Issues with implementation of strain gradient plasticity framework of Niordson and Hutchinson (2003) in Abaqus UEL subroutine

I have implemented the rate-independent strain gradient plasticity framework of Niordson and Hutchinson (2003) in UEL. I was able to validate the results of the shear of infinitely wide thin layer between two rigid surfaces (Niordson and Hutchinson, 2003). However, the UEL subroutine fails to converge for the more complex problem of the shear of finite width slab. I suspect the convergence issues are arising out of internal elastic-plastic boundary conditions and the yield criteria. I have few doubts in this regard.

1. It is mentioned in Niordson and Hutchinson (2003) that whenever a negative plastic increment is obtained at a gauss point in the previous increment, the gauss point is elastically unloaded. What would be the stress update method in this situation ? Would it be dQ=0, d\tau=0 and d\sigma = L*d\epsilon (since d\epsilon^p =0). Also, in the elastic region q=\sigma_e. Hence, during an elastic unloading (negative plastic strain increment) one should also assign Q=\sigma_e.

2. During such a step, where one has found a negative plastic increment in the previous increment, the stiffness matrix K_p should also be set to a very large number such as K_p = 10^8 E, for that gauss point.

3. It is also mentioned that the plastic yielding starts again when Q reaches Qy. Does it mean that once an elastic unloading of a gauss point is done, one would solve the system with the condition mentioned in the previous item (2) =\s Q reaches Qy. And during this period is Q=\sigma_e althrough ? Also, does Qy correspond to the value of Q at the increment when the elastic unloading is initiated ?

It would be of great help if anyone can shed some light on this issue.

Many thanks.