Efficient inexact Newton-like methods with application to problems of the deformation theory of plasticity (Q1321610)

The authors consider inexact Newton-like methods for numerically solving large nonlinear equations of the form: \(A(u)u=b\), arising from the discretization of nonlinear elliptic problems. The Newton-like methods consist of steps of the form: solve the linear system \(B_ i d_ i=b- A(u_ i)u_ i\) for \(d_ i\) and set \(u_{i+1}:=u_ i+(r_ i)d_ i\), where \(r_ i>0\) is a damping parameter and \(B_ i\) is a suitable matrix, e.g., \(B_ i=A(u_ 0)\) or \(A(u_ i)\). The former case is called the generalized Picard method and the latter is called the secant modulus method. The linear systems are solved via an iterative solver such as preconditioned conjugate gradients. The authors study the convergence of such methods and the efficient control of accuracy in the inexact correction. Numerical experiments show the efficiency of the suggested techniques when problems involving the deformation theory of plasticity are treated.