Documentation

Expdb.ExponentialSums.ExponentSumGrowth

Exponential sum growth exponents #

This module formalizes the asymptotic definition of the exponential sum growth exponent from the blueprint's Exponential sum growth exponents chapter (beta-chapter). It defines admissible model-phase bounds at a fixed scale and defines β(α) as the least such exponent.

Overview #

For α : ℝ≥0, an exponent β is admissible if every model-phase exponential sum at scale N = T ^ (α + o(1)) is bounded by T ^ (β + o(1)). We show that the admissible exponents form [β(α), ∞) and then specialize the bound to the logarithmic model phase.

Asymptotic definition #

N = T ^ (α + o(1)), expressed using a variable exponent that is equal to α up to an infinitesimal.

Equations
  • One or more equations did not get rendered due to their size.
Instances For

    X ≪ T ^ (β + o(1)), expressed using a variable exponent that is equal to β up to an infinitesimal.

    Equations
    • One or more equations did not get rendered due to their size.
    Instances For

      The variable exponential sum ∑ n ∈ [a, b], e(T F(n / N)).

      Equations
      Instances For

        The assertion that β is an admissible exponential-sum growth exponent at scale α.

        Equations
        • One or more equations did not get rendered due to their size.
        Instances For
          noncomputable def Expdb.exponentSumGrowthExponent (α : NNReal) :

          The exponential sum growth exponent β(α) from the blueprint definition beta-def.

          Equations
          Instances For

            Power asymptotics #

            theorem Expdb.IsPowerAsymptotic.eventually_between {N T : VariableObject } {α ε : } (hNT : IsPowerAsymptotic N T α) (hT : ∀ᶠ (i : ) in Filter.atTop, 1 T i) ( : 0 < ε) :
            ∀ᶠ (i : ) in Filter.atTop, T i ^ (α - ε) N i N i T i ^ (α + ε)

            If N = T ^ (α + o(1)), then N eventually lies between the powers with exponents α - ε and α + ε.

            theorem Expdb.isPowerAsymptotic_of_logb_tendsto {N T : VariableObject } {α : } (hT : ∀ᶠ (i : ) in Filter.atTop, 1 < T i) (hN : ∀ᶠ (i : ) in Filter.atTop, 0 < N i) (hexponent : Filter.Tendsto (fun (i : ) => Real.logb (T i) (N i)) Filter.atTop (nhds α)) :

            Convergence of the logarithmic exponent gives a power asymptotic.

            theorem Expdb.isPowerAsymptotic_of_between {N T δ : VariableObject } {α : } (hT : ∀ (i : ), 1 < T i) (hN : ∀ (i : ), 0 < N i) (hδnonneg : ∀ (i : ), 0 δ i) ( : δ.IsInfinitesimal) (hbetween : ∀ (i : ), T i ^ (α - δ i) N i N i T i ^ (α + δ i)) :

            A variable power sandwich with an infinitesimal error determines a power asymptotic.

            theorem Expdb.isPowerAsymptotic_of_between_const_rpow {N T : VariableObject } {α A B : } ( : 0 < α) (hA : 0 < A) (hB : 0 < B) (hNtop : Filter.Tendsto N Filter.atTop Filter.atTop) (hT : ∀ᶠ (i : ) in Filter.atTop, A * N i ^ α⁻¹ T i T i B * N i ^ α⁻¹) :

            Fixed positive multiplicative uncertainty in T = N ^ α⁻¹ does not affect the power asymptotic N = T ^ (α + o(1)).

            Power bounds #

            theorem Expdb.isPowerBounded_iff_forall_pos {E : Type u_1} [SeminormedAddCommGroup E] (X : VariableObject E) (T : VariableObject ) (β : ) (hT : ∀ (i : ), 1 T i) (hTunbounded : T.IsUnbounded) :
            IsPowerBounded X T β ∀ (ε : ), 0 < εX =O[Filter.atTop] fun (i : ) => T i ^ (β + ε)

            If T is at least one and unbounded, then X ≪ T ^ (β + o(1)) iff X = O(T ^ (β + ε)) for every fixed ε > 0.

            theorem Expdb.exponent_le_of_isPowerBounded_of_eventually_norm_ge_rpow {E : Type u_1} [SeminormedAddCommGroup E] {X : VariableObject E} {T : VariableObject } {β γ c : } (hT : ∀ (i : ), 1 T i) (hTunbounded : T.IsUnbounded) (hbound : IsPowerBounded X T β) (hc : 0 < c) (hlower : ∀ᶠ (i : ) in Filter.atTop, c * T i ^ γ X i) :
            γ β

            A power lower bound for an unbounded object forces its exponent to be no larger than any admissible power-bound exponent.

            Admissible exponents #

            The candidate set is upward closed, nonempty, and bounded below.

            Monotonicity #

            theorem Expdb.IsExponentSumBound.mono {α : NNReal} {β γ : } ( : IsExponentSumBound α β) (hβγ : β γ) :

            Admissible exponential-sum bounds are monotone in the exponent.

            Nonemptiness and lower bound #

            The triangle inequality gives the admissible exponent α.

            theorem Expdb.IsExponentSumBound.nonneg {α : NNReal} {β : } ( : IsExponentSumBound α β) :
            0 β

            Every admissible exponent at a nonnegative scale is nonnegative.

            The least admissible exponent #

            By underspill, the infimum of the admissible exponents is itself admissible.

            The defining universal property of the exponential sum growth exponent.

            The exponent sum growth exponent is the least admissible exponent.

            At scale α, the admissible exponents form the interval [β(α), ∞).

            The set of admissible exponential-sum exponents at scale α is closed.

            Logarithmic model phase #

            theorem Expdb.isPowerBounded_logPhase (α : NNReal) {N T : VariableObject } {a b : VariableObject } (hN : ∀ (i : ), 1 N i) (hT : ∀ (i : ), 1 T i) (hTunbounded : T.IsUnbounded) (hNT : IsPowerAsymptotic N T α) (hab : ∀ (i : ), N i (a i) (b i) 2 * N i) :

            Logarithmic-phase sums at scale α are power-bounded by the least admissible exponent. This specializes isExponentSumBound_exponentSumGrowthExponent to logPhase.