User:Tomoyuki627298297/sandbox

Rank-based Particle Systems
Among interacting particle systems, a rank-based particle model is defined as a multi-dimensional Markov_process whose dynamics depend  on the order in which the coordinates are ranked. Let $$X_{i}(t)$$  be the position of particle $$\, i = 1, \ldots, n \,$$  on the real line at time $$\, t\ge0\,$$, and denote by  $$\, \max_{1 \le i \le n} X_i(t) = X_{(1)}(t) \ge X_{(2)}(t) \ge \ldots \ge X_{(n)}(t)=\min_{1 \le i \le n} X_i(t) \,\,$$  the descending order statistics of the positions. Ties are resolved  lexicographically,    or by randomization. The diffusion processes in the Atlas models are prototypical examples of  rank-based particle systems. A general such system can be rewritten as

$$\begin{align} d X_{i}(t) = b_{i} (F(X_{i}(t); \rho^{n}(t))) d t + \sigma_{i} (F(X_{i}(t); \rho^{n}(t))) d W_{i}(t) \,, \qquad i=1, \cdots, n \, , \end{align} $$

where $$\rho^{n}(t) = \sum_{i=1}^{n} \delta (X_{i}(t)) \in {\mathbf M}\,$$ is the empirical measure of the particle system, $$F(\cdot \,; \rho^{n}(t))$$ is the corresponding cumulative distribution function, $$ (W_1(\cdot), \ldots, W_n(\cdot)) $$ is an $$n$$-dimensional Brownian motion and $$\, b_{i}:[0, 1] \times {\mathbf M} \to {\mathbf R}\,$$,  $$\sigma_{i}:[0,1] \times {\mathbf M} \to (0, \infty)$$ are given functions  (Shkolnikov (2012)).

Collisions of Particles
Since the collisions of particles are incidents of rank change, it is significant to understand how and when the particles collide (Harris (1965)). Three independent Brownian particles on the real line (components of three-dimensional standard Brownian motion) never collide at the same time; however, there may be collisions of three or more particles with some positive probability at a finite time for general rank-based particle systems (Bass and Pardoux (1987), Ichiba and Karatzas (2010)). A Brownian motion in a Weyl chamber is an example of an interacting particle system, where the process terminates if two particles collide.

Long-term Behaviors
The long-term asymptotic behavior of rank-based particle systems can be understood through  projections onto lower-dimensional manifolds. For example, the $$n$$-dimensional rank-based particle process in the Atlas models is projected on the hyperplane $$\, \{x \in {\mathbf R}^{n}: x_{1} + \cdots + x_{n} = 0\}\, $$; then the projected process is the centered process obtained by subtracting the average from each of the coordinates. Another example of projected processes is the gap process $$\, (X_{(1)}(\cdot) - X_{(2)}(\cdot), \ldots, X_{(n-1)}(\cdot) - X_{(n)}(\cdot))\, $$.

Under appropriate conditions, the projected process is positive recurrent in   neighborhoods of the origin, and thus   has a unique stationary distribution (e.g., Banner, Fernholz and Karatzas (2005), Ichiba et al. (2011), Chatterjee and Pal (2011) for the finite-dimensional case, and Pal and Pitman (2008), Chatterjee and Pal (2010) for the infinite-dimensional case). A similar quasi-invariant distribution is considered by Ruzmaikina and Aizenman (2005) and Arguin and Aizenman (2009), who connect the Ruelle Probability Cascades and hierarchically nested Poisson-Dirichlet processes with infinite interacting particle systems. The rate of convergence to the stationary distribution in the Atlas type models is studied by Jourdain and Malrieu (2008), Pal and Shkolnikov (2011), Ichiba, Pal and Shkolnikov (2013).

Infinite Dimension
Nonlinear evolution equations for porous media emerge as limits of finite-dimensional rank-based particle systems (Jourdain (2000 a,b), Shkolnikov (2011), Jourdain and Reygner (2012), Chatterjee and Pal (2010)). Under appropriate conditions, as the number of particles $$\, n\, $$ tends to infinity,  the limiting particle density follows the McKean-Vlasov equation, and the corresponding cumulative distribution functions evolve according to the porous medium equation with convection (Shkolnikov (2012), Dembo, Shkolnikov, Varadhan and Zeitouni (2012)).