François Bachoc
We consider a Gaussian process subjected to inequality constraints (for instance boundedness, monotonicity or convexity). These types of inequality constraints correspond to additional information on the Gaussian process realization, that are regularly available in applications. We explain how to simulate from the conditional distribution of a constrained Gaussian process, given observed values. We also introduce a maximum likelihood estimator taking the constraints into account. We obtain its asymptotic properties and compare it to the usual maximum likelihood estimator, that does not take the constraints into account.
