Documentation

CsdLean4.Mathlib.Analysis.Matrix.SumOverPaths

The sum over paths: Feynman's formulation at finite dimension #

Category: 1-Mathlib (CSD-free; staged for upstream).

For a splitting A + B of a skew-Hermitian generator, the (i, j) entry of exp (A + B) is the limit, as the number of steps grows, of a sum over all discrete paths from i to j of the product of the one-step amplitudes along the path. This is Feynman's sum over histories as a theorem rather than a postulate, in the finite-dimensional setting where every object is a matrix: the "paths" are sequences of basis states, the "action" is the product of transfer-matrix entries, and the limit is the Lie–Trotter product formula read entry by entry through the path expansion of a matrix power.

Honest scope #

⚠️ Finite dimension, discrete paths. This is the sum over paths of a finite-level system: no continuum of positions, no measure on path space, no action functional. It is what the corpus's finite-dimensional reconstruction can say, and it is exact. The continuum path integral (Wiener measure and Feynman–Kac in imaginary time, or the Trotter–Kato limit for unbounded generators) is a different rung and is not here.

⚠️ The limit, not a rate. The convergence is inherited from trotter_skew, whose bound is O(1/n) in the operator norm; no rate is restated here entrywise.

References: R. P. Feynman, Space-time approach to non-relativistic quantum mechanics, Rev. Mod. Phys. 20, 367 (1948); E. Nelson, Feynman integrals and the Schrödinger equation, J. Math. Phys. 5, 332 (1964), the Trotter route; Analysis/Matrix/TrotterProduct.lean (trotter_skew); LinearAlgebra/Matrix/PathSum.lean (pow_succ_apply_eq_sum_pathWeight); specs/future-work.md, specs/BACKLOG.md #36.

noncomputable def Matrix.trotterStep {m : Type u_1} [Fintype m] [DecidableEq m] (A B : Matrix m m ℂ) (n : ℕ) :

The one-step transfer matrix of the splitting A + B into n steps: exp (A / n) * exp (B / n). Its (i, j) entry is the amplitude of one step from i to j, summed over the intermediate state between the A-half and the B-half of the step (Matrix.mul_apply). The quantitative entry bound norm_entry_le_l2_opNorm (Analysis/Matrix/L2OpNormEntry.lean) is the companion of tendsto_apply_of_tendsto below.

Equations
Instances For
    theorem Matrix.trotterStep_def {m : Type u_1} [Fintype m] [DecidableEq m] (A B : Matrix m m ℂ) (n : ℕ) :

    trotterStep, unfolded.

    theorem Matrix.trotterStep_apply {m : Type u_1} [Fintype m] [DecidableEq m] (A B : Matrix m m ℂ) (n : ℕ) (i j : m) :
    A.trotterStep B n i j = ∑ k : m, NormedSpace.exp ((↑n)⁻¹ • A) i k * NormedSpace.exp ((↑n)⁻¹ • B) k j

    The one-step amplitude, as the sum over the intermediate state of the step.

    theorem Matrix.tendsto_apply_of_tendsto {m : Type u_1} {f : ℕ → Matrix m m ℂ} {X : Matrix m m ℂ} (h : Filter.Tendsto f Filter.atTop (nhds X)) (i j : m) :
    Filter.Tendsto (fun (n : ℕ) => f n i j) Filter.atTop (nhds (X i j))

    Convergence of a sequence of matrices in the operator norm gives convergence of every entry (the operator-norm topology is the product topology, and the entry map is continuous).

    theorem Matrix.trotterStep_pow_apply_tendsto {m : Type u_1} [Fintype m] [DecidableEq m] [Nonempty m] {A B : Matrix m m ℂ} (hA : A.conjTranspose = -A) (hB : B.conjTranspose = -B) (i j : m) :
    Filter.Tendsto (fun (n : ℕ) => (A.trotterStep B n ^ n) i j) Filter.atTop (nhds (NormedSpace.exp (A + B) i j))

    The Trotter approximants converge entry by entry.

    theorem Matrix.exp_add_apply_tendsto_sum_pathWeight {m : Type u_1} [Fintype m] [DecidableEq m] [Nonempty m] {A B : Matrix m m ℂ} (hA : A.conjTranspose = -A) (hB : B.conjTranspose = -B) (i j : m) :
    Filter.Tendsto (fun (n : ℕ) => ∑ p : Fin n → m, (A.trotterStep B (n + 1)).pathWeight (Fin.cons i (Fin.snoc p j))) Filter.atTop (nhds (NormedSpace.exp (A + B) i j))

    ★★ The sum over paths. For skew-Hermitian A and B, the (i, j) entry of exp (A + B) is the limit, as n → ∞, of the sum over every sequence p of n intermediate basis states of the product of the n + 1 one-step amplitudes of trotterStep A B (n + 1) along the path i, p 0, …, p (n - 1), j.

    For Hermitian H and real t, the Schrödinger generator -(i t) H is skew-Hermitian.

    theorem Matrix.exp_neg_I_mul_smul_add_apply_tendsto_sum_pathWeight {m : Type u_1} [Fintype m] [DecidableEq m] [Nonempty m] {H₁ H₂ : Matrix m m ℂ} (h₁ : H₁.IsHermitian) (h₂ : H₂.IsHermitian) (t : ℝ) (i j : m) :
    Filter.Tendsto (fun (n : ℕ) => ∑ p : Fin n → m, (((-↑t * Complex.I) • H₁).trotterStep ((-↑t * Complex.I) • H₂) (n + 1)).pathWeight (Fin.cons i (Fin.snoc p j))) Filter.atTop (nhds (NormedSpace.exp ((-↑t * Complex.I) • (H₁ + H₂)) i j))

    ★★ The sum over paths for a split Hamiltonian. For Hermitian H₁, H₂ and a real time t, the matrix element of the propagator exp (-(i t) (H₁ + H₂)) between the basis states i and j is the limit of the sum over every sequence of n intermediate states of the product of the short-time propagators exp (-(i t/(n+1)) H₁) exp (-(i t/(n+1)) H₂) along the path.