Fourier Sobolev optimization on the two-dimensional torus

DefinitionRescaling the torus to unit period#✓fully proved

The homeomorphism from the torus \(\mathbb{T}^2 = (\mathbb{R}/2\pi\mathbb{Z})^2\) onto the unit torus \((\mathbb{R}/\mathbb{Z})^2\) that sends \((x_1, x_2)\) to \((x_1/(2\pi), x_2/(2\pi))\).

def unitRescaling : Torus ≃ₜ UnitAddTorus (Fin 2) :=
  Homeomorph.piCongrRight fun _ ↦
    AddCircle.homeomorphAddCircle (2 * Real.pi) 1 (by positivity) one_ne_zero
Uses 1 · Used by 9

DefinitionA function on the torus as a function on the unit torus#✓fully proved

For a continuous function \(u : \mathbb{T}^2 \to \mathbb{R}\) on the torus \(\mathbb{T}^2 = (\mathbb{R}/2\pi\mathbb{Z})^2\), its unit representative is the continuous function \(\tilde u : (\mathbb{R}/\mathbb{Z})^2 \to \mathbb{C}\) given by \(\tilde u(y) = u(2\pi y)\), the composition of \(u\) with the inverse of the rescaling to unit period, with its real values regarded as complex numbers.

def unitRepresentative (u : C(Torus, ℝ)) : C(UnitAddTorus (Fin 2), ℂ) where
  toFun x := (u (unitRescaling.symm x) : ℂ)
  continuous_toFun := Complex.continuous_ofReal.comp
    (u.continuous.comp unitRescaling.symm.continuous)
Uses 2 · Used by 6

DefinitionFourier coefficients on the torus#✓fully proved

Let \(u : (\mathbb{R}/2\pi\mathbb{Z})^2 \to \mathbb{R}\) be continuous and let \(k \in \mathbb{Z}^2\) be a frequency. The Fourier coefficient of \(u\) at \(k\) is \(a_k = 2\pi c_k\), where \(c_k = \int_{[0,1]^2} u(2\pi y)\, e^{-2\pi i\, k \cdot y}\, dy\) is the Fourier coefficient of the rescaled function \(y \mapsto u(2\pi y)\) on the unit torus with respect to normalized Haar measure. Equivalently, \(a_k = \frac{1}{2\pi} \int_{[0,2\pi]^2} u(x)\, e^{-i\, k \cdot x}\, dx\).

def fourierCoefficient (u : C(Torus, ℝ)) (k : Frequency) : ℂ :=
  (2 * Real.pi : ℂ) * UnitAddTorus.mFourierCoeff (unitRepresentative u) k
Uses 3 · Used by 11

DefinitionSquared frequency#✓fully proved

For a frequency \(k = (k_1, k_2) \in \mathbb{Z}^2\), the frequency weight is \(|k|^2 = k_1^2 + k_2^2\), the squared frequency of the mode \(e^{i k \cdot x}\) on the torus \((\mathbb{R}/2\pi\mathbb{Z})^2\).

def frequencyWeight (k : Frequency) : ℝ := ∑ i, (k i : ℝ) ^ 2
Uses 1 · Used by 11

DefinitionGradient energy#✓fully proved

For a continuous function \(u : (\mathbb{R}/2\pi\mathbb{Z})^2 \to \mathbb{R}\) with Fourier coefficients \(a_k\), the gradient energy is \(\sum_{k \in \mathbb{Z}^2} |k|^2 \, |a_k|^2\), where \(|k|^2 = k_1^2 + k_2^2\); the sum is taken to be \(0\) when the series diverges.

By Parseval's identity, for smooth \(u\) this equals \(\int_{[0,2\pi]^2} |\nabla u(x)|^2 \, dx\).

def gradientEnergy (u : C(Torus, ℝ)) : ℝ :=
  ∑' k, frequencyWeight k * ‖fourierCoefficient u k‖ ^ 2
Uses 4 · Used by 7

DefinitionLaplacian energy#✓fully proved

For a continuous function \(u : (\mathbb{R}/2\pi\mathbb{Z})^2 \to \mathbb{R}\) with Fourier coefficients \(a_k\), the Laplacian energy is \(\sum_{k \in \mathbb{Z}^2} |k|^4 \, |a_k|^2\), where \(|k|^2 = k_1^2 + k_2^2\); the sum is taken to be \(0\) when the series diverges.

By Parseval's identity, for smooth \(u\) this equals \(\int_{[0,2\pi]^2} |\Delta u(x)|^2 \, dx\).

def laplaceEnergy (u : C(Torus, ℝ)) : ℝ :=
  ∑' k, frequencyWeight k ^ 2 * ‖fourierCoefficient u k‖ ^ 2
Uses 4 · Used by 4

DefinitionFunctions in \(H^2\)#✓fully proved

A continuous function \(u : (\mathbb{R}/2\pi\mathbb{Z})^2 \to \mathbb{R}\) with Fourier coefficients \(a_k\) is in \(H^2\) when the series \(\sum_{k \in \mathbb{Z}^2} (1 + |k|^2)^2 \, |a_k|^2\) converges, where \(|k|^2 = k_1^2 + k_2^2\).

This is the full spectral \(H^2\) condition, not a restriction to trigonometric polynomials.

def IsH2 (u : C(Torus, ℝ)) : Prop :=
  Summable fun k ↦ (1 + frequencyWeight k) ^ 2 * ‖fourierCoefficient u k‖ ^ 2
Uses 4 · Used by 5

DefinitionNormalized Haar measure on the unit torus#✓fully proved

The measure on the unit torus \((\mathbb{R}/\mathbb{Z})^2\) that is the product of two copies of the Haar probability measure on the circle \(\mathbb{R}/\mathbb{Z}\).

Used by 2

DefinitionMean of a function on the torus#✓fully proved

For a continuous function \(u : \mathbb{T}^2 \to \mathbb{R}\) on the torus \(\mathbb{T}^2 = (\mathbb{R}/2\pi\mathbb{Z})^2\), its mean is \(\int_{(\mathbb{R}/\mathbb{Z})^2} u(2\pi y)\, d\mu(y)\), where \(\mu\) is the normalized Haar measure on the unit torus \((\mathbb{R}/\mathbb{Z})^2\) and \(y \mapsto 2\pi y\) is the inverse of the rescaling to unit period.

Since \(\mu\) is a probability measure, this is the average \((2\pi)^{-2} \int_{[0, 2\pi]^2} u(x)\, dx\); the Lebesgue integral of \(u\) over a period cell is \((2\pi)^2\) times the mean.

def mean (u : C(Torus, ℝ)) : ℝ :=
  ∫ y, u (unitRescaling.symm y) ∂unitHaar
Uses 3 · Used by 4

DefinitionZero-mean \(H^2\) functions#✓fully proved

A continuous function \(u : (\mathbb{R}/2\pi\mathbb{Z})^2 \to \mathbb{R}\) is a zero-mean \(H^2\) function when it is in \(H^2\) and its mean is \(0\).

def IsZeroMeanH2 (u : C(Torus, ℝ)) : Prop := IsH2 u ∧ mean u = 0
Uses 3 · Used by 6

DefinitionConditionally admissible functions#✓fully proved

Let \(\delta\) be a real number. A continuous function \(u : (\mathbb{R}/2\pi\mathbb{Z})^2 \to \mathbb{R}\) is admissible at level \(\delta\) when it is a zero-mean \(H^2\) function, its gradient energy equals \(1\), and its Laplacian energy equals \(\delta\).

def ConditionalAdmissible (u : C(Torus, ℝ)) (δ : ℝ) : Prop :=
  IsZeroMeanH2 u ∧ gradientEnergy u = 1 ∧ laplaceEnergy u = δ
Uses 4 · Used by 3

DefinitionSquared values at the origin of admissible functions#✓fully proved

For a real number \(\delta\), this is the set of extended real numbers \(u(0)^2\), where \(u\) ranges over the continuous functions \(u : (\mathbb{R}/2\pi\mathbb{Z})^2 \to \mathbb{R}\) that are conditionally admissible at level \(\delta\).

def conditionalPointValues (δ : ℝ) : Set EReal :=
  (fun u : C(Torus, ℝ) ↦ (((u 0) ^ 2 : ℝ) : EReal)) '' {u | ConditionalAdmissible u δ}
Uses 2 · Used by 2

DefinitionConditional supremum of the squared value at the origin#✓fully proved

For real \(\delta\), \(\theta(\delta)\) is the supremum, taken in the extended reals \([-\infty, \infty]\), of the conditional point values at \(\delta\): the numbers \(u(0)^2\) over all continuous \(u : \mathbb{T}^2 \to \mathbb{R}\) on the torus \(\mathbb{T}^2 = (\mathbb{R}/2\pi\mathbb{Z})^2\) that are conditionally admissible at \(\delta\), that is, of mean zero and in \(H^2\) with gradient energy \(1\) and Laplace energy \(\delta\).

The supremum may be \(+\infty\), and it is \(-\infty\) when no function is admissible.

def theta (δ : ℝ) : EReal := sSup (conditionalPointValues δ)
Uses 1 · Used by 2

DefinitionFrequency ratio#✓fully proved

For a continuous function \(u : (\mathbb{R}/2\pi\mathbb{Z})^2 \to \mathbb{R}\), the frequency ratio is the Laplacian energy of \(u\) divided by the gradient energy of \(u\).

Applications assume positive gradient energy; when the gradient energy is \(0\), the quotient is taken to be \(0\).

def frequencyRatio (u : C(Torus, ℝ)) : ℝ := laplaceEnergy u / gradientEnergy u
Uses 3 · Used by 3

DefinitionLogarithmic remainder#✓fully proved

For real \(\delta\), the remainder is the extended real number \(R(\delta) = 4\pi\,\theta(\delta) - \log \delta - \log(1 + \log \delta)\), where \(\theta(\delta)\) is the conditional supremum of \(u(0)^2\) over the functions conditionally admissible at \(\delta\).

Here \(\log\) is the real logarithm with the conventions \(\log 0 = 0\) and \(\log x = \log |x|\) for \(x < 0\), so \(R(\delta)\) is defined for every real \(\delta\).

def remainder (δ : ℝ) : EReal :=
  ((4 * Real.pi : ℝ) : EReal) * theta δ - (Real.log δ : EReal) -
    (Real.log (1 + Real.log δ) : EReal)
Uses 1 · Used by 2

DefinitionSet of remainder values#✓fully proved

The set of extended real numbers \(R(\delta)\) for all real \(\delta \ge 1\), where \(R(\delta) = 4\pi\,\theta(\delta) - \log \delta - \log(1 + \log \delta)\) is the logarithmic remainder and \(\theta(\delta)\) is the conditional supremum of \(u(0)^2\) over the functions conditionally admissible at \(\delta\).

Every \(\delta \ge 1\) is included, not only sufficiently large ratios.

def sharpRemainderValues : Set EReal := remainder '' Set.Ici 1
Uses 1 · Used by 1

DefinitionSharp logarithmic remainder#✓fully proved

The sharp remainder is the supremum, in the extended reals \([-\infty, \infty]\), of the remainders \(R(\delta) = 4\pi\,\theta(\delta) - \log \delta - \log(1 + \log \delta)\) over all real \(\delta \ge 1\), where \(\theta(\delta)\) is the conditional supremum of \(u(0)^2\) over the functions conditionally admissible at \(\delta\).

The definition does not assume that the supremum is finite or attained.

def sharpRemainderValue : EReal := sSup sharpRemainderValues
Uses 1 · Used by 1

DefinitionUniform logarithmic remainder bound#✓fully proved

For a real constant \(L\), the bound with constant \(L\) holds when, for every continuous \(u : \mathbb{T}^2 \to \mathbb{R}\) on the torus \(\mathbb{T}^2 = (\mathbb{R}/2\pi\mathbb{Z})^2\) that has mean zero and lies in \(H^2\), has positive gradient energy \(E(u)\), and, writing \(\Delta(u)\) for its Laplace energy, has frequency ratio \(\Lambda(u) = \Delta(u)/E(u) \ge 1\), every point \(x\) of the torus satisfies \(u(x)^2 \le \frac{E(u)}{4\pi}\bigl(\log \Lambda(u) + \log(1 + \log \Lambda(u)) + L\bigr)\).

def UniformRemainderBound (L : ℝ) : Prop :=
  ∀ u : C(Torus, ℝ), IsZeroMeanH2 u → 0 < gradientEnergy u → 1 ≤ frequencyRatio u →
    ∀ x : Torus, (u x) ^ 2 ≤ gradientEnergy u / (4 * Real.pi) *
      (Real.log (frequencyRatio u) + Real.log (1 + Real.log (frequencyRatio u)) + L)
Uses 4
20 fully proved

The dependency explorer needs JavaScript and its graph data.

Download the complete graph data.