Astronomy · White dwarfs, neutron stars and black holes

What remnant observations do and do not constrain

Trace the path from a detector signal to a remnant claim, distinguishing measurement uncertainty, incomplete identification and conditional physical histories.

Continue in the approved chapter order. Necessary vocabulary, quantities and assumptions are explained in the lesson.

Back to the chapter

After reading, you can

Separate measured quantities from model-dependent remnant properties.
Explain why smaller measurement errors need not remove a physical degeneracy.
Evaluate conditional age and uncertainty examples without turning them into universal measurements.

A recorded signal and an inferred property are different things

An observing instrument records signals such as detector counts, spectral features or pulse arrival times. Calibration and an observing model connect those records to fluxes, velocities or timing quantities. A further physical model connects them to masses, radii, energy reservoirs or evolutionary histories.

Each connection has assumptions. A pulse-profile radius inference can require surface-emission and atmosphere descriptions, geometry and relativistic light propagation. A dynamical mass inference requires a suitable orbit and constraints on orientation and other system properties.

An observed period can be measured precisely while its interpretation still depends on whether orbital motion, kinematic effects or changing emission features have been accounted for. Precision in one recorded quantity does not automatically establish every inferred property.

Begin by stating the claim and identifying which observations constrain it. For example, “the system has this fitted orbital mass function” is a narrower statement than “the unseen object has this exact mass and an observed event horizon”. The distinction helps assess both strong evidence and its remaining limits.

An unresolved model relation is not merely a noisy measurement

The preceding mass-function example constrained a combination of mass, mass ratio and inclination. The accretion example constrained the product of radiative efficiency and supplied rest-mass rate. In each case, additional information is needed to separate the unknown inputs.

Imagine knowing an emitted accretion luminosity perfectly while retaining L_acc = ηṀ_sup c² with both positive η and Ṁ_sup unknown. Halving η and doubling Ṁ_sup leaves the same luminosity. Smaller observational error alone cannot remove that mathematical ambiguity.

Independent constraints can help, but their assumptions must be retained. An inclination inferred from a light curve depends on its geometry and extra light; an efficiency inferred from a flow model depends on its boundary and radiative treatment. A prior assumption is useful only when it is identified as such.

Selection adds another layer. A missed beam does not rule out a neutron star, and a detected jet alone does not establish a black hole. Comparisons must include which kinds of sources and signals the observation could have recovered.

An initial condition can change the age interpretation

For an original timing example, take present P = 0.1 s and intrinsic Ṗ = 10⁻¹⁵ s s⁻¹. Suppose the model has constant A =PṖ through the relevant history and constant moment of inertia. Integrating d(P²)/dt = 2A gives elapsed time t = (P² − P₀²)/(2PṖ), with initial spin period P₀.

Neglecting P₀ relative to P gives the familiar characteristic value τ_c =P/(2Ṗ) = 5 × 10¹³ s. Using one Julian year =31557600 s, this is about 1584404.391 years. It is a calculated value under those assumptions, not an observed birth date.

Instead stipulate P₀ = 0.08 s within the same constant-A history. Then t = 1.8 × 10¹³ s, or about 570385.581 Julian years, which is 0.36τ_c. Both calculations use the same present timing inputs; their different initial-period assumption changes the inferred elapsed time.

Real torques need not follow this constant law, and abrupt timing changes, accretion or a changing moment of inertia can require another model. The example illustrates sensitivity to an initial condition rather than supplying a universal correction factor or a fitted age for an actual pulsar.

Original conditional timing ages for P = 0.1 s and intrinsic Ṗ = 10⁻¹⁵ s s⁻¹. Constant PṖ and moment of inertia are assumed. Years are Julian; no real object is dated.
Initial-period treatment	Elapsed-time result (s)	Elapsed Julian years; rounded
Neglect P₀/P; characteristic approximation	5 × 10¹³	1584404.391
Supply P₀ = 0.08 s	1.8 × 10¹³	570385.581

Correlation changes a local uncertainty calculation

Consider a separate forward calculation L = ηṀc² with positive input estimates η and Ṁ and exact c. Supply relative standard uncertainties u_r(η) = 0.05 and u_r(Ṁ) = 0.10. These are synthetic input assignments, not quantities inferred by fitting the same luminosity twice.

When local linearization is adequate, the relative output variance is approximately u_r²(L) = u_r²(η) + u_r²(Ṁ) +2r u_r(η)u_r(Ṁ), where r is the input correlation coefficient. The covariance term must not be dropped merely for convenience.

With r = 0, the result is √(0.05² +0.10²) ≈ 0.111803399. With r = 0.5, it is √(0.05² +0.10² +0.005) ≈ 0.132287566. The larger second value follows from the supplied positive correlation.

These are first-order relative standard uncertainties in the declared forward model. They do not automatically supply Gaussian coverage, confidence intervals or probabilities for competing remnant types. They also do not include an unspecified model discrepancy or make the efficiency–flow product uniquely identifiable.

Original first-order uncertainty propagation for positive η and Ṁ with relative input standard uncertainties 0.05 and 0.10. Exact c is used; no coverage interval or observational calibration is asserted.
Assumed correlation r	Relative variance; first order	Relative standard uncertainty; first order
0	0.0125	0.111803399
0.5	0.0175	0.132287566

A defensible result reports what remains conditional

Independent measurements can test whether a single physical model explains several kinds of evidence. Compatibility of a dynamical mass, a pulse profile and a spectrum can be informative when each inference uses appropriate assumptions and uncertainties.

Shared calibration or modelling choices can also link nominally separate estimates. Repeating one assumption across several analyses does not create several independent validations of that assumption. Compare the actual evidence and dependence structure.

A clear report identifies measured signals, calibrated quantities, adopted physical relations, fitted parameters, uncertainty sources and plausible alternatives. It distinguishes an upper or lower bound from a unique value and distinguishes a model-dependent interpretation from a directly recorded signal.

Compact remnants thus provide powerful tests of stellar and gravitational physics while requiring careful inference. The next chapter changes scale to our view inside the Milky Way, where distance, obscuration and observing selection again determine how a larger physical system is reconstructed.

Worked examples

Compare histories with the same present timing

Why does the supplied P₀ = 0.08 s give only0.36 of the characteristic time for present P = 0.1 s?

The integrated constant-PṖ result is t = τ_c[1 − (P₀/P)²].
The supplied ratio is 0.08/0.1 = 0.8.
Calculate 1 −0.8² = 0.36.

The elapsed model time is 0.36τ_c. The difference comes from the initial condition, not a change in the measured present period.

Retain the covariance term

What first-order relative variance follows from input uncertainties 0.05 and 0.10 with correlation 0.5?

The two squared contributions are 0.0025 and 0.0100.
The covariance contribution is 2 ×0.5 ×0.05 ×0.10 = 0.0050.
Their sum is 0.0175, whose square root is approximately 0.132287566.

The relative variance is 0.0175 and relative standard uncertainty about 0.132287566 in this declared local model; no coverage interval is thereby supplied.

Try the reasoning

Does a precisely measured signal make every remnant property inferred from it equally certain?
Show the answer and reasoning
No.
The physical connection can depend on additional uncertain parameters, model assumptions and calibration.
With L =ηṀc² and both positive η and Ṁ unknown, does a perfect L measurement alone identify each input?
Show the answer and reasoning
No.
Many efficiency–flow pairs give the same product, so additional constraints are needed.
May a known nonzero input correlation be omitted from uncertainty propagation simply to obtain a smaller result?
Show the answer and reasoning
No.
The covariance term belongs to the declared model and changes the propagated variance.
Does a calculated standard uncertainty automatically specify a Gaussian confidence or coverage interval?
Show the answer and reasoning
No.
A coverage interpretation requires further distributional and inferential assumptions.

What to carry forward

Measured signals and inferred remnant properties belong to different stages of reasoning.
Observational precision and physical identification are distinct limitations.
Initial conditions, covariance, selection and shared model assumptions must remain visible when evidence is combined.

References

Teaching resources and research supporting this lesson’s explanations and examples.