Characterizing the interior of a rocky exoplanet rests on a surprisingly thin observational foundation: almost everything scientists can infer about a distant planet's core, mantle, and possible volatile layers begins with just two measured numbers, its mass and its radius. From that sparse starting point, researchers have developed a toolkit of methods—mass-radius scaling relationships, Bayesian inference models, and machine-learning approaches—that trade simplicity for detail. This guide compares what each method requires, what it can reveal, and where it fails, with a central caveat that recurs throughout: different interior compositions can produce the same observed mass and radius, so no current method yields a unique answer.
Introduction to Exoplanet Interior Characterization
The goal of studying exoplanet interiors, as Sara Seager's research group frames it, is "to learn what individual exoplanets are made of, and even what their internal structure might be." That ambition collides immediately with a hard constraint: only a planet's mass and radius can be measured directly, with no other information about the interior available from a distance. Mass and radius, as a review of interior modeling fundamentals explains, "encapsulate information about the integrated interior structure and composition of an exoplanet" and can constrain first-order characteristics—but they compress an entire planet into two numbers.
Everything beyond those numbers is inference, and the methods below differ chiefly in how much structure they can responsibly extract from that inference, and how honestly they report their own uncertainty. This article focuses on terrestrial and rocky planets, the domain where these methods are developed and applied.
Foundational Observational Data: Mass and Radius
Planetary mass is measured through radial velocity or transit timing variations, while radius comes from transit photometry—the slight dimming of the host star as the planet crosses its face. These measurements inherit their precision from our knowledge of the host star itself. Stellar parameters impose strict limits on what is achievable: the best uncertainties are approximately 2% for stellar radius and 5–10% for stellar mass, values that depend on the star's type, age, and the detection method used, according to the interior modeling review. Those stellar uncertainties propagate directly into every derived planetary quantity, which means every downstream interior estimate carries an irreducible observational floor. As the review of terrestrial exoplanet interiors notes, astrophysical observations are generally expensive, especially from space missions, so thorough and careful use of the available data is essential for comprehensive interior characterization.
The Core Challenge: Parameter Degeneracy
Seager's group is blunt about the problem, describing "fatal limitations to uncovering the interior composition of an exoplanet because only the mass and radius can be measured." The technical name for this obstacle is degeneracy: multiple combinations of core size, mantle composition, and volatile content can produce the same observed mass and radius. The terrestrial exoplanet review states that degeneracy of interior parameters is "generally large," making it difficult to identify a specific interior type such as a purely rocky composition. The problem is not merely theoretical.
Even with excellent data, planetary mass and radius alone are insufficient to distinguish between thick gas layers and oceans, and quantifying water mass fractions below 1%—the level needed to identify Earth-like planets with liquid surface water and exposed continents—lies far below current observational uncertainties on planetary mass. Degeneracy is the reason the field has moved from simple scaling laws to statistical and computational methods.
Method 1: Mass-Radius Relationships
The oldest and most widely used approach compares a planet's measured mass and radius against theoretical composition curves. Mass-radius scaling laws have long formed a cornerstone of exoplanet interior modeling, and these composition curves can be well approximated by power laws of the form R ∝ Mβ, where the exponent β is derived by fitting the power law to interior models. For rocky planets, typical values of β fall between 0.25 and 0.3, according to the interior modeling review citing work by Valencia and colleagues and by Sotin.
The method's strength is its simplicity: from two numbers, a researcher can estimate bulk density and place a planet into a broad compositional category—rocky, iron-rich, or water-rich. Its weakness is resolution. A mass-radius comparison cannot resolve specific internal layers, detailed mineralogy, or the difference between an ocean and a thick atmosphere. It is foundational, but as Seager's group emphasizes, often insufficient on its own for detailed characterization.
Method 2: Bayesian Inference Models
Bayesian inference attacks the degeneracy problem statistically rather than trying to eliminate it. Inversion methods based on Bayesian techniques, developed by researchers including Rogers and Seager and by Dorn and colleagues, constrain the interior structure and composition of rocky exoplanets by comparing observations against a large suite of theoretical interior models. As a 2023 Astrophysical Journal Supplement Series paper by Zhao, Ni, and Liu explains, Bayesian inference analysis is suitable for nonlinear and degenerate problems with limited observed data. The inputs are planetary mass and radius, often supplemented by host star refractory element abundances, which inform prior distributions about the planet's building blocks.
The outputs are probability distributions rather than single values: core-mantle ratios, bulk composition, and the possible presence of volatile layers. The trade-offs are real. The calculations are computationally intensive, the quality of prior knowledge significantly influences the results, and the choice of equations of state—the physics linking pressure, temperature, and density—can shape the inferred interior.
Method 3: Machine Learning Models
Machine learning offers a way to make interior inference dramatically faster. Zhao, Ni, and Liu's 2023 work uses mixture density networks (MDNs) and deep neural networks trained on large libraries of pre-computed interior models. Their MDN takes planet bulk parameters as inputs—mass, radius, the bulk molar ratio Fe/(Mg + Si), and the static Love number k2, which measures how a planet deforms under tidal forces—and outputs the radial fractions of each layer and properties at the core-mantle boundary, including the water radial fraction, mantle radial fraction, core radial fraction, core mass fraction, and the pressure and temperature at the boundary.
A companion deep neural network trained on mass, radius, and bulk molar ratios Fe/Mg and Mg/Si predicts static Love numbers, with its accuracy tracked through mean squared error loss across training epochs. The approach trades the heavy per-planet computation of Bayesian sampling for an upfront training cost, but it remains dependent on the same underlying model physics and inherits the degeneracy of the problem it emulates.
Comparative Overview of Exoplanet Interior Characterization Methods
The table below compares the three computational approaches across uniform criteria: what each requires, what it yields, and where it breaks down. The pattern that emerges is that mass and radius are universal inputs, while computational models buy more detailed structural insight at the price of complexity and dependence on prior assumptions.
| Method | Observational Inputs | Derived Parameters | Key Limitations |
|---|---|---|---|
| Mass-radius relationships | Planetary mass (radial velocity or transit timing variations) and radius (transit photometry); uncertainties propagate from stellar parameters | Bulk density and broad compositional categories (rocky, iron-rich, water-rich) via power-law scaling with β ≈ 0.25–0.3 for rocky planets | Parameter degeneracy—different compositions yield the same mass and radius; limited resolution of internal layers or mineralogy; foundational but often insufficient |
| Bayesian inference models | Mass and radius, often supplemented by host star refractory element abundances informing priors | Probability distributions for core-mantle ratios, bulk composition, and volatile layers | Computationally intensive; results shaped by prior assumptions and choice of equations of state; degeneracy persists, especially with large observational uncertainties |
| Machine learning models (MDN/DNN) | Mass, radius, bulk molar ratios Fe/(Mg + Si) or Fe/Mg and Mg/Si, and static Love number k2 | Radial fractions of water, mantle, and core layers; core mass fraction; core-mantle boundary pressure and temperature | Depends on training libraries of interior models, inheriting their physics and degeneracies; accuracy limited by training data coverage |
Read side by side, the comparison shows a clear progression: the mass-radius relationship gives a first-order classification in minutes; Bayesian inference converts the same data plus stellar abundances into structured probability distributions; and machine-learning models extend the input set with quantities like the Love number to map layer-by-layer structure at a fraction of the computational cost of full sampling. None of them escapes the underlying problem, because all of them are constrained by the same two fundamental measurements and the stellar uncertainties attached to them.
Navigating the Uncertainties of Exoplanet Interiors
Scientists must continue to refine observational precision and develop more sophisticated computational models to reduce the impact of parameter degeneracy and improve the accuracy of exoplanet interior characterization. The path forward is visible in the evidence itself: future studies, as the terrestrial exoplanet review anticipates, will quantify and compare the merit of different data types and determine the precision needed to further enhance constraints on planetary composition and structure. The measurable indicators to watch are improved precision in mass and radius measurements from next-generation telescopes, and the development of new modeling techniques that better constrain interior parameters. Until those arrive, every interior picture of a rocky exoplanet should be read as a set of probabilities, not a portrait.
