Methods

inf(A::IntervalMatrix{T}) where {T}

Return the infimum of an interval matrix A, which corresponds to taking the element-wise infimum of A.

Input

• A – interval matrix

Output

A scalar matrix whose coefficients are the infima of each element in A.

sup(A::IntervalMatrix{T}) where {T}

Return the supremum of an interval matrix A, which corresponds to taking the element-wise supremum of A.

Input

• A – interval matrix

Output

A scalar matrix whose coefficients are the suprema of each element in A.

mid(A::IntervalMatrix{T}) where {T}

Return the midpoint of an interval matrix A, which corresponds to taking the element-wise midpoint of A.

Input

• A – interval matrix

Output

A scalar matrix whose coefficients are the midpoints of each element in A.

diam(A::IntervalMatrix{T}) where {T}

Return a matrix whose entries describe the diameters of the intervals.

Input

• A – interval matrix

Output

A matrix B of the same shape as A such that B[i, j] == diam(A[i, j]) for each i and j.

radius(A::IntervalMatrix{T}) where {T}

Return the radius of an interval matrix A, which corresponds to taking the element-wise radius of A.

Input

• A – interval matrix

Output

A scalar matrix whose coefficients are the radii of each element in A.

midpoint_radius(A::IntervalMatrix{T}) where {T}

Split an interval matrix $A$ into two scalar matrices $C$ and $S$ such that $A = C + [-S, S]$.

Input

• A – interval matrix

Output

A pair (C, S) such that the entries of C are the central points and the entries of S are the (nonnegative) radii of the intervals in A.

rand(::Type{IntervalMatrix}, m::Int=2, [n]::Int=m;
     N=Float64, rng::AbstractRNG=GLOBAL_RNG)

Return a random interval matrix of the given size and numeric type.

Input

• IntervalMatrix – type, used for dispatch
• m – (optional, default: 2) number of rows
• n – (optional, default: m) number of columns
• rng – (optional, default: GLOBAL_RNG) random-number generator

Output

An interval matrix of size $m × n$ whose coefficients are normally-distributed intervals of type N with mean 0 and standard deviation 1.

Notes

If this function is called with only one argument, it creates a square matrix, because the number of columns defaults to the number of rows.

sample(A::IntervalMatrix{T}; rng::AbstractRNG=GLOBAL_RNG) where {T}

Return a sample of the given random interval matrix.

Input

• A – interval matrix
• m – (optional, default: 2) number of rows
• n – (optional, default: 2) number of columns
• rng – (optional, default: GLOBAL_RNG) random-number generator

Output

An interval matrix of size $m × n$ whose coefficients are normally-distributed intervals of type N with mean 0 and standard deviation 1.

∈(M::AbstractMatrix, A::AbstractIntervalMatrix)

Check whether a concrete matrix is an instance of an interval matrix.

Input

• M – concrete matrix
• A – interval matrix

Output

true iff M is an instance of A

Algorithm

We check for each entry in M whether it belongs to the corresponding interval in A.

±(C::MT, S::MT) where {T, MT<:AbstractMatrix{T}}

Return an interval matrix such that the center and radius of the intervals is given by the matrices C and S respectively.

Input

• C – center matrix
• S – radii matrix

Output

An interval matrix M such that M[i, j] corresponds to the interval whose center is C[i, j] and whose radius is S[i, j], for each i and j. That is, $M = C + [-S, S]$.

Notes

The radii matrix should be nonnegative, i.e. S[i, j] ≥ 0 for each i and j.

Examples

julia> [1 2; 3 4] ± [1 2; 4 5]
2×2 IntervalMatrix{Float64, Interval{Float64}, Matrix{Interval{Float64}}}:
  [0.0, 2.0]   [0.0, 4.0]
 [-1.0, 7.0]  [-1.0, 9.0]

⊆(A::AbstractIntervalMatrix, B::AbstractIntervalMatrix)

Check whether an interval matrix is contained in another interval matrix.

Input

• A – interval matrix
• B – interval matrix

Output

true iff A[i, j] ⊆ B[i, j] for all i, j.

∩(A::IntervalMatrix, B::IntervalMatrix)

Intersect two interval matrices.

Input

• A – interval matrix
• B – interval matrix (of the same shape as A)

Output

A new matrix C of the same shape as A such that C[i, j] = A[i, j] ∩ B[i, j] for each i and j.

∪(A::IntervalMatrix, B::IntervalMatrix)

Finds the interval union (hull) of two interval matrices. This is equivalent to hull.

Input

• A – interval matrix
• B – interval matrix (of the same shape as A)

Output

A new matrix C of the same shape as A such that C[i, j] = A[i, j] ∪ B[i, j] for each i and j.

hull(A::IntervalMatrix, B::IntervalMatrix)

Finds the interval hull of two interval matrices. This is equivalent to ∪.

Input

• A – interval matrix
• B – interval matrix (of the same shape as A)

Output

A new matrix C of the same shape as A such that C[i, j] = hull(A[i, j], B[i, j]) for each i and j.

square(A::IntervalMatrix)

Compute the square of an interval matrix.

Input

• A – interval matrix

Output

An interval matrix equivalent to A * A.

Algorithm

We follow [1, Section 6].

[1] Kosheleva, Kreinovich, Mayer, Nguyen. Computing the cube of an interval matrix is NP-hard. SAC 2005.

scale(A::IntervalMatrix{T}, α::T) where {T}

Return a new interval matrix whose entries are scaled by the given factor.

Input

• A – interval matrix
• α – scaling factor

Output

A new matrix B of the same shape as A such that B[i, j] = α*A[i, j] for each i and j.

Notes

See scale! for the in-place version of this function.

scale!(A::IntervalMatrix{T}, α::T) where {T}

Modifies the given interval matrix, scaling its entries by the given factor.

Input

• A – interval matrix
• α – scaling factor

Output

The matrix A such that for each i and j, the new value of A[i, j] is α*A[i, j].

Notes

This is the in-place version of scale.

set_multiplication_mode(multype)

Sets the algorithm used to perform matrix multiplication with interval matrices.

Input

• multype – symbol describing the algorithm used
  • :slow – uses traditional matrix multiplication algorithm.
  • :fast – computes an enclosure of the matrix product using the midpoint-radius notation of the matrix [RUM10].

Notes

• By default, :slow is used.
• Using fast is generally significantly faster, but it may return larger intervals, especially if midpoint and radius have the same order of magnitude (50% overestimate at most) [RUM99].

increment!(pow::IntervalMatrixPower; [algorithm=default_algorithm])

Increment a matrix power in-place (i.e., storing the result in pow).

Input

• pow – wrapper of a matrix power (modified in this function)
• algorithm – (optional; default: default_algorithm) algorithm to compute the matrix power; available options:
  • "multiply" – fast computation using * from the previous result
  • "power" – recomputation using ^
  • "decompose_binary" – decompose k = 2a + b
  • "intersect" – combination of "multiply"/"power"/"decompose_binary"

Output

The next matrix power, reflected in the modified wrapper.

Notes

Independent of "algorithm", if the index is a power of two, we compute the exact result using squaring.

increment(pow::IntervalMatrixPower; [algorithm=default_algorithm])

Increment a matrix power without modifying pow.

Input

• pow – wrapper of a matrix power
• algorithm – (optional; default: default_algorithm) algorithm to compute the matrix power; see increment! for available options

Output

The next matrix power.

matrix(pow::IntervalMatrixPower)

Return the matrix represented by a wrapper of a matrix power.

Input

• pow – wrapper of a matrix power

Output

The matrix power represented by the wrapper.

base(pow::IntervalMatrixPower)

Return the original matrix represented by a wrapper of a matrix power.

Input

• pow – wrapper of a matrix power

Output

The matrix $M$ being the basis of the matrix power $M^k$ represented by the wrapper.

index(pow::IntervalMatrixPower)

Return the current index of the wrapper of a matrix power.

Input

• pow – wrapper of a matrix power

Output

The index k of the wrapper representing $M^k$.

Horner <: AbstractExponentiationMethod

Matrix exponential using Horner's method.

Fields

• K – number of expansions in the Horner scheme

ScaleAndSquare <: AbstractExponentiationMethod

Fields

• l – scaling-and-squaring order
• p – order of the approximation

TaylorOverapproximation <: AbstractExponentiationMethod

Matrix exponential overapproximation using a truncated Taylor series.

Fields

• p – order of the approximation

TaylorUnderapproximation <: AbstractExponentiationMethod

Matrix exponential underapproximation using a truncated Taylor series.

Fields

• p – order of the approximation

exp_overapproximation(A::IntervalMatrix{T}, t, p)

Overapproximation of the exponential of an interval matrix, exp(A*t), using a truncated Taylor series.

Input

• A – interval matrix
• t – exponentiation factor
• p – order of the approximation

Output

A matrix enclosure of exp(A*t), i.e. an interval matrix M = (m_{ij}) such that [exp(A*t)]_{ij} ⊆ m_{ij}.

Algorithm

See Theorem 1 in Reachability Analysis of Linear Systems with Uncertain Parameters and Inputs by M. Althoff, O. Stursberg, M. Buss.

horner(A::IntervalMatrix{T}, K::Integer; [validate]::Bool=true)

Compute the matrix exponential using the Horner scheme.

Input

• A – interval matrix
• K – number of expansions in the Horner scheme
• validate – (optional; default: true) option to validate the precondition of the algorithm

Algorithm

We use the algorithm in [1, Section 4.2].

[1] Goldsztejn, Alexandre, Arnold Neumaier. "On the exponentiation of interval matrices". Reliable Computing. 2014.

scale_and_square(A::IntervalMatrix{T}, l::Integer, t, p;
                 [validate]::Bool=true)

Compute the matrix exponential using scaling and squaring.

Input

• A – interval matrix
• l – scaling-and-squaring order
• t – non-negative time value
• p – order of the approximation
• validate – (optional; default: true) option to validate the precondition of the algorithm

Algorithm

We use the algorithm in [1, Section 4.3], which first scales A by factor $2^{-l}$, computes the matrix exponential for the scaled matrix, and then squares the result $l$ times.

\[ \exp(A * 2^{-l})^{2^l}\]

[1] Goldsztejn, Alexandre, Arnold Neumaier. "On the exponentiation of interval matrices". Reliable Computing. 2014.

exp_underapproximation(A::IntervalMatrix{T}, t, p) where {T}

Underapproximation of the exponential of an interval matrix, exp(A*t), using a truncated Taylor series expansion.

Input

• A – interval matrix
• t – exponentiation factor
• p – order of the approximation

Output

An underapproximation of exp(A*t), i.e. an interval matrix M = (m_{ij}) such that m_{ij} ⊆ [exp(A*t)]_{ij}.

Algorithm

See Theorem 2 in Reachability Analysis of Linear Systems with Uncertain Parameters and Inputs by M. Althoff, O. Stursberg, M. Buss.

quadratic_expansion(A::IntervalMatrix, α::Real, β::Real)

Compute the quadratic expansion of an interval matrix, $αA + βA^2$, using interval arithmetic.

Input

• A – interval matrix
• α – linear coefficient
• β – quadratic coefficient

Output

An interval matrix that encloses $B := αA + βA^2$.

Algorithm

This a variation of the algorithm in [1, Section 6]. If $A = (aᵢⱼ)$ and $B := αA + βA^2 = (bᵢⱼ)$, the idea is to compute each $bᵢⱼ$ by factoring out repeated expressions (thus the term single-use expressions).

First, let $i = j$. In this case,

\[bⱼⱼ = β\sum_\{k, k ≠ j} a_{jk} a_{kj} + (α + βa_{jj}) a_{jj}.\]

Now consider $i ≠ j$. Then,

\[bᵢⱼ = β\sum_\{k, k ≠ i, k ≠ j} a_{ik} a_{kj} + (α + βa_{ii} + βa_{jj}) a_{ij}.\]

[1] Kosheleva, Kreinovich, Mayer, Nguyen. Computing the cube of an interval matrix is NP-hard. SAC 2005.

correction_hull(A::IntervalMatrix{T}, t, p) where {T}

Compute the correction term for the convex hull of a point and its linear map with an interval matrix in order to contain all trajectories of a linear system.

Input

• A – interval matrix
• t – non-negative time value
• p – order of the approximation

Output

An interval matrix representing the correction term.

Algorithm

See Theorem 3 in [1].

[1] M. Althoff, O. Stursberg, M. Buss. Reachability Analysis of Linear Systems with Uncertain Parameters and Inputs. CDC 2007.

input_correction(A::IntervalMatrix{T}, t, p) where {T}

Compute the input correction matrix for discretizing an inhomogeneous affine dynamical system with an interval matrix and an input domain not containing the origin.

Input

• A – interval matrix
• t – non-negative time value
• p – order of the Taylor approximation

Output

An interval matrix representing the correction matrix.

Algorithm

See Proposition 3.4 in [1].

[1] M. Althoff. Reachability analysis and its application to the safety assessment of autonomous cars. 2010.

opnorm(A::IntervalMatrix, p::Real=Inf)

The matrix norm of an interval matrix.

Input

• A – interval matrix
• p – (optional, default: Inf) the class of p-norm

Notes

The matrix $p$-norm of an interval matrix $A$ is defined as

\[ ‖A‖_p := ‖\max(|\text{inf}(A)|, |\text{sup}(A)|)‖_p\]

where $\max$ and $|·|$ are taken elementwise.

diam_norm(A::IntervalMatrix, p=Inf)

Return the diameter norm of the interval matrix.

Input

• A – interval matrix
• p – (optional, default: Inf) the p-norm used; valid options are: 1, 2, Inf

Output

The operator norm, in the p-norm, of the scalar matrix obtained by taking the element-wise diam function, where diam(x) := sup(x) - inf(x) for an interval x.

Notes

This function gives a measure of the width of the interval matrix.