Comparison between the deterministic and stochastic models of nonlocal diffusion (Q6195984): Difference between revisions

The paper under review is to study a nonlocal reaction-diffusion equation and a multi-dimensional jump Markov process to analyze the difference between the deterministic and stochastic models, and finds the difference converges to zero in probability with a supremum norm for a sizeable network and further converges to a stochastic process in distribution on the Skorokhod space. Fife (2003) describes the nonlocal reaction-diffusion equation in Hutson et al., (2003) on the integral term as the rate at individuals arrive at position from all other positions, and the other non-integral term as the rate at which they leave position to travel to any other site plus the last reaction term. The equation can be used for neural firing in dispersal in Murray (2003), for pattern formation in biology in Kondo (2017), for the SIS or SIR epidemic model in Yang et al., (2017). Kurtz (1970, 1971) first conducted the study on a Markov process as a discretized model of the diffusion that converges to its differential equation, and Arnold and Theodosopulu (1980) constructed a space time jump Markov process for a chemical reaction that involves a simple random walk on a discrete torus with a diffusion rate. The methods and proofs in this paper are originally come from Blount (1987, 1991, 1992) that can be applied not only local diffusion but also nonlocal case. Section 2 starts with a deterministic model of nonlocal diffusion (referring to Pazy (1983)'s book Chapter 6), and Section 3 gives the level of particle change accumulated in the k-th site, the number of events in the k-th site, the number of simultaneous transition event between k-th and i-th sites as a sigma-algebra martingales in Lemma 3.1, and writes out the stochastic process \(X^N(t)\). Using the martingale and stopping time properties, Lemma 3.2 formulates a mean 0 martingale which is important for evaluating certain formulae. Section 4 compares the solution of deterministic model in Section 2 and the stochastic process \(X^N(t)\) using the law of large numbers. The main result Theorem 4.1 shows that \(X^N(t)\) converges to the solution of deterministic model in probability for a particular regime. With aids of some preparation lemmas, the proof of Theorem 4.1 is completed by representing the function \(\overline{X}^N(t)\) starting from the value \(X^N\) at the stopping time \(\tau\) and comparing the difference with this replacing, using the semigroup and Gronwell's inequality to squeeze the convergence in probability. Section 5 checks whether the nonlocal diffusion using a stochastic differential equation at another scale to have the weak convergence result. With some extra assumptions for the scale ratio and \(L^2\)-converging , Theorem 5.1 states the weak convergence result, where thanks for Blount (1991) for the existence of a stochastic process \(M(t)\) as a stochastic fluctuation with desired properties for a compact support in probability, relative compact on \(D_{L^2}[0, \infty)\) for the sequence of \(M^N(t)\), and convergence term by term from various lemmas. Appendix as Section 6 provides technical proofs of Lemma 4.3 and 5.2. Hence the paper provides explicit form of semigroup generated by diffusion operator.