On 21 Nov 2013, at 12:00, [email protected] wrote: > Send fenics mailing list submissions to > [email protected] > > To subscribe or unsubscribe via the World Wide Web, visit > http://fenicsproject.org/mailman/listinfo/fenics > or, via email, send a message with subject or body 'help' to > [email protected] > > You can reach the person managing the list at > [email protected] > > When replying, please edit your Subject line so it is more specific > than "Re: Contents of fenics digest..." > > > Today's Topics: > > 1. Newton and SNES parameters (Anders Logg) > > > ---------------------------------------------------------------------- > > Message: 1 > Date: Thu, 21 Nov 2013 09:23:48 +0100 > From: Anders Logg <[email protected]> > To: FEniCS mailing list <[email protected]> > Subject: [FEniCS] Newton and SNES parameters > Message-ID: <20131121082348.GP7436@ancalagon> > Content-Type: text/plain; charset="us-ascii" > > I have made some changes to the handling of parameters in NewtonSolver > and PETScSNESSolver in order to fix this bug: > > https://bitbucket.org/fenics-project/dolfin/issue/10/nonlinearvariationalsolver-does-not-pass > > The change involves pushing the linear solver parameters down one > level from NonlinearVariationalSolver to NewtonSolver and PETScSNESSolver > where these parameters are actually used. > > Please have a look to see if the changes are ok, especially regarding > PETScSNESSolver which I have never used myself: > > https://bitbucket.org/fenics-project/dolfin/branch/logg/fix-issue-10 > > The SNES demo has been updated and seems to work fine. > I tested the part on the linear_solver and and preconditioner and it works well. I agree that it is much more logical to have them as parameters of NewtonSolver and PETScSNESSolver. You just need to modify also the test in test/unit/nls/python/PETScSNESSolver.py for the new parameter layout.
However, the issue is larger than that. Many other parameters are not passed, although existing in snes and newton_solver. For example it is not possible to set the tolerance or the max iterations of the krylov_solver used in the newton or snes iterations, which is very bad in my opinion. I proposed a (ugly) fix some time ago which was declined ( https://bitbucket.org/fenics-project/dolfin/pull-request/34/updating-of-ksp-parameters-in/diff ). However I think that something must be done, because many parameters are shown and modifiable, but they are not taken into account. To test on snes_solver, run demo_contact_vi-snes.py with snes_solver_parameters = {"nonlinear_solver": "snes", "snes_solver" : { "linear_solver" : "gmres", "preconditioner" : "amg", "maximum_iterations": 20, "report": True, "error_on_nonconvergence": False, "krylov_solver": {"maximum_iterations": 1, "report": True} }} The issue is that "krylov_solver": {...} are ignored. To test on newton_solver, consider the following modify version of demo_hyperelasticity, where par_new.krylov_solver are ignored: from dolfin import * # Optimization options for the form compiler parameters["form_compiler"]["cpp_optimize"] = True ffc_options = {"optimize": True, \ "eliminate_zeros": True, \ "precompute_basis_const": True, \ "precompute_ip_const": True} # Create mesh and define function space mesh = UnitCubeMesh(24, 16, 16) V = VectorFunctionSpace(mesh, "Lagrange", 1) # Mark boundary subdomians left = CompiledSubDomain("near(x[0], side) && on_boundary", side = 0.0) right = CompiledSubDomain("near(x[0], side) && on_boundary", side = 1.0) # Define Dirichlet boundary (x = 0 or x = 1) c = Expression(("0.0", "0.0", "0.0")) r = Expression(("scale*0.0", "scale*(y0 + (x[1] - y0)*cos(theta) - (x[2] - z0)*sin(theta) - x[1])", "scale*(z0 + (x[1] - y0)*sin(theta) + (x[2] - z0)*cos(theta) - x[2])"), scale = 0.5, y0 = 0.5, z0 = 0.5, theta = pi/3) bcl = DirichletBC(V, c, left) bcr = DirichletBC(V, r, right) bcs = [bcl, bcr] # Define functions du = TrialFunction(V) # Incremental displacement v = TestFunction(V) # Test function u = Function(V) # Displacement from previous iteration B = Constant((0.0, -0.5, 0.0)) # Body force per unit volume T = Constant((0.1, 0.0, 0.0)) # Traction force on the boundary # Kinematics I = Identity(V.cell().d) # Identity tensor F = I + grad(u) # Deformation gradient C = F.T*F # Right Cauchy-Green tensor # Invariants of deformation tensors Ic = tr(C) J = det(F) # Elasticity parameters E, nu = 10.0, 0.3 mu, lmbda = Constant(E/(2*(1 + nu))), Constant(E*nu/((1 + nu)*(1 - 2*nu))) # Stored strain energy density (compressible neo-Hookean model) psi = (mu/2)*(Ic - 3) - mu*ln(J) + (lmbda/2)*(ln(J))**2 # Total potential energy Pi = psi*dx - dot(B, u)*dx - dot(T, u)*ds # Compute first variation of Pi (directional derivative about u in the direction of v) F = derivative(Pi, u, v) # Compute Jacobian of F J = derivative(F, u, du) # Solve variational problem problem = NonlinearVariationalProblem(F,u,bcs,J,form_compiler_parameters=ffc_options) solver = NonlinearVariationalSolver(problem) par_new = solver.parameters.newton_solver par_new.linear_solver = "gmres" par_new.preconditioner = "amg" par_new.krylov_solver["report"] = True par_new.krylov_solver["monitor_convergence"] = True par_new.krylov_solver["maximum_iterations"] = 5 info(par_new, True) solver.solve() > -- > Anders > > > ------------------------------ > > _______________________________________________ > fenics mailing list > [email protected] > http://fenicsproject.org/mailman/listinfo/fenics > > > End of fenics Digest, Vol 8, Issue 17 > ************************************* Corrado Maurini [email protected] _______________________________________________ fenics mailing list [email protected] http://fenicsproject.org/mailman/listinfo/fenics
