Vertex representation propagator (VREP)


This algorithm solves the recurrence relation

\[X_{k+1} = Φ X_k \oplus V_k\]

using the vertex representation of the sets involved. If the system is homogeneous, i.e. $V_k = ∅$ for all $k ≥ 0$, then the number of vertices of the sequence doesn't increase. On the other hand, the number of vertices of $X_k$ increases hence the method requires doing some conservative reduction strategy.

The method is aimed towards small dimensional systems – typically $n < 15$ – which show a substantial benefit by using heap-allocated statically sized arrays through the StaticArrays.jl package. To use static arrays, pass the option static=true. Optionally, a dimension field can be passed to the VREP algorithm constructor too using the dim keyword argument. (It is sometimes useful that Julia is able to infer the set representation, or type, of the output flowpipe based only on the algorithm choice, but since the length of a vector is stored as type information for the case of static arrays, it has to be given in advance.)

Specifying the backend

If the dimension of the system is two, VREP uses efficient algorithms for convex polygons (e.g. convex hull, Minkowski sum) implemented in LazySets.jl. On the other hand, for systems of dimension higher than two, concrete polyhedral computations use the Polyhedra.jl library which itself relies on specific backends, which can be specified with the backend keyword argument in the VREP algorithm constructor. Such backend is used in the discretization phase. Actually, Polyhedra.jl features a default solver (hence it doesn't require additional packages apart from Polyhedra.jl itself), although for efficiency reasons we suggest to use CDDLib.jl. For example, to use CDDLib in a 4-dimensional system, do

using ReachabilityAnalysis, Polyhedra, CDDLib

# ... define model ...
# prob = @ivp(...)
alg = VREP(δ=1e-3, static=true, dim=4, backend=CDDLib.Library())
solve(prob, tspan=(0.0, 1.0), alg=alg)

Other backends are available e.g. QHull.jl.