Herbert Egger,Felix Engertsberger,Lukas Domenig,Klaus Roppert,Manfred Kaltenbacher
Abstract:Fixed-point or Newton-methods are typically employed for the numerical solution of nonlinear systems arising from discretization of nonlinear magnetic field problems. We here discuss an alternative strategy which uses local Quasi-Newton updates to construct appropriate linearizations of the material behavior during the nonlinear iteration. The resulting scheme shows similar fast convergence as the Newton-method but, like the fixed-point methods, does not require derivative information of the underlying material law. As a consequence, the method can be used for the efficient solution of models with hysteresis which involve nonsmooth material behavior. The implementation of the proposed scheme can be realized in standard finite-element codes in parallel to the fixed-point and the Newton method. A full convergence analysis of all three methods is established proving global mesh-independent convergence. The theoretical results and the performance of the nonlinear iterative schemes are evaluated by computational tests for a typical benchmark problem.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is how to efficiently construct a linearization method for material behavior when numerically solving nonlinear magnetics problems. Specifically, the paper explores a method using local Quasi - Newton updates to construct the linearization of material behavior in order to accelerate the convergence rate in the nonlinear iteration process. This method combines the advantages of fixed - point methods and Newton methods:
1. **Only need to access the material law**: Unlike the Newton method, this method does not require derivative information of the material law, which enables it to be applied to problems with non - differentiable material behavior, such as hysteresis phenomena.
2. **Fast convergence**: Similar to the Newton method, this method can achieve fast convergence, while the fixed - point method may converge slowly when dealing with strongly nonlinear materials.
### Main contributions
- **Local Quasi - Newton updates**: By applying the low - rank update formula on each grid cell, significant information about the linearized material behavior, namely the Jacobian matrix $\frac{\partial \mathbf{b}}{\partial \mathbf{h}}$ or its generalized form, is extracted from the previous magnetic field intensity $\mathbf{h}$ and magnetic flux density $\mathbf{b}(\mathbf{h})$ values.
- **Global convergence**: Through an appropriate step - size selection strategy, the global r - linear convergence of this method is proved, and the convergence rate is independent of the finite - element mesh.
- **Wide applicability**: This method is applicable not only to the discretization of scalar potentials, but also to the discretization of vector potentials and high - order approximations, and can even be generalized to the numerical solution of other nonlinear partial differential equations.
### Numerical experiments
The paper verifies the effectiveness of the proposed method through multiple numerical experiments:
- **Nonlinear magnetostatic problems**: For nonlinear magnetostatic problems without hysteresis, the Newton method, the fixed - point method, and two methods using local Quasi - Newton updates (BFGS and DFP) are compared. The results show that the convergence speed of the local Quasi - Newton method is close to that of the Newton method and much faster than that of the fixed - point method.
- **Models with hysteresis**: For material models with hysteresis effects, the local Quasi - Newton method also performs well, being able to achieve a relatively fast convergence speed while ensuring global convergence.
- **Typical load - cycle simulations**: In practical applications, a series of excitation situations of load currents are considered. The results indicate that when dealing with multi - step loading problems, the average number of iterations of the local Quasi - Newton method is significantly less than that of the fixed - point method and is comparable to that of the Newton method.
### Conclusion
This paper proposes a method using local Quasi - Newton updates to construct the linearization of material behavior in nonlinear magnetics problems. This method, while maintaining fast convergence, avoids the need for derivative information of the material law and is applicable to complex material models with hysteresis effects. Through detailed theoretical analysis and numerical experiments, the effectiveness and robustness of this method are proved.