Investigate nondeterministic behavior

[schillic]

julia> em = ExponentialMap(SparseMatrixExp(SparseMatrixCSC([2.0 0.0; 0.0 1.0])),
                           BallInf([1., 1.], 1.));
julia> LazySets.Approximations.box_approximation(em)

This nondeterministically results in either

LazySets.Hyperrectangle{Float64}([7.38906, 2.71828], [7.38906, 2.71828])

or

ERROR: AssertionError: radius must not be negative

[schillic]

Narrowing it down:

julia> em = ExponentialMap(SparseMatrixExp(SparseMatrixCSC([2.0 0.0; 0.0 1.0])),
                           BallInf([1., 1.], 1.));
julia> d = [1., 0.];

julia> σ(d, em)
2-element Array{Float64,1}:
 14.7781 
  5.43656

julia> σ(d, em)
2-element Array{Float64,1}:
 NaN      
   5.43656

[schillic]

The problem is in Expokit.jl.

julia> using Expokit

julia> t = 1.; A = SparseMatrixCSC([2.0 0.0; 0.0 1.0]); vec = [2., 2.];

julia> for i in 1:10
           println(expmv(t, A, vec))
       end
[14.7781, 5.43656]
[14.7781, 5.43656]
[14.7781, 5.43656]
[14.7781, 5.43656]
[14.7781, 5.43656]
[14.7781, 5.43656]
[14.7781, 5.43656]
[NaN, 5.43656]
[14.7781, 5.43656]
[14.7781, 5.43656]

[mforets]

julia> using Expokit

julia> t = 1.; A = SparseMatrixCSC([2.0 0.0; 0.0 1.0]); v = [2., 2.];

julia> P = [expmv(t, A, v) for i in 1:10^4];

julia> any([isnan(Pi[1]) for Pi in P])
false

[schillic]

In the master branch of Expokit.jl I also cannot reproduce. If you want to try, go back to 6f5b0b6 (git checkout 6f5b0b6).
Should we ask for a new release?

[mforets]

good! i can ask.