# State-dependent diffusions: Fick's law vs. Fokker-Planck's, which and why?

Consider a "state-dependent diffusion": a diffusion process for which the diffusion coefficient $D(x)$ depends on the (stochastic) state $x$ of the system. (An example is provided by the diffusion of tracers in a spatially inhomogeneous bath, e.g. with varying viscosity or temperature.) What is the associated flux? If $c(x)$ is the probability to find the system in the state $x$ at a given time, should the flux read $j(x)=-D(x)\nabla c(x)$, as in Fick's law? Or rather $j(x)=-\nabla(Dc)(x)$, as in the Fokker-Planck equation? (I'm using the terms "Fick" and "Fokker-Planck" here only as tags for each alternative.)

It is well known that this question does not have a definite answer, and should be answered case by case. My actual question is threefold:

1. Do you know examples of physically relevant state-dependent diffusions? What category do they fall in? Fick, or Fokker-Planck?
2. Do you have an intuition for the physics underlying this alternative?
3. The Fokker-Planck case has the peculiarity to lead to equilibrium distributions which are not of the Boltzmann-Gibbs form (the steady-state $c(x)$ depends on $D(x)$ and not just on the state's energy). What does this tell us about the foundations of statistical mechanics?
-

I don't agree with the premise of the question (that there is some mysterious disagreement between hydrodynamics, Fick's law, and the Fokker-Plank equation). I am not entirely certain what you mean by state dependent''. I will assume that the system is in local thermodynamic equilibrium. This is the basic assumption in hydrodynamics, and the Fokker-Planck equation is a possible microscopic model of how equilibrium is reached. In this case the diffusion constant is a function of $x$ only through its dependence on thermodynamic variables. In a simple fluid with a single type of impurity these are $T(x)$ and $\mu(x)$.

Note that $j(x)=-D(\mu(x),T(x))\nabla c(x)$ is not the most general statement of Fick's law. In general, there is also thermal diffusion, and there are extra terms in the presence of an external potential. This is explained in standard texts on fluid dynamics (like Landau). The RHS of the diffusion equation is $\nabla [D(\mu(x),T(x))\nabla c(x)]$.

I think that this is indeed the form one obtains from a stochastic model, see for example equ.(312) in Chandrasekhar's review http://rmp.aps.org/abstract/RMP/v15/i1/p1_1 (he calls this the Smoluchowski equation), or equ. (4.19) in these lecture notes http://www.ks.uiuc.edu/~kosztin/PHYCS498NSM/LectureNotes/chp4.pdf . There is an extra term in Fick's law, but this term is related to the external potential, $j(x)\sim Dc/T\nabla V$. I also think that this had to be the case. As long as I consider the most general hydro equation any stochastic model that relaxes to local thermodynamic equilibrium should reduce to this equation in the appropriate limit.

-
I didn't mention any mysterious disagreement, and do not think there is one. My point is, contrary to what many people think, not all diffusion processes are consistent with Fick's law, even with a drift term. I'd like to understand this phenomenon better. – Orwell Oct 4 '12 at 21:11
A simple experiment with anybody can make (I did) is the following: prepare two equal volumes of (i) water and (ii) a water+gelatine mixture. Dissolve the same quantity of food coloring in both, and put them in contact without mixing them (typically the gel at the bottom of a tube, the water on top). Initially the food coloring is uniformly distributed. After a day, not so: the coloring has accumulated in the gel. No temperature gradient, no potential, and yet the equilibrium c(x) is not homogeneous: Fick's law does not apply. Credit: [B Ph van Milligen et al 2005 Eur. J. Phys. 26 913] – Orwell Oct 4 '12 at 21:16