The diffusion equation for non-Markovian Gaussian stochastic processes Preprint: APS/123-QED
Alessandro Taloni, Gianni Pagnini, Aleksei Chechkin · 2026
Read The diffusion equation for non-Markovian Gaussian stochastic processes Preprint: APS/123-QED by Alessandro Taloni, Gianni Pagnini, Aleksei Chechkin (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-10561.txt.
Introduction
Alessandro Taloni and two colleagues derive the diffusion equation for non-Markovian Gaussian processes. They derive the evolution equation for the chance density of particle shifts from any Gaussian velocity processes, without Markov property or stationarity. Their 2026 paper builds the result from the key function of the position density.
The approach builds a ladder of equations from Wick's theorem, where motion follows sums of shape-linked Wick contractions. That yields a closed diffusion equation for such processes beyond the Fokker-Planck description, keeping Gaussian shape only in the endless-order limit. In the long-time limit or under thermal initial conditions, velocity reduces to the stationary Gaussian Markov process with exponential self-link, the Ornstein-Uhlenbeck process. Position shows lasting effects due to finite time correlations of velocity. Readers get a diffusion equation holding without the Markov or steady-state needs behind Brownian motion theory.
Catalog record
| CREATOR | Alessandro Taloni, Gianni Pagnini, Aleksei Chechkin |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 27 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.10561
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
The diffusion equation for non-Markovian Gaussian stochastic processes Preprint: APS/123-QED Alessandro Taloni, Gianni Pagnini, Aleksei Chechkin Abstract We derive the exact evolution equation for the probability density function of particle displacements generated by arbitrary Gaussian velocity processes, when neither Markovianity and nor stationarity are assumed. Starting from the characteristic function of the density of the position, we construct a systematic hierarchy of equations based on Wick’s theorem, in which the dynamics is governed by sums of geometrically connected Wick contractions. This approach yields a closed non‑Markovian diffusion equation that generalizes the Fokker–Planck description and preserves Gaussianity only in the infinite‑order limit. The Fokker-Planck equation \frac{\partial P(v,t)}{\partial t}=\gamma\frac{\partial}{\partial v}(vP(v,t))+\frac{k_{B}T\gamma}{m}\frac{\partial^{2}P(v,t)}{\partial v^{2}} (1) was independently derived by Fokker 11 ; 12 , Planck 31 , and Smoluchowski 34 , building on the earlier insight of Lord Rayleigh, who investigated the velocity dynamics of a massive particle m colliding with a bath of lighter particles in thermal equilibrium at temperature T . This setting, later known as the Rayleigh piston 20 ; 10 , leads at the macroscopic level to linear damping, \dot{v}=-\gamma v 35 , and, upon Langevin’s introduction of a stochastic force \xi(t) , to the classical description of Brownian motion 24 : \displaystyle\dot{x}(t) \displaystyle=v(t) (2a) \displaystyle\dot{v}(t) \displaystyle=-\gamma v+\xi(t)\,. (2b) The statistical properties of the velocity v(t) are entirely determined by those of the stochastic force \xi(t) . In most physical applications one specifies only \langle\xi(t)\rangle=0 and \langle\xi(t_{1})\xi(t_{2})\rangle=2\gamma k_{B}T\,\delta(t_{1}-t_{2}) , thereby ensuring consistency with the fluctuation–dissipation theorem. However, full equivalence between the Langevin description ( 2b ) and the associated Fokker–Planck equation ( 1 ) requires the process v(t) to be Gaussian at all times. This, in turn, demands that \xi(t) be Gaussian white noise, implying an additional postulate beyond the first two moments 37 (see the Supplementary Material – SM). As a Gaussian process, all statistical properties of v(t) are completely determined by its two-time correlation function. In particular, any 2n -point velocity correlation admits a Wick decomposition 39 , \langle v(t_{1})\cdots v(t_{2n})\rangle=\sum_{\text{all pairings}}\prod_{\begin{subarray}{c}1\leq i<j\leq 2n\\ (i,j)\,\text{paired}\end{subarray}}\langle v(t_{i})v(t_{j})\rangle\,, (3) with (2n-1)!! distinct pairings. In the long-time limit or under thermal initial conditions, v(t) reduces to the unique stationary Gaussian Markov process with exponential autocorrelation, \langle v(t_{1})v(t_{2})\rangle=\frac{k_{B}T}{m}\rm{e}^{-\gamma\left|t_{1}-t_{2}\right|}\,, (4) namely the Ornstein–Uhlenbeck process 36 ; 9 ; 8 . Therefore, also the position variable x(t) , defined by the linear kinematic relation ( 2a ), is Gaussian. Its even moments follow directly from Wick’s theorem and obey the identity \langle x^{2n}(t)\rangle=(2n-1)!!\,\langle x^{2}(t)\rangle^{n} . However, unlike v(t) , x(t) exhibits non-Markovian dynamics due to the finite temporal correlations of the velocity ( 4 ) 1 1 1 In general, time integration of a Markov process generates non‑Markovian dynamics, except in the singular case of white noise. We recall that Markovianity, as well as stationarity or ergodicity, is a property of a chosen observable, not an intrinsic property of the underlying stochastic process. . On long time scales, t\gg 1/\gamma , or in the large damping limit, the probability density P(x,t) obeys the standard Smoluchowski diffusion equation 34 ; 37 ; 33 . At shorter time scales, however, the correct evolution equation remains nontrivial. The coupled system defined by Eqs. ( 2a ) and ( 4 ) thus provides a prototypical example of a non-Markovian Gaussian process. In this work we derive the exact diffusion equation governing P(x,t) for an arbitrary Gaussian non-Markovian velocity process, not necessarily stationary. Rather than restricting to Ornstein–Uhlenbeck statistics, we consider a general Gaussian velocity process v(t) fully characterized by its two-time correlation function \langle v(t_{1})v(t_{2})\rangle , thereby encompassing a wide class of internal and external colored noises 37 ; 27 ; 19 . Although linear systems driven by Gaussian colored noise have been extensively studied 19 ; 37 ; 42 ; 33 , our starting point follows Balescu’s physicist’s compromise 2 ; 3 ; 7 , whereby the velocity v(t) is treated as a prescribed Gaussian process, rather than being generated explicitly by an underlying Langevin equation, i.e., Eq. ( 2b ). Within the Balescu’s hybrid kinetic framework, the quasi-linear approximation leads to the diffusion equation \frac{\partial P(x,t)}{\partial t}=\int_{0}^{t}dt_{1}\,\langle v(t)v(t_{1})\rangle\frac{\partial^{2}P(x,t_{1})}{\partial x^{2}}\,, (5) which first emerged from projection-operator coarse-graining of Hamiltonian systems in the work of Zwanzig 41 ; 28 ; 42 . We therefore refer to Eq. ( 5 ) as the Zwanzig–Balescu equation (ZBE). An alternative route, based on a non-Markovian generalization of the Chapman–Kolmogorov equation, was developed by Hänggi and leads to 17 ; 18 ; 16 ; 19 \frac{\partial P(x,t)}{\partial t}=\int_{t_{0}}^{t}dt_{1}\,\langle v(t)v(t_{1})\rangle\frac{\partial^{2}P(x,t)}{\partial x^{2}}. (6) This equation, also referred to as a bona fide Fokker–Planck equation 1 ; 13 , was originally introduced by Batchelor in the context of turbulent diffusion 4 . We therefore refer to Eq. ( 6 ) as the Batchelor–Hänggi equation (BHE). Although both equations apply to Gaussian stationary processes, the BHE is explicitly dependent on the initial time t_{0} and may violate positivity 14 ; 19 . More fundamentally, neither equation is exact : both fail to reproduce Gaussian position moments beyond the second order, as shown in the SM. This structural shortcoming motivates the exact diffusion equation derived in the next section. The truncated characteristic function hierarchy. We introduce the characteristic function as \hat{P}(q,t)=\int_{-\infty}^{\infty}\!dx\,P(x,t)\,e^{iqx}, whose Taylor expansion around q=0 reads \hat{P}(q,t)=\sum_{n=0}^{\infty}\frac{(iq)^{2n}}{(2n)!}\,\langle x^{2n}(t)\rangle. (7) Taking the time derivative of Eq. ( 7 ) yields the exact identity \displaystyle\frac{\partial}{\partial t}\hat{P}(q,t)=\sum_{n=1}^{\infty}(iq)^{2n}\int_{0}^{t}dt_{1}\int_{0}^{t_{1}}dt_{2}\cdots \displaystyle\int_{0}^{t_{2n-2}}dt_{2n-1}\langle v(t)v(t_{1})\cdots v(t_{2n-1})\rangle, (8) with t_{0}\equiv t (see SM). Eq.( 8 ) holds for any stochastic process, independently of Markovianity, Gaussianity, or stationarity (with the exception of Lévy–stable processes). We now define the moment–truncated characteristic function (which coincides with the truncated cumulant expansion in the present Gaussian setting) as \hat{\rho}_{N}(q,t)=\sum_{n=0}^{N}\frac{(iq)^{2n}}{(2n)!}\langle x^{2n}(t)\rangle, (9) with the consistency condition \lim_{N\to\infty}\hat{\rho}_{N}(q,t)=\hat{P}(q,t). Finally, making use of the Wick decomposition ( 3 ), the time evolution of the truncated characteristic function can be written as \displaystyle\frac{\partial}{\partial t}\hat{\rho}_{N}(q,t)= \displaystyle\sum_{n=1}^{N}(iq)^{2n}\sum_{\text{all pairings}}\int_{0}^{t}dt_{1}\int_{0}^{t_{1}}dt_{2}\cdots \displaystyle\int_{0}^{t_{2n-2}}dt_{2n-1}\prod_{\begin{subarray}{c}0\leq i<j\leq 2n-1\\ (i,j)\,\text{paired}\end{subarray}}\langle v(t_{i})v(t_{j})\rangle. (10) This equation admits a natural diagrammatic interpretation in terms of Wick contractions: each term in the sum ( 3 ) corresponds to a specific pairing of velocity variables and can be represented by a diagram in which lines connect the associated times. In this representation, the full 2n ‑point correlation function is given by the sum over all possible pairwise connections among the points t_{1},t_{2},\ldots,t_{2n} , each link denoting a two‑time correlation \langle v(t_{i})v(t_{j})\rangle . By instance the 4 -times velocity correlation function is depicted as \displaystyle\langle v(t_{1})v(t_{2})v(t_{3})v(t_{4})\rangle=\underbrace{\hbox to117.19pt{\vbox to20.07pt{\pgfpicture\makeatletter\hbox{\hskip 7.38162pt\lower-7.1584pt\hbox to0.0pt{\lxSVG@begingroup@{_scopebegin} \lxSVG@begingroup@{stroke} \lxSVG@begingroup@{fill} \lxSVG@setlinewidth{\the\pgflinewidth}\lxSVG@begingroup@{stroke-width} \lx@inpgf@ignorespaces\nullfont\hbox to0.0pt{\lxSVG@begingroup@{_scopebegin} {{}}\lx@inpgf@ignorespaces\hbox{\hbox{{\lxSVG@begingroup@{_scopebegin} {{}{}{{ {}{}}}{ {}{}} {{}{{\lx@inpgf@ignorespaces}}}{{}{\lx@inpgf@ignorespaces}}{}{{}{\lx@inpgf@ignorespaces}} {\lx@inpgf@ignorespaces }{{{{\lx@inpgf@ignorespaces}}\lxSVG@begingroup@{_scopebegin} \lxSVG@transformcm{1.0}{0.0}{0.0}{1.0}{-4.04861pt}{-2.3254pt}\lxSVG@begingroup@{transform} \pgfsys@hbox{66}\lxSVG@closescope }}} \lxSVG@closescope }}} {{}}\lx@inpgf@ignorespaces\hbox{\hbox{{\lxSVG@begingroup@{_scopebegin} {{}{}{{ {}{}}}{ {}{}} {{}{{\lx@inpgf@ignorespaces}}}{{}{\lx@inpgf@ignorespaces}}{}{{}{\lx@inpgf@ignorespaces}} {\lx@inpgf@ignorespaces }{{{{\lx@inpgf@ignorespaces}}\lxSVG@begingroup@{_scopebegin} \lxSVG@transformcm{1.0}{0.0}{0.0}{1.0}{30.0946pt}{-2.3254pt}\lxSVG@begingroup@{transform} \pgfsys@hbox{66}\lxSVG@closescope }}} \lxSVG@closescope }}} {{}}\lx@inpgf@ignorespaces\hbox{\hbox{{\lxSVG@begingroup@{_scopebegin} {{}{}{{ {}{}}}{ {}{}} {{}{{\lx@inpgf@ignorespaces}}}{{}{\lx@inpgf@ignorespaces}}{}{{}{\lx@inpgf@ignorespaces}} {\lx@inpgf@ignorespaces }{{{{\lx@inpgf@ignorespaces}}\lxSVG@begingroup@{_scopebegin} \lxSVG@transformcm{1.0}{0.0}{0.0}{1.0}{64.23782pt}{-2.3254pt}\lxSVG@begingroup@{transform} \pgfsys@hbox{66}\lxSVG@closescope }}} \lxSVG@closescope }}} {{}}\lx@inpgf@ignorespaces\hbox{\hbox{{\lxSVG@begingroup@{_scopebegin} {{}{}{{ {}{}}}{ {}{}} {{}{{\lx@inpgf@ignorespaces}}}{{}{\lx@inpgf@ignorespaces}}{}{{}{\lx@inpgf@ignorespaces}} {\lx@inpgf@ignorespaces }{{{{\lx@inpgf@ignorespaces}}\lxSVG@begingroup@{_scopebegin} \lxSVG@transformcm{1.0}{0.0}{0.0}{1.0}{98.38148pt}{-2.3254pt}\lxSVG@begingroup@{transform} \pgfsys@hbox{66}\lxSVG@closescope }}} \lxSVG@closescope }}} { {}{}{}}{}{ {}{}{}}{ {}{}{}}{{{{}}{{\lx@inpgf@ignorespaces}}{{}}{{}}}{{{{}}{ {}{}}{}{}{{}{}}}} }{{{{}}{{\lx@inpgf@ignorespaces}}{{\lx@inpgf@ignorespaces}}{{}}{{}}}{{{{}}{ {}{}}{}{}{{}{}}}} }{{}{}\lx@inpgf@ignorespaces}{{}} {}{}{}{{{}}{{\lx@inpgf@ignorespaces}}{{}}} {{{}}{{\lx@inpgf@ignorespaces}}{{}}} {\lx@inpgf@ignorespaces}{{}}{}{{\lx@inpgf@ignorespaces}}{\lx@inpgf@ignorespaces}{{\lx@inpgf@ignorespaces}}{}{}{}{}\lxSVG@stroke\lxSVG@drawpath@unclipped{M 10.18 10.18 C 17.59 17.6 29.65 17.6 37.06 10.18}{fill:none} \lx@inpgf@ignorespaces { {}{}{}}{}{ {}{}{}}{ {}{}{}}{{{{}}{{\lx@inpgf@ignorespaces}}{{}}{{}}}{{{{}}{ {}{}}{}{}{{}{}}}} }{{{{}}{{\lx@inpgf@ignorespaces}}{{\lx@inpgf@ignorespaces}}{{}}{{}}}{{{{}}{ {}{}}{}{}{{}{}}}} }{{}{}\lx@inpgf@ignorespaces}{{}} {}{}{}{{{}}{{\lx@inpgf@ignorespaces}}{{}}} {{{}}{{\lx@inpgf@ignorespaces}}{{}}} {\lx@inpgf@ignorespaces}{{}}{}{{\lx@inpgf@ignorespaces}}{\lx@inpgf@ignorespaces}{{\lx@inpgf@ignorespaces}}{}{}{}{}\lxSVG@stroke\lxSVG@drawpath@unclipped{M 104.67 10.18 C 112.08 17.6 124.14 17.6 131.55 10.18}{fill:none} \lx@inpgf@ignorespaces \lxSVG@closescope {\lx@inpgf@ignorespaces}{\lx@inpgf@ignorespaces}{\lx@inpgf@ignorespaces}\hss}\lxSVG@discardpath\lxSVG@closescope \hss}}\lxSVG@closescope\endpgfpicture}}}_{\;\langle v(t_{1})v(t_{2})\rangle\langle v(t_{3})v(t_{4})\rangle\;}\;+\; (11) \displaystyle\underbrace{\hbox to117.19pt{\vbox to29.49pt{\pgfpicture\makeatletter\hbox{\hskip 7.38162pt\lower-7.1584pt\hbox to0.0pt{\lxSVG@begingroup@{_scopebegin} \lxSVG@begingroup@{stroke} \lxSVG@begingroup@{fill} \lxSVG@setlinewidth{\the\pgflinewidth}\lxSVG@begingroup@{stroke-width} \lx@inpgf@ignorespaces\nullfont\hbox to0.0pt{\lxSVG@begingroup@{_scopebegin} {{}}\lx@inpgf@ignorespaces\hbox{\hbox{{\lxSVG@begingroup@{_scopebegin} {{}{}{{ {}{}}}{ {}{}} {{}{{\lx@inpgf@ignorespaces}}}{{}{\lx@inpgf@ignorespaces}}{}{{}{\lx@inpgf@ignorespaces}} {\lx@inpgf@ignorespaces }{{{{\lx@inpgf@ignorespaces}}\lxSVG@begingroup@{_scopebegin} \lxSVG@transformcm{1.0}{0.0}{0.0}{1.0}{-4.04861pt}{-2.3254pt}\lxSVG@begingroup@{transform} \pgfsys@hbox{66}\lxSVG@closescope }}} \lxSVG@closescope }}} {{}}\lx@inpgf@ignorespaces\hbox{\hbox{{\lxSVG@begingroup@{_scopebegin} {{}{}{{ {}{}}}{ {}{}} {{}{{\lx@inpgf@ignorespaces}}}{{}{\lx@inpgf@ignorespaces}}{}{{}{\lx@inpgf@ignorespaces}} {\lx@inpgf@ignorespaces }{{{{\lx@inpgf@ignorespaces}}\lxSVG@begingroup@{_scopebegin} \lxSVG@transformcm{1.0}{0.0}{0.0}{1.0}{30.0946pt}{-2.3254pt}\lxSVG@begingroup@{transform} \pgfsys@hbox{66}\lxSVG@closescope