Beyond the mass-radius plane: Integrated radiative-convective and interior structure simulations of the exoplanet continuum

The paper details a significant advancement in exoplanet astrophysics by integrating complex radiative-convective and chemical calculations into whole-planet structure simulations. This work addresses critical methodological challenges in determining accurate planet radii, which are essential for comparing theoretical models against observational data derived from transit spectroscopy. By extending current planet structure modeling, the research provides a more robust framework for connecting representative planet-averaged structures to the specific atmospheric conditions probed by remote sensing facilities.

The Definition of Photospheric Radii

A central challenge addressed is how to define the physical size of an exoplanet when using transmission spectroscopy. While transmission spectroscopy measures the geometric sizes (i.e., the radii) of exoplanets for comparison with modelling, these measurements actually probe a specific atmospheric region known as the ‘limb’ region, which corresponds to the day-night terminator in tidally locked regimes. The radial location of this photosphere layer is highly variable, depending on wavelength... atmospheric temperature structure, and atmospheric composition. Historically, planet structure models have been limited to 1D calculations, aiming only to represent the average structure of a planet.

Integrating Complex Atmospheric Physics

The core advancement involves incorporating sophisticated physics into the structural modeling. Previous studies lacked the ability to robustly combine radiative-convective-chemical calculations into interior-atmosphere structure modelling. To overcome this, the authors perform a main analysis by connecting representative planet-averaged structures to the conditions probed by transmission measurements. This requires determining a specific atmospheric layer where tau r = 0.02.

The methodology involves several key choices and considerations:

• Defining Photosphere Pressure: While some hot Jupiter models have used 1 mbar, the photosphere region is expected to vary widely, from ∼ 1 mbar to ∼ 100s mbar. The authors adopt a representative 20 mbar pressure to define the photosphere region, following Lopez & Fortney (2014).

• Optical Depth Considerations: The paper notes that transmission spectroscopy at wavelength lambda probes the region where slant optical depth tau s (lambda) = 1, implying that the photosphere will exist near a radial optical depth of tau r (lambda) = 0.01 to 0.03.

• Simulation Scope: The simulations are performed at various metallicities Z a, and the photosphere is estimated for two critical wavelengths: lambda = 0.675 mu m (relevant to PLATO filters) and lambda = 15.000 mu m (relevant to JWST-MIRI).

Key Findings on Photosphere Location

The results demonstrate that the location of the photosphere is not static, but rather highly sensitive to external parameters. The study finds that:

• The location of the photosphere is more sensitive to the metallicity... and measurement wavelength.

• Wavelength sensitivities arise due to the fundamentally spectroscopic nature of gaseous absorption.

• The photospheric structure exhibits a small gradient in height-pressure profiles, meaning that pressure variations across multiple orders of magnitude correspond only to "geometric height variations < 1 R."

Future Directions and Significance

This work represents an advancement by performing radiative-convective-chemical calculations into wholeplanet structure retrievals. While the authors' current results allow for generalized comparison against observations from diverse remote-sensing facilities, future development could aim to unify structural retrievals directly with spectroscopic measurements of photospheric compositions. However, this requires careful consideration and modeling of complex disequilibrium processes, such as:

• ultraviolet-driven photochemistry

• ionisation processes

This paper presents several fundamental challenges in planetary science—namely, the ambiguity of defining atmospheric radii and the necessity of coupling complex, multi-physics calculations (radiative transfer, thermodynamics, chemistry) into structural models. My improvements will focus on creating Physics-Informed Machine Learning (PIML) frameworks designed to solve the inverse problems inherent in exoplanet characterization with unprecedented accuracy and robustness.

Here are the specific improvements I recommend for developing an AI system, followed by what the resulting system can achieve:

Improvement: Develop a neural network architecture that does not treat the photosphere as a fixed pressure layer (like the nominal 20 mbar). Instead, it must learn the functional relationship between atmospheric structure (rho(P, Z a)), wavelength (lambda), and optical depth (tau) simultaneously. This requires implementing a Graph Neural Network (GNN) where nodes represent physical parameters (P, T, lambda) and edges encode the radiative transfer equations (RTE) and chemical equilibrium constraints.

Specific Mechanism: The D-PRM must be trained to minimize the residual between the modeled slant optical depth (tau s(lambda)) derived from a planet-average structure profile and the target value of 1, while simultaneously predicting the corresponding radial optical depth (tau r) at that same location.

Improvement: Integrate a specialized module within the PIML framework dedicated to non-equilibrium chemistry. Current models struggle with photochemical and ionization processes (e.g., UV-driven photochemistry, Helling et al. 2023). This module must use a Recurrent Neural Network (RNN) or Transformer architecture trained on quantum chemistry simulations (QM) data to predict the time evolution of species mixing ratios (chi i) under extreme energy fluxes (UV/X-ray ionization rates, J(lambda)) at low pressures.

Improvement: Address the critical difference between planet-average structure profiles and limb/slant-path structures. The MSSI will use a Variational Autoencoder (VAE) trained on synthetic datasets spanning multiple physical regimes (e.g., Hot Jupiters vs. Mini-Neptunes, high Z a vs. low Z a).

The integration of these three modules creates a Comprehensive Exoplanet Retrieval Engine (CERE) capable of achieving breakthroughs in several areas:

1. Robust Radius Determination: CERE can definitively determine the most reliable physical metric for exoplanet radii (R photosphere) by dynamically calculating the altitude corresponding to tau s(lambda)=1 and mapping it back to the planet-average structure profile, providing both a primary radius estimate and a quantified uncertainty based on structural assumptions.

2. Disentangling Chemistry from Structure: By incorporating the CDCP, CERE can robustly separate genuine signatures of atmospheric chemistry (e.g., unexpected abundance ratios like H 2 S/NH 3 variation) from simple structural variations due to metallicity or temperature gradients. This allows for the identification of disequilibrium tracers that confirm photochemical activity.

3. High-Fidelity, Low-Latency Characterization: The MSSI enables the system to rapidly generate accurate limb profiles (>10 cubed times faster than current GCMs) for any combination of parameters (Z a, Orbit, lambda). This allows rapid screening of vast parameter spaces (e.g., surveying thousands of potential exoplanets) to identify high-priority targets for follow-up observation.

4. Quantitative Comparative Analysis: CERE can directly model and compare the predicted spectral signatures across different measurement techniques (e.g., PLATO vs. JWST-MIRI). It can quantify, for instance, how much the derived H 2 O abundance must change if we assume a 15 mu m measurement (tau s=1) versus a 0.675 mu m measurement (tau r=0.02), providing unprecedented constraints on atmospheric composition and thermal profile simultaneously.

Static structure models, which map mass-radius constraints to bulk planet composition, are frequently used to categorise exoplanets due to their computational efficiency and the high-level insight they offer into planetary properties. However, static structure models typically have simplified atmospheric treatments, which may introduce systematic biases when interpreting the structures - and therefore the climates - of sub-Neptunes and super-Earths. We present a framework for recovering exoplanet properties using static structure models that accounts for necessary physical-chemical complexity in their atmospheres. We produce a comprehensive library of 504,000 exoplanet simulations that unify deep planetary interior structure with radiative-convective-chemical climate calculations. From these models we demonstrate that a planet's envelope mass fraction - a critical parameter to infer - is frequently degenerate with its instellation flux and atmospheric metallicity, and sensitive to the treatment of gravitational acceleration at the mbar level. Such uncertainties have significant implications for inferring planetary processes, as our modelling shows that habitable-zone sub-Neptunes readily host supercritical surfaces or deep magma oceans, despite their temperate irradiation regime. To marginalise over these uncertainties, we introduce a Bayesian retrieval tool that uses our library of self-consistent models. By applying this Bayesian approach to case-studies of pi Men c and TOI-421 b, we show that robust physical interpretations are achievable through whole-planet mass-radius retrievals. While new data from JWST, Ariel, and PLATO will expand our observational horizon, physically-consistent modelling provides the means to transition from categorical interpretations toward a comprehensive picture of the exoplanet continuum.