JuliaApproximation / JuliaApproximation/ApproxFun.jl
Rounding errors lead to empty domain error when using abs
Nobody has claimed this yet.
- Dominant language
- Julia
- Stars
- 559
- Forks
- 71
- PR merge metrics
- No merged PRs in 30d
Description
Because of rounding (I assume), when calculating roots we do not recover, say, the domain endpoint but something very close, ie 0.9999999999999994 instead of 1.0. If we try to take the abs of that Fun, it will throw an error complaining that the "Domain cannot be empty". The following shows both the bad behavior and, by perturbing one of the coefficients, the good behavior.
macd@macd-NUC9:~$ julia19
_
_ _ _(_)_ | Documentation: https://docs.julialang.org
(_) | (_) (_) |
_ _ _| |_ __ _ | Type "?" for help, "]?" for Pkg help.
| | | | | | |/ _` | |
| | |_| | | | (_| | | Version 1.9.2 (2023-07-05)
_/ |\__'_|_|_|\__'_| | Official https://julialang.org/ release
|__/ |
julia @v1.9> using ApproxFun
julia @v1.9> rcoefs = [7//32, -23//64, 9//32, -9//64]
4-element Vector{Rational{Int64}}:
7//32
-23//64
9//32
-9//64
julia @v1.9> f1 = Fun(Chebyshev(), Float64.(rcoefs))
Fun(Chebyshev(), [0.21875, -0.359375, 0.28125, -0.140625])
julia @v1.9> f2 = Fun(Chebyshev(), Float64.(rcoefs))
Fun(Chebyshev(), [0.21875, -0.359375, 0.28125, -0.140625])
julia @v1.9> f2.coefficients[2] += eps(1.0)
-0.3593749999999998
julia @v1.9> roots(f1)
3-element Vector{Float64}:
-0.33333333333333326
0.3333333333333333
0.9999999999999994
julia @v1.9> roots(f2)
3-element Vector{Float64}:
-0.3333333333333335
0.33333333333333265
1.0
julia @v1.9> abs(f2)
Fun(ContinuousSpace{Float64, Float64, PiecewiseSegment{Float64, Vector{Float64}}}(PiecewiseSegment{Float64, Vector{Float64}}([-1.0, -0.3333333333333335, 0.33333333333333265, 1.0])), [1.0, 1.11022e-16, 8.67362e-17, 1.91687e-16, 0.09375, -0.03125, -0.03125, -0.00520833, 0.00520833, -0.00520833])
julia @v1.9> abs(f1)
ERROR: ArgumentError: Domain cannot be empty
Stacktrace:
[1] Chebyshev
@ ~/julia-versions/dot_julia/packages/ApproxFunOrthogonalPolynomials/hVSG5/src/Spaces/Chebyshev/Chebyshev.jl:16 [inlined]
[2] Chebyshev
@ ~/julia-versions/dot_julia/packages/ApproxFunOrthogonalPolynomials/hVSG5/src/Spaces/Chebyshev/Chebyshev.jl:22 [inlined]
[3] canonicalspace
@ ~/julia-versions/dot_julia/packages/ApproxFunOrthogonalPolynomials/hVSG5/src/Spaces/Ultraspherical/DirichletSpace.jl:30 [inlined]
[4] checkcanonicalspace
@ ~/julia-versions/dot_julia/packages/ApproxFunBase/rVDQs/src/Space.jl:477 [inlined]
[5] CanonicalTransformPlan
@ ~/julia-versions/dot_julia/packages/ApproxFunBase/rVDQs/src/Space.jl:484 [inlined]
[6] plan_transform
@ ~/julia-versions/dot_julia/packages/ApproxFunBase/rVDQs/src/Space.jl:487 [inlined]
[7] transform(S::ChebyshevDirichlet{1, 1, Segment{Float64}, Float64}, vals::Vector{Float64})
@ ApproxFunBase ~/julia-versions/dot_julia/packages/ApproxFunBase/rVDQs/src/Space.jl:532
[8] *(P::ApproxFunBase.TransformPlan{Float64, ContinuousSpace{Float64, Float64, PiecewiseSegment{Float64, Vector{Float64}}}, false, Nothing}, vals::Vector{Float64})
@ ApproxFunBase ~/julia-versions/dot_julia/packages/ApproxFunBase/rVDQs/src/Spaces/ContinuousSpace.jl:73
[9] transform(S::ContinuousSpace{Float64, Float64, PiecewiseSegment{Float64, Vector{Float64}}}, vals::Vector{Float64})
@ ApproxFunBase ~/julia-versions/dot_julia/packages/ApproxFunBase/rVDQs/src/Space.jl:532
[10] default_Fun(T::Type, f::Function, d::ContinuousSpace{Float64, Float64, PiecewiseSegment{Float64, Vector{Float64}}}, pts::Vector{Float64}, shouldsplat::Val{false})
@ ApproxFunBase ~/julia-versions/dot_julia/packages/ApproxFunBase/rVDQs/src/constructors.jl:48
[11] default_Fun(f::ComposedFunction{typeof(abs), Fun{Chebyshev{ChebyshevInterval{Float64}, Float64}, Float64, Vector{Float64}}}, d::ContinuousSpace{Float64, Float64, PiecewiseSegment{Float64, Vector{Float64}}}, n::Int64, shouldsplat::Val{false})
@ ApproxFunBase ~/julia-versions/dot_julia/packages/ApproxFunBase/rVDQs/src/constructors.jl:62
[12] _default_Fun(f::ComposedFunction{typeof(abs), Fun{Chebyshev{ChebyshevInterval{Float64}, Float64}, Float64, Vector{Float64}}}, d::ContinuousSpace{Float64, Float64, PiecewiseSegment{Float64, Vector{Float64}}})
@ ApproxFunBase ~/julia-versions/dot_julia/packages/ApproxFunBase/rVDQs/src/constructors.jl:117
[13] default_Fun(f::Function, d::ContinuousSpace{Float64, Float64, PiecewiseSegment{Float64, Vector{Float64}}})
@ ApproxFunBase ~/julia-versions/dot_julia/packages/ApproxFunBase/rVDQs/src/constructors.jl:96
[14] Fun(f::Function, d::ContinuousSpace{Float64, Float64, PiecewiseSegment{Float64, Vector{Float64}}})
@ ApproxFunBase ~/julia-versions/dot_julia/packages/ApproxFunBase/rVDQs/src/constructors.jl:225
[15] Fun(f::Function, d::PiecewiseSegment{Float64, Vector{Float64}})
@ ApproxFunBase ~/julia-versions/dot_julia/packages/ApproxFunBase/rVDQs/src/constructors.jl:205
[16] splitmap(g::Function, d::ChebyshevInterval{Float64}, pts::Vector{Float64})
@ ApproxFunBase ~/julia-versions/dot_julia/packages/ApproxFunBase/rVDQs/src/specialfunctions.jl:2
[17] abs(f::Fun{Chebyshev{ChebyshevInterval{Float64}, Float64}, Float64, Vector{Float64}})
@ ApproxFunBase ~/julia-versions/dot_julia/packages/ApproxFunBase/rVDQs/src/specialfunctions.jl:31
[18] top-level scope
@ REPL[9]:1
julia @v1.9>
Contributor guide
No contributing guide indexed for this repository
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start with abs and splitmap in ApproxFunBase/src/specialfunctions.jl, then trace the Chebyshev and Dirichlet canonical-space path shown in the stack trace. Reproduce the endpoint-rounding case in Julia 1.9.2 and compare it with the perturbed f2 example; done means abs no longer reports an empty domain for the unperturbed coefficients.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- julia
- Domain
- tooling
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Mostly clear
- Newbie friendliness
- 35/100