The p-type q-tuple model with migration
isomorfic to K
p −1
, it is then possible to prove strong uniqueness of solutions to 1.1.48 by induction. For this technique, see Example 3.1.8 below.
For c sufficiently large, it is possible to show that solutions to the martingale problem for A
p,q
+ B
θ, c
never hit the boundary ∂ K
p
, so that solutions to 1.1.48 are strongly unique for all time. For this technique, see Theorem 2.2.9 below.
However, for c positive but not too large the process X with probability one reaches the boundary ∂ K
p
in a finite time. In fact, it hits ∂ K infinitely often in a finite time, each time bouncing back from it. About processes with such behavior
very little is known. Certainly the standard strong uniqueness results do not apply, except in the one-dimensional i.e. 2-type case.
For the p-type 2-tuple model, there are several ways to circumvent this prob- lem. For example, it is possible to find a root σ that is lower-triangular. This
corresponds the fact that each subselection of the colors is itself a Markov process following a Wright-Fisher diffusion. Thus, the idea is that one can first prove
strong uniqueness for one color, using one-dimensional techniques, then prove strong uniqueness for the second color conditional on the first one, and so on.
In another approach, one can prove weak uniqueness for the p-type 2-tuple model with migration by more or less explicitly calculating all moments of X t.
Here one uses the fact that A
p,2
+ B
θ, c
maps a polynomial of degree n into a polynomial of degree at most n. Thus, the time evolution of all moments up to
n-th order is described by a closed system of equations, that is easily seen to have a unique solution.
3
This technique has the advantage that it also proves that the closure of A
p,2
+ B
θ, c
generates a Feller semigroup. For general p-type q-tuple models, one can see that A
p,q
maps a polynomial of degree n into a polynomial of degree n
− 2 + q, and hence for q ≥ 3 the time evolution of moments up to n-th order cannot be expressed in a closed system.
It seems that duality techniques involving moments that are known to work in certain other models also fail here, and uniqueness of solutions to the martingale
problem for A
p,q
+ B
θ, c
with p, q ≥ 3, for general θ ∈ K
p
and c 0, is still an open problem.