Third Virial Coefficient of uv-BH implementation returns NaN

[anreimer]

Example in Python (as in lj_models.ipynb):

sigma = 3.7039
eps_k = 150.03
parameters = UVParameters.new_simple(12.0, 6.0, sigma, eps_k)
uvtheory_bh = EquationOfState.uvtheory(parameters, perturbation=Perturbation.BarkerHenderson)
uvtheory_bh.third_virial_coefficient(6 * KELVIN)

returns NaN m³/mol²
Works for uv-B3 & uv-WCA implementation.

[g-bauer]

The u-fraction returns NaN. The reduced_density is zero and reduced_density.powf(a) (see here) returns NaN. Looks like this is a bug in num-dual.
This can be reproduced with:

from num_dual import Dual64

Dual64.from_re(0.0)**1.0 # 0 + [0]ε
Dual64.from_re(0.0)**2.0 # 0 + [0]ε
Dual64.from_re(0.0)**1.5 # NaN + [NaN]ε

[longemen3000]

Related:
autodiff/autodiff#242 (comment)

JuliaDiff/ForwardDiff.jl#480

[prehner]

Since this seems to be caused by the uv-BH model actually not being differentiable sufficiently, there is nothing really to solve here, and I'm closing this issue.