Inconsistent number of pairs between theory.wp and theory.DDrppi #345
Description
General information
- Corrfunc version: 2.5.3
- platform: Ubuntu 20.04.6 LTS (GNU/Linux 5.4.0-215-generic x86_64)
- installation method (pip/source/other?): pip
Issue description
The number of pairs ("npairs") return by theory.wp and theory.DDrppi is slightly different on large-scales.
Expected behavior
Should be exactly the same.
Actual behavior
They are different on the order of 1e-7 on large-scales.
What have you tried so far?
Tried performing the same calculation but using float32 rather than float64, the inconsistency would be even larger. So I suspect it's due to precision loss during the calculation.
Minimal failing example
import numpy as np
from Corrfunc.theory import DDrppi, wp, DD, xi
testpos = np.load("testpos.npy")
rp_bins = np.logspace(-2, np.log10(30), 16)
boxsize=400
ncpu = 100
pimax = 40
testpos = testpos.astype(np.float64)
wp_res = wp(boxsize, pimax, ncpu, rp_bins, X=testpos[0], Y=testpos[1], Z=testpos[2])
ddrppi_res = DDrppi(1, ncpu, pimax, rp_bins, X1=testpos[0], Y1=testpos[1], Z1=testpos[2], boxsize=boxsize, periodic=True)
ddrppi_in_rpbins = ddrppi_res['npairs'].reshape(rp_bins.size-1, 40).sum(axis=1)
print(wp_res['npairs']), print(ddrppi_in_rpbins), print(wp_res['npairs'] - ddrppi_in_rpbins)