Goal¶
Extend the elliptic distributed-control codes to a boundary-control problem where the control is the Dirichlet value imposed on part or all of the boundary.
This is the first project where the control does not act as a volume source. The main challenge is the correct treatment of traces, liftings, and boundary degrees of freedom.
Mathematical Formulation¶
Let be a bounded Lipschitz domain and let . Given , , and a reference control , solve
subject to
A simpler numerical variant replaces the norm by a discrete boundary mass norm:
This is easier to implement but should be discussed as a variational crime.
Using a lifting with , write
and solve
The adjoint satisfies
The boundary gradient can be derived through the lifting or through a transposition argument. In the formal smooth case,
Implementation Hints¶
Start from
codes/dealii/execs/laplacian.ccorcodes/dealii/execs/unconstrained_poisson_optimization.cc.Use boundary DoFs as control DoFs.
First implement a simple boundary mass regularization.
Compare reduced-gradient updates with a monolithic formulation.
Relevant Examples¶
Course material:
jupyterbook/lectures/lecture14.mdon boundary controls.Course code:
codes/dealii/execs/laplacian.cc,codes/dealii/execs/unconstrained_poisson_optimization.cc.deal.II:
step-4andstep-6for Poisson and boundary conditions;step-7for boundary integrals.
Deliverables¶
A short derivation of the reduced gradient.
A working 2D implementation.
Numerical tests with manufactured targets.
A discussion of the difference between and discrete regularization.