As far as I have seen, solving problems of nonlinear elasticity using the finite element method proceeds by linearizing, either around the initial configuration (total Lagrangian approach) or around the most recently solved configuration (updated Lagrangian approach). This produces a system of linear equations, and represents an example of the Newton-Raphson method.

My question is about possible generalizations to other methods of solving systems of nonlinear equations. I realize that for engineering problems, the linearization may not cause any problems, because the goal is often to study deformations starting from a known, unstressed configuration, which can then be gradually loaded. In other words, there is a good small parameter, so linearization is fine. However, I'm interested in the case that external forces (loadings) depend on the deformation, and the dependence can exhibit strong nonlinearities. In this case, the global solution may not be anywhere near the undeformed configuration.

I think that attempts to address this problem are known as "globalization strategies", and I'm interested in the application of these to nonlinear elasticity. I know about methods like arc-length control, but I'm interested in other methods such as full multigrid.

I don't know enough about the lingo or the literature to even search the literature properly - so far my naive attempts have not yielded anything. I also don't know good examples of problems with strong nonlinearities that might be solved with finite element - maybe Navier-Stokes? I don't even see how to properly formulate the problem as a system of nonlinear equations - do I just numerically integrate the principal of virtual work?

  • 1
    $\begingroup$ I think that your equilibrium configuration being "far away" from your initial one does not render Newton-Raphson useless. What might happen is that your mesh get too distorted to be useful and you need to remesh. I think this is one of the reasons that motivated the development of meshfree methods. $\endgroup$ – nicoguaro Mar 13 '18 at 14:03
  • $\begingroup$ @nicoguaro - thanks for your comment, but it's not really an answer to my question. I don't see Newton-Raphson as useless; rather, I'm interested in considering alternatives. Can you recommend a book or article on meshfree methods? $\endgroup$ – OskarM Mar 13 '18 at 16:23
  • 1
    $\begingroup$ The case for deformation dependent loading is a relevant engineering problem (i.e. inflation of membrane-like structures). Raphson-Newton methods are usually used to obtain the solution. The problem with deformation-dependent loads is that the tangent matrix will be non-symmetric which will impact the convergence. There are alternatives to the Raphson-Newton method which probably won't perform better. $\endgroup$ – P. G. Mar 14 '18 at 12:13
  • $\begingroup$ @P.G. - thanks for your comment. This statement about non-symmetric stiffness matrices arising from deformation-dependent loads has been floating around for a long time, but I think it's settled. If the forces are derived from a potential, i.e. are conservative, then the stiffness matrix remains symmetric, as the part derived from the potential is essentially the Hessian, which is symmetric. Can you say more about actually employing the alternatives? $\endgroup$ – OskarM Mar 14 '18 at 15:10

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.