Sylvia BiscoveanuComparison of calibration with LALInference
GW150914: Greg's direct comparison
At the request of Matt, here is a direct comparison of bilby and LALInference with and without calibration. These runs where done on bilby v0.5.5 and LALInference version:2ece0713fc5c9f27233696b5c459c23c1d9eae66,2019-07-12 19:13:12.
For interest, here is a table listing some properties derived from each run for comparison.
| Run setup | ini file | Run-time | logZ-ratio |
|---|---|---|---|
| Bilby (no calibration) | link | 15hr 43m | 312.9 +/- 0.3 |
| Bilby (with calibration) | link | 16hr 10m | 312.37 +/- 0.3 |
| LALInference (no calibration) | link | 36hr 65m | 311.093 |
| LALInference (with calibration) | link | 56hr | 311.097 |
GW150914: Sylvia's work
LALInference Calibration envelopes have extra content from 5 - 17.555959 Hz and from 1155.0649 - 5000.0000 Hz (flow = 20Hz, fmax = 1024 Hz)
Generated with this script plot_compare.py using the following results files bilby lalinference
Bilby posteriors are in red, LALInference in blue, Bilby prior in green, LALInference prior in turquoise.
After changing the interpolation function and switching to interpolate in log space in the frequencies to match LALInference, the priors between LALInference and bilby match when reading the GW150914 calibration envelope. The bilby posterior, however, appears to be narrower than the prior and the LALInference posterior. In the following plots, the bilby posterior is in red and the prior in green, and the LALInferenece posterior is in dark blue and the prior in turquoise.
Adding reflective boundary conditions to the calibration priors has eliminated the minor discrepancies between the posteriors above:
S190521r
Bilby posterior in red, LALinference posterior in blue.
Calibration plots:
Bilby:
LALInference:
