Why does the plot of EnsembleProblem differ from the original ODEProblem?

Specific Domains Numerics
, , , ,
1

I assume that if I don’t change the problem in EnsembleProblem, the EnsembleSummary should produce the same plot as the original ODEProblem solution.

Simple example:

module test

using Plots, StatsPlots
using LinearAlgebra
using OrdinaryDiffEq
using SciMLExpectations
using Distributions

function f(u, p, t)
    return [
        u[2],
        -1 * u[1] + u[3] * (1 - u[1]^2) * u[2],
        0
    ]
end

tspan = (0.0, 2000.0)
u0 = [2.0, 0, 1000]

p = nothing
ref = ODEProblem(f, u0, tspan, p)
refsol = solve(ref, AutoTsit5(Rosenbrock23()))
display(plot(refsol, idxs=[1, 2], title="Van der Pol"))

prob_func(prob, i, repeat) = remake(prob,
    p=nothing)
ensemble_prob = EnsembleProblem(ref, prob_func=prob_func)

ensemblesol = solve(ensemble_prob,
    AutoTsit5(Rosenbrock23()),
    EnsembleThreads(),
    trajectories=3)

summ = EnsembleSummary(ensemblesol)
display(plot(summ, idxs=[1, 2], title="Ensemble"))

# there is no difference
for uu in ensemblesol.u
    display(maximum(norm.(uu.u - refsol.u)))
end

end

The last loop confirms that the solutions are the same:

0.0
0.0
0.0

However, the plots are different:
1
2

What am I missing?

2

What about with safetycopy on?

3

Same.