Exact simulation of the Wright-Fisher diffusion (Q2403134)

Monte Carlo simulation of diffusion processes is of great interest, as it underlies methods of statistical inference from discrete observation in a variety of applications. In the present paper the Wright-Fisher diffusion is considered. This process is widely used for inference, especially in genetics, where it serves as a model for the evolution of the frequency. It is demonstrated that is possible to simulate exactly from a broad class of Wright-Fisher diffusion processes and their bridges. The key idea of the paper is to exploit an eigenfunction expansion of the transition function. An exact rejection algorithm for processes with more general drift functions, including those modelling natural selection, is developed. The proposed approach also yields methods for exact simulation of the moment dual of the Wright-Fisher diffusion, the ancestral process of an infinite-leaf Kingman's coalescent tree. The frequency \(X_t \in [0,1]\) is given as a solution of a one-dimensional stochastic differential equation \[ d X_t = \gamma(X_t) + \sqrt{X_t(1 - X_t)} d B_t, \;\;X_0 = x_0, \;\;t \in [0,T], \tag{1} \] with a drift coefficient \(\gamma,\) which can encompass a variety of evolutionary forces. In the introduction of the paper some simulations of the equation (1) are discussed. In Section 2 it is demonstrated how exact simulation from the neutral Wright-Fisher diffusion can be achieved. In Algorithm 1 a simulation from the transition density of the neutral Wright-Fisher diffusion is given. In Algorithm 2 a simulation from the ancestral process of Kingman's coalescent with mutation is given. Also, a transition density expansion in higher dimensions are discussed. In Algorithm 3 a simulation from the transition density of the neutral Fleming-Viot process with parent-independent mutation is developed. In Section 3 it is demonstrated how exact simulation from the neutral Wright-Fisher diffusion bridges can be achieved, via a new probabilistic description of its transition density. In Algorithm 4 a simulation from the transition density of a bridge of the neutral Wright-Fisher diffusion with mutation is given. In Algorithm 5 a simulation from the discrete random variable is given. In Section 4 the developed algorithms are applied to neutral processes. The correctness of the proposed algorithms and their performance are investigated by an application to the Euler-Maruyama simulation. The obtained results are graphically illustrated. In Section 5 an exact rejection algorithm for simulation of the Wright-Fisher diffusion with general drift is developed. First, a small overview of the concept for exact algorithm is made and the exact algorithm for simulating the path of a diffusion process with law is proposed. Second, the exact algorithm is applied to the Wright-Fisher diffusion. In Section 6 an algorithm for simulating a nonneutral Wright-Fisher bridge is provided. In Section 7 it is discussed how to simulate exactly from the scalar Wright-Fisher diffusion, as well as a number of important and closely related processes: the ancestral process of an \(\infty\)-leaf Kingman' s coalescent tree, the Fleming-Viot process with parent-independent mutation, the nonneutral Wright-Fisher diffusion and neutral and nonneutral Wright-Fisher bridge. In Section 8 the proofs of some propositions are given.