---
title: 'Sonora Elf Owl: Disequilibrium Chemistry Model'
url: https://www.emergentmind.com/topics/sonora-elf-owl
type: topic
---

# Sonora Elf Owl: Disequilibrium Chemistry Model

Sonora Elf Owl is the disequilibrium-chemistry branch of the Sonora substellar atmosphere model family, designed for brown dwarfs and self-luminous giant exoplanets. It comprises self-consistent, cloud-free, one-dimensional radiative–convective equilibrium atmospheres built with PICASO 3.0, with free metallicity, carbon-to-oxygen ratio, and eddy diffusion coefficient $K_{\rm zz}$, and was developed for spectra in regimes where vertical mixing drives departures from local chemical equilibrium. Within the Sonora series, it extends earlier cloud-free grids by coupling composition variation and quench chemistry over a broad parameter space, and version 2 further corrected the treatment of disequilibrium CO$_2$ and removed PH$_3$ opacity from the released spectra [2402.00756, 2505.03994].

## 1. Position within the Sonora model family

Sonora Elf Owl is the fourth major release in the Sonora Substellar Atmosphere Models series. Sonora Bobcat provides cloud-free rainout equilibrium atmospheres spanning $200$–$2400$ K with metallicity and C/O variations but no disequilibrium chemistry. Sonora Cholla adds cloud-free disequilibrium chemistry at solar composition for approximately $500$–$1300$ K. Sonora Elf Owl extends disequilibrium chemistry self-consistently over $275$–$2400$ K, multiple gravities, seven orders of magnitude in $K_{\rm zz}$, sub- to super-solar metallicities, and C/O from oxygen-rich to carbon-rich. Opacity mixing is performed “on-the-fly” for all gases, which improves self-consistency relative to previous grids.

Its intended regime is the cloud-free low-temperature atmosphere domain, especially mid- and late-T dwarfs and cold Y dwarfs, where condensates are largely absent from the photosphere. Later applications explicitly describe it as a cloud-free grid optimized especially for T–Y dwarfs. The same cloud-free assumption also defines its principal failure mode: it does not reproduce dusty L-dwarf spectral energy distributions or the L/T transition, where condensate opacity and cloud clearing modify the continuum shape and band depths. In those regimes, studies comparing Sonora Elf Owl with cloudy alternatives report that cloud-free models compensate by pushing metallicity and C/O toward grid edges or by returning physically problematic surface gravities, while residuals remain large [2402.00756, 2505.19993, 2510.02026].

## 2. Physical framework and disequilibrium chemistry

The grid is formulated as a $1$D, plane-parallel atmosphere with $90$ pressure layers under hydrostatic equilibrium,
$$
\frac{dP}{dz} = -\rho g.
$$
PICASO iterates the temperature–pressure profile until radiative–convective equilibrium is achieved. Convective energy transport is handled by mixing-length theory inside the equilibrium solver, while disequilibrium abundances are imposed through vertical mixing parameterized by a constant-with-pressure eddy diffusion coefficient $K_{\rm zz}$ for each model.

Tracer transport is represented by an eddy-diffusion flux,
$$
F_i = -K_{\rm zz}\, n \frac{d f_i}{dz},
$$
where $n$ is the total number density and $f_i$ the mixing ratio of species $i$. The vertical mixing timescale is
$$
\tau_{\rm mix} = \frac{H^2}{K_{\rm zz}}, \qquad H=\frac{k_B T}{\mu m_H g},
$$
and quenching occurs where
$$
\tau_{\rm mix}(P_Q)=\tau_{\rm chem}(P_Q).
$$
Below the quench pressure, equilibrium is maintained; above it, abundances are frozen in.

The disequilibrium network includes CH$_4$, CO, CO$_2$, NH$_3$, N$_2$, HCN, PH$_3$, and H$_2$O in the original release. The governing net reactions include
$$
{\rm CH_4 + H_2O \rightleftarrows CO + 3H_2},
$$
$$
{\rm CO + H_2O \rightleftarrows CO_2 + H_2},
$$
$$
{\rm 2NH_3 \rightleftarrows N_2 + 3H_2},
$$
$$
{\rm CH_4 + NH_3 \rightleftarrows HCN + 3H_2},
$$
$$
{\rm 2CO + N_2 + 3H_2 \rightleftarrows 2HCN + 2H_2O},
$$
$$
{\rm CO + NH_3 \rightleftarrows HCN + H_2O},
$$
and, for phosphorus,
$$
{\rm P_4O_6 + 12H_2 \rightleftarrows 4PH_3 + 6H_2O}.
$$
Chemical timescales are taken from Zahnle & Marley (2014), with Visscher & Fegley formulations for phosphorus chemistry. Equilibrium abundances below quench levels are obtained from precomputed tables across [M/H] and C/O, using solar elemental abundances from Lodders (2009), and C/O is varied by holding C+O fixed and changing their ratio before metallicity scaling so that C/O and [M/H] are not conflated [2402.00756].

## 3. Grid architecture and parameterization

The full Sonora Elf Owl grid spans effective temperature $T_{\rm eff}=275$–$2400$ K, surface gravity $\log g=3.25$–$5.5$ in cgs, $\log_{10}(K_{\rm zz}/{\rm cgs})=2,4,7,8,9$, metallicities [M/H] $=-1.0,-0.5,0.0,+0.5,+0.7,+1.0$, and C/O $=0.22,0.458,0.687,\approx1.12$–$1.14$. The temperature sampling is $25$ K from $275$–$600$ K, $50$ K from $600$–$1000$ K, and $100$ K from $1000$–$2400$ K. The atmosphere files provide $T(P)$ structures, equilibrium and quenched abundance profiles, and emission spectra from $0.5$–$30\,\mu{\rm m}$ at $R=5000$, for a total of $43{,}200$ models.

Published fitting studies use subsets or instrument-matched interpolants rather than the raw grid alone. In a comparative study of low-temperature atmosphere fitting methodologies, the working Sonora Elf Owl subset covered $T_{\rm eff}=600$–$2400$ K, $\log g=4.5$–$5.5$, [M/H] $=-0.5$ to $+0.5$, C/O $=0.5$–$1.5$ relative to solar, and $\log \kappa_{\rm zz}=2$–$9$, with $5175$ models after down-selection. In the JWST study of $20$ cold brown dwarfs, forward modeling used linear interpolation across eight fitted parameters: $T_{\rm eff}$, $\log g$, [M/H], C/O, $\log K_{\rm zz}$, two MIRI wavelength-correction parameters, and a global flux-scaling parameter $\log(R^2/D^2)$. In the later JWST/NIRSpec distant-brown-dwarf survey, Sonora Elf Owl version 2 was interpolated and convolved to PRISM/CLEAR resolution over $0.6$–$5.3\,\mu{\rm m}$, with flat priors on the physical parameters and a logarithmic prior on the scale factor [2402.00756, 2409.19191, 2510.02026].

## 4. Secondary CO$_2$ quenching and the version 2 correction

Version 1 applied quench chemistry species-by-species: for each species $i$, a quench pressure $P_{q,i}$ was obtained from $\tau_{\rm chem}=\tau_{\rm mix}$, the atmosphere was assumed to remain in chemical equilibrium for $P>P_{q,i}$, and the volume mixing ratio for $P<P_{q,i}$ was held constant at the equilibrium value evaluated at $P_{q,i}$. This procedure worked well for species such as CO, CH$_4$, and NH$_3$, but it systematically underpredicted CO$_2$ because the observable CO$_2$ abundance is controlled by equilibrium with CO and H$_2$O, while CO commonly quenches deeper than CO$_2$.

The controlling reaction is
$$
{\rm CO_2 + H_2 \rightleftarrows CO + H_2O},
$$
with mass-action relation
$$
f_{\rm CO_2} = K_{\rm eq}(T)\frac{f_{\rm CO}f_{\rm H_2O}}{f_{\rm H_2}},
$$
and
$$
K_{\rm eq} = 18.3\exp\!\left(-\frac{2376}{T} - \left(\frac{932}{T}\right)^2\right).
$$
For the CO$\leftrightarrow$CH$_4$ system, the characteristic timescale used to locate the CO quench level is
$$
\tau_{\rm CO}=3\times10^{-6}P^{-1}\exp(42000/T),
$$
with $\tau_{\rm CO}$ in seconds, $P$ in bars, and $T$ in Kelvin. The version 2 correction therefore adopts “secondary quenching”: first determine the CO and H$_2$O quench level, freeze their abundances there, then compute CO$_2$ in equilibrium with those already-quenched abundances until CO$_2$ itself quenches, after which CO$_2$ is frozen.

Operationally, version 2 implements four steps. First, compute $\tau_{\rm mix}$ from $K_{\rm zz}$ and $H$ and locate quench pressures for CO/H$_2$O and CO$_2$. Second, at $P_{q,{\rm CO}}$, set $f_{\rm CO}$ and $f_{\rm H_2O}$ above that level equal to their equilibrium values at $P_{q,{\rm CO}}$. Third, for pressures between $P_{q,{\rm CO}}$ and $P_{q,{\rm CO_2}}$, set
$$
f_{\rm CO_2}(P)=K_{\rm eq}(T(P))\frac{f_{\rm CO}^q f_{\rm H_2O}^q}{f_{\rm H_2}(P)}.
$$
Fourth, freeze CO$_2$ at its own quench level:
$$
f_{\rm CO_2}(P<P_{q,{\rm CO_2}})=f_{\rm CO_2}(P_{q,{\rm CO_2}}).
$$

The quantitative effect is substantial for the coldest objects. The correction increases CO$_2$ by over one order of magnitude for objects with $T_{\rm eff}\lesssim600$ K and by up to two orders of magnitude in some cases. In an illustrative $500$ K brown dwarf model with $g=31\,{\rm m\,s^{-2}}$, metallicity $0.1\times$ solar, solar C/O, and vertically constant $K_{\rm zz}=10^8\,{\rm cm^2\,s^{-1}}$, a full-kinetics Photochem calculation shows CO quenching near $P\approx20$ bar and CO$_2$ quenching near $P\approx7$ bar, with CO$_2$ freezing at $f_{\rm CO_2}\approx10^{-10}$. In that case, version 1 underpredicted CO$_2$ above $\sim7$ bar by roughly two orders of magnitude. Version 2 also removes PH$_3$ opacity because observations indicated that version 1 consistently produced too much PH$_3$ absorption. The updated abundances and spectra were posted as new versions of the original Zenodo records, but the radiative–convective climate simulations were not re-run, so the published version 2 pressure–temperature profiles are not strictly self-consistent with the corrected composition [2505.03994].

## 5. Spectral behavior, parameter sensitivities, and degeneracies

The influence of $K_{\rm zz}$ on the thermal structure and emergent spectrum depends strongly on metallicity and $T_{\rm eff}$. At solar metallicity, vigorous mixing with $K_{\rm zz}\approx10^9$ cools the $T(P)$ profile by approximately $100$ K relative to chemical equilibrium. At [M/H] $=+1.0$, the same mixing can cool the profile by approximately $300$–$400$ K. At [M/H] $=-1.0$, the effect on $T(P)$ is minimal because the atmosphere is more transparent and disequilibrium versus equilibrium optical depths differ by less than an order of magnitude across most wavelengths. The temperature range of strongest sensitivity also shifts with composition: metal-poor atmospheres are most $K_{\rm zz}$-sensitive at approximately $1200$–$1800$ K, solar-composition atmospheres at approximately $900$–$1600$ K, and super-solar atmospheres at approximately $500$–$1200$ K.

Key wavelength regions encode different combinations of physics. The $3.0$–$4.0\,\mu{\rm m}$ region is dominated by CH$_4$; increasing $K_{\rm zz}$ weakens CH$_4$ bands by lowering CH$_4$ aloft, higher metallicity can also weaken them, and high C/O deepens them. The $4.0$–$4.5\,\mu{\rm m}$ region contains CO$_2$, which grows strongly with metallicity and has weaker dependence on $K_{\rm zz}$ and C/O; this is the principal band for breaking the $K_{\rm zz}$–[M/H] degeneracy. The $4.5$–$4.8\,\mu{\rm m}$ region contains the CO fundamental and is degenerate because both higher $K_{\rm zz}$ and higher metallicity enhance CO absorption. The $1.1$–$2.0\,\mu{\rm m}$ region is dominated by H$_2$O and alkali line wings and is more sensitive to metallicity than to $K_{\rm zz}$. The $8.5$–$12\,\mu{\rm m}$ region probes NH$_3$ and becomes important in combined near- and mid-infrared analyses.

Synthetic fitting tests define a practical observing strategy. M band alone at $R\sim100$ generally cannot uniquely constrain $K_{\rm zz}$, [M/H], C/O, or $\log g$. Coverage from $5$–$14\,\mu{\rm m}$ can constrain $T_{\rm eff}$, $K_{\rm zz}$, [M/H], and C/O, but $\log g$ remains loose. Coverage from $4$–$14\,\mu{\rm m}$ constrains all parameters, including gravity, and JWST NIRSpec Prism with broad $0.6$–$5.3\,\mu{\rm m}$ coverage yields the tightest posteriors in the published tests. In the original nine-object T-dwarf application, inferred $K_{\rm zz}$ values were consistently low, approximately $10^1$–$10^4\,{\rm cm^2\,s^{-1}}$, and were interpreted as evidence that the observations probe quenching in a deep detached radiative zone rather than vigorous convective transport [2402.00756].

## 6. Inference workflows and observational performance

Sonora Elf Owl has been used in both direct grid fitting and machine-learning-assisted inference. In benchmark SpeX prism analyses at $0.8$–$2.5\,\mu{\rm m}$ and $R\approx150$, one study compared a Markov Chain Monte Carlo algorithm that interpolates across spectral fluxes with a Random Forest Retrieval trained on the grid. The MCMC method uses a $\chi^2$ statistic with an optimal multiplicative scale factor $\beta$, linear interpolation in log flux over logarithmic parameter axes, a Metropolis–Hastings acceptance scheme, and posterior sampling under
$$
L(\theta)\propto\exp\!\left(-\tfrac12\chi^2(\theta)\right), \qquad p(\theta|D)\propto p(D|\theta)p(\theta).
$$
The Random Forest Retrieval uses absolute-calibrated flux densities as features and predicts $T_{\rm eff}$, $\log g$, [M/H], C/O, $\log\kappa_{\rm zz}$, and a flux-scaling parameter $f$, with ensemble averaging
$$
\hat{y}=\frac{1}{N_{\rm trees}}\sum_t T_t(x).
$$
That study found the MCMC approach to yield higher fit quality and more precise parameters, while the Random Forest Retrieval was orders of magnitude faster after training. The recommended workflow was therefore rapid multi-grid screening with the random forest followed by MCMC refinement on the chosen grid. In that benchmark sample, Sonora Elf Owl was generally optimal for mid- and late-T dwarfs and not preferred for L dwarfs or many early-T dwarfs, because the absence of cloud opacity is decisive in the latter regimes [2505.19993].

JWST applications confirm both the utility and the systematic structure of the grid. In a sample of $20$ T and Y dwarfs fitted jointly to NIRSpec CLEAR/PRISM and MIRI LRS spectra, Sonora Elf Owl and ATMO2020++ returned relatively consistent effective temperatures, typically differing by tens of Kelvin, and both yielded metallicities close to solar values for nearby objects. The largest disagreement was in surface gravity: Sonora Elf Owl values were typically about $1$ dex lower than ATMO2020++, with $12$ of $20$ objects at the lower Elf Owl grid boundary $\log g=3.25$ and $17$ objects having $\log g<4$. Radii inferred from the scaling parameter and parallax were typically $\sim0.8$–$1.2\,R_{\rm Jup}$, and the corresponding masses and ages derived from evolutionary tracks indicated that most objects were below $30\,M_{\rm Jup}$ and younger than $6$ Gyr, with Y dwarfs at $2$–$20\,M_{\rm Jup}$ and $0.1$–$6.7$ Gyr. The same study found that NIRSpec-only fits usually recovered $T_{\rm eff}$ and $\log g$ close to the combined-spectrum values, whereas MIRI-only fits gave more uncertain and often lower gravities unless $T_{\rm eff}$ was fixed [2409.19191].

A later JWST/NIRSpec PRISM/CLEAR survey of $41{,}283$ public spectra used Sonora Elf Owl version 2 alongside LOWZ and SAND. In that pipeline, the spectra were truncated at the blue end where ${\rm S/N}<5$, cross-calibrated to NIRCam photometry to mitigate MOS mis-centering, convolved to the wavelength-dependent PRISM resolution, and analyzed with UltraNest nested sampling using a jitter term in the likelihood. Sonora Elf Owl version 2 performed strongly for cool T and Y dwarfs, but cloud-free grids failed to reproduce L/T transition spectra, whereas SAND provided a better match in those cases. For the metallicity-gradient analysis in that survey, Sonora-based results showed no evident trend of [M/H] with Galactic height $|Z|$ after excluding cluster sources, L/T transition objects, and objects with $T_{\rm eff}<600$ K [2510.02026].

## 7. Limitations, systematic biases, and future development

The main limitations of Sonora Elf Owl follow directly from its simplifying assumptions. It is a cloud-free, one-dimensional grid with constant-with-pressure $K_{\rm zz}$ in each model. Real atmospheres can have pressure-dependent mixing that differs by orders of magnitude between radiative and convective zones, and clouds can alter both the thermal profile and quenched abundances by warming deep adiabats. These omissions are central to the grid’s poor performance for L dwarfs, the L/T transition, and some early-T objects.

Several systematic biases recur across applications. Surface gravities inferred with Sonora Elf Owl are often lower than those from ATMO2020++ or from evolutionary expectations derived from benchmark primaries. In the JWST $20$-object study, this discrepancy was attributed to differences in convective treatment and opacity databases, particularly the ATMO2020++ use of an effective adiabatic index $\gamma_{\rm eff}=1.25$, which reddens colors and lowers J-band peaks in a way partly degenerate with gravity. In benchmark SpeX fitting, late-T dwarf gravities from Sonora Elf Owl were again systematically low relative to expectations, a behavior the authors associated with pressure-dependent chemistry or collision-induced H$_2$ absorption modeling. For very cold objects, metallicities derived from NIRSpec-only Sonora fits can differ from those obtained with combined NIRSpec+MIRI coverage, and both C/O and $\log K_{\rm zz}$ are especially sensitive to interpolation error and spectral mismatch.

The uncertainties quoted in fits are also often optimistic if only observed flux errors are propagated. Published interpolation tests show that most parameters behave well under linear interpolation, but the $4$–$5\,\mu{\rm m}$ region is more nonlinear because CO$_2$, CO, CH$_4$, and H$_2$O overlap there, and biases are strongest for $\log K_{\rm zz}$ and C/O. The large NIRSpec survey therefore added half-grid spacing as an uncertainty term and used a free jitter parameter; including that jitter increased posterior uncertainties by approximately $20$–$200\%$ depending on the parameter.

Version 2 resolves the specific CO$_2$ underprediction in the original release and removes PH$_3$ opacity, but it does not yet provide fully self-consistent recomputed pressure–temperature profiles. The correction note states that a fully self-consistent grid incorporating corrected CO$_2$ will be released in future work. Other suggested developments across the literature include more finely sampled grids, improved interpolation or emulation, pressure-sensitive opacity revisions, methane-chemistry updates, expansion or validation of metallicity coverage for NIRSpec-only analyses, and incorporation of condensate physics across the L/T transition [2505.03994, 2505.19993, 2409.19191].

Source: https://www.emergentmind.com/topics/sonora-elf-owl