6.1. Newton Solver

BEGIN SOLVER
  BEGIN NEWTON
     #
     # convergence criteria commands
     #
     TARGET RESIDUAL = <real>target_resid
       [DURING <string list>period_names]
     TARGET RELATIVE RESIDUAL = <real>target_rel_resid(1.0e-4)
       [DURING <string list>period_names]
     ACCEPTABLE RESIDUAL = <real>accept_resid
       [DURING <string list>period_names]
     ACCEPTABLE RELATIVE RESIDUAL = <real>accept_rel_resid
       [DURING <string list>period_names]
     REFERENCE = EXTERNAL|INTERNAL|BELYTSCHKO|RESIDUAL|ENERGY
       (EXTERNAL) [DURING <string list>period_names]
     RESIDUAL NORM TYPE = ALL|TRANSLATION|SCALE_RB_ROTATIONS
       (ALL) [DURING <string list>period_names]
     #
     # iteration control
     #
     MINIMUM ITERATIONS = <integer>min_iter(0)
       [DURING <string list>period_names]
     MAXIMUM ITERATIONS = <integer>max_iter
       [DURING <string list>period_names]
     #
     # Selection of the linear solver for use in solving
     # linearized Newton iterations
     #
     LINEAR SOLVER = <string>linear_solver_name
  END
END

The Newton solver is an nonlinear equation solver that is an alternative to the default conjugate gradient (CG) solver. Each iteration of the Newton solver consists of reforming current tangent stiffness matrix and re-solving the equation set with that current tangent. The Newton solver is typically significantly more expensive than the CG solver but may be more robust if there is substantial nonlinearity occurring over a time step. The Newton solver may also potentially prevent overshooting of yielding or other material nonlinearity. If a model has highly nonlinear materials and is failing to converge with the CG solver the Newton solver may be worth trying.

The convergence criteria and iteration control commands in the Newton solver behave identically to the equivalent commands in the CG solver. The LINEAR SOLVER specifies the solver to use during the linearized equation solution step each Newton iteration. The FETI solver is recommended but any available linear may work.