ClassicalOrthogonalPolynomials.jl
A Julia package for classical orthogonal polynomials and expansions
Definitions
We follow the Digital Library of Mathematical Functions, which defines the following classical orthogonal polynomials:
- Legendre: $P_n(x)$
- Chebyshev (1st kind, 2nd kind): $T_n(x)$, $U_n(x)$
- Ultraspherical: $C_n^{(λ)}(x)$
- Jacobi: $P_n^{(a,b)}(x)$
- Laguerre: $L_n^{(α)}(x)$
- Hermite: $H_n(x)$
Evaluation
The simplest usage of this package is to evaluate classical orthogonal polynomials:
julia> using ClassicalOrthogonalPolynomials
julia> n, x = 5, 0.1;
julia> legendrep(n, x) # P_n(x)
0.17882875
julia> chebyshevt(n, x) # T_n(x) == cos(n*acos(x))
0.48016
julia> chebyshevu(n, x) # U_n(x) == sin((n+1)*acos(x))/sin(acos(x))
0.56832
julia> λ = 0.3; ultrasphericalc(n, λ, x) # C_n^(λ)(x)
0.08578714248
julia> a,b = 0.1,0.2; jacobip(n, a, b, x) # P_n^(a,b)(x)
0.17459116797117194
julia> laguerrel(n, x) # L_n(x)
0.5483540833333331
julia> α = 0.1; laguerrel(n, α, x) # L_n^(α)(x)
0.732916666666666
julia> hermiteh(n, x) # H_n(x)
11.84032
Continuum arrays
For expansions, recurrence relationships, and other operations linked with linear equations, it is useful to treat the families of orthogonal polynomials as continuum arrays, as implemented in ContinuumArrays.jl. The continuum arrays implementation is accessed as follows:
julia> T = ChebyshevT() # Or just Chebyshev()
ChebyshevT()
julia> axes(T) # [-1,1] by 1:∞
(Inclusion(-1.0..1.0 (Chebyshev)), OneToInf())
julia> T[x, n+1] # T_n(x) = cos(n*acos(x))
0.48016
We can thereby access many points and indices efficiently using array-like language:
julia> x = range(-1, 1; length=1000);
julia> T[x, 1:1000] # [T_j(x[k]) for k=1:1000, j=0:999]
1000×1000 Matrix{Float64}:
1.0 -1.0 1.0 -1.0 1.0 -1.0 1.0 … -1.0 1.0 -1.0 1.0 -1.0
1.0 -0.997998 0.992 -0.98203 0.968128 -0.95035 0.928766 -0.99029 0.979515 -0.964818 0.946258 -0.92391
1.0 -0.995996 0.984016 -0.964156 0.936575 -0.901494 0.859194 -0.448975 0.367296 -0.282676 0.195792 -0.107341
1.0 -0.993994 0.976048 -0.946378 0.90534 -0.853427 0.791262 0.660163 -0.738397 0.807761 -0.867423 0.916664
1.0 -0.991992 0.968096 -0.928695 0.874421 -0.806141 0.72495 -0.942051 0.892136 -0.827934 0.750471 -0.660989
1.0 -0.98999 0.96016 -0.911108 0.843816 -0.75963 0.660237 … 0.891882 -0.946786 0.982736 -0.999011 0.995286
1.0 -0.987988 0.952241 -0.893616 0.813524 -0.713888 0.597101 0.905338 -0.828835 0.73242 -0.618409 0.489542
⋮ ⋮ ⋱ ⋮
1.0 0.987988 0.952241 0.893616 0.813524 0.713888 0.597101 -0.905338 -0.828835 -0.73242 -0.618409 -0.489542
1.0 0.98999 0.96016 0.911108 0.843816 0.75963 0.660237 -0.891882 -0.946786 -0.982736 -0.999011 -0.995286
1.0 0.991992 0.968096 0.928695 0.874421 0.806141 0.72495 … 0.942051 0.892136 0.827934 0.750471 0.660989
1.0 0.993994 0.976048 0.946378 0.90534 0.853427 0.791262 -0.660163 -0.738397 -0.807761 -0.867423 -0.916664
1.0 0.995996 0.984016 0.964156 0.936575 0.901494 0.859194 0.448975 0.367296 0.282676 0.195792 0.107341
1.0 0.997998 0.992 0.98203 0.968128 0.95035 0.928766 0.99029 0.979515 0.964818 0.946258 0.92391
1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0
Expansions
We view a function expansion in say Chebyshev polynomials in terms of continuum arrays as follows:
\[f(x) = \sum_{k=0}^∞ c_k T_k(x) = \begin{bmatrix}T_0(x) | T_1(x) | … \end{bmatrix} \begin{bmatrix}c_0\\ c_1 \\ \vdots \end{bmatrix} = T[x,:] * 𝐜\]
To be more precise, we think of functions as continuum-vectors. Here is a simple example:
julia> f = T * [1; 2; 3; zeros(∞)]; # T_0(x) + T_1(x) + T_2(x)
julia> f[0.1]
-1.74
To find the coefficients for a given function we consider this as the problem of finding $𝐜$ such that $T*𝐜 == f$, that is:
julia> T \ f
vcat(3-element Vector{Float64}, ℵ₀-element FillArrays.Zeros{Float64, 1, Tuple{InfiniteArrays.OneToInf{Int64}}} with indices OneToInf()) with indices OneToInf():
1.0
2.0
3.0
⋅
⋅
⋅
⋅
⋮
For a function given only pointwise we broadcast over x
, e.g., the following are the coefficients of $\exp(x)$:
julia> x = axes(T, 1);
julia> c = T \ exp.(x)
vcat(14-element Vector{Float64}, ℵ₀-element FillArrays.Zeros{Float64, 1, Tuple{InfiniteArrays.OneToInf{Int64}}} with indices OneToInf()) with indices OneToInf():
1.2660658777520084
1.1303182079849703
0.27149533953407656
0.04433684984866379
0.0054742404420936785
0.0005429263119139232
4.497732295427654e-5
⋮
julia> f = T*c; f[0.1] # ≈ exp(0.1)
1.1051709180756477
With a little cheeky usage of Julia's order-of-operations this can be written succicently as:
julia> f = T / T \ exp.(x);
julia> f[0.1]
1.1051709180756477
(Or for more clarity just write T * (T \ exp.(x))
.)
Jacobi matrices
Orthogonal polynomials satisfy well-known three-term recurrences:
\[x p_n(x) = c_{n-1} p_{n-1}(x) + a_n p_n(x) + b_n p_{n+1}(x).\]
In continuum-array language this has the form of a comuting relationship:
\[x \begin{bmatrix} p_0 | p_1 | \cdots \end{bmatrix} = \begin{bmatrix} p_0 | p_1 | \cdots \end{bmatrix} \begin{bmatrix} a_0 & c_0 \\ b_0 & a_1 & c_1 \\ & b_1 & a_2 & \ddots \\ &&\ddots & \ddots \end{bmatrix}\]
We can therefore find the Jacobi matrix naturally as follows:
julia> T \ (x .* T)
ℵ₀×ℵ₀ LazyBandedMatrices.Tridiagonal{Float64, LazyArrays.ApplyArray{Float64, 1, typeof(vcat), Tuple{Float64, FillArrays.Fill{Float64, 1, Tuple{InfiniteArrays.OneToInf{Int64}}}}}, FillArrays.Zeros{Float64, 1, Tuple{InfiniteArrays.OneToInf{Int64}}}, FillArrays.Fill{Float64, 1, Tuple{InfiniteArrays.OneToInf{Int64}}}} with indices OneToInf()×OneToInf():
0.0 0.5 ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ …
1.0 0.0 0.5 ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅
⋅ 0.5 0.0 0.5 ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅
⋅ ⋅ 0.5 0.0 0.5 ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅
⋅ ⋅ ⋅ 0.5 0.0 0.5 ⋅ ⋅ ⋅ ⋅ ⋅ ⋅
⋅ ⋅ ⋅ ⋅ 0.5 0.0 0.5 ⋅ ⋅ ⋅ ⋅ ⋅ …
⋅ ⋅ ⋅ ⋅ ⋅ 0.5 0.0 0.5 ⋅ ⋅ ⋅ ⋅
⋮ ⋮ ⋮ ⋱
Alternatively, just call jacobimatrix(T)
(noting its the transpose of the more traditional convention).
Derivatives
The derivatives of classical orthogonal polynomials are also classical OPs, and this can be seen as follows:
julia> U = ChebyshevU();
julia> D = Derivative(x);
julia> U\D*T
ℵ₀×ℵ₀ BandedMatrix{Float64} with bandwidths (-1, 1) with data 1×ℵ₀ adjoint(::InfiniteArrays.InfStepRange{Float64, Float64}) with eltype Float64 with indices Base.OneTo(1)×OneToInf() with indices OneToInf()×OneToInf():
⋅ 1.0 ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ …
⋅ ⋅ 2.0 ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅
⋅ ⋅ ⋅ 3.0 ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅
⋅ ⋅ ⋅ ⋅ 4.0 ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅
⋅ ⋅ ⋅ ⋅ ⋅ 5.0 ⋅ ⋅ ⋅ ⋅ ⋅ ⋅
⋅ ⋅ ⋅ ⋅ ⋅ ⋅ 6.0 ⋅ ⋅ ⋅ ⋅ ⋅ …
⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ 7.0 ⋅ ⋅ ⋅ ⋅
⋮ ⋮ ⋮ ⋱
Similarly, the derivative of weighted classical OPs are weighted classical OPs:
lia> Weighted(T)\D*Weighted(U)
ℵ₀×ℵ₀ BandedMatrix{Float64} with bandwidths (1, -1) with data 1×ℵ₀ adjoint(::InfiniteArrays.InfStepRange{Float64, Float64}) with eltype Float64 with indices Base.OneTo(1)×OneToInf() with indices OneToInf()×OneToInf():
⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ …
-1.0 ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅
⋅ -2.0 ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅
⋅ ⋅ -3.0 ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅
⋅ ⋅ ⋅ -4.0 ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅
⋅ ⋅ ⋅ ⋅ -5.0 ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ …
⋅ ⋅ ⋅ ⋅ ⋅ -6.0 ⋅ ⋅ ⋅ ⋅ ⋅
⋮ ⋮ ⋮ ⋱
Other recurrence relationships
Many other sparse recurrence relationships are implemented. Here's one:
julia> U\T
ℵ₀×ℵ₀ BandedMatrix{Float64} with bandwidths (0, 2) with data vcat(1×ℵ₀ FillArrays.Fill{Float64, 2, Tuple{Base.OneTo{Int64}, InfiniteArrays.OneToInf{Int64}}} with indices Base.OneTo(1)×OneToInf(), 1×ℵ₀ FillArrays.Zeros{Float64, 2, Tuple{Base.OneTo{Int64}, InfiniteArrays.OneToInf{Int64}}} with indices Base.OneTo(1)×OneToInf(), hcat(1×1 Ones{Float64}, 1×ℵ₀ FillArrays.Fill{Float64, 2, Tuple{Base.OneTo{Int64}, InfiniteArrays.OneToInf{Int64}}} with indices Base.OneTo(1)×OneToInf()) with indices Base.OneTo(1)×OneToInf()) with indices Base.OneTo(3)×OneToInf() with indices OneToInf()×OneToInf():
1.0 0.0 -0.5 ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ …
⋅ 0.5 0.0 -0.5 ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅
⋅ ⋅ 0.5 0.0 -0.5 ⋅ ⋅ ⋅ ⋅ ⋅ ⋅
⋅ ⋅ ⋅ 0.5 0.0 -0.5 ⋅ ⋅ ⋅ ⋅ ⋅
⋅ ⋅ ⋅ ⋅ 0.5 0.0 -0.5 ⋅ ⋅ ⋅ ⋅
⋅ ⋅ ⋅ ⋅ ⋅ 0.5 0.0 -0.5 ⋅ ⋅ ⋅ …
⋅ ⋅ ⋅ ⋅ ⋅ ⋅ 0.5 0.0 -0.5 ⋅ ⋅
⋮ ⋮ ⋮ ⋱
(Probably best to ignore the type signature 😅)