POLYFLOW  convergence/divergence, accuracy, precision for a power law or BirdCarreau fluid model with a low index (nonNewtonian inelastic or generalised Newtonian)
Flow simulations that have a generalised Newtonian (nonNewtonian inelastic) fluid model, such as power law or BirdCarreau, with a low powerlaw index may quickly diverge. What can be done to improve the success of these calculations? For a shearratedependent viscosity, POLYFLOW uses Newton iterations to compute the viscosity from the local shear rate, which is the default option. For powerlaw or BirdCarreau fluids with a low power index, Picard (fixedpoint) iteration is recommended when evolution is not activated. The Picard scheme typically requires 20 to 30 iterations to converge to a tolerance of 1e04, while the Newton scheme often requires 4 to 5 iterations. However, the radius of convergence for the Newton scheme is smaller for powerlaw fluids with a low index, i.e. a better initial solution guess is needed to ensure convergence. The Picard scheme provides better convergence behaviour when the powerlaw index is low. For powerlaw or BirdCarreau fluids with a low powerlaw index, the default Newton iteration can be used in combination with an evolution scheme that gradually decreases the powerlaw index. See Section 10.3.3 in the User's manual for details. 

