Warning
You are reading a version of the website built against the unstable main branch. This content is liable to change without notice and may be inappropriate for your use case.
You can find the documentation for the current stable release here.
firedrake.adjoint_utils.blocks package¶
Submodules¶
firedrake.adjoint_utils.blocks.assembly module¶
- class firedrake.adjoint_utils.blocks.assembly.AssembleBlock(form, ad_block_tag=None)[source]¶
Bases:
Block- adj_state¶
- block_helper¶
- compute_action_adjoint(adj_input, arity_form, form=None, c_rep=None, space=None, dform=None)[source]¶
This computes the action of the adjoint of the derivative of \(form\) wrt \(c_rep\) on \(adj_input\). In other words, it returns: \(<(dform/dc_rep)*, adj_input>\)
If \(form\) has arity 0 => \(dform/dc_rep\) is a 1-form and \(adj_input\) a float, we can simply use the \(*\) operator.
If \(form\) has arity 1 => \(dform/dc_rep\) is a 2-form and we can symbolically take its adjoint and then apply the action on \(adj_input\), to finally assemble the result.
- evaluate_hessian_component(inputs, hessian_inputs, adj_inputs, block_variable, idx, relevant_dependencies, prepared=None)[source]¶
- tag¶
firedrake.adjoint_utils.blocks.block_utils module¶
firedrake.adjoint_utils.blocks.constant module¶
- class firedrake.adjoint_utils.blocks.constant.ConstantAssignBlock(other, ad_block_tag=None)[source]¶
Bases:
Block- adj_state¶
- block_helper¶
- evaluate_hessian_component(inputs, hessian_inputs, adj_inputs, block_variable, idx, relevant_dependencies, prepared=None)[source]¶
- tag¶
- firedrake.adjoint_utils.blocks.constant.constant_from_values(constant, values=None)[source]¶
Returns a new Constant with \(constant.values()\).
If the optional argument \(values\) is provided, then \(values\) will be the values of the new Constant instead, while still preserving the ufl_shape of \(constant\).
- Parameters:
constant – A constant with the ufl_shape to preserve. values
(numpy.array) – An optional argument to use instead of
constant.values().
- Returns:
The created Constant of the same type as \(constant\).
- Return type:
firedrake.adjoint_utils.blocks.dirichlet_bc module¶
firedrake.adjoint_utils.blocks.function module¶
- class firedrake.adjoint_utils.blocks.function.CofunctionAssignBlock(lhs, rhs, ad_block_tag=None, rhs_from_assemble=False)[source]¶
Bases:
BlockClass specifically for the case b.assign(a).
All other cofunction assignment operations are annotated via Assemble. In effect this means that this is the annotation of an identity operation.
- Parameters:
lhs – The target of the assignment.
rhs – The cofunction being assigned.
- adj_state¶
- block_helper¶
- evaluate_hessian_component(inputs, hessian_inputs, adj_inputs, block_variable, idx, relevant_dependencies, prepared=None)[source]¶
- recompute_component(inputs, block_variable, idx, prepared=None)[source]¶
Recompute the assignment.
- Parameters:
inputs (list of firedrake.function.Function or Constant) – The variable in the RHS of the assignment.
block_variable (pyadjoint.block_variable.BlockVariable) – The output block variable.
idx (int) – Index associated to the inputs list.
prepared – The precomputed RHS value.
Notes
Recomputes the block_variable only if the checkpoint was not delegated to another
Function.- Returns:
Return either the firedrake cofunction or \(BlockVariable\) checkpoint to which was delegated the checkpointing.
- Return type:
- tag¶
- class firedrake.adjoint_utils.blocks.function.FunctionAssignBlock(func, other, ad_block_tag=None)[source]¶
Bases:
Block- adj_state¶
- block_helper¶
- evaluate_hessian_component(inputs, hessian_inputs, adj_inputs, block_variable, idx, relevant_dependencies, prepared=None)[source]¶
- recompute_component(inputs, block_variable, idx, prepared=None)[source]¶
Recompute the assignment.
- Parameters:
inputs (list of firedrake.function.Function or Constant) – The variables in the RHS of the assignment.
block_variable (pyadjoint.block_variable.BlockVariable) – The output block variable.
idx (int) – Index associated to the inputs list.
prepared – The precomputed RHS value.
Notes
Recomputes the block_variable only if the checkpoint was not delegated to another
firedrake.function.Function.- Returns:
Return either the firedrake function or \(BlockVariable\) checkpoint to which was delegated the checkpointing.
- Return type:
- tag¶
- class firedrake.adjoint_utils.blocks.function.FunctionMergeBlock(func, idx, ad_block_tag=None)[source]¶
Bases:
Block- adj_state¶
- block_helper¶
- evaluate_hessian_component(inputs, hessian_inputs, adj_inputs, block_variable, idx, relevant_dependencies, prepared=None)[source]¶
- tag¶
firedrake.adjoint_utils.blocks.mesh module¶
- class firedrake.adjoint_utils.blocks.mesh.MeshInputBlock(mesh, ad_block_tag=None)[source]¶
Bases:
BlockBlock which links a MeshGeometry to its coordinates, which is a firedrake function.
- adj_state¶
- block_helper¶
- evaluate_hessian_component(inputs, hessian_inputs, adj_inputs, idx, block_variable, relevant_dependencies, prepared=None)[source]¶
- tag¶
- class firedrake.adjoint_utils.blocks.mesh.MeshOutputBlock(func, mesh, ad_block_tag=None)[source]¶
Bases:
BlockBlock which is called when the coordinates of a mesh are changed.
- adj_state¶
- block_helper¶
- evaluate_hessian_component(inputs, hessian_inputs, adj_inputs, idx, block_variable, relevant_dependencies, prepared=None)[source]¶
- tag¶
firedrake.adjoint_utils.blocks.solving module¶
- class firedrake.adjoint_utils.blocks.solving.GenericSolveBlock(lhs, rhs, func, bcs, *args, **kwargs)[source]¶
Bases:
Block- property adj_sol¶
- adj_state¶
- block_helper¶
- evaluate_hessian_component(inputs, hessian_inputs, adj_inputs, block_variable, idx, relevant_dependencies, prepared=None)[source]¶
- pop_kwargs_keys = ['adj_cb', 'adj_bdy_cb', 'adj2_cb', 'adj2_bdy_cb', 'forward_args', 'forward_kwargs', 'adj_args', 'adj_kwargs']¶
- tag¶
- class firedrake.adjoint_utils.blocks.solving.NonlinearVariationalSolveBlock(equation, func, bcs, adj_cache, problem_J, solver_kwargs, **kwargs)[source]¶
Bases:
GenericSolveBlock- adj_state¶
- block_helper¶
- tag¶
- class firedrake.adjoint_utils.blocks.solving.ProjectBlock(v, V, output, bcs=[], *args, **kwargs)[source]¶
Bases:
SolveVarFormBlock- adj_state¶
- block_helper¶
- tag¶
- class firedrake.adjoint_utils.blocks.solving.SolveLinearSystemBlock(A, u, b, *args, **kwargs)[source]¶
Bases:
GenericSolveBlock- adj_state¶
- block_helper¶
- tag¶
- class firedrake.adjoint_utils.blocks.solving.SolveVarFormBlock(equation, func, bcs=[], *args, **kwargs)[source]¶
Bases:
GenericSolveBlock- adj_state¶
- block_helper¶
- tag¶
- class firedrake.adjoint_utils.blocks.solving.Solver(*values)[source]¶
Bases:
EnumEnum for solver types.
- ADJOINT = 1¶
- FORWARD = 0¶
- class firedrake.adjoint_utils.blocks.solving.SupermeshProjectBlock(source, target_space, target, bcs=[], **kwargs)[source]¶
Bases:
BlockAnnotates supermesh projection.
Suppose we have a source space, \(V_A\), and a target space, \(V_B\). Projecting a source from \(V_A\) to \(V_B\) amounts to solving the linear system
\[M_B * v_B = M_{AB} * v_A,\]- where
\(M_B\) is the mass matrix on \(V_B\),
\(M_{AB}\) is the mixed mass matrix for \(V_A\) and \(V_B\),
\(v_A\) and \(v_B\) are vector representations of the source and target
firedrake.function.Functions.
- This can be broken into two steps:
Step 1. form RHS, multiplying the source with the mixed mass matrix;
Step 2. solve linear system.
- adj_state¶
- block_helper¶
- evaluate_adj_component(inputs, adj_inputs, block_variable, idx, prepared=None)[source]¶
Evaluate the adjoint to one output of the block
- Recall that the forward propagation can be broken down as:
Step 1. multiply \(w := M_{AB} * v_A\);
Step 2. solve \(M_B * v_B = w\).
For a seed vector \(v_B^{seed}\) from the target space, the adjoint is given by:
Adjoint of step 2. solve \(M_B^T * w = v_B^{seed}\) for \(w\);
Adjoint of step 1. multiply \(v_A^{adj} := M_{AB}^T * w\).
- evaluate_hessian_component(inputs, hessian_inputs, adj_inputs, block_variable, idx, relevant_dependencies, prepared=None)[source]¶
- evaluate_tlm_component(inputs, tlm_inputs, block_variable, idx, prepared=None)[source]¶
Given that the input is a \(Function\), we just have a linear operation. As such, the tlm is just the sum of each tlm input projected into the target space.
- tag¶
