assemble_matrix assemble_rhs compute_hardav_staggered compute_nuH_norm compute_nuH_staggered compute_nuH_staggered_cfbc diagnostics_impl fracture_induced_softening init_impl integrated_viscosity is_marginal m_A m_b m_default_pc_failure_count m_default_pc_failure_max_count m_hardness m_KSP m_nuH m_nuH_old m_nuh_viewer m_nuh_viewer_size m_scaling m_velocity_old m_view_nuh m_work pc_setup_asm pc_setup_bjacobi picard_iteration picard_manager picard_strategy_regularization solve SSAFD update_nuH_viewers write_system_petsc ~SSAFD

compute_nuH_norm() void pism::stressbalance::SSAFD::compute_nuH_norm ( double &  norm, double &  norm_change  ) protectedvirtual Compute the norm of nu H and the change in nu H. Verification and PST experiments suggest that an \(L^1\) criterion for convergence is best. For verification there seems to be little difference, presumably because the solutions are smooth and the norms are roughly equivalent on a subspace of smooth functions. For PST, the \(L^1\) criterion gives faster runs with essentially the same results. Presumably that is because rapid (temporal and spatial) variation in \(\nu H\) occurs at margins, occupying very few horizontal grid cells. For the significant (e.g.~in terms of flux) parts of the flow, it is o.k. to ignore a bit of bad behavior at these few places, and \(L^1\) ignores it more than \(L^2\) (much less \(L^\infty\), which might not work at all). Definition at line 1256 of file SSAFD.cc. References pism::array::Array::add(), pism::Component::m_grid, m_nuH, m_nuH_old, and pism::array::Array::norm(). Referenced by picard_manager().