Several discretization methods apply the set operations lazily and return these lazy sets. But often one wants to apply them concretely because the iteration becomes expensive otherwise.
I checked two cases, but there may be more.
Forward has an option for Ω0 but not for V (called Ud here), which I believe is a bug:
|
Ud = δ * U ⊕ Eψ0 |
|
In = IdentityMultiple(one(eltype(A)), size(A, 1)) |
|
|
|
Ω0 = ConvexHull(X0, Φ * X0 ⊕ Ud ⊕ E⁺) |
|
Ω0 = _apply_setops(Ω0, alg.setops) |
FirstOrder does not offer an option at all, which should be changed:
|
Ω0 = ConvexHull(X0, Xδ + δ * U + BallInf(zeros(n), α)) |
|
|
|
# discretize inputs |
|
β = factor * norm_U_over_A |
|
V = δ * U + BallInf(zeros(n), β) |
Several discretization methods apply the set operations lazily and return these lazy sets. But often one wants to apply them concretely because the iteration becomes expensive otherwise.
I checked two cases, but there may be more.
Forwardhas an option forΩ0but not forV(calledUdhere), which I believe is a bug:ReachabilityAnalysis.jl/src/Discretization/ForwardModule.jl
Lines 144 to 148 in 4d1ef9c
FirstOrderdoes not offer an option at all, which should be changed:ReachabilityAnalysis.jl/src/Discretization/FirstOrderModule.jl
Lines 98 to 102 in 4d1ef9c