Composition gradients in Uranus, Neptune, and sub-Neptunes

We advocate for a more agnostic approach to modelling the interiors of Uranus, Neptune, and sub-Neptune exoplanets. In this talk, we examine what can and cannot be inferred from the limited observables currently available.It is well established that mass and radius alone provide insufficient information to meaningfully constrain the internal distributions of materials such as hydrogen and helium, water, rocky silicates, and iron. We show that this degeneracy persists even within the Solar System, despite precise measurements of the masses, radii, and even gravitational fields of Uranus and Neptune. Using interior models that do not impose layered structures or homogeneous adiabatic profiles, we demonstrate that both rock-dominated and water-dominated solutions remain consistent with existing constraints for the Solar System ice giants (rock-to-water mass ratios between 0.04-3.92 for Uranus and 0.20-1.78 for Neptune).  Their traditional classification as “ice giants” is therefore more historical than reflective of physical reality.For sub-Neptune exoplanets, the problem becomes substantially more severe because the available observational constraints are far more limited. We use K2-18 b, TOI-270 d, and LHS 1140 b as case studies and apply the same agnostic interior inference framework used for Uranus and Neptune. For K2-18 b and TOI-270 d, only weak upper limits can be robustly established: the hydrogen and helium mass fraction must remain below approximately 10% for K2-18 b and 5% for TOI-270 d, and neither planet can consist purely of iron. Beyond these simple limits, a broad range of interior structures remains admissible.We further show that many solutions excluded by homogeneous adiabatic models reappear once composition gradients are permitted. As an example, low-temperature and water-rich interiors can be artificially ruled out within simplified homogeneous models when observational uncertainties are reduced. When composition gradients are included however, these solutions become viable again, demonstrating that the apparent reduction in degeneracy arises primarily from modelling assumptions rather than from the data themselves.As a further example, we also investigate the effect of the assumed 1-bar temperature boundary condition. For K2-18 b, changing this assumption from 300 K to 600 K significantly alters the inferred compositions in homogeneous adiabatic models. However, this sensitivity is strongly reduced once composition gradients are allowed.LHS 1140 b, which lies more in the super-Earth rather than sub-Neptune regime, is more strongly constrained even in the presence of composition gradients. All viable models require at least ~70% combined rocky silicates and iron, less than ~30% water, and less than ~2% hydrogen and helium. Nevertheless, composition gradients still broaden the allowed ranges of internal temperature, entropy, and density profiles relative to homogeneous models.Overall, we argue that the dominant limitation in current interior inference is not observational precision, but the modelling assumptions themselves. Homogeneous adiabatic models systematically overestimate the information content of measurements and can create the illusion of tighter constraints where none physically exist. If composition gradients, mixed materials, and non-convective regions are allowed, the interiors of Uranus, Neptune, and sub-Neptunes remain substantially more degenerate than is often assumed. These results suggest that interior modelling should adopt agnostic frameworks by default, and that questions of formation, evolution, and habitability must be addressed with explicit recognition of the broad range of admissible internal states.