Fourier Transforms for QFT #
This file contains Fourier transform identities needed for the Gaussian Free Field, particularly for proving reflection positivity of the free covariance.
Main Results #
fourier_lorentzian_1d: The 1D Fourier transform of the Lorentzian 1/(k² + μ²) gives exponential decay: ∫ dk e^{ikx} / (k² + μ²) = (π/μ) e^{-μ|x|}fourier_exponential_decay: The inverse transform - Fourier transform of e^{-μ|x|} gives the Lorentzian: ∫ dx e^{-ikx} e^{-μ|x|} = 2μ/(k² + μ²)exp_factorization_reflection: Key factorization for reflection positivity
Mathematical Background #
The key identity for reflection positivity is:
C(x - y) = ∫ d⁴k / (2π)⁴ · e^{ik·(x-y)} / (k² + m²)
For the time direction k₀, we need:
∫ dk₀ e^{ik₀(x₀ - y₀)} / (k₀² + μ²) = (π/μ) e^{-μ|x₀ - y₀|}
where μ² = |pbar|² + m² (with pbar the spatial momentum).
When x₀ > 0 and y₀ < 0 (positive and negative time respectively), we have x₀ - y₀ > 0, so the exponential is e^{-μ(x₀ - y₀)} which factorizes as e^{-μx₀} · e^{μy₀}, enabling the reflection positivity argument.
Proof Strategy #
The 1D result follows from Fourier inversion:
- Compute half-line integrals using the fundamental theorem of calculus
- Sum to get FT[e^{-μ|x|}] = 2μ/(k² + μ²)
- Apply Fourier inversion to derive the Lorentzian result
Dependencies #
No axioms declared in this file.
Fubini reordering for triple integrals.
For integrable functions on α × α × α, we can reorder the integration from ∫x ∫y ∫k to ∫k ∫x ∫y.
This follows from Fubini-Tonelli: if F is integrable on the product space, then we can integrate in any order.
The exponential decay function is integrable when μ > 0. Proof: Split ℝ into (-∞, 0] ∪ (0, ∞) and use:
- integrableOn_exp_mul_Iic for exp(μx) on (-∞, 0] (since μ > 0)
- integrableOn_exp_mul_Ioi for exp(-μx) on (0, ∞) (since -μ < 0)
The Fourier integrand of exponential decay is integrable. Proof: |exp(ikx)| = 1, so the norm of the integrand equals exp(-μ|x|), which is integrable by integrable_exponential_decay.
Half-Line Integrals and Antiderivatives #
The half-line integrals are the fundamental building blocks. We compute these using the fundamental theorem of calculus for improper integrals, then combine them to get the Fourier transform of e^{-μ|x|}, and finally use Fourier inversion to derive the Lorentzian result.
The antiderivative of e^{(ik+α)x} for α ≠ 0. This is the indefinite integral: ∫ e^{(ik+α)x} dx = e^{(ik+α)x} / (ik + α)
The denominator ik + α is never zero since Re(ik + α) = α ≠ 0.
Special cases:
- α = -μ (μ > 0): gives decay on [0,∞), converges at +∞
- α = +μ (μ > 0): gives growth on (-∞,0], converges at -∞
Complex exponential e^{cx} tends to 0 as x → +∞ when Re(c) < 0. Proof: ‖e^{cx}‖ = e^{Re(c)·x} → 0 since Re(c) < 0 and x → +∞.
Complex exponential e^{cx} tends to 0 as x → -∞ when Re(c) > 0. Proof: ‖e^{cx}‖ = e^{Re(c)·x} → 0 since Re(c) > 0 and x → -∞.
The integrand e^{(ik-μ)x} is integrable on [0, ∞) when μ > 0. This follows from the exponential decay since Re(ik - μ) = -μ < 0.
Exponential e^{bx} is integrable on (-∞, a) when b > 0. Proved by change of variables from exp_neg_integrableOn_Ioi.
Exponential e^{bx} is integrable on (-∞, a] when b > 0. Follows from Iio version since measure of a point is 0.
The integrand e^{(ik+μ)x} is integrable on (-∞, 0] when μ > 0. This follows from the exponential decay since Re(ik + μ) = μ > 0.
The integral over the positive half-line [0, ∞): ∫₀^∞ e^{ikx} e^{-μx} dx = 1/(μ - ik)
Proof: Use FTC with antiderivative e^{(ik-μ)x}/(ik-μ). At +∞: e^{(ik-μ)x} → 0 since Re(ik-μ) = -μ < 0. At 0: e^0/(ik-μ) = 1/(ik-μ). Result: 0 - 1/(ik-μ) = -1/(ik-μ) = 1/(μ-ik).
The integral over the negative half-line (-∞, 0]: ∫_{-∞}^0 e^{ikx} e^{μx} dx = 1/(μ + ik)
Proof: Use FTC with antiderivative e^{(ik+μ)x}/(ik+μ). At -∞: e^{(ik+μ)x} → 0 since Re(ik+μ) = μ > 0. At 0: e^0/(ik+μ) = 1/(ik+μ) = 1/(μ+ik).
The full integral as sum of half-line integrals. This is the key decomposition: ∫{-∞}^∞ e^{ikx} e^{-μ|x|} dx = ∫{-∞}^0 e^{ikx} e^{μx} dx + ∫_0^∞ e^{ikx} e^{-μx} dx = 1/(μ+ik) + 1/(μ-ik) = 2μ/(k² + μ²)
Fourier Transform of Exponential Decay #
The Fourier transform of e^{-μ|x|} gives the Lorentzian 2μ/(k² + μ²). This is the "forward" direction of the Fourier pair.
The Fourier transform of the exponential decay function e^{-μ|x|}. ∫_{-∞}^{∞} e^{ikx} e^{-μ|x|} dx = 2μ/(k² + μ²)
This follows from splitting at x = 0 (see fourier_exponential_decay_split).
Fourier Inversion and the Lorentzian Transform #
By the Fourier inversion theorem, if FTf = g(k), then IFTg = f(x). Since FTe^{-μ|x|} = 2μ/(k² + μ²), the inverse gives: (1/2π) ∫ e^{ikx} · 2μ/(k² + μ²) dk = e^{-μ|x|} Rearranging: ∫ e^{ikx} / (k² + μ²) dk = (π/μ) e^{-μ|x|}
Relation to Mathlib's Fourier Transform #
Mathlib's Fourier transform uses the convention: FT_mathlib(f)(ξ) = ∫ exp(2πi⟨x,ξ⟩) f(x) dx
For ℝ with standard inner product ⟨x,ξ⟩ = xξ, this becomes: FT_mathlib(f)(ξ) = ∫ exp(2πixξ) f(x) dx
Our convention uses: FT(f)(k) = ∫ exp(ikx) f(x) dx
The relationship is: FT(f)(k) = FT_mathlib(f)(k/(2π))
Mathlib provides MeasureTheory.Integrable.fourierInv_fourier_eq which says:
If f is integrable, FT(f) is integrable, and f is continuous at x, then
IFT(FT(f))(x) = f(x)
The exponential decay function is continuous.
The exponential decay function is integrable over ℝ.
Mathlib's Fourier transform of expDecayFun equals the scaled Lorentzian. FT_mathlib(e^{-μ|x|})(ξ) = 2μ/(4π²ξ² + μ²) This follows from fourier_exponential_decay' via the substitution k = -2πξ.
The Mathlib Fourier transform of expDecayFun is integrable.
Fourier inversion theorem for the exponential decay / Lorentzian pair. If FTe^{-μ|x|} = 2μ/(k² + μ²), then the inverse transform gives: (1/2π) ∫ e^{ikx} · 2μ/(k² + μ²) dk = e^{-μ|x|}
This follows from Mathlib's Fourier inversion theorem applied to the exponential decay function, combined with the explicit formula for its Fourier transform and a change of variables.
The Lorentzian Fourier Transform (Main Result) #
We derive the Lorentzian Fourier transform from the Fourier inversion theorem.
The 1D Fourier transform of the Lorentzian/Cauchy distribution. This is the fundamental identity: ∫_{-∞}^{∞} e^{ikx} / (k² + μ²) dk = (π/μ) e^{-μ|x|}
Derivation from Fourier inversion:
From fourier_inversion_exp_decay: (1/2π) ∫ e^{ikx} · 2μ/(k² + μ²) dk = e^{-μ|x|}
Multiply both sides by π/μ:
(1/2π) · (π/μ) · 2μ ∫ e^{ikx} / (k² + μ²) dk = (π/μ) e^{-μ|x|}
∫ e^{ikx} / (k² + μ²) dk = (π/μ) e^{-μ|x|}
The exponential from the Lorentzian Fourier transform factorizes. For x, y with x ≥ 0 and y ≤ 0, we have |x - y| = x - y = x + |y|, so e^{-μ|x-y|} = e^{-μx} · e^{-μ|y|} = e^{-μx} · e^{μy}.
Negative phase variant: ∫ e^{-ikx} / (k² + μ²) dk = (π/μ) e^{-μ|x|}
This follows from fourier_lorentzian_1d by the substitution k ↦ -k.
Since (-k)² = k² and the integral over ℝ is symmetric, we get the same result.
Application to Free Field Propagator #
The free field propagator in d=4 Euclidean dimensions is:
C(x) = ∫ d⁴k/(2π)⁴ · e^{ik·x} / (k² + m²)
We can split this into time and spatial parts. After integrating over the time component k₀, we get an exponential factor e^{-μ|x₀|} where μ = √(|pbar|² + m²).
Proof Strategy: Fourier Inversion Approach #
Overview #
The main result fourier_lorentzian_1d is derived via Fourier inversion:
Half-line integrals (FTC): Compute ∫₀^∞ e^{(ik-μ)x} dx and ∫_{-∞}^0 e^{(ik+μ)x} dx using the fundamental theorem of calculus for improper integrals.
Fourier transform of e^{-μ|x|}: Split into half-lines and sum: FTe^{-μ|x|} = 2μ/(k² + μ²)
Fourier inversion: Apply (1/2π) ∫ e^{ikx} · FTf dk = f(x) to get: (1/2π) ∫ e^{ikx} · 2μ/(k² + μ²) dk = e^{-μ|x|}
Lorentzian result: Rearrange to obtain: ∫ e^{ikx} / (k² + μ²) dk = (π/μ) e^{-μ|x|}
Key Lemmas (in dependency order) #
antideriv_exp_complex_linear: d/dx[e^{cx}/c] = e^{cx}tendsto_cexp_atTop_zero,tendsto_cexp_atBot_zero: Limits at ±∞integrableOn_exp_decay_Ioi,integrableOn_exp_growth_Iic: Integrabilityfourier_exp_decay_positive_halfline,fourier_exp_decay_negative_halfline: Half-line integralsfourier_exponential_decay_split: Sum to get 2μ/(k² + μ²)fourier_inversion_exp_decay: Inversion gives e^{-μ|x|}fourier_lorentzian_1d: Main result
Bridge to OS3 #
The lemma contour_integral_propagator in OS3 has a negative phase convention.
To prove it, use fourier_lorentzian_1d with substitution k ↦ -k:
theorem fourier_lorentzian_1d_neg (μ : ℝ) (hμ : 0 < μ) (x : ℝ) :
∫ k : ℝ, Complex.exp (-Complex.I * k * x) / (k^2 + μ^2) = (π / μ) * Real.exp (-μ * |x|)
Then coerce to complex exponential and instantiate with μ = omega m k.