Enumeration

NumCosmoXcorMethod

Declaration

enum NumCosmo.XcorMethod

Description [src]

Methods to compute integrals.

The kernel-space methods share one closure across a whole NcXcor:ell-batch-size block, fitted to an $L^2$ norm over the block. With NcXcor:ell-batch-size 1 each multipole gets its own closure, the independent check that does not share a closure with anything.

NC_XCOR_METHOD_KERNEL_EXACT needs no tolerance and cannot fail to converge. The name is meant literally rather than as “fixed-order”: on the closures it is handed, the quadrature carries no error at all, and refining its panels changes nothing beyond rounding. It uses the same per-kernel closures as NC_XCOR_METHOD_KERNEL_CUBATURE and differs only in the outer quadrature: each kernel’s $W(k)$ is a cubic spline, so on the common refinement of a pair’s two knot sets the outer integrand $k^2 W_i W_j$ is a degree-8 polynomial on every panel, and a 5-node Gauss-Legendre rule integrates it exactly. The adaptive alternatives target a tolerance the integrand may not be able to support, and abort when they cannot reach it.

Measured over 28 pairs of 7 top-hat bins it is also slightly faster than NC_XCOR_METHOD_KERNEL_CUBATURE (1.05x at $\ell = 0$, 1.18x over $\ell = 0\dots26$), since the exact rule replaces adaptive refinement on splines that have already been built.

Members

NC_XCOR_METHOD_LIMBER_Z_GSL

Use GSL numerical integration.

  • Value: 0
  • Available since: 1.0
NC_XCOR_METHOD_LIMBER_Z_CUBATURE

Use cubature numerical integration.

  • Value: 1
  • Available since: 1.0
NC_XCOR_METHOD_KERNEL_CUBATURE

Use cubature numerical integration over kernel variables.

  • Value: 2
  • Available since: 1.0
NC_XCOR_METHOD_KERNEL_EXACT

Integrate the kernel closures exactly, on the union of their own knots.

  • Value: 3
  • Available since: 1.0

Type functions

nc_xcor_method_get_name
No description available.

nc_xcor_method_has_error_estimate

Whether meth fills the vp_err of nc_xcor_compute_full() and nc_xcor_integrate_block(). Every other method leaves it alone rather than writing a zero that would read as “no error”, so a caller that wants an estimate has to ask this first.

nc_xcor_method_is_kernel_space

Whether meth integrates over $k$ with a pair of fitted closures, as opposed to the redshift-space Limber tier.