Feynman's kernel for the free propagator #
Category: 1-Mathlib (CSD-free; staged for upstream).
Analysis/Semigroup/SchrodingerSchwartz.lean builds the free propagator U₀(t) = e^{−itH₀},
H₀ = −½ d²/dx², as a Fourier multiplier: U₀(t) = 𝓕⁻¹ e^{−2π²itξ²} 𝓕. This module (BACKLOG #44,
FC-5′) writes the kernel of that operator — the one the path integral is built from — by
analytic continuation of the heat kernel in the time variable:
- the kernel at complex time
w,fresnelKernel w x = (4πw)^{−1/2} e^{−x²/(4w)}, is a Schwartz function forRe w > 0(fresnelS, on top ofGaussianPacket.lean's complex-parameter Gaussian) and its Fourier transform is the multipliere^{−4π²wξ²}(★fourier_fresnelS): the amplitude(4πw)^{−1/2}is exactly what cancels thea^{−1/2}of the Gaussian transform; - the operator
fresnelOp wis convolution with that kernel, so it has both the kernel form∫ K_w(x − y) ψ(y) dy(★fresnelOp_apply) and the multiplier form (★fourier_fresnelOp); it is a semigroup in the complex time (★★fresnelOp_fresnelOp), and therefore every time-slicing is exact:n+1slices ofw/(n+1)compose tow(★★fresnelOp_iterate_div), each slice one kernel integral (★fresnelOp_iterate_succ_apply); - ★★★
tendsto_fresnelOp— the Gaussian-regularised Fresnel integral is the propagator:(U₀(t)ψ)(x) = lim_{ε ↓ 0} ∫ K_{ε + it/2}(x − y) ψ(y) dy, by dominated convergence on the Fourier side (|e^{−4π²(ε+it/2)ξ²}| ≤ 1,𝓕ψintegrable). This is the continuationt ↦ itof the heat kernel, taken where it converges absolutely and carried to the imaginary axis; - ★
fresnelKernel_I_mul— atw = it/2the kernel is Feynman's(2πit)^{−1/2} e^{ix²/(2t)}, of constant modulus (norm_fresnelKernel_I_mul): it does not decay, which is why the integral is improper for generalL²data — and absolutely convergent against theL¹Schwartz functions used here (★integrable_fresnelKernel_mul); - ★★★
freeSchrodingerS_eq_integral_fresnel— Feynman's kernel formula, exactly:(U₀(t)ψ)(x) = (2πit)^{−1/2} ∫ e^{i(x−y)²/(2t)} ψ(y) dyfort ≠ 0and Schwartzψ, with no regularisation left in the statement. The regularised integrals converge both to the propagator and to this integral, and limits are unique; - ★★
freeSchrodingerS_iterate—U₀(t) = (U₀(t/(n+1)))^{n+1}, so then-slice iterated Feynman integral is exact for everyn; and ★★★prod_fresnelKernel_eq_exp_discreteAction— the product of thenkernels in that iterated integral is((2πit/n)^{−1/2})^n e^{iS_n}withS_nthe discrete action∑ (q_{k+1} − q_k)²/(2Δt)of the sampled path (discreteAction). The phase of the path integral is a theorem here, not a notation.
Honest scope #
⚠️ One dimension, Schwartz data. The scoping note's ψ ∈ L¹ ∩ L² is not claimed: Fourier
inversion at this pin is the Schwartz theorem, and for ψ ∈ L² \ L¹ the Feynman integral is
genuinely improper — it needs oscillatory-integral technology the pin does not have
(MATHLIB-ABSENT(fresnelIntegral)). Nothing here evaluates an oscillatory integral as a limit of
truncations ∫_{−R}^{R}; the ε ↓ 0 regularisation is the whole method.
⚠️ The n-slice identity is exact for every n, so no limit over paths is taken and none is
claimed: there is no measure on path space here, and no statement that S_n converges to the
classical action. The imaginary-time counterpart, where the path measure does exist, is the
corpus's Feynman–Kac chain.
⚠️ fresnelS is defined by a dite on Re w > 0, following Mathlib's own
SchwartzMap.smulLeftCLM: off the half-plane the kernel is not Schwartz, and every lemma carries
the hypothesis.
References: M. Reed, B. Simon, Methods of Modern Mathematical Physics II §IX.7;
R. P. Feynman, A. R. Hibbs, Quantum Mechanics and Path Integrals (1965) §2-2, §3-1;
Analysis/Semigroup/SchrodingerSchwartz.lean (FC-5″), GaussianPacket.lean (FC-5‴);
specs/feynman-continuum-scoping.md §5 and §8 D4; specs/BACKLOG.md #44;
specs/future-work.md FP-1.
The kernel at complex time #
The amplitude (4πw)^{−1/2} of the heat kernel at complex time w.
Instances For
The heat kernel at complex time w: (4πw)^{−1/2} e^{−x²/(4w)}. At w = it/2 this is
Feynman's kernel (2πit)^{−1/2} e^{ix²/(2t)} (fresnelKernel_I_mul).
Equations
- SchrodingerGroup.fresnelKernel w x = SchrodingerGroup.fresnelAmp w * Complex.exp (-↑x ^ 2 / (4 * w))
Instances For
The Fourier multiplier of the complex-time kernel: e^{−4π²wξ²}. At w = it/2 this is the
free Schrödinger phase e^{−2π²itξ²} (fresnelSymbol_I_mul).
Equations
- SchrodingerGroup.fresnelSymbol w ξ = Complex.exp (-(4 * ↑Real.pi ^ 2 * w) * ↑ξ ^ 2)
Instances For
The kernel at complex time w, Re w > 0, as a Schwartz function: the Gaussian e^{−πax²} of
parameter a = (4πw)⁻¹, scaled by (4πw)^{−1/2}. The dite follows Mathlib's own
SchwartzMap.smulLeftCLM: outside the half-plane the Gaussian is not Schwartz, and every lemma
below carries 0 < w.re.
Equations
- SchrodingerGroup.fresnelS w = if hw : 0 < w.re then SchrodingerGroup.fresnelAmp w • SchrodingerGroup.gaussianS (4 * ↑Real.pi * w)⁻¹ ⋯ else 0
Instances For
The Schwartz function is the kernel.
★ The Fourier transform of the complex-time kernel is the multiplier e^{−4π²wξ²}:
the amplitude (4πw)^{−1/2} is exactly what cancels the a^{−1/2} of the Gaussian transform.
The operator #
Convolution with the complex-time kernel, on Schwartz functions: e^{−wH₀} for
H₀ = −½ d²/dx², read through Re w > 0 (heat) and continued to w = it/2 (Schrödinger).
Equations
Instances For
★ The kernel form of the operator: (fresnelOp w ψ) x = ∫ K_w(x − y) ψ(y) dy, an
absolutely convergent integral (Mathlib's Schwartz convolution).
★ The multiplier form: the Fourier transform of the operator is multiplication by
e^{−4π²wξ²}.
Fourier inversion, pointwise: a Schwartz function is the inverse transform of its transform.
The multiplier is a one-parameter semigroup in the complex time.
★★ The complex-time semigroup: two slices compose into one,
fresnelOp w₁ ∘ fresnelOp w₂ = fresnelOp (w₁ + w₂). On the Fourier side it is
e^{−4π²w₁ξ²} e^{−4π²w₂ξ²} = e^{−4π²(w₁+w₂)ξ²}; on the kernel side it is the Gaussian convolution
identity, continued in the time.
★ The time-slicing recursion: each slice is one kernel integral, so the n+1-slice operator
is an n+1-fold iterated kernel integral.
The free propagator is the ε ↓ 0 limit #
At purely imaginary time the multiplier is the free Schrödinger phase e^{−2π²itξ²}.
The Fourier transform of the free propagator applied to a Schwartz function.
★★★ The Gaussian-regularised Fresnel integral is the free propagator: for Schwartz data,
(U₀(t)ψ)(x) = lim_{ε ↓ 0} (e^{−(ε + it/2)H₀} ψ)(x), every term on the left an absolutely
convergent kernel integral. The proof is dominated convergence on the Fourier side: the multiplier
e^{−4π²(ε + it/2)ξ²} has modulus ≤ 1 and tends pointwise to the free phase, and 𝓕ψ is
integrable.
★★★ The same limit as the kernel integral the physics writes:
(U₀(t)ψ)(x) = lim_{ε ↓ 0} ∫ (4π(ε + it/2))^{−1/2} e^{−(x−y)²/(4(ε + it/2))} ψ(y) dy.
★★ Feynman's time-slicing, regularised: for every number of slices the n+1-fold iterated
kernel integral has the same ε ↓ 0 limit, the propagator.
Feynman's kernel #
★ Feynman's kernel: at purely imaginary time w = it/2 the complex-time kernel is
(2πit)^{−1/2} e^{ix²/(2t)}, the kernel of the path integral.
The Feynman kernel has constant modulus: it does not decay, which is why the Fresnel integral is
improper for general L² data — and why it is nevertheless absolutely convergent against the L¹
Schwartz functions used here.
★ The Feynman integral converges absolutely for Schwartz data: the kernel is bounded and ψ
is integrable.
The ε ↓ 0 limit of the regularised kernel integrals is the Feynman integral: the amplitude
converges by continuity of the principal square root off the cut, and the Gaussian factor by
dominated convergence (its modulus is ≤ 1 for Re w ≥ 0).
★★★ Feynman's kernel formula, exactly: for Schwartz data and t ≠ 0,
(U₀(t)ψ)(x) = (2πit)^{−1/2} ∫ e^{i(x−y)²/(2t)} ψ(y) dy, an absolutely convergent integral — no
regularisation is left in the statement. The Gaussian regularisation proves it: the regularised
integrals converge to the propagator (tendsto_fresnelOp) and to this integral
(tendsto_integral_fresnelKernel), and limits in ℂ are unique.
The free propagator is a one-parameter group on Schwartz functions.
★★ The propagator is its own n-slice product: U₀(t) = (U₀(t/(n+1)))^{n+1}, so the
n-slice iterated Feynman integral is exact for every n, not only in a limit.
The discrete action #
The discrete action of a path q : ℕ → ℝ sampled at n steps of duration t/n:
S_n = ∑_{k<n} (q_{k+1} − q_k)²/(2Δt), the Riemann sum of ∫ ½ q̇² ds for the free Lagrangian.
Equations
- SchrodingerGroup.discreteAction t n q = ∑ k ∈ Finset.range n, (q (k + 1) - q k) ^ 2 / (2 * (t / ↑n))
Instances For
★★★ e^{iS_n} is the integrand: the product of the n Feynman kernels that make up the
n-slice iterated integral is ((2πit/n)^{−1/2})^n e^{iS_n}, with S_n the discrete action of the
sampled path. The phase of the path integral is a theorem, not a notation.