Documentation

Atlas.HighDimensionalStatistics.code.Chapter11.Defs_and_Props

Chapter 11: Constant-Coefficient Differential Operators and Tempered Distributions #

This file formalizes the key definitions and results from Chapter 11 on constant-coefficient differential operators acting on tempered distributions.

Main definitions and results #

Constant-coefficient differential operators #

A constant-coefficient differential operator P(D) on tempered distributions is modeled as a continuous ℂ-linear endomorphism of the space of tempered distributions. The requirement that the operator be non-zero is encoded in the structure.

A constant-coefficient linear partial differential operator on S'(E, ℂ), modeled as a continuous linear endomorphism of the space of tempered distributions.

Instances For

    Definition 11.3: Tempered fundamental solution (Goal 77) #

    Definition 11.3 (Tempered fundamental solution): An element E_sol ∈ S'(ℝⁿ) satisfying P(D) E_sol = δ is said to be a (tempered) fundamental solution of P(D).

    Here P is a constant-coefficient differential operator modeled as a continuous ℂ-linear endomorphism of S'(E, ℂ), and δ is the Dirac delta distribution at the origin.

    Instances For

      Proposition 11.1: Schwartz space is weakly dense in S'(Rⁿ) (Goal 75) #

      Proposition 11.1 (Schwartz space is weakly dense in tempered distributions): The subspace S(ℝⁿ) ⊂ S'(ℝⁿ) is weakly dense, i.e., each u ∈ S'(ℝⁿ) is the weak limit of a sequence uⱼ ∈ S(ℝⁿ).

      The proof is not given in the textbook, so we axiomatize this result.

      Proposition 11.2: Weak limits preserve operations (Goal 76) #

      If uⱼ → u and u'ⱼ → u' weakly in S'(ℝⁿ) then:

      all weakly in S'(ℝⁿ). These follow from the fact that the respective operations are continuous linear maps on 𝓢'(E, ℂ) (which carries the pointwise convergence topology), so preservation under limits is automatic.

      theorem Chapter11.prop_11_2_add {E : Type u_1} [NormedAddCommGroup E] [NormedSpace E] (u v : TemperedDistribution E ) (u₀ v₀ : TemperedDistribution E ) (hu : Filter.Tendsto u Filter.atTop (nhds u₀)) (hv : Filter.Tendsto v Filter.atTop (nhds v₀)) :
      Filter.Tendsto (fun (j : ) => u j + v j) Filter.atTop (nhds (u₀ + v₀))

      Goal 76

      Theorem 11.4: Malgrange-Ehrenpreis (Goal 78) #

      Every non-zero constant-coefficient differential operator has a tempered fundamental solution. This is the celebrated Malgrange–Ehrenpreis theorem. The proof is not given in the textbook, so we axiomatize this result.

      Theorem 11.4 (Malgrange–Ehrenpreis): Every non-zero constant-coefficient differential operator P(D) has a tempered fundamental solution, i.e., there exists E ∈ S'(ℝⁿ) such that P(D) E = δ.

      This is a deep existence theorem; the proof is not given in the textbook, so we state it as an axiom.

      Lemma 11.5: Cauchy-Riemann fundamental solution (Goal 79) #

      E(x,y) = (1/2π)(x + iy)⁻¹ is locally integrable, defines E ∈ S'(R²), and is a tempered fundamental solution of ∂̄ (the Cauchy-Riemann operator).

      noncomputable def Chapter11.e₁ :

      Standard basis vector e₁ in ℝ².

      Instances For
        noncomputable def Chapter11.e₂ :

        Standard basis vector e₂ in ℝ².

        Instances For

          The Cauchy-Riemann operator ∂̄ = (1/2)(∂_x + i∂_y), modeled as a continuous linear map on tempered distributions over ℝ².

          Instances For
            noncomputable def Chapter11.cauchyRiemannKernel :

            The function E(x, y) = (1/(2π)) * (x + iy)⁻¹, the Cauchy-Riemann kernel.

            Instances For

              Lemma 11.5 (Cauchy-Riemann fundamental solution): E(x, y) = (1/(2π))(x + iy)⁻¹ is locally integrable and so defines E ∈ S'(ℝ²) by E(φ) = (1/(2π)) ∫_{ℝ²} (x + iy)⁻¹ φ(x,y) dx dy, and E so defined is a tempered fundamental solution of ∂̄, i.e., ∂̄ E = δ.

              The kernel is given by cauchyRiemannKernel and the operator ∂̄ by dbarOp. The proof is not given in the textbook, so we axiomatize this result.

              Definition 11.8: Parametrix and hypoelliptic operators #

              A parametrix for P(D) is F ∈ S'(ℝⁿ) such that P(D)F = δ + ψ where ψ ∈ C^∞(ℝⁿ). An operator P(D) is hypoelliptic if it has a parametrix whose singular support is contained in {0}.

              The singular support of a distribution u is the smallest closed set outside of which u agrees with a smooth function. A point x₀ is NOT in sing supp(u) if there exists a smooth compactly supported function φ with φ(x₀) ≠ 0 such that φ · u is smooth.

              Singular support of a tempered distribution. The singular support of u ∈ S'(E, ℂ) is the set of points x₀ ∈ E at which u is not smooth. Formally, x₀ ∈ SingularSupport E u if and only if for every smooth compactly supported function φ : E → ℂ with φ(x₀) ≠ 0, the product φ · u is NOT representable by a Schwartz function (i.e., is not in the range of the Schwartz-to-distribution embedding).

              Instances For

                Definition 11.8 (Parametrix). A parametrix for a constant-coefficient differential operator P(D) (modeled as a continuous linear endomorphism of S'(E, ℂ)) is a tempered distribution F ∈ S'(E, ℂ) such that P(D)F = δ + ψ where ψ is smooth. Equivalently, P(D)F - δ lies in the range of the Schwartz-to-distribution embedding.

                Instances For

                  Definition 11.8 (Hypoelliptic operator). A constant-coefficient differential operator P(D) is hypoelliptic if it admits a parametrix F whose singular support is contained in {0}. In other words, F is smooth away from the origin.

                  Instances For

                    Definition 11.11: Elliptic differential operator #

                    def Chapter11.IsElliptic {n : } (P : MvPolynomial (Fin n) ) (m : ) :

                    Definition 11.11 (Elliptic polynomial/operator): A polynomial P(ξ) in n variables over is said to be elliptic of order m provided its principal part P_m(ξ) (the homogeneous component of degree m) is nonzero for all nonzero ξ ∈ ℝⁿ.

                    In the formalization, the polynomial is an element of MvPolynomial (Fin n) ℂ, and ξ ranges over Fin n → ℝ, with evaluation performed via MvPolynomial.eval after embedding ξ into ℂⁿ componentwise.

                    Instances For
                      @[reducible, inline]

                      Alias for the evaluator: def_11_11_elliptic is IsElliptic.

                      Instances For