          @[simp]

          Reindex a Hessian on R^(n+n) as a block Hessian on R^n × R^n.

          Equations
          Instances For

            Reindex a block Hessian on R^n × R^n as a Hessian on R^(n+n).

            Equations
            Instances For

              Reindexing a Hermitian Hessian on R^(n+n) as a block Hessian on R^n × R^n preserves Hermitian symmetry.

              Reindexing a Hermitian block Hessian on R^n × R^n as a Hessian on R^(n+n) preserves Hermitian symmetry.

              The coordinate reindexing from Point (n + n) to BlockPoint n preserves dot products.

              Matrix-vector multiplication is preserved by reindexing a Hessian on R^(n+n) as a block Hessian on R^n × R^n.

              The quadratic form of a Hessian is unchanged by the coordinate equivalence between R^(n+n) and R^n × R^n.

              The first coordinate of a point of R^n × R^n.

              Equations
              Instances For

                The second coordinate of a point of R^n × R^n.

                Equations
                Instances For
                  theorem ViscositySolns.quadraticForm_mono_of_matrix_le {m : Type u_1} [Fintype m] {A B : Matrix m m ℝ} (hAB : A ≤ B) (z : m → ℝ) :

                  Loewner order implies the corresponding inequality of quadratic forms: if A ≤ B, then for every vector z, z · A z ≤ z · B z.

                  theorem ViscositySolns.matrix_le_submatrix {m : Type u_1} {k : Type u_2} {A B : Matrix m m ℝ} (hAB : A ≤ B) (e : k → m) :
                  A.submatrix e e ≤ B.submatrix e e

                  Taking the same principal submatrix on both sides preserves the Loewner order.

                  The coordinate reindexing from R^(n+n) to R^n × R^n preserves the Loewner order.

                  The block diagonal matrix [X 0; 0 -Y].

                  Equations
                  Instances For

                    The block identity matrix [I 0; 0 I].

                    Equations
                    Instances For

                      The block identity matrix is the ordinary identity matrix on Fin n ⊕ Fin n.

                      The identity matrix on R^(n+n) becomes the block identity matrix on R^n × R^n.

                      A scalar multiple of the identity matrix on R^(n+n) becomes the same scalar multiple of the block identity matrix on R^n × R^n.

                      Reindex the two-sided matrix bounds supplied by the semiconvex matrix theorem from R^(n+n) to R^n × R^n.

                      In quantified mathematical form, if a I ≤ A ≤ B as matrices on R^(n+n), then, after identifying R^(n+n) with R^n × R^n, a [I 0; 0 I] ≤ A_block ≤ B_block.

                      The Hessian matrix of the quadratic penalty (α / 2) * ∑ i, (x i - y i)^2, written in block form.

                      Equations
                      Instances For

                        The square of the Hessian matrix of the coordinate quadratic penalty.

                        In standard mathematical notation, if

                        A = [ α I -α I ], [ -α I α I ]

                        then A^2 = 2α A.

                        The quadratic-penalty Hessian algebra used in the proof of Ishii's lemma.

                        In standard mathematical notation, if

                        A = [ α I -α I ], [ -α I α I ]

                        and α ≠ 0, then A + α⁻¹ A^2 = 3A.

                        The lower matrix bound -(3 * α) [I 0; 0 I].

                        Equations
                        Instances For

                          The upper matrix bound 3 times the Hessian of the quadratic penalty.

                          Equations
                          Instances For
                            def ViscositySolns.IshiiMatrixRelation {n : ℕ} (α : ℝ) (_x _y : Point n) (X Y : Hessian n) :

                            The matrix relation appearing in the comparison theorem.

                            For α > 0, matrices X and Y satisfy this relation if [X 0; 0 -Y] lies between the two block matrices displayed at the top of this file. The points x and y are arguments so that this relation can be used as a ComparisonMatrixRelation; the inequality itself does not depend on them.

                            Equations
                            • One or more equations did not get rendered due to their size.
                            Instances For
                              def ViscositySolns.doubledSameVector {n : ℕ} (z : Fin n → ℝ) :
                              Fin n ⊕ Fin n → ℝ

                              The vector (z, z) in the product coordinate space R^n × R^n.

                              Equations
                              Instances For

                                If X and Y satisfy IshiiMatrixRelation α x y X Y, then for every z : Fin n ⊕ Fin n -> Real,

                                zᵀ (ishiiLowerBlock α) z ≤ zᵀ (comparisonBlockDiagonal X Y) z.

                                If X and Y satisfy IshiiMatrixRelation α x y X Y, then for every z : Fin n ⊕ Fin n -> Real,

                                zᵀ (comparisonBlockDiagonal X Y) z ≤ zᵀ (ishiiUpperBlock α) z.

                                For every vector z : Fin n -> Real, the quadratic form of comparisonBlockDiagonal X Y on (z, z) is zᵀ X z - zᵀ Y z.

                                theorem ViscositySolns.dotProduct_self_nonneg {m : Type u_1} [Fintype m] (x : m → ℝ) :
                                0 ≤ x ⬝ᵥ x

                                The coordinate dot product of a vector with itself is nonnegative.

                                theorem ViscositySolns.hessian_quadraticForm_le_of_entrywise_abs_le {n : ℕ} (A : Hessian n) {eta : ℝ} (heta : 0 ≤ eta) (hA : ∀ (i j : Fin n), |A i j| ≤ eta) (v : Point n) :
                                Matrix.mulVec A v ⬝ᵥ v ≤ eta * ↑n * ↑n * v ⬝ᵥ v

                                A finite entrywise bound on a Hessian gives an upper bound for its quadratic form.

                                In quantified mathematical form, if η ≥ 0 and |Aᵢⱼ| ≤ η for every matrix entry, then for every vector v,

                                ⟪Av, v⟫ ≤ η n^2 ⟪v, v⟫.

                                Every real symmetric Hessian is bounded above by a scalar multiple of the identity in the Loewner order.

                                In quantified mathematical form, for every symmetric matrix A on R^n, there exists μ ∈ R such that A ≤ μ I.

                                Every real symmetric block Hessian on R^n × R^n is bounded above by a scalar multiple of the block identity in the Loewner order.

                                In quantified mathematical form, for every symmetric matrix A on R^n × R^n, there exists μ ∈ R such that

                                A ≤ μ [I 0; 0 I].

                                theorem ViscositySolns.dotProduct_le_young {m : Type u_1} [Fintype m] {epsilon : ℝ} (hepsilon : 0 < epsilon) (u v : m → ℝ) :
                                u ⬝ᵥ v ≤ epsilon / 2 * u ⬝ᵥ u + epsilon⁻¹ / 2 * v ⬝ᵥ v

                                Young's inequality for the coordinate dot product.

                                In standard mathematical terms, for ε > 0 and finite-coordinate vectors u and v,

                                ⟪u, v⟫ ≤ (ε / 2) * |u|^2 + (ε⁻¹ / 2) * |v|^2.