Return Vector for evaluate_adj_component and evaluate_hessian_component

Pyadjoint expects Vectors to be returned for evaluate_adj_component and
evaluate_hessian_component, not Functions as present.

The tests have been changed to only pass when this is the case.

Author: ReubenHill

firedrake/adjoint/blocks.py

@@ -402,7 +402,7 @@ def evaluate_adj_component(self, inputs, adj_inputs, block_variable, idx, prepar
         if len(adj_inputs) > 1:
             raise(NotImplementedError("Interpolate block must have a single output"))
         dJdm = self.backend.derivative(prepared, inputs[idx])
-        return self.backend.Interpolator(dJdm, self.V).interpolate(adj_inputs[0], transpose=True)
+        return self.backend.Interpolator(dJdm, self.V).interpolate(adj_inputs[0], transpose=True).vector()
 
     def prepare_evaluate_tlm(self, inputs, tlm_inputs, relevant_outputs):
         return replace(self.expr, self._replace_map())
@@ -580,7 +580,7 @@ def evaluate_hessian_component(self, inputs, hessian_inputs, adj_inputs,
         # left multiply by dJ/dv (adj_inputs[0]) - i.e. interpolate using the
         # transpose operator
         component += self.backend.Interpolator(d2exprdudu, self.V).interpolate(adj_inputs[0], transpose=True)
-        return component
+        return component.vector()
 
     def prepare_recompute_component(self, inputs, relevant_outputs):
         return replace(self.expr, self._replace_map())

tests/regression/test_adjoint_interpolate.py

@@ -291,7 +291,9 @@ def test_interpolate_hessian_linear_expr():
     g = f.copy(deepcopy=True)
 
     dJdm = J.block_variable.tlm_value
-    Hm = f.original_block_variable.hessian_value.vector().inner(h.vector())
+    assert isinstance(f.original_block_variable.adj_value, Vector)
+    assert isinstance(f.original_block_variable.hessian_value, Vector)
+    Hm = f.original_block_variable.hessian_value.inner(h.vector())
     # If the new interpolate block has the right hessian, taylor test
     # convergence rate should be as for the unmodified test.
     assert taylor_test(Jhat, g, h, dJdm=dJdm, Hm=Hm) > 2.9
@@ -347,7 +349,9 @@ def test_interpolate_hessian_nonlinear_expr():
     g = f.copy(deepcopy=True)
 
     dJdm = J.block_variable.tlm_value
-    Hm = f.original_block_variable.hessian_value.vector().inner(h.vector())
+    assert isinstance(f.original_block_variable.adj_value, Vector)
+    assert isinstance(f.original_block_variable.hessian_value, Vector)
+    Hm = f.original_block_variable.hessian_value.inner(h.vector())
     # If the new interpolate block has the right hessian, taylor test
     # convergence rate should be as for the unmodified test.
     assert taylor_test(Jhat, g, h, dJdm=dJdm, Hm=Hm) > 2.9
@@ -407,7 +411,9 @@ def test_interpolate_hessian_nonlinear_expr_multi():
     g = f.copy(deepcopy=True)
 
     dJdm = J.block_variable.tlm_value
-    Hm = f.original_block_variable.hessian_value.vector().inner(h.vector())
+    assert isinstance(f.original_block_variable.adj_value, Vector)
+    assert isinstance(f.original_block_variable.hessian_value, Vector)
+    Hm = f.original_block_variable.hessian_value.inner(h.vector())
     # If the new interpolate block has the right hessian, taylor test
     # convergence rate should be as for the unmodified test.
     assert taylor_test(Jhat, g, h, dJdm=dJdm, Hm=Hm) > 2.9