Parametric estimation of pairwise Gibbs point processes with infinite range interaction (Q520700)

The authors study infinite range interaction Gibbs point processes. The processes are observed on a sequence of bounded domains converging to \(\mathbb{R}^d\). The case of parametric Gibbs measures is studied. The main objects of this study are the pseudo-likelihood and the logistic regression estimators. Their advantage compared to the maximum likelihood is that they do not involve a normalizing constant. Numerous results about properties of the pseudo-likelihood and logistic regression estimators were established in the literature under finite range interaction assumptions. The authors extend them to the case of infinite range interactions. In particular, the strong consistency of the estimator is proved for Ruelle superstable and lower regular models. The paper derives a central limit theorem for triangular arrays of almost conditionally centered random fields. Then, this result is applied to obtain the asymptotic normality of the estimators. A simulation study is conducted to study the performance of maximum pseudo-likelihood estimators for Lennard-Jones model.