IMRPhenomPv3(HM) failures near equal mass
Hi @sebastian-khan during the PhenomXP(HM) review I noticed some PhenomPv3(HM) failures in the pycbc test_lalsim.py
test case test_almost_equal_mass_waveform
when perturbing around an equal-mass configuration. Here's an extracted example ported to lalsim-inspiral
for simplicity:
lalsim-inspiral -a IMRPhenomPv3 -F --m1 23.000000536119625 --m2 23.000000592237015
XLAL Error - init_useful_powers (LALSimIMRPhenomD_internals.c:216): number must be non-negative
XLAL Error - init_useful_powers (LALSimIMRPhenomD_internals.c:216): Input domain error
backtrace:
/home/dkeitel/bin//lalsuite-opt/lib/liblal.so.18(XLALBacktraceErrorHandler+0x5b)[0x7f557456703f]
/home/dkeitel/bin//lalsuite-opt/lib/liblal.so.18(XLALError+0x5f)[0x7f5574566f75]
/home/dkeitel/bin//lalsuite-opt/lib/liblalsimulation.so.22(+0x32c8da)[0x7f55755f38da]
/home/dkeitel/bin//lalsuite-opt/lib/liblalsimulation.so.22(IMRPhenomDPhase_OneFrequency+0x37)[0x7f55755ff55c]
/home/dkeitel/bin//lalsuite-opt/lib/liblalsimulation.so.22(IMRPhenomHMPhasePreComp+0x5bc)[0x7f557560aae5]
/home/dkeitel/bin//lalsuite-opt/lib/liblalsimulation.so.22(IMRPhenomHMPhase+0x67)[0x7f557560dbb3]
/home/dkeitel/bin//lalsuite-opt/lib/liblalsimulation.so.22(IMRPhenomHMEvaluateOnehlmMode+0x72)[0x7f557560d2c8]
/home/dkeitel/bin//lalsuite-opt/lib/liblalsimulation.so.22(XLALSimIMRPhenomHMGethlmModes+0xe86)[0x7f557560cff9]
/home/dkeitel/bin//lalsuite-opt/lib/liblalsimulation.so.22(XLALSimIMRPhenomPv3HMGetHplusHcross+0x82e)[0x7f557562a664]
/home/dkeitel/bin//lalsuite-opt/lib/liblalsimulation.so.22(XLALSimIMRPhenomPv3+0x140)[0x7f557562b54b]
/home/dkeitel/bin//lalsuite-opt/lib/liblalsimulation.so.22(XLALSimInspiralChooseFDWaveform+0x9b66)[0x7f55756d8ac6]
lalsim-inspiral(+0x4e76)[0x565069639e76]
lalsim-inspiral(+0x3802)[0x565069638802]
/lib/x86_64-linux-gnu/libc.so.6(__libc_start_main+0xeb)[0x7f5573b3109b]
lalsim-inspiral(+0x350a)[0x56506963850a]
Aborted
Neither D, Pv2 nor PX(HM) fail on this. No failure for --m1 23.0 --m2 23.0
nor --m1 23.000000536119625 --m2 23.000000536119625
, but the failing values came out of random perturbations and when rerunning the test it happens every now and then, not just for one combination of magic numbers.
This seems to happen on master both before (8e40d942) and after (a28c2f71) the changes from !1270 (merged), and the failure appears even for 0-spin anyway.
cc @maria.haney @sascha.husa @cecilio.garcia-quiros @geraint.pratten