BUG: BNS parameter conversion can overwrite valid `lambda_1 = 0` with a small negative value during post-processing

[physicish]

Describe the bug

When convert_to_lal_binary_neutron_star_parameters receives a parameter dictionary containing both the component tidal deformabilities (lambda_1 and lambda_2) and the derived tidal parameters (lambda_tilde and delta_lambda_tilde), it reconstructs and overwrites the existing component values from the derived representation. Due to floating-point round-off, this conversion can change an explicitly supplied lambda_1 = 0.0 into a very small negative value.

I think this occurs during run_sampler post-processing because coversion_function=conversion_function=bilby.gw.conversion.generate_all_bns_parameters is applied to the injection parameters twice. The first call adds lambda_tilde and delta_lambda_tilde and the second call passes the filled dictionary back through convert_to_lal_binary_neutron_star_parameters, triggering the overwrite.

In my case, the (NSBH) injection specified:

lambda_1 = 0.0
lambda_2 = 1097.4026147742702

The sampler completed successfully and produced posterior samples. During post-processing, Bilby converted the injection parameters a second time. Since the first conversion had added lambda_tilde and delta_lambda_tilde, the second conversion reconstructed the component tidal deformabilities from those derived quantities. Due to floating-point round-off, the reconstructed value was approximately:

lambda_1 = -2.46e-14

LALSimulation printed this as:

lambda1 = -0.000000

and rejected the waveform because tidal deformabilities are required to be non-negative.

This appears to be caused by the following logic in convert_to_lal_binary_neutron_star_parameters:

if 'delta_lambda_tilde' in converted_parameters.keys():
    converted_parameters['lambda_1'], converted_parameters['lambda_2'] =\
        lambda_tilde_delta_lambda_tilde_to_lambda_1_lambda_2(
            converted_parameters['lambda_tilde'],
            parameters['delta_lambda_tilde'],
            converted_parameters['mass_1'],
            converted_parameters['mass_2'])
elif 'lambda_tilde' in converted_parameters.keys():
    converted_parameters['lambda_1'], converted_parameters['lambda_2'] =\
        lambda_tilde_to_lambda_1_lambda_2(
            converted_parameters['lambda_tilde'],
            converted_parameters['mass_1'],
            converted_parameters['mass_2'])

These lines overwrite lambda_1 and lambda_2 even when both component values are already present and valid.

To Reproduce

A minimal reproducer:

import bilby

parameters = {
    "mass_1": 7.320188129088971,
    "mass_2": 1.4957247709090036,
    "lambda_1": 0.0,
    "lambda_2": 1097.4026147742702,
}

first_conversion = bilby.gw.conversion.generate_all_bns_parameters(
    parameters
)

print("After first conversion:")
print("lambda_1 =", repr(first_conversion["lambda_1"]))
print("lambda_2 =", repr(first_conversion["lambda_2"]))
print("lambda_tilde =", repr(first_conversion["lambda_tilde"]))
print(
    "delta_lambda_tilde =",
    repr(first_conversion["delta_lambda_tilde"]),
)

second_conversion, _ = (
    bilby.gw.conversion.convert_to_lal_binary_neutron_star_parameters(
        first_conversion
    )
)

print("After second conversion:")
print("lambda_1 =", repr(second_conversion["lambda_1"]))
print("lambda_2 =", repr(second_conversion["lambda_2"]))

For the values above, the second conversion produces a value close to:

lambda_1 = -2.45587692656649e-14

In my run, the relevant runtime sequence was:

bilby INFO    : Rejection sampling nested samples to obtain 6335 posterior samples
bilby INFO    : Sampling time: 1 day, 3:24:28.088555
bilby WARNING : Result.save_to_file called with extension=True. This will default to json, and ignore the extension from the filename.

The failure then occurred after sampling.

Error message

XLAL Error - XLALStoreTidalParametersFromLambdas (LALSimIMRPhenomXHM_tidal.c:71): Tidal deformabilities cannot be negative. Value in LALDict are lambda1 = -0.000000, lambda2 = 1097.402615.
XLAL Error - XLALStoreTidalParametersFromLambdas (LALSimIMRPhenomXHM_tidal.c:71): Invalid argument
XLAL Error - IMRPhenomXSetWaveformVariables (LALSimIMRPhenomX_internals.c:306): Failed to set tidal parameters from lambdas.

XLAL Error - IMRPhenomXSetWaveformVariables (LALSimIMRPhenomX_internals.c:306): Internal function call failed: Invalid argument
XLAL Error - XLALSimIMRPhenomXHMFrequencySequence (LALSimIMRPhenomXHM.c:1785): Error: IMRPhenomXSetWaveformVariables failed.

XLAL Error - XLALSimIMRPhenomXHMFrequencySequence (LALSimIMRPhenomXHM.c:1785): Internal function call failed: Invalid argument
XLAL Error - XLALSimInspiralChooseFDWaveformSequence (LALSimInspiralWaveformCache.c:1710): Internal function call failed: Invalid argument
Traceback (most recent call last):
  File "/ligo/home/ligo.org/ish.gupta/Projects/nsbh_hom/tests/test1/phenom-nsbh/run_1/ECC/runs/run_1/run.py", line 199, in <module>
    result = bilby.run_sampler(
             ^^^^^^^^^^^^^^^^^^
  File "/scratch2/ligo.org/ish.gupta/conda-envs/nsbh-hom/src/bilby/bilby/core/sampler/__init__.py", line 358, in run_sampler
    result.injection_parameters = conversion_function(
                                  ^^^^^^^^^^^^^^^^^^^^
  File "/scratch2/ligo.org/ish.gupta/conda-envs/nsbh-hom/src/bilby/bilby/gw/conversion.py", line 1758, in generate_all_bns_parameters
    output_sample = _generate_all_cbc_parameters(
                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/scratch2/ligo.org/ish.gupta/conda-envs/nsbh-hom/src/bilby/bilby/gw/conversion.py", line 1653, in _generate_all_cbc_parameters
    compute_per_detector_log_likelihoods(
  File "/scratch2/ligo.org/ish.gupta/conda-envs/nsbh-hom/src/bilby/bilby/gw/conversion.py", line 2317, in compute_per_detector_log_likelihoods
    samples = likelihood.compute_per_detector_log_likelihood(samples)
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/scratch2/ligo.org/ish.gupta/conda-envs/nsbh-hom/src/bilby/bilby/gw/likelihood/base.py", line 482, in compute_per_detector_log_likelihood
    self.waveform_generator.frequency_domain_strain(parameters)
  File "/scratch2/ligo.org/ish.gupta/conda-envs/nsbh-hom/src/bilby/bilby/gw/waveform_generator.py", line 135, in frequency_domain_strain
    return self._calculate_strain(model=self.frequency_domain_source_model,
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/scratch2/ligo.org/ish.gupta/conda-envs/nsbh-hom/src/bilby/bilby/gw/waveform_generator.py", line 197, in _calculate_strain
    model_strain = self._strain_from_model(model_data_points, model, parameters, xp=xp)
                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/scratch2/ligo.org/ish.gupta/conda-envs/nsbh-hom/src/bilby/bilby/gw/waveform_generator.py", line 214, in _strain_from_model
    return model(model_data_points, **parameters)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/scratch2/ligo.org/ish.gupta/conda-envs/nsbh-hom/src/bilby/bilby/gw/source.py", line 795, in lal_binary_neutron_star_relative_binning
    return _base_waveform_frequency_sequence(
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/scratch2/ligo.org/ish.gupta/conda-envs/nsbh-hom/src/bilby/bilby/gw/source.py", line 1125, in _base_waveform_frequency_sequence
    h_plus, h_cross = lalsim_SimInspiralChooseFDWaveformSequence(
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/scratch2/ligo.org/ish.gupta/conda-envs/nsbh-hom/src/bilby/bilby/gw/utils.py", line 743, in lalsim_SimInspiralChooseFDWaveformSequence
    return SimInspiralChooseFDWaveformSequence(
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
RuntimeError: Internal function call failed: Invalid argument

Expected behavior

When both lambda_1 and lambda_2 are already present and non-None, Bilby should preserve those component values.

Derived quantities such as lambda_tilde and delta_lambda_tilde should not take precedence over explicitly supplied component tidal deformabilities during a subsequent conversion.

Suggested solution (optional)

Only reconstruct lambda_1 and lambda_2 from the combined tidal quantities when valid component lambdas are not already available.

For example:

have_component_lambdas = all(
    key in converted_parameters
    and converted_parameters[key] is not None
    for key in ("lambda_1", "lambda_2")
)

if (
    "delta_lambda_tilde" in converted_parameters
    and not have_component_lambdas
):
    converted_parameters["lambda_1"], converted_parameters["lambda_2"] = (
        lambda_tilde_delta_lambda_tilde_to_lambda_1_lambda_2(
            converted_parameters["lambda_tilde"],
            converted_parameters["delta_lambda_tilde"],
            converted_parameters["mass_1"],
            converted_parameters["mass_2"],
        )
    )

elif (
    "lambda_tilde" in converted_parameters
    and not have_component_lambdas
):
    converted_parameters["lambda_1"], converted_parameters["lambda_2"] = (
        lambda_tilde_to_lambda_1_lambda_2(
            converted_parameters["lambda_tilde"],
            converted_parameters["mass_1"],
            converted_parameters["mass_2"],
        )
    )

This would:

1. Preserve explicitly available lambda_1 and lambda_2.
2. Reconstruct component lambdas from lambda_tilde and delta_lambda_tilde only when the component representation is unavailable.

Environment

• Bilby version: 2.8.0.dev60+g28fca2270
• Installation method: source installation inside a Conda environment
• Python version: 3.11
• Dynesty version: 3.0.0
• LAL version: 7.6.1.1
• LALSimulation version: 6.1.0.1
• Waveform approximant: IMRPhenomXHM_NSBH (private repo)
• Source model: bilby.gw.source.lal_binary_neutron_star_relative_binning