Merge pull request #275 from mfem/mfem_48_dev_ex40

Mfem 48 dev ex40

Author: sshiraiwa

examples/ex39.py

@@ -249,7 +249,7 @@ def run(order=1,
 if __name__ == "__main__":
     from mfem.common.arg_parser import ArgParser
 
-    parser = ArgParser(description='Ex1 (Laplace Problem)')
+    parser = ArgParser(description='Ex39 (Named attribute)')
     parser.add_argument('-m', '--mesh',
                         default='compass.msh',
                         action='store', type=str,

examples/ex40.py

@@ -0,0 +1,327 @@
+'''
+   MFEM example 40 (converted from ex40.cpp)
+
+   See c++ version in the MFEM library for more detail
+
+   Sample runs: python ex40.py -step 10.0 -gr 2.0
+                python ex40.py -step 10.0 -gr 2.0 -o 3 -r 1
+                python ex40.py -step 10.0 -gr 2.0 -r 4 -m ../data/l-shape.mesh
+                python ex40.py -step 10.0 -gr 2.0 -r 2 -m ../data/fichera.mesh
+
+   Description: This example code demonstrates how to use MFEM to solve the
+                eikonal equation,
+
+                        |∇𝑢| = 1 in Ω,  𝑢 = 0 on ∂Ω.
+
+                The viscosity solution of this problem coincides with the unique optimum
+                of the nonlinear program
+
+                     maximize ∫_Ω 𝑢 d𝑥 subject to |∇𝑢| ≤ 1 in Ω, 𝑢 = 0 on ∂Ω,    (⋆)
+
+                which is the foundation for method implemented below.
+
+                Following the proximal Galerkin methodology [1,2] (see also Example
+                36), we construct a Legendre function for the closed unit ball
+                𝐵₁ := {𝑥 ∈ Rⁿ | |𝑥| ≤ 1}. Our choice is the Hellinger entropy,
+
+                      R(𝑥) = −( 1 − |𝑥|² )^{1/2},
+
+                although other choices are possible, each leading to a slightly
+                different algorithm. We then adaptively regularize the optimization
+                problem (⋆) with the Bregman divergence of the Hellinger entropy,
+
+                   maximize  ∫_Ω 𝑢 d𝑥 - αₖ⁻¹ D(∇𝑢,∇𝑢ₖ₋₁)  subject to  𝑢 = 0 on Ω.
+
+                This results in a sequence of functions ( 𝜓ₖ , 𝑢ₖ ),
+
+                        𝑢ₖ → 𝑢,    𝜓ₖ/|𝜓ₖ| → ∇𝑢    as k → ∞,
+
+                defined by the nonlinear saddle-point problems
+
+                 Find 𝜓ₖ ∈ H(div,Ω) and 𝑢ₖ ∈ L²(Ω) such that
+                 ( (∇R)⁻¹(𝜓ₖ) , τ ) + ( 𝑢ₖ , ∇⋅τ ) = 0                     ∀ τ ∈ H(div,Ω)
+                 ( ∇⋅𝜓ₖ , v )                     = ( ∇⋅𝜓ₖ₋₁ - αₖ , v )    ∀ v ∈ L²(Ω)
+
+                where (∇R)⁻¹(𝜓) = 𝜓 / ( 1 + |𝜓|² )^{1/2} and αₖ = α₀rᵏ, where r ≥ 1
+                is a prescribed growth rate. (r = 1 is the most stable.) The
+                saddle-point problems are solved using a damped quasi-Newton method
+                with a tunable regularization parameter 0 ≤ ϵ << 1.
+
+                [1] Keith, B. and Surowiec, T. (2024) Proximal Galerkin: A structure-
+                    preserving finite element method for pointwise bound constraints.
+                    Foundations of Computational Mathematics, 1–97.
+                [2] Dokken, J., Farrell, P., Keith, B., Papadopoulos, I., and
+                    Surowiec, T. (2025) The latent variable proximal point algorithm
+                    for variational problems with inequality constraints. (To appear.)
+
+'''
+import os
+from os.path import expanduser, join
+import numpy as np
+
+import mfem.ser as mfem
+
+if hasattr(mfem, "UMFPackSolver"):
+    use_umfpack = True if not args.no_use_umfpack else False
+else:
+    use_umfpack = False
+
+
+def run(meshfile="",
+        order=1,
+        max_it=5,
+        ref_levels=3,
+        alpha=1.0,
+        growth_rate=1.0,
+        newton_scaling=0.8,
+        eps=1e-6,
+        tol=1e-4,
+        visualization=True):
+
+    # 2. Read the mesh from the mesh file.
+    mesh = mfem.Mesh(meshfile, 1, 1)
+    dim = mesh.Dimension()
+    sdim = mesh.SpaceDimension()
+
+    # 3. Postprocess the mesh.
+    # 3A. Refine the mesh to increase the resolution.
+    for i in range(ref_levels):
+        mesh.UniformRefinement()
+
+    # 3B. Interpolate the geometry after refinement to control geometry error.
+    # NOTE: Minimum second-order interpolation is used to improve the accuracy.
+    curvature_order = max(order, 2)
+    mesh.SetCurvature(curvature_order)
+
+    # 4. Define the necessary finite element spaces on the mesh.
+    RTfec = mfem.RT_FECollection(order, dim)
+    RTfes = mfem.FiniteElementSpace(mesh, RTfec)
+
+    L2fec = mfem.L2_FECollection(order, dim)
+    L2fes = mfem.FiniteElementSpace(mesh, L2fec)
+
+    print("Number of H(div) dofs: " + str(RTfes.GetTrueVSize()))
+    print("Number of L² dofs: " + str(L2fes.GetTrueVSize()))
+
+    # 5. Define the offsets for the block matrices
+    offsets = mfem.intArray([0, RTfes.GetVSize(), L2fes.GetVSize()])
+    offsets.PartialSum()
+
+    x = mfem.BlockVector(offsets)
+    rhs = mfem.BlockVector(offsets)
+    x.Assign(0.0)
+    rhs.Assign(0.0)
+
+    # 6. Define the solution vectors as a finite element grid functions
+    #    corresponding to the fespaces.
+
+    u_gf = mfem.GridFunction()
+    delta_psi_gf = mfem.GridFunction()
+    delta_psi_gf.MakeRef(RTfes, x, offsets[0])
+    u_gf.MakeRef(L2fes, x, offsets[1])
+
+    psi_old_gf = mfem.GridFunction(RTfes)
+    psi_gf = mfem.GridFunction(RTfes)
+    u_old_gf = mfem.GridFunction(L2fes)
+
+    # 7. Define initial guesses for the solution variables.
+    delta_psi_gf.Assign(0.0)
+    psi_gf.Assign(0.0)
+    u_gf.Assign(0.0)
+    psi_old_gf.Assign(0.0)
+    u_old_gf.Assign(0.0)
+
+    # 8. Prepare for glvis output.
+    # 14. Send the solution by socket to a GLVis server.
+    if visualization:
+        sol_sock = mfem.socketstream("localhost", 19916)
+        sol_sock.precision(8)
+
+    # 9. Coefficients to be used later.
+    neg_alpha_cf = mfem.ConstantCoefficient(-1.0*alpha)
+    zero_cf = mfem.ConstantCoefficient(0.0)
+
+    psigf_cf = mfem.VectorGridFunctionCoefficient(psi_gf)
+
+    @mfem.jit.vector(shape=(sdim,), dependency=(psigf_cf,))
+    def Z(_ptx, psi_vals):
+        norm = np.linalg.norm(psi_vals)
+        phi = 1.0 / np.sqrt(1.0 + norm*norm)
+        vvec = psi_vals*phi
+        return vvec
+
+    @mfem.jit.matrix(shape=(sdim, sdim), dependency=(psigf_cf,))
+    def DZ(_ptx, psi_vals):
+        norm = np.linalg.norm(psi_vals)
+        phi = 1.0 / np.sqrt(1.0 + norm*norm)
+
+        kmat = np.zeros((sdim, sdim))
+        for i in range(sdim):
+            kmat[i, i] = phi + eps
+            for j in range(sdim):
+                kmat[i, j] -= psi_vals[i]*psi_vals[j] * phi**3
+        return kmat
+
+    neg_Z = mfem.ScalarVectorProductCoefficient(-1.0, Z)
+    div_psi_cf = mfem.DivergenceGridFunctionCoefficient(psi_gf)
+    div_psi_old_cf = mfem.DivergenceGridFunctionCoefficient(psi_old_gf)
+    psi_old_minus_psi = mfem.SumCoefficient(
+        div_psi_old_cf, div_psi_cf, 1.0, -1.0)
+
+    # 10. Assemble constant matrices/vectors to avoid reassembly in the loop.
+    b0 = mfem.LinearForm()
+    b1 = mfem.LinearForm()
+    b0.MakeRef(RTfes, rhs.GetBlock(0), 0)
+    b1.MakeRef(L2fes, rhs.GetBlock(1), 0)
+
+    b0.AddDomainIntegrator(mfem.VectorFEDomainLFIntegrator(neg_Z))
+    b1.AddDomainIntegrator(mfem.DomainLFIntegrator(neg_alpha_cf))
+    b1.AddDomainIntegrator(mfem.DomainLFIntegrator(psi_old_minus_psi))
+
+    a00 = mfem.BilinearForm(RTfes)
+    a00.AddDomainIntegrator(mfem.VectorFEMassIntegrator(DZ))
+
+    a10 = mfem.MixedBilinearForm(RTfes, L2fes)
+    a10.AddDomainIntegrator(mfem.VectorFEDivergenceIntegrator())
+    a10.Assemble()
+    a10.Finalize()
+    A10 = a10.SpMat()
+    A01 = mfem.Transpose(A10)
+
+    # 11. Iterate.
+    total_iterations = 0
+    increment_u = 0.1
+    u_tmp = mfem.GridFunction(L2fes)
+
+    for k in range(max_it):
+        u_tmp.Assign(u_old_gf)
+
+        print("\nOUTER ITERATION " + str(k+1))
+
+        for j in range(5):
+            total_iterations += 1
+
+            b0.Assemble()
+            b1.Assemble()
+
+            a00.Assemble(0)
+            a00.Finalize(0)
+            A00 = a00.SpMat()
+
+            # Construct Schur-complement preconditioner
+            A00_diag = mfem.Vector(a00.Height())
+            A00.GetDiag(A00_diag)
+            A00_diag.Reciprocal()
+            S = mfem.Mult_AtDA(A01, A00_diag)
+
+            # Python note:
+            #    owns_blocks should be 0 in Python
+            #    because wrapper class will delete blocks
+            prec = mfem.BlockDiagonalPreconditioner(offsets)
+            prec.owns_blocks = 0
+
+            prec.SetDiagonalBlock(0, mfem.DSmoother(A00))
+
+            if not use_umfpack:
+                prec.SetDiagonalBlock(1, mfem.GSSmoother(S))
+            else:
+                prec.SetDiagonalBlock(1, mfem.UMFPackSolver(S))
+
+            A = mfem.BlockOperator(offsets)
+            A.SetBlock(0, 0, A00)
+            A.SetBlock(1, 0, A10)
+            A.SetBlock(0, 1, A01)
+
+            mfem.MINRES(A, prec, rhs, x, 0, 2000, 1e-12)
+
+            del S
+
+            u_tmp -= u_gf  # u_tmp = u_tmp - u_gf
+
+            Newton_update_size = u_tmp.ComputeL2Error(zero_cf)
+            u_tmp.Assign(u_gf)
+
+            # Damped Newton update
+            psi_gf.Add(newton_scaling, delta_psi_gf)
+            a00.Update()
+
+            if visualization:
+                sol_sock << "solution\n" << mesh << u_gf
+                sol_sock << "window_title 'Discrete solution'"
+                sol_sock.flush()
+
+            print("Newton_update_size = " + "{:g}".format(Newton_update_size))
+
+            if Newton_update_size < increment_u:
+                break
+
+        u_tmp.Assign(u_gf)
+        u_tmp -= u_old_gf
+        increment_u = u_tmp.ComputeL2Error(zero_cf)
+
+        print("Number of Newton iterations = " + str(j+1))
+        print("Increment (|| uₕ - uₕ_prvs||) = " + "{:g}".format(increment_u))
+
+        u_old_gf.Assign(u_gf)
+        psi_old_gf.Assign(psi_gf)
+
+        if increment_u < tol or k == max_it-1:
+            break
+
+        alpha *= max(growth_rate, 1.0)
+        neg_alpha_cf.constant = -alpha
+
+    print("\n Outer iterations: " + str(k+1) +
+          "\n Total iterations: " + str(total_iterations) +
+          "\n Total dofs:       " + str(RTfes.GetTrueVSize() + L2fes.GetTrueVSize()))
+
+
+if __name__ == "__main__":
+    from mfem.common.arg_parser import ArgParser
+
+    parser = ArgParser(description='Ex40 (Eikonal queation)')
+    parser.add_argument('-m', '--mesh',
+                        default='star.mesh',
+                        action='store', type=str,
+                        help='Mesh file to use.')
+    parser.add_argument('-o', '--order',
+                        action='store', default=1, type=int,
+                        help="Finite element order (polynomial degree).")
+    parser.add_argument('-r', '--refs',
+                        action='store', default=3, type=int,
+                        help="Number of h-refinements.")
+    parser.add_argument('-mi', '--max-it',
+                        action='store', default=5, type=int,
+                        help="Maximum number of iterations")
+    parser.add_argument("-tol", "--tol",
+                        action='store', default=1e-4, type=float,
+                        help="Stopping criteria based on the difference between.")
+    parser.add_argument('-step', '--step',
+                        action='store', default=1.0, type=float,
+                        help="Initial size alpha")
+    parser.add_argument("-gr", "--growth-rate",
+                        action='store', default=1.0, type=float,
+                        help="Growth rate of the step size alpha")
+    parser.add_argument('-no-vis', '--no-visualization',
+                        action='store_true',
+                        default=False,
+                        help='Disable or disable GLVis visualization')
+
+    args = parser.parse_args()
+    parser.print_options(args)
+
+    meshfile = expanduser(
+        join(os.path.dirname(__file__), '..', 'data', args.mesh))
+    visualization = not args.no_visualization
+    in_alpha = args.step
+
+    run(meshfile=meshfile,
+        order=args.order,
+        max_it=args.max_it,
+        ref_levels=args.refs,
+        alpha=in_alpha,
+        growth_rate=args.growth_rate,
+        newton_scaling=0.8,
+        eps=1e-6,
+        tol=args.tol,
+        visualization=visualization)