Set Interfaces

This section of the manual describes the interfaces for different set types. Every set that fits the description of an interface should also implement it. This helps in several ways:

  • avoid code duplicates,
  • provide functions for many sets at once,
  • allow changes in the source code without changing the API.

The interface functions are outlined in the interface documentation. For implementations of the interfaces see the corresponding sub-pages linked in the respective sections.

Note

The naming convention is such that all interface names (with the exception of the main abstract type ConvexSet) should be preceded by Abstract.

The following diagram shows the interface hierarchy.

../assets/interfaces.png

General set (LazySet)

Every set in this library is a subtype of the abstract type LazySet.

LazySets.LazySetType
LazySet{N}

Abstract type for the set types in LazySets.

Notes

LazySet types should be parameterized with a type N, typically N<:Real, for using different numeric types.

Every concrete LazySet must define the following functions:

  • dim(S::LazySet) – the ambient dimension of S

The subtypes of LazySet (including abstract interfaces):

julia> subtypes(LazySet, false)
8-element Vector{Any}:
 AbstractPolynomialZonotope
 Complement
 ConvexSet
 LazySets.AbstractStar
 QuadraticMap
 Rectification
 UnionSet
 UnionSetArray

If we only consider concrete subtypes, then:

julia> concrete_subtypes = subtypes(LazySet, true);

julia> length(concrete_subtypes)
53

julia> println.(concrete_subtypes);
AffineMap
Ball1
Ball2
BallInf
Ballp
Bloating
CachedMinkowskiSumArray
CartesianProduct
CartesianProductArray
Complement
ConvexHull
ConvexHullArray
DensePolynomialZonotope
Ellipsoid
EmptySet
ExponentialMap
ExponentialProjectionMap
HParallelotope
HPolygon
HPolygonOpt
HPolyhedron
HPolytope
HalfSpace
Hyperplane
Hyperrectangle
Intersection
IntersectionArray
Interval
InverseLinearMap
LazySets.AbstractStar
Line
Line2D
LineSegment
LinearMap
MinkowskiSum
MinkowskiSumArray
QuadraticMap
Rectification
ResetMap
RotatedHyperrectangle
SimpleSparsePolynomialZonotope
Singleton
SparsePolynomialZonotope
Star
SymmetricIntervalHull
Translation
UnionSet
UnionSetArray
Universe
VPolygon
VPolytope
ZeroSet
Zonotope
source

General sets (ConvexSet)

Every convex set in this library implements this interface.

LazySets.ConvexSetType
ConvexSet{N} <: LazySet{N}

Abstract type for convex sets, i.e., sets characterized by a (possibly infinite) intersection of halfspaces, or equivalently, sets $S$ such that for any two elements $x, y ∈ S$ and $0 ≤ λ ≤ 1$ it holds that $λ·x + (1-λ)·y ∈ S$.

Notes

ConvexSet types should be parameterized with a type N, typically N<:Real, for using different numeric types.

Every concrete ConvexSet must define the following function:

  • σ(d::AbstractVector, S::ConvexSet) – the support vector of S in a given direction d

The function

  • ρ(d::AbstractVector, S::ConvexSet) – the support function of S in a given direction d

is optional because there is a fallback implementation relying on σ. However, for unbounded sets (which includes most lazy set types) this fallback cannot be used and an explicit method must be implemented.

source

Support function and support vector

Every ConvexSet type must define a function σ to compute the support vector.

LazySets.ρMethod
ρ(d::AbstractVector, S::ConvexSet)

Evaluate the support function of a set in a given direction.

Input

  • d – direction
  • S – convex set

Output

The support function of the set S for the direction d.

source
LazySets.singleton_listMethod
singleton_list(P::ConvexSet)

Return the vertices of a polytopic set as a list of singletons.

Input

  • P – polytopic set

Output

The list of vertices of P, as Singleton.

Notes

This function relies on vertices_list, which raises an error if the set is not polytopic (e.g., unbounded).

source
LazySets.constraintsMethod
constraints(X::ConvexSet)

Construct an iterator over the constraints of a polyhedral set.

Input

  • X – polyhedral set

Output

An iterator over the constraints of X.

source
LazySets.verticesMethod
vertices(X::ConvexSet)

Construct an iterator over the vertices of a polyhedral set.

Input

  • X – polyhedral set

Output

An iterator over the vertices of X.

source
MiniQhull.delaunayFunction
delaunay(X::ConvexSet)

Compute the Delaunay triangulation of the given convex set.

Input

  • X – set

Output

A union of polytopes in vertex representation.

Notes

This function requires that you have properly installed the package MiniQhull.jl, including the library Qhull.

The method works in arbitrary dimension and the requirement is that the list of vertices of X can be obtained.

source

Other globally defined set functions

LazySets.basetypeFunction
basetype(T::Type{<:ConvexSet})

Return the base type of the given set type (i.e., without type parameters).

Input

  • T – set type, used for dispatch

Output

The base type of T.

source
basetype(S::ConvexSet)

Return the base type of the given set (i.e., without type parameters).

Input

  • S – set instance, used for dispatch

Output

The base type of S.

Examples

julia> z = rand(Zonotope);

julia> basetype(z)
Zonotope

julia> basetype(z + z)
MinkowskiSum

julia> basetype(LinearMap(rand(2, 2), z + z))
LinearMap
source
LinearAlgebra.normFunction
norm(S::ConvexSet, [p]::Real=Inf)

Return the norm of a convex set. It is the norm of the enclosing ball (of the given $p$-norm) of minimal volume that is centered in the origin.

Input

  • S – convex set
  • p – (optional, default: Inf) norm

Output

A real number representing the norm.

source
IntervalArithmetic.radiusFunction
radius(S::ConvexSet, [p]::Real=Inf)

Return the radius of a convex set. It is the radius of the enclosing ball (of the given $p$-norm) of minimal volume with the same center.

Input

  • S – convex set
  • p – (optional, default: Inf) norm

Output

A real number representing the radius.

source
LazySets.diameterFunction
diameter(S::ConvexSet, [p]::Real=Inf)

Return the diameter of a convex set. It is the maximum distance between any two elements of the set, or, equivalently, the diameter of the enclosing ball (of the given $p$-norm) of minimal volume with the same center.

Input

  • S – convex set
  • p – (optional, default: Inf) norm

Output

A real number representing the diameter.

source
LazySets.isboundedtypeMethod
isboundedtype(::Type{<:ConvexSet})

Determine whether a set type only represents bounded sets.

Input

  • ConvexSet – set type for dispatch

Output

true if the set type only represents bounded sets. Note that some sets may still represent an unbounded set even though their type actually does not (example: HPolytope, because the construction with non-bounding linear constraints is allowed).

Notes

By default this function returns false. All set types that can determine boundedness should override this behavior.

source
LazySets.isboundedMethod
isbounded(S::ConvexSet)

Determine whether a set is bounded.

Input

  • S – set
  • algorithm – (optional, default: "support_function") algorithm choice, possible options are "support_function" and "stiemke"

Output

true iff the set is bounded.

Algorithm

See the documentation of _isbounded_unit_dimensions or _isbounded_stiemke for details.

source
LazySets._isbounded_unit_dimensionsMethod
_isbounded_unit_dimensions(S::ConvexSet{N}) where {N}

Determine whether a set is bounded in each unit dimension.

Input

  • S – set

Output

true iff the set is bounded in each unit dimension.

Algorithm

This function performs $2n$ support function checks, where $n$ is the ambient dimension of S.

source
LazySets.is_polyhedralMethod
is_polyhedral(S::ConvexSet)

Trait for polyhedral sets.

Input

  • S – set

Output

true only if the set behaves like an AbstractPolyhedron. The answer is conservative, i.e., may sometimes be false even if the set is polyhedral.

source
LazySets.an_elementMethod
an_element(S::ConvexSet{N}) where {N}

Return some element of a convex set.

Input

  • S – convex set

Output

An element of a convex set.

Algorithm

An element of the set is obtained by evaluating its support vector along direction $[1, 0, …, 0]$.

source
an_element(P::AbstractPolyhedron{N};
           [solver]=default_lp_solver(N)) where {N}

Return some element of a polyhedron.

Input

  • P – polyhedron
  • solver – (optional, default: default_lp_solver(N)) LP solver

Output

An element of the polyhedron, or an error if the polyhedron is empty.

Algorithm

An element of the polyhedron is obtained by solving a feasibility linear program.

source
an_element(L::Line2D{N}) where {N}

Return some element of a line.

Input

  • L – line

Output

An element on the line.

Algorithm

If the $b$ value of the line is zero, the result is the origin. Otherwise the result is some $x = [x1, x2]$ such that $a·[x1, x2] = b$. We first find out in which dimension $a$ is nonzero, say, dimension 1, and then choose $x1 = 1$ and accordingly $x2 = \frac{b - a1}{a2}$.

source
an_element(U::Universe{N}) where {N}

Return some element of a universe.

Input

  • U – universe

Output

The origin.

source
LazySets.tosimplehrepMethod
tosimplehrep(S::ConvexSet)

Return the simple H-representation $Ax ≤ b$ of a set from its list of linear constraints.

Input

  • S – set

Output

The tuple (A, b) where A is the matrix of normal directions and b is the vector of offsets.

Notes

This function only works for sets that can be represented exactly by a finite list of linear constraints. This fallback implementation relies on constraints_list(S).

source
LazySets.isuniversalMethod
isuniversal(X::ConvexSet{N}, [witness]::Bool=false) where {N}

Check whether a given convex set is universal, and otherwise optionally compute a witness.

Input

  • X – convex set
  • witness – (optional, default: false) compute a witness if activated

Output

  • If witness option is deactivated: true iff $X$ is universal
  • If witness option is activated:
    • (true, []) iff $X$ is universal
    • (false, v) iff $X$ is not universal and $v ∉ X$

Notes

This is a naive fallback implementation.

source
LazySets.affine_mapMethod
affine_map(M::AbstractMatrix, X::ConvexSet, v::AbstractVector; kwargs...)

Compute a concrete affine map.

Input

  • M – linear map
  • X – convex set
  • v – translation vector

Output

A set representing the affine map of X.

Algorithm

The implementation applies the functions linear_map and translate.

source
LazySets.exponential_mapMethod
exponential_map(M::AbstractMatrix, X::ConvexSet)

Compute the concrete exponential map of M and X, i.e., exp(M) * X.

Input

  • M – matrix
  • X – set

Output

A set representing the exponential map of M and X.

Algorithm

The implementation applies the functions exp and linear_map.

source
LazySets.reflectMethod
reflect(P::ConvexSet)

Concrete reflection of a convex set P, resulting in the reflected set -P.

Note

This function requires that the list of constraints of the set P is available, i.e. such that it can be written as $P = \{z ∈ ℝⁿ: ⋂ sᵢᵀz ≤ rᵢ, i = 1, ..., N\}.$

This function can be used to implement the alternative definition of the Minkowski Difference, which writes as

\[A ⊖ B = \{a − b | a ∈ A, b ∈ B\} = A ⊕ (-B)\]

by calling minkowski_sum(A, reflect(B)).

source
LazySets.is_interior_pointMethod
is_interior_point(d::AbstractVector{N}, P::ConvexSet{N};
                  p=N(Inf), ε=_rtol(N)) where {N<:Real}

Check if the point d is contained in the interior of the convex set P.

Input

  • d – point
  • P – set
  • p – (optional; default: N(Inf)) norm of the ball used to apply the error tolerance
  • ε – (optional; default: _rtol(N)) error tolerance of check

Output

Boolean which indicates if the point d is contained in P.

Algorithm

The implementation checks if a Ballp of norm p with center d and radius ε is contained in the set P. This is a numerical check for d ∈ interior(P) with error tolerance ε.

source
LazySets.isoperationtypeMethod
isoperationtype(X::Type{<:ConvexSet})

Check whether the given ConvexSet type is an operation or not.

Input

  • X – subtype of ConvexSet

Output

true if the given set type is a set-based operation and false otherwise.

Notes

The fallback for this function returns an error that isoperationtype is not implemented. Subtypes of ConvexSet should dispatch on this function as required.

See also isoperation(X<:ConvexSet).

Examples

julia> isoperationtype(BallInf)
false

julia> isoperationtype(LinearMap)
true
source
LazySets.isoperationMethod
isoperation(X::ConvexSet)

Check whether the given ConvexSet is an instance of a set operation or not.

Input

  • X – a ConvexSet

Output

true if X is an instance of a set-based operation and false otherwise.

Notes

The fallback implementation returns whether the set type of the input is an operation or not using isoperationtype.

See also isoperationtype(X::Type{<:ConvexSet}).

Examples

julia> B = BallInf([0.0, 0.0], 1.0);

julia> isoperation(B)
false

julia> isoperation(B ⊕ B)
true
source
LazySets.isequivalentMethod
isequivalent(X::ConvexSet, Y::ConvexSet)

Return whether two LazySets are equal in the mathematical sense, i.e. equivalent.

Input

  • X – any ConvexSet
  • Y – another ConvexSet

Output

true iff X is equivalent to Y.

Algorithm

First we check X ≈ Y, which returns true if and only if X and Y have the same type and approximately the same values (checked with LazySets._isapprox). If that fails, we check the double inclusion X ⊆ Y && Y ⊆ X.

Examples

julia> X = BallInf([0.1, 0.2], 0.3);

julia> Y = convert(HPolytope, X);

julia> X == Y
false

julia> isequivalent(X, Y)
true
source
LazySets.isconvextypeMethod
isconvextype(X::Type{<:ConvexSet})

Check whether the given ConvexSet type is convex.

Input

  • X – subtype of ConvexSet

Output

true if the given set type is guaranteed to be convex by using only type information, and false otherwise.

Notes

Since this operation only acts on types (not on values), it can return false negatives, i.e. there may be instances where the set is convex, even though the answer of this function is false. The examples below illustrate this point.

Examples

A ball in the infinity norm is always convex, hence we get:

julia> isconvextype(BallInf)
true

For instance, the union (UnionSet) of two sets may in general be either convex or not, since convexity cannot be decided by just using type information. Hence, isconvextype returns false if X is Type{<:UnionSet}.

julia> isconvextype(UnionSet)
false

However, the type parameters from the set operations allow to decide convexity in some cases, by falling back to the convexity of the type of its arguments. Consider for instance the lazy intersection. The intersection of two convex sets is always convex, hence we can get:

julia> isconvextype(Intersection{Float64, BallInf{Float64}, BallInf{Float64}})
true
source
LazySets.lowMethod
low(X::ConvexSet, i::Int)

Return the lower coordinate of a set in a given dimension.

Input

  • X – set
  • i – dimension of interest

Output

The lower coordinate of the set in the given dimension.

source
LazySets.highMethod
high(X::ConvexSet, i::Int)

Return the higher coordinate of a set in a given dimension.

Input

  • X – set
  • i – dimension of interest

Output

The higher coordinate of the set in the given dimension.

source
Base.extremaMethod
extrema(X::ConvexSet, i::Int)

Return the lower and higher coordinate of a set in a given dimension.

Input

  • X – set
  • i – dimension of interest

Output

The lower and higher coordinate of the set in the given dimension.

Notes

The result is equivalent to (low(X, i), high(X, i)), but sometimes it can be computed more efficiently.

source
LazySets.lowMethod
low(X::ConvexSet)

Return a vector with the lowest coordinates of the set for each canonical direction.

Input

  • X – set

Output

A vector with the lower coordinate of the set for each dimension.

Notes

See also low(X::ConvexSet, i::Int).

source
LazySets.highMethod
high(X::ConvexSet)

Return a vector with the highest coordinate of the set for each canonical direction.

Input

  • X – set

Output

A vector with the highest coordinate of the set for each dimension.

Notes

See also high(X::ConvexSet, i::Int).

source
Base.extremaMethod
extrema(X::ConvexSet)

Return two vectors with the lowest and highest coordinate of X for each dimension.

Input

  • X – set

Output

Two vectors with the lowest and highest coordinates of X for each dimension.

Notes

The result is equivalent to (low(X), high(X)), but sometimes it can be computed more efficiently.

source
LazySets.surfaceMethod
surface(X::ConvexSet{N}) where {N}

Compute the surface area of a set.

Input

  • X – set

Output

A number representing the surface area of X.

source
LazySets.areaMethod
area(X::ConvexSet{N}) where {N}

Compute the area of a two-dimensional polytopic set using the Shoelace formula.

Input

  • X – two-dimensional set

Output

A number representing the area of X.

Notes

This algorithm is applicable to any lazy set X such that its list of vertices, vertices_list, can be computed.

Algorithm

Let m be the number of vertices of X. The following instances are considered:

  • m = 0, 1, 2: the output is zero.
  • m = 3: the triangle case is computed using the Shoelace formula with 3 points.
  • m = 4: the quadrilateral case is obtained by the factored version of the Shoelace formula with 4 points.

Otherwise, the general Shoelace formula is used; for detals see the wikipedia article Shoelace formula.

source
area(∅::EmptySet{N}) where {N}

Return the area of an empty set.

Input

  • – empty set

Output

The zero element of type N.

source
LazySets.concretizeMethod
concretize(X::ConvexSet)

Construct a concrete representation of a (possibly lazy) set.

Input

  • X – set

Output

A concrete representation of X (as far as possible).

Notes

Since not every lazy set has a concrete set representation in this library, the result may be partially lazy.

source
LazySets.complementMethod
complement(X::ConvexSet)

Return the complement of a set.

Input

  • X – set

Output

A UnionSetArray of half-spaces, i.e. the output is the union of the linear constraints which are obtained by complementing each constraint of X.

Algorithm

The principle used in this function is that if $X$ and $Y$ are any pair of sets, then $(X ∩ Y)^C = X^C ∪ Y^C$. In particular, we can apply this rule for each constraint that defines a polyhedral set, hence the concrete complement can be represented as the set union of the complement of each constraint.

source
LazySets.projectMethod
project(S::ConvexSet{N},
        block::AbstractVector{Int},
        [::Nothing=nothing],
        [n]::Int=dim(S);
        [kwargs...]
       ) where {N}

Project a high-dimensional set to a given block by using a concrete linear map.

Input

  • S – set
  • block – block structure - a vector with the dimensions of interest
  • nothing – (default: nothing) used for dispatch
  • n – (optional, default: dim(S)) ambient dimension of the set S

Output

A set representing the projection of the set S to block block.

Algorithm

We apply the function linear_map.

source
LazySets.projectMethod
project(S::ConvexSet,
        block::AbstractVector{Int},
        set_type::Type{TS},
        [n]::Int=dim(S);
        [kwargs...]
       ) where {TS<:ConvexSet}

Project a high-dimensional set to a given block and set type, possibly involving an overapproximation.

Input

  • S – set
  • block – block structure - a vector with the dimensions of interest
  • set_type – target set type
  • n – (optional, default: dim(S)) ambient dimension of the set S

Output

A set of type set_type representing an overapproximation of the projection of S.

Algorithm

  1. Project the set S with M⋅S, where M is the identity matrix in the block

coordinates and zero otherwise.

  1. Overapproximate the projected lazy set using overapproximate and

set_type.

source
LazySets.projectMethod
project(S::ConvexSet,
        block::AbstractVector{Int},
        set_type_and_precision::Pair{T, N},
        [n]::Int=dim(S);
        [kwargs...]
       ) where {T<:UnionAll, N<:Real}

Project a high-dimensional set to a given block and set type with a certified error bound.

Input

  • S – set
  • block – block structure - a vector with the dimensions of interest
  • set_type_and_precision – pair (T, ε) of a target set type T and an error bound ε for approximation
  • n – (optional, default: dim(S)) ambient dimension of the set S

Output

A set representing the epsilon-close approximation of the projection of S.

Notes

Currently we only support HPolygon as set type, which implies that the set must be two-dimensional.

Algorithm

  1. Project the set S with M⋅S, where M is the identity matrix in the block

coordinates and zero otherwise.

  1. Overapproximate the projected lazy set with the given error bound ε.
source
LazySets.projectFunction
project(S::ConvexSet,
        block::AbstractVector{Int},
        ε::Real,
        [n]::Int=dim(S);
        [kwargs...]
       )

Project a high-dimensional set to a given block and set type with a certified error bound.

Input

  • S – set
  • block – block structure - a vector with the dimensions of interest
  • ε – error bound for approximation
  • n – (optional, default: dim(S)) ambient dimension of the set S

Output

A set representing the epsilon-close approximation of the projection of S.

Algorithm

  1. Project the set S with M⋅S, where M is the identity matrix in the block

coordinates and zero otherwise.

  1. Overapproximate the projected lazy set with the given error bound ε.

The target set type is chosen automatically.

source
ReachabilityBase.Arrays.rectifyFunction
rectify(X::ConvexSet, [concrete_intersection]::Bool=false)

Concrete rectification of a set.

Input

  • X – set
  • concrete_intersection – (optional, default: false) flag to compute concrete intersections for intermediate results

Output

A set corresponding to the rectification of X, which is in general a union of linear maps of intersections.

Algorithm

For each dimension in which X is both positive and negative we split X into these two parts. Additionally we project the negative part to zero.

source
SparseArrays.permuteFunction
permute(X::ConvexSet, p::AbstractVector{Int})

Permute the dimensions of a set according to a given permutation vector.

Input

  • X – set
  • p – permutation vector

Output

A new set corresponding to X where the dimensions have been permuted according to p.

source
Base.rationalizeMethod
rationalize(::Type{T}, X::ConvexSet{N}, tol::Real) where {T<:Integer, N<:AbstractFloat}

Approximate a ConvexSet of floating point numbers as a set whose entries are rationals of the given integer type.

Input

  • T – (optional, default: Int) integer type to represent the rationals
  • X – set which has floating-point components
  • tol – (optional, default: eps(N)) tolerance of the result; each rationalized component will differ by no more than tol with respect to the floating-point value

Output

A ConvexSet of the same base type of X where each numerical component is of type Rational{T}.

source

Plotting is available for general one- or two-dimensional ConvexSets, provided that the overapproximation using iterative refinement is available:

LazySets.plot_recipeMethod
plot_recipe(X::ConvexSet{N}, [ε]=N(PLOT_PRECISION)) where {N}

Convert a convex set to a pair (x, y) of points for plotting.

Input

  • X – convex set
  • ε – (optional, default: PLOT_PRECISION) approximation error bound

Output

A pair (x, y) of points that can be plotted.

Algorithm

We first assert that X is bounded.

One-dimensional sets are converted to an Interval. We do not support three-dimensional or higher-dimensional sets at the moment.

For two-dimensional sets, we first compute a polygonal overapproximation. The second argument, ε, corresponds to the error in Hausdorff distance between the overapproximating set and X. The default value PLOT_PRECISION is chosen such that the unit ball in the 2-norm is approximated with reasonable accuracy. On the other hand, if you only want to produce a fast box-overapproximation of X, pass ε=Inf. Finally, we use the plot recipe for polygons.

source
RecipesBase.apply_recipeMethod
plot_lazyset(X::ConvexSet{N}, [ε]::N=N(PLOT_PRECISION); ...) where {N}

Plot a convex set.

Input

  • X – convex set
  • ε – (optional, default: PLOT_PRECISION) approximation error bound

Notes

See plot_recipe(::ConvexSet).

For polyhedral set types (subtypes of AbstractPolyhedron), the argument ε is ignored.

Examples

julia> B = Ball2(ones(2), 0.1);

julia> plot(B, 1e-3)  # default accuracy value (explicitly given for clarity)

julia> plot(B, 1e-2)  # faster but less accurate than the previous call
source
RecipesBase.apply_recipeMethod
plot_list(list::AbstractVector{VN}, [ε]::N=N(PLOT_PRECISION),
          [Nφ]::Int=PLOT_POLAR_DIRECTIONS; [same_recipe]=false; ...)
    where {N, VN<:ConvexSet{N}}

Plot a list of convex sets.

Input

  • list – list of convex sets (1D or 2D)
  • ε – (optional, default: PLOT_PRECISION) approximation error bound
  • – (optional, default: PLOT_POLAR_DIRECTIONS) number of polar directions (used to plot lazy intersections)
  • same_recipe – (optional, default: false) switch for faster plotting but without individual plot recipes (see notes below)

Notes

For each set in the list we apply an individual plot recipe.

The option same_recipe provides access to a faster plotting scheme where all sets in the list are first converted to polytopes and then plotted in one single run. This, however, is not suitable when plotting flat sets (line segments, singletons) because then the polytope plot recipe does not deliver good results. Hence by default we do not use this option. For plotting a large number of (non-flat) polytopes, we highly advise activating this option.

Examples

julia> B1 = BallInf(zeros(2), 0.4);

julia> B2 = BallInf(ones(2), 0.4);

julia> plot([B1, B2])

Some of the sets in the list may not be plotted precisely but rather overapproximated first. The second argument ε controls the accuracy of this overapproximation.

julia> Bs = [BallInf(zeros(2), 0.4), Ball2(ones(2), 0.4)];

julia> plot(Bs, 1e-3)  # default accuracy value (explicitly given for clarity)

julia> plot(Bs, 1e-2)  # faster but less accurate than the previous call
source

For three-dimensional sets, we support Makie:

LazySets.plot3dFunction
plot3d(S::ConvexSet; backend=default_polyhedra_backend(S),
       alpha=1.0, color=:blue, colormap=:viridis, colorrange=nothing,
       interpolate=false, linewidth=1, overdraw=false, shading=true,
       transparency=true, visible=true)

Plot a three-dimensional convex set using Makie.

Input

  • S – convex set
  • backend – (optional, default: default_polyhedra_backend(S)) polyhedral computations backend
  • alpha – (optional, default: 1.0) float in [0,1]; the alpha or transparency value
  • color – (optional, default: :blue) Symbol or Colorant; the color of the main plot element (markers, lines, etc.) and it can be a color symbol/string like :red
  • colormap – (optional, default: :viridis) the color map of the main plot; call available_gradients() to see what gradients are available, and it can also be used as [:red, :black]
  • colorrange – (optional, default: nothing, which falls back to Makie.Automatic()) a tuple (min, max) where min and max specify the data range to be used for indexing the colormap
  • interpolate – (optional, default: false) a bool for heatmap and images, it toggles color interpolation between nearby pixels
  • linewidth – (optional, default: 1) a number that specifies the width of the line in line and linesegments plots
  • overdraw – (optional, default: false)
  • shading – (optional, default: true) a boolean that specifies if shading should be on or not (for meshes)
  • transparency – (optional, default: true) if true, the set is transparent otherwise it is displayed as a solid object
  • visible – (optional, default: true) a bool that toggles visibility of the plot

For a complete list of attributes and usage see Makie's documentation.

Notes

This plot recipe works by computing the list of constraints of S and converting to a polytope in H-representation. Then, this polytope is transformed with Polyhedra.Mesh and it is plotted using the mesh function.

If the function constraints_list is not applicable to your set S, try overapproximation first; e.g. via

julia> Sapprox = overapproximate(S, SphericalDirections(10))

julia> using Polyhedra, GLMakie

julia> plot3d(Sapprox)

The number 10 above corresponds to the number of directions considered; for better resolution use higher values (but it will take longer).

For efficiency consider using the CDDLib backend, as in

julia> using CDDLib

julia> plot3d(Sapprox, backend=CDDLib.Library())

Examples

The functionality requires both Polyhedra and a Makie backend. After loading LazySets, do using Polyhedra, GLMakie (or another Makie backend).

julia> using LazySets, Polyhedra, GLMakie

julia> plot3d(10. * rand(Hyperrectangle, dim=3))

julia> plot3d!(10. * rand(Hyperrectangle, dim=3), color=:red)
source
LazySets.plot3d!Function
plot3d!(S::ConvexSet; backend=default_polyhedra_backend(S),
        alpha=1.0, color=:blue, colormap=:viridis, colorrange=nothing, interpolate=false,
        linewidth=1, overdraw=false, shading=true, transparency=true, visible=true)

Plot a three-dimensional convex set using Makie.

Input

See plot3d for the description of the inputs. For a complete list of attributes and usage see Makie's documentation.

Notes

See the documentation of plot3d for examples.

source

Set functions that override Base functions

Base.:==Method
==(X::ConvexSet, Y::ConvexSet)

Return whether two LazySets of the same type are exactly equal.

Input

  • X – any ConvexSet
  • Y – another ConvexSet of the same type as X

Output

  • true iff X is equal to Y.

Notes

The check is purely syntactic and the sets need to have the same base type. For instance, X::VPolytope == Y::HPolytope returns false even if X and Y represent the same polytope. However X::HPolytope{Int64} == Y::HPolytope{Float64} is a valid comparison.

Algorithm

We recursively compare the fields of X and Y until a mismatch is found.

Examples

julia> HalfSpace([1], 1) == HalfSpace([1], 1)
true

julia> HalfSpace([1], 1) == HalfSpace([1.0], 1.0)
true

julia> Ball1([0.], 1.) == Ball2([0.], 1.)
false
source
Base.:≈Method
≈(X::ConvexSet, Y::ConvexSet)

Return whether two LazySets of the same type are approximately equal.

Input

  • X – any ConvexSet
  • Y – another ConvexSet of the same type as X

Output

  • true iff X is equal to Y.

Notes

The check is purely syntactic and the sets need to have the same base type. For instance, X::VPolytope ≈ Y::HPolytope returns false even if X and Y represent the same polytope. However X::HPolytope{Int64} ≈ Y::HPolytope{Float64} is a valid comparison.

Algorithm

We recursively compare the fields of X and Y until a mismatch is found.

Examples

julia> HalfSpace([1], 1) ≈ HalfSpace([1], 1)
true

julia> HalfSpace([1], 1) ≈ HalfSpace([1.00000001], 0.99999999)
true

julia> HalfSpace([1], 1) ≈ HalfSpace([1.0], 1.0)
true

julia> Ball1([0.], 1.) ≈ Ball2([0.], 1.)
false
source
Base.copyMethod
copy(S::ConvexSet)

Return a copy of the given set by copying its values recursively.

Input

  • S – any ConvexSet

Output

A copy of S.

Notes

This function performs a copy of each field in S. See the documentation of ?copy for further details.

source
Base.eltypeFunction
eltype(::Type{<:ConvexSet{N}}) where {N}

Return the numeric type (N) of the given set type.

Input

  • T – set type, used for dispatch

Output

The numeric type of T.

source
eltype(::ConvexSet{N}) where {N}

Return the numeric type (N) of the given set.

Input

  • X – set instance, used for dispatch

Output

The numeric type of X.

source

Aliases for set types

LazySets.CompactSetType
CompactSet

An alias for compact set types.

Notes

Most lazy operations are not captured by this alias because whether their result is compact or not depends on the argument(s).

source
LazySets.NonCompactSetType
NonCompactSet

An alias for non-compact set types.

Notes

Most lazy operations are not captured by this alias because whether their result is non-compact or not depends on the argument(s).

source

Implementations

Concrete set representations:

Lazy set operations:

Centrally symmetric sets (AbstractCentrallySymmetric)

Centrally symmetric sets such as balls of different norms are characterized by a center. Note that there is a special interface combination Centrally symmetric polytope.

LazySets.AbstractCentrallySymmetricType
AbstractCentrallySymmetric{N} <: ConvexSet{N}

Abstract type for centrally symmetric sets.

Notes

Every concrete AbstractCentrallySymmetric must define the following functions:

  • center(::AbstractCentrallySymmetric) – return the center point
  • center(::AbstractCentrallySymmetric, i::Int) – return the center point at index i
julia> subtypes(AbstractCentrallySymmetric)
3-element Vector{Any}:
 Ball2
 Ballp
 Ellipsoid
source

This interface defines the following functions:

LazySets.dimMethod
dim(S::AbstractCentrallySymmetric)

Return the ambient dimension of a centrally symmetric set.

Input

  • S – set

Output

The ambient dimension of the set.

source
LazySets.isboundedMethod
isbounded(S::AbstractCentrallySymmetric)

Determine whether a centrally symmetric set is bounded.

Input

  • S – centrally symmetric set

Output

true (since a set with a unique center must be bounded).

source
LazySets.isuniversalMethod
isuniversal(S::AbstractCentrallySymmetric{N}, [witness]::Bool=false
           ) where {N}

Check whether a centrally symmetric set is universal.

Input

  • S – centrally symmetric set
  • witness – (optional, default: false) compute a witness if activated

Output

  • If witness option is deactivated: false
  • If witness option is activated: (false, v) where $v ∉ S$

Algorithm

A witness is obtained by computing the support vector in direction d = [1, 0, …, 0] and adding d on top.

source
LazySets.an_elementMethod
an_element(S::AbstractCentrallySymmetric)

Return some element of a centrally symmetric set.

Input

  • S – centrally symmetric set

Output

The center of the centrally symmetric set.

source
Base.isemptyMethod
isempty(S::AbstractCentrallySymmetric)

Return if a centrally symmetric set is empty or not.

Input

  • S – centrally symmetric set

Output

false.

source
LazySets.centerMethod
center(H::AbstractCentrallySymmetric, i::Int)

Return the center along a given dimension of a centrally symmetric set.

Input

  • S – centrally symmetric set
  • i – dimension of interest

Output

The center along a given dimension of the centrally symmetric set.

source

Implementations

Polyhedra (AbstractPolyhedron)

A polyhedron has finitely many facets (H-representation) and is not necessarily bounded.

LazySets.AbstractPolyhedronType
AbstractPolyhedron{N} <: ConvexSet{N}

Abstract type for compact convex polyhedral sets.

Notes

Every concrete AbstractPolyhedron must define the following functions:

  • constraints_list(::AbstractPolyhedron{N}) – return a list of all facet constraints
julia> subtypes(AbstractPolyhedron)
8-element Vector{Any}:
 AbstractPolytope
 HPolyhedron
 HalfSpace
 Hyperplane
 Line
 Line2D
 Star
 Universe

Polyhedra are defined as the intersection of a finite number of closed half-spaces. As such, polyhedra are closed and convex but not necessarily bounded. Bounded polyhedra are called polytopes (see AbstractPolytope).

source

This interface defines the following functions:

Base.:∈Method
∈(x::AbstractVector, P::AbstractPolyhedron)

Check whether a given point is contained in a polyhedron.

Input

  • x – point/vector
  • P – polyhedron

Output

true iff $x ∈ P$.

Algorithm

This implementation checks if the point lies inside each defining half-space.

source
LazySets.isuniversalMethod
isuniversal(P::AbstractPolyhedron{N}, [witness]::Bool=false) where {N}

Check whether a polyhedron is universal.

Input

  • P – polyhedron
  • witness – (optional, default: false) compute a witness if activated

Output

  • If witness option is deactivated: true iff $P$ is universal
  • If witness option is activated:
    • (true, []) iff $P$ is universal
    • (false, v) iff $P$ is not universal and $v ∉ P$

Algorithm

P is universal iff it has no constraints.

A witness is produced using isuniversal(H) where H is the first linear constraint of P.

source
LazySets.constrained_dimensionsMethod
constrained_dimensions(P::AbstractPolyhedron)

Return the indices in which a polyhedron is constrained.

Input

  • P – polyhedron

Output

A vector of ascending indices i such that the polyhedron is constrained in dimension i.

Examples

A 2D polyhedron with constraint $x1 ≥ 0$ is constrained in dimension 1 only.

source
LazySets.linear_mapMethod
linear_map(M::AbstractMatrix{NM},
           P::AbstractPolyhedron{NP};
           [algorithm]::Union{String, Nothing}=nothing,
           [check_invertibility]::Bool=true,
           [cond_tol]::Number=DEFAULT_COND_TOL,
           [inverse]::Union{AbstractMatrix{N}, Nothing}=nothing,
           [backend]=nothing,
           [elimination_method]=nothing) where {NM, NP}

Concrete linear map of a polyhedral set.

Input

  • M – matrix

  • P – polyhedral set

  • algorithm – (optional; default: nothing) algorithm to be used; for the description see the Algorithm section below; possible choices are:

    • "inverse", alias: "inv"
    • "inverse_right", alias: "inv_right"
    • "elimination", alias: "elim"
    • "lift"
    • "vrep"
    • "vrep_chull"
  • check_invertibility – (optional, default: true) if true check whether given matrix M is invertible; set to false only if you know that M is invertible

  • cond_tol – (optional; default: DEFAULT_COND_TOL) tolerance of matrix condition (used to check whether the matrix is invertible)

  • inverse – (optional; default: nothing) matrix inverse M⁻¹; use this option if you have already computed the inverse matrix of M

  • backend – (optional: default: nothing) polyhedra backend

  • elimination_method – (optional: default: nothing) elimination method for the "elimination" algorithm

Output

The type of the result is "as close as possible" to the the type of P. Let (m, n) be the size of M, where m ≠ n is allowed for rectangular maps.

To fix the type of the output to something different than the default value, consider post-processing the result of this function with a call to a suitable convert method.

In particular, the output depends on the type of P, on m, and the algorithm that was used:

  • If the vertex-based approach was used:

    • If P is a VPolygon and m = 2 then the output is a VPolygon.
    • If P is a VPolytope then the output is a VPolytope.
    • Otherwise, the output is an Interval if m = 1, a VPolygon if m = 2 and a VPolytope in other cases.
  • If the invertibility criterion was used:

    • The types of HalfSpace, Hyperplane, Line2D and AbstractHPolygon are preserved.
    • If P is an AbstractPolytope, then the output is an Interval if m = 1, an HPolygon if m = 2 and an HPolytope in other cases.
    • Otherwise, the output is an HPolyhedron.

Notes

Since the different linear map algorithms work at the level of constraints (not sets representations), this function uses dispatch on two stages: once the algorithm has been defined, first the helper functions _linear_map_hrep_helper (resp. _linear_map_vrep) are invoked, which dispatch on the set type. Then, each helper function calls the concrete implementation of _linear_map_hrep, which dispatches on the algorithm, and returns a list of constraints.

To simplify working with different algorithms and options, the types <: AbstractLinearMapAlgorithm are used. These types are singleton type or types that carry only the key data for the given algorithm, such as the matrix inverse or the polyhedra backend.

New subtypes of the AbstractPolyhedron interface may define their own helper functions _linear_map_vrep, respectively _linear_map_hrep_helper for special handling of the constraints returned by the implementations of _linear_map_hrep; otherwise the fallback implementation for AbstractPolyhedron is used, which instantiates an HPolyhedron.

Algorithm

This function mainly implements several approaches for the linear map: inverse, right inverse, transformation to the vertex representation, variable elimination, and variable lifting. Depending on the properties of M and P, one algorithm may be preferable over the other. Details on the algorithms are given in the following subsections.

Otherwise, if the algorithm argument is not specified, a default option is chosen based on heuristics on the types and values of M and P:

  • If the "inverse" algorithm applies, it is used.
  • If the "inverse_right" algorithm applies, it is used.
  • Otherwise, if the "lift" algorithm applies, it is used.
  • Otherwise, the "elimination" algorithm is used.

Note that "inverse" does not require the external library Polyhedra, and neither does "inverse_right". However, the fallback method "elimination" requires Polyhedra as well as the library CDDLib.

The optional keyword arguments inverse and check_invertibility modify the default behavior:

  • If an inverse matrix is passed in inverse, the given algorithm is applied, and if none is given, either "inverse" or "inverse_right" is applied (in that order of preference).
  • If check_invertibility is set to false, the given algorithm is applied, and if none is given, either "inverse" or "inverse_right" is applied (in that order of preference).

Inverse

This algorithm is invoked with the keyword argument algorithm="inverse" (or algorithm="inv"). The algorithm requires that M is invertible, square, and dense. If you know a priori that M is invertible, set the flag check_invertibility=false, such that no extra checks are done within linear_map. Otherwise, we check the sufficient condition that the condition number of M is not too high. The threshold for the condition number can be modified from its default value, DEFAULT_COND_TOL, by passing a custom cond_tol.

The algorithm is described next. Assuming that the matrix $M$ is invertible (which we check via a sufficient condition,), $y = M x$ implies $x = \text{inv}(M) y$ and we can transform the polyhedron $A x ≤ b$ to the polyhedron $A \text{inv}(M) y ≤ b$.

If the dense condition on M is not fullfilled, there are two suggested workarounds: either transform to dense matrix, i.e. calling linear_map with Matrix(M), or use the "inverse_right" algorithm, which does not compute the inverse matrix explicitly, but uses a polyalgorithm; see the documentation of ? for details.

Inverse-right

This algorithm is invoked with the keyword argument algorithm="inverse_right" (or algorithm="inv_right"). This algorithm applies to square and invertible matrices M. The idea is essentially the same as for the "inverse" algorithm; the difference is that in "inverse" the full matrix inverse is computed, and in "inverse_right" only the left division on the normal vectors is used. In particular, "inverse_right" is good as a workaround when M is sparse (since the inv function is not available for sparse matrices).

Elimination

This algorithm is invoked with the keyword argument algorithm = "elimination" or algorithm = "elim". The algorithm applies to any matrix M (invertible or not), and any polyhedron P (bounded or not).

The idea is described next. If P : Ax <= b and y = Mx denote the polyhedron and the linear map respectively, we consider the vector z = [y, x], write the given equalities and the inequalities, and then eliminate the last x variables (there are length(x) in total) using a call to Polyhedra.eliminate to a backend library that can do variable elimination, typically CDDLib with the BlockElimination() algorithm. In this way we have eliminated the "old" variables x and kept the "new" or transformed variables "y".

The default elimination method is block elimination. For possible options we refer to the documentation of Polyhedra, projection/elimination.

Lift

This algorithm is invoked with the keyword argument algorithm="lift". The algorithm applies if M is rectangular of size m × n with m > n and full rank (i.e. of rank n).

The idea is to embed the polyhedron into the m-dimensional space by appending zeros, i.e. extending all constraints of P to m dimensions, and constraining the last m - n dimensions to 0. The matrix resulting matrix is extended to an invertible m × m matrix and the algorithm using the inverse of the linear map is applied. For the technical details of the extension of M to a higher-dimensional invertible matrix, see ReachabilityBase.Arrays.extend.

Vertex representation

This algorithm is invoked with the keyword argument algorithm either "vrep" or "vrep_chull". The idea is to convert the polyhedron to its vertex representation and apply the linear map to each vertex of P.

The returned set is a polytope in vertex representation. Note that conversion of the result back to half-space representation is not computed by default, since this may be costly. If you used this algorithm and still want to convert back to half-space representation, apply tohrep to the result of this function. Note that this method only works for bounded polyhedra.

source
LazySets.chebyshev_centerMethod
chebyshev_center(P::AbstractPolyhedron{N};
                 [get_radius]::Bool=false,
                 [backend]=default_polyhedra_backend(P),
                 [solver]=default_lp_solver_polyhedra(N; presolve=true)
                 ) where {N}

Compute the Chebyshev center of a polytope.

Input

  • P – polytope
  • get_radius – (optional; default: false) option to additionally return the radius of the largest ball enclosed by P around the Chebyshev center
  • backend – (optional; default: default_polyhedra_backend(P)) the backend for polyhedral computations
  • solver – (optional; default: default_lp_solver_polyhedra(N; presolve=true)) the LP solver passed to Polyhedra

Output

If get_radius is false, the result is the Chebyshev center of P. If get_radius is true, the result is the pair (c, r) where c is the Chebyshev center of P and r is the radius of the largest ball with center c enclosed by P.

Notes

The Chebyshev center is the center of a largest Euclidean ball enclosed by P. In general, the center of such a ball is not unique (but the radius is).

source
LazySets.an_elementMethod
an_element(S::ConvexSet{N}) where {N}

Return some element of a convex set.

Input

  • S – convex set

Output

An element of a convex set.

Algorithm

An element of the set is obtained by evaluating its support vector along direction $[1, 0, …, 0]$.

source
an_element(P::AbstractPolyhedron{N};
           [solver]=default_lp_solver(N)) where {N}

Return some element of a polyhedron.

Input

  • P – polyhedron
  • solver – (optional, default: default_lp_solver(N)) LP solver

Output

An element of the polyhedron, or an error if the polyhedron is empty.

Algorithm

An element of the polyhedron is obtained by solving a feasibility linear program.

source
an_element(L::Line2D{N}) where {N}

Return some element of a line.

Input

  • L – line

Output

An element on the line.

Algorithm

If the $b$ value of the line is zero, the result is the origin. Otherwise the result is some $x = [x1, x2]$ such that $a·[x1, x2] = b$. We first find out in which dimension $a$ is nonzero, say, dimension 1, and then choose $x1 = 1$ and accordingly $x2 = \frac{b - a1}{a2}$.

source
an_element(U::Universe{N}) where {N}

Return some element of a universe.

Input

  • U – universe

Output

The origin.

source
LazySets.isboundedMethod
isbounded(P::AbstractPolyhedron{N}; [solver]=default_lp_solver(N)) where {N}

Determine whether a polyhedron is bounded.

Input

  • P – polyhedron
  • solver – (optional, default: default_lp_solver(N)) the backend used to solve the linear program

Output

true iff the polyhedron is bounded

Algorithm

We first check if the polyhedron has more than max(dim(P), 1) constraints, which is a necessary condition for boundedness.

If so, we check boundedness via _isbounded_stiemke.

source
isbounded(r::Rectification{N}) where {N}

Determine whether a rectification is bounded.

Input

  • r – rectification

Output

true iff the rectification is bounded.

Algorithm

Let $X$ be the set wrapped by rectification $r$. We first check whether $X$ is bounded (because then $r$ is bounded). Otherwise, we check unboundedness of $X$ in direction $(1, 1, …, 1)$, which is sufficient for unboundedness of $r$; this step is not necessary but rather a heuristics. Otherwise, we check boundedness of $X$ in every positive unit direction, which is sufficient and necessary for boundedness of $r$.

source
LazySets.vertices_listMethod
vertices_list(P::AbstractPolyhedron; check_boundedness::Bool=true)

Return the list of vertices of a polyhedron in constraint representation.

Input

  • P – polyhedron in constraint representation
  • check_boundedness – (optional, default: true) if true, check whether the polyhedron is bounded

Output

The list of vertices of P, or an error if P is unbounded.

Notes

This function returns an error if the polyhedron is unbounded. Otherwise, the polyhedron is converted to an HPolytope and its list of vertices is computed.

Examples

julia> P = HPolyhedron([HalfSpace([1.0, 0.0], 1.0),
                        HalfSpace([0.0, 1.0], 1.0),
                        HalfSpace([-1.0, 0.0], 1.0),
                        HalfSpace([0.0, -1.0], 1.0)]);

julia> length(vertices_list(P))
4
source
LazySets.projectMethod
project(P::AbstractPolyhedron{N}, block::AbstractVector{Int};
        [kwargs...]) where {N}

Concrete projection of a polyhedral set.

Input

  • P – set
  • block – block structure, a vector with the dimensions of interest

Output

A polyhedron representing the projection of P on the dimensions specified by block. If P was bounded, the result is an HPolytope; otherwise the result is an HPolyhedron. Note that there are more specific methods for specific input types, which give a different output type; e.g., projecting a Ball1 results in a Ball1.

Algorithm

  • We first try to exploit the special case where each of the constraints of P and block are compatible, which is one of the two cases described below. Let c be a constraint of P and let $D_c$ and $D_b$ be the set of dimensions in which c resp. block are constrained.
    • If $D_c ⊆ D_b$, then one can project the normal vector of c.
    • If $D_c ∩ D_b = ∅$, then the constraint becomes redundant.
  • In the general case, we compute the concrete linear map of the projection matrix associated to the given block structure.

Examples

Consider the four-dimensional cross-polytope (unit ball in the 1-norm):

julia> P = convert(HPolytope, Ball1(zeros(4), 1.0));

All dimensions are constrained, and computing the (trivial) projection on the whole space behaves as expected:

julia> constrained_dimensions(P)
4-element Vector{Int64}:
 1
 2
 3
 4

julia> project(P, [1, 2, 3, 4]) == P
true

Each constraint of the cross polytope is constrained in all dimensions.

Now let's take a ball in the infinity norm and remove some constraints:

julia> B = BallInf(zeros(4), 1.0);

julia> c = constraints_list(B)[1:2]
2-element Vector{HalfSpace{Float64, ReachabilityBase.Arrays.SingleEntryVector{Float64}}}:
 HalfSpace{Float64, ReachabilityBase.Arrays.SingleEntryVector{Float64}}([1.0, 0.0, 0.0, 0.0], 1.0)
 HalfSpace{Float64, ReachabilityBase.Arrays.SingleEntryVector{Float64}}([0.0, 1.0, 0.0, 0.0], 1.0)

julia> P = HPolyhedron(c);

julia> constrained_dimensions(P)
2-element Vector{Int64}:
 1
 2

Finally we take the concrete projection onto variables 1 and 2:

julia> project(P, [1, 2]) |> constraints_list
2-element Vector{HalfSpace{Float64, Vector{Float64}}}:
 HalfSpace{Float64, Vector{Float64}}([1.0, 0.0], 1.0)
 HalfSpace{Float64, Vector{Float64}}([0.0, 1.0], 1.0)
source

Plotting (bounded) polyhedra is available, too:

LazySets.plot_recipeMethod
plot_recipe(P::AbstractPolyhedron{N}, [ε]=zero(N)) where {N}

Convert a (bounded) polyhedron to a pair (x, y) of points for plotting.

Input

  • P – bounded polyhedron
  • ε – (optional, default: 0) ignored, used for dispatch

Output

A pair (x, y) of points that can be plotted, where x is the vector of x-coordinates and y is the vector of y-coordinates.

Algorithm

We first assert that P is bounded (i.e., that P is a polytope).

One-dimensional polytopes are converted to an Interval. Three-dimensional or higher-dimensional polytopes are not supported.

For two-dimensional polytopes (i.e., polygons) we compute their set of vertices using vertices_list and then plot the convex hull of these vertices.

source

Implementations

Polytopes (AbstractPolytope)

A polytope is a bounded set with finitely many vertices (V-representation) resp. facets (H-representation). Note that there is a special interface combination Centrally symmetric polytope.

LazySets.AbstractPolytopeType
AbstractPolytope{N} <: AbstractPolyhedron{N}

Abstract type for compact convex polytopic sets.

Notes

Every concrete AbstractPolytope must define the following functions:

  • vertices_list(::AbstractPolytope{N}) – return a list of all vertices
julia> subtypes(AbstractPolytope)
4-element Vector{Any}:
 AbstractCentrallySymmetricPolytope
 AbstractPolygon
 HPolytope
 VPolytope

A polytope is a bounded polyhedron (see AbstractPolyhedron). Polytopes are compact convex sets with either of the following equivalent properties:

  1. They are the intersection of a finite number of closed half-spaces.
  2. They are the convex hull of finitely many vertices.
source

This interface defines the following functions:

LazySets.isboundedMethod
isbounded(P::AbstractPolytope)

Determine whether a polytopic set is bounded.

Input

  • P – polytopic set

Output

true (since a polytope must be bounded).

source
LazySets.isuniversalMethod
isuniversal(P::AbstractPolytope{N}, [witness]::Bool=false) where {N}

Check whether a polyhedron is universal.

Input

  • P – polyhedron
  • witness – (optional, default: false) compute a witness if activated

Output

  • If witness option is deactivated: false
  • If witness option is activated: (false, v) where $v ∉ P$

Algorithm

A witness is produced using isuniversal(H) where H is the first linear constraint of P.

source
Base.isemptyMethod
isempty(P::AbstractPolytope)

Determine whether a polytope is empty.

Input

  • P – abstract polytope

Output

true if the given polytope contains no vertices, and false otherwise.

Algorithm

This algorithm checks whether the vertices_list of the given polytope is empty or not.

source
LazySets.volumeMethod
volume(P::AbstractPolytope; backend=default_polyhedra_backend(P))

Compute the volume of a polytope.

Input

  • P – polytope
  • backend – (optional, default: default_polyhedra_backend(P)) the backend for polyhedral computations; see Polyhedra's documentation for further information

Output

The volume of P.

Algorithm

The volume is computed by the Polyhedra library.

source

Implementations

Polygons (AbstractPolygon)

A polygon is a two-dimensional polytope.

LazySets.AbstractPolygonType
AbstractPolygon{N} <: AbstractPolytope{N}

Abstract type for polygons (i.e., 2D polytopes).

Notes

Every concrete AbstractPolygon must define the following functions:

  • tovrep(::AbstractPolygon{N}) – transform into V-representation
  • tohrep(::AbstractPolygon{N}) where {S<:AbstractHPolygon{N}} – transform into H-representation
julia> subtypes(AbstractPolygon)
2-element Vector{Any}:
 AbstractHPolygon
 VPolygon
source

This interface defines the following functions:

LazySets.dimMethod
dim(P::AbstractPolygon)

Return the ambient dimension of a polygon.

Input

  • P – polygon

Output

The ambient dimension of the polygon, which is 2.

source

The following helper functions are used for sorting directions:

LazySets.jump2piFunction
jump2pi(x::N) where {N<:AbstractFloat}

Return $x + 2π$ if $x$ is negative, otherwise return $x$.

Input

  • x – real scalar

Output

$x + 2π$ if $x$ is negative, $x$ otherwise.

Examples

julia> using LazySets: jump2pi

julia> jump2pi(0.0)
0.0

julia> jump2pi(-0.5)
5.783185307179586

julia> jump2pi(0.5)
0.5
source
Base.:<=Method
<=(u::AbstractVector, v::AbstractVector)

Compare two 2D vectors by their direction.

Input

  • u – first 2D direction
  • v – second 2D direction

Output

true iff $\arg(u) [2π] ≤ \arg(v) [2π]$.

Notes

The argument is measured in counter-clockwise fashion, with the 0 being the direction (1, 0).

Algorithm

The implementation checks the quadrant of each direction, and compares directions using the right-hand rule. In particular, this method does not use the arctangent.

source
LazySets._leq_trigMethod
_leq_trig(u::AbstractVector{N}, v::AbstractVector{N}) where {N<:AbstractFloat}

Compares two 2D vectors by their direction.

Input

  • u – first 2D direction
  • v – second 2D direction

Output

true iff $\arg(u) [2π] ≤ \arg(v) [2π]$.

Notes

The argument is measured in counter-clockwise fashion, with the 0 being the direction (1, 0).

Algorithm

The implementation uses the arctangent function with sign, atan, which for two arguments implements the atan2 function.

source
LazySets.quadrantMethod
quadrant(w::AbstractVector{N}) where {N}

Compute the quadrant where the direction w belongs.

Input

  • w – direction

Output

An integer from 0 to 3, with the following convention:

     ^
   1 | 0
  ---+-->
   2 | 3

Algorithm

The idea is to encode the following logic function: $11 ↦ 0, 01 ↦ 1, 00 ↦ 2, 10 ↦ 3$, according to the convention of above.

This function is inspired from AGPX's answer in: Sort points in clockwise order?

source

Implementations

Polygons in constraint representation (AbstractHPolygon)

An HPolygon is a polygon in H-representation (or constraint representation).

LazySets.AbstractHPolygonType
AbstractHPolygon{N} <: AbstractPolygon{N}

Abstract type for polygons in H-representation (i.e., constraints).

Notes

All subtypes must satisfy the invariant that constraints are sorted counter-clockwise.

Every concrete AbstractHPolygon must have the following fields:

  • constraints::Vector{LinearConstraint{N, AbstractVector{N}}} – the constraints

New subtypes should be added to the convert method in order to be convertible.

julia> subtypes(AbstractHPolygon)
2-element Vector{Any}:
 HPolygon
 HPolygonOpt
source

This interface defines the following functions:

LazySets.an_elementMethod
an_element(P::AbstractHPolygon)

Return some element of a polygon in constraint representation.

Input

  • P – polygon in constraint representation

Output

A vertex of the polygon in constraint representation (the first one in the order of the constraints).

source
Base.:∈Method
∈(x::AbstractVector, P::AbstractHPolygon)

Check whether a given 2D point is contained in a polygon in constraint representation.

Input

  • x – two-dimensional point/vector
  • P – polygon in constraint representation

Output

true iff $x ∈ P$.

Algorithm

This implementation checks if the point lies on the outside of each edge.

source
Base.randMethod
rand(::Type{HPOLYGON}; [N]::Type=Float64, [dim]::Int=2,
     [rng]::AbstractRNG=GLOBAL_RNG, [seed]::Union{Int, Nothing}=nothing,
     [num_constraints]::Int=-1) where {HPOLYGON<:AbstractHPolygon}

Create a random polygon in constraint representation.

Input

  • HPOLYGON – type for dispatch
  • N – (optional, default: Float64) numeric type
  • dim – (optional, default: 2) dimension
  • rng – (optional, default: GLOBAL_RNG) random number generator
  • seed – (optional, default: nothing) seed for reseeding
  • num_constraints – (optional, default: -1) number of constraints of the polygon (must be 3 or bigger; see comment below)

Output

A random polygon in constraint representation.

Algorithm

We create a random polygon in vertex representation and convert it to constraint representation. See rand(::Type{VPolygon}). For non-flat polygons the number of vertices and the number of constraints are identical.

source
LazySets.tohrepMethod
tohrep(P::HPOLYGON) where {HPOLYGON<:AbstractHPolygon}

Build a contraint representation of the given polygon.

Input

  • P – polygon in constraint representation

Output

The identity, i.e., the same polygon instance.

source
LazySets.tovrepMethod
tovrep(P::AbstractHPolygon)

Build a vertex representation of the given polygon.

Input

  • P – polygon in constraint representation

Output

The same polygon but in vertex representation, a VPolygon.

source
LazySets.addconstraint!Method
addconstraint!(P::AbstractHPolygon,
               constraint::LinearConstraint;
               [linear_search]::Bool=(length(P.constraints) <
                                      BINARY_SEARCH_THRESHOLD),
               [prune]::Bool=true)

Add a linear constraint to a polygon in constraint representation, keeping the constraints sorted by their normal directions.

Input

  • P – polygon in constraint representation
  • constraint – linear constraint to add
  • linear_search – (optional, default: length(constraints) < BINARY_SEARCH_THRESHOLD) flag to choose between linear and binary search
  • prune – (optional, default: true) flag for removing redundant constraints in the end
source
LazySets.addconstraint!Method
addconstraint!(constraints::Vector{LC},
               new_constraint::LinearConstraint;
               [linear_search]::Bool=(length(P.constraints) <
                                      BINARY_SEARCH_THRESHOLD),
               [prune]::Bool=true
              ) where {LC<:LinearConstraint}

Add a linear constraint to a sorted vector of constrains, keeping the constraints sorted by their normal directions.

Input

  • constraints – vector of linear constraintspolygon in constraint representation
  • new_constraint – linear constraint to add
  • linear_search – (optional, default: length(constraints) < BINARY_SEARCH_THRESHOLD) flag to choose between linear and binary search
  • prune – (optional, default: true) flag for removing redundant constraints in the end

Algorithm

If prune is active, we check if the new constraint is redundant. If the constraint is not redundant, we perform the same check to the left and to the right until we find the first constraint that is not redundant.

source
LinearAlgebra.normalizeMethod
normalize(P::AbstractHPolygon{N}, p=N(2)) where {N}

Normalize a polygon in constraint representation.

Input

  • P – polygon in constraint representation
  • p – (optional, default: 2) norm

Output

A new polygon in constraint representation whose normal directions $a_i$ are normalized, i.e., such that $‖a_i‖_p = 1$ holds.

source
LazySets.isredundantMethod
isredundant(cmid::LinearConstraint, cright::LinearConstraint,
            cleft::LinearConstraint)

Check whether a linear constraint is redundant wrt. two surrounding constraints.

Input

  • cmid – linear constraint of concern
  • cright – linear constraint to the right (clockwise turn)
  • cleft – linear constraint to the left (counter-clockwise turn)

Output

true iff the constraint is redundant.

Algorithm

We first check whether the angle between the surrounding constraints is < 180°, which is a necessary condition (unless the direction is identical to one of the other two constraints). If so, we next check if the angle is 0°, in which case the constraint cmid is redundant unless it is strictly tighter than the other two constraints. If the angle is strictly between 0° and 180°, the constraint cmid is redundant if and only if the vertex defined by the other two constraints lies inside the set defined by cmid.

source
LazySets.remove_redundant_constraints!Method
remove_redundant_constraints!(P::AbstractHPolygon)

Remove all redundant constraints of a polygon in constraint representation.

Input

  • P – polygon in constraint representation

Output

The same polygon with all redundant constraints removed.

Notes

Since we only consider bounded polygons and a polygon needs at least three constraints to be bounded, we stop removing redundant constraints if there are three or less constraints left. This means that for non-bounded polygons the result may be unexpected.

Algorithm

We go through all consecutive triples of constraints and check if the one in the middle is redundant. For this we assume that the constraints are sorted.

source
LazySets.constraints_listMethod
constraints_list(P::AbstractHPolygon)

Return the list of constraints defining a polygon in H-representation.

Input

  • P – polygon in H-representation

Output

The list of constraints of the polygon. The implementation guarantees that the constraints are sorted counter-clockwise.

source
LazySets.vertices_listMethod
vertices_list(P::AbstractHPolygon{N};
              apply_convex_hull::Bool=true,
              check_feasibility::Bool=true) where {N}

Return the list of vertices of a polygon in constraint representation.

Input

  • P – polygon in constraint representation
  • apply_convex_hull – (optional, default: true) flag to post-process the intersection of constraints with a convex hull
  • check_feasibility – (optional, default: true) flag to check whether the polygon was empty (required for correctness in case of empty polygons)

Output

List of vertices.

Algorithm

We compute each vertex as the intersection of consecutive lines defined by the half-spaces. If check_feasibility is active, we then check if the constraints of the polygon were actually feasible (i.e., they pointed in the right direction). For this we compute the average of all vertices and check membership in each constraint.

source
vertices_list(B::Ball1{N, VN}) where {N, VN<:AbstractVector}

Return the list of vertices of a ball in the 1-norm.

Input

  • B – ball in the 1-norm

Output

A list containing the vertices of the ball in the 1-norm.

source
vertices_list(∅::EmptySet{N}) where {N}

Return the list of vertices of an empty set.

Input

  • – empty set

Output

The empty list of vertices, as the empty set does not contain any vertices.

source
vertices_list(P::HPolytope{N};
              [backend]=nothing, [prune]::Bool=true) where {N}

Return the list of vertices of a polytope in constraint representation.

Input

  • P – polytope in constraint representation
  • backend – (optional, default: nothing) the polyhedral computations backend
  • prune – (optional, default: true) flag to remove redundant vertices

Output

List of vertices.

Algorithm

If the polytope is two-dimensional, the polytope is converted to a polygon in H-representation and then its vertices_list function is used. This ensures that, by default, the optimized two-dimensional methods are used.

It is possible to use the Polyhedra backend in two-dimensions as well by passing, e.g. backend=CDDLib.Library().

If the polytope is not two-dimensional, the concrete polyhedra manipulation library Polyhedra is used. The actual computation is performed by a given backend; for the default backend used in LazySets see default_polyhedra_backend(P). For further information on the supported backends see Polyhedra's documentation.

source
vertices_list(cp::CartesianProduct{N}) where {N}

Return the list of vertices of a (polytopic) Cartesian product.

Input

  • cp – Cartesian product

Output

A list of vertices.

Algorithm

We assume that the underlying sets are polytopic. Then the high-dimensional set of vertices is just the Cartesian product of the low-dimensional sets of vertices.

source

vertices_list(cpa::CartesianProductArray{N}) where {N}

Return the list of vertices of a (polytopic) Cartesian product of a finite number of sets.

Input

  • cpa – Cartesian product array

Output

A list of vertices.

Algorithm

We assume that the underlying sets are polytopic. Then the high-dimensional set of vertices is just the Cartesian product of the low-dimensional sets of vertices.

source
vertices_list(em::ExponentialMap{N};
              [backend]=get_exponential_backend()) where {N}

Return the list of vertices of a (polytopic) exponential map.

Input

  • em – exponential map
  • backend – (optional; default: get_exponential_backend()) exponentiation backend

Output

A list of vertices.

Algorithm

We assume that the underlying set X is polytopic. Then the result is just the exponential map applied to the vertices of X.

source
LazySets.isboundedFunction
isbounded(P::AbstractHPolygon, [use_type_assumption]::Bool=true)

Determine whether a polygon in constraint representation is bounded.

Input

  • P – polygon in constraint representation
  • use_type_assumption – (optional, default: true) flag for ignoring the type assumption that polygons are bounded

Output

true if use_type_assumption is activated. Otherwise, true iff P is bounded.

Algorithm

If !use_type_assumption, we convert P to an HPolyhedron P2 and then use isbounded(P2).

source

Implementations

Centrally symmetric polytopes (AbstractCentrallySymmetricPolytope)

A centrally symmetric polytope is a combination of two other interfaces: Centrally symmetric sets and Polytope.

LazySets.AbstractCentrallySymmetricPolytopeType
AbstractCentrallySymmetricPolytope{N} <: AbstractPolytope{N}

Abstract type for centrally symmetric, polytopic sets. It combines the AbstractCentrallySymmetric and AbstractPolytope interfaces. Such a type combination is necessary as long as Julia does not support multiple inheritance.

Notes

Every concrete AbstractCentrallySymmetricPolytope must define the following functions:

  • from AbstractCentrallySymmetric:
    • center(::AbstractCentrallySymmetricPolytope) – return the center point
  • from AbstractPolytope:
    • vertices_list(::AbstractCentrallySymmetricPolytope) – return a list of all vertices
julia> subtypes(AbstractCentrallySymmetricPolytope)
2-element Vector{Any}:
 AbstractZonotope
 Ball1
source

This interface defines the following functions:

LazySets.dimMethod
dim(P::AbstractCentrallySymmetricPolytope)

Return the ambient dimension of a centrally symmetric, polytopic set.

Input

  • P – centrally symmetric, polytopic set

Output

The ambient dimension of the polytopic set.

source
LazySets.an_elementMethod
an_element(P::AbstractCentrallySymmetricPolytope)

Return some element of a centrally symmetric polytope.

Input

  • P – centrally symmetric polytope

Output

The center of the centrally symmetric polytope.

source
Base.isemptyMethod
isempty(P::AbstractCentrallySymmetricPolytope)

Return if a centrally symmetric, polytopic set is empty or not.

Input

  • P – centrally symmetric, polytopic set

Output

false.

source
LazySets.isuniversalMethod
isuniversal(S::AbstractCentrallySymmetricPolytope{N},
            [witness]::Bool=false) where {N}

Check whether a centrally symmetric polytope is universal.

Input

  • S – centrally symmetric polytope
  • witness – (optional, default: false) compute a witness if activated

Output

  • If witness option is deactivated: false
  • If witness option is activated: (false, v) where $v ∉ S$

Algorithm

A witness is obtained by computing the support vector in direction d = [1, 0, …, 0] and adding d on top.

source
LazySets.centerMethod
center(S::AbstractCentrallySymmetricPolytope, i::Int)

Return the center along a given dimension of a centrally symmetric polytope.

Input

  • S – centrally symmetric polytope
  • i – dimension of interest

Output

The center along a given dimension of the centrally symmetric polytope.

source

Implementations

Zonotopes (AbstractZonotope)

A zonotope is a specific centrally symmetric polytope characterized by a center and a collection of generators.

LazySets.AbstractZonotopeType
AbstractZonotope{N} <: AbstractCentrallySymmetricPolytope{N}

Abstract type for zonotopic sets.

Notes

Mathematically, a zonotope is defined as the set

\[Z = \left\{ c + ∑_{i=1}^p ξ_i g_i,~~ ξ_i \in [-1, 1]~~ ∀ i = 1,…, p \right\},\]

where $c \in \mathbb{R}^n$ is its center and $\{g_i\}_{i=1}^p$, $g_i \in \mathbb{R}^n$, is the set of generators. This characterization defines a zonotope as the finite Minkowski sum of line segments. Zonotopes can be equivalently described as the image of a unit infinity-norm ball in $\mathbb{R}^n$ by an affine transformation.

See Zonotope for a standard implementation of this interface.

Every concrete AbstractZonotope must define the following functions:

  • genmat(::AbstractZonotope{N}) – return the generator matrix
  • generators(::AbstractZonotope{N}) – return an iterator over the generators

Since the functions genmat and generators can be defined in terms of each other, it is sufficient to only genuinely implement one of them and let the implementation of the other function call the fallback implementation genmat_fallback resp. generators_fallback.

julia> subtypes(AbstractZonotope)
5-element Vector{Any}:
 AbstractHyperrectangle
 HParallelotope
 LineSegment
 RotatedHyperrectangle
 Zonotope
source

This interface defines the following functions:

LazySets.ngensMethod
ngens(Z::AbstractZonotope)

Return the number of generators of a zonotopic set.

Input

  • Z – zonotopic set

Output

An integer representing the number of generators.

source
LazySets.genmat_fallbackMethod
genmat_fallback(Z::AbstractZonotope{N};
                [gens]=generators(Z),
                [ngens]=nothing) where {N}

Fallback definition of genmat for zonotopic sets.

Input

  • Z – zonotopic set
  • gens – (optional; default: generators(Z)) iterator over generators
  • ngens – (optional; default: nothing) number of generators or nothing if unknown

Output

A matrix where each column represents one generator of Z.

Notes

Passing the number of generators is much more efficient as otherwise the generators have to be obtained from the iterator (gens) and stored in an intermediate vector until the final result matrix can be allocated.

source
LazySets.generators_fallbackMethod
generators_fallback(Z::AbstractZonotope)

Fallback definition of generators for zonotopic sets.

Input

  • Z – zonotopic set

Output

An iterator over the generators of Z.

source
LazySets.ρMethod
ρ(d::AbstractVector, Z::AbstractZonotope)

Return the support function of a zonotopic set in a given direction.

Input

  • d – direction
  • Z – zonotopic set

Output

The support function of the zonotopic set in the given direction.

Algorithm

The support value is $cᵀ d + ‖Gᵀ d‖₁$ where $c$ is the center and $G$ is the generator matrix of Z.

source
LazySets.σMethod
σ(d::AbstractVector, Z::AbstractZonotope)

Return the support vector of a zonotopic set in a given direction.

Input

  • d – direction
  • Z – zonotopic set

Output

A support vector in the given direction. If the direction has norm zero, the vertex with $ξ_i = 1 \ \ ∀ i = 1,…, p$ is returned.

source
Base.:∈Method
∈(x::AbstractVector, Z::AbstractZonotope; solver=nothing)

Check whether a given point is contained in a zonotopic set.

Input

  • x – point/vector
  • Z – zonotopic set
  • solver – (optional, default: nothing) the backend used to solve the linear program

Output

true iff $x ∈ Z$.

Examples

julia> Z = Zonotope([1.0, 0.0], [0.1 0.0; 0.0 0.1]);

julia> [1.0, 0.2] ∈ Z
false
julia> [1.0, 0.1] ∈ Z
true

Notes

If solver == nothing, we fall back to default_lp_solver(N).

Algorithm

The membership problem is computed by stating and solving the following linear program. Let $p$ and $n$ be the number of generators and ambient dimension, respectively. We consider the minimization of $x_0$ in the $p+1$-dimensional space of elements $(x_0, ξ_1, …, ξ_p)$ constrained to $0 ≤ x_0 ≤ ∞$, $ξ_i ∈ [-1, 1]$ for all $i = 1, …, p$, and such that $x-c = Gξ$ holds. If a feasible solution exists, the optimal value $x_0 = 0$ is achieved.

source
LazySets.linear_mapMethod
linear_map(M::AbstractMatrix, Z::AbstractZonotope)

Concrete linear map of a zonotopic set.

Input

  • M – matrix
  • Z – zonotopic set

Output

The zonotope obtained by applying the linear map to the center and generators of $Z$.

source
LazySets.translate!Method
translate!(Z::AbstractZonotope, v::AbstractVector)

Translate (i.e., shift) a zonotope by a given vector in-place.

Input

  • Z – zonotope
  • v – translation vector

Output

A translated zonotope.

Notes

See also translate(Z::AbstractZonotope, v::AbstractVector) for the out-of-place version.

Algorithm

We add the translation vector to the center of the zonotope.

source
Base.splitMethod
split(Z::AbstractZonotope, j::Int)

Return two zonotopes obtained by splitting the given zonotope.

Input

  • Z – zonotope
  • j – index of the generator to be split

Output

The zonotope obtained by splitting Z into two zonotopes such that their union is Z and their intersection is possibly non-empty.

Algorithm

This function implements [Prop. 3, 1], that we state next. The zonotope $Z = ⟨c, g^{(1, …, p)}⟩$ is split into:

\[Z₁ = ⟨c - \frac{1}{2}g^{(j)}, (g^{(1, …,j-1)}, \frac{1}{2}g^{(j)}, g^{(j+1, …, p)})⟩ \\ Z₂ = ⟨c + \frac{1}{2}g^{(j)}, (g^{(1, …,j-1)}, \frac{1}{2}g^{(j)}, g^{(j+1, …, p)})⟩,\]

such that $Z₁ ∪ Z₂ = Z$ and $Z₁ ∩ Z₂ = Z^*$, where

\[Z^* = ⟨c, (g^{(1,…,j-1)}, g^{(j+1,…, p)})⟩.\]

[1] Althoff, M., Stursberg, O., & Buss, M. (2008). Reachability analysis of nonlinear systems with uncertain parameters using conservative linearization. In Proc. of the 47th IEEE Conference on Decision and Control.

source
Base.splitMethod
split(Z::AbstractZonotope, gens::AbstractVector{Int}, nparts::AbstractVector{Int})

Split a zonotope along the given generators into a vector of zonotopes.

Input

  • Z – zonotope
  • gens – vector of indices of the generators to be split
  • n – vector of integers describing the number of partitions in the corresponding generator

Output

The zonotopes obtained by splitting Z into 2^{n_i} zonotopes for each generator i such that their union is Z and their intersection is possibly non-empty.

Examples

Splitting of a two-dimensional zonotope along its first generator:

julia> Z = Zonotope([1.0, 0.0], [0.1 0.0; 0.0 0.1])
Zonotope{Float64, Vector{Float64}, Matrix{Float64}}([1.0, 0.0], [0.1 0.0; 0.0 0.1])

julia> split(Z, [1], [1])
2-element Vector{Zonotope{Float64, Vector{Float64}, Matrix{Float64}}}:
 Zonotope{Float64, Vector{Float64}, Matrix{Float64}}([0.95, 0.0], [0.05 0.0; 0.0 0.1])
 Zonotope{Float64, Vector{Float64}, Matrix{Float64}}([1.05, 0.0], [0.05 0.0; 0.0 0.1])

Here, the first vector in the arguments corresponds to the zonotope's generator to be split, and the second vector corresponds to the exponent of 2^n parts that the zonotope will be split into along the corresponding generator.

Splitting of a two-dimensional zonotope along its generators:

julia> Z = Zonotope([1.0, 0.0], [0.1 0.0; 0.0 0.1])
Zonotope{Float64, Vector{Float64}, Matrix{Float64}}([1.0, 0.0], [0.1 0.0; 0.0 0.1])

julia> split(Z, [1, 2], [2, 2])
16-element Vector{Zonotope{Float64, Vector{Float64}, Matrix{Float64}}}:
Zonotope{Float64, Vector{Float64}, Matrix{Float64}}([0.925, -0.075], [0.025 0.0; 0.0 0.025])
Zonotope{Float64, Vector{Float64}, Matrix{Float64}}([0.925, -0.025], [0.025 0.0; 0.0 0.025])
Zonotope{Float64, Vector{Float64}, Matrix{Float64}}([0.925, 0.025], [0.025 0.0; 0.0 0.025])
Zonotope{Float64, Vector{Float64}, Matrix{Float64}}([0.925, 0.075], [0.025 0.0; 0.0 0.025])
Zonotope{Float64, Vector{Float64}, Matrix{Float64}}([0.975, -0.075], [0.025 0.0; 0.0 0.025])
Zonotope{Float64, Vector{Float64}, Matrix{Float64}}([0.975, -0.025], [0.025 0.0; 0.0 0.025])
Zonotope{Float64, Vector{Float64}, Matrix{Float64}}([0.975, 0.025], [0.025 0.0; 0.0 0.025])
Zonotope{Float64, Vector{Float64}, Matrix{Float64}}([0.975, 0.075], [0.025 0.0; 0.0 0.025])
Zonotope{Float64, Vector{Float64}, Matrix{Float64}}([1.025, -0.075], [0.025 0.0; 0.0 0.025])
Zonotope{Float64, Vector{Float64}, Matrix{Float64}}([1.025, -0.025], [0.025 0.0; 0.0 0.025])
Zonotope{Float64, Vector{Float64}, Matrix{Float64}}([1.025, 0.025], [0.025 0.0; 0.0 0.025])
Zonotope{Float64, Vector{Float64}, Matrix{Float64}}([1.025, 0.075], [0.025 0.0; 0.0 0.025])
Zonotope{Float64, Vector{Float64}, Matrix{Float64}}([1.075, -0.075], [0.025 0.0; 0.0 0.025])
Zonotope{Float64, Vector{Float64}, Matrix{Float64}}([1.075, -0.025], [0.025 0.0; 0.0 0.025])
Zonotope{Float64, Vector{Float64}, Matrix{Float64}}([1.075, 0.025], [0.025 0.0; 0.0 0.025])
Zonotope{Float64, Vector{Float64}, Matrix{Float64}}([1.075, 0.075], [0.025 0.0; 0.0 0.025])

Here the zonotope is split along both of its generators, each time into four parts.

source
LazySets.constraints_listMethod
constraints_list(P::AbstractZonotope)

Return the list of constraints defining a zonotopic set.

Input

  • Z – zonotopic set

Output

The list of constraints of the zonotopic set.

Algorithm

This is the (inefficient) fallback implementation for rational numbers. It first computes the vertices and then converts the corresponding polytope to constraint representation.

source
LazySets.constraints_listMethod
constraints_list(Z::AbstractZonotope{N}) where {N<:AbstractFloat}

Return the list of constraints defining a zonotopic set.

Input

  • Z – zonotopic set

Output

The list of constraints of the zonotopic set.

Notes

The algorithm assumes that no generator is redundant. The result has $2 \binom{p}{n-1}$ (with $p$ being the number of generators and $n$ being the ambient dimension) constraints, which is optimal under this assumption.

If $p < n$ or the generator matrix is not full rank, we fall back to the (slower) computation based on the vertex representation.

Algorithm

We follow the algorithm presented in Althoff, Stursberg, Buss: Computing Reachable Sets of Hybrid Systems Using a Combination of Zonotopes and Polytopes. 2009.

The one-dimensional case is not covered by that algorithm; we manually handle this case.

source
LazySets.vertices_listMethod
vertices_list(Z::AbstractZonotope; [apply_convex_hull]::Bool=true)

Return the vertices of a zonotopic set.

Input

  • Z – zonotopic set
  • apply_convex_hull – (optional, default: true) if true, post-process the computation with the convex hull of the points

Output

List of vertices as a vector of vectors.

Algorithm

Two-dimensional case

We use a trick to speed up enumerating vertices of 2-dimensional zonotopic sets with all generators in the first quadrant or third quadrant (same sign). Namely, sort the generators in angle and add them clockwise in increasing order and anticlockwise in decreasing order, the algorithm detail: https://math.stackexchange.com/q/3356460

To avoid cumulative sum from both directions separately, we build a 2d index matrix to sum generators for both directions in one matrix-vector product.

General case

If the zonotopic set has $p$ generators, each vertex is the result of summing the center with some linear combination of generators, where the combination factors are $ξ_i ∈ \{-1, 1\}$.

There are at most $2^p$ distinct vertices. Use the flag apply_convex_hull to control whether a convex hull algorithm is applied to the vertices computed by this method; otherwise, redundant vertices may be present.

source
LazySets.orderMethod
order(Z::AbstractZonotope)

Return the order of a zonotope.

Input

  • Z – zonotope

Output

A rational number representing the order of the zonotope.

Notes

The order of a zonotope is defined as the quotient of its number of generators and its dimension.

source
LazySets.togrepMethod
togrep(Z::AbstractZonotope)

Return a generator representation of a zonotopic set.

Input

  • Z – zonotopic set

Output

The same set in generator representation. This fallback implementation returns a Zonotope; however, more specific implementations may return other generator representations.

source
LazySets.remove_redundant_generatorsMethod
remove_redundant_generators(Z::AbstractZonotope)

Remove all redundant (pairwise linearly dependent) generators of a zonotope.

Input

  • Z – zonotope

Output

A new zonotope with fewer generators, or the same zonotope if no generator could be removed.

Algorithm

By default this function returns the input zonotope. Subtypes of AbstractZonotope where generators can be removed have to define a new method.

source
LazySets.reduce_orderMethod
reduce_order(Z::AbstractZonotope, r::Number, method::AbstractReductionMethod=GIR05())

Reduce the order of a zonotope by overapproximating with a zonotope with fewer generators.

Input

  • Z – zonotope
  • r – desired order
  • method – (optional, default: GIR05()) the reduction method used

Output

A new zonotope with fewer generators, if possible.

Algorithm

The available algorithms are:

julia> subtypes(AbstractReductionMethod)
2-element Vector{Any}:
 LazySets.COMB03
 LazySets.GIR05

See the documentation of each algorithm for references. These methods split the given zonotope Z into two zonotopes, K and L, where K contains the most "representative" generators and L contains the generators that are reduced, Lred, using a box overapproximation. We follow the notation from [YS18]. See also [KSA17].

  • [KSA17] Kopetzki, A. K., Schürmann, B., & Althoff, M. (2017). Methods for order reduction of zonotopes. In 2017 IEEE 56th Annual CDC (pp. 5626-5633). IEEE.
  • [YS18] Yang, X., & Scott, J. K. (2018). A comparison of zonotope order reduction techniques. Automatica, 95, 378-384.
source

Implementations

Hyperrectangles (AbstractHyperrectangle)

A hyperrectangle is a special centrally symmetric polytope with axis-aligned facets.

LazySets.AbstractHyperrectangleType
AbstractHyperrectangle{N} <: AbstractZonotope{N}

Abstract type for hyperrectangular sets.

Notes

See Hyperrectangle for a standard implementation of this interface.

Every concrete AbstractHyperrectangle must define the following functions:

  • radius_hyperrectangle(::AbstractHyperrectangle) – return the hyperrectangle's radius, which is a full-dimensional vector

  • radius_hyperrectangle(::AbstractHyperrectangle, i::Int) – return the hyperrectangle's radius in the i-th dimension

  • isflat(::AbstractHyperrectangle) – determine whether the hyperrectangle's radius is zero in some dimension

Every hyperrectangular set is also a zonotopic set; see AbstractZonotope.

julia> subtypes(AbstractHyperrectangle)
5-element Vector{Any}:
 AbstractSingleton
 BallInf
 Hyperrectangle
 Interval
 SymmetricIntervalHull
source

This interface defines the following functions:

LinearAlgebra.normFunction
norm(H::AbstractHyperrectangle, [p]::Real=Inf)

Return the norm of a hyperrectangular set.

The norm of a hyperrectangular set is defined as the norm of the enclosing ball, of the given $p$-norm, of minimal volume that is centered in the origin.

Input

  • H – hyperrectangular set
  • p – (optional, default: Inf) norm

Output

A real number representing the norm.

Algorithm

Recall that the norm is defined as

\[‖ X ‖ = \max_{x ∈ X} ‖ x ‖_p = max_{x ∈ \text{vertices}(X)} ‖ x ‖_p.\]

The last equality holds because the optimum of a convex function over a polytope is attained at one of its vertices.

This implementation uses the fact that the maximum is achieved in the vertex $c + \text{diag}(\text{sign}(c)) r$, for any $p$-norm, hence it suffices to take the $p$-norm of this particular vertex. This statement is proved below. Note that, in particular, there is no need to compute the $p$-norm for each vertex, which can be very expensive.

If $X$ is an axis-aligned hyperrectangle and the $n$-dimensional vectors center and radius of the hyperrectangle are denoted $c$ and $r$ respectively, then reasoning on the $2^n$ vertices we have that:

\[\max_{x ∈ \text{vertices}(X)} ‖ x ‖_p = \max_{α_1, …, α_n ∈ \{-1, 1\}} (|c_1 + α_1 r_1|^p + ... + |c_n + α_n r_n|^p)^{1/p}.\]

The function $x ↦ x^p$, $p > 0$, is monotonically increasing and thus the maximum of each term $|c_i + α_i r_i|^p$ is given by $|c_i + \text{sign}(c_i) r_i|^p$ for each $i$. Hence, $x^* := \text{argmax}_{x ∈ X} ‖ x ‖_p$ is the vertex $c + \text{diag}(\text{sign}(c)) r$.

source
IntervalArithmetic.radiusFunction
radius(H::AbstractHyperrectangle, [p]::Real=Inf)

Return the radius of a hyperrectangular set.

Input

  • H – hyperrectangular set
  • p – (optional, default: Inf) norm

Output

A real number representing the radius.

Notes

The radius is defined as the radius of the enclosing ball of the given $p$-norm of minimal volume with the same center. It is the same for all corners of a hyperrectangular set.

source
LazySets.σMethod
σ(d::AbstractVector, H::AbstractHyperrectangle)

Return the support vector of a hyperrectangular set in a given direction.

Input

  • d – direction
  • H – hyperrectangular set

Output

The support vector in the given direction. If the direction has norm zero, the vertex with biggest values is returned.

source
LazySets.ρMethod
ρ(d::AbstractVector, H::AbstractHyperrectangle)

Evaluate the support function of a hyperrectangular set in a given direction.

Input

  • d – direction
  • H – hyperrectangular set

Output

Evaluation of the support function in the given direction.

source
Base.:∈Method
∈(x::AbstractVector, H::AbstractHyperrectangle)

Check whether a given point is contained in a hyperrectangular set.

Input

  • x – point/vector
  • H – hyperrectangular set

Output

true iff $x ∈ H$.

Algorithm

Let $H$ be an $n$-dimensional hyperrectangular set, $c_i$ and $r_i$ be the box's center and radius and $x_i$ be the vector $x$ in dimension $i$, respectively. Then $x ∈ H$ iff $|c_i - x_i| ≤ r_i$ for all $i=1,…,n$.

source
LazySets.vertices_listMethod
vertices_list(H::AbstractHyperrectangle)

Return the list of vertices of a hyperrectangular set.

Input

  • H – hyperrectangular set

Output

A list of vertices. Zeros in the radius are correctly handled, i.e., the result does not contain any duplicate vertices.

Notes

For high dimensions, it is preferable to develop a vertex_iterator approach.

Algorithm

First we identify the dimensions where H is flat, i.e., its radius is zero. We also compute the number of vertices that we have to create.

Next we create the vertices. We do this by enumerating all vectors v of length n (the dimension of H) with entries -1/0/1 and construct the corresponding vertex as follows:

\[ \text{vertex}(v)(i) = \begin{cases} c(i) + r(i) & v(i) = 1 \\ c(i) & v(i) = 0 \\ c(i) - r(i) & v(i) = -1. \end{cases}\]

For enumerating the vectors v, we modify the current v from left to right by changing entries -1 to 1, skipping entries 0, and stopping at the first entry 1 (but changing it to -1). This way we only need to change the vertex in those dimensions where v has changed, which usually is a smaller number than n.

source
LazySets.constraints_listMethod
constraints_list(H::AbstractHyperrectangle{N}) where {N}

Return the list of constraints of an axis-aligned hyperrectangular set.

Input

  • H – hyperrectangular set

Output

A list of linear constraints.

source
constraints_list(P::Ball1{N}) where {N}

Return the list of constraints defining a ball in the 1-norm.

Input

  • B – ball in the 1-norm

Output

The list of constraints of the ball.

Algorithm

The constraints can be defined as $d_i^T (x-c) ≤ r$ for all $d_i$, where $d_i$ is a vector with elements $1$ or $-1$ in $n$ dimensions. To span all possible $d_i$, the function Iterators.product is used.

source
constraints_list(x::Interval{N}) where {N}

Return the list of constraints of the given interval.

Input

  • x – interval

Output

The list of constraints of the interval represented as two one-dimensional half-spaces.

source
constraints_list(L::Line{N, VN}) where {N, VN}

Return the list of constraints of a line.

Input

  • L – line

Output

A list containing 2n-2 half-spaces whose intersection is L, where n is the ambient dimension of L.

source
constraints_list(U::Universe{N}) where {N}

Return the list of constraints defining a universe.

Input

  • U – universe

Output

The empty list of constraints, as the universe is unconstrained.

source
constraints_list(P::HParallelotope{N, VN}) where {N, VN}

Return the list of constraints of the given parallelotope.

Input

  • P – parallelotope in constraint representation

Output

The list of constraints of P.

source

constraints_list(cpa::CartesianProductArray{N}) where {N}

Return the list of constraints of a (polyhedral) Cartesian product of a finite number of sets.

Input

  • cpa – Cartesian product array

Output

A list of constraints.

source
constraints_list(ia::IntersectionArray{N}) where {N}

Return the list of constraints of an intersection of a finite number of (polyhedral) sets.

Input

  • ia – intersection of a finite number of (polyhedral) sets

Output

The list of constraints of the intersection.

Notes

We assume that the underlying sets are polyhedral, i.e., offer a method constraints_list.

Algorithm

We create the polyhedron from the constraints_lists of the sets and remove redundant constraints.

source
constraints_list(rm::ResetMap{N}) where {N}

Return the list of constraints of a polytopic reset map.

Input

  • rm – reset map of a polytope

Output

The list of constraints of the reset map.

Notes

We assume that the underlying set X is a polytope, i.e., is bounded and offers a method constraints_list(X).

Algorithm

We fall back to constraints_list of a LinearMap of the A-matrix in the affine-map view of a reset map. Each reset dimension $i$ is projected to zero, expressed by two constraints for each reset dimension. Then it remains to shift these constraints to the new value.

For instance, if the dimension $5$ was reset to $4$, then there will be constraints $x₅ ≤ 0$ and $-x₅ ≤ 0$. We then modify the right-hand side of these constraints to $x₅ ≤ 4$ and $-x₅ ≤ -4$, respectively.

source
constraints_list(rm::ResetMap{N, S}) where {N, S<:AbstractHyperrectangle}

Return the list of constraints of a hyperrectangular reset map.

Input

  • rm – reset map of a hyperrectangular set

Output

The list of constraints of the reset map.

Algorithm

We iterate through all dimensions. If there is a reset, we construct the corresponding (flat) constraints. Otherwise, we construct the corresponding constraints of the underlying set.

source
LazySets.highMethod
high(H::AbstractHyperrectangle)

Return the higher coordinates of a hyperrectangular set.

Input

  • H – hyperrectangular set

Output

A vector with the higher coordinates of the hyperrectangular set.

source
LazySets.highMethod
high(H::AbstractHyperrectangle, i::Int)

Return the higher coordinate of a hyperrectangular set in a given dimension.

Input

  • H – hyperrectangular set
  • i – dimension of interest

Output

The higher coordinate of the hyperrectangular set in the given dimension.

source
LazySets.lowMethod
low(H::AbstractHyperrectangle)

Return the lower coordinates of a hyperrectangular set.

Input

  • H – hyperrectangular set

Output

A vector with the lower coordinates of the hyperrectangular set.

source
LazySets.lowMethod
low(H::AbstractHyperrectangle, i::Int)

Return the lower coordinate of a hyperrectangular set in a given dimension.

Input

  • H – hyperrectangular set
  • i – dimension of interest

Output

The lower coordinate of the hyperrectangular set in the given dimension.

source
LazySets.isflatMethod
isflat(H::AbstractHyperrectangle)

Determine whether a hyperrectangular set is flat, i.e. whether its radius is zero in some dimension.

Input

  • H – hyperrectangular set

Output

true iff the hyperrectangular set is flat.

Notes

For robustness with respect to floating-point inputs, this function relies on the result of isapproxzero when applied to the radius in some dimension. Hence, this function depends on the absolute zero tolerance ABSZTOL.

source
Base.splitMethod
split(H::AbstractHyperrectangle{N}, num_blocks::AbstractVector{Int}
     ) where {N}

Partition a hyperrectangular set into uniform sub-hyperrectangles.

Input

  • H – hyperrectangular set
  • num_blocks – number of blocks in the partition for each dimension

Output

A list of Hyperrectangles.

source
LazySets.generatorsMethod
generators(H::AbstractHyperrectangle)

Return an iterator over the generators of a hyperrectangular set.

Input

  • H – hyperrectangular set

Output

An iterator over the generators of H.

source
LazySets.genmatMethod

genmat(H::AbstractHyperrectangle)

Return the generator matrix of a hyperrectangular set.

Input

  • H – hyperrectangular set

Output

A matrix where each column represents one generator of H.

source
LazySets.ngensMethod
ngens(H::AbstractHyperrectangle{N}) where {N}

Return the number of generators of a hyperrectangular set.

Input

  • H – hyperrectangular set

Output

The number of generators.

Algorithm

A hyperrectangular set has one generator for each non-flat dimension.

source
ReachabilityBase.Arrays.rectifyMethod
rectify(H::AbstractHyperrectangle)

Concrete rectification of a hyperrectangular set.

Input

  • H – hyperrectangular set

Output

The Hyperrectangle that corresponds to the rectification of H.

source
LazySets.volumeMethod
volume(H::AbstractHyperrectangle)

Return the volume of a hyperrectangular set.

Input

  • H – hyperrectangular set

Output

The volume of $H$.

Algorithm

The volume of the $n$-dimensional hyperrectangle $H$ with vector radius $r$ is $2ⁿ ∏ᵢ rᵢ$ where $rᵢ$ denotes the $i$-th component of $r$.

source
ReachabilityBase.Arrays.distanceMethod
distance(x::AbstractVector, H::AbstractHyperrectangle{N};
         [p]::Real=N(2)) where {N}

Compute the distance between point x and hyperrectangle H with respect to the given p-norm.

Input

  • x – vector
  • H – hyperrectangle

Output

A scalar representing the distance between point x and hyperrectangle H.

source

Implementations

Concrete set representations:

Lazy set operations:

Singletons (AbstractSingleton)

A singleton is a special hyperrectangle consisting of only one point.

LazySets.AbstractSingletonType
AbstractSingleton{N} <: AbstractHyperrectangle{N}

Abstract type for sets with a single value.

Notes

Every concrete AbstractSingleton must define the following functions:

  • element(::AbstractSingleton{N}) – return the single element
  • element(::AbstractSingleton{N}, i::Int) – return the single element's entry in the i-th dimension
julia> subtypes(AbstractSingleton)
2-element Vector{Any}:
 Singleton
 ZeroSet
source

This interface defines the following functions:

LazySets.σMethod
σ(d::AbstractVector, S::AbstractSingleton)

Return the support vector of a set with a single value.

Input

  • d – direction
  • S – set with a single value

Output

The support vector, which is the set's vector itself, irrespective of the given direction.

source
LazySets.ρMethod
ρ(d::AbstractVector, S::AbstractSingleton)

Evaluate the support function of a set with a single value in a given direction.

Input

  • d – direction
  • S – set with a single value

Output

Evaluation of the support function in the given direction.

source
Base.:∈Method
∈(x::AbstractVector, S::AbstractSingleton)

Check whether a given point is contained in a set with a single value.

Input

  • x – point/vector
  • S – set with a single value

Output

true iff $x ∈ S$.

Notes

This implementation performs an exact comparison, which may be insufficient with floating point computations.

source
LazySets.centerMethod
center(S::AbstractSingleton)

Return the center of a set with a single value.

Input

  • S – set with a single value

Output

The only element of the set.

source
LazySets.verticesMethod
vertices(S::AbstractSingleton{N}) where {N}

Construct an iterator over the vertices of a set with a single value.

Input

  • S – set with a single value

Output

An iterator with a single value.

source
vertices(∅::EmptySet{N}) where {N}

Construct an iterator over the vertices of an empty set.

Input

  • – empty set

Output

The empty iterator, as the empty set does not contain any vertices.

source
LazySets.vertices_listMethod
vertices_list(S::AbstractSingleton)

Return the list of vertices of a set with a single value.

Input

  • S – set with a single value

Output

A list containing only a single vertex.

source
LazySets.radius_hyperrectangleMethod
radius_hyperrectangle(S::AbstractSingleton{N}) where {N}

Return the box radius of a set with a single value in every dimension.

Input

  • S – set with a single value

Output

The zero vector.

source
LazySets.radius_hyperrectangleMethod
radius_hyperrectangle(S::AbstractSingleton{N}, i::Int) where {N}

Return the box radius of a set with a single value in a given dimension.

Input

  • S – set with a single value
  • i – dimension of interest

Output

Zero.

source
LazySets.highMethod
high(S::AbstractSingleton)

Return the higher coordinates of a set with a single value.

Input

  • S – set with a single value

Output

A vector with the higher coordinates of the set with a single value.

source
LazySets.highMethod
high(S::AbstractSingleton, i::Int)

Return the higher coordinate of a set with a single value in the given dimension.

Input

  • S – set with a single value
  • i – dimension of interest

Output

The higher coordinate of the set with a single value in the given dimension.

source
LazySets.lowMethod
low(S::AbstractSingleton)

Return the lower coordinates of a set with a single value.

Input

  • S – set with a single value

Output

A vector with the lower coordinates of the set with a single value.

source
LazySets.lowMethod
low(S::AbstractSingleton, i::Int)

Return the lower coordinate of a set with a single value in the given dimension.

Input

  • S – set with a single value
  • i – dimension of interest

Output

The lower coordinate of the set with a single value in the given dimension.

source
LazySets.linear_mapMethod
linear_map(M::AbstractMatrix, S::AbstractSingleton)

Concrete linear map of an abstract singleton.

Input

  • M – matrix
  • S – abstract singleton

Output

The abstract singleton of the same type of $S$ obtained by applying the linear map to the element in $S$.

source
LazySets.generatorsMethod
generators(S::AbstractSingleton{N}) where {N}

Return an (empty) iterator over the generators of a set with a single value.

Input

  • S – set with a single value

Output

An empty iterator.

source
generators(L::LineSegment{N}) where {N}

Return an iterator over the (single) generator of a line segment.

Input

  • L – line segment

Output

A one-element iterator over the generator of L.

source
LazySets.genmatMethod

genmat(S::AbstractSingleton{N}) where {N}

Return the (empty) generator matrix of a set with a single value.

Input

  • S – set with a single value

Output

A matrix with no columns representing the generators of S.

source
LazySets.ngensMethod
ngens(S::AbstractSingleton)

Return the number of generators of a set with a single value.

Input

  • H – set with a single value

Output

The number of generators.

Algorithm

A set with a single value has no generators, so the result is $0$.

source
LazySets.plot_recipeMethod
plot_recipe(S::AbstractSingleton{N}, [ε]=zero(N)) where {N}

Convert a singleton to a pair (x, y) of points for plotting.

Input

  • S – singleton
  • ε – (optional, default: 0) ignored, used for dispatch

Output

A pair (x, y) of one point that can be plotted.

source
RecipesBase.apply_recipeMethod
plot_singleton(S::AbstractSingleton{N}, [ε]::N=zero(N); ...) where {N}

Plot a singleton.

Input

  • S – singleton
  • ε – (optional, default: 0) ignored, used for dispatch

Examples

julia> plot(Singleton([0.5, 1.0]))
source

Implementations

Affine maps (AbstractAffineMap)

An affine map consists of a linear map and a translation.

LazySets.AbstractAffineMapType
AbstractAffineMap{N, S<:ConvexSet{N}} <: ConvexSet{N}

Abstract type for affine maps.

Notes

See AffineMap for a standard implementation of this interface.

Every concrete AbstractAffineMap must define the following functions:

  • matrix(::AbstractAffineMap) – return the linear map
  • vector(::AbstractAffineMap) – return the affine translation vector
  • set(::AbstractAffineMap) – return the set that the map is applied to
julia> subtypes(AbstractAffineMap)
7-element Vector{Any}:
 AffineMap
 ExponentialMap
 ExponentialProjectionMap
 InverseLinearMap
 LinearMap
 ResetMap
 Translation
source

This interface defines the following functions:

LazySets.dimMethod
dim(am::AbstractAffineMap)

Return the dimension of an affine map.

Input

  • am – affine map

Output

The dimension of an affine map.

source
LazySets.σMethod
σ(d::AbstractVector, am::AbstractAffineMap)

Return the support vector of an affine map.

Input

  • d – direction
  • am – affine map

Output

The support vector in the given direction.

source
LazySets.ρMethod
ρ(d::AbstractVector, am::AbstractAffineMap)

Return the support function of an affine map.

Input

  • d – direction
  • am – affine map

Output

The support function in the given direction.

source
LazySets.an_elementMethod
an_element(am::AbstractAffineMap)

Return some element of an affine map.

Input

  • am – affine map

Output

An element of the affine map. It relies on the an_element function of the wrapped set.

source
Base.isemptyMethod
isempty(am::AbstractAffineMap)

Return whether an affine map is empty or not.

Input

  • am – affine map

Output

true iff the wrapped set is empty.

source
LazySets.isboundedMethod
isbounded(am::AbstractAffineMap; cond_tol::Number=DEFAULT_COND_TOL)

Determine whether an affine map is bounded.

Input

  • am – affine map
  • cond_tol – (optional) tolerance of matrix condition (used to check whether the matrix is invertible)

Output

true iff the affine map is bounded.

Algorithm

We first check if the matrix is zero or the wrapped set is bounded. If not, we perform a sufficient check whether the matrix is invertible. If the matrix is invertible, then the map being bounded is equivalent to the wrapped set being bounded, and hence the map is unbounded. Otherwise, we check boundedness via _isbounded_unit_dimensions.

source
Base.:∈Method
∈(x::AbstractVector, am::AbstractAffineMap)

Check whether a given point is contained in the affine map of a convex set.

Input

  • x – point/vector
  • am – affine map of a convex set

Output

true iff $x ∈ am$.

Algorithm

Note that $x ∈ M⋅S ⊕ v$ iff $M^{-1}⋅(x - v) ∈ S$. This implementation does not explicitly invert the matrix, which is why it also works for non-square matrices.

Examples

julia> am = AffineMap([2.0 0.0; 0.0 1.0], BallInf([1., 1.], 1.), [-1.0, -1.0]);

julia> [5.0, 1.0] ∈ am
false

julia> [3.0, 1.0] ∈ am
true

An example with a non-square matrix:

julia> B = BallInf(zeros(4), 1.);

julia> M = [1. 0 0 0; 0 1 0 0]/2;

julia> [0.5, 0.5] ∈ M*B
true
source
LazySets.centerMethod
center(am::AbstractAffineMap)

Return the center of an affine map of a centrally-symmetric set.

Input

  • cp – affine map of a centrally-symmetric set

Output

The center of the affine map.

source
LazySets.vertices_listMethod
vertices_list(am::AbstractAffineMap; [apply_convex_hull]::Bool)

Return the list of vertices of a (polyhedral) affine map.

Input

  • am – affine map
  • apply_convex_hull – (optional, default: true) if true, apply the convex hull operation to the list of vertices transformed by the affine map

Output

A list of vertices.

Algorithm

This implementation computes all vertices of X, then transforms them through the affine map, i.e. x ↦ M*x + v for each vertex x of X. By default, the convex hull operation is taken before returning this list. For dimensions three or higher, this operation relies on the functionality through the concrete polyhedra library Polyhedra.jl.

If you are not interested in taking the convex hull of the resulting vertices under the affine map, pass apply_convex_hull=false as a keyword argument.

Note that we assume that the underlying set X is polyhedral, either concretely or lazily, i.e. there the function vertices_list should be applicable.

source
LazySets.constraints_listMethod
constraints_list(am::AbstractAffineMap)

Return the list of constraints of a (polyhedral) affine map.

Input

  • am – affine map

Output

The list of constraints of the affine map.

Notes

We assume that the underlying set X is polyhedral, i.e., offers a method constraints_list(X).

Algorithm

Falls back to the list of constraints of the translation of a lazy linear map.

source
LazySets.linear_mapMethod
linear_map(M::AbstractMatrix, am::AbstractAffineMap)

Return the linear map of a lazy affine map.

Input

  • M – matrix
  • am – affine map

Output

A set corresponding to the linear map of the lazy affine map of a set.

source

Implementations

Star sets (AbstractStar)

LazySets.AbstractStarType
AbstractStar{N} <: LazySet{N}

Abstract supertype for all star set types.

Notes

A set $X$ is star-like (also known as generalized star) if it can be represented by a center $x₀ ∈ \mathbb{R}^n$ and $m$ vectors $v₁, …, vₘ$ forming the basis, and a predicate $P : \mathbb{R}^n → \{⊤, ⊥\}$ such that

\[ X = \{x ∈ \mathbb{R}^n : x = x₀ + \sum_{i=1}^m α_i v_i,,~~\textrm{s.t. } P(α) = ⊤ \}.\]

source

Implementations

Polynomial zonotope sets (AbstractPolynomialZonotope)

LazySets.AbstractPolynomialZonotopeType
AbstractPolynomialZonotope{N} <: LazySet{N}

Abstract type for polynomial zonotope sets.

Notes

Polynomial zonotopes are in general non-convex. They are always bounded.

julia> subtypes(LazySets.AbstractPolynomialZonotope)
3-element Vector{Any}:
 DensePolynomialZonotope
 SimpleSparsePolynomialZonotope
 SparsePolynomialZonotope
source

Implementations