5
$\begingroup$

Consider the aggregation-diffusion equation $$ \frac{\partial \rho}{\partial t} = \nabla (\rho \nabla(W\star \rho)) + \nu \Delta \rho, $$ where $W:\mathbb{R}^d \to \mathbb{R}$ is a twice continuously differentiable interaction potential, $\nu \ge 0$, and where $\star$ denotes convolution, that is, $(W\star \rho) (x) = \int_{\mathbb{R}^d} W(x-y) \rho(y) \mathrm{d}y$. It seems that the case where $W$ is an attractive potential (that is $W(x)$ increases as $x\to \infty$) is well-studied, see for example the article I cite below. I suppose that makes sense as it is called aggregation equation.

However, I am interested in the case where $W$ is repulsive, on which I have found very little. In that case the solution should converge vaguely to zero, and I would be interested in asymptotic upper bounds on $\rho_t(0)$ in the case where $\rho$ is started from something nice and symmetric (say the standard normal density or something compactly supported), and where $W$ is a symmetric repulsive potential, like a standard normal density, a power law, anything really. I would even be interested in the case where $\nu = 0$, and specifically in $d = 2$.


Carrillo, José A.; Craig, Katy; Yao, Yao, Aggregation-diffusion equations: dynamics, asymptotics, and singular limits, Bellomo, Nicola (ed.) et al., Active particles, Volume 2. Advances in theory, models, and applications. Cham: Birkhäuser. Model. Simul. Sci. Eng. Technol., 65-108 (2019). ZBL1451.76117..

$\endgroup$
2
  • $\begingroup$ What exactly is $W\star \rho$ ? $\left(W\star \rho\right)(x)=\int \mathrm{d}^dx' \:W(x-x')\rho(x')$ ? $\endgroup$ Dec 1, 2021 at 12:52
  • $\begingroup$ Yes exactly, I added this to the question. @DanielCastro $\endgroup$ Dec 1, 2021 at 13:08

1 Answer 1

1
$\begingroup$

(Too long to be a comment.) A hand-waving, physicist calculation of a specific example for $d=3$ goes as follows. It may be put in more rigours terms for higher but not for lower dimensions. Let's take

\begin{equation} W(\mathbf{x}-\mathbf{x}')=\frac{\alpha}{|\mathbf{x}-\mathbf{x}'|}, \end{equation}

with $\alpha>0$ for a repulsive force. The one can show that $\nabla^2 W(\mathbf{x}-\mathbf{x}')=-4\pi\alpha\delta(\mathbf{x}-\mathbf{x}')$. Since the potential is spherically symmetric let's assume that $\rho(\mathbf{x})=\rho(r)$ and just for notation omit the time dependence, then

\begin{align} \nabla\cdot\left(\rho(\mathbf{x})\nabla\left( W\star \rho\right)(\mathbf{x}) \right)&=\nabla\rho\cdot\nabla\left( W\star \rho\right)+\rho\nabla^2\left( W\star \rho\right)\\ &=\nabla\rho\cdot\int d^3x'\:\rho(r') \nabla W(\mathbf{x}-\mathbf{x}')+\rho \int d^3x'\:\rho(r')\nabla^2 W(\mathbf{x}-\mathbf{x}')\\ &=-\nabla\rho\cdot\int d^3x'\:\rho(r') \nabla' W(\mathbf{x}-\mathbf{x}')-4\pi\alpha\:\rho \int d^3x'\:\rho(r')\delta(\mathbf{x}-\mathbf{x}')\\ &=\alpha\int d^3x'\frac{\nabla\rho(r)\cdot\nabla'\rho(r')}{|\mathbf{x}-\mathbf{x}'|}-4\pi \alpha\:\rho^2(r). \end{align}

It's not hard to show that the remaining integral vanishes by explicit calculation. Then we have

\begin{align} \frac{\partial\rho(r;t)}{\partial t}=-4\pi\alpha\: \rho^2(r;t)+\frac{\nu}{r^2}\frac{\partial}{\partial r}\left(r^2\frac{\partial \rho(r;t)}{\partial r} \right). \end{align}

Note that we cannot put $\nu=0$ (i.e. vanishing diffusion) because then we lose the solution. Since $4\pi\:\alpha>0$ you can see that the solution indeed dies away very rapidly, much more that typically linear Fokker–Planck solutions, but the non-linearity makes it hard to see analytically how this happens.

$\endgroup$
3
  • $\begingroup$ Thanks for the answer! Going through it in detail I don't actually see why $\int dx' \nabla' \rho(x') / |x-x'| = 0$, could you elaborate on this? Clearly $\nabla \rho$ is antisymmetric and $1 / |x|$ is symmetric, but since they are integrated with a shift I don't see why everything must cancel. $\endgroup$ Dec 2, 2021 at 11:12
  • $\begingroup$ @Peter . Since the gradient has only radial component it is proportional to $ \hat{\mathbf{r}}=\sin\theta'\cos\varphi'\hat{\mathbf{x}}+\sin\theta'\sin\varphi'\hat{\mathbf{y}}+\cos\theta'\hat{\mathbf{z}}$. The first two terms do not contribute because the $\varphi'$ integral vanishes, and you are left with $ \int_0^{2\pi}d\theta'\:\cos\theta'\sin\theta'\Big/\sqrt{r^2+r'^2-2rr'\cos\theta'}=0 $. $\endgroup$ Dec 2, 2021 at 13:11
  • $\begingroup$ Ok, and the integral over $\theta'$ (which should be from $0$ to $\pi$, right?) is $\int_{-1}^1 (x / \sqrt{r^2 + r'^2 - 2rr'x}) dx$, which should not always give zero, or am I missing something? Say if $r = r' = 1$, then it is $2 / 3$. $\endgroup$ Dec 2, 2021 at 13:29

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.