The MSD equation
ProvedLangevinHarmonicTrap.msd_odeThe goal of the mission, part (b) of the source. For an ensemble of realizations of Newton's equation , assume with Langevin that ensemble averaging commutes with time differentiation, that the random force is uncorrelated with the instantaneous position, , and that the velocity is thermalized, . Then the mean-squared displacement obeys the closed, deterministic, linear equation
import Mathlib import Definitions.Def_LangevinHarmonicTrap_dotSum import Definitions.Def_LangevinHarmonicTrap_IsLangevinPath open Finset MeasureTheory
namespace LangevinHarmonicTrap
theorem msd_ode {d : ℕ} {m zeta k kB T : ℝ}
{Ω : Type*} [MeasurableSpace Ω] {mu : Measure Ω} [IsProbabilityMeasure mu]
{x v a f : Ω → Fin d → ℝ → ℝ}
(hdyn : ∀ w : Ω, IsLangevinPath d m zeta k (x w) (v w) (a w) (f w))
(M M1 M2 : ℝ → ℝ)
(hM : ∀ t : ℝ, M t = ∫ w, dotSum d (x w) (x w) t ∂mu)
(hM1 : ∀ t : ℝ, M1 t = ∫ w, 2 * dotSum d (x w) (v w) t ∂mu)
(hM2 : ∀ t : ℝ, M2 t = ∫ w, 2 * (dotSum d (v w) (v w) t + dotSum d (x w) (a w) t) ∂mu)
(hMderiv : ∀ t : ℝ, HasDerivAt M (M1 t) t)
(hM1deriv : ∀ t : ℝ, HasDerivAt M1 (M2 t) t)
(hxx : ∀ t : ℝ, Integrable (fun w => dotSum d (x w) (x w) t) mu)
(hxv : ∀ t : ℝ, Integrable (fun w => dotSum d (x w) (v w) t) mu)
(hvv : ∀ t : ℝ, Integrable (fun w => dotSum d (v w) (v w) t) mu)
(hxa : ∀ t : ℝ, Integrable (fun w => dotSum d (x w) (a w) t) mu)
(hnoise : ∀ t : ℝ, ∫ w, dotSum d (x w) (f w) t ∂mu = 0)
(hequip : ∀ t : ℝ, m * ∫ w, dotSum d (v w) (v w) t ∂mu = d * kB * T) :
∀ t : ℝ, m * deriv (deriv M) t + zeta * deriv M t + 2 * k * M t
= 2 * (d * kB * T) := by sorry
end LangevinHarmonicTrap
Read-back
What the Lean code literally says, in plain math · Aristotle (Harmonic)
Disclosure — non-blind read-back. This read-back was written by the same agent that drafted the Lean statement above, not by an independent auditor working from the code alone. It is therefore not independent testimony: it may reproduce the author's intended reading rather than an outside reading of the code, and a reviewer should not treat it as a second, corroborating opinion.
Fix a dimension , reals , , , , , a measurable space and a probability measure on it. Fix families , , , assigning to each sample point and index a real function of time, and assume that for every the model predicate holds: has derivative at every , has derivative at every , and for all , .
Fix real functions , , of time with, for every real ,
and assume in addition that is differentiable at every with derivative , and that is differentiable at every with derivative . These last two hypotheses are precisely the interchange of averaging and differentiation: they are assumed, not derived. Assume also that for every the four functions of given by , , and are -integrable, and the two physical inputs: for every ,
where is coerced from a natural number to a real. (The second hypothesis is imposed at every time and, with and arbitrary reals, it is a genuine restriction on the ensemble rather than a definition of .)
The conclusion is that for every real ,
where denotes the derivative function of and the derivative of . Because and are assumed differentiable everywhere, these derivatives are the genuine ones and no junk-value convention is in play. Nothing asserts that the hypotheses are simultaneously satisfiable for a given parameter set; if they are not, the statement is vacuous for that set. No positivity is assumed of , , , or , and no initial condition on is imposed.
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.