Compute the space-concentration kernel of an arbitrary mask on the sphere.
dij = ComputeDMap (
dij: float, dimension ( (
- The space-concentration kernel corresponding to the mask dh_mask.
dh_mask: integer, dimension (
- A Driscoll and Healy (1994) sampled grid describing the concentration region R. All elements should either be 1 (for inside the concentration region) or 0 (for outside R).
- The maximum spherical harmonic degree of the matrix
n: optional, integer, default =
- The number of latitudinal samples in
dh_mask. The effective spherical harmonic bandwidth of this grid is
degrees: integer, optional, dimension (
lmax+1), default = 1
- List of degrees to use when computing the space-concentration kernel. Only those degrees where
degrees[l]is non-zero will be employed.
sampling: optional, integer, default determined by dimensions of
- For 1,
nsamples. For 2,
ComputeDMap will calculate the space-concentration kernel for a generic mask defined on the sphere. The input mask
dh_mask must be sampled according to the Driscoll and Healy (1994) sampling theorem with
n samples in latitude, and possess a value of 1 inside the concentration region, and 0 elsewhere.
dh_mask can either possess
n samples in longitude (
2n samples in longitude (
sampling=2). Given the approximate way in which the elements of
dij are calculated (see below),
sampling=2 should be preferred.
dij is symmetric, and the elements are ordered according to the scheme described in
YilmIndexVector. See Simons et al. (2006) for further details. If the optional vector
degrees is specified, then the matrix will be computed only for elements where
degrees(l) is not zero.
The elements of DIJ are explicitly given by
Dlm,l'm' = 1/(4pi) Integral_R Ylm Yl'm' dOmega,
R is the concentration region. In this routine, all values of
l'm' are calculated in a single spherical harmonic transform for a given value of
lm according to
Dl'm' = 1/(4pi) Integral_Omega F Yl'm' dOmega.
F = Ylm dh_mask.
F is in general not a polynomial, and thus the coefficients
Dl'm' should not be expected to be exact. For this reason, the effective spherical harmonic degree of the input mask (
L=n/2-1) should be greater than
lmax. The exact value of
n should be chosen such that further increases in
n do not alter the returned eigenvalues. The routine prints out the fractional area of the mask computed in the pixel domain divided by
D(0,0) (the fractional area computed by the spherical harmonic transforms), and the ratio of the two should be close to 1. Experience suggests that
l should be about 5 times
Driscoll, J.R. and D.M. Healy, Computing Fourier transforms and convolutions on the 2-sphere, Adv. Appl. Math., 15, 202-250, 1994.
Simons, F.J., F.A. Dahlen, and M.A. Wieczorek, Spatiospectral concentration on a sphere, SIAM Review, 48, 504-536, 2006.