Tolerances
The absolute tolerance used for floating point comparisons is hard-coded in the project to eps(T) ^ (3 // 4)
where T
is either Float64
or Float32
. This formula preserves roughly 3/4 of the significant digits of the type T
. You can use ScopedValues.jl to customize these tolerance values in specific computations:
using Meshes
using ScopedValues
with(Meshes.ATOL64 => 1e-9, Meshes.ATOL32 => 1f-4) do
# do your computations with custom tolerances
end