Documentation

Atlas.ComplexVariables.code.Lecture17

Lecture 17: Mittag-Leffler's Theorem and Series Convergence #

This file formalizes results from Lecture 17:

Main results #

Implementation notes #

The existence of correcting polynomials with the required Taylor remainder decay bounds is proved using the Cauchy power series and the geometric approximation lemma for power series (HasFPowerSeriesOnBall.uniform_geometric_approx). For each ν with b ν ≠ 0, the singular part Pᵥ(1/(z-bᵥ)) is analytic at z=0 with a power series converging on a ball of radius ‖bᵥ‖/2. The Taylor polynomial (partial sum) of sufficiently high degree approximates the function to within 2⁻ᵛ on the smaller ball of radius ‖bᵥ‖/4. The corrected series then converges uniformly on compact subsets by the Weierstrass M-test.

References #

theorem summation_by_parts_convergence (a v : ) (hA : ∃ (M : ), ∀ (n : ), kFinset.range (n + 1), a k M) (hv : Filter.Tendsto v Filter.atTop (nhds 0)) (hvs : Summable fun (n : ) => v n - v (n + 1)) :
∃ (l : ), Filter.Tendsto (fun (N : ) => iFinset.range N, a i * v i) Filter.atTop (nhds l)

Lemma 1, Lecture 17 (Summation by parts test for convergence). If the partial sums Aₙ = ∑_{k=0}^{n} aₖ are bounded (in norm), vₙ → 0, and ∑ ‖vₙ - vₙ₊₁‖ < ∞, then ∑ aₙvₙ converges (the partial sums have a limit).

The proof uses Abel's summation by parts formula: ∑_{i=0}^{N} aᵢvᵢ = vₙAₙ + ∑_{i=0}^{N-1} Aᵢ(vᵢ - vᵢ₊₁) The first term tends to zero (bounded × vanishing), and the second series converges absolutely since ‖Aᵢ(vᵢ - vᵢ₊₁)‖ ≤ M‖vᵢ - vᵢ₊₁‖ and ∑‖vᵢ - vᵢ₊₁‖ < ∞.

Mittag-Leffler's Theorem #

noncomputable def MittagLeffler.singularPart (P : Polynomial ) (b z : ) :

The singular part function: z ↦ P(1/(z - b)).

Instances For

    A polynomial has no constant term if P(0) = 0.

    Instances For

      f has singular part P(1/(z-b)) at b.

      Instances For
        theorem MittagLeffler.singularPart_analyticAt (P : Polynomial ) (b z₀ : ) (hz : z₀ b) :

        The singular part z ↦ P(1/(z - b)) is analytic at any point z₀ ≠ b.

        In the 1D case (ℂ → ℂ), partial sums of a formal power series equal polynomial evaluation.

        theorem MittagLeffler.exists_correcting_poly_single (P : Polynomial ) (b : ) (hb : b 0) (ε : ) ( : 0 < ε) :
        ∃ (p : Polynomial ), ∀ (z : ), z b / 4singularPart P b z - Polynomial.eval z p ε

        For b ≠ 0, there exists a polynomial approximating singularPart P b to within ε on the ball of radius ‖b‖/4 around 0. The polynomial is a partial sum of the Cauchy power series of singularPart P b centered at 0.

        theorem MittagLeffler.exists_correcting_polynomials (b : ) (P : Polynomial ) (hb : Filter.Tendsto (fun (n : ) => b n) Filter.atTop Filter.atTop) (hP : ∀ (n : ), HasNoConstantTerm (P n)) :
        ∃ (p : Polynomial ), ∀ (ν : ) (z : ), z b ν / 4singularPart (P ν) (b ν) z - Polynomial.eval z (p ν) 2⁻¹ ^ ν

        Taylor remainder bound for correcting polynomials (Theorem 8, p.125 and formula (29), p.126 from the textbook).

        For each ν, Pᵥ(1/(z - bᵥ)) is analytic on {z : ‖z‖ < ‖bᵥ‖}. By the Cauchy power series and geometric approximation (HasFPowerSeriesOnBall.uniform_geometric_approx), choosing the truncation degree nᵥ large enough, we obtain a polynomial pᵥ (partial sum of the power series of Pᵥ(1/(z-bᵥ)) at z=0) satisfying ‖Pᵥ(1/(z-bᵥ)) - pᵥ(z)‖ ≤ 2⁻ᵛ for ‖z‖ ≤ ‖bᵥ‖/4.

        The singular part z ↦ P(1/(z - b)) is meromorphic at its pole b.

        Polynomial evaluation z ↦ eval z p is analytic at every point.

        theorem MittagLeffler.correctedTerm_analyticAt (P : Polynomial ) (b : ) (p : Polynomial ) (z₀ : ) (hz : z₀ b) :
        AnalyticAt (fun (z : ) => singularPart P b z - Polynomial.eval z p) z₀

        Each corrected term Pᵥ(1/(z-bᵥ)) - pᵥ(z) is analytic at points z₀ ≠ bᵥ.

        theorem MittagLeffler.summable_correctedTerms (b : ) (P p : Polynomial ) (hbound : ∀ (ν : ) (z : ), z b ν / 4singularPart (P ν) (b ν) z - Polynomial.eval z (p ν) 2⁻¹ ^ ν) (hb : Filter.Tendsto (fun (n : ) => b n) Filter.atTop Filter.atTop) (z : ) :
        Summable fun (ν : ) => singularPart (P ν) (b ν) z - Polynomial.eval z (p ν)

        The corrected terms are summable at each point.

        theorem MittagLeffler.h_analyticAt_away (b : ) (P p : Polynomial ) (hbound : ∀ (ν : ) (z : ), z b ν / 4singularPart (P ν) (b ν) z - Polynomial.eval z (p ν) 2⁻¹ ^ ν) (hb : Filter.Tendsto (fun (n : ) => b n) Filter.atTop Filter.atTop) (z₀ : ) (hz₀ : ∀ (n : ), z₀ b n) :
        AnalyticAt (fun (z : ) => ∑' (ν : ), (singularPart (P ν) (b ν) z - Polynomial.eval z (p ν))) z₀

        h is analytic away from all poles (Lecture 17 proof, Step 4).

        For z₀ away from all poles, split h(z) = Σ_{ν<N} + Σ_{ν≥N} where N is chosen so that all poles bᵥ with ν ≥ N satisfy ‖bᵥ‖/4 > ‖z₀‖. The finite sum is analytic at z₀ (each term has its pole at bᵥ ≠ z₀). The tail converges uniformly on B(0, ‖z₀‖+1) by the Weierstrass M-test (bound by 2⁻ᵛ), so its sum is holomorphic there by the Weierstrass theorem on uniform limits of holomorphic functions.

        theorem MittagLeffler.remaining_tsum_analyticAt (b : ) (P p : Polynomial ) (hbound : ∀ (ν : ) (z : ), z b ν / 4singularPart (P ν) (b ν) z - Polynomial.eval z (p ν) 2⁻¹ ^ ν) (hb : Filter.Tendsto (fun (n : ) => b n) Filter.atTop Filter.atTop) (hinj : Function.Injective b) (n : ) :
        AnalyticAt (fun (z : ) => ∑' (ν : ), if ν = n then 0 else singularPart (P ν) (b ν) z - Polynomial.eval z (p ν)) (b n)

        The "if-tsum" ∑' ν, (if ν = n then 0 else f ν z) (remaining corrected terms with the nth term zeroed out) is analytic at b n when b is injective.

        This uses the same Weierstrass M-test argument as h_analyticAt_away, noting that each term f ν with ν ≠ n is analytic at b n (since b is injective, so b n ≠ b ν), and the bound 2⁻ᵛ still applies.

        noncomputable def MittagLeffler.hFunc (b : ) (P p : Polynomial ) (z : ) :

        The function h(z) = ∑' ν, (Pᵥ(1/(z-bᵥ)) - pᵥ(z)).

        Instances For
          theorem MittagLeffler.mittag_leffler (b : ) (P : Polynomial ) (hb : Filter.Tendsto (fun (n : ) => b n) Filter.atTop Filter.atTop) (hP : ∀ (n : ), HasNoConstantTerm (P n)) (hinj : Function.Injective b) :
          ∃ (h : ), Meromorphic h (∀ (n : ), HasSingularPartAt h (P n) (b n)) ∀ (f : ), (∀ (n : ), HasSingularPartAt f (P n) (b n))(∀ zSet.range b, AnalyticAt f z)∀ (z : ), AnalyticAt (fun (w : ) => f w - h w) z

          Theorem 1, Lecture 17 (Mittag-Leffler's Theorem).

          Let {bᵥ} be a sequence in with ‖bᵥ‖ → ∞ (and the bᵥ distinct), and let Pᵥ(ζ) be polynomials without constant term. Then there exist functions f meromorphic in with poles at exactly the points bᵥ and corresponding singular parts Pᵥ(1/(z - bᵥ)).

          The most general such meromorphic function is f(z) = g(z) + ∑ᵥ [Pᵥ(1/(z - bᵥ)) - pᵥ(z)] where g is holomorphic everywhere and the pᵥ are correcting polynomials.

          The proof constructs h(z) = ∑ᵥ [Pᵥ(1/(z-bᵥ)) - pᵥ(z)] using correcting polynomials pᵥ with ‖Pᵥ(1/(z-bᵥ)) - pᵥ(z)‖ ≤ 2⁻ᵛ for ‖z‖ ≤ ‖bᵥ‖/4, then verifies: (a) h is meromorphic on (analytic away from all poles by the Weierstrass M-test, meromorphic at each pole by splitting off the finite singular term), (b) h has singular part Pₙ(1/(z-bₙ)) at each bₙ (the tsum minus the singular part reduces to an analytic correction plus a remaining tsum that is analytic), (c) any other meromorphic function with the same singular parts differs from h by an entire function (at poles, the singular parts cancel; away from poles, both are analytic).