Transport 1D PDE
This documentation is for the 1D Transport PDE Environment defined by the boundary control problem
where \(\beta(x)\) is the nonlinear recirculation plant coefficient and control can be applied as either
This problem is the simplest possible Hyperbolic PDE benchmark and thus is a good indicator of potentially algorithmic success on more difficult problems. The goal of the problem is to stabilize the system (\(\lim_{t \to \infty} \|u(x, t)\| = 0\)) where the norm can be varied depending on the problem formulation and continuity of the plant coefficient \(\beta(x)\). A variety of sensing options are supported for the implementation as well if one wants to attempt to perform output feedback stabilization. See the parameter list below.
Numerical Implementation
We derive the numerical implementation scheme for those looking for inner details of the environment. We use a first-order finite-difference scheme to approximate the PDE leading to super fast implementation speeds - although with sacrifice on spatial and temporal timestep parameterization.
Consider the first-order taylor approximation as
with finite spatial derviatves approximated by first-order differences where \(u_{j}^{n}\) is shorthand for \(u(x_{j}, t_{n})\)
where \(\Delta t=dt=\text{time step}\), \(\Delta x=dx=\text{spatial step}\), \(n=0, ..., Nt\), \(j=0, ..., Nx\), where \(Nt\) and \(Nx\) are the total number of discretized temporal and spatial steps respectively. Then substituting \(u_x\) and \(u_t\) into the taylor approximation yields
Now, the last thing to consider the is boundary conditions. Begin with the harder Neumann boundary condition. Let \(u_\zeta^n|_{\zeta=Nx}\) represent the spatial derivative at time \(t\) of spatial point \(Nx=X\) which is given by the user as control input. Then, we have
which is rearranged for the final boundary point
In the case of Dirchilet boundary conditions, the computation is straightforward as \(u_{Nx}^n\) is directly set as the given control input.