I am implementing a strain-gradient plasticity model as a UEL in Abaqus. My unknowns are nodal displacements and nodal plastic strains, hence my stiffness matrix calculations involve both displacement(K_u,u & K_u,eps) and plastic strain(K_eps,u & K_eps,eps).  The problem I have is that the first iteration of the first increment approximates the unknowns to be zeros, which is correct, but this causes my calculations for K_eps,eps to have NaN(undefined) values which then causes the subsequent iterations and increments to also have NaN values. The Abaqus/Standard analysis executes and completes successfully but all I have are undefined values to my unknowns and a nodal force vector(RHS)  of undefined values.

To try and resolve this issue, I want to define my stiffness matrix(AMATRX) using only the displacements contribution(i.e  directly prescribe K_eps,eps to be zeros) for the first iteration of the first increment so that I get the first approximations of the nodal displacement and nodal plastic strain. Then use this approximations to calculate the updated stiffness matrix( including K_eps,eps).

My question is how do I define the first iteration of an increment?