Sobolev_coefs function

Transformation between different coefficients in Sobolev statistics

Transformation between different coefficients in Sobolev statistics

Given a Sobolev statistic [REMOVE_ME]Sn,p=i,j=1nψ(cos1(XiXj)),Sn,p=i,j=1nψ(cos1(XiXj)),[REMOVEME2] S_{n, p} = \sum_{i, j = 1}^n \psi(\cos^{-1}({\bf X}_i'{\bf X}_j)),S_{n, p} = \sum_{i, j = 1}^n \psi(\cos^{-1}(X_i'X_j)), [REMOVE_ME_2]

for a sample c("X1,ldots,XninSp1:=\nX_1, \\ldots, X_n \\in S^{p - 1} :=\n", "xinRp:x=1{x \\in R^p : ||x|| = 1}"), p2p\ge 2, three important sequences are related to Sn,pS_{n, p}.

  • Gegenbauer coefficients {bk,p}\{b_{k, p}\} of ψp\psi_p (see, e.g., the projected-ecdf statistics ), given by [REMOVE_ME]bk,p:=1ck,p0πψp(θ)Ckp/21(cosθ)dθ.bk,p:=1ck,p0πψp(θ)Ck(p/21)(cosθ)dθ.[REMOVEME2] b_{k, p} := \frac{1}{c_{k, p}}\int_0^\pi \psi_p(\theta)C_k^{p / 2 - 1}(\cos\theta)\,\mathrm{d}\theta.b_{k, p} := \frac{1}{c_{k, p}} \int_0^\pi \psi_p(\theta)C_k^(p / 2 - 1)(\cos\theta) d\theta. [REMOVE_ME_2]
  • Weights {vk,p2}\{v_{k, p}^2\} of the asymptotic distribution of the Sobolev statistic, k=1vk2χdp,k2\sum_{k = 1}^\infty v_k^2 \chi^2_{d_{p, k}}, given by [REMOVE_ME]vk,p2=(1+2kp2)1bk,p,p3.vk,p2=(1+2k/(p2))1bk,p,p3.[REMOVEME2] v_{k, p}^2 = \left(1 + \frac{2k}{p - 2}\right)^{-1} b_{k, p},\quad p \ge 3.v_{k, p}^2 = (1 + 2k / (p - 2))^{-1} b_{k, p}, p \ge 3. [REMOVE_ME_2]
  • Gegenbauer coefficients {uk,p}\{u_{k, p}\} of the local projected alternative associated to Sn,pS_{n, p}, given by [REMOVE_ME]uk,p=(1+2kp2)vk,p,p3.uk,p=(1+2k/(p2))bk,p,p3.[REMOVEME2] u_{k, p} = \left(1 + \frac{2k}{p - 2}\right) v_{k, p},\quad p \ge 3.u_{k, p} = (1 + 2k / (p - 2)) b_{k, p}, p \ge 3. [REMOVE_ME_2]

For p=2p = 2, the factor (1+2k/(p2))(1 + 2k / (p - 2)) is replaced by 22.

bk_to_vk2(bk, p, log = FALSE) bk_to_uk(bk, p, signs = 1) vk2_to_bk(vk2, p, log = FALSE) vk2_to_uk(vk2, p, signs = 1) uk_to_vk2(uk, p) uk_to_bk(uk, p)

Arguments

  • bk: coefficients bk,pb_{k, p} associated to the indexes 1:length(bk), a vector.
  • p: integer giving the dimension of the ambient space RpR^p that contains Sp1S^{p-1}.
  • log: do operations in log scale (log-in, log-out)? Defaults to FALSE.
  • signs: signs of the coefficients uk,pu_{k, p}, a vector of the same size as vk2 or bk, or a scalar. Defaults to 1.
  • vk2: squared coefficients vk,p2v_{k, p}^2 associated to the indexes 1:length(vk2), a vector.
  • uk: coefficients uk,pu_{k, p} associated to the indexes 1:length(uk), a vector.

Returns

The corresponding vectors of coefficients vk2, bk, or uk, depending on the call.

Description

Given a Sobolev statistic

Sn,p=i,j=1nψ(cos1(XiXj)),Sn,p=i,j=1nψ(cos1(XiXj)), S_{n, p} = \sum_{i, j = 1}^n \psi(\cos^{-1}({\bf X}_i'{\bf X}_j)),S_{n, p} = \sum_{i, j = 1}^n \psi(\cos^{-1}(X_i'X_j)),

for a sample c("X1,ldots,XninSp1:=\nX_1, \\ldots, X_n \\in S^{p - 1} :=\n", "xinRp:x=1{x \\in R^p : ||x|| = 1}"), p2p\ge 2, three important sequences are related to Sn,pS_{n, p}.

  • Gegenbauer coefficients {bk,p}\{b_{k, p}\} of ψp\psi_p (see, e.g., the projected-ecdf statistics ), given by
bk,p:=1ck,p0πψp(θ)Ckp/21(cosθ)dθ.bk,p:=1ck,p0πψp(θ)Ck(p/21)(cosθ)dθ. b_{k, p} := \frac{1}{c_{k, p}}\int_0^\pi \psi_p(\theta)C_k^{p / 2 - 1}(\cos\theta)\,\mathrm{d}\theta.b_{k, p} := \frac{1}{c_{k, p}} \int_0^\pi \psi_p(\theta)C_k^(p / 2 - 1)(\cos\theta) d\theta.
  • Weights {vk,p2}\{v_{k, p}^2\} of the asymptotic distribution of the Sobolev statistic, k=1vk2χdp,k2\sum_{k = 1}^\infty v_k^2 \chi^2_{d_{p, k}}, given by
vk,p2=(1+2kp2)1bk,p,p3.vk,p2=(1+2k/(p2))1bk,p,p3. v_{k, p}^2 = \left(1 + \frac{2k}{p - 2}\right)^{-1} b_{k, p},\quad p \ge 3.v_{k, p}^2 = (1 + 2k / (p - 2))^{-1} b_{k, p}, p \ge 3.
  • Gegenbauer coefficients {uk,p}\{u_{k, p}\} of the local projected alternative associated to Sn,pS_{n, p}, given by
uk,p=(1+2kp2)vk,p,p3.uk,p=(1+2k/(p2))bk,p,p3. u_{k, p} = \left(1 + \frac{2k}{p - 2}\right) v_{k, p},\quad p \ge 3.u_{k, p} = (1 + 2k / (p - 2)) b_{k, p}, p \ge 3.

For p=2p = 2, the factor (1+2k/(p2))(1 + 2k / (p - 2)) is replaced by 22.

Details

See more details in Prentice (1978) and García-Portugués et al. (2023). The adequate signs of uk for the "PRt" Rothman test

can be retrieved with akx and sqr = TRUE, see the examples.

Examples

# bk, vk2, and uk for the PCvM test in p = 3 (bk <- Gegen_coefs_Pn(k = 1:5, type = "PCvM", p = 3)) (vk2 <- bk_to_vk2(bk = bk, p = 3)) (uk <- bk_to_uk(bk = bk, p = 3)) # vk2 is the same as weights_dfs_Sobolev(K_max = 10, thre = 0, p = 3, type = "PCvM")$weights # bk and uk for the Rothman test in p = 3, with adequate signs t <- 1 / 3 (bk <- Gegen_coefs_Pn(k = 1:5, type = "PRt", p = 3, Rothman_t = t)) (ak <- akx(x = drop(q_proj_unif(t, p = 3)), p = 3, k = 1:5, sqr = TRUE)) (uk <- bk_to_uk(bk = bk, p = 3, signs = ak))

References

García-Portugués, E., Navarro-Esteban, P., Cuesta-Albertos, J. A. (2023) On a projection-based class of uniformity tests on the hypersphere. Bernoulli, 29(1):181--204. tools:::Rd_expr_doi("10.3150/21-BEJ1454") .

Prentice, M. J. (1978). On invariant tests of uniformity for directions and orientations. The Annals of Statistics, 6(1):169--176. tools:::Rd_expr_doi("10.1214/aos/1176344075")

  • Maintainer: Eduardo García-Portugués
  • License: GPL-3
  • Last published: 2024-05-24