Пример #1
0
 def cube(k):
     cglob = -1j * k
     cloc = cloc0 + cloc1 * k + cloc2 * k**2
     print("k: ", k, "cloc: ", cloc)
     mus = {'k': k, 'c_glob': cglob, 'c_loc': cloc}
     resolution = int(
         np.ceil(float(k * 1.5 + 50) / coarse_grid_resolution) *
         coarse_grid_resolution)
     n = int(k / 5 + 30)
     gq, lq = localize_problem(p,
                               coarse_grid_resolution,
                               resolution,
                               mus=mus)
     d = gq["d"]
     u = d.solve(mus)
     e_r = []
     e_d = []
     for i in range(it):
         print(i, )
         sys.stdout.flush()
         bases = create_bases2(gq, lq, n, transfer='robin')
         ru_r = reconstruct_solution(gq, lq, bases)
         del bases
         dif_r = u - ru_r
         e_r.append(gq["full_norm"](dif_r)[0] / gq["full_norm"](u)[0])
         bases = create_bases2(gq, lq, n, transfer='dirichlet')
         ru_d = reconstruct_solution(gq, lq, bases)
         del bases
         dif_d = u - ru_d
         e_d.append(gq["full_norm"](dif_d)[0] / gq["full_norm"](u)[0])
     return np.mean(e_d), np.mean(e_r)
Пример #2
0
    def cube(n):
        print("n: ", n)
        Q0 = []
        Q1 = []
        Q2 = []
        for j in range(it):
            print(j, )
            sys.stdout.flush()
            basis_robin = create_bases3(gq, lq, n, 0)
            ru_robin = reconstruct_solution(gq, lq, basis_robin)
            del basis_robin
            dif_robin = u - ru_robin
            Q0.append(gq["full_norm"](dif_robin)[0] / gq["full_norm"](u)[0])

            basis_robin = create_bases3(gq, lq, n, 1)
            ru_robin = reconstruct_solution(gq, lq, basis_robin)
            del basis_robin
            dif_robin = u - ru_robin
            Q1.append(gq["full_norm"](dif_robin)[0] / gq["full_norm"](u)[0])

            basis_robin = create_bases3(gq, lq, n, 2)
            ru_robin = reconstruct_solution(gq, lq, basis_robin)
            del basis_robin
            dif_robin = u - ru_robin
            Q2.append(gq["full_norm"](dif_robin)[0] / gq["full_norm"](u)[0])
        return [np.mean(Q0), np.mean(Q1), np.mean(Q2)]
Пример #3
0
def evaluation(it,
               lim,
               k,
               boundary,
               save,
               cglob=0,
               cloc=0,
               plot=False,
               resolution=200,
               coarse_grid_resolution=10):
    p = helmholtz(boundary=boundary)
    mus = {'k': k, 'c_glob': cglob, 'c_loc': cloc}
    gq, lq = localize_problem(p, coarse_grid_resolution, resolution, mus=mus)
    d = gq["d"]
    u = d.solve(mus)
    h1_dirichlet = []
    h1_robin = []
    nrang = np.arange(0, lim, 5)
    for n in nrang:
        print("n: ", n)
        h1d = []
        h1r = []
        for j in range(it):
            print(j, )
            sys.stdout.flush()
            basis_dirichlet = create_bases2(gq, lq, n)
            ru_dirichlet = reconstruct_solution(gq, lq, basis_dirichlet)
            del basis_dirichlet
            basis_robin = create_bases2(gq, lq, n, transfer='robin')
            ru_robin = reconstruct_solution(gq, lq, basis_robin)
            del basis_robin
            dif_dirichlet = u - ru_dirichlet
            dif_robin = u - ru_robin
            h1d.append(gq["full_norm"](dif_dirichlet)[0] /
                       gq["full_norm"](u)[0])
            h1r.append(gq["full_norm"](dif_robin)[0] / gq["full_norm"](u)[0])
        h1_dirichlet.append(h1d)
        h1_robin.append(h1r)
    means_dirichlet_h1 = np.mean(h1_dirichlet, axis=1)
    means_robin_h1 = np.mean(h1_robin, axis=1)
    data = np.vstack([nrang, means_dirichlet_h1, means_robin_h1]).T
    open(save, "w").writelines([" ".join(map(str, v)) + "\n" for v in data])
    if plot:
        from matplotlib import pyplot as plt
        plt.figure()
        plt.semilogy(nrang, means_dirichlet_h1, label="Dirichlet h1")
        plt.semilogy(nrang, means_robin_h1, label="Robin h1")
        plt.legend(loc='upper right')
        plt.xlabel('Basis size')
        plt.show()
Пример #4
0
def knerr2D(it,
            boundary,
            save,
            cglob=None,
            cloc0=0,
            cloc1=1,
            cloc2=1,
            krang=np.arange(0.1, 200.1, 10.),
            nrang=np.arange(0, 100, 5),
            plot=False,
            resolution=100,
            coarse_grid_resolution=10):
    err_r = np.zeros((len(krang), len(nrang)))
    p = helmholtz(boundary=boundary)
    usecglob = (cglob is None)
    xi = 0
    for k in krang:
        yi = 0
        if usecglob:
            cglob = -1j * k
        cloc = cloc0 + cloc1 * k + cloc2 * k**2
        mus = {'k': k, 'c_glob': cglob, 'c_loc': cloc}
        gq, lq = localize_problem(p,
                                  coarse_grid_resolution,
                                  resolution,
                                  mus=mus)
        d = gq["d"]
        u = d.solve(mus)
        for n in nrang:
            print(k, n)
            e_r = []
            for i in range(min(20 - n / 5, it)):
                print(i, )
                sys.stdout.flush()
                bases = create_bases2(gq, lq, n, transfer='robin')
                ru_r = reconstruct_solution(gq, lq, bases)
                del bases
                dif_r = u - ru_r
                e_r.append(gq["full_norm"](dif_r)[0] / gq["full_norm"](u)[0])
            err_r[xi][yi] = np.mean(e_r)
            yi += 1
        xi += 1
    X, Y = np.meshgrid(krang, nrang)
    data = np.vstack([X.T.ravel(), Y.T.ravel(), err_r.ravel()]).T
    open(save, "w").writelines([" ".join(map(str, v)) + "\n" for v in data])
    if plot:
        import matplotlib.pyplot as plt
        from matplotlib import cm
        fig = plt.figure()
        ax = fig.gca(projection='3d')
        surf = ax.plot_surface(X,
                               Y,
                               err_r,
                               cstride=1,
                               rstride=1,
                               cmap=cm.coolwarm,
                               linewidth=0,
                               antialiased=False)
        fig.colorbar(surf, shrink=0.5, aspect=5)
        plt.show()
Пример #5
0
def resolution(it,
               k,
               n,
               boundary,
               save,
               cloc=0,
               returnvals=False,
               coarse_grid_resolution=10):
    cglob = -1j * k
    mus = {'k': k, 'c_glob': cglob, 'c_loc': cloc}
    p = helmholtz(boundary=boundary)
    space = np.arange(20, 160, 10)
    err = []
    for resolution in space:
        print(resolution)
        gq, lq = localize_problem(p,
                                  coarse_grid_resolution,
                                  resolution,
                                  mus=mus)
        d = gq["d"]
        u = d.solve(mus)
        E = []
        for i in range(it):
            bases = create_bases2(gq, lq, n, transfer='robin')
            ru = reconstruct_solution(gq, lq, bases)
            E.append(gq["full_norm"](u - ru)[0] / gq["full_norm"](u)[0])
        err.append(E)
    errs = np.mean(err, axis=1)
    data = np.vstack([space, errs]).T
    open(save, "w").writelines([" ".join(map(str, v)) + "\n" for v in data])
    if returnvals:
        return [space, errs]
Пример #6
0
def test2(transfer='robin',
          boundary='dirichlet',
          acc=1e-2,
          k=6.,
          cloc=6.,
          title='test',
          resolution=100,
          coarse_grid_resolution=10):
    cglob = -1j * k
    mus = {'k': k, 'c_glob': cglob, 'c_loc': cloc}
    p = helmholtz(boundary=boundary)
    gq, lq = localize_problem(p,
                              coarse_grid_resolution,
                              resolution,
                              mus,
                              calQ=True)
    basis = create_bases(gq,
                         lq,
                         20,
                         transfer=transfer,
                         target_accuracy=acc,
                         silent=False)
    ru = reconstruct_solution(gq, lq, basis, silent=False)
    d = gq["d"]
    u = d.solve(mus)
    dif = u - ru
    print(gq["full_norm"](dif)[0] / gq["full_norm"](u)[0])
    d.visualize((dif.real, dif.imag, u.real, u.imag, ru.real, ru.imag),
                legend=('dif.real', 'dif.imag', 'u.real', 'u.imag', 'ru.real',
                        'ru.imag'),
                separate_colorbars=True,
                title=title)
Пример #7
0
 def cube():
     bases = create_bases2(gq, lq, n, transfer='robin')
     ru_r = reconstruct_solution(gq, lq, bases)
     del bases
     dif_r = u - ru_r
     return gq["full_norm"](dif_r)[0] / gq["full_norm"](u)[0]
Пример #8
0
    def cube(k):
        print(k)
        cglob = -1j * k
        cloc = cloc0 + cloc1 * k + cloc2 * k**2
        mus = {'k': k, 'c_glob': cglob, 'c_loc': cloc}
        resolution = int(
            np.ceil(float(k * 1.5 + 50) / coarse_grid_resolution) *
            coarse_grid_resolution)
        gq, lq = localize_problem(p,
                                  coarse_grid_resolution,
                                  resolution,
                                  mus=mus,
                                  calT=True,
                                  calQ=True)
        calculate_continuity_constant(gq, lq)
        calculate_inf_sup_constant2(gq, lq)
        # calculate_lambda_min(gq, lq)
        calculate_csis(gq, lq)
        calculate_Psi_norm(gq, lq)
        d = gq["d"]
        u = d.solve(mus)
        norm = gq["full_norm"]
        localizer = gq["localizer"]
        pou = gq["pou"]
        dats = []
        for j in range(it):
            print(j, )
            sys.stdout.flush()
            bases = create_bases(gq,
                                 lq,
                                 num_testvecs=20,
                                 transfer='robin',
                                 target_accuracy=acc,
                                 calC=False)
            ru = reconstruct_solution(gq, lq, bases)
            ls = norm(u - ru)[0] / norm(u)[0]
            sum1 = u * 0
            sum2 = 0
            sum3 = 0
            sum4 = 0
            sum51 = []
            sum52 = 0
            for space in gq["spaces"]:
                ldict = lq[space]
                B = bases[space]
                range_product = ldict["range_product"]
                source_product = ldict["source_product"]
                T = ldict["robin_transfer"].as_range_array()
                range_space = ldict["range_space"]
                omega_star_space = ldict["omega_star_space"]
                omega_star_product = ldict["omega_star_product"]
                u_s = ldict["local_solution_robin"]
                u_s2 = ldict["local_sol2"]

                u_loc = pou[range_space](localizer.localize_vector_array(
                    u, range_space))
                u_loc2 = localizer.localize_vector_array(u, omega_star_space)
                u_dif = u_loc - u_s
                u_dif2 = u_loc2 - u_s2
                term = u_dif - B.lincomb(B.inner(u_dif, range_product).T)

                T1 = T - B.lincomb(B.inner(T, range_product).T)
                maxval = operator_svd2(T1.data.T, source_product.matrix,
                                       range_product.matrix)[0][0]

                sum1 += localizer.globalize_vector_array(term, range_space)
                sum2 += term.norm(range_product)[0]**2
                sum3 += maxval**2 * ldict["Psi_norm"]**2 * u_dif2.norm(
                    omega_star_product)[0]**2
                sum4 += maxval**2 * ldict["Psi_norm"]**2 * u_loc2.norm(
                    omega_star_product)[0]**2 * ldict["csi"]**2
                sum51.append(maxval * ldict["Psi_norm"] * ldict["csi"])
                sum52 += u_loc2.norm(omega_star_product)[0]**2
            rs1 = gq["continuity_constant"] / gq["inf_sup_constant"] * norm(
                sum1)[0] / norm(u)[0]
            rs2 = gq["continuity_constant"] / gq[
                "inf_sup_constant"] * 2 * np.sqrt(sum2) / norm(u)[0]
            rs3 = gq["continuity_constant"] / gq[
                "inf_sup_constant"] * 2 * np.sqrt(sum3) / norm(u)[0]
            rs4 = gq["continuity_constant"] / gq[
                "inf_sup_constant"] * 2 * np.sqrt(sum4) / norm(u)[0]
            rs5 = gq["continuity_constant"] / gq["inf_sup_constant"] * 2 * max(
                sum51) * np.sqrt(sum52) / norm(u)[0]
            rs6 = gq["continuity_constant"] / gq["inf_sup_constant"] * max(
                sum51) * 8
            ccs = gq["continuity_constant"] / gq["inf_sup_constant"]
            dats.append([ls, rs1, rs2, rs3, rs4, rs5, rs6, ccs])
        return dats