Large systems of interacting particles and the porous medium equation (Q811033)

Let \(X^ k_ N(t)\), \(k=1,...,N\), denote the positions of N interacting particles in \(R^ d\), which evolve according to \[ (d/dt)X^ k_ N(t)=-N^{-1}\sum^{N}_{m=1;m\neq k}\nabla V_ N(X^ k_ N(t)-X^ m_ N(t)),\quad k=1,...,N, \] for some nice symmetric interaction potential \(V_ N\), obtained from a fixed function \(V_ 1\) by the scaling \[ V_ N(x)=N^{\beta}V_ 1(N^{\beta /d}x),\quad \beta \in (0,1]. \] Let \(X_ N(t)\) denote the probability measure valued empirical processes \[ X_ N(t)=N^{-1}\sum^{N}_{k=1}\delta_{X^ k_ N(t)}. \] Under the assumptions of convergence of \(X_ N(0)\) to the function \(p_ 0\) as \(N\to \infty\), existence of an asymptotic density, and a hypothesis of local equilibrium, it is shown that \(X_ N(t)\) converges for each \(t\geq 0\), as \(N\to \infty\), to p(\(\cdot,t)\), where p is the solution of a porous medium equation with initial datum \(p_ 0\). Depending on the scaling parameter \(\beta\), different versions are obtained of the porous medium equation as limit dynamics. In particular, Boussinesq's equation is obtained for \(\beta\in (0,1)\), and the hydrodynamic limit of a one-dimensional particle system with gradient dynamics is obtained for \(\beta =1.\) Similar questions have been treated by the author when the dynamics of the \(X^ k_ N\) involve added Brownian motions [Z. Wahrscheinlichkeitstheor. Verw. Geb. 69, 279-322 (1985; Zbl 0549.60071) and Probab. Theory Relat. Fields 74, 591-616 (1987; Zbl 0592.60064)].