Merge pull request #77 from firedrakeproject/pbrubeck/hct/hierarchical

HCT: ensure hierarchical edge basis functions

FIAT/hct.py

@@ -53,8 +53,8 @@ def __init__(self, ref_complex, degree, reduced=False):
                 entity_ids[1][e].extend(range(cur, len(nodes)))
         else:
             x = 2.0*qpts - 1
-            phis = eval_jacobi_batch(2, 2, k, x)
-            dphis = eval_jacobi_deriv_batch(2, 2, k, x)
+            phis = eval_jacobi_batch(1, 1, k, x)
+            dphis = eval_jacobi_deriv_batch(1, 1, k, x)
             for e in sorted(top[1]):
                 Q_mapped = FacetQuadratureRule(ref_el, 1, e, Q)
                 scale = 2 / Q_mapped.jacobian_determinant()

test/unit/test_argyris.py

@@ -1,7 +1,7 @@
 import pytest
 import numpy
 
-from FIAT import Argyris
+from FIAT import Argyris, HsiehCloughTocher
 from FIAT.jacobi import eval_jacobi_batch, eval_jacobi_deriv_batch
 from FIAT.quadrature import FacetQuadratureRule
 from FIAT.quadrature_schemes import create_quadrature
@@ -13,60 +13,78 @@ def cell():
     return ufc_simplex(2)
 
 
-@pytest.mark.parametrize("degree", range(6, 10))
-def test_argyris_basis_functions(cell, degree):
-    fe = Argyris(cell, degree)
+def directional_derivative(direction, tab):
+    return sum(direction[alpha.index(1)] * tab[alpha]
+               for alpha in tab if sum(alpha) == 1)
 
+
+def inner(u, v, wts):
+    return numpy.dot(numpy.multiply(u, wts), v.T)
+
+
+@pytest.mark.parametrize("family, degree", [
+    *((Argyris, k) for k in range(6, 10)),
+    *((HsiehCloughTocher, k) for k in range(4, 8))])
+def test_argyris_basis_functions(cell, family, degree):
+    fe = family(cell, degree)
+
+    ref_el = fe.get_reference_element()
+    sd = ref_el.get_spatial_dimension()
+    top = ref_el.get_topology()
     entity_ids = fe.entity_dofs()
-    sd = cell.get_spatial_dimension()
-    top = cell.get_topology()
+
+    degree = fe.degree()
+    lowest_p = 5 if isinstance(fe, Argyris) else 3
+    a = (lowest_p - 1) // 2
+    q = degree - lowest_p
+
     rline = ufc_simplex(1)
     Qref = create_quadrature(rline, 2*degree)
+    xref = 2.0 * Qref.get_points() - 1
 
-    q = degree - 5
-    xref = 2.0*Qref.get_points()-1
     ell_at_qpts = eval_jacobi_batch(0, 0, degree, xref)
-    P_at_qpts = eval_jacobi_batch(2, 2, degree, xref)
-    DP_at_qpts = eval_jacobi_deriv_batch(2, 2, degree, xref)
+    P_at_qpts = eval_jacobi_batch(a, a, degree, xref)
+    DP_at_qpts = eval_jacobi_deriv_batch(a, a, degree, xref)
 
     for e in top[1]:
-        n = cell.compute_normal(e)
-        t = cell.compute_edge_tangent(e)
-        Q = FacetQuadratureRule(cell, 1, e, Qref)
+        n = ref_el.compute_normal(e)
+        t = ref_el.compute_edge_tangent(e)
+        Q = FacetQuadratureRule(ref_el, 1, e, Qref)
         qpts, qwts = Q.get_points(), Q.get_weights()
 
         ids = entity_ids[1][e]
         ids1 = ids[1:-q]
         ids0 = ids[-q:]
 
         tab = fe.tabulate(1, qpts)
-        # Test that normal derivative moment bfs have vanishing trace
-        assert numpy.allclose(tab[(0,) * sd][ids[:-q]], 0)
+        phi = tab[(0,) * sd]
+        phi_n = directional_derivative(n, tab)
+        phi_t = directional_derivative(t, tab)
 
-        phi = tab[(0,) * sd][ids0]
-
-        phi_n = sum(n[alpha.index(1)] * tab[alpha][ids1]
-                    for alpha in tab if sum(alpha) == 1)
+        # Test that normal derivative moment bfs have vanishing trace
+        assert numpy.allclose(phi[ids[:-q]], 0)
 
-        phi_t = sum(t[alpha.index(1)] * tab[alpha][ids0]
-                    for alpha in tab if sum(alpha) == 1)
+        # Test that trace moment bfs have vanishing normal derivative
+        assert numpy.allclose(phi_n[ids0], 0)
 
         # Test that facet bfs are hierarchical
-        coeffs = numpy.dot(numpy.multiply(phi, qwts), ell_at_qpts[6:].T)
-        assert numpy.allclose(numpy.triu(coeffs, k=1), 0)
+        coeffs = inner(ell_at_qpts[1+lowest_p:], phi[ids0], qwts)
+        assert numpy.allclose(coeffs, numpy.triu(coeffs))
+        coeffs = inner(ell_at_qpts[1+lowest_p:], phi_n[ids1], qwts)
+        assert numpy.allclose(coeffs, numpy.triu(coeffs))
 
-        # Test duality of normal derivarive moments
-        coeffs = numpy.dot(numpy.multiply(phi_n, qwts), P_at_qpts[1:].T)
-        assert numpy.allclose(coeffs[:, q:], 0)
-        assert numpy.allclose(coeffs[:, :q], numpy.diag(numpy.diag(coeffs[:, :q])))
+        # Test duality of normal derivative moments
+        coeffs = inner(P_at_qpts[1:], phi_n[ids1], qwts)
+        assert numpy.allclose(coeffs[q:], 0)
+        assert numpy.allclose(coeffs[:q], numpy.diag(numpy.diag(coeffs[:q])))
 
         # Test duality of trace moments
-        coeffs = numpy.dot(numpy.multiply(phi, qwts), DP_at_qpts[1:].T)
-        assert numpy.allclose(coeffs[:, q:], 0)
-        assert numpy.allclose(coeffs[:, :q], numpy.diag(numpy.diag(coeffs[:, :q])))
+        coeffs = inner(DP_at_qpts[1:], phi[ids0], qwts)
+        assert numpy.allclose(coeffs[q:], 0)
+        assert numpy.allclose(coeffs[:q], numpy.diag(numpy.diag(coeffs[:q])))
 
         # Test the integration by parts property arising from the choice
         # of normal derivative and trace moments DOFs.
         # The normal derivative of the normal derviative bfs must be equal
         # to minus the tangential derivative of the trace moment bfs
-        assert numpy.allclose(numpy.dot((phi_n + phi_t)**2, qwts), 0)
+        assert numpy.allclose(numpy.dot((phi_n[ids1] + phi_t[ids0])**2, qwts), 0)